This is to support cached and cached-coherent memory types in vulkan.
v2: - added patches 2/3 to enable using dma_ops_bypass - changed DRM_MSM_GEM_SYNC_CACHE patch to use dma_sync_sg_for_device() and dma_sync_sg_for_cpu(), and renamed sync flags.
Not sure I did the right thing with for the dma_ops_bypass part, this is what I came up with reading the emails.
Jonathan Marek (5): drm/msm: add MSM_BO_CACHED_COHERENT dma-direct: add dma_direct_bypass() to force direct ops drm/msm: call dma_direct_bypass() drm/msm: add DRM_MSM_GEM_SYNC_CACHE for non-coherent cache maintenance drm/msm: bump up the uapi version
drivers/gpu/drm/msm/Kconfig | 1 + drivers/gpu/drm/msm/adreno/adreno_device.c | 1 + drivers/gpu/drm/msm/msm_drv.c | 32 +++++++++++++++++++--- drivers/gpu/drm/msm/msm_drv.h | 3 ++ drivers/gpu/drm/msm/msm_gem.c | 31 +++++++++++++++++++++ include/linux/dma-direct.h | 9 ++++++ include/uapi/drm/msm_drm.h | 25 +++++++++++++++-- kernel/dma/direct.c | 23 ++++++++++++++++ 8 files changed, 118 insertions(+), 7 deletions(-)
Add a new cache mode for creating coherent host-cached BOs.
Signed-off-by: Jonathan Marek jonathan@marek.ca Reviewed-by: Jordan Crouse jcrouse@codeaurora.org --- drivers/gpu/drm/msm/adreno/adreno_device.c | 1 + drivers/gpu/drm/msm/msm_drv.h | 1 + drivers/gpu/drm/msm/msm_gem.c | 8 ++++++++ include/uapi/drm/msm_drm.h | 5 ++--- 4 files changed, 12 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/msm/adreno/adreno_device.c b/drivers/gpu/drm/msm/adreno/adreno_device.c index 58e03b20e1c7..21c9bc954f38 100644 --- a/drivers/gpu/drm/msm/adreno/adreno_device.c +++ b/drivers/gpu/drm/msm/adreno/adreno_device.c @@ -410,6 +410,7 @@ static int adreno_bind(struct device *dev, struct device *master, void *data) config.rev.minor, config.rev.patchid);
priv->is_a2xx = config.rev.core == 2; + priv->has_cached_coherent = config.rev.core >= 6;
gpu = info->init(drm); if (IS_ERR(gpu)) { diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h index f33281ac7913..22ebecb28349 100644 --- a/drivers/gpu/drm/msm/msm_drv.h +++ b/drivers/gpu/drm/msm/msm_drv.h @@ -168,6 +168,7 @@ struct msm_drm_private { struct msm_file_private *lastctx; /* gpu is only set on open(), but we need this info earlier */ bool is_a2xx; + bool has_cached_coherent;
struct drm_fb_helper *fbdev;
diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index 04be4cfcccc1..3d8254b5de16 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -420,6 +420,9 @@ static int msm_gem_pin_iova(struct drm_gem_object *obj, if (msm_obj->flags & MSM_BO_MAP_PRIV) prot |= IOMMU_PRIV;
+ if (msm_obj->flags & MSM_BO_CACHED_COHERENT) + prot |= IOMMU_CACHE; + WARN_ON(!mutex_is_locked(&msm_obj->lock));
if (WARN_ON(msm_obj->madv != MSM_MADV_WILLNEED)) @@ -1004,6 +1007,7 @@ static int msm_gem_new_impl(struct drm_device *dev, uint32_t size, uint32_t flags, struct drm_gem_object **obj) { + struct msm_drm_private *priv = dev->dev_private; struct msm_gem_object *msm_obj;
switch (flags & MSM_BO_CACHE_MASK) { @@ -1011,6 +1015,10 @@ static int msm_gem_new_impl(struct drm_device *dev, case MSM_BO_CACHED: case MSM_BO_WC: break; + case MSM_BO_CACHED_COHERENT: + if (priv->has_cached_coherent) + break; + /* fallthrough */ default: DRM_DEV_ERROR(dev->dev, "invalid cache flag: %x\n", (flags & MSM_BO_CACHE_MASK)); diff --git a/include/uapi/drm/msm_drm.h b/include/uapi/drm/msm_drm.h index a6c1f3eb2623..474497e8743a 100644 --- a/include/uapi/drm/msm_drm.h +++ b/include/uapi/drm/msm_drm.h @@ -94,12 +94,11 @@ struct drm_msm_param { #define MSM_BO_CACHED 0x00010000 #define MSM_BO_WC 0x00020000 #define MSM_BO_UNCACHED 0x00040000 +#define MSM_BO_CACHED_COHERENT 0x080000
#define MSM_BO_FLAGS (MSM_BO_SCANOUT | \ MSM_BO_GPU_READONLY | \ - MSM_BO_CACHED | \ - MSM_BO_WC | \ - MSM_BO_UNCACHED) + MSM_BO_CACHE_MASK)
struct drm_msm_gem_new { __u64 size; /* in */
Always use direct dma ops and no swiotlb.
Note: arm-smmu-qcom already avoids creating iommu dma ops, but not everything uses arm-smmu-qcom and this also sets the dma mask.
Signed-off-by: Jonathan Marek jonathan@marek.ca --- drivers/gpu/drm/msm/Kconfig | 1 + drivers/gpu/drm/msm/msm_drv.c | 8 +++++--- 2 files changed, 6 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/msm/Kconfig b/drivers/gpu/drm/msm/Kconfig index e5816b498494..07c50405970a 100644 --- a/drivers/gpu/drm/msm/Kconfig +++ b/drivers/gpu/drm/msm/Kconfig @@ -20,6 +20,7 @@ config DRM_MSM select SND_SOC_HDMI_CODEC if SND_SOC select SYNC_FILE select PM_OPP + select DMA_OPS_BYPASS help DRM/KMS driver for MSM/snapdragon.
diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c index 49685571dc0e..bae48afca82e 100644 --- a/drivers/gpu/drm/msm/msm_drv.c +++ b/drivers/gpu/drm/msm/msm_drv.c @@ -6,6 +6,7 @@ */
#include <linux/dma-mapping.h> +#include <linux/dma-direct.h> #include <linux/kthread.h> #include <linux/uaccess.h> #include <uapi/linux/sched/types.h> @@ -1288,10 +1289,11 @@ static int msm_pdev_probe(struct platform_device *pdev) if (ret) goto fail;
- /* on all devices that I am aware of, iommu's which can map - * any address the cpu can see are used: + /* always use direct dma ops and no swiotlb + * note: arm-smmu-qcom already avoids creating iommu dma ops, but + * not everything uses arm-smmu-qcom and this also sets the dma mask */ - ret = dma_set_mask_and_coherent(&pdev->dev, ~0); + ret = dma_direct_bypass(&pdev->dev); if (ret) goto fail;
This makes it possible to use the non-coherent cached MSM_BO_CACHED mode, which otherwise doesn't provide any method for cleaning/invalidating the cache to sync with the device.
Signed-off-by: Jonathan Marek jonathan@marek.ca --- drivers/gpu/drm/msm/msm_drv.c | 21 +++++++++++++++++++++ drivers/gpu/drm/msm/msm_drv.h | 2 ++ drivers/gpu/drm/msm/msm_gem.c | 23 +++++++++++++++++++++++ include/uapi/drm/msm_drm.h | 20 ++++++++++++++++++++ 4 files changed, 66 insertions(+)
diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c index bae48afca82e..3f17acdf6594 100644 --- a/drivers/gpu/drm/msm/msm_drv.c +++ b/drivers/gpu/drm/msm/msm_drv.c @@ -959,6 +959,26 @@ static int msm_ioctl_submitqueue_close(struct drm_device *dev, void *data, return msm_submitqueue_remove(file->driver_priv, id); }
+static int msm_ioctl_gem_sync_cache(struct drm_device *dev, void *data, + struct drm_file *file) +{ + struct drm_msm_gem_sync_cache *args = data; + struct drm_gem_object *obj; + + if (args->flags & ~MSM_GEM_SYNC_CACHE_FLAGS) + return -EINVAL; + + obj = drm_gem_object_lookup(file, args->handle); + if (!obj) + return -ENOENT; + + msm_gem_sync_cache(obj, args->flags, args->offset, args->end); + + drm_gem_object_put(obj); + + return 0; +} + static const struct drm_ioctl_desc msm_ioctls[] = { DRM_IOCTL_DEF_DRV(MSM_GET_PARAM, msm_ioctl_get_param, DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(MSM_GEM_NEW, msm_ioctl_gem_new, DRM_RENDER_ALLOW), @@ -971,6 +991,7 @@ static const struct drm_ioctl_desc msm_ioctls[] = { DRM_IOCTL_DEF_DRV(MSM_SUBMITQUEUE_NEW, msm_ioctl_submitqueue_new, DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(MSM_SUBMITQUEUE_CLOSE, msm_ioctl_submitqueue_close, DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(MSM_SUBMITQUEUE_QUERY, msm_ioctl_submitqueue_query, DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(MSM_GEM_SYNC_CACHE, msm_ioctl_gem_sync_cache, DRM_RENDER_ALLOW), };
static const struct vm_operations_struct vm_ops = { diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h index 22ebecb28349..f170f843010e 100644 --- a/drivers/gpu/drm/msm/msm_drv.h +++ b/drivers/gpu/drm/msm/msm_drv.h @@ -318,6 +318,8 @@ void msm_gem_active_get(struct drm_gem_object *obj, struct msm_gpu *gpu); void msm_gem_active_put(struct drm_gem_object *obj); int msm_gem_cpu_prep(struct drm_gem_object *obj, uint32_t op, ktime_t *timeout); int msm_gem_cpu_fini(struct drm_gem_object *obj); +void msm_gem_sync_cache(struct drm_gem_object *obj, uint32_t flags, + size_t range_start, size_t range_end); void msm_gem_free_object(struct drm_gem_object *obj); int msm_gem_new_handle(struct drm_device *dev, struct drm_file *file, uint32_t size, uint32_t flags, uint32_t *handle, char *name); diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index 3d8254b5de16..039738696f9a 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -797,6 +797,29 @@ int msm_gem_cpu_fini(struct drm_gem_object *obj) return 0; }
+void msm_gem_sync_cache(struct drm_gem_object *obj, uint32_t flags, + size_t range_start, size_t range_end) +{ + struct msm_gem_object *msm_obj = to_msm_bo(obj); + struct device *dev = msm_obj->base.dev->dev; + + /* exit early if get_pages() hasn't been called yet */ + if (!msm_obj->pages) + return; + + /* TODO: sync only the specified range */ + + if (flags & MSM_GEM_SYNC_FOR_DEVICE) { + dma_sync_sg_for_device(dev, msm_obj->sgt->sgl, + msm_obj->sgt->nents, DMA_TO_DEVICE); + } + + if (flags & MSM_GEM_SYNC_FOR_CPU) { + dma_sync_sg_for_cpu(dev, msm_obj->sgt->sgl, + msm_obj->sgt->nents, DMA_FROM_DEVICE); + } +} + #ifdef CONFIG_DEBUG_FS static void describe_fence(struct dma_fence *fence, const char *type, struct seq_file *m) diff --git a/include/uapi/drm/msm_drm.h b/include/uapi/drm/msm_drm.h index 474497e8743a..c8288f328528 100644 --- a/include/uapi/drm/msm_drm.h +++ b/include/uapi/drm/msm_drm.h @@ -319,6 +319,24 @@ struct drm_msm_submitqueue_query { __u32 pad; };
+/* + * Host cache maintenance (relevant for MSM_BO_CACHED) + * driver may both clean/invalidate (flush) for clean + */ + +#define MSM_GEM_SYNC_FOR_DEVICE 0x1 +#define MSM_GEM_SYNC_FOR_CPU 0x2 + +#define MSM_GEM_SYNC_CACHE_FLAGS (MSM_GEM_SYNC_FOR_DEVICE | \ + MSM_GEM_SYNC_FOR_CPU) + +struct drm_msm_gem_sync_cache { + __u32 handle; + __u32 flags; + __u64 offset; + __u64 end; /* offset + size */ +}; + #define DRM_MSM_GET_PARAM 0x00 /* placeholder: #define DRM_MSM_SET_PARAM 0x01 @@ -336,6 +354,7 @@ struct drm_msm_submitqueue_query { #define DRM_MSM_SUBMITQUEUE_NEW 0x0A #define DRM_MSM_SUBMITQUEUE_CLOSE 0x0B #define DRM_MSM_SUBMITQUEUE_QUERY 0x0C +#define DRM_MSM_GEM_SYNC_CACHE 0x0D
#define DRM_IOCTL_MSM_GET_PARAM DRM_IOWR(DRM_COMMAND_BASE + DRM_MSM_GET_PARAM, struct drm_msm_param) #define DRM_IOCTL_MSM_GEM_NEW DRM_IOWR(DRM_COMMAND_BASE + DRM_MSM_GEM_NEW, struct drm_msm_gem_new) @@ -348,6 +367,7 @@ struct drm_msm_submitqueue_query { #define DRM_IOCTL_MSM_SUBMITQUEUE_NEW DRM_IOWR(DRM_COMMAND_BASE + DRM_MSM_SUBMITQUEUE_NEW, struct drm_msm_submitqueue) #define DRM_IOCTL_MSM_SUBMITQUEUE_CLOSE DRM_IOW (DRM_COMMAND_BASE + DRM_MSM_SUBMITQUEUE_CLOSE, __u32) #define DRM_IOCTL_MSM_SUBMITQUEUE_QUERY DRM_IOW (DRM_COMMAND_BASE + DRM_MSM_SUBMITQUEUE_QUERY, struct drm_msm_submitqueue_query) +#define DRM_IOCTL_MSM_GEM_SYNC_CACHE DRM_IOW (DRM_COMMAND_BASE + DRM_MSM_GEM_SYNC_CACHE, struct drm_msm_gem_sync_cache)
#if defined(__cplusplus) }
Increase the minor version to indicate the presence of new features.
Signed-off-by: Jonathan Marek jonathan@marek.ca --- drivers/gpu/drm/msm/msm_drv.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c index 3f17acdf6594..7230d3c0eee5 100644 --- a/drivers/gpu/drm/msm/msm_drv.c +++ b/drivers/gpu/drm/msm/msm_drv.c @@ -39,9 +39,10 @@ * GEM object's debug name * - 1.5.0 - Add SUBMITQUERY_QUERY ioctl * - 1.6.0 - Syncobj support + * - 1.7.0 - MSM_BO_CACHED_COHERENT and DRM_IOCTL_MSM_GEM_SYNC_CACHE */ #define MSM_VERSION_MAJOR 1 -#define MSM_VERSION_MINOR 6 +#define MSM_VERSION_MINOR 7 #define MSM_VERSION_PATCHLEVEL 0
static const struct drm_mode_config_funcs mode_config_funcs = {
dri-devel@lists.freedesktop.org
-
Jonathan Marek