Revert "drm/ttm: add a way to bo_wait for either the last read or last write"

This reverts commit dfadbbdb57b3f2bb33e14f129a43047c6f0caefa.

Further upstream discussion between Marek and Thomas decided this wasn't
fully baked and needed further work, so revert it before it hits mainline.

Signed-off-by: Dave Airlie <airlied@redhat.com>

10 files changed, 38 insertions, 120 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 424dff5d0ab..7226f419e17 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1104,8 +1104,7 @@ nouveau_bo_vma_del(struct nouveau_bo *nvbo, struct nouveau_vma *vma)
 	if (vma->node) {
 		if (nvbo->bo.mem.mem_type != TTM_PL_SYSTEM) {
 			spin_lock(&nvbo->bo.bdev->fence_lock);
-			ttm_bo_wait(&nvbo->bo, false, false, false,
-				    TTM_USAGE_READWRITE);
+			ttm_bo_wait(&nvbo->bo, false, false, false);
 			spin_unlock(&nvbo->bo.bdev->fence_lock);
 			nouveau_vm_unmap(vma);
 		}
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
index 322bf62a064..5f0bc57fdaa 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -589,8 +589,7 @@ nouveau_gem_pushbuf_reloc_apply(struct drm_device *dev,
 		}
 
 		spin_lock(&nvbo->bo.bdev->fence_lock);
-		ret = ttm_bo_wait(&nvbo->bo, false, false, false,
-				  TTM_USAGE_READWRITE);
+		ret = ttm_bo_wait(&nvbo->bo, false, false, false);
 		spin_unlock(&nvbo->bo.bdev->fence_lock);
 		if (ret) {
 			NV_ERROR(dev, "reloc wait_idle failed: %d\n", ret);
@@ -826,7 +825,7 @@ nouveau_gem_ioctl_cpu_prep(struct drm_device *dev, void *data,
 	nvbo = nouveau_gem_object(gem);
 
 	spin_lock(&nvbo->bo.bdev->fence_lock);
-	ret = ttm_bo_wait(&nvbo->bo, true, true, no_wait, TTM_USAGE_READWRITE);
+	ret = ttm_bo_wait(&nvbo->bo, true, true, no_wait);
 	spin_unlock(&nvbo->bo.bdev->fence_lock);
 	drm_gem_object_unreference_unlocked(gem);
 	return ret;
diff --git a/drivers/gpu/drm/radeon/radeon_cs.c b/drivers/gpu/drm/radeon/radeon_cs.c
index 14e85315110..fae00c0d75a 100644
--- a/drivers/gpu/drm/radeon/radeon_cs.c
+++ b/drivers/gpu/drm/radeon/radeon_cs.c
@@ -80,7 +80,6 @@ int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
 			p->relocs[i].lobj.wdomain = r->write_domain;
 			p->relocs[i].lobj.rdomain = r->read_domains;
 			p->relocs[i].lobj.tv.bo = &p->relocs[i].robj->tbo;
-			p->relocs[i].lobj.tv.usage = TTM_USAGE_READWRITE;
 			p->relocs[i].handle = r->handle;
 			p->relocs[i].flags = r->flags;
 			radeon_bo_list_add_object(&p->relocs[i].lobj,
diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
index b8f75f5d344..1c851521f45 100644
--- a/drivers/gpu/drm/radeon/radeon_object.c
+++ b/drivers/gpu/drm/radeon/radeon_object.c
@@ -527,7 +527,7 @@ int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type, bool no_wait)
 	if (mem_type)
 		*mem_type = bo->tbo.mem.mem_type;
 	if (bo->tbo.sync_obj)
-		r = ttm_bo_wait(&bo->tbo, true, true, no_wait, false);
+		r = ttm_bo_wait(&bo->tbo, true, true, no_wait);
 	spin_unlock(&bo->tbo.bdev->fence_lock);
 	ttm_bo_unreserve(&bo->tbo);
 	return r;
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 50fc8e4c9a3..617b64678fc 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -499,7 +499,7 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
 	int ret;
 
 	spin_lock(&bdev->fence_lock);
-	(void) ttm_bo_wait(bo, false, false, true, TTM_USAGE_READWRITE);
+	(void) ttm_bo_wait(bo, false, false, true);
 	if (!bo->sync_obj) {
 
 		spin_lock(&glob->lru_lock);
@@ -567,8 +567,7 @@ static int ttm_bo_cleanup_refs(struct ttm_buffer_object *bo,
 
 retry:
 	spin_lock(&bdev->fence_lock);
-	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu,
-			  TTM_USAGE_READWRITE);
+	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
 	spin_unlock(&bdev->fence_lock);
 
 	if (unlikely(ret != 0))
@@ -727,8 +726,7 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
 	int ret = 0;
 
 	spin_lock(&bdev->fence_lock);
-	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu,
-			  TTM_USAGE_READWRITE);
+	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
 	spin_unlock(&bdev->fence_lock);
 
 	if (unlikely(ret != 0)) {
@@ -1075,8 +1073,7 @@ int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
 	 * instead of doing it here.
 	 */
 	spin_lock(&bdev->fence_lock);
-	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu,
-			  TTM_USAGE_READWRITE);
+	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
 	spin_unlock(&bdev->fence_lock);
 	if (ret)
 		return ret;
@@ -1697,83 +1694,34 @@ out_unlock:
 	return ret;
 }
 
-static void ttm_bo_unref_sync_obj_locked(struct ttm_buffer_object *bo,
-					 void *sync_obj,
-					 void **extra_sync_obj)
-{
-	struct ttm_bo_device *bdev = bo->bdev;
-	struct ttm_bo_driver *driver = bdev->driver;
-	void *tmp_obj = NULL, *tmp_obj_read = NULL, *tmp_obj_write = NULL;
-
-	/* We must unref the sync obj wherever it's ref'd.
-	 * Note that if we unref bo->sync_obj, we can unref both the read
-	 * and write sync objs too, because they can't be newer than
-	 * bo->sync_obj, so they are no longer relevant. */
-	if (sync_obj == bo->sync_obj ||
-	    sync_obj == bo->sync_obj_read) {
-		tmp_obj_read = bo->sync_obj_read;
-		bo->sync_obj_read = NULL;
-	}
-	if (sync_obj == bo->sync_obj ||
-	    sync_obj == bo->sync_obj_write) {
-		tmp_obj_write = bo->sync_obj_write;
-		bo->sync_obj_write = NULL;
-	}
-	if (sync_obj == bo->sync_obj) {
-		tmp_obj = bo->sync_obj;
-		bo->sync_obj = NULL;
-	}
-
-	clear_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags);
-	spin_unlock(&bdev->fence_lock);
-	if (tmp_obj)
-		driver->sync_obj_unref(&tmp_obj);
-	if (tmp_obj_read)
-		driver->sync_obj_unref(&tmp_obj_read);
-	if (tmp_obj_write)
-		driver->sync_obj_unref(&tmp_obj_write);
-	if (extra_sync_obj)
-		driver->sync_obj_unref(extra_sync_obj);
-	spin_lock(&bdev->fence_lock);
-}
-
 int ttm_bo_wait(struct ttm_buffer_object *bo,
-		bool lazy, bool interruptible, bool no_wait,
-		enum ttm_buffer_usage usage)
+		bool lazy, bool interruptible, bool no_wait)
 {
 	struct ttm_bo_driver *driver = bo->bdev->driver;
 	struct ttm_bo_device *bdev = bo->bdev;
 	void *sync_obj;
 	void *sync_obj_arg;
 	int ret = 0;
-	void **bo_sync_obj;
 
-	switch (usage) {
-	case TTM_USAGE_READ:
-		bo_sync_obj = &bo->sync_obj_read;
-		break;
-	case TTM_USAGE_WRITE:
-		bo_sync_obj = &bo->sync_obj_write;
-		break;
-	case TTM_USAGE_READWRITE:
-	default:
-		bo_sync_obj = &bo->sync_obj;
-	}
-
-	if (likely(*bo_sync_obj == NULL))
+	if (likely(bo->sync_obj == NULL))
 		return 0;
 
-	while (*bo_sync_obj) {
+	while (bo->sync_obj) {
 
-		if (driver->sync_obj_signaled(*bo_sync_obj, bo->sync_obj_arg)) {
-			ttm_bo_unref_sync_obj_locked(bo, *bo_sync_obj, NULL);
+		if (driver->sync_obj_signaled(bo->sync_obj, bo->sync_obj_arg)) {
+			void *tmp_obj = bo->sync_obj;
+			bo->sync_obj = NULL;
+			clear_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags);
+			spin_unlock(&bdev->fence_lock);
+			driver->sync_obj_unref(&tmp_obj);
+			spin_lock(&bdev->fence_lock);
 			continue;
 		}
 
 		if (no_wait)
 			return -EBUSY;
 
-		sync_obj = driver->sync_obj_ref(*bo_sync_obj);
+		sync_obj = driver->sync_obj_ref(bo->sync_obj);
 		sync_obj_arg = bo->sync_obj_arg;
 		spin_unlock(&bdev->fence_lock);
 		ret = driver->sync_obj_wait(sync_obj, sync_obj_arg,
@@ -1784,9 +1732,16 @@ int ttm_bo_wait(struct ttm_buffer_object *bo,
 			return ret;
 		}
 		spin_lock(&bdev->fence_lock);
-		if (likely(*bo_sync_obj == sync_obj &&
+		if (likely(bo->sync_obj == sync_obj &&
 			   bo->sync_obj_arg == sync_obj_arg)) {
-			ttm_bo_unref_sync_obj_locked(bo, *bo_sync_obj, &sync_obj);
+			void *tmp_obj = bo->sync_obj;
+			bo->sync_obj = NULL;
+			clear_bit(TTM_BO_PRIV_FLAG_MOVING,
+				  &bo->priv_flags);
+			spin_unlock(&bdev->fence_lock);
+			driver->sync_obj_unref(&sync_obj);
+			driver->sync_obj_unref(&tmp_obj);
+			spin_lock(&bdev->fence_lock);
 		} else {
 			spin_unlock(&bdev->fence_lock);
 			driver->sync_obj_unref(&sync_obj);
@@ -1810,7 +1765,7 @@ int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait)
 	if (unlikely(ret != 0))
 		return ret;
 	spin_lock(&bdev->fence_lock);
-	ret = ttm_bo_wait(bo, false, true, no_wait, TTM_USAGE_READWRITE);
+	ret = ttm_bo_wait(bo, false, true, no_wait);
 	spin_unlock(&bdev->fence_lock);
 	if (likely(ret == 0))
 		atomic_inc(&bo->cpu_writers);
@@ -1884,7 +1839,7 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
 	 */
 
 	spin_lock(&bo->bdev->fence_lock);
-	ret = ttm_bo_wait(bo, false, false, false, TTM_USAGE_READWRITE);
+	ret = ttm_bo_wait(bo, false, false, false);
 	spin_unlock(&bo->bdev->fence_lock);
 
 	if (unlikely(ret != 0))
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index 6135f58169c..ae3c6f5dd2b 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -436,8 +436,6 @@ static int ttm_buffer_object_transfer(struct ttm_buffer_object *bo,
 	atomic_set(&fbo->cpu_writers, 0);
 
 	fbo->sync_obj = driver->sync_obj_ref(bo->sync_obj);
-	fbo->sync_obj_read = driver->sync_obj_ref(bo->sync_obj_read);
-	fbo->sync_obj_write = driver->sync_obj_ref(bo->sync_obj_write);
 	kref_init(&fbo->list_kref);
 	kref_init(&fbo->kref);
 	fbo->destroy = &ttm_transfered_destroy;
@@ -620,30 +618,20 @@ int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
 	struct ttm_mem_reg *old_mem = &bo->mem;
 	int ret;
 	struct ttm_buffer_object *ghost_obj;
-	void *tmp_obj = NULL, *tmp_obj_read = NULL, *tmp_obj_write = NULL;
+	void *tmp_obj = NULL;
 
 	spin_lock(&bdev->fence_lock);
-	if (bo->sync_obj)
+	if (bo->sync_obj) {
 		tmp_obj = bo->sync_obj;
-	if (bo->sync_obj_read)
-		tmp_obj_read = bo->sync_obj_read;
-	if (bo->sync_obj_write)
-		tmp_obj_write = bo->sync_obj_write;
-
+		bo->sync_obj = NULL;
+	}
 	bo->sync_obj = driver->sync_obj_ref(sync_obj);
-	bo->sync_obj_read = driver->sync_obj_ref(sync_obj);
-	bo->sync_obj_write = driver->sync_obj_ref(sync_obj);
 	bo->sync_obj_arg = sync_obj_arg;
 	if (evict) {
-		ret = ttm_bo_wait(bo, false, false, false,
-				  TTM_USAGE_READWRITE);
+		ret = ttm_bo_wait(bo, false, false, false);
 		spin_unlock(&bdev->fence_lock);
 		if (tmp_obj)
 			driver->sync_obj_unref(&tmp_obj);
-		if (tmp_obj_read)
-			driver->sync_obj_unref(&tmp_obj_read);
-		if (tmp_obj_write)
-			driver->sync_obj_unref(&tmp_obj_write);
 		if (ret)
 			return ret;
 
@@ -667,10 +655,6 @@ int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
 		spin_unlock(&bdev->fence_lock);
 		if (tmp_obj)
 			driver->sync_obj_unref(&tmp_obj);
-		if (tmp_obj_read)
-			driver->sync_obj_unref(&tmp_obj_read);
-		if (tmp_obj_write)
-			driver->sync_obj_unref(&tmp_obj_write);
 
 		ret = ttm_buffer_object_transfer(bo, &ghost_obj);
 		if (ret)
diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c
index ff1e26f4b09..221b924aceb 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
@@ -122,7 +122,7 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 
 	spin_lock(&bdev->fence_lock);
 	if (test_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags)) {
-		ret = ttm_bo_wait(bo, false, true, false, TTM_USAGE_READWRITE);
+		ret = ttm_bo_wait(bo, false, true, false);
 		spin_unlock(&bdev->fence_lock);
 		if (unlikely(ret != 0)) {
 			retval = (ret != -ERESTARTSYS) ?
diff --git a/drivers/gpu/drm/ttm/ttm_execbuf_util.c b/drivers/gpu/drm/ttm/ttm_execbuf_util.c
index 36d111a8823..3832fe10b4d 100644
--- a/drivers/gpu/drm/ttm/ttm_execbuf_util.c
+++ b/drivers/gpu/drm/ttm/ttm_execbuf_util.c
@@ -221,18 +221,8 @@ void ttm_eu_fence_buffer_objects(struct list_head *list, void *sync_obj)
 
 	list_for_each_entry(entry, list, head) {
 		bo = entry->bo;
-		entry->old_sync_obj_read = NULL;
-		entry->old_sync_obj_write = NULL;
 		entry->old_sync_obj = bo->sync_obj;
 		bo->sync_obj = driver->sync_obj_ref(sync_obj);
-		if (entry->usage & TTM_USAGE_READ) {
-			entry->old_sync_obj_read = bo->sync_obj_read;
-			bo->sync_obj_read = driver->sync_obj_ref(sync_obj);
-		}
-		if (entry->usage & TTM_USAGE_WRITE) {
-			entry->old_sync_obj_write = bo->sync_obj_write;
-			bo->sync_obj_write = driver->sync_obj_ref(sync_obj);
-		}
 		bo->sync_obj_arg = entry->new_sync_obj_arg;
 		ttm_bo_unreserve_locked(bo);
 		entry->reserved = false;
@@ -241,15 +231,8 @@ void ttm_eu_fence_buffer_objects(struct list_head *list, void *sync_obj)
 	spin_unlock(&bdev->fence_lock);
 
 	list_for_each_entry(entry, list, head) {
-		if (entry->old_sync_obj) {
+		if (entry->old_sync_obj)
 			driver->sync_obj_unref(&entry->old_sync_obj);
-		}
-		if (entry->old_sync_obj_read) {
-			driver->sync_obj_unref(&entry->old_sync_obj_read);
-		}
-		if (entry->old_sync_obj_write) {
-			driver->sync_obj_unref(&entry->old_sync_obj_write);
-		}
 	}
 }
 EXPORT_SYMBOL(ttm_eu_fence_buffer_objects);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index b8eb8cdcfb7..13afddc1f03 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -244,7 +244,7 @@ static void vmw_dummy_query_bo_prepare(struct vmw_private *dev_priv)
 
 	ttm_bo_reserve(bo, false, false, false, 0);
 	spin_lock(&bdev->fence_lock);
-	ret = ttm_bo_wait(bo, false, false, false, TTM_USAGE_READWRITE);
+	ret = ttm_bo_wait(bo, false, false, false);
 	spin_unlock(&bdev->fence_lock);
 	if (unlikely(ret != 0))
 		(void) vmw_fallback_wait(dev_priv, false, true, 0, false,
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 28e1c35aec6..40932fbdac0 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -89,7 +89,6 @@ static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
 	if (unlikely(val_node == sw_context->cur_val_buf)) {
 		val_buf->new_sync_obj_arg = NULL;
 		val_buf->bo = ttm_bo_reference(bo);
-		val_buf->usage = TTM_USAGE_READWRITE;
 		list_add_tail(&val_buf->head, &sw_context->validate_nodes);
 		++sw_context->cur_val_buf;
 	}