@@ -1454,26 +1454,19 @@ nouveau_ttm_tt_unpopulate(struct ttm_tt *ttm)
ttm_pool_unpopulate(ttm);
}
+static void
+nouveau_bo_fence_unref(void **sync_obj)
+{
+ nouveau_fence_unref((struct nouveau_fence **)sync_obj);
+}
+
void
nouveau_bo_fence(struct nouveau_bo *nvbo, struct nouveau_fence *fence)
{
- struct nouveau_fence *new_fence = nouveau_fence_ref(fence);
- struct nouveau_fence *old_fence = NULL;
-
lockdep_assert_held(&nvbo->bo.resv->lock.base);
- spin_lock(&nvbo->bo.bdev->fence_lock);
- old_fence = nvbo->bo.sync_obj;
- nvbo->bo.sync_obj = new_fence;
- spin_unlock(&nvbo->bo.bdev->fence_lock);
-
- nouveau_fence_unref(&old_fence);
-}
-
-static void
-nouveau_bo_fence_unref(void **sync_obj)
-{
- nouveau_fence_unref((struct nouveau_fence **)sync_obj);
+ nouveau_bo_fence_unref(&nvbo->bo.sync_obj);
+ nvbo->bo.sync_obj = nouveau_fence_ref(fence);
}
static void *
@@ -615,11 +615,7 @@ nouveau_crtc_page_flip(struct drm_crtc *crtc, struct drm_framebuffer *fb,
goto fail_unpin;
/* synchronise rendering channel with the kernel's channel */
- spin_lock(&new_bo->bo.bdev->fence_lock);
- fence = nouveau_fence_ref(new_bo->bo.sync_obj);
- spin_unlock(&new_bo->bo.bdev->fence_lock);
- ret = nouveau_fence_sync(fence, chan);
- nouveau_fence_unref(&fence);
+ ret = nouveau_fence_sync(new_bo->bo.sync_obj, chan);
if (ret) {
ttm_bo_unreserve(&new_bo->bo);
@@ -105,9 +105,7 @@ nouveau_gem_object_unmap(struct nouveau_bo *nvbo, struct nouveau_vma *vma)
list_del(&vma->head);
if (mapped) {
- spin_lock(&nvbo->bo.bdev->fence_lock);
fence = nouveau_fence_ref(nvbo->bo.sync_obj);
- spin_unlock(&nvbo->bo.bdev->fence_lock);
}
if (fence) {
@@ -434,17 +432,11 @@ retry:
static int
validate_sync(struct nouveau_channel *chan, struct nouveau_bo *nvbo)
{
- struct nouveau_fence *fence = NULL;
+ struct nouveau_fence *fence = nvbo->bo.sync_obj;
int ret = 0;
- spin_lock(&nvbo->bo.bdev->fence_lock);
- fence = nouveau_fence_ref(nvbo->bo.sync_obj);
- spin_unlock(&nvbo->bo.bdev->fence_lock);
-
- if (fence) {
+ if (fence)
ret = nouveau_fence_sync(fence, chan);
- nouveau_fence_unref(&fence);
- }
return ret;
}
@@ -669,9 +661,7 @@ nouveau_gem_pushbuf_reloc_apply(struct nouveau_cli *cli,
data |= r->vor;
}
- spin_lock(&nvbo->bo.bdev->fence_lock);
ret = ttm_bo_wait(&nvbo->bo, false, false, false);
- spin_unlock(&nvbo->bo.bdev->fence_lock);
if (ret) {
NV_ERROR(cli, "reloc wait_idle failed: %d\n", ret);
break;
@@ -904,11 +894,9 @@ nouveau_gem_ioctl_cpu_prep(struct drm_device *dev, void *data,
ret = ttm_bo_reserve(&nvbo->bo, true, false, false, 0);
if (!ret) {
- spin_lock(&nvbo->bo.bdev->fence_lock);
ret = ttm_bo_wait(&nvbo->bo, true, true, true);
if (!no_wait && ret)
fence = nouveau_fence_ref(nvbo->bo.sync_obj);
- spin_unlock(&nvbo->bo.bdev->fence_lock);
ttm_bo_unreserve(&nvbo->bo);
}
@@ -628,9 +628,7 @@ static int qxl_reap_surf(struct qxl_device *qdev, struct qxl_bo *surf, bool stal
if (stall)
mutex_unlock(&qdev->surf_evict_mutex);
- spin_lock(&surf->tbo.bdev->fence_lock);
ret = ttm_bo_wait(&surf->tbo, true, true, !stall);
- spin_unlock(&surf->tbo.bdev->fence_lock);
if (stall)
mutex_lock(&qdev->surf_evict_mutex);
@@ -60,9 +60,6 @@ int qxl_fence_remove_release(struct qxl_fence *qfence, uint32_t rel_id)
{
void *ret;
int retval = 0;
- struct qxl_bo *bo = container_of(qfence, struct qxl_bo, fence);
-
- spin_lock(&bo->tbo.bdev->fence_lock);
ret = radix_tree_delete(&qfence->tree, rel_id);
if (ret == qfence)
@@ -71,7 +68,6 @@ int qxl_fence_remove_release(struct qxl_fence *qfence, uint32_t rel_id)
DRM_DEBUG("didn't find fence in radix tree for %d\n", rel_id);
retval = -ENOENT;
}
- spin_unlock(&bo->tbo.bdev->fence_lock);
return retval;
}
@@ -76,12 +76,10 @@ static inline int qxl_bo_wait(struct qxl_bo *bo, u32 *mem_type,
}
return r;
}
- spin_lock(&bo->tbo.bdev->fence_lock);
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);
- spin_unlock(&bo->tbo.bdev->fence_lock);
ttm_bo_unreserve(&bo->tbo);
return r;
}
@@ -337,7 +337,6 @@ void qxl_release_fence_buffer_objects(struct qxl_release *release)
glob = bo->glob;
spin_lock(&glob->lru_lock);
- spin_lock(&bdev->fence_lock);
list_for_each_entry(entry, &release->bos, head) {
bo = entry->bo;
@@ -351,7 +350,6 @@ void qxl_release_fence_buffer_objects(struct qxl_release *release)
ttm_bo_add_to_lru(bo);
ww_mutex_unlock(&bo->resv->lock);
}
- spin_unlock(&bdev->fence_lock);
spin_unlock(&glob->lru_lock);
ww_acquire_fini(&release->ticket);
}
@@ -380,10 +380,8 @@ static int radeon_crtc_page_flip(struct drm_crtc *crtc,
obj = new_radeon_fb->obj;
rbo = gem_to_radeon_bo(obj);
- spin_lock(&rbo->tbo.bdev->fence_lock);
if (rbo->tbo.sync_obj)
work->fence = radeon_fence_ref(rbo->tbo.sync_obj);
- spin_unlock(&rbo->tbo.bdev->fence_lock);
INIT_WORK(&work->work, radeon_unpin_work_func);
@@ -612,12 +612,10 @@ int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type, bool no_wait)
r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, 0);
if (unlikely(r != 0))
return r;
- spin_lock(&bo->tbo.bdev->fence_lock);
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);
- spin_unlock(&bo->tbo.bdev->fence_lock);
ttm_bo_unreserve(&bo->tbo);
return r;
}
@@ -412,24 +412,20 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
spin_lock(&glob->lru_lock);
ret = ttm_bo_reserve_nolru(bo, false, true, false, 0);
- spin_lock(&bdev->fence_lock);
- (void) ttm_bo_wait(bo, false, false, true);
- if (!ret && !bo->sync_obj) {
- spin_unlock(&bdev->fence_lock);
- put_count = ttm_bo_del_from_lru(bo);
+ if (!ret) {
+ (void) ttm_bo_wait(bo, false, false, true);
- spin_unlock(&glob->lru_lock);
- ttm_bo_cleanup_memtype_use(bo);
+ if (!bo->sync_obj) {
+ put_count = ttm_bo_del_from_lru(bo);
- ttm_bo_list_ref_sub(bo, put_count, true);
+ spin_unlock(&glob->lru_lock);
+ ttm_bo_cleanup_memtype_use(bo);
- return;
- }
- if (bo->sync_obj)
- sync_obj = driver->sync_obj_ref(bo->sync_obj);
- spin_unlock(&bdev->fence_lock);
+ ttm_bo_list_ref_sub(bo, put_count, true);
- if (!ret) {
+ return;
+ }
+ sync_obj = driver->sync_obj_ref(bo->sync_obj);
/*
* Make NO_EVICT bos immediately available to
@@ -478,7 +474,6 @@ static int ttm_bo_cleanup_refs_and_unlock(struct ttm_buffer_object *bo,
int put_count;
int ret;
- spin_lock(&bdev->fence_lock);
ret = ttm_bo_wait(bo, false, false, true);
if (ret && !no_wait_gpu) {
@@ -490,7 +485,6 @@ static int ttm_bo_cleanup_refs_and_unlock(struct ttm_buffer_object *bo,
* no new sync objects can be attached.
*/
sync_obj = driver->sync_obj_ref(bo->sync_obj);
- spin_unlock(&bdev->fence_lock);
ww_mutex_unlock(&bo->resv->lock);
spin_unlock(&glob->lru_lock);
@@ -520,11 +514,9 @@ static int ttm_bo_cleanup_refs_and_unlock(struct ttm_buffer_object *bo,
* remove sync_obj with ttm_bo_wait, the wait should be
* finished, and no new wait object should have been added.
*/
- spin_lock(&bdev->fence_lock);
ret = ttm_bo_wait(bo, false, false, true);
WARN_ON(ret);
}
- spin_unlock(&bdev->fence_lock);
if (ret || unlikely(list_empty(&bo->ddestroy))) {
ww_mutex_unlock(&bo->resv->lock);
@@ -662,9 +654,7 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
struct ttm_placement placement;
int ret = 0;
- spin_lock(&bdev->fence_lock);
ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
- spin_unlock(&bdev->fence_lock);
if (unlikely(ret != 0)) {
if (ret != -ERESTARTSYS) {
@@ -961,7 +951,6 @@ static int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
{
int ret = 0;
struct ttm_mem_reg mem;
- struct ttm_bo_device *bdev = bo->bdev;
lockdep_assert_held(&bo->resv->lock.base);
@@ -970,9 +959,7 @@ static int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
* Have the driver move function wait for idle when necessary,
* instead of doing it here.
*/
- spin_lock(&bdev->fence_lock);
ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
- spin_unlock(&bdev->fence_lock);
if (ret)
return ret;
mem.num_pages = bo->num_pages;
@@ -1471,7 +1458,6 @@ int ttm_bo_device_init(struct ttm_bo_device *bdev,
bdev->glob = glob;
bdev->need_dma32 = need_dma32;
bdev->val_seq = 0;
- spin_lock_init(&bdev->fence_lock);
mutex_lock(&glob->device_list_mutex);
list_add_tail(&bdev->device_list, &glob->device_list);
mutex_unlock(&glob->device_list_mutex);
@@ -1529,7 +1515,6 @@ int ttm_bo_wait(struct ttm_buffer_object *bo,
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;
int ret = 0;
@@ -1538,53 +1523,33 @@ int ttm_bo_wait(struct ttm_buffer_object *bo,
if (likely(bo->sync_obj == NULL))
return 0;
- while (bo->sync_obj) {
-
+ if (bo->sync_obj) {
if (driver->sync_obj_signaled(bo->sync_obj)) {
- void *tmp_obj = bo->sync_obj;
- bo->sync_obj = NULL;
+ driver->sync_obj_unref(&bo->sync_obj);
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;
+ return 0;
}
if (no_wait)
return -EBUSY;
sync_obj = driver->sync_obj_ref(bo->sync_obj);
- spin_unlock(&bdev->fence_lock);
ret = driver->sync_obj_wait(sync_obj,
lazy, interruptible);
- if (unlikely(ret != 0)) {
- driver->sync_obj_unref(&sync_obj);
- spin_lock(&bdev->fence_lock);
- return ret;
- }
- spin_lock(&bdev->fence_lock);
- if (likely(bo->sync_obj == sync_obj)) {
- void *tmp_obj = bo->sync_obj;
- bo->sync_obj = NULL;
+
+ if (likely(ret == 0)) {
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);
- spin_lock(&bdev->fence_lock);
+ driver->sync_obj_unref(&bo->sync_obj);
}
+ driver->sync_obj_unref(&sync_obj);
}
- return 0;
+ return ret;
}
EXPORT_SYMBOL(ttm_bo_wait);
int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait)
{
- struct ttm_bo_device *bdev = bo->bdev;
int ret = 0;
/*
@@ -1594,9 +1559,7 @@ int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait)
ret = ttm_bo_reserve(bo, true, no_wait, false, 0);
if (unlikely(ret != 0))
return ret;
- spin_lock(&bdev->fence_lock);
ret = ttm_bo_wait(bo, false, true, no_wait);
- spin_unlock(&bdev->fence_lock);
if (likely(ret == 0))
atomic_inc(&bo->cpu_writers);
ttm_bo_unreserve(bo);
@@ -1653,9 +1616,7 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
* Wait for GPU, then move to system cached.
*/
- spin_lock(&bo->bdev->fence_lock);
ret = ttm_bo_wait(bo, false, false, false);
- spin_unlock(&bo->bdev->fence_lock);
if (unlikely(ret != 0))
goto out;
@@ -466,12 +466,10 @@ static int ttm_buffer_object_transfer(struct ttm_buffer_object *bo,
drm_vma_node_reset(&fbo->vma_node);
atomic_set(&fbo->cpu_writers, 0);
- spin_lock(&bdev->fence_lock);
if (bo->sync_obj)
fbo->sync_obj = driver->sync_obj_ref(bo->sync_obj);
else
fbo->sync_obj = NULL;
- spin_unlock(&bdev->fence_lock);
kref_init(&fbo->list_kref);
kref_init(&fbo->kref);
fbo->destroy = &ttm_transfered_destroy;
@@ -657,7 +655,6 @@ int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
struct ttm_buffer_object *ghost_obj;
void *tmp_obj = NULL;
- spin_lock(&bdev->fence_lock);
if (bo->sync_obj) {
tmp_obj = bo->sync_obj;
bo->sync_obj = NULL;
@@ -665,7 +662,6 @@ int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
bo->sync_obj = driver->sync_obj_ref(sync_obj);
if (evict) {
ret = ttm_bo_wait(bo, false, false, false);
- spin_unlock(&bdev->fence_lock);
if (tmp_obj)
driver->sync_obj_unref(&tmp_obj);
if (ret)
@@ -688,7 +684,6 @@ int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
*/
set_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags);
- spin_unlock(&bdev->fence_lock);
if (tmp_obj)
driver->sync_obj_unref(&tmp_obj);
@@ -45,10 +45,8 @@ static int ttm_bo_vm_fault_idle(struct ttm_buffer_object *bo,
struct vm_area_struct *vma,
struct vm_fault *vmf)
{
- struct ttm_bo_device *bdev = bo->bdev;
int ret = 0;
- spin_lock(&bdev->fence_lock);
if (likely(!test_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags)))
goto out_unlock;
@@ -82,7 +80,6 @@ static int ttm_bo_vm_fault_idle(struct ttm_buffer_object *bo,
VM_FAULT_NOPAGE;
out_unlock:
- spin_unlock(&bdev->fence_lock);
return ret;
}
@@ -181,7 +181,6 @@ void ttm_eu_fence_buffer_objects(struct ww_acquire_ctx *ticket,
glob = bo->glob;
spin_lock(&glob->lru_lock);
- spin_lock(&bdev->fence_lock);
list_for_each_entry(entry, list, head) {
bo = entry->bo;
@@ -190,7 +189,6 @@ void ttm_eu_fence_buffer_objects(struct ww_acquire_ctx *ticket,
ttm_bo_add_to_lru(bo);
ww_mutex_unlock(&bo->resv->lock);
}
- spin_unlock(&bdev->fence_lock);
spin_unlock(&glob->lru_lock);
if (ticket)
ww_acquire_fini(ticket);
@@ -829,11 +829,7 @@ static void vmw_move_notify(struct ttm_buffer_object *bo,
*/
static void vmw_swap_notify(struct ttm_buffer_object *bo)
{
- struct ttm_bo_device *bdev = bo->bdev;
-
- spin_lock(&bdev->fence_lock);
ttm_bo_wait(bo, false, false, false);
- spin_unlock(&bdev->fence_lock);
}
@@ -554,12 +554,13 @@ static int vmw_user_dmabuf_synccpu_grab(struct vmw_user_dma_buffer *user_bo,
int ret;
if (flags & drm_vmw_synccpu_allow_cs) {
- struct ttm_bo_device *bdev = bo->bdev;
+ int try = !!(flags & drm_vmw_synccpu_dontblock);
- spin_lock(&bdev->fence_lock);
- ret = ttm_bo_wait(bo, false, true,
- !!(flags & drm_vmw_synccpu_dontblock));
- spin_unlock(&bdev->fence_lock);
+ ret = ttm_bo_reserve(bo, false, try, false, NULL);
+ if (!ret) {
+ ret = ttm_bo_wait(bo, false, true, try);
+ ttm_bo_unreserve(bo);
+ }
return ret;
}
@@ -1419,12 +1420,10 @@ void vmw_fence_single_bo(struct ttm_buffer_object *bo,
else
driver->sync_obj_ref(fence);
- spin_lock(&bdev->fence_lock);
old_fence_obj = bo->sync_obj;
bo->sync_obj = fence;
- spin_unlock(&bdev->fence_lock);
if (old_fence_obj)
vmw_fence_obj_unreference(&old_fence_obj);
@@ -1465,7 +1464,6 @@ void vmw_resource_move_notify(struct ttm_buffer_object *bo,
if (mem->mem_type != VMW_PL_MOB) {
struct vmw_resource *res, *n;
- struct ttm_bo_device *bdev = bo->bdev;
struct ttm_validate_buffer val_buf;
val_buf.bo = bo;
@@ -1481,9 +1479,7 @@ void vmw_resource_move_notify(struct ttm_buffer_object *bo,
list_del_init(&res->mob_head);
}
- spin_lock(&bdev->fence_lock);
(void) ttm_bo_wait(bo, false, false, false);
- spin_unlock(&bdev->fence_lock);
}
}
@@ -227,10 +227,7 @@ struct ttm_buffer_object {
struct list_head io_reserve_lru;
/**
- * Members protected by struct buffer_object_device::fence_lock
- * In addition, setting sync_obj to anything else
- * than NULL requires bo::reserved to be held. This allows for
- * checking NULL while reserved but not holding the mentioned lock.
+ * Members protected by a bo reservation.
*/
void *sync_obj;
@@ -518,8 +518,6 @@ struct ttm_bo_global {
*
* @driver: Pointer to a struct ttm_bo_driver struct setup by the driver.
* @man: An array of mem_type_managers.
- * @fence_lock: Protects the synchronizing members on *all* bos belonging
- * to this device.
* @vma_manager: Address space manager
* lru_lock: Spinlock that protects the buffer+device lru lists and
* ddestroy lists.
@@ -539,7 +537,6 @@ struct ttm_bo_device {
struct ttm_bo_global *glob;
struct ttm_bo_driver *driver;
struct ttm_mem_type_manager man[TTM_NUM_MEM_TYPES];
- spinlock_t fence_lock;
/*
* Protected by internal locks.
No users are left, kill it off! :D Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com> --- Fixed to apply on top of drm-next. Also, for nouveau "[PATCH 1/2] drm/nouveau: hold mutex while syncing to kernel channel" is required, else this patch will fail to apply. drivers/gpu/drm/nouveau/nouveau_bo.c | 23 ++++------ drivers/gpu/drm/nouveau/nouveau_display.c | 6 +-- drivers/gpu/drm/nouveau/nouveau_gem.c | 16 +------ drivers/gpu/drm/qxl/qxl_cmd.c | 2 - drivers/gpu/drm/qxl/qxl_fence.c | 4 -- drivers/gpu/drm/qxl/qxl_object.h | 2 - drivers/gpu/drm/qxl/qxl_release.c | 2 - drivers/gpu/drm/radeon/radeon_display.c | 2 - drivers/gpu/drm/radeon/radeon_object.c | 2 - drivers/gpu/drm/ttm/ttm_bo.c | 75 ++++++++----------------------- drivers/gpu/drm/ttm/ttm_bo_util.c | 5 --- drivers/gpu/drm/ttm/ttm_bo_vm.c | 3 -- drivers/gpu/drm/ttm/ttm_execbuf_util.c | 2 - drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c | 4 -- drivers/gpu/drm/vmwgfx/vmwgfx_resource.c | 16 +++---- include/drm/ttm/ttm_bo_api.h | 5 +-- include/drm/ttm/ttm_bo_driver.h | 3 -- 17 files changed, 36 insertions(+), 136 deletions(-)