drm/ipvr is a new GEM driver for baytrail's vxd392, which accelerates VP8 video decoding. The driver name "ipvr" means the PowerVR's IP wrapped by Intel. In the future, ipvr may support other platforms such as Merrifield. Code is placed at drivers/gpu/drm/ipvr and the following two new Kconfig are added: CONFIG_DRM_IPVR: Build option for ipvr module CONFIG_DRM_IPVR_EC: Experimental feature of error concealment
User mode drm helper "libdrm_ipvr.so" and simple test are also included.
Yao Cheng (3): [1/3] drm/i915: add vxd392 bridge in i915 on baytrail [2/3] drm/ipvr: ipvr drm driver for vxd392 [3/3] libdrm: user mode helper for ipvr drm driver
Setup following resources during i915_driver_load: 1. create a child platform and resource 2. allocate a new IRQ line and irq chip 3. set up IRQ mask/unmask callbacks vxd392 driver (if installed) will bind itself to the platform device and create new drm device
Signed-off-by: Yao Cheng yao.cheng@intel.com --- drivers/gpu/drm/i915/i915_dma.c | 98 +++++++++++++++++++++++++++++++++++++++ drivers/gpu/drm/i915/i915_drv.h | 6 +++ drivers/gpu/drm/i915/i915_irq.c | 70 ++++++++++++++++++++++++++++ drivers/gpu/drm/i915/i915_reg.h | 4 ++ drivers/gpu/drm/i915/i915_trace.h | 15 ++++++ 5 files changed, 193 insertions(+)
diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c index 85d14e1..73c78d1 100644 --- a/drivers/gpu/drm/i915/i915_dma.c +++ b/drivers/gpu/drm/i915/i915_dma.c @@ -85,6 +85,96 @@ intel_read_legacy_status_page(struct drm_i915_private *dev_priv, int reg) #define READ_BREADCRUMB(dev_priv) READ_HWSP(dev_priv, I915_BREADCRUMB_INDEX) #define I915_BREADCRUMB_INDEX 0x21
+static int valleyview_ved_init(struct drm_device *dev) +{ + int ret; + int irq = -1; + struct resource *rsc = NULL; + struct drm_i915_private *dev_priv = dev->dev_private;; + + dev_priv->ved_platdev = platform_device_alloc("ipvr-ved", -1); + if (unlikely(!dev_priv->ved_platdev)) { + DRM_ERROR("Failed to allocate VED platform device\n"); + ret = -ENOMEM; + goto err; + } + + rsc = kzalloc(sizeof(*rsc) * 3, GFP_KERNEL); + if (unlikely(!rsc)) { + DRM_ERROR("Failed to allocate resource for VED platform device\n"); + ret = -ENOMEM; + goto err; + } + + /* init IRQ number and chip/callbacks */ + irq = irq_alloc_descs(-1, 0, 1, 0); + if (unlikely(irq < 0)) { + DRM_ERROR("Failed to allocate IRQ number: %d\n", irq); + ret = -ENOMEM; + goto err; + } + + ret = valleyview_initialize_ved_irq(dev, irq); + if (unlikely(ret)) { + DRM_ERROR("Failed to initialize VED IRQ: %d\n", ret); + goto err; + } + + dev_priv->ved_irq = irq; + rsc[0].start = rsc[0].end = irq; + rsc[0].flags = IORESOURCE_IRQ; + rsc[0].name = "ipvr-ved-irq"; + + /* MMIO/REG for child's use */ + rsc[1].start = pci_resource_start(dev->pdev, 0); + rsc[1].end = pci_resource_start(dev->pdev, 0) + 2*1024*1024; /* gen7 */ + rsc[1].flags = IORESOURCE_MEM; + rsc[1].name = "ipvr-ved-mmio"; + + rsc[2].start = VLV_VED_BASE; + rsc[2].end = VLV_VED_BASE + VLV_VED_SIZE; + rsc[2].flags = IORESOURCE_REG; + rsc[2].name = "ipvr-ved-reg"; + + ret = platform_device_add_resources(dev_priv->ved_platdev, rsc, 3); + if (unlikely(ret)) { + DRM_ERROR("Failed to add resource for VED platform device: %d\n", ret); + goto err; + } + + /* Runtime-PM hook */ + dev_priv->ved_platdev->dev.parent = dev->dev; + ret = platform_device_add(dev_priv->ved_platdev); + if (unlikely(ret)) { + DRM_ERROR("Failed to add VED platform device: %d\n", ret); + goto err; + } + + kfree(rsc); + DRM_INFO("Successfully initialized Valleyview-VED\n"); + return 0; +err: + if (rsc) + kfree(rsc); + if (dev_priv->ved_platdev) + platform_device_unregister(dev_priv->ved_platdev); + if (irq >= 0) + irq_free_desc(irq); + return ret; +} + +static void valleyview_ved_cleanup(struct drm_device *dev) +{ + int irq; + struct drm_i915_private *dev_priv = dev->dev_private; + + irq = platform_get_irq(dev_priv->ved_platdev, 0); + if (irq >= 0) + irq_free_desc(irq); + + platform_device_unregister(dev_priv->ved_platdev); +} + void i915_update_dri1_breadcrumb(struct drm_device *dev) { struct drm_i915_private *dev_priv = dev->dev_private; @@ -1793,6 +1883,10 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags)
intel_runtime_pm_enable(dev_priv);
+ if (IS_VALLEYVIEW(dev)) { + BUG_ON(valleyview_ved_init(dev)); + } + return 0;
out_power_well: @@ -1833,6 +1927,10 @@ int i915_driver_unload(struct drm_device *dev) struct drm_i915_private *dev_priv = dev->dev_private; int ret;
+ if (IS_VALLEYVIEW(dev)) { + valleyview_ved_cleanup(dev); + } + ret = i915_gem_suspend(dev); if (ret) { DRM_ERROR("failed to idle hardware: %d\n", ret); diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 821ba26..aa8a183 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -1709,6 +1709,10 @@ struct drm_i915_private {
uint32_t bios_vgacntr;
+ /* used for setup sub device for valleyview */ + struct platform_device *ved_platdev; + int ved_irq; + /* Old dri1 support infrastructure, beware the dragons ya fools entering * here! */ struct i915_dri1_state dri1; @@ -2921,6 +2925,8 @@ void vlv_flisdsi_write(struct drm_i915_private *dev_priv, u32 reg, u32 val); int vlv_gpu_freq(struct drm_i915_private *dev_priv, int val); int vlv_freq_opcode(struct drm_i915_private *dev_priv, int val);
+extern int valleyview_initialize_ved_irq(struct drm_device *dev, int irq); + #define FORCEWAKE_RENDER (1 << 0) #define FORCEWAKE_MEDIA (1 << 1) #define FORCEWAKE_ALL (FORCEWAKE_RENDER | FORCEWAKE_MEDIA) diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c index 737b239..25c8cde 100644 --- a/drivers/gpu/drm/i915/i915_irq.c +++ b/drivers/gpu/drm/i915/i915_irq.c @@ -2142,12 +2142,75 @@ static void i9xx_hpd_irq_handler(struct drm_device *dev) } }
+static void valleyview_enable_ved_irq(struct irq_data *d) +{ + struct drm_device *dev = d->chip_data; + struct drm_i915_private *dev_priv = (struct drm_i915_private *) dev->dev_private; + unsigned long irqflags; + u32 imr, ier; + spin_lock_irqsave(&dev_priv->irq_lock, irqflags); + + ier = I915_READ(VLV_IER); + ier |= VLV_VED_BLOCK_INTERRUPT; + DRM_DEBUG_DRIVER("%s IER=>0x%08x\n", __func__, ier); + I915_WRITE(VLV_IER, ier); + POSTING_READ(VLV_IER); + + imr = I915_READ(VLV_IMR); + imr &= ~VLV_VED_BLOCK_INTERRUPT; + dev_priv->irq_mask = imr; + DRM_DEBUG_DRIVER("%s IMR=>0x%08x\n", __func__, imr); + I915_WRITE(VLV_IMR, imr); + POSTING_READ(VLV_IMR); + + spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags); +} + +static void valleyview_disable_ved_irq(struct irq_data *d) +{ + struct drm_device *dev = d->chip_data; + struct drm_i915_private *dev_priv = (struct drm_i915_private *) dev->dev_private; + unsigned long irqflags; + u32 imr, ier; + spin_lock_irqsave(&dev_priv->irq_lock, irqflags); + + ier = I915_READ(VLV_IER); + ier &= ~VLV_VED_BLOCK_INTERRUPT; + DRM_DEBUG_DRIVER("%s IER=>0x%08x\n", __func__, ier); + I915_WRITE(VLV_IER, ier); + POSTING_READ(VLV_IER); + + imr = I915_READ(VLV_IMR); + imr |= VLV_VED_BLOCK_INTERRUPT; + dev_priv->irq_mask = imr; + DRM_DEBUG_DRIVER("%s IMR=>0x%08x\n", __func__, imr); + I915_WRITE(VLV_IMR, imr); + POSTING_READ(VLV_IMR); + + spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags); +} + +int valleyview_initialize_ved_irq(struct drm_device *dev, int irq) +{ + static struct irq_chip ved_irqchip = { + .name = "ipvr_ved_irqchip", + .irq_mask = valleyview_disable_ved_irq, + .irq_unmask = valleyview_enable_ved_irq, + }; + irq_set_chip_and_handler_name(irq, + &ved_irqchip, + handle_simple_irq, + "ipvr_ved_irq_handler"); + return irq_set_chip_data(irq, dev); +} + static irqreturn_t valleyview_irq_handler(int irq, void *arg) { struct drm_device *dev = arg; struct drm_i915_private *dev_priv = dev->dev_private; u32 iir, gt_iir, pm_iir; irqreturn_t ret = IRQ_NONE; + int ved_ret;
while (true) { /* Find, clear, then process each source of interrupt */ @@ -2177,6 +2240,13 @@ static irqreturn_t valleyview_irq_handler(int irq, void *arg) snb_gt_irq_handler(dev, dev_priv, gt_iir); if (pm_iir) gen6_rps_irq_handler(dev_priv, pm_iir); + if (IS_VALLEYVIEW(dev) && dev_priv->ved_irq >= 0 + && (iir & VLV_VED_BLOCK_INTERRUPT)) { + ved_ret = generic_handle_irq(dev_priv->ved_irq); + if (ved_ret) + DRM_ERROR("Error forwarding VED irq: %d\n", ved_ret); + trace_valleyview_ved_event(iir); + } /* Call regardless, as some status bits might not be * signalled in iir */ valleyview_pipestat_irq_handler(dev, iir); diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h index 2ed02c3..89c8a06 100644 --- a/drivers/gpu/drm/i915/i915_reg.h +++ b/drivers/gpu/drm/i915/i915_reg.h @@ -1281,9 +1281,13 @@ enum punit_power_well { #define GFX_PSMI_GRANULARITY (1<<10) #define GFX_PPGTT_ENABLE (1<<9)
+#define VLV_VED_BASE 0x170000 +#define VLV_VED_SIZE 0x010000 #define VLV_DISPLAY_BASE 0x180000 #define VLV_MIPI_BASE VLV_DISPLAY_BASE
+#define VLV_VED_BLOCK_INTERRUPT (1 << 23) + #define VLV_GU_CTL0 (VLV_DISPLAY_BASE + 0x2030) #define VLV_GU_CTL1 (VLV_DISPLAY_BASE + 0x2034) #define SCPD0 0x0209c /* 915+ only */ diff --git a/drivers/gpu/drm/i915/i915_trace.h b/drivers/gpu/drm/i915/i915_trace.h index f5aa006..522bd1d 100644 --- a/drivers/gpu/drm/i915/i915_trace.h +++ b/drivers/gpu/drm/i915/i915_trace.h @@ -587,6 +587,21 @@ TRACE_EVENT(intel_gpu_freq_change, TP_printk("new_freq=%u", __entry->freq) );
+TRACE_EVENT(valleyview_ved_event, + TP_PROTO(u32 iir), + TP_ARGS(iir), + + TP_STRUCT__entry( + __field(u32, iir) + ), + + TP_fast_assign( + __entry->iir = iir; + ), + + TP_printk("forwarded with iir 0x%08x", __entry->iir) +); + #endif /* _I915_TRACE_H_ */
/* This part must be outside protection */
On Mon, 13 Oct 2014, Yao Cheng yao.cheng@intel.com wrote:
Setup following resources during i915_driver_load:
- create a child platform and resource
- allocate a new IRQ line and irq chip
- set up IRQ mask/unmask callbacks
vxd392 driver (if installed) will bind itself to the platform device and create new drm device
Signed-off-by: Yao Cheng yao.cheng@intel.com
drivers/gpu/drm/i915/i915_dma.c | 98 +++++++++++++++++++++++++++++++++++++++ drivers/gpu/drm/i915/i915_drv.h | 6 +++ drivers/gpu/drm/i915/i915_irq.c | 70 ++++++++++++++++++++++++++++ drivers/gpu/drm/i915/i915_reg.h | 4 ++ drivers/gpu/drm/i915/i915_trace.h | 15 ++++++ 5 files changed, 193 insertions(+)
diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c index 85d14e1..73c78d1 100644 --- a/drivers/gpu/drm/i915/i915_dma.c +++ b/drivers/gpu/drm/i915/i915_dma.c @@ -85,6 +85,96 @@ intel_read_legacy_status_page(struct drm_i915_private *dev_priv, int reg) #define READ_BREADCRUMB(dev_priv) READ_HWSP(dev_priv, I915_BREADCRUMB_INDEX) #define I915_BREADCRUMB_INDEX 0x21
+static int valleyview_ved_init(struct drm_device *dev) +{
- int ret;
- int irq = -1;
- struct resource *rsc = NULL;
- struct drm_i915_private *dev_priv = dev->dev_private;;
- dev_priv->ved_platdev = platform_device_alloc("ipvr-ved", -1);
- if (unlikely(!dev_priv->ved_platdev)) {
DRM_ERROR("Failed to allocate VED platform device\n");ret = -ENOMEM;goto err;- }
- rsc = kzalloc(sizeof(*rsc) * 3, GFP_KERNEL);
- if (unlikely(!rsc)) {
DRM_ERROR("Failed to allocate resource for VED platform device\n");ret = -ENOMEM;goto err;- }
- /* init IRQ number and chip/callbacks */
- irq = irq_alloc_descs(-1, 0, 1, 0);
- if (unlikely(irq < 0)) {
DRM_ERROR("Failed to allocate IRQ number: %d\n", irq);ret = -ENOMEM;goto err;- }
- ret = valleyview_initialize_ved_irq(dev, irq);
- if (unlikely(ret)) {
DRM_ERROR("Failed to initialize VED IRQ: %d\n", ret);goto err;- }
- dev_priv->ved_irq = irq;
- rsc[0].start = rsc[0].end = irq;
- rsc[0].flags = IORESOURCE_IRQ;
- rsc[0].name = "ipvr-ved-irq";
- /* MMIO/REG for child's use */
- rsc[1].start = pci_resource_start(dev->pdev, 0);
- rsc[1].end = pci_resource_start(dev->pdev, 0) + 2*1024*1024; /* gen7 */
- rsc[1].flags = IORESOURCE_MEM;
- rsc[1].name = "ipvr-ved-mmio";
- rsc[2].start = VLV_VED_BASE;
- rsc[2].end = VLV_VED_BASE + VLV_VED_SIZE;
- rsc[2].flags = IORESOURCE_REG;
- rsc[2].name = "ipvr-ved-reg";
- ret = platform_device_add_resources(dev_priv->ved_platdev, rsc, 3);
- if (unlikely(ret)) {
DRM_ERROR("Failed to add resource for VED platform device: %d\n", ret);goto err;- }
- /* Runtime-PM hook */
- dev_priv->ved_platdev->dev.parent = dev->dev;
- ret = platform_device_add(dev_priv->ved_platdev);
- if (unlikely(ret)) {
DRM_ERROR("Failed to add VED platform device: %d\n", ret);goto err;- }
- kfree(rsc);
- DRM_INFO("Successfully initialized Valleyview-VED\n");
- return 0;
+err:
- if (rsc)
kfree(rsc);- if (dev_priv->ved_platdev)
platform_device_unregister(dev_priv->ved_platdev);- if (irq >= 0)
irq_free_desc(irq);- return ret;
+}
+static void valleyview_ved_cleanup(struct drm_device *dev) +{
- int irq;
- struct drm_i915_private *dev_priv = dev->dev_private;
- irq = platform_get_irq(dev_priv->ved_platdev, 0);
- if (irq >= 0)
irq_free_desc(irq);- platform_device_unregister(dev_priv->ved_platdev);
+}
void i915_update_dri1_breadcrumb(struct drm_device *dev) { struct drm_i915_private *dev_priv = dev->dev_private; @@ -1793,6 +1883,10 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags)
intel_runtime_pm_enable(dev_priv);
- if (IS_VALLEYVIEW(dev)) {
BUG_ON(valleyview_ved_init(dev));
Just a quick drive-by comment, please don't bring down the machine if this one fails.
Thanks, Jani.
- }
- return 0;
out_power_well: @@ -1833,6 +1927,10 @@ int i915_driver_unload(struct drm_device *dev) struct drm_i915_private *dev_priv = dev->dev_private; int ret;
- if (IS_VALLEYVIEW(dev)) {
valleyview_ved_cleanup(dev);- }
- ret = i915_gem_suspend(dev); if (ret) { DRM_ERROR("failed to idle hardware: %d\n", ret);
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 821ba26..aa8a183 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -1709,6 +1709,10 @@ struct drm_i915_private {
uint32_t bios_vgacntr;
- /* used for setup sub device for valleyview */
- struct platform_device *ved_platdev;
- int ved_irq;
- /* Old dri1 support infrastructure, beware the dragons ya fools entering
struct i915_dri1_state dri1;
- here! */
@@ -2921,6 +2925,8 @@ void vlv_flisdsi_write(struct drm_i915_private *dev_priv, u32 reg, u32 val); int vlv_gpu_freq(struct drm_i915_private *dev_priv, int val); int vlv_freq_opcode(struct drm_i915_private *dev_priv, int val);
+extern int valleyview_initialize_ved_irq(struct drm_device *dev, int irq);
#define FORCEWAKE_RENDER (1 << 0) #define FORCEWAKE_MEDIA (1 << 1) #define FORCEWAKE_ALL (FORCEWAKE_RENDER | FORCEWAKE_MEDIA) diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c index 737b239..25c8cde 100644 --- a/drivers/gpu/drm/i915/i915_irq.c +++ b/drivers/gpu/drm/i915/i915_irq.c @@ -2142,12 +2142,75 @@ static void i9xx_hpd_irq_handler(struct drm_device *dev) } }
+static void valleyview_enable_ved_irq(struct irq_data *d) +{
- struct drm_device *dev = d->chip_data;
- struct drm_i915_private *dev_priv = (struct drm_i915_private *) dev->dev_private;
- unsigned long irqflags;
- u32 imr, ier;
- spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
- ier = I915_READ(VLV_IER);
- ier |= VLV_VED_BLOCK_INTERRUPT;
- DRM_DEBUG_DRIVER("%s IER=>0x%08x\n", __func__, ier);
- I915_WRITE(VLV_IER, ier);
- POSTING_READ(VLV_IER);
- imr = I915_READ(VLV_IMR);
- imr &= ~VLV_VED_BLOCK_INTERRUPT;
- dev_priv->irq_mask = imr;
- DRM_DEBUG_DRIVER("%s IMR=>0x%08x\n", __func__, imr);
- I915_WRITE(VLV_IMR, imr);
- POSTING_READ(VLV_IMR);
- spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
+}
+static void valleyview_disable_ved_irq(struct irq_data *d) +{
- struct drm_device *dev = d->chip_data;
- struct drm_i915_private *dev_priv = (struct drm_i915_private *) dev->dev_private;
- unsigned long irqflags;
- u32 imr, ier;
- spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
- ier = I915_READ(VLV_IER);
- ier &= ~VLV_VED_BLOCK_INTERRUPT;
- DRM_DEBUG_DRIVER("%s IER=>0x%08x\n", __func__, ier);
- I915_WRITE(VLV_IER, ier);
- POSTING_READ(VLV_IER);
- imr = I915_READ(VLV_IMR);
- imr |= VLV_VED_BLOCK_INTERRUPT;
- dev_priv->irq_mask = imr;
- DRM_DEBUG_DRIVER("%s IMR=>0x%08x\n", __func__, imr);
- I915_WRITE(VLV_IMR, imr);
- POSTING_READ(VLV_IMR);
- spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
+}
+int valleyview_initialize_ved_irq(struct drm_device *dev, int irq) +{
- static struct irq_chip ved_irqchip = {
.name = "ipvr_ved_irqchip",.irq_mask = valleyview_disable_ved_irq,.irq_unmask = valleyview_enable_ved_irq,- };
- irq_set_chip_and_handler_name(irq,
&ved_irqchip,handle_simple_irq,"ipvr_ved_irq_handler");- return irq_set_chip_data(irq, dev);
+}
static irqreturn_t valleyview_irq_handler(int irq, void *arg) { struct drm_device *dev = arg; struct drm_i915_private *dev_priv = dev->dev_private; u32 iir, gt_iir, pm_iir; irqreturn_t ret = IRQ_NONE;
int ved_ret;
while (true) { /* Find, clear, then process each source of interrupt */
@@ -2177,6 +2240,13 @@ static irqreturn_t valleyview_irq_handler(int irq, void *arg) snb_gt_irq_handler(dev, dev_priv, gt_iir); if (pm_iir) gen6_rps_irq_handler(dev_priv, pm_iir);
if (IS_VALLEYVIEW(dev) && dev_priv->ved_irq >= 0&& (iir & VLV_VED_BLOCK_INTERRUPT)) {ved_ret = generic_handle_irq(dev_priv->ved_irq);if (ved_ret)DRM_ERROR("Error forwarding VED irq: %d\n", ved_ret);trace_valleyview_ved_event(iir);}valleyview_pipestat_irq_handler(dev, iir);
- signalled in iir */
diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h index 2ed02c3..89c8a06 100644 --- a/drivers/gpu/drm/i915/i915_reg.h +++ b/drivers/gpu/drm/i915/i915_reg.h @@ -1281,9 +1281,13 @@ enum punit_power_well { #define GFX_PSMI_GRANULARITY (1<<10) #define GFX_PPGTT_ENABLE (1<<9)
+#define VLV_VED_BASE 0x170000 +#define VLV_VED_SIZE 0x010000 #define VLV_DISPLAY_BASE 0x180000 #define VLV_MIPI_BASE VLV_DISPLAY_BASE
+#define VLV_VED_BLOCK_INTERRUPT (1 << 23)
#define VLV_GU_CTL0 (VLV_DISPLAY_BASE + 0x2030) #define VLV_GU_CTL1 (VLV_DISPLAY_BASE + 0x2034) #define SCPD0 0x0209c /* 915+ only */ diff --git a/drivers/gpu/drm/i915/i915_trace.h b/drivers/gpu/drm/i915/i915_trace.h index f5aa006..522bd1d 100644 --- a/drivers/gpu/drm/i915/i915_trace.h +++ b/drivers/gpu/drm/i915/i915_trace.h @@ -587,6 +587,21 @@ TRACE_EVENT(intel_gpu_freq_change, TP_printk("new_freq=%u", __entry->freq) );
+TRACE_EVENT(valleyview_ved_event,
TP_PROTO(u32 iir),TP_ARGS(iir),TP_STRUCT__entry(__field(u32, iir)),TP_fast_assign(__entry->iir = iir;),TP_printk("forwarded with iir 0x%08x", __entry->iir)+);
#endif /* _I915_TRACE_H_ */
/* This part must be outside protection */
1.9.1
Intel-gfx mailing list Intel-gfx@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Nikula, thx for reminder, will update it.
-----Original Message----- From: Jani Nikula [mailto:jani.nikula@linux.intel.com] Sent: Tuesday, October 14, 2014 9:27 PM To: Cheng, Yao; intel-gfx@lists.freedesktop.org Cc: Jiang, Fei; dri-devel@lists.freedesktop.org; Vetter, Daniel Subject: Re: [Intel-gfx] [RFC PATCH 1/3] drm/i915: add vxd392 bridge in i915
On Mon, 13 Oct 2014, Yao Cheng yao.cheng@intel.com wrote:
Setup following resources during i915_driver_load:
- create a child platform and resource 2. allocate a new IRQ line and
irq chip 3. set up IRQ mask/unmask callbacks vxd392 driver (if installed) will bind itself to the platform device and create new drm device
Signed-off-by: Yao Cheng yao.cheng@intel.com
drivers/gpu/drm/i915/i915_dma.c | 98 +++++++++++++++++++++++++++++++++++++++ drivers/gpu/drm/i915/i915_drv.h | 6 +++ drivers/gpu/drm/i915/i915_irq.c | 70 ++++++++++++++++++++++++++++ drivers/gpu/drm/i915/i915_reg.h | 4 ++ drivers/gpu/drm/i915/i915_trace.h | 15 ++++++ 5 files changed, 193 insertions(+)
diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c index 85d14e1..73c78d1 100644 --- a/drivers/gpu/drm/i915/i915_dma.c +++ b/drivers/gpu/drm/i915/i915_dma.c @@ -85,6 +85,96 @@ intel_read_legacy_status_page(struct drm_i915_private *dev_priv, int reg) #define READ_BREADCRUMB(dev_priv) READ_HWSP(dev_priv, I915_BREADCRUMB_INDEX) #define I915_BREADCRUMB_INDEX 0x21
+static int valleyview_ved_init(struct drm_device *dev) {
- int ret;
- int irq = -1;
- struct resource *rsc = NULL;
- struct drm_i915_private *dev_priv = dev->dev_private;;
- dev_priv->ved_platdev = platform_device_alloc("ipvr-ved", -1);
- if (unlikely(!dev_priv->ved_platdev)) {
DRM_ERROR("Failed to allocate VED platform device\n");ret = -ENOMEM;goto err;- }
- rsc = kzalloc(sizeof(*rsc) * 3, GFP_KERNEL);
- if (unlikely(!rsc)) {
DRM_ERROR("Failed to allocate resource for VED platform device\n");ret = -ENOMEM;goto err;- }
- /* init IRQ number and chip/callbacks */
- irq = irq_alloc_descs(-1, 0, 1, 0);
- if (unlikely(irq < 0)) {
DRM_ERROR("Failed to allocate IRQ number: %d\n", irq);ret = -ENOMEM;goto err;- }
- ret = valleyview_initialize_ved_irq(dev, irq);
- if (unlikely(ret)) {
DRM_ERROR("Failed to initialize VED IRQ: %d\n", ret);goto err;- }
- dev_priv->ved_irq = irq;
- rsc[0].start = rsc[0].end = irq;
- rsc[0].flags = IORESOURCE_IRQ;
- rsc[0].name = "ipvr-ved-irq";
- /* MMIO/REG for child's use */
- rsc[1].start = pci_resource_start(dev->pdev, 0);
- rsc[1].end = pci_resource_start(dev->pdev, 0) + 2*1024*1024; /* gen7 */
- rsc[1].flags = IORESOURCE_MEM;
- rsc[1].name = "ipvr-ved-mmio";
- rsc[2].start = VLV_VED_BASE;
- rsc[2].end = VLV_VED_BASE + VLV_VED_SIZE;
- rsc[2].flags = IORESOURCE_REG;
- rsc[2].name = "ipvr-ved-reg";
- ret = platform_device_add_resources(dev_priv->ved_platdev, rsc, 3);
- if (unlikely(ret)) {
DRM_ERROR("Failed to add resource for VED platform device: %d\n", ret);goto err;- }
- /* Runtime-PM hook */
- dev_priv->ved_platdev->dev.parent = dev->dev;
- ret = platform_device_add(dev_priv->ved_platdev);
- if (unlikely(ret)) {
DRM_ERROR("Failed to add VED platform device: %d\n", ret);goto err;- }
- kfree(rsc);
- DRM_INFO("Successfully initialized Valleyview-VED\n");
- return 0;
+err:
- if (rsc)
kfree(rsc);- if (dev_priv->ved_platdev)
platform_device_unregister(dev_priv->ved_platdev);- if (irq >= 0)
irq_free_desc(irq);- return ret;
+}
+static void valleyview_ved_cleanup(struct drm_device *dev) {
- int irq;
- struct drm_i915_private *dev_priv = dev->dev_private;
- irq = platform_get_irq(dev_priv->ved_platdev, 0);
- if (irq >= 0)
irq_free_desc(irq);- platform_device_unregister(dev_priv->ved_platdev);
+}
void i915_update_dri1_breadcrumb(struct drm_device *dev) { struct drm_i915_private *dev_priv = dev->dev_private; @@ -1793,6 +1883,10 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags)
intel_runtime_pm_enable(dev_priv);
- if (IS_VALLEYVIEW(dev)) {
BUG_ON(valleyview_ved_init(dev));
Just a quick drive-by comment, please don't bring down the machine if this one fails.
Thanks, Jani.
- }
- return 0;
out_power_well: @@ -1833,6 +1927,10 @@ int i915_driver_unload(struct drm_device *dev) struct drm_i915_private *dev_priv = dev->dev_private; int ret;
- if (IS_VALLEYVIEW(dev)) {
valleyview_ved_cleanup(dev);- }
- ret = i915_gem_suspend(dev); if (ret) { DRM_ERROR("failed to idle hardware: %d\n", ret); diff --git
a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 821ba26..aa8a183 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -1709,6 +1709,10 @@ struct drm_i915_private {
uint32_t bios_vgacntr;
- /* used for setup sub device for valleyview */
- struct platform_device *ved_platdev;
- int ved_irq;
- /* Old dri1 support infrastructure, beware the dragons ya fools entering
struct i915_dri1_state dri1;
- here! */
@@ -2921,6 +2925,8 @@ void vlv_flisdsi_write(struct drm_i915_private *dev_priv, u32 reg, u32 val); int vlv_gpu_freq(struct drm_i915_private *dev_priv, int val); int vlv_freq_opcode(struct drm_i915_private *dev_priv, int val);
+extern int valleyview_initialize_ved_irq(struct drm_device *dev, int +irq);
#define FORCEWAKE_RENDER (1 << 0) #define FORCEWAKE_MEDIA (1 << 1) #define FORCEWAKE_ALL (FORCEWAKE_RENDER | FORCEWAKE_MEDIA) diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c index 737b239..25c8cde 100644 --- a/drivers/gpu/drm/i915/i915_irq.c +++ b/drivers/gpu/drm/i915/i915_irq.c @@ -2142,12 +2142,75 @@ static void i9xx_hpd_irq_handler(struct drm_device *dev) } }
+static void valleyview_enable_ved_irq(struct irq_data *d) {
- struct drm_device *dev = d->chip_data;
- struct drm_i915_private *dev_priv = (struct drm_i915_private *) dev->dev_private;
- unsigned long irqflags;
- u32 imr, ier;
- spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
- ier = I915_READ(VLV_IER);
- ier |= VLV_VED_BLOCK_INTERRUPT;
- DRM_DEBUG_DRIVER("%s IER=>0x%08x\n", __func__, ier);
- I915_WRITE(VLV_IER, ier);
- POSTING_READ(VLV_IER);
- imr = I915_READ(VLV_IMR);
- imr &= ~VLV_VED_BLOCK_INTERRUPT;
- dev_priv->irq_mask = imr;
- DRM_DEBUG_DRIVER("%s IMR=>0x%08x\n", __func__, imr);
- I915_WRITE(VLV_IMR, imr);
- POSTING_READ(VLV_IMR);
- spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags); }
+static void valleyview_disable_ved_irq(struct irq_data *d) {
- struct drm_device *dev = d->chip_data;
- struct drm_i915_private *dev_priv = (struct drm_i915_private *) dev->dev_private;
- unsigned long irqflags;
- u32 imr, ier;
- spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
- ier = I915_READ(VLV_IER);
- ier &= ~VLV_VED_BLOCK_INTERRUPT;
- DRM_DEBUG_DRIVER("%s IER=>0x%08x\n", __func__, ier);
- I915_WRITE(VLV_IER, ier);
- POSTING_READ(VLV_IER);
- imr = I915_READ(VLV_IMR);
- imr |= VLV_VED_BLOCK_INTERRUPT;
- dev_priv->irq_mask = imr;
- DRM_DEBUG_DRIVER("%s IMR=>0x%08x\n", __func__, imr);
- I915_WRITE(VLV_IMR, imr);
- POSTING_READ(VLV_IMR);
- spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags); }
+int valleyview_initialize_ved_irq(struct drm_device *dev, int irq) {
- static struct irq_chip ved_irqchip = {
.name = "ipvr_ved_irqchip",.irq_mask = valleyview_disable_ved_irq,.irq_unmask = valleyview_enable_ved_irq,- };
- irq_set_chip_and_handler_name(irq,
&ved_irqchip,handle_simple_irq,"ipvr_ved_irq_handler");- return irq_set_chip_data(irq, dev);
+}
static irqreturn_t valleyview_irq_handler(int irq, void *arg) { struct drm_device *dev = arg; struct drm_i915_private *dev_priv = dev->dev_private; u32 iir, gt_iir, pm_iir; irqreturn_t ret = IRQ_NONE;
int ved_ret;
while (true) { /* Find, clear, then process each source of interrupt */ @@ -2177,6
+2240,13 @@ static irqreturn_t valleyview_irq_handler(int irq, void *arg) snb_gt_irq_handler(dev, dev_priv, gt_iir); if (pm_iir) gen6_rps_irq_handler(dev_priv, pm_iir);
if (IS_VALLEYVIEW(dev) && dev_priv->ved_irq >= 0&& (iir & VLV_VED_BLOCK_INTERRUPT)) {ved_ret = generic_handle_irq(dev_priv->ved_irq);if (ved_ret)DRM_ERROR("Error forwarding VED irq: %d\n", ved_ret);trace_valleyview_ved_event(iir);}valleyview_pipestat_irq_handler(dev, iir); diff --git
- signalled in iir */
a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h index 2ed02c3..89c8a06 100644 --- a/drivers/gpu/drm/i915/i915_reg.h +++ b/drivers/gpu/drm/i915/i915_reg.h @@ -1281,9 +1281,13 @@ enum punit_power_well { #define GFX_PSMI_GRANULARITY (1<<10) #define GFX_PPGTT_ENABLE (1<<9)
+#define VLV_VED_BASE 0x170000 +#define VLV_VED_SIZE 0x010000 #define VLV_DISPLAY_BASE 0x180000 #define VLV_MIPI_BASE VLV_DISPLAY_BASE
+#define VLV_VED_BLOCK_INTERRUPT (1 << 23)
#define VLV_GU_CTL0 (VLV_DISPLAY_BASE + 0x2030) #define VLV_GU_CTL1 (VLV_DISPLAY_BASE + 0x2034) #define SCPD0 0x0209c /* 915+ only */ diff --git a/drivers/gpu/drm/i915/i915_trace.h b/drivers/gpu/drm/i915/i915_trace.h index f5aa006..522bd1d 100644 --- a/drivers/gpu/drm/i915/i915_trace.h +++ b/drivers/gpu/drm/i915/i915_trace.h @@ -587,6 +587,21 @@ TRACE_EVENT(intel_gpu_freq_change, TP_printk("new_freq=%u", __entry->freq) );
+TRACE_EVENT(valleyview_ved_event,
TP_PROTO(u32 iir),TP_ARGS(iir),TP_STRUCT__entry(__field(u32, iir)),TP_fast_assign(__entry->iir = iir;),TP_printk("forwarded with iir 0x%08x", __entry->iir) );#endif /* _I915_TRACE_H_ */
/* This part must be outside protection */
1.9.1
Intel-gfx mailing list Intel-gfx@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/intel-gfx
-- Jani Nikula, Intel Open Source Technology Center
On Mon, Oct 13, 2014 at 08:15:01PM +0800, Yao Cheng wrote:
Setup following resources during i915_driver_load:
- create a child platform and resource
- allocate a new IRQ line and irq chip
- set up IRQ mask/unmask callbacks
vxd392 driver (if installed) will bind itself to the platform device and create new drm device
Signed-off-by: Yao Cheng yao.cheng@intel.com
Ok, bunch of comments. First a high-level one: I think this qualifies as a new subsystem of i915, and so it would be good to extract this into a new file (i915_ved.c maybe), including adding kerneldoc for the setup function, a short DOC: overview section and pulling it all into the drm kerneldoc (probably a new subsection in the driver core section).
Aside from the lack of documentation just a few small comments below. Overall I really like how cleanly we can integrate vxd support into i915, so good work.
Cheers, Daniel
drivers/gpu/drm/i915/i915_dma.c | 98 +++++++++++++++++++++++++++++++++++++++ drivers/gpu/drm/i915/i915_drv.h | 6 +++ drivers/gpu/drm/i915/i915_irq.c | 70 ++++++++++++++++++++++++++++ drivers/gpu/drm/i915/i915_reg.h | 4 ++ drivers/gpu/drm/i915/i915_trace.h | 15 ++++++ 5 files changed, 193 insertions(+)
diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c index 85d14e1..73c78d1 100644 --- a/drivers/gpu/drm/i915/i915_dma.c +++ b/drivers/gpu/drm/i915/i915_dma.c @@ -85,6 +85,96 @@ intel_read_legacy_status_page(struct drm_i915_private *dev_priv, int reg) #define READ_BREADCRUMB(dev_priv) READ_HWSP(dev_priv, I915_BREADCRUMB_INDEX) #define I915_BREADCRUMB_INDEX 0x21
+static int valleyview_ved_init(struct drm_device *dev) +{
- int ret;
- int irq = -1;
- struct resource *rsc = NULL;
- struct drm_i915_private *dev_priv = dev->dev_private;;
- dev_priv->ved_platdev = platform_device_alloc("ipvr-ved", -1);
I think we should put vlv into the platform name, in case this driver will ever be reused on other platforms. So maybe "ipvr-ved-vxd"?
- if (unlikely(!dev_priv->ved_platdev)) {
DRM_ERROR("Failed to allocate VED platform device\n");ret = -ENOMEM;goto err;- }
- rsc = kzalloc(sizeof(*rsc) * 3, GFP_KERNEL);
- if (unlikely(!rsc)) {
DRM_ERROR("Failed to allocate resource for VED platform device\n");ret = -ENOMEM;goto err;- }
- /* init IRQ number and chip/callbacks */
- irq = irq_alloc_descs(-1, 0, 1, 0);
- if (unlikely(irq < 0)) {
DRM_ERROR("Failed to allocate IRQ number: %d\n", irq);ret = -ENOMEM;goto err;- }
- ret = valleyview_initialize_ved_irq(dev, irq);
- if (unlikely(ret)) {
DRM_ERROR("Failed to initialize VED IRQ: %d\n", ret);goto err;- }
- dev_priv->ved_irq = irq;
- rsc[0].start = rsc[0].end = irq;
- rsc[0].flags = IORESOURCE_IRQ;
- rsc[0].name = "ipvr-ved-irq";
- /* MMIO/REG for child's use */
- rsc[1].start = pci_resource_start(dev->pdev, 0);
- rsc[1].end = pci_resource_start(dev->pdev, 0) + 2*1024*1024; /* gen7 */
- rsc[1].flags = IORESOURCE_MEM;
- rsc[1].name = "ipvr-ved-mmio";
- rsc[2].start = VLV_VED_BASE;
- rsc[2].end = VLV_VED_BASE + VLV_VED_SIZE;
- rsc[2].flags = IORESOURCE_REG;
- rsc[2].name = "ipvr-ved-reg";
- ret = platform_device_add_resources(dev_priv->ved_platdev, rsc, 3);
- if (unlikely(ret)) {
DRM_ERROR("Failed to add resource for VED platform device: %d\n", ret);goto err;- }
- /* Runtime-PM hook */
- dev_priv->ved_platdev->dev.parent = dev->dev;
- ret = platform_device_add(dev_priv->ved_platdev);
- if (unlikely(ret)) {
DRM_ERROR("Failed to add VED platform device: %d\n", ret);goto err;- }
- kfree(rsc);
- DRM_INFO("Successfully initialized Valleyview-VED\n");
- return 0;
+err:
- if (rsc)
kfree(rsc);- if (dev_priv->ved_platdev)
platform_device_unregister(dev_priv->ved_platdev);- if (irq >= 0)
irq_free_desc(irq);- return ret;
+}
+static void valleyview_ved_cleanup(struct drm_device *dev) +{
- int irq;
- struct drm_i915_private *dev_priv = dev->dev_private;
- irq = platform_get_irq(dev_priv->ved_platdev, 0);
- if (irq >= 0)
irq_free_desc(irq);- platform_device_unregister(dev_priv->ved_platdev);
I think you should unregister the platform device _before_ you free the irq. Otherwise the driver cleanup might freak out. Aside: Does the module reload test for i915 in i-g-t still work with the vxd driver loaded on vlv? Iirc you need to manually unload the vxd driver first to avoid inherit races in the platform device support code, so if that's the case you need to supply a patch for igt, too.
+}
void i915_update_dri1_breadcrumb(struct drm_device *dev) { struct drm_i915_private *dev_priv = dev->dev_private; @@ -1793,6 +1883,10 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags)
intel_runtime_pm_enable(dev_priv);
- if (IS_VALLEYVIEW(dev)) {
BUG_ON(valleyview_ved_init(dev));- }
As Jani said, now BUG_ON here please. Also, I think you should just print an error here but not even fail driver load for i915 (i.e. still return 0). We try to only fail i915 load if there's a hard issue with the modeset part of the driver, if render/GT/... parts fail to init then we'll continue. The idea is that if the user at least has a working display he can grab a lot more useful information about the init failure than when i915 is completely dead.
- return 0;
out_power_well: @@ -1833,6 +1927,10 @@ int i915_driver_unload(struct drm_device *dev) struct drm_i915_private *dev_priv = dev->dev_private; int ret;
- if (IS_VALLEYVIEW(dev)) {
valleyview_ved_cleanup(dev);- }
- ret = i915_gem_suspend(dev); if (ret) { DRM_ERROR("failed to idle hardware: %d\n", ret);
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 821ba26..aa8a183 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -1709,6 +1709,10 @@ struct drm_i915_private {
uint32_t bios_vgacntr;
- /* used for setup sub device for valleyview */
- struct platform_device *ved_platdev;
- int ved_irq;
- /* Old dri1 support infrastructure, beware the dragons ya fools entering
struct i915_dri1_state dri1;
- here! */
@@ -2921,6 +2925,8 @@ void vlv_flisdsi_write(struct drm_i915_private *dev_priv, u32 reg, u32 val); int vlv_gpu_freq(struct drm_i915_private *dev_priv, int val); int vlv_freq_opcode(struct drm_i915_private *dev_priv, int val);
+extern int valleyview_initialize_ved_irq(struct drm_device *dev, int irq);
#define FORCEWAKE_RENDER (1 << 0) #define FORCEWAKE_MEDIA (1 << 1) #define FORCEWAKE_ALL (FORCEWAKE_RENDER | FORCEWAKE_MEDIA) diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c index 737b239..25c8cde 100644 --- a/drivers/gpu/drm/i915/i915_irq.c +++ b/drivers/gpu/drm/i915/i915_irq.c @@ -2142,12 +2142,75 @@ static void i9xx_hpd_irq_handler(struct drm_device *dev) } }
+static void valleyview_enable_ved_irq(struct irq_data *d) +{
- struct drm_device *dev = d->chip_data;
- struct drm_i915_private *dev_priv = (struct drm_i915_private *) dev->dev_private;
- unsigned long irqflags;
- u32 imr, ier;
- spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
- ier = I915_READ(VLV_IER);
- ier |= VLV_VED_BLOCK_INTERRUPT;
- DRM_DEBUG_DRIVER("%s IER=>0x%08x\n", __func__, ier);
- I915_WRITE(VLV_IER, ier);
- POSTING_READ(VLV_IER);
- imr = I915_READ(VLV_IMR);
- imr &= ~VLV_VED_BLOCK_INTERRUPT;
- dev_priv->irq_mask = imr;
- DRM_DEBUG_DRIVER("%s IMR=>0x%08x\n", __func__, imr);
- I915_WRITE(VLV_IMR, imr);
- POSTING_READ(VLV_IMR);
- spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
+}
+static void valleyview_disable_ved_irq(struct irq_data *d) +{
- struct drm_device *dev = d->chip_data;
- struct drm_i915_private *dev_priv = (struct drm_i915_private *) dev->dev_private;
- unsigned long irqflags;
- u32 imr, ier;
- spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
- ier = I915_READ(VLV_IER);
- ier &= ~VLV_VED_BLOCK_INTERRUPT;
- DRM_DEBUG_DRIVER("%s IER=>0x%08x\n", __func__, ier);
- I915_WRITE(VLV_IER, ier);
- POSTING_READ(VLV_IER);
- imr = I915_READ(VLV_IMR);
- imr |= VLV_VED_BLOCK_INTERRUPT;
- dev_priv->irq_mask = imr;
- DRM_DEBUG_DRIVER("%s IMR=>0x%08x\n", __func__, imr);
- I915_WRITE(VLV_IMR, imr);
- POSTING_READ(VLV_IMR);
- spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
+}
+int valleyview_initialize_ved_irq(struct drm_device *dev, int irq) +{
- static struct irq_chip ved_irqchip = {
.name = "ipvr_ved_irqchip",.irq_mask = valleyview_disable_ved_irq,.irq_unmask = valleyview_enable_ved_irq,- };
Please move this struct out of the function, we generally keep vtables global. Also please add a WARN_ON(!intel_irqs_enabled()) here so that we make sure that the driver load ordering is correct. Same for the unregister side, interrupts should still be working when we remove the platform device.
- irq_set_chip_and_handler_name(irq,
&ved_irqchip,handle_simple_irq,"ipvr_ved_irq_handler");- return irq_set_chip_data(irq, dev);
+}
static irqreturn_t valleyview_irq_handler(int irq, void *arg) { struct drm_device *dev = arg; struct drm_i915_private *dev_priv = dev->dev_private; u32 iir, gt_iir, pm_iir; irqreturn_t ret = IRQ_NONE;
int ved_ret;
while (true) { /* Find, clear, then process each source of interrupt */
@@ -2177,6 +2240,13 @@ static irqreturn_t valleyview_irq_handler(int irq, void *arg) snb_gt_irq_handler(dev, dev_priv, gt_iir); if (pm_iir) gen6_rps_irq_handler(dev_priv, pm_iir);
if (IS_VALLEYVIEW(dev) && dev_priv->ved_irq >= 0&& (iir & VLV_VED_BLOCK_INTERRUPT)) {ved_ret = generic_handle_irq(dev_priv->ved_irq);if (ved_ret)DRM_ERROR("Error forwarding VED irq: %d\n", ved_ret);trace_valleyview_ved_event(iir);
I don't see a value in this tracepoint here - we don't have any other tracepoints in i915 for irq events. If you want this and the irq core doesn't provide any tracpoints (haven't checked but would be really suprising), then I think the right place to add this is in the vxd driver itself.
}valleyview_pipestat_irq_handler(dev, iir);
- signalled in iir */
diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h index 2ed02c3..89c8a06 100644 --- a/drivers/gpu/drm/i915/i915_reg.h +++ b/drivers/gpu/drm/i915/i915_reg.h @@ -1281,9 +1281,13 @@ enum punit_power_well { #define GFX_PSMI_GRANULARITY (1<<10) #define GFX_PPGTT_ENABLE (1<<9)
+#define VLV_VED_BASE 0x170000 +#define VLV_VED_SIZE 0x010000 #define VLV_DISPLAY_BASE 0x180000 #define VLV_MIPI_BASE VLV_DISPLAY_BASE
+#define VLV_VED_BLOCK_INTERRUPT (1 << 23)
#define VLV_GU_CTL0 (VLV_DISPLAY_BASE + 0x2030) #define VLV_GU_CTL1 (VLV_DISPLAY_BASE + 0x2034) #define SCPD0 0x0209c /* 915+ only */ diff --git a/drivers/gpu/drm/i915/i915_trace.h b/drivers/gpu/drm/i915/i915_trace.h index f5aa006..522bd1d 100644 --- a/drivers/gpu/drm/i915/i915_trace.h +++ b/drivers/gpu/drm/i915/i915_trace.h @@ -587,6 +587,21 @@ TRACE_EVENT(intel_gpu_freq_change, TP_printk("new_freq=%u", __entry->freq) );
+TRACE_EVENT(valleyview_ved_event,
TP_PROTO(u32 iir),TP_ARGS(iir),TP_STRUCT__entry(__field(u32, iir)),TP_fast_assign(__entry->iir = iir;),TP_printk("forwarded with iir 0x%08x", __entry->iir)+);
#endif /* _I915_TRACE_H_ */
/* This part must be outside protection */
1.9.1
Intel-gfx mailing list Intel-gfx@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Ok, bunch of comments. First a high-level one: I think this qualifies as a new subsystem of i915, and so it would be good to extract this into a new file (i915_ved.c maybe), including adding kerneldoc for the setup function, a short DOC: overview section and pulling it all into the drm kerneldoc (probably a new subsection in the driver core section).
Aside from the lack of documentation just a few small comments below. Overall I really like how cleanly we can integrate vxd support into i915, so good work.
I915_ved.c sounds to be a good place for these code, thx for this suggestion!
For the kerneldoc, I'll add a subsection in the core section for your review.
+static void valleyview_ved_cleanup(struct drm_device *dev) {
- int irq;
- struct drm_i915_private *dev_priv = dev->dev_private;
- irq = platform_get_irq(dev_priv->ved_platdev, 0);
- if (irq >= 0)
irq_free_desc(irq);- platform_device_unregister(dev_priv->ved_platdev);
I think you should unregister the platform device _before_ you free the irq. Otherwise the driver cleanup might freak out. Aside: Does the module reload test for i915 in i-g-t still work with the vxd driver loaded on vlv? Iirc you need to manually unload the vxd driver first to avoid inherit races in the platform device support code, so if that's the case you need to supply a patch for igt, too.
Sorry, what is i-g-t? I'll follow your suggestion, test i-g-t, and patch it if needed.
+}
void i915_update_dri1_breadcrumb(struct drm_device *dev) { struct drm_i915_private *dev_priv = dev->dev_private; @@ -1793,6 +1883,10 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags)
intel_runtime_pm_enable(dev_priv);
- if (IS_VALLEYVIEW(dev)) {
BUG_ON(valleyview_ved_init(dev));- }
As Jani said, now BUG_ON here please. Also, I think you should just print an error here but not even fail driver load for i915 (i.e. still return 0). We try to only fail i915 load if there's a hard issue with the modeset part of the driver, if render/GT/... parts fail to init then we'll continue. The idea is that if the user at least has a working display he can grab a lot more useful information about the init failure than when i915 is completely dead.
Sorry for this BUG_ON. Will replace with useful return value and message.
- return 0;
out_power_well: @@ -1833,6 +1927,10 @@ int i915_driver_unload(struct drm_device *dev) struct drm_i915_private *dev_priv = dev->dev_private; int ret;
- if (IS_VALLEYVIEW(dev)) {
valleyview_ved_cleanup(dev);- }
- ret = i915_gem_suspend(dev); if (ret) { DRM_ERROR("failed to idle hardware: %d\n", ret); diff --git
a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 821ba26..aa8a183 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -1709,6 +1709,10 @@ struct drm_i915_private {
uint32_t bios_vgacntr;
- /* used for setup sub device for valleyview */
- struct platform_device *ved_platdev;
- int ved_irq;
- /* Old dri1 support infrastructure, beware the dragons ya fools
entering
* here! */struct i915_dri1_state dri1; @@ -2921,6 +2925,8 @@ void vlv_flisdsi_write(struct drm_i915_private *dev_priv, u32 reg, u32 val); int vlv_gpu_freq(struct drm_i915_private *dev_priv, int val); int vlv_freq_opcode(struct drm_i915_private *dev_priv, int val);
+extern int valleyview_initialize_ved_irq(struct drm_device *dev, int +irq);
#define FORCEWAKE_RENDER (1 << 0) #define FORCEWAKE_MEDIA (1 << 1) #define FORCEWAKE_ALL (FORCEWAKE_RENDER |
FORCEWAKE_MEDIA)
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c index 737b239..25c8cde 100644 --- a/drivers/gpu/drm/i915/i915_irq.c +++ b/drivers/gpu/drm/i915/i915_irq.c @@ -2142,12 +2142,75 @@ static void i9xx_hpd_irq_handler(struct
drm_device *dev)
} }
+static void valleyview_enable_ved_irq(struct irq_data *d) {
- struct drm_device *dev = d->chip_data;
- struct drm_i915_private *dev_priv = (struct drm_i915_private *)
dev->dev_private;
- unsigned long irqflags;
- u32 imr, ier;
- spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
- ier = I915_READ(VLV_IER);
- ier |= VLV_VED_BLOCK_INTERRUPT;
- DRM_DEBUG_DRIVER("%s IER=>0x%08x\n", __func__, ier);
- I915_WRITE(VLV_IER, ier);
- POSTING_READ(VLV_IER);
- imr = I915_READ(VLV_IMR);
- imr &= ~VLV_VED_BLOCK_INTERRUPT;
- dev_priv->irq_mask = imr;
- DRM_DEBUG_DRIVER("%s IMR=>0x%08x\n", __func__, imr);
- I915_WRITE(VLV_IMR, imr);
- POSTING_READ(VLV_IMR);
- spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags); }
+static void valleyview_disable_ved_irq(struct irq_data *d) {
- struct drm_device *dev = d->chip_data;
- struct drm_i915_private *dev_priv = (struct drm_i915_private *)
dev->dev_private;
- unsigned long irqflags;
- u32 imr, ier;
- spin_lock_irqsave(&dev_priv->irq_lock, irqflags);
- ier = I915_READ(VLV_IER);
- ier &= ~VLV_VED_BLOCK_INTERRUPT;
- DRM_DEBUG_DRIVER("%s IER=>0x%08x\n", __func__, ier);
- I915_WRITE(VLV_IER, ier);
- POSTING_READ(VLV_IER);
- imr = I915_READ(VLV_IMR);
- imr |= VLV_VED_BLOCK_INTERRUPT;
- dev_priv->irq_mask = imr;
- DRM_DEBUG_DRIVER("%s IMR=>0x%08x\n", __func__, imr);
- I915_WRITE(VLV_IMR, imr);
- POSTING_READ(VLV_IMR);
- spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags); }
+int valleyview_initialize_ved_irq(struct drm_device *dev, int irq) {
- static struct irq_chip ved_irqchip = {
.name = "ipvr_ved_irqchip",.irq_mask = valleyview_disable_ved_irq,.irq_unmask = valleyview_enable_ved_irq,- };
Please move this struct out of the function, we generally keep vtables global. Also please add a WARN_ON(!intel_irqs_enabled()) here so that we make sure that the driver load ordering is correct. Same for the unregister side, interrupts should still be working when we remove the platform device.
Sure. Will move this to i915_ved.c as you suggested.
- irq_set_chip_and_handler_name(irq,
&ved_irqchip,handle_simple_irq,"ipvr_ved_irq_handler");- return irq_set_chip_data(irq, dev);
+}
static irqreturn_t valleyview_irq_handler(int irq, void *arg) { struct drm_device *dev = arg; struct drm_i915_private *dev_priv = dev->dev_private; u32 iir, gt_iir, pm_iir; irqreturn_t ret = IRQ_NONE;
int ved_ret;
while (true) { /* Find, clear, then process each source of interrupt */ @@ -
2177,6
+2240,13 @@ static irqreturn_t valleyview_irq_handler(int irq, void *arg) snb_gt_irq_handler(dev, dev_priv, gt_iir); if (pm_iir) gen6_rps_irq_handler(dev_priv, pm_iir);
if (IS_VALLEYVIEW(dev) && dev_priv->ved_irq >= 0&& (iir & VLV_VED_BLOCK_INTERRUPT)) {ved_ret = generic_handle_irq(dev_priv->ved_irq);if (ved_ret)DRM_ERROR("Error forwarding VEDirq: %d\n", ved_ret);
trace_valleyview_ved_event(iir);I don't see a value in this tracepoint here - we don't have any other tracepoints in i915 for irq events. If you want this and the irq core doesn't provide any tracpoints (haven't checked but would be really suprising), then I think the right place to add this is in the vxd driver itself.
OK vxd driver has its own irq trace_points so will remove this one.
}valleyview_pipestat_irq_handler(dev, iir); diff --git
- signalled in iir */
a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h index 2ed02c3..89c8a06 100644 --- a/drivers/gpu/drm/i915/i915_reg.h +++ b/drivers/gpu/drm/i915/i915_reg.h @@ -1281,9 +1281,13 @@ enum punit_power_well { #define GFX_PSMI_GRANULARITY (1<<10) #define GFX_PPGTT_ENABLE (1<<9)
+#define VLV_VED_BASE 0x170000 +#define VLV_VED_SIZE 0x010000 #define VLV_DISPLAY_BASE 0x180000 #define VLV_MIPI_BASE VLV_DISPLAY_BASE
+#define VLV_VED_BLOCK_INTERRUPT (1 << 23)
#define VLV_GU_CTL0 (VLV_DISPLAY_BASE + 0x2030) #define VLV_GU_CTL1 (VLV_DISPLAY_BASE + 0x2034) #define SCPD0 0x0209c /* 915+ only */ diff --git a/drivers/gpu/drm/i915/i915_trace.h b/drivers/gpu/drm/i915/i915_trace.h index f5aa006..522bd1d 100644 --- a/drivers/gpu/drm/i915/i915_trace.h +++ b/drivers/gpu/drm/i915/i915_trace.h @@ -587,6 +587,21 @@ TRACE_EVENT(intel_gpu_freq_change, TP_printk("new_freq=%u", __entry->freq) );
+TRACE_EVENT(valleyview_ved_event,
TP_PROTO(u32 iir),TP_ARGS(iir),TP_STRUCT__entry(__field(u32, iir)),TP_fast_assign(__entry->iir = iir;),TP_printk("forwarded with iir 0x%08x", __entry->iir) );#endif /* _I915_TRACE_H_ */
/* This part must be outside protection */
1.9.1
Intel-gfx mailing list Intel-gfx@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/intel-gfx
-- Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
On Thu, Oct 16, 2014 at 03:05:07PM +0000, Cheng, Yao wrote:
Ok, bunch of comments. First a high-level one: I think this qualifies as a new subsystem of i915, and so it would be good to extract this into a new file (i915_ved.c maybe), including adding kerneldoc for the setup function, a short DOC: overview section and pulling it all into the drm kerneldoc (probably a new subsection in the driver core section).
Aside from the lack of documentation just a few small comments below. Overall I really like how cleanly we can integrate vxd support into i915, so good work.
I915_ved.c sounds to be a good place for these code, thx for this suggestion!
For the kerneldoc, I'll add a subsection in the core section for your review.
Yeah, i915 driver core section sounds like the right place. Btw there's a small blog post from me with some tips for how to go about this:
http://blog.ffwll.ch/2014/06/documentation-for-drmi915.html
+static void valleyview_ved_cleanup(struct drm_device *dev) {
- int irq;
- struct drm_i915_private *dev_priv = dev->dev_private;
- irq = platform_get_irq(dev_priv->ved_platdev, 0);
- if (irq >= 0)
irq_free_desc(irq);- platform_device_unregister(dev_priv->ved_platdev);
I think you should unregister the platform device _before_ you free the irq. Otherwise the driver cleanup might freak out. Aside: Does the module reload test for i915 in i-g-t still work with the vxd driver loaded on vlv? Iirc you need to manually unload the vxd driver first to avoid inherit races in the platform device support code, so if that's the case you need to supply a patch for igt, too.
Sorry, what is i-g-t? I'll follow your suggestion, test i-g-t, and patch it if needed.
i-g-t is the i915 kernel driver regression test suite. For a quick intro see:
http://blog.ffwll.ch/2013/08/recent-drmi915-testsuite-improvements.html
The igt repo is at
http://cgit.freedesktop.org/xorg/app/intel-gpu-tools/
The README file in there also has some good information.
Cheers, Daniel
Probes vxd392 on Baytrail platform and create a new drm device for hardware accelerated video decoding.
Signed-off-by: Yao Cheng yao.cheng@intel.com --- drivers/gpu/drm/Kconfig | 2 + drivers/gpu/drm/Makefile | 1 + drivers/gpu/drm/ipvr/Kconfig | 16 + drivers/gpu/drm/ipvr/Makefile | 21 + drivers/gpu/drm/ipvr/ipvr_buffer.c | 1177 +++++++++++++++++++++++++++++++++ drivers/gpu/drm/ipvr/ipvr_buffer.h | 184 ++++++ drivers/gpu/drm/ipvr/ipvr_debug.c | 263 ++++++++ drivers/gpu/drm/ipvr/ipvr_debug.h | 50 ++ drivers/gpu/drm/ipvr/ipvr_drm.h | 265 ++++++++ drivers/gpu/drm/ipvr/ipvr_drv.c | 776 ++++++++++++++++++++++ drivers/gpu/drm/ipvr/ipvr_drv.h | 464 +++++++++++++ drivers/gpu/drm/ipvr/ipvr_exec.c | 530 +++++++++++++++ drivers/gpu/drm/ipvr/ipvr_exec.h | 68 ++ drivers/gpu/drm/ipvr/ipvr_fence.c | 550 ++++++++++++++++ drivers/gpu/drm/ipvr/ipvr_fence.h | 68 ++ drivers/gpu/drm/ipvr/ipvr_gem.c | 248 +++++++ drivers/gpu/drm/ipvr/ipvr_gem.h | 66 ++ drivers/gpu/drm/ipvr/ipvr_mmu.c | 807 +++++++++++++++++++++++ drivers/gpu/drm/ipvr/ipvr_mmu.h | 135 ++++ drivers/gpu/drm/ipvr/ipvr_trace.c | 11 + drivers/gpu/drm/ipvr/ipvr_trace.h | 296 +++++++++ drivers/gpu/drm/ipvr/ved_cmd.c | 1269 ++++++++++++++++++++++++++++++++++++ drivers/gpu/drm/ipvr/ved_cmd.h | 104 +++ drivers/gpu/drm/ipvr/ved_ec.c | 584 +++++++++++++++++ drivers/gpu/drm/ipvr/ved_ec.h | 207 ++++++ drivers/gpu/drm/ipvr/ved_fw.c | 660 +++++++++++++++++++ drivers/gpu/drm/ipvr/ved_fw.h | 73 +++ drivers/gpu/drm/ipvr/ved_init.c | 829 +++++++++++++++++++++++ drivers/gpu/drm/ipvr/ved_init.h | 61 ++ drivers/gpu/drm/ipvr/ved_msg.h | 364 +++++++++++ drivers/gpu/drm/ipvr/ved_pm.c | 392 +++++++++++ drivers/gpu/drm/ipvr/ved_pm.h | 55 ++ drivers/gpu/drm/ipvr/ved_reg.h | 609 +++++++++++++++++ 33 files changed, 11205 insertions(+) create mode 100644 drivers/gpu/drm/ipvr/Kconfig create mode 100644 drivers/gpu/drm/ipvr/Makefile create mode 100644 drivers/gpu/drm/ipvr/ipvr_buffer.c create mode 100644 drivers/gpu/drm/ipvr/ipvr_buffer.h create mode 100644 drivers/gpu/drm/ipvr/ipvr_debug.c create mode 100644 drivers/gpu/drm/ipvr/ipvr_debug.h create mode 100644 drivers/gpu/drm/ipvr/ipvr_drm.h create mode 100644 drivers/gpu/drm/ipvr/ipvr_drv.c create mode 100644 drivers/gpu/drm/ipvr/ipvr_drv.h create mode 100644 drivers/gpu/drm/ipvr/ipvr_exec.c create mode 100644 drivers/gpu/drm/ipvr/ipvr_exec.h create mode 100644 drivers/gpu/drm/ipvr/ipvr_fence.c create mode 100644 drivers/gpu/drm/ipvr/ipvr_fence.h create mode 100644 drivers/gpu/drm/ipvr/ipvr_gem.c create mode 100644 drivers/gpu/drm/ipvr/ipvr_gem.h create mode 100644 drivers/gpu/drm/ipvr/ipvr_mmu.c create mode 100644 drivers/gpu/drm/ipvr/ipvr_mmu.h create mode 100644 drivers/gpu/drm/ipvr/ipvr_trace.c create mode 100644 drivers/gpu/drm/ipvr/ipvr_trace.h create mode 100644 drivers/gpu/drm/ipvr/ved_cmd.c create mode 100644 drivers/gpu/drm/ipvr/ved_cmd.h create mode 100644 drivers/gpu/drm/ipvr/ved_ec.c create mode 100644 drivers/gpu/drm/ipvr/ved_ec.h create mode 100644 drivers/gpu/drm/ipvr/ved_fw.c create mode 100644 drivers/gpu/drm/ipvr/ved_fw.h create mode 100644 drivers/gpu/drm/ipvr/ved_init.c create mode 100644 drivers/gpu/drm/ipvr/ved_init.h create mode 100644 drivers/gpu/drm/ipvr/ved_msg.h create mode 100644 drivers/gpu/drm/ipvr/ved_pm.c create mode 100644 drivers/gpu/drm/ipvr/ved_pm.h create mode 100644 drivers/gpu/drm/ipvr/ved_reg.h
diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig index e3b4b0f..ad7585d 100644 --- a/drivers/gpu/drm/Kconfig +++ b/drivers/gpu/drm/Kconfig @@ -165,6 +165,8 @@ config DRM_SAVAGE Choose this option if you have a Savage3D/4/SuperSavage/Pro/Twister chipset. If M is selected the module will be called savage.
+source "drivers/gpu/drm/ipvr/Kconfig" + source "drivers/gpu/drm/exynos/Kconfig"
source "drivers/gpu/drm/vmwgfx/Kconfig" diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile index 9292a76..8ec4bda 100644 --- a/drivers/gpu/drm/Makefile +++ b/drivers/gpu/drm/Makefile @@ -41,6 +41,7 @@ obj-$(CONFIG_DRM_RADEON)+= radeon/ obj-$(CONFIG_DRM_MGA) += mga/ obj-$(CONFIG_DRM_I810) += i810/ obj-$(CONFIG_DRM_I915) += i915/ +obj-$(CONFIG_DRM_IPVR) += ipvr/ obj-$(CONFIG_DRM_MGAG200) += mgag200/ obj-$(CONFIG_DRM_CIRRUS_QEMU) += cirrus/ obj-$(CONFIG_DRM_SIS) += sis/ diff --git a/drivers/gpu/drm/ipvr/Kconfig b/drivers/gpu/drm/ipvr/Kconfig new file mode 100644 index 0000000..6516590 --- /dev/null +++ b/drivers/gpu/drm/ipvr/Kconfig @@ -0,0 +1,16 @@ +config DRM_IPVR + tristate "Baytrail VP8 video decode driver" + depends on DRM + depends on DRM_I915 + select SHMEM + select TMPFS + default m + help + Choose this option if you want to enable hardware accelerated VP8 decoding on Baytrail platform + +config DRM_IPVR_EC + bool "[EXPERIMENTAL] error concealment for VP8" + depends on DRM_IPVR + default y + help + Choose this option if you want to enable hardware error concealment for IPVR VP8 diff --git a/drivers/gpu/drm/ipvr/Makefile b/drivers/gpu/drm/ipvr/Makefile new file mode 100644 index 0000000..e25a7d3 --- /dev/null +++ b/drivers/gpu/drm/ipvr/Makefile @@ -0,0 +1,21 @@ + +ccflags-y := -Iinclude/drm + +ipvr-y := \ + ipvr_gem.o \ + ipvr_drv.o \ + ipvr_buffer.o \ + ipvr_exec.o \ + ipvr_fence.o \ + ipvr_mmu.o \ + ipvr_debug.o \ + ipvr_trace.o \ + ved_pm.o \ + ved_ec.o \ + ved_cmd.o \ + ved_fw.o \ + ved_init.o + +obj-$(CONFIG_DRM_IPVR) += ipvr.o + +CFLAGS_ipvr_trace.o := -I$(src) diff --git a/drivers/gpu/drm/ipvr/ipvr_buffer.c b/drivers/gpu/drm/ipvr/ipvr_buffer.c new file mode 100644 index 0000000..a2fe587 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_buffer.c @@ -0,0 +1,1177 @@ +/************************************************************************** + * ipvr_buffer.c: IPVR buffer creation, destory, map etc + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + +#include <drmP.h> +#include "ipvr_buffer.h" +#include "ipvr_trace.h" + +extern int fake_bo_debug; + +/* Ensure that the associated pages are gathered from the backing storage + * and pinned into our object. ipvr_gem_object_get_pages() may be called + * multiple times before they are released by a single call to + * ipvr_gem_object_put_pages() - once the pages are no longer referenced + * either as a result of memory pressure (reaping pages under the shrinker) + * or as the object is itself released. + */ +/* for fake debug, will allocate bo with alloc_page, instead of using sg */ +int32_t ipvr_gem_object_get_pages(struct drm_ipvr_gem_object *obj) +{ + struct drm_ipvr_private *dev_priv = obj->base.dev->dev_private; + const struct drm_ipvr_gem_object_ops *ops = obj->ops; + int32_t ret, r; + + if (obj->sg_table) + return 0; + + BUG_ON(obj->pages_pin_count); + + if (fake_bo_debug) { + gfp_t gfp_flags = GFP_USER | GFP_DMA32 | __GFP_IO; + struct page *p = NULL; + obj->fake_pages = + drm_calloc_large(obj->base.size / 4096, sizeof(void*)); + IPVR_DEBUG_GENERAL("fake call alloc_page to alloc pages.\n"); + for (r = 0; r < obj->base.size / 4096; ++r) { + p = alloc_page(gfp_flags); + if (!p) { + IPVR_ERROR("Unable to allocate page\n"); + return -ENOMEM; + } + obj->fake_pages[r] = p; + switch (obj->cache_level) { + case IPVR_CACHE_NONE: + ret = set_pages_uc(p, 1); + break; + case IPVR_CACHE_WC: + ret = set_memory_wc( + (unsigned long)page_address(p), 1); + break; + case IPVR_CACHE_WB: + ret = set_pages_wb(p, 1); + break; + default: + ret = -EINVAL; + break; + } + if (ret) { + IPVR_DEBUG_WARN("failed to set page cache.\n"); + return -ENOMEM; + } + } + } + /* will call ipvr_gem_object_get_pages_mmu */ + ret = ops->get_pages(obj); + if (ret) { + IPVR_ERROR("failed to call ops->get_pages\n"); + return ret; + } + + /* do we need lock here? */ + list_add_tail(&obj->global_list, &dev_priv->ipvr_mm.unbound_list); + return 0; +} + +int32_t ipvr_gem_object_put_pages(struct drm_ipvr_gem_object *obj) +{ + const struct drm_ipvr_gem_object_ops *ops = obj->ops; + + if (obj->sg_table == NULL) + return 0; + + if (obj->pages_pin_count) + return -EBUSY; + + list_del(&obj->global_list); + + ops->put_pages(obj); + obj->sg_table = NULL; + + return 0; +} + +static int32_t ipvr_gem_object_get_pages_sg(struct drm_ipvr_gem_object *obj) +{ + uint32_t page_count, i; + struct address_space *mapping; + struct sg_table *st; + struct scatterlist *sg; + struct sg_page_iter sg_iter; + struct page *page = NULL; + struct page **page_array = NULL; + int carray = 0; + unsigned long last_pfn = 0; /* suppress gcc warning */ + gfp_t gfp; + int32_t ret; + + /* Assert that the object is not currently in any GPU domain. As it + * wasn't in the MMU, there shouldn't be any way it could have been in + * a GPU cache + */ + BUG_ON(obj->base.read_domains & IPVR_GEM_DOMAIN_GPU); + BUG_ON(obj->base.write_domain & IPVR_GEM_DOMAIN_GPU); + + st = kmalloc(sizeof(*st), GFP_KERNEL); + if (st == NULL) + return -ENOMEM; + + page_count = obj->base.size / PAGE_SIZE; + + if (sg_alloc_table(st, page_count, GFP_KERNEL)) { + sg_free_table(st); + kfree(st); + return -ENOMEM; + } + + page_array = kmalloc(page_count * sizeof(struct page *), GFP_KERNEL); + if (unlikely(!page_array)) { + sg_free_table(st); + kfree(st); + return -ENOMEM; + } + /* Get the list of pages out of our struct file. They'll be pinned + * at this point until we release them. + * + * Fail silently without starting the shrinker + */ + mapping = file_inode(obj->base.filp)->i_mapping; + gfp = mapping_gfp_mask(mapping); + /* todo: need set correct gfp */ + gfp |= __GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD; + gfp |= GFP_DMA32 | __GFP_IO; + sg = st->sgl; + st->nents = 0; + for (i = 0; i < page_count; i++) { + page = shmem_read_mapping_page_gfp(mapping, i, gfp); + if (IS_ERR(page)) { + /* doesn't support shrink + * since ipvr has 4GiB address space */ + goto err_pages; + } + page_array[carray++] = page; + +#ifdef CONFIG_SWIOTLB + if (swiotlb_nr_tbl()) { + st->nents++; + sg_set_page(sg, page, PAGE_SIZE, 0); + sg = sg_next(sg); + continue; + } +#endif + if (!i || page_to_pfn(page) != last_pfn + 1) { + if (i) + sg = sg_next(sg); + st->nents++; + sg_set_page(sg, page, PAGE_SIZE, 0); + } else { + sg->length += PAGE_SIZE; + } + last_pfn = page_to_pfn(page); + } +#ifdef CONFIG_SWIOTLB + if (!swiotlb_nr_tbl()) +#endif + sg_mark_end(sg); + obj->sg_table = st; + + switch (obj->cache_level) { + case IPVR_CACHE_NONE: + ret = set_pages_array_uc(page_array, carray); + break; + case IPVR_CACHE_WC: + ret = set_pages_array_wc(page_array, carray); + break; + case IPVR_CACHE_WB: + ret = set_pages_array_wb(page_array, carray); + break; + default: + ret = -EINVAL; + break; + } + if (ret) { + IPVR_DEBUG_WARN("failed to set page cache.\n"); + goto err_pages; + } + kfree(page_array); + return 0; + +err_pages: + sg_mark_end(sg); + for_each_sg_page(st->sgl, &sg_iter, st->nents, 0) + page_cache_release(sg_page_iter_page(&sg_iter)); + sg_free_table(st); + kfree(st); + if (page_array) + kfree(page_array); + return PTR_ERR(page); +} + +static bool cpu_cache_is_coherent(struct drm_device *dev, + enum ipvr_cache_level level) +{ + /* todo: need check if cache snoop is enabled */ + if (level == IPVR_CACHE_WB) + return false; + else + return true; +} + +bool ipvr_gem_clflush_object(struct drm_ipvr_gem_object *obj, bool force) +{ + /* If we don't have a page list set up, then we're not pinned + * to GPU, and we can ignore the cache flush because it'll happen + * again at bind time. + */ + if (obj->sg_table == NULL) + return false; + + /* If the GPU is snooping the contents of the CPU cache, + * we do not need to manually clear the CPU cache lines. However, + * the caches are only snooped when the render cache is + * flushed/invalidated. As we always have to emit invalidations + * and flushes when moving into and out of the RENDER domain, correct + * snooping behaviour occurs naturally as the result of our domain + * tracking. + */ + if (!force && cpu_cache_is_coherent(obj->base.dev, obj->cache_level)) + return false; + + drm_clflush_sg(obj->sg_table); + + return true; +} + +/** + * Moves a single object to the CPU read, and possibly write domain. + * + * This function returns when the move is complete, including waiting on + * flushes to occur. + */ +int32_t +ipvr_gem_object_set_to_cpu_domain(struct drm_ipvr_gem_object *obj, bool write) +{ + uint32_t old_write_domain, old_read_domains; + int32_t ret; + + if (obj->base.write_domain == IPVR_GEM_DOMAIN_CPU) + return 0; + + ret = ipvr_fence_wait(obj->fence, true, false); + if (ret) + return ret; + + old_write_domain = obj->base.write_domain; + old_read_domains = obj->base.read_domains; + + /* Flush the CPU cache if it's still invalid. */ + if ((obj->base.read_domains & IPVR_GEM_DOMAIN_CPU) == 0) { + ipvr_gem_clflush_object(obj, false); + + obj->base.read_domains |= IPVR_GEM_DOMAIN_CPU; + } + + /* It should now be out of any other write domains, and we can update + * the domain values for our changes. + */ + BUG_ON((obj->base.write_domain & ~IPVR_GEM_DOMAIN_CPU) != 0); + + /* If we're writing through the CPU, then the GPU read domains will + * need to be invalidated at next use. + */ + if (write) { + obj->base.read_domains = IPVR_GEM_DOMAIN_CPU; + obj->base.write_domain = IPVR_GEM_DOMAIN_CPU; + } + + return 0; +} + +static void ipvr_gem_object_put_pages_sg(struct drm_ipvr_gem_object *obj) +{ + struct sg_page_iter sg_iter; + int32_t ret; + + ret = ipvr_gem_object_set_to_cpu_domain(obj, true); + if (ret) { + /* In the event of a disaster, abandon all caches and + * hope for the best. + */ + WARN_ON(ret != -EIO); + ipvr_gem_clflush_object(obj, true); + obj->base.read_domains = IPVR_GEM_DOMAIN_CPU; + obj->base.write_domain = IPVR_GEM_DOMAIN_CPU; + } + + IPVR_DEBUG_GENERAL("start to free pages.\n"); + for_each_sg_page(obj->sg_table->sgl, + &sg_iter, obj->sg_table->nents, 0) { + struct page *page = sg_page_iter_page(&sg_iter); + + if (obj->dirty) + set_page_dirty(page); + /* need set back to wb before release page */ + ret = set_pages_wb(page, 1); + if (ret) + IPVR_ERROR("failed to set page as wb.\n"); + page_cache_release(page); + } + obj->dirty = 0; + + sg_free_table(obj->sg_table); + kfree(obj->sg_table); +} + +static const struct drm_ipvr_gem_object_ops ipvr_gem_object_ops = { + .get_pages = ipvr_gem_object_get_pages_sg, + .put_pages = ipvr_gem_object_put_pages_sg, +}; + +/* All the new VM stuff */ +unsigned long ipvr_gem_obj_mmu_offset(struct drm_ipvr_gem_object *obj) +{ + struct drm_ipvr_private *dev_priv = obj->base.dev->dev_private; + struct ipvr_vma *vma; + + struct ipvr_address_space *vm = &dev_priv->addr_space; + + BUG_ON(list_empty(&obj->vma_list)); + list_for_each_entry(vma, &obj->vma_list, vma_link) { + /* todo: for user ptr obj, need consider offset in page in future */ + if (vma->vm == vm) { + IPVR_DEBUG_GENERAL("node start is 0x%lx.\n", + vma->node.start); + return vma->node.start; + } + } + + IPVR_ERROR("failed to find vma corresponding to this bo.\n"); + return IPVR_ERR_OFFSET(-EINVAL); +} + +void *ipvr_gem_object_alloc(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + if (dev_priv == NULL) + IPVR_ERROR("error: dev_prive is NULL.\n"); + return kmem_cache_alloc(dev_priv->ipvr_bo_slab, GFP_KERNEL | __GFP_ZERO); +} + +void ipvr_gem_object_free(struct drm_ipvr_gem_object *obj) +{ + struct drm_ipvr_private *dev_priv = obj->base.dev->dev_private; + kmem_cache_free(dev_priv->ipvr_bo_slab, obj); +} + +/* some bookkeeping */ +static void +ipvr_gem_info_add_obj(struct drm_ipvr_private *dev_priv, size_t size) +{ + spin_lock(&dev_priv->ipvr_mm.object_stat_lock); + dev_priv->ipvr_mm.object_count++; + dev_priv->ipvr_mm.object_memory += size; + spin_unlock(&dev_priv->ipvr_mm.object_stat_lock); +} + +static void +ipvr_gem_info_remove_obj(struct drm_ipvr_private *dev_priv, size_t size) +{ + spin_lock(&dev_priv->ipvr_mm.object_stat_lock); + dev_priv->ipvr_mm.object_count--; + dev_priv->ipvr_mm.object_memory -= size; + spin_unlock(&dev_priv->ipvr_mm.object_stat_lock); +} + +void ipvr_gem_object_init(struct drm_ipvr_gem_object *obj, + const struct drm_ipvr_gem_object_ops *ops) +{ + INIT_LIST_HEAD(&obj->global_list); + INIT_LIST_HEAD(&obj->vma_list); + + obj->ops = ops; + + ipvr_gem_info_add_obj(obj->base.dev->dev_private, obj->base.size); +} + +struct drm_ipvr_gem_object * +ipvr_gem_alloc_object(struct drm_device *dev, size_t size) +{ + struct drm_ipvr_gem_object *obj; + struct address_space *mapping; + gfp_t mask; + + obj = ipvr_gem_object_alloc(dev); + if (obj == NULL) + return NULL; + memset(obj, 0, sizeof(*obj)); + + if (drm_gem_object_init(dev, &obj->base, size) != 0) { + ipvr_gem_object_free(obj); + return NULL; + } + + init_waitqueue_head(&obj->event_queue); + atomic_set(&obj->cpu_writers, 0); + /* todo: need set correct mask */ + mask = GFP_HIGHUSER | __GFP_RECLAIMABLE; + + /* ipvr cannot relocate objects above 4GiB. */ + mask &= ~__GFP_HIGHMEM; + mask |= __GFP_DMA32; + + mapping = file_inode(obj->base.filp)->i_mapping; + mapping_set_gfp_mask(mapping, mask); + + ipvr_gem_object_init(obj, &ipvr_gem_object_ops); + + obj->base.write_domain = IPVR_GEM_DOMAIN_CPU; + obj->base.read_domains = IPVR_GEM_DOMAIN_CPU; + + return obj; +} + +int32_t ipvr_gem_mmu_bind_object(struct drm_ipvr_gem_object *obj) +{ + struct drm_ipvr_private *dev_priv = obj->base.dev->dev_private; + uint32_t ret, type = 0; + const unsigned long entry = ipvr_gem_obj_mmu_offset(obj); + + if (IPVR_IS_ERR(entry)) { + return IPVR_OFFSET_ERR(entry); + } + + IPVR_DEBUG_GENERAL("entry is 0x%lx, size is %zu, nents is %d.\n", + entry, obj->base.size, obj->sg_table->nents); + + /* todo: it is better also pass RO, WO info */ + type = (obj->cache_level == IPVR_CACHE_WB) ? + IPVR_MMU_CACHED_MEMORY : 0; + + /* should not use dev_priv->pf_pd */ + if (fake_bo_debug) { + ret = ipvr_mmu_insert_pages(dev_priv->mmu->default_pd, + obj->fake_pages, + entry, obj->base.size >> PAGE_SHIFT, + 0, + 0, type); + } else { + uint32_t i = 0; + struct page **pages; + struct sg_page_iter sg_iter; + pages = drm_malloc_ab(obj->base.size >> PAGE_SHIFT, + sizeof(*pages)); + if (pages == NULL) { + IPVR_ERROR("Failed to get space for pages\n"); + return -ENOMEM; + } + for_each_sg_page(obj->sg_table->sgl, &sg_iter, + obj->sg_table->nents, 0) + pages[i++] = sg_page_iter_page(&sg_iter); + ret = ipvr_mmu_insert_pages(dev_priv->mmu->default_pd, + pages, entry, obj->base.size >> PAGE_SHIFT, + 0, 0, type); + if (pages) + drm_free_large(pages); + } + return ret; +} + +int32_t ipvr_gem_create(struct drm_file *file_priv, struct drm_device *dev, + uint64_t size, uint32_t tiling, uint32_t cache_level, + uint64_t *rsize_p, uint32_t *handle_p, + uint64_t *offset_p) +{ + struct drm_ipvr_gem_object *obj; + int32_t ret; + uint32_t handle; + struct drm_ipvr_private *dev_priv = dev->dev_private; + IPVR_DEBUG_GENERAL("create bo size is %lld, tiling is %d, " + "cache level is %d.\n", + size, tiling, cache_level); + + size = roundup(size, PAGE_SIZE); + if (size == 0) + return -EINVAL; + *rsize_p = size; + + /* Allocate the new object */ + obj = ipvr_gem_alloc_object(dev, size); + if (obj == NULL) + return -ENOMEM; + obj->drv_name = "ipvr"; + obj->fence = NULL; + obj->tiling = tiling; + obj->cache_level = cache_level; + ret = drm_gem_handle_create(file_priv, &obj->base, &handle); + /* drop reference from allocate - handle holds it now */ + drm_gem_object_unreference_unlocked(&obj->base); + if (ret) + return ret; + *handle_p = handle; + + /* force to get pages at allocation stage */ + ret = ipvr_gem_object_bind_to_vm(obj, &dev_priv->addr_space, 4096); + if (ret) { + ipvr_gem_free_object(&obj->base); + return ret; + } + + ret = ipvr_gem_mmu_bind_object(obj); + if (ret) { + ipvr_gem_free_object(&obj->base); + return ret; + } + *offset_p = ipvr_gem_obj_mmu_offset(obj); + if (IPVR_IS_ERR(*offset_p)) + return IPVR_OFFSET_ERR(*offset_p); + + IPVR_DEBUG_GENERAL("bo create done, gpu offset: 0x%llx.\n", *offset_p); + + trace_ipvr_gem_create(obj, *offset_p); + + return 0; +} + +struct drm_ipvr_gem_object * +ipvr_gem_obj_create_and_bind(struct drm_device *dev, size_t size) +{ + uint32_t ret; + struct drm_ipvr_gem_object *obj; + struct drm_ipvr_private *dev_priv = dev->dev_private; + + size = roundup(size, PAGE_SIZE); + if (size == 0) { + IPVR_ERROR("Passed size is 0, not correct.\n"); + return NULL; + } + + /* Allocate the new object */ + obj = ipvr_gem_alloc_object(dev, size); + if (obj == NULL) { + IPVR_ERROR("Failed to allocate ipvr object.\n"); + return NULL; + } + + ret = ipvr_gem_object_bind_to_vm(obj, &dev_priv->addr_space, 4096); + if (ret) { + IPVR_ERROR("Failed to bind obj to vm.\n"); + ipvr_gem_free_object(&obj->base); + return NULL; + } + ret = ipvr_gem_mmu_bind_object(obj); + if (ret) { + IPVR_ERROR("Failed to bind obj to mmu.\n"); + ipvr_gem_free_object(&obj->base); + return NULL; + } + + return obj; +} + +struct ipvr_vma *ipvr_gem_obj_to_vma(struct drm_ipvr_gem_object *obj, + struct ipvr_address_space *vm) +{ + struct ipvr_vma *vma; + list_for_each_entry(vma, &obj->vma_list, vma_link) + if (vma->vm == vm) + return vma; + + return NULL; +} + +struct ipvr_vma *ipvr_gem_vma_create(struct drm_ipvr_gem_object *obj, + struct ipvr_address_space *vm) +{ + struct ipvr_vma *vma = kzalloc(sizeof(*vma), GFP_KERNEL); + if (vma == NULL) + return ERR_PTR(-ENOMEM); + + INIT_LIST_HEAD(&vma->vma_link); + INIT_LIST_HEAD(&vma->mm_list); + vma->vm = vm; + vma->obj = obj; + + list_add_tail(&vma->vma_link, &obj->vma_list); + + return vma; +} + +static void ipvr_gem_vma_destroy(struct ipvr_vma *vma) +{ + WARN_ON(vma->node.allocated); + list_del(&vma->vma_link); + kfree(vma); +} + +struct ipvr_vma * +ipvr_gem_obj_lookup_or_create_vma(struct drm_ipvr_gem_object *obj, + struct ipvr_address_space *vm) +{ + struct ipvr_vma *vma; + + vma = ipvr_gem_obj_to_vma(obj, vm); + if (!vma) + vma = ipvr_gem_vma_create(obj, vm); + + return vma; +} + +void ipvr_gem_object_pin_pages(struct drm_ipvr_gem_object *obj) +{ + BUG_ON(obj->sg_table == NULL); + obj->pages_pin_count++; +} + +void ipvr_gem_object_unpin_pages(struct drm_ipvr_gem_object *obj) +{ + BUG_ON(obj->pages_pin_count == 0); + obj->pages_pin_count--; +} + +static bool ipvr_gem_valid_mmu_space(struct drm_device *dev, + struct drm_mm_node *mmu_space, + uint32_t cache_level) +{ + struct drm_mm_node *other; + + if (!drm_mm_node_allocated(mmu_space)) + return true; + + if (list_empty(&mmu_space->node_list)) + return true; + + other = list_entry(mmu_space->node_list.prev, + struct drm_mm_node, node_list); + if (other->allocated && !other->hole_follows && + other->color != cache_level) + return false; + + other = list_entry(mmu_space->node_list.next, + struct drm_mm_node, node_list); + if (other->allocated && !mmu_space->hole_follows && + other->color != cache_level) + return false; + + return true; +} + +/** + * Finds free space in the MMU aperture and binds the object there. + */ +int32_t ipvr_gem_object_bind_to_vm(struct drm_ipvr_gem_object *obj, + struct ipvr_address_space *vm, + uint32_t alignment) +{ + struct drm_device *dev = obj->base.dev; + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ipvr_vma *vma; + unsigned long start, end, size; + int32_t ret; + struct drm_mm *mm; + if (obj->tiling) { + mm = &vm->tiling_mm; + start = vm->tiling_start; + end = vm->tiling_start + vm->tiling_total; + } else { + mm = &vm->linear_mm; + start = vm->linear_start; + end = vm->linear_start + vm->linear_total; + } + size = obj->base.size; + + ret = ipvr_gem_object_get_pages(obj); + if (ret) { + IPVR_ERROR("failed to call ipvr_gem_object_get_pages.\n"); + return ret; + } + + ipvr_gem_object_pin_pages(obj); + + vma = ipvr_gem_obj_lookup_or_create_vma(obj, vm); + if (IS_ERR(vma)) { + ret = PTR_ERR(vma); + IPVR_ERROR("failed on ipvr_gem_obj_lookup_or_create_vma.\n"); + goto err_unpin; + } + + /* For now we only ever use 1 vma per object */ + BUG_ON(!list_is_singular(&obj->vma_list)); + + IPVR_DEBUG_GENERAL("call drm_mm_insert_node_in_range_generic.\n"); + ret = drm_mm_insert_node_in_range_generic(mm, &vma->node, size, + alignment, obj->cache_level, + start, end, + DRM_MM_SEARCH_DEFAULT, + DRM_MM_CREATE_DEFAULT); + if (ret) { + /* + * ipvr doesn't support shrink so far since it has 4GiB addr space + */ + IPVR_ERROR("failed on drm_mm_insert_node_in_range_generic.\n"); + goto err_free_vma; + } + + if (unlikely(!ipvr_gem_valid_mmu_space(dev, &vma->node, + obj->cache_level))) { + ret = -EINVAL; + IPVR_ERROR("failed on ipvr_gem_valid_mmu_space.\n"); + goto err_remove_node; + } + + list_move_tail(&obj->global_list, &dev_priv->ipvr_mm.bound_list); + list_add_tail(&vma->mm_list, &vm->inactive_list); + + /* ipvr_gem_verify_gtt can be added here for debug */ + return 0; + +err_remove_node: + drm_mm_remove_node(&vma->node); +err_free_vma: + ipvr_gem_vma_destroy(vma); +err_unpin: + ipvr_gem_object_unpin_pages(obj); + return ret; +} + +void *ipvr_gem_object_vmap(struct drm_ipvr_gem_object *obj) +{ + int32_t i; + void *addr = NULL; + struct sg_page_iter sg_iter; + struct page **pages; + + if (fake_bo_debug) { + addr = vmap(obj->fake_pages, obj->base.size / 4096, + VM_MAP, pgprot_writecombine(PAGE_KERNEL)); + if (addr == NULL) { + IPVR_ERROR("Failed to vmap pages\n"); + return ERR_PTR(-ENOMEM); + } + return addr; + } + + pages = drm_malloc_ab(obj->base.size >> PAGE_SHIFT, sizeof(*pages)); + if (pages == NULL) { + IPVR_ERROR("Failed to get space for pages\n"); + goto finish; + } + + i = 0; + for_each_sg_page(obj->sg_table->sgl, &sg_iter, + obj->sg_table->nents, 0) { + pages[i] = sg_page_iter_page(&sg_iter); + i++; + } + + addr = vmap(pages, i, VM_MAP, pgprot_writecombine(PAGE_KERNEL)); + if (addr == NULL) { + IPVR_ERROR("Failed to vmap pages\n"); + goto finish; + } + +finish: + if (pages) + drm_free_large(pages); + return addr; +} + +int32_t ipvr_gem_object_finish_gpu(struct drm_ipvr_gem_object *obj) +{ + int32_t ret = 0; + + if ((obj->base.read_domains & IPVR_GEM_DOMAIN_GPU) == 0) + return ret; + + /* Ensure that we invalidate the GPU's caches and TLBs. */ + obj->base.read_domains &= ~IPVR_GEM_DOMAIN_GPU; + return ret; +} + +/** + * ipvr_gem_release_mmap - remove physical page mappings + * @obj: obj in question + * + * If the object has been moved out of the aperture, + * then pages mapped into userspace must be revoked. Removing the + * mapping will then trigger a page fault on the next user access, allowing + * fixup by ipvr_gem_fault(). + */ +void ipvr_gem_release_mmap(struct drm_ipvr_gem_object *obj) +{ + /* todo: + * will not call drm_gem_create_mmap_offset, + * ipvr dont' need use mmap for dev map + */ + drm_vma_node_unmap(&obj->base.vma_node, + obj->base.dev->anon_inode->i_mapping); + + /* todo: + * it seems it need cause kernel panic, disable vm_munmap as workaround + */ + if (obj->mmap_base) { + IPVR_DEBUG_GENERAL("call vm_munmap to unmap bo, base is 0x%lx.\n", + obj->mmap_base); + obj->mmap_base = 0; + } +} + +int32_t ipvr_gem_object_vma_unbind(struct ipvr_vma *vma) +{ + struct drm_ipvr_gem_object *obj = vma->obj; + drm_ipvr_private_t *dev_priv = obj->base.dev->dev_private; + unsigned long entry; + int32_t ret; + + if (list_empty(&vma->vma_link)) + return 0; + + if (!drm_mm_node_allocated(&vma->node)) + goto destroy; + + if (obj->pin_count) + return -EBUSY; + + BUG_ON(obj->sg_table == NULL); + entry = ipvr_gem_obj_mmu_offset(obj); + if (IPVR_IS_ERR(entry)) + return IPVR_OFFSET_ERR(entry); + + ret = ipvr_gem_object_finish_gpu(obj); + if (ret) + return ret; + + /* Continue on if we fail due to EIO, the GPU is hung so we + * should be safe and we need to cleanup or else we might + * cause memory corruption through use-after-free. + */ + ipvr_gem_release_mmap(obj); + + /* remove the bo from the mmu */ + ipvr_mmu_remove_pages(dev_priv->mmu->default_pd, entry, + obj->base.size >> PAGE_SHIFT, 0, 0); + + ipvr_gem_object_unpin_pages(obj); + + list_del(&vma->mm_list); + + drm_mm_remove_node(&vma->node); + +destroy: + ipvr_gem_vma_destroy(vma); + + /* Since the unbound list is global, only move to that list if + * no more VMAs exist. + * NB: Until we have real VMAs there will only ever be one */ + WARN_ON(!list_empty(&obj->vma_list)); + if (list_empty(&obj->vma_list)) + list_move_tail(&obj->global_list, + &dev_priv->ipvr_mm.unbound_list); + + return 0; +} + +static void ipvr_gem_object_free_mmap_offset(struct drm_ipvr_gem_object *obj) +{ + drm_gem_free_mmap_offset(&obj->base); +} + +/** + * Called after the last reference to the object has been lost. + * Must be called holding struct_ mutex + * + * Frees the object + */ +static void ipvr__gem_free_object(struct kref *kref) +{ + struct drm_gem_object *gem_obj = (struct drm_gem_object *)kref; + struct drm_device *dev = gem_obj->dev; + struct drm_ipvr_gem_object *obj = to_ipvr_bo(gem_obj); + drm_ipvr_private_t *dev_priv = dev->dev_private; + struct ipvr_vma *vma, *next; + int32_t ret; + + // FIXME: consider unlocked case + WARN_ON(!mutex_is_locked(&dev->struct_mutex)); + + if (obj->fence) { + ret = ipvr_fence_wait(obj->fence, true, false); + if (ret) + IPVR_DEBUG_WARN("Failed to wait fence signaled.\n"); + } + + obj->pin_count = 0; + /* NB: 0 or 1 elements */ + WARN_ON(!list_empty(&obj->vma_list) && + !list_is_singular(&obj->vma_list)); + list_for_each_entry_safe(vma, next, &obj->vma_list, vma_link) { + int ret = ipvr_gem_object_vma_unbind(vma); + if (WARN_ON(ret == -ERESTARTSYS)) { + bool was_interruptible; + + was_interruptible = dev_priv->ipvr_mm.interruptible; + dev_priv->ipvr_mm.interruptible = false; + + WARN_ON(ipvr_gem_object_vma_unbind(vma)); + + dev_priv->ipvr_mm.interruptible = was_interruptible; + } + } + + if (WARN_ON(obj->pages_pin_count)) + obj->pages_pin_count = 0; + ipvr_gem_object_put_pages(obj); + ipvr_gem_object_free_mmap_offset(obj); + + BUG_ON(obj->sg_table); + + if (obj->base.import_attach) + drm_prime_gem_destroy(&obj->base, NULL); + + if (obj->ops->release) + obj->ops->release(obj); + + drm_gem_object_release(&obj->base); + ipvr_gem_info_remove_obj(dev_priv, obj->base.size); + + trace_ipvr__gem_free_object(obj); + + ipvr_gem_object_free(obj); +} + +/* + * When the last reference to a GEM object is released the GEM core calls the + * drm_driver .gem_free_object() operation. That operation is mandatory for + * GEM-enabled drivers and must free the GEM object and all associated + * resources. +*/ +void ipvr_gem_free_object(struct drm_gem_object *gem_obj) +{ + ipvr__gem_free_object((struct kref *)gem_obj); +} + +static struct +ipvr_gem_userptr_object *to_userptr_object(struct drm_ipvr_gem_object *obj) +{ + return container_of(obj, struct ipvr_gem_userptr_object, gem); +} + +static int32_t ipvr_gem_userptr_get_pages_sg(struct drm_ipvr_gem_object *obj) { + struct ipvr_gem_userptr_object *vmap = to_userptr_object(obj); + int32_t num_pages = obj->base.size >> PAGE_SHIFT; + struct sg_table *st; + struct scatterlist *sg; + struct page **pvec; + int32_t n, pinned, ret; + + if (!access_ok(vmap->read_only ? VERIFY_READ : VERIFY_WRITE, + (char __user *)vmap->user_ptr, vmap->user_size)) { + IPVR_ERROR("access_ok check failed.\n"); + return -EFAULT; + } + + /* If userspace should engineer that these pages are replaced in + * the vma between us binding this page into the GTT and completion + * of rendering... Their loss. If they change the mapping of their + * pages they need to create a new bo to point to the new vma. + * + * However, that still leaves open the possibility of the vma + * being copied upon fork. Which falls under the same userspace + * synchronisation issue as a regular bo, except that this time + * the process may not be expecting that a particular piece of + * memory is tied to the GPU. + * + * Fortunately, we can hook into the mmu_notifier in order to + * discard the page references prior to anything nasty happening + * to the vma (discard or cloning) which should prevent the more + * egregious cases from causing harm. + */ + + pvec = kmalloc(num_pages * sizeof(struct page *), + GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY); + if (pvec == NULL) { + pvec = drm_malloc_ab(num_pages, sizeof(struct page *)); + if (pvec == NULL) { + IPVR_ERROR("pvec alloc failure\n"); + return -ENOMEM; + } + } + + pinned = __get_user_pages_fast(vmap->user_ptr, num_pages, + !vmap->read_only, pvec); + if (pinned < num_pages) { + struct mm_struct *mm = current->mm; + ret = 0; + down_read(&mm->mmap_sem); + ret = get_user_pages(current, mm, + vmap->user_ptr + (pinned << PAGE_SHIFT), + num_pages - pinned, + !vmap->read_only, 0, + pvec + pinned, + NULL); + up_read(&mm->mmap_sem); + if (ret > 0) + pinned += ret; + + if (obj->sg_table || pinned < num_pages) { + ret = obj->sg_table ? 0 : -EFAULT; + IPVR_ERROR("obj->sg_table is NULL\n"); + goto cleanup_pinned; + } + } + + st = kmalloc(sizeof(*st), GFP_KERNEL); + if (st == NULL) { + ret = -ENOMEM; + goto cleanup_pinned; + } + + if (sg_alloc_table(st, num_pages, GFP_KERNEL)) { + ret = -ENOMEM; + goto cleanup_st; + } + + for_each_sg(st->sgl, sg, num_pages, n) + sg_set_page(sg, pvec[n], PAGE_SIZE, 0); + drm_free_large(pvec); + + obj->sg_table = st; + return 0; + +cleanup_st: + kfree(st); +cleanup_pinned: + release_pages(pvec, pinned, 0); + drm_free_large(pvec); + return ret; +} + +static void ipvr_gem_userptr_put_pages_sg(struct drm_ipvr_gem_object *obj) { + struct scatterlist *sg; + int32_t i; + + for_each_sg(obj->sg_table->sgl, sg, obj->sg_table->nents, i) { + struct page *page = sg_page(sg); + + if (obj->dirty) + set_page_dirty(page); + + mark_page_accessed(page); + page_cache_release(page); + } + obj->dirty = 0; + + sg_free_table(obj->sg_table); + kfree(obj->sg_table); +} + +static const struct drm_ipvr_gem_object_ops ipvr_gem_userptr_ops = { + .get_pages = ipvr_gem_userptr_get_pages_sg, + .put_pages = ipvr_gem_userptr_put_pages_sg, +}; + +int32_t ipvr_gem_userptr(struct drm_file *file_priv, struct drm_device *dev, + uint64_t user_ptr, uint64_t user_size, + uint32_t cache_level, uint32_t tiling, uint32_t *handle_p, + uint64_t *offset_p) +{ + struct ipvr_gem_userptr_object *obj; + int32_t ret = 0; + uint32_t handle, num_pages; + loff_t first_data_page, last_data_page; + + struct drm_ipvr_private *dev_priv = dev->dev_private; + IPVR_DEBUG_GENERAL("create user bo userptr is %llx, size is %lld, " + "cache level is %d.\n", + user_ptr, user_size, cache_level); + + first_data_page = user_ptr / PAGE_SIZE; + last_data_page = (user_ptr + user_size - 1) / PAGE_SIZE; + num_pages = last_data_page - first_data_page + 1; + + /* only support page aligned buffer for now */ + if (offset_in_page(user_ptr) != 0 || (user_size & (PAGE_SIZE - 1)) != 0) + return -EINVAL; + + /* Allocate the new object */ + obj = ipvr_gem_object_alloc(dev); + if (obj == NULL) + return -ENOMEM; + memset(obj, 0, sizeof(*obj)); + + drm_gem_private_object_init(dev, &obj->gem.base, + num_pages * PAGE_SIZE); + + ipvr_gem_object_init(&obj->gem, &ipvr_gem_userptr_ops); + + init_waitqueue_head(&obj->gem.event_queue); + atomic_set(&obj->gem.cpu_writers, 0); + + obj->gem.drv_name = "ipvr"; + obj->gem.fence = NULL; + obj->gem.cache_level = cache_level; + obj->gem.tiling = tiling; + + obj->user_ptr = user_ptr; + obj->user_size = user_size; + + ret = drm_gem_handle_create(file_priv, &obj->gem.base, &handle); + /* drop reference from allocate - handle holds it now */ + drm_gem_object_unreference_unlocked(&obj->gem.base); + if (ret) { + IPVR_ERROR("failed drm_gem_handle_create for obj 0x%lx\n", + (unsigned long)obj); + goto err; + } + + *handle_p = handle; + + /* force to get pages at allocation stage */ + ret = ipvr_gem_object_bind_to_vm(&obj->gem, + &dev_priv->addr_space, 4096); + if (ret) { + IPVR_ERROR("failed ipvr_gem_object_bind_to_vm for obj 0x%lx\n", + (unsigned long)obj); + goto err; + } + + ret = ipvr_gem_mmu_bind_object(&obj->gem); + if (ret) { + IPVR_ERROR("failed ipvr_gem_mmu_bind_object for obj 0x%lx\n", + (unsigned long)obj); + goto err; + } + + *offset_p = ipvr_gem_obj_mmu_offset(&obj->gem) + offset_in_page(user_ptr); + if (IPVR_IS_ERR(*offset_p)) { + ret = IPVR_OFFSET_ERR(*offset_p); + goto err; + } + + IPVR_DEBUG_GENERAL("bo create done, gpu offset: 0x%llx.\n", *offset_p); + +err: + if (ret) + ipvr_gem_object_free(&obj->gem); + return ret; +} diff --git a/drivers/gpu/drm/ipvr/ipvr_buffer.h b/drivers/gpu/drm/ipvr/ipvr_buffer.h new file mode 100644 index 0000000..245ff8c --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_buffer.h @@ -0,0 +1,184 @@ +/************************************************************************** + * ipvr_buffer.h: IPVR buffer handling header file + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + + +#ifndef _IPVR_BUFFER_H_ +#define _IPVR_BUFFER_H_ + +#include <drmP.h> +#include <drm_gem.h> +#include <linux/shmem_fs.h> +#include "ipvr_drv.h" +#include "ipvr_drm.h" +#include "ipvr_fence.h" + +struct ipvr_fence; + +struct drm_ipvr_gem_object { + struct drm_gem_object base; + + /* used to disinguish between i915 and ipvr */ + char *drv_name; + + const struct drm_ipvr_gem_object_ops *ops; + + /** List of VMAs backed by this object */ + struct list_head vma_list; + + /* for ipvr case, buffer is bound into mmu, will not be unbound */ + struct list_head global_list; + + /** + * if this bo is a tiling buffer + * we can support multiple mode of tiling, which has different stride + * but only one tiling mode can be enabled. + */ + bool tiling; + + enum ipvr_cache_level cache_level; + + /** + * This is set if the object is on the active lists (has pending + * rendering and so a non-zero seqno), and is not set if it is on + * inactive (ready to be unbound) list. + */ + uint32_t active:1; + + /** + * This is set if the object has been written to since last bound + * to the GTT + */ + uint32_t dirty:1; + + uint32_t pin_count:5; +#define DRM_IPVR_GEM_OBJECT_MAX_PIN_COUNT 0x1f + + struct sg_table *sg_table; + int32_t pages_pin_count; + + /** User space pin count and filp owning the pin */ + uint32_t user_pin_count; + struct drm_file *pin_filp; + + struct ipvr_fence *fence; + atomic_t reserved; + wait_queue_head_t event_queue; + atomic_t cpu_writers; + + unsigned long mmap_base; + uint64_t mmap_size; + + struct page **fake_pages; +}; + +struct ipvr_gem_userptr_object { + struct drm_ipvr_gem_object gem; + uintptr_t user_ptr; + size_t user_size; + int32_t read_only; + struct mm_struct *mm; +#if defined(CONFIG_MMU_NOTIFIER) + struct mmu_notifier mn; +#endif +}; + +union drm_ipvr_gem_objects { + struct drm_ipvr_gem_object base; + struct ipvr_gem_userptr_object vmap; +}; + +struct drm_ipvr_gem_object_ops { + /* + * Interface between the GEM object and its backing storage. + * get_pages() is called once prior to the use of the associated set + * of pages before to binding them into the GTT, and put_pages() is + * called after we no longer need them. As we expect there to be + * associated cost with migrating pages between the backing storage + * and making them available for the GPU (e.g. clflush), we may hold + * onto the pages after they are no longer referenced by the GPU + * in case they may be used again shortly (for example migrating the + * pages to a different memory domain within the GTT). put_pages() + * will therefore most likely be called when the object itself is + * being released or under memory pressure (where we attempt to + * reap pages for the shrinker). + */ + int (*get_pages)(struct drm_ipvr_gem_object *); + void (*put_pages)(struct drm_ipvr_gem_object *); + void (*release)(struct drm_ipvr_gem_object *); + bool (*is_userptr_obj)(void); +}; + +/* + * A VMA represents a GEM BO that is bound into an address space. Therefore, a + * VMA's presence cannot be guaranteed before binding, or after unbinding the + * object into/from the address space. + * + * To make things as simple as possible (ie. no refcounting), a VMA's lifetime + * will always be <= an objects lifetime. So object refcounting should cover us. + */ +struct ipvr_vma { + struct drm_mm_node node; + struct drm_ipvr_gem_object *obj; + struct ipvr_address_space *vm; + + /* This object's place on the active/inactive lists */ + struct list_head mm_list; + + struct list_head vma_link; /* Link in the object's VMA list */ +}; + +int32_t ipvr_gem_create(struct drm_file *file_priv, struct drm_device *dev, + uint64_t size, uint32_t tiling, uint32_t cache_level, + uint64_t *rsize_p, uint32_t *handle_p, + uint64_t *offset_p); + +struct drm_ipvr_gem_object * +ipvr_gem_obj_create_and_bind(struct drm_device *dev, size_t size); + +struct drm_ipvr_gem_object * +ipvr_gem_alloc_object(struct drm_device *dev, size_t size); + +int32_t ipvr_gem_object_bind_to_vm(struct drm_ipvr_gem_object *obj, + struct ipvr_address_space *vm, + uint32_t alignment); + +unsigned long ipvr_gem_obj_mmu_offset(struct drm_ipvr_gem_object *obj); + +int32_t ipvr_gem_mmu_bind_object(struct drm_ipvr_gem_object *obj); + +void *ipvr_gem_object_vmap(struct drm_ipvr_gem_object *obj); + +int ipvr_gem_init_object(struct drm_gem_object *obj); + +void ipvr_gem_free_object(struct drm_gem_object *obj); + +bool ipvr_gem_clflush_object(struct drm_ipvr_gem_object *obj, bool force); + +int32_t ipvr_gem_userptr(struct drm_file *file_priv, struct drm_device *dev, + uint64_t user_ptr, uint64_t user_size, + uint32_t cache_level, uint32_t tiling, + uint32_t *handle_p, uint64_t *offset_p); + +#endif diff --git a/drivers/gpu/drm/ipvr/ipvr_debug.c b/drivers/gpu/drm/ipvr/ipvr_debug.c new file mode 100644 index 0000000..eecf8de --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_debug.c @@ -0,0 +1,263 @@ +/************************************************************************** + * ipvr_debug.c: IPVR debugfs support to assist bug triage + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * + **************************************************************************/ + +#if defined(CONFIG_DEBUG_FS) + +#include <linux/seq_file.h> +#include <linux/debugfs.h> +#include "ipvr_debug.h" +#include "ipvr_drv.h" +#include "ved_reg.h" + +union ipvr_debugfs_vars debugfs_vars; + +static int32_t ipvr_debug_info(struct seq_file *m, void *data) +{ + struct drm_info_node *node = (struct drm_info_node *) m->private; + struct drm_device *dev = node->minor->dev; + struct drm_ipvr_private *dev_priv = dev->dev_private; + + seq_printf(m, "ipvr platform revison id: 0x%x\n", + dev_priv->platform_rev_id); + + return 0; +} + +static int32_t ipvr_debug_gem_object_info(struct seq_file *m, void* data) +{ + struct drm_info_node *node = (struct drm_info_node *) m->private; + struct drm_device *dev = node->minor->dev; + struct drm_ipvr_private *dev_priv = dev->dev_private; + int32_t ret; + + ret = mutex_lock_interruptible(&dev->struct_mutex); + if (ret) + return ret; + + seq_printf(m, "ipvr total allocate %u objects, %zu bytes\n", + dev_priv->ipvr_mm.object_count, + dev_priv->ipvr_mm.object_memory); + + mutex_unlock(&dev->struct_mutex); + + return 0; +} + +static int32_t ipvr_debug_gem_seqno_info(struct seq_file *m, void *data) +{ + struct drm_info_node *node = (struct drm_info_node *) m->private; + struct drm_device *dev = node->minor->dev; + drm_ipvr_private_t *dev_priv = dev->dev_private; + int32_t ret; + + ret = mutex_lock_interruptible(&dev->struct_mutex); + if (ret) + return ret; + + seq_printf(m, "last signaled seq is %d, last emitted seq is %d\n", + atomic_read(&dev_priv->fence_drv.signaled_seq), + dev_priv->fence_drv.sync_seq); + + mutex_unlock(&dev->struct_mutex); + + return 0; +} + +static ssize_t ipvr_debug_ved_reg_read(struct file *filp, char __user *ubuf, + size_t max, loff_t *ppos) +{ + struct drm_device *dev = filp->private_data; + drm_ipvr_private_t *dev_priv = dev->dev_private; + char buf[200], offset[20], operation[10], format[20], val[20]; + int32_t len = 0, ret, no_of_tokens; + unsigned long reg_offset, reg_to_write; + + if (debugfs_vars.reg.reg_input == 0) + return len; + + snprintf(format, sizeof(format), "%%%zus %%%zus %%%zus", + sizeof(operation), sizeof(offset), sizeof(val)); + + no_of_tokens = sscanf(debugfs_vars.reg.reg_vars, + format, operation, offset, val); + + if (no_of_tokens < 3) + return len; + + len = sizeof(debugfs_vars.reg.reg_vars); + + if (strcmp(operation, IPVR_READ_TOKEN) == 0) { + ret = kstrtoul(offset, 16, ®_offset); + if (ret) + return -EINVAL; + + len = scnprintf(buf, sizeof(buf), "0x%x: 0x%x\n", + (uint32_t)reg_offset, + VED_REG_READ32((uint32_t)reg_offset)); + } else if (strcmp(operation, IPVR_WRITE_TOKEN) == 0) { + ret = kstrtoul(offset, 16, ®_offset); + if (ret) + return -EINVAL; + + ret = kstrtoul(val, 16, ®_to_write); + if (ret) + return -EINVAL; + + VED_REG_WRITE32(reg_offset, reg_to_write); + len = scnprintf(buf, sizeof(buf), + "0x%x: 0x%x\n", + (uint32_t)reg_offset, + (uint32_t)VED_REG_READ32(reg_offset)); + } else { + len = scnprintf(buf, sizeof(buf), "Operation Not Supported\n"); + } + + debugfs_vars.reg.reg_input = 0; + + simple_read_from_buffer(ubuf, max, ppos, buf, len); + + return len; +} + +static ssize_t +ipvr_debug_ved_reg_write(struct file *filp,const char __user *ubuf, + size_t cnt, loff_t *ppos) +{ + /* reset the string */ + memset(debugfs_vars.reg.reg_vars, 0, IPVR_MAX_BUFFER_STR_LEN); + + if (cnt > 0) { + if (cnt > sizeof(debugfs_vars.reg.reg_vars) - 1) + return -EINVAL; + + if (copy_from_user(debugfs_vars.reg.reg_vars, ubuf, cnt)) + return -EFAULT; + + debugfs_vars.reg.reg_vars[cnt] = 0; + + /* Enable Read */ + debugfs_vars.reg.reg_input = 1; + } + + return cnt; +} + +/* As the drm_debugfs_init() routines are called before dev->dev_private is + * allocated we need to hook into the minor for release. */ +static int32_t ipvr_add_fake_info_node(struct drm_minor *minor, + struct dentry *ent, const void *key) +{ + struct drm_info_node *node; + + node = kmalloc(sizeof(struct drm_info_node), GFP_KERNEL); + if (node == NULL) { + debugfs_remove(ent); + return -ENOMEM; + } + + node->minor = minor; + node->dent = ent; + node->info_ent = (void *) key; + + mutex_lock(&minor->debugfs_lock); + list_add(&node->list, &minor->debugfs_list); + mutex_unlock(&minor->debugfs_lock); + + return 0; +} + +static int32_t ipvr_debugfs_create(struct dentry *root, + struct drm_minor *minor, + const char *name, + const struct file_operations *fops) +{ + struct drm_device *dev = minor->dev; + struct dentry *ent; + + ent = debugfs_create_file(name, + S_IRUGO | S_IWUSR, + root, dev, + fops); + if (IS_ERR(ent)) + return PTR_ERR(ent); + + return ipvr_add_fake_info_node(minor, ent, fops); +} + +static const struct file_operations ipvr_ved_reg_fops = { + .owner = THIS_MODULE, + .open = simple_open, + .read = ipvr_debug_ved_reg_read, + .write = ipvr_debug_ved_reg_write, + .llseek = default_llseek, +}; + +static struct drm_info_list ipvr_debugfs_list[] = { + {"ipvr_capabilities", ipvr_debug_info, 0}, + {"ipvr_gem_objects", ipvr_debug_gem_object_info, 0}, + {"ipvr_gem_seqno", ipvr_debug_gem_seqno_info, 0}, + +}; +#define IPVR_DEBUGFS_ENTRIES ARRAY_SIZE(ipvr_debugfs_list) + +static struct ipvr_debugfs_files { + const char *name; + const struct file_operations *fops; +} ipvr_debugfs_files[] = { + {"ipvr_ved_reg_api", &ipvr_ved_reg_fops}, +}; + +int32_t ipvr_debugfs_init(struct drm_minor *minor) +{ + int32_t ret, i; + + for (i = 0; i < ARRAY_SIZE(ipvr_debugfs_files); i++) { + ret = ipvr_debugfs_create(minor->debugfs_root, minor, + ipvr_debugfs_files[i].name, + ipvr_debugfs_files[i].fops); + if (ret) + return ret; + } + + return drm_debugfs_create_files(ipvr_debugfs_list, + IPVR_DEBUGFS_ENTRIES, + minor->debugfs_root, minor); +} + +void ipvr_debugfs_cleanup(struct drm_minor *minor) +{ + int32_t i; + + drm_debugfs_remove_files(ipvr_debugfs_list, + IPVR_DEBUGFS_ENTRIES, minor); + + for (i = 0; i < ARRAY_SIZE(ipvr_debugfs_files); i++) { + struct drm_info_list *info_list = + (struct drm_info_list *)ipvr_debugfs_files[i].fops; + + drm_debugfs_remove_files(info_list, 1, minor); + } +} + +#endif /* CONFIG_DEBUG_FS */ diff --git a/drivers/gpu/drm/ipvr/ipvr_debug.h b/drivers/gpu/drm/ipvr/ipvr_debug.h new file mode 100644 index 0000000..3cbeb73 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_debug.h @@ -0,0 +1,50 @@ +/************************************************************************** + * ipvr_debug.h: IPVR debugfs support header file + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * + **************************************************************************/ + + +#ifndef _IPVR_DEBUG_H_ +#define _IPVR_DEBUG_H_ + +#include "drmP.h" + +/* Operations supported */ +#define IPVR_MAX_BUFFER_STR_LEN 200 + +#define IPVR_READ_TOKEN "READ" +#define IPVR_WRITE_TOKEN "WRITE" + +/* DebugFS Variable declaration */ +struct ipvr_debugfs_reg_vars { + char reg_vars[IPVR_MAX_BUFFER_STR_LEN]; + uint32_t reg_input; +}; + +union ipvr_debugfs_vars { + struct ipvr_debugfs_reg_vars reg; +}; + +int32_t ipvr_debugfs_init(struct drm_minor *minor); +void ipvr_debugfs_cleanup(struct drm_minor *minor); + +#endif diff --git a/drivers/gpu/drm/ipvr/ipvr_drm.h b/drivers/gpu/drm/ipvr/ipvr_drm.h new file mode 100644 index 0000000..1825d9f --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_drm.h @@ -0,0 +1,265 @@ +/************************************************************************** + * ipvr_drm.h: IPVR header file exported to user space + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + + +/* this file only define structs and macro which need export to user space */ +#ifndef _IPVR_DRM_H_ +#define _IPVR_DRM_H_ + +#include <drm/drm.h> +struct drm_ipvr_context_create { + /* passed ctx_info, including codec, profile info */ + uint32_t ctx_type; + /* returned back ctx_id */ + uint32_t ctx_id; + /* + * following tiling strides for VED are supported: + * stride 0: 512 for scheme 0, 1024 for scheme 1 + * stride 1: 1024 for scheme 0, 2048 for scheme 1 + * stride 2: 2048 for scheme 0, 4096 for scheme 1 + * stride 3: 4096 for scheme 0 + */ + uint32_t tiling_stride; + /* + * scheme 0: tile is 256x16, while minimal tile stride is 512 + * scheme 1: tile is 512x8, while minimal tile stride is 1024 + */ + uint32_t tiling_scheme; +}; + +struct drm_ipvr_context_destroy { + uint32_t ctx_id; + uint32_t pad64; +}; + +enum drm_ipvr_misc_key { + IPVR_DEVICE_INFO, + IPVR_UPDATE_TILING, + IPVR_SET_DISPLAYING_FRAME, + IPVR_GET_DISPLAYING_FRAME, + IPVR_QUERY_ENTRY +}; + +/* + * different context maybe has different tiling stride, + * then tiling info need be bound with ctx + */ +struct drm_ipvr_update_tiling { + uint32_t ctx_id; + uint32_t tiling_stride; + uint32_t tiling_scheme; + uint32_t pad64; +}; + +/* Ioctl to set/get misc params: + */ +struct drm_ipvr_misc { + uint64_t key; + uint64_t arg; /* argument pointer */ + uint64_t value; /* feed back pointer */ +}; + +struct ipvr_validate_arg { + /* point to next ipvr_validate_arg */ + uint64_t next; + uint64_t presumed_gpu_offset; + /* User's handle for a buffer */ + uint32_t handle; + uint32_t pad64; + /* fencing */ + uint32_t skip_fence; + int32_t fence_fd; +}; + +#define MAX_SLICES_PER_PICTURE 72 +struct drm_ipvr_mb_region { + uint32_t start; + uint32_t end; +}; + +struct drm_ipvr_decode_status { + uint32_t num_region; + struct drm_ipvr_mb_region mb_regions[MAX_SLICES_PER_PICTURE]; +}; + +struct drm_ipvr_gem_execbuffer { + /* + * List of ipvr_validate_arg structs + */ + uint64_t buffer_list; + uint32_t buffer_count; + + /* from user space point, this is msg buffer actually */ + uint32_t cmdbuf_handle; + uint32_t cmdbuf_size; + + uint32_t ctx_id; +}; + +enum ipvr_cache_level { + IPVR_CACHE_NONE = 0, /* uncacheable */ + IPVR_CACHE_WB, /* write back cacheable */ + IPVR_CACHE_WC, /* write combine, uncacheable */ + IPVR_CACHE_MAX, +}; + +struct drm_ipvr_gem_create { + /* + * Requested size for the object. + * The (page-aligned) allocated size for the object will be returned. + */ + uint64_t size; + uint64_t rounded_size; + uint64_t gpu_offset; + /* + * Returned handle for the object. + * Object handles are nonzero. + */ + uint32_t handle; + uint32_t tiling; + uint32_t cache_level; + uint32_t pad64; +}; + +struct drm_ipvr_gem_busy { + /* Handle of the buffer to check for busy */ + uint32_t handle; + + /* + * Return busy status (1 if busy, 0 if idle). + * The high word is used to indicate on which rings the object + * currently resides: + * 16:31 - busy (r or r/w) rings (16 render, 17 bsd, 18 blt, etc) + */ + uint32_t busy; +}; + +struct drm_ipvr_gem_mmap { + /* Handle for the object being mapped. */ + uint32_t handle; + uint32_t pad64; + /** Offset in the object to map. */ + uint64_t offset; + /* + * Length of data to map. + * The value will be page-aligned. + */ + uint64_t size; + /* + * Returned pointer the data was mapped at. + * This is a fixed-size type for 32/64 compatibility. + */ + uint64_t addr_ptr; +}; + +/* + * ACCESS mode flags for SYNCCPU. + * + * IPVR_SYNCCPU_MODE_READ will guarantee that the GPU is not + * writing to the buffer. + * + * IPVR_SYNCCPU_MODE_WRITE will guarantee that the GPU is not + * accessing the buffer. + * + * IPVR_SYNCCPU_MODE_NO_BLOCK makes sure the call does not wait + * for GPU accesses to finish but return -EBUSY. + * + * IPVR_SYNCCPU_MODE_TRYCACHED Try to place the buffer in cacheable + * memory while synchronized for CPU. + * + */ + +#define IPVR_SYNCCPU_MODE_READ (1 << 0) +#define IPVR_SYNCCPU_MODE_WRITE (1 << 1) +#define IPVR_SYNCCPU_MODE_NO_BLOCK (1 << 2) +#define IPVR_SYNCCPU_MODE_TRYCACHED (1 << 3) + +struct drm_ipvr_sync_cpu { + /* Handle for the object */ + uint32_t handle; + uint32_t access_mode; + enum { + IPVR_SYNCCPU_OP_GRAB, + IPVR_SYNCCPU_OP_RELEASE + } op; + uint32_t pad64; +}; + +struct drm_ipvr_gem_wait { + /* Handle of BO we shall wait on */ + uint32_t handle; + uint32_t flags; +}; + +struct drm_ipvr_gem_userptr { + uint64_t user_ptr; + uint64_t user_size; + uint64_t gpu_offset; + uint32_t cache_level; + uint32_t tiling; + /* + * Returned handle for the object. + * Object handles are nonzero. + */ + uint32_t handle; +}; + +/* + * IPVR GEM specific ioctls + * The device specific ioctl range is 0x50 to 0x5f. + */ +#define DRM_IPVR_CONTEXT_CREATE 0x00 +#define DRM_IPVR_CONTEXT_DESTROY 0x01 +#define DRM_IPVR_MISC 0x02 +#define DRM_IPVR_GEM_EXECBUFFER 0x03 +#define DRM_IPVR_GEM_BUSY 0x04 +#define DRM_IPVR_GEM_CREATE 0x05 +#define DRM_IPVR_GEM_MMAP 0x06 +#define DRM_IPVR_SYNC_CPU 0x07 +#define DRM_IPVR_GEM_WAIT 0x08 +#define DRM_IPVR_GEM_USERPTR 0x09 + +#define DRM_IOCTL_IPVR_CONTEXT_CREATE \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_CONTEXT_CREATE, struct drm_ipvr_context_create) +#define DRM_IOCTL_IPVR_CONTEXT_DESTROY \ + DRM_IOW(DRM_COMMAND_BASE + DRM_IPVR_CONTEXT_DESTROY, struct drm_ipvr_context_destroy) +#define DRM_IOCTL_IPVR_MISC \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_MISC, struct drm_ipvr_misc) +#define DRM_IOCTL_IPVR_GEM_EXECBUFFER \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_GEM_EXECBUFFER, struct drm_ipvr_gem_execbuffer) +#define DRM_IOCTL_IPVR_GEM_BUSY \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_GEM_BUSY, struct drm_ipvr_gem_busy) +#define DRM_IOCTL_IPVR_GEM_CREATE \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_GEM_CREATE, struct drm_ipvr_gem_create) +#define DRM_IOCTL_IPVR_GEM_MMAP \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_GEM_MMAP, struct drm_ipvr_gem_mmap) +#define DRM_IOCTL_IPVR_SYNC_CPU \ + DRM_IOW(DRM_COMMAND_BASE + DRM_IPVR_SYNC_CPU, struct drm_ipvr_sync_cpu) +#define DRM_IOCTL_IPVR_GEM_WAIT \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_GEM_WAIT, struct drm_ipvr_gem_wait) +#define DRM_IOCTL_IPVR_GEM_USERPTR \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_GEM_USERPTR, struct drm_ipvr_gem_userptr) + +#endif diff --git a/drivers/gpu/drm/ipvr/ipvr_drv.c b/drivers/gpu/drm/ipvr/ipvr_drv.c new file mode 100644 index 0000000..7a7ccbf --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_drv.c @@ -0,0 +1,776 @@ +/************************************************************************** + * ipvr_drv.c: IPVR driver common file for initialization/de-initialization + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + + +#include <linux/device.h> +#include <linux/version.h> +#include <uapi/drm/drm.h> +#include <linux/pm_runtime.h> +#include <linux/console.h> +#include <linux/module.h> +#include <asm/uaccess.h> + +#include "ipvr_drv.h" +#include "ipvr_gem.h" +#include "ipvr_mmu.h" +#include "ipvr_exec.h" +#include "ipvr_buffer.h" +#include "ipvr_debug.h" +#include "ved_init.h" +#include "ved_pm.h" +#include "ved_reg.h" +#include "ved_cmd.h" +#include "ipvr_trace.h" + +int32_t drm_ipvr_cpurelax = 0; +int32_t drm_ipvr_udelaydivider = 1; +int32_t drm_ipvr_trap_pagefaults = 0; +int32_t drm_ipvr_tiling = 1; +int32_t drm_ipvr_debug = 0x80; +bool fake_bo_debug = false; +int32_t drm_ipvr_freq = IP_FREQ_266_67; + +module_param_named(trap_pagefaults, drm_ipvr_trap_pagefaults, int, 0600); +module_param_named(debug, drm_ipvr_debug, int, 0600); +module_param_named(freq, drm_ipvr_freq, int, 0600); + +MODULE_PARM_DESC(debug, + "control debug info output" + "default: 0" + "0:IPVR_D_GENERAL, 1:IPVR_D_INIT, 2:IPVR_D_IRQ, 3:IPVR_D_ENTRY" + "4:IPVR_D_PM, 5:IPVR_D_REG, 6:IPVR_D_VED, 7:IPVR_D_WARN"); +MODULE_PARM_DESC(freq, + "control VED frequency" + "default: 0x0b (266.67 MHz)" + "0x1f: 100.00 MHz" + "0x1d: 106.67 MHz" + "0x17: 133.30 MHz" + "0x13: 160.00 MHz" + "0x11: 177.78 MHz" + "0x0f: 200.00 MHz" + "0x0e: 213.33 MHz" + "0x0b: 266.67 MHz" + "0x09: 320.00 MHz"); + +uint32_t REGISTER_READ(struct drm_device *dev, uint32_t reg) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + return ioread32(dev_priv->ved_reg_base + dev_priv->ved_reg_offset + reg); +} + +static int32_t +ipvr_context_create_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv) +{ + struct drm_ipvr_context_create *args = data; + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ipvr_context *ipvr_ctx = NULL; + unsigned long irq_flags; + int32_t ctx_id, ret = 0; + + IPVR_DEBUG_ENTRY("enter\n"); + /* add video decode context */ + ipvr_ctx = kzalloc(sizeof(struct ipvr_context), GFP_KERNEL); + if (ipvr_ctx == NULL) + return -ENOMEM; + + ctx_id = idr_alloc(&dev_priv->ipvr_ctx_idr, ipvr_ctx , + IPVR_MIN_CONTEXT_ID, IPVR_MAX_CONTEXT_ID, + GFP_KERNEL); + if (ctx_id < 0) + return -ENOMEM; + ipvr_ctx->ctx_id = ctx_id; + + INIT_LIST_HEAD(&ipvr_ctx->head); + ipvr_ctx->ctx_type = args->ctx_type; + ipvr_ctx->ipvr_fpriv = file_priv->driver_priv; + spin_lock_irqsave(&dev_priv->ipvr_ctx_lock, irq_flags); + list_add(&ipvr_ctx ->head, &dev_priv->ipvr_ctx_list); + spin_unlock_irqrestore(&dev_priv->ipvr_ctx_lock, irq_flags); + args->ctx_id = ctx_id; + IPVR_DEBUG_INIT("add ctx profile %d, entry %d, ctx_id is %d, " + "protected is %d.\n", + (ipvr_ctx->ctx_type >> 8) & 0xff, + ipvr_ctx->ctx_type & 0xff, ctx_id, + ipvr_ctx->ctx_type & VA_RT_FORMAT_PROTECTED); + /* + * todo: only one tiling region is supported now, + * maybe we need create additional tiling region for rotation case, + * which has different tiling stride + */ + if ((args->tiling_scheme == 0 && args->tiling_stride <= 3) || + (args->tiling_scheme == 1 && args->tiling_stride <= 2)) { + ipvr_ctx->tiling_scheme = args->tiling_scheme; + ipvr_ctx->tiling_stride = args->tiling_stride; + } else { + IPVR_DEBUG_WARN("unsupported tiling scheme %d and stide %d.\n", + args->tiling_scheme, args->tiling_stride); + ret = -EINVAL; + } + + return ret; +} + +static int32_t +ipvr_context_destroy_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv) +{ + struct drm_ipvr_context_destroy *args = data; + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct drm_ipvr_file_private *fpriv = file_priv->driver_priv; + struct ved_private *ved_priv = dev_priv->ved_private; + struct ipvr_context *ipvr_ctx = NULL; + unsigned long irq_flags; + + IPVR_DEBUG_ENTRY("enter\n"); + ipvr_ctx = (struct ipvr_context *) + idr_find(&dev_priv->ipvr_ctx_idr, args->ctx_id); + if (!ipvr_ctx || (ipvr_ctx->ipvr_fpriv != file_priv->driver_priv)) { + return -ENOENT; + } + IPVR_DEBUG_PM("Video:remove context profile %d, entrypoint %d\n", + (ipvr_ctx->ctx_type >> 8) & 0xff, + (ipvr_ctx->ctx_type & 0xff)); + mutex_lock(&ved_priv->ved_mutex); + if (ved_priv->ipvr_ctx == ipvr_ctx ) + ved_priv->ipvr_ctx = NULL; + mutex_unlock(&ved_priv->ved_mutex); + + spin_lock_irqsave(&dev_priv->ipvr_ctx_lock, irq_flags); + list_del(&ipvr_ctx->head); + fpriv->ctx_id = IPVR_CONTEXT_INVALID_ID; + spin_unlock_irqrestore(&dev_priv->ipvr_ctx_lock, irq_flags); + + idr_remove(&dev_priv->ipvr_ctx_idr, ipvr_ctx->ctx_id); + + kfree(ipvr_ctx); + return 0; +} + +static int32_t +ipvr_misc_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct drm_ipvr_misc *args = data; + int32_t ret = 0; + uint64_t value; + + IPVR_DEBUG_ENTRY("enter\n"); + if (!dev_priv) { + IPVR_DEBUG_WARN("called with no initialization.\n"); + return -EINVAL; + } + switch (args->key) { + case IPVR_DEVICE_INFO: { + /* todo: remove hard-coding */ + uint32_t device_info = 0xf31 << 16; + ret = copy_to_user((void __user *)((unsigned long)args->value), + &device_info, sizeof(device_info)); + break; + } + case IPVR_UPDATE_TILING: { + struct drm_ipvr_update_tiling tiling; + struct ipvr_context *ipvr_ctx = NULL; + ret = copy_from_user(&tiling, + (void __user *)((unsigned long)args->arg), + sizeof(tiling)); + if (ret) + break; + + ipvr_ctx = (struct ipvr_context *) + idr_find(&dev_priv->ipvr_ctx_idr, tiling.ctx_id); + if (!ipvr_ctx || + (ipvr_ctx->ipvr_fpriv != file_priv->driver_priv)) { + IPVR_DEBUG_WARN("fail to find ctx %d", tiling.ctx_id); + return -ENOENT; + } + IPVR_DEBUG_GENERAL("Video: update video tiling for ctx %d, " + "old tiling scheme is %d, old tiling stride is %d, " + "new tiling scheme is %d, new tiling stride is %d.\n", + tiling.ctx_id, + ipvr_ctx ->tiling_scheme, ipvr_ctx ->tiling_stride, + tiling.tiling_scheme, tiling.tiling_stride); + if ((tiling.tiling_scheme == 0 && tiling.tiling_stride <= 3) || + (tiling.tiling_scheme == 1 && tiling.tiling_stride <= 2)) { + ipvr_ctx ->tiling_scheme = tiling.tiling_scheme; + ipvr_ctx ->tiling_stride = tiling.tiling_stride; + } else { + IPVR_ERROR("unsupported tile scheme: %d, stide: %d.\n", + tiling.tiling_scheme, tiling.tiling_stride); + ret = -EINVAL; + } + break; + } + default: + if (copy_from_user(&value, + (void __user *)((unsigned long)args->value), + sizeof(value))) { + IPVR_DEBUG_WARN("copy_from_user failed\n"); + return -EFAULT; + } + if (copy_to_user((void __user *)((unsigned long)args->value), + &value, sizeof(value))) { + IPVR_DEBUG_WARN("copy_to_user failed\n"); + return -EFAULT; + } + ret = -EFAULT; + break; + } + return ret; +} + +static struct drm_ioctl_desc ipvr_gem_ioctls[] = { + DRM_IOCTL_DEF_DRV(IPVR_CONTEXT_CREATE, + ipvr_context_create_ioctl, DRM_UNLOCKED), + DRM_IOCTL_DEF_DRV(IPVR_CONTEXT_DESTROY, + ipvr_context_destroy_ioctl, DRM_UNLOCKED), + DRM_IOCTL_DEF_DRV(IPVR_MISC, + ipvr_misc_ioctl, DRM_AUTH), + DRM_IOCTL_DEF_DRV(IPVR_GEM_EXECBUFFER, + ipvr_gem_execbuffer, DRM_AUTH|DRM_UNLOCKED), + DRM_IOCTL_DEF_DRV(IPVR_GEM_BUSY, + ipvr_gem_busy_ioctl, DRM_AUTH|DRM_UNLOCKED), + DRM_IOCTL_DEF_DRV(IPVR_GEM_CREATE, + ipvr_gem_create_ioctl, DRM_UNLOCKED), + DRM_IOCTL_DEF_DRV(IPVR_GEM_MMAP, + ipvr_gem_mmap_ioctl, DRM_UNLOCKED), + DRM_IOCTL_DEF_DRV(IPVR_SYNC_CPU, + ipvr_sync_cpu_ioctl, DRM_UNLOCKED), + DRM_IOCTL_DEF_DRV(IPVR_GEM_WAIT, + ipvr_gem_wait_ioctl, DRM_AUTH|DRM_UNLOCKED), + DRM_IOCTL_DEF_DRV(IPVR_GEM_USERPTR, + ipvr_gem_userptr_ioctl, DRM_UNLOCKED), +}; + +static void ipvr_gem_init(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + + dev_priv->ipvr_bo_slab = kmem_cache_create("ipvr_gem_object", + sizeof(union drm_ipvr_gem_objects), 0, + SLAB_HWCACHE_ALIGN, NULL); + + INIT_LIST_HEAD(&dev_priv->ipvr_mm.unbound_list); + INIT_LIST_HEAD(&dev_priv->ipvr_mm.bound_list); + spin_lock_init(&dev_priv->ipvr_mm.object_stat_lock); + + dev_priv->ipvr_mm.interruptible = true; +} + +static void ipvr_gem_setup_mmu(struct drm_device *dev, + unsigned long linear_start, + unsigned long linear_end, + unsigned long tiling_start, + unsigned long tiling_end) +{ + /* Let GEM Manage all of the aperture. + * + * However, leave one page at the end still bound to the scratch page. + * There are a number of places where hardware apparently prefetches + * past the end of the object, and we've seen multiple hangs with the + * GPU head pointer stuck in a batchbuffer bound at last page of the + * aperture. One page should be enough to keep any prefetching inside + * of the aperture. + */ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ipvr_address_space *addr_space = &dev_priv->addr_space; + + /* todo: add sanity check */ + addr_space->dev = dev_priv->dev; + INIT_LIST_HEAD(&addr_space->active_list); + INIT_LIST_HEAD(&addr_space->inactive_list); + + /* Subtract the guard page ... */ + drm_mm_init(&addr_space->linear_mm, linear_start, + linear_end - linear_start - PAGE_SIZE); + dev_priv->addr_space.linear_start = linear_start; + dev_priv->addr_space.linear_total = linear_end - linear_start; + + drm_mm_init(&addr_space->tiling_mm, tiling_start, + tiling_end - tiling_start - PAGE_SIZE); + dev_priv->addr_space.tiling_start = tiling_start; + dev_priv->addr_space.tiling_total = tiling_end - tiling_start; +} + +static void ipvr_do_takedown(struct drm_device *dev) +{ + /* todo: need check if need clean up mm here */ + ipvr_ved_uninit(dev); +} + +static int32_t ipvr_drm_unload(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + + BUG_ON(!dev->platformdev); + BUG_ON(atomic_read(&dev_priv->ved_power_usage)); + + IPVR_DEBUG_ENTRY("entered."); + if (dev_priv) { + WARN_ON(pm_runtime_get_sync(&dev->platformdev->dev) < 0); + + if (dev_priv->ipvr_bo_slab) + kmem_cache_destroy(dev_priv->ipvr_bo_slab); + ipvr_fence_driver_fini(dev_priv); + + ipvr_do_takedown(dev); + + WARN_ON(pm_runtime_put_sync_suspend(&dev->platformdev->dev) < 0); + + if (dev_priv->validate_ctx.buffers) + vfree(dev_priv->validate_ctx.buffers); + + if (dev_priv->pf_pd) { + ipvr_mmu_free_pagedir(dev_priv->pf_pd); + dev_priv->pf_pd = NULL; + } + if (dev_priv->mmu) { + ipvr_mmu_driver_takedown(dev_priv->mmu); + dev_priv->mmu = NULL; + } + + if (dev_priv->ved_reg_base) { + iounmap(dev_priv->ved_reg_base - dev_priv->ved_reg_offset); + dev_priv->ved_reg_base = NULL; + dev_priv->ved_reg_offset = 0; + } + + list_del(&dev_priv->default_ctx.head); + idr_remove(&dev_priv->ipvr_ctx_idr, dev_priv->default_ctx.ctx_id); + kfree(dev_priv); + + } + + pm_runtime_disable(&dev->platformdev->dev); + + return 0; +} + +static int32_t ipvr_drm_load(struct drm_device *dev, unsigned long flags) +{ + struct drm_ipvr_private *dev_priv; + int32_t ctx_id, ret = 0; + struct platform_device *platdev; + struct resource *res_mmio, *res_reg; + void __iomem* mmio_addr; + + dev_priv = kzalloc(sizeof(*dev_priv), GFP_KERNEL); + if (dev_priv == NULL) + return -ENOMEM; + + dev->dev_private = dev_priv; + dev_priv->dev = dev; + + BUG_ON(!dev->platformdev); + platdev = dev->platformdev; + + mutex_init(&dev_priv->cmdbuf_mutex); + INIT_LIST_HEAD(&dev_priv->validate_ctx.validate_list); + + dev_priv->pci_root = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0)); + if (!dev_priv->pci_root) { + kfree(dev_priv); + return -ENODEV; + } + + dev->irq = platform_get_irq(platdev, 0); + if (dev->irq < 0) { + ret = -ENODEV; + goto out_err; + } + + res_mmio = platform_get_resource(platdev, IORESOURCE_MEM, 0); + res_reg = platform_get_resource(platdev, IORESOURCE_REG, 0); + if (!res_mmio || !res_reg) { + ret = -ENXIO; + goto out_err; + } + + mmio_addr = ioremap_wc(res_mmio->start, + res_mmio->end - res_mmio->start); + if (IS_ERR(mmio_addr)) { + ret = -EACCES; + goto out_err; + } + + dev_priv->ved_reg_base = mmio_addr + res_reg->start; + dev_priv->ved_reg_offset = res_reg->start; + IPVR_DEBUG_VED("ved_reg_base is %p, range is 0x%llx - 0x%llx.\n", + dev_priv->ved_reg_base, + res_reg->start, res_reg->end); + + platform_set_drvdata(dev->platformdev, dev); + pm_runtime_enable(&dev->platformdev->dev); + + if (pm_runtime_get_sync(&dev->platformdev->dev) < 0) { + ret = -EBUSY; + goto out_err; + } + + IPVR_DEBUG_INIT("MSVDX_CORE_REV_OFFSET by readl is 0x%x.\n", + readl(dev_priv->ved_reg_base + 0x640)); + IPVR_DEBUG_INIT("MSVDX_CORE_REV_OFFSET by VED_REG_READ32 is 0x%x.\n", + VED_REG_READ32(MSVDX_CORE_REV_OFFSET)); + + /* mmu init */ + dev_priv->mmu = ipvr_mmu_driver_init((void *)0, + drm_ipvr_trap_pagefaults, + 0, dev_priv); + if (!dev_priv->mmu) { + ret = -EBUSY; + goto out_err; + } + + dev_priv->pf_pd = ipvr_mmu_alloc_pd(dev_priv->mmu, 1, 0); + if (!dev_priv->pf_pd) { + ret = -ENOMEM; + goto out_err; + } + + ipvr_mmu_set_pd_context(ipvr_mmu_get_default_pd(dev_priv->mmu), 0); + ipvr_mmu_set_pd_context(dev_priv->pf_pd, 1); + + /* + * Initialize sequence numbers for the different command + * submission mechanisms. + */ + dev_priv->last_seq = 1; + + ipvr_gem_init(dev); + + ipvr_gem_setup_mmu(dev, + IPVR_MEM_MMU_LINEAR_START, + IPVR_MEM_MMU_LINEAR_END, + IPVR_MEM_MMU_TILING_START, + IPVR_MEM_MMU_TILING_END); + + ipvr_ved_init(dev); + + WARN_ON(pm_runtime_put_sync_suspend(&dev->platformdev->dev) < 0); + + dev_priv->ved_private->ved_needs_reset = 1; + mutex_init(&dev_priv->ved_pm_mutex); + atomic_set(&dev_priv->ved_power_usage, 0); + + ipvr_fence_driver_init(dev_priv); + + dev_priv->validate_ctx.buffers = + vmalloc(IPVR_NUM_VALIDATE_BUFFERS * + sizeof(struct ipvr_validate_buffer)); + if (!dev_priv->validate_ctx.buffers) { + ret = -ENOMEM; + goto out_err; + } + + /* ipvr context initialization */ + INIT_LIST_HEAD(&dev_priv->ipvr_ctx_list); + spin_lock_init(&dev_priv->ipvr_ctx_lock); + idr_init(&dev_priv->ipvr_ctx_idr); + /* default ipvr context is used for scaling, rotation case */ + ctx_id = idr_alloc(&dev_priv->ipvr_ctx_idr, &dev_priv->default_ctx, + IPVR_MIN_CONTEXT_ID, IPVR_MAX_CONTEXT_ID, + GFP_KERNEL); + if (ctx_id < 0) { + return -ENOMEM; + goto out_err; + } + dev_priv->default_ctx.ctx_id = ctx_id; + INIT_LIST_HEAD(&dev_priv->default_ctx.head); + dev_priv->default_ctx.ctx_type = 0; + dev_priv->default_ctx.ipvr_fpriv = NULL; + + /* don't need protect with spinlock during module load stage */ + list_add(&dev_priv->default_ctx.head, &dev_priv->ipvr_ctx_list); + dev_priv->default_ctx.tiling_scheme = 0; + dev_priv->default_ctx.tiling_stride = 0; + + return 0; +out_err: + ipvr_drm_unload(dev); + return ret; + +} + +/* + * The .open() method is called every time the device is opened by an + * application. Drivers can allocate per-file private data in this method and + * store them in the struct drm_file::driver_priv field. Note that the .open() + * method is called before .firstopen(). + */ +static int32_t +ipvr_drm_open(struct drm_device *dev, struct drm_file *file_priv) +{ + struct drm_ipvr_file_private *ipvr_fp; + IPVR_DEBUG_ENTRY("enter\n"); + + ipvr_fp = kzalloc(sizeof(*ipvr_fp), GFP_KERNEL); + if (unlikely(ipvr_fp == NULL)) + return -ENOMEM; + + file_priv->driver_priv = ipvr_fp; + + return 0; +} + +/* + * The close operation is split into .preclose() and .postclose() methods. + * Drivers must stop and cleanup all per-file operations in the .preclose() + * method. For instance pending vertical blanking and page flip events must be + * cancelled. No per-file operation is allowed on the file handle after + * returning from the .preclose() method. + */ +static void +ipvr_drm_preclose(struct drm_device *dev, struct drm_file *file_priv) +{ + /* if user didn't destory ctx explicitly, remove ctx here */ + struct drm_ipvr_private *dev_priv; + struct drm_ipvr_file_private *ipvr_fpriv; + struct ved_private *ved_priv; + struct ipvr_context *ipvr_ctx = NULL; + unsigned long irq_flags; + + IPVR_DEBUG_ENTRY("enter\n"); + dev_priv = dev->dev_private; + ipvr_fpriv = file_priv->driver_priv; + ved_priv = dev_priv->ved_private; + + if (ipvr_fpriv->ctx_id == IPVR_CONTEXT_INVALID_ID) + return; + ipvr_ctx = (struct ipvr_context *) + idr_find(&dev_priv->ipvr_ctx_idr, ipvr_fpriv->ctx_id); + if (!ipvr_ctx || (ipvr_ctx->ipvr_fpriv != ipvr_fpriv)) { + IPVR_DEBUG_GENERAL("ctx for id %d has already destroyed\n", + ipvr_fpriv->ctx_id); + return; + } + IPVR_DEBUG_PM("Video:remove context profile %d, entrypoint %d\n", + (ipvr_ctx->ctx_type >> 8) & 0xff, + (ipvr_ctx->ctx_type & 0xff)); + mutex_lock(&ved_priv->ved_mutex); + if (ved_priv->ipvr_ctx == ipvr_ctx ) + ved_priv->ipvr_ctx = NULL; + mutex_unlock(&ved_priv->ved_mutex); + + spin_lock_irqsave(&dev_priv->ipvr_ctx_lock, irq_flags); + list_del(&ipvr_ctx->head); + ipvr_fpriv->ctx_id = IPVR_CONTEXT_INVALID_ID; + spin_unlock_irqrestore(&dev_priv->ipvr_ctx_lock, irq_flags); + + idr_remove(&dev_priv->ipvr_ctx_idr, ipvr_ctx->ctx_id); + + kfree(ipvr_ctx ); +} + +/* + * Finally the .postclose() method is called as the last step of the close + * operation, right before calling the .lastclose() method if no other open + * file handle exists for the device. Drivers that have allocated per-file + * private data in the .open() method should free it here. + */ +static void +ipvr_drm_postclose(struct drm_device *dev, struct drm_file *file_priv) +{ + struct drm_ipvr_file_private *ipvr_fpriv = file_priv->driver_priv; + IPVR_DEBUG_ENTRY("enter\n"); + kfree(ipvr_fpriv); +} + +static irqreturn_t ipvr_irq_handler(int32_t irq, void *arg) +{ + struct drm_device *dev = (struct drm_device *) arg; + WARN_ON(ved_irq_handler(dev)); + return IRQ_HANDLED; +} + +static const struct file_operations ipvr_fops = { + .owner = THIS_MODULE, + .open = drm_open, + .release = drm_release, + .unlocked_ioctl = drm_ioctl, +#ifdef CONFIG_COMPAT + .compat_ioctl = drm_ioctl, +#endif + /* no need to define mmap. User space maps bo with DRM_IPVR_GEM_MMAP */ +}; + +static int32_t ipvr_drm_freeze(struct drm_device *dev) +{ + int32_t ret; + int32_t power_usage; + struct drm_ipvr_private *dev_priv = dev->dev_private; + + IPVR_DEBUG_ENTRY("enter\n"); + power_usage = atomic_read(&dev_priv->ved_power_usage); + BUG_ON(power_usage < 0); + if (power_usage > 0) { + IPVR_DEBUG_PM("VED power usage is %d, skip freezing\n", power_usage); + return 0; + } + + ret = ved_check_idle(dev); + if (ret) { + IPVR_DEBUG_PM("VED check idle fail: %d, skip freezing\n", ret); + return 0; + } + + if (dev->irq_enabled) { + ret = drm_irq_uninstall(dev); + if (unlikely(ret)) { + IPVR_ERROR("Error uninstalling IRQ handler: %d\n", ret); + return -EFAULT; + } + IPVR_DEBUG_PM("Successfully uninstalled IRQ\n"); + } + else + IPVR_DEBUG_PM("irq_enabled is %d\n", dev->irq_enabled); + + if (is_ved_on(dev)) { + if (!ved_power_off(dev)) { + IPVR_ERROR("Failed to power off VED\n"); + return -EFAULT; + } + IPVR_DEBUG_PM("Successfully powered off\n"); + } else { + IPVR_DEBUG_PM("Skiped power-off since already powered off\n"); + } + + return 0; +} + +static int32_t ipvr_drm_thaw(struct drm_device *dev) +{ + int ret; + IPVR_DEBUG_ENTRY("enter\n"); + if (!is_ved_on(dev)) { + if (!ved_power_on(dev)) { + IPVR_ERROR("Failed to power on VED\n"); + return -EFAULT; + } + IPVR_DEBUG_PM("Successfully powered on\n"); + } else { + IPVR_DEBUG_PM("Skiped power-on since already powered on\n"); + } + + if (!dev->irq_enabled) { + ret = drm_irq_install(dev, dev->irq); + if (ret) { + IPVR_ERROR("Error installing IRQ handler: %d\n", ret); + return -EFAULT; + } + IPVR_DEBUG_PM("Successfully installed IRQ\n"); + } + else + IPVR_DEBUG_PM("irq_enabled is %d\n", dev->irq_enabled); + + return 0; +} + +static int32_t ipvr_pm_suspend(struct device *dev) +{ + struct platform_device *platformdev = to_platform_device(dev); + struct drm_device *drm_dev = platform_get_drvdata(platformdev); + IPVR_DEBUG_PM("PM suspend called\n"); + BUG_ON(!drm_dev); + return ipvr_drm_freeze(drm_dev); +} +static int32_t ipvr_pm_resume(struct device *dev) +{ + struct platform_device *platformdev = to_platform_device(dev); + struct drm_device *drm_dev = platform_get_drvdata(platformdev); + IPVR_DEBUG_PM("PM resume called\n"); + BUG_ON(!drm_dev); + return ipvr_drm_thaw(drm_dev); +} + +/* + * dump GEM API is mainly for dumb buffers suitable for scanout, + * it is not needed for ipvr driver. + * gem_vm_ops is used for mmap case, also not needed for ipvr + * todo: prime support can be enabled later + */ +static struct drm_driver ipvr_drm_driver = { + .driver_features = DRIVER_HAVE_IRQ | DRIVER_GEM, + .load = ipvr_drm_load, + .unload = ipvr_drm_unload, + .open = ipvr_drm_open, + .preclose = ipvr_drm_preclose, + .postclose = ipvr_drm_postclose, + .irq_handler = ipvr_irq_handler, + .set_busid = drm_platform_set_busid, + .gem_free_object = ipvr_gem_free_object, +#ifdef CONFIG_DEBUG_FS + .debugfs_init = ipvr_debugfs_init, + .debugfs_cleanup = ipvr_debugfs_cleanup, +#endif + .ioctls = ipvr_gem_ioctls, + .num_ioctls = ARRAY_SIZE(ipvr_gem_ioctls), + .fops = &ipvr_fops, + .name = IPVR_DRIVER_NAME, + .desc = IPVR_DRIVER_DESC, + .date = IPVR_DRIVER_DATE, + .major = IPVR_DRIVER_MAJOR, + .minor = IPVR_DRIVER_MINOR, + .patchlevel = IPVR_DRIVER_PATCHLEVEL, +}; + +static int32_t ipvr_plat_probe(struct platform_device *device) +{ + return drm_platform_init(&ipvr_drm_driver, device); +} + +static int32_t ipvr_plat_remove(struct platform_device *device) +{ + drm_put_dev(platform_get_drvdata(device)); + return 0; +} + +static struct dev_pm_ops ipvr_pm_ops = { + .suspend = ipvr_pm_suspend, + .resume = ipvr_pm_resume, + .freeze = ipvr_pm_suspend, + .thaw = ipvr_pm_resume, + .poweroff = ipvr_pm_suspend, + .restore = ipvr_pm_resume, +#ifdef CONFIG_PM_RUNTIME + .runtime_suspend = ipvr_pm_suspend, + .runtime_resume = ipvr_pm_resume, +#endif +}; + +static struct platform_driver ipvr_plat_driver = { + .driver = { + .name = "ipvr-ved", + .owner = THIS_MODULE, +#ifdef CONFIG_PM + .pm = &ipvr_pm_ops, +#endif + }, + .probe = ipvr_plat_probe, + .remove = ipvr_plat_remove, +}; + +module_platform_driver(ipvr_plat_driver); +MODULE_LICENSE("GPL"); diff --git a/drivers/gpu/drm/ipvr/ipvr_drv.h b/drivers/gpu/drm/ipvr/ipvr_drv.h new file mode 100644 index 0000000..280307b --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_drv.h @@ -0,0 +1,464 @@ +/************************************************************************** + * ipvr_drv.h: IPVR driver common header file + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + +#ifndef _IPVR_DRV_H_ +#define _IPVR_DRV_H_ +#include "drmP.h" +#include "ipvr_drm.h" +#include "ipvr_mmu.h" +#include <linux/version.h> +#include <linux/io-mapping.h> +#include <linux/i2c.h> +#include <linux/i2c-algo-bit.h> +#include <linux/backlight.h> +#include <linux/intel-iommu.h> +#include <linux/kref.h> +#include <linux/pm_qos.h> +#include <linux/mmu_notifier.h> + +#define IPVR_DRIVER_AUTHOR "Intel, Inc." +#define IPVR_DRIVER_NAME "ipvr" +#define IPVR_DRIVER_DESC "Vxd392 VP8 driver for Baytrail" +#define IPVR_DRIVER_DATE "20140609" +#define IPVR_DRIVER_MAJOR 0 +#define IPVR_DRIVER_MINOR 1 +#define IPVR_DRIVER_PATCHLEVEL 0 + +/* General customization: + */ + +#define IPVR_MMU_CACHED_MEMORY 0x0001 /* Bind to MMU only */ +#define IPVR_MMU_RO_MEMORY 0x0002 /* MMU RO memory */ +#define IPVR_MMU_WO_MEMORY 0x0004 /* MMU WO memory */ + +/* + *PTE's and PDE's + */ + +#define IPVR_PDE_MASK 0x003FFFFF +#define IPVR_PDE_SHIFT 22 +#define IPVR_PTE_SHIFT 12 + +#define IPVR_PTE_VALID 0x0001 /* PTE / PDE valid */ +#define IPVR_PTE_WO 0x0002 /* Write only */ +#define IPVR_PTE_RO 0x0004 /* Read only */ +#define IPVR_PTE_CACHED 0x0008 /* CPU cache coherent */ + +/* + * linear MMU size is 512M : 0 - 512M + * tiling MMU size is 512M : 512M - 1024M + */ +#define IPVR_MEM_MMU_LINEAR_START 0x00000000 +#define IPVR_MEM_MMU_LINEAR_END 0x20000000 +#define IPVR_MEM_MMU_TILING_START 0x20000000 +#define IPVR_MEM_MMU_TILING_END 0x40000000 + +#define IPVR_GEM_DOMAIN_CPU 0x00000001 +#define IPVR_GEM_DOMAIN_GPU 0x00000002 + +/* video context */ +#define VA_RT_FORMAT_PROTECTED 0x80000000 + +#define IPVR_CONTEXT_INVALID_ID 0 +#define IPVR_MIN_CONTEXT_ID 1 +#define IPVR_MAX_CONTEXT_ID 0xff + +/* video error status report */ +#define MAX_SLICES_PER_PICTURE 72 +#define MAX_DECODE_BUFFERS (24) +#define VED_MAX_EC_INSTANCE (4) + +/* + *Debug print bits setting + */ +#define IPVR_D_GENERAL (1 << 0) +#define IPVR_D_INIT (1 << 1) +#define IPVR_D_IRQ (1 << 2) +#define IPVR_D_ENTRY (1 << 3) +#define IPVR_D_PM (1 << 4) +#define IPVR_D_REG (1 << 5) +#define IPVR_D_VED (1 << 6) +#define IPVR_D_WARN (1 << 7) + +#define IPVR_DEBUG_GENERAL(_fmt, _arg...) \ + IPVR_DEBUG(IPVR_D_GENERAL, _fmt, ##_arg) +#define IPVR_DEBUG_INIT(_fmt, _arg...) \ + IPVR_DEBUG(IPVR_D_INIT, _fmt, ##_arg) +#define IPVR_DEBUG_IRQ(_fmt, _arg...) \ + IPVR_DEBUG(IPVR_D_IRQ, _fmt, ##_arg) +#define IPVR_DEBUG_ENTRY(_fmt, _arg...) \ + IPVR_DEBUG(IPVR_D_ENTRY, _fmt, ##_arg) +#define IPVR_DEBUG_PM(_fmt, _arg...) \ + IPVR_DEBUG(IPVR_D_PM, _fmt, ##_arg) +#define IPVR_DEBUG_REG(_fmt, _arg...) \ + IPVR_DEBUG(IPVR_D_REG, _fmt, ##_arg) +#define IPVR_DEBUG_VED(_fmt, _arg...) \ + IPVR_DEBUG(IPVR_D_VED, _fmt, ##_arg) +#define IPVR_DEBUG_WARN(_fmt, _arg...) \ + IPVR_DEBUG(IPVR_D_WARN, _fmt, ##_arg) + +#define IPVR_DEBUG(_flag, _fmt, _arg...) \ + do { \ + if (unlikely((_flag) & drm_ipvr_debug)) \ + printk(KERN_INFO \ + "[ipvr:0x%02x:%s] " _fmt , _flag, \ + __func__ , ##_arg); \ + } while (0) + +#define IPVR_ERROR(_fmt, _arg...) \ + do { \ + printk(KERN_ERR \ + "[ipvr:ERROR:%s] " _fmt , \ + __func__ , ##_arg); \ + } while (0) + +/* + * set cpu_relax = 1 in sysfs to use cpu_relax instead of udelay bysy loop + * set udelay_divider to reduce the udelay values,e.g.= 10, reduce 10 times + */ +#define IPVR_UDELAY(usec) \ +do { \ + if (drm_ipvr_cpurelax == 0) \ + DRM_UDELAY(usec / drm_ipvr_udelaydivider); \ + else \ + cpu_relax(); \ +} while (0) + +typedef struct ipvr_validate_buffer ipvr_validate_buffer_t; + +uint32_t REGISTER_READ(struct drm_device *dev, uint32_t reg); +#define REG_READ(reg) REGISTER_READ(dev, (reg)) + +#define to_ipvr_bo(x) container_of(x, struct drm_ipvr_gem_object, base) + +extern int drm_ipvr_debug; +extern int drm_ipvr_udelaydivider; +extern int drm_ipvr_cpurelax; +/** + *struct ipvr_validate_context + * + *@buffers: array of pre-allocated validate buffers. + *@used_buffers: number of buffers in @buffers array currently in use. + *@validate_buffer: buffers validated from user-space. + *@kern_validate_buffers : buffers validated from kernel-space. + *@fence_flags : Fence flags to be used for fence creation. + * + *This structure is used during execbuf validation. + */ +struct ipvr_validate_context { + ipvr_validate_buffer_t *buffers; + uint32_t used_buffers; + struct list_head validate_list; +}; + +struct ipvr_mmu_driver; +struct ipvr_mmu_pd; + +/* + * may be not needed, BOs are always bound into mmu, + * so there is no need for bound_list and unbound_list + */ +struct ipvr_gem_mm { + /** List of all objects in mmu space. Used to restore mmu + * mappings on resume */ + struct list_head bound_list; + /** + * List of objects which are not bound to the mmu (thus + * are idle and not used by the GPU) but still have + * (presumably uncached) pages still attached. + */ + struct list_head unbound_list; + + /** + * Are we in a non-interruptible section of code like + * modesetting? + */ + bool interruptible; + + /* accounting, useful for userland debugging */ + spinlock_t object_stat_lock; + size_t object_memory; + uint32_t object_count; +}; + +struct ipvr_address_space { + struct drm_mm linear_mm; + struct drm_mm tiling_mm; + struct drm_device *dev; + unsigned long linear_start; + size_t linear_total; + unsigned long tiling_start; + size_t tiling_total; + + /* need it during clear_range */ + struct { + dma_addr_t addr; + struct page *page; + } scratch; + + /** + * List of objects currently involved in rendering. + * + * Includes buffers having the contents of their GPU caches + * flushed, not necessarily primitives. last_rendering_seqno + * represents when the rendering involved will be completed. + * + * A reference is held on the buffer while on this list. + */ + struct list_head active_list; + + /** + * LRU list of objects which are not in the ringbuffer and + * are ready to unbind, but are still in the GTT. + * + * last_rendering_seqno is 0 while an object is in this list. + * + * A reference is not held on the buffer while on this list, + * as merely being GTT-bound shouldn't prevent its being + * freed, and we'll pull it off the list in the free path. + */ + struct list_head inactive_list; + + /* todo: it should be needed to avoid security problem, + * when destroy bo, need set scratch page the bo mmu entry + */ + void (*clear_range)(struct ipvr_address_space *vm, + unsigned int first_entry, + unsigned int num_entries); + /* todo: directly call mmu function, the func ptr is not needed */ + void (*insert_entries)(struct ipvr_address_space *vm, + struct sg_table *st, + unsigned int first_entry, + int cache_level); + void (*cleanup)(struct ipvr_address_space *vm); +}; + +struct ipvr_fence_driver { + uint16_t sync_seq; + atomic_t signaled_seq; + unsigned long last_activity; + bool initialized; + spinlock_t fence_lock; +}; + +struct ipvr_context { + /* used to link into ipvr_ctx_list */ + struct list_head head; + uint32_t ctx_id; + /* used to double check ctx when find with idr, may be removed */ + struct drm_ipvr_file_private *ipvr_fpriv; /* DRM device file pointer */ + uint32_t ctx_type; /* profile << 8 | entrypoint */ + + uint16_t cur_seq; + + /* for IMG DDK, only use tiling for 2k and 4k buffer stride */ + /* + * following tiling strides for VED are supported: + * stride 0: 512 for scheme 0, 1024 for scheme 1 + * stride 1: 1024 for scheme 0, 2048 for scheme 1 + * stride 2: 2048 for scheme 0, 4096 for scheme 1 + * stride 3: 4096 for scheme 0 + */ + uint8_t tiling_stride; + /* + * scheme 0: tile is 256x16, while minimal tile stride is 512 + * scheme 1: tile is 512x8, while minimal tile stride is 1024 + */ + uint8_t tiling_scheme; +}; + +typedef struct drm_ipvr_private { + struct drm_device *dev; + struct pci_dev *pci_root; + /* pci revision id for B0:D2:F0 */ + uint8_t platform_rev_id; + + uint32_t device_id; + + /* IMG video context */ + struct list_head ipvr_ctx_list; + spinlock_t ipvr_ctx_lock; + struct idr ipvr_ctx_idr; + struct ipvr_context default_ctx; + + /* PM related */ + struct mutex ved_pm_mutex; + atomic_t ved_power_usage; + + /* exec related */ + struct mutex cmdbuf_mutex; + struct ipvr_validate_context validate_ctx; + + /* IMG MMU specific */ + struct ipvr_mmu_driver *mmu; + struct ipvr_mmu_pd *pf_pd; + atomic_t ipvr_mmu_invaldc; + + /* GEM mm related */ + struct ipvr_gem_mm ipvr_mm; + struct kmem_cache *ipvr_bo_slab; + struct ipvr_address_space addr_space; + + /* fence related */ + uint32_t last_seq; + wait_queue_head_t fence_queue; + struct ipvr_fence_driver fence_drv; + + /* + * VED specific + */ + uint8_t __iomem* ved_reg_base; + unsigned long ved_reg_offset; + struct ved_private *ved_private; +}drm_ipvr_private_t; + +struct ved_mb_region { + uint32_t start; + uint32_t end; +}; + +struct ved_decode_status { + uint32_t num_region; + struct ved_mb_region mb_regions[MAX_SLICES_PER_PICTURE]; +}; + +struct ved_frame_info { + uint32_t handle; + uint32_t surface_id; + uint16_t fence; + uint32_t buffer_stride; + uint32_t buffer_size; + uint32_t picture_width_mb; + uint32_t fw_status; + uint32_t size_mb; + struct ved_decode_status decode_status; +}; + +struct drm_ipvr_gem_object; + +/* VED private structure */ +struct ved_private { + struct drm_device *dev; + struct drm_ipvr_private *dev_priv; + + /* used to record seq got from irq fw-to-host msg */ + uint16_t ved_cur_seq; + + /* + * VED Rendec Memory + */ + struct drm_ipvr_gem_object *ccb0; + uint32_t base_addr0; + struct drm_ipvr_gem_object *ccb1; + uint32_t base_addr1; + bool rendec_initialized; + + /* VED firmware related */ + struct drm_ipvr_gem_object *fw_bo; + uint32_t fw_offset; + uint32_t mtx_mem_size; + bool fw_loaded_to_bo; + bool ved_fw_loaded; + void *ved_fw_ptr; + int ved_fw_size; + uint32_t fw_b0_uploaded; + /* + * there are two ways to load fw: + * 1, load fw directly by kernel driver, byt use this way + * 2, load by punit, which has security benefits + */ + bool fw_loaded_by_punit; + + /* + * ved command queue + */ + spinlock_t ved_lock; + struct mutex ved_mutex; + struct list_head ved_queue; + /* busy means cmd submitted to fw, while irq hasn't been receieved */ + bool ved_busy; + /* VED status read from register 0x20D0 */ + uint32_t ved_hw_busy; + + uint32_t ved_dash_access_ctrl; + uint32_t decoding_err; + + struct ved_frame_info frame_info[MAX_DECODE_BUFFERS]; + struct ved_decode_status decode_status; + uint32_t host_be_opp_enabled; + + /* error concealment related */ + struct work_struct ec_work; + struct drm_file *tfile; /* protected by cmdbuf_mutex */ + struct ved_ec_context *ved_ec_ctx[VED_MAX_EC_INSTANCE]; + struct ved_ec_context *cur_msvdx_ec_ctx; + uint32_t deblock_cmd_offset; + int num_cmd; + uint32_t vec_ec_mem_data[5]; + uint32_t vec_ec_mem_saved; + + /* pm related */ + int ved_needs_reset; + unsigned int ved_pmstate; + struct kernfs_node *sysfs_pmstate; + uint32_t pm_gating_count; + /* pm suspend wq, use wq for current implementation */ + struct delayed_work ved_suspend_wq; + struct tasklet_struct ved_suspend_tq; + + /* protected by ved_mutex */ + /* current ved decode context */ + struct ipvr_context *ipvr_ctx; + + struct page *mmu_recover_page; +}; + +struct drm_ipvr_file_private { + uint32_t ctx_id; +}; + +/* public ipvr gem functions for bo/page operation */ +void ipvr_gem_object_pin_pages(struct drm_ipvr_gem_object *obj); +void ipvr_gem_object_unpin_pages(struct drm_ipvr_gem_object *obj); +void *ipvr_gem_object_alloc(struct drm_device *dev); +void ipvr_gem_object_free(struct drm_ipvr_gem_object *obj); +int32_t ipvr_gem_object_get_pages(struct drm_ipvr_gem_object *obj); +int32_t ipvr_gem_object_put_pages(struct drm_ipvr_gem_object *obj); + +int32_t ipvr_sync_cpu_grab(struct drm_device *dev, + struct drm_ipvr_gem_object *obj); +int32_t ipvr_sync_cpu_release(struct drm_device *dev, + struct drm_ipvr_gem_object *obj); + +/* PRIME callbacks */ +struct sg_table *ipvr_gem_prime_get_sg_table(struct drm_gem_object *obj); +void *ipvr_gem_prime_vmap(struct drm_gem_object *obj); +void ipvr_gem_prime_vunmap(struct drm_gem_object *obj, void *vaddr); +struct drm_gem_object *ipvr_gem_prime_import_sg_table(struct drm_device *dev, + size_t size, struct sg_table *sg); +int ipvr_gem_prime_pin(struct drm_gem_object *obj); +void ipvr_gem_prime_unpin(struct drm_gem_object *obj); + +#endif diff --git a/drivers/gpu/drm/ipvr/ipvr_exec.c b/drivers/gpu/drm/ipvr/ipvr_exec.c new file mode 100644 index 0000000..8b9d638 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_exec.c @@ -0,0 +1,530 @@ +/************************************************************************** + * ipvr_exec.c: IPVR command buffer execution + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + +#include "ipvr_exec.h" +#include "ipvr_gem.h" +#include "ipvr_mmu.h" +#include "ipvr_buffer.h" +#include "ipvr_fence.h" +#include "ipvr_trace.h" +#include "ved_fw.h" +#include "ved_msg.h" +#include "ved_reg.h" +#include "ved_ec.h" +#include "ved_init.h" +#include "ved_pm.h" +#include "ved_cmd.h" + +#include <linux/io.h> +#include <linux/delay.h> +#include <linux/pm_runtime.h> + +static inline bool ipvr_bo_is_reserved(struct drm_ipvr_gem_object *obj) +{ + return atomic_read(&obj->reserved); +} + +static int32_t +ipvr_bo_wait_unreserved(struct drm_ipvr_gem_object *obj, bool interruptible) +{ + if (interruptible) { + return wait_event_interruptible(obj->event_queue, + !ipvr_bo_is_reserved(obj)); + } else { + wait_event(obj->event_queue, !ipvr_bo_is_reserved(obj)); + return 0; + } +} + +/** + * ipvr_bo_reserve - reserve the given bo + * + * @obj: The buffer object to reserve. + * @interruptible: whether the waiting is interruptible or not. + * @no_wait: flag to indicate returning immediately + * + * Returns: + * -EBUSY: if bo is busy and @no_wait is true. + * -ERESTARTSYS if waiting was interrupted by a signal. + * 0 if reserving succeeded. + */ +int32_t ipvr_bo_reserve(struct drm_ipvr_gem_object *obj, + bool interruptible, bool no_wait) +{ + int32_t ret; + + while (unlikely(atomic_xchg(&obj->reserved, 1) != 0)) { + if (no_wait) + return -EBUSY; + IPVR_DEBUG_GENERAL("wait bo unreserved, add to wait queue.\n"); + ret = ipvr_bo_wait_unreserved(obj, interruptible); + if (unlikely(ret)) + return ret; + } + + return 0; +} + +/** + * ipvr_bo_unreserve - unreserve the given bo + * + * @obj: The buffer object to reserve. + * + * No return value. + */ +void ipvr_bo_unreserve(struct drm_ipvr_gem_object *obj) +{ + atomic_set(&obj->reserved, 0); + wake_up_all(&obj->event_queue); +} + +static void ipvr_backoff_reservation(struct list_head *list) +{ + struct ipvr_validate_buffer *entry; + + list_for_each_entry(entry, list, head) { + struct drm_ipvr_gem_object *obj = entry->ipvr_gem_bo; + if (!atomic_read(&obj->reserved)) + continue; + atomic_set(&obj->reserved, 0); + wake_up_all(&obj->event_queue); + } +} + +/* + * ipvr_reserve_buffers - Reserve buffers for validation. + * + * @list: points to a bo list to be backoffed + * + * If a buffer in the list is marked for CPU access, we back off and + * wait for that buffer to become free for GPU access. + * + * If a buffer is reserved for another validation, the validator with + * the highest validation sequence backs off and waits for that buffer + * to become unreserved. This prevents deadlocks when validating multiple + * buffers in different orders. + * + * Returns: + * -EBUSY: if bo is busy and @no_wait is true. + * -ERESTARTSYS if waiting was interrupted by a signal. + * 0 if reserving succeeded. + */ +int32_t ipvr_reserve_buffers(struct list_head *list) +{ + struct ipvr_validate_buffer *entry; + int32_t ret; + + if (list_empty(list)) + return 0; + + list_for_each_entry(entry, list, head) { + struct drm_ipvr_gem_object *bo = entry->ipvr_gem_bo; + + ret = ipvr_bo_reserve(bo, true, true); + switch (ret) { + case 0: + break; + case -EBUSY: + ret = ipvr_bo_reserve(bo, true, false); + if (!ret) + break; + else + goto err; + default: + goto err; + } + + if (unlikely(atomic_read(&bo->cpu_writers) > 0)) { + ret = -EBUSY; + goto err; + } + } + + return 0; +err: + ipvr_backoff_reservation(list); + return ret; +} + +/** + * ipvr_set_tile - global setting of tiling info + * + * @dev: the ipvr drm device + * @tiling_scheme: see ipvr_drm.h for details + * @tiling_stride: see ipvr_drm.h for details + * + * vxd392 hardware supports only one tile region so this configuration + * is global. + */ +void ipvr_set_tile(struct drm_device *dev, + uint8_t tiling_scheme, uint8_t tiling_stride) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + uint32_t cmd; + uint32_t start = IPVR_MEM_MMU_TILING_START; + uint32_t end = IPVR_MEM_MMU_TILING_END; + + /* Enable memory tiling */ + cmd = ((start >> 20) + (((end >> 20) - 1) << 12) + + ((0x8 | tiling_stride) << 24)); + IPVR_DEBUG_GENERAL("VED: MMU Tiling register0 %08x.\n", cmd); + IPVR_DEBUG_GENERAL("Region 0x%08x-0x%08x.\n", start, end); + VED_REG_WRITE32(cmd, MSVDX_MMU_TILE_BASE0_OFFSET); + + /* we need set tile format as 512x8 on Baytrail, which is shceme 1 */ + VED_REG_WRITE32(tiling_scheme << 3, MSVDX_MMU_CONTROL2_OFFSET); +} + +/** + * ipvr_find_ctx_with_fence - lookup the context with given fence seqno + * + * @dev_priv: the ipvr drm device + * @fence: fence seqno generated by the context + * + * Returns: + * context pointer if found. + * NULL if not found. + */ +struct ipvr_context* +ipvr_find_ctx_with_fence(struct drm_ipvr_private *dev_priv, uint16_t fence) +{ + struct ipvr_context *pos = NULL, *n = NULL; + + if (unlikely(dev_priv == NULL)) { + return NULL; + } + + spin_lock(&dev_priv->ipvr_ctx_lock); + list_for_each_entry_safe(pos, n, &dev_priv->ipvr_ctx_list, head) { + if (pos->cur_seq == fence) { + spin_unlock(&dev_priv->ipvr_ctx_lock); + return pos; + } + } + spin_unlock(&dev_priv->ipvr_ctx_lock); + + return NULL; +} + +static void ipvr_unreference_buffers(struct ipvr_validate_context *context) +{ + struct ipvr_validate_buffer *entry, *next; + struct drm_ipvr_gem_object *obj; + struct list_head *list = &context->validate_list; + + list_for_each_entry_safe(entry, next, list, head) { + obj = entry->ipvr_gem_bo; + list_del(&entry->head); + drm_gem_object_unreference(&obj->base); + context->used_buffers--; + } +} + +static int ipvr_update_buffers(struct drm_file *file_priv, + struct ipvr_validate_context *context, + uint64_t buffer_list, + int32_t count) +{ + struct ipvr_validate_buffer *entry; + struct ipvr_validate_arg *val_arg = + (struct ipvr_validate_arg __user *)(unsigned long) buffer_list; + + if (list_empty(&context->validate_list)) + return 0; + + list_for_each_entry(entry, &context->validate_list, head) { + if (!val_arg) { + IPVR_DEBUG_WARN("unexpected end of val_arg list!!!\n"); + return -EINVAL; + } + if (unlikely(copy_to_user(val_arg, &entry->val_req, + sizeof(entry->val_req)))) { + IPVR_ERROR("copy_to_user fault.\n"); + return -EFAULT; + } + val_arg = (struct ipvr_validate_arg __user *) + (unsigned long)entry->val_req.next; + } + return 0; +} + +static int ipvr_reference_buffers(struct drm_file *file_priv, + struct ipvr_validate_context *context, + uint64_t buffer_list, + int32_t count) +{ + struct drm_device *dev = file_priv->minor->dev; + struct ipvr_validate_arg *val_arg = + (struct ipvr_validate_arg __user *)(unsigned long) buffer_list; + struct ipvr_validate_buffer *item; + struct drm_ipvr_gem_object *obj; + int32_t ret = 0; + + while (likely(val_arg != 0) && (context->used_buffers <= count)) { + if (unlikely(context->used_buffers >= + IPVR_NUM_VALIDATE_BUFFERS)) { + IPVR_ERROR("Too many buffers " + "on validate list.\n"); + ret = -EINVAL; + goto out_err; + } + item = &context->buffers[context->used_buffers]; + if (unlikely(copy_from_user(&item->val_req, val_arg, + sizeof(item->val_req)) != 0)) { + IPVR_ERROR("copy_from_user fault.\n"); + ret = -EFAULT; + goto out_err; + } + INIT_LIST_HEAD(&item->head); + obj = to_ipvr_bo(drm_gem_object_lookup(dev, file_priv, + item->val_req.handle)); + if (&obj->base == NULL) { + ret = -ENOENT; + goto out_err; + } + item->ipvr_gem_bo = obj; + + list_add_tail(&item->head, &context->validate_list); + context->used_buffers++; + + val_arg = (struct ipvr_validate_arg __user *) + (unsigned long)item->val_req.next; + } + + return 0; + +out_err: + ipvr_unreference_buffers(context); + return ret; +} + +static int32_t ipvr_validate_buffer_list(struct drm_file *file_priv, + struct ipvr_validate_context *context) +{ + struct ipvr_validate_buffer *entry; + struct drm_ipvr_gem_object *obj; + struct list_head *list = &context->validate_list; + int32_t ret = 0; + + list_for_each_entry(entry, list, head) { + obj = entry->ipvr_gem_bo; + /** + * need validate bo locate in the mmu space + * check if presumed offset is correct + * with ved_check_presumed, if presume is not correct, + * call fixup relocs with ved_fixup_relocs. + * current implementation doesn't support shrink/evict, + * so needn't validate mmu offset. + * need be implemented in the future if shrink/evict + * is supported. + */ + } + + return ret; +} + +/** + * ipvr_gem_do_execbuffer - lookup the context with given fence seqno + * + * @dev: the ipvr drm device + * @file_priv: the ipvr drm file pointer + * @args: input argument passed from userland + * @vm: ipvr address space for all the bo to bind to + * + * Returns: + * -ENOENT if context not found, or cmdbuf bo not found + * -EINVAL if referencing buffer fails, or executing cmdbuf fails + * -EINTR if fails to lock mutex + * -EBUSY if fails to get power well, or execution fails + * -ERESTARTSYS if reservating buffer fails + * -ENOMEM if execution fails + * -EFAULT if execution fails + * 0 if successful + */ +static int32_t ipvr_gem_do_execbuffer(struct drm_device *dev, + struct drm_file *file_priv, + struct drm_ipvr_gem_execbuffer *args, + struct ipvr_address_space *vm) +{ + drm_ipvr_private_t *dev_priv = dev->dev_private; + struct ipvr_validate_context *context = &dev_priv->validate_ctx; + struct ved_private *ved_priv = dev_priv->ved_private; + struct drm_ipvr_gem_object *cmd_buffer; + struct ipvr_context *ipvr_ctx = NULL; + int32_t ret, ctx_id; + + /* if not pass 0, use default context instead */ + if (args->ctx_id == 0) + ctx_id = dev_priv->default_ctx.ctx_id; + else + ctx_id = args->ctx_id; + + IPVR_DEBUG_GENERAL("try to find ctx according ctx_id %d.\n", ctx_id); + ipvr_ctx = (struct ipvr_context *) + idr_find(&dev_priv->ipvr_ctx_idr, ctx_id); + if (!ipvr_ctx) { + IPVR_DEBUG_WARN("video ctx is not found.\n"); + return -ENOENT; + } + + IPVR_DEBUG_GENERAL("reference all buffers passed through buffer_list.\n"); + ret = ipvr_reference_buffers(file_priv, context, + args->buffer_list, args->buffer_count); + if (unlikely(ret != 0)) { + IPVR_DEBUG_WARN("reference buffer failed.\n"); + return -EINVAL; + } + + IPVR_DEBUG_GENERAL("reserve all buffers to make them not accessed " + "by other threads.\n"); + ret = ipvr_reserve_buffers(&context->validate_list); + if (unlikely(ret != 0)) { + IPVR_DEBUG_WARN("reserve buffers failed.\n"); + /* -EBUSY or -ERESTARTSYS */ + goto out_unref_buf; + } + + IPVR_DEBUG_GENERAL("validate buffer list, mainly check " + "the bo gpu offset.\n"); + ret = ipvr_validate_buffer_list(file_priv, context); + if (ret) { + IPVR_DEBUG_WARN("validate buffers failed.\n"); + goto out_backoff_reserv; + } + + cmd_buffer = to_ipvr_bo(idr_find(&file_priv->object_idr, + args->cmdbuf_handle)); + if (!cmd_buffer) { + IPVR_DEBUG_WARN("Invalid cmd object handle 0x%x.\n", + args->cmdbuf_handle); + ret = -ENOENT; + goto out_backoff_reserv; + } + + /* + * here VED is supported currently + * when support VEC and VSP, need implement a general way to + * call sub-engine functions + */ + atomic_inc(&dev_priv->ved_power_usage); + IPVR_DEBUG_GENERAL("get VED power with usage=%d.\n", + atomic_read(&dev->platformdev->dev.power.usage_count)); + ret = pm_runtime_get_sync(&dev->platformdev->dev); + if (unlikely(ret < 0)) { + IPVR_ERROR("Error get VED power: %d\n", ret); + ret = -EBUSY; + goto out_backoff_reserv; + } + + ret = mutex_lock_interruptible(&ved_priv->ved_mutex); + if (unlikely(ret)) { + IPVR_ERROR("Error get VED mutex: %d\n", ret); + /* -EINTR */ + goto out_power_put; + } + + IPVR_DEBUG_GENERAL("parse cmd buffer and send to VED.\n"); + ret = ved_cmdbuf_video(file_priv, cmd_buffer, + args->cmdbuf_size, ipvr_ctx ); + if (unlikely(ret)) { + IPVR_DEBUG_WARN("ved_cmdbuf_video returns %d.\n", ret); + /* -EINVAL, -ENOMEM, -EFAULT, -EBUSY */ + goto out_power_put; + } + + mutex_unlock(&ved_priv->ved_mutex); + + /** + * update mmu_offsets and fence fds to user + */ + ret = ipvr_update_buffers(file_priv, context, + args->buffer_list, args->buffer_count); + if (unlikely(ret)) { + IPVR_DEBUG_WARN("ipvr_update_buffers returns error %d.\n", ret); + ret = 0; + } + +out_power_put: + if (ret) { + ret = pm_runtime_put(&dev->platformdev->dev); + if (unlikely(ret < 0)) + IPVR_ERROR("Error put VED power: %d\n", ret); + IPVR_DEBUG_GENERAL("VED power put. usage became %d.\n", + atomic_read(&dev->platformdev->dev.power.usage_count)); + } + +out_backoff_reserv: + IPVR_DEBUG_GENERAL("unreserve buffer list.\n"); + ipvr_backoff_reservation(&context->validate_list); + +out_unref_buf: + IPVR_DEBUG_GENERAL("unref bufs which are refered during bo lookup.\n"); + ipvr_unreference_buffers(context); + + return ret; +} + +/** + * ipvr_gem_do_execbuffer - lookup the context with given fence seqno + * + * ioctl entry for DRM_IPVR_GEM_EXECBUFFER + * + * Returns: + * -ENOENT if context not found, or cmdbuf bo not found + * -EINVAL if referencing buffer fails, or executing cmdbuf fails + * -EINTR if fails to lock mutex + * -EBUSY if fails to get power well, or execution fails + * -ERESTARTSYS if reservating buffer fails + * -ENOMEM if execution fails + * -EFAULT if execution fails + * 0 if successful + */ +int32_t ipvr_gem_execbuffer(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct drm_ipvr_gem_execbuffer *args = data; + int32_t ret; + struct ipvr_validate_context *context = &dev_priv->validate_ctx; + + if (!context || !context->buffers) { + ret = -EINVAL; + return ret; + } + + context->used_buffers = 0; + + if (args->buffer_count < 1 || + args->buffer_count > + (UINT_MAX / sizeof(struct ipvr_validate_buffer))) { + IPVR_ERROR("validate %d buffers.\n", args->buffer_count); + return -EINVAL; + } + + trace_ipvr_gem_exec_ioctl(args); + ret = ipvr_gem_do_execbuffer(dev, file_priv, args, + &dev_priv->addr_space); + return ret; +} diff --git a/drivers/gpu/drm/ipvr/ipvr_exec.h b/drivers/gpu/drm/ipvr/ipvr_exec.h new file mode 100644 index 0000000..68405df --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_exec.h @@ -0,0 +1,68 @@ +/************************************************************************** + * ipvr_exec.h: IPVR header file for command buffer execution + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * + **************************************************************************/ + +#ifndef _IPVR_EXEC_H_ +#define _IPVR_EXEC_H_ + +#include "ipvr_drv.h" +#include "ipvr_drm.h" +#include "ipvr_gem.h" +#include "ipvr_fence.h" + +struct drm_ipvr_private; + +#define IPVR_NUM_VALIDATE_BUFFERS 2048 + +#define IPVR_MAX_RELOC_PAGES 1024 + +/* status of the command sent to the ipvr device */ +enum ipvr_cmd_status { + IPVR_CMD_SUCCESS, + IPVR_CMD_FAILED, + IPVR_CMD_LOCKUP, + IPVR_CMD_SKIP +}; + +struct ipvr_validate_buffer { + struct ipvr_validate_arg val_req; + struct list_head head; + struct drm_ipvr_gem_object *ipvr_gem_bo; + struct ipvr_fence *old_fence; +}; + +int ipvr_bo_reserve(struct drm_ipvr_gem_object *obj, + bool interruptible, bool no_wait); + +void ipvr_bo_unreserve(struct drm_ipvr_gem_object *obj); + +struct ipvr_context* +ipvr_find_ctx_with_fence(struct drm_ipvr_private *dev_priv, uint16_t fence); + +void ipvr_set_tile(struct drm_device *dev, + uint8_t tiling_scheme, uint8_t tiling_stride); + +int ipvr_cmdbuf_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); + +#endif diff --git a/drivers/gpu/drm/ipvr/ipvr_fence.c b/drivers/gpu/drm/ipvr/ipvr_fence.c new file mode 100644 index 0000000..838e0d4 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_fence.c @@ -0,0 +1,550 @@ +/************************************************************************** + * ipvr_fence.c: IPVR fence handling to track command exectuion status + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + +#include "ipvr_fence.h" +#include "ved_reg.h" +#include "ved_fw.h" +#include "ved_cmd.h" +#include "ipvr_exec.h" +#include "ipvr_buffer.h" +#include "ipvr_trace.h" +#include <linux/debugfs.h> +#include <linux/export.h> +#include <linux/file.h> +#include <linux/fs.h> +#include <linux/kernel.h> +#include <linux/poll.h> +#include <linux/sched.h> +#include <linux/seq_file.h> +#include <linux/slab.h> +#include <linux/uaccess.h> +#include <linux/anon_inodes.h> + +static int32_t ipvr_usrfence_release(struct inode *inode, struct file *file); +static uint32_t ipvr_usrfence_poll(struct file *file, poll_table *wait); +static const struct file_operations ipvr_usrfence_fops = { + .release = ipvr_usrfence_release, + .poll = ipvr_usrfence_poll, +}; + +/** + * ipvr_fence_create - create and init a fence + * + * @dev_priv: drm_ipvr_private pointer + * @fence: ipvr fence object + * @fence_fd: file descriptor for exporting fence + * + * Create a fence, actually the fence is written to ipvr through msg. + * exporting a new file descriptor to userspace. + * Returns 0 on success, -ENOMEM on failure. + */ +int32_t +ipvr_fence_create(struct drm_ipvr_private *dev_priv, + struct ipvr_fence **fence, + int *fence_fd) +{ + unsigned long irq_flags; + uint16_t old_seq; + int fd; + struct ved_private *ved_priv; + BUG_ON(!fence_fd || *fence_fd >= 0); + fd = get_unused_fd(); + ved_priv = dev_priv->ved_private; + + if (fd < 0) { + IPVR_ERROR("ALARM!!! no unused FD found!\n"); + return fd; + } + + *fence = kmalloc(sizeof(struct ipvr_fence), GFP_KERNEL); + if ((*fence) == NULL) { + put_unused_fd(fd); + return -ENOMEM; + } + (*fence)->file = anon_inode_getfile("ipvr_usrfence", &ipvr_usrfence_fops, + *fence, 0); + if (IS_ERR((*fence)->file)) { + kfree(*fence); + put_unused_fd(fd); + IPVR_ERROR("ALARM!!! anon_inode_getfile call failed\n"); + return -ENOMEM; + } + + kref_init(&((*fence)->kref)); + (*fence)->dev_priv = dev_priv; + + spin_lock_irqsave(&dev_priv->fence_drv.fence_lock, irq_flags); + /* cmds in one batch use different fence value */ + old_seq = dev_priv->fence_drv.sync_seq; + dev_priv->fence_drv.sync_seq = dev_priv->last_seq++; + dev_priv->fence_drv.sync_seq <<= 4; + dev_priv->fence_drv.sync_seq += ved_priv->num_cmd; + (*fence)->seq = dev_priv->fence_drv.sync_seq; + + spin_unlock_irqrestore(&dev_priv->fence_drv.fence_lock, irq_flags); + + fd_install(fd, (*fence)->file); + kref_get(&(*fence)->kref); + IPVR_DEBUG_GENERAL("fence is created and its seq is %u (0x%04x), fd is %d.\n", + (*fence)->seq, (*fence)->seq, fd); + *fence_fd = fd; + return 0; +} + +/** + * ipvr_fence_destroy - destroy a fence + * + * @kref: fence kref + * + * Frees the fence object (all asics). + */ +static void ipvr_fence_destroy(struct kref *kref) +{ + struct ipvr_fence *fence; + + fence = container_of(kref, struct ipvr_fence, kref); + kfree(fence); +} + +/** + * ipvr_fence_process - process a fence + * + * @dev_priv: drm_ipvr_private pointer + * @seq: indicate the fence seq has been signaled + * @err: indicate if err happened, for future use + * + * Checks the current fence value and wakes the fence queue + * if the sequence number has increased (all asics). + */ +void ipvr_fence_process(struct drm_ipvr_private *dev_priv, + uint16_t seq, uint8_t err) +{ + int signaled_seq_int; + uint16_t signaled_seq; + uint16_t last_emitted; + + signaled_seq_int = atomic_read(&dev_priv->fence_drv.signaled_seq); + signaled_seq = (uint16_t)signaled_seq_int; + last_emitted = dev_priv->fence_drv.sync_seq; + + if (ipvr_seq_after(seq, last_emitted)) { + IPVR_DEBUG_WARN("seq error, seq is %u, signaled_seq is %u, " + "last_emitted is %u.\n", + seq, signaled_seq, last_emitted); + return; + } + if (ipvr_seq_after(seq, signaled_seq)) { + atomic_xchg(&dev_priv->fence_drv.signaled_seq, seq); + dev_priv->fence_drv.last_activity = jiffies; + IPVR_DEBUG_GENERAL("last emitted seq %u is updated.\n", seq); + wake_up_all(&dev_priv->fence_queue); + } +} + +/** + * ipvr_fence_signaled - check if a fence sequeuce number has signaled + * + * @dev_priv: ipvr device pointer + * @seq: sequence number + * + * Check if the last singled fence sequnce number is >= the requested + * sequence number (all asics). + * Returns true if the fence has signaled (current fence value + * is >= requested value) or false if it has not (current fence + * value is < the requested value. + */ +static bool ipvr_fence_signaled(struct drm_ipvr_private *dev_priv, uint16_t seq) +{ + uint16_t curr_seq, signaled_seq; + unsigned long irq_flags; + spin_lock_irqsave(&dev_priv->fence_drv.fence_lock, irq_flags); + curr_seq = dev_priv->ved_private->ved_cur_seq; + signaled_seq = atomic_read(&dev_priv->fence_drv.signaled_seq); + + if (ipvr_seq_after(seq, signaled_seq)) { + /* poll new last sequence at least once */ + ipvr_fence_process(dev_priv, curr_seq, IPVR_CMD_SUCCESS); + signaled_seq = atomic_read(&dev_priv->fence_drv.signaled_seq); + if (ipvr_seq_after(seq, signaled_seq)) { + spin_unlock_irqrestore(&dev_priv->fence_drv.fence_lock, + irq_flags); + return false; + } + } + spin_unlock_irqrestore(&dev_priv->fence_drv.fence_lock, irq_flags); + return true; +} + +/** + * ipvr_fence_lockup - ipvr lockup is detected + * + * @dev_priv: ipvr device pointer + * @fence: lockup detected when wait the specific fence + * + * During the calling of ipvr_fence_wait, if wait to timeout, + * indicate lockup happened, need flush cmd queue and reset ved + * If ipvr_fence_wait_empty_locked encounter lockup, fence is NULL + */ + +static void +ipvr_fence_lockup(struct drm_ipvr_private *dev_priv, struct ipvr_fence *fence) +{ + unsigned long irq_flags; + struct drm_device *dev = (struct drm_device *)dev_priv->dev; + struct ved_private *ved_priv = dev_priv->ved_private; + + IPVR_DEBUG_WARN("timeout detected, flush queued cmd, maybe lockup.\n"); + IPVR_DEBUG_WARN("MSVDX_COMMS_FW_STATUS reg is 0x%x.\n", + VED_REG_READ32(MSVDX_COMMS_FW_STATUS)); + + if (fence) { + spin_lock_irqsave(&dev_priv->fence_drv.fence_lock, irq_flags); + ipvr_fence_process(dev_priv, fence->seq, IPVR_CMD_LOCKUP); + spin_unlock_irqrestore(&dev_priv->fence_drv.fence_lock, irq_flags); + } + ved_flush_cmd_queue(dev); + + if (ved_priv->fw_loaded_by_punit) + ved_priv->ved_needs_reset |= MSVDX_RESET_NEEDS_REUPLOAD_FW | + MSVDX_RESET_NEEDS_INIT_FW; + else + ved_priv->ved_needs_reset = 1; +} + +/** + * ipvr_fence_wait_seq - wait for a specific sequence number + * + * @dev_priv: ipvr device pointer + * @target_seq: sequence number we want to wait for + * @intr: use interruptable sleep + * + * Wait for the requested sequence number to be written. + * @intr selects whether to use interruptable (true) or non-interruptable + * (false) sleep when waiting for the sequence number. + * Returns 0 if the sequence number has passed, error for all other cases. + * -EDEADLK is returned when a GPU lockup has been detected. + */ +static int32_t ipvr_fence_wait_seq(struct drm_ipvr_private *dev_priv, + uint16_t target_seq, bool intr) +{ + struct ipvr_fence_driver *fence_drv = &dev_priv->fence_drv; + unsigned long timeout, last_activity; + uint16_t signaled_seq; + int32_t ret; + unsigned long irq_flags; + bool signaled; + spin_lock_irqsave(&dev_priv->fence_drv.fence_lock, irq_flags); + + while (ipvr_seq_after(target_seq, + (uint16_t)atomic_read(&fence_drv->signaled_seq))) { + /* seems the fence_drv->last_activity is useless? */ + timeout = IPVR_FENCE_JIFFIES_TIMEOUT; + signaled_seq = atomic_read(&fence_drv->signaled_seq); + /* save last activity valuee, used to check for GPU lockups */ + last_activity = fence_drv->last_activity; + + spin_unlock_irqrestore(&dev_priv->fence_drv.fence_lock, irq_flags); + if (intr) { + ret = wait_event_interruptible_timeout( + dev_priv->fence_queue, + (signaled = ipvr_fence_signaled(dev_priv, target_seq)), + timeout); + } else { + ret = wait_event_timeout( + dev_priv->fence_queue, + (signaled = ipvr_fence_signaled(dev_priv, target_seq)), + timeout); + } + spin_lock_irqsave(&dev_priv->fence_drv.fence_lock, irq_flags); + + if (unlikely(!signaled)) { + /* we were interrupted for some reason and fence + * isn't signaled yet, resume waiting until timeout */ + if (unlikely(ret < 0)) { + /* should return -ERESTARTSYS, + * interrupted by a signal */ + continue; + } + + /* check if sequence value has changed since + * last_activity */ + if (signaled_seq != + atomic_read(&fence_drv->signaled_seq)) { + continue; + } + + if (last_activity != fence_drv->last_activity) { + continue; + } + + /* lockup happen, it is better have some reg to check */ + IPVR_DEBUG_WARN("GPU lockup (waiting for 0x%0x last " + "signaled fence id 0x%x).\n", + target_seq, signaled_seq); + + /* change last activity so nobody else + * think there is a lockup */ + fence_drv->last_activity = jiffies; + spin_unlock_irqrestore(&dev_priv->fence_drv.fence_lock, + irq_flags); + return -EDEADLK; + + } + } + spin_unlock_irqrestore(&dev_priv->fence_drv.fence_lock, irq_flags); + return 0; +} + +/** + * ipvr_fence_wait - wait for a fence to signal + * + * @fence: ipvr fence object + * @intr: use interruptable sleep + * @no_wait: not signaled, if need add into wait queue + * + * Wait for the requested fence to signal (all asics). + * @intr selects whether to use interruptable (true) or non-interruptable + * (false) sleep when waiting for the fence. + * Returns 0 if the fence has passed, error for all other cases. + */ +int32_t ipvr_fence_wait(struct ipvr_fence *fence, bool intr, bool no_wait) +{ + int32_t ret; + struct drm_ipvr_private *dev_priv; + + if (fence == NULL || fence->seq == IPVR_FENCE_SIGNALED_SEQ) { + IPVR_DEBUG_GENERAL("fence is NULL or has been singaled.\n"); + return 0; + } + dev_priv = fence->dev_priv; + + IPVR_DEBUG_GENERAL("wait fence seq %u, last signaled seq is %d, " + "last emitted seq is %u.\n", fence->seq, + atomic_read(&dev_priv->fence_drv.signaled_seq), + dev_priv->fence_drv.sync_seq); + trace_ipvr_fence_wait(fence, + atomic_read(&dev_priv->fence_drv.signaled_seq), + dev_priv->fence_drv.sync_seq); + + if (ipvr_fence_signaled(dev_priv, fence->seq)) { + IPVR_DEBUG_GENERAL("fence has been signaled.\n"); + /* + * compare with ttm_bo_wait, don't need create a tmp_obj + * it is better we also set bo->fence = NULL + */ + fence->seq = IPVR_FENCE_SIGNALED_SEQ; + ipvr_fence_unref(&fence); + return 0; + } + + if (no_wait) + return -EBUSY; + + ret = ipvr_fence_wait_seq(dev_priv, fence->seq, intr); + if (ret) { + if (ret == -EDEADLK) + ipvr_fence_lockup(dev_priv, fence); + return ret; + } + fence->seq = IPVR_FENCE_SIGNALED_SEQ; + + return 0; +} + +/** + * ipvr_fence_driver_init - init the fence driver + * + * @dev_priv: ipvr device pointer + * + * Init the fence driver, will not fail + */ +void ipvr_fence_driver_init(struct drm_ipvr_private *dev_priv) +{ + spin_lock_init(&dev_priv->fence_drv.fence_lock); + init_waitqueue_head(&dev_priv->fence_queue); + dev_priv->fence_drv.sync_seq = 0; + atomic_set(&dev_priv->fence_drv.signaled_seq, 0); + dev_priv->fence_drv.last_activity = jiffies; + dev_priv->fence_drv.initialized = false; +} + +/** + * ipvr_fence_wait_empty_locked - wait for all fences to signal + * + * @dev_priv: ipvr device pointer + * + * Wait for all fences to be signalled. + */ +void ipvr_fence_wait_empty_locked(struct drm_ipvr_private *dev_priv) +{ + uint16_t seq; + + seq = dev_priv->fence_drv.sync_seq; + + while(1) { + int32_t ret; + ret = ipvr_fence_wait_seq(dev_priv, seq, false); + if (ret == 0) { + return; + } else if (ret == -EDEADLK) { + ipvr_fence_lockup(dev_priv, NULL); + IPVR_DEBUG_WARN("Lockup found waiting for seq %d.\n", + seq); + return; + } else { + continue; + } + } +} + +/** + * ipvr_fence_driver_fini - tear down the fence driver + * for all possible rings. + * + * @dev_priv: ipvr device pointer + * + * Tear down the fence driver for all possible rings (all asics). + */ +void ipvr_fence_driver_fini(struct drm_ipvr_private *dev_priv) +{ + if (!dev_priv->fence_drv.initialized) + return; + ipvr_fence_wait_empty_locked(dev_priv); + wake_up_all(&dev_priv->fence_queue); + dev_priv->fence_drv.initialized = false; +} + +/** + * ipvr_fence_ref - take a ref on a fence + * + * @fence: fence object + * + * Take a reference on a fence (all asics). + * Returns the fence. + */ +struct ipvr_fence *ipvr_fence_ref(struct ipvr_fence *fence) +{ + kref_get(&fence->kref); + return fence; +} + +/** + * ipvr_fence_unref - remove a ref on a fence + * + * @fence: ipvr fence object + * + * Remove a reference on a fence, if ref == 0, destory the fence. + */ +void ipvr_fence_unref(struct ipvr_fence **fence) +{ + struct ipvr_fence *tmp = *fence; + + *fence = NULL; + if (tmp) { + kref_put(&tmp->kref, &ipvr_fence_destroy); + } +} + +/** + * ipvr_fence_buffer_objects - bind fence to buffer list + * + * @list: validation buffer list + * @fence: ipvr fence object + * + * bind a fence to all obj in the validation list + */ +void +ipvr_fence_buffer_objects(struct list_head *list, struct ipvr_fence *fence, int fence_fd) +{ + struct ipvr_validate_buffer *entry; + struct drm_ipvr_gem_object *obj; + + if (list_empty(list)) + return; + + list_for_each_entry(entry, list, head) { + obj = entry->ipvr_gem_bo; + /** + * do not update fence if val_args specifies so + */ + if (!entry->val_req.skip_fence) { + entry->old_fence = obj->fence; + obj->fence = ipvr_fence_ref(fence); + entry->val_req.fence_fd = fence_fd; + } + else { + IPVR_DEBUG_GENERAL("obj 0x%lx marked as non-fence\n", + ipvr_gem_obj_mmu_offset(obj)); + } + ipvr_bo_unreserve(obj); + } + + list_for_each_entry(entry, list, head) { + if (!entry->val_req.skip_fence && entry->old_fence) + ipvr_fence_unref(&entry->old_fence); + } +} + +static int32_t ipvr_usrfence_release(struct inode *inode, struct file *file) +{ + struct ipvr_fence *fence = file->private_data; + struct drm_ipvr_private *dev_priv = NULL; + int ret; + BUG_ON(!fence); + dev_priv = fence->dev_priv; + ret = kref_put(&fence->kref, &ipvr_fence_destroy); + IPVR_DEBUG_GENERAL("fence for seq %u is released by usr space\n", fence->seq); + return ret; +} + +static uint32_t ipvr_usrfence_poll(struct file *file, poll_table *wait) +{ + struct ipvr_fence *fence = file->private_data; + struct drm_ipvr_private *dev_priv = NULL; + BUG_ON(!fence); + dev_priv = fence->dev_priv; + if (ipvr_fence_signaled(dev_priv, fence->seq)) { + IPVR_DEBUG_GENERAL("seq %u is already signalled, return POLLIN immediately\n", fence->seq); + return POLLIN; + } + IPVR_DEBUG_GENERAL("seq %u is not signalled, wait for fence_queue at %lu\n", fence->seq, jiffies); + poll_wait(file, &dev_priv->fence_queue, wait); + + /* + * Make sure that reads to fence->status are ordered with the + * wait queue event triggering + */ + /* smp_rmb(); */ + IPVR_DEBUG_GENERAL("seq %u, fence_queue awaken up at %lu\n", fence->seq, jiffies); + + if (ipvr_fence_signaled(dev_priv, fence->seq)) { + IPVR_DEBUG_GENERAL("seq %u is signalled after waiting\n", fence->seq); + return POLLIN; + } + else { + IPVR_DEBUG_GENERAL("seq %u is still NOT signalled after waiting\n", fence->seq); + return 0; + } +} diff --git a/drivers/gpu/drm/ipvr/ipvr_fence.h b/drivers/gpu/drm/ipvr/ipvr_fence.h new file mode 100644 index 0000000..54846da --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_fence.h @@ -0,0 +1,68 @@ +/************************************************************************** + * ipvr_fence.h: IPVR header file for fence handling + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + +#ifndef _IPVR_FENCE_H_ +#define _IPVR_FENCE_H_ + +#include "ipvr_drv.h" + +/* seq_after(a,b) returns true if the seq a is after seq b.*/ +#define ipvr_seq_after(a,b) \ + (typecheck(uint16_t, a) && \ + typecheck(uint16_t, b) && \ + ((int16_t)(a - b) > 0)) + +#define IPVR_FENCE_JIFFIES_TIMEOUT (HZ / 2) +/* fence seq are set to this number when signaled */ +#define IPVR_FENCE_SIGNALED_SEQ 0LL + +struct ipvr_fence { + struct drm_ipvr_private *dev_priv; + struct kref kref; + /* protected by dev_priv->fence_drv.fence_lock */ + uint16_t seq; + /* fields for usrfence */ + struct file *file; + char name[32]; +}; + +int32_t ipvr_fence_wait(struct ipvr_fence *fence, bool intr, bool no_wait); + +void ipvr_fence_process(struct drm_ipvr_private *dev_priv, + uint16_t seq, uint8_t err); + +void ipvr_fence_driver_init(struct drm_ipvr_private *dev_priv); + +void ipvr_fence_driver_fini(struct drm_ipvr_private *dev_priv); + +int32_t ipvr_fence_create(struct drm_ipvr_private *dev_priv, + struct ipvr_fence **fence, int *fence_fd); + +void +ipvr_fence_buffer_objects(struct list_head *list, struct ipvr_fence *fence, int fence_fd); + +void ipvr_fence_unref(struct ipvr_fence **fence); + +#endif diff --git a/drivers/gpu/drm/ipvr/ipvr_gem.c b/drivers/gpu/drm/ipvr/ipvr_gem.c new file mode 100644 index 0000000..888e369 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_gem.c @@ -0,0 +1,248 @@ +/************************************************************************** + * ipvr_gem.c: IPVR hook file for gem ioctls + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + +#include "ipvr_gem.h" +#include "ipvr_buffer.h" +#include "ipvr_fence.h" +#include "ipvr_exec.h" +#include "ipvr_trace.h" +#include <linux/slab.h> +#include <linux/swap.h> +#include <linux/pci.h> +#include <linux/dma-buf.h> +#include <drm_gem.h> +/** + * Creates a new mm object and returns a handle to it. + */ +int32_t ipvr_gem_create_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_ipvr_gem_create *args = data; + if (args->cache_level >= IPVR_CACHE_MAX) + return -EINVAL; + return ipvr_gem_create(file_priv, dev, args->size, args->tiling, + args->cache_level, &args->rounded_size, + &args->handle, &args->gpu_offset); +} + +int32_t ipvr_gem_busy_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_ipvr_gem_busy *args = data; + struct drm_ipvr_gem_object *obj; + int32_t ret = 0; + + obj = to_ipvr_bo(drm_gem_object_lookup(dev, file_priv, args->handle)); + if (!obj || &obj->base == NULL) { + return -ENOENT; + } + IPVR_DEBUG_GENERAL("Checking bo %p (fence %p seq %u) busy status\n", + obj, obj->fence, ((obj->fence)? obj->fence->seq: 0)); + + ret = ipvr_bo_reserve(obj, true, false); + if (unlikely(ret != 0)) + goto out; + ret = ipvr_fence_wait(obj->fence, true, true); + ipvr_bo_unreserve(obj); + + args->busy = ret? 1: 0; +out: + drm_gem_object_unreference(&obj->base); + return ret; +} + +int32_t ipvr_gem_mmap_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv) +{ + struct drm_ipvr_gem_mmap *args = data; + struct drm_ipvr_gem_object *obj; + struct vm_area_struct *vma; + struct mm_struct *mm = current->mm; + uint16_t signaled_seq, seq; + struct drm_ipvr_private *dev_priv = (struct drm_ipvr_private*)dev->dev_private; + int32_t ret = 0; + IPVR_DEBUG_ENTRY("mmap is called on handle %u offset 0x%llx size %llu.\n", + args->handle, args->offset, args->size); + + obj = to_ipvr_bo(drm_gem_object_lookup(dev, file_priv, args->handle)); + if (!obj) { + return -ENOENT; + } + + /* sanity check */ + if (unlikely(obj->fence)) { + signaled_seq = atomic_read(&dev_priv->fence_drv.signaled_seq); + seq = obj->fence->seq; + if (ipvr_seq_after(seq, signaled_seq)) + IPVR_DEBUG_WARN("mmaping an obj %lx which is currently fenced!!!\n", + ipvr_gem_obj_mmu_offset(obj)); + } + + /* prime objects have no backing filp to GEM mmap + * pages from. + */ + if (unlikely(!obj->base.filp)) { + IPVR_ERROR("has no backing flip.\n"); + drm_gem_object_unreference_unlocked(&obj->base); + return -EINVAL; + } + + /* need assume mmap size is same as bo size */ + if (obj->mmap_base == 0) { + obj->mmap_size = args->size; + obj->mmap_base = vm_mmap(obj->base.filp, 0, args->size, + PROT_READ | PROT_WRITE, MAP_SHARED, + args->offset); + } + + if (IS_ERR((void *)obj->mmap_base)) { + IPVR_ERROR("addr is not correct: %d.\n", (int)obj->mmap_base); + ret = obj->mmap_base; + goto err; + } + + vma = find_vma(mm, obj->mmap_base); + if (!vma || vma->vm_start > obj->mmap_base || + obj->mmap_base + obj->mmap_size > vma->vm_end) { + IPVR_ERROR("failed to find the vma.\n"); + ret = -EFAULT; + } + + if (obj->cache_level == IPVR_CACHE_NONE) { + vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); + } else if (obj->cache_level == IPVR_CACHE_WC) { + vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot); + } + args->addr_ptr = obj->mmap_base; + IPVR_DEBUG_GENERAL("mmap got %p.\n", (void*)args->addr_ptr); + trace_ipvr_gem_mmap(obj, obj->mmap_base); +err: + drm_gem_object_unreference_unlocked(&obj->base); + return ret; +} + +int32_t ipvr_sync_cpu_grab(struct drm_device *dev, + struct drm_ipvr_gem_object *obj) +{ + int32_t ret = 0; + ret = ipvr_bo_reserve(obj, true, false); + if (unlikely(ret != 0)) + goto out; + ret = ipvr_fence_wait(obj->fence, true, false); + if (likely(ret == 0)) + atomic_inc(&obj->cpu_writers); + else + IPVR_DEBUG_WARN("Failed to call ipvr_fence_wait.\n"); + ipvr_bo_unreserve(obj); +out: + drm_gem_object_unreference(&obj->base); + return ret; +} +int32_t ipvr_sync_cpu_release(struct drm_device *dev, + struct drm_ipvr_gem_object *obj) +{ + ipvr_gem_clflush_object(obj, false); + atomic_dec(&obj->cpu_writers); + drm_gem_object_unreference(&obj->base); + return 0; +} +int32_t ipvr_sync_cpu_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv) +{ + struct drm_ipvr_gem_object *obj; + struct drm_ipvr_sync_cpu *args = data; + + IPVR_DEBUG_ENTRY("enter\n"); + obj = to_ipvr_bo(drm_gem_object_lookup(dev, file_priv, args->handle)); + if (&obj->base == NULL) + return -ENOENT; + + switch (args->op) { + case IPVR_SYNCCPU_OP_GRAB: + return ipvr_sync_cpu_grab(dev, obj); + case IPVR_SYNCCPU_OP_RELEASE: + return ipvr_sync_cpu_release(dev, obj); + default: + return -EINVAL; + } +} + +/** + * ipvr_gem_wait_ioctl - implements DRM_IOCTL_IPVR_GEM_WAIT + * @DRM_IOCTL_ARGS: standard ioctl arguments + * + * Returns 0 if successful, else an error is returned with the remaining time in + * the timeout parameter. + * -ETIME: object is still busy after timeout + * -ERESTARTSYS: signal interrupted the wait + * -ENONENT: object doesn't exist + * Also possible, but rare: + * -EAGAIN: GPU wedged + * -ENOMEM: damn + * -ENODEV: Internal IRQ fail + * -E?: The add request failed + * + * The wait ioctl with a timeout of 0 reimplements the busy ioctl. With any + * non-zero timeout parameter the wait ioctl will wait for the given number of + * nanoseconds on an object becoming unbusy. Since the wait itself does so + * without holding struct_mutex the object may become re-busied before this + * function completes. A similar but shorter * race condition exists in the busy + * ioctl + */ +int32_t ipvr_gem_wait_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv) +{ + struct drm_ipvr_gem_wait *args = data; + struct drm_ipvr_gem_object *obj; + int32_t ret = 0; + + IPVR_DEBUG_ENTRY("wait %d buffer to finish execution.\n", args->handle); + obj = to_ipvr_bo(drm_gem_object_lookup(dev, file_priv, args->handle)); + if (&obj->base == NULL) { + return -ENOENT; + } + + ret = ipvr_bo_reserve(obj, true, false); + if (unlikely(ret != 0)) + goto out; + + trace_ipvr_gem_wait_ioctl(obj); + ret = ipvr_fence_wait(obj->fence, true, false); + + ipvr_bo_unreserve(obj); + +out: + drm_gem_object_unreference(&obj->base); + return ret; +} + +int32_t ipvr_gem_userptr_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv) +{ + struct drm_ipvr_gem_userptr *args = data; + return ipvr_gem_userptr(file_priv, dev, args->user_ptr, + args->user_size, args->cache_level, args->tiling, + &args->handle, &args->gpu_offset); +} diff --git a/drivers/gpu/drm/ipvr/ipvr_gem.h b/drivers/gpu/drm/ipvr/ipvr_gem.h new file mode 100644 index 0000000..9e7b4e7 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_gem.h @@ -0,0 +1,66 @@ +/************************************************************************** + * ipvr_gem.h: IPVR header file for GEM ioctls + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * + **************************************************************************/ + +#ifndef _IPVR_GEM_H_ +#define _IPVR_GEM_H_ + +#include "ipvr_drv.h" + +/* ipvr_gem.c */ +int32_t ipvr_gem_execbuffer(struct drm_device *dev, + void *data, struct drm_file *file_priv); + +int32_t ipvr_gem_pin_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv); + +int32_t ipvr_gem_unpin_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv); + +int32_t ipvr_gem_busy_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv); + +int32_t ipvr_gem_create_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv); + +int32_t ipvr_gem_pread_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv); + +int32_t ipvr_gem_pwrite_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv); +int32_t ipvr_gem_mmap_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv); + +int32_t ipvr_gem_set_domain_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv); + +int32_t ipvr_sync_cpu_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv); + +int32_t ipvr_gem_wait_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv); + +int32_t ipvr_gem_userptr_ioctl(struct drm_device *dev, + void *data, struct drm_file *file_priv); + +#endif diff --git a/drivers/gpu/drm/ipvr/ipvr_mmu.c b/drivers/gpu/drm/ipvr/ipvr_mmu.c new file mode 100644 index 0000000..4d6f536 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_mmu.c @@ -0,0 +1,807 @@ +/************************************************************************** + * ipvr_mmu.c: IPVR MMU handling to support VED, VEC, VSP buffer access + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * Copyright (c) Imagination Technologies Limited, UK + * Copyright (c) 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * + **************************************************************************/ + +#include "ipvr_mmu.h" + +/* + * Code for the VED MMU, maybe also enabled for VEC/VSP + */ + +/* + * clflush on one processor only: + * clflush should apparently flush the cache line on all processors in an + * SMP system. + */ + +/* + * kmap atomic: + * Usage of the slots must be completely encapsulated within a spinlock, and + * no other functions that may be using the locks for other purposed may be + * called from within the locked region. + * Since the slots are per processor, this will guarantee that we are the only + * user. + */ + +static inline uint32_t ipvr_mmu_pt_index(uint32_t offset) +{ + return (offset >> IPVR_PTE_SHIFT) & 0x3FF; +} + +static inline uint32_t ipvr_mmu_pd_index(uint32_t offset) +{ + return offset >> IPVR_PDE_SHIFT; +} + +#if defined(CONFIG_X86) +static inline void ipvr_clflush(void *addr) +{ + __asm__ __volatile__("clflush (%0)\n" : : "r"(addr) : "memory"); +} + +static inline void +ipvr_mmu_clflush(struct ipvr_mmu_driver *driver, void *addr) +{ + if (!driver->has_clflush) + return; + + mb(); + ipvr_clflush(addr); + mb(); +} + +static void +ipvr_mmu_page_clflush(struct ipvr_mmu_driver *driver, struct page* page) +{ + uint32_t clflush_add = driver->clflush_add >> PAGE_SHIFT; + uint32_t clflush_count = PAGE_SIZE / clflush_add; + int i; + uint8_t *clf; + + clf = kmap_atomic(page); + + mb(); + for (i = 0; i < clflush_count; ++i) { + ipvr_clflush(clf); + clf += clflush_add; + } + mb(); + + kunmap_atomic(clf); +} + +static void ipvr_mmu_pages_clflush(struct ipvr_mmu_driver *driver, + struct page *page[], unsigned long num_pages) +{ + int i; + + if (!driver->has_clflush) + return ; + + for (i = 0; i < num_pages; i++) + ipvr_mmu_page_clflush(driver, *page++); +} +#else + +static inline void +ipvr_mmu_clflush(struct ipvr_mmu_driver *driver, void *addr) +{ + ; +} + +static void ipvr_mmu_pages_clflush(struct ipvr_mmu_driver *driver, + struct page *page[], unsigned long num_pages) +{ + printk("Dumy ipvr_mmu_pages_clflush\n"); +} + +#endif + +static void +ipvr_mmu_flush_pd_locked(struct ipvr_mmu_driver *driver, int32_t force) +{ + if (atomic_read(&driver->needs_tlbflush) || force) { + if (!driver->dev_priv) + goto out; + + atomic_set( + &driver->dev_priv->ipvr_mmu_invaldc, + 1); + + } +out: + atomic_set(&driver->needs_tlbflush, 0); +} + +void ipvr_mmu_flush(struct ipvr_mmu_driver *driver, int32_t rc_prot) +{ + if (rc_prot) + down_write(&driver->sem); + + if (!driver->dev_priv) + goto out; + + atomic_set(&driver->dev_priv->ipvr_mmu_invaldc, 1); + +out: + if (rc_prot) + up_write(&driver->sem); +} + +void ipvr_mmu_set_pd_context(struct ipvr_mmu_pd *pd, int32_t hw_context) +{ + /*ttm_tt_cache_flush(&pd->p, 1);*/ + ipvr_mmu_pages_clflush(pd->driver, &pd->p, 1); + down_write(&pd->driver->sem); + wmb(); + ipvr_mmu_flush_pd_locked(pd->driver, 1); + pd->hw_context = hw_context; + up_write(&pd->driver->sem); + +} + +static inline unsigned long +ipvr_pd_addr_end(unsigned long addr, unsigned long end) +{ + + addr = (addr + IPVR_PDE_MASK + 1) & ~IPVR_PDE_MASK; + return (addr < end) ? addr : end; +} + +static inline uint32_t ipvr_mmu_mask_pte(uint32_t pfn, int32_t type) +{ + uint32_t mask = IPVR_PTE_VALID; + + if (type & IPVR_MMU_CACHED_MEMORY) + mask |= IPVR_PTE_CACHED; + if (type & IPVR_MMU_RO_MEMORY) + mask |= IPVR_PTE_RO; + if (type & IPVR_MMU_WO_MEMORY) + mask |= IPVR_PTE_WO; + + return (pfn << PAGE_SHIFT) | mask; +} + +struct ipvr_mmu_pd *ipvr_mmu_alloc_pd(struct ipvr_mmu_driver *driver, + int32_t trap_pagefaults, int32_t invalid_type) +{ + struct ipvr_mmu_pd *pd = kmalloc(sizeof(*pd), GFP_KERNEL); + uint32_t *v; + int32_t i; + + if (!pd) + return NULL; + + pd->p = alloc_page(GFP_DMA32); + if (!pd->p) + goto out_err1; + pd->dummy_pt = alloc_page(GFP_DMA32); + if (!pd->dummy_pt) + goto out_err2; + pd->dummy_page = alloc_page(GFP_DMA32); + if (!pd->dummy_page) + goto out_err3; + + if (!trap_pagefaults) { + pd->invalid_pde = + ipvr_mmu_mask_pte(page_to_pfn(pd->dummy_pt), + invalid_type); + pd->invalid_pte = + ipvr_mmu_mask_pte(page_to_pfn(pd->dummy_page), + invalid_type); + } else { + pd->invalid_pde = 0; + pd->invalid_pte = 0; + } + + v = kmap(pd->dummy_pt); + if (!v) + goto out_err4; + for (i = 0; i < (PAGE_SIZE / sizeof(uint32_t)); ++i) + v[i] = pd->invalid_pte; + + kunmap(pd->dummy_pt); + + v = kmap(pd->p); + if (!v) + goto out_err4; + for (i = 0; i < (PAGE_SIZE / sizeof(uint32_t)); ++i) + v[i] = pd->invalid_pde; + + kunmap(pd->p); + + v = kmap(pd->dummy_page); + if (!v) + goto out_err4; + clear_page(v); + kunmap(pd->dummy_page); + + pd->tables = vmalloc_user(sizeof(struct ipvr_mmu_pt *) * 1024); + if (!pd->tables) + goto out_err4; + + pd->hw_context = -1; + pd->pd_mask = IPVR_PTE_VALID; + pd->driver = driver; + + return pd; + +out_err4: + __free_page(pd->dummy_page); +out_err3: + __free_page(pd->dummy_pt); +out_err2: + __free_page(pd->p); +out_err1: + kfree(pd); + return NULL; +} + +void ipvr_mmu_free_pt(struct ipvr_mmu_pt *pt) +{ + __free_page(pt->p); + kfree(pt); +} + +void ipvr_mmu_free_pagedir(struct ipvr_mmu_pd *pd) +{ + struct ipvr_mmu_driver *driver = pd->driver; + struct ipvr_mmu_pt *pt; + int32_t i; + + down_write(&driver->sem); + if (pd->hw_context != -1) + ipvr_mmu_flush_pd_locked(driver, 1); + + /* Should take the spinlock here, but we don't need to do that + since we have the semaphore in write mode. */ + + for (i = 0; i < 1024; ++i) { + pt = pd->tables[i]; + if (pt) + ipvr_mmu_free_pt(pt); + } + + vfree(pd->tables); + __free_page(pd->dummy_page); + __free_page(pd->dummy_pt); + __free_page(pd->p); + kfree(pd); + up_write(&driver->sem); +} + +static struct ipvr_mmu_pt *ipvr_mmu_alloc_pt(struct ipvr_mmu_pd *pd) +{ + struct ipvr_mmu_pt *pt = kmalloc(sizeof(*pt), GFP_KERNEL); + void *v; + uint32_t clflush_add = pd->driver->clflush_add >> PAGE_SHIFT; + uint32_t clflush_count = PAGE_SIZE / clflush_add; + spinlock_t *lock = &pd->driver->lock; + uint8_t *clf; + uint32_t *ptes; + int32_t i; + + if (!pt) + return NULL; + + pt->p = alloc_page(GFP_DMA32); + if (!pt->p) { + kfree(pt); + return NULL; + } + + spin_lock(lock); + + v = kmap_atomic(pt->p); + + clf = (uint8_t *) v; + ptes = (uint32_t *) v; + for (i = 0; i < (PAGE_SIZE / sizeof(uint32_t)); ++i) + *ptes++ = pd->invalid_pte; + + +#if defined(CONFIG_X86) + if (pd->driver->has_clflush && pd->hw_context != -1) { + mb(); + for (i = 0; i < clflush_count; ++i) { + ipvr_clflush(clf); + clf += clflush_add; + } + mb(); + } +#endif + kunmap_atomic(v); + + spin_unlock(lock); + + pt->count = 0; + pt->pd = pd; + pt->index = 0; + + return pt; +} + +struct ipvr_mmu_pt * +ipvr_mmu_pt_alloc_map_lock(struct ipvr_mmu_pd *pd, unsigned long addr) +{ + uint32_t index = ipvr_mmu_pd_index(addr); + struct ipvr_mmu_pt *pt; + uint32_t *v; + spinlock_t *lock = &pd->driver->lock; + + spin_lock(lock); + pt = pd->tables[index]; + while (!pt) { + spin_unlock(lock); + pt = ipvr_mmu_alloc_pt(pd); + if (!pt) + return NULL; + spin_lock(lock); + + if (pd->tables[index]) { + spin_unlock(lock); + ipvr_mmu_free_pt(pt); + spin_lock(lock); + pt = pd->tables[index]; + continue; + } + + v = kmap_atomic(pd->p); + + pd->tables[index] = pt; + v[index] = (page_to_pfn(pt->p) << 12) | + pd->pd_mask; + + + pt->index = index; + + kunmap_atomic((void *) v); + + if (pd->hw_context != -1) { + ipvr_mmu_clflush(pd->driver, (void *) &v[index]); + atomic_set(&pd->driver->needs_tlbflush, 1); + } + } + + pt->v = kmap_atomic(pt->p); + + return pt; +} + +static struct ipvr_mmu_pt * +ipvr_mmu_pt_map_lock(struct ipvr_mmu_pd *pd, unsigned long addr) +{ + uint32_t index = ipvr_mmu_pd_index(addr); + struct ipvr_mmu_pt *pt; + spinlock_t *lock = &pd->driver->lock; + + spin_lock(lock); + pt = pd->tables[index]; + if (!pt) { + spin_unlock(lock); + return NULL; + } + + pt->v = kmap_atomic(pt->p); + + return pt; +} + +static void ipvr_mmu_pt_unmap_unlock(struct ipvr_mmu_pt *pt) +{ + struct ipvr_mmu_pd *pd = pt->pd; + uint32_t *v; + + kunmap_atomic(pt->v); + + if (pt->count == 0) { + v = kmap_atomic(pd->p); + + v[pt->index] = pd->invalid_pde; + pd->tables[pt->index] = NULL; + + if (pd->hw_context != -1) { + ipvr_mmu_clflush(pd->driver, + (void *) &v[pt->index]); + atomic_set(&pd->driver->needs_tlbflush, 1); + } + + kunmap_atomic(pt->v); + + spin_unlock(&pd->driver->lock); + ipvr_mmu_free_pt(pt); + return; + } + spin_unlock(&pd->driver->lock); +} + +static inline void +ipvr_mmu_set_pte(struct ipvr_mmu_pt *pt, unsigned long addr, uint32_t pte) +{ + pt->v[ipvr_mmu_pt_index(addr)] = pte; +} + +static inline void +ipvr_mmu_invalidate_pte(struct ipvr_mmu_pt *pt, unsigned long addr) +{ + pt->v[ipvr_mmu_pt_index(addr)] = pt->pd->invalid_pte; +} + +struct ipvr_mmu_pd *ipvr_mmu_get_default_pd(struct ipvr_mmu_driver *driver) +{ + struct ipvr_mmu_pd *pd; + + /* down_read(&driver->sem); */ + pd = driver->default_pd; + /* up_read(&driver->sem); */ + + return pd; +} + +/* Returns the physical address of the PD shared by sgx/msvdx */ +uint32_t ipvr_get_default_pd_addr(struct ipvr_mmu_driver *driver) +{ + struct ipvr_mmu_pd *pd; + + pd = ipvr_mmu_get_default_pd(driver); + return page_to_pfn(pd->p) << PAGE_SHIFT; +} + +void ipvr_mmu_driver_takedown(struct ipvr_mmu_driver *driver) +{ + ipvr_mmu_free_pagedir(driver->default_pd); + kfree(driver); +} + +struct ipvr_mmu_driver * +ipvr_mmu_driver_init(uint8_t __iomem * registers, int32_t trap_pagefaults, + int32_t invalid_type, struct drm_ipvr_private *dev_priv) +{ + struct ipvr_mmu_driver *driver; + + driver = kmalloc(sizeof(*driver), GFP_KERNEL); + if (!driver) + return NULL; + + driver->dev_priv = dev_priv; + + driver->default_pd = + ipvr_mmu_alloc_pd(driver, trap_pagefaults, invalid_type); + if (!driver->default_pd) + goto out_err1; + + spin_lock_init(&driver->lock); + init_rwsem(&driver->sem); + down_write(&driver->sem); + driver->register_map = registers; + atomic_set(&driver->needs_tlbflush, 1); + + driver->has_clflush = 0; + +#if defined(CONFIG_X86) + if (cpu_has_clflush) { + uint32_t tfms, misc, cap0, cap4, clflush_size; + + /* + * clflush size is determined at kernel setup for x86_64 + * but not for i386. We have to do it here. + */ + + cpuid(0x00000001, &tfms, &misc, &cap0, &cap4); + clflush_size = ((misc >> 8) & 0xff) * 8; + driver->has_clflush = 1; + driver->clflush_add = + PAGE_SIZE * clflush_size / sizeof(uint32_t); + driver->clflush_mask = driver->clflush_add - 1; + driver->clflush_mask = ~driver->clflush_mask; + } +#endif + + up_write(&driver->sem); + return driver; + +out_err1: + kfree(driver); + return NULL; +} + +#if defined(CONFIG_X86) +static void ipvr_mmu_flush_ptes(struct ipvr_mmu_pd *pd, + unsigned long address, + uint32_t num_pages, + uint32_t desired_tile_stride, + uint32_t hw_tile_stride) +{ + struct ipvr_mmu_pt *pt; + uint32_t rows = 1; + uint32_t i; + unsigned long addr; + unsigned long end; + unsigned long next; + unsigned long add; + unsigned long row_add; + unsigned long clflush_add = pd->driver->clflush_add; + unsigned long clflush_mask = pd->driver->clflush_mask; + IPVR_DEBUG_GENERAL("call x86 ipvr_mmu_flush_ptes, address is 0x%lx, " + "num pages is %d.\n", address, num_pages); + if (!pd->driver->has_clflush) { + IPVR_DEBUG_GENERAL("call ipvr_mmu_pages_clflush.\n"); + ipvr_mmu_pages_clflush(pd->driver, &pd->p, num_pages); + return; + } + + if (hw_tile_stride) + rows = num_pages / desired_tile_stride; + else + desired_tile_stride = num_pages; + + add = desired_tile_stride << PAGE_SHIFT; + row_add = hw_tile_stride << PAGE_SHIFT; + mb(); + for (i = 0; i < rows; ++i) { + addr = address; + end = addr + add; + + do { + next = ipvr_pd_addr_end(addr, end); + pt = ipvr_mmu_pt_map_lock(pd, addr); + if (!pt) + continue; + do { + ipvr_clflush(&pt->v + [ipvr_mmu_pt_index(addr)]); + } while (addr += + clflush_add, + (addr & clflush_mask) < next); + + ipvr_mmu_pt_unmap_unlock(pt); + } while (addr = next, next != end); + address += row_add; + } + mb(); +} +#else + +static void ipvr_mmu_flush_ptes(struct ipvr_mmu_pd *pd, + unsigned long address, + uint32_t num_pages, + uint32_t desired_tile_stride, + uint32_t hw_tile_stride) +{ + IPVR_DEBUG_GENERAL("call non-x86 ipvr_mmu_flush_ptes.\n"); +} +#endif + +void ipvr_mmu_remove_pfn_sequence(struct ipvr_mmu_pd *pd, + unsigned long address, + uint32_t num_pages) +{ + struct ipvr_mmu_pt *pt; + unsigned long addr; + unsigned long end; + unsigned long next; + unsigned long f_address = address; + + down_read(&pd->driver->sem); + + addr = address; + end = addr + (num_pages << PAGE_SHIFT); + + do { + next = ipvr_pd_addr_end(addr, end); + pt = ipvr_mmu_pt_alloc_map_lock(pd, addr); + if (!pt) + goto out; + do { + ipvr_mmu_invalidate_pte(pt, addr); + --pt->count; + } while (addr += PAGE_SIZE, addr < next); + ipvr_mmu_pt_unmap_unlock(pt); + + } while (addr = next, next != end); + +out: + if (pd->hw_context != -1) + ipvr_mmu_flush_ptes(pd, f_address, num_pages, 1, 1); + + up_read(&pd->driver->sem); + + if (pd->hw_context != -1) + ipvr_mmu_flush(pd->driver, 0); + + return; +} + +void ipvr_mmu_remove_pages(struct ipvr_mmu_pd *pd, unsigned long address, + uint32_t num_pages, uint32_t desired_tile_stride, + uint32_t hw_tile_stride) +{ + struct ipvr_mmu_pt *pt; + uint32_t rows = 1; + uint32_t i; + unsigned long addr; + unsigned long end; + unsigned long next; + unsigned long add; + unsigned long row_add; + unsigned long f_address = address; + + if (hw_tile_stride) + rows = num_pages / desired_tile_stride; + else + desired_tile_stride = num_pages; + + add = desired_tile_stride << PAGE_SHIFT; + row_add = hw_tile_stride << PAGE_SHIFT; + + /* down_read(&pd->driver->sem); */ + + /* Make sure we only need to flush this processor's cache */ + + for (i = 0; i < rows; ++i) { + + addr = address; + end = addr + add; + + do { + next = ipvr_pd_addr_end(addr, end); + pt = ipvr_mmu_pt_map_lock(pd, addr); + if (!pt) + continue; + do { + ipvr_mmu_invalidate_pte(pt, addr); + --pt->count; + + } while (addr += PAGE_SIZE, addr < next); + ipvr_mmu_pt_unmap_unlock(pt); + + } while (addr = next, next != end); + address += row_add; + } + if (pd->hw_context != -1) + ipvr_mmu_flush_ptes(pd, f_address, num_pages, + desired_tile_stride, hw_tile_stride); + + /* up_read(&pd->driver->sem); */ + + if (pd->hw_context != -1) + ipvr_mmu_flush(pd->driver, 0); +} + +int32_t ipvr_mmu_insert_pfn_sequence(struct ipvr_mmu_pd *pd, + uint32_t start_pfn, + unsigned long address, + uint32_t num_pages, + int32_t type) +{ + struct ipvr_mmu_pt *pt; + uint32_t pte; + unsigned long addr; + unsigned long end; + unsigned long next; + unsigned long f_address = address; + int32_t ret = 0; + + down_read(&pd->driver->sem); + + addr = address; + end = addr + (num_pages << PAGE_SHIFT); + + do { + next = ipvr_pd_addr_end(addr, end); + pt = ipvr_mmu_pt_alloc_map_lock(pd, addr); + if (!pt) { + ret = -ENOMEM; + goto out; + } + do { + pte = ipvr_mmu_mask_pte(start_pfn++, type); + + ipvr_mmu_set_pte(pt, addr, pte); + pt->count++; + } while (addr += PAGE_SIZE, addr < next); + ipvr_mmu_pt_unmap_unlock(pt); + + } while (addr = next, next != end); + +out: + if (pd->hw_context != -1) + ipvr_mmu_flush_ptes(pd, f_address, num_pages, 1, 1); + + up_read(&pd->driver->sem); + + if (pd->hw_context != -1) + ipvr_mmu_flush(pd->driver, 1); + + return ret; +} + +int32_t ipvr_mmu_insert_pages(struct ipvr_mmu_pd *pd, struct page **pages, + unsigned long address, uint32_t num_pages, + uint32_t desired_tile_stride, + uint32_t hw_tile_stride, int32_t type) +{ + struct ipvr_mmu_pt *pt; + uint32_t rows = 1; + uint32_t i; + uint32_t pte; + unsigned long addr; + unsigned long end; + unsigned long next; + unsigned long add; + unsigned long row_add; + unsigned long f_address = address; + int32_t ret = 0; + + if (hw_tile_stride) { + if (num_pages % desired_tile_stride != 0) + return -EINVAL; + rows = num_pages / desired_tile_stride; + } else { + desired_tile_stride = num_pages; + } + + add = desired_tile_stride << PAGE_SHIFT; + row_add = hw_tile_stride << PAGE_SHIFT; + + down_read(&pd->driver->sem); + + for (i = 0; i < rows; ++i) { + + addr = address; + end = addr + add; + + do { + next = ipvr_pd_addr_end(addr, end); + pt = ipvr_mmu_pt_alloc_map_lock(pd, addr); + if (!pt) { + ret = -ENOMEM; + goto out; + } + do { + pte = ipvr_mmu_mask_pte( + page_to_pfn(*pages++), + type); + ipvr_mmu_set_pte(pt, addr, pte); + pt->count++; + } while (addr += PAGE_SIZE, addr < next); + ipvr_mmu_pt_unmap_unlock(pt); + + } while (addr = next, next != end); + + address += row_add; + } +out: + if (pd->hw_context != -1) + ipvr_mmu_flush_ptes(pd, f_address, num_pages, + desired_tile_stride, hw_tile_stride); + + up_read(&pd->driver->sem); + + if (pd->hw_context != -1) + ipvr_mmu_flush(pd->driver, 1); + + return ret; +} diff --git a/drivers/gpu/drm/ipvr/ipvr_mmu.h b/drivers/gpu/drm/ipvr/ipvr_mmu.h new file mode 100644 index 0000000..9ba4b69 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_mmu.h @@ -0,0 +1,135 @@ +/************************************************************************** + * ipvr_mmu.h: IPVR header file for VED/VEC/VSP MMU handling + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * Copyright (c) Imagination Technologies Limited, UK + * Copyright (c) 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Eric Anholt eric@anholt.net + * + **************************************************************************/ + +#ifndef _IPVR_MMU_H_ +#define _IPVR_MMU_H_ + +#include "ipvr_drv.h" + +static inline bool __must_check IPVR_IS_ERR(__force const unsigned long offset) +{ + return unlikely((offset) >= (unsigned long)-MAX_ERRNO); +} + +static inline int32_t __must_check IPVR_OFFSET_ERR(__force const unsigned long offset) +{ + return (int32_t)offset; +} + +static inline unsigned long __must_check IPVR_ERR_OFFSET(__force const int32_t err) +{ + return (unsigned long)err; +} + +struct ipvr_mmu_pd; + +struct ipvr_mmu_pt { + struct ipvr_mmu_pd *pd; + uint32_t index; + uint32_t count; + struct page *p; + uint32_t *v; +}; + +struct ipvr_mmu_driver { + /* protects driver- and pd structures. Always take in read mode + * before taking the page table spinlock. + */ + struct rw_semaphore sem; + + /* protects page tables, directory tables and pt tables. + * and pt structures. + */ + spinlock_t lock; + + atomic_t needs_tlbflush; + + uint8_t __iomem *register_map; + struct ipvr_mmu_pd *default_pd; + + int32_t has_clflush; + int32_t clflush_add; + unsigned long clflush_mask; + + struct drm_ipvr_private *dev_priv; +}; + +struct ipvr_mmu_pd { + struct ipvr_mmu_driver *driver; + int32_t hw_context; + struct ipvr_mmu_pt **tables; + struct page *p; + struct page *dummy_pt; + struct page *dummy_page; + uint32_t pd_mask; + uint32_t invalid_pde; + uint32_t invalid_pte; +}; + +struct ipvr_mmu_driver *ipvr_mmu_driver_init(uint8_t __iomem *registers, + int32_t trap_pagefaults, + int32_t invalid_type, + struct drm_ipvr_private *dev_priv); + +void ipvr_mmu_driver_takedown(struct ipvr_mmu_driver *driver); + +struct ipvr_mmu_pd * +ipvr_mmu_get_default_pd(struct ipvr_mmu_driver *driver); + +struct ipvr_mmu_pd *ipvr_mmu_alloc_pd(struct ipvr_mmu_driver *driver, + int32_t trap_pagefaults, + int32_t invalid_type); + +void ipvr_mmu_free_pagedir(struct ipvr_mmu_pd *pd); + +void ipvr_mmu_flush(struct ipvr_mmu_driver *driver, int rc_prot); + +void ipvr_mmu_remove_pfn_sequence(struct ipvr_mmu_pd *pd, + unsigned long address, + uint32_t num_pages); + +int32_t ipvr_mmu_insert_pfn_sequence(struct ipvr_mmu_pd *pd, + uint32_t start_pfn, + unsigned long address, + uint32_t num_pages, + int32_t type); + +void ipvr_mmu_set_pd_context(struct ipvr_mmu_pd *pd, int32_t hw_context); + +int32_t ipvr_mmu_insert_pages(struct ipvr_mmu_pd *pd, struct page **pages, + unsigned long address, uint32_t num_pages, + uint32_t desired_tile_stride, + uint32_t hw_tile_stride, int32_t type); + +void ipvr_mmu_remove_pages(struct ipvr_mmu_pd *pd, + unsigned long address, uint32_t num_pages, + uint32_t desired_tile_stride, + uint32_t hw_tile_stride); + +uint32_t ipvr_get_default_pd_addr(struct ipvr_mmu_driver *driver); + +#endif diff --git a/drivers/gpu/drm/ipvr/ipvr_trace.c b/drivers/gpu/drm/ipvr/ipvr_trace.c new file mode 100644 index 0000000..91c0bda --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_trace.c @@ -0,0 +1,11 @@ +/* + * Copyright © 2014 Intel Corporation + * + * Authors: + * Yao Cheng yao.cheng@intel.com> + */ + +#ifndef __CHECKER__ +#define CREATE_TRACE_POINTS +#include "ipvr_trace.h" +#endif diff --git a/drivers/gpu/drm/ipvr/ipvr_trace.h b/drivers/gpu/drm/ipvr/ipvr_trace.h new file mode 100644 index 0000000..d3209df --- /dev/null +++ b/drivers/gpu/drm/ipvr/ipvr_trace.h @@ -0,0 +1,296 @@ +/************************************************************************** + * ipvr_trace.h: IPVR header file for trace support + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + +#if !defined(_IPVR_TRACE_H_) || defined(TRACE_HEADER_MULTI_READ) +#define _IPVR_TRACE_H_ + +#include <linux/stringify.h> +#include <linux/types.h> +#include <linux/tracepoint.h> + +#include <drm/drmP.h> +#include "ipvr_buffer.h" +#include "ipvr_fence.h" +#include "ved_msg.h" + +#undef TRACE_SYSTEM +#define TRACE_SYSTEM ipvr +#define TRACE_SYSTEM_STRING __stringify(TRACE_SYSTEM) +#define TRACE_INCLUDE_FILE ipvr_trace + +/* object tracking */ + +TRACE_EVENT(ipvr_gem_create, + TP_PROTO(struct drm_ipvr_gem_object *obj, uint64_t gpu_offset), + TP_ARGS(obj, gpu_offset), + TP_STRUCT__entry( + __field(struct drm_ipvr_gem_object *, obj) + __field(uint32_t, size) + __field(bool, tiling) + __field(uint32_t, cache_level) + __field(uint64_t, gpu_offset) + ), + TP_fast_assign( + __entry->obj = obj; + __entry->size = obj->base.size; + __entry->tiling = obj->tiling; + __entry->cache_level = obj->cache_level; + __entry->gpu_offset = gpu_offset; + ), + TP_printk("obj=0x%p, size=%u, tiling=%u, cache=%u, gpu_offset=0x%llx", + __entry->obj, __entry->size, __entry->tiling, + __entry->cache_level, __entry->gpu_offset) +); + +TRACE_EVENT(ipvr__gem_free_object, + TP_PROTO(struct drm_ipvr_gem_object *obj), + TP_ARGS(obj), + TP_STRUCT__entry( + __field(struct drm_ipvr_gem_object *, obj) + ), + TP_fast_assign( + __entry->obj = obj; + ), + TP_printk("obj=0x%p", __entry->obj) +); + +TRACE_EVENT(ipvr_gem_wait_ioctl, + TP_PROTO(struct drm_ipvr_gem_object *obj), + TP_ARGS(obj), + TP_STRUCT__entry( + __field(struct drm_ipvr_gem_object *, obj) + __field(struct ipvr_fence *, fence) + ), + TP_fast_assign( + __entry->obj = obj; + __entry->fence = obj->fence; + ), + TP_printk("obj=%p, fence=%p ", __entry->obj, __entry->fence) +); + +TRACE_EVENT(ipvr_fence_wait, + TP_PROTO(struct ipvr_fence *fence, + uint32_t signaled_seq, + uint16_t sync_seq), + TP_ARGS(fence, signaled_seq, sync_seq), + TP_STRUCT__entry( + __field(struct ipvr_fence *, fence) + __field(uint16_t, fence_seq) + __field(uint32_t, signaled_seq) + __field(uint16_t, sync_seq) + ), + TP_fast_assign( + __entry->fence = fence; + __entry->fence_seq = fence->seq; + __entry->signaled_seq = signaled_seq; + __entry->sync_seq = sync_seq; + ), + TP_printk("fence=%p, fence_seq=%d, signaled_seq=%d, sync_seq=%d", + __entry->fence, __entry->fence_seq, + __entry->signaled_seq, __entry->sync_seq) +); + +TRACE_EVENT(ipvr_gem_mmap, + TP_PROTO(struct drm_ipvr_gem_object *obj, unsigned long mmap_base), + TP_ARGS(obj, mmap_base), + TP_STRUCT__entry( + __field(struct drm_ipvr_gem_object *, obj) + __field(unsigned long, mmap_base) + ), + TP_fast_assign( + __entry->obj = obj; + __entry->mmap_base = mmap_base; + ), + TP_printk("obj=%p, mmap_base=0x%lx", __entry->obj, __entry->mmap_base) +); + + +TRACE_EVENT(ipvr_gem_exec_ioctl, + TP_PROTO(struct drm_ipvr_gem_execbuffer *exec), + TP_ARGS(exec), + TP_STRUCT__entry( + __field(uint64_t, buffer_list) + __field(uint32_t, buffer_count) + __field(uint32_t, cmdbuf_handle) + __field(uint32_t, cmdbuf_size) + __field(uint32_t, ctx_id) + ), + TP_fast_assign( + __entry->buffer_list = exec->buffer_list; + __entry->buffer_count = exec->buffer_count; + __entry->cmdbuf_handle = exec->cmdbuf_handle; + __entry->cmdbuf_size = exec->cmdbuf_size; + __entry->ctx_id = exec->ctx_id; + ), + TP_printk("buffer_list=0x%llx, buffer_count=0x%d, " + "cmdbuf_handle=0x%x, cmdbuf_size=%u, ctx_id=%d", + __entry->buffer_list, __entry->buffer_count, + __entry->cmdbuf_handle, __entry->cmdbuf_size, + __entry->ctx_id) +); + +TRACE_EVENT(ved_cmd_send, + TP_PROTO(uint32_t cmd_id, uint32_t seq), + TP_ARGS(cmd_id, seq), + TP_STRUCT__entry( + __field(uint32_t, cmd_id) + __field(uint32_t, seq) + ), + TP_fast_assign( + __entry->cmd_id = cmd_id; + __entry->seq = seq; + ), + TP_printk("cmd_id=0x%08x, seq=0x%08x", + __entry->cmd_id, __entry->seq) +); + +TRACE_EVENT(ved_power_on, + TP_PROTO(int freq), + TP_ARGS(freq), + TP_STRUCT__entry( + __field(int, freq) + ), + TP_fast_assign( + __entry->freq = freq; + ), + TP_printk("frequency %d MHz", __entry->freq) +); + +TRACE_EVENT(ved_power_off, + TP_PROTO(int freq), + TP_ARGS(freq), + TP_STRUCT__entry( + __field(int, freq) + ), + TP_fast_assign( + __entry->freq = freq; + ), + TP_printk("frequency %d MHz", __entry->freq) +); + +TRACE_EVENT(ved_irq_completed, + TP_PROTO(struct fw_completed_msg *completed_msg), + TP_ARGS(completed_msg), + TP_STRUCT__entry( + __field(uint16_t, seqno) + __field(uint32_t, flags) + __field(uint32_t, vdebcr) + __field(uint16_t, start_mb) + __field(uint16_t, last_mb) + ), + TP_fast_assign( + __entry->seqno = completed_msg->header.bits.msg_fence; + __entry->flags = completed_msg->flags; + __entry->vdebcr = completed_msg->vdebcr; + __entry->start_mb = completed_msg->mb.bits.start_mb; + __entry->last_mb = completed_msg->mb.bits.last_mb; + ), + TP_printk("seq=0x%04x, flags=0x%08x, vdebcr=0x%08x, mb=[%u, %u]", + __entry->seqno, + __entry->flags, + __entry->vdebcr, + __entry->start_mb, + __entry->last_mb) +); + +TRACE_EVENT(ved_irq_panic, + TP_PROTO(struct fw_panic_msg *panic_msg, uint32_t err_trig, + uint32_t irq_status, uint32_t mmu_status, uint32_t dmac_status), + TP_ARGS(panic_msg, err_trig, irq_status, mmu_status, dmac_status), + TP_STRUCT__entry( + __field(uint16_t, seqno) + __field(uint32_t, fe_status) + __field(uint32_t, be_status) + __field(uint16_t, rsvd) + __field(uint16_t, last_mb) + __field(uint32_t, err_trig) + __field(uint32_t, irq_status) + __field(uint32_t, mmu_status) + __field(uint32_t, dmac_status) + + ), + TP_fast_assign( + __entry->seqno = panic_msg->header.bits.msg_fence; + __entry->fe_status = panic_msg->fe_status; + __entry->be_status = panic_msg->be_status; + __entry->rsvd = panic_msg->mb.bits.reserved2; + __entry->last_mb = panic_msg->mb.bits.last_mb; + __entry->err_trig = err_trig; + __entry->irq_status = irq_status; + __entry->mmu_status = mmu_status; + __entry->dmac_status = dmac_status; + ), + TP_printk("seq=0x%04x, status=[fe 0x%08x be 0x%08x], rsvd=0x%04x, " + "last_mb=%u, err_trig=0x%08x, irq_status=0x%08x, " + "mmu_status=0x%08x, dmac_status=0x%08x", + __entry->seqno, + __entry->fe_status, + __entry->be_status, + __entry->rsvd, + __entry->last_mb, + __entry->err_trig, + __entry->irq_status, + __entry->mmu_status, + __entry->dmac_status) +); + +TRACE_EVENT(ved_irq_contiguity, + TP_PROTO(struct fw_contiguity_msg *msg), + TP_ARGS(msg), + TP_STRUCT__entry( + __field(uint16_t, seqno) + __field(uint16_t, begin_mb) + __field(uint16_t, end_mb) + ), + TP_fast_assign( + __entry->seqno = msg->header.bits.msg_fence; + __entry->begin_mb = msg->mb.bits.begin_mb_num; + __entry->end_mb = msg->mb.bits.end_mb_num; + ), + TP_printk("seq=0x%04x, mb=[%u, %u]", + __entry->seqno, + __entry->begin_mb, + __entry->end_mb) +); + +TRACE_EVENT(ved_irq_deblock_required, + TP_PROTO(struct fw_deblock_required_msg *msg), + TP_ARGS(msg), + TP_STRUCT__entry( + __field(uint16_t, seqno) + ), + TP_fast_assign( + __entry->seqno = msg->header.bits.msg_fence; + ), + TP_printk("seq=0x%04x", + __entry->seqno) +); + +#endif /* _IPVR_TRACE_H_ */ + + /* This part must be outside protection */ +#undef TRACE_INCLUDE_PATH +#define TRACE_INCLUDE_PATH . +#include <trace/define_trace.h> diff --git a/drivers/gpu/drm/ipvr/ved_cmd.c b/drivers/gpu/drm/ipvr/ved_cmd.c new file mode 100644 index 0000000..0ae46e5 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ved_cmd.c @@ -0,0 +1,1269 @@ +/************************************************************************** + * ved_cmd.c: VED command handling between host driver and VED firmware + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * Copyright (c) Imagination Technologies Limited, UK + * Copyright (c) 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + +#include "ipvr_gem.h" +#include "ipvr_mmu.h" +#include "ipvr_buffer.h" +#include "ipvr_trace.h" +#include "ipvr_fence.h" +#include "ved_cmd.h" +#include "ved_fw.h" +#include "ved_msg.h" +#include "ved_reg.h" +#include "ved_ec.h" +#include "ved_init.h" +#include "ved_pm.h" +#include <linux/pm_runtime.h> + +#include <linux/io.h> +#include <linux/delay.h> + +#ifndef list_first_entry +#define list_first_entry(ptr, type, member) \ + list_entry((ptr)->next, type, member) +#endif + +int32_t ved_mtx_send(struct drm_ipvr_private *dev_priv, const void *msg) +{ + static struct fw_padding_msg pad_msg; + const uint32_t *p_msg = (uint32_t *)msg; + uint32_t msg_num, words_free, ridx, widx, buf_size, buf_offset; + int32_t ret = 0; + int i; + union msg_header *header; + header = (union msg_header *)msg; + + IPVR_DEBUG_ENTRY("enter.\n"); + + /* we need clocks enabled before we touch VEC local ram, + * but fw will take care of the clock after fw is loaded + */ + + msg_num = (header->bits.msg_size + 3) / 4; + + /* debug code for msg dump */ + IPVR_DEBUG_VED("MSVDX: ved_mtx_send is %dDW\n", msg_num); + + for (i = 0; i < msg_num; i++) + IPVR_DEBUG_VED(" 0x%08x\n", p_msg[i]); + + buf_size = VED_REG_READ32(MSVDX_COMMS_TO_MTX_BUF_SIZE) & + ((1 << 16) - 1); + + if (msg_num > buf_size) { + ret = -EINVAL; + IPVR_ERROR("VED: message exceed maximum, ret:%d\n", ret); + goto out; + } + + ridx = VED_REG_READ32(MSVDX_COMMS_TO_MTX_RD_INDEX); + widx = VED_REG_READ32(MSVDX_COMMS_TO_MTX_WRT_INDEX); + + + buf_size = VED_REG_READ32(MSVDX_COMMS_TO_MTX_BUF_SIZE) & + ((1 << 16) - 1); + /*0x2000 is VEC Local Ram offset*/ + buf_offset = + (VED_REG_READ32(MSVDX_COMMS_TO_MTX_BUF_SIZE) >> 16) + 0x2000; + + /* message would wrap, need to send a pad message */ + if (widx + msg_num > buf_size) { + /* Shouldn't happen for a PAD message itself */ + if (header->bits.msg_type == MTX_MSGID_PADDING) + IPVR_DEBUG_WARN("VED: should not wrap pad msg, " + "buf_size is %d, widx is %d, msg_num is %d.\n", + buf_size, widx, msg_num); + + /* if the read pointer is at zero then we must wait for it to + * change otherwise the write pointer will equal the read + * pointer,which should only happen when the buffer is empty + * + * This will only happens if we try to overfill the queue, + * queue management should make + * sure this never happens in the first place. + */ + if (0 == ridx) { + ret = -EINVAL; + IPVR_ERROR("MSVDX: RIndex=0, ret:%d\n", ret); + goto out; + } + + /* Send a pad message */ + pad_msg.header.bits.msg_size = (buf_size - widx) << 2; + pad_msg.header.bits.msg_type = MTX_MSGID_PADDING; + ved_mtx_send(dev_priv, (void *)&pad_msg); + widx = VED_REG_READ32(MSVDX_COMMS_TO_MTX_WRT_INDEX); + } + + if (widx >= ridx) + words_free = buf_size - (widx - ridx) - 1; + else + words_free = ridx - widx - 1; + + if (msg_num > words_free) { + ret = -EINVAL; + IPVR_ERROR("MSVDX: msg_num > words_free, ret:%d\n", ret); + goto out; + } + while (msg_num > 0) { + VED_REG_WRITE32(*p_msg++, buf_offset + (widx << 2)); + msg_num--; + widx++; + if (buf_size == widx) + widx = 0; + } + + VED_REG_WRITE32(widx, MSVDX_COMMS_TO_MTX_WRT_INDEX); + + /* Make sure clocks are enabled before we kick + * but fw will take care of the clock after fw is loaded + */ + + /* signal an interrupt to let the mtx know there is a new message */ + VED_REG_WRITE32(1, MTX_KICK_INPUT_OFFSET); + + /* Read MSVDX Register several times in case Idle signal assert */ + VED_REG_READ32(MSVDX_INTERRUPT_STATUS_OFFSET); + VED_REG_READ32(MSVDX_INTERRUPT_STATUS_OFFSET); + VED_REG_READ32(MSVDX_INTERRUPT_STATUS_OFFSET); + VED_REG_READ32(MSVDX_INTERRUPT_STATUS_OFFSET); + +out: + return ret; +} + +static int32_t ved_cmd_send(struct drm_device *dev, void *cmd, + uint32_t cmd_size, struct ipvr_context *ipvr_ctx) +{ + int32_t ret = 0; + struct drm_ipvr_private *dev_priv = dev->dev_private; + union msg_header *header; + uint32_t cur_seq = 0xffffffff; + + while (cmd_size > 0) { + uint32_t cur_cmd_size, cur_cmd_id; + header = (union msg_header *)cmd; + cur_cmd_size = header->bits.msg_size; + cur_cmd_id = header->bits.msg_type; + + cur_seq = ((struct fw_msg_header *)cmd)->header.bits.msg_fence; + + if (cur_seq != 0xffffffff) { + ipvr_ctx->cur_seq = cur_seq; + } + + if (cur_cmd_size > cmd_size) { + ret = -EINVAL; + IPVR_ERROR("VED: cmd_size %u cur_cmd_size %u.\n", + cmd_size, cur_cmd_size); + goto out; + } + + /* Send the message to h/w */ + trace_ved_cmd_send(cur_cmd_id, cur_seq); + ret = ved_mtx_send(dev_priv, cmd); + if (ret) { + IPVR_DEBUG_WARN("VED: ret:%d\n", ret); + goto out; + } + cmd += cur_cmd_size; + cmd_size -= cur_cmd_size; + if (cur_cmd_id == MTX_MSGID_HOST_BE_OPP || + cur_cmd_id == MTX_MSGID_DEBLOCK || + cur_cmd_id == MTX_MSGID_INTRA_OOLD) { + cmd += (sizeof(struct fw_deblock_msg) - cur_cmd_size); + cmd_size -= + (sizeof(struct fw_deblock_msg) - cur_cmd_size); + } + } +out: + IPVR_DEBUG_VED("VED: ret:%d\n", ret); + return ret; +} + +int32_t ved_cmd_dequeue_send(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ved_cmd_queue *ved_cmd = NULL; + int32_t ret = 0; + struct ved_private *ved_priv = dev_priv->ved_private; + unsigned long irq_flags; + + spin_lock_irqsave(&ved_priv->ved_lock, irq_flags); + if (list_empty(&ved_priv->ved_queue)) { + IPVR_DEBUG_VED("VED: ved cmd queue empty.\n"); + ved_priv->ved_busy = 0; + spin_unlock_irqrestore(&ved_priv->ved_lock, irq_flags); + return -EINVAL; + } + + ved_cmd = list_first_entry(&ved_priv->ved_queue, + struct ved_cmd_queue, head); + list_del(&ved_cmd->head); + spin_unlock_irqrestore(&ved_priv->ved_lock, irq_flags); + + IPVR_DEBUG_VED("VED: cmd queue seq is %08x.\n", ved_cmd->cmd_seq); + + if (drm_ipvr_tiling) + ipvr_set_tile(dev, ved_cmd->tiling_scheme, + ved_cmd->tiling_stride); + +#ifdef CONFIG_DRM_IPVR_EC + /* Seperate update frame and backup cmds because if a batch of cmds + * doesn't have * host_be_opp message, no need to update frame info + * but still need to backup cmds. + * This case can happen if an batch of cmds is not the entire frame + */ + if (ved_cmd->host_be_opp_enabled) + ved_update_frame_info(ved_priv, ved_cmd->tfile, + ved_cmd->cmd + ved_cmd->deblock_cmd_offset); + + ved_backup_cmd(ved_priv, ved_cmd->tfile, + ved_cmd->cmd, + ved_cmd->cmd_size, + ved_cmd->deblock_cmd_offset); +#endif + ret = ved_cmd_send(dev, ved_cmd->cmd, + ved_cmd->cmd_size, ved_cmd->ipvr_ctx); + if (ret) { + IPVR_ERROR("VED: ved_cmd_send failed.\n"); + ret = -EINVAL; + } + + kfree(ved_cmd->cmd); + kfree(ved_cmd); + + return ret; +} + +void ved_flush_cmd_queue(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ved_cmd_queue *ved_cmd; + struct list_head *list, *next; + struct ved_private *ved_priv = dev_priv->ved_private; + unsigned long irq_flags; + spin_lock_irqsave(&ved_priv->ved_lock, irq_flags); + /* Flush the VED cmd queue and signal all fences in the queue */ + list_for_each_safe(list, next, &ved_priv->ved_queue) { + ved_cmd = list_entry(list, struct ved_cmd_queue, head); + list_del(list); + IPVR_DEBUG_VED("VED: flushing sequence:0x%08x.\n", + ved_cmd->cmd_seq); + ved_priv->ved_cur_seq = ved_cmd->cmd_seq; + + ipvr_fence_process(dev_priv, ved_cmd->cmd_seq, IPVR_CMD_SKIP); + + kfree(ved_cmd->cmd); + kfree(ved_cmd); + } + ved_priv->ved_busy = 0; + spin_unlock_irqrestore(&ved_priv->ved_lock, irq_flags); +} + +/* + * Returns: + * -EINVAL if invalid argument found + * -EFAULT if mapping command buffer fails + * -ENOMEM if memory not enough for copying command + * 0 if successful. + */ +static int32_t +ved_map_command(struct drm_device *dev, + struct drm_ipvr_gem_object *cmd_buffer, + uint32_t cmd_size, void **ved_cmd, + uint16_t sequence, int32_t copy_cmd, + struct ipvr_context *ipvr_ctx) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ved_private *ved_priv = dev_priv->ved_private; + int32_t ret = 0; + uint32_t cmd_size_remain; + void *cmd, *cmd_copy, *cmd_start; + union msg_header *header; + struct ipvr_fence *fence = NULL; + int fence_fd = -1; + + /* command buffers may not exceed page boundary */ + if (cmd_size > PAGE_SIZE) + return -EINVAL; + + cmd_start = kmap(sg_page(cmd_buffer->sg_table->sgl)); + if (!cmd_start) { + IPVR_ERROR("VED: kmap failed.\n"); + return -EFAULT; + } + + cmd = cmd_start; + cmd_size_remain = cmd_size; + + ved_priv->host_be_opp_enabled = 0; + ved_priv->deblock_cmd_offset = VED_INVALID_OFFSET; + + while (cmd_size_remain > 0) { + uint32_t cur_cmd_size, cur_cmd_id, mmu_ptd, msvdx_mmu_invalid; + if (cmd_size_remain < MTX_GENMSG_HEADER_SIZE) { + ret = -EINVAL; + goto out; + } + header = (union msg_header *)cmd; + cur_cmd_size = header->bits.msg_size; + cur_cmd_id = header->bits.msg_type; + mmu_ptd = 0; + msvdx_mmu_invalid = 0; + + IPVR_DEBUG_VED("cmd start at %lx cur_cmd_size = %d" + " cur_cmd_id = %02x fence = %08x\n", + (unsigned long)cmd, cur_cmd_size, + cur_cmd_id, sequence); + if ((cur_cmd_size % sizeof(uint32_t)) + || (cur_cmd_size > cmd_size_remain)) { + ret = -EINVAL; + IPVR_ERROR("VED: cmd size err, ret:%d.\n", ret); + goto out; + } + + switch (cur_cmd_id) { + case MTX_MSGID_DECODE_FE: { + struct fw_decode_msg *decode_msg; + if (sizeof(struct fw_decode_msg) > cmd_size_remain) { + /* Msg size is not correct */ + ret = -EINVAL; + IPVR_DEBUG_VED("MSVDX: wrong msg size.\n"); + goto out; + } + decode_msg = (struct fw_decode_msg *)cmd; + decode_msg->header.bits.msg_fence = sequence; + + mmu_ptd = ipvr_get_default_pd_addr(dev_priv->mmu); + msvdx_mmu_invalid = + atomic_cmpxchg(&dev_priv->ipvr_mmu_invaldc, + 1, 0); + if (msvdx_mmu_invalid == 1) { + decode_msg->flag_size.bits.flags |= + FW_INVALIDATE_MMU; + IPVR_DEBUG_VED("VED: Set MMU invalidate\n"); + } + /* + if (msvdx_mmu_invalid == 1) + ipvr_mmu_pgtable_dump(dev); + */ + /* if ctx_id is not passed, use default id */ + if (decode_msg->mmu_context.bits.context == 0) + decode_msg->mmu_context.bits.context = + dev_priv->default_ctx.ctx_id; + + decode_msg->mmu_context.bits.mmu_ptd = mmu_ptd >> 8; + IPVR_DEBUG_VED("VED: MSGID_DECODE_FE:" + " - fence: %08x" + " - flags: %08x - buffer_size: %08x" + " - crtl_alloc_addr: %08x" + " - context: %08x - mmu_ptd: %08x" + " - operating_mode: %08x.\n", + decode_msg->header.bits.msg_fence, + decode_msg->flag_size.bits.flags, + decode_msg->flag_size.bits.buffer_size, + decode_msg->crtl_alloc_addr, + decode_msg->mmu_context.bits.context, + decode_msg->mmu_context.bits.mmu_ptd, + decode_msg->operating_mode); + break; + } + + case MTX_MSGID_HOST_BE_OPP_MFLD: + ved_priv->host_be_opp_enabled = 1; + ved_priv->deblock_cmd_offset = + cmd_size - cmd_size_remain; + case MTX_MSGID_INTRA_OOLD_MFLD: + case MTX_MSGID_DEBLOCK_MFLD: { + struct fw_deblock_msg *deblock_msg; + if (sizeof(struct fw_deblock_msg) > cmd_size_remain) { + /* Msg size is not correct */ + ret = -EINVAL; + IPVR_DEBUG_VED("MSVDX: wrong msg size.\n"); + goto out; + } + deblock_msg = (struct fw_deblock_msg *)cmd; + mmu_ptd = ipvr_get_default_pd_addr(dev_priv->mmu); + msvdx_mmu_invalid = + atomic_cmpxchg(&dev_priv->ipvr_mmu_invaldc, + 1, 0); + if (msvdx_mmu_invalid == 1) { + deblock_msg->flag_type.bits.flags |= + FW_INVALIDATE_MMU; + IPVR_DEBUG_VED("VED: Set MMU invalidate\n"); + } + + /* patch to right cmd type */ + deblock_msg->header.bits.msg_type = + cur_cmd_id - + MTX_MSGID_DEBLOCK_MFLD + + MTX_MSGID_DEBLOCK; + + deblock_msg->header.bits.msg_fence = sequence & 0xffff; + deblock_msg->mmu_context.bits.mmu_ptd = (mmu_ptd >> 8); + /* if ctx_id is not passed, use default id */ + if (deblock_msg->mmu_context.bits.context == 0) + deblock_msg->mmu_context.bits.context = + dev_priv->default_ctx.ctx_id; + IPVR_DEBUG_VED("VED: MSGID_DEBLOCK:" + " - fence: %08x" + " - flags: %08x - slice_field_type: %08x" + " - operating_mode: %08x" + " - context: %08x - mmu_ptd: %08x" + " - frame_height_mb: %08x - pic_width_mb: %08x" + " - address_a0: %08x - address_a1: %08x" + " - mb_param_address: %08x" + " - ext_stride_a: %08x" + " - address_b0: %08x - address_b1: %08x" + " - alt_output_flags_b: %08x.\n", + deblock_msg->header.bits.msg_fence, + deblock_msg->flag_type.bits.flags, + deblock_msg->flag_type.bits.slice_field_type, + deblock_msg->operating_mode, + deblock_msg->mmu_context.bits.context, + deblock_msg->mmu_context.bits.mmu_ptd, + deblock_msg->pic_size.bits.frame_height_mb, + deblock_msg->pic_size.bits.pic_width_mb, + deblock_msg->address_a0, + deblock_msg->address_a1, + deblock_msg->mb_param_address, + deblock_msg->ext_stride_a, + deblock_msg->address_b0, + deblock_msg->address_b1, + deblock_msg->alt_output_flags_b); + cmd += (sizeof(struct fw_deblock_msg) - cur_cmd_size); + cmd_size_remain -= (sizeof(struct fw_deblock_msg) - + cur_cmd_size); + break; + } + default: + /* Msg not supported */ + ret = -EINVAL; + IPVR_DEBUG_WARN("VED: msg not supported.\n"); + goto out; + } + + cmd += cur_cmd_size; + cmd_size_remain -= cur_cmd_size; + if (((sequence++) & 0xf) == 0xf) { + ret = -EINVAL; + IPVR_DEBUG_WARN("VED: too many cmds, abort.\n"); + goto out; + } + } + + ved_priv->num_cmd = ((--sequence) & 0xf); + + ipvr_fence_create(dev_priv, &fence, &fence_fd); + + ipvr_fence_buffer_objects(&dev_priv->validate_ctx.validate_list, + fence, fence_fd); + + if (copy_cmd) { + IPVR_DEBUG_VED("VED: copying command.\n"); + + cmd_copy = kzalloc(cmd_size, GFP_KERNEL); + if (cmd_copy == NULL) { + ret = -ENOMEM; + IPVR_ERROR("VED: fail to callc, ret=:%d\n", ret); + goto out; + } + memcpy(cmd_copy, cmd_start, cmd_size); + *ved_cmd = cmd_copy; + } else { + IPVR_DEBUG_VED("VED: did NOT copy command.\n"); + if (drm_ipvr_tiling) + ipvr_set_tile(dev, ved_priv->ipvr_ctx->tiling_scheme, + ved_priv->ipvr_ctx->tiling_stride); + +#ifdef CONFIG_DRM_IPVR_EC + if (ved_priv->host_be_opp_enabled) { + ved_update_frame_info(ved_priv, + ved_priv->tfile, + cmd_start + ved_priv->deblock_cmd_offset); + } + ved_backup_cmd(ved_priv, ved_priv->tfile, + cmd_start, + cmd_size, + ved_priv->deblock_cmd_offset); +#endif + ret = ved_cmd_send(dev, cmd_start, cmd_size, ipvr_ctx); + if (ret) { + IPVR_ERROR("VED: ved_cmd_send failed\n"); + ret = -EINVAL; + } + } + +out: + kunmap(sg_page(cmd_buffer->sg_table->sgl)); + + return ret; +} + +/* + * Returns: + * -EINVAL + * -EFAULT + * -ENOMEM + * 0 + */ +static int32_t +ved_submit_cmdbuf_copy(struct drm_device *dev, + struct drm_ipvr_gem_object *cmd_buffer, + uint32_t cmd_size, + struct ipvr_context *ipvr_ctx, + uint32_t fence_flag) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ved_private *ved_priv = dev_priv->ved_private; + struct ved_cmd_queue *ved_cmd; + uint16_t sequence = (dev_priv->last_seq << 4); + unsigned long irq_flags; + void *cmd = NULL; + int32_t ret; + + /* queue the command to be sent when the h/w is ready */ + IPVR_DEBUG_VED("VED: queueing sequence:%08x.\n", + sequence); + ved_cmd = kzalloc(sizeof(struct ved_cmd_queue), + GFP_KERNEL); + if (ved_cmd == NULL) { + IPVR_ERROR("MSVDXQUE: Out of memory...\n"); + return -ENOMEM; + } + + ret = ved_map_command(dev, cmd_buffer, cmd_size, + &cmd, sequence, 1, ipvr_ctx); + if (ret) { + IPVR_ERROR("VED: Failed to extract cmd\n"); + kfree(ved_cmd); + /* -EINVAL or -EFAULT or -ENOMEM */ + return ret; + } + ved_cmd->cmd = cmd; + ved_cmd->cmd_size = cmd_size; + ved_cmd->cmd_seq = sequence; + + ved_cmd->tiling_scheme = ved_priv->ipvr_ctx->tiling_scheme; + ved_cmd->tiling_stride = ved_priv->ipvr_ctx->tiling_stride; + ved_cmd->deblock_cmd_offset = + ved_priv->deblock_cmd_offset; + ved_cmd->host_be_opp_enabled = + ved_priv->host_be_opp_enabled; + ved_cmd->tfile = + ved_priv->tfile; + ved_cmd->ipvr_ctx = ipvr_ctx; + spin_lock_irqsave(&ved_priv->ved_lock, irq_flags); + list_add_tail(&ved_cmd->head, &ved_priv->ved_queue); + spin_unlock_irqrestore(&ved_priv->ved_lock, irq_flags); + if (!ved_priv->ved_busy) { + ved_priv->ved_busy = 1; + IPVR_DEBUG_VED("VED: Need immediate dequeue.\n"); + ved_cmd_dequeue_send(dev); + } + + return ret; +} + +int32_t +ved_submit_video_cmdbuf(struct drm_device *dev, + struct drm_ipvr_gem_object *cmd_buffer, + uint32_t cmd_size, + struct ipvr_context *ipvr_ctx, + uint32_t fence_flag) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + uint16_t sequence = (dev_priv->last_seq << 4) & 0xffff; + unsigned long irq_flags; + int32_t ret = 0; + struct ved_private *ved_priv = dev_priv->ved_private; +#ifdef CONFIG_DRM_IPVR_EC + int offset = 0; +#endif + + if (sequence == IPVR_FENCE_SIGNALED_SEQ) { + sequence = (++dev_priv->last_seq << 4) & 0xffff; + } + + if (!ved_priv->fw_b0_uploaded){ + /* not needed for baytrail */ + } + + if (!ipvr_ctx) { + IPVR_ERROR("VED: null ctx\n"); + return -EFAULT; + } + + spin_lock_irqsave(&ved_priv->ved_lock, irq_flags); + + ved_priv->ipvr_ctx = ipvr_ctx; + + IPVR_DEBUG_VED("sequence is 0x%x, needs_reset is 0x%x.\n", + sequence, ved_priv->ved_needs_reset); + + if (ved_priv->ved_busy) { + spin_unlock_irqrestore(&ved_priv->ved_lock, irq_flags); + ret = ved_submit_cmdbuf_copy(dev, cmd_buffer, + cmd_size, ipvr_ctx, fence_flag); + + return ret; + } + + if (ved_priv->ved_needs_reset) { + spin_unlock_irqrestore(&ved_priv->ved_lock, irq_flags); + IPVR_DEBUG_VED("VED: will reset msvdx.\n"); + + if (!ved_priv->fw_loaded_by_punit) { + if (ved_core_reset(dev_priv)) { + ret = -EBUSY; + IPVR_ERROR("VED: Reset failed.\n"); + return ret; + } + } + + ved_priv->ved_needs_reset = 0; + ved_priv->ved_busy = 0; + + if (ved_priv->fw_loaded_by_punit){ + ret = ved_post_init(dev); + if (ret) { + ret = -EBUSY; + IPVR_DEBUG_WARN("VED: ved_post_init fail.\n"); + return ret; + } + } + else{ + if (ipvr_ved_init(dev)) { + ret = -EBUSY; + IPVR_DEBUG_WARN("VED: ipvr_ved_init fail.\n"); + return ret; + } + } + +#ifdef CONFIG_DRM_IPVR_EC + /* restore the state when power up during EC */ + if (ved_priv->vec_ec_mem_saved) { + for (offset = 0; offset < 4; ++offset) + VED_REG_WRITE32( + ved_priv->vec_ec_mem_data[offset], + 0x2cb0 + offset * 4); + + VED_REG_WRITE32(ved_priv->vec_ec_mem_data[4], + 0x2cc4); + ved_priv->vec_ec_mem_saved = 0; + } +#endif + + spin_lock_irqsave(&ved_priv->ved_lock, irq_flags); + } + + if (ved_priv->fw_loaded_by_punit && !ved_priv->rendec_initialized) { + spin_unlock_irqrestore(&ved_priv->ved_lock, irq_flags); + IPVR_DEBUG_VED("VED: setup msvdx.\n"); + ret = ved_post_boot_init(dev); + if (ret) { + IPVR_ERROR("VED: fail to setup msvdx.\n"); + /* FIXME: find a proper return value */ + return -EFAULT; + } + ved_priv->rendec_initialized = 1; + + IPVR_DEBUG_VED("VED: setup msvdx successfully.\n"); + spin_lock_irqsave(&ved_priv->ved_lock, irq_flags); + } + + if (!ved_priv->fw_loaded_by_punit && !ved_priv->ved_fw_loaded) { + spin_unlock_irqrestore(&ved_priv->ved_lock, irq_flags); + IPVR_DEBUG_VED("VED: reload FW to MTX\n"); + ret = ved_setup_fw(dev); + if (ret) { + IPVR_ERROR("VED: fail to load FW\n"); + /* FIXME: find a proper return value */ + return -EFAULT; + } + ved_priv->ved_fw_loaded = 1; + + IPVR_DEBUG_VED("VED: load firmware successfully\n"); + spin_lock_irqsave(&ved_priv->ved_lock, irq_flags); + } + + ved_priv->ved_busy = 1; + spin_unlock_irqrestore(&ved_priv->ved_lock, irq_flags); + IPVR_DEBUG_VED("VED: commit command to HW,seq=0x%08x\n", + sequence); + ret = ved_map_command(dev, cmd_buffer, cmd_size, + NULL, sequence, 0, ipvr_ctx); + if (ret) + IPVR_ERROR("VED: Failed to extract cmd.\n"); + + return ret; +} + +/* Returns: + * -EINVAL + * -ENOMEM + * -EFAULT + * -EBUSY + */ +int32_t ved_cmdbuf_video(struct drm_file *file_priv, + struct drm_ipvr_gem_object *cmd_buffer, + uint32_t cmdbuf_size, + struct ipvr_context *ipvr_ctx) +{ + struct drm_device *dev = file_priv->minor->dev; + int32_t ret; + + /* + * Check this. Doesn't seem right. Have fencing done AFTER command + * submission and make sure drm_ved_idle idles the VED completely. + */ + ret = ved_submit_video_cmdbuf(dev, cmd_buffer, cmdbuf_size, ipvr_ctx, 0); + if (ret) + return ret; + + return 0; +} + +static int32_t ved_handle_panic_msg(struct ved_private *ved_priv, + struct fw_panic_msg *panic_msg) +{ + /* For VXD385 firmware, fence value is not validate here */ + uint32_t i, diff = 0; + uint16_t fence; + uint32_t err_trig, irq_sts, mmu_sts, dmac_sts; + struct ved_frame_info *failed_frame = NULL; + struct drm_ipvr_private *dev_priv = ved_priv->dev_priv; + struct drm_device *dev = dev_priv->dev; + IPVR_DEBUG_WARN("MSVDX: MSGID_CMD_HW_PANIC:" + "Fault detected" + " - Fence: %08x" + " - fe_status mb: %08x" + " - be_status mb: %08x" + " - reserved2: %08x" + " - last mb: %08x" + " - resetting and ignoring error\n", + panic_msg->header.bits.msg_fence, + panic_msg->fe_status, + panic_msg->be_status, + panic_msg->mb.bits.reserved2, + panic_msg->mb.bits.last_mb); + /* + * If bit 8 of MSVDX_INTERRUPT_STATUS is set the fault + * was caused in the DMAC. In this case you should + * check bits 20:22 of MSVDX_INTERRUPT_STATUS. + * If bit 20 is set there was a problem DMAing the buffer + * back to host. If bit 22 is set you'll need to get the + * value of MSVDX_DMAC_STREAM_STATUS (0x648). + * If bit 1 is set then there was an issue DMAing + * the bitstream or termination code for parsing. + */ + err_trig = VED_REG_READ32(MSVDX_COMMS_ERROR_TRIG); + irq_sts = VED_REG_READ32(MSVDX_INTERRUPT_STATUS_OFFSET); + mmu_sts = VED_REG_READ32(MSVDX_MMU_STATUS_OFFSET); + dmac_sts = VED_REG_READ32(MSVDX_DMAC_STREAM_STATUS_OFFSET); + IPVR_DEBUG_VED("MSVDX: MSVDX_COMMS_ERROR_TRIG is 0x%x," + "MSVDX_INTERRUPT_STATUS is 0x%x," + "MSVDX_MMU_STATUS is 0x%x," + "MSVDX_DMAC_STREAM_STATUS is 0x%x.\n", + err_trig, irq_sts, mmu_sts, dmac_sts); + + trace_ved_irq_panic(panic_msg, err_trig, irq_sts, mmu_sts, dmac_sts); + + fence = panic_msg->header.bits.msg_fence; + + if (ved_priv->fw_loaded_by_punit) + ved_priv->ved_needs_reset |= MSVDX_RESET_NEEDS_REUPLOAD_FW | + MSVDX_RESET_NEEDS_INIT_FW; + else + ved_priv->ved_needs_reset = 1; + + diff = ved_priv->ved_cur_seq - dev_priv->last_seq; + if (diff > 0x0FFFFFFF) + ved_priv->ved_cur_seq++; + + IPVR_DEBUG_WARN("VED: Fence ID missing, assuming %08x\n", + ved_priv->ved_cur_seq); + + ipvr_fence_process(dev_priv, ved_priv->ved_cur_seq, IPVR_CMD_FAILED); + + /* Flush the command queue */ + ved_flush_cmd_queue(dev); + if (ved_priv->host_be_opp_enabled) { + /* get the frame_info struct for ec frame */ + for (i = 0; i < MAX_DECODE_BUFFERS; i++) { + /* by default fence is 0, so there is problem here */ + if (ved_priv->frame_info[i].fence == fence) { + failed_frame = &ved_priv->frame_info[i]; + break; + } + } + if (!failed_frame) { + IPVR_ERROR("VED: didn't find frame_info which matched " + "the fence %d in panic message\n", fence); + return -EINVAL; + } + + failed_frame->fw_status = 1; /* set ERROR flag */ + } + ved_priv->decoding_err = 1; + return 0; +} + +static int32_t +ved_handle_completed_msg(struct ved_private *ved_priv, + struct fw_completed_msg *completed_msg) +{ + struct drm_ipvr_private *dev_priv = ved_priv->dev_priv; + struct drm_device *dev = dev_priv->dev; + uint16_t fence, flags; + struct ipvr_context *ipvr_ctx; + + IPVR_DEBUG_VED("VED: MSGID_CMD_COMPLETED:" + " - Fence: %08x - flags: %08x - vdebcr: %08x" + " - first_mb : %d - last_mb: %d\n", + completed_msg->header.bits.msg_fence, + completed_msg->flags, completed_msg->vdebcr, + completed_msg->mb.bits.start_mb, + completed_msg->mb.bits.last_mb); + + trace_ved_irq_completed(completed_msg); + + flags = completed_msg->flags; + fence = completed_msg->header.bits.msg_fence; + + ved_priv->ved_cur_seq = fence; + +#if 0 + if (IS_MRFLD(dev)) + ved_fw_error_detected(dev, fence, flags); +#endif + + ipvr_fence_process(dev_priv, fence, IPVR_CMD_SUCCESS); + + ipvr_ctx = ipvr_find_ctx_with_fence(dev_priv, fence); + if (unlikely(ipvr_ctx == NULL)) { + IPVR_DEBUG_GENERAL("abnormal complete msg.\n"); + return -EINVAL; + } + + if (flags & FW_VA_RENDER_HOST_INT) { + /* Now send the next command from the msvdx cmd queue */ + ved_cmd_dequeue_send(dev); + } + return 0; +} + +static int32_t +ved_handle_contiguity_msg(struct ved_private *ved_priv, + struct fw_contiguity_msg *contiguity_msg) +{ + struct ved_decode_status *fault_region = NULL; + struct ved_ec_context *ved_ec_ctx = NULL; + uint32_t reg_idx, i, fence, start, end; + int32_t found = 0; + struct ved_decode_status *cur_dec_status; + + IPVR_DEBUG_VED("VED: MSGID_CONTIGUITY_WARNING:"); + IPVR_DEBUG_VED("- Fence: %08x - end_mb: %08x - begin_mb: %08x\n", + contiguity_msg->header.bits.msg_fence, + contiguity_msg->mb.bits.end_mb_num, + contiguity_msg->mb.bits.begin_mb_num); + + trace_ved_irq_contiguity(contiguity_msg); + /* get erro info */ + fence = contiguity_msg->header.bits.msg_fence; + start = contiguity_msg->mb.bits.begin_mb_num; + end = contiguity_msg->mb.bits.end_mb_num; + + /*get the frame_info struct for error concealment frame*/ + for (i = 0; i < VED_MAX_EC_INSTANCE; i++) + if (ved_priv->ved_ec_ctx[i]->fence == (fence & (~0xf))) { + ved_ec_ctx = ved_priv->ved_ec_ctx[i]; + found++; + } + /* ved_mtx_message_dump(dev); */ + if (!ved_ec_ctx || !(ved_ec_ctx->tfile) || found > 1) { + IPVR_DEBUG_VED("no matched ctx: fence 0x%x, " + "found %d, ctx 0x%08lx\n", + fence, found, (unsigned long)ved_ec_ctx); + return -EINVAL; + } + + if (ved_ec_ctx->cur_frame_info) + cur_dec_status = &ved_ec_ctx->cur_frame_info->decode_status; + + fault_region = &ved_ec_ctx->decode_status; + if (start > end) + start = end; + if (start < VED_EC_ROLLBACK) + start = 0; + else + start -= VED_EC_ROLLBACK; + + if (fault_region->num_region) { + reg_idx = fault_region->num_region - 1; + if ((start <= fault_region->mb_regions[reg_idx].end) && + (end > fault_region->mb_regions[reg_idx].end)) { + fault_region->mb_regions[reg_idx].end = end; + if (ved_ec_ctx->cur_frame_info) { + cur_dec_status->mb_regions[reg_idx].end = end; + } + } + else { + reg_idx = fault_region->num_region++; + if (unlikely(reg_idx >= MAX_SLICES_PER_PICTURE)) { + IPVR_DEBUG_VED("too many fault regions.\n"); + return -EINVAL; + } + fault_region->mb_regions[reg_idx].start = start; + fault_region->mb_regions[reg_idx].end = end; + if (ved_ec_ctx->cur_frame_info) { + cur_dec_status->num_region = + fault_region->num_region; + cur_dec_status->mb_regions[reg_idx].start = + start; + cur_dec_status->mb_regions[reg_idx].end = end; + } + } + } else { + fault_region->num_region++; + fault_region->mb_regions[0].start = start; + fault_region->mb_regions[0].end = end; + if (ved_ec_ctx->cur_frame_info) { + cur_dec_status->num_region = fault_region->num_region; + cur_dec_status->mb_regions[0].start = start; + cur_dec_status->mb_regions[0].end = end; + } + } + return 0; +} + +static int32_t +ved_handle_deblock_required_msg(struct ved_private *ved_priv, + struct fw_deblock_required_msg *deblock_required_msg) +{ + uint32_t i; + int32_t found = 0; + struct ved_ec_context *ved_ec_ctx = NULL; + struct drm_ipvr_private *dev_priv = ved_priv->dev_priv; + uint16_t fence = deblock_required_msg->header.bits.msg_fence; + IPVR_DEBUG_VED("VED: MTX_MSGID_DEBLOCK_REQUIRED Fence=%08x.\n", fence); + IPVR_DEBUG_VED("Get deblock required msg for ec.\n"); + for (i = 0; i < VED_MAX_EC_INSTANCE; i++) + if (ved_priv->ved_ec_ctx[i]->fence == (fence & (~0xf))) { + ved_ec_ctx = ved_priv->ved_ec_ctx[i]; + found++; + } + + trace_ved_irq_deblock_required(deblock_required_msg); + + /* if found > 1, fence wrapping happens */ + if (!ved_ec_ctx || !(ved_ec_ctx->tfile) || found > 1) { + IPVR_DEBUG_VED("no matched ctx: fence 0x%x, " + "found %d, ctx 0x%08lx\n", + fence, found, (unsigned long)ved_ec_ctx); + VED_REG_WRITE32(0, MSVDX_CMDS_END_SLICE_PICTURE_OFFSET); + VED_REG_WRITE32(1, MSVDX_CMDS_END_SLICE_PICTURE_OFFSET); + return -EINVAL; + } + + ved_ec_ctx->cur_frame_info->fw_status = 1; + ved_priv->cur_msvdx_ec_ctx = ved_ec_ctx; + + /*do error concealment with hw*/ + schedule_work(&ved_priv->ec_work); + return 0; +} + +/* + * MSVDX MTX interrupt + */ +static void ved_mtx_interrupt(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + static uint32_t buf[128]; /* message buffer */ + uint32_t ridx, widx, buf_size, buf_offset; + uint32_t num, ofs; /* message num and offset */ + struct ved_private *ved_priv = dev_priv->ved_private; + union msg_header *header; + int32_t cmd_complete = 0; + int ret; + IPVR_DEBUG_VED("VED: Got a VED MTX interrupt.\n"); + + /* we need clocks enabled before we touch VEC local ram, + * but fw will take care of the clock after fw is loaded + */ + +loop: /* just for coding style check */ + ridx = VED_REG_READ32(MSVDX_COMMS_TO_HOST_RD_INDEX); + widx = VED_REG_READ32(MSVDX_COMMS_TO_HOST_WRT_INDEX); + + /* Get out of here if nothing */ + if (ridx == widx) + goto done; + + buf_size = VED_REG_READ32(MSVDX_COMMS_TO_HOST_BUF_SIZE) & + ((1 << 16) - 1); + /*0x2000 is VEC Local Ram offset*/ + buf_offset = (VED_REG_READ32(MSVDX_COMMS_TO_HOST_BUF_SIZE) >> 16) + + 0x2000; + + ofs = 0; + buf[ofs] = VED_REG_READ32(buf_offset + (ridx << 2)); + header = (union msg_header *)buf; + + /* round to nearest word */ + num = (header->bits.msg_size + 3) / 4; + + /* ASSERT(num <= sizeof(buf) / sizeof(uint32_t)); */ + + if (++ridx >= buf_size) + ridx = 0; + + for (ofs++; ofs < num; ofs++) { + buf[ofs] = VED_REG_READ32(buf_offset + (ridx << 2)); + + if (++ridx >= buf_size) + ridx = 0; + } + + /* Update the Read index */ + VED_REG_WRITE32(ridx, MSVDX_COMMS_TO_HOST_RD_INDEX); + + if (ved_priv->ved_needs_reset) + goto loop; + + switch (header->bits.msg_type) { + case MTX_MSGID_HW_PANIC: { + struct fw_panic_msg *panic_msg = (struct fw_panic_msg *)buf; + cmd_complete = 1; + ved_handle_panic_msg(ved_priv, panic_msg); + goto done; + } + + case MTX_MSGID_COMPLETED: { + struct fw_completed_msg *completed_msg = + (struct fw_completed_msg *)buf; + cmd_complete = 1; + if (ved_handle_completed_msg(ved_priv, completed_msg)) + cmd_complete = 0; + break; + } + + case MTX_MSGID_CONTIGUITY_WARNING: { + struct fw_contiguity_msg *contiguity_msg = + (struct fw_contiguity_msg *)buf; + ved_handle_contiguity_msg(ved_priv, contiguity_msg); + break; + + } + + case MTX_MSGID_DEBLOCK_REQUIRED: { + struct fw_deblock_required_msg *deblock_required_msg = + (struct fw_deblock_required_msg *)buf; + ved_handle_deblock_required_msg(ved_priv, deblock_required_msg); + break; + } + default: + IPVR_ERROR("VED: unknown message from MTX, ID:0x%08x.\n", + header->bits.msg_type); + goto done; + } + +done: + IPVR_DEBUG_VED("VED Interrupt: finish process a message.\n"); + if (ridx != widx) { + IPVR_DEBUG_VED("VED: there are more message to be read.\n"); + goto loop; + } + + atomic_dec(&dev_priv->ved_power_usage); + ret = pm_runtime_put(&dev->platformdev->dev); + if (unlikely(ret < 0)) { + IPVR_ERROR("Error put VED power: %d\n", ret); + } + IPVR_DEBUG_PM("VED power put, usage became %d\n", + atomic_read(&dev->platformdev->dev.power.usage_count)); + + mb(); /* TBD check this... */ +} + +/* + * MSVDX interrupt. + */ +int32_t ved_irq_handler(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv; + struct ved_private *ved_priv; + uint32_t msvdx_stat; + + if (dev == NULL) { + IPVR_ERROR("VED: invalid dev.\n"); + return -EINVAL; + } + + dev_priv = dev->dev_private; + + ved_priv = dev_priv->ved_private; + msvdx_stat = VED_REG_READ32(MSVDX_INTERRUPT_STATUS_OFFSET); + + /* driver only needs to handle mtx irq + * For MMU fault irq, there's always a HW PANIC generated + * if HW/FW is totally hang, the lockup function will handle + * the reseting + */ + if (msvdx_stat & MSVDX_INTERRUPT_STATUS_MMU_FAULT_IRQ_MASK) { + /*Ideally we should we should never get to this */ + IPVR_DEBUG_IRQ("VED: MMU Fault:0x%x\n", msvdx_stat); + + /* Pause MMU */ + VED_REG_WRITE32(MSVDX_MMU_CONTROL0_MMU_PAUSE_MASK, + MSVDX_MMU_CONTROL0_OFFSET); + wmb(); + + /* Clear this interupt bit only */ + VED_REG_WRITE32(MSVDX_INTERRUPT_STATUS_MMU_FAULT_IRQ_MASK, + MSVDX_INTERRUPT_CLEAR_OFFSET); + VED_REG_READ32(MSVDX_INTERRUPT_CLEAR_OFFSET); + rmb(); + + ved_priv->ved_needs_reset = 1; + } else if (msvdx_stat & MSVDX_INTERRUPT_STATUS_MTX_IRQ_MASK) { + IPVR_DEBUG_IRQ("VED: msvdx_stat: 0x%x(MTX)\n", msvdx_stat); + + /* Clear all interupt bits */ + if (ved_priv->fw_loaded_by_punit) + VED_REG_WRITE32(MSVDX_INTERRUPT_STATUS_MTX_IRQ_MASK, + MSVDX_INTERRUPT_CLEAR_OFFSET); + else + VED_REG_WRITE32(0xffff, MSVDX_INTERRUPT_CLEAR_OFFSET); + + VED_REG_READ32(MSVDX_INTERRUPT_CLEAR_OFFSET); + rmb(); + + ved_mtx_interrupt(dev); + } + + return 0; +} + +int32_t ved_check_idle(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv; + struct ved_private *ved_priv; + uint32_t loop, ret; + + dev_priv = dev->dev_private; + if (!dev_priv) + return -ENODEV; + + ved_priv = dev_priv->ved_private; + if (!ved_priv) + return 0; + + if (ved_priv->fw_loaded_by_punit && ved_priv->rendec_initialized == 0) + return 0; + + if (!ved_priv->fw_loaded_by_punit && ved_priv->ved_fw_loaded == 0) + return 0; + + if (ved_priv->ved_busy) { + IPVR_DEBUG_PM("VED: ved_busy was set, return busy.\n"); + return -EBUSY; + } + + if (ved_priv->fw_loaded_by_punit) { + if (!(VED_REG_READ32(MSVDX_COMMS_FW_STATUS) & + MSVDX_FW_STATUS_HW_IDLE)) { + IPVR_DEBUG_PM("MSVDX_COMMS_SIGNATURE reg is 0x%x,\n" + "MSVDX_COMMS_FW_STATUS reg is 0x%x,\n" + "indicate hw is busy.\n", + VED_REG_READ32(MSVDX_COMMS_SIGNATURE), + VED_REG_READ32(MSVDX_COMMS_FW_STATUS)); + return -EBUSY; + } + } + + /* on some cores below 50502, there is one instance that + * read requests may not go to zero is in the case of a page fault, + * check core revision by reg MSVDX_CORE_REV, 385 core is 0x20001 + * check if mmu page fault happend by reg MSVDX_INTERRUPT_STATUS, + * check was it a page table rather than a protection fault + * by reg MSVDX_MMU_STATUS, for such case, + * need call ved_core_reset as the work around */ + if ((VED_REG_READ32(MSVDX_CORE_REV_OFFSET) < 0x00050502) && + (VED_REG_READ32(MSVDX_INTERRUPT_STATUS_OFFSET) + & MSVDX_INTERRUPT_STATUS_MMU_FAULT_IRQ_MASK) && + (VED_REG_READ32(MSVDX_MMU_STATUS_OFFSET) & 1)) { + IPVR_DEBUG_WARN("mmu page fault, recover by core_reset.\n"); + return 0; + } + + /* check MSVDX_MMU_MEM_REQ to confirm there's no memory requests */ + for (loop = 0; loop < 10; loop++) + ret = ved_wait_for_register(dev_priv, + MSVDX_MMU_MEM_REQ_OFFSET, + 0, 0xff, 100, 1); + if (ret) { + IPVR_DEBUG_WARN("MSVDX: MSVDX_MMU_MEM_REQ reg is 0x%x,\n" + "indicate mem busy, prevent power off ved," + "MSVDX_COMMS_FW_STATUS reg is 0x%x," + "MSVDX_COMMS_ERROR_TRIG reg is 0x%x,", + VED_REG_READ32(MSVDX_MMU_MEM_REQ_OFFSET), + VED_REG_READ32(MSVDX_COMMS_FW_STATUS), + VED_REG_READ32(MSVDX_COMMS_ERROR_TRIG)); + return -EBUSY; + } + /* + if (ved_priv->ved_hw_busy) { + IPVR_DEBUG_PM("VED: %s, HW is busy\n", __func__); + return -EBUSY; + } + */ + return 0; +} + +void ved_check_reset_fw(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ved_private *ved_priv = dev_priv->ved_private; + unsigned long irq_flags; + + spin_lock_irqsave(&ved_priv->ved_lock, irq_flags); + + /* handling fw upload here if required */ + /* power off first, then hw_begin will power up/upload FW correctly */ + if (ved_priv->ved_needs_reset & MSVDX_RESET_NEEDS_REUPLOAD_FW) { + ved_priv->ved_needs_reset &= ~MSVDX_RESET_NEEDS_REUPLOAD_FW; + spin_unlock_irqrestore(&ved_priv->ved_lock, irq_flags); + IPVR_DEBUG_VED("VED: force power off VED due to decode err\n"); + /*ospm_apm_power_down_ved(dev, 1);*/ + spin_lock_irqsave(&ved_priv->ved_lock, irq_flags); + } + spin_unlock_irqrestore(&ved_priv->ved_lock, irq_flags); +} diff --git a/drivers/gpu/drm/ipvr/ved_cmd.h b/drivers/gpu/drm/ipvr/ved_cmd.h new file mode 100644 index 0000000..f3b5138 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ved_cmd.h @@ -0,0 +1,104 @@ +/************************************************************************** + * ved_cmd.h: VED header file to support command buffer handling + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * Copyright (c) Imagination Technologies Limited, UK + * Copyright (c) 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * + **************************************************************************/ + +#ifndef _VED_CMD_H_ +#define _VED_CMD_H_ + +#include "ipvr_drv.h" +#include "ipvr_drm.h" +#include "ipvr_gem.h" +#include "ved_reg.h" +#include "ved_ec.h" +#include "ved_pm.h" +#include "ipvr_fence.h" +#include "ipvr_exec.h" + +extern int32_t drm_ipvr_tiling; + +/* HOST_BE_OPP parameters */ +struct HOST_BE_OPP_PARAMS { + uint32_t handle; /* struct ttm_buffer_object * of REGIO */ + uint32_t buffer_stride; + uint32_t buffer_size; + uint32_t picture_width_mb; + uint32_t size_mb; +}; + +struct ved_cmd_queue { + struct list_head head; + void *cmd; + uint32_t cmd_size; + uint16_t cmd_seq; + uint32_t fence_flag; + uint8_t tiling_scheme; + uint8_t tiling_stride; + uint32_t host_be_opp_enabled; + uint32_t deblock_cmd_offset; + struct drm_file *tfile; + struct ipvr_context *ipvr_ctx; + int32_t frame_boundary; +}; + +#define VED_PMSTATE_POWERUP 0 +#define VED_PMSTATE_CLOCKGATED 1 +#define VED_PMSTATE_POWERDOWN 2 + +#define VED_NEW_PMSTATE(drm_dev, ved_priv, new_state) \ +do { \ + ved_priv->ved_pmstate = new_state; \ + if (new_state == VED_PMSTATE_POWERDOWN) \ + ved_priv->pm_gating_count++; \ + sysfs_notify_dirent(ved_priv->sysfs_pmstate); \ + IPVR_DEBUG_PM("VED: %s, power gating count 0x%08x\n", \ + (new_state == VED_PMSTATE_POWERUP) ? "powerup" \ + : ((new_state == VED_PMSTATE_POWERDOWN) ? "powerdown" \ + : "clockgated"), ved_priv->pm_gating_count); \ +} while (0) + +int32_t ved_irq_handler(struct drm_device *dev); + +int32_t ved_mtx_send(struct drm_ipvr_private *dev_priv, const void *msg); + +int32_t ved_check_idle(struct drm_device *dev); + +void ved_check_reset_fw(struct drm_device *dev); + +void ved_flush_cmd_queue(struct drm_device *dev); + +int32_t ved_cmdbuf_video(struct drm_file *file_priv, + struct drm_ipvr_gem_object *cmd_buffer, + uint32_t cmdbuf_size, + struct ipvr_context *ipvr_ctx); + +int32_t ved_submit_video_cmdbuf(struct drm_device *dev, + struct drm_ipvr_gem_object *cmd_buffer, + uint32_t cmd_size, + struct ipvr_context *ipvr_ctx, + uint32_t fence_flag); + +int32_t ved_cmd_dequeue_send(struct drm_device *dev); + +#endif diff --git a/drivers/gpu/drm/ipvr/ved_ec.c b/drivers/gpu/drm/ipvr/ved_ec.c new file mode 100644 index 0000000..2aa52cc --- /dev/null +++ b/drivers/gpu/drm/ipvr/ved_ec.c @@ -0,0 +1,584 @@ +/************************************************************************** + * ved_ec.c: VED error concealment support when decoding error happened + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * Copyright (c) Imagination Technologies Limited, UK + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Li Zeng li.zeng@intel.com + * + **************************************************************************/ + +#include "ved_ec.h" +#include "ved_cmd.h" +#include "ved_msg.h" +#include "ved_reg.h" +#include "ved_fw.h" +#include "ved_pm.h" +#include <linux/pm_runtime.h> + +#define MAX_SIZE_IN_MB (4096 / 16) + +static inline int32_t +ved_cmd_port_write(struct drm_ipvr_private *dev_priv, + uint32_t offset, uint32_t value, uint32_t *cmd_space) +{ + uint32_t max_attempts = 0xff; + uint32_t attempts = 0; + + max_attempts = 0xff; + while (*cmd_space == 0) { + *cmd_space = VED_REG_READ32( + MSVDX_CORE_CR_MSVDX_COMMAND_SPACE_OFFSET + + MSVDX_CORE_BASE); + if (*cmd_space) + break; + IPVR_UDELAY(2); + attempts++; + if (attempts > max_attempts) { + IPVR_DEBUG_WARN("VED: poll cmd space timeout.\n"); + return -1; + } + } + + VED_REG_WRITE32(value, offset + MSVDX_CMDS_BASE); + (*cmd_space)--; + /* + *IPVR_DEBUG_WARN("VED: poll cmd space attempts %d\n", attempts); + */ + return 0; +} + +#define VED_CMDPORT_WRITE(_dev_priv_, _offset_, _cmd_, _cmd_space_) \ + do { \ + ret = ved_cmd_port_write(_dev_priv_, \ + _offset_, _cmd_, &_cmd_space_); \ + if (ret) { \ + IPVR_DEBUG_WARN("VED: write cmd fail, abort\n");\ + goto ec_done; \ + } \ + } while (0); + +#define VED_CMDPORT_WRITE_FAST(_dev_priv_, _offset_, _cmd_, _cmd_space_)\ + ved_cmd_port_write(_dev_priv_, _offset_, _cmd_, &_cmd_space_); \ + +void ved_do_concealment(struct work_struct *work) +{ + struct ved_private *ved_priv = + container_of(work, struct ved_private, ec_work); + struct drm_ipvr_private *dev_priv = NULL; + struct ved_ec_context *ved_ec_ctx = ved_priv->cur_msvdx_ec_ctx; + struct ved_decode_status *fault_region = NULL; + struct fw_deblock_msg *deblock_msg = + (struct fw_deblock_msg *)(ved_ec_ctx->unfenced_cmd + + ved_ec_ctx->deblock_cmd_offset); + uint32_t width_in_mb, height_in_mb, cmd; + int32_t conceal_above_row = 0, loop, mb_loop; + uint32_t cmd_space = 0; + int32_t ret = 0; + int32_t pm_ret = 0; + + pm_ret = pm_runtime_get_sync(&dev_priv->dev->platformdev->dev); + if (unlikely(pm_ret < 0)) { + IPVR_ERROR("Error get VED power: %d\n", pm_ret); + return; + } + + dev_priv = ved_priv->dev_priv; + fault_region = &ved_ec_ctx->decode_status; + + /* Concealment should be done in time, + * otherwise panic msg will be signaled in msvdx + */ + preempt_disable(); + + if (ved_ec_ctx->deblock_cmd_offset == VED_INVALID_OFFSET) { + IPVR_DEBUG_WARN("VED: invalid msg offset, abort conceal.\n"); + goto ec_done; + } + + if (fault_region->num_region == 0) { + IPVR_DEBUG_VED("VED: no fault region.\n"); + goto ec_done; + } + + width_in_mb = deblock_msg->pic_size.bits.pic_width_mb; + height_in_mb = deblock_msg->pic_size.bits.frame_height_mb; + + { + int32_t i; + for (i = 0; i < fault_region->num_region; i++) + IPVR_DEBUG_VED("[region %d] is %d to %d\n", + i, + fault_region->mb_regions[i].start, + fault_region->mb_regions[i].end); + IPVR_DEBUG_VED("MSVDX: MSGID_DEBLOCK:" + " - fence: %08x" + " - flags: %08x - slice_field_type: %08x" + " - operating_mode: %08x" + " - context: %08x - mmu_ptd: %08x" + " - frame_height_mb: %08x - pic_width_mb: %08x" + " - address_a0: %08x - address_a1: %08x" + " - mb_param_address: %08x" + " - ext_stride_a: %08x" + " - address_b0: %08x - address_b1: %08x" + " - alt_output_flags_b: %08x.\n", + deblock_msg->header.bits.msg_fence, + deblock_msg->flag_type.bits.flags, + deblock_msg->flag_type.bits.slice_field_type, + deblock_msg->operating_mode, + deblock_msg->mmu_context.bits.context, + deblock_msg->mmu_context.bits.mmu_ptd, + deblock_msg->pic_size.bits.frame_height_mb, + deblock_msg->pic_size.bits.pic_width_mb, + deblock_msg->address_a0, + deblock_msg->address_a1, + deblock_msg->mb_param_address, + deblock_msg->ext_stride_a, + deblock_msg->address_b0, + deblock_msg->address_b1, + deblock_msg->alt_output_flags_b); + IPVR_DEBUG_VED("deblock addr_c0 is 0x%08x\n", + deblock_msg->address_c0); + IPVR_DEBUG_VED("deblock addr_c1 is 0x%08x\n", + deblock_msg->address_c1); + } + + if (unlikely(!width_in_mb || !height_in_mb || + width_in_mb > MAX_SIZE_IN_MB || + height_in_mb > MAX_SIZE_IN_MB)) { + IPVR_DEBUG_VED("wrong pic size\n"); + goto ec_done; + } + + cmd = 0; + REGIO_WRITE_FIELD_LITE(cmd, MSVDX_CMDS, + DISPLAY_PICTURE_SIZE_DISPLAY_PICTURE_HEIGHT, + (height_in_mb * 16) - 1); + REGIO_WRITE_FIELD_LITE(cmd, MSVDX_CMDS, + DISPLAY_PICTURE_SIZE_DISPLAY_PICTURE_WIDTH, + (width_in_mb * 16) - 1); + VED_CMDPORT_WRITE(dev_priv, + MSVDX_CMDS_DISPLAY_PICTURE_SIZE_OFFSET, + cmd, cmd_space); + + cmd = 0; + REGIO_WRITE_FIELD_LITE(cmd, MSVDX_CMDS, + CODED_PICTURE_SIZE_CODED_PICTURE_HEIGHT, + (height_in_mb * 16) - 1); + REGIO_WRITE_FIELD_LITE(cmd, MSVDX_CMDS, + CODED_PICTURE_SIZE_CODED_PICTURE_WIDTH, + (width_in_mb * 16) - 1); + VED_CMDPORT_WRITE(dev_priv, + MSVDX_CMDS_CODED_PICTURE_SIZE_OFFSET, + cmd, cmd_space); + + cmd = deblock_msg->operating_mode; + REGIO_WRITE_FIELD(cmd, MSVDX_CMDS_OPERATING_MODE, + CHROMA_FORMAT, 1); + REGIO_WRITE_FIELD(cmd, MSVDX_CMDS_OPERATING_MODE, + ASYNC_MODE, 1); + REGIO_WRITE_FIELD(cmd, MSVDX_CMDS_OPERATING_MODE, + CODEC_MODE, 3); + REGIO_WRITE_FIELD(cmd, MSVDX_CMDS_OPERATING_MODE, + CODEC_PROFILE, 1); + VED_CMDPORT_WRITE(dev_priv, MSVDX_CMDS_OPERATING_MODE_OFFSET, + cmd, cmd_space); + + /* dest frame address */ + VED_CMDPORT_WRITE(dev_priv, + MSVDX_CMDS_LUMA_RECONSTRUCTED_PICTURE_BASE_ADDRESSES_OFFSET, + deblock_msg->address_a0, + cmd_space); + + VED_CMDPORT_WRITE(dev_priv, + MSVDX_CMDS_CHROMA_RECONSTRUCTED_PICTURE_BASE_ADDRESSES_OFFSET, + deblock_msg->address_a1, + cmd_space); + + /* conceal frame address */ + VED_CMDPORT_WRITE(dev_priv, + MSVDX_CMDS_REFERENCE_PICTURE_BASE_ADDRESSES_OFFSET, + deblock_msg->address_b0, + cmd_space); + VED_CMDPORT_WRITE(dev_priv, + MSVDX_CMDS_REFERENCE_PICTURE_BASE_ADDRESSES_OFFSET + 4, + deblock_msg->address_b1, + cmd_space); + cmd = 0; + REGIO_WRITE_FIELD(cmd, MSVDX_CMDS_SLICE_PARAMS, SLICE_FIELD_TYPE, 2); + REGIO_WRITE_FIELD(cmd, MSVDX_CMDS_SLICE_PARAMS, SLICE_CODE_TYPE, 1); + + VED_CMDPORT_WRITE(dev_priv, + MSVDX_CMDS_SLICE_PARAMS_OFFSET, + cmd, cmd_space); + + cmd = deblock_msg->alt_output_flags_b; + if ((cmd & 3) != 0) { + IPVR_DEBUG_VED("MSVDX: conceal to rotate surface\n"); + } else { + VED_CMDPORT_WRITE(dev_priv, + MSVDX_CMDS_ALTERNATIVE_OUTPUT_PICTURE_ROTATION_OFFSET, + cmd, cmd_space); + + VED_CMDPORT_WRITE(dev_priv, + MSVDX_CMDS_VC1_LUMA_RANGE_MAPPING_BASE_ADDRESS_OFFSET, + 0, cmd_space); + + VED_CMDPORT_WRITE(dev_priv, + MSVDX_CMDS_VC1_CHROMA_RANGE_MAPPING_BASE_ADDRESS_OFFSET, + 0, cmd_space); + + VED_CMDPORT_WRITE(dev_priv, + MSVDX_CMDS_VC1_RANGE_MAPPING_FLAGS_OFFSET, + 0, cmd_space); + } + + cmd = deblock_msg->ext_stride_a; + VED_CMDPORT_WRITE(dev_priv, + MSVDX_CMDS_EXTENDED_ROW_STRIDE_OFFSET, + cmd, cmd_space); + + for (loop = 0; loop < fault_region->num_region; loop++) { + + uint32_t start = fault_region->mb_regions[loop].start; + uint32_t end = fault_region->mb_regions[loop].end; + uint32_t x, y; + + IPVR_DEBUG_VED("MSVDX: region(%d) is %d~%d\n", + loop, start, end); + + if (conceal_above_row) + start -= width_in_mb; + if (end > (width_in_mb * height_in_mb - 1)) + end = (width_in_mb * height_in_mb - 1); + if (start > end) + start = 0; + + IPVR_DEBUG_VED("MSVDX: modify region(%d) is %d~%d\n", + loop, start, end); + + x = start % width_in_mb; + y = start / width_in_mb; + + for (mb_loop = start; mb_loop <= end; mb_loop++, x++) { + if (x >= width_in_mb) { + x = 0; + y++; + } + + /* IPVR_DEBUG_VED("MSVDX: concleament (%d,%d)\n", + x, y); */ + if ((x == 0) && (mb_loop != start)) + VED_CMDPORT_WRITE_FAST(dev_priv, + MSVDX_CMDS_END_SLICE_PICTURE_OFFSET, + 0, cmd_space); + cmd = 0; + REGIO_WRITE_FIELD_LITE(cmd, + MSVDX_CMDS_MACROBLOCK_NUMBER, + MB_CODE_TYPE, 1); + REGIO_WRITE_FIELD_LITE(cmd, + MSVDX_CMDS_MACROBLOCK_NUMBER, + MB_NO_X, x); + REGIO_WRITE_FIELD_LITE(cmd, + MSVDX_CMDS_MACROBLOCK_NUMBER, + MB_NO_Y, y); + VED_CMDPORT_WRITE_FAST(dev_priv, + MSVDX_CMDS_MACROBLOCK_NUMBER_OFFSET, + cmd, cmd_space); + VED_CMDPORT_WRITE_FAST(dev_priv, + MSVDX_CMDS_MACROBLOCK_RESIDUAL_FORMAT_OFFSET, + 0, cmd_space); + cmd = 0; + REGIO_WRITE_FIELD_LITE(cmd, + MSVDX_CMDS_INTER_BLOCK_PREDICTION, + REF_INDEX_A_VALID, 1); + REGIO_WRITE_FIELD_LITE(cmd, + MSVDX_CMDS_INTER_BLOCK_PREDICTION, + INTER_PRED_BLOCK_SIZE, 0); + REGIO_WRITE_FIELD_LITE(cmd, + MSVDX_CMDS_INTER_BLOCK_PREDICTION, + REF_INDEX_A, 0); + REGIO_WRITE_FIELD_LITE(cmd, + MSVDX_CMDS_INTER_BLOCK_PREDICTION, + REF_INDEX_B, 0); + VED_CMDPORT_WRITE_FAST(dev_priv, + MSVDX_CMDS_INTER_BLOCK_PREDICTION_OFFSET, + cmd, cmd_space); + VED_CMDPORT_WRITE_FAST(dev_priv, + MSVDX_CMDS_MOTION_VECTOR_OFFSET, + 0, cmd_space); + } + + VED_CMDPORT_WRITE(dev_priv, + MSVDX_CMDS_END_SLICE_PICTURE_OFFSET, + 0, cmd_space); + } + +ec_done: + /* try to unblock rendec */ + ret = VED_CMDPORT_WRITE_FAST(dev_priv, + MSVDX_CMDS_END_SLICE_PICTURE_OFFSET, + 1, cmd_space); + + fault_region->num_region = 0; + + preempt_enable(); + + pm_ret = pm_runtime_put(&dev_priv->dev->platformdev->dev); + if (unlikely(pm_ret < 0)) + IPVR_ERROR("Error put VED power: %d\n", pm_ret); + + IPVR_DEBUG_VED("VED: EC done, unlock msvdx ret %d.\n", ret); + + return; +} + +struct ved_ec_context *ved_find_ec_ctx(struct ved_private *ved_priv, + struct drm_file *tfile, void *cmd) +{ + int32_t i, free_idx; + struct ved_ec_context *ec_ctx = NULL; + struct fw_deblock_msg *deblock_msg = (struct fw_deblock_msg *)cmd; + + free_idx = -1; + for (i = 0; i < VED_MAX_EC_INSTANCE; i++) { + if (ved_priv->ved_ec_ctx[i]->tfile == tfile) + break; + else if (free_idx < 0 && + ved_priv->ved_ec_ctx[i]->tfile == NULL) + free_idx = i; + } + + if (i < VED_MAX_EC_INSTANCE) + ec_ctx = ved_priv->ved_ec_ctx[i]; + else if (free_idx >= 0 && cmd) { + IPVR_DEBUG_VED("acquire ec ctx idx %d for tfile 0x%08lx.\n", + free_idx, (unsigned long)tfile); + ec_ctx = ved_priv->ved_ec_ctx[free_idx]; + memset(ec_ctx, 0, sizeof(*ec_ctx)); + ec_ctx->tfile = tfile; + ec_ctx->context_id = deblock_msg->mmu_context.bits.context; + } else { + IPVR_DEBUG_VED("Available ec ctx is not found.\n"); + } + + return ec_ctx; +} + +void ved_update_frame_info(struct ved_private *ved_priv, + struct drm_file *tfile, void *cmd) +{ + + int32_t i, free_idx; + struct ved_frame_info *frame_info; + struct fw_deblock_msg *deblock_msg = (struct fw_deblock_msg *)cmd; + uint32_t buffer_handle = deblock_msg->mb_param_address; + + struct ved_ec_context *ec_ctx; + + IPVR_DEBUG_VED( + "update frame info (handle 0x%08x) for error concealment\n", + buffer_handle); + + ec_ctx = ved_find_ec_ctx(ved_priv, tfile, cmd); + + if (!ec_ctx) + return; + + free_idx = -1; + for (i = 0; i < MAX_DECODE_BUFFERS; i++) { + if (buffer_handle == ec_ctx->frame_info[i].handle) + break; + if (free_idx < 0 && ec_ctx->frame_info[i].handle == 0) + free_idx = i; + } + + if (i < MAX_DECODE_BUFFERS) + frame_info = &ec_ctx->frame_info[i]; + else if (free_idx >= 0) { + IPVR_DEBUG_VED("acquire frame_info solt idx %d\n", free_idx); + frame_info = &ec_ctx->frame_info[free_idx]; + } else { + IPVR_DEBUG_VED("%d solts occupied, abort update frame_info\n", + MAX_DECODE_BUFFERS); + return; + } + + frame_info->fw_status = 0; + frame_info->handle = buffer_handle; + frame_info->fence = (deblock_msg->header.bits.msg_fence & (~0xf)); + frame_info->decode_status.num_region = 0; + ec_ctx->cur_frame_info = frame_info; +} + +void ved_backup_cmd(struct ved_private *ved_priv, struct drm_file *tfile, + void *cmd, uint32_t cmd_size, + uint32_t deblock_cmd_offset) +{ + struct fw_deblock_msg *deblock_msg = NULL; + + struct ved_ec_context *ec_ctx; + union msg_header *header; + + IPVR_DEBUG_VED("backup cmd for ved error concealment\n"); + + ec_ctx = ved_find_ec_ctx(ved_priv, tfile, NULL); + + if (!ec_ctx) { + IPVR_DEBUG_VED("this is not a ec ctx, abort backup cmd\n"); + return; + } + + if (deblock_cmd_offset != VED_INVALID_OFFSET) + deblock_msg = (struct fw_deblock_msg *) + (cmd + deblock_cmd_offset); + + if (deblock_msg && + ec_ctx->context_id != deblock_msg->mmu_context.bits.context) { + IPVR_DEBUG_VED("backup cmd but find mis-match context id\n"); + return; + } + + ec_ctx->cmd_size = cmd_size; + ec_ctx->deblock_cmd_offset = deblock_cmd_offset; + memcpy(ec_ctx->unfenced_cmd, cmd, cmd_size); + ec_ctx->fence = VED_INVALID_FENCE; + header = (union msg_header *)ec_ctx->unfenced_cmd; + if (cmd_size) + ec_ctx->fence = header->bits.msg_fence; + ec_ctx->fence &= (~0xf); + IPVR_DEBUG_VED("backup cmd for ved: fence 0x%08x, cmd_size %d\n", + ec_ctx->fence, cmd_size); +} + +void ved_mtx_message_dump(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + int32_t i, buf_size, buf_offset; + buf_size = + VED_REG_READ32(MSVDX_COMMS_TO_HOST_BUF_SIZE) & ((1 << 16) - 1); + buf_offset = + (VED_REG_READ32(MSVDX_COMMS_TO_HOST_BUF_SIZE) >> 16) + 0x2000; + + IPVR_DEBUG_VED("Dump to HOST message buffer (offset:size)%04x:%04x\n", + buf_offset, buf_size); + for (i = 0; i < buf_size; i += 4) { + uint32_t reg1, reg2, reg3, reg4; + reg1 = VED_REG_READ32(buf_offset + i * 4); + reg2 = VED_REG_READ32(buf_offset + i * 4 + 4); + reg3 = VED_REG_READ32(buf_offset + i * 4 + 8); + reg4 = VED_REG_READ32(buf_offset + i * 4 + 12); + IPVR_DEBUG_VED("VED: 0x%04x: 0x%08x 0x%08x 0x%08x 0x%08x\n", + (buf_offset + i * 4), reg1, reg2, reg3, reg4); + } + + buf_size = VED_REG_READ32(MSVDX_COMMS_TO_MTX_BUF_SIZE) & ((1 << 16) - 1); + buf_offset = (VED_REG_READ32(MSVDX_COMMS_TO_MTX_BUF_SIZE) >> 16) + 0x2000; + + IPVR_DEBUG_VED("Dump to MTX message buffer (offset:size)%04x:%04x\n", + buf_offset, buf_size); + for (i = 0; i < buf_size; i += 4) { + uint32_t reg1, reg2, reg3, reg4; + reg1 = VED_REG_READ32(buf_offset + i * 4); + reg2 = VED_REG_READ32(buf_offset + i * 4 + 4); + reg3 = VED_REG_READ32(buf_offset + i * 4 + 8); + reg4 = VED_REG_READ32(buf_offset + i * 4 + 12); + IPVR_DEBUG_VED("VED: 0x%04x: 0x%08x 0x%08x 0x%08x 0x%08x\n", + (buf_offset + i * 4), reg1, reg2, reg3, reg4); + } + + buf_size = 12; + buf_offset = 0xFD0 + 0x2000; + + IPVR_DEBUG_VED("VED: Comm header (offset:size)%04x:%04x\n", + buf_offset, buf_size); + for (i = 0; i < buf_size; i += 4) { + uint32_t reg1, reg2, reg3, reg4; + reg1 = VED_REG_READ32(buf_offset + i * 4); + reg2 = VED_REG_READ32(buf_offset + i * 4 + 4); + reg3 = VED_REG_READ32(buf_offset + i * 4 + 8); + reg4 = VED_REG_READ32(buf_offset + i * 4 + 12); + IPVR_DEBUG_VED("VED: 0x%04x: 0x%08x 0x%08x 0x%08x 0x%08x\n", + (buf_offset + i * 4), reg1, reg2, reg3, reg4); + } + + IPVR_DEBUG_VED("VED: Error status 0x2cc4: 0x%08x\n", + VED_REG_READ32(0x2cc4)); +} + +void ved_fw_error_detected(struct drm_device *dev, + uint32_t fence, uint32_t flags) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ved_private *ved_priv = dev_priv->ved_private; + struct ved_ec_context *ved_ec_ctx = NULL; + struct ved_frame_info *frame_info = NULL; + int32_t i, found = 0; + + if (!(flags & FW_DEVA_ERROR_DETECTED)) + return; + + /*get the frame_info struct for error concealment frame*/ + for (i = 0; i < VED_MAX_EC_INSTANCE; i++) + if (ved_priv->ved_ec_ctx[i]->fence == + (fence & (~0xf))) { + ved_ec_ctx = ved_priv->ved_ec_ctx[i]; + found++; + } + /* ved_mtx_message_dump(dev); */ + if (!ved_ec_ctx || !(ved_ec_ctx->tfile) || found > 1) { + IPVR_DEBUG_VED( + "no matched ctx: fence 0x%x, found %d, ctx 0x%08lx\n", + fence, found, (unsigned long)ved_ec_ctx); + return; + } + + if (ved_ec_ctx->cur_frame_info && + ved_ec_ctx->cur_frame_info->fence == (fence & (~0xf))) { + frame_info = ved_ec_ctx->cur_frame_info; + } else { + if (ved_ec_ctx->cur_frame_info) + IPVR_DEBUG_VED( + "cur_frame_info fence(%x) doesn't match fence (%x).\n", + ved_ec_ctx->cur_frame_info->fence, fence); + else + IPVR_DEBUG_VED( + "The pointer ved_ec_ctx->cur_frame_info is null\n"); + return; + } + + if (frame_info->decode_status.num_region) { + IPVR_DEBUG_VED("Error already record, no need record again\n"); + return; + } + + IPVR_DEBUG_VED("record error as first fault region.\n"); + frame_info->decode_status.num_region++; + frame_info->decode_status.mb_regions[0].start = 0; + frame_info->decode_status.mb_regions[0].end = 0; + + /* + for (i = 0; i < MAX_DECODE_BUFFERS; i++) { + if (ved_ec_ctx->frame_info[i].fence == (fence & (~0xf))) { + break; + } + + } + */ +} diff --git a/drivers/gpu/drm/ipvr/ved_ec.h b/drivers/gpu/drm/ipvr/ved_ec.h new file mode 100644 index 0000000..c5f9fe6 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ved_ec.h @@ -0,0 +1,207 @@ +/************************************************************************** + * ved_ec.h: VED error concealment header file + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * Copyright (c) Imagination Technologies Limited, UK + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Li Zeng li.zeng@intel.com + * + **************************************************************************/ + +#ifndef _VED_EC_H_ +#define _VED_EC_H_ + +#include "ipvr_drv.h" + +#define VED_INVALID_FENCE (0xffff) +#define VED_INVALID_OFFSET (0xffffffff) +#define VED_EC_ROLLBACK (9) + +struct ved_ec_context { + struct drm_file *tfile; /* protected by cmdbuf_mutex */ + uint32_t context_id; + struct ved_frame_info frame_info[MAX_DECODE_BUFFERS]; + struct ved_frame_info *cur_frame_info; + int32_t frame_idx; + + /* 12 render msg + 1 deblock msg + * 12 * 20 + 1 * 48 = 288; + */ + uint8_t unfenced_cmd[300]; + uint32_t cmd_size; + uint32_t deblock_cmd_offset; + uint16_t fence; + struct ved_decode_status decode_status; +}; + +#define _PSB_MSVDX_EC_H_ + +#define MSVDX_CMDS_BASE 0x1000 +#define MSVDX_CMDS_DISPLAY_PICTURE_SIZE_OFFSET (0x0000) + +/* MSVDX_CMDS, DISPLAY_PICTURE_SIZE, DISPLAY_PICTURE_HEIGHT */ +#define MSVDX_CMDS_DISPLAY_PICTURE_SIZE_DISPLAY_PICTURE_HEIGHT_MASK \ + (0x00FFF000) +#define MSVDX_CMDS_DISPLAY_PICTURE_SIZE_DISPLAY_PICTURE_HEIGHT_SHIFT (12) + +/* MSVDX_CMDS, DISPLAY_PICTURE_SIZE, DISPLAY_PICTURE_WIDTH */ +#define MSVDX_CMDS_DISPLAY_PICTURE_SIZE_DISPLAY_PICTURE_WIDTH_MASK (0x00000FFF) +#define MSVDX_CMDS_DISPLAY_PICTURE_SIZE_DISPLAY_PICTURE_WIDTH_SHIFT (0) + +#define MSVDX_CMDS_CODED_PICTURE_SIZE_OFFSET (0x0004) + +/* MSVDX_CMDS, CODED_PICTURE_SIZE, CODED_PICTURE_HEIGHT */ +#define MSVDX_CMDS_CODED_PICTURE_SIZE_CODED_PICTURE_HEIGHT_MASK (0x00FFF000) +#define MSVDX_CMDS_CODED_PICTURE_SIZE_CODED_PICTURE_HEIGHT_SHIFT (12) + +/* MSVDX_CMDS, CODED_PICTURE_SIZE, CODED_PICTURE_WIDTH */ +#define MSVDX_CMDS_CODED_PICTURE_SIZE_CODED_PICTURE_WIDTH_MASK (0x00000FFF) +#define MSVDX_CMDS_CODED_PICTURE_SIZE_CODED_PICTURE_WIDTH_SHIFT (0) + +#define MSVDX_CMDS_OPERATING_MODE_OFFSET (0x0008) + +/* MSVDX_CMDS, OPERATING_MODE, RPR_ENABLE */ +#define MSVDX_CMDS_OPERATING_MODE_RPR_ENABLE_MASK (0x20000000) +#define MSVDX_CMDS_OPERATING_MODE_RPR_ENABLE_SHIFT (29) + +/* MSVDX_CMDS, OPERATING_MODE, USE_EXT_ROW_STRIDE */ +#define MSVDX_CMDS_OPERATING_MODE_USE_EXT_ROW_STRIDE_MASK (0x10000000) +#define MSVDX_CMDS_OPERATING_MODE_USE_EXT_ROW_STRIDE_SHIFT (28) + +/* MSVDX_CMDS, OPERATING_MODE, CHROMA_INTERLEAVED */ +#define MSVDX_CMDS_OPERATING_MODE_CHROMA_INTERLEAVED_MASK (0x08000000) +#define MSVDX_CMDS_OPERATING_MODE_CHROMA_INTERLEAVED_SHIFT (27) +/* MSVDX_CMDS, OPERATING_MODE, ROW_STRIDE */ +#define MSVDX_CMDS_OPERATING_MODE_ROW_STRIDE_MASK (0x07000000) +#define MSVDX_CMDS_OPERATING_MODE_ROW_STRIDE_SHIFT (24) + +/* MSVDX_CMDS, OPERATING_MODE, CODEC_PROFILE */ +#define MSVDX_CMDS_OPERATING_MODE_CODEC_PROFILE_MASK (0x00300000) +#define MSVDX_CMDS_OPERATING_MODE_CODEC_PROFILE_SHIFT (20) + +/* MSVDX_CMDS, OPERATING_MODE, CODEC_MODE */ +#define MSVDX_CMDS_OPERATING_MODE_CODEC_MODE_MASK (0x000F0000) +#define MSVDX_CMDS_OPERATING_MODE_CODEC_MODE_SHIFT (16) + +/* MSVDX_CMDS, OPERATING_MODE, ASYNC_MODE */ +#define MSVDX_CMDS_OPERATING_MODE_ASYNC_MODE_MASK (0x00006000) +#define MSVDX_CMDS_OPERATING_MODE_ASYNC_MODE_SHIFT (13) + +/* MSVDX_CMDS, OPERATING_MODE, CHROMA_FORMAT */ +#define MSVDX_CMDS_OPERATING_MODE_CHROMA_FORMAT_MASK (0x00001000) +#define MSVDX_CMDS_OPERATING_MODE_CHROMA_FORMAT_SHIFT (12) + +/* MSVDX_CMDS, OPERATING_MODE, INTERLACED */ +#define MSVDX_CMDS_OPERATING_MODE_INTERLACED_MASK (0x00000800) +#define MSVDX_CMDS_OPERATING_MODE_INTERLACED_SHIFT (11) + +/* MSVDX_CMDS, OPERATING_MODE, OVERLAP */ +#define MSVDX_CMDS_OPERATING_MODE_OVERLAP_MASK (0x00000400) +#define MSVDX_CMDS_OPERATING_MODE_OVERLAP_SHIFT (10) + +/* MSVDX_CMDS, OPERATING_MODE, PIC_CONDOVER */ +#define MSVDX_CMDS_OPERATING_MODE_PIC_CONDOVER_MASK (0x00000300) +#define MSVDX_CMDS_OPERATING_MODE_PIC_CONDOVER_SHIFT (8) +/* MSVDX_CMDS, OPERATING_MODE, DEBLOCK_STRENGTH */ +#define MSVDX_CMDS_OPERATING_MODE_DEBLOCK_STRENGTH_MASK (0x000000E0) +#define MSVDX_CMDS_OPERATING_MODE_DEBLOCK_STRENGTH_SHIFT (5) + +/* MSVDX_CMDS, OPERATING_MODE, PIC_QUANT */ +#define MSVDX_CMDS_OPERATING_MODE_PIC_QUANT_MASK (0x0000001F) +#define MSVDX_CMDS_OPERATING_MODE_PIC_QUANT_SHIFT (0) + +#define MSVDX_CMDS_LUMA_RECONSTRUCTED_PICTURE_BASE_ADDRESSES_OFFSET (0x000C) +#define MSVDX_CMDS_CHROMA_RECONSTRUCTED_PICTURE_BASE_ADDRESSES_OFFSET (0x0010) + +#define MSVDX_CMDS_REFERENCE_PICTURE_BASE_ADDRESSES_OFFSET (0x0100) + +#define MSVDX_CMDS_SLICE_PARAMS_OFFSET (0x0400) + +/* MSVDX_CMDS, SLICE_PARAMS, SLICE_FIELD_TYPE */ +#define MSVDX_CMDS_SLICE_PARAMS_SLICE_FIELD_TYPE_MASK (0x0000000C) +#define MSVDX_CMDS_SLICE_PARAMS_SLICE_FIELD_TYPE_SHIFT (2) + + +/* MSVDX_CMDS, SLICE_PARAMS, SLICE_CODE_TYPE */ +#define MSVDX_CMDS_SLICE_PARAMS_SLICE_CODE_TYPE_MASK (0x00000003) +#define MSVDX_CMDS_SLICE_PARAMS_SLICE_CODE_TYPE_SHIFT (0) + +#define MSVDX_CMDS_ALTERNATIVE_OUTPUT_PICTURE_ROTATION_OFFSET (0x003C) + +#define MSVDX_CMDS_VC1_LUMA_RANGE_MAPPING_BASE_ADDRESS_OFFSET (0x0028) +#define MSVDX_CMDS_VC1_CHROMA_RANGE_MAPPING_BASE_ADDRESS_OFFSET (0x002C) +#define MSVDX_CMDS_VC1_RANGE_MAPPING_FLAGS_OFFSET (0x0030) + +#define MSVDX_CMDS_EXTENDED_ROW_STRIDE_OFFSET (0x0040) + +#define MSVDX_CMDS_MACROBLOCK_NUMBER_OFFSET (0x0408) + +/* MSVDX_CMDS, MACROBLOCK_NUMBER, MB_CODE_TYPE */ +#define MSVDX_CMDS_MACROBLOCK_NUMBER_MB_CODE_TYPE_MASK (0x00030000) +#define MSVDX_CMDS_MACROBLOCK_NUMBER_MB_CODE_TYPE_SHIFT (16) + +/* MSVDX_CMDS, MACROBLOCK_NUMBER, MB_NO_Y */ +#define MSVDX_CMDS_MACROBLOCK_NUMBER_MB_NO_Y_MASK (0x0000FF00) +#define MSVDX_CMDS_MACROBLOCK_NUMBER_MB_NO_Y_SHIFT (8) + +/* MSVDX_CMDS, MACROBLOCK_NUMBER, MB_NO_X */ +#define MSVDX_CMDS_MACROBLOCK_NUMBER_MB_NO_X_MASK (0x000000FF) +#define MSVDX_CMDS_MACROBLOCK_NUMBER_MB_NO_X_SHIFT (0) + +#define MSVDX_CMDS_MACROBLOCK_RESIDUAL_FORMAT_OFFSET (0x0418) + +#define MSVDX_CMDS_INTER_BLOCK_PREDICTION_OFFSET (0x0430) + +/* MSVDX_CMDS, INTER_BLOCK_PREDICTION, REF_INDEX_A_VALID */ +#define MSVDX_CMDS_INTER_BLOCK_PREDICTION_REF_INDEX_A_VALID_MASK (0x00000020) +#define MSVDX_CMDS_INTER_BLOCK_PREDICTION_REF_INDEX_A_VALID_SHIFT (5) + +/* MSVDX_CMDS, INTER_BLOCK_PREDICTION, INTER_PRED_BLOCK_SIZE */ +#define MSVDX_CMDS_INTER_BLOCK_PREDICTION_INTER_PRED_BLOCK_SIZE_MASK (0x70000) +#define MSVDX_CMDS_INTER_BLOCK_PREDICTION_INTER_PRED_BLOCK_SIZE_SHIFT (16) + +/* MSVDX_CMDS, INTER_BLOCK_PREDICTION, REF_INDEX_A */ +#define MSVDX_CMDS_INTER_BLOCK_PREDICTION_REF_INDEX_A_MASK (0x0000000F) +#define MSVDX_CMDS_INTER_BLOCK_PREDICTION_REF_INDEX_A_SHIFT (0) + +/* MSVDX_CMDS, INTER_BLOCK_PREDICTION, REF_INDEX_B */ +#define MSVDX_CMDS_INTER_BLOCK_PREDICTION_REF_INDEX_B_MASK (0x00000F00) +#define MSVDX_CMDS_INTER_BLOCK_PREDICTION_REF_INDEX_B_SHIFT (8) + +#define MSVDX_CMDS_MOTION_VECTOR_OFFSET (0x0500) + +#define MSVDX_CORE_CR_MSVDX_COMMAND_SPACE_OFFSET (0x0028) + +#define MSVDX_CORE_BASE (0x600) + +void ved_update_frame_info(struct ved_private *ved_priv, + struct drm_file *tfile, void *cmd); + +void ved_backup_cmd(struct ved_private *ved_priv, + struct drm_file *tfile, void *cmd, + uint32_t cmd_size, uint32_t deblock_cmd_offset); + +void ved_mtx_message_dump(struct drm_device *dev); + +void ved_do_concealment(struct work_struct *work); + +void ved_fw_error_detected(struct drm_device *dev, + uint32_t fence, uint32_t flags); + +#endif diff --git a/drivers/gpu/drm/ipvr/ved_fw.c b/drivers/gpu/drm/ipvr/ved_fw.c new file mode 100644 index 0000000..47fbad4 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ved_fw.c @@ -0,0 +1,660 @@ +/************************************************************************** + * ved_fw.c: VED initialization and mtx-firmware upload + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * Copyright (c) Imagination Technologies Limited, UK + * Copyright (c) 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * + **************************************************************************/ + +#include "ved_fw.h" +#include "ipvr_gem.h" +#include "ved_cmd.h" +#include "ved_reg.h" +#include "ved_init.h" +#include "ipvr_buffer.h" +#include "ipvr_mmu.h" +#include <linux/firmware.h> +#include <asm/cacheflush.h> +#include <linux/module.h> + +#define UPLOAD_FW_BY_DMA 1 +#define STACKGUARDWORD 0x10101010 +#define MSVDX_MTX_DATA_LOCATION 0x82880000 +#define UNINITILISE_MEM 0xcdcdcdcd +#define FIRMWARE_NAME "msvdx_fw_mfld_DE2.0.bin" + +/* VED FW header */ +struct ved_fw { + uint32_t ver; + uint32_t text_size; + uint32_t data_size; + uint32_t data_location; +}; + +#if UPLOAD_FW_BY_DMA + +static void ved_get_mtx_control_from_dash(struct drm_ipvr_private *dev_priv) +{ + struct ved_private *ved_priv = dev_priv->ved_private; + int32_t count = 0; + uint32_t reg_val = 0; + + REGIO_WRITE_FIELD(reg_val, MSVDX_MTX_DEBUG, MTX_DBG_IS_SLAVE, 1); + REGIO_WRITE_FIELD(reg_val, MSVDX_MTX_DEBUG, MTX_DBG_GPIO_IN, 0x02); + VED_REG_WRITE32(reg_val, MSVDX_MTX_DEBUG_OFFSET); + + do { + reg_val = VED_REG_READ32(MSVDX_MTX_DEBUG_OFFSET); + count++; + } while (((reg_val & 0x18) != 0) && count < 50000); + + if (count >= 50000) + IPVR_DEBUG_VED("VED: timeout in get_mtx_control_from_dash.\n"); + + /* Save the access control register...*/ + ved_priv->ved_dash_access_ctrl = VED_REG_READ32(MTX_RAM_ACCESS_CONTROL_OFFSET); +} + +static void +ved_release_mtx_control_from_dash(struct drm_ipvr_private *dev_priv) +{ + struct ved_private *ved_priv = dev_priv->ved_private; + + /* restore access control */ + VED_REG_WRITE32(ved_priv->ved_dash_access_ctrl, MTX_RAM_ACCESS_CONTROL_OFFSET); + /* release bus */ + VED_REG_WRITE32(0x4, MSVDX_MTX_DEBUG_OFFSET); +} + +/* for future debug info of msvdx related registers */ +static void +ved_setup_fw_dump(struct drm_ipvr_private *dev_priv, uint32_t dma_channel) +{ + IPVR_DEBUG_REG("dump registers during fw upload for debug:\n"); + /* for DMAC REGISTER */ + IPVR_DEBUG_REG("MTX_SYSC_CDMAA is 0x%x\n", + VED_REG_READ32(MTX_SYSC_CDMAA_OFFSET)); + IPVR_DEBUG_REG("MTX_SYSC_CDMAC value is 0x%x\n", + VED_REG_READ32(MTX_SYSC_CDMAC_OFFSET)); + IPVR_DEBUG_REG("DMAC_SETUP value is 0x%x\n", + VED_REG_READ32(DMAC_DMAC_SETUP_OFFSET + dma_channel)); + IPVR_DEBUG_REG("DMAC_DMAC_COUNT value is 0x%x\n", + VED_REG_READ32(DMAC_DMAC_COUNT_OFFSET + dma_channel)); + IPVR_DEBUG_REG("DMAC_DMAC_PERIPH_OFFSET value is 0x%x\n", + VED_REG_READ32(DMAC_DMAC_PERIPH_OFFSET + dma_channel)); + IPVR_DEBUG_REG("DMAC_DMAC_PERIPHERAL_ADDR value is 0x%x\n", + VED_REG_READ32(DMAC_DMAC_PERIPHERAL_ADDR_OFFSET + + dma_channel)); + IPVR_DEBUG_REG("MSVDX_CONTROL value is 0x%x\n", + VED_REG_READ32(MSVDX_CONTROL_OFFSET)); + IPVR_DEBUG_REG("DMAC_DMAC_IRQ_STAT value is 0x%x\n", + VED_REG_READ32(DMAC_DMAC_IRQ_STAT_OFFSET)); + IPVR_DEBUG_REG("MSVDX_MMU_CONTROL0 value is 0x%x\n", + VED_REG_READ32(MSVDX_MMU_CONTROL0_OFFSET)); + IPVR_DEBUG_REG("DMAC_DMAC_COUNT 2222 value is 0x%x\n", + VED_REG_READ32(DMAC_DMAC_COUNT_OFFSET + dma_channel)); + + /* for MTX REGISTER */ + IPVR_DEBUG_REG("MTX_ENABLE_OFFSET is 0x%x\n", + VED_REG_READ32(MTX_ENABLE_OFFSET)); + IPVR_DEBUG_REG("MTX_KICK_INPUT_OFFSET value is 0x%x\n", + VED_REG_READ32(MTX_KICK_INPUT_OFFSET)); + IPVR_DEBUG_REG("MTX_REG_READ_WRITE_REQUEST_OFFSET value is 0x%x\n", + VED_REG_READ32(MTX_REGISTER_READ_WRITE_REQUEST_OFFSET)); + IPVR_DEBUG_REG("MTX_RAM_ACCESS_CONTROL_OFFSET value is 0x%x\n", + VED_REG_READ32(MTX_RAM_ACCESS_CONTROL_OFFSET)); + IPVR_DEBUG_REG("MTX_RAM_ACCESS_STATUS_OFFSET value is 0x%x\n", + VED_REG_READ32(MTX_RAM_ACCESS_STATUS_OFFSET)); + IPVR_DEBUG_REG("MTX_SYSC_TIMERDIV_OFFSET value is 0x%x\n", + VED_REG_READ32(MTX_SYSC_TIMERDIV_OFFSET)); + IPVR_DEBUG_REG("MTX_SYSC_CDMAC_OFFSET value is 0x%x\n", + VED_REG_READ32(MTX_SYSC_CDMAC_OFFSET)); + IPVR_DEBUG_REG("MTX_SYSC_CDMAA_OFFSET value is 0x%x\n", + VED_REG_READ32(MTX_SYSC_CDMAA_OFFSET)); + IPVR_DEBUG_REG("MTX_SYSC_CDMAS0_OFFSET value is 0x%x\n", + VED_REG_READ32(MTX_SYSC_CDMAS0_OFFSET)); + IPVR_DEBUG_REG("MTX_SYSC_CDMAT_OFFSET value is 0x%x\n", + VED_REG_READ32(MTX_SYSC_CDMAT_OFFSET)); + + /* for MSVDX CORE REGISTER */ + IPVR_DEBUG_REG("MSVDX_CONTROL_OFFSET is 0x%x\n", + VED_REG_READ32(MSVDX_CONTROL_OFFSET)); + IPVR_DEBUG_REG("MSVDX_INTERRUPT_CLEAR_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_INTERRUPT_CLEAR_OFFSET)); + IPVR_DEBUG_REG("MSVDX_INTERRUPT_STATUS_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_INTERRUPT_STATUS_OFFSET)); + IPVR_DEBUG_REG("MMSVDX_HOST_INTERRUPT_ENABLE_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_HOST_INTERRUPT_ENABLE_OFFSET)); + IPVR_DEBUG_REG("MSVDX_MAN_CLK_ENABLE_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_MAN_CLK_ENABLE_OFFSET)); + IPVR_DEBUG_REG("MSVDX_CORE_ID_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_CORE_ID_OFFSET)); + IPVR_DEBUG_REG("MSVDX_MMU_STATUS_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_MMU_STATUS_OFFSET)); + IPVR_DEBUG_REG("FE_MSVDX_WDT_CONTROL_OFFSET value is 0x%x\n", + VED_REG_READ32(FE_MSVDX_WDT_CONTROL_OFFSET)); + IPVR_DEBUG_REG("FE_MSVDX_WDTIMER_OFFSET value is 0x%x\n", + VED_REG_READ32(FE_MSVDX_WDTIMER_OFFSET)); + IPVR_DEBUG_REG("BE_MSVDX_WDT_CONTROL_OFFSET value is 0x%x\n", + VED_REG_READ32(BE_MSVDX_WDT_CONTROL_OFFSET)); + IPVR_DEBUG_REG("BE_MSVDX_WDTIMER_OFFSET value is 0x%x\n", + VED_REG_READ32(BE_MSVDX_WDTIMER_OFFSET)); + + /* for MSVDX RENDEC REGISTER */ + IPVR_DEBUG_REG("VEC_SHIFTREG_CONTROL_OFFSET is 0x%x\n", + VED_REG_READ32(VEC_SHIFTREG_CONTROL_OFFSET)); + IPVR_DEBUG_REG("MSVDX_RENDEC_CONTROL0_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_RENDEC_CONTROL0_OFFSET)); + IPVR_DEBUG_REG("MSVDX_RENDEC_BUFFER_SIZE_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_RENDEC_BUFFER_SIZE_OFFSET)); + IPVR_DEBUG_REG("MSVDX_RENDEC_BASE_ADDR0_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_RENDEC_BASE_ADDR0_OFFSET)); + IPVR_DEBUG_REG("MMSVDX_RENDEC_BASE_ADDR1_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_RENDEC_BASE_ADDR1_OFFSET)); + IPVR_DEBUG_REG("MSVDX_RENDEC_READ_DATA_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_RENDEC_READ_DATA_OFFSET)); + IPVR_DEBUG_REG("MSVDX_RENDEC_CONTEXT0_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_RENDEC_CONTEXT0_OFFSET)); + IPVR_DEBUG_REG("MSVDX_RENDEC_CONTEXT1_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_RENDEC_CONTEXT1_OFFSET)); + IPVR_DEBUG_REG("MSVDX_CMDS_END_SLICE_PICTURE_OFFSET value is 0x%x\n", + VED_REG_READ32(MSVDX_CMDS_END_SLICE_PICTURE_OFFSET)); + + IPVR_DEBUG_REG("MSVDX_MMU_MEM_REQ value is 0x%x\n", + VED_REG_READ32(MSVDX_MMU_MEM_REQ_OFFSET)); + IPVR_DEBUG_REG("MSVDX_SYS_MEMORY_DEBUG2 value is 0x%x\n", + VED_REG_READ32(0x6fc)); +} + +static void ved_upload_fw(struct drm_ipvr_private *dev_priv, + uint32_t address, const uint32_t words) +{ + uint32_t reg_val = 0; + uint32_t cmd; + uint32_t uCountReg, offset, mmu_ptd; + uint32_t size = words * 4; /* byte count */ + uint32_t dma_channel = 0; /* Setup a Simple DMA for Ch0 */ + struct ved_private *ved_priv = dev_priv->ved_private; + + IPVR_DEBUG_VED("VED: Upload firmware by DMA.\n"); + ved_get_mtx_control_from_dash(dev_priv); + + /* + * dma transfers to/from the mtx have to be 32-bit aligned and + * in multiples of 32 bits + */ + VED_REG_WRITE32(address, MTX_SYSC_CDMAA_OFFSET); + + /* burst size in multiples of 64 bits (allowed values are 2 or 4) */ + REGIO_WRITE_FIELD_LITE(reg_val, MTX_SYSC_CDMAC, BURSTSIZE, 4); + /* false means write to mtx mem, true means read from mtx mem */ + REGIO_WRITE_FIELD_LITE(reg_val, MTX_SYSC_CDMAC, RNW, 0); + /* begin transfer */ + REGIO_WRITE_FIELD_LITE(reg_val, MTX_SYSC_CDMAC, ENABLE, 1); + /* specifies transfer size of the DMA operation by 32-bit words */ + REGIO_WRITE_FIELD_LITE(reg_val, MTX_SYSC_CDMAC, LENGTH, words); + VED_REG_WRITE32(reg_val, MTX_SYSC_CDMAC_OFFSET); + + /* toggle channel 0 usage between mtx and other msvdx peripherals */ + { + reg_val = VED_REG_READ32(MSVDX_CONTROL_OFFSET); + REGIO_WRITE_FIELD(reg_val, MSVDX_CONTROL, DMAC_CH0_SELECT, 0); + VED_REG_WRITE32(reg_val, MSVDX_CONTROL_OFFSET); + } + + /* Clear the DMAC Stats */ + VED_REG_WRITE32(0 , DMAC_DMAC_IRQ_STAT_OFFSET + dma_channel); + + offset = ved_priv->fw_offset; + IPVR_DEBUG_VED("fw gpu offset is 0x%x.\n", offset); + + /* use bank 0 */ + cmd = 0; + VED_REG_WRITE32(cmd, MSVDX_MMU_BANK_INDEX_OFFSET); + + /* Write PTD to mmu base 0*/ + mmu_ptd = ipvr_get_default_pd_addr(dev_priv->mmu); + VED_REG_WRITE32(mmu_ptd, MSVDX_MMU_DIR_LIST_BASE_OFFSET + 0); + IPVR_DEBUG_VED("mmu_ptd is %d.\n", mmu_ptd); + + /* Invalidate */ + reg_val = VED_REG_READ32(MSVDX_MMU_CONTROL0_OFFSET); + reg_val &= ~0xf; + REGIO_WRITE_FIELD(reg_val, MSVDX_MMU_CONTROL0, MMU_INVALDC, 1); + VED_REG_WRITE32(reg_val, MSVDX_MMU_CONTROL0_OFFSET); + + VED_REG_WRITE32(offset, DMAC_DMAC_SETUP_OFFSET + dma_channel); + + /* Only use a single dma - assert that this is valid */ + if ((size / 4) >= (1 << 15)) { + IPVR_ERROR("DMA size beyond limit, abort firmware upload.\n"); + return; + } + + uCountReg = PSB_DMAC_VALUE_COUNT(PSB_DMAC_BSWAP_NO_SWAP, 0, + PSB_DMAC_DIR_MEM_TO_PERIPH, 0, + (size / 4)); + /* Set the number of bytes to dma*/ + VED_REG_WRITE32(uCountReg, DMAC_DMAC_COUNT_OFFSET + dma_channel); + + cmd = PSB_DMAC_VALUE_PERIPH_PARAM(PSB_DMAC_ACC_DEL_0, + PSB_DMAC_INCR_OFF, + PSB_DMAC_BURST_2); + VED_REG_WRITE32(cmd, DMAC_DMAC_PERIPH_OFFSET + dma_channel); + + /* Set destination port for dma */ + cmd = 0; + REGIO_WRITE_FIELD(cmd, DMAC_DMAC_PERIPHERAL_ADDR, ADDR, + MTX_SYSC_CDMAT_OFFSET); + VED_REG_WRITE32(cmd, DMAC_DMAC_PERIPHERAL_ADDR_OFFSET + dma_channel); + + + /* Finally, rewrite the count register with the enable bit set */ + VED_REG_WRITE32(uCountReg | DMAC_DMAC_COUNT_EN_MASK, + DMAC_DMAC_COUNT_OFFSET + dma_channel); + + /* Wait for all to be done */ + if (ved_wait_for_register(dev_priv, + DMAC_DMAC_IRQ_STAT_OFFSET + dma_channel, + DMAC_DMAC_IRQ_STAT_TRANSFER_FIN_MASK, + DMAC_DMAC_IRQ_STAT_TRANSFER_FIN_MASK, + 2000000, 5)) { + ved_setup_fw_dump(dev_priv, dma_channel); + ved_release_mtx_control_from_dash(dev_priv); + return; + } + + /* Assert that the MTX DMA port is all done aswell */ + if (ved_wait_for_register(dev_priv, + MTX_SYSC_CDMAS0_OFFSET, + 1, 1, 2000000, 5)) { + ved_release_mtx_control_from_dash(dev_priv); + return; + } + + ved_release_mtx_control_from_dash(dev_priv); + + IPVR_DEBUG_VED("VED: Upload done\n"); +} + +#else + +static void ved_upload_fw(struct drm_ipvr_private *dev_priv, + const uint32_t data_mem, + uint32_t ram_bank_size, uint32_t address, + const uint32_t words, + const uint32_t * const data) +{ + uint32_t loop, ctrl, ram_id, addr, cur_bank = (uint32_t) ~0; + uint32_t access_ctrl; + + IPVR_DEBUG_VED("VED: Upload firmware by register interface.\n"); + /* Save the access control register... */ + access_ctrl = VED_REG_READ32(MTX_RAM_ACCESS_CONTROL_OFFSET); + + /* Wait for MCMSTAT to become be idle 1 */ + ved_wait_for_register(dev_priv, MTX_RAM_ACCESS_STATUS_OFFSET, + 1, /* Required Value */ + 0xffffffff, /* Enables */ + 2000000, 5); + + for (loop = 0; loop < words; loop++) { + ram_id = data_mem + (address / ram_bank_size); + if (ram_id != cur_bank) { + addr = address >> 2; + ctrl = 0; + REGIO_WRITE_FIELD_LITE(ctrl, + MTX_RAM_ACCESS_CONTROL, + MTX_MCMID, ram_id); + REGIO_WRITE_FIELD_LITE(ctrl, + MTX_RAM_ACCESS_CONTROL, + MTX_MCM_ADDR, addr); + REGIO_WRITE_FIELD_LITE(ctrl, + MTX_RAM_ACCESS_CONTROL, + MTX_MCMAI, 1); + VED_REG_WRITE32(ctrl, MTX_RAM_ACCESS_CONTROL_OFFSET); + cur_bank = ram_id; + } + address += 4; + + VED_REG_WRITE32(data[loop], + MTX_RAM_ACCESS_DATA_TRANSFER_OFFSET); + + /* Wait for MCMSTAT to become be idle 1 */ + ved_wait_for_register(dev_priv, MTX_RAM_ACCESS_STATUS_OFFSET, + 1, /* Required Value */ + 0xffffffff, /* Enables */ + 2000000, 5); + } + IPVR_DEBUG_VED("VED: Upload done\n"); + + /* Restore the access control register... */ + VED_REG_WRITE32(access_ctrl, MSVDX_MTX_RAM_ACCESS_CONTROL); +} + +#endif + +static int32_t ved_get_fw_bo(struct drm_device *dev, + const struct firmware **raw, char *name) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + int32_t rc, fw_size; + void *ptr = NULL; + struct ved_private *ved_priv = dev_priv->ved_private; + void *fw_bo_addr = NULL; + uint32_t *last_word; + struct ved_fw *fw; + + rc = request_firmware(raw, name, &dev->platformdev->dev); + if (*raw == NULL || rc < 0) { + IPVR_ERROR("VED: %s request_firmware failed: Reason %d.\n", + name, rc); + return 1; + } + + if ((*raw)->size < sizeof(struct ved_fw)) { + IPVR_ERROR("VED: %s is is not correct size(%zd).\n", + name, (*raw)->size); + return 1; + } + + ptr = (void *)((*raw))->data; + if (!ptr) { + IPVR_ERROR("VED: Failed to load %s.\n", name); + return 1; + } + + /* another sanity check... */ + fw_size = sizeof(struct ved_fw) + + sizeof(uint32_t) * ((struct ved_fw *) ptr)->text_size + + sizeof(uint32_t) * ((struct ved_fw *) ptr)->data_size; + if ((*raw)->size < fw_size) { + IPVR_ERROR("VED: %s is is not correct size(%zd).\n", + name, (*raw)->size); + return 1; + } + + fw_bo_addr = ipvr_gem_object_vmap(ved_priv->fw_bo); + if (IS_ERR(fw_bo_addr)) { + IPVR_ERROR("VED: kmap failed for fw buffer.\n"); + return 1; + } + + fw = (struct ved_fw *)ptr; + memset(fw_bo_addr, UNINITILISE_MEM, ved_priv->mtx_mem_size); + memcpy(fw_bo_addr, ptr + sizeof(struct ved_fw), + sizeof(uint32_t) * fw->text_size); + memcpy(fw_bo_addr + (fw->data_location - MSVDX_MTX_DATA_LOCATION), + (void *)ptr + sizeof(struct ved_fw) + sizeof(uint32_t) * fw->text_size, + sizeof(uint32_t) * fw->data_size); + last_word = (uint32_t *)(fw_bo_addr + ved_priv->mtx_mem_size - 4); + /* + * Write a know value to last word in mtx memory + * Usefull for detection of stack overrun + */ + *last_word = STACKGUARDWORD; + + vunmap(fw_bo_addr); + IPVR_DEBUG_VED("VED: releasing firmware resouces.\n"); + IPVR_DEBUG_VED("VED: Load firmware into BO successfully.\n"); + release_firmware(*raw); + return rc; +} + +static uint32_t * +ved_get_fw(struct drm_device *dev, const struct firmware **raw, char *name) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + int32_t rc, fw_size; + void *ptr = NULL; + struct ved_fw *fw; + struct ved_private *ved_priv = dev_priv->ved_private; + + rc = request_firmware(raw, name, &dev->platformdev->dev); + if (*raw == NULL || rc < 0) { + IPVR_ERROR("VED: %s request_firmware failed: Reason %d\n", + name, rc); + return NULL; + } + + if ((*raw)->size < sizeof(struct ved_fw)) { + IPVR_ERROR("VED: %s is is not correct size(%zd)\n", + name, (*raw)->size); + release_firmware(*raw); + return NULL; + } + + ptr = (int *)((*raw))->data; + if (!ptr) { + IPVR_ERROR("VED: Failed to load %s.\n", name); + release_firmware(*raw); + return NULL; + } + fw = (struct ved_fw *)ptr; + + /* another sanity check... */ + fw_size = sizeof(fw) + + sizeof(uint32_t) * fw->text_size + + sizeof(uint32_t) * fw->data_size; + if ((*raw)->size < fw_size) { + IPVR_ERROR("VED: %s is is not correct size(%zd).\n", + name, (*raw)->size); + release_firmware(*raw); + return NULL; + } + + ved_priv->ved_fw_ptr = kzalloc(fw_size, GFP_KERNEL); + if (ved_priv->ved_fw_ptr == NULL) + IPVR_ERROR("VED: allocate FW buffer failed.\n"); + else { + memcpy(ved_priv->ved_fw_ptr, ptr, fw_size); + ved_priv->ved_fw_size = fw_size; + } + + IPVR_DEBUG_VED("VED: releasing firmware resouces.\n"); + release_firmware(*raw); + + return ved_priv->ved_fw_ptr; +} + +static void +ved_write_mtx_core_reg(struct drm_ipvr_private *dev_priv, + const uint32_t core_reg, const uint32_t val) +{ + uint32_t reg = 0; + + /* Put data in MTX_RW_DATA */ + VED_REG_WRITE32(val, MTX_REGISTER_READ_WRITE_DATA_OFFSET); + + /* DREADY is set to 0 and request a write */ + reg = core_reg; + REGIO_WRITE_FIELD_LITE(reg, MTX_REGISTER_READ_WRITE_REQUEST, + MTX_RNW, 0); + REGIO_WRITE_FIELD_LITE(reg, MTX_REGISTER_READ_WRITE_REQUEST, + MTX_DREADY, 0); + VED_REG_WRITE32(reg, MTX_REGISTER_READ_WRITE_REQUEST_OFFSET); + + ved_wait_for_register(dev_priv, + MTX_REGISTER_READ_WRITE_REQUEST_OFFSET, + MTX_REGISTER_READ_WRITE_REQUEST_MTX_DREADY_MASK, + MTX_REGISTER_READ_WRITE_REQUEST_MTX_DREADY_MASK, + 2000000, 5); +} + +int32_t ved_alloc_fw_bo(struct drm_ipvr_private *dev_priv) +{ + uint32_t core_rev; + struct ved_private *ved_priv = dev_priv->ved_private; + + core_rev = VED_REG_READ32(MSVDX_CORE_REV_OFFSET); + + if ((core_rev & 0xffffff) < 0x020000) + ved_priv->mtx_mem_size = 16 * 1024; + else + ved_priv->mtx_mem_size = 56 * 1024; + + IPVR_DEBUG_INIT("VED: MTX mem size is 0x%08x bytes," + "allocate firmware BO size 0x%08x.\n", + ved_priv->mtx_mem_size, + ved_priv->mtx_mem_size + 4096); + + /* Allocate the new object */ + ved_priv->fw_bo = ipvr_gem_obj_create_and_bind(dev_priv->dev, + ved_priv->mtx_mem_size + 4096); + if (ved_priv->fw_bo == NULL) { + IPVR_ERROR("VED: failed to allocate fw buffer.\n"); + ved_priv->fw_bo = NULL; + return -ENOMEM; + } + ved_priv->fw_offset = ipvr_gem_obj_mmu_offset(ved_priv->fw_bo); + if (IPVR_IS_ERR(ved_priv->fw_offset)) { + ved_priv->fw_bo = NULL; + return -ENOMEM; + } + + return 0; +} + +int32_t ved_setup_fw(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + uint32_t ram_bank_size; + struct ved_private *ved_priv = dev_priv->ved_private; + int32_t ret = 0; + struct ved_fw *fw; + uint32_t *fw_ptr = NULL; + uint32_t *text_ptr = NULL; + uint32_t *data_ptr = NULL; + const struct firmware *raw = NULL; + + /* todo : Assert the clock is on - if not turn it on to upload code */ + IPVR_DEBUG_VED("VED: ved_setup_fw.\n"); + + ved_set_clocks(dev_priv->dev, clk_enable_all); + + /* Reset MTX */ + VED_REG_WRITE32(MTX_SOFT_RESET_MTX_RESET_MASK, + MTX_SOFT_RESET_OFFSET); + + VED_REG_WRITE32(FIRMWAREID, MSVDX_COMMS_FIRMWARE_ID); + + VED_REG_WRITE32(0, MSVDX_COMMS_ERROR_TRIG); + VED_REG_WRITE32(199, MTX_SYSC_TIMERDIV_OFFSET); /* MTX_SYSC_TIMERDIV */ + VED_REG_WRITE32(0, MSVDX_EXT_FW_ERROR_STATE); /* EXT_FW_ERROR_STATE */ + VED_REG_WRITE32(0, MSVDX_COMMS_MSG_COUNTER); + VED_REG_WRITE32(0, MSVDX_COMMS_SIGNATURE); + VED_REG_WRITE32(0, MSVDX_COMMS_TO_HOST_RD_INDEX); + VED_REG_WRITE32(0, MSVDX_COMMS_TO_HOST_WRT_INDEX); + VED_REG_WRITE32(0, MSVDX_COMMS_TO_MTX_RD_INDEX); + VED_REG_WRITE32(0, MSVDX_COMMS_TO_MTX_WRT_INDEX); + VED_REG_WRITE32(0, MSVDX_COMMS_FW_STATUS); + VED_REG_WRITE32(DSIABLE_IDLE_GPIO_SIG | + DSIABLE_Auto_CLOCK_GATING | + RETURN_VDEB_DATA_IN_COMPLETION | + NOT_ENABLE_ON_HOST_CONCEALMENT, + MSVDX_COMMS_OFFSET_FLAGS); + VED_REG_WRITE32(0, MSVDX_COMMS_SIGNATURE); + + /* read register bank size */ + { + uint32_t bank_size, reg; + reg = VED_REG_READ32(MSVDX_MTX_RAM_BANK_OFFSET); + bank_size = + REGIO_READ_FIELD(reg, MSVDX_MTX_RAM_BANK, + MTX_RAM_BANK_SIZE); + ram_bank_size = (uint32_t)(1 << (bank_size + 2)); + } + + IPVR_DEBUG_VED("VED: RAM bank size = %d bytes\n", ram_bank_size); + + /* if FW already loaded from storage */ + if (ved_priv->ved_fw_ptr) { + fw_ptr = ved_priv->ved_fw_ptr; + } else { + fw_ptr = ved_get_fw(dev, &raw, FIRMWARE_NAME); + IPVR_DEBUG_VED("VED:load msvdx_fw_mfld_DE2.0.bin by udevd\n"); + } + if (!fw_ptr) { + IPVR_ERROR("VED:load ved_fw.bin failed,is udevd running?\n"); + ret = 1; + goto out; + } + + if (!ved_priv->fw_loaded_to_bo) { /* Load firmware into BO */ + IPVR_DEBUG_VED("MSVDX:load ved_fw.bin by udevd into BO\n"); + ret = ved_get_fw_bo(dev, &raw, FIRMWARE_NAME); + if (ret) { + IPVR_ERROR("VED: failed to call ved_get_fw_bo.\n"); + ret = 1; + goto out; + } + ved_priv->fw_loaded_to_bo = 1; + } + + fw = (struct ved_fw *) fw_ptr; + + /* need check fw->ver? */ + text_ptr = (uint32_t *)((uint8_t *) fw_ptr + sizeof(struct ved_fw)); + data_ptr = text_ptr + fw->text_size; + + /* maybe we can judge fw version according to fw text size */ + + IPVR_DEBUG_VED("VED: Retrieved pointers for firmware\n"); + IPVR_DEBUG_VED("VED: text_size: %d\n", fw->text_size); + IPVR_DEBUG_VED("VED: data_size: %d\n", fw->data_size); + IPVR_DEBUG_VED("VED: data_location: 0x%x\n", fw->data_location); + IPVR_DEBUG_VED("VED: First 4 bytes of text: 0x%x\n", *text_ptr); + IPVR_DEBUG_VED("VED: First 4 bytes of data: 0x%x\n", *data_ptr); + IPVR_DEBUG_VED("VED: Uploading firmware\n"); + +#if UPLOAD_FW_BY_DMA + ved_upload_fw(dev_priv, 0, ved_priv->mtx_mem_size / 4); +#else + ved_upload_fw(dev_priv, MTX_CORE_CODE_MEM, ram_bank_size, + PC_START_ADDRESS - MTX_CODE_BASE, fw->text_size, + text_ptr); + ved_upload_fw(dev_priv, MTX_CORE_DATA_MEM, ram_bank_size, + fw->data_location - MTX_DATA_BASE, fw->data_size, + data_ptr); +#endif + + /* -- Set starting PC address */ + ved_write_mtx_core_reg(dev_priv, MTX_PC, PC_START_ADDRESS); + + /* -- Turn on the thread */ + VED_REG_WRITE32(MTX_ENABLE_MTX_ENABLE_MASK, MTX_ENABLE_OFFSET); + + /* Wait for the signature value to be written back */ + ret = ved_wait_for_register(dev_priv, MSVDX_COMMS_SIGNATURE, + MSVDX_COMMS_SIGNATURE_VALUE, + 0xffffffff, /* Enabled bits */ + 2000000, 5); + if (ret) { + IPVR_ERROR("VED: firmware fails to initialize.\n"); + goto out; + } + + IPVR_DEBUG_VED("VED: MTX Initial indications OK.\n"); + IPVR_DEBUG_VED("VED: MSVDX_COMMS_AREA_ADDR = %08x.\n", + MSVDX_COMMS_AREA_ADDR); +out: + return ret; +} diff --git a/drivers/gpu/drm/ipvr/ved_fw.h b/drivers/gpu/drm/ipvr/ved_fw.h new file mode 100644 index 0000000..8c2da56 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ved_fw.h @@ -0,0 +1,73 @@ +/************************************************************************** + * ved_fw.h: VED firmware support header file + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * Copyright (c) Imagination Technologies Limited, UK + * Copyright (c) 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * + **************************************************************************/ + + +#ifndef _VED_FW_H_ +#define _VED_FW_H_ + +#include "ipvr_drv.h" + +#define FIRMWAREID 0x014d42ab + +/* Non-Optimal Invalidation is not default */ +#define MSVDX_DEVICE_NODE_FLAGS_MMU_NONOPT_INV 2 + +#define FW_VA_RENDER_HOST_INT 0x00004000 +#define MSVDX_DEVICE_NODE_FLAGS_MMU_HW_INVALIDATION 0x00000020 +#define FW_DEVA_ERROR_DETECTED 0x08000000 + +/* There is no work currently underway on the hardware */ +#define MSVDX_FW_STATUS_HW_IDLE 0x00000001 +#define MSVDX_DEVICE_NODE_FLAG_BRN23154_BLOCK_ON_FE 0x00000200 +#define MSVDX_DEVICE_NODE_FLAGS_DEFAULT_D0 \ + (MSVDX_DEVICE_NODE_FLAGS_MMU_NONOPT_INV | \ + MSVDX_DEVICE_NODE_FLAGS_MMU_HW_INVALIDATION | \ + MSVDX_DEVICE_NODE_FLAG_BRN23154_BLOCK_ON_FE) + +#define MSVDX_DEVICE_NODE_FLAGS_DEFAULT_D1 \ + (MSVDX_DEVICE_NODE_FLAGS_MMU_HW_INVALIDATION | \ + MSVDX_DEVICE_NODE_FLAG_BRN23154_BLOCK_ON_FE) + +#define MTX_CODE_BASE (0x80900000) +#define MTX_DATA_BASE (0x82880000) +#define PC_START_ADDRESS (0x80900000) + +#define MTX_CORE_CODE_MEM (0x10) +#define MTX_CORE_DATA_MEM (0x18) + +#define RENDEC_A_SIZE (4 * 1024 * 1024) +#define RENDEC_B_SIZE (1024 * 1024) + +#define TERMINATION_SIZE 48 + +#define MSVDX_RESET_NEEDS_REUPLOAD_FW (0x2) +#define MSVDX_RESET_NEEDS_INIT_FW (0x1) + +int32_t ved_alloc_fw_bo(struct drm_ipvr_private *dev_priv); + +int32_t ved_setup_fw(struct drm_device *dev); + +#endif diff --git a/drivers/gpu/drm/ipvr/ved_init.c b/drivers/gpu/drm/ipvr/ved_init.c new file mode 100644 index 0000000..bf82dfd --- /dev/null +++ b/drivers/gpu/drm/ipvr/ved_init.c @@ -0,0 +1,829 @@ +/************************************************************************** + * ved_init.c: VED initialization and deinitialization + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * Copyright (c) Imagination Technologies Limited, UK + * Copyright (c) 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Li Zeng li.zeng@intel.com + * Binglin Chen binglin.chen@intel.com + **************************************************************************/ + +#include "ipvr_gem.h" +#include "ipvr_buffer.h" +#include "ved_init.h" +#include "ved_cmd.h" +#include "ved_msg.h" +#include "ved_reg.h" +#include "ved_ec.h" +#include "ved_pm.h" +#include "ved_fw.h" +#include <linux/firmware.h> + +static ssize_t +ved_pmstate_show(struct device *dev, struct device_attribute *attr, char *buf) +{ + struct drm_device *drm_dev = dev_get_drvdata(dev); + struct drm_ipvr_private *dev_priv = NULL; + struct ved_private *ved_priv = NULL; + int32_t ret = -EINVAL; + + if (drm_dev == NULL) + return 0; + + dev_priv = drm_dev->dev_private; + ved_priv = dev_priv->ved_private; + return ret; +} + +static DEVICE_ATTR(ved_pmstate, 0444, ved_pmstate_show, NULL); + +void ved_clear_irq(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + uint32_t mtx_int = 0; + + /* Clear MTX interrupt */ + REGIO_WRITE_FIELD_LITE(mtx_int, MSVDX_INTERRUPT_STATUS, MTX_IRQ, 1); + VED_REG_WRITE32(mtx_int, MSVDX_INTERRUPT_CLEAR_OFFSET); +} + +/* following two functions also works for CLV and MFLD */ +/* IPVR_INT_ENABLE_R is set in ipvr_irq_(un)install_islands */ +void ved_disable_irq(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + /*uint32_t ier = dev_priv->vdc_irq_mask & (~_PSB_IRQ_MSVDX_FLAG); */ + + uint32_t enables = 0; + + REGIO_WRITE_FIELD_LITE(enables, MSVDX_INTERRUPT_STATUS, MTX_IRQ, 0); + VED_REG_WRITE32(enables, MSVDX_HOST_INTERRUPT_ENABLE_OFFSET); + + /* write in sysirq.c */ + /* PSB_WVDC32(ier, PSB_INT_ENABLE_R); /* essential */ */ +} + +void ved_enable_irq(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + /* uint32_t ier = dev_priv->vdc_irq_mask | _PSB_IRQ_MSVDX_FLAG; */ + uint32_t enables = 0; + + /* Only enable the master core IRQ*/ + REGIO_WRITE_FIELD_LITE(enables, MSVDX_INTERRUPT_STATUS, MTX_IRQ, + 1); + VED_REG_WRITE32(enables, MSVDX_HOST_INTERRUPT_ENABLE_OFFSET); + + /* write in sysirq.c */ + /* PSB_WVDC32(ier, PSB_INT_ENABLE_R); /* essential */ */ +} + +/* + * the original 1000 of udelay is derive from reference driver + * From Liu, Haiyang, changed the originial udelay value from 1000 to 5 + * can save 3% C0 residence + */ +int32_t +ved_wait_for_register(struct drm_ipvr_private *dev_priv, + uint32_t offset, uint32_t value, uint32_t enable, + uint32_t poll_cnt, uint32_t timeout) +{ + uint32_t reg_value = 0; + while (poll_cnt) { + reg_value = VED_REG_READ32(offset); + if (value == (reg_value & enable)) + return 0; + + /* Wait a bit */ + IPVR_UDELAY(timeout); + poll_cnt--; + } + IPVR_DEBUG_REG("MSVDX: Timeout while waiting for register %08x:" + " expecting %08x (mask %08x), got %08x\n", + offset, value, enable, reg_value); + + return -EFAULT; +} + +void +ved_set_clocks(struct drm_device *dev, uint32_t clock_state) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + uint32_t old_clock_state = 0; + /* IPVR_DEBUG_VED("SetClocks to %x.\n", clock_state); */ + old_clock_state = VED_REG_READ32(MSVDX_MAN_CLK_ENABLE_OFFSET); + if (old_clock_state == clock_state) + return; + + if (clock_state == 0) { + /* Turn off clocks procedure */ + if (old_clock_state) { + /* Turn off all the clocks except core */ + VED_REG_WRITE32( + MSVDX_MAN_CLK_ENABLE_CORE_MAN_CLK_ENABLE_MASK, + MSVDX_MAN_CLK_ENABLE_OFFSET); + + /* Make sure all the clocks are off except core */ + ved_wait_for_register(dev_priv, + MSVDX_MAN_CLK_ENABLE_OFFSET, + MSVDX_MAN_CLK_ENABLE_CORE_MAN_CLK_ENABLE_MASK, + 0xffffffff, 2000000, 5); + + /* Turn off core clock */ + VED_REG_WRITE32(0, MSVDX_MAN_CLK_ENABLE_OFFSET); + } + } else { + uint32_t clocks_en = clock_state; + + /*Make sure that core clock is not accidentally turned off */ + clocks_en |= MSVDX_MAN_CLK_ENABLE_CORE_MAN_CLK_ENABLE_MASK; + + /* If all clocks were disable do the bring up procedure */ + if (old_clock_state == 0) { + /* turn on core clock */ + VED_REG_WRITE32( + MSVDX_MAN_CLK_ENABLE_CORE_MAN_CLK_ENABLE_MASK, + MSVDX_MAN_CLK_ENABLE_OFFSET); + + /* Make sure core clock is on */ + ved_wait_for_register(dev_priv, + MSVDX_MAN_CLK_ENABLE_OFFSET, + MSVDX_MAN_CLK_ENABLE_CORE_MAN_CLK_ENABLE_MASK, + 0xffffffff, 2000000, 5); + + /* turn on the other clocks as well */ + VED_REG_WRITE32(clocks_en, MSVDX_MAN_CLK_ENABLE_OFFSET); + + /* Make sure that all they are on */ + ved_wait_for_register(dev_priv, + MSVDX_MAN_CLK_ENABLE_OFFSET, + clocks_en, 0xffffffff, 2000000, 5); + } else { + VED_REG_WRITE32(clocks_en, MSVDX_MAN_CLK_ENABLE_OFFSET); + + /* Make sure that they are on */ + ved_wait_for_register(dev_priv, + MSVDX_MAN_CLK_ENABLE_OFFSET, + clocks_en, 0xffffffff, 2000000, 5); + } + } +} + +int32_t ved_core_reset(struct drm_ipvr_private *dev_priv) +{ + int32_t ret = 0; + int32_t loop; + uint32_t cmd; + struct ved_private *ved_priv = dev_priv->ved_private; + /* Enable Clocks */ + IPVR_DEBUG_GENERAL("Enabling clocks.\n"); + ved_set_clocks(dev_priv->dev, clk_enable_all); + + /* Always pause the MMU as the core may be still active + * when resetting. It is very bad to have memory + * activity at the same time as a reset - Very Very bad + */ + VED_REG_WRITE32(2, MSVDX_MMU_CONTROL0_OFFSET); + + /* BRN26106, BRN23944, BRN33671 */ + /* This is neccessary for all cores up to Tourmaline */ + if ((VED_REG_READ32(MSVDX_CORE_REV_OFFSET) < 0x00050502) && + (VED_REG_READ32(MSVDX_INTERRUPT_STATUS_OFFSET) + & MSVDX_INTERRUPT_STATUS_MMU_FAULT_IRQ_MASK) && + (VED_REG_READ32(MSVDX_MMU_STATUS_OFFSET) & 1)) { + uint32_t *pptd; + uint32_t loop; + uint32_t ptd_addr; + + /* do work around */ + ptd_addr = page_to_pfn(ved_priv->mmu_recover_page) + << PAGE_SHIFT; + pptd = kmap(ved_priv->mmu_recover_page); + if (!pptd) { + IPVR_ERROR("failed to kmap mmu recover page.\n"); + return -1; + } + for (loop = 0; loop < 1024; loop++) + pptd[loop] = ptd_addr | 0x00000003; + VED_REG_WRITE32(ptd_addr, MSVDX_MMU_DIR_LIST_BASE_OFFSET + 0); + VED_REG_WRITE32(ptd_addr, MSVDX_MMU_DIR_LIST_BASE_OFFSET + 4); + VED_REG_WRITE32(ptd_addr, MSVDX_MMU_DIR_LIST_BASE_OFFSET + 8); + VED_REG_WRITE32(ptd_addr, MSVDX_MMU_DIR_LIST_BASE_OFFSET + 12); + + VED_REG_WRITE32(6, MSVDX_MMU_CONTROL0_OFFSET); + VED_REG_WRITE32(MSVDX_INTERRUPT_STATUS_MMU_FAULT_IRQ_MASK, + MSVDX_INTERRUPT_STATUS_OFFSET); + kunmap(ved_priv->mmu_recover_page); + } + + /* make sure *ALL* outstanding reads have gone away */ + for (loop = 0; loop < 10; loop++) + ret = ved_wait_for_register(dev_priv, MSVDX_MMU_MEM_REQ_OFFSET, + 0, 0xff, 100, 1); + if (ret) { + IPVR_DEBUG_WARN("MSVDX_MMU_MEM_REQ is %d,\n" + "indicate outstanding read request 0.\n", + VED_REG_READ32(MSVDX_MMU_MEM_REQ_OFFSET)); + ret = -1; + return ret; + } + /* disconnect RENDEC decoders from memory */ + cmd = VED_REG_READ32(MSVDX_RENDEC_CONTROL1_OFFSET); + REGIO_WRITE_FIELD(cmd, MSVDX_RENDEC_CONTROL1, RENDEC_DEC_DISABLE, 1); + VED_REG_WRITE32(cmd, MSVDX_RENDEC_CONTROL1_OFFSET); + + /* Issue software reset for all but core */ + VED_REG_WRITE32((unsigned int)~MSVDX_CONTROL_MSVDX_SOFT_RESET_MASK, + MSVDX_CONTROL_OFFSET); + VED_REG_READ32(MSVDX_CONTROL_OFFSET); + /* bit format is set as little endian */ + VED_REG_WRITE32(0, MSVDX_CONTROL_OFFSET); + /* make sure read requests are zero */ + ret = ved_wait_for_register(dev_priv, MSVDX_MMU_MEM_REQ_OFFSET, + 0, 0xff, 100, 100); + if (!ret) { + /* Issue software reset */ + VED_REG_WRITE32(MSVDX_CONTROL_MSVDX_SOFT_RESET_MASK, + MSVDX_CONTROL_OFFSET); + + ret = ved_wait_for_register(dev_priv, MSVDX_CONTROL_OFFSET, 0, + MSVDX_CONTROL_MSVDX_SOFT_RESET_MASK, + 2000000, 5); + if (!ret) { + /* Clear interrupt enabled flag */ + VED_REG_WRITE32(0, MSVDX_HOST_INTERRUPT_ENABLE_OFFSET); + + /* Clear any pending interrupt flags */ + VED_REG_WRITE32(0xFFFFFFFF, MSVDX_INTERRUPT_CLEAR_OFFSET); + } else { + IPVR_DEBUG_WARN("MSVDX_CONTROL_OFFSET is %d,\n" + "indicate software reset failed.\n", + VED_REG_READ32(MSVDX_CONTROL_OFFSET)); + } + } else { + IPVR_DEBUG_WARN("MSVDX_MMU_MEM_REQ is %d,\n" + "indicate outstanding read request 1.\n", + VED_REG_READ32(MSVDX_MMU_MEM_REQ_OFFSET)); + } + return ret; +} + +/* + * Reset chip and disable interrupts. + * Return 0 success, 1 failure + * use ved_core_reset instead of ved_reset + */ +int32_t ved_reset(struct drm_ipvr_private *dev_priv) +{ + int32_t ret = 0; + + /* Issue software reset */ + /* VED_REG_WRITE32(msvdx_sw_reset_all, MSVDX_CONTROL); */ + VED_REG_WRITE32(MSVDX_CONTROL_MSVDX_SOFT_RESET_MASK, + MSVDX_CONTROL_OFFSET); + + ret = ved_wait_for_register(dev_priv, MSVDX_CONTROL_OFFSET, 0, + MSVDX_CONTROL_MSVDX_SOFT_RESET_MASK, 2000000, 5); + if (!ret) { + /* Clear interrupt enabled flag */ + VED_REG_WRITE32(0, MSVDX_HOST_INTERRUPT_ENABLE_OFFSET); + + /* Clear any pending interrupt flags */ + VED_REG_WRITE32(0xFFFFFFFF, MSVDX_INTERRUPT_CLEAR_OFFSET); + } else { + IPVR_DEBUG_WARN("MSVDX_CONTROL_OFFSET is %d,\n" + "indicate software reset failed.\n", + VED_REG_READ32(MSVDX_CONTROL_OFFSET)); + } + + return ret; +} + +static int32_t ved_alloc_ccb_for_rendec(struct ved_private *ved_priv, + int32_t ccb0_size, + int32_t ccb1_size) +{ + struct drm_device *dev = ved_priv->dev_priv->dev; + size_t size; + uint8_t *ccb0_addr = NULL; + uint8_t *ccb1_addr = NULL; + + IPVR_DEBUG_INIT("VED: setting up RENDEC, allocate CCB 0/1\n"); + + /*handling for ccb0*/ + if (ved_priv->ccb0 == NULL) { + size = roundup(ccb0_size, PAGE_SIZE); + if (size == 0) + return -EINVAL; + + /* Allocate the new object */ + ved_priv->ccb0 = ipvr_gem_obj_create_and_bind(dev, size); + if (ved_priv->ccb0 == NULL) { + IPVR_ERROR("VED: failed to allocate ccb0 buffer.\n"); + ved_priv->ccb0 = NULL; + return -ENOMEM; + } + + ved_priv->base_addr0 = ipvr_gem_obj_mmu_offset(ved_priv->ccb0); + + ccb0_addr = ipvr_gem_object_vmap(ved_priv->ccb0); + if (IS_ERR(ccb0_addr)) { + IPVR_ERROR("VED: kmap failed for ccb0 buffer.\n"); + return PTR_ERR(ccb0_addr); + } + + memset(ccb0_addr, 0, size); + vunmap(ccb0_addr); + } + + /*handling for ccb1*/ + if (ved_priv->ccb1 == NULL) { + size = roundup(ccb1_size, PAGE_SIZE); + if (size == 0) + return -EINVAL; + + /* Allocate the new object */ + ved_priv->ccb1 = ipvr_gem_obj_create_and_bind(dev, size); + if (ved_priv->ccb1 == NULL) { + IPVR_ERROR("VED: failed to allocate ccb1 buffer.\n"); + ved_priv->ccb1 = NULL; + return -ENOMEM; + } + + ved_priv->base_addr1 = ipvr_gem_obj_mmu_offset(ved_priv->ccb1); + + ccb1_addr = ipvr_gem_object_vmap(ved_priv->ccb1); + if (IS_ERR(ccb1_addr)) { + IPVR_ERROR("VED: kmap failed for ccb1 buffer.\n"); + return PTR_ERR(ccb1_addr); + } + + memset(ccb1_addr, 0, size); + vunmap(ccb1_addr); + } + + IPVR_DEBUG_INIT("VED: RENDEC A: %08x RENDEC B: %08x\n", + ved_priv->base_addr0, ved_priv->base_addr1); + + return 0; +} + +static void ved_free_ccb(struct ved_private *ved_priv) +{ + if (ved_priv->ccb0) { + drm_gem_object_unreference_unlocked(&ved_priv->ccb0->base); + ved_priv->ccb0 = NULL; + } + if (ved_priv->ccb1) { + drm_gem_object_unreference_unlocked(&ved_priv->ccb1->base); + ved_priv->ccb1 = NULL; + } +} + +static void ved_rendec_init_by_reg(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ved_private *ved_priv = dev_priv->ved_private; + uint32_t cmd; + + VED_REG_WRITE32(ved_priv->base_addr0, MSVDX_RENDEC_BASE_ADDR0_OFFSET); + VED_REG_WRITE32(ved_priv->base_addr1, MSVDX_RENDEC_BASE_ADDR1_OFFSET); + + cmd = 0; + REGIO_WRITE_FIELD(cmd, MSVDX_RENDEC_BUFFER_SIZE, + RENDEC_BUFFER_SIZE0, RENDEC_A_SIZE / 4096); + REGIO_WRITE_FIELD(cmd, MSVDX_RENDEC_BUFFER_SIZE, + RENDEC_BUFFER_SIZE1, RENDEC_B_SIZE / 4096); + VED_REG_WRITE32(cmd, MSVDX_RENDEC_BUFFER_SIZE_OFFSET); + + cmd = 0; + REGIO_WRITE_FIELD(cmd, MSVDX_RENDEC_CONTROL1, + RENDEC_DECODE_START_SIZE, 0); + REGIO_WRITE_FIELD(cmd, MSVDX_RENDEC_CONTROL1, + RENDEC_BURST_SIZE_W, 1); + REGIO_WRITE_FIELD(cmd, MSVDX_RENDEC_CONTROL1, + RENDEC_BURST_SIZE_R, 1); + REGIO_WRITE_FIELD(cmd, MSVDX_RENDEC_CONTROL1, + RENDEC_EXTERNAL_MEMORY, 1); + VED_REG_WRITE32(cmd, MSVDX_RENDEC_CONTROL1_OFFSET); + + cmd = 0x00101010; + VED_REG_WRITE32(cmd, MSVDX_RENDEC_CONTEXT0_OFFSET); + VED_REG_WRITE32(cmd, MSVDX_RENDEC_CONTEXT1_OFFSET); + VED_REG_WRITE32(cmd, MSVDX_RENDEC_CONTEXT2_OFFSET); + VED_REG_WRITE32(cmd, MSVDX_RENDEC_CONTEXT3_OFFSET); + VED_REG_WRITE32(cmd, MSVDX_RENDEC_CONTEXT4_OFFSET); + VED_REG_WRITE32(cmd, MSVDX_RENDEC_CONTEXT5_OFFSET); + + cmd = 0; + REGIO_WRITE_FIELD(cmd, MSVDX_RENDEC_CONTROL0, RENDEC_INITIALISE, + 1); + VED_REG_WRITE32(cmd, MSVDX_RENDEC_CONTROL0_OFFSET); +} + +int32_t ved_rendec_init_by_msg(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ved_private *ved_priv = dev_priv->ved_private; + + /* at this stage, FW is uplaoded successfully, + * can send RENDEC init message */ + struct fw_init_msg init_msg; + init_msg.header.bits.msg_size = sizeof(struct fw_init_msg); + init_msg.header.bits.msg_type = MTX_MSGID_INIT; + init_msg.rendec_addr0 = ved_priv->base_addr0; + init_msg.rendec_addr1 = ved_priv->base_addr1; + init_msg.rendec_size.bits.rendec_size0 = RENDEC_A_SIZE / (4 * 1024); + init_msg.rendec_size.bits.rendec_size1 = RENDEC_B_SIZE / (4 * 1024); + return ved_mtx_send(dev_priv, (void *)&init_msg); +} + +#ifdef CONFIG_DRM_IPVR_EC +static void ved_init_ec(struct ved_private *ved_priv) +{ + struct drm_ipvr_private *dev_priv = ved_priv->dev_priv; + + /* we should restore the state, if we power down/up during EC */ + VED_REG_WRITE32(0, 0x2000 + 0xcc4); /* EXT_FW_ERROR_STATE */ + VED_REG_WRITE32(0, 0x2000 + 0xcb0); /* EXT_FW_LAST_MBS */ + VED_REG_WRITE32(0, 0x2000 + 0xcb4); /* EXT_FW_LAST_MBS */ + VED_REG_WRITE32(0, 0x2000 + 0xcb8); /* EXT_FW_LAST_MBS */ + VED_REG_WRITE32(0, 0x2000 + 0xcbc); /* EXT_FW_LAST_MBS */ + + ved_priv->vec_ec_mem_saved = 1; + + ved_priv->ved_ec_ctx[0] = + kzalloc(sizeof(struct ved_ec_context) * + VED_MAX_EC_INSTANCE, + GFP_KERNEL); + if (ved_priv->ved_ec_ctx[0] == NULL) { + IPVR_ERROR("VED: fail to allocate memory for ec ctx.\n"); + } else { + int i; + for (i = 1; i < VED_MAX_EC_INSTANCE; i++) + ved_priv->ved_ec_ctx[i] = + ved_priv->ved_ec_ctx[0] + i; + for (i = 0; i < VED_MAX_EC_INSTANCE; i++) + ved_priv->ved_ec_ctx[i]->fence = + VED_INVALID_FENCE; + } + INIT_WORK(&(ved_priv->ec_work), ved_do_concealment); + return; +} +#endif + +static int32_t ved_startup_init(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ved_private *ved_priv; + + ved_priv = kmalloc(sizeof(struct ved_private), GFP_KERNEL); + if (ved_priv == NULL) { + IPVR_ERROR("VED: alloc ved_private failed.\n"); + return -ENOMEM; + } + + dev_priv->ved_private = ved_priv; + memset(ved_priv, 0, sizeof(struct ved_private)); + ved_priv->dev_priv = dev_priv; + ved_priv->dev = dev; + ved_priv->fw_loaded_by_punit = 0; + + ved_priv->pm_gating_count = 0; + + /* get device --> drm_device --> drm_ipvr_private --> ved_priv + * for ved_pmstate_show: ved_pmpolicy + * if not pci_set_drvdata, can't get drm_device from device + */ + /* pci_set_drvdata(dev->pdev, dev); */ + if (device_create_file(&dev->platformdev->dev, + &dev_attr_ved_pmstate)) + IPVR_ERROR("VED: could not create sysfs file\n"); + + ved_priv->sysfs_pmstate = sysfs_get_dirent( + dev->platformdev->dev.kobj.sd, + "ved_pmstate"); + +#ifdef CONFIG_DRM_IPVR_EC + ved_init_ec(ved_priv); +#endif + + /* Initialize comand ved queueing */ + INIT_LIST_HEAD(&ved_priv->ved_queue); + mutex_init(&ved_priv->ved_mutex); + spin_lock_init(&ved_priv->ved_lock); + ved_priv->mmu_recover_page = alloc_page(GFP_DMA32); + if (!ved_priv->mmu_recover_page) + goto err_exit; + + return 0; + +err_exit: + IPVR_ERROR("VED: init one time failed.\n"); + kfree(dev_priv->ved_private); + + return 1; +} + +/* This value is hardcoded in FW */ +#define WDT_CLOCK_DIVIDER 128 +int32_t ved_post_boot_init(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ved_private *ved_priv = dev_priv->ved_private; + uint32_t device_node_flags = + DSIABLE_IDLE_GPIO_SIG | DSIABLE_Auto_CLOCK_GATING | + RETURN_VDEB_DATA_IN_COMPLETION | + NOT_ENABLE_ON_HOST_CONCEALMENT; + + /* DDK set fe_wdt_clks as 0x820 and be_wdt_clks as 0x8200 */ + uint32_t fe_wdt_clks = 0x334 * WDT_CLOCK_DIVIDER; + uint32_t be_wdt_clks = 0x2008 * WDT_CLOCK_DIVIDER; + + VED_REG_WRITE32(FIRMWAREID, MSVDX_COMMS_FIRMWARE_ID); + VED_REG_WRITE32(device_node_flags, MSVDX_COMMS_OFFSET_FLAGS); + + /* read register bank size */ + { + uint32_t ram_bank_size; + uint32_t bank_size, reg; + reg = VED_REG_READ32(MSVDX_MTX_RAM_BANK_OFFSET); + bank_size = + REGIO_READ_FIELD(reg, MSVDX_MTX_RAM_BANK, + MTX_RAM_BANK_SIZE); + ram_bank_size = (uint32_t)(1 << (bank_size + 2)); + IPVR_DEBUG_INIT("VED: RAM bank size = %d bytes\n", + ram_bank_size); + } + /* host end */ + + /* DDK setup tiling region here */ + /* DDK set MMU_CONTROL2 register */ + + /* set watchdog timer here */ + if (!ved_priv->fw_loaded_by_punit) { + int reg_val = 0; + REGIO_WRITE_FIELD(reg_val, FE_MSVDX_WDT_CONTROL, + FE_WDT_CNT_CTRL, 0x3); + REGIO_WRITE_FIELD(reg_val, FE_MSVDX_WDT_CONTROL, + FE_WDT_ENABLE, 0); + REGIO_WRITE_FIELD(reg_val, FE_MSVDX_WDT_CONTROL, + FE_WDT_ACTION0, 1); + REGIO_WRITE_FIELD(reg_val, FE_MSVDX_WDT_CONTROL, + FE_WDT_CLEAR_SELECT, 1); + REGIO_WRITE_FIELD(reg_val, FE_MSVDX_WDT_CONTROL, + FE_WDT_CLKDIV_SELECT, 7); + VED_REG_WRITE32(fe_wdt_clks / WDT_CLOCK_DIVIDER, + FE_MSVDX_WDT_COMPAREMATCH_OFFSET); + VED_REG_WRITE32(reg_val, FE_MSVDX_WDT_CONTROL_OFFSET); + + reg_val = 0; + /* DDK set BE_WDT_CNT_CTRL as 0x5 and BE_WDT_CLEAR_SELECT as 0x1 */ + REGIO_WRITE_FIELD(reg_val, BE_MSVDX_WDT_CONTROL, + BE_WDT_CNT_CTRL, 0x7); + REGIO_WRITE_FIELD(reg_val, BE_MSVDX_WDT_CONTROL, + BE_WDT_ENABLE, 0); + REGIO_WRITE_FIELD(reg_val, BE_MSVDX_WDT_CONTROL, + BE_WDT_ACTION0, 1); + REGIO_WRITE_FIELD(reg_val, BE_MSVDX_WDT_CONTROL, + BE_WDT_CLEAR_SELECT, 0xd); + REGIO_WRITE_FIELD(reg_val, BE_MSVDX_WDT_CONTROL, + BE_WDT_CLKDIV_SELECT, 7); + + VED_REG_WRITE32(be_wdt_clks / WDT_CLOCK_DIVIDER, + BE_MSVDX_WDT_COMPAREMATCH_OFFSET); + VED_REG_WRITE32(reg_val, BE_MSVDX_WDT_CONTROL_OFFSET); + } else { + /* for the other two, use the default value punit set */ + VED_REG_WRITE32(fe_wdt_clks / WDT_CLOCK_DIVIDER, + FE_MSVDX_WDT_COMPAREMATCH_OFFSET); + VED_REG_WRITE32(be_wdt_clks / WDT_CLOCK_DIVIDER, + BE_MSVDX_WDT_COMPAREMATCH_OFFSET); + } + + return ved_rendec_init_by_msg(dev); +} + +static void ved_post_powerup_core_reset(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + ved_set_clocks(dev_priv->dev, clk_enable_all); + + /* ved_clear_irq only clear CR_MTX_IRQ int, + * while DDK set 0xFFFFFFFF */ + ved_clear_irq(dev); + ved_enable_irq(dev); +} + +static int32_t ved_mtx_init(struct drm_device *dev, int32_t error_reset) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + uint32_t clk_divider = 200; + int32_t ret; + struct ved_private *ved_priv = dev_priv->ved_private; + + /* These should not be reprogrames after a error reset */ + if (!error_reset) { + VED_REG_WRITE32(0, MSVDX_COMMS_MSG_COUNTER); + VED_REG_WRITE32(0, MSVDX_EXT_FW_ERROR_STATE); + } + + VED_REG_WRITE32(0, MSVDX_COMMS_ERROR_TRIG); + VED_REG_WRITE32(0, MSVDX_COMMS_TO_HOST_RD_INDEX); + VED_REG_WRITE32(0, MSVDX_COMMS_TO_HOST_WRT_INDEX); + VED_REG_WRITE32(0, MSVDX_COMMS_TO_MTX_RD_INDEX); + VED_REG_WRITE32(0, MSVDX_COMMS_TO_MTX_WRT_INDEX); + VED_REG_WRITE32(0, MSVDX_COMMS_FIRMWARE_ID); + /* + * IMG DDK set as: gui32DeviceNodeFlags & 0x4000, + * while it is not set in fw spec + * The bit neede to be set preboot is the performce data bit since this + * controls caused the firmware to rebalance the message queues. + */ + VED_REG_WRITE32(0, MSVDX_COMMS_OFFSET_FLAGS); + + /* DDK: check device_node_flags with 0x400 to | (1<<16), + * while it is not set in fw spec */ + VED_REG_WRITE32(clk_divider - 1, MTX_SYSC_TIMERDIV_OFFSET); + + /* DDK: LLDMA upload fw, which is now done by gunit */ + + /* DDK: redefine toHost and toMTX msg buffer, seems not needed */ + + /* Wait for the signature value to be written back */ + ret = ved_wait_for_register(dev_priv, MSVDX_COMMS_SIGNATURE, + MSVDX_COMMS_SIGNATURE_VALUE, + 0xffffffff, + 1000, 1000); + if (ret) { + IPVR_DEBUG_WARN("WARN: Gunit upload fw failure,\n" + "MSVDX_COMMS_SIGNATURE reg is 0x%x," + "MSVDX_COMMS_FW_STATUS reg is 0x%x," + "MTX_ENABLE reg is 0x%x.\n", + VED_REG_READ32(MSVDX_COMMS_SIGNATURE), + VED_REG_READ32(MSVDX_COMMS_FW_STATUS), + VED_REG_READ32(MTX_ENABLE_OFFSET)); + ved_priv->ved_needs_reset |= + MSVDX_RESET_NEEDS_REUPLOAD_FW | + MSVDX_RESET_NEEDS_INIT_FW; + } + return ret; +} + +int32_t ved_post_init(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + uint32_t cmd; + int ret; + struct ved_private *ved_priv; + + if (!dev_priv || !dev_priv->ved_private) + return -EINVAL; + + ved_priv = dev_priv->ved_private; + + ved_priv->ved_busy = 0; + ved_priv->ved_hw_busy = 1; + + if (ved_priv->fw_loaded_by_punit) { + /* DDK: Configure MSVDX Memory Stalling iwth the min, max and ratio of access */ + ved_post_powerup_core_reset(dev); + } + + if (!ved_priv->fw_loaded_by_punit) { + /* Enable MMU by removing all bypass bits */ + VED_REG_WRITE32(0, MSVDX_MMU_CONTROL0_OFFSET); + } else { + ved_priv->rendec_initialized = 0; + ret = ved_mtx_init(dev, ved_priv->decoding_err); + if (ret) { + IPVR_ERROR("VED: ved_mtx_init failed.\n"); + return ret; + } + } + + if (!ved_priv->fw_loaded_by_punit) { + ved_rendec_init_by_reg(dev); + if (!ved_priv->fw_bo) { + ret = ved_alloc_fw_bo(dev_priv); + if (ret) { + IPVR_ERROR("VED: ved_alloc_fw_bo failed.\n"); + return ret; + } + } + /* move fw loading to the place receiving first cmd buffer */ + ved_priv->ved_fw_loaded = 0; /* need to load firware */ + /* it should be set at punit post boot init phase */ + VED_REG_WRITE32(820, FE_MSVDX_WDT_COMPAREMATCH_OFFSET); + VED_REG_WRITE32(8200, BE_MSVDX_WDT_COMPAREMATCH_OFFSET); + + VED_REG_WRITE32(820, FE_MSVDX_WDT_COMPAREMATCH_OFFSET); + VED_REG_WRITE32(8200, BE_MSVDX_WDT_COMPAREMATCH_OFFSET); + + ved_clear_irq(dev); + ved_enable_irq(dev); + + cmd = 0; + cmd = VED_REG_READ32(VEC_SHIFTREG_CONTROL_OFFSET); + REGIO_WRITE_FIELD(cmd, VEC_SHIFTREG_CONTROL, + SR_MASTER_SELECT, 1); /* Host */ + VED_REG_WRITE32(cmd, VEC_SHIFTREG_CONTROL_OFFSET); + } + + return 0; +} + +int32_t ipvr_ved_init(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + int32_t ret; + + if (!dev_priv->ved_private) { + ret = ved_startup_init(dev); + if (unlikely(ret)) { + IPVR_ERROR("VED: ved_startup_init failed.\n"); + return ret; + } + } + + if (unlikely(dev_priv->ved_private == NULL)) + return -ENOMEM; + + ret = ved_alloc_ccb_for_rendec(dev_priv->ved_private, + RENDEC_A_SIZE, RENDEC_B_SIZE); + if (unlikely(ret)) { + IPVR_ERROR("VED: msvdx_alloc_ccb_for_rendec failed.\n"); + ved_free_ccb(dev_priv->ved_private); + return ret; + } + + ret = ved_post_init(dev); + if (unlikely(ret)) { + IPVR_ERROR("VED: ved_post_init failed.\n"); + return ret; + } + + return 0; +} + +int32_t ipvr_ved_uninit(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ved_private *ved_priv = dev_priv->ved_private; + + /* VED_REG_WRITE32 (clk_enable_minimal, MSVDX_MAN_CLK_ENABLE); */ + IPVR_DEBUG_INIT("VED: set the VED clock to 0.\n"); + ved_set_clocks(dev_priv->dev, 0); + + if (NULL == ved_priv) { + IPVR_ERROR("VED: ipvr_ved_uninit: ved_priv is NULL!\n"); + return -1; + } + + if (ved_priv->ccb0 || ved_priv->ccb1) + ved_free_ccb(ved_priv); + + if (ved_priv->fw_bo) { + drm_gem_object_unreference_unlocked(&ved_priv->fw_bo->base); + ved_priv->fw_bo = NULL; + } + + if (!ved_priv->fw_loaded_by_punit) { + if (ved_priv->ved_fw_ptr) + kfree(ved_priv->ved_fw_ptr); + } + + kfree(ved_priv->ved_ec_ctx[0]); + + if (ved_priv->mmu_recover_page) + __free_page(ved_priv->mmu_recover_page); + + if (ved_priv) { + /* pci_set_drvdata(dev->pdev, NULL); */ + device_remove_file(&dev->platformdev->dev, + &dev_attr_ved_pmstate); + sysfs_put(ved_priv->sysfs_pmstate); + ved_priv->sysfs_pmstate = NULL; + kfree(ved_priv); + dev_priv->ved_private = NULL; + } + + return 0; +} diff --git a/drivers/gpu/drm/ipvr/ved_init.h b/drivers/gpu/drm/ipvr/ved_init.h new file mode 100644 index 0000000..ab57f025 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ved_init.h @@ -0,0 +1,61 @@ +/************************************************************************** + * ved_init.h: VED initialization header file + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * Copyright (c) Imagination Technologies Limited, UK + * Copyright (c) 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * + **************************************************************************/ + + +#ifndef _VED_INIT_H_ +#define _VED_INIT_H_ + +#include "ipvr_drv.h" + +void ved_clear_irq(struct drm_device *dev); + +void ved_disable_irq(struct drm_device *dev); + +void ved_enable_irq(struct drm_device *dev); + +int32_t ved_wait_for_register(struct drm_ipvr_private *dev_priv, + uint32_t offset, uint32_t value, uint32_t enable, + uint32_t poll_cnt, uint32_t timeout); + +void ved_set_clocks(struct drm_device *dev, uint32_t clock_state); + +int32_t ved_core_reset(struct drm_ipvr_private *dev_priv); + +/* todo: ved_reset is used for the case of fw loading by driver + * Later we can test if it can be removed. */ +int32_t ved_reset(struct drm_ipvr_private *dev_priv); + +int32_t ved_rendec_init_by_msg(struct drm_device *dev); + +int32_t ved_post_init(struct drm_device *dev); + +int32_t ved_post_boot_init(struct drm_device *dev); + +int32_t ipvr_ved_init(struct drm_device *dev); + +int32_t ipvr_ved_uninit(struct drm_device *dev); + +#endif diff --git a/drivers/gpu/drm/ipvr/ved_msg.h b/drivers/gpu/drm/ipvr/ved_msg.h new file mode 100644 index 0000000..1ffba7e --- /dev/null +++ b/drivers/gpu/drm/ipvr/ved_msg.h @@ -0,0 +1,364 @@ +/************************************************************************** + * ved_msg.h: VED message definition + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * Copyright (c) 2003 Imagination Technologies Limited, UK + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Li Zeng li.zeng@intel.com + * + **************************************************************************/ + +#ifndef _VED_MSG_H_ +#define _VED_MSG_H_ + +/* Start of parser specific Host->MTX messages. */ +#define FWRK_MSGID_START_PSR_HOSTMTX_MSG (0x80) + +/* Start of parser specific MTX->Host messages. */ +#define FWRK_MSGID_START_PSR_MTXHOST_MSG (0xC0) + +/* Host defined msg, just for host use, MTX not recgnize */ +#define FWRK_MSGID_HOST_EMULATED (0x40) + +/* This type defines the framework specified message ids */ +enum { + /* ! Sent by the VA driver on the host to the mtx firmware. + */ + MTX_MSGID_PADDING = 0, + MTX_MSGID_INIT = FWRK_MSGID_START_PSR_HOSTMTX_MSG, + MTX_MSGID_DECODE_FE, + MTX_MSGID_DEBLOCK, + MTX_MSGID_INTRA_OOLD, + MTX_MSGID_DECODE_BE, + MTX_MSGID_HOST_BE_OPP, + + /*! Sent by the mtx firmware to itself. + */ + MTX_MSGID_RENDER_MC_INTERRUPT, + + /* used to ditinguish mrst and mfld */ + MTX_MSGID_DEBLOCK_MFLD = FWRK_MSGID_HOST_EMULATED, + MTX_MSGID_INTRA_OOLD_MFLD, + MTX_MSGID_DECODE_BE_MFLD, + MTX_MSGID_HOST_BE_OPP_MFLD, + + /*! Sent by the DXVA firmware on the MTX to the host. + */ + MTX_MSGID_COMPLETED = FWRK_MSGID_START_PSR_MTXHOST_MSG, + MTX_MSGID_COMPLETED_BATCH, + MTX_MSGID_DEBLOCK_REQUIRED, + MTX_MSGID_TEST_RESPONCE, + MTX_MSGID_ACK, + MTX_MSGID_FAILED, + MTX_MSGID_CONTIGUITY_WARNING, + MTX_MSGID_HW_PANIC, +}; + +#define MTX_GENMSG_SIZE_TYPE uint8_t +#define MTX_GENMSG_SIZE_MASK (0xFF) +#define MTX_GENMSG_SIZE_SHIFT (0) +#define MTX_GENMSG_SIZE_OFFSET (0x0000) + +#define MTX_GENMSG_ID_TYPE uint8_t +#define MTX_GENMSG_ID_MASK (0xFF) +#define MTX_GENMSG_ID_SHIFT (0) +#define MTX_GENMSG_ID_OFFSET (0x0001) + +#define MTX_GENMSG_HEADER_SIZE 2 + +#define MTX_GENMSG_FENCE_TYPE uint16_t +#define MTX_GENMSG_FENCE_MASK (0xFFFF) +#define MTX_GENMSG_FENCE_OFFSET (0x0002) +#define MTX_GENMSG_FENCE_SHIFT (0) + +#define FW_INVALIDATE_MMU (0x0010) + +union msg_header { + struct { + uint32_t msg_size:8; + uint32_t msg_type:8; + uint32_t msg_fence:16; + } bits; + uint32_t value; +}; + +struct fw_init_msg { + union { + struct { + uint32_t msg_size:8; + uint32_t msg_type:8; + uint32_t reserved:16; + } bits; + uint32_t value; + } header; + uint32_t rendec_addr0; + uint32_t rendec_addr1; + union { + struct { + uint32_t rendec_size0:16; + uint32_t rendec_size1:16; + } bits; + uint32_t value; + } rendec_size; +}; + +struct fw_decode_msg { + union { + struct { + uint32_t msg_size:8; + uint32_t msg_type:8; + uint32_t msg_fence:16; + } bits; + uint32_t value; + } header; + union { + struct { + uint32_t flags:16; + uint32_t buffer_size:16; + } bits; + uint32_t value; + } flag_size; + uint32_t crtl_alloc_addr; + union { + struct { + uint32_t context:8; + uint32_t mmu_ptd:24; + } bits; + uint32_t value; + } mmu_context; + uint32_t operating_mode; +}; + +struct fw_deblock_msg { + union { + struct { + uint32_t msg_size:8; + uint32_t msg_type:8; + uint32_t msg_fence:16; + } bits; + uint32_t value; + } header; + union { + struct { + uint32_t flags:16; + uint32_t slice_field_type:2; + uint32_t reserved:14; + } bits; + uint32_t value; + } flag_type; + uint32_t operating_mode; + union { + struct { + uint32_t context:8; + uint32_t mmu_ptd:24; + } bits; + uint32_t value; + } mmu_context; + union { + struct { + uint32_t frame_height_mb:16; + uint32_t pic_width_mb:16; + } bits; + uint32_t value; + } pic_size; + uint32_t address_a0; + uint32_t address_a1; + uint32_t mb_param_address; + uint32_t ext_stride_a; + uint32_t address_b0; + uint32_t address_b1; + uint32_t alt_output_flags_b; + /* additional msg outside of IMG msg */ + uint32_t address_c0; + uint32_t address_c1; +}; + +#define MTX_PADMSG_SIZE 2 +struct fw_padding_msg { + union { + struct { + uint32_t msg_size:8; + uint32_t msg_type:8; + } bits; + uint16_t value; + } header; +}; + +struct fw_msg_header { + union { + struct { + uint32_t msg_size:8; + uint32_t msg_type:8; + uint32_t msg_fence:16; + } bits; + uint32_t value; + } header; +}; + +struct fw_completed_msg { + union { + struct { + uint32_t msg_size:8; + uint32_t msg_type:8; + uint32_t msg_fence:16; + } bits; + uint32_t value; + } header; + union { + struct { + uint32_t start_mb:16; + uint32_t last_mb:16; + } bits; + uint32_t value; + } mb; + uint32_t flags; + uint32_t vdebcr; +}; + +struct fw_deblock_required_msg { + union { + struct { + uint32_t msg_size:8; + uint32_t msg_type:8; + uint32_t msg_fence:16; + } bits; + uint32_t value; + } header; +}; + +struct fw_panic_msg { + union { + struct { + uint32_t msg_size:8; + uint32_t msg_type:8; + uint32_t msg_fence:16; + } bits; + uint32_t value; + } header; + uint32_t fe_status; + uint32_t be_status; + union { + struct { + uint32_t last_mb:16; + uint32_t reserved2:16; + } bits; + uint32_t value; + } mb; +}; + +struct fw_contiguity_msg { + union { + struct { + uint32_t msg_size:8; + uint32_t msg_type:8; + uint32_t msg_fence:16; + } bits; + uint32_t value; + } header; + union { + struct { + uint32_t end_mb_num:16; + uint32_t begin_mb_num:16; + } bits; + uint32_t value; + } mb; +}; + +struct fw_slice_header_extract_msg { + union { + struct { + uint32_t msg_size:8; + uint32_t msg_type:8; + uint32_t msg_fence:16; + } bits; + uint32_t value; + } header; + + union { + struct { + uint32_t flags:16; + uint32_t res:16; + } bits; + uint32_t value; + } flags; + + uint32_t src; + + union { + struct { + uint32_t context:8; + uint32_t mmu_ptd:24; + } bits; + uint32_t value; + } mmu_context; + + uint32_t dst; + uint32_t src_size; + uint32_t dst_size; + + union { + struct { + uint32_t expected_pps_id:8; + uint32_t nalu_header_unit_type:5; + uint32_t nalu_header_ref_idc:2; + uint32_t nalu_header_reserved:1; + uint32_t continue_parse_flag:1; + uint32_t frame_mbs_only_flag:1; + uint32_t pic_order_present_flag:1; + uint32_t delta_pic_order_always_zero_flag:1; + uint32_t redundant_pic_cnt_present_flag:1; + uint32_t weighted_pred_flag:1; + uint32_t entropy_coding_mode_flag:1; + uint32_t deblocking_filter_control_present_flag:1; + uint32_t weighted_bipred_idc:2; + uint32_t residual_colour_transform_flag:1; + uint32_t chroma_format_idc:2; + uint32_t idr_flag:1; + uint32_t pic_order_cnt_type:2; + } bits; + uint32_t value; + } flag_bitfield; + + union { + struct { + uint8_t num_slice_groups_minus1:3; + uint8_t num_ref_idc_l1_active_minus1:5; + uint8_t slice_group_map_type:3; + uint8_t num_ref_idc_l0_active_minus1:5; + uint8_t log2_slice_group_change_cycle:4; + uint8_t slice_header_bit_offset:4; + uint8_t log2_max_frame_num_minus4:4; + uint8_t logs_max_pic_order_cnt_lsb_minus4:4; + } bits; + uint32_t value; + } pic_param0; +}; + +struct fw_slice_header_extract_done_msg { + union { + struct { + uint32_t msg_size:8; + uint32_t msg_type:8; + uint32_t msg_fence:16; + } bits; + uint32_t value; + } header; +}; + +#endif diff --git a/drivers/gpu/drm/ipvr/ved_pm.c b/drivers/gpu/drm/ipvr/ved_pm.c new file mode 100644 index 0000000..9aee650 --- /dev/null +++ b/drivers/gpu/drm/ipvr/ved_pm.c @@ -0,0 +1,392 @@ +/************************************************************************** + * ved_pm.c: VED power management support + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + + +#include "ved_pm.h" +#include "ved_init.h" +#include "ved_reg.h" +#include "ved_cmd.h" +#include "ved_fw.h" +#include "ipvr_trace.h" + +#ifdef CONFIG_INTEL_SOC_PMC +#include <linux/intel_mid_pm.h> +#endif +#include <linux/module.h> +#include <linux/pm_runtime.h> + +extern int32_t drm_ipvr_freq; + +#ifdef CONFIG_INTEL_SOC_PMC +extern int pmc_nc_set_power_state(int islands, int state_type, int reg); +extern int pmc_nc_get_power_state(int islands, int reg); +#endif + +#define PCI_ROOT_MSGBUS_CTRL_REG 0xD0 +#define PCI_ROOT_MSGBUS_DATA_REG 0xD4 +#define PCI_ROOT_MSGBUS_CTRL_EXT_REG 0xD8 +#define PCI_ROOT_MSGBUS_READ 0x10 +#define PCI_ROOT_MSGBUS_WRITE 0x11 +#define PCI_ROOT_MSGBUS_DWORD_ENABLE 0xf0 + +/* VED power state set/get */ +#define PUNIT_PORT 0x04 +#define VEDSSPM0 0x32 +#define VEDSSPM1 0x33 +#define VEDSSC 0x1 + +/* VED frequency set/get */ +#define IP_FREQ_VALID 0x80 /* Freq is valid bit */ + +#define IP_FREQ_SIZE 5 /* number of bits in freq fields */ +#define IP_FREQ_MASK 0x1f /* Bit mask for freq field */ + +/* Positions of various frequency fields */ +#define IP_FREQ_POS 0 /* Freq control [4:0] */ +#define IP_FREQ_GUAR_POS 8 /* Freq guar [12:8] */ +#define IP_FREQ_STAT_POS 24 /* Freq status [28:24] */ + +#define IP_FREQ_RESUME_SET 0x64 + +#define IPVR_VED_CLOCKGATING_OFFSET 0x2064 + +enum APM_VED_STATUS { + VED_APM_STS_D0 = 0, + VED_APM_STS_D1, + VED_APM_STS_D2, + VED_APM_STS_D3 +}; + +static int32_t ved_save_context(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = dev->dev_private; + struct ved_private *ved_priv = dev_priv->ved_private; + int32_t offset; + int ret; + + if (ved_priv->fw_loaded_by_punit) + ved_priv->ved_needs_reset = MSVDX_RESET_NEEDS_INIT_FW; + else + ved_priv->ved_needs_reset = 1; + +#ifdef CONFIG_DRM_IPVR_EC + /* we should restore the state, if we power down/up during EC */ + for (offset = 0; offset < 4; ++offset) + ved_priv->vec_ec_mem_data[offset] = + VED_REG_READ32(0x2cb0 + offset * 4); + + ved_priv->vec_ec_mem_data[4] = VED_REG_READ32(0x2cc4); + + ved_priv->vec_ec_mem_saved = 1; + IPVR_DEBUG_VED("ec last mb %d %d %d %d\n", + ved_priv->vec_ec_mem_data[0], + ved_priv->vec_ec_mem_data[1], + ved_priv->vec_ec_mem_data[2], + ved_priv->vec_ec_mem_data[3]); + IPVR_DEBUG_VED("ec error state %d\n", ved_priv->vec_ec_mem_data[4]); +#endif + + /* Reset MTX */ + VED_REG_WRITE32(MTX_SOFT_RESET_MTXRESET, MTX_SOFT_RESET_OFFSET); + + /* why need reset msvdx before power off it, need check IMG */ + ret = ved_core_reset(dev_priv); + if (unlikely(ret)) + IPVR_DEBUG_WARN("failed to call ved_core_reset: %d\n", ret); + + /* Initialize VEC Local RAM */ + for (offset = 0; offset < VEC_LOCAL_MEM_BYTE_SIZE / 4; ++offset) + VED_REG_WRITE32(0, VEC_LOCAL_MEM_OFFSET + offset * 4); + + if (ved_priv->fw_loaded_by_punit) { + VED_REG_WRITE32(0, MTX_ENABLE_OFFSET); + ved_set_clocks(dev_priv->dev, 0); + } + + return 0; +} + +static u32 __ipvr_msgbus_read32(struct pci_dev *pci_root, u8 port, u32 addr) +{ + uint32_t data; + uint32_t cmd; + uint32_t cmdext; + + cmd = (PCI_ROOT_MSGBUS_READ << 24) | (port << 16) | + ((addr & 0xff) << 8) | PCI_ROOT_MSGBUS_DWORD_ENABLE; + cmdext = addr & 0xffffff00; + + if (cmdext) { + /* This resets to 0 automatically, no need to write 0 */ + pci_write_config_dword(pci_root, PCI_ROOT_MSGBUS_CTRL_EXT_REG, + cmdext); + } + + pci_write_config_dword(pci_root, PCI_ROOT_MSGBUS_CTRL_REG, cmd); + pci_read_config_dword(pci_root, PCI_ROOT_MSGBUS_DATA_REG, &data); + + return data; +} + +static void __ipvr_msgbus_write32(struct pci_dev *pci_root, u8 port, u32 addr, u32 data) +{ + uint32_t cmd; + uint32_t cmdext; + + cmd = (PCI_ROOT_MSGBUS_WRITE << 24) | (port << 16) | + ((addr & 0xFF) << 8) | PCI_ROOT_MSGBUS_DWORD_ENABLE; + cmdext = addr & 0xffffff00; + + pci_write_config_dword(pci_root, PCI_ROOT_MSGBUS_DATA_REG, data); + + if (cmdext) { + /* This resets to 0 automatically, no need to write 0 */ + pci_write_config_dword(pci_root, PCI_ROOT_MSGBUS_CTRL_EXT_REG, + cmdext); + } + + pci_write_config_dword(pci_root, PCI_ROOT_MSGBUS_CTRL_REG, cmd); +} + +static int __ipvr_pm_cmd_freq_wait(struct pci_dev *pci_root, u32 reg_freq, u32 *freq_code_rlzd) +{ + int32_t tcount; + uint32_t freq_val; + + for (tcount = 0; ; tcount++) { + freq_val = __ipvr_msgbus_read32(pci_root, PUNIT_PORT, reg_freq); + if ((freq_val & IP_FREQ_VALID) == 0) + break; + if (tcount > 500) { + IPVR_ERROR("P-Unit freq request wait timeout %x", + freq_val); + return -EBUSY; + } + udelay(1); + } + + if (freq_code_rlzd) { + *freq_code_rlzd = ((freq_val >> IP_FREQ_STAT_POS) & + IP_FREQ_MASK); + } + + return 0; +} + +static int32_t __ipvr_pm_cmd_freq_get(struct pci_dev *pci_root, u32 reg_freq) +{ + uint32_t freq_val; + int32_t freq_code=0; + + __ipvr_pm_cmd_freq_wait(pci_root, reg_freq, NULL); + + freq_val = __ipvr_msgbus_read32(pci_root, PUNIT_PORT, reg_freq); + freq_code =(int)((freq_val>>IP_FREQ_STAT_POS) & ~IP_FREQ_VALID); + return freq_code; +} + +static int32_t __ipvr_pm_cmd_freq_set(struct pci_dev *pci_root, u32 reg_freq, u32 freq_code, u32 *p_freq_code_rlzd) +{ + uint32_t freq_val; + uint32_t freq_code_realized; + int32_t rva; + + rva = __ipvr_pm_cmd_freq_wait(pci_root, reg_freq, NULL); + if (rva < 0) { + IPVR_ERROR("pm_cmd_freq_wait 1 failed: %d\n", rva); + return rva; + } + + freq_val = IP_FREQ_VALID | freq_code; + __ipvr_msgbus_write32(pci_root, PUNIT_PORT, reg_freq, freq_val); + + rva = __ipvr_pm_cmd_freq_wait(pci_root, reg_freq, &freq_code_realized); + if (rva < 0) { + IPVR_ERROR("pm_cmd_freq_wait 2 failed: %d\n", rva); + return rva; + } + + if (p_freq_code_rlzd) + *p_freq_code_rlzd = freq_code_realized; + + return rva; +} + +static int32_t ved_set_freq(struct drm_device *dev, u32 freq_code) +{ + uint32_t freq_code_rlzd = 0; + int32_t ret; + struct drm_ipvr_private *dev_priv = (struct drm_ipvr_private*)dev->dev_private; + + ret = __ipvr_pm_cmd_freq_set(dev_priv->pci_root, VEDSSPM1, freq_code, &freq_code_rlzd); + if (ret < 0) { + IPVR_ERROR("failed to set freqency, current is %x\n", + freq_code_rlzd); + } + + return ret; +} + +static int32_t ved_get_freq(struct drm_device *dev) +{ + struct drm_ipvr_private *dev_priv = (struct drm_ipvr_private*)dev->dev_private; + return __ipvr_pm_cmd_freq_get(dev_priv->pci_root, VEDSSPM1); +} + +#ifdef CONFIG_INTEL_SOC_PMC +bool ved_power_on(struct drm_device *dev) +{ + int32_t ved_freq; + int32_t ret; + IPVR_DEBUG_PM("VED: power on msvdx using ATOM_PMC.\n"); + + ret = pmc_nc_set_power_state(VEDSSC, 0, VEDSSPM0); + if (unlikely(ret)) { + IPVR_ERROR("VED: pmu_nc_set_power_state ON fail!\n"); + return false; + } + + ved_freq = ved_get_freq(dev); + + IPVR_DEBUG_PM("VED freqency: code %d (%dMHz)\n", ved_freq, GET_VED_FREQUENCY(ved_freq)); + + trace_ved_power_on(GET_VED_FREQUENCY(ved_freq)); + return true; +} +bool ved_power_off(struct drm_device *dev) +{ + int32_t ved_freq; + int32_t ret; + struct drm_ipvr_private *dev_priv = (struct drm_ipvr_private*)dev->dev_private; + IPVR_DEBUG_PM("VED: power off msvdx using ATOM_PMC.\n"); + + if (dev_priv->ved_private) { + ret = ved_save_context(dev); + if (unlikely(ret)) { + IPVR_ERROR("Failed to save VED context, stop powering off\n"); + return false; + } + } + + ved_freq = ved_get_freq(dev); + IPVR_DEBUG_PM("VED freqency: code %d (%dMHz)\n", ved_freq, GET_VED_FREQUENCY(ved_freq)); + + ret = pmc_nc_set_power_state(VEDSSC, 1, VEDSSPM0); + if (unlikely(ret)) { + IPVR_ERROR("VED: pmu_nc_set_power_state DOWN fail!\n"); + return false; + } + + trace_ved_power_off(GET_VED_FREQUENCY(ved_freq)); + return true; +} +#else +bool ved_power_on(struct drm_device *dev) +{ + int32_t ved_freq_before, ved_freq_after; + uint32_t pwr_sts; + struct drm_ipvr_private *dev_priv = (struct drm_ipvr_private*)dev->dev_private; + IPVR_DEBUG_PM("VED: power on msvdx.\n"); + + /* FIXME: add timeout check */ + do { + __ipvr_msgbus_write32(dev_priv->pci_root, PUNIT_PORT, VEDSSPM0, VED_APM_STS_D0); + udelay(10); + pwr_sts = __ipvr_msgbus_read32(dev_priv->pci_root, PUNIT_PORT, VEDSSPM0); + } while (pwr_sts != 0x0); + + do { + __ipvr_msgbus_write32(dev_priv->pci_root, PUNIT_PORT, VEDSSPM0, VED_APM_STS_D3); + udelay(10); + pwr_sts = __ipvr_msgbus_read32(dev_priv->pci_root, PUNIT_PORT, VEDSSPM0); + } while (pwr_sts != 0x03000003); + + do { + __ipvr_msgbus_write32(dev_priv->pci_root, PUNIT_PORT, VEDSSPM0, VED_APM_STS_D0); + udelay(10); + pwr_sts = __ipvr_msgbus_read32(dev_priv->pci_root, PUNIT_PORT, VEDSSPM0); + } while (pwr_sts != 0x0); + + ved_freq_before = ved_get_freq(dev); + + if (ved_set_freq(dev, drm_ipvr_freq)) { + IPVR_ERROR("Failed to set VED frequency\n"); + } + + ved_freq_after = ved_get_freq(dev); + IPVR_DEBUG_PM("VED freqency: %dMHz => %dMHz\n", + GET_VED_FREQUENCY(ved_freq_before), GET_VED_FREQUENCY(ved_freq_after)); + + trace_ved_power_on(GET_VED_FREQUENCY(ved_freq_after)); + return true; +} + +bool ved_power_off(struct drm_device *dev) +{ + uint32_t pwr_sts; + int32_t ved_freq; + int32_t ret; + struct drm_ipvr_private *dev_priv = (struct drm_ipvr_private*)dev->dev_private; + IPVR_DEBUG_PM("VED: power off msvdx.\n"); + + if (dev_priv->ved_private) { + ret = ved_save_context(dev); + if (unlikely(ret)) { + IPVR_ERROR("Failed to save VED context: %d, stop powering off\n", ret); + return false; + } + } + + ved_freq = ved_get_freq(dev); + IPVR_DEBUG_PM("VED freqency: code %d (%dMHz)\n", ved_freq, GET_VED_FREQUENCY(ved_freq)); + + /* FIXME: add timeout check */ + do { + __ipvr_msgbus_write32(dev_priv->pci_root, PUNIT_PORT, VEDSSPM0, VED_APM_STS_D3); + udelay(10); + pwr_sts = __ipvr_msgbus_read32(dev_priv->pci_root, PUNIT_PORT, VEDSSPM0); + } while (pwr_sts != 0x03000003); + + trace_ved_power_off(GET_VED_FREQUENCY(ved_freq)); + return true; +} + +#endif +/** + * is_island_on + * + * Description: checks to see if the island is up + * returns true if hw_island is ON + * returns false if hw_island is OFF + */ +bool is_ved_on(struct drm_device *dev) +{ + uint32_t pwr_sts; + struct drm_ipvr_private *dev_priv = dev->dev_private; + + pwr_sts = __ipvr_msgbus_read32(dev_priv->pci_root, PUNIT_PORT, VEDSSPM0); + + return (pwr_sts == VED_APM_STS_D0); +} diff --git a/drivers/gpu/drm/ipvr/ved_pm.h b/drivers/gpu/drm/ipvr/ved_pm.h new file mode 100644 index 0000000..1f00f8b --- /dev/null +++ b/drivers/gpu/drm/ipvr/ved_pm.h @@ -0,0 +1,55 @@ +/************************************************************************** + * ved_pm.h: VED power management header file + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + +#ifndef _VED_PM_H_ +#define _VED_PM_H_ + +#include "ipvr_drv.h" + +#define IP_FREQ_100_00 0x1f /* 0b11111 100.00 */ +#define IP_FREQ_106_67 0x1d /* 0b11101 106.67 */ +#define IP_FREQ_133_30 0x17 /* 0b10111 133.30 */ +#define IP_FREQ_160_00 0x13 /* 0b10011 160.00 */ +#define IP_FREQ_177_78 0x11 /* 0b10001 177.78 */ +#define IP_FREQ_200_00 0x0f /* 0b01111 200.00 */ +#define IP_FREQ_213_33 0x0e /* 0b01110 213.33 */ +#define IP_FREQ_266_67 0x0b /* 0b01011 266.67 */ +#define IP_FREQ_320_00 0x09 /* 0b01001 320.00 */ +#define IP_FREQ_355_56 0x08 /* 0b01000 355.56 */ +#define IP_FREQ_400_00 0x07 /* 0b00111 400.00 */ +#define IP_FREQ_457_14 0x06 /* 0b00110 457.14 */ +#define IP_FREQ_533_33 0x05 /* 0b00101 533.33 */ +#define IP_FREQ_640_00 0x04 /* 0b00100 640.00 */ +#define IP_FREQ_800_00 0x03 /* 0b00011 800.00 */ + +#define GET_VED_FREQUENCY(freq_code) ((1600 * 2)/((freq_code) + 1)) + +bool is_ved_on(struct drm_device *dev); + +bool ved_power_on(struct drm_device *dev); + +bool ved_power_off(struct drm_device *dev); + +#endif diff --git a/drivers/gpu/drm/ipvr/ved_reg.h b/drivers/gpu/drm/ipvr/ved_reg.h new file mode 100644 index 0000000..1d55f2a --- /dev/null +++ b/drivers/gpu/drm/ipvr/ved_reg.h @@ -0,0 +1,609 @@ +/************************************************************************** + * ved_reg.h: VED register definition + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * Copyright (c) Imagination Technologies Limited, UK + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * + **************************************************************************/ + +#ifndef _VED_REG_H_ +#define _VED_REG_H_ + +#include "ipvr_drv.h" + +#if (defined MFLD_MSVDX_FABRIC_DEBUG) && MFLD_MSVDX_FABRIC_DEBUG +#define VED_REG_WRITE32(_val, _offs) \ +do { \ + if (ipvr_get_power_state(OSPM_VIDEO_DEC_ISLAND) == 0) \ + panic("msvdx reg 0x%x write failed.\n", \ + (unsigned int)(_offs)); \ + else \ + iowrite32(_val, dev_priv->ved_reg_base + (_offs)); \ +} while (0) + +static inline uint32_t VED_REG_READ32(uint32_t _offs) +{ + struct drm_ipvr_private *dev_priv = + (struct drm_ipvr_private *)gpDrmDevice->dev_private; + if (ipvr_get_power_state(OSPM_VIDEO_DEC_ISLAND) == 0) { + panic("msvdx reg 0x%x read failed.\n", (unsigned int)(_offs)); + return 0; + } else { + return ioread32(dev_priv->ved_reg_base + (_offs)); + } +} + +#elif (defined MSVDX_REG_DUMP) && MSVDX_REG_DUMP + +#define VED_REG_WRITE32(_val, _offs) \ +do { \ + printk(KERN_INFO"MSVDX: write %08x to reg 0x%08x\n", \ + (unsigned int)(_val), \ + (unsigned int)(_offs)); \ + iowrite32(_val, dev_priv->ved_reg_base + (_offs)); \ +} while (0) + +static inline uint32_t VED_REG_READ32(uint32_t _offs) +{ + uint32_t val = ioread32(dev_priv->ved_reg_base + (_offs)); + printk(KERN_INFO"MSVDX: read reg 0x%08x, get %08x\n", + (unsigned int)(_offs), val); + return val; +} + +#else + +#define VED_REG_WRITE32(_val, _offs) \ + iowrite32(_val, dev_priv->ved_reg_base + (_offs)) +#define VED_REG_READ32(_offs) \ + ioread32(dev_priv->ved_reg_base + (_offs)) + +#endif + +#define REGISTER(__group__, __reg__) (__group__##_##__reg__##_OFFSET) + +#define MTX_INTERNAL_REG(R_SPECIFIER , U_SPECIFIER) \ + (((R_SPECIFIER)<<4) | (U_SPECIFIER)) +#define MTX_PC MTX_INTERNAL_REG(0, 5) + +#define MEMIO_READ_FIELD(vpMem, field) \ + ((uint32_t)(((*((field##_TYPE*)(((uint32_t)vpMem) + field##_OFFSET))) \ + & field##_MASK) >> field##_SHIFT)) \ + +#define MEMIO_WRITE_FIELD(vpMem, field, value) \ +do { \ + ((*((field##_TYPE*)(((uint32_t)vpMem) + field##_OFFSET))) = \ + ((*((field##_TYPE*)(((uint32_t)vpMem) + field##_OFFSET))) \ + & (field##_TYPE)~field##_MASK) | \ + (field##_TYPE)(((uint32_t)(value) << field##_SHIFT) & field##_MASK)); \ +} while (0) + +#define MEMIO_WRITE_FIELD_LITE(vpMem, field, value) \ +do { \ + (*((field##_TYPE*)(((uint32_t)vpMem) + field##_OFFSET))) = \ + ((*((field##_TYPE*)(((uint32_t)vpMem) + field##_OFFSET))) | \ + (field##_TYPE)(((uint32_t)(value) << field##_SHIFT))); \ +} while (0) + +#define REGIO_READ_FIELD(reg_val, reg, field) \ + ((reg_val & reg##_##field##_MASK) >> reg##_##field##_SHIFT) + +#define REGIO_WRITE_FIELD(reg_val, reg, field, value) \ +do { \ + (reg_val) = \ + ((reg_val) & ~(reg##_##field##_MASK)) | \ + (((value) << (reg##_##field##_SHIFT)) & (reg##_##field##_MASK)); \ +} while (0) + + +#define REGIO_WRITE_FIELD_LITE(reg_val, reg, field, value) \ +do { \ + (reg_val) = ((reg_val) | ((value) << (reg##_##field##_SHIFT))); \ +} while (0) + +/****** MSVDX.Technical Reference Manual.2.0.2.4.External VXD38x ************** +Offset address Name Identifier +0x0000 - 0x03FF (1024B) MTX Register REG_MSVDX_MTX +0x0400 - 0x047F (128B) VDMC Register REG_MSVDX _VDMC +0x0480 - 0x04FF (128B) VDEB Register REG_MSVDX _VDEB +0x0500 - 0x05FF (256B) DMAC Register REG_MSVDX _DMAC +0x0600 - 0x06FF (256B) MSVDX Core Register REG_MSVDX _SYS +0x0700 - 0x07FF (256B) VEC iQ Matrix RAM REG_MSVDX_VEC_IQRAM +0x0800 - 0x0FFF (2048B) VEC Registers REG_MSVDX _VEC +0x1000 - 0x1FFF (4kB) Command Register REG_MSVDX _CMD +0x2000 - 0x2FFF (4kB) VEC Local RAM REG_MSVDX _VEC_RAM +0x3000 - 0x4FFF (8kB) VEC VLC Table RAM REG_MSVDX _VEC_VLC +0x5000 - 0x5FFF (4kB) AXI Register REG_MSVDX _AXI +******************************************************************************/ + +/*************** MTX registers start: 0x0000 - 0x03FF (1024B) ****************/ +#define MTX_ENABLE_OFFSET (0x0000) +#define MTX_ENABLE_MTX_ENABLE_MASK (0x00000001) +#define MTX_ENABLE_MTX_ENABLE_SHIFT (0) + +#define MTX_KICK_INPUT_OFFSET (0x0080) + +#define MTX_REGISTER_READ_WRITE_REQUEST_OFFSET (0x00FC) +#define MTX_REGISTER_READ_WRITE_REQUEST_MTX_DREADY_MASK (0x80000000) +#define MTX_REGISTER_READ_WRITE_REQUEST_MTX_DREADY_SHIFT (31) +#define MTX_REGISTER_READ_WRITE_REQUEST_MTX_RNW_MASK (0x00010000) +#define MTX_REGISTER_READ_WRITE_REQUEST_MTX_RNW_SHIFT (16) + +#define MTX_REGISTER_READ_WRITE_DATA_OFFSET (0x00F8) + +#define MTX_RAM_ACCESS_DATA_TRANSFER_OFFSET (0x0104) + +#define MTX_RAM_ACCESS_CONTROL_OFFSET (0x0108) +#define MTX_RAM_ACCESS_CONTROL_MTX_MCMID_MASK (0x0FF00000) +#define MTX_RAM_ACCESS_CONTROL_MTX_MCMID_SHIFT (20) +#define MTX_RAM_ACCESS_CONTROL_MTX_MCM_ADDR_MASK (0x000FFFFC) +#define MTX_RAM_ACCESS_CONTROL_MTX_MCM_ADDR_SHIFT (2) +#define MTX_RAM_ACCESS_CONTROL_MTX_MCMAI_MASK (0x00000002) +#define MTX_RAM_ACCESS_CONTROL_MTX_MCMAI_SHIFT (1) +#define MTX_RAM_ACCESS_CONTROL_MTX_MCMR_MASK (0x00000001) +#define MTX_RAM_ACCESS_CONTROL_MTX_MCMR_SHIFT (0) + +#define MTX_RAM_ACCESS_STATUS_OFFSET (0x010C) + +#define MTX_SOFT_RESET_OFFSET (0x0200) +#define MTX_SOFT_RESET_MTX_RESET_MASK (0x00000001) +#define MTX_SOFT_RESET_MTX_RESET_SHIFT (0) +#define MTX_SOFT_RESET_MTXRESET (0x00000001) + +#define MTX_SYSC_TIMERDIV_OFFSET (0x0208) + +#define MTX_SYSC_CDMAC_OFFSET (0x0340) +#define MTX_SYSC_CDMAC_BURSTSIZE_MASK (0x07000000) +#define MTX_SYSC_CDMAC_BURSTSIZE_SHIFT (24) +#define MTX_SYSC_CDMAC_RNW_MASK (0x00020000) +#define MTX_SYSC_CDMAC_RNW_SHIFT (17) +#define MTX_SYSC_CDMAC_ENABLE_MASK (0x00010000) +#define MTX_SYSC_CDMAC_ENABLE_SHIFT (16) +#define MTX_SYSC_CDMAC_LENGTH_MASK (0x0000FFFF) +#define MTX_SYSC_CDMAC_LENGTH_SHIFT (0) + +#define MTX_SYSC_CDMAA_OFFSET (0x0344) + +#define MTX_SYSC_CDMAS0_OFFSET (0x0348) + +#define MTX_SYSC_CDMAT_OFFSET (0x0350) +/************************** MTX registers end **************************/ + +/**************** DMAC Registers: 0x0500 - 0x05FF (256B) ***************/ +#define DMAC_DMAC_COUNT_EN_MASK (0x00010000) +#define DMAC_DMAC_IRQ_STAT_TRANSFER_FIN_MASK (0x00020000) + +#define DMAC_DMAC_SETUP_OFFSET (0x0500) + +#define DMAC_DMAC_COUNT_OFFSET (0x0504) +#define DMAC_DMAC_COUNT_BSWAP_LSBMASK (0x00000001) +#define DMAC_DMAC_COUNT_BSWAP_SHIFT (30) +#define DMAC_DMAC_COUNT_PW_LSBMASK (0x00000003) +#define DMAC_DMAC_COUNT_PW_SHIFT (27) +#define DMAC_DMAC_COUNT_DIR_LSBMASK (0x00000001) +#define DMAC_DMAC_COUNT_DIR_SHIFT (26) +#define DMAC_DMAC_COUNT_PI_LSBMASK (0x00000003) +#define DMAC_DMAC_COUNT_PI_SHIFT (24) +#define DMAC_DMAC_COUNT_CNT_LSBMASK (0x0000FFFF) +#define DMAC_DMAC_COUNT_CNT_SHIFT (0) +#define DMAC_DMAC_COUNT_EN_MASK (0x00010000) +#define DMAC_DMAC_COUNT_EN_SHIFT (16) + +#define DMAC_DMAC_PERIPH_OFFSET (0x0508) +#define DMAC_DMAC_PERIPH_ACC_DEL_LSBMASK (0x00000007) +#define DMAC_DMAC_PERIPH_ACC_DEL_SHIFT (29) +#define DMAC_DMAC_PERIPH_INCR_LSBMASK (0x00000001) +#define DMAC_DMAC_PERIPH_INCR_SHIFT (27) +#define DMAC_DMAC_PERIPH_BURST_LSBMASK (0x00000007) +#define DMAC_DMAC_PERIPH_BURST_SHIFT (24) + +#define DMAC_DMAC_IRQ_STAT_OFFSET (0x050C) +#define DMAC_DMAC_IRQ_STAT_TRANSFER_FIN_MASK (0x00020000) + +#define DMAC_DMAC_PERIPHERAL_ADDR_OFFSET (0x0514) +#define DMAC_DMAC_PERIPHERAL_ADDR_ADDR_MASK (0x007FFFFF) +#define DMAC_DMAC_PERIPHERAL_ADDR_ADDR_LSBMASK (0x007FFFFF) +#define DMAC_DMAC_PERIPHERAL_ADDR_ADDR_SHIFT (0) + +/* DMAC control */ +#define PSB_DMAC_VALUE_COUNT(BSWAP, PW, DIR, PERIPH_INCR, COUNT) \ + ((((BSWAP) & DMAC_DMAC_COUNT_BSWAP_LSBMASK) << \ + DMAC_DMAC_COUNT_BSWAP_SHIFT) | \ + (((PW) & DMAC_DMAC_COUNT_PW_LSBMASK) << \ + DMAC_DMAC_COUNT_PW_SHIFT) | \ + (((DIR) & DMAC_DMAC_COUNT_DIR_LSBMASK) << \ + DMAC_DMAC_COUNT_DIR_SHIFT) | \ + (((PERIPH_INCR) & DMAC_DMAC_COUNT_PI_LSBMASK) << \ + DMAC_DMAC_COUNT_PI_SHIFT) | \ + (((COUNT) & DMAC_DMAC_COUNT_CNT_LSBMASK) << \ + DMAC_DMAC_COUNT_CNT_SHIFT)) + +#define PSB_DMAC_VALUE_PERIPH_PARAM(ACC_DEL, INCR, BURST) \ + ((((ACC_DEL) & DMAC_DMAC_PERIPH_ACC_DEL_LSBMASK) << \ + DMAC_DMAC_PERIPH_ACC_DEL_SHIFT) | \ + (((INCR) & DMAC_DMAC_PERIPH_INCR_LSBMASK) << \ + DMAC_DMAC_PERIPH_INCR_SHIFT) | \ + (((BURST) & DMAC_DMAC_PERIPH_BURST_LSBMASK) << \ + DMAC_DMAC_PERIPH_BURST_SHIFT)) + +typedef enum { + /* !< No byte swapping will be performed. */ + PSB_DMAC_BSWAP_NO_SWAP = 0x0, + /* !< Byte order will be reversed. */ + PSB_DMAC_BSWAP_REVERSE = 0x1, +} DMAC_eBSwap; + +typedef enum { + /* !< Data from memory to peripheral. */ + PSB_DMAC_DIR_MEM_TO_PERIPH = 0x0, + /* !< Data from peripheral to memory. */ + PSB_DMAC_DIR_PERIPH_TO_MEM = 0x1, +} DMAC_eDir; + +typedef enum { + PSB_DMAC_ACC_DEL_0 = 0x0, /* !< Access delay zero clock cycles */ + PSB_DMAC_ACC_DEL_256 = 0x1, /* !< Access delay 256 clock cycles */ + PSB_DMAC_ACC_DEL_512 = 0x2, /* !< Access delay 512 clock cycles */ + PSB_DMAC_ACC_DEL_768 = 0x3, /* !< Access delay 768 clock cycles */ + PSB_DMAC_ACC_DEL_1024 = 0x4, /* !< Access delay 1024 clock cycles */ + PSB_DMAC_ACC_DEL_1280 = 0x5, /* !< Access delay 1280 clock cycles */ + PSB_DMAC_ACC_DEL_1536 = 0x6, /* !< Access delay 1536 clock cycles */ + PSB_DMAC_ACC_DEL_1792 = 0x7, /* !< Access delay 1792 clock cycles */ +} DMAC_eAccDel; + +typedef enum { + PSB_DMAC_INCR_OFF = 0, /* !< Static peripheral address. */ + PSB_DMAC_INCR_ON = 1, /* !< Incrementing peripheral address. */ +} DMAC_eIncr; + +typedef enum { + PSB_DMAC_BURST_0 = 0x0, /* !< burst size of 0 */ + PSB_DMAC_BURST_1 = 0x1, /* !< burst size of 1 */ + PSB_DMAC_BURST_2 = 0x2, /* !< burst size of 2 */ + PSB_DMAC_BURST_3 = 0x3, /* !< burst size of 3 */ + PSB_DMAC_BURST_4 = 0x4, /* !< burst size of 4 */ + PSB_DMAC_BURST_5 = 0x5, /* !< burst size of 5 */ + PSB_DMAC_BURST_6 = 0x6, /* !< burst size of 6 */ + PSB_DMAC_BURST_7 = 0x7, /* !< burst size of 7 */ +} DMAC_eBurst; +/************************** DMAC Registers end **************************/ + +/**************** MSVDX Core Registers: 0x0600 - 0x06FF (256B) ***************/ +#define MSVDX_CONTROL_OFFSET (0x0600) +#define MSVDX_CONTROL_MSVDX_SOFT_RESET_MASK (0x00000100) +#define MSVDX_CONTROL_MSVDX_SOFT_RESET_SHIFT (8) +#define MSVDX_CONTROL_DMAC_CH0_SELECT_MASK (0x00001000) +#define MSVDX_CONTROL_DMAC_CH0_SELECT_SHIFT (12) +#define MSVDX_CONTROL_MSVDX_SOFT_RESET_MASK (0x00000100) +#define MSVDX_CONTROL_MSVDX_SOFT_RESET_SHIFT (8) +#define MSVDX_CONTROL_MSVDX_FE_SOFT_RESET_MASK (0x00010000) +#define MSVDX_CONTROL_MSVDX_BE_SOFT_RESET_MASK (0x00100000) +#define MSVDX_CONTROL_MSVDX_VEC_MEMIF_SOFT_RESET_MASK (0x01000000) +#define MSVDX_CONTROL_MSVDX_VEC_RENDEC_DEC_SOFT_RESET_MASK (0x10000000) +#define msvdx_sw_reset_all \ + (MSVDX_CONTROL_MSVDX_SOFT_RESET_MASK | \ + MSVDX_CONTROL_MSVDX_FE_SOFT_RESET_MASK | \ + MSVDX_CONTROL_MSVDX_BE_SOFT_RESET_MASK | \ + MSVDX_CONTROL_MSVDX_VEC_MEMIF_SOFT_RESET_MASK | \ + MSVDX_CONTROL_MSVDX_VEC_RENDEC_DEC_SOFT_RESET_MASK) + +#define MSVDX_INTERRUPT_CLEAR_OFFSET (0x060C) + +#define MSVDX_INTERRUPT_STATUS_OFFSET (0x0608) +#define MSVDX_INTERRUPT_STATUS_MMU_FAULT_IRQ_MASK (0x00000F00) +#define MSVDX_INTERRUPT_STATUS_MMU_FAULT_IRQ_SHIFT (8) +#define MSVDX_INTERRUPT_STATUS_MTX_IRQ_MASK (0x00004000) +#define MSVDX_INTERRUPT_STATUS_MTX_IRQ_SHIFT (14) + +#define MSVDX_HOST_INTERRUPT_ENABLE_OFFSET (0x0610) + +#define MSVDX_MAN_CLK_ENABLE_OFFSET (0x0620) +#define MSVDX_MAN_CLK_ENABLE_CORE_MAN_CLK_ENABLE_MASK (0x00000001) +#define MSVDX_MAN_CLK_ENABLE_VDEB_PROCESS_MAN_CLK_ENABLE_MASK (0x00000002) +#define MSVDX_MAN_CLK_ENABLE_VDEB_ACCESS_MAN_CLK_ENABLE_MASK (0x00000004) +#define MSVDX_MAN_CLK_ENABLE_VDMC_MAN_CLK_ENABLE_MASK (0x00000008) +#define MSVDX_MAN_CLK_ENABLE_VEC_ENTDEC_MAN_CLK_ENABLE_MASK (0x00000010) +#define MSVDX_MAN_CLK_ENABLE_VEC_ITRANS_MAN_CLK_ENABLE_MASK (0x00000020) +#define MSVDX_MAN_CLK_ENABLE_MTX_MAN_CLK_ENABLE_MASK (0x00000040) +#define MSVDX_MAN_CLK_ENABLE_VDEB_PROCESS_AUTO_CLK_ENABLE_MASK (0x00020000) +#define MSVDX_MAN_CLK_ENABLE_VDEB_ACCESS_AUTO_CLK_ENABLE_MASK (0x00040000) +#define MSVDX_MAN_CLK_ENABLE_VDMC_AUTO_CLK_ENABLE_MASK (0x00080000) +#define MSVDX_MAN_CLK_ENABLE_VEC_ENTDEC_AUTO_CLK_ENABLE_MASK (0x00100000) +#define MSVDX_MAN_CLK_ENABLE_VEC_ITRANS_AUTO_CLK_ENABLE_MASK (0x00200000) + +#define clk_enable_all \ + (MSVDX_MAN_CLK_ENABLE_CORE_MAN_CLK_ENABLE_MASK | \ + MSVDX_MAN_CLK_ENABLE_VDEB_PROCESS_MAN_CLK_ENABLE_MASK | \ + MSVDX_MAN_CLK_ENABLE_VDEB_ACCESS_MAN_CLK_ENABLE_MASK | \ + MSVDX_MAN_CLK_ENABLE_VDMC_MAN_CLK_ENABLE_MASK | \ + MSVDX_MAN_CLK_ENABLE_VEC_ENTDEC_MAN_CLK_ENABLE_MASK | \ + MSVDX_MAN_CLK_ENABLE_VEC_ITRANS_MAN_CLK_ENABLE_MASK | \ + MSVDX_MAN_CLK_ENABLE_MTX_MAN_CLK_ENABLE_MASK) + +#define clk_enable_minimal \ + (MSVDX_MAN_CLK_ENABLE_CORE_MAN_CLK_ENABLE_MASK | \ + MSVDX_MAN_CLK_ENABLE_MTX_MAN_CLK_ENABLE_MASK) + +#define clk_enable_auto \ + (MSVDX_MAN_CLK_ENABLE_VDEB_PROCESS_AUTO_CLK_ENABLE_MASK | \ + MSVDX_MAN_CLK_ENABLE_VDEB_ACCESS_AUTO_CLK_ENABLE_MASK | \ + MSVDX_MAN_CLK_ENABLE_VDMC_AUTO_CLK_ENABLE_MASK | \ + MSVDX_MAN_CLK_ENABLE_VEC_ENTDEC_AUTO_CLK_ENABLE_MASK | \ + MSVDX_MAN_CLK_ENABLE_VEC_ITRANS_AUTO_CLK_ENABLE_MASK | \ + MSVDX_MAN_CLK_ENABLE_CORE_MAN_CLK_ENABLE_MASK | \ + MSVDX_MAN_CLK_ENABLE_MTX_MAN_CLK_ENABLE_MASK) + +#define MSVDX_CORE_ID_OFFSET (0x0630) +#define MSVDX_CORE_REV_OFFSET (0x0640) + +#define MSVDX_DMAC_STREAM_STATUS_OFFSET (0x0648) + +#define MSVDX_MMU_CONTROL0_OFFSET (0x0680) +#define MSVDX_MMU_CONTROL0_MMU_PAUSE_MASK (0x00000002) +#define MSVDX_MMU_CONTROL0_MMU_PAUSE_SHIFT (1) +#define MSVDX_MMU_CONTROL0_MMU_INVALDC_MASK (0x00000008) +#define MSVDX_MMU_CONTROL0_MMU_INVALDC_SHIFT (3) + +#define MSVDX_MMU_BANK_INDEX_OFFSET (0x0688) + +#define MSVDX_MMU_STATUS_OFFSET (0x068C) + +#define MSVDX_MMU_CONTROL2_OFFSET (0x0690) + +#define MSVDX_MMU_DIR_LIST_BASE_OFFSET (0x0694) + +#define MSVDX_MMU_MEM_REQ_OFFSET (0x06D0) + +#define MSVDX_MMU_TILE_BASE0_OFFSET (0x06D4) + +#define MSVDX_MMU_TILE_BASE1_OFFSET (0x06D8) + +#define MSVDX_MTX_RAM_BANK_OFFSET (0x06F0) +#define MSVDX_MTX_RAM_BANK_MTX_RAM_BANK_SIZE_MASK (0x000F0000) +#define MSVDX_MTX_RAM_BANK_MTX_RAM_BANK_SIZE_SHIFT (16) + +#define MSVDX_MTX_DEBUG_OFFSET MSVDX_MTX_RAM_BANK_OFFSET +#define MSVDX_MTX_DEBUG_MTX_DBG_IS_SLAVE_MASK (0x00000004) +#define MSVDX_MTX_DEBUG_MTX_DBG_IS_SLAVE_LSBMASK (0x00000001) +#define MSVDX_MTX_DEBUG_MTX_DBG_IS_SLAVE_SHIFT (2) +#define MSVDX_MTX_DEBUG_MTX_DBG_GPIO_IN_MASK (0x00000003) +#define MSVDX_MTX_DEBUG_MTX_DBG_GPIO_IN_LSBMASK (0x00000003) +#define MSVDX_MTX_DEBUG_MTX_DBG_GPIO_IN_SHIFT (0) + +/*watch dog for FE and BE*/ +#define FE_MSVDX_WDT_CONTROL_OFFSET (0x0664) +/* MSVDX_CORE, CR_FE_MSVDX_WDT_CONTROL, FE_WDT_CNT_CTRL */ +#define FE_MSVDX_WDT_CONTROL_FE_WDT_CNT_CTRL_MASK (0x00060000) +#define FE_MSVDX_WDT_CONTROL_FE_WDT_CNT_CTRL_LSBMASK (0x00000003) +#define FE_MSVDX_WDT_CONTROL_FE_WDT_CNT_CTRL_SHIFT (17) +/* MSVDX_CORE, CR_FE_MSVDX_WDT_CONTROL, FE_WDT_ENABLE */ +#define FE_MSVDX_WDT_CONTROL_FE_WDT_ENABLE_MASK (0x00010000) +#define FE_MSVDX_WDT_CONTROL_FE_WDT_ENABLE_LSBMASK (0x00000001) +#define FE_MSVDX_WDT_CONTROL_FE_WDT_ENABLE_SHIFT (16) +/* MSVDX_CORE, CR_FE_MSVDX_WDT_CONTROL, FE_WDT_ACTION1 */ +#define FE_MSVDX_WDT_CONTROL_FE_WDT_ACTION1_MASK (0x00003000) +#define FE_MSVDX_WDT_CONTROL_FE_WDT_ACTION1_LSBMASK (0x00000003) +#define FE_MSVDX_WDT_CONTROL_FE_WDT_ACTION1_SHIFT (12) +/* MSVDX_CORE, CR_FE_MSVDX_WDT_CONTROL, FE_WDT_ACTION0 */ +#define FE_MSVDX_WDT_CONTROL_FE_WDT_ACTION0_MASK (0x00000100) +#define FE_MSVDX_WDT_CONTROL_FE_WDT_ACTION0_LSBMASK (0x00000001) +#define FE_MSVDX_WDT_CONTROL_FE_WDT_ACTION0_SHIFT (8) +/* MSVDX_CORE, CR_FE_MSVDX_WDT_CONTROL, FE_WDT_CLEAR_SELECT */ +#define FE_MSVDX_WDT_CONTROL_FE_WDT_CLEAR_SELECT_MASK (0x00000030) +#define FE_MSVDX_WDT_CONTROL_FE_WDT_CLEAR_SELECT_LSBMASK (0x00000003) +#define FE_MSVDX_WDT_CONTROL_FE_WDT_CLEAR_SELECT_SHIFT (4) +/* MSVDX_CORE, CR_FE_MSVDX_WDT_CONTROL, FE_WDT_CLKDIV_SELECT */ +#define FE_MSVDX_WDT_CONTROL_FE_WDT_CLKDIV_SELECT_MASK (0x00000007) +#define FE_MSVDX_WDT_CONTROL_FE_WDT_CLKDIV_SELECT_LSBMASK (0x00000007) +#define FE_MSVDX_WDT_CONTROL_FE_WDT_CLKDIV_SELECT_SHIFT (0) + +#define FE_MSVDX_WDTIMER_OFFSET (0x0668) +/* MSVDX_CORE, CR_FE_MSVDX_WDTIMER, FE_WDT_COUNTER */ +#define FE_MSVDX_WDTIMER_FE_WDT_COUNTER_MASK (0x0000FFFF) +#define FE_MSVDX_WDTIMER_FE_WDT_COUNTER_LSBMASK (0x0000FFFF) +#define FE_MSVDX_WDTIMER_FE_WDT_COUNTER_SHIFT (0) + +#define FE_MSVDX_WDT_COMPAREMATCH_OFFSET (0x066c) +/* MSVDX_CORE, CR_FE_MSVDX_WDT_COMPAREMATCH, FE_WDT_CM1 */ +#define FE_MSVDX_WDT_COMPAREMATCH_FE_WDT_CM1_MASK (0xFFFF0000) +#define FE_MSVDX_WDT_COMPAREMATCH_FE_WDT_CM1_LSBMASK (0x0000FFFF) +#define FE_MSVDX_WDT_COMPAREMATCH_FE_WDT_CM1_SHIFT (16) +/* MSVDX_CORE, CR_FE_MSVDX_WDT_COMPAREMATCH, FE_WDT_CM0 */ +#define FE_MSVDX_WDT_COMPAREMATCH_FE_WDT_CM0_MASK (0x0000FFFF) +#define FE_MSVDX_WDT_COMPAREMATCH_FE_WDT_CM0_LSBMASK (0x0000FFFF) +#define FE_MSVDX_WDT_COMPAREMATCH_FE_WDT_CM0_SHIFT (0) + +#define BE_MSVDX_WDT_CONTROL_OFFSET (0x0670) +/* MSVDX_CORE, CR_BE_MSVDX_WDT_CONTROL, BE_WDT_CNT_CTRL */ +#define BE_MSVDX_WDT_CONTROL_BE_WDT_CNT_CTRL_MASK (0x001E0000) +#define BE_MSVDX_WDT_CONTROL_BE_WDT_CNT_CTRL_LSBMASK (0x0000000F) +#define BE_MSVDX_WDT_CONTROL_BE_WDT_CNT_CTRL_SHIFT (17) +/* MSVDX_CORE, CR_BE_MSVDX_WDT_CONTROL, BE_WDT_ENABLE */ +#define BE_MSVDX_WDT_CONTROL_BE_WDT_ENABLE_MASK (0x00010000) +#define BE_MSVDX_WDT_CONTROL_BE_WDT_ENABLE_LSBMASK (0x00000001) +#define BE_MSVDX_WDT_CONTROL_BE_WDT_ENABLE_SHIFT (16) +/* MSVDX_CORE, CR_BE_MSVDX_WDT_CONTROL, BE_WDT_ACTION0 */ +#define BE_MSVDX_WDT_CONTROL_BE_WDT_ACTION0_MASK (0x00000100) +#define BE_MSVDX_WDT_CONTROL_BE_WDT_ACTION0_LSBMASK (0x00000001) +#define BE_MSVDX_WDT_CONTROL_BE_WDT_ACTION0_SHIFT (8) +/* MSVDX_CORE, CR_BE_MSVDX_WDT_CONTROL, BE_WDT_CLEAR_SELECT */ +#define BE_MSVDX_WDT_CONTROL_BE_WDT_CLEAR_SELECT_MASK (0x000000F0) +#define BE_MSVDX_WDT_CONTROL_BE_WDT_CLEAR_SELECT_LSBMASK (0x0000000F) +#define BE_MSVDX_WDT_CONTROL_BE_WDT_CLEAR_SELECT_SHIFT (4) +/* MSVDX_CORE, CR_BE_MSVDX_WDT_CONTROL, BE_WDT_CLKDIV_SELECT */ +#define BE_MSVDX_WDT_CONTROL_BE_WDT_CLKDIV_SELECT_MASK (0x00000007) +#define BE_MSVDX_WDT_CONTROL_BE_WDT_CLKDIV_SELECT_LSBMASK (0x00000007) +#define BE_MSVDX_WDT_CONTROL_BE_WDT_CLKDIV_SELECT_SHIFT (0) + +#define BE_MSVDX_WDTIMER_OFFSET (0x0674) +/* MSVDX_CORE, CR_BE_MSVDX_WDTIMER, BE_WDT_COUNTER */ +#define BE_MSVDX_WDTIMER_BE_WDT_COUNTER_MASK (0x0000FFFF) +#define BE_MSVDX_WDTIMER_BE_WDT_COUNTER_LSBMASK (0x0000FFFF) +#define BE_MSVDX_WDTIMER_BE_WDT_COUNTER_SHIFT (0) + +#define BE_MSVDX_WDT_COMPAREMATCH_OFFSET (0x678) +/* MSVDX_CORE, CR_BE_MSVDX_WDT_COMPAREMATCH, BE_WDT_CM0 */ +#define BE_MSVDX_WDT_COMPAREMATCH_BE_WDT_CM0_MASK (0x0000FFFF) +#define BE_MSVDX_WDT_COMPAREMATCH_BE_WDT_CM0_LSBMASK (0x0000FFFF) +#define BE_MSVDX_WDT_COMPAREMATCH_BE_WDT_CM0_SHIFT (0) + +/*watch dog end*/ +/************************** MSVDX Core Registers end *************************/ + +/******************* VEC Registers: 0x0800 - 0x0FFF (2048B) ******************/ +#define VEC_SHIFTREG_CONTROL_OFFSET (0x0818) +#define VEC_SHIFTREG_CONTROL_SR_MASTER_SELECT_MASK (0x00000300) +#define VEC_SHIFTREG_CONTROL_SR_MASTER_SELECT_SHIFT (8) +/************************** VEC Registers end **************************/ + +/************************** RENDEC Registers **************************/ +#define MSVDX_RENDEC_CONTROL0_OFFSET (0x0868) +#define MSVDX_RENDEC_CONTROL0_RENDEC_INITIALISE_MASK (0x00000001) +#define MSVDX_RENDEC_CONTROL0_RENDEC_INITIALISE_SHIFT (0) + +#define MSVDX_RENDEC_CONTROL1_OFFSET (0x086C) +#define MSVDX_RENDEC_CONTROL1_RENDEC_DECODE_START_SIZE_MASK (0x000000FF) +#define MSVDX_RENDEC_CONTROL1_RENDEC_DECODE_START_SIZE_SHIFT (0) +#define MSVDX_RENDEC_CONTROL1_RENDEC_BURST_SIZE_W_MASK (0x000C0000) +#define MSVDX_RENDEC_CONTROL1_RENDEC_BURST_SIZE_W_SHIFT (18) +#define MSVDX_RENDEC_CONTROL1_RENDEC_BURST_SIZE_R_MASK (0x00030000) +#define MSVDX_RENDEC_CONTROL1_RENDEC_BURST_SIZE_R_SHIFT (16) +#define MSVDX_RENDEC_CONTROL1_RENDEC_EXTERNAL_MEMORY_MASK (0x01000000) +#define MSVDX_RENDEC_CONTROL1_RENDEC_EXTERNAL_MEMORY_SHIFT (24) +#define MSVDX_RENDEC_CONTROL1_RENDEC_DEC_DISABLE_MASK (0x08000000) +#define MSVDX_RENDEC_CONTROL1_RENDEC_DEC_DISABLE_SHIFT (27) + +#define MSVDX_RENDEC_BUFFER_SIZE_OFFSET (0x0870) +#define MSVDX_RENDEC_BUFFER_SIZE_RENDEC_BUFFER_SIZE0_MASK (0x0000FFFF) +#define MSVDX_RENDEC_BUFFER_SIZE_RENDEC_BUFFER_SIZE0_SHIFT (0) +#define MSVDX_RENDEC_BUFFER_SIZE_RENDEC_BUFFER_SIZE1_MASK (0xFFFF0000) +#define MSVDX_RENDEC_BUFFER_SIZE_RENDEC_BUFFER_SIZE1_SHIFT (16) + +#define MSVDX_RENDEC_BASE_ADDR0_OFFSET (0x0874) + +#define MSVDX_RENDEC_BASE_ADDR1_OFFSET (0x0878) + +#define MSVDX_RENDEC_READ_DATA_OFFSET (0x0898) + +#define MSVDX_RENDEC_CONTEXT0_OFFSET (0x0950) + +#define MSVDX_RENDEC_CONTEXT1_OFFSET (0x0954) + +#define MSVDX_RENDEC_CONTEXT2_OFFSET (0x0958) + +#define MSVDX_RENDEC_CONTEXT3_OFFSET (0x095C) + +#define MSVDX_RENDEC_CONTEXT4_OFFSET (0x0960) + +#define MSVDX_RENDEC_CONTEXT5_OFFSET (0x0964) +/*************************** RENDEC registers end ****************************/ + +/******************** CMD Register: 0x1000 - 0x1FFF (4kB) ********************/ +#define MSVDX_CMDS_END_SLICE_PICTURE_OFFSET (0x1404) +/****************************** CMD Register end *****************************/ + +/******************** VEC Local RAM: 0x2000 - 0x2FFF (4kB) *******************/ +/* vec local MEM save/restore */ +#define VEC_LOCAL_MEM_BYTE_SIZE (4 * 1024) +#define VEC_LOCAL_MEM_OFFSET 0x2000 + +#define MSVDX_EXT_FW_ERROR_STATE (0x2CC4) +/* Decode operations in progress or not complete */ +#define MSVDX_FW_STATUS_IN_PROGRESS 0x00000000 +/* there's no work underway on the hardware, idle, can be powered down */ +#define MSVDX_FW_STATUS_HW_IDLE 0x00000001 +/* Panic, waiting to be reloaded */ +#define MSVDX_FW_STATUS_HW_PANIC 0x00000003 + +/* + * This defines the MSVDX communication buffer + */ +#define MSVDX_COMMS_SIGNATURE_VALUE (0xA5A5A5A5) /*!< Signature value */ +/*!< Host buffer size (in 32-bit words) */ +#define NUM_WORDS_HOST_BUF (100) +/*!< MTX buffer size (in 32-bit words) */ +#define NUM_WORDS_MTX_BUF (100) + +#define MSVDX_COMMS_AREA_ADDR (0x02fe0) +#define MSVDX_COMMS_CORE_WTD (MSVDX_COMMS_AREA_ADDR - 0x08) +#define MSVDX_COMMS_ERROR_TRIG (MSVDX_COMMS_AREA_ADDR - 0x08) +#define MSVDX_COMMS_FIRMWARE_ID (MSVDX_COMMS_AREA_ADDR - 0x0C) +#define MSVDX_COMMS_OFFSET_FLAGS (MSVDX_COMMS_AREA_ADDR + 0x18) +#define MSVDX_COMMS_MSG_COUNTER (MSVDX_COMMS_AREA_ADDR - 0x04) +#define MSVDX_COMMS_FW_STATUS (MSVDX_COMMS_AREA_ADDR - 0x10) +#define MSVDX_COMMS_SIGNATURE (MSVDX_COMMS_AREA_ADDR + 0x00) +#define MSVDX_COMMS_TO_HOST_BUF_SIZE (MSVDX_COMMS_AREA_ADDR + 0x04) +#define MSVDX_COMMS_TO_HOST_RD_INDEX (MSVDX_COMMS_AREA_ADDR + 0x08) +#define MSVDX_COMMS_TO_HOST_WRT_INDEX (MSVDX_COMMS_AREA_ADDR + 0x0C) +#define MSVDX_COMMS_TO_MTX_BUF_SIZE (MSVDX_COMMS_AREA_ADDR + 0x10) +#define MSVDX_COMMS_TO_MTX_RD_INDEX (MSVDX_COMMS_AREA_ADDR + 0x14) +#define MSVDX_COMMS_TO_MTX_CB_RD_INDEX (MSVDX_COMMS_AREA_ADDR + 0x18) +#define MSVDX_COMMS_TO_MTX_WRT_INDEX (MSVDX_COMMS_AREA_ADDR + 0x1C) +#define MSVDX_COMMS_TO_HOST_BUF (MSVDX_COMMS_AREA_ADDR + 0x20) +#define MSVDX_COMMS_TO_MTX_BUF \ + (MSVDX_COMMS_TO_HOST_BUF + (NUM_WORDS_HOST_BUF << 2)) + +/* + * FW FLAGs: it shall be written by the host prior to starting the Firmware. + */ +/* Disable Firmware based Watch dog timers. */ +#define DSIABLE_FW_WDT 0x0008 + /* Abort Immediately on errors */ +#define ABORT_ON_ERRORS_IMMEDIATE 0x0010 + /* Aborts faulted slices as soon as possible. Allows non faulted slices + * to reach backend but faulted slice will not be allowed to start. */ +#define ABORT_FAULTED_SLICE_IMMEDIATE 0x0020 + /* Flush faulted slices - Debug option */ +#define FLUSH_FAULTED_SLICES 0x0080 + /* Don't interrupt host when to host buffer becomes full. + * Stall until space is freed up by host on it's own. */ +#define NOT_INTERRUPT_WHEN_HOST_IS_FULL 0x0200 + /* Contiguity warning msg will be send to host for stream with + * FW_ERROR_DETECTION_AND_RECOVERY flag set if non-contiguous + * macroblocks are detected. */ +#define NOT_ENABLE_ON_HOST_CONCEALMENT 0x0400 + /* Return VDEB Signature Value in Completion message. + * This requires a VDEB data flush every slice for constant results.*/ +#define RETURN_VDEB_DATA_IN_COMPLETION 0x0800 + /* Disable Auto Clock Gating. */ +#define DSIABLE_Auto_CLOCK_GATING 0x1000 + /* Disable Idle GPIO signal. */ +#define DSIABLE_IDLE_GPIO_SIG 0x2000 + /* Enable Setup, FE and BE Time stamps in completion message. + * Used by IMG only for firmware profiling. */ +#define ENABLE_TIMESTAMPS_IN_COMPLETE_MSG 0x4000 + /* Disable off-host 2nd pass Deblocking in Firmware. */ +#define DSIABLE_OFFHOST_SECOND_DEBLOCK 0x20000 + /* Sum address signature to data signature + * when returning VDEB signature values. */ +#define SUM_ADD_SIG_TO_DATA_SIGNATURE 0x80000 + +/* +#define MSVDX_COMMS_AREA_END \ + (MSVDX_COMMS_TO_MTX_BUF + (NUM_WORDS_HOST_BUF << 2)) +*/ +#define MSVDX_COMMS_AREA_END 0x03000 + +#if (MSVDX_COMMS_AREA_END != 0x03000) +#error +#endif +/***************************** VEC Local RAM end *****************************/ + +#endif
Hi
+static struct drm_ioctl_desc ipvr_gem_ioctls[] = {
DRM_IOCTL_DEF_DRV(IPVR_CONTEXT_CREATE,ipvr_context_create_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_CONTEXT_DESTROY,ipvr_context_destroy_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_MISC,ipvr_misc_ioctl, DRM_AUTH),DRM_IOCTL_DEF_DRV(IPVR_GEM_EXECBUFFER,ipvr_gem_execbuffer, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_BUSY,ipvr_gem_busy_ioctl, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_CREATE,ipvr_gem_create_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_MMAP,ipvr_gem_mmap_ioctl, DRM_UNLOCKED),
Why do you need this ioctl? mmap() should work perfectly fine. I don't see why you require people to use a ipvr specific ioctl to map buffers.
DRM_IOCTL_DEF_DRV(IPVR_SYNC_CPU,ipvr_sync_cpu_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_WAIT,ipvr_gem_wait_ioctl, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_USERPTR,ipvr_gem_userptr_ioctl, DRM_UNLOCKED),+};
+static void ipvr_gem_init(struct drm_device *dev) +{
struct drm_ipvr_private *dev_priv = dev->dev_private;dev_priv->ipvr_bo_slab = kmem_cache_create("ipvr_gem_object",sizeof(union drm_ipvr_gem_objects), 0,SLAB_HWCACHE_ALIGN, NULL);INIT_LIST_HEAD(&dev_priv->ipvr_mm.unbound_list);INIT_LIST_HEAD(&dev_priv->ipvr_mm.bound_list);spin_lock_init(&dev_priv->ipvr_mm.object_stat_lock);dev_priv->ipvr_mm.interruptible = true;+}
+static void ipvr_gem_setup_mmu(struct drm_device *dev,
unsigned long linear_start,unsigned long linear_end,unsigned long tiling_start,unsigned long tiling_end)+{
/* Let GEM Manage all of the aperture.** However, leave one page at the end still bound to the scratch page.* There are a number of places where hardware apparently prefetches* past the end of the object, and we've seen multiple hangs with the* GPU head pointer stuck in a batchbuffer bound at last page of the* aperture. One page should be enough to keep any prefetching inside* of the aperture.*/struct drm_ipvr_private *dev_priv = dev->dev_private;struct ipvr_address_space *addr_space = &dev_priv->addr_space;/* todo: add sanity check */addr_space->dev = dev_priv->dev;INIT_LIST_HEAD(&addr_space->active_list);INIT_LIST_HEAD(&addr_space->inactive_list);/* Subtract the guard page ... */drm_mm_init(&addr_space->linear_mm, linear_start,linear_end - linear_start - PAGE_SIZE);dev_priv->addr_space.linear_start = linear_start;dev_priv->addr_space.linear_total = linear_end - linear_start;drm_mm_init(&addr_space->tiling_mm, tiling_start,tiling_end - tiling_start - PAGE_SIZE);dev_priv->addr_space.tiling_start = tiling_start;dev_priv->addr_space.tiling_total = tiling_end - tiling_start;+}
+static void ipvr_do_takedown(struct drm_device *dev) +{
/* todo: need check if need clean up mm here */ipvr_ved_uninit(dev);+}
+static int32_t ipvr_drm_unload(struct drm_device *dev) +{
struct drm_ipvr_private *dev_priv = dev->dev_private;BUG_ON(!dev->platformdev);BUG_ON(atomic_read(&dev_priv->ved_power_usage));IPVR_DEBUG_ENTRY("entered.");if (dev_priv) {WARN_ON(pm_runtime_get_sync(&dev->platformdev->dev) < 0);if (dev_priv->ipvr_bo_slab)kmem_cache_destroy(dev_priv->ipvr_bo_slab);ipvr_fence_driver_fini(dev_priv);ipvr_do_takedown(dev);WARN_ON(pm_runtime_put_sync_suspend(&dev->platformdev->dev) < 0);if (dev_priv->validate_ctx.buffers)vfree(dev_priv->validate_ctx.buffers);if (dev_priv->pf_pd) {ipvr_mmu_free_pagedir(dev_priv->pf_pd);dev_priv->pf_pd = NULL;}if (dev_priv->mmu) {ipvr_mmu_driver_takedown(dev_priv->mmu);dev_priv->mmu = NULL;}if (dev_priv->ved_reg_base) {iounmap(dev_priv->ved_reg_base - dev_priv->ved_reg_offset);dev_priv->ved_reg_base = NULL;dev_priv->ved_reg_offset = 0;}list_del(&dev_priv->default_ctx.head);idr_remove(&dev_priv->ipvr_ctx_idr, dev_priv->default_ctx.ctx_id);kfree(dev_priv);}pm_runtime_disable(&dev->platformdev->dev);return 0;+}
+static int32_t ipvr_drm_load(struct drm_device *dev, unsigned long flags) +{
struct drm_ipvr_private *dev_priv;int32_t ctx_id, ret = 0;struct platform_device *platdev;struct resource *res_mmio, *res_reg;void __iomem* mmio_addr;dev_priv = kzalloc(sizeof(*dev_priv), GFP_KERNEL);if (dev_priv == NULL)return -ENOMEM;dev->dev_private = dev_priv;dev_priv->dev = dev;BUG_ON(!dev->platformdev);platdev = dev->platformdev;mutex_init(&dev_priv->cmdbuf_mutex);INIT_LIST_HEAD(&dev_priv->validate_ctx.validate_list);dev_priv->pci_root = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));if (!dev_priv->pci_root) {kfree(dev_priv);return -ENODEV;}dev->irq = platform_get_irq(platdev, 0);if (dev->irq < 0) {ret = -ENODEV;goto out_err;}res_mmio = platform_get_resource(platdev, IORESOURCE_MEM, 0);res_reg = platform_get_resource(platdev, IORESOURCE_REG, 0);if (!res_mmio || !res_reg) {ret = -ENXIO;goto out_err;}mmio_addr = ioremap_wc(res_mmio->start,res_mmio->end - res_mmio->start);if (IS_ERR(mmio_addr)) {ret = -EACCES;goto out_err;}dev_priv->ved_reg_base = mmio_addr + res_reg->start;dev_priv->ved_reg_offset = res_reg->start;IPVR_DEBUG_VED("ved_reg_base is %p, range is 0x%llx - 0x%llx.\n",dev_priv->ved_reg_base,res_reg->start, res_reg->end);platform_set_drvdata(dev->platformdev, dev);pm_runtime_enable(&dev->platformdev->dev);if (pm_runtime_get_sync(&dev->platformdev->dev) < 0) {ret = -EBUSY;goto out_err;}IPVR_DEBUG_INIT("MSVDX_CORE_REV_OFFSET by readl is 0x%x.\n",readl(dev_priv->ved_reg_base + 0x640));IPVR_DEBUG_INIT("MSVDX_CORE_REV_OFFSET by VED_REG_READ32 is 0x%x.\n",VED_REG_READ32(MSVDX_CORE_REV_OFFSET));/* mmu init */dev_priv->mmu = ipvr_mmu_driver_init((void *)0,drm_ipvr_trap_pagefaults,0, dev_priv);if (!dev_priv->mmu) {ret = -EBUSY;goto out_err;}dev_priv->pf_pd = ipvr_mmu_alloc_pd(dev_priv->mmu, 1, 0);if (!dev_priv->pf_pd) {ret = -ENOMEM;goto out_err;}ipvr_mmu_set_pd_context(ipvr_mmu_get_default_pd(dev_priv->mmu), 0);ipvr_mmu_set_pd_context(dev_priv->pf_pd, 1);/** Initialize sequence numbers for the different command* submission mechanisms.*/dev_priv->last_seq = 1;ipvr_gem_init(dev);ipvr_gem_setup_mmu(dev,IPVR_MEM_MMU_LINEAR_START,IPVR_MEM_MMU_LINEAR_END,IPVR_MEM_MMU_TILING_START,IPVR_MEM_MMU_TILING_END);ipvr_ved_init(dev);WARN_ON(pm_runtime_put_sync_suspend(&dev->platformdev->dev) < 0);dev_priv->ved_private->ved_needs_reset = 1;mutex_init(&dev_priv->ved_pm_mutex);atomic_set(&dev_priv->ved_power_usage, 0);ipvr_fence_driver_init(dev_priv);dev_priv->validate_ctx.buffers =vmalloc(IPVR_NUM_VALIDATE_BUFFERS *sizeof(struct ipvr_validate_buffer));if (!dev_priv->validate_ctx.buffers) {ret = -ENOMEM;goto out_err;}/* ipvr context initialization */INIT_LIST_HEAD(&dev_priv->ipvr_ctx_list);spin_lock_init(&dev_priv->ipvr_ctx_lock);idr_init(&dev_priv->ipvr_ctx_idr);/* default ipvr context is used for scaling, rotation case */ctx_id = idr_alloc(&dev_priv->ipvr_ctx_idr, &dev_priv->default_ctx,IPVR_MIN_CONTEXT_ID, IPVR_MAX_CONTEXT_ID,GFP_KERNEL);if (ctx_id < 0) {return -ENOMEM;goto out_err;}dev_priv->default_ctx.ctx_id = ctx_id;INIT_LIST_HEAD(&dev_priv->default_ctx.head);dev_priv->default_ctx.ctx_type = 0;dev_priv->default_ctx.ipvr_fpriv = NULL;/* don't need protect with spinlock during module load stage */list_add(&dev_priv->default_ctx.head, &dev_priv->ipvr_ctx_list);dev_priv->default_ctx.tiling_scheme = 0;dev_priv->default_ctx.tiling_stride = 0;return 0;+out_err:
ipvr_drm_unload(dev);return ret;+}
+/*
- The .open() method is called every time the device is opened by an
- application. Drivers can allocate per-file private data in this method and
- store them in the struct drm_file::driver_priv field. Note that the .open()
- method is called before .firstopen().
- */
+static int32_t +ipvr_drm_open(struct drm_device *dev, struct drm_file *file_priv) +{
struct drm_ipvr_file_private *ipvr_fp;IPVR_DEBUG_ENTRY("enter\n");ipvr_fp = kzalloc(sizeof(*ipvr_fp), GFP_KERNEL);if (unlikely(ipvr_fp == NULL))return -ENOMEM;file_priv->driver_priv = ipvr_fp;return 0;+}
+/*
- The close operation is split into .preclose() and .postclose() methods.
- Drivers must stop and cleanup all per-file operations in the .preclose()
- method. For instance pending vertical blanking and page flip events must be
- cancelled. No per-file operation is allowed on the file handle after
- returning from the .preclose() method.
- */
+static void +ipvr_drm_preclose(struct drm_device *dev, struct drm_file *file_priv) +{
/* if user didn't destory ctx explicitly, remove ctx here */struct drm_ipvr_private *dev_priv;struct drm_ipvr_file_private *ipvr_fpriv;struct ved_private *ved_priv;struct ipvr_context *ipvr_ctx = NULL;unsigned long irq_flags;IPVR_DEBUG_ENTRY("enter\n");dev_priv = dev->dev_private;ipvr_fpriv = file_priv->driver_priv;ved_priv = dev_priv->ved_private;if (ipvr_fpriv->ctx_id == IPVR_CONTEXT_INVALID_ID)return;ipvr_ctx = (struct ipvr_context *)idr_find(&dev_priv->ipvr_ctx_idr, ipvr_fpriv->ctx_id);if (!ipvr_ctx || (ipvr_ctx->ipvr_fpriv != ipvr_fpriv)) {IPVR_DEBUG_GENERAL("ctx for id %d has already destroyed\n",ipvr_fpriv->ctx_id);return;}IPVR_DEBUG_PM("Video:remove context profile %d, entrypoint %d\n",(ipvr_ctx->ctx_type >> 8) & 0xff,(ipvr_ctx->ctx_type & 0xff));mutex_lock(&ved_priv->ved_mutex);if (ved_priv->ipvr_ctx == ipvr_ctx )ved_priv->ipvr_ctx = NULL;mutex_unlock(&ved_priv->ved_mutex);spin_lock_irqsave(&dev_priv->ipvr_ctx_lock, irq_flags);list_del(&ipvr_ctx->head);ipvr_fpriv->ctx_id = IPVR_CONTEXT_INVALID_ID;spin_unlock_irqrestore(&dev_priv->ipvr_ctx_lock, irq_flags);idr_remove(&dev_priv->ipvr_ctx_idr, ipvr_ctx->ctx_id);kfree(ipvr_ctx );+}
+/*
- Finally the .postclose() method is called as the last step of the close
- operation, right before calling the .lastclose() method if no other open
- file handle exists for the device. Drivers that have allocated per-file
- private data in the .open() method should free it here.
- */
+static void +ipvr_drm_postclose(struct drm_device *dev, struct drm_file *file_priv) +{
struct drm_ipvr_file_private *ipvr_fpriv = file_priv->driver_priv;IPVR_DEBUG_ENTRY("enter\n");kfree(ipvr_fpriv);+}
postclose() is deprecated. You can safely move this into preclose().
+static irqreturn_t ipvr_irq_handler(int32_t irq, void *arg) +{
struct drm_device *dev = (struct drm_device *) arg;WARN_ON(ved_irq_handler(dev));return IRQ_HANDLED;+}
+static const struct file_operations ipvr_fops = {
.owner = THIS_MODULE,.open = drm_open,.release = drm_release,.unlocked_ioctl = drm_ioctl,+#ifdef CONFIG_COMPAT
.compat_ioctl = drm_ioctl,+#endif
/* no need to define mmap. User space maps bo with DRM_IPVR_GEM_MMAP */+};
+static int32_t ipvr_drm_freeze(struct drm_device *dev) +{
int32_t ret;int32_t power_usage;struct drm_ipvr_private *dev_priv = dev->dev_private;IPVR_DEBUG_ENTRY("enter\n");power_usage = atomic_read(&dev_priv->ved_power_usage);BUG_ON(power_usage < 0);if (power_usage > 0) {IPVR_DEBUG_PM("VED power usage is %d, skip freezing\n", power_usage);return 0;}ret = ved_check_idle(dev);if (ret) {IPVR_DEBUG_PM("VED check idle fail: %d, skip freezing\n", ret);return 0;}if (dev->irq_enabled) {ret = drm_irq_uninstall(dev);if (unlikely(ret)) {IPVR_ERROR("Error uninstalling IRQ handler: %d\n", ret);return -EFAULT;}IPVR_DEBUG_PM("Successfully uninstalled IRQ\n");}elseIPVR_DEBUG_PM("irq_enabled is %d\n", dev->irq_enabled);if (is_ved_on(dev)) {if (!ved_power_off(dev)) {IPVR_ERROR("Failed to power off VED\n");return -EFAULT;}IPVR_DEBUG_PM("Successfully powered off\n");} else {IPVR_DEBUG_PM("Skiped power-off since already powered off\n");}return 0;+}
+static int32_t ipvr_drm_thaw(struct drm_device *dev) +{
int ret;IPVR_DEBUG_ENTRY("enter\n");if (!is_ved_on(dev)) {if (!ved_power_on(dev)) {IPVR_ERROR("Failed to power on VED\n");return -EFAULT;}IPVR_DEBUG_PM("Successfully powered on\n");} else {IPVR_DEBUG_PM("Skiped power-on since already powered on\n");}if (!dev->irq_enabled) {ret = drm_irq_install(dev, dev->irq);if (ret) {IPVR_ERROR("Error installing IRQ handler: %d\n", ret);return -EFAULT;}IPVR_DEBUG_PM("Successfully installed IRQ\n");}elseIPVR_DEBUG_PM("irq_enabled is %d\n", dev->irq_enabled);return 0;+}
+static int32_t ipvr_pm_suspend(struct device *dev) +{
struct platform_device *platformdev = to_platform_device(dev);struct drm_device *drm_dev = platform_get_drvdata(platformdev);IPVR_DEBUG_PM("PM suspend called\n");BUG_ON(!drm_dev);return ipvr_drm_freeze(drm_dev);+} +static int32_t ipvr_pm_resume(struct device *dev) +{
struct platform_device *platformdev = to_platform_device(dev);struct drm_device *drm_dev = platform_get_drvdata(platformdev);IPVR_DEBUG_PM("PM resume called\n");BUG_ON(!drm_dev);return ipvr_drm_thaw(drm_dev);+}
+/*
- dump GEM API is mainly for dumb buffers suitable for scanout,
- it is not needed for ipvr driver.
- gem_vm_ops is used for mmap case, also not needed for ipvr
- todo: prime support can be enabled later
- */
+static struct drm_driver ipvr_drm_driver = {
.driver_features = DRIVER_HAVE_IRQ | DRIVER_GEM,.load = ipvr_drm_load,.unload = ipvr_drm_unload,.open = ipvr_drm_open,.preclose = ipvr_drm_preclose,.postclose = ipvr_drm_postclose,.irq_handler = ipvr_irq_handler,.set_busid = drm_platform_set_busid,
This should be NULL for new drivers.
.gem_free_object = ipvr_gem_free_object,+#ifdef CONFIG_DEBUG_FS
.debugfs_init = ipvr_debugfs_init,.debugfs_cleanup = ipvr_debugfs_cleanup,+#endif
.ioctls = ipvr_gem_ioctls,.num_ioctls = ARRAY_SIZE(ipvr_gem_ioctls),.fops = &ipvr_fops,.name = IPVR_DRIVER_NAME,.desc = IPVR_DRIVER_DESC,.date = IPVR_DRIVER_DATE,.major = IPVR_DRIVER_MAJOR,.minor = IPVR_DRIVER_MINOR,.patchlevel = IPVR_DRIVER_PATCHLEVEL,+};
+static int32_t ipvr_plat_probe(struct platform_device *device) +{
return drm_platform_init(&ipvr_drm_driver, device);
drm_platform_init() should be used. Please use drm_dev_alloc(), drm_dev_register() directly. udl and tegra should serve as example.
+}
+static int32_t ipvr_plat_remove(struct platform_device *device) +{
drm_put_dev(platform_get_drvdata(device));
Again, please use drm_dev_unregister() + drm_dev_unref() directly.
Thanks David
return 0;+}
+static struct dev_pm_ops ipvr_pm_ops = {
.suspend = ipvr_pm_suspend,.resume = ipvr_pm_resume,.freeze = ipvr_pm_suspend,.thaw = ipvr_pm_resume,.poweroff = ipvr_pm_suspend,.restore = ipvr_pm_resume,+#ifdef CONFIG_PM_RUNTIME
.runtime_suspend = ipvr_pm_suspend,.runtime_resume = ipvr_pm_resume,+#endif +};
+static struct platform_driver ipvr_plat_driver = {
.driver = {.name = "ipvr-ved",.owner = THIS_MODULE,+#ifdef CONFIG_PM
.pm = &ipvr_pm_ops,+#endif
},.probe = ipvr_plat_probe,.remove = ipvr_plat_remove,+};
+module_platform_driver(ipvr_plat_driver); +MODULE_LICENSE("GPL");
Hi Herrmann
-----Original Message----- From: David Herrmann [mailto:dh.herrmann@gmail.com] Sent: Monday, October 13, 2014 10:27 PM To: Cheng, Yao Cc: Intel Graphics Development; Jiang, Fei; dri-devel@lists.freedesktop.org; Vetter, Daniel Subject: Re: [RFC PATCH 2/3] drm/ipvr: drm driver for vxd392
Hi
+static struct drm_ioctl_desc ipvr_gem_ioctls[] = {
DRM_IOCTL_DEF_DRV(IPVR_CONTEXT_CREATE,ipvr_context_create_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_CONTEXT_DESTROY,ipvr_context_destroy_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_MISC,ipvr_misc_ioctl, DRM_AUTH),DRM_IOCTL_DEF_DRV(IPVR_GEM_EXECBUFFER,ipvr_gem_execbuffer, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_BUSY,ipvr_gem_busy_ioctl, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_CREATE,ipvr_gem_create_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_MMAP,ipvr_gem_mmap_ioctl, DRM_UNLOCKED),Why do you need this ioctl? mmap() should work perfectly fine. I don't see why you require people to use a ipvr specific ioctl to map buffers.
Many thanks to your comments, in our existing libdrm helper and userspace drivers, mmap_ioctl was the interface of mapping objects. We continued using the ioctl way for compatibility. Is it mandatory to implement mmap() to replace mmap_ioctl for GEM drivers?
DRM_IOCTL_DEF_DRV(IPVR_SYNC_CPU,ipvr_sync_cpu_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_WAIT,ipvr_gem_wait_ioctl, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_USERPTR,ipvr_gem_userptr_ioctl, DRM_UNLOCKED), };+static void ipvr_gem_init(struct drm_device *dev) {
struct drm_ipvr_private *dev_priv = dev->dev_private;dev_priv->ipvr_bo_slab = kmem_cache_create("ipvr_gem_object",sizeof(union drm_ipvr_gem_objects), 0,SLAB_HWCACHE_ALIGN, NULL);INIT_LIST_HEAD(&dev_priv->ipvr_mm.unbound_list);INIT_LIST_HEAD(&dev_priv->ipvr_mm.bound_list);spin_lock_init(&dev_priv->ipvr_mm.object_stat_lock);dev_priv->ipvr_mm.interruptible = true; }+static void ipvr_gem_setup_mmu(struct drm_device *dev,
unsigned long linear_start,unsigned long linear_end,unsigned long tiling_start,unsigned long tiling_end) {/* Let GEM Manage all of the aperture.** However, leave one page at the end still bound to the scratch page.* There are a number of places where hardware apparentlyprefetches
* past the end of the object, and we've seen multiple hangs with the* GPU head pointer stuck in a batchbuffer bound at last page of the* aperture. One page should be enough to keep any prefetchinginside
* of the aperture.*/struct drm_ipvr_private *dev_priv = dev->dev_private;struct ipvr_address_space *addr_space = &dev_priv->addr_space;/* todo: add sanity check */addr_space->dev = dev_priv->dev;INIT_LIST_HEAD(&addr_space->active_list);INIT_LIST_HEAD(&addr_space->inactive_list);/* Subtract the guard page ... */drm_mm_init(&addr_space->linear_mm, linear_start,linear_end - linear_start - PAGE_SIZE);dev_priv->addr_space.linear_start = linear_start;dev_priv->addr_space.linear_total = linear_end - linear_start;drm_mm_init(&addr_space->tiling_mm, tiling_start,tiling_end - tiling_start - PAGE_SIZE);dev_priv->addr_space.tiling_start = tiling_start;dev_priv->addr_space.tiling_total = tiling_end - tiling_start;+}
+static void ipvr_do_takedown(struct drm_device *dev) {
/* todo: need check if need clean up mm here */ipvr_ved_uninit(dev);+}
+static int32_t ipvr_drm_unload(struct drm_device *dev) {
struct drm_ipvr_private *dev_priv = dev->dev_private;BUG_ON(!dev->platformdev);BUG_ON(atomic_read(&dev_priv->ved_power_usage));IPVR_DEBUG_ENTRY("entered.");if (dev_priv) {WARN_ON(pm_runtime_get_sync(&dev->platformdev->dev) <- 0);
if (dev_priv->ipvr_bo_slab)kmem_cache_destroy(dev_priv->ipvr_bo_slab);ipvr_fence_driver_fini(dev_priv);ipvr_do_takedown(dev);- WARN_ON(pm_runtime_put_sync_suspend(&dev->platformdev->dev)
< 0);
if (dev_priv->validate_ctx.buffers)vfree(dev_priv->validate_ctx.buffers);if (dev_priv->pf_pd) {ipvr_mmu_free_pagedir(dev_priv->pf_pd);dev_priv->pf_pd = NULL;}if (dev_priv->mmu) {ipvr_mmu_driver_takedown(dev_priv->mmu);dev_priv->mmu = NULL;}if (dev_priv->ved_reg_base) {iounmap(dev_priv->ved_reg_base - dev_priv-ved_reg_offset);
dev_priv->ved_reg_base = NULL;dev_priv->ved_reg_offset = 0;}list_del(&dev_priv->default_ctx.head);idr_remove(&dev_priv->ipvr_ctx_idr, dev_priv-default_ctx.ctx_id);
kfree(dev_priv);}pm_runtime_disable(&dev->platformdev->dev);return 0;+}
+static int32_t ipvr_drm_load(struct drm_device *dev, unsigned long +flags) {
struct drm_ipvr_private *dev_priv;int32_t ctx_id, ret = 0;struct platform_device *platdev;struct resource *res_mmio, *res_reg;void __iomem* mmio_addr;dev_priv = kzalloc(sizeof(*dev_priv), GFP_KERNEL);if (dev_priv == NULL)return -ENOMEM;dev->dev_private = dev_priv;dev_priv->dev = dev;BUG_ON(!dev->platformdev);platdev = dev->platformdev;mutex_init(&dev_priv->cmdbuf_mutex);INIT_LIST_HEAD(&dev_priv->validate_ctx.validate_list);dev_priv->pci_root = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));if (!dev_priv->pci_root) {kfree(dev_priv);return -ENODEV;}dev->irq = platform_get_irq(platdev, 0);if (dev->irq < 0) {ret = -ENODEV;goto out_err;}res_mmio = platform_get_resource(platdev, IORESOURCE_MEM, 0);res_reg = platform_get_resource(platdev, IORESOURCE_REG, 0);if (!res_mmio || !res_reg) {ret = -ENXIO;goto out_err;}mmio_addr = ioremap_wc(res_mmio->start,res_mmio->end - res_mmio->start);if (IS_ERR(mmio_addr)) {ret = -EACCES;goto out_err;}dev_priv->ved_reg_base = mmio_addr + res_reg->start;dev_priv->ved_reg_offset = res_reg->start;IPVR_DEBUG_VED("ved_reg_base is %p, range is 0x%llx - 0x%llx.\n",dev_priv->ved_reg_base,res_reg->start, res_reg->end);platform_set_drvdata(dev->platformdev, dev);pm_runtime_enable(&dev->platformdev->dev);if (pm_runtime_get_sync(&dev->platformdev->dev) < 0) {ret = -EBUSY;goto out_err;}IPVR_DEBUG_INIT("MSVDX_CORE_REV_OFFSET by readl is 0x%x.\n",readl(dev_priv->ved_reg_base + 0x640));IPVR_DEBUG_INIT("MSVDX_CORE_REV_OFFSET byVED_REG_READ32 is 0x%x.\n",
VED_REG_READ32(MSVDX_CORE_REV_OFFSET));/* mmu init */dev_priv->mmu = ipvr_mmu_driver_init((void *)0,drm_ipvr_trap_pagefaults,0, dev_priv);if (!dev_priv->mmu) {ret = -EBUSY;goto out_err;}dev_priv->pf_pd = ipvr_mmu_alloc_pd(dev_priv->mmu, 1, 0);if (!dev_priv->pf_pd) {ret = -ENOMEM;goto out_err;}ipvr_mmu_set_pd_context(ipvr_mmu_get_default_pd(dev_priv-mmu), 0);
ipvr_mmu_set_pd_context(dev_priv->pf_pd, 1);/** Initialize sequence numbers for the different command* submission mechanisms.*/dev_priv->last_seq = 1;ipvr_gem_init(dev);ipvr_gem_setup_mmu(dev,IPVR_MEM_MMU_LINEAR_START,IPVR_MEM_MMU_LINEAR_END,IPVR_MEM_MMU_TILING_START,IPVR_MEM_MMU_TILING_END);ipvr_ved_init(dev);WARN_ON(pm_runtime_put_sync_suspend(&dev->platformdev-dev) <
- 0);
dev_priv->ved_private->ved_needs_reset = 1;mutex_init(&dev_priv->ved_pm_mutex);atomic_set(&dev_priv->ved_power_usage, 0);ipvr_fence_driver_init(dev_priv);dev_priv->validate_ctx.buffers =vmalloc(IPVR_NUM_VALIDATE_BUFFERS *sizeof(struct ipvr_validate_buffer));if (!dev_priv->validate_ctx.buffers) {ret = -ENOMEM;goto out_err;}/* ipvr context initialization */INIT_LIST_HEAD(&dev_priv->ipvr_ctx_list);spin_lock_init(&dev_priv->ipvr_ctx_lock);idr_init(&dev_priv->ipvr_ctx_idr);/* default ipvr context is used for scaling, rotation case */ctx_id = idr_alloc(&dev_priv->ipvr_ctx_idr, &dev_priv->default_ctx,IPVR_MIN_CONTEXT_ID, IPVR_MAX_CONTEXT_ID,GFP_KERNEL);if (ctx_id < 0) {return -ENOMEM;goto out_err;}dev_priv->default_ctx.ctx_id = ctx_id;INIT_LIST_HEAD(&dev_priv->default_ctx.head);dev_priv->default_ctx.ctx_type = 0;dev_priv->default_ctx.ipvr_fpriv = NULL;/* don't need protect with spinlock during module load stage */list_add(&dev_priv->default_ctx.head, &dev_priv->ipvr_ctx_list);dev_priv->default_ctx.tiling_scheme = 0;dev_priv->default_ctx.tiling_stride = 0;return 0;+out_err:
ipvr_drm_unload(dev);return ret;+}
+/*
- The .open() method is called every time the device is opened by an
- application. Drivers can allocate per-file private data in this
+method and
- store them in the struct drm_file::driver_priv field. Note that
+the .open()
- method is called before .firstopen().
- */
+static int32_t +ipvr_drm_open(struct drm_device *dev, struct drm_file *file_priv) {
struct drm_ipvr_file_private *ipvr_fp;IPVR_DEBUG_ENTRY("enter\n");ipvr_fp = kzalloc(sizeof(*ipvr_fp), GFP_KERNEL);if (unlikely(ipvr_fp == NULL))return -ENOMEM;file_priv->driver_priv = ipvr_fp;return 0;+}
+/*
- The close operation is split into .preclose() and .postclose() methods.
- Drivers must stop and cleanup all per-file operations in the
+.preclose()
- method. For instance pending vertical blanking and page flip
+events must be
- cancelled. No per-file operation is allowed on the file handle
+after
- returning from the .preclose() method.
- */
+static void +ipvr_drm_preclose(struct drm_device *dev, struct drm_file *file_priv) +{
/* if user didn't destory ctx explicitly, remove ctx here */struct drm_ipvr_private *dev_priv;struct drm_ipvr_file_private *ipvr_fpriv;struct ved_private *ved_priv;struct ipvr_context *ipvr_ctx = NULL;unsigned long irq_flags;IPVR_DEBUG_ENTRY("enter\n");dev_priv = dev->dev_private;ipvr_fpriv = file_priv->driver_priv;ved_priv = dev_priv->ved_private;if (ipvr_fpriv->ctx_id == IPVR_CONTEXT_INVALID_ID)return;ipvr_ctx = (struct ipvr_context *)idr_find(&dev_priv->ipvr_ctx_idr, ipvr_fpriv->ctx_id);if (!ipvr_ctx || (ipvr_ctx->ipvr_fpriv != ipvr_fpriv)) {IPVR_DEBUG_GENERAL("ctx for id %d has already destroyed\n",ipvr_fpriv->ctx_id);return;}IPVR_DEBUG_PM("Video:remove context profile %d,entrypoint %d\n",
(ipvr_ctx->ctx_type >> 8) & 0xff,(ipvr_ctx->ctx_type & 0xff));mutex_lock(&ved_priv->ved_mutex);if (ved_priv->ipvr_ctx == ipvr_ctx )ved_priv->ipvr_ctx = NULL;mutex_unlock(&ved_priv->ved_mutex);spin_lock_irqsave(&dev_priv->ipvr_ctx_lock, irq_flags);list_del(&ipvr_ctx->head);ipvr_fpriv->ctx_id = IPVR_CONTEXT_INVALID_ID;spin_unlock_irqrestore(&dev_priv->ipvr_ctx_lock, irq_flags);idr_remove(&dev_priv->ipvr_ctx_idr, ipvr_ctx->ctx_id);kfree(ipvr_ctx );+}
+/*
- Finally the .postclose() method is called as the last step of the
+close
- operation, right before calling the .lastclose() method if no
+other open
- file handle exists for the device. Drivers that have allocated
+per-file
- private data in the .open() method should free it here.
- */
+static void +ipvr_drm_postclose(struct drm_device *dev, struct drm_file +*file_priv) {
struct drm_ipvr_file_private *ipvr_fpriv = file_priv->driver_priv;IPVR_DEBUG_ENTRY("enter\n");kfree(ipvr_fpriv);+}
postclose() is deprecated. You can safely move this into preclose().
Thx, will update it in next version.
+static irqreturn_t ipvr_irq_handler(int32_t irq, void *arg) {
struct drm_device *dev = (struct drm_device *) arg;WARN_ON(ved_irq_handler(dev));return IRQ_HANDLED;+}
+static const struct file_operations ipvr_fops = {
.owner = THIS_MODULE,.open = drm_open,.release = drm_release,.unlocked_ioctl = drm_ioctl,+#ifdef CONFIG_COMPAT
.compat_ioctl = drm_ioctl,+#endif
/* no need to define mmap. User space maps bo with+DRM_IPVR_GEM_MMAP */ };
+static int32_t ipvr_drm_freeze(struct drm_device *dev) {
int32_t ret;int32_t power_usage;struct drm_ipvr_private *dev_priv = dev->dev_private;IPVR_DEBUG_ENTRY("enter\n");power_usage = atomic_read(&dev_priv->ved_power_usage);BUG_ON(power_usage < 0);if (power_usage > 0) {IPVR_DEBUG_PM("VED power usage is %d, skip freezing\n",power_usage);
return 0;}ret = ved_check_idle(dev);if (ret) {IPVR_DEBUG_PM("VED check idle fail: %d, skip freezing\n", ret);return 0;}if (dev->irq_enabled) {ret = drm_irq_uninstall(dev);if (unlikely(ret)) {IPVR_ERROR("Error uninstalling IRQ handler: %d\n", ret);return -EFAULT;}IPVR_DEBUG_PM("Successfully uninstalled IRQ\n");}elseIPVR_DEBUG_PM("irq_enabled is %d\n",- dev->irq_enabled);
if (is_ved_on(dev)) {if (!ved_power_off(dev)) {IPVR_ERROR("Failed to power off VED\n");return -EFAULT;}IPVR_DEBUG_PM("Successfully powered off\n");} else {IPVR_DEBUG_PM("Skiped power-off since already poweredoff\n");
}return 0;+}
+static int32_t ipvr_drm_thaw(struct drm_device *dev) {
int ret;IPVR_DEBUG_ENTRY("enter\n");if (!is_ved_on(dev)) {if (!ved_power_on(dev)) {IPVR_ERROR("Failed to power on VED\n");return -EFAULT;}IPVR_DEBUG_PM("Successfully powered on\n");} else {IPVR_DEBUG_PM("Skiped power-on since already poweredon\n");
}if (!dev->irq_enabled) {ret = drm_irq_install(dev, dev->irq);if (ret) {IPVR_ERROR("Error installing IRQ handler: %d\n", ret);return -EFAULT;}IPVR_DEBUG_PM("Successfully installed IRQ\n");}elseIPVR_DEBUG_PM("irq_enabled is %d\n",- dev->irq_enabled);
return 0;+}
+static int32_t ipvr_pm_suspend(struct device *dev) {
struct platform_device *platformdev = to_platform_device(dev);struct drm_device *drm_dev = platform_get_drvdata(platformdev);IPVR_DEBUG_PM("PM suspend called\n");BUG_ON(!drm_dev);return ipvr_drm_freeze(drm_dev); } static int32_t+ipvr_pm_resume(struct device *dev) {
struct platform_device *platformdev = to_platform_device(dev);struct drm_device *drm_dev = platform_get_drvdata(platformdev);IPVR_DEBUG_PM("PM resume called\n");BUG_ON(!drm_dev);return ipvr_drm_thaw(drm_dev); }+/*
- dump GEM API is mainly for dumb buffers suitable for scanout,
- it is not needed for ipvr driver.
- gem_vm_ops is used for mmap case, also not needed for ipvr
- todo: prime support can be enabled later */ static struct
+drm_driver ipvr_drm_driver = {
.driver_features = DRIVER_HAVE_IRQ | DRIVER_GEM,.load = ipvr_drm_load,.unload = ipvr_drm_unload,.open = ipvr_drm_open,.preclose = ipvr_drm_preclose,.postclose = ipvr_drm_postclose,.irq_handler = ipvr_irq_handler,.set_busid = drm_platform_set_busid,This should be NULL for new drivers.
.gem_free_object = ipvr_gem_free_object, #ifdef+CONFIG_DEBUG_FS
.debugfs_init = ipvr_debugfs_init,.debugfs_cleanup = ipvr_debugfs_cleanup, #endif.ioctls = ipvr_gem_ioctls,.num_ioctls = ARRAY_SIZE(ipvr_gem_ioctls),.fops = &ipvr_fops,.name = IPVR_DRIVER_NAME,.desc = IPVR_DRIVER_DESC,.date = IPVR_DRIVER_DATE,.major = IPVR_DRIVER_MAJOR,.minor = IPVR_DRIVER_MINOR,.patchlevel = IPVR_DRIVER_PATCHLEVEL, };+static int32_t ipvr_plat_probe(struct platform_device *device) {
return drm_platform_init(&ipvr_drm_driver, device);drm_platform_init() should be used. Please use drm_dev_alloc(), drm_dev_register() directly. udl and tegra should serve as example.
Thx, will update it in next version.
+}
+static int32_t ipvr_plat_remove(struct platform_device *device) {
drm_put_dev(platform_get_drvdata(device));Again, please use drm_dev_unregister() + drm_dev_unref() directly.
Thx, will update it in next version.
Thanks David
return 0;+}
+static struct dev_pm_ops ipvr_pm_ops = {
.suspend = ipvr_pm_suspend,.resume = ipvr_pm_resume,.freeze = ipvr_pm_suspend,.thaw = ipvr_pm_resume,.poweroff = ipvr_pm_suspend,.restore = ipvr_pm_resume,+#ifdef CONFIG_PM_RUNTIME
.runtime_suspend = ipvr_pm_suspend,.runtime_resume = ipvr_pm_resume, #endif };+static struct platform_driver ipvr_plat_driver = {
.driver = {.name = "ipvr-ved",.owner = THIS_MODULE,+#ifdef CONFIG_PM
.pm = &ipvr_pm_ops,+#endif
},.probe = ipvr_plat_probe,.remove = ipvr_plat_remove,+};
+module_platform_driver(ipvr_plat_driver); +MODULE_LICENSE("GPL");
Hi
On Wed, Oct 15, 2014 at 4:14 AM, Cheng, Yao yao.cheng@intel.com wrote:
Hi Herrmann
-----Original Message----- From: David Herrmann [mailto:dh.herrmann@gmail.com] Sent: Monday, October 13, 2014 10:27 PM To: Cheng, Yao Cc: Intel Graphics Development; Jiang, Fei; dri-devel@lists.freedesktop.org; Vetter, Daniel Subject: Re: [RFC PATCH 2/3] drm/ipvr: drm driver for vxd392
Hi
+static struct drm_ioctl_desc ipvr_gem_ioctls[] = {
DRM_IOCTL_DEF_DRV(IPVR_CONTEXT_CREATE,ipvr_context_create_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_CONTEXT_DESTROY,ipvr_context_destroy_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_MISC,ipvr_misc_ioctl, DRM_AUTH),DRM_IOCTL_DEF_DRV(IPVR_GEM_EXECBUFFER,ipvr_gem_execbuffer, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_BUSY,ipvr_gem_busy_ioctl, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_CREATE,ipvr_gem_create_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_MMAP,ipvr_gem_mmap_ioctl, DRM_UNLOCKED),Why do you need this ioctl? mmap() should work perfectly fine. I don't see why you require people to use a ipvr specific ioctl to map buffers.
Many thanks to your comments, in our existing libdrm helper and userspace drivers, mmap_ioctl was the interface of mapping objects. We continued using the ioctl way for compatibility. Is it mandatory to implement mmap() to replace mmap_ioctl for GEM drivers?
Usually we don't care for existing user-space. We want APIs to be cleanly designed when they go upstream. But I'm not the authority to decide that. All I do is recommend hooking it up with mmap() like everyone else does. If that's not what you want, you have to convince Dave (and probably Daniel). The least you can do is hook up mmap() and make it work. Whether there is a legacy ioctl to do the same, I don't care..
Thanks David
On Wed, Oct 15, 2014 at 02:14:33AM +0000, Cheng, Yao wrote:
Hi Herrmann
-----Original Message----- From: David Herrmann [mailto:dh.herrmann@gmail.com] Sent: Monday, October 13, 2014 10:27 PM To: Cheng, Yao Cc: Intel Graphics Development; Jiang, Fei; dri-devel@lists.freedesktop.org; Vetter, Daniel Subject: Re: [RFC PATCH 2/3] drm/ipvr: drm driver for vxd392
Hi
+static struct drm_ioctl_desc ipvr_gem_ioctls[] = {
DRM_IOCTL_DEF_DRV(IPVR_CONTEXT_CREATE,ipvr_context_create_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_CONTEXT_DESTROY,ipvr_context_destroy_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_MISC,ipvr_misc_ioctl, DRM_AUTH),DRM_IOCTL_DEF_DRV(IPVR_GEM_EXECBUFFER,ipvr_gem_execbuffer, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_BUSY,ipvr_gem_busy_ioctl, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_CREATE,ipvr_gem_create_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_MMAP,ipvr_gem_mmap_ioctl, DRM_UNLOCKED),Why do you need this ioctl? mmap() should work perfectly fine. I don't see why you require people to use a ipvr specific ioctl to map buffers.
Many thanks to your comments, in our existing libdrm helper and userspace drivers, mmap_ioctl was the interface of mapping objects. We continued using the ioctl way for compatibility. Is it mandatory to implement mmap() to replace mmap_ioctl for GEM drivers?
Yeah, the new way is to have a ioctl in the drm driver to create an mmap offset (look at the gtt_mmap_offset ioctl for i915), and the use mmap on the drm fd. Doing a direct driver ioctl to forward the mmap is deprecated.
The reason for that is that a) this is the new standard way used by all drm drivers b) if you use the standard mmap system call debug tools like valgrind will understand it without any special actions.
I'll do a patch to i915 so that people stop copy-pasting that old misdesign. -Daniel
DRM_IOCTL_DEF_DRV(IPVR_SYNC_CPU,ipvr_sync_cpu_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_WAIT,ipvr_gem_wait_ioctl, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_USERPTR,ipvr_gem_userptr_ioctl, DRM_UNLOCKED), };+static void ipvr_gem_init(struct drm_device *dev) {
struct drm_ipvr_private *dev_priv = dev->dev_private;dev_priv->ipvr_bo_slab = kmem_cache_create("ipvr_gem_object",sizeof(union drm_ipvr_gem_objects), 0,SLAB_HWCACHE_ALIGN, NULL);INIT_LIST_HEAD(&dev_priv->ipvr_mm.unbound_list);INIT_LIST_HEAD(&dev_priv->ipvr_mm.bound_list);spin_lock_init(&dev_priv->ipvr_mm.object_stat_lock);dev_priv->ipvr_mm.interruptible = true; }+static void ipvr_gem_setup_mmu(struct drm_device *dev,
unsigned long linear_start,unsigned long linear_end,unsigned long tiling_start,unsigned long tiling_end) {/* Let GEM Manage all of the aperture.** However, leave one page at the end still bound to the scratch page.* There are a number of places where hardware apparentlyprefetches
* past the end of the object, and we've seen multiple hangs with the* GPU head pointer stuck in a batchbuffer bound at last page of the* aperture. One page should be enough to keep any prefetchinginside
* of the aperture.*/struct drm_ipvr_private *dev_priv = dev->dev_private;struct ipvr_address_space *addr_space = &dev_priv->addr_space;/* todo: add sanity check */addr_space->dev = dev_priv->dev;INIT_LIST_HEAD(&addr_space->active_list);INIT_LIST_HEAD(&addr_space->inactive_list);/* Subtract the guard page ... */drm_mm_init(&addr_space->linear_mm, linear_start,linear_end - linear_start - PAGE_SIZE);dev_priv->addr_space.linear_start = linear_start;dev_priv->addr_space.linear_total = linear_end - linear_start;drm_mm_init(&addr_space->tiling_mm, tiling_start,tiling_end - tiling_start - PAGE_SIZE);dev_priv->addr_space.tiling_start = tiling_start;dev_priv->addr_space.tiling_total = tiling_end - tiling_start;+}
+static void ipvr_do_takedown(struct drm_device *dev) {
/* todo: need check if need clean up mm here */ipvr_ved_uninit(dev);+}
+static int32_t ipvr_drm_unload(struct drm_device *dev) {
struct drm_ipvr_private *dev_priv = dev->dev_private;BUG_ON(!dev->platformdev);BUG_ON(atomic_read(&dev_priv->ved_power_usage));IPVR_DEBUG_ENTRY("entered.");if (dev_priv) {WARN_ON(pm_runtime_get_sync(&dev->platformdev->dev) <- 0);
if (dev_priv->ipvr_bo_slab)kmem_cache_destroy(dev_priv->ipvr_bo_slab);ipvr_fence_driver_fini(dev_priv);ipvr_do_takedown(dev);- WARN_ON(pm_runtime_put_sync_suspend(&dev->platformdev->dev)
< 0);
if (dev_priv->validate_ctx.buffers)vfree(dev_priv->validate_ctx.buffers);if (dev_priv->pf_pd) {ipvr_mmu_free_pagedir(dev_priv->pf_pd);dev_priv->pf_pd = NULL;}if (dev_priv->mmu) {ipvr_mmu_driver_takedown(dev_priv->mmu);dev_priv->mmu = NULL;}if (dev_priv->ved_reg_base) {iounmap(dev_priv->ved_reg_base - dev_priv-ved_reg_offset);
dev_priv->ved_reg_base = NULL;dev_priv->ved_reg_offset = 0;}list_del(&dev_priv->default_ctx.head);idr_remove(&dev_priv->ipvr_ctx_idr, dev_priv-default_ctx.ctx_id);
kfree(dev_priv);}pm_runtime_disable(&dev->platformdev->dev);return 0;+}
+static int32_t ipvr_drm_load(struct drm_device *dev, unsigned long +flags) {
struct drm_ipvr_private *dev_priv;int32_t ctx_id, ret = 0;struct platform_device *platdev;struct resource *res_mmio, *res_reg;void __iomem* mmio_addr;dev_priv = kzalloc(sizeof(*dev_priv), GFP_KERNEL);if (dev_priv == NULL)return -ENOMEM;dev->dev_private = dev_priv;dev_priv->dev = dev;BUG_ON(!dev->platformdev);platdev = dev->platformdev;mutex_init(&dev_priv->cmdbuf_mutex);INIT_LIST_HEAD(&dev_priv->validate_ctx.validate_list);dev_priv->pci_root = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));if (!dev_priv->pci_root) {kfree(dev_priv);return -ENODEV;}dev->irq = platform_get_irq(platdev, 0);if (dev->irq < 0) {ret = -ENODEV;goto out_err;}res_mmio = platform_get_resource(platdev, IORESOURCE_MEM, 0);res_reg = platform_get_resource(platdev, IORESOURCE_REG, 0);if (!res_mmio || !res_reg) {ret = -ENXIO;goto out_err;}mmio_addr = ioremap_wc(res_mmio->start,res_mmio->end - res_mmio->start);if (IS_ERR(mmio_addr)) {ret = -EACCES;goto out_err;}dev_priv->ved_reg_base = mmio_addr + res_reg->start;dev_priv->ved_reg_offset = res_reg->start;IPVR_DEBUG_VED("ved_reg_base is %p, range is 0x%llx - 0x%llx.\n",dev_priv->ved_reg_base,res_reg->start, res_reg->end);platform_set_drvdata(dev->platformdev, dev);pm_runtime_enable(&dev->platformdev->dev);if (pm_runtime_get_sync(&dev->platformdev->dev) < 0) {ret = -EBUSY;goto out_err;}IPVR_DEBUG_INIT("MSVDX_CORE_REV_OFFSET by readl is 0x%x.\n",readl(dev_priv->ved_reg_base + 0x640));IPVR_DEBUG_INIT("MSVDX_CORE_REV_OFFSET byVED_REG_READ32 is 0x%x.\n",
VED_REG_READ32(MSVDX_CORE_REV_OFFSET));/* mmu init */dev_priv->mmu = ipvr_mmu_driver_init((void *)0,drm_ipvr_trap_pagefaults,0, dev_priv);if (!dev_priv->mmu) {ret = -EBUSY;goto out_err;}dev_priv->pf_pd = ipvr_mmu_alloc_pd(dev_priv->mmu, 1, 0);if (!dev_priv->pf_pd) {ret = -ENOMEM;goto out_err;}ipvr_mmu_set_pd_context(ipvr_mmu_get_default_pd(dev_priv-mmu), 0);
ipvr_mmu_set_pd_context(dev_priv->pf_pd, 1);/** Initialize sequence numbers for the different command* submission mechanisms.*/dev_priv->last_seq = 1;ipvr_gem_init(dev);ipvr_gem_setup_mmu(dev,IPVR_MEM_MMU_LINEAR_START,IPVR_MEM_MMU_LINEAR_END,IPVR_MEM_MMU_TILING_START,IPVR_MEM_MMU_TILING_END);ipvr_ved_init(dev);WARN_ON(pm_runtime_put_sync_suspend(&dev->platformdev-dev) <
- 0);
dev_priv->ved_private->ved_needs_reset = 1;mutex_init(&dev_priv->ved_pm_mutex);atomic_set(&dev_priv->ved_power_usage, 0);ipvr_fence_driver_init(dev_priv);dev_priv->validate_ctx.buffers =vmalloc(IPVR_NUM_VALIDATE_BUFFERS *sizeof(struct ipvr_validate_buffer));if (!dev_priv->validate_ctx.buffers) {ret = -ENOMEM;goto out_err;}/* ipvr context initialization */INIT_LIST_HEAD(&dev_priv->ipvr_ctx_list);spin_lock_init(&dev_priv->ipvr_ctx_lock);idr_init(&dev_priv->ipvr_ctx_idr);/* default ipvr context is used for scaling, rotation case */ctx_id = idr_alloc(&dev_priv->ipvr_ctx_idr, &dev_priv->default_ctx,IPVR_MIN_CONTEXT_ID, IPVR_MAX_CONTEXT_ID,GFP_KERNEL);if (ctx_id < 0) {return -ENOMEM;goto out_err;}dev_priv->default_ctx.ctx_id = ctx_id;INIT_LIST_HEAD(&dev_priv->default_ctx.head);dev_priv->default_ctx.ctx_type = 0;dev_priv->default_ctx.ipvr_fpriv = NULL;/* don't need protect with spinlock during module load stage */list_add(&dev_priv->default_ctx.head, &dev_priv->ipvr_ctx_list);dev_priv->default_ctx.tiling_scheme = 0;dev_priv->default_ctx.tiling_stride = 0;return 0;+out_err:
ipvr_drm_unload(dev);return ret;+}
+/*
- The .open() method is called every time the device is opened by an
- application. Drivers can allocate per-file private data in this
+method and
- store them in the struct drm_file::driver_priv field. Note that
+the .open()
- method is called before .firstopen().
- */
+static int32_t +ipvr_drm_open(struct drm_device *dev, struct drm_file *file_priv) {
struct drm_ipvr_file_private *ipvr_fp;IPVR_DEBUG_ENTRY("enter\n");ipvr_fp = kzalloc(sizeof(*ipvr_fp), GFP_KERNEL);if (unlikely(ipvr_fp == NULL))return -ENOMEM;file_priv->driver_priv = ipvr_fp;return 0;+}
+/*
- The close operation is split into .preclose() and .postclose() methods.
- Drivers must stop and cleanup all per-file operations in the
+.preclose()
- method. For instance pending vertical blanking and page flip
+events must be
- cancelled. No per-file operation is allowed on the file handle
+after
- returning from the .preclose() method.
- */
+static void +ipvr_drm_preclose(struct drm_device *dev, struct drm_file *file_priv) +{
/* if user didn't destory ctx explicitly, remove ctx here */struct drm_ipvr_private *dev_priv;struct drm_ipvr_file_private *ipvr_fpriv;struct ved_private *ved_priv;struct ipvr_context *ipvr_ctx = NULL;unsigned long irq_flags;IPVR_DEBUG_ENTRY("enter\n");dev_priv = dev->dev_private;ipvr_fpriv = file_priv->driver_priv;ved_priv = dev_priv->ved_private;if (ipvr_fpriv->ctx_id == IPVR_CONTEXT_INVALID_ID)return;ipvr_ctx = (struct ipvr_context *)idr_find(&dev_priv->ipvr_ctx_idr, ipvr_fpriv->ctx_id);if (!ipvr_ctx || (ipvr_ctx->ipvr_fpriv != ipvr_fpriv)) {IPVR_DEBUG_GENERAL("ctx for id %d has already destroyed\n",ipvr_fpriv->ctx_id);return;}IPVR_DEBUG_PM("Video:remove context profile %d,entrypoint %d\n",
(ipvr_ctx->ctx_type >> 8) & 0xff,(ipvr_ctx->ctx_type & 0xff));mutex_lock(&ved_priv->ved_mutex);if (ved_priv->ipvr_ctx == ipvr_ctx )ved_priv->ipvr_ctx = NULL;mutex_unlock(&ved_priv->ved_mutex);spin_lock_irqsave(&dev_priv->ipvr_ctx_lock, irq_flags);list_del(&ipvr_ctx->head);ipvr_fpriv->ctx_id = IPVR_CONTEXT_INVALID_ID;spin_unlock_irqrestore(&dev_priv->ipvr_ctx_lock, irq_flags);idr_remove(&dev_priv->ipvr_ctx_idr, ipvr_ctx->ctx_id);kfree(ipvr_ctx );+}
+/*
- Finally the .postclose() method is called as the last step of the
+close
- operation, right before calling the .lastclose() method if no
+other open
- file handle exists for the device. Drivers that have allocated
+per-file
- private data in the .open() method should free it here.
- */
+static void +ipvr_drm_postclose(struct drm_device *dev, struct drm_file +*file_priv) {
struct drm_ipvr_file_private *ipvr_fpriv = file_priv->driver_priv;IPVR_DEBUG_ENTRY("enter\n");kfree(ipvr_fpriv);+}
postclose() is deprecated. You can safely move this into preclose().
Thx, will update it in next version.
+static irqreturn_t ipvr_irq_handler(int32_t irq, void *arg) {
struct drm_device *dev = (struct drm_device *) arg;WARN_ON(ved_irq_handler(dev));return IRQ_HANDLED;+}
+static const struct file_operations ipvr_fops = {
.owner = THIS_MODULE,.open = drm_open,.release = drm_release,.unlocked_ioctl = drm_ioctl,+#ifdef CONFIG_COMPAT
.compat_ioctl = drm_ioctl,+#endif
/* no need to define mmap. User space maps bo with+DRM_IPVR_GEM_MMAP */ };
+static int32_t ipvr_drm_freeze(struct drm_device *dev) {
int32_t ret;int32_t power_usage;struct drm_ipvr_private *dev_priv = dev->dev_private;IPVR_DEBUG_ENTRY("enter\n");power_usage = atomic_read(&dev_priv->ved_power_usage);BUG_ON(power_usage < 0);if (power_usage > 0) {IPVR_DEBUG_PM("VED power usage is %d, skip freezing\n",power_usage);
return 0;}ret = ved_check_idle(dev);if (ret) {IPVR_DEBUG_PM("VED check idle fail: %d, skip freezing\n", ret);return 0;}if (dev->irq_enabled) {ret = drm_irq_uninstall(dev);if (unlikely(ret)) {IPVR_ERROR("Error uninstalling IRQ handler: %d\n", ret);return -EFAULT;}IPVR_DEBUG_PM("Successfully uninstalled IRQ\n");}elseIPVR_DEBUG_PM("irq_enabled is %d\n",- dev->irq_enabled);
if (is_ved_on(dev)) {if (!ved_power_off(dev)) {IPVR_ERROR("Failed to power off VED\n");return -EFAULT;}IPVR_DEBUG_PM("Successfully powered off\n");} else {IPVR_DEBUG_PM("Skiped power-off since already poweredoff\n");
}return 0;+}
+static int32_t ipvr_drm_thaw(struct drm_device *dev) {
int ret;IPVR_DEBUG_ENTRY("enter\n");if (!is_ved_on(dev)) {if (!ved_power_on(dev)) {IPVR_ERROR("Failed to power on VED\n");return -EFAULT;}IPVR_DEBUG_PM("Successfully powered on\n");} else {IPVR_DEBUG_PM("Skiped power-on since already poweredon\n");
}if (!dev->irq_enabled) {ret = drm_irq_install(dev, dev->irq);if (ret) {IPVR_ERROR("Error installing IRQ handler: %d\n", ret);return -EFAULT;}IPVR_DEBUG_PM("Successfully installed IRQ\n");}elseIPVR_DEBUG_PM("irq_enabled is %d\n",- dev->irq_enabled);
return 0;+}
+static int32_t ipvr_pm_suspend(struct device *dev) {
struct platform_device *platformdev = to_platform_device(dev);struct drm_device *drm_dev = platform_get_drvdata(platformdev);IPVR_DEBUG_PM("PM suspend called\n");BUG_ON(!drm_dev);return ipvr_drm_freeze(drm_dev); } static int32_t+ipvr_pm_resume(struct device *dev) {
struct platform_device *platformdev = to_platform_device(dev);struct drm_device *drm_dev = platform_get_drvdata(platformdev);IPVR_DEBUG_PM("PM resume called\n");BUG_ON(!drm_dev);return ipvr_drm_thaw(drm_dev); }+/*
- dump GEM API is mainly for dumb buffers suitable for scanout,
- it is not needed for ipvr driver.
- gem_vm_ops is used for mmap case, also not needed for ipvr
- todo: prime support can be enabled later */ static struct
+drm_driver ipvr_drm_driver = {
.driver_features = DRIVER_HAVE_IRQ | DRIVER_GEM,.load = ipvr_drm_load,.unload = ipvr_drm_unload,.open = ipvr_drm_open,.preclose = ipvr_drm_preclose,.postclose = ipvr_drm_postclose,.irq_handler = ipvr_irq_handler,.set_busid = drm_platform_set_busid,This should be NULL for new drivers.
.gem_free_object = ipvr_gem_free_object, #ifdef+CONFIG_DEBUG_FS
.debugfs_init = ipvr_debugfs_init,.debugfs_cleanup = ipvr_debugfs_cleanup, #endif.ioctls = ipvr_gem_ioctls,.num_ioctls = ARRAY_SIZE(ipvr_gem_ioctls),.fops = &ipvr_fops,.name = IPVR_DRIVER_NAME,.desc = IPVR_DRIVER_DESC,.date = IPVR_DRIVER_DATE,.major = IPVR_DRIVER_MAJOR,.minor = IPVR_DRIVER_MINOR,.patchlevel = IPVR_DRIVER_PATCHLEVEL, };+static int32_t ipvr_plat_probe(struct platform_device *device) {
return drm_platform_init(&ipvr_drm_driver, device);drm_platform_init() should be used. Please use drm_dev_alloc(), drm_dev_register() directly. udl and tegra should serve as example.
Thx, will update it in next version.
+}
+static int32_t ipvr_plat_remove(struct platform_device *device) {
drm_put_dev(platform_get_drvdata(device));Again, please use drm_dev_unregister() + drm_dev_unref() directly.
Thx, will update it in next version.
Thanks David
return 0;+}
+static struct dev_pm_ops ipvr_pm_ops = {
.suspend = ipvr_pm_suspend,.resume = ipvr_pm_resume,.freeze = ipvr_pm_suspend,.thaw = ipvr_pm_resume,.poweroff = ipvr_pm_suspend,.restore = ipvr_pm_resume,+#ifdef CONFIG_PM_RUNTIME
.runtime_suspend = ipvr_pm_suspend,.runtime_resume = ipvr_pm_resume, #endif };+static struct platform_driver ipvr_plat_driver = {
.driver = {.name = "ipvr-ved",.owner = THIS_MODULE,+#ifdef CONFIG_PM
.pm = &ipvr_pm_ops,+#endif
},.probe = ipvr_plat_probe,.remove = ipvr_plat_remove,+};
+module_platform_driver(ipvr_plat_driver); +MODULE_LICENSE("GPL");
Intel-gfx mailing list Intel-gfx@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Hi Daniel/Hermann,
-----Original Message----- From: Daniel Vetter [mailto:daniel.vetter@ffwll.ch] On Behalf Of Daniel Vetter Sent: Thursday, October 16, 2014 6:29 PM To: Cheng, Yao Cc: David Herrmann; Vetter, Daniel; Intel Graphics Development; Jiang, Fei; dri-devel@lists.freedesktop.org Subject: Re: [Intel-gfx] [RFC PATCH 2/3] drm/ipvr: drm driver for vxd392
On Wed, Oct 15, 2014 at 02:14:33AM +0000, Cheng, Yao wrote:
Hi Herrmann
-----Original Message----- From: David Herrmann [mailto:dh.herrmann@gmail.com] Sent: Monday, October 13, 2014 10:27 PM To: Cheng, Yao Cc: Intel Graphics Development; Jiang, Fei; dri-devel@lists.freedesktop.org; Vetter, Daniel Subject: Re: [RFC PATCH 2/3] drm/ipvr: drm driver for vxd392
Hi
+static struct drm_ioctl_desc ipvr_gem_ioctls[] = {
DRM_IOCTL_DEF_DRV(IPVR_CONTEXT_CREATE,ipvr_context_create_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_CONTEXT_DESTROY,ipvr_context_destroy_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_MISC,ipvr_misc_ioctl, DRM_AUTH),DRM_IOCTL_DEF_DRV(IPVR_GEM_EXECBUFFER,ipvr_gem_execbuffer, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_BUSY,ipvr_gem_busy_ioctl, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_CREATE,ipvr_gem_create_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_MMAP,ipvr_gem_mmap_ioctl, DRM_UNLOCKED),Why do you need this ioctl? mmap() should work perfectly fine. I don't see why you require people to use a ipvr specific ioctl to map
buffers.
Many thanks to your comments, in our existing libdrm helper and userspace drivers, mmap_ioctl was the interface of mapping objects. We continued using the ioctl way for compatibility. Is it mandatory to implement mmap() to replace mmap_ioctl for GEM drivers?
Yeah, the new way is to have a ioctl in the drm driver to create an mmap offset (look at the gtt_mmap_offset ioctl for i915), and the use mmap on the drm fd. Doing a direct driver ioctl to forward the mmap is deprecated.
The reason for that is that a) this is the new standard way used by all drm drivers b) if you use the standard mmap system call debug tools like valgrind will understand it without any special actions.
I'll do a patch to i915 so that people stop copy-pasting that old misdesign. -Daniel
Accepted :) I will update the patch to implement the mmap interface and remove the legacy MMAP_IOCTL. BTW I didn't see a field to get mmap_offset in struct drm_gem_open, I guess something like a new "GET_MMAP_OFFSET_IOCTL" need be added to support mapping flinked/primed BO, is it?
DRM_IOCTL_DEF_DRV(IPVR_SYNC_CPU,ipvr_sync_cpu_ioctl, DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_WAIT,ipvr_gem_wait_ioctl, DRM_AUTH|DRM_UNLOCKED),DRM_IOCTL_DEF_DRV(IPVR_GEM_USERPTR,ipvr_gem_userptr_ioctl, DRM_UNLOCKED), };+static void ipvr_gem_init(struct drm_device *dev) {
struct drm_ipvr_private *dev_priv = dev->dev_private;dev_priv->ipvr_bo_slab =kmem_cache_create("ipvr_gem_object",
sizeof(union drm_ipvr_gem_objects), 0,SLAB_HWCACHE_ALIGN, NULL);INIT_LIST_HEAD(&dev_priv->ipvr_mm.unbound_list);INIT_LIST_HEAD(&dev_priv->ipvr_mm.bound_list);spin_lock_init(&dev_priv->ipvr_mm.object_stat_lock);dev_priv->ipvr_mm.interruptible = true; }+static void ipvr_gem_setup_mmu(struct drm_device *dev,
unsigned long linear_start,unsigned long linear_end,unsigned long tiling_start,unsigned long tiling_end) {/* Let GEM Manage all of the aperture.** However, leave one page at the end still bound to the scratchpage.
* There are a number of places where hardware apparentlyprefetches
* past the end of the object, and we've seen multiple hangs withthe
* GPU head pointer stuck in a batchbuffer bound at last page ofthe
* aperture. One page should be enough to keep any- prefetching
inside
* of the aperture.*/struct drm_ipvr_private *dev_priv = dev->dev_private;struct ipvr_address_space *addr_space =- &dev_priv->addr_space;
/* todo: add sanity check */addr_space->dev = dev_priv->dev;INIT_LIST_HEAD(&addr_space->active_list);INIT_LIST_HEAD(&addr_space->inactive_list);/* Subtract the guard page ... */drm_mm_init(&addr_space->linear_mm, linear_start,linear_end - linear_start - PAGE_SIZE);dev_priv->addr_space.linear_start = linear_start;dev_priv->addr_space.linear_total = linear_end -- linear_start;
drm_mm_init(&addr_space->tiling_mm, tiling_start,tiling_end - tiling_start - PAGE_SIZE);dev_priv->addr_space.tiling_start = tiling_start;dev_priv->addr_space.tiling_total = tiling_end -+tiling_start; }
+static void ipvr_do_takedown(struct drm_device *dev) {
/* todo: need check if need clean up mm here */ipvr_ved_uninit(dev);+}
+static int32_t ipvr_drm_unload(struct drm_device *dev) {
struct drm_ipvr_private *dev_priv = dev->dev_private;BUG_ON(!dev->platformdev);BUG_ON(atomic_read(&dev_priv->ved_power_usage));IPVR_DEBUG_ENTRY("entered.");if (dev_priv) {- WARN_ON(pm_runtime_get_sync(&dev->platformdev->dev) < 0);
if (dev_priv->ipvr_bo_slab)kmem_cache_destroy(dev_priv->ipvr_bo_slab);ipvr_fence_driver_fini(dev_priv);ipvr_do_takedown(dev);- WARN_ON(pm_runtime_put_sync_suspend(&dev->platformdev-
dev)
< 0);
if (dev_priv->validate_ctx.buffers)vfree(dev_priv->validate_ctx.buffers);if (dev_priv->pf_pd) {ipvr_mmu_free_pagedir(dev_priv->pf_pd);dev_priv->pf_pd = NULL;}if (dev_priv->mmu) {ipvr_mmu_driver_takedown(dev_priv->mmu);dev_priv->mmu = NULL;}if (dev_priv->ved_reg_base) {iounmap(dev_priv->ved_reg_base - dev_priv-ved_reg_offset);
dev_priv->ved_reg_base = NULL;dev_priv->ved_reg_offset = 0;}list_del(&dev_priv->default_ctx.head);idr_remove(&dev_priv->ipvr_ctx_idr, dev_priv-default_ctx.ctx_id);
kfree(dev_priv);}pm_runtime_disable(&dev->platformdev->dev);return 0;+}
+static int32_t ipvr_drm_load(struct drm_device *dev, unsigned +long +flags) {
struct drm_ipvr_private *dev_priv;int32_t ctx_id, ret = 0;struct platform_device *platdev;struct resource *res_mmio, *res_reg;void __iomem* mmio_addr;dev_priv = kzalloc(sizeof(*dev_priv), GFP_KERNEL);if (dev_priv == NULL)return -ENOMEM;dev->dev_private = dev_priv;dev_priv->dev = dev;BUG_ON(!dev->platformdev);platdev = dev->platformdev;mutex_init(&dev_priv->cmdbuf_mutex);INIT_LIST_HEAD(&dev_priv->validate_ctx.validate_list);dev_priv->pci_root = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));if (!dev_priv->pci_root) {kfree(dev_priv);return -ENODEV;}dev->irq = platform_get_irq(platdev, 0);if (dev->irq < 0) {ret = -ENODEV;goto out_err;}res_mmio = platform_get_resource(platdev, IORESOURCE_MEM,0);
res_reg = platform_get_resource(platdev, IORESOURCE_REG, 0);if (!res_mmio || !res_reg) {ret = -ENXIO;goto out_err;}mmio_addr = ioremap_wc(res_mmio->start,res_mmio->end - res_mmio->start);if (IS_ERR(mmio_addr)) {ret = -EACCES;goto out_err;}dev_priv->ved_reg_base = mmio_addr + res_reg->start;dev_priv->ved_reg_offset = res_reg->start;IPVR_DEBUG_VED("ved_reg_base is %p, range is 0x%llx -0x%llx.\n",
dev_priv->ved_reg_base,res_reg->start, res_reg->end);platform_set_drvdata(dev->platformdev, dev);pm_runtime_enable(&dev->platformdev->dev);if (pm_runtime_get_sync(&dev->platformdev->dev) < 0) {ret = -EBUSY;goto out_err;}IPVR_DEBUG_INIT("MSVDX_CORE_REV_OFFSET by readl is0x%x.\n",
readl(dev_priv->ved_reg_base + 0x640));IPVR_DEBUG_INIT("MSVDX_CORE_REV_OFFSET byVED_REG_READ32 is 0x%x.\n",
VED_REG_READ32(MSVDX_CORE_REV_OFFSET));/* mmu init */dev_priv->mmu = ipvr_mmu_driver_init((void *)0,drm_ipvr_trap_pagefaults,0, dev_priv);if (!dev_priv->mmu) {ret = -EBUSY;goto out_err;}dev_priv->pf_pd = ipvr_mmu_alloc_pd(dev_priv->mmu, 1, 0);if (!dev_priv->pf_pd) {ret = -ENOMEM;goto out_err;}ipvr_mmu_set_pd_context(ipvr_mmu_get_default_pd(dev_priv-mmu), 0);
ipvr_mmu_set_pd_context(dev_priv->pf_pd, 1);/** Initialize sequence numbers for the different command* submission mechanisms.*/dev_priv->last_seq = 1;ipvr_gem_init(dev);ipvr_gem_setup_mmu(dev,IPVR_MEM_MMU_LINEAR_START,IPVR_MEM_MMU_LINEAR_END,IPVR_MEM_MMU_TILING_START,IPVR_MEM_MMU_TILING_END);ipvr_ved_init(dev);WARN_ON(pm_runtime_put_sync_suspend(&dev->platformdev-dev) <
- 0);
dev_priv->ved_private->ved_needs_reset = 1;mutex_init(&dev_priv->ved_pm_mutex);atomic_set(&dev_priv->ved_power_usage, 0);ipvr_fence_driver_init(dev_priv);dev_priv->validate_ctx.buffers =vmalloc(IPVR_NUM_VALIDATE_BUFFERS *sizeof(struct ipvr_validate_buffer));if (!dev_priv->validate_ctx.buffers) {ret = -ENOMEM;goto out_err;}/* ipvr context initialization */INIT_LIST_HEAD(&dev_priv->ipvr_ctx_list);spin_lock_init(&dev_priv->ipvr_ctx_lock);idr_init(&dev_priv->ipvr_ctx_idr);/* default ipvr context is used for scaling, rotation case */ctx_id = idr_alloc(&dev_priv->ipvr_ctx_idr, &dev_priv-default_ctx,
IPVR_MIN_CONTEXT_ID, IPVR_MAX_CONTEXT_ID,GFP_KERNEL);if (ctx_id < 0) {return -ENOMEM;goto out_err;}dev_priv->default_ctx.ctx_id = ctx_id;INIT_LIST_HEAD(&dev_priv->default_ctx.head);dev_priv->default_ctx.ctx_type = 0;dev_priv->default_ctx.ipvr_fpriv = NULL;/* don't need protect with spinlock during module load stage */list_add(&dev_priv->default_ctx.head, &dev_priv->ipvr_ctx_list);dev_priv->default_ctx.tiling_scheme = 0;dev_priv->default_ctx.tiling_stride = 0;return 0;+out_err:
ipvr_drm_unload(dev);return ret;+}
+/*
- The .open() method is called every time the device is opened
+by an
- application. Drivers can allocate per-file private data in
+this method and
- store them in the struct drm_file::driver_priv field. Note
+that the .open()
- method is called before .firstopen().
- */
+static int32_t +ipvr_drm_open(struct drm_device *dev, struct drm_file *file_priv) {
struct drm_ipvr_file_private *ipvr_fp;IPVR_DEBUG_ENTRY("enter\n");ipvr_fp = kzalloc(sizeof(*ipvr_fp), GFP_KERNEL);if (unlikely(ipvr_fp == NULL))return -ENOMEM;file_priv->driver_priv = ipvr_fp;return 0;+}
+/*
- The close operation is split into .preclose() and .postclose() methods.
- Drivers must stop and cleanup all per-file operations in the
+.preclose()
- method. For instance pending vertical blanking and page flip
+events must be
- cancelled. No per-file operation is allowed on the file handle
+after
- returning from the .preclose() method.
- */
+static void +ipvr_drm_preclose(struct drm_device *dev, struct drm_file +*file_priv) {
/* if user didn't destory ctx explicitly, remove ctx here */struct drm_ipvr_private *dev_priv;struct drm_ipvr_file_private *ipvr_fpriv;struct ved_private *ved_priv;struct ipvr_context *ipvr_ctx = NULL;unsigned long irq_flags;IPVR_DEBUG_ENTRY("enter\n");dev_priv = dev->dev_private;ipvr_fpriv = file_priv->driver_priv;ved_priv = dev_priv->ved_private;if (ipvr_fpriv->ctx_id == IPVR_CONTEXT_INVALID_ID)return;ipvr_ctx = (struct ipvr_context *)idr_find(&dev_priv->ipvr_ctx_idr, ipvr_fpriv->ctx_id);if (!ipvr_ctx || (ipvr_ctx->ipvr_fpriv != ipvr_fpriv)) {IPVR_DEBUG_GENERAL("ctx for id %d has alreadydestroyed\n",
ipvr_fpriv->ctx_id);return;}IPVR_DEBUG_PM("Video:remove context profile %d,entrypoint %d\n",
(ipvr_ctx->ctx_type >> 8) & 0xff,(ipvr_ctx->ctx_type & 0xff));mutex_lock(&ved_priv->ved_mutex);if (ved_priv->ipvr_ctx == ipvr_ctx )ved_priv->ipvr_ctx = NULL;mutex_unlock(&ved_priv->ved_mutex);spin_lock_irqsave(&dev_priv->ipvr_ctx_lock, irq_flags);list_del(&ipvr_ctx->head);ipvr_fpriv->ctx_id = IPVR_CONTEXT_INVALID_ID;spin_unlock_irqrestore(&dev_priv->ipvr_ctx_lock,- irq_flags);
idr_remove(&dev_priv->ipvr_ctx_idr, ipvr_ctx->ctx_id);kfree(ipvr_ctx );+}
+/*
- Finally the .postclose() method is called as the last step of
+the close
- operation, right before calling the .lastclose() method if no
+other open
- file handle exists for the device. Drivers that have allocated
+per-file
- private data in the .open() method should free it here.
- */
+static void +ipvr_drm_postclose(struct drm_device *dev, struct drm_file +*file_priv) {
struct drm_ipvr_file_private *ipvr_fpriv = file_priv->driver_priv;IPVR_DEBUG_ENTRY("enter\n");kfree(ipvr_fpriv);+}
postclose() is deprecated. You can safely move this into preclose().
Thx, will update it in next version.
+static irqreturn_t ipvr_irq_handler(int32_t irq, void *arg) {
struct drm_device *dev = (struct drm_device *) arg;WARN_ON(ved_irq_handler(dev));return IRQ_HANDLED;+}
+static const struct file_operations ipvr_fops = {
.owner = THIS_MODULE,.open = drm_open,.release = drm_release,.unlocked_ioctl = drm_ioctl, #ifdef CONFIG_COMPAT.compat_ioctl = drm_ioctl, #endif/* no need to define mmap. User space maps bo with+DRM_IPVR_GEM_MMAP */ };
+static int32_t ipvr_drm_freeze(struct drm_device *dev) {
int32_t ret;int32_t power_usage;struct drm_ipvr_private *dev_priv = dev->dev_private;IPVR_DEBUG_ENTRY("enter\n");power_usage = atomic_read(&dev_priv->ved_power_usage);BUG_ON(power_usage < 0);if (power_usage > 0) {IPVR_DEBUG_PM("VED power usage is %d, skip- freezing\n",
power_usage);
return 0;}ret = ved_check_idle(dev);if (ret) {IPVR_DEBUG_PM("VED check idle fail: %d, skip freezing\n",ret);
return 0;}if (dev->irq_enabled) {ret = drm_irq_uninstall(dev);if (unlikely(ret)) {IPVR_ERROR("Error uninstalling IRQ handler: %d\n", ret);return -EFAULT;}IPVR_DEBUG_PM("Successfully uninstalled IRQ\n");}elseIPVR_DEBUG_PM("irq_enabled is %d\n",- dev->irq_enabled);
if (is_ved_on(dev)) {if (!ved_power_off(dev)) {IPVR_ERROR("Failed to power off VED\n");return -EFAULT;}IPVR_DEBUG_PM("Successfully powered off\n");} else {IPVR_DEBUG_PM("Skiped power-off since already- powered
off\n");
}return 0;+}
+static int32_t ipvr_drm_thaw(struct drm_device *dev) {
int ret;IPVR_DEBUG_ENTRY("enter\n");if (!is_ved_on(dev)) {if (!ved_power_on(dev)) {IPVR_ERROR("Failed to power on VED\n");return -EFAULT;}IPVR_DEBUG_PM("Successfully powered on\n");} else {IPVR_DEBUG_PM("Skiped power-on since already+powered
on\n");
}if (!dev->irq_enabled) {ret = drm_irq_install(dev, dev->irq);if (ret) {IPVR_ERROR("Error installing IRQ handler: %d\n", ret);return -EFAULT;}IPVR_DEBUG_PM("Successfully installed IRQ\n");}elseIPVR_DEBUG_PM("irq_enabled is %d\n",- dev->irq_enabled);
return 0;+}
+static int32_t ipvr_pm_suspend(struct device *dev) {
struct platform_device *platformdev = to_platform_device(dev);struct drm_device *drm_dev =platform_get_drvdata(platformdev);
IPVR_DEBUG_PM("PM suspend called\n");BUG_ON(!drm_dev);return ipvr_drm_freeze(drm_dev); } static int32_t+ipvr_pm_resume(struct device *dev) {
struct platform_device *platformdev = to_platform_device(dev);struct drm_device *drm_dev =platform_get_drvdata(platformdev);
IPVR_DEBUG_PM("PM resume called\n");BUG_ON(!drm_dev);return ipvr_drm_thaw(drm_dev); }+/*
- dump GEM API is mainly for dumb buffers suitable for scanout,
- it is not needed for ipvr driver.
- gem_vm_ops is used for mmap case, also not needed for ipvr
- todo: prime support can be enabled later */ static struct
+drm_driver ipvr_drm_driver = {
.driver_features = DRIVER_HAVE_IRQ | DRIVER_GEM,.load = ipvr_drm_load,.unload = ipvr_drm_unload,.open = ipvr_drm_open,.preclose = ipvr_drm_preclose,.postclose = ipvr_drm_postclose,.irq_handler = ipvr_irq_handler,.set_busid = drm_platform_set_busid,This should be NULL for new drivers.
.gem_free_object = ipvr_gem_free_object, #ifdef+CONFIG_DEBUG_FS
.debugfs_init = ipvr_debugfs_init,.debugfs_cleanup = ipvr_debugfs_cleanup, #endif.ioctls = ipvr_gem_ioctls,.num_ioctls = ARRAY_SIZE(ipvr_gem_ioctls),.fops = &ipvr_fops,.name = IPVR_DRIVER_NAME,.desc = IPVR_DRIVER_DESC,.date = IPVR_DRIVER_DATE,.major = IPVR_DRIVER_MAJOR,.minor = IPVR_DRIVER_MINOR,.patchlevel = IPVR_DRIVER_PATCHLEVEL, };+static int32_t ipvr_plat_probe(struct platform_device *device) {
return drm_platform_init(&ipvr_drm_driver, device);drm_platform_init() should be used. Please use drm_dev_alloc(), drm_dev_register() directly. udl and tegra should serve as example.
Thx, will update it in next version.
+}
+static int32_t ipvr_plat_remove(struct platform_device *device) {
drm_put_dev(platform_get_drvdata(device));Again, please use drm_dev_unregister() + drm_dev_unref() directly.
Thx, will update it in next version.
Thanks David
return 0;+}
+static struct dev_pm_ops ipvr_pm_ops = {
.suspend = ipvr_pm_suspend,.resume = ipvr_pm_resume,.freeze = ipvr_pm_suspend,.thaw = ipvr_pm_resume,.poweroff = ipvr_pm_suspend,.restore = ipvr_pm_resume, #ifdef CONFIG_PM_RUNTIME.runtime_suspend = ipvr_pm_suspend,.runtime_resume = ipvr_pm_resume, #endif };+static struct platform_driver ipvr_plat_driver = {
.driver = {.name = "ipvr-ved",.owner = THIS_MODULE, #ifdef CONFIG_PM.pm = &ipvr_pm_ops, #endif},.probe = ipvr_plat_probe,.remove = ipvr_plat_remove, };+module_platform_driver(ipvr_plat_driver); +MODULE_LICENSE("GPL");
Intel-gfx mailing list Intel-gfx@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/intel-gfx
-- Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Hi
On Thu, Oct 16, 2014 at 3:39 PM, Cheng, Yao yao.cheng@intel.com wrote:
Accepted :) I will update the patch to implement the mmap interface and remove the legacy MMAP_IOCTL. BTW I didn't see a field to get mmap_offset in struct drm_gem_open, I guess something like a new "GET_MMAP_OFFSET_IOCTL" need be added to support mapping flinked/primed BO, is it?
Yes. There is currently no generic interface to get mmap offsets for any gem bo (because some of them may not be mapped from user-space or require extra parameters).
So, either support dumb-buffers (if you don't need any advanced parameters like tiling or cache constraints), or add an MMAP_OFFSET_IOCTL like you proposed (and most drivers do so).
Thanks David
On Thu, Oct 16, 2014 at 03:47:08PM +0200, David Herrmann wrote:
Hi
On Thu, Oct 16, 2014 at 3:39 PM, Cheng, Yao yao.cheng@intel.com wrote:
Accepted :) I will update the patch to implement the mmap interface and remove the legacy MMAP_IOCTL. BTW I didn't see a field to get mmap_offset in struct drm_gem_open, I guess something like a new "GET_MMAP_OFFSET_IOCTL" need be added to support mapping flinked/primed BO, is it?
Yes. There is currently no generic interface to get mmap offsets for any gem bo (because some of them may not be mapped from user-space or require extra parameters).
So, either support dumb-buffers (if you don't need any advanced parameters like tiling or cache constraints), or add an MMAP_OFFSET_IOCTL like you proposed (and most drivers do so).
Dave Airlie _really_ doesn't like it if userspace abuses the dumb buffer interface for rendering. That really should only be used for sw rendering for boot splash and similar. Some armsoc drivers don't really follow that though. -Daniel
On Mon, Oct 13, 2014 at 08:15:00PM +0800, Yao Cheng wrote:
drm/ipvr is a new GEM driver for baytrail's vxd392, which accelerates VP8 video decoding. The driver name "ipvr" means the PowerVR's IP wrapped by Intel. In the future, ipvr may support other platforms such as Merrifield. Code is placed at drivers/gpu/drm/ipvr and the following two new Kconfig are added: CONFIG_DRM_IPVR: Build option for ipvr module CONFIG_DRM_IPVR_EC: Experimental feature of error concealment
User mode drm helper "libdrm_ipvr.so" and simple test are also included.
Yao Cheng (3): [1/3] drm/i915: add vxd392 bridge in i915 on baytrail [2/3] drm/ipvr: ipvr drm driver for vxd392
If this is Intel-specific, why doesn't it live under the i915 driver?
Thierry
On Tue, Oct 14, 2014 at 4:53 AM, Thierry Reding thierry.reding@gmail.com wrote:
On Mon, Oct 13, 2014 at 08:15:00PM +0800, Yao Cheng wrote:
drm/ipvr is a new GEM driver for baytrail's vxd392, which accelerates VP8 video decoding. The driver name "ipvr" means the PowerVR's IP wrapped by Intel. In the future, ipvr may support other platforms such as Merrifield. Code is placed at drivers/gpu/drm/ipvr and the following two new Kconfig are added: CONFIG_DRM_IPVR: Build option for ipvr module CONFIG_DRM_IPVR_EC: Experimental feature of error concealment
User mode drm helper "libdrm_ipvr.so" and simple test are also included.
Yao Cheng (3): [1/3] drm/i915: add vxd392 bridge in i915 on baytrail [2/3] drm/ipvr: ipvr drm driver for vxd392
If this is Intel-specific, why doesn't it live under the i915 driver?
It is an entirely unrelated HW IP block, VXD392. Nothing to do with GEN aside from DRM based.
Sean
Thierry
Intel-gfx mailing list Intel-gfx@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Thx Kelley/Reding for comments, Yes, the vxd392 is a totally different IP and it also exist on Intel's non-GEN platforms such as Merrifield. If we put it inside i915 there'll be issue if we someday supports Merrifield.
-----Original Message----- From: Sean V Kelley [mailto:sean.v.kelley@intel.com] Sent: Tuesday, October 14, 2014 11:51 PM To: Thierry Reding Cc: Cheng, Yao; Vetter, Daniel; intel-gfx@lists.freedesktop.org; Jiang, Fei; DRI mailing list Subject: Re: [Intel-gfx] [RFC PATCH 0/3] drm driver for baytrail's vxd392
On Tue, Oct 14, 2014 at 4:53 AM, Thierry Reding thierry.reding@gmail.com wrote:
On Mon, Oct 13, 2014 at 08:15:00PM +0800, Yao Cheng wrote:
drm/ipvr is a new GEM driver for baytrail's vxd392, which accelerates VP8 video decoding. The driver name "ipvr" means the PowerVR's IP wrapped by Intel. In the future, ipvr may support other platforms such as Merrifield. Code is placed at drivers/gpu/drm/ipvr and the following two new Kconfig are added: CONFIG_DRM_IPVR: Build option for ipvr module CONFIG_DRM_IPVR_EC: Experimental feature of error concealment
User mode drm helper "libdrm_ipvr.so" and simple test are also included.
Yao Cheng (3): [1/3] drm/i915: add vxd392 bridge in i915 on baytrail [2/3] drm/ipvr: ipvr drm driver for vxd392
If this is Intel-specific, why doesn't it live under the i915 driver?
It is an entirely unrelated HW IP block, VXD392. Nothing to do with GEN aside from DRM based.
Sean
Thierry
Intel-gfx mailing list Intel-gfx@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/intel-gfx
-- Sean V. Kelley sean.v.kelley@intel.com Open Source Technology Center / SSG Intel Corp.
On Tue, Oct 14, 2014 at 08:50:35AM -0700, Sean V Kelley wrote:
On Tue, Oct 14, 2014 at 4:53 AM, Thierry Reding thierry.reding@gmail.com wrote:
On Mon, Oct 13, 2014 at 08:15:00PM +0800, Yao Cheng wrote:
drm/ipvr is a new GEM driver for baytrail's vxd392, which accelerates VP8 video decoding. The driver name "ipvr" means the PowerVR's IP wrapped by Intel. In the future, ipvr may support other platforms such as Merrifield. Code is placed at drivers/gpu/drm/ipvr and the following two new Kconfig are added: CONFIG_DRM_IPVR: Build option for ipvr module CONFIG_DRM_IPVR_EC: Experimental feature of error concealment
User mode drm helper "libdrm_ipvr.so" and simple test are also included.
Yao Cheng (3): [1/3] drm/i915: add vxd392 bridge in i915 on baytrail [2/3] drm/ipvr: ipvr drm driver for vxd392
If this is Intel-specific, why doesn't it live under the i915 driver?
It is an entirely unrelated HW IP block, VXD392. Nothing to do with GEN aside from DRM based.
With GEN you're referring to the Intel integrated GPU? And VXD392 I take it is the IP block licensed by Imagination? Baytrail and others then wrap some additional logic around this as it is integrated into the SoC?
How much wrapping actually happens here? I worry that this is going to lead to a lot of duplication if we ever want to support another SoC that uses the VXD392 IP. Could the code be split into a VXD392 "library" and some driver that implements the Intel-specific glue?
Finally, if this IP block is a VP8 video decoding engine only, I'm not sure DRM is the best subsystem for it. Traditionally video decoding has been done primarily in V4L2. I'm not sure that's the best fit given that it was originally designed for video capturing, but they've evolved some infrastructure to deal with encoding/decoding, whereas we have nothing like that at all in DRM.
Thierry
On Wed, Oct 15, 2014 at 1:13 AM, Thierry Reding thierry.reding@gmail.com wrote:
On Tue, Oct 14, 2014 at 08:50:35AM -0700, Sean V Kelley wrote:
On Tue, Oct 14, 2014 at 4:53 AM, Thierry Reding thierry.reding@gmail.com wrote:
On Mon, Oct 13, 2014 at 08:15:00PM +0800, Yao Cheng wrote:
drm/ipvr is a new GEM driver for baytrail's vxd392, which accelerates
VP8 video decoding.
The driver name "ipvr" means the PowerVR's IP wrapped by Intel. In
the future, ipvr may support other platforms such as Merrifield.
Code is placed at drivers/gpu/drm/ipvr and the following two new
Kconfig are added:
CONFIG_DRM_IPVR: Build option for ipvr module CONFIG_DRM_IPVR_EC: Experimental feature of error concealment
User mode drm helper "libdrm_ipvr.so" and simple test are also
included.
Yao Cheng (3): [1/3] drm/i915: add vxd392 bridge in i915 on baytrail [2/3] drm/ipvr: ipvr drm driver for vxd392
If this is Intel-specific, why doesn't it live under the i915 driver?
It is an entirely unrelated HW IP block, VXD392. Nothing to do with GEN aside from DRM based.
With GEN you're referring to the Intel integrated GPU? And VXD392 I take it is the IP block licensed by Imagination? Baytrail and others then wrap some additional logic around this as it is integrated into the SoC?
How much wrapping actually happens here? I worry that this is going to lead to a lot of duplication if we ever want to support another SoC that uses the VXD392 IP. Could the code be split into a VXD392 "library" and some driver that implements the Intel-specific glue?
Finally, if this IP block is a VP8 video decoding engine only, I'm not sure DRM is the best subsystem for it. Traditionally video decoding has been done primarily in V4L2. I'm not sure that's the best fit given that it was originally designed for video capturing, but they've evolved some infrastructure to deal with encoding/decoding, whereas we have nothing like that at all in DRM.
That isn't true. i915, nouveau and radeon drm drivers all support video decoding user space in some form.
Stéphane
On Wed, Oct 15, 2014 at 01:18:02AM -0700, Stéphane Marchesin wrote:
On Wed, Oct 15, 2014 at 1:13 AM, Thierry Reding thierry.reding@gmail.com wrote:
On Tue, Oct 14, 2014 at 08:50:35AM -0700, Sean V Kelley wrote:
On Tue, Oct 14, 2014 at 4:53 AM, Thierry Reding thierry.reding@gmail.com wrote:
On Mon, Oct 13, 2014 at 08:15:00PM +0800, Yao Cheng wrote:
drm/ipvr is a new GEM driver for baytrail's vxd392, which accelerates
VP8 video decoding.
The driver name "ipvr" means the PowerVR's IP wrapped by Intel. In
the future, ipvr may support other platforms such as Merrifield.
Code is placed at drivers/gpu/drm/ipvr and the following two new
Kconfig are added:
CONFIG_DRM_IPVR: Build option for ipvr module CONFIG_DRM_IPVR_EC: Experimental feature of error concealment
User mode drm helper "libdrm_ipvr.so" and simple test are also
included.
Yao Cheng (3): [1/3] drm/i915: add vxd392 bridge in i915 on baytrail [2/3] drm/ipvr: ipvr drm driver for vxd392
If this is Intel-specific, why doesn't it live under the i915 driver?
It is an entirely unrelated HW IP block, VXD392. Nothing to do with GEN aside from DRM based.
With GEN you're referring to the Intel integrated GPU? And VXD392 I take it is the IP block licensed by Imagination? Baytrail and others then wrap some additional logic around this as it is integrated into the SoC?
How much wrapping actually happens here? I worry that this is going to lead to a lot of duplication if we ever want to support another SoC that uses the VXD392 IP. Could the code be split into a VXD392 "library" and some driver that implements the Intel-specific glue?
Finally, if this IP block is a VP8 video decoding engine only, I'm not sure DRM is the best subsystem for it. Traditionally video decoding has been done primarily in V4L2. I'm not sure that's the best fit given that it was originally designed for video capturing, but they've evolved some infrastructure to deal with encoding/decoding, whereas we have nothing like that at all in DRM.
That isn't true. i915, nouveau and radeon drm drivers all support video decoding user space in some form.
Well yes, because they are GPUs that happen to have video decoding engines on the same board and use the same method of command-stream submission that they use for 2D or 3D work.
For standalone video decoding hardware, the only drivers that I know of are in drivers/media.
Thierry
On Wed, Oct 15, 2014 at 10:13:15AM +0200, Thierry Reding wrote:
On Tue, Oct 14, 2014 at 08:50:35AM -0700, Sean V Kelley wrote:
On Tue, Oct 14, 2014 at 4:53 AM, Thierry Reding thierry.reding@gmail.com wrote:
On Mon, Oct 13, 2014 at 08:15:00PM +0800, Yao Cheng wrote:
drm/ipvr is a new GEM driver for baytrail's vxd392, which accelerates VP8 video decoding. The driver name "ipvr" means the PowerVR's IP wrapped by Intel. In the future, ipvr may support other platforms such as Merrifield. Code is placed at drivers/gpu/drm/ipvr and the following two new Kconfig are added: CONFIG_DRM_IPVR: Build option for ipvr module CONFIG_DRM_IPVR_EC: Experimental feature of error concealment
User mode drm helper "libdrm_ipvr.so" and simple test are also included.
Yao Cheng (3): [1/3] drm/i915: add vxd392 bridge in i915 on baytrail [2/3] drm/ipvr: ipvr drm driver for vxd392
If this is Intel-specific, why doesn't it live under the i915 driver?
It is an entirely unrelated HW IP block, VXD392. Nothing to do with GEN aside from DRM based.
With GEN you're referring to the Intel integrated GPU? And VXD392 I take it is the IP block licensed by Imagination? Baytrail and others then wrap some additional logic around this as it is integrated into the SoC?
How much wrapping actually happens here? I worry that this is going to lead to a lot of duplication if we ever want to support another SoC that uses the VXD392 IP. Could the code be split into a VXD392 "library" and some driver that implements the Intel-specific glue?
Well if someone cares for it the driver ioctl interface should be portable. But without anyone else chiming in I honestly don't care. We did try to design the entire beast like that though, and it should pretty much work with few changes on gma500 (if anyone bothers).
Finally, if this IP block is a VP8 video decoding engine only, I'm not sure DRM is the best subsystem for it. Traditionally video decoding has been done primarily in V4L2. I'm not sure that's the best fit given that it was originally designed for video capturing, but they've evolved some infrastructure to deal with encoding/decoding, whereas we have nothing like that at all in DRM.
Shuffling massive piles of state submission code into the kernel is imo not a sane design. So imo the sane way is to put encode/decode drivers into drm, and the kernel part just has minimal validation for security (if required at all if the iommu isn't good enough). The common interface for applications would then be gstreamer (since we don't even have a cross-vendor video api with at least libva and vdpau having rather decent open drivers available). -Daniel
On Wed, Oct 15, 2014 at 4:13 AM, Thierry Reding thierry.reding@gmail.com wrote:
Finally, if this IP block is a VP8 video decoding engine only, I'm not sure DRM is the best subsystem for it. Traditionally video decoding has been done primarily in V4L2. I'm not sure that's the best fit given that it was originally designed for video capturing, but they've evolved some infrastructure to deal with encoding/decoding, whereas we have nothing like that at all in DRM.
v4l maybe works for some vid dec/enc hw if the hw does enough for you.. the common ioctl approach might work for some hw, but I think not really all. For the same reason we don't try to standardize the 3d/2d cmd submission ioctls across drivers..
BR, -R
The reason why we use DRM interface is mainly we want to use GEM for video memory manager.
Thanks, Fei -----Original Message----- From: Rob Clark [mailto:robdclark@gmail.com] Sent: Friday, October 17, 2014 2:43 AM To: Thierry Reding Cc: Kelley, Sean V; Vetter, Daniel; intel-gfx@lists.freedesktop.org; DRI mailing list; Jiang, Fei Subject: Re: [Intel-gfx] [RFC PATCH 0/3] drm driver for baytrail's vxd392
On Wed, Oct 15, 2014 at 4:13 AM, Thierry Reding thierry.reding@gmail.com wrote:
Finally, if this IP block is a VP8 video decoding engine only, I'm not sure DRM is the best subsystem for it. Traditionally video decoding has been done primarily in V4L2. I'm not sure that's the best fit given that it was originally designed for video capturing, but they've evolved some infrastructure to deal with encoding/decoding, whereas we have nothing like that at all in DRM.
v4l maybe works for some vid dec/enc hw if the hw does enough for you.. the common ioctl approach might work for some hw, but I think not really all. For the same reason we don't try to standardize the 3d/2d cmd submission ioctls across drivers..
BR, -R
On Mon, Oct 13, 2014 at 08:15:00PM +0800, Yao Cheng wrote:
drm/ipvr is a new GEM driver for baytrail's vxd392, which accelerates VP8 video decoding. The driver name "ipvr" means the PowerVR's IP wrapped by Intel. In the future, ipvr may support other platforms such as Merrifield. Code is placed at drivers/gpu/drm/ipvr and the following two new Kconfig are added: CONFIG_DRM_IPVR: Build option for ipvr module CONFIG_DRM_IPVR_EC: Experimental feature of error concealment
User mode drm helper "libdrm_ipvr.so" and simple test are also included.
git repos/patches for the userspace side seems to be missing. -Daniel
Hi Daniel, this patch [RFC PATCH 3/3] libdrm: user mode helper for ipvr drm driver adds the ipvr part into libdrm. Do you mean this patch is missing?
-----Original Message----- From: Daniel Vetter [mailto:daniel.vetter@ffwll.ch] On Behalf Of Daniel Vetter Sent: Thursday, October 16, 2014 7:46 PM To: Cheng, Yao Cc: intel-gfx@lists.freedesktop.org; Jiang, Fei; dri- devel@lists.freedesktop.org; Vetter, Daniel Subject: Re: [Intel-gfx] [RFC PATCH 0/3] drm driver for baytrail's vxd392
On Mon, Oct 13, 2014 at 08:15:00PM +0800, Yao Cheng wrote:
drm/ipvr is a new GEM driver for baytrail's vxd392, which accelerates VP8
video decoding.
The driver name "ipvr" means the PowerVR's IP wrapped by Intel. In the
future, ipvr may support other platforms such as Merrifield.
Code is placed at drivers/gpu/drm/ipvr and the following two new Kconfig
are added:
CONFIG_DRM_IPVR: Build option for ipvr module CONFIG_DRM_IPVR_EC: Experimental feature of error concealment
User mode drm helper "libdrm_ipvr.so" and simple test are also included.
git repos/patches for the userspace side seems to be missing.
-Daniel
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
On Thu, Oct 16, 2014 at 7:36 AM, Cheng, Yao yao.cheng@intel.com wrote:
Hi Daniel, this patch [RFC PATCH 3/3] libdrm: user mode helper for ipvr drm driver adds the ipvr part into libdrm. Do you mean this patch is missing?
The umd (ipvr_vaapi_driver) is needed to fully exercise the APIs exposed. That code is in flight and I'll work with you to get it into the repos.
Sean
-----Original Message----- From: Daniel Vetter [mailto:daniel.vetter@ffwll.ch] On Behalf Of Daniel Vetter Sent: Thursday, October 16, 2014 7:46 PM To: Cheng, Yao Cc: intel-gfx@lists.freedesktop.org; Jiang, Fei; dri- devel@lists.freedesktop.org; Vetter, Daniel Subject: Re: [Intel-gfx] [RFC PATCH 0/3] drm driver for baytrail's vxd392
On Mon, Oct 13, 2014 at 08:15:00PM +0800, Yao Cheng wrote:
drm/ipvr is a new GEM driver for baytrail's vxd392, which accelerates VP8
video decoding.
The driver name "ipvr" means the PowerVR's IP wrapped by Intel. In the
future, ipvr may support other platforms such as Merrifield.
Code is placed at drivers/gpu/drm/ipvr and the following two new Kconfig
are added:
CONFIG_DRM_IPVR: Build option for ipvr module CONFIG_DRM_IPVR_EC: Experimental feature of error concealment
User mode drm helper "libdrm_ipvr.so" and simple test are also included.
git repos/patches for the userspace side seems to be missing.
-Daniel
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
dri-devel mailing list dri-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/dri-devel
dri-devel@lists.freedesktop.org
-
Cheng, Yao -
Daniel Vetter -
David Herrmann -
Jani Nikula -
Jiang, Fei -
Rob Clark -
Sean V Kelley -
Stéphane Marchesin -
Thierry Reding -
Yao Cheng