drm/ttm: make ttm reservation calls behave like reservation calls

This commit converts the source of the val_seq counter to
the ww_mutex api. The reservation objects are converted later,
because there is still a lockdep splat in nouveau that has to
resolved first.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drivers/gpu/drm/nouveau/nouveau_gem.c

@@ -277,10 +277,12 @@ struct validate_op {
  struct list_head vram_list;
  struct list_head gart_list;
  struct list_head both_list;
+ struct ww_acquire_ctx ticket;
 };
 
 static void
-validate_fini_list(struct list_head *list, struct nouveau_fence *fence)
+validate_fini_list(struct list_head *list, struct nouveau_fence *fence,
+ struct ww_acquire_ctx *ticket)
 {
  struct list_head *entry, *tmp;
  struct nouveau_bo *nvbo;
@@ -297,17 +299,24 @@ validate_fini_list(struct list_head *list, struct nouveau_fence *fence)
 
  list_del(&nvbo->entry);
  nvbo->reserved_by = NULL;
- ttm_bo_unreserve(&nvbo->bo);
+ ttm_bo_unreserve_ticket(&nvbo->bo, ticket);
  drm_gem_object_unreference_unlocked(nvbo->gem);
  }
 }
 
 static void
-validate_fini(struct validate_op *op, struct nouveau_fence* fence)
+validate_fini_no_ticket(struct validate_op *op, struct nouveau_fence *fence)
 {
- validate_fini_list(&op->vram_list, fence);
- validate_fini_list(&op->gart_list, fence);
- validate_fini_list(&op->both_list, fence);
+ validate_fini_list(&op->vram_list, fence, &op->ticket);
+ validate_fini_list(&op->gart_list, fence, &op->ticket);
+ validate_fini_list(&op->both_list, fence, &op->ticket);
+}
+
+static void
+validate_fini(struct validate_op *op, struct nouveau_fence *fence)
+{
+ validate_fini_no_ticket(op, fence);
+ ww_acquire_fini(&op->ticket);
 }
 
 static int
@@ -317,13 +326,11 @@ validate_init(struct nouveau_channel *chan, struct drm_file *file_priv,
 {
  struct nouveau_cli *cli = nouveau_cli(file_priv);
  struct drm_device *dev = chan->drm->dev;
- struct nouveau_drm *drm = nouveau_drm(dev);
- uint32_t sequence;
  int trycnt = 0;
  int ret, i;
  struct nouveau_bo *res_bo = NULL;
 
- sequence = atomic_add_return(1, &drm->ttm.validate_sequence);
+ ww_acquire_init(&op->ticket, &reservation_ww_class);
 retry:
  if (++trycnt > 100000) {
  NV_ERROR(cli, "%s failed and gave up.\n", __func__);
@@ -338,6 +345,7 @@ validate_init(struct nouveau_channel *chan, struct drm_file *file_priv,
  gem = drm_gem_object_lookup(dev, file_priv, b->handle);
  if (!gem) {
  NV_ERROR(cli, "Unknown handle 0x%08x\n", b->handle);
+ ww_acquire_done(&op->ticket);
  validate_fini(op, NULL);
  return -ENOENT;
  }
@@ -352,21 +360,23 @@ validate_init(struct nouveau_channel *chan, struct drm_file *file_priv,
  NV_ERROR(cli, "multiple instances of buffer %d on "
  "validation list\n", b->handle);
  drm_gem_object_unreference_unlocked(gem);
+ ww_acquire_done(&op->ticket);
  validate_fini(op, NULL);
  return -EINVAL;
  }
 
- ret = ttm_bo_reserve(&nvbo->bo, true, false, true, sequence);
+ ret = ttm_bo_reserve(&nvbo->bo, true, false, true, &op->ticket);
  if (ret) {
- validate_fini(op, NULL);
+ validate_fini_no_ticket(op, NULL);
  if (unlikely(ret == -EAGAIN)) {
- sequence = atomic_add_return(1, &drm->ttm.validate_sequence);
  ret = ttm_bo_reserve_slowpath(&nvbo->bo, true,
- sequence);
+ &op->ticket);
  if (!ret)
  res_bo = nvbo;
  }
  if (unlikely(ret)) {
+ ww_acquire_done(&op->ticket);
+ ww_acquire_fini(&op->ticket);
  drm_gem_object_unreference_unlocked(gem);
  if (ret != -ERESTARTSYS)
  NV_ERROR(cli, "fail reserve\n");
@@ -390,13 +400,15 @@ validate_init(struct nouveau_channel *chan, struct drm_file *file_priv,
  NV_ERROR(cli, "invalid valid domains: 0x%08x\n",
  b->valid_domains);
  list_add_tail(&nvbo->entry, &op->both_list);
+ ww_acquire_done(&op->ticket);
  validate_fini(op, NULL);
  return -EINVAL;
  }
  if (nvbo == res_bo)
  goto retry;
  }
 
+ ww_acquire_done(&op->ticket);
  return 0;
 }
 

drivers/gpu/drm/radeon/radeon.h

@@ -979,6 +979,7 @@ struct radeon_cs_parser {
  u32 cs_flags;
  u32 ring;
  s32 priority;
+ struct ww_acquire_ctx ticket;
 };
 
 extern int radeon_cs_finish_pages(struct radeon_cs_parser *p);

drivers/gpu/drm/radeon/radeon_cs.c

@@ -106,7 +106,7 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
  radeon_bo_list_add_object(&p->relocs[i].lobj,
  &p->validated);
  }
- return radeon_bo_list_validate(&p->validated, p->ring);
+ return radeon_bo_list_validate(&p->ticket, &p->validated, p->ring);
 }
 
 static int radeon_cs_get_ring(struct radeon_cs_parser *p, u32 ring, s32 priority)
@@ -314,15 +314,17 @@ int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data)
  * If error is set than unvalidate buffer, otherwise just free memory
  * used by parsing context.
  **/
-static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error)
+static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error, bool backoff)
 {
  unsigned i;
 
  if (!error) {
- ttm_eu_fence_buffer_objects(&parser->validated,
+ ttm_eu_fence_buffer_objects(&parser->ticket,
+ &parser->validated,
  parser->ib.fence);
- } else {
- ttm_eu_backoff_reservation(&parser->validated);
+ } else if (backoff) {
+ ttm_eu_backoff_reservation(&parser->ticket,
+ &parser->validated);
  }
 
  if (parser->relocs != NULL) {
@@ -535,7 +537,7 @@ int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
  r = radeon_cs_parser_init(&parser, data);
  if (r) {
  DRM_ERROR("Failed to initialize parser !\n");
- radeon_cs_parser_fini(&parser, r);
+ radeon_cs_parser_fini(&parser, r, false);
  up_read(&rdev->exclusive_lock);
  r = radeon_cs_handle_lockup(rdev, r);
  return r;
@@ -544,7 +546,7 @@ int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
  if (r) {
  if (r != -ERESTARTSYS)
  DRM_ERROR("Failed to parse relocation %d!\n", r);
- radeon_cs_parser_fini(&parser, r);
+ radeon_cs_parser_fini(&parser, r, false);
  up_read(&rdev->exclusive_lock);
  r = radeon_cs_handle_lockup(rdev, r);
  return r;
@@ -563,7 +565,7 @@ int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
  goto out;
  }
 out:
- radeon_cs_parser_fini(&parser, r);
+ radeon_cs_parser_fini(&parser, r, true);
  up_read(&rdev->exclusive_lock);
  r = radeon_cs_handle_lockup(rdev, r);
  return r;

drivers/gpu/drm/radeon/radeon_object.c

@@ -349,14 +349,15 @@ void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
  }
 }
 
-int radeon_bo_list_validate(struct list_head *head, int ring)
+int radeon_bo_list_validate(struct ww_acquire_ctx *ticket,
+ struct list_head *head, int ring)
 {
  struct radeon_bo_list *lobj;
  struct radeon_bo *bo;
  u32 domain;
  int r;
 
- r = ttm_eu_reserve_buffers(head);
+ r = ttm_eu_reserve_buffers(ticket, head);
  if (unlikely(r != 0)) {
  return r;
  }

drivers/gpu/drm/radeon/radeon_object.h

@@ -128,7 +128,8 @@ extern int radeon_bo_init(struct radeon_device *rdev);
 extern void radeon_bo_fini(struct radeon_device *rdev);
 extern void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
  struct list_head *head);
-extern int radeon_bo_list_validate(struct list_head *head, int ring);
+extern int radeon_bo_list_validate(struct ww_acquire_ctx *ticket,
+ struct list_head *head, int ring);
 extern int radeon_bo_fbdev_mmap(struct radeon_bo *bo,
  struct vm_area_struct *vma);
 extern int radeon_bo_set_tiling_flags(struct radeon_bo *bo,

drivers/gpu/drm/radeon/radeon_uvd.c

@@ -550,6 +550,7 @@ static int radeon_uvd_send_msg(struct radeon_device *rdev,
  struct radeon_fence **fence)
 {
  struct ttm_validate_buffer tv;
+ struct ww_acquire_ctx ticket;
  struct list_head head;
  struct radeon_ib ib;
  uint64_t addr;
@@ -561,24 +562,20 @@ static int radeon_uvd_send_msg(struct radeon_device *rdev,
  INIT_LIST_HEAD(&head);
  list_add(&tv.head, &head);
 
- r = ttm_eu_reserve_buffers(&head);
+ r = ttm_eu_reserve_buffers(&ticket, &head);
  if (r)
  return r;
 
  radeon_ttm_placement_from_domain(bo, RADEON_GEM_DOMAIN_VRAM);
  radeon_uvd_force_into_uvd_segment(bo);
 
  r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
- if (r) {
- ttm_eu_backoff_reservation(&head);
- return r;
- }
+ if (r) 
+ goto err;
 
  r = radeon_ib_get(rdev, ring, &ib, NULL, 16);
- if (r) {
- ttm_eu_backoff_reservation(&head);
- return r;
- }
+ if (r)
+ goto err;
 
  addr = radeon_bo_gpu_offset(bo);
  ib.ptr[0] = PACKET0(UVD_GPCOM_VCPU_DATA0, 0);
@@ -592,18 +589,20 @@ static int radeon_uvd_send_msg(struct radeon_device *rdev,
  ib.length_dw = 16;
 
  r = radeon_ib_schedule(rdev, &ib, NULL);
- if (r) {
- ttm_eu_backoff_reservation(&head);
- return r;
- }
- ttm_eu_fence_buffer_objects(&head, ib.fence);
+ if (r)
+ goto err;
+ ttm_eu_fence_buffer_objects(&ticket, &head, ib.fence);
 
  if (fence)
  *fence = radeon_fence_ref(ib.fence);
 
  radeon_ib_free(rdev, &ib);
  radeon_bo_unref(&bo);
  return 0;
+
+err:
+ ttm_eu_backoff_reservation(&ticket, &head);
+ return r;
 }
 
 /* multiple fence commands without any stream commands in between can