[PATCH 0/5] drm/ttm, amdgpu: fix crashes due to shadow page table evictions

[PATCH 0/5] drm/ttm, amdgpu: fix crashes due to shadow page table evictions

[Nicolai Hähnle]

Fix a bunch of related crashes in amdgpu that occur when shadow page tables
are kicked out of the GART.

One of the issues was that during command submission, we rely on a
device-global evictions counter to skip some of the work of page-table
validation. The driver was never informed of evictions from GART, so this
series adds an evict parameter to ttm_bo_driver::move_notify.

There's still the evict parameter on ttm_bo_driver::move which is used by
radeon and nouveau for the call to ttm_bo_move_accel_cleanup. The 'evict'
parameter there should probably be more accurately called 'wait', but unless
that wait can always be avoided, the evict parameter on ttm_bo_driver::move
needs to stay.

Please review!
Nicolai
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c    | 20 ++++++++++++++------
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c |  5 +++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 12 +++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    |  4 ----
 drivers/gpu/drm/nouveau/nouveau_bo.c       |  3 ++-
 drivers/gpu/drm/qxl/qxl_ttm.c              |  1 +
 drivers/gpu/drm/radeon/radeon_object.c     |  1 +
 drivers/gpu/drm/radeon/radeon_object.h     |  1 +
 drivers/gpu/drm/ttm/ttm_bo.c               |  8 ++++----
 drivers/gpu/drm/virtio/virtgpu_ttm.c       |  1 +
 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c     |  1 +
 include/drm/ttm/ttm_bo_driver.h            | 10 ++++++++--
 12 files changed, 49 insertions(+), 18 deletions(-)

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 1/5] drm/ttm: add evict parameter to ttm_bo_driver::move_notify

[Nicolai Hähnle]

From: Nicolai Hähnle <nicolai.haehnle@amd.com>

Ensure that the driver can listen to evictions even when they don't take the
path through ttm_bo_driver::move.

This is crucial for amdgpu, which relies on an eviction counter to skip
re-binding page tables when possible.

Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h |  3 ++-
 drivers/gpu/drm/nouveau/nouveau_bo.c       |  3 ++-
 drivers/gpu/drm/qxl/qxl_ttm.c              |  1 +
 drivers/gpu/drm/radeon/radeon_object.c     |  1 +
 drivers/gpu/drm/radeon/radeon_object.h     |  1 +
 drivers/gpu/drm/ttm/ttm_bo.c               |  8 ++++----
 drivers/gpu/drm/virtio/virtgpu_ttm.c       |  1 +
 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c     |  1 +
 include/drm/ttm/ttm_bo_driver.h            | 10 ++++++++--
 10 files changed, 22 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index bf79b73..c29db99 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -842,20 +842,21 @@ int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
 
 	if (metadata_size)
 		*metadata_size = bo->metadata_size;
 	if (flags)
 		*flags = bo->metadata_flags;
 
 	return 0;
 }
 
 void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
+			   bool evict,
 			   struct ttm_mem_reg *new_mem)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
 	struct amdgpu_bo *abo;
 	struct ttm_mem_reg *old_mem = &bo->mem;
 
 	if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
 		return;
 
 	abo = container_of(bo, struct amdgpu_bo, tbo);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index 5cbf59e..4306b2f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -148,21 +148,22 @@ void amdgpu_bo_fini(struct amdgpu_device *adev);
 int amdgpu_bo_fbdev_mmap(struct amdgpu_bo *bo,
 				struct vm_area_struct *vma);
 int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64 tiling_flags);
 void amdgpu_bo_get_tiling_flags(struct amdgpu_bo *bo, u64 *tiling_flags);
 int amdgpu_bo_set_metadata (struct amdgpu_bo *bo, void *metadata,
 			    uint32_t metadata_size, uint64_t flags);
 int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
 			   size_t buffer_size, uint32_t *metadata_size,
 			   uint64_t *flags);
 void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
-				  struct ttm_mem_reg *new_mem);
+			   bool evict,
+			   struct ttm_mem_reg *new_mem);
 int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
 void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence,
 		     bool shared);
 u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo);
 int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev,
 			       struct amdgpu_ring *ring,
 			       struct amdgpu_bo *bo,
 			       struct reservation_object *resv,
 			       struct dma_fence **fence, bool direct);
 int amdgpu_bo_restore_from_shadow(struct amdgpu_device *adev,
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index e0c0007..6fa1521 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1187,21 +1187,22 @@ nouveau_bo_move_flips(struct ttm_buffer_object *bo, bool evict, bool intr,
 	ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_gpu, new_mem);
 	if (ret)
 		goto out;
 
 out:
 	ttm_bo_mem_put(bo, &tmp_mem);
 	return ret;
 }
 
 static void
-nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem)
+nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
+		     struct ttm_mem_reg *new_mem)
 {
 	struct nouveau_bo *nvbo = nouveau_bo(bo);
 	struct nvkm_vma *vma;
 
 	/* ttm can now (stupidly) pass the driver bos it didn't create... */
 	if (bo->destroy != nouveau_bo_del_ttm)
 		return;
 
 	list_for_each_entry(vma, &nvbo->vma_list, head) {
 		if (new_mem && new_mem->mem_type != TTM_PL_SYSTEM &&
diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c b/drivers/gpu/drm/qxl/qxl_ttm.c
index 1176133..f3939a9 100644
--- a/drivers/gpu/drm/qxl/qxl_ttm.c
+++ b/drivers/gpu/drm/qxl/qxl_ttm.c
@@ -360,20 +360,21 @@ static int qxl_bo_move(struct ttm_buffer_object *bo,
 
 	if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
 		qxl_move_null(bo, new_mem);
 		return 0;
 	}
 	return ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu,
 				  new_mem);
 }
 
 static void qxl_bo_move_notify(struct ttm_buffer_object *bo,
+			       bool evict,
 			       struct ttm_mem_reg *new_mem)
 {
 	struct qxl_bo *qbo;
 	struct qxl_device *qdev;
 
 	if (!qxl_ttm_bo_is_qxl_bo(bo))
 		return;
 	qbo = to_qxl_bo(bo);
 	qdev = qbo->gem_base.dev->dev_private;
 
diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
index 41b72ce..74b2760 100644
--- a/drivers/gpu/drm/radeon/radeon_object.c
+++ b/drivers/gpu/drm/radeon/radeon_object.c
@@ -758,20 +758,21 @@ int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
 		return 0;
 	}
 
 	if ((bo->surface_reg >= 0) && !has_moved)
 		return 0;
 
 	return radeon_bo_get_surface_reg(bo);
 }
 
 void radeon_bo_move_notify(struct ttm_buffer_object *bo,
+			   bool evict,
 			   struct ttm_mem_reg *new_mem)
 {
 	struct radeon_bo *rbo;
 
 	if (!radeon_ttm_bo_is_radeon_bo(bo))
 		return;
 
 	rbo = container_of(bo, struct radeon_bo, tbo);
 	radeon_bo_check_tiling(rbo, 0, 1);
 	radeon_vm_bo_invalidate(rbo->rdev, rbo);
diff --git a/drivers/gpu/drm/radeon/radeon_object.h b/drivers/gpu/drm/radeon/radeon_object.h
index a10bb3d..9ffd821 100644
--- a/drivers/gpu/drm/radeon/radeon_object.h
+++ b/drivers/gpu/drm/radeon/radeon_object.h
@@ -143,20 +143,21 @@ extern void radeon_bo_fini(struct radeon_device *rdev);
 extern int radeon_bo_list_validate(struct radeon_device *rdev,
 				   struct ww_acquire_ctx *ticket,
 				   struct list_head *head, int ring);
 extern int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
 				u32 tiling_flags, u32 pitch);
 extern void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
 				u32 *tiling_flags, u32 *pitch);
 extern int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
 				bool force_drop);
 extern void radeon_bo_move_notify(struct ttm_buffer_object *bo,
+				  bool evict,
 				  struct ttm_mem_reg *new_mem);
 extern int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
 extern int radeon_bo_get_surface_reg(struct radeon_bo *bo);
 extern void radeon_bo_fence(struct radeon_bo *bo, struct radeon_fence *fence,
 			    bool shared);
 
 /*
  * sub allocation
  */
 
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index d506361..d4973e9c 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -335,45 +335,45 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 			goto out_err;
 
 		if (mem->mem_type != TTM_PL_SYSTEM) {
 			ret = ttm_tt_bind(bo->ttm, mem);
 			if (ret)
 				goto out_err;
 		}
 
 		if (bo->mem.mem_type == TTM_PL_SYSTEM) {
 			if (bdev->driver->move_notify)
-				bdev->driver->move_notify(bo, mem);
+				bdev->driver->move_notify(bo, evict, mem);
 			bo->mem = *mem;
 			mem->mm_node = NULL;
 			goto moved;
 		}
 	}
 
 	if (bdev->driver->move_notify)
-		bdev->driver->move_notify(bo, mem);
+		bdev->driver->move_notify(bo, evict, mem);
 
 	if (!(old_man->flags & TTM_MEMTYPE_FLAG_FIXED) &&
 	    !(new_man->flags & TTM_MEMTYPE_FLAG_FIXED))
 		ret = ttm_bo_move_ttm(bo, interruptible, no_wait_gpu, mem);
 	else if (bdev->driver->move)
 		ret = bdev->driver->move(bo, evict, interruptible,
 					 no_wait_gpu, mem);
 	else
 		ret = ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu, mem);
 
 	if (ret) {
 		if (bdev->driver->move_notify) {
 			struct ttm_mem_reg tmp_mem = *mem;
 			*mem = bo->mem;
 			bo->mem = tmp_mem;
-			bdev->driver->move_notify(bo, mem);
+			bdev->driver->move_notify(bo, false, mem);
 			bo->mem = *mem;
 			*mem = tmp_mem;
 		}
 
 		goto out_err;
 	}
 
 moved:
 	if (bo->evicted) {
 		if (bdev->driver->invalidate_caches) {
@@ -407,21 +407,21 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
  * Call bo::reserved.
  * Will release GPU memory type usage on destruction.
  * This is the place to put in driver specific hooks to release
  * driver private resources.
  * Will release the bo::reserved lock.
  */
 
 static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo)
 {
 	if (bo->bdev->driver->move_notify)
-		bo->bdev->driver->move_notify(bo, NULL);
+		bo->bdev->driver->move_notify(bo, false, NULL);
 
 	ttm_tt_destroy(bo->ttm);
 	bo->ttm = NULL;
 	ttm_bo_mem_put(bo, &bo->mem);
 
 	ww_mutex_unlock (&bo->resv->lock);
 }
 
 static void ttm_bo_flush_all_fences(struct ttm_buffer_object *bo)
 {
diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c b/drivers/gpu/drm/virtio/virtgpu_ttm.c
index 4a1de9f..10387d7 100644
--- a/drivers/gpu/drm/virtio/virtgpu_ttm.c
+++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c
@@ -379,20 +379,21 @@ static int virtio_gpu_bo_move(struct ttm_buffer_object *bo,
 
 	ret = ttm_bo_wait(bo, interruptible, no_wait_gpu);
 	if (ret)
 		return ret;
 
 	virtio_gpu_move_null(bo, new_mem);
 	return 0;
 }
 
 static void virtio_gpu_bo_move_notify(struct ttm_buffer_object *tbo,
+				      bool evict,
 				      struct ttm_mem_reg *new_mem)
 {
 	struct virtio_gpu_object *bo;
 	struct virtio_gpu_device *vgdev;
 
 	bo = container_of(tbo, struct virtio_gpu_object, tbo);
 	vgdev = (struct virtio_gpu_device *)bo->gem_base.dev->dev_private;
 
 	if (!new_mem || (new_mem->placement & TTM_PL_FLAG_SYSTEM)) {
 		if (bo->hw_res_handle)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index c894a48..caa279b 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -818,20 +818,21 @@ static int vmw_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
  * vmw_move_notify - TTM move_notify_callback
  *
  * @bo: The TTM buffer object about to move.
  * @mem: The struct ttm_mem_reg indicating to what memory
  *       region the move is taking place.
  *
  * Calls move_notify for all subsystems needing it.
  * (currently only resources).
  */
 static void vmw_move_notify(struct ttm_buffer_object *bo,
+			    bool evict,
 			    struct ttm_mem_reg *mem)
 {
 	vmw_resource_move_notify(bo, mem);
 	vmw_query_move_notify(bo, mem);
 }
 
 
 /**
  * vmw_swap_notify - TTM move_notify_callback
  *
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index cdbdb40..8b4bec7 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -424,23 +424,29 @@ struct ttm_bo_driver {
 	 *
 	 * Called from the map / write / read methods to verify that the
 	 * caller is permitted to access the buffer object.
 	 * This member may be set to NULL, which will refuse this kind of
 	 * access for all buffer objects.
 	 * This function should return 0 if access is granted, -EPERM otherwise.
 	 */
 	int (*verify_access)(struct ttm_buffer_object *bo,
 			     struct file *filp);
 
-	/* hook to notify driver about a driver move so it
-	 * can do tiling things */
+	/**
+	 * Hook to notify driver about a driver move so it
+	 * can do tiling things and book-keeping.
+	 *
+	 * @evict: whether this move is evicting the buffer from the graphics
+	 * address space
+	 */
 	void (*move_notify)(struct ttm_buffer_object *bo,
+			    bool evict,
 			    struct ttm_mem_reg *new_mem);
 	/* notify the driver we are taking a fault on this BO
 	 * and have reserved it */
 	int (*fault_reserve_notify)(struct ttm_buffer_object *bo);
 
 	/**
 	 * notify the driver that we're about to swap out this bo
 	 */
 	void (*swap_notify)(struct ttm_buffer_object *bo);
 
-- 
2.7.4

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 1/5] drm/ttm: add evict parameter to ttm_bo_driver::move_notify

[zhoucm1]

On 2016年12月16日 01:10, Nicolai Hähnle wrote:
> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>
> Ensure that the driver can listen to evictions even when they don't take the
> path through ttm_bo_driver::move.
>
> This is crucial for amdgpu, which relies on an eviction counter to skip
> re-binding page tables when possible.
>
> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Acked-by: Chunming Zhou <david1.zhou@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.c |  1 +
>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.h |  3 ++-
>   drivers/gpu/drm/nouveau/nouveau_bo.c       |  3 ++-
>   drivers/gpu/drm/qxl/qxl_ttm.c              |  1 +
>   drivers/gpu/drm/radeon/radeon_object.c     |  1 +
>   drivers/gpu/drm/radeon/radeon_object.h     |  1 +
>   drivers/gpu/drm/ttm/ttm_bo.c               |  8 ++++----
>   drivers/gpu/drm/virtio/virtgpu_ttm.c       |  1 +
>   drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c     |  1 +
>   include/drm/ttm/ttm_bo_driver.h            | 10 ++++++++--
>   10 files changed, 22 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> index bf79b73..c29db99 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> @@ -842,20 +842,21 @@ int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
>   
>   	if (metadata_size)
>   		*metadata_size = bo->metadata_size;
>   	if (flags)
>   		*flags = bo->metadata_flags;
>   
>   	return 0;
>   }
>   
>   void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
> +			   bool evict,
>   			   struct ttm_mem_reg *new_mem)
>   {
>   	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
>   	struct amdgpu_bo *abo;
>   	struct ttm_mem_reg *old_mem = &bo->mem;
>   
>   	if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
>   		return;
>   
>   	abo = container_of(bo, struct amdgpu_bo, tbo);
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> index 5cbf59e..4306b2f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> @@ -148,21 +148,22 @@ void amdgpu_bo_fini(struct amdgpu_device *adev);
>   int amdgpu_bo_fbdev_mmap(struct amdgpu_bo *bo,
>   				struct vm_area_struct *vma);
>   int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64 tiling_flags);
>   void amdgpu_bo_get_tiling_flags(struct amdgpu_bo *bo, u64 *tiling_flags);
>   int amdgpu_bo_set_metadata (struct amdgpu_bo *bo, void *metadata,
>   			    uint32_t metadata_size, uint64_t flags);
>   int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
>   			   size_t buffer_size, uint32_t *metadata_size,
>   			   uint64_t *flags);
>   void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
> -				  struct ttm_mem_reg *new_mem);
> +			   bool evict,
> +			   struct ttm_mem_reg *new_mem);
>   int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
>   void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence,
>   		     bool shared);
>   u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo);
>   int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev,
>   			       struct amdgpu_ring *ring,
>   			       struct amdgpu_bo *bo,
>   			       struct reservation_object *resv,
>   			       struct dma_fence **fence, bool direct);
>   int amdgpu_bo_restore_from_shadow(struct amdgpu_device *adev,
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index e0c0007..6fa1521 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -1187,21 +1187,22 @@ nouveau_bo_move_flips(struct ttm_buffer_object *bo, bool evict, bool intr,
>   	ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_gpu, new_mem);
>   	if (ret)
>   		goto out;
>   
>   out:
>   	ttm_bo_mem_put(bo, &tmp_mem);
>   	return ret;
>   }
>   
>   static void
> -nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem)
> +nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
> +		     struct ttm_mem_reg *new_mem)
>   {
>   	struct nouveau_bo *nvbo = nouveau_bo(bo);
>   	struct nvkm_vma *vma;
>   
>   	/* ttm can now (stupidly) pass the driver bos it didn't create... */
>   	if (bo->destroy != nouveau_bo_del_ttm)
>   		return;
>   
>   	list_for_each_entry(vma, &nvbo->vma_list, head) {
>   		if (new_mem && new_mem->mem_type != TTM_PL_SYSTEM &&
> diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c b/drivers/gpu/drm/qxl/qxl_ttm.c
> index 1176133..f3939a9 100644
> --- a/drivers/gpu/drm/qxl/qxl_ttm.c
> +++ b/drivers/gpu/drm/qxl/qxl_ttm.c
> @@ -360,20 +360,21 @@ static int qxl_bo_move(struct ttm_buffer_object *bo,
>   
>   	if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
>   		qxl_move_null(bo, new_mem);
>   		return 0;
>   	}
>   	return ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu,
>   				  new_mem);
>   }
>   
>   static void qxl_bo_move_notify(struct ttm_buffer_object *bo,
> +			       bool evict,
>   			       struct ttm_mem_reg *new_mem)
>   {
>   	struct qxl_bo *qbo;
>   	struct qxl_device *qdev;
>   
>   	if (!qxl_ttm_bo_is_qxl_bo(bo))
>   		return;
>   	qbo = to_qxl_bo(bo);
>   	qdev = qbo->gem_base.dev->dev_private;
>   
> diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
> index 41b72ce..74b2760 100644
> --- a/drivers/gpu/drm/radeon/radeon_object.c
> +++ b/drivers/gpu/drm/radeon/radeon_object.c
> @@ -758,20 +758,21 @@ int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
>   		return 0;
>   	}
>   
>   	if ((bo->surface_reg >= 0) && !has_moved)
>   		return 0;
>   
>   	return radeon_bo_get_surface_reg(bo);
>   }
>   
>   void radeon_bo_move_notify(struct ttm_buffer_object *bo,
> +			   bool evict,
>   			   struct ttm_mem_reg *new_mem)
>   {
>   	struct radeon_bo *rbo;
>   
>   	if (!radeon_ttm_bo_is_radeon_bo(bo))
>   		return;
>   
>   	rbo = container_of(bo, struct radeon_bo, tbo);
>   	radeon_bo_check_tiling(rbo, 0, 1);
>   	radeon_vm_bo_invalidate(rbo->rdev, rbo);
> diff --git a/drivers/gpu/drm/radeon/radeon_object.h b/drivers/gpu/drm/radeon/radeon_object.h
> index a10bb3d..9ffd821 100644
> --- a/drivers/gpu/drm/radeon/radeon_object.h
> +++ b/drivers/gpu/drm/radeon/radeon_object.h
> @@ -143,20 +143,21 @@ extern void radeon_bo_fini(struct radeon_device *rdev);
>   extern int radeon_bo_list_validate(struct radeon_device *rdev,
>   				   struct ww_acquire_ctx *ticket,
>   				   struct list_head *head, int ring);
>   extern int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
>   				u32 tiling_flags, u32 pitch);
>   extern void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
>   				u32 *tiling_flags, u32 *pitch);
>   extern int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
>   				bool force_drop);
>   extern void radeon_bo_move_notify(struct ttm_buffer_object *bo,
> +				  bool evict,
>   				  struct ttm_mem_reg *new_mem);
>   extern int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
>   extern int radeon_bo_get_surface_reg(struct radeon_bo *bo);
>   extern void radeon_bo_fence(struct radeon_bo *bo, struct radeon_fence *fence,
>   			    bool shared);
>   
>   /*
>    * sub allocation
>    */
>   
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index d506361..d4973e9c 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -335,45 +335,45 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
>   			goto out_err;
>   
>   		if (mem->mem_type != TTM_PL_SYSTEM) {
>   			ret = ttm_tt_bind(bo->ttm, mem);
>   			if (ret)
>   				goto out_err;
>   		}
>   
>   		if (bo->mem.mem_type == TTM_PL_SYSTEM) {
>   			if (bdev->driver->move_notify)
> -				bdev->driver->move_notify(bo, mem);
> +				bdev->driver->move_notify(bo, evict, mem);
>   			bo->mem = *mem;
>   			mem->mm_node = NULL;
>   			goto moved;
>   		}
>   	}
>   
>   	if (bdev->driver->move_notify)
> -		bdev->driver->move_notify(bo, mem);
> +		bdev->driver->move_notify(bo, evict, mem);
>   
>   	if (!(old_man->flags & TTM_MEMTYPE_FLAG_FIXED) &&
>   	    !(new_man->flags & TTM_MEMTYPE_FLAG_FIXED))
>   		ret = ttm_bo_move_ttm(bo, interruptible, no_wait_gpu, mem);
>   	else if (bdev->driver->move)
>   		ret = bdev->driver->move(bo, evict, interruptible,
>   					 no_wait_gpu, mem);
>   	else
>   		ret = ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu, mem);
>   
>   	if (ret) {
>   		if (bdev->driver->move_notify) {
>   			struct ttm_mem_reg tmp_mem = *mem;
>   			*mem = bo->mem;
>   			bo->mem = tmp_mem;
> -			bdev->driver->move_notify(bo, mem);
> +			bdev->driver->move_notify(bo, false, mem);
>   			bo->mem = *mem;
>   			*mem = tmp_mem;
>   		}
>   
>   		goto out_err;
>   	}
>   
>   moved:
>   	if (bo->evicted) {
>   		if (bdev->driver->invalidate_caches) {
> @@ -407,21 +407,21 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
>    * Call bo::reserved.
>    * Will release GPU memory type usage on destruction.
>    * This is the place to put in driver specific hooks to release
>    * driver private resources.
>    * Will release the bo::reserved lock.
>    */
>   
>   static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo)
>   {
>   	if (bo->bdev->driver->move_notify)
> -		bo->bdev->driver->move_notify(bo, NULL);
> +		bo->bdev->driver->move_notify(bo, false, NULL);
>   
>   	ttm_tt_destroy(bo->ttm);
>   	bo->ttm = NULL;
>   	ttm_bo_mem_put(bo, &bo->mem);
>   
>   	ww_mutex_unlock (&bo->resv->lock);
>   }
>   
>   static void ttm_bo_flush_all_fences(struct ttm_buffer_object *bo)
>   {
> diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c b/drivers/gpu/drm/virtio/virtgpu_ttm.c
> index 4a1de9f..10387d7 100644
> --- a/drivers/gpu/drm/virtio/virtgpu_ttm.c
> +++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c
> @@ -379,20 +379,21 @@ static int virtio_gpu_bo_move(struct ttm_buffer_object *bo,
>   
>   	ret = ttm_bo_wait(bo, interruptible, no_wait_gpu);
>   	if (ret)
>   		return ret;
>   
>   	virtio_gpu_move_null(bo, new_mem);
>   	return 0;
>   }
>   
>   static void virtio_gpu_bo_move_notify(struct ttm_buffer_object *tbo,
> +				      bool evict,
>   				      struct ttm_mem_reg *new_mem)
>   {
>   	struct virtio_gpu_object *bo;
>   	struct virtio_gpu_device *vgdev;
>   
>   	bo = container_of(tbo, struct virtio_gpu_object, tbo);
>   	vgdev = (struct virtio_gpu_device *)bo->gem_base.dev->dev_private;
>   
>   	if (!new_mem || (new_mem->placement & TTM_PL_FLAG_SYSTEM)) {
>   		if (bo->hw_res_handle)
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> index c894a48..caa279b 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> @@ -818,20 +818,21 @@ static int vmw_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
>    * vmw_move_notify - TTM move_notify_callback
>    *
>    * @bo: The TTM buffer object about to move.
>    * @mem: The struct ttm_mem_reg indicating to what memory
>    *       region the move is taking place.
>    *
>    * Calls move_notify for all subsystems needing it.
>    * (currently only resources).
>    */
>   static void vmw_move_notify(struct ttm_buffer_object *bo,
> +			    bool evict,
>   			    struct ttm_mem_reg *mem)
>   {
>   	vmw_resource_move_notify(bo, mem);
>   	vmw_query_move_notify(bo, mem);
>   }
>   
>   
>   /**
>    * vmw_swap_notify - TTM move_notify_callback
>    *
> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
> index cdbdb40..8b4bec7 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -424,23 +424,29 @@ struct ttm_bo_driver {
>   	 *
>   	 * Called from the map / write / read methods to verify that the
>   	 * caller is permitted to access the buffer object.
>   	 * This member may be set to NULL, which will refuse this kind of
>   	 * access for all buffer objects.
>   	 * This function should return 0 if access is granted, -EPERM otherwise.
>   	 */
>   	int (*verify_access)(struct ttm_buffer_object *bo,
>   			     struct file *filp);
>   
> -	/* hook to notify driver about a driver move so it
> -	 * can do tiling things */
> +	/**
> +	 * Hook to notify driver about a driver move so it
> +	 * can do tiling things and book-keeping.
> +	 *
> +	 * @evict: whether this move is evicting the buffer from the graphics
> +	 * address space
> +	 */
>   	void (*move_notify)(struct ttm_buffer_object *bo,
> +			    bool evict,
>   			    struct ttm_mem_reg *new_mem);
>   	/* notify the driver we are taking a fault on this BO
>   	 * and have reserved it */
>   	int (*fault_reserve_notify)(struct ttm_buffer_object *bo);
>   
>   	/**
>   	 * notify the driver that we're about to swap out this bo
>   	 */
>   	void (*swap_notify)(struct ttm_buffer_object *bo);
>   

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 1/5] drm/ttm: add evict parameter to ttm_bo_driver::move_notify

[Nicolai Hähnle]

On 16.12.2016 03:49, zhoucm1 wrote:
> On 2016年12月16日 01:10, Nicolai Hähnle wrote:
>> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>>
>> Ensure that the driver can listen to evictions even when they don't
>> take the
>> path through ttm_bo_driver::move.
>>
>> This is crucial for amdgpu, which relies on an eviction counter to skip
>> re-binding page tables when possible.
>>
>> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
> Acked-by: Chunming Zhou <david1.zhou@amd.com>

Thanks. Ping for feedback from non-AMD people?

Nicolai


>> ---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.c |  1 +
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.h |  3 ++-
>>   drivers/gpu/drm/nouveau/nouveau_bo.c       |  3 ++-
>>   drivers/gpu/drm/qxl/qxl_ttm.c              |  1 +
>>   drivers/gpu/drm/radeon/radeon_object.c     |  1 +
>>   drivers/gpu/drm/radeon/radeon_object.h     |  1 +
>>   drivers/gpu/drm/ttm/ttm_bo.c               |  8 ++++----
>>   drivers/gpu/drm/virtio/virtgpu_ttm.c       |  1 +
>>   drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c     |  1 +
>>   include/drm/ttm/ttm_bo_driver.h            | 10 ++++++++--
>>   10 files changed, 22 insertions(+), 8 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> index bf79b73..c29db99 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> @@ -842,20 +842,21 @@ int amdgpu_bo_get_metadata(struct amdgpu_bo *bo,
>> void *buffer,
>>         if (metadata_size)
>>           *metadata_size = bo->metadata_size;
>>       if (flags)
>>           *flags = bo->metadata_flags;
>>         return 0;
>>   }
>>     void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
>> +               bool evict,
>>                  struct ttm_mem_reg *new_mem)
>>   {
>>       struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
>>       struct amdgpu_bo *abo;
>>       struct ttm_mem_reg *old_mem = &bo->mem;
>>         if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
>>           return;
>>         abo = container_of(bo, struct amdgpu_bo, tbo);
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>> index 5cbf59e..4306b2f 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>> @@ -148,21 +148,22 @@ void amdgpu_bo_fini(struct amdgpu_device *adev);
>>   int amdgpu_bo_fbdev_mmap(struct amdgpu_bo *bo,
>>                   struct vm_area_struct *vma);
>>   int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64 tiling_flags);
>>   void amdgpu_bo_get_tiling_flags(struct amdgpu_bo *bo, u64
>> *tiling_flags);
>>   int amdgpu_bo_set_metadata (struct amdgpu_bo *bo, void *metadata,
>>                   uint32_t metadata_size, uint64_t flags);
>>   int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
>>                  size_t buffer_size, uint32_t *metadata_size,
>>                  uint64_t *flags);
>>   void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
>> -                  struct ttm_mem_reg *new_mem);
>> +               bool evict,
>> +               struct ttm_mem_reg *new_mem);
>>   int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
>>   void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence,
>>                bool shared);
>>   u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo);
>>   int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev,
>>                      struct amdgpu_ring *ring,
>>                      struct amdgpu_bo *bo,
>>                      struct reservation_object *resv,
>>                      struct dma_fence **fence, bool direct);
>>   int amdgpu_bo_restore_from_shadow(struct amdgpu_device *adev,
>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> index e0c0007..6fa1521 100644
>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> @@ -1187,21 +1187,22 @@ nouveau_bo_move_flips(struct ttm_buffer_object
>> *bo, bool evict, bool intr,
>>       ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_gpu, new_mem);
>>       if (ret)
>>           goto out;
>>     out:
>>       ttm_bo_mem_put(bo, &tmp_mem);
>>       return ret;
>>   }
>>     static void
>> -nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, struct ttm_mem_reg
>> *new_mem)
>> +nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
>> +             struct ttm_mem_reg *new_mem)
>>   {
>>       struct nouveau_bo *nvbo = nouveau_bo(bo);
>>       struct nvkm_vma *vma;
>>         /* ttm can now (stupidly) pass the driver bos it didn't
>> create... */
>>       if (bo->destroy != nouveau_bo_del_ttm)
>>           return;
>>         list_for_each_entry(vma, &nvbo->vma_list, head) {
>>           if (new_mem && new_mem->mem_type != TTM_PL_SYSTEM &&
>> diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c
>> b/drivers/gpu/drm/qxl/qxl_ttm.c
>> index 1176133..f3939a9 100644
>> --- a/drivers/gpu/drm/qxl/qxl_ttm.c
>> +++ b/drivers/gpu/drm/qxl/qxl_ttm.c
>> @@ -360,20 +360,21 @@ static int qxl_bo_move(struct ttm_buffer_object
>> *bo,
>>         if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
>>           qxl_move_null(bo, new_mem);
>>           return 0;
>>       }
>>       return ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu,
>>                     new_mem);
>>   }
>>     static void qxl_bo_move_notify(struct ttm_buffer_object *bo,
>> +                   bool evict,
>>                      struct ttm_mem_reg *new_mem)
>>   {
>>       struct qxl_bo *qbo;
>>       struct qxl_device *qdev;
>>         if (!qxl_ttm_bo_is_qxl_bo(bo))
>>           return;
>>       qbo = to_qxl_bo(bo);
>>       qdev = qbo->gem_base.dev->dev_private;
>>   diff --git a/drivers/gpu/drm/radeon/radeon_object.c
>> b/drivers/gpu/drm/radeon/radeon_object.c
>> index 41b72ce..74b2760 100644
>> --- a/drivers/gpu/drm/radeon/radeon_object.c
>> +++ b/drivers/gpu/drm/radeon/radeon_object.c
>> @@ -758,20 +758,21 @@ int radeon_bo_check_tiling(struct radeon_bo *bo,
>> bool has_moved,
>>           return 0;
>>       }
>>         if ((bo->surface_reg >= 0) && !has_moved)
>>           return 0;
>>         return radeon_bo_get_surface_reg(bo);
>>   }
>>     void radeon_bo_move_notify(struct ttm_buffer_object *bo,
>> +               bool evict,
>>                  struct ttm_mem_reg *new_mem)
>>   {
>>       struct radeon_bo *rbo;
>>         if (!radeon_ttm_bo_is_radeon_bo(bo))
>>           return;
>>         rbo = container_of(bo, struct radeon_bo, tbo);
>>       radeon_bo_check_tiling(rbo, 0, 1);
>>       radeon_vm_bo_invalidate(rbo->rdev, rbo);
>> diff --git a/drivers/gpu/drm/radeon/radeon_object.h
>> b/drivers/gpu/drm/radeon/radeon_object.h
>> index a10bb3d..9ffd821 100644
>> --- a/drivers/gpu/drm/radeon/radeon_object.h
>> +++ b/drivers/gpu/drm/radeon/radeon_object.h
>> @@ -143,20 +143,21 @@ extern void radeon_bo_fini(struct radeon_device
>> *rdev);
>>   extern int radeon_bo_list_validate(struct radeon_device *rdev,
>>                      struct ww_acquire_ctx *ticket,
>>                      struct list_head *head, int ring);
>>   extern int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
>>                   u32 tiling_flags, u32 pitch);
>>   extern void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
>>                   u32 *tiling_flags, u32 *pitch);
>>   extern int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
>>                   bool force_drop);
>>   extern void radeon_bo_move_notify(struct ttm_buffer_object *bo,
>> +                  bool evict,
>>                     struct ttm_mem_reg *new_mem);
>>   extern int radeon_bo_fault_reserve_notify(struct ttm_buffer_object
>> *bo);
>>   extern int radeon_bo_get_surface_reg(struct radeon_bo *bo);
>>   extern void radeon_bo_fence(struct radeon_bo *bo, struct
>> radeon_fence *fence,
>>                   bool shared);
>>     /*
>>    * sub allocation
>>    */
>>   diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>> b/drivers/gpu/drm/ttm/ttm_bo.c
>> index d506361..d4973e9c 100644
>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>> @@ -335,45 +335,45 @@ static int ttm_bo_handle_move_mem(struct
>> ttm_buffer_object *bo,
>>               goto out_err;
>>             if (mem->mem_type != TTM_PL_SYSTEM) {
>>               ret = ttm_tt_bind(bo->ttm, mem);
>>               if (ret)
>>                   goto out_err;
>>           }
>>             if (bo->mem.mem_type == TTM_PL_SYSTEM) {
>>               if (bdev->driver->move_notify)
>> -                bdev->driver->move_notify(bo, mem);
>> +                bdev->driver->move_notify(bo, evict, mem);
>>               bo->mem = *mem;
>>               mem->mm_node = NULL;
>>               goto moved;
>>           }
>>       }
>>         if (bdev->driver->move_notify)
>> -        bdev->driver->move_notify(bo, mem);
>> +        bdev->driver->move_notify(bo, evict, mem);
>>         if (!(old_man->flags & TTM_MEMTYPE_FLAG_FIXED) &&
>>           !(new_man->flags & TTM_MEMTYPE_FLAG_FIXED))
>>           ret = ttm_bo_move_ttm(bo, interruptible, no_wait_gpu, mem);
>>       else if (bdev->driver->move)
>>           ret = bdev->driver->move(bo, evict, interruptible,
>>                        no_wait_gpu, mem);
>>       else
>>           ret = ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu, mem);
>>         if (ret) {
>>           if (bdev->driver->move_notify) {
>>               struct ttm_mem_reg tmp_mem = *mem;
>>               *mem = bo->mem;
>>               bo->mem = tmp_mem;
>> -            bdev->driver->move_notify(bo, mem);
>> +            bdev->driver->move_notify(bo, false, mem);
>>               bo->mem = *mem;
>>               *mem = tmp_mem;
>>           }
>>             goto out_err;
>>       }
>>     moved:
>>       if (bo->evicted) {
>>           if (bdev->driver->invalidate_caches) {
>> @@ -407,21 +407,21 @@ static int ttm_bo_handle_move_mem(struct
>> ttm_buffer_object *bo,
>>    * Call bo::reserved.
>>    * Will release GPU memory type usage on destruction.
>>    * This is the place to put in driver specific hooks to release
>>    * driver private resources.
>>    * Will release the bo::reserved lock.
>>    */
>>     static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo)
>>   {
>>       if (bo->bdev->driver->move_notify)
>> -        bo->bdev->driver->move_notify(bo, NULL);
>> +        bo->bdev->driver->move_notify(bo, false, NULL);
>>         ttm_tt_destroy(bo->ttm);
>>       bo->ttm = NULL;
>>       ttm_bo_mem_put(bo, &bo->mem);
>>         ww_mutex_unlock (&bo->resv->lock);
>>   }
>>     static void ttm_bo_flush_all_fences(struct ttm_buffer_object *bo)
>>   {
>> diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c
>> b/drivers/gpu/drm/virtio/virtgpu_ttm.c
>> index 4a1de9f..10387d7 100644
>> --- a/drivers/gpu/drm/virtio/virtgpu_ttm.c
>> +++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c
>> @@ -379,20 +379,21 @@ static int virtio_gpu_bo_move(struct
>> ttm_buffer_object *bo,
>>         ret = ttm_bo_wait(bo, interruptible, no_wait_gpu);
>>       if (ret)
>>           return ret;
>>         virtio_gpu_move_null(bo, new_mem);
>>       return 0;
>>   }
>>     static void virtio_gpu_bo_move_notify(struct ttm_buffer_object *tbo,
>> +                      bool evict,
>>                         struct ttm_mem_reg *new_mem)
>>   {
>>       struct virtio_gpu_object *bo;
>>       struct virtio_gpu_device *vgdev;
>>         bo = container_of(tbo, struct virtio_gpu_object, tbo);
>>       vgdev = (struct virtio_gpu_device *)bo->gem_base.dev->dev_private;
>>         if (!new_mem || (new_mem->placement & TTM_PL_FLAG_SYSTEM)) {
>>           if (bo->hw_res_handle)
>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>> b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>> index c894a48..caa279b 100644
>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>> @@ -818,20 +818,21 @@ static int vmw_ttm_fault_reserve_notify(struct
>> ttm_buffer_object *bo)
>>    * vmw_move_notify - TTM move_notify_callback
>>    *
>>    * @bo: The TTM buffer object about to move.
>>    * @mem: The struct ttm_mem_reg indicating to what memory
>>    *       region the move is taking place.
>>    *
>>    * Calls move_notify for all subsystems needing it.
>>    * (currently only resources).
>>    */
>>   static void vmw_move_notify(struct ttm_buffer_object *bo,
>> +                bool evict,
>>                   struct ttm_mem_reg *mem)
>>   {
>>       vmw_resource_move_notify(bo, mem);
>>       vmw_query_move_notify(bo, mem);
>>   }
>>       /**
>>    * vmw_swap_notify - TTM move_notify_callback
>>    *
>> diff --git a/include/drm/ttm/ttm_bo_driver.h
>> b/include/drm/ttm/ttm_bo_driver.h
>> index cdbdb40..8b4bec7 100644
>> --- a/include/drm/ttm/ttm_bo_driver.h
>> +++ b/include/drm/ttm/ttm_bo_driver.h
>> @@ -424,23 +424,29 @@ struct ttm_bo_driver {
>>        *
>>        * Called from the map / write / read methods to verify that the
>>        * caller is permitted to access the buffer object.
>>        * This member may be set to NULL, which will refuse this kind of
>>        * access for all buffer objects.
>>        * This function should return 0 if access is granted, -EPERM
>> otherwise.
>>        */
>>       int (*verify_access)(struct ttm_buffer_object *bo,
>>                    struct file *filp);
>>   -    /* hook to notify driver about a driver move so it
>> -     * can do tiling things */
>> +    /**
>> +     * Hook to notify driver about a driver move so it
>> +     * can do tiling things and book-keeping.
>> +     *
>> +     * @evict: whether this move is evicting the buffer from the
>> graphics
>> +     * address space
>> +     */
>>       void (*move_notify)(struct ttm_buffer_object *bo,
>> +                bool evict,
>>                   struct ttm_mem_reg *new_mem);
>>       /* notify the driver we are taking a fault on this BO
>>        * and have reserved it */
>>       int (*fault_reserve_notify)(struct ttm_buffer_object *bo);
>>         /**
>>        * notify the driver that we're about to swap out this bo
>>        */
>>       void (*swap_notify)(struct ttm_buffer_object *bo);
>>
>
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 1/5] drm/ttm: add evict parameter to ttm_bo_driver::move_notify

[Christian König]

Am 21.12.2016 um 16:12 schrieb Nicolai Hähnle:
> On 16.12.2016 03:49, zhoucm1 wrote:
>> On 2016年12月16日 01:10, Nicolai Hähnle wrote:
>>> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>>>
>>> Ensure that the driver can listen to evictions even when they don't
>>> take the
>>> path through ttm_bo_driver::move.
>>>
>>> This is crucial for amdgpu, which relies on an eviction counter to skip
>>> re-binding page tables when possible.
>>>
>>> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
>> Acked-by: Chunming Zhou <david1.zhou@amd.com>

Feel free to add my Reviewed-by: Christian König 
<christian.koenig@amd.com> to patch #1-#4 and V2 of patch #5 as well.

>
> Thanks. Ping for feedback from non-AMD people?

Ping once more. Would be nice if somebody else can take a look as well.

Christian.

>
> Nicolai
>
>
>>> ---
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.c |  1 +
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.h |  3 ++-
>>>   drivers/gpu/drm/nouveau/nouveau_bo.c       |  3 ++-
>>>   drivers/gpu/drm/qxl/qxl_ttm.c              |  1 +
>>>   drivers/gpu/drm/radeon/radeon_object.c     |  1 +
>>>   drivers/gpu/drm/radeon/radeon_object.h     |  1 +
>>>   drivers/gpu/drm/ttm/ttm_bo.c               |  8 ++++----
>>>   drivers/gpu/drm/virtio/virtgpu_ttm.c       |  1 +
>>>   drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c     |  1 +
>>>   include/drm/ttm/ttm_bo_driver.h            | 10 ++++++++--
>>>   10 files changed, 22 insertions(+), 8 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>> index bf79b73..c29db99 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>> @@ -842,20 +842,21 @@ int amdgpu_bo_get_metadata(struct amdgpu_bo *bo,
>>> void *buffer,
>>>         if (metadata_size)
>>>           *metadata_size = bo->metadata_size;
>>>       if (flags)
>>>           *flags = bo->metadata_flags;
>>>         return 0;
>>>   }
>>>     void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
>>> +               bool evict,
>>>                  struct ttm_mem_reg *new_mem)
>>>   {
>>>       struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
>>>       struct amdgpu_bo *abo;
>>>       struct ttm_mem_reg *old_mem = &bo->mem;
>>>         if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
>>>           return;
>>>         abo = container_of(bo, struct amdgpu_bo, tbo);
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>>> index 5cbf59e..4306b2f 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>>> @@ -148,21 +148,22 @@ void amdgpu_bo_fini(struct amdgpu_device *adev);
>>>   int amdgpu_bo_fbdev_mmap(struct amdgpu_bo *bo,
>>>                   struct vm_area_struct *vma);
>>>   int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64 
>>> tiling_flags);
>>>   void amdgpu_bo_get_tiling_flags(struct amdgpu_bo *bo, u64
>>> *tiling_flags);
>>>   int amdgpu_bo_set_metadata (struct amdgpu_bo *bo, void *metadata,
>>>                   uint32_t metadata_size, uint64_t flags);
>>>   int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
>>>                  size_t buffer_size, uint32_t *metadata_size,
>>>                  uint64_t *flags);
>>>   void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
>>> -                  struct ttm_mem_reg *new_mem);
>>> +               bool evict,
>>> +               struct ttm_mem_reg *new_mem);
>>>   int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
>>>   void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence,
>>>                bool shared);
>>>   u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo);
>>>   int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev,
>>>                      struct amdgpu_ring *ring,
>>>                      struct amdgpu_bo *bo,
>>>                      struct reservation_object *resv,
>>>                      struct dma_fence **fence, bool direct);
>>>   int amdgpu_bo_restore_from_shadow(struct amdgpu_device *adev,
>>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> index e0c0007..6fa1521 100644
>>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> @@ -1187,21 +1187,22 @@ nouveau_bo_move_flips(struct ttm_buffer_object
>>> *bo, bool evict, bool intr,
>>>       ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_gpu, new_mem);
>>>       if (ret)
>>>           goto out;
>>>     out:
>>>       ttm_bo_mem_put(bo, &tmp_mem);
>>>       return ret;
>>>   }
>>>     static void
>>> -nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, struct ttm_mem_reg
>>> *new_mem)
>>> +nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
>>> +             struct ttm_mem_reg *new_mem)
>>>   {
>>>       struct nouveau_bo *nvbo = nouveau_bo(bo);
>>>       struct nvkm_vma *vma;
>>>         /* ttm can now (stupidly) pass the driver bos it didn't
>>> create... */
>>>       if (bo->destroy != nouveau_bo_del_ttm)
>>>           return;
>>>         list_for_each_entry(vma, &nvbo->vma_list, head) {
>>>           if (new_mem && new_mem->mem_type != TTM_PL_SYSTEM &&
>>> diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c
>>> b/drivers/gpu/drm/qxl/qxl_ttm.c
>>> index 1176133..f3939a9 100644
>>> --- a/drivers/gpu/drm/qxl/qxl_ttm.c
>>> +++ b/drivers/gpu/drm/qxl/qxl_ttm.c
>>> @@ -360,20 +360,21 @@ static int qxl_bo_move(struct ttm_buffer_object
>>> *bo,
>>>         if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
>>>           qxl_move_null(bo, new_mem);
>>>           return 0;
>>>       }
>>>       return ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu,
>>>                     new_mem);
>>>   }
>>>     static void qxl_bo_move_notify(struct ttm_buffer_object *bo,
>>> +                   bool evict,
>>>                      struct ttm_mem_reg *new_mem)
>>>   {
>>>       struct qxl_bo *qbo;
>>>       struct qxl_device *qdev;
>>>         if (!qxl_ttm_bo_is_qxl_bo(bo))
>>>           return;
>>>       qbo = to_qxl_bo(bo);
>>>       qdev = qbo->gem_base.dev->dev_private;
>>>   diff --git a/drivers/gpu/drm/radeon/radeon_object.c
>>> b/drivers/gpu/drm/radeon/radeon_object.c
>>> index 41b72ce..74b2760 100644
>>> --- a/drivers/gpu/drm/radeon/radeon_object.c
>>> +++ b/drivers/gpu/drm/radeon/radeon_object.c
>>> @@ -758,20 +758,21 @@ int radeon_bo_check_tiling(struct radeon_bo *bo,
>>> bool has_moved,
>>>           return 0;
>>>       }
>>>         if ((bo->surface_reg >= 0) && !has_moved)
>>>           return 0;
>>>         return radeon_bo_get_surface_reg(bo);
>>>   }
>>>     void radeon_bo_move_notify(struct ttm_buffer_object *bo,
>>> +               bool evict,
>>>                  struct ttm_mem_reg *new_mem)
>>>   {
>>>       struct radeon_bo *rbo;
>>>         if (!radeon_ttm_bo_is_radeon_bo(bo))
>>>           return;
>>>         rbo = container_of(bo, struct radeon_bo, tbo);
>>>       radeon_bo_check_tiling(rbo, 0, 1);
>>>       radeon_vm_bo_invalidate(rbo->rdev, rbo);
>>> diff --git a/drivers/gpu/drm/radeon/radeon_object.h
>>> b/drivers/gpu/drm/radeon/radeon_object.h
>>> index a10bb3d..9ffd821 100644
>>> --- a/drivers/gpu/drm/radeon/radeon_object.h
>>> +++ b/drivers/gpu/drm/radeon/radeon_object.h
>>> @@ -143,20 +143,21 @@ extern void radeon_bo_fini(struct radeon_device
>>> *rdev);
>>>   extern int radeon_bo_list_validate(struct radeon_device *rdev,
>>>                      struct ww_acquire_ctx *ticket,
>>>                      struct list_head *head, int ring);
>>>   extern int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
>>>                   u32 tiling_flags, u32 pitch);
>>>   extern void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
>>>                   u32 *tiling_flags, u32 *pitch);
>>>   extern int radeon_bo_check_tiling(struct radeon_bo *bo, bool 
>>> has_moved,
>>>                   bool force_drop);
>>>   extern void radeon_bo_move_notify(struct ttm_buffer_object *bo,
>>> +                  bool evict,
>>>                     struct ttm_mem_reg *new_mem);
>>>   extern int radeon_bo_fault_reserve_notify(struct ttm_buffer_object
>>> *bo);
>>>   extern int radeon_bo_get_surface_reg(struct radeon_bo *bo);
>>>   extern void radeon_bo_fence(struct radeon_bo *bo, struct
>>> radeon_fence *fence,
>>>                   bool shared);
>>>     /*
>>>    * sub allocation
>>>    */
>>>   diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>> index d506361..d4973e9c 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>> @@ -335,45 +335,45 @@ static int ttm_bo_handle_move_mem(struct
>>> ttm_buffer_object *bo,
>>>               goto out_err;
>>>             if (mem->mem_type != TTM_PL_SYSTEM) {
>>>               ret = ttm_tt_bind(bo->ttm, mem);
>>>               if (ret)
>>>                   goto out_err;
>>>           }
>>>             if (bo->mem.mem_type == TTM_PL_SYSTEM) {
>>>               if (bdev->driver->move_notify)
>>> -                bdev->driver->move_notify(bo, mem);
>>> +                bdev->driver->move_notify(bo, evict, mem);
>>>               bo->mem = *mem;
>>>               mem->mm_node = NULL;
>>>               goto moved;
>>>           }
>>>       }
>>>         if (bdev->driver->move_notify)
>>> -        bdev->driver->move_notify(bo, mem);
>>> +        bdev->driver->move_notify(bo, evict, mem);
>>>         if (!(old_man->flags & TTM_MEMTYPE_FLAG_FIXED) &&
>>>           !(new_man->flags & TTM_MEMTYPE_FLAG_FIXED))
>>>           ret = ttm_bo_move_ttm(bo, interruptible, no_wait_gpu, mem);
>>>       else if (bdev->driver->move)
>>>           ret = bdev->driver->move(bo, evict, interruptible,
>>>                        no_wait_gpu, mem);
>>>       else
>>>           ret = ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu, 
>>> mem);
>>>         if (ret) {
>>>           if (bdev->driver->move_notify) {
>>>               struct ttm_mem_reg tmp_mem = *mem;
>>>               *mem = bo->mem;
>>>               bo->mem = tmp_mem;
>>> -            bdev->driver->move_notify(bo, mem);
>>> +            bdev->driver->move_notify(bo, false, mem);
>>>               bo->mem = *mem;
>>>               *mem = tmp_mem;
>>>           }
>>>             goto out_err;
>>>       }
>>>     moved:
>>>       if (bo->evicted) {
>>>           if (bdev->driver->invalidate_caches) {
>>> @@ -407,21 +407,21 @@ static int ttm_bo_handle_move_mem(struct
>>> ttm_buffer_object *bo,
>>>    * Call bo::reserved.
>>>    * Will release GPU memory type usage on destruction.
>>>    * This is the place to put in driver specific hooks to release
>>>    * driver private resources.
>>>    * Will release the bo::reserved lock.
>>>    */
>>>     static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object 
>>> *bo)
>>>   {
>>>       if (bo->bdev->driver->move_notify)
>>> -        bo->bdev->driver->move_notify(bo, NULL);
>>> +        bo->bdev->driver->move_notify(bo, false, NULL);
>>>         ttm_tt_destroy(bo->ttm);
>>>       bo->ttm = NULL;
>>>       ttm_bo_mem_put(bo, &bo->mem);
>>>         ww_mutex_unlock (&bo->resv->lock);
>>>   }
>>>     static void ttm_bo_flush_all_fences(struct ttm_buffer_object *bo)
>>>   {
>>> diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c
>>> b/drivers/gpu/drm/virtio/virtgpu_ttm.c
>>> index 4a1de9f..10387d7 100644
>>> --- a/drivers/gpu/drm/virtio/virtgpu_ttm.c
>>> +++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c
>>> @@ -379,20 +379,21 @@ static int virtio_gpu_bo_move(struct
>>> ttm_buffer_object *bo,
>>>         ret = ttm_bo_wait(bo, interruptible, no_wait_gpu);
>>>       if (ret)
>>>           return ret;
>>>         virtio_gpu_move_null(bo, new_mem);
>>>       return 0;
>>>   }
>>>     static void virtio_gpu_bo_move_notify(struct ttm_buffer_object 
>>> *tbo,
>>> +                      bool evict,
>>>                         struct ttm_mem_reg *new_mem)
>>>   {
>>>       struct virtio_gpu_object *bo;
>>>       struct virtio_gpu_device *vgdev;
>>>         bo = container_of(tbo, struct virtio_gpu_object, tbo);
>>>       vgdev = (struct virtio_gpu_device 
>>> *)bo->gem_base.dev->dev_private;
>>>         if (!new_mem || (new_mem->placement & TTM_PL_FLAG_SYSTEM)) {
>>>           if (bo->hw_res_handle)
>>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>>> b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>>> index c894a48..caa279b 100644
>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>>> @@ -818,20 +818,21 @@ static int vmw_ttm_fault_reserve_notify(struct
>>> ttm_buffer_object *bo)
>>>    * vmw_move_notify - TTM move_notify_callback
>>>    *
>>>    * @bo: The TTM buffer object about to move.
>>>    * @mem: The struct ttm_mem_reg indicating to what memory
>>>    *       region the move is taking place.
>>>    *
>>>    * Calls move_notify for all subsystems needing it.
>>>    * (currently only resources).
>>>    */
>>>   static void vmw_move_notify(struct ttm_buffer_object *bo,
>>> +                bool evict,
>>>                   struct ttm_mem_reg *mem)
>>>   {
>>>       vmw_resource_move_notify(bo, mem);
>>>       vmw_query_move_notify(bo, mem);
>>>   }
>>>       /**
>>>    * vmw_swap_notify - TTM move_notify_callback
>>>    *
>>> diff --git a/include/drm/ttm/ttm_bo_driver.h
>>> b/include/drm/ttm/ttm_bo_driver.h
>>> index cdbdb40..8b4bec7 100644
>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>> @@ -424,23 +424,29 @@ struct ttm_bo_driver {
>>>        *
>>>        * Called from the map / write / read methods to verify that the
>>>        * caller is permitted to access the buffer object.
>>>        * This member may be set to NULL, which will refuse this kind of
>>>        * access for all buffer objects.
>>>        * This function should return 0 if access is granted, -EPERM
>>> otherwise.
>>>        */
>>>       int (*verify_access)(struct ttm_buffer_object *bo,
>>>                    struct file *filp);
>>>   -    /* hook to notify driver about a driver move so it
>>> -     * can do tiling things */
>>> +    /**
>>> +     * Hook to notify driver about a driver move so it
>>> +     * can do tiling things and book-keeping.
>>> +     *
>>> +     * @evict: whether this move is evicting the buffer from the
>>> graphics
>>> +     * address space
>>> +     */
>>>       void (*move_notify)(struct ttm_buffer_object *bo,
>>> +                bool evict,
>>>                   struct ttm_mem_reg *new_mem);
>>>       /* notify the driver we are taking a fault on this BO
>>>        * and have reserved it */
>>>       int (*fault_reserve_notify)(struct ttm_buffer_object *bo);
>>>         /**
>>>        * notify the driver that we're about to swap out this bo
>>>        */
>>>       void (*swap_notify)(struct ttm_buffer_object *bo);
>>>
>>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx


_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 2/5] drm/amd/amdgpu: move eviction counting to amdgpu_bo_move_notify

[Nicolai Hähnle]

From: Nicolai Hähnle <nicolai.haehnle@amd.com>

This catches evictions of shadow page tables from the GART. Since shadow
page tables are always stored in system memory, amdgpu_bo_move is never
called for them.

This fixes a crash during command submission that occurs when only a shadow
page table and no other BOs were evicted since the last submission.

Fixes: 1baa439fb2f4e586 ("drm/amdgpu: allocate shadow for pd/pt bo V2")
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 4 ++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 4 ----
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index c29db99..d94cdef 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -855,20 +855,24 @@ void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
 	struct amdgpu_bo *abo;
 	struct ttm_mem_reg *old_mem = &bo->mem;
 
 	if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
 		return;
 
 	abo = container_of(bo, struct amdgpu_bo, tbo);
 	amdgpu_vm_bo_invalidate(adev, abo);
 
+	/* remember the eviction */
+	if (evict)
+		atomic64_inc(&adev->num_evictions);
+
 	/* update statistics */
 	if (!new_mem)
 		return;
 
 	/* move_notify is called before move happens */
 	amdgpu_update_memory_usage(adev, &bo->mem, new_mem);
 
 	trace_amdgpu_ttm_bo_move(abo, new_mem->mem_type, old_mem->mem_type);
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 8f18b8e..80924c2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -460,24 +460,20 @@ static int amdgpu_bo_move(struct ttm_buffer_object *bo,
 	struct ttm_mem_reg *old_mem = &bo->mem;
 	int r;
 
 	/* Can't move a pinned BO */
 	abo = container_of(bo, struct amdgpu_bo, tbo);
 	if (WARN_ON_ONCE(abo->pin_count > 0))
 		return -EINVAL;
 
 	adev = amdgpu_ttm_adev(bo->bdev);
 
-	/* remember the eviction */
-	if (evict)
-		atomic64_inc(&adev->num_evictions);
-
 	if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
 		amdgpu_move_null(bo, new_mem);
 		return 0;
 	}
 	if ((old_mem->mem_type == TTM_PL_TT &&
 	     new_mem->mem_type == TTM_PL_SYSTEM) ||
 	    (old_mem->mem_type == TTM_PL_SYSTEM &&
 	     new_mem->mem_type == TTM_PL_TT)) {
 		/* bind is enough */
 		amdgpu_move_null(bo, new_mem);
-- 
2.7.4

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 2/5] drm/amd/amdgpu: move eviction counting to amdgpu_bo_move_notify

[zhoucm1]

On 2016年12月16日 01:10, Nicolai Hähnle wrote:
> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>
> This catches evictions of shadow page tables from the GART. Since shadow
> page tables are always stored in system memory, amdgpu_bo_move is never
> called for them.
>
> This fixes a crash during command submission that occurs when only a shadow
> page table and no other BOs were evicted since the last submission.
>
> Fixes: 1baa439fb2f4e586 ("drm/amdgpu: allocate shadow for pd/pt bo V2")
> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Acked-by: Chunming Zhou <david1.zhou@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 4 ++++
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 4 ----
>   2 files changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> index c29db99..d94cdef 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> @@ -855,20 +855,24 @@ void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
>   	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
>   	struct amdgpu_bo *abo;
>   	struct ttm_mem_reg *old_mem = &bo->mem;
>   
>   	if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
>   		return;
>   
>   	abo = container_of(bo, struct amdgpu_bo, tbo);
>   	amdgpu_vm_bo_invalidate(adev, abo);
>   
> +	/* remember the eviction */
> +	if (evict)
> +		atomic64_inc(&adev->num_evictions);
> +
>   	/* update statistics */
>   	if (!new_mem)
>   		return;
>   
>   	/* move_notify is called before move happens */
>   	amdgpu_update_memory_usage(adev, &bo->mem, new_mem);
>   
>   	trace_amdgpu_ttm_bo_move(abo, new_mem->mem_type, old_mem->mem_type);
>   }
>   
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index 8f18b8e..80924c2 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -460,24 +460,20 @@ static int amdgpu_bo_move(struct ttm_buffer_object *bo,
>   	struct ttm_mem_reg *old_mem = &bo->mem;
>   	int r;
>   
>   	/* Can't move a pinned BO */
>   	abo = container_of(bo, struct amdgpu_bo, tbo);
>   	if (WARN_ON_ONCE(abo->pin_count > 0))
>   		return -EINVAL;
>   
>   	adev = amdgpu_ttm_adev(bo->bdev);
>   
> -	/* remember the eviction */
> -	if (evict)
> -		atomic64_inc(&adev->num_evictions);
> -
>   	if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
>   		amdgpu_move_null(bo, new_mem);
>   		return 0;
>   	}
>   	if ((old_mem->mem_type == TTM_PL_TT &&
>   	     new_mem->mem_type == TTM_PL_SYSTEM) ||
>   	    (old_mem->mem_type == TTM_PL_SYSTEM &&
>   	     new_mem->mem_type == TTM_PL_TT)) {
>   		/* bind is enough */
>   		amdgpu_move_null(bo, new_mem);

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 3/5] drm/amd/amdgpu: add amdgpu_bo_gpu_accessible helper function

[Nicolai Hähnle]

From: Nicolai Hähnle <nicolai.haehnle@amd.com>

Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index 4306b2f..15a723a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -107,20 +107,29 @@ static inline unsigned amdgpu_bo_gpu_page_alignment(struct amdgpu_bo *bo)
  * amdgpu_bo_mmap_offset - return mmap offset of bo
  * @bo:	amdgpu object for which we query the offset
  *
  * Returns mmap offset of the object.
  */
 static inline u64 amdgpu_bo_mmap_offset(struct amdgpu_bo *bo)
 {
 	return drm_vma_node_offset_addr(&bo->tbo.vma_node);
 }
 
+/**
+ * amdgpu_bo_gpu_accessible - return whether the bo is currently in memory that
+ * is accessible to the GPU.
+ */
+static inline bool amdgpu_bo_gpu_accessible(struct amdgpu_bo *bo)
+{
+	return bo->tbo.mem.mem_type != TTM_PL_SYSTEM;
+}
+
 int amdgpu_bo_create(struct amdgpu_device *adev,
 			    unsigned long size, int byte_align,
 			    bool kernel, u32 domain, u64 flags,
 			    struct sg_table *sg,
 			    struct reservation_object *resv,
 			    struct amdgpu_bo **bo_ptr);
 int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
 				unsigned long size, int byte_align,
 				bool kernel, u32 domain, u64 flags,
 				struct sg_table *sg,
-- 
2.7.4

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 3/5] drm/amd/amdgpu: add amdgpu_bo_gpu_accessible helper function

[zhoucm1]

On 2016年12月16日 01:10, Nicolai Hähnle wrote:
> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>
> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 9 +++++++++
>   1 file changed, 9 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> index 4306b2f..15a723a 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> @@ -107,20 +107,29 @@ static inline unsigned amdgpu_bo_gpu_page_alignment(struct amdgpu_bo *bo)
>    * amdgpu_bo_mmap_offset - return mmap offset of bo
>    * @bo:	amdgpu object for which we query the offset
>    *
>    * Returns mmap offset of the object.
>    */
>   static inline u64 amdgpu_bo_mmap_offset(struct amdgpu_bo *bo)
>   {
>   	return drm_vma_node_offset_addr(&bo->tbo.vma_node);
>   }
>   
> +/**
> + * amdgpu_bo_gpu_accessible - return whether the bo is currently in memory that
> + * is accessible to the GPU.
> + */
> +static inline bool amdgpu_bo_gpu_accessible(struct amdgpu_bo *bo)
> +{
> +	return bo->tbo.mem.mem_type != TTM_PL_SYSTEM;
> +}
> +
>   int amdgpu_bo_create(struct amdgpu_device *adev,
>   			    unsigned long size, int byte_align,
>   			    bool kernel, u32 domain, u64 flags,
>   			    struct sg_table *sg,
>   			    struct reservation_object *resv,
>   			    struct amdgpu_bo **bo_ptr);
>   int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
>   				unsigned long size, int byte_align,
>   				bool kernel, u32 domain, u64 flags,
>   				struct sg_table *sg,

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 4/5] drm/amd/amdgpu: add check that shadow page directory is GPU-accessible

[Nicolai Hähnle]

From: Nicolai Hähnle <nicolai.haehnle@amd.com>

Skip amdgpu_gem_va_update_vm when shadow the page directory is swapped out.
Clean up the check for non-shadow BOs as well using the new helper function.

This fixes a crash with the stack trace:

amdgpu_gem_va_update_vm
-> amdgpu_vm_update_page_directory
 -> amdgpu_ttm_bind
  -> amdgpu_gtt_mgr_alloc

Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index cd62f6f..4e1eb05 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -489,44 +489,49 @@ static int amdgpu_gem_va_check(void *param, struct amdgpu_bo *bo)
  * vital here, so they are not reported back to userspace.
  */
 static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
 				    struct amdgpu_bo_va *bo_va,
 				    uint32_t operation)
 {
 	struct ttm_validate_buffer tv, *entry;
 	struct amdgpu_bo_list_entry vm_pd;
 	struct ww_acquire_ctx ticket;
 	struct list_head list, duplicates;
-	unsigned domain;
 	int r;
 
 	INIT_LIST_HEAD(&list);
 	INIT_LIST_HEAD(&duplicates);
 
 	tv.bo = &bo_va->bo->tbo;
 	tv.shared = true;
 	list_add(&tv.head, &list);
 
 	amdgpu_vm_get_pd_bo(bo_va->vm, &list, &vm_pd);
 
 	/* Provide duplicates to avoid -EALREADY */
 	r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates);
 	if (r)
 		goto error_print;
 
 	list_for_each_entry(entry, &list, head) {
-		domain = amdgpu_mem_type_to_domain(entry->bo->mem.mem_type);
+		struct amdgpu_bo *bo =
+			container_of(entry->bo, struct amdgpu_bo, tbo);
+
 		/* if anything is swapped out don't swap it in here,
 		   just abort and wait for the next CS */
-		if (domain == AMDGPU_GEM_DOMAIN_CPU)
+		if (!amdgpu_bo_gpu_accessible(bo))
+			goto error_unreserve;
+
+		if (bo->shadow && !amdgpu_bo_gpu_accessible(bo->shadow))
 			goto error_unreserve;
 	}
+
 	r = amdgpu_vm_validate_pt_bos(adev, bo_va->vm, amdgpu_gem_va_check,
 				      NULL);
 	if (r)
 		goto error_unreserve;
 
 	r = amdgpu_vm_update_page_directory(adev, bo_va->vm);
 	if (r)
 		goto error_unreserve;
 
 	r = amdgpu_vm_clear_freed(adev, bo_va->vm);
-- 
2.7.4

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 4/5] drm/amd/amdgpu: add check that shadow page directory is GPU-accessible

[zhoucm1]

On 2016年12月16日 01:10, Nicolai Hähnle wrote:
> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>
> Skip amdgpu_gem_va_update_vm when shadow the page directory is swapped out.
> Clean up the check for non-shadow BOs as well using the new helper function.
>
> This fixes a crash with the stack trace:
>
> amdgpu_gem_va_update_vm
> -> amdgpu_vm_update_page_directory
>   -> amdgpu_ttm_bind
>    -> amdgpu_gtt_mgr_alloc
>
> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 11 ++++++++---
>   1 file changed, 8 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> index cd62f6f..4e1eb05 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> @@ -489,44 +489,49 @@ static int amdgpu_gem_va_check(void *param, struct amdgpu_bo *bo)
>    * vital here, so they are not reported back to userspace.
>    */
>   static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
>   				    struct amdgpu_bo_va *bo_va,
>   				    uint32_t operation)
>   {
>   	struct ttm_validate_buffer tv, *entry;
>   	struct amdgpu_bo_list_entry vm_pd;
>   	struct ww_acquire_ctx ticket;
>   	struct list_head list, duplicates;
> -	unsigned domain;
>   	int r;
>   
>   	INIT_LIST_HEAD(&list);
>   	INIT_LIST_HEAD(&duplicates);
>   
>   	tv.bo = &bo_va->bo->tbo;
>   	tv.shared = true;
>   	list_add(&tv.head, &list);
>   
>   	amdgpu_vm_get_pd_bo(bo_va->vm, &list, &vm_pd);
>   
>   	/* Provide duplicates to avoid -EALREADY */
>   	r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates);
>   	if (r)
>   		goto error_print;
>   
>   	list_for_each_entry(entry, &list, head) {
> -		domain = amdgpu_mem_type_to_domain(entry->bo->mem.mem_type);
> +		struct amdgpu_bo *bo =
> +			container_of(entry->bo, struct amdgpu_bo, tbo);
> +
>   		/* if anything is swapped out don't swap it in here,
>   		   just abort and wait for the next CS */
> -		if (domain == AMDGPU_GEM_DOMAIN_CPU)
> +		if (!amdgpu_bo_gpu_accessible(bo))
> +			goto error_unreserve;
> +
> +		if (bo->shadow && !amdgpu_bo_gpu_accessible(bo->shadow))
>   			goto error_unreserve;
>   	}
> +
>   	r = amdgpu_vm_validate_pt_bos(adev, bo_va->vm, amdgpu_gem_va_check,
>   				      NULL);
>   	if (r)
>   		goto error_unreserve;
>   
>   	r = amdgpu_vm_update_page_directory(adev, bo_va->vm);
>   	if (r)
>   		goto error_unreserve;
>   
>   	r = amdgpu_vm_clear_freed(adev, bo_va->vm);

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 5/5] drm/amd/amdgpu: add check that shadow page tables are GPU-accessible

[Nicolai Hähnle]

From: Nicolai Hähnle <nicolai.haehnle@amd.com>

Skip amdgpu_gem_va_update_vm otherwise. Also clean up the check for the
non-shadow page tables using the new helper function.

This fixes a crash with the stack trace:

amdgpu_gem_va_update_vm
-> amdgpu_vm_update_page_directory
 -> amdgpu_ttm_bind
  -> amdgpu_gtt_mgr_alloc

Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 4e1eb05..d91c80b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -464,26 +464,29 @@ int amdgpu_gem_metadata_ioctl(struct drm_device *dev, void *data,
 
 unreserve:
 	amdgpu_bo_unreserve(robj);
 out:
 	drm_gem_object_unreference_unlocked(gobj);
 	return r;
 }
 
 static int amdgpu_gem_va_check(void *param, struct amdgpu_bo *bo)
 {
-	unsigned domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
-
 	/* if anything is swapped out don't swap it in here,
 	   just abort and wait for the next CS */
+	if (!amdgpu_bo_gpu_accessible(bo))
+		return -ERESTARTSYS;
+
+	if (bo->shadow && !amdgpu_bo_gpu_accessible(bo))
+		return -ERESTARTSYS;
 
-	return domain == AMDGPU_GEM_DOMAIN_CPU ? -ERESTARTSYS : 0;
+	return 0;
 }
 
 /**
  * amdgpu_gem_va_update_vm -update the bo_va in its VM
  *
  * @adev: amdgpu_device pointer
  * @bo_va: bo_va to update
  *
  * Update the bo_va directly after setting it's address. Errors are not
  * vital here, so they are not reported back to userspace.
-- 
2.7.4

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2] drm/amd/amdgpu: add check that shadow page tables are GPU-accessible

[Nicolai Hähnle]

From: Nicolai Hähnle <nicolai.haehnle@amd.com>

Skip amdgpu_gem_va_update_vm otherwise. Also clean up the check for the
non-shadow page tables using the new helper function.

This fixes a crash with the stack trace:

amdgpu_gem_va_update_vm
-> amdgpu_vm_update_page_directory
 -> amdgpu_ttm_bind
  -> amdgpu_gtt_mgr_alloc

v2: actually check bo->shadow instead of just checking bo twice

Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 4e1eb05..9bd1b4e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -464,26 +464,29 @@ int amdgpu_gem_metadata_ioctl(struct drm_device *dev, void *data,
 
 unreserve:
 	amdgpu_bo_unreserve(robj);
 out:
 	drm_gem_object_unreference_unlocked(gobj);
 	return r;
 }
 
 static int amdgpu_gem_va_check(void *param, struct amdgpu_bo *bo)
 {
-	unsigned domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
-
 	/* if anything is swapped out don't swap it in here,
 	   just abort and wait for the next CS */
+	if (!amdgpu_bo_gpu_accessible(bo))
+		return -ERESTARTSYS;
+
+	if (bo->shadow && !amdgpu_bo_gpu_accessible(bo->shadow))
+		return -ERESTARTSYS;
 
-	return domain == AMDGPU_GEM_DOMAIN_CPU ? -ERESTARTSYS : 0;
+	return 0;
 }
 
 /**
  * amdgpu_gem_va_update_vm -update the bo_va in its VM
  *
  * @adev: amdgpu_device pointer
  * @bo_va: bo_va to update
  *
  * Update the bo_va directly after setting it's address. Errors are not
  * vital here, so they are not reported back to userspace.
-- 
2.7.4

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH v2] drm/amd/amdgpu: add check that shadow page tables are GPU-accessible

[zhoucm1]

On 2016年12月16日 01:59, Nicolai Hähnle wrote:
> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>
> Skip amdgpu_gem_va_update_vm otherwise. Also clean up the check for the
> non-shadow page tables using the new helper function.
>
> This fixes a crash with the stack trace:
>
> amdgpu_gem_va_update_vm
> -> amdgpu_vm_update_page_directory
>   -> amdgpu_ttm_bind
>    -> amdgpu_gtt_mgr_alloc
>
> v2: actually check bo->shadow instead of just checking bo twice
>
> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 9 ++++++---
>   1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> index 4e1eb05..9bd1b4e 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> @@ -464,26 +464,29 @@ int amdgpu_gem_metadata_ioctl(struct drm_device *dev, void *data,
>   
>   unreserve:
>   	amdgpu_bo_unreserve(robj);
>   out:
>   	drm_gem_object_unreference_unlocked(gobj);
>   	return r;
>   }
>   
>   static int amdgpu_gem_va_check(void *param, struct amdgpu_bo *bo)
>   {
> -	unsigned domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
> -
>   	/* if anything is swapped out don't swap it in here,
>   	   just abort and wait for the next CS */
> +	if (!amdgpu_bo_gpu_accessible(bo))
> +		return -ERESTARTSYS;
> +
> +	if (bo->shadow && !amdgpu_bo_gpu_accessible(bo->shadow))
> +		return -ERESTARTSYS;
>   
> -	return domain == AMDGPU_GEM_DOMAIN_CPU ? -ERESTARTSYS : 0;
> +	return 0;
>   }
>   
>   /**
>    * amdgpu_gem_va_update_vm -update the bo_va in its VM
>    *
>    * @adev: amdgpu_device pointer
>    * @bo_va: bo_va to update
>    *
>    * Update the bo_va directly after setting it's address. Errors are not
>    * vital here, so they are not reported back to userspace.

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 5/5] drm/amd/amdgpu: add check that shadow page tables are GPU-accessible

[zhoucm1]

On 2016年12月16日 01:10, Nicolai Hähnle wrote:
> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>
> Skip amdgpu_gem_va_update_vm otherwise. Also clean up the check for the
> non-shadow page tables using the new helper function.
>
> This fixes a crash with the stack trace:
>
> amdgpu_gem_va_update_vm
> -> amdgpu_vm_update_page_directory
>   -> amdgpu_ttm_bind
>    -> amdgpu_gtt_mgr_alloc
>
> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 9 ++++++---
>   1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> index 4e1eb05..d91c80b 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> @@ -464,26 +464,29 @@ int amdgpu_gem_metadata_ioctl(struct drm_device *dev, void *data,
>   
>   unreserve:
>   	amdgpu_bo_unreserve(robj);
>   out:
>   	drm_gem_object_unreference_unlocked(gobj);
>   	return r;
>   }
>   
>   static int amdgpu_gem_va_check(void *param, struct amdgpu_bo *bo)
>   {
> -	unsigned domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
> -
>   	/* if anything is swapped out don't swap it in here,
>   	   just abort and wait for the next CS */
> +	if (!amdgpu_bo_gpu_accessible(bo))
> +		return -ERESTARTSYS;
> +
> +	if (bo->shadow && !amdgpu_bo_gpu_accessible(bo))

this should be 'if (bo->shadow && !amdgpu_bo_gpu_accessible(bo->shadow))'.

Regards,
David Zhou

> +		return -ERESTARTSYS;
>   
> -	return domain == AMDGPU_GEM_DOMAIN_CPU ? -ERESTARTSYS : 0;
> +	return 0;
>   }
>   
>   /**
>    * amdgpu_gem_va_update_vm -update the bo_va in its VM
>    *
>    * @adev: amdgpu_device pointer
>    * @bo_va: bo_va to update
>    *
>    * Update the bo_va directly after setting it's address. Errors are not
>    * vital here, so they are not reported back to userspace.

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx