Currently the behavior of a command stream is provided by the user application during submission and the application is expected to internally maintain the settings for each 'context' or 'rendering queue' and specify the correct ones.

This works okay for simple cases but as applications become more complex we will want to set context specific flags and do various permission checks to allow certain contexts to enable additional privileges.

Add kernel-side submit queues to be analogous to 'contexts' or 'rendering queues' on the application side. Each file descriptor instance will maintain its own list of queues. Queues cannot be shared between file descriptors.

For backwards compatibility context id '0' is defined as a default context specifying no priority and no special flags. This is intended to be the usual configuration for 99% of applications so that a garden variety application can function correctly without creating a queue. Only those applications requiring the specific benefit of different queues need create one.

Signed-off-by: Jordan Crouse jcrouse@codeaurora.org --- drivers/gpu/drm/msm/Makefile | 3 +- drivers/gpu/drm/msm/msm_drv.c | 56 ++++++++++++-- drivers/gpu/drm/msm/msm_drv.h | 20 +++-- drivers/gpu/drm/msm/msm_gem.h | 1 + drivers/gpu/drm/msm/msm_gem_submit.c | 14 +++- drivers/gpu/drm/msm/msm_gpu.h | 15 ++++ drivers/gpu/drm/msm/msm_submitqueue.c | 135 ++++++++++++++++++++++++++++++++++ include/uapi/drm/msm_drm.h | 22 ++++++ 8 files changed, 250 insertions(+), 16 deletions(-) create mode 100644 drivers/gpu/drm/msm/msm_submitqueue.c

diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile index 33008fa..3c234e7 100644 --- a/drivers/gpu/drm/msm/Makefile +++ b/drivers/gpu/drm/msm/Makefile @@ -57,7 +57,8 @@ msm-y := \ msm_iommu.o \ msm_perf.o \ msm_rd.o \ - msm_ringbuffer.o + msm_ringbuffer.o \ + msm_submitqueue.o

msm-$(CONFIG_DRM_FBDEV_EMULATION) += msm_fbdev.o msm-$(CONFIG_COMMON_CLK) += mdp/mdp4/mdp4_lvds_pll.o diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c index dda55b6..cf743c7 100644 --- a/drivers/gpu/drm/msm/msm_drv.c +++ b/drivers/gpu/drm/msm/msm_drv.c @@ -29,9 +29,12 @@ * - 1.0.0 - initial interface * - 1.1.0 - adds madvise, and support for submits with > 4 cmd buffers * - 1.2.0 - adds explicit fence support for submit ioctl + * - 1.3.0 - adds GMEM_BASE + NR_RINGS params, SUBMITQUEUE_NEW + + * SUBMITQUEUE_CLOSE ioctls, and MSM_INFO_IOVA flag for + * MSM_GEM_INFO ioctl. */ #define MSM_VERSION_MAJOR 1 -#define MSM_VERSION_MINOR 2 +#define MSM_VERSION_MINOR 3 #define MSM_VERSION_PATCHLEVEL 0

static void msm_fb_output_poll_changed(struct drm_device *dev) @@ -514,24 +517,37 @@ static void load_gpu(struct drm_device *dev) mutex_unlock(&init_lock); }

-static int msm_open(struct drm_device *dev, struct drm_file *file) +static int context_init(struct drm_file *file) { struct msm_file_private *ctx;

- /* For now, load gpu on open.. to avoid the requirement of having - * firmware in the initrd. - */ - load_gpu(dev); - ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); if (!ctx) return -ENOMEM;

+ msm_submitqueue_init(ctx); + file->driver_priv = ctx;

return 0; }

+static int msm_open(struct drm_device *dev, struct drm_file *file) +{ + /* For now, load gpu on open.. to avoid the requirement of having + * firmware in the initrd. + */ + load_gpu(dev); + + return context_init(file); +} + +static void context_close(struct msm_file_private *ctx) +{ + msm_submitqueue_close(ctx); + kfree(ctx); +} + static void msm_postclose(struct drm_device *dev, struct drm_file *file) { struct msm_drm_private *priv = dev->dev_private; @@ -542,7 +558,7 @@ static void msm_postclose(struct drm_device *dev, struct drm_file *file) priv->lastctx = NULL; mutex_unlock(&dev->struct_mutex);

- kfree(ctx); + context_close(ctx); }

static void msm_lastclose(struct drm_device *dev) @@ -787,6 +803,28 @@ static int msm_ioctl_gem_madvise(struct drm_device *dev, void *data, return ret; }

+ +static int msm_ioctl_submitqueue_new(struct drm_device *dev, void *data, + struct drm_file *file) +{ + struct drm_msm_submitqueue *args = data; + + if (args->flags & ~MSM_SUBMITQUEUE_FLAGS) + return -EINVAL; + + return msm_submitqueue_create(file->driver_priv, args->prio, + args->flags, &args->id); +} + + +static int msm_ioctl_submitqueue_close(struct drm_device *dev, void *data, + struct drm_file *file) +{ + u32 id = *(u32 *) data; + + return msm_submitqueue_remove(file->driver_priv, id); +} + static const struct drm_ioctl_desc msm_ioctls[] = { DRM_IOCTL_DEF_DRV(MSM_GET_PARAM, msm_ioctl_get_param, DRM_AUTH|DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(MSM_GEM_NEW, msm_ioctl_gem_new, DRM_AUTH|DRM_RENDER_ALLOW), @@ -796,6 +834,8 @@ static int msm_ioctl_gem_madvise(struct drm_device *dev, void *data, DRM_IOCTL_DEF_DRV(MSM_GEM_SUBMIT, msm_ioctl_gem_submit, DRM_AUTH|DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(MSM_WAIT_FENCE, msm_ioctl_wait_fence, DRM_AUTH|DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(MSM_GEM_MADVISE, msm_ioctl_gem_madvise, DRM_AUTH|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(MSM_SUBMITQUEUE_NEW, msm_ioctl_submitqueue_new, DRM_AUTH|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(MSM_SUBMITQUEUE_CLOSE, msm_ioctl_submitqueue_close, DRM_AUTH|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 5e8109c..b3b8f20 100644 --- a/drivers/gpu/drm/msm/msm_drv.h +++ b/drivers/gpu/drm/msm/msm_drv.h @@ -56,11 +56,9 @@ struct msm_gem_vma;

struct msm_file_private { - /* currently we don't do anything useful with this.. but when - * per-context address spaces are supported we'd keep track of - * the context's page-tables here. - */ - int dummy; + rwlock_t queuelock; + struct list_head submitqueues; + int queueid; };

enum msm_mdp_plane_property { @@ -319,6 +317,18 @@ void __iomem *msm_ioremap(struct platform_device *pdev, const char *name, void msm_writel(u32 data, void __iomem *addr); u32 msm_readl(const void __iomem *addr);

+struct msm_gpu_submitqueue; +int msm_submitqueue_init(struct msm_file_private *ctx); +struct msm_gpu_submitqueue *msm_submitqueue_get(struct msm_file_private *ctx, + u32 id); +int msm_submitqueue_create(struct msm_file_private *ctx, u32 prio, + u32 flags, u32 *id); +int msm_submitqueue_remove(struct msm_file_private *ctx, u32 id); +void msm_submitqueue_close(struct msm_file_private *ctx); + +void msm_submitqueue_destroy(struct kref *kref); + + #define DBG(fmt, ...) DRM_DEBUG_DRIVER(fmt"\n", ##__VA_ARGS__) #define VERB(fmt, ...) if (0) DRM_DEBUG_DRIVER(fmt"\n", ##__VA_ARGS__)

diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h index 91c210d..17f8a6c 100644 --- a/drivers/gpu/drm/msm/msm_gem.h +++ b/drivers/gpu/drm/msm/msm_gem.h @@ -142,6 +142,7 @@ struct msm_gem_submit { struct list_head bo_list; struct ww_acquire_ctx ticket; struct dma_fence *fence; + struct msm_gpu_submitqueue *queue; struct pid *pid; /* submitting process */ bool valid; /* true if no cmdstream patching needed */ unsigned int nr_cmds; diff --git a/drivers/gpu/drm/msm/msm_gem_submit.c b/drivers/gpu/drm/msm/msm_gem_submit.c index 8a75c0b..0973884 100644 --- a/drivers/gpu/drm/msm/msm_gem_submit.c +++ b/drivers/gpu/drm/msm/msm_gem_submit.c @@ -31,7 +31,8 @@ #define BO_PINNED 0x2000

static struct msm_gem_submit *submit_create(struct drm_device *dev, - struct msm_gpu *gpu, uint32_t nr_bos, uint32_t nr_cmds) + struct msm_gpu *gpu, struct msm_gpu_submitqueue *queue, + uint32_t nr_bos, uint32_t nr_cmds) { struct msm_gem_submit *submit; uint64_t sz = sizeof(*submit) + ((u64)nr_bos * sizeof(submit->bos[0])) + @@ -49,6 +50,7 @@ static struct msm_gem_submit *submit_create(struct drm_device *dev, submit->fence = NULL; submit->pid = get_pid(task_pid(current)); submit->cmd = (void *)&submit->bos[nr_bos]; + submit->queue = queue;

/* initially, until copy_from_user() and bo lookup succeeds: */ submit->nr_bos = 0; @@ -66,6 +68,8 @@ void msm_gem_submit_free(struct msm_gem_submit *submit) dma_fence_put(submit->fence); list_del(&submit->node); put_pid(submit->pid); + msm_submitqueue_put(submit->queue); + kfree(submit); }

@@ -391,6 +395,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, struct msm_gpu *gpu = priv->gpu; struct dma_fence *in_fence = NULL; struct sync_file *sync_file = NULL; + struct msm_gpu_submitqueue *queue; int out_fence_fd = -1; unsigned i; int ret; @@ -407,6 +412,10 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, if (MSM_PIPE_FLAGS(args->flags) & ~MSM_SUBMIT_FLAGS) return -EINVAL;

+ queue = msm_submitqueue_get(ctx, args->queueid); + if (!queue) + return -ENOENT; + if (args->flags & MSM_SUBMIT_FENCE_FD_IN) { in_fence = sync_file_get_fence(args->fence_fd);

@@ -437,7 +446,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, } priv->struct_mutex_task = current;

- submit = submit_create(dev, gpu, args->nr_bos, args->nr_cmds); + submit = submit_create(dev, gpu, queue, args->nr_bos, args->nr_cmds); if (!submit) { ret = -ENOMEM; goto out_unlock; @@ -523,6 +532,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, submit->nr_cmds = i;

submit->fence = msm_fence_alloc(gpu->fctx); + if (IS_ERR(submit->fence)) { ret = PTR_ERR(submit->fence); submit->fence = NULL; diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h index df4e277..a890176 100644 --- a/drivers/gpu/drm/msm/msm_gpu.h +++ b/drivers/gpu/drm/msm/msm_gpu.h @@ -150,6 +150,15 @@ struct msm_gpu_perfcntr { const char *name; };

+struct msm_gpu_submitqueue { + int id; + u32 flags; + u32 prio; + int faults; + struct list_head node; + struct kref ref; +}; + static inline void gpu_write(struct msm_gpu *gpu, u32 reg, u32 data) { msm_writel(data, gpu->mmio + (reg << 2)); @@ -223,4 +232,10 @@ int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev, void __init adreno_register(void); void __exit adreno_unregister(void);

+static inline void msm_submitqueue_put(struct msm_gpu_submitqueue *queue) +{ + if (queue) + kref_put(&queue->ref, msm_submitqueue_destroy); +} + #endif /* __MSM_GPU_H__ */ diff --git a/drivers/gpu/drm/msm/msm_submitqueue.c b/drivers/gpu/drm/msm/msm_submitqueue.c new file mode 100644 index 0000000..593c3b5 --- /dev/null +++ b/drivers/gpu/drm/msm/msm_submitqueue.c @@ -0,0 +1,135 @@ +/* Copyright (c) 2017 The Linux Foundation. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 and + * only version 2 as published by the Free Software Foundation. + * + * This program is distributed in the hope that 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. + * + */ + +#include <linux/kref.h> +#include "msm_gpu.h" + +void msm_submitqueue_destroy(struct kref *kref) +{ + struct msm_gpu_submitqueue *queue = container_of(kref, + struct msm_gpu_submitqueue, ref); + + kfree(queue); +} + +struct msm_gpu_submitqueue *msm_submitqueue_get(struct msm_file_private *ctx, + u32 id) +{ + struct msm_gpu_submitqueue *entry; + + if (!ctx) + return NULL; + + read_lock(&ctx->queuelock); + + list_for_each_entry(entry, &ctx->submitqueues, node) { + if (entry->id == id) { + kref_get(&entry->ref); + read_unlock(&ctx->queuelock); + + return entry; + } + } + + read_unlock(&ctx->queuelock); + return NULL; +} + +void msm_submitqueue_close(struct msm_file_private *ctx) +{ + struct msm_gpu_submitqueue *entry, *tmp; + + if (!ctx) + return; + + /* + * No lock needed in close and there won't + * be any more user ioctls coming our way + */ + list_for_each_entry_safe(entry, tmp, &ctx->submitqueues, node) + msm_submitqueue_put(entry); +} + +int msm_submitqueue_create(struct msm_file_private *ctx, u32 prio, u32 flags, + u32 *id) +{ + struct msm_gpu_submitqueue *queue; + + if (!ctx) + return -ENODEV; + + queue = kzalloc(sizeof(*queue), GFP_KERNEL); + + if (!queue) + return -ENOMEM; + + kref_init(&queue->ref); + queue->flags = flags; + queue->prio = prio; + + write_lock(&ctx->queuelock); + + queue->id = ctx->queueid++; + + if (id) + *id = queue->id; + + list_add_tail(&queue->node, &ctx->submitqueues); + + write_unlock(&ctx->queuelock); + + return 0; +} + +int msm_submitqueue_init(struct msm_file_private *ctx) +{ + if (!ctx) + return 0; + + INIT_LIST_HEAD(&ctx->submitqueues); + + rwlock_init(&ctx->queuelock); + + return msm_submitqueue_create(ctx, 2, 0, NULL); +} + +int msm_submitqueue_remove(struct msm_file_private *ctx, u32 id) +{ + struct msm_gpu_submitqueue *entry; + + if (!ctx) + return 0; + + /* + * id 0 is the "default" queue and can't be destroyed + * by the user + */ + if (!id) + return -ENOENT; + + write_lock(&ctx->queuelock); + + list_for_each_entry(entry, &ctx->submitqueues, node) { + if (entry->id == id) { + list_del(&entry->node); + write_unlock(&ctx->queuelock); + + msm_submitqueue_put(entry); + return 0; + } + } + + write_unlock(&ctx->queuelock); + return -ENOENT; +} + diff --git a/include/uapi/drm/msm_drm.h b/include/uapi/drm/msm_drm.h index ad4eb28..9fb4b81 100644 --- a/include/uapi/drm/msm_drm.h +++ b/include/uapi/drm/msm_drm.h @@ -218,6 +218,7 @@ struct drm_msm_gem_submit { __u64 bos; /* in, ptr to array of submit_bo's */ __u64 cmds; /* in, ptr to array of submit_cmd's */ __s32 fence_fd; /* in/out fence fd (see MSM_SUBMIT_FENCE_FD_IN/OUT) */ + __u32 queueid; /* in, submitqueue id */ };

/* The normal way to synchronize with the GPU is just to CPU_PREP on @@ -254,6 +255,20 @@ struct drm_msm_gem_madvise { __u32 retained; /* out, whether backing store still exists */ };

+/* + * Draw queues allow the user to set specific submission parameter. Command + * submissions specify a specific submitqueue to use. ID 0 is reserved for + * backwards compatibility as a "default" submitqueue + */ + +#define MSM_SUBMITQUEUE_FLAGS (0) + +struct drm_msm_submitqueue { + __u32 flags; /* in, MSM_SUBMITQUEUE_x */ + __u32 prio; /* in, Priority level */ + __u32 id; /* out, identifier */ +}; + #define DRM_MSM_GET_PARAM 0x00 /* placeholder: #define DRM_MSM_SET_PARAM 0x01 @@ -265,6 +280,11 @@ struct drm_msm_gem_madvise { #define DRM_MSM_GEM_SUBMIT 0x06 #define DRM_MSM_WAIT_FENCE 0x07 #define DRM_MSM_GEM_MADVISE 0x08 +/* placeholder: +#define DRM_MSM_GEM_SVM_NEW 0x09 + */ +#define DRM_MSM_SUBMITQUEUE_NEW 0x0A +#define DRM_MSM_SUBMITQUEUE_CLOSE 0x0B

#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) @@ -274,6 +294,8 @@ struct drm_msm_gem_madvise { #define DRM_IOCTL_MSM_GEM_SUBMIT DRM_IOWR(DRM_COMMAND_BASE + DRM_MSM_GEM_SUBMIT, struct drm_msm_gem_submit) #define DRM_IOCTL_MSM_WAIT_FENCE DRM_IOW (DRM_COMMAND_BASE + DRM_MSM_WAIT_FENCE, struct drm_msm_wait_fence) #define DRM_IOCTL_MSM_GEM_MADVISE DRM_IOWR(DRM_COMMAND_BASE + DRM_MSM_GEM_MADVISE, struct drm_msm_gem_madvise) +#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)

#if defined(__cplusplus) }

Add the infrastructure to support the idea of multiple ringbuffers. Assign each ringbuffer an id and use that as an index for the various ring specific operations.

The biggest delta is to support legacy fences. Each fence gets its own sequence number but the legacy functions expect to use a unique integer. To handle this we return a unique identifier for each submission but map it to a specific ring/sequence under the covers. Newer users use a dma_fence pointer anyway so they don't care about the actual sequence ID or ring.

The actual mechanics for multiple ringbuffers are very target specific so this code just allows for the possibility but still only defines one ringbuffer for each target family.

Signed-off-by: Jordan Crouse jcrouse@codeaurora.org --- drivers/gpu/drm/msm/adreno/a3xx_gpu.c | 9 +- drivers/gpu/drm/msm/adreno/a4xx_gpu.c | 9 +- drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 54 ++++++----- drivers/gpu/drm/msm/adreno/a5xx_gpu.h | 2 +- drivers/gpu/drm/msm/adreno/a5xx_power.c | 6 +- drivers/gpu/drm/msm/adreno/adreno_gpu.c | 137 ++++++++++++++++----------- drivers/gpu/drm/msm/adreno/adreno_gpu.h | 20 ++-- drivers/gpu/drm/msm/msm_drv.c | 23 +++-- drivers/gpu/drm/msm/msm_drv.h | 8 +- drivers/gpu/drm/msm/msm_fence.c | 2 +- drivers/gpu/drm/msm/msm_fence.h | 2 +- drivers/gpu/drm/msm/msm_gem.h | 4 +- drivers/gpu/drm/msm/msm_gem_submit.c | 12 ++- drivers/gpu/drm/msm/msm_gpu.c | 163 ++++++++++++++++++++++---------- drivers/gpu/drm/msm/msm_gpu.h | 42 ++++---- drivers/gpu/drm/msm/msm_ringbuffer.c | 35 +++++-- drivers/gpu/drm/msm/msm_ringbuffer.h | 20 +++- drivers/gpu/drm/msm/msm_submitqueue.c | 27 +++++- include/uapi/drm/msm_drm.h | 1 + 19 files changed, 366 insertions(+), 210 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c index 789f7fb..4baef27 100644 --- a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c +++ b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c @@ -44,7 +44,7 @@

static bool a3xx_me_init(struct msm_gpu *gpu) { - struct msm_ringbuffer *ring = gpu->rb; + struct msm_ringbuffer *ring = gpu->rb[0];

OUT_PKT3(ring, CP_ME_INIT, 17); OUT_RING(ring, 0x000003f7); @@ -65,7 +65,7 @@ static bool a3xx_me_init(struct msm_gpu *gpu) OUT_RING(ring, 0x00000000); OUT_RING(ring, 0x00000000);

- gpu->funcs->flush(gpu); + gpu->funcs->flush(gpu, ring); return a3xx_idle(gpu); }

@@ -339,7 +339,7 @@ static void a3xx_destroy(struct msm_gpu *gpu) static bool a3xx_idle(struct msm_gpu *gpu) { /* wait for ringbuffer to drain: */ - if (!adreno_idle(gpu)) + if (!adreno_idle(gpu, gpu->rb[0])) return false;

/* then wait for GPU to finish: */ @@ -446,6 +446,7 @@ static void a3xx_dump(struct msm_gpu *gpu) .recover = a3xx_recover, .submit = adreno_submit, .flush = adreno_flush, + .active_ring = adreno_active_ring, .irq = a3xx_irq, .destroy = a3xx_destroy, #ifdef CONFIG_DEBUG_FS @@ -491,7 +492,7 @@ struct msm_gpu *a3xx_gpu_init(struct drm_device *dev) adreno_gpu->registers = a3xx_registers; adreno_gpu->reg_offsets = a3xx_register_offsets;

- ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs); + ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs, 1); if (ret) goto fail;

diff --git a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c index f87c4312..8199a4b 100644 --- a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c +++ b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c @@ -116,7 +116,7 @@ static void a4xx_enable_hwcg(struct msm_gpu *gpu)

static bool a4xx_me_init(struct msm_gpu *gpu) { - struct msm_ringbuffer *ring = gpu->rb; + struct msm_ringbuffer *ring = gpu->rb[0];

OUT_PKT3(ring, CP_ME_INIT, 17); OUT_RING(ring, 0x000003f7); @@ -137,7 +137,7 @@ static bool a4xx_me_init(struct msm_gpu *gpu) OUT_RING(ring, 0x00000000); OUT_RING(ring, 0x00000000);

- gpu->funcs->flush(gpu); + gpu->funcs->flush(gpu, ring); return a4xx_idle(gpu); }

@@ -337,7 +337,7 @@ static void a4xx_destroy(struct msm_gpu *gpu) static bool a4xx_idle(struct msm_gpu *gpu) { /* wait for ringbuffer to drain: */ - if (!adreno_idle(gpu)) + if (!adreno_idle(gpu, gpu->rb[0])) return false;

/* then wait for GPU to finish: */ @@ -534,6 +534,7 @@ static int a4xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value) .recover = a4xx_recover, .submit = adreno_submit, .flush = adreno_flush, + .active_ring = adreno_active_ring, .irq = a4xx_irq, .destroy = a4xx_destroy, #ifdef CONFIG_DEBUG_FS @@ -573,7 +574,7 @@ struct msm_gpu *a4xx_gpu_init(struct drm_device *dev) adreno_gpu->registers = a4xx_registers; adreno_gpu->reg_offsets = a4xx_register_offsets;

- ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs); + ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs, 1); if (ret) goto fail;

diff --git a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c index 32a5b55..b7b9f4f 100644 --- a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c +++ b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c @@ -99,7 +99,7 @@ static void a5xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, struct msm_file_private *ctx) { struct msm_drm_private *priv = gpu->dev->dev_private; - struct msm_ringbuffer *ring = gpu->rb; + struct msm_ringbuffer *ring = submit->ring; unsigned int i, ibs = 0;

for (i = 0; i < submit->nr_cmds; i++) { @@ -120,15 +120,15 @@ static void a5xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, }

OUT_PKT4(ring, REG_A5XX_CP_SCRATCH_REG(2), 1); - OUT_RING(ring, submit->fence->seqno); + OUT_RING(ring, submit->seqno);

OUT_PKT7(ring, CP_EVENT_WRITE, 4); OUT_RING(ring, CACHE_FLUSH_TS | (1 << 31)); - OUT_RING(ring, lower_32_bits(rbmemptr(gpu, fence))); - OUT_RING(ring, upper_32_bits(rbmemptr(gpu, fence))); - OUT_RING(ring, submit->fence->seqno); + OUT_RING(ring, lower_32_bits(rbmemptr(ring, fence))); + OUT_RING(ring, upper_32_bits(rbmemptr(ring, fence))); + OUT_RING(ring, submit->seqno);

- gpu->funcs->flush(gpu); + gpu->funcs->flush(gpu, ring); }

static const struct { @@ -244,7 +244,7 @@ void a5xx_set_hwcg(struct msm_gpu *gpu, bool state) static int a5xx_me_init(struct msm_gpu *gpu) { struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); - struct msm_ringbuffer *ring = gpu->rb; + struct msm_ringbuffer *ring = gpu->rb[0];

OUT_PKT7(ring, CP_ME_INIT, 8);

@@ -275,9 +275,8 @@ static int a5xx_me_init(struct msm_gpu *gpu) OUT_RING(ring, 0x00000000); OUT_RING(ring, 0x00000000);

- gpu->funcs->flush(gpu); - - return a5xx_idle(gpu) ? 0 : -EINVAL; + gpu->funcs->flush(gpu, ring); + return a5xx_idle(gpu, ring) ? 0 : -EINVAL; }

static struct drm_gem_object *a5xx_ucode_load_bo(struct msm_gpu *gpu, @@ -564,11 +563,11 @@ static int a5xx_hw_init(struct msm_gpu *gpu) * ticking correctly */ if (adreno_is_a530(adreno_gpu)) { - OUT_PKT7(gpu->rb, CP_EVENT_WRITE, 1); - OUT_RING(gpu->rb, 0x0F); + OUT_PKT7(gpu->rb[0], CP_EVENT_WRITE, 1); + OUT_RING(gpu->rb[0], 0x0F);

- gpu->funcs->flush(gpu); - if (!a5xx_idle(gpu)) + gpu->funcs->flush(gpu, gpu->rb[0]); + if (!a5xx_idle(gpu, gpu->rb[0])) return -EINVAL; }

@@ -581,11 +580,11 @@ static int a5xx_hw_init(struct msm_gpu *gpu) */ ret = a5xx_zap_shader_init(gpu); if (!ret) { - OUT_PKT7(gpu->rb, CP_SET_SECURE_MODE, 1); - OUT_RING(gpu->rb, 0x00000000); + OUT_PKT7(gpu->rb[0], CP_SET_SECURE_MODE, 1); + OUT_RING(gpu->rb[0], 0x00000000);

- gpu->funcs->flush(gpu); - if (!a5xx_idle(gpu)) + gpu->funcs->flush(gpu, gpu->rb[0]); + if (!a5xx_idle(gpu, gpu->rb[0])) return -EINVAL; } else { /* Print a warning so if we die, we know why */ @@ -659,18 +658,19 @@ static inline bool _a5xx_check_idle(struct msm_gpu *gpu) A5XX_RBBM_INT_0_MASK_MISC_HANG_DETECT); }

-bool a5xx_idle(struct msm_gpu *gpu) +bool a5xx_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring) { /* wait for CP to drain ringbuffer: */ - if (!adreno_idle(gpu)) + if (!adreno_idle(gpu, ring)) return false;

if (spin_until(_a5xx_check_idle(gpu))) { - DRM_ERROR("%s: %ps: timeout waiting for GPU to idle: status %8.8X irq %8.8X\n", + DRM_ERROR("%s: %ps: timeout waiting for GPU to idle: status %8.8X irq %8.8X rptr/wptr %d/%d\n", gpu->name, __builtin_return_address(0), gpu_read(gpu, REG_A5XX_RBBM_STATUS), - gpu_read(gpu, REG_A5XX_RBBM_INT_0_STATUS)); - + gpu_read(gpu, REG_A5XX_RBBM_INT_0_STATUS), + gpu_read(gpu, REG_A5XX_CP_RB_RPTR), + gpu_read(gpu, REG_A5XX_CP_RB_WPTR)); return false; }

@@ -801,9 +801,10 @@ static void a5xx_fault_detect_irq(struct msm_gpu *gpu) { struct drm_device *dev = gpu->dev; struct msm_drm_private *priv = dev->dev_private; + struct msm_ringbuffer *ring = gpu->funcs->active_ring(gpu);

- dev_err(dev->dev, "gpu fault fence %x status %8.8X rb %4.4x/%4.4x ib1 %16.16llX/%4.4x ib2 %16.16llX/%4.4x\n", - gpu->memptrs->fence, + dev_err(dev->dev, "gpu fault ring %d fence %x status %8.8X rb %4.4x/%4.4x ib1 %16.16llX/%4.4x ib2 %16.16llX/%4.4x\n", + ring ? ring->id : -1, ring ? ring->seqno : 0, gpu_read(gpu, REG_A5XX_RBBM_STATUS), gpu_read(gpu, REG_A5XX_CP_RB_RPTR), gpu_read(gpu, REG_A5XX_CP_RB_WPTR), @@ -993,6 +994,7 @@ static void a5xx_show(struct msm_gpu *gpu, struct seq_file *m) .recover = a5xx_recover, .submit = a5xx_submit, .flush = adreno_flush, + .active_ring = adreno_active_ring, .irq = a5xx_irq, .destroy = a5xx_destroy, #ifdef CONFIG_DEBUG_FS @@ -1028,7 +1030,7 @@ struct msm_gpu *a5xx_gpu_init(struct drm_device *dev)

a5xx_gpu->lm_leakage = 0x4E001A;

- ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs); + ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs, 1); if (ret) { a5xx_destroy(&(a5xx_gpu->base.base)); return ERR_PTR(ret); diff --git a/drivers/gpu/drm/msm/adreno/a5xx_gpu.h b/drivers/gpu/drm/msm/adreno/a5xx_gpu.h index e944516..44db48d 100644 --- a/drivers/gpu/drm/msm/adreno/a5xx_gpu.h +++ b/drivers/gpu/drm/msm/adreno/a5xx_gpu.h @@ -55,7 +55,7 @@ static inline int spin_usecs(struct msm_gpu *gpu, uint32_t usecs, return -ETIMEDOUT; }

-bool a5xx_idle(struct msm_gpu *gpu); +bool a5xx_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring); void a5xx_set_hwcg(struct msm_gpu *gpu, bool state);

#endif /* __A5XX_GPU_H__ */ diff --git a/drivers/gpu/drm/msm/adreno/a5xx_power.c b/drivers/gpu/drm/msm/adreno/a5xx_power.c index 04aab1d..d09a716 100644 --- a/drivers/gpu/drm/msm/adreno/a5xx_power.c +++ b/drivers/gpu/drm/msm/adreno/a5xx_power.c @@ -173,7 +173,7 @@ static int a5xx_gpmu_init(struct msm_gpu *gpu) { struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu); - struct msm_ringbuffer *ring = gpu->rb; + struct msm_ringbuffer *ring = gpu->rb[0];

if (!a5xx_gpu->gpmu_dwords) return 0; @@ -192,9 +192,9 @@ static int a5xx_gpmu_init(struct msm_gpu *gpu) OUT_PKT7(ring, CP_SET_PROTECTED_MODE, 1); OUT_RING(ring, 1);

- gpu->funcs->flush(gpu); + gpu->funcs->flush(gpu, ring);

- if (!a5xx_idle(gpu)) { + if (!a5xx_idle(gpu, ring)) { DRM_ERROR("%s: Unable to load GPMU firmware. GPMU will not be active\n", gpu->name); return -EINVAL; diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c index b9238ac..fad946946 100644 --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c @@ -21,7 +21,6 @@ #include "msm_gem.h" #include "msm_mmu.h"

-#define RB_SIZE SZ_32K #define RB_BLKSIZE 32

int adreno_get_param(struct msm_gpu *gpu, uint32_t param, uint64_t *value) @@ -67,38 +66,47 @@ int adreno_get_param(struct msm_gpu *gpu, uint32_t param, uint64_t *value) int adreno_hw_init(struct msm_gpu *gpu) { struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); - int ret; + int i;

DBG("%s", gpu->name);

- ret = msm_gem_get_iova(gpu->rb->bo, gpu->aspace, &gpu->rb_iova); - if (ret) { - gpu->rb_iova = 0; - dev_err(gpu->dev->dev, "could not map ringbuffer: %d\n", ret); - return ret; - } + for (i = 0; i < gpu->nr_rings; i++) { + struct msm_ringbuffer *ring = gpu->rb[i]; + int ret; + + if (!ring) + continue;

- /* reset ringbuffer: */ - gpu->rb->cur = gpu->rb->start; + ret = msm_gem_get_iova(ring->bo, gpu->aspace, &ring->iova); + if (ret) { + ring->iova = 0; + dev_err(gpu->dev->dev, + "could not map ringbuffer %d: %d\n", i, ret); + return ret; + } + + ring->cur = ring->start;

- /* reset completed fence seqno: */ - gpu->memptrs->fence = gpu->fctx->completed_fence; - gpu->memptrs->rptr = 0; + /* reset completed fence seqno: */ + ring->memptrs->fence = ring->seqno; + ring->memptrs->rptr = 0; + }

/* Setup REG_CP_RB_CNTL: */ adreno_gpu_write(adreno_gpu, REG_ADRENO_CP_RB_CNTL, - /* size is log2(quad-words): */ - AXXX_CP_RB_CNTL_BUFSZ(ilog2(gpu->rb->size / 8)) | - AXXX_CP_RB_CNTL_BLKSZ(ilog2(RB_BLKSIZE / 8)) | - (adreno_is_a430(adreno_gpu) ? AXXX_CP_RB_CNTL_NO_UPDATE : 0)); + /* size is log2(quad-words): */ + AXXX_CP_RB_CNTL_BUFSZ(ilog2(MSM_GPU_RINGBUFFER_SZ / 8)) | + AXXX_CP_RB_CNTL_BLKSZ(ilog2(RB_BLKSIZE / 8)) | + (adreno_is_a430(adreno_gpu) ? AXXX_CP_RB_CNTL_NO_UPDATE : 0));

- /* Setup ringbuffer address: */ + /* Setup ringbuffer address - use ringbuffer[0] for GPU init */ adreno_gpu_write64(adreno_gpu, REG_ADRENO_CP_RB_BASE, - REG_ADRENO_CP_RB_BASE_HI, gpu->rb_iova); + REG_ADRENO_CP_RB_BASE_HI, gpu->rb[0]->iova);

if (!adreno_is_a430(adreno_gpu)) { adreno_gpu_write64(adreno_gpu, REG_ADRENO_CP_RB_RPTR_ADDR, - REG_ADRENO_CP_RB_RPTR_ADDR_HI, rbmemptr(gpu, rptr)); + REG_ADRENO_CP_RB_RPTR_ADDR_HI, + rbmemptr(gpu->rb[0], rptr)); }

return 0; @@ -110,15 +118,19 @@ static uint32_t get_wptr(struct msm_ringbuffer *ring) }

/* Use this helper to read rptr, since a430 doesn't update rptr in memory */ -static uint32_t get_rptr(struct adreno_gpu *adreno_gpu) +static uint32_t get_rptr(struct adreno_gpu *adreno_gpu, + struct msm_ringbuffer *ring) { - struct msm_gpu *gpu = &adreno_gpu->base; - if (adreno_is_a430(adreno_gpu)) - return gpu->memptrs->rptr = adreno_gpu_read( + return ring->memptrs->rptr = adreno_gpu_read( adreno_gpu, REG_ADRENO_CP_RB_RPTR); else - return gpu->memptrs->rptr; + return ring->memptrs->rptr; +} + +struct msm_ringbuffer *adreno_active_ring(struct msm_gpu *gpu) +{ + return gpu->rb[0]; }

void adreno_recover(struct msm_gpu *gpu) @@ -144,7 +156,7 @@ void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, { struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); struct msm_drm_private *priv = gpu->dev->dev_private; - struct msm_ringbuffer *ring = gpu->rb; + struct msm_ringbuffer *ring = submit->ring; unsigned i;

for (i = 0; i < submit->nr_cmds; i++) { @@ -167,7 +179,7 @@ void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, }

OUT_PKT0(ring, REG_AXXX_CP_SCRATCH_REG2, 1); - OUT_RING(ring, submit->fence->seqno); + OUT_RING(ring, submit->seqno);

if (adreno_is_a3xx(adreno_gpu) || adreno_is_a4xx(adreno_gpu)) { /* Flush HLSQ lazy updates to make sure there is nothing @@ -183,8 +195,8 @@ void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,

OUT_PKT3(ring, CP_EVENT_WRITE, 3); OUT_RING(ring, CACHE_FLUSH_TS); - OUT_RING(ring, rbmemptr(gpu, fence)); - OUT_RING(ring, submit->fence->seqno); + OUT_RING(ring, rbmemptr(ring, fence)); + OUT_RING(ring, submit->seqno);

/* we could maybe be clever and only CP_COND_EXEC the interrupt: */ OUT_PKT3(ring, CP_INTERRUPT, 1); @@ -210,10 +222,10 @@ void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, } #endif

- gpu->funcs->flush(gpu); + gpu->funcs->flush(gpu, ring); }

-void adreno_flush(struct msm_gpu *gpu) +void adreno_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring) { struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); uint32_t wptr; @@ -223,7 +235,7 @@ void adreno_flush(struct msm_gpu *gpu) * to account for the possibility that the last command fit exactly into * the ringbuffer and rb->next hasn't wrapped to zero yet */ - wptr = get_wptr(gpu->rb) & ((gpu->rb->size / 4) - 1); + wptr = get_wptr(ring) % (MSM_GPU_RINGBUFFER_SZ >> 2);

/* ensure writes to ringbuffer have hit system memory: */ mb(); @@ -231,17 +243,18 @@ void adreno_flush(struct msm_gpu *gpu) adreno_gpu_write(adreno_gpu, REG_ADRENO_CP_RB_WPTR, wptr); }

-bool adreno_idle(struct msm_gpu *gpu) +bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring) { struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); - uint32_t wptr = get_wptr(gpu->rb); + uint32_t wptr = get_wptr(ring);

/* wait for CP to drain ringbuffer: */ - if (!spin_until(get_rptr(adreno_gpu) == wptr)) + if (!spin_until(get_rptr(adreno_gpu, ring) == wptr)) return true;

/* TODO maybe we need to reset GPU here to recover from hang? */ - DRM_ERROR("%s: timeout waiting to drain ringbuffer!\n", gpu->name); + DRM_ERROR("%s: timeout waiting to drain ringbuffer %d!\n", gpu->name, + ring->id); return false; }

@@ -256,10 +269,16 @@ void adreno_show(struct msm_gpu *gpu, struct seq_file *m) adreno_gpu->rev.major, adreno_gpu->rev.minor, adreno_gpu->rev.patchid);

- seq_printf(m, "fence: %d/%d\n", gpu->memptrs->fence, - gpu->fctx->last_fence); - seq_printf(m, "rptr: %d\n", get_rptr(adreno_gpu)); - seq_printf(m, "rb wptr: %d\n", get_wptr(gpu->rb)); + for (i = 0; i < gpu->nr_rings; i++) { + struct msm_ringbuffer *ring = gpu->rb[i]; + + seq_printf(m, "rb %d: fence: %d/%d\n", i, + ring->memptrs->fence, ring->seqno); + + seq_printf(m, " rptr: %d\n", + get_rptr(adreno_gpu, ring)); + seq_printf(m, "rb wptr: %d\n", get_wptr(ring)); + }

/* dump these out in a form that can be parsed by demsm: */ seq_printf(m, "IO:region %s 00000000 00020000\n", gpu->name); @@ -285,16 +304,23 @@ void adreno_show(struct msm_gpu *gpu, struct seq_file *m) void adreno_dump_info(struct msm_gpu *gpu) { struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); + int i;

printk("revision: %d (%d.%d.%d.%d)\n", adreno_gpu->info->revn, adreno_gpu->rev.core, adreno_gpu->rev.major, adreno_gpu->rev.minor, adreno_gpu->rev.patchid);

- printk("fence: %d/%d\n", gpu->memptrs->fence, - gpu->fctx->last_fence); - printk("rptr: %d\n", get_rptr(adreno_gpu)); - printk("rb wptr: %d\n", get_wptr(gpu->rb)); + for (i = 0; i < gpu->nr_rings; i++) { + struct msm_ringbuffer *ring = gpu->rb[i]; + + printk("rb %d: fence: %d/%d\n", i, + ring->memptrs->fence, + ring->seqno); + + printk("rptr: %d\n", get_rptr(adreno_gpu, ring)); + printk("rb wptr: %d\n", get_wptr(ring)); + } }

/* would be nice to not have to duplicate the _show() stuff with printk(): */ @@ -317,23 +343,26 @@ void adreno_dump(struct msm_gpu *gpu) } }

-static uint32_t ring_freewords(struct msm_gpu *gpu) +static uint32_t ring_freewords(struct msm_ringbuffer *ring) { - struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); - uint32_t size = gpu->rb->size / 4; - uint32_t wptr = get_wptr(gpu->rb); - uint32_t rptr = get_rptr(adreno_gpu); + struct adreno_gpu *adreno_gpu = to_adreno_gpu(ring->gpu); + uint32_t size = MSM_GPU_RINGBUFFER_SZ >> 2; + uint32_t wptr = get_wptr(ring); + uint32_t rptr = get_rptr(adreno_gpu, ring); return (rptr + (size - 1) - wptr) % size; }

-void adreno_wait_ring(struct msm_gpu *gpu, uint32_t ndwords) +void adreno_wait_ring(struct msm_ringbuffer *ring, uint32_t ndwords) { - if (spin_until(ring_freewords(gpu) >= ndwords)) - DRM_ERROR("%s: timeout waiting for ringbuffer space\n", gpu->name); + if (spin_until(ring_freewords(ring) >= ndwords)) + DRM_DEV_ERROR(ring->gpu->dev->dev, + "timeout waiting for space in ringubffer %d\n", + ring->id); }

int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev, - struct adreno_gpu *adreno_gpu, const struct adreno_gpu_funcs *funcs) + struct adreno_gpu *adreno_gpu, + const struct adreno_gpu_funcs *funcs, int nr_rings) { struct adreno_platform_config *config = pdev->dev.platform_data; struct msm_gpu_config adreno_gpu_config = { 0 }; @@ -361,7 +390,7 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev, adreno_gpu_config.va_start = SZ_16M; adreno_gpu_config.va_end = 0xffffffff;

- adreno_gpu_config.ringsz = RB_SIZE; + adreno_gpu_config.nr_rings = nr_rings;

pm_runtime_set_autosuspend_delay(&pdev->dev, DRM_MSM_INACTIVE_PERIOD); pm_runtime_use_autosuspend(&pdev->dev); diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h index 5ee1d6a..2a51ac4 100644 --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h @@ -185,17 +185,19 @@ static inline int adreno_is_a530(struct adreno_gpu *gpu) void adreno_recover(struct msm_gpu *gpu); void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, struct msm_file_private *ctx); -void adreno_flush(struct msm_gpu *gpu); -bool adreno_idle(struct msm_gpu *gpu); +void adreno_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring); +bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring); #ifdef CONFIG_DEBUG_FS void adreno_show(struct msm_gpu *gpu, struct seq_file *m); #endif void adreno_dump_info(struct msm_gpu *gpu); void adreno_dump(struct msm_gpu *gpu); -void adreno_wait_ring(struct msm_gpu *gpu, uint32_t ndwords); +void adreno_wait_ring(struct msm_ringbuffer *ring, uint32_t ndwords); +struct msm_ringbuffer *adreno_active_ring(struct msm_gpu *gpu);

int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev, - struct adreno_gpu *gpu, const struct adreno_gpu_funcs *funcs); + struct adreno_gpu *gpu, const struct adreno_gpu_funcs *funcs, + int nr_rings); void adreno_gpu_cleanup(struct adreno_gpu *gpu);

@@ -204,7 +206,7 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev, static inline void OUT_PKT0(struct msm_ringbuffer *ring, uint16_t regindx, uint16_t cnt) { - adreno_wait_ring(ring->gpu, cnt+1); + adreno_wait_ring(ring, cnt+1); OUT_RING(ring, CP_TYPE0_PKT | ((cnt-1) << 16) | (regindx & 0x7FFF)); }

@@ -212,14 +214,14 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev, static inline void OUT_PKT2(struct msm_ringbuffer *ring) { - adreno_wait_ring(ring->gpu, 1); + adreno_wait_ring(ring, 1); OUT_RING(ring, CP_TYPE2_PKT); }

static inline void OUT_PKT3(struct msm_ringbuffer *ring, uint8_t opcode, uint16_t cnt) { - adreno_wait_ring(ring->gpu, cnt+1); + adreno_wait_ring(ring, cnt+1); OUT_RING(ring, CP_TYPE3_PKT | ((cnt-1) << 16) | ((opcode & 0xFF) << 8)); }

@@ -241,14 +243,14 @@ static inline u32 PM4_PARITY(u32 val) static inline void OUT_PKT4(struct msm_ringbuffer *ring, uint16_t regindx, uint16_t cnt) { - adreno_wait_ring(ring->gpu, cnt + 1); + adreno_wait_ring(ring, cnt + 1); OUT_RING(ring, PKT4(regindx, cnt)); }

static inline void OUT_PKT7(struct msm_ringbuffer *ring, uint8_t opcode, uint16_t cnt) { - adreno_wait_ring(ring->gpu, cnt + 1); + adreno_wait_ring(ring, cnt + 1); OUT_RING(ring, CP_TYPE7_PKT | (cnt << 0) | (PM4_PARITY(cnt) << 15) | ((opcode & 0x7F) << 16) | (PM4_PARITY(opcode) << 23)); } diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c index cf743c7..45f1170 100644 --- a/drivers/gpu/drm/msm/msm_drv.c +++ b/drivers/gpu/drm/msm/msm_drv.c @@ -517,7 +517,7 @@ static void load_gpu(struct drm_device *dev) mutex_unlock(&init_lock); }

-static int context_init(struct drm_file *file) +static int context_init(struct drm_device *dev, struct drm_file *file) { struct msm_file_private *ctx;

@@ -525,7 +525,7 @@ static int context_init(struct drm_file *file) if (!ctx) return -ENOMEM;

- msm_submitqueue_init(ctx); + msm_submitqueue_init(dev, ctx);

file->driver_priv = ctx;

@@ -539,7 +539,7 @@ static int msm_open(struct drm_device *dev, struct drm_file *file) */ load_gpu(dev);

- return context_init(file); + return context_init(dev, file); }

static void context_close(struct msm_file_private *ctx) @@ -753,16 +753,27 @@ static int msm_ioctl_wait_fence(struct drm_device *dev, void *data, struct msm_drm_private *priv = dev->dev_private; struct drm_msm_wait_fence *args = data; ktime_t timeout = to_ktime(args->timeout); + struct msm_gpu_submitqueue *queue; + struct msm_gpu *gpu = priv->gpu; + int ret;

if (args->pad) { DRM_ERROR("invalid pad: %08x\n", args->pad); return -EINVAL; }

- if (!priv->gpu) + if (!gpu) return 0;

- return msm_wait_fence(priv->gpu->fctx, args->fence, &timeout, true); + queue = msm_submitqueue_get(file->driver_priv, args->queueid); + if (!queue) + return -ENOENT; + + ret = msm_wait_fence(gpu->rb[queue->prio]->fctx, args->fence, &timeout, + true); + + msm_submitqueue_put(queue); + return ret; }

static int msm_ioctl_gem_madvise(struct drm_device *dev, void *data, @@ -812,7 +823,7 @@ static int msm_ioctl_submitqueue_new(struct drm_device *dev, void *data, if (args->flags & ~MSM_SUBMITQUEUE_FLAGS) return -EINVAL;

- return msm_submitqueue_create(file->driver_priv, args->prio, + return msm_submitqueue_create(dev, file->driver_priv, args->prio, args->flags, &args->id); }

diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h index b3b8f20..2821f57 100644 --- a/drivers/gpu/drm/msm/msm_drv.h +++ b/drivers/gpu/drm/msm/msm_drv.h @@ -74,6 +74,8 @@ struct msm_vblank_ctrl { spinlock_t lock; };

+#define MSM_GPU_MAX_RINGS 1 + struct msm_drm_private {

struct drm_device *dev; @@ -318,11 +320,11 @@ void __iomem *msm_ioremap(struct platform_device *pdev, const char *name, u32 msm_readl(const void __iomem *addr);

struct msm_gpu_submitqueue; -int msm_submitqueue_init(struct msm_file_private *ctx); +int msm_submitqueue_init(struct drm_device *drm, struct msm_file_private *ctx); struct msm_gpu_submitqueue *msm_submitqueue_get(struct msm_file_private *ctx, u32 id); -int msm_submitqueue_create(struct msm_file_private *ctx, u32 prio, - u32 flags, u32 *id); +int msm_submitqueue_create(struct drm_device *drm, struct msm_file_private *ctx, + u32 prio, u32 flags, u32 *id); int msm_submitqueue_remove(struct msm_file_private *ctx, u32 id); void msm_submitqueue_close(struct msm_file_private *ctx);

diff --git a/drivers/gpu/drm/msm/msm_fence.c b/drivers/gpu/drm/msm/msm_fence.c index a2f89ba..349c12f 100644 --- a/drivers/gpu/drm/msm/msm_fence.c +++ b/drivers/gpu/drm/msm/msm_fence.c @@ -31,7 +31,7 @@ struct msm_fence_context * return ERR_PTR(-ENOMEM);

fctx->dev = dev; - fctx->name = name; + strncpy(fctx->name, name, sizeof(fctx->name)); fctx->context = dma_fence_context_alloc(1); init_waitqueue_head(&fctx->event); spin_lock_init(&fctx->spinlock); diff --git a/drivers/gpu/drm/msm/msm_fence.h b/drivers/gpu/drm/msm/msm_fence.h index 56061aa..1aa6a4c 100644 --- a/drivers/gpu/drm/msm/msm_fence.h +++ b/drivers/gpu/drm/msm/msm_fence.h @@ -22,7 +22,7 @@

struct msm_fence_context { struct drm_device *dev; - const char *name; + char name[32]; unsigned context; /* last_fence == completed_fence --> no pending work */ uint32_t last_fence; /* last assigned fence */ diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h index 17f8a6c..9320e18 100644 --- a/drivers/gpu/drm/msm/msm_gem.h +++ b/drivers/gpu/drm/msm/msm_gem.h @@ -138,13 +138,15 @@ enum msm_gem_lock { struct msm_gem_submit { struct drm_device *dev; struct msm_gpu *gpu; - struct list_head node; /* node in gpu submit_list */ + struct list_head node; /* node in ring submit list */ struct list_head bo_list; struct ww_acquire_ctx ticket; + uint32_t seqno; /* Sequence number of the submit on the ring */ struct dma_fence *fence; struct msm_gpu_submitqueue *queue; struct pid *pid; /* submitting process */ bool valid; /* true if no cmdstream patching needed */ + struct msm_ringbuffer *ring; unsigned int nr_cmds; unsigned int nr_bos; struct { diff --git a/drivers/gpu/drm/msm/msm_gem_submit.c b/drivers/gpu/drm/msm/msm_gem_submit.c index 0973884..9e087cf 100644 --- a/drivers/gpu/drm/msm/msm_gem_submit.c +++ b/drivers/gpu/drm/msm/msm_gem_submit.c @@ -51,6 +51,7 @@ static struct msm_gem_submit *submit_create(struct drm_device *dev, submit->pid = get_pid(task_pid(current)); submit->cmd = (void *)&submit->bos[nr_bos]; submit->queue = queue; + submit->ring = gpu->rb[queue->prio];

/* initially, until copy_from_user() and bo lookup succeeds: */ submit->nr_bos = 0; @@ -233,7 +234,8 @@ static int submit_fence_sync(struct msm_gem_submit *submit) struct msm_gem_object *msm_obj = submit->bos[i].obj; bool write = submit->bos[i].flags & MSM_SUBMIT_BO_WRITE;

- ret = msm_gem_sync_object(&msm_obj->base, submit->gpu->fctx, write); + ret = msm_gem_sync_object(&msm_obj->base, submit->ring->fctx, + write); if (ret) break; } @@ -396,6 +398,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, struct dma_fence *in_fence = NULL; struct sync_file *sync_file = NULL; struct msm_gpu_submitqueue *queue; + struct msm_ringbuffer *ring; int out_fence_fd = -1; unsigned i; int ret; @@ -416,6 +419,8 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, if (!queue) return -ENOENT;

+ ring = gpu->rb[queue->prio]; + if (args->flags & MSM_SUBMIT_FENCE_FD_IN) { in_fence = sync_file_get_fence(args->fence_fd);

@@ -426,7 +431,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, * Wait if the fence is from a foreign context, or if the fence * array contains any fence from a foreign context. */ - if (!dma_fence_match_context(in_fence, gpu->fctx->context)) { + if (!dma_fence_match_context(in_fence, ring->fctx->context)) { ret = dma_fence_wait(in_fence, true); if (ret) return ret; @@ -531,8 +536,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data,

submit->nr_cmds = i;

- submit->fence = msm_fence_alloc(gpu->fctx); - + submit->fence = msm_fence_alloc(ring->fctx); if (IS_ERR(submit->fence)) { ret = PTR_ERR(submit->fence); submit->fence = NULL; diff --git a/drivers/gpu/drm/msm/msm_gpu.c b/drivers/gpu/drm/msm/msm_gpu.c index 9e645ca..69c62c0 100644 --- a/drivers/gpu/drm/msm/msm_gpu.c +++ b/drivers/gpu/drm/msm/msm_gpu.c @@ -221,6 +221,20 @@ int msm_gpu_hw_init(struct msm_gpu *gpu) * Hangcheck detection for locked gpu: */

+static void update_fences(struct msm_gpu *gpu, struct msm_ringbuffer *ring, + uint32_t fence) +{ + struct msm_gem_submit *submit; + + list_for_each_entry(submit, &ring->submits, node) { + if (submit->seqno > fence) + break; + + msm_update_fence(submit->ring->fctx, + submit->fence->seqno); + } +} + static void retire_submits(struct msm_gpu *gpu);

static void recover_worker(struct work_struct *work) @@ -228,15 +242,34 @@ static void recover_worker(struct work_struct *work) struct msm_gpu *gpu = container_of(work, struct msm_gpu, recover_work); struct drm_device *dev = gpu->dev; struct msm_gem_submit *submit; - uint32_t fence = gpu->memptrs->fence; + struct msm_ringbuffer *cur_ring = gpu->funcs->active_ring(gpu); + uint64_t fence; + int i; + + /* Update all the rings with the latest and greatest fence */ + for (i = 0; i < ARRAY_SIZE(gpu->rb); i++) { + struct msm_ringbuffer *ring = gpu->rb[i];

- msm_update_fence(gpu->fctx, fence + 1); + fence = ring->memptrs->fence; + + /* + * For the current (faulting?) ring/submit advance the fence by + * one more to clear the faulting submit + */ + if (ring == cur_ring) + fence = fence + 1; + + update_fences(gpu, ring, fence); + }

mutex_lock(&dev->struct_mutex);

+ dev_err(dev->dev, "%s: hangcheck recover!\n", gpu->name); - list_for_each_entry(submit, &gpu->submit_list, node) { - if (submit->fence->seqno == (fence + 1)) { + fence = cur_ring->memptrs->fence + 1; + + list_for_each_entry(submit, &cur_ring->submits, node) { + if (submit->seqno == fence) { struct task_struct *task;

rcu_read_lock(); @@ -258,9 +291,16 @@ static void recover_worker(struct work_struct *work) gpu->funcs->recover(gpu); pm_runtime_put_sync(&gpu->pdev->dev);

- /* replay the remaining submits after the one that hung: */ - list_for_each_entry(submit, &gpu->submit_list, node) { - gpu->funcs->submit(gpu, submit, NULL); + /* + * Replay all remaining submits starting with highest priority + * ring + */ + + for (i = gpu->nr_rings - 1; i >= 0; i--) { + struct msm_ringbuffer *ring = gpu->rb[i]; + + list_for_each_entry(submit, &ring->submits, node) + gpu->funcs->submit(gpu, submit, NULL); } }

@@ -281,25 +321,27 @@ static void hangcheck_handler(unsigned long data) struct msm_gpu *gpu = (struct msm_gpu *)data; struct drm_device *dev = gpu->dev; struct msm_drm_private *priv = dev->dev_private; - uint32_t fence = gpu->memptrs->fence; + struct msm_ringbuffer *ring = gpu->funcs->active_ring(gpu); + uint32_t fence = ring->memptrs->fence;

- if (fence != gpu->hangcheck_fence) { + if (fence != ring->hangcheck_fence) { /* some progress has been made.. ya! */ - gpu->hangcheck_fence = fence; - } else if (fence < gpu->fctx->last_fence) { + ring->hangcheck_fence = fence; + } else if (fence < ring->seqno) { /* no progress and not done.. hung! */ - gpu->hangcheck_fence = fence; - dev_err(dev->dev, "%s: hangcheck detected gpu lockup!\n", - gpu->name); + ring->hangcheck_fence = fence; + dev_err(dev->dev, "%s: hangcheck detected gpu lockup rb %d!\n", + gpu->name, ring->id); dev_err(dev->dev, "%s: completed fence: %u\n", gpu->name, fence); dev_err(dev->dev, "%s: submitted fence: %u\n", - gpu->name, gpu->fctx->last_fence); + gpu->name, ring->seqno); + queue_work(priv->wq, &gpu->recover_work); }

/* if still more pending work, reset the hangcheck timer: */ - if (gpu->fctx->last_fence > gpu->hangcheck_fence) + if (ring->seqno > ring->hangcheck_fence) hangcheck_timer_reset(gpu);

/* workaround for missing irq: */ @@ -428,19 +470,18 @@ static void retire_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit) static void retire_submits(struct msm_gpu *gpu) { struct drm_device *dev = gpu->dev; + struct msm_gem_submit *submit, *tmp; + int i;

WARN_ON(!mutex_is_locked(&dev->struct_mutex));

- while (!list_empty(&gpu->submit_list)) { - struct msm_gem_submit *submit; - - submit = list_first_entry(&gpu->submit_list, - struct msm_gem_submit, node); + /* Retire the commits starting with highest priority */ + for (i = gpu->nr_rings - 1; i >= 0; i--) { + struct msm_ringbuffer *ring = gpu->rb[i];

- if (dma_fence_is_signaled(submit->fence)) { - retire_submit(gpu, submit); - } else { - break; + list_for_each_entry_safe(submit, tmp, &ring->submits, node) { + if (dma_fence_is_signaled(submit->fence)) + retire_submit(gpu, submit); } } } @@ -449,9 +490,10 @@ static void retire_worker(struct work_struct *work) { struct msm_gpu *gpu = container_of(work, struct msm_gpu, retire_work); struct drm_device *dev = gpu->dev; - uint32_t fence = gpu->memptrs->fence; + int i;

- msm_update_fence(gpu->fctx, fence); + for (i = 0; i < gpu->nr_rings; i++) + update_fences(gpu, gpu->rb[i], gpu->rb[i]->memptrs->fence);

mutex_lock(&dev->struct_mutex); retire_submits(gpu); @@ -472,6 +514,7 @@ void msm_gpu_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, { struct drm_device *dev = gpu->dev; struct msm_drm_private *priv = dev->dev_private; + struct msm_ringbuffer *ring = submit->ring; int i;

WARN_ON(!mutex_is_locked(&dev->struct_mutex)); @@ -480,7 +523,9 @@ void msm_gpu_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,

msm_gpu_hw_init(gpu);

- list_add_tail(&submit->node, &gpu->submit_list); + submit->seqno = ++ring->seqno; + + list_add_tail(&submit->node, &ring->submits);

msm_rd_dump_submit(submit);

@@ -605,7 +650,9 @@ int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev, struct msm_gpu *gpu, const struct msm_gpu_funcs *funcs, const char *name, struct msm_gpu_config *config) { - int ret; + int i, ret, nr_rings = config->nr_rings; + void *memptrs; + uint64_t memptrs_iova;

if (WARN_ON(gpu->num_perfcntrs > ARRAY_SIZE(gpu->last_cntrs))) gpu->num_perfcntrs = ARRAY_SIZE(gpu->last_cntrs); @@ -613,18 +660,11 @@ int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev, gpu->dev = drm; gpu->funcs = funcs; gpu->name = name; - gpu->fctx = msm_fence_context_alloc(drm, name); - if (IS_ERR(gpu->fctx)) { - ret = PTR_ERR(gpu->fctx); - gpu->fctx = NULL; - goto fail; - }

INIT_LIST_HEAD(&gpu->active_list); INIT_WORK(&gpu->retire_work, retire_worker); INIT_WORK(&gpu->recover_work, recover_worker);

- INIT_LIST_HEAD(&gpu->submit_list);

setup_timer(&gpu->hangcheck_timer, hangcheck_handler, (unsigned long)gpu); @@ -689,29 +729,47 @@ int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev, goto fail; }

- gpu->memptrs = msm_gem_kernel_new(drm, sizeof(*gpu->memptrs_bo), + memptrs = msm_gem_kernel_new(drm, sizeof(*gpu->memptrs_bo), MSM_BO_UNCACHED, gpu->aspace, &gpu->memptrs_bo, - &gpu->memptrs_iova); + &memptrs_iova);

- if (IS_ERR(gpu->memptrs)) { - ret = PTR_ERR(gpu->memptrs); - gpu->memptrs = NULL; + if (IS_ERR(memptrs)) { + ret = PTR_ERR(memptrs); dev_err(drm->dev, "could not allocate memptrs: %d\n", ret); goto fail; }

- /* Create ringbuffer: */ - gpu->rb = msm_ringbuffer_new(gpu, config->ringsz); - if (IS_ERR(gpu->rb)) { - ret = PTR_ERR(gpu->rb); - gpu->rb = NULL; - dev_err(drm->dev, "could not create ringbuffer: %d\n", ret); - goto fail; + if (nr_rings > ARRAY_SIZE(gpu->rb)) { + DRM_DEV_INFO_ONCE(drm->dev, "Only creating %lu ringbuffers\n", + ARRAY_SIZE(gpu->rb)); + nr_rings = ARRAY_SIZE(gpu->rb); }

+ /* Create ringbuffer(s): */ + for (i = 0; i < nr_rings; i++) { + gpu->rb[i] = msm_ringbuffer_new(gpu, i, memptrs, memptrs_iova); + + if (IS_ERR(gpu->rb[i])) { + ret = PTR_ERR(gpu->rb[i]); + dev_err(drm->dev, + "could not create ringbuffer %d: %d\n", i, ret); + goto fail; + } + + memptrs += sizeof(struct msm_rbmemptrs); + memptrs_iova += sizeof(struct msm_rbmemptrs); + } + + gpu->nr_rings = nr_rings; + return 0;

fail: + for (i = 0; i < ARRAY_SIZE(gpu->rb); i++) { + msm_ringbuffer_destroy(gpu->rb[i]); + gpu->rb[i] = NULL; + } + if (gpu->memptrs_bo) { msm_gem_put_vaddr(gpu->memptrs_bo); msm_gem_put_iova(gpu->memptrs_bo, gpu->aspace); @@ -724,16 +782,17 @@ int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev,

void msm_gpu_cleanup(struct msm_gpu *gpu) { + int i; + DBG("%s", gpu->name);

WARN_ON(!list_empty(&gpu->active_list));

bs_fini(gpu);

- if (gpu->rb) { - if (gpu->rb_iova) - msm_gem_put_iova(gpu->rb->bo, gpu->aspace); - msm_ringbuffer_destroy(gpu->rb); + for (i = 0; i < ARRAY_SIZE(gpu->rb); i++) { + msm_ringbuffer_destroy(gpu->rb[i]); + gpu->rb[i] = NULL; }

if (gpu->memptrs_bo) { diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h index 8ddda05..1be0317 100644 --- a/drivers/gpu/drm/msm/msm_gpu.h +++ b/drivers/gpu/drm/msm/msm_gpu.h @@ -33,7 +33,7 @@ struct msm_gpu_config { const char *irqname; uint64_t va_start; uint64_t va_end; - unsigned int ringsz; + unsigned int nr_rings; };

/* So far, with hardware that I've seen to date, we can have: @@ -57,8 +57,9 @@ struct msm_gpu_funcs { int (*pm_resume)(struct msm_gpu *gpu); void (*submit)(struct msm_gpu *gpu, struct msm_gem_submit *submit, struct msm_file_private *ctx); - void (*flush)(struct msm_gpu *gpu); + void (*flush)(struct msm_gpu *gpu, struct msm_ringbuffer *ring); irqreturn_t (*irq)(struct msm_gpu *irq); + struct msm_ringbuffer *(*active_ring)(struct msm_gpu *gpu); void (*recover)(struct msm_gpu *gpu); void (*destroy)(struct msm_gpu *gpu); #ifdef CONFIG_DEBUG_FS @@ -67,14 +68,6 @@ struct msm_gpu_funcs { #endif };

-#define rbmemptr(gpu, member) \ - ((gpu)->memptrs_iova + offsetof(struct msm_rbmemptrs, member)) - -struct msm_rbmemptrs { - volatile uint32_t rptr; - volatile uint32_t fence; -}; - struct msm_gpu { const char *name; struct drm_device *dev; @@ -93,16 +86,12 @@ struct msm_gpu { const struct msm_gpu_perfcntr *perfcntrs; uint32_t num_perfcntrs;

- /* ringbuffer: */ - struct msm_ringbuffer *rb; - uint64_t rb_iova; + struct msm_ringbuffer *rb[MSM_GPU_MAX_RINGS]; + int nr_rings;

/* list of GEM active objects: */ struct list_head active_list;

- /* fencing: */ - struct msm_fence_context *fctx; - /* does gpu need hw_init? */ bool needs_hw_init;

@@ -133,21 +122,26 @@ struct msm_gpu { #define DRM_MSM_HANGCHECK_PERIOD 500 /* in ms */ #define DRM_MSM_HANGCHECK_JIFFIES msecs_to_jiffies(DRM_MSM_HANGCHECK_PERIOD) struct timer_list hangcheck_timer; - uint32_t hangcheck_fence; struct work_struct recover_work;

- struct list_head submit_list; - - struct msm_rbmemptrs *memptrs; struct drm_gem_object *memptrs_bo; - uint64_t memptrs_iova; - - };

+/* It turns out that all targets use the same ringbuffer size */ +#define MSM_GPU_RINGBUFFER_SZ SZ_32K + static inline bool msm_gpu_active(struct msm_gpu *gpu) { - return gpu->fctx->last_fence > gpu->memptrs->fence; + int i; + + for (i = 0; i < gpu->nr_rings; i++) { + struct msm_ringbuffer *ring = gpu->rb[i]; + + if (ring->seqno > ring->memptrs->fence) + return true; + } + + return false; }

/* Perf-Counters: diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.c b/drivers/gpu/drm/msm/msm_ringbuffer.c index bf065a5..febf2541 100644 --- a/drivers/gpu/drm/msm/msm_ringbuffer.c +++ b/drivers/gpu/drm/msm/msm_ringbuffer.c @@ -18,13 +18,15 @@ #include "msm_ringbuffer.h" #include "msm_gpu.h"

-struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int size) +struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int id, + void *memptrs, uint64_t memptrs_iova) { struct msm_ringbuffer *ring; + char name[32]; int ret;

- if (WARN_ON(!is_power_of_2(size))) - return ERR_PTR(-EINVAL); + /* We assume everwhere that MSM_GPU_RINGBUFFER_SZ is a power of 2 */ + BUILD_BUG_ON(!is_power_of_2(MSM_GPU_RINGBUFFER_SZ));

ring = kzalloc(sizeof(*ring), GFP_KERNEL); if (!ring) { @@ -33,32 +35,45 @@ struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int size) }

ring->gpu = gpu; - + ring->id = id; /* Pass NULL for the iova pointer - we will map it later */ - ring->start = msm_gem_kernel_new(gpu->dev, size, MSM_BO_WC, - gpu->aspace, &ring->bo, NULL); + ring->start = msm_gem_kernel_new(gpu->dev, MSM_GPU_RINGBUFFER_SZ, + MSM_BO_WC, gpu->aspace, &ring->bo, NULL);

if (IS_ERR(ring->start)) { ret = PTR_ERR(ring->start); ring->start = 0; goto fail; } - ring->end = ring->start + (size / 4); + ring->end = ring->start + (MSM_GPU_RINGBUFFER_SZ >> 2); ring->cur = ring->start;

- ring->size = size; + ring->memptrs = memptrs; + ring->memptrs_iova = memptrs_iova; + + INIT_LIST_HEAD(&ring->submits); + + snprintf(name, sizeof(name), "gpu-ring-%d", ring->id); + + /* FIXME: need struct drm_device */ + ring->fctx = msm_fence_context_alloc(gpu->dev, name);

return ring;

fail: - if (ring) - msm_ringbuffer_destroy(ring); + msm_ringbuffer_destroy(ring); return ERR_PTR(ret); }

void msm_ringbuffer_destroy(struct msm_ringbuffer *ring) { + if (IS_ERR_OR_NULL(ring)) + return; + + msm_fence_context_free(ring->fctx); + if (ring->bo) { + msm_gem_put_iova(ring->bo, ring->gpu->aspace); msm_gem_put_vaddr(ring->bo); drm_gem_object_unreference_unlocked(ring->bo); } diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.h b/drivers/gpu/drm/msm/msm_ringbuffer.h index 6e0e104..ec44251 100644 --- a/drivers/gpu/drm/msm/msm_ringbuffer.h +++ b/drivers/gpu/drm/msm/msm_ringbuffer.h @@ -20,14 +20,30 @@

#include "msm_drv.h"

+#define rbmemptr(ring, member) \ + ((ring)->memptrs_iova + offsetof(struct msm_rbmemptrs, member)) + +struct msm_rbmemptrs { + volatile uint32_t rptr; + volatile uint32_t fence; +}; + struct msm_ringbuffer { struct msm_gpu *gpu; - int size; + int id; struct drm_gem_object *bo; uint32_t *start, *end, *cur; + struct list_head submits; + uint64_t iova; + uint32_t seqno; + uint32_t hangcheck_fence; + struct msm_rbmemptrs *memptrs; + uint64_t memptrs_iova; + struct msm_fence_context *fctx; };

-struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int size); +struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int id, + void *memptrs, uint64_t memptrs_iova); void msm_ringbuffer_destroy(struct msm_ringbuffer *ring);

/* ringbuffer helpers (the parts that are same for a3xx/a2xx/z180..) */ diff --git a/drivers/gpu/drm/msm/msm_submitqueue.c b/drivers/gpu/drm/msm/msm_submitqueue.c index 593c3b5..5115f75 100644 --- a/drivers/gpu/drm/msm/msm_submitqueue.c +++ b/drivers/gpu/drm/msm/msm_submitqueue.c @@ -60,9 +60,10 @@ void msm_submitqueue_close(struct msm_file_private *ctx) msm_submitqueue_put(entry); }

-int msm_submitqueue_create(struct msm_file_private *ctx, u32 prio, u32 flags, - u32 *id) +int msm_submitqueue_create(struct drm_device *drm, struct msm_file_private *ctx, + u32 prio, u32 flags, u32 *id) { + struct msm_drm_private *priv = drm->dev_private; struct msm_gpu_submitqueue *queue;

if (!ctx) @@ -75,7 +76,13 @@ int msm_submitqueue_create(struct msm_file_private *ctx, u32 prio, u32 flags,

kref_init(&queue->ref); queue->flags = flags; - queue->prio = prio; + + if (priv->gpu) { + if (prio >= priv->gpu->nr_rings) + return -EINVAL; + + queue->prio = prio; + }

write_lock(&ctx->queuelock);

@@ -91,16 +98,26 @@ int msm_submitqueue_create(struct msm_file_private *ctx, u32 prio, u32 flags, return 0; }

-int msm_submitqueue_init(struct msm_file_private *ctx) +int msm_submitqueue_init(struct drm_device *drm, struct msm_file_private *ctx) { + struct msm_drm_private *priv = drm->dev_private; + int default_prio; + if (!ctx) return 0;

+ /* + * Select priority 2 as the "default priority" unless nr_rings is less + * than 2 and then pick the lowest pirority + */ + default_prio = priv->gpu ? + clamp_t(uint32_t, 2, 0, priv->gpu->nr_rings - 1) : 0; + INIT_LIST_HEAD(&ctx->submitqueues);

rwlock_init(&ctx->queuelock);

- return msm_submitqueue_create(ctx, 2, 0, NULL); + return msm_submitqueue_create(drm, ctx, default_prio, 0, NULL); }

int msm_submitqueue_remove(struct msm_file_private *ctx, u32 id) diff --git a/include/uapi/drm/msm_drm.h b/include/uapi/drm/msm_drm.h index 9fb4b81..42fa781 100644 --- a/include/uapi/drm/msm_drm.h +++ b/include/uapi/drm/msm_drm.h @@ -232,6 +232,7 @@ struct drm_msm_wait_fence { __u32 fence; /* in */ __u32 pad; struct drm_msm_timespec timeout; /* in */ + __u32 queueid; /* in, submitqueue id */ };

/* madvise provides a way to tell the kernel in case a buffers contents

Add a shadow pointer to track the current command being written into the ring. Don't commit it as 'cur' until the command is submitted. Because 'cur' is used to construct the software copy of the wptr this ensures that somebody peeking in on the ring doesn't assume that a command is inflight while it is being written. This isn't a huge deal with a single ring (though technically the hangcheck could assume the system is prematurely busy when it isn't) but it will be rather important for preemption where the decision to preempt is based on a non-empty ringbuffer. Without a shadow an aggressive preemption scheme could assume that the ringbuffer is non empty and switch to it before the CPU is done writing the command and boom.

Even though preemption won't be supported for all targets because of the way the code is organized it is simpler to make this generic for all targets. The extra load for non-preemption targets should be minimal.

Signed-off-by: Jordan Crouse jcrouse@codeaurora.org --- drivers/gpu/drm/msm/adreno/adreno_gpu.c | 9 +++++++-- drivers/gpu/drm/msm/msm_ringbuffer.c | 1 + drivers/gpu/drm/msm/msm_ringbuffer.h | 12 ++++++++---- 3 files changed, 16 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c index cf91840..77eaa46 100644 --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c @@ -89,6 +89,7 @@ int adreno_hw_init(struct msm_gpu *gpu) }

ring->cur = ring->start; + ring->next = ring->start;

/* reset completed fence seqno: */ ring->memptrs->fence = ring->seqno; @@ -233,12 +234,15 @@ void adreno_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring) struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); uint32_t wptr;

+ /* Copy the shadow to the actual register */ + ring->cur = ring->next; + /* * Mask wptr value that we calculate to fit in the HW range. This is * to account for the possibility that the last command fit exactly into * the ringbuffer and rb->next hasn't wrapped to zero yet */ - wptr = get_wptr(ring) % (MSM_GPU_RINGBUFFER_SZ >> 2); + wptr = (ring->cur - ring->start) % (MSM_GPU_RINGBUFFER_SZ >> 2);

/* ensure writes to ringbuffer have hit system memory: */ mb(); @@ -350,7 +354,8 @@ static uint32_t ring_freewords(struct msm_ringbuffer *ring) { struct adreno_gpu *adreno_gpu = to_adreno_gpu(ring->gpu); uint32_t size = MSM_GPU_RINGBUFFER_SZ >> 2; - uint32_t wptr = get_wptr(ring); + /* Use ring->next to calculate free size */ + uint32_t wptr = ring->next - ring->start; uint32_t rptr = get_rptr(adreno_gpu, ring); return (rptr + (size - 1) - wptr) % size; } diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.c b/drivers/gpu/drm/msm/msm_ringbuffer.c index febf2541..6015959 100644 --- a/drivers/gpu/drm/msm/msm_ringbuffer.c +++ b/drivers/gpu/drm/msm/msm_ringbuffer.c @@ -46,6 +46,7 @@ struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int id, goto fail; } ring->end = ring->start + (MSM_GPU_RINGBUFFER_SZ >> 2); + ring->next = ring->start; ring->cur = ring->start;

ring->memptrs = memptrs; diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.h b/drivers/gpu/drm/msm/msm_ringbuffer.h index ec44251..3749764 100644 --- a/drivers/gpu/drm/msm/msm_ringbuffer.h +++ b/drivers/gpu/drm/msm/msm_ringbuffer.h @@ -32,7 +32,7 @@ struct msm_ringbuffer { struct msm_gpu *gpu; int id; struct drm_gem_object *bo; - uint32_t *start, *end, *cur; + uint32_t *start, *end, *cur, *next; struct list_head submits; uint64_t iova; uint32_t seqno; @@ -51,9 +51,13 @@ struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int id, static inline void OUT_RING(struct msm_ringbuffer *ring, uint32_t data) { - if (ring->cur == ring->end) - ring->cur = ring->start; - *(ring->cur++) = data; + /* + * ring->next points to the current command being written - it won't be + * committed as ring->cur until the flush + */ + if (ring->next == ring->end) + ring->next = ring->start; + *(ring->next++) = data; }

#endif /* __MSM_RINGBUFFER_H__ */

Implement preemption for A5XX targets - this allows multiple ringbuffers for different priorities with automatic preemption of a lower priority ringbuffer if a higher one is ready.

Signed-off-by: Jordan Crouse jcrouse@codeaurora.org --- drivers/gpu/drm/msm/Makefile | 1 + drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 176 ++++++++++++++++- drivers/gpu/drm/msm/adreno/a5xx_gpu.h | 107 ++++++++++- drivers/gpu/drm/msm/adreno/a5xx_preempt.c | 305 ++++++++++++++++++++++++++++++ drivers/gpu/drm/msm/adreno/adreno_gpu.c | 14 +- drivers/gpu/drm/msm/adreno/adreno_gpu.h | 7 +- drivers/gpu/drm/msm/msm_drv.h | 2 +- drivers/gpu/drm/msm/msm_gpu.c | 5 +- drivers/gpu/drm/msm/msm_ringbuffer.c | 1 + drivers/gpu/drm/msm/msm_ringbuffer.h | 1 + 10 files changed, 599 insertions(+), 20 deletions(-) create mode 100644 drivers/gpu/drm/msm/adreno/a5xx_preempt.c

diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile index 3c234e7..d0b26dd 100644 --- a/drivers/gpu/drm/msm/Makefile +++ b/drivers/gpu/drm/msm/Makefile @@ -8,6 +8,7 @@ msm-y := \ adreno/a4xx_gpu.o \ adreno/a5xx_gpu.o \ adreno/a5xx_power.o \ + adreno/a5xx_preempt.o \ hdmi/hdmi.o \ hdmi/hdmi_audio.o \ hdmi/hdmi_bridge.o \ diff --git a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c index b7b9f4f..a0f69b2 100644 --- a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c +++ b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c @@ -95,13 +95,65 @@ static int zap_shader_load_mdt(struct device *dev, const char *fwname) return ret; }

+static void a5xx_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring) +{ + struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); + struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu); + uint32_t wptr; + unsigned long flags; + + spin_lock_irqsave(&ring->lock, flags); + + /* Copy the shadow to the actual register */ + ring->cur = ring->next; + + /* Make sure to wrap wptr if we need to */ + wptr = get_wptr(ring); + + spin_unlock_irqrestore(&ring->lock, flags); + + /* Make sure everything is posted before making a decision */ + mb(); + + /* Update HW if this is the current ring and we are not in preempt */ + if (a5xx_gpu->cur_ring == ring && !a5xx_in_preempt(a5xx_gpu)) + gpu_write(gpu, REG_A5XX_CP_RB_WPTR, wptr); +} + static void a5xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, struct msm_file_private *ctx) { + struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); + struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu); struct msm_drm_private *priv = gpu->dev->dev_private; struct msm_ringbuffer *ring = submit->ring; unsigned int i, ibs = 0;

+ OUT_PKT7(ring, CP_PREEMPT_ENABLE_GLOBAL, 1); + OUT_RING(ring, 0x02); + + /* Turn off protected mode to write to special registers */ + OUT_PKT7(ring, CP_SET_PROTECTED_MODE, 1); + OUT_RING(ring, 0); + + /* Set the save preemption record for the ring/command */ + OUT_PKT4(ring, REG_A5XX_CP_CONTEXT_SWITCH_SAVE_ADDR_LO, 2); + OUT_RING(ring, lower_32_bits(a5xx_gpu->preempt_iova[submit->ring->id])); + OUT_RING(ring, upper_32_bits(a5xx_gpu->preempt_iova[submit->ring->id])); + + /* Turn back on protected mode */ + OUT_PKT7(ring, CP_SET_PROTECTED_MODE, 1); + OUT_RING(ring, 1); + + /* Enable local preemption for finegrain preemption */ + OUT_PKT7(ring, CP_PREEMPT_ENABLE_GLOBAL, 1); + OUT_RING(ring, 0x02); + + /* Allow CP_CONTEXT_SWITCH_YIELD packets in the IB2 */ + OUT_PKT7(ring, CP_YIELD_ENABLE, 1); + OUT_RING(ring, 0x02); + + /* Submit the commands */ for (i = 0; i < submit->nr_cmds; i++) { switch (submit->cmd[i].type) { case MSM_SUBMIT_CMD_IB_TARGET_BUF: @@ -119,16 +171,54 @@ static void a5xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, } }

+ /* + * Write the render mode to NULL (0) to indicate to the CP that the IBs + * are done rendering - otherwise a lucky preemption would start + * replaying from the last checkpoint + */ + OUT_PKT7(ring, CP_SET_RENDER_MODE, 5); + OUT_RING(ring, 0); + OUT_RING(ring, 0); + OUT_RING(ring, 0); + OUT_RING(ring, 0); + OUT_RING(ring, 0); + + /* Turn off IB level preemptions */ + OUT_PKT7(ring, CP_YIELD_ENABLE, 1); + OUT_RING(ring, 0x01); + + /* Write the fence to the scratch register */ OUT_PKT4(ring, REG_A5XX_CP_SCRATCH_REG(2), 1); OUT_RING(ring, submit->seqno);

+ /* + * Execute a CACHE_FLUSH_TS event. This will ensure that the + * timestamp is written to the memory and then triggers the interrupt + */ OUT_PKT7(ring, CP_EVENT_WRITE, 4); OUT_RING(ring, CACHE_FLUSH_TS | (1 << 31)); OUT_RING(ring, lower_32_bits(rbmemptr(ring, fence))); OUT_RING(ring, upper_32_bits(rbmemptr(ring, fence))); OUT_RING(ring, submit->seqno);

- gpu->funcs->flush(gpu, ring); + /* Yield the floor on command completion */ + OUT_PKT7(ring, CP_CONTEXT_SWITCH_YIELD, 4); + /* + * If dword[2:1] are non zero, they specify an address for the CP to + * write the value of dword[3] to on preemption complete. Write 0 to + * skip the write + */ + OUT_RING(ring, 0x00); + OUT_RING(ring, 0x00); + /* Data value - not used if the address above is 0 */ + OUT_RING(ring, 0x01); + /* Set bit 0 to trigger an interrupt on preempt complete */ + OUT_RING(ring, 0x01); + + a5xx_flush(gpu, ring); + + /* Check to see if we need to start preemption */ + a5xx_preempt_trigger(gpu); }

static const struct { @@ -279,6 +369,50 @@ static int a5xx_me_init(struct msm_gpu *gpu) return a5xx_idle(gpu, ring) ? 0 : -EINVAL; }

+static int a5xx_preempt_start(struct msm_gpu *gpu) +{ + struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); + struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu); + struct msm_ringbuffer *ring = gpu->rb[0]; + + if (gpu->nr_rings == 1) + return 0; + + /* Turn off protected mode to write to special registers */ + OUT_PKT7(ring, CP_SET_PROTECTED_MODE, 1); + OUT_RING(ring, 0); + + /* Set the save preemption record for the ring/command */ + OUT_PKT4(ring, REG_A5XX_CP_CONTEXT_SWITCH_SAVE_ADDR_LO, 2); + OUT_RING(ring, lower_32_bits(a5xx_gpu->preempt_iova[ring->id])); + OUT_RING(ring, upper_32_bits(a5xx_gpu->preempt_iova[ring->id])); + + /* Turn back on protected mode */ + OUT_PKT7(ring, CP_SET_PROTECTED_MODE, 1); + OUT_RING(ring, 1); + + OUT_PKT7(ring, CP_PREEMPT_ENABLE_GLOBAL, 1); + OUT_RING(ring, 0x00); + + OUT_PKT7(ring, CP_PREEMPT_ENABLE_LOCAL, 1); + OUT_RING(ring, 0x01); + + OUT_PKT7(ring, CP_YIELD_ENABLE, 1); + OUT_RING(ring, 0x01); + + /* Yield the floor on command completion */ + OUT_PKT7(ring, CP_CONTEXT_SWITCH_YIELD, 4); + OUT_RING(ring, 0x00); + OUT_RING(ring, 0x00); + OUT_RING(ring, 0x01); + OUT_RING(ring, 0x01); + + gpu->funcs->flush(gpu, ring); + + return a5xx_idle(gpu, ring) ? 0 : -EINVAL; +} + + static struct drm_gem_object *a5xx_ucode_load_bo(struct msm_gpu *gpu, const struct firmware *fw, u64 *iova) { @@ -394,6 +528,7 @@ static int a5xx_zap_shader_init(struct msm_gpu *gpu) A5XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNC_OVERFLOW | \ A5XX_RBBM_INT_0_MASK_CP_HW_ERROR | \ A5XX_RBBM_INT_0_MASK_MISC_HANG_DETECT | \ + A5XX_RBBM_INT_0_MASK_CP_SW | \ A5XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS | \ A5XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS | \ A5XX_RBBM_INT_0_MASK_GPMU_VOLTAGE_DROOP) @@ -541,6 +676,8 @@ static int a5xx_hw_init(struct msm_gpu *gpu) if (ret) return ret;

+ a5xx_preempt_hw_init(gpu); + ret = a5xx_ucode_init(gpu); if (ret) return ret; @@ -593,6 +730,9 @@ static int a5xx_hw_init(struct msm_gpu *gpu) gpu_write(gpu, REG_A5XX_RBBM_SECVID_TRUST_CNTL, 0x0); }

+ /* Last step - yield the ringbuffer */ + a5xx_preempt_start(gpu); + return 0; }

@@ -623,6 +763,8 @@ static void a5xx_destroy(struct msm_gpu *gpu)

DBG("%s", gpu->name);

+ a5xx_preempt_fini(gpu); + if (a5xx_gpu->pm4_bo) { if (a5xx_gpu->pm4_iova) msm_gem_put_iova(a5xx_gpu->pm4_bo, gpu->aspace); @@ -660,6 +802,14 @@ static inline bool _a5xx_check_idle(struct msm_gpu *gpu)

bool a5xx_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring) { + struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); + struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu); + + if (ring != a5xx_gpu->cur_ring) { + WARN(1, "Tried to idle a non-current ringbuffer\n"); + return false; + } + /* wait for CP to drain ringbuffer: */ if (!adreno_idle(gpu, ring)) return false; @@ -854,8 +1004,13 @@ static irqreturn_t a5xx_irq(struct msm_gpu *gpu) if (status & A5XX_RBBM_INT_0_MASK_GPMU_VOLTAGE_DROOP) a5xx_gpmu_err_irq(gpu);

- if (status & A5XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS) + if (status & A5XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS) { + a5xx_preempt_trigger(gpu); msm_gpu_retire(gpu); + } + + if (status & A5XX_RBBM_INT_0_MASK_CP_SW) + a5xx_preempt_irq(gpu);

return IRQ_HANDLED; } @@ -985,6 +1140,14 @@ static void a5xx_show(struct msm_gpu *gpu, struct seq_file *m) } #endif

+static struct msm_ringbuffer *a5xx_active_ring(struct msm_gpu *gpu) +{ + struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); + struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu); + + return a5xx_gpu->cur_ring; +} + static const struct adreno_gpu_funcs funcs = { .base = { .get_param = adreno_get_param, @@ -993,8 +1156,8 @@ static void a5xx_show(struct msm_gpu *gpu, struct seq_file *m) .pm_resume = a5xx_pm_resume, .recover = a5xx_recover, .submit = a5xx_submit, - .flush = adreno_flush, - .active_ring = adreno_active_ring, + .flush = a5xx_flush, + .active_ring = a5xx_active_ring, .irq = a5xx_irq, .destroy = a5xx_destroy, #ifdef CONFIG_DEBUG_FS @@ -1030,7 +1193,7 @@ struct msm_gpu *a5xx_gpu_init(struct drm_device *dev)

a5xx_gpu->lm_leakage = 0x4E001A;

- ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs, 1); + ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs, 4); if (ret) { a5xx_destroy(&(a5xx_gpu->base.base)); return ERR_PTR(ret); @@ -1039,5 +1202,8 @@ struct msm_gpu *a5xx_gpu_init(struct drm_device *dev) if (gpu->aspace) msm_mmu_set_fault_handler(gpu->aspace->mmu, gpu, a5xx_fault_handler);

+ /* Set up the preemption specific bits and pieces for each ringbuffer */ + a5xx_preempt_init(gpu); + return gpu; } diff --git a/drivers/gpu/drm/msm/adreno/a5xx_gpu.h b/drivers/gpu/drm/msm/adreno/a5xx_gpu.h index 44db48d..6fb8c2f 100644 --- a/drivers/gpu/drm/msm/adreno/a5xx_gpu.h +++ b/drivers/gpu/drm/msm/adreno/a5xx_gpu.h @@ -1,4 +1,4 @@ -/* Copyright (c) 2016 The Linux Foundation. All rights reserved. +/* Copyright (c) 2016-2017 The Linux Foundation. All rights reserved. * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 and @@ -35,10 +35,100 @@ struct a5xx_gpu { uint32_t gpmu_dwords;

uint32_t lm_leakage; + + struct msm_ringbuffer *cur_ring; + struct msm_ringbuffer *next_ring; + + struct drm_gem_object *preempt_bo[MSM_GPU_MAX_RINGS]; + struct a5xx_preempt_record *preempt[MSM_GPU_MAX_RINGS]; + uint64_t preempt_iova[MSM_GPU_MAX_RINGS]; + + atomic_t preempt_state; + struct timer_list preempt_timer; };

#define to_a5xx_gpu(x) container_of(x, struct a5xx_gpu, base)

+/* + * In order to do lockless preemption we use a simple state machine to progress + * through the process. + * + * PREEMPT_NONE - no preemption in progress. Next state START. + * PREEMPT_START - The trigger is evaulating if preemption is possible. Next + * states: TRIGGERED, NONE + * PREEMPT_ABORT - An intermediate state before moving back to NONE. Next + * state: NONE. + * PREEMPT_TRIGGERED: A preemption has been executed on the hardware. Next + * states: FAULTED, PENDING + * PREEMPT_FAULTED: A preemption timed out (never completed). This will trigger + * recovery. Next state: N/A + * PREEMPT_PENDING: Preemption complete interrupt fired - the callback is + * checking the success of the operation. Next state: FAULTED, NONE. + */ + +enum preempt_state { + PREEMPT_NONE = 0, + PREEMPT_START, + PREEMPT_ABORT, + PREEMPT_TRIGGERED, + PREEMPT_FAULTED, + PREEMPT_PENDING, +}; + +/* + * struct a5xx_preempt_record is a shared buffer between the microcode and the + * CPU to store the state for preemption. The record itself is much larger + * (64k) but most of that is used by the CP for storage. + * + * There is a preemption record assigned per ringbuffer. When the CPU triggers a + * preemption, it fills out the record with the useful information (wptr, ring + * base, etc) and the microcode uses that information to set up the CP following + * the preemption. When a ring is switched out, the CP will save the ringbuffer + * state back to the record. In this way, once the records are properly set up + * the CPU can quickly switch back and forth between ringbuffers by only + * updating a few registers (often only the wptr). + * + * These are the CPU aware registers in the record: + * @magic: Must always be 0x27C4BAFC + * @info: Type of the record - written 0 by the CPU, updated by the CP + * @data: Data field from SET_RENDER_MODE or a checkpoint. Written and used by + * the CP + * @cntl: Value of RB_CNTL written by CPU, save/restored by CP + * @rptr: Value of RB_RPTR written by CPU, save/restored by CP + * @wptr: Value of RB_WPTR written by CPU, save/restored by CP + * @rptr_addr: Value of RB_RPTR_ADDR written by CPU, save/restored by CP + * @rbase: Value of RB_BASE written by CPU, save/restored by CP + * @counter: GPU address of the storage area for the performance counters + */ +struct a5xx_preempt_record { + uint32_t magic; + uint32_t info; + uint32_t data; + uint32_t cntl; + uint32_t rptr; + uint32_t wptr; + uint64_t rptr_addr; + uint64_t rbase; + uint64_t counter; +}; + +/* Magic identifier for the preemption record */ +#define A5XX_PREEMPT_RECORD_MAGIC 0x27C4BAFCUL + +/* + * Even though the structure above is only a few bytes, we need a full 64k to + * store the entire preemption record from the CP + */ +#define A5XX_PREEMPT_RECORD_SIZE (64 * 1024) + +/* + * The preemption counter block is a storage area for the value of the + * preemption counters that are saved immediately before context switch. We + * append it on to the end of the allocation for the preemption record. + */ +#define A5XX_PREEMPT_COUNTER_SIZE (16 * 4) + + int a5xx_power_init(struct msm_gpu *gpu); void a5xx_gpmu_ucode_init(struct msm_gpu *gpu);

@@ -58,4 +148,19 @@ static inline int spin_usecs(struct msm_gpu *gpu, uint32_t usecs, bool a5xx_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring); void a5xx_set_hwcg(struct msm_gpu *gpu, bool state);

+void a5xx_preempt_init(struct msm_gpu *gpu); +void a5xx_preempt_hw_init(struct msm_gpu *gpu); +void a5xx_preempt_trigger(struct msm_gpu *gpu); +void a5xx_preempt_irq(struct msm_gpu *gpu); +void a5xx_preempt_fini(struct msm_gpu *gpu); + +/* Return true if we are in a preempt state */ +static inline bool a5xx_in_preempt(struct a5xx_gpu *a5xx_gpu) +{ + int preempt_state = atomic_read(&a5xx_gpu->preempt_state); + + return !(preempt_state == PREEMPT_NONE || + preempt_state == PREEMPT_ABORT); +} + #endif /* __A5XX_GPU_H__ */ diff --git a/drivers/gpu/drm/msm/adreno/a5xx_preempt.c b/drivers/gpu/drm/msm/adreno/a5xx_preempt.c new file mode 100644 index 0000000..40f4840 --- /dev/null +++ b/drivers/gpu/drm/msm/adreno/a5xx_preempt.c @@ -0,0 +1,305 @@ +/* Copyright (c) 2017 The Linux Foundation. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 and + * only version 2 as published by the Free Software Foundation. + * + * This program is distributed in the hope that 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. + * + */ + +#include "msm_gem.h" +#include "a5xx_gpu.h" + +/* + * Try to transition the preemption state from old to new. Return + * true on success or false if the original state wasn't 'old' + */ +static inline bool try_preempt_state(struct a5xx_gpu *a5xx_gpu, + enum preempt_state old, enum preempt_state new) +{ + enum preempt_state cur = atomic_cmpxchg(&a5xx_gpu->preempt_state, + old, new); + + return (cur == old); +} + +/* + * Force the preemption state to the specified state. This is used in cases + * where the current state is known and won't change + */ +static inline void set_preempt_state(struct a5xx_gpu *gpu, + enum preempt_state new) +{ + /* + * preempt_state may be read by other cores trying to trigger a + * preemption or in the interrupt handler so barriers are needed + * before... + */ + smp_mb__before_atomic(); + atomic_set(&gpu->preempt_state, new); + /* ... and after*/ + smp_mb__after_atomic(); +} + +/* Write the most recent wptr for the given ring into the hardware */ +static inline void update_wptr(struct msm_gpu *gpu, struct msm_ringbuffer *ring) +{ + unsigned long flags; + uint32_t wptr; + + if (!ring) + return; + + spin_lock_irqsave(&ring->lock, flags); + wptr = get_wptr(ring); + spin_unlock_irqrestore(&ring->lock, flags); + + gpu_write(gpu, REG_A5XX_CP_RB_WPTR, wptr); +} + +/* Return the highest priority ringbuffer with something in it */ +static struct msm_ringbuffer *get_next_ring(struct msm_gpu *gpu) +{ + unsigned long flags; + int i; + + for (i = 0; i < gpu->nr_rings; i++) { + bool empty; + struct msm_ringbuffer *ring = gpu->rb[i]; + + spin_lock_irqsave(&ring->lock, flags); + empty = (get_wptr(ring) == ring->memptrs->rptr); + spin_unlock_irqrestore(&ring->lock, flags); + + if (!empty) + return ring; + } + + return NULL; +} + +static void a5xx_preempt_timer(unsigned long data) +{ + struct a5xx_gpu *a5xx_gpu = (struct a5xx_gpu *) data; + struct msm_gpu *gpu = &a5xx_gpu->base.base; + struct drm_device *dev = gpu->dev; + struct msm_drm_private *priv = dev->dev_private; + + if (!try_preempt_state(a5xx_gpu, PREEMPT_TRIGGERED, PREEMPT_FAULTED)) + return; + + dev_err(dev->dev, "%s: preemption timed out\n", gpu->name); + queue_work(priv->wq, &gpu->recover_work); +} + +/* Try to trigger a preemption switch */ +void a5xx_preempt_trigger(struct msm_gpu *gpu) +{ + struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); + struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu); + unsigned long flags; + struct msm_ringbuffer *ring; + + if (gpu->nr_rings == 1) + return; + + /* + * Try to start preemption by moving from NONE to START. If + * unsuccessful, a preemption is already in flight + */ + if (!try_preempt_state(a5xx_gpu, PREEMPT_NONE, PREEMPT_START)) + return; + + /* Get the next ring to preempt to */ + ring = get_next_ring(gpu); + + /* + * If no ring is populated or the highest priority ring is the current + * one do nothing except to update the wptr to the latest and greatest + */ + if (!ring || (a5xx_gpu->cur_ring == ring)) { + /* + * Its possible that while a preemption request is in progress + * from an irq context, a user context trying to submit might + * fail to update the write pointer, because it determines + * that the preempt state is not PREEMPT_NONE. + * + * Close the race by introducing an intermediate + * state PREEMPT_ABORT to let the submit path + * know that the ringbuffer is not going to change + * and can safely update the write pointer. + */ + + set_preempt_state(a5xx_gpu, PREEMPT_ABORT); + update_wptr(gpu, a5xx_gpu->cur_ring); + set_preempt_state(a5xx_gpu, PREEMPT_NONE); + return; + } + + /* Make sure the wptr doesn't update while we're in motion */ + spin_lock_irqsave(&ring->lock, flags); + a5xx_gpu->preempt[ring->id]->wptr = get_wptr(ring); + spin_unlock_irqrestore(&ring->lock, flags); + + /* Set the address of the incoming preemption record */ + gpu_write64(gpu, REG_A5XX_CP_CONTEXT_SWITCH_RESTORE_ADDR_LO, + REG_A5XX_CP_CONTEXT_SWITCH_RESTORE_ADDR_HI, + a5xx_gpu->preempt_iova[ring->id]); + + a5xx_gpu->next_ring = ring; + + /* Start a timer to catch a stuck preemption */ + mod_timer(&a5xx_gpu->preempt_timer, jiffies + msecs_to_jiffies(10000)); + + /* Set the preemption state to triggered */ + set_preempt_state(a5xx_gpu, PREEMPT_TRIGGERED); + + /* Make sure everything is written before hitting the button */ + wmb(); + + /* And actually start the preemption */ + gpu_write(gpu, REG_A5XX_CP_CONTEXT_SWITCH_CNTL, 1); +} + +void a5xx_preempt_irq(struct msm_gpu *gpu) +{ + uint32_t status; + struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); + struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu); + struct drm_device *dev = gpu->dev; + struct msm_drm_private *priv = dev->dev_private; + + if (!try_preempt_state(a5xx_gpu, PREEMPT_TRIGGERED, PREEMPT_PENDING)) + return; + + /* Delete the preemption watchdog timer */ + del_timer(&a5xx_gpu->preempt_timer); + + /* + * The hardware should be setting CP_CONTEXT_SWITCH_CNTL to zero before + * firing the interrupt, but there is a non zero chance of a hardware + * condition or a software race that could set it again before we have a + * chance to finish. If that happens, log and go for recovery + */ + status = gpu_read(gpu, REG_A5XX_CP_CONTEXT_SWITCH_CNTL); + if (unlikely(status)) { + set_preempt_state(a5xx_gpu, PREEMPT_FAULTED); + dev_err(dev->dev, "%s: Preemption failed to complete\n", + gpu->name); + queue_work(priv->wq, &gpu->recover_work); + return; + } + + a5xx_gpu->cur_ring = a5xx_gpu->next_ring; + a5xx_gpu->next_ring = NULL; + + update_wptr(gpu, a5xx_gpu->cur_ring); + + set_preempt_state(a5xx_gpu, PREEMPT_NONE); +} + +void a5xx_preempt_hw_init(struct msm_gpu *gpu) +{ + struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); + struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu); + int i; + + for (i = 0; i < gpu->nr_rings; i++) { + a5xx_gpu->preempt[i]->wptr = 0; + a5xx_gpu->preempt[i]->rptr = 0; + a5xx_gpu->preempt[i]->rbase = gpu->rb[i]->iova; + } + + /* Write a 0 to signal that we aren't switching pagetables */ + gpu_write64(gpu, REG_A5XX_CP_CONTEXT_SWITCH_SMMU_INFO_LO, + REG_A5XX_CP_CONTEXT_SWITCH_SMMU_INFO_HI, 0); + + /* Reset the preemption state */ + set_preempt_state(a5xx_gpu, PREEMPT_NONE); + + /* Always come up on rb 0 */ + a5xx_gpu->cur_ring = gpu->rb[0]; +} + +static int preempt_init_ring(struct a5xx_gpu *a5xx_gpu, + struct msm_ringbuffer *ring) +{ + struct adreno_gpu *adreno_gpu = &a5xx_gpu->base; + struct msm_gpu *gpu = &adreno_gpu->base; + struct a5xx_preempt_record *ptr; + struct drm_gem_object *bo = NULL; + u64 iova = 0; + + ptr = msm_gem_kernel_new(gpu->dev, + A5XX_PREEMPT_RECORD_SIZE + A5XX_PREEMPT_COUNTER_SIZE, + MSM_BO_UNCACHED, gpu->aspace, &bo, &iova); + + if (IS_ERR(ptr)) + return PTR_ERR(ptr); + + a5xx_gpu->preempt_bo[ring->id] = bo; + a5xx_gpu->preempt_iova[ring->id] = iova; + a5xx_gpu->preempt[ring->id] = ptr; + + /* Set up the defaults on the preemption record */ + + ptr->magic = A5XX_PREEMPT_RECORD_MAGIC; + ptr->info = 0; + ptr->data = 0; + ptr->cntl = MSM_GPU_RB_CNTL_DEFAULT; + ptr->rptr_addr = rbmemptr(ring, rptr); + ptr->counter = iova + A5XX_PREEMPT_RECORD_SIZE; + + return 0; +} + +void a5xx_preempt_fini(struct msm_gpu *gpu) +{ + struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); + struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu); + int i; + + for (i = 0; i < gpu->nr_rings; i++) { + if (!a5xx_gpu->preempt_bo[i]) + continue; + + msm_gem_put_vaddr(a5xx_gpu->preempt_bo[i]); + + if (a5xx_gpu->preempt_iova[i]) + msm_gem_put_iova(a5xx_gpu->preempt_bo[i], gpu->aspace); + + drm_gem_object_unreference(a5xx_gpu->preempt_bo[i]); + a5xx_gpu->preempt_bo[i] = NULL; + } +} + +void a5xx_preempt_init(struct msm_gpu *gpu) +{ + struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); + struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu); + int i; + + /* No preemption if we only have one ring */ + if (gpu->nr_rings <= 1) + return; + + for (i = 0; i < gpu->nr_rings; i++) { + if (preempt_init_ring(a5xx_gpu, gpu->rb[i])) { + /* + * On any failure our adventure is over. Clean up and + * set nr_rings to 1 to force preemption off + */ + a5xx_preempt_fini(gpu); + gpu->nr_rings = 1; + + return; + } + } + + setup_timer(&a5xx_gpu->preempt_timer, a5xx_preempt_timer, + (unsigned long) a5xx_gpu); +} diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c index 72f1132..1b06eef 100644 --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c @@ -118,11 +118,6 @@ int adreno_hw_init(struct msm_gpu *gpu) return 0; }

-static uint32_t get_wptr(struct msm_ringbuffer *ring) -{ - return ring->cur - ring->start; -} - /* Use this helper to read rptr, since a430 doesn't update rptr in memory */ static uint32_t get_rptr(struct adreno_gpu *adreno_gpu, struct msm_ringbuffer *ring) @@ -177,7 +172,7 @@ void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, case MSM_SUBMIT_CMD_BUF: OUT_PKT3(ring, adreno_is_a430(adreno_gpu) ? CP_INDIRECT_BUFFER_PFE : CP_INDIRECT_BUFFER_PFD, 2); - OUT_RING(ring, submit->cmd[i].iova); + OUT_RING(ring, lower_32_bits(submit->cmd[i].iova)); OUT_RING(ring, submit->cmd[i].size); OUT_PKT2(ring); break; @@ -244,7 +239,7 @@ void adreno_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring) * to account for the possibility that the last command fit exactly into * the ringbuffer and rb->next hasn't wrapped to zero yet */ - wptr = (ring->cur - ring->start) % (MSM_GPU_RINGBUFFER_SZ >> 2); + wptr = get_wptr(ring);

/* ensure writes to ringbuffer have hit system memory: */ mb(); @@ -262,8 +257,9 @@ bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring) return true;

/* TODO maybe we need to reset GPU here to recover from hang? */ - DRM_ERROR("%s: timeout waiting to drain ringbuffer %d!\n", gpu->name, - ring->id); + DRM_ERROR("%s: timeout waiting to drain ringbuffer %d rptr/wptr = %X/%X\n", + gpu->name, ring->id, get_rptr(adreno_gpu, ring), wptr); + return false; }

diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h index 2a51ac4..c81434c 100644 --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h @@ -2,7 +2,7 @@ * Copyright (C) 2013 Red Hat * Author: Rob Clark robdclark@gmail.com * - * Copyright (c) 2014 The Linux Foundation. All rights reserved. + * Copyright (c) 2014,2017 The Linux Foundation. All rights reserved. * * This program is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License version 2 as published by @@ -309,6 +309,11 @@ static inline void adreno_gpu_write64(struct adreno_gpu *gpu, adreno_gpu_write(gpu, hi, upper_32_bits(data)); }

+static inline uint32_t get_wptr(struct msm_ringbuffer *ring) +{ + return (ring->cur - ring->start) % (MSM_GPU_RINGBUFFER_SZ >> 2); +} + /* * Given a register and a count, return a value to program into * REG_CP_PROTECT_REG(n) - this will block both reads and writes for _len diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h index 2821f57..9853e3e 100644 --- a/drivers/gpu/drm/msm/msm_drv.h +++ b/drivers/gpu/drm/msm/msm_drv.h @@ -74,7 +74,7 @@ struct msm_vblank_ctrl { spinlock_t lock; };

-#define MSM_GPU_MAX_RINGS 1 +#define MSM_GPU_MAX_RINGS 4

struct msm_drm_private {

diff --git a/drivers/gpu/drm/msm/msm_gpu.c b/drivers/gpu/drm/msm/msm_gpu.c index 69c62c0..c675439 100644 --- a/drivers/gpu/drm/msm/msm_gpu.c +++ b/drivers/gpu/drm/msm/msm_gpu.c @@ -295,8 +295,7 @@ static void recover_worker(struct work_struct *work) * Replay all remaining submits starting with highest priority * ring */ - - for (i = gpu->nr_rings - 1; i >= 0; i--) { + for (i = 0; i < gpu->nr_rings; i++) { struct msm_ringbuffer *ring = gpu->rb[i];

list_for_each_entry(submit, &ring->submits, node) @@ -476,7 +475,7 @@ static void retire_submits(struct msm_gpu *gpu) WARN_ON(!mutex_is_locked(&dev->struct_mutex));

/* Retire the commits starting with highest priority */ - for (i = gpu->nr_rings - 1; i >= 0; i--) { + for (i = 0; i < gpu->nr_rings; i++) { struct msm_ringbuffer *ring = gpu->rb[i];

list_for_each_entry_safe(submit, tmp, &ring->submits, node) { diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.c b/drivers/gpu/drm/msm/msm_ringbuffer.c index 6015959..d67955d 100644 --- a/drivers/gpu/drm/msm/msm_ringbuffer.c +++ b/drivers/gpu/drm/msm/msm_ringbuffer.c @@ -53,6 +53,7 @@ struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int id, ring->memptrs_iova = memptrs_iova;

INIT_LIST_HEAD(&ring->submits); + spin_lock_init(&ring->lock);

snprintf(name, sizeof(name), "gpu-ring-%d", ring->id);

diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.h b/drivers/gpu/drm/msm/msm_ringbuffer.h index 3749764..cffce09 100644 --- a/drivers/gpu/drm/msm/msm_ringbuffer.h +++ b/drivers/gpu/drm/msm/msm_ringbuffer.h @@ -40,6 +40,7 @@ struct msm_ringbuffer { struct msm_rbmemptrs *memptrs; uint64_t memptrs_iova; struct msm_fence_context *fctx; + spinlock_t lock; };

struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int id,

Currently the rd dump avoids any buffers marked as WRITE under the assumption that the contents are not interesting. While it is true that the contents are uninteresting we should still print the iova and size for all buffers so that any listening replay tools can correctly construct the submission.

Print the header for all buffers but only dump the contents for buffers marked as READ.

Signed-off-by: Jordan Crouse jcrouse@codeaurora.org --- drivers/gpu/drm/msm/msm_rd.c | 30 +++++++++++++++--------------- 1 file changed, 15 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/msm/msm_rd.c b/drivers/gpu/drm/msm/msm_rd.c index 0366b80..4c858d8 100644 --- a/drivers/gpu/drm/msm/msm_rd.c +++ b/drivers/gpu/drm/msm/msm_rd.c @@ -268,10 +268,6 @@ static void snapshot_buf(struct msm_rd_state *rd, struct msm_gem_object *obj = submit->bos[idx].obj; const char *buf;

- buf = msm_gem_get_vaddr(&obj->base); - if (IS_ERR(buf)) - return; - if (iova) { buf += iova - submit->bos[idx].iova; } else { @@ -279,8 +275,21 @@ static void snapshot_buf(struct msm_rd_state *rd, size = obj->base.size; }

+ /* + * Always write the GPUADDR header so can get a complete list of all the + * buffers in the cmd + */ rd_write_section(rd, RD_GPUADDR, (uint32_t[3]){ iova, size, iova >> 32 }, 12); + + /* But only dump the contents of buffers marked READ */ + if (!(submit->bos[idx].flags & MSM_SUBMIT_BO_READ)) + return; + + buf = msm_gem_get_vaddr(&obj->base); + if (IS_ERR(buf)) + return; + rd_write_section(rd, RD_BUFFER_CONTENTS, buf, size);

msm_gem_put_vaddr(&obj->base); @@ -309,17 +318,8 @@ void msm_rd_dump_submit(struct msm_gem_submit *submit)

rd_write_section(rd, RD_CMD, msg, ALIGN(n, 4));

- if (rd_full) { - for (i = 0; i < submit->nr_bos; i++) { - /* buffers that are written to probably don't start out - * with anything interesting: - */ - if (submit->bos[i].flags & MSM_SUBMIT_BO_WRITE) - continue; - - snapshot_buf(rd, submit, i, 0, 0); - } - } + for (i = 0; rd_full && i < submit->nr_bos; i++) + snapshot_buf(rd, submit, i, 0, 0);

for (i = 0; i < submit->nr_cmds; i++) { uint64_t iova = submit->cmd[i].iova;