--- drivers/gpu/drm/nouveau/nouveau_bo.c | 48 +------- drivers/gpu/drm/nouveau/nouveau_fence.c | 24 +--- drivers/gpu/drm/nouveau/nouveau_fence.h | 2 drivers/gpu/drm/nouveau/nouveau_gem.c | 16 ++- drivers/gpu/drm/qxl/qxl_debugfs.c | 6 + drivers/gpu/drm/qxl/qxl_drv.h | 2 drivers/gpu/drm/qxl/qxl_kms.c | 1 drivers/gpu/drm/qxl/qxl_object.h | 4 - drivers/gpu/drm/qxl/qxl_release.c | 3 - drivers/gpu/drm/qxl/qxl_ttm.c | 104 ------------------ drivers/gpu/drm/radeon/radeon_cs.c | 10 +- drivers/gpu/drm/radeon/radeon_display.c | 18 +++ drivers/gpu/drm/radeon/radeon_object.c | 4 - drivers/gpu/drm/radeon/radeon_ttm.c | 34 ------ drivers/gpu/drm/radeon/radeon_uvd.c | 8 + drivers/gpu/drm/radeon/radeon_vm.c | 2 drivers/gpu/drm/ttm/ttm_bo.c | 171 +++++++++++++++++++++--------- drivers/gpu/drm/ttm/ttm_bo_util.c | 23 +--- drivers/gpu/drm/ttm/ttm_execbuf_util.c | 10 -- drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c | 40 ------- drivers/gpu/drm/vmwgfx/vmwgfx_resource.c | 14 +- include/drm/ttm/ttm_bo_api.h | 2 include/drm/ttm/ttm_bo_driver.h | 26 ----- include/drm/ttm/ttm_execbuf_util.h | 10 +- 24 files changed, 197 insertions(+), 385 deletions(-)
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c index 84aba3fa1bd0..5b8ccc39a282 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -92,13 +92,13 @@ nv10_bo_get_tile_region(struct drm_device *dev, int i)
static void nv10_bo_put_tile_region(struct drm_device *dev, struct nouveau_drm_tile *tile, - struct nouveau_fence *fence) + struct fence *fence) { struct nouveau_drm *drm = nouveau_drm(dev);
if (tile) { spin_lock(&drm->tile.lock); - tile->fence = nouveau_fence_ref(fence); + tile->fence = nouveau_fence_ref((struct nouveau_fence *)fence); tile->used = false; spin_unlock(&drm->tile.lock); } @@ -965,7 +965,8 @@ nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr, if (ret == 0) { ret = nouveau_fence_new(chan, false, &fence); if (ret == 0) { - ret = ttm_bo_move_accel_cleanup(bo, fence, + ret = ttm_bo_move_accel_cleanup(bo, + &fence->base, evict, no_wait_gpu, new_mem); @@ -1151,8 +1152,9 @@ nouveau_bo_vm_cleanup(struct ttm_buffer_object *bo, { struct nouveau_drm *drm = nouveau_bdev(bo->bdev); struct drm_device *dev = drm->dev; + struct fence *fence = reservation_object_get_excl(bo->resv);
- nv10_bo_put_tile_region(dev, *old_tile, bo->sync_obj); + nv10_bo_put_tile_region(dev, *old_tile, fence); *old_tile = new_tile; }
@@ -1423,47 +1425,14 @@ 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 reservation_object *resv = nvbo->bo.resv;
- nouveau_bo_fence_unref(&nvbo->bo.sync_obj); - nvbo->bo.sync_obj = nouveau_fence_ref(fence); - reservation_object_add_excl_fence(resv, &fence->base); }
-static void * -nouveau_bo_fence_ref(void *sync_obj) -{ - return nouveau_fence_ref(sync_obj); -} - -static bool -nouveau_bo_fence_signalled(void *sync_obj) -{ - return nouveau_fence_done(sync_obj); -} - -static int -nouveau_bo_fence_wait(void *sync_obj, bool lazy, bool intr) -{ - return nouveau_fence_wait(sync_obj, lazy, intr); -} - -static int -nouveau_bo_fence_flush(void *sync_obj) -{ - return 0; -} - struct ttm_bo_driver nouveau_bo_driver = { .ttm_tt_create = &nouveau_ttm_tt_create, .ttm_tt_populate = &nouveau_ttm_tt_populate, @@ -1474,11 +1443,6 @@ struct ttm_bo_driver nouveau_bo_driver = { .move_notify = nouveau_bo_move_ntfy, .move = nouveau_bo_move, .verify_access = nouveau_bo_verify_access, - .sync_obj_signaled = nouveau_bo_fence_signalled, - .sync_obj_wait = nouveau_bo_fence_wait, - .sync_obj_flush = nouveau_bo_fence_flush, - .sync_obj_unref = nouveau_bo_fence_unref, - .sync_obj_ref = nouveau_bo_fence_ref, .fault_reserve_notify = &nouveau_ttm_fault_reserve_notify, .io_mem_reserve = &nouveau_ttm_io_mem_reserve, .io_mem_free = &nouveau_ttm_io_mem_free, diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.c b/drivers/gpu/drm/nouveau/nouveau_fence.c index 9a9e04985826..b1aba6f79605 100644 --- a/drivers/gpu/drm/nouveau/nouveau_fence.c +++ b/drivers/gpu/drm/nouveau/nouveau_fence.c @@ -139,17 +139,18 @@ static bool nouveau_fence_is_signaled(struct fence *f) }
void -nouveau_fence_work(struct nouveau_fence *fence, +nouveau_fence_work(struct fence *fence, void (*func)(void *), void *data) { struct nouveau_fence_work *work;
- if (fence_is_signaled(&fence->base)) + if (fence_is_signaled(fence)) goto err;
work = kmalloc(sizeof(*work), GFP_KERNEL); if (!work) { - WARN_ON(nouveau_fence_wait(fence, false, false)); + WARN_ON(nouveau_fence_wait((struct nouveau_fence *)fence, + false, false)); goto err; }
@@ -157,7 +158,7 @@ nouveau_fence_work(struct nouveau_fence *fence, work->func = func; work->data = data;
- if (fence_add_callback(&fence->base, &work->cb, nouveau_fence_work_cb) < 0) + if (fence_add_callback(fence, &work->cb, nouveau_fence_work_cb) < 0) goto err_free; return;
@@ -322,14 +323,9 @@ nouveau_fence_sync(struct nouveau_bo *nvbo, struct nouveau_channel *chan) struct reservation_object_list *fobj; int ret = 0, i;
- fence = nvbo->bo.sync_obj; - if (fence && fence_is_signaled(fence)) { - nouveau_fence_unref((struct nouveau_fence **) - &nvbo->bo.sync_obj); - fence = NULL; - } + fence = reservation_object_get_excl(resv);
- if (fence) { + if (fence && !fence_is_signaled(fence)) { struct nouveau_fence *f = container_of(fence, struct nouveau_fence, base); @@ -345,12 +341,8 @@ nouveau_fence_sync(struct nouveau_bo *nvbo, struct nouveau_channel *chan) if (ret) return ret;
- fence = reservation_object_get_excl(resv); - if (fence && !nouveau_local_fence(fence, chan->drm)) - ret = fence_wait(fence, true); - fobj = reservation_object_get_list(resv); - if (!fobj || ret) + if (!fobj) return ret;
for (i = 0; i < fobj->shared_count && !ret; ++i) { diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.h b/drivers/gpu/drm/nouveau/nouveau_fence.h index 1989ec22e66e..41abc8a44e3c 100644 --- a/drivers/gpu/drm/nouveau/nouveau_fence.h +++ b/drivers/gpu/drm/nouveau/nouveau_fence.h @@ -26,7 +26,7 @@ void nouveau_fence_unref(struct nouveau_fence **);
int nouveau_fence_emit(struct nouveau_fence *, struct nouveau_channel *); bool nouveau_fence_done(struct nouveau_fence *); -void nouveau_fence_work(struct nouveau_fence *, void (*)(void *), void *); +void nouveau_fence_work(struct fence *, void (*)(void *), void *); int nouveau_fence_wait(struct nouveau_fence *, bool lazy, bool intr); int nouveau_fence_sync(struct nouveau_bo *, struct nouveau_channel *);
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c index a61530becfb9..4beaa897adad 100644 --- a/drivers/gpu/drm/nouveau/nouveau_gem.c +++ b/drivers/gpu/drm/nouveau/nouveau_gem.c @@ -100,13 +100,12 @@ static void nouveau_gem_object_unmap(struct nouveau_bo *nvbo, struct nouveau_vma *vma) { const bool mapped = nvbo->bo.mem.mem_type != TTM_PL_SYSTEM; - struct nouveau_fence *fence = NULL; + struct fence *fence = NULL;
list_del(&vma->head);
- if (mapped) { - fence = nouveau_fence_ref(nvbo->bo.sync_obj); - } + if (mapped) + fence = reservation_object_get_excl(nvbo->bo.resv);
if (fence) { nouveau_fence_work(fence, nouveau_gem_object_delete, vma); @@ -116,7 +115,6 @@ nouveau_gem_object_unmap(struct nouveau_bo *nvbo, struct nouveau_vma *vma) nouveau_vm_put(vma); kfree(vma); } - nouveau_fence_unref(&fence); }
void @@ -876,8 +874,12 @@ nouveau_gem_ioctl_cpu_prep(struct drm_device *dev, void *data, ret = ttm_bo_reserve(&nvbo->bo, true, false, false, 0); if (!ret) { ret = ttm_bo_wait(&nvbo->bo, true, true, true); - if (!no_wait && ret) - fence = nouveau_fence_ref(nvbo->bo.sync_obj); + if (!no_wait && ret) { + struct fence *excl; + + excl = reservation_object_get_excl(nvbo->bo.resv); + fence = nouveau_fence_ref((struct nouveau_fence *)excl); + }
ttm_bo_unreserve(&nvbo->bo); } diff --git a/drivers/gpu/drm/qxl/qxl_debugfs.c b/drivers/gpu/drm/qxl/qxl_debugfs.c index 0d144e0646d6..a4a63fd84803 100644 --- a/drivers/gpu/drm/qxl/qxl_debugfs.c +++ b/drivers/gpu/drm/qxl/qxl_debugfs.c @@ -67,9 +67,9 @@ qxl_debugfs_buffers_info(struct seq_file *m, void *data) rel = fobj ? fobj->shared_count : 0; rcu_read_unlock();
- seq_printf(m, "size %ld, pc %d, sync obj %p, num releases %d\n", - (unsigned long)bo->gem_base.size, bo->pin_count, - bo->tbo.sync_obj, rel); + seq_printf(m, "size %ld, pc %d, num releases %d\n", + (unsigned long)bo->gem_base.size, + bo->pin_count, rel); } spin_unlock(&qdev->release_lock); return 0; diff --git a/drivers/gpu/drm/qxl/qxl_drv.h b/drivers/gpu/drm/qxl/qxl_drv.h index d547cbdebeb4..74e2117ee0e6 100644 --- a/drivers/gpu/drm/qxl/qxl_drv.h +++ b/drivers/gpu/drm/qxl/qxl_drv.h @@ -280,9 +280,7 @@ struct qxl_device { uint8_t slot_gen_bits; uint64_t va_slot_mask;
- /* XXX: when rcu becomes available, release_lock can be killed */ spinlock_t release_lock; - spinlock_t fence_lock; struct idr release_idr; uint32_t release_seqno; spinlock_t release_idr_lock; diff --git a/drivers/gpu/drm/qxl/qxl_kms.c b/drivers/gpu/drm/qxl/qxl_kms.c index a9e7c30e92c5..7234561e09d9 100644 --- a/drivers/gpu/drm/qxl/qxl_kms.c +++ b/drivers/gpu/drm/qxl/qxl_kms.c @@ -224,7 +224,6 @@ static int qxl_device_init(struct qxl_device *qdev, idr_init(&qdev->release_idr); spin_lock_init(&qdev->release_idr_lock); spin_lock_init(&qdev->release_lock); - spin_lock_init(&qdev->fence_lock);
idr_init(&qdev->surf_id_idr); spin_lock_init(&qdev->surf_id_idr_lock); diff --git a/drivers/gpu/drm/qxl/qxl_object.h b/drivers/gpu/drm/qxl/qxl_object.h index 98395b223ad0..9da7becbdb34 100644 --- a/drivers/gpu/drm/qxl/qxl_object.h +++ b/drivers/gpu/drm/qxl/qxl_object.h @@ -78,8 +78,8 @@ static inline int qxl_bo_wait(struct qxl_bo *bo, u32 *mem_type, } 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); + + r = ttm_bo_wait(&bo->tbo, true, true, no_wait); ttm_bo_unreserve(&bo->tbo); return r; } diff --git a/drivers/gpu/drm/qxl/qxl_release.c b/drivers/gpu/drm/qxl/qxl_release.c index 3b1398d735f4..cfd4b8036269 100644 --- a/drivers/gpu/drm/qxl/qxl_release.c +++ b/drivers/gpu/drm/qxl/qxl_release.c @@ -464,9 +464,6 @@ void qxl_release_fence_buffer_objects(struct qxl_release *release) bo = entry->bo; qbo = to_qxl_bo(bo);
- if (!entry->bo->sync_obj) - entry->bo->sync_obj = qbo; - reservation_object_add_shared_fence(bo->resv, &release->base); ttm_bo_add_to_lru(bo); __ttm_bo_unreserve(bo); diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c b/drivers/gpu/drm/qxl/qxl_ttm.c index 80879e38e447..439e2b23465d 100644 --- a/drivers/gpu/drm/qxl/qxl_ttm.c +++ b/drivers/gpu/drm/qxl/qxl_ttm.c @@ -361,105 +361,6 @@ static int qxl_bo_move(struct ttm_buffer_object *bo, return ttm_bo_move_memcpy(bo, evict, no_wait_gpu, new_mem); }
-static bool qxl_sync_obj_signaled(void *sync_obj); - -static int qxl_sync_obj_wait(void *sync_obj, - bool lazy, bool interruptible) -{ - struct qxl_bo *bo = (struct qxl_bo *)sync_obj; - struct qxl_device *qdev = bo->gem_base.dev->dev_private; - struct reservation_object_list *fobj; - int count = 0, sc = 0, num_release = 0; - bool have_drawable_releases; - -retry: - if (sc == 0) { - if (bo->type == QXL_GEM_DOMAIN_SURFACE) - qxl_update_surface(qdev, bo); - } else if (sc >= 1) { - qxl_io_notify_oom(qdev); - } - - sc++; - - for (count = 0; count < 10; count++) { - if (qxl_sync_obj_signaled(sync_obj)) - return 0; - - if (!qxl_queue_garbage_collect(qdev, true)) - break; - } - - have_drawable_releases = false; - num_release = 0; - - spin_lock(&qdev->release_lock); - fobj = bo->tbo.resv->fence; - for (count = 0; fobj && count < fobj->shared_count; count++) { - struct qxl_release *release; - - release = container_of(fobj->shared[count], - struct qxl_release, base); - - if (fence_is_signaled(&release->base)) - continue; - - num_release++; - - if (release->type == QXL_RELEASE_DRAWABLE) - have_drawable_releases = true; - } - spin_unlock(&qdev->release_lock); - - qxl_queue_garbage_collect(qdev, true); - - if (have_drawable_releases || sc < 4) { - if (sc > 2) - /* back off */ - usleep_range(500, 1000); - if (have_drawable_releases && sc > 300) { - WARN(1, "sync obj %d still has outstanding releases %d %d %d %ld %d\n", sc, bo->surface_id, bo->is_primary, bo->pin_count, (unsigned long)bo->gem_base.size, num_release); - return -EBUSY; - } - goto retry; - } - return 0; -} - -static int qxl_sync_obj_flush(void *sync_obj) -{ - return 0; -} - -static void qxl_sync_obj_unref(void **sync_obj) -{ - *sync_obj = NULL; -} - -static void *qxl_sync_obj_ref(void *sync_obj) -{ - return sync_obj; -} - -static bool qxl_sync_obj_signaled(void *sync_obj) -{ - struct qxl_bo *qbo = (struct qxl_bo *)sync_obj; - struct qxl_device *qdev = qbo->gem_base.dev->dev_private; - struct reservation_object_list *fobj; - bool ret = true; - unsigned i; - - spin_lock(&qdev->release_lock); - fobj = qbo->tbo.resv->fence; - for (i = 0; fobj && i < fobj->shared_count; ++i) { - ret = fence_is_signaled(fobj->shared[i]); - if (!ret) - break; - } - spin_unlock(&qdev->release_lock); - return ret; -} - static void qxl_bo_move_notify(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem) { @@ -486,11 +387,6 @@ static struct ttm_bo_driver qxl_bo_driver = { .verify_access = &qxl_verify_access, .io_mem_reserve = &qxl_ttm_io_mem_reserve, .io_mem_free = &qxl_ttm_io_mem_free, - .sync_obj_signaled = &qxl_sync_obj_signaled, - .sync_obj_wait = &qxl_sync_obj_wait, - .sync_obj_flush = &qxl_sync_obj_flush, - .sync_obj_unref = &qxl_sync_obj_unref, - .sync_obj_ref = &qxl_sync_obj_ref, .move_notify = &qxl_bo_move_notify, };
diff --git a/drivers/gpu/drm/radeon/radeon_cs.c b/drivers/gpu/drm/radeon/radeon_cs.c index 2b6e0ebcc13a..af1dae9ce829 100644 --- a/drivers/gpu/drm/radeon/radeon_cs.c +++ b/drivers/gpu/drm/radeon/radeon_cs.c @@ -222,11 +222,17 @@ static void radeon_cs_sync_rings(struct radeon_cs_parser *p) int i;
for (i = 0; i < p->nrelocs; i++) { + struct reservation_object *resv; + struct fence *fence; + if (!p->relocs[i].robj) continue;
+ resv = p->relocs[i].robj->tbo.resv; + fence = reservation_object_get_excl(resv); + radeon_semaphore_sync_to(p->ib.semaphore, - p->relocs[i].robj->tbo.sync_obj); + (struct radeon_fence *)fence); } }
@@ -389,7 +395,7 @@ static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error, bo
ttm_eu_fence_buffer_objects(&parser->ticket, &parser->validated, - parser->ib.fence); + &parser->ib.fence->base); } else if (backoff) { ttm_eu_backoff_reservation(&parser->ticket, &parser->validated); diff --git a/drivers/gpu/drm/radeon/radeon_display.c b/drivers/gpu/drm/radeon/radeon_display.c index 6a7340289ddd..f977b552e1ca 100644 --- a/drivers/gpu/drm/radeon/radeon_display.c +++ b/drivers/gpu/drm/radeon/radeon_display.c @@ -363,6 +363,7 @@ static int radeon_crtc_page_flip(struct drm_crtc *crtc, struct drm_gem_object *obj; struct radeon_bo *rbo; struct radeon_unpin_work *work; + struct fence *fence; unsigned long flags; u32 tiling_flags, pitch_pixels; u64 base; @@ -386,9 +387,6 @@ static int radeon_crtc_page_flip(struct drm_crtc *crtc, obj = new_radeon_fb->obj; rbo = gem_to_radeon_bo(obj);
- if (rbo->tbo.sync_obj) - work->fence = radeon_fence_ref(rbo->tbo.sync_obj); - INIT_WORK(&work->work, radeon_unpin_work_func);
/* We borrow the event spin lock for protecting unpin_work */ @@ -411,6 +409,20 @@ static int radeon_crtc_page_flip(struct drm_crtc *crtc, DRM_ERROR("failed to reserve new rbo buffer before flip\n"); goto pflip_cleanup; } + + fence = reservation_object_get_excl(rbo->tbo.resv); + if (fence) { + fence_get(fence); + work->fence = (struct radeon_fence *)fence; + + if (!fence->ops->signaled) + /* + * make sure if this fence doesn't belong to this + * device that it will still signal completion + */ + fence_enable_sw_signaling(fence); + } + /* Only 27 bit offset for legacy CRTC */ r = radeon_bo_pin_restricted(rbo, RADEON_GEM_DOMAIN_VRAM, ASIC_IS_AVIVO(rdev) ? 0 : 1 << 27, &base); diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c index 3b27cac9240b..e082feae53d9 100644 --- a/drivers/gpu/drm/radeon/radeon_object.c +++ b/drivers/gpu/drm/radeon/radeon_object.c @@ -727,8 +727,8 @@ int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type, bool no_wait) return r; 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); + + r = ttm_bo_wait(&bo->tbo, true, true, no_wait); ttm_bo_unreserve(&bo->tbo); return r; } diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c index c8a8a5144ec1..715e29f984c1 100644 --- a/drivers/gpu/drm/radeon/radeon_ttm.c +++ b/drivers/gpu/drm/radeon/radeon_ttm.c @@ -265,12 +265,12 @@ static int radeon_move_blit(struct ttm_buffer_object *bo, BUILD_BUG_ON((PAGE_SIZE % RADEON_GPU_PAGE_SIZE) != 0);
/* sync other rings */ - fence = bo->sync_obj; + fence = (struct radeon_fence *)reservation_object_get_excl(bo->resv); r = radeon_copy(rdev, old_start, new_start, new_mem->num_pages * (PAGE_SIZE / RADEON_GPU_PAGE_SIZE), /* GPU pages */ &fence); /* FIXME: handle copy error */ - r = ttm_bo_move_accel_cleanup(bo, (void *)fence, + r = ttm_bo_move_accel_cleanup(bo, &fence->base, evict, no_wait_gpu, new_mem); radeon_fence_unref(&fence); return r; @@ -483,31 +483,6 @@ static void radeon_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_re { }
-static int radeon_sync_obj_wait(void *sync_obj, bool lazy, bool interruptible) -{ - return radeon_fence_wait((struct radeon_fence *)sync_obj, interruptible); -} - -static int radeon_sync_obj_flush(void *sync_obj) -{ - return 0; -} - -static void radeon_sync_obj_unref(void **sync_obj) -{ - radeon_fence_unref((struct radeon_fence **)sync_obj); -} - -static void *radeon_sync_obj_ref(void *sync_obj) -{ - return radeon_fence_ref((struct radeon_fence *)sync_obj); -} - -static bool radeon_sync_obj_signaled(void *sync_obj) -{ - return radeon_fence_signaled((struct radeon_fence *)sync_obj); -} - /* * TTM backend functions. */ @@ -685,11 +660,6 @@ static struct ttm_bo_driver radeon_bo_driver = { .evict_flags = &radeon_evict_flags, .move = &radeon_bo_move, .verify_access = &radeon_verify_access, - .sync_obj_signaled = &radeon_sync_obj_signaled, - .sync_obj_wait = &radeon_sync_obj_wait, - .sync_obj_flush = &radeon_sync_obj_flush, - .sync_obj_unref = &radeon_sync_obj_unref, - .sync_obj_ref = &radeon_sync_obj_ref, .move_notify = &radeon_bo_move_notify, .fault_reserve_notify = &radeon_bo_fault_reserve_notify, .io_mem_reserve = &radeon_ttm_io_mem_reserve, diff --git a/drivers/gpu/drm/radeon/radeon_uvd.c b/drivers/gpu/drm/radeon/radeon_uvd.c index 2f93fef15aab..08eb88067ffa 100644 --- a/drivers/gpu/drm/radeon/radeon_uvd.c +++ b/drivers/gpu/drm/radeon/radeon_uvd.c @@ -356,6 +356,7 @@ static int radeon_uvd_cs_msg(struct radeon_cs_parser *p, struct radeon_bo *bo, { int32_t *msg, msg_type, handle; unsigned img_size = 0; + struct fence *f; void *ptr;
int i, r; @@ -365,8 +366,9 @@ static int radeon_uvd_cs_msg(struct radeon_cs_parser *p, struct radeon_bo *bo, return -EINVAL; }
- if (bo->tbo.sync_obj) { - r = radeon_fence_wait(bo->tbo.sync_obj, false); + f = reservation_object_get_excl(bo->tbo.resv); + if (f) { + r = radeon_fence_wait((struct radeon_fence *)f, false); if (r) { DRM_ERROR("Failed waiting for UVD message (%d)!\n", r); return r; @@ -649,7 +651,7 @@ static int radeon_uvd_send_msg(struct radeon_device *rdev, r = radeon_ib_schedule(rdev, &ib, NULL); if (r) goto err; - ttm_eu_fence_buffer_objects(&ticket, &head, ib.fence); + ttm_eu_fence_buffer_objects(&ticket, &head, &ib.fence->base);
if (fence) *fence = radeon_fence_ref(ib.fence); diff --git a/drivers/gpu/drm/radeon/radeon_vm.c b/drivers/gpu/drm/radeon/radeon_vm.c index f4fd72477a71..8bf660e5bcc3 100644 --- a/drivers/gpu/drm/radeon/radeon_vm.c +++ b/drivers/gpu/drm/radeon/radeon_vm.c @@ -388,7 +388,7 @@ static int radeon_vm_clear_bo(struct radeon_device *rdev, if (r) goto error;
- ttm_eu_fence_buffer_objects(&ticket, &head, ib.fence); + ttm_eu_fence_buffer_objects(&ticket, &head, &ib.fence->base); radeon_ib_free(rdev, &ib);
return 0; diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index ce0434377223..31c4a6dd722d 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -40,6 +40,7 @@ #include <linux/file.h> #include <linux/module.h> #include <linux/atomic.h> +#include <linux/reservation.h>
#define TTM_ASSERT_LOCKED(param) #define TTM_DEBUG(fmt, arg...) @@ -141,7 +142,6 @@ static void ttm_bo_release_list(struct kref *list_kref) BUG_ON(atomic_read(&bo->list_kref.refcount)); BUG_ON(atomic_read(&bo->kref.refcount)); BUG_ON(atomic_read(&bo->cpu_writers)); - BUG_ON(bo->sync_obj != NULL); BUG_ON(bo->mem.mm_node != NULL); BUG_ON(!list_empty(&bo->lru)); BUG_ON(!list_empty(&bo->ddestroy)); @@ -402,12 +402,30 @@ static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo) ww_mutex_unlock (&bo->resv->lock); }
+static void ttm_bo_flush_all_fences(struct ttm_buffer_object *bo) +{ + struct reservation_object_list *fobj; + struct fence *fence; + int i; + + fobj = reservation_object_get_list(bo->resv); + fence = reservation_object_get_excl(bo->resv); + if (fence && !fence->ops->signaled) + fence_enable_sw_signaling(fence); + + for (i = 0; fobj && i < fobj->shared_count; ++i) { + fence = rcu_dereference_protected(fobj->shared[i], + reservation_object_held(bo->resv)); + + if (!fence->ops->signaled) + fence_enable_sw_signaling(fence); + } +} + static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo) { struct ttm_bo_device *bdev = bo->bdev; struct ttm_bo_global *glob = bo->glob; - struct ttm_bo_driver *driver = bdev->driver; - void *sync_obj = NULL; int put_count; int ret;
@@ -415,9 +433,7 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo) ret = __ttm_bo_reserve(bo, false, true, false, 0);
if (!ret) { - (void) ttm_bo_wait(bo, false, false, true); - - if (!bo->sync_obj) { + if (!ttm_bo_wait(bo, false, false, true)) { put_count = ttm_bo_del_from_lru(bo);
spin_unlock(&glob->lru_lock); @@ -426,8 +442,8 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo) ttm_bo_list_ref_sub(bo, put_count, true);
return; - } - sync_obj = driver->sync_obj_ref(bo->sync_obj); + } else + ttm_bo_flush_all_fences(bo);
/* * Make NO_EVICT bos immediately available to @@ -446,14 +462,70 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo) list_add_tail(&bo->ddestroy, &bdev->ddestroy); spin_unlock(&glob->lru_lock);
- if (sync_obj) { - driver->sync_obj_flush(sync_obj); - driver->sync_obj_unref(&sync_obj); - } schedule_delayed_work(&bdev->wq, ((HZ / 100) < 1) ? 1 : HZ / 100); }
+static int ttm_bo_unreserve_and_wait(struct ttm_buffer_object *bo, + bool interruptible) +{ + struct ttm_bo_global *glob = bo->glob; + struct reservation_object_list *fobj; + struct fence *excl = NULL; + struct fence **shared = NULL; + u32 shared_count = 0, i; + int ret = 0; + + fobj = reservation_object_get_list(bo->resv); + if (fobj && fobj->shared_count) { + shared = kmalloc(sizeof(*shared) * fobj->shared_count, + GFP_KERNEL); + + if (!shared) { + ret = -ENOMEM; + __ttm_bo_unreserve(bo); + spin_unlock(&glob->lru_lock); + return ret; + } + + for (i = 0; i < fobj->shared_count; ++i) { + if (!fence_is_signaled(fobj->shared[i])) { + fence_get(fobj->shared[i]); + shared[shared_count++] = fobj->shared[i]; + } + } + if (!shared_count) { + kfree(shared); + shared = NULL; + } + } + + excl = reservation_object_get_excl(bo->resv); + if (excl && !fence_is_signaled(excl)) + fence_get(excl); + else + excl = NULL; + + __ttm_bo_unreserve(bo); + spin_unlock(&glob->lru_lock); + + if (excl) { + ret = fence_wait(excl, interruptible); + fence_put(excl); + } + + if (shared_count > 0) { + for (i = 0; i < shared_count; ++i) { + if (!ret) + ret = fence_wait(shared[i], interruptible); + fence_put(shared[i]); + } + kfree(shared); + } + + return ret; +} + /** * function ttm_bo_cleanup_refs_and_unlock * If bo idle, remove from delayed- and lru lists, and unref. @@ -470,8 +542,6 @@ static int ttm_bo_cleanup_refs_and_unlock(struct ttm_buffer_object *bo, bool interruptible, bool no_wait_gpu) { - struct ttm_bo_device *bdev = bo->bdev; - struct ttm_bo_driver *driver = bdev->driver; struct ttm_bo_global *glob = bo->glob; int put_count; int ret; @@ -479,20 +549,7 @@ static int ttm_bo_cleanup_refs_and_unlock(struct ttm_buffer_object *bo, ret = ttm_bo_wait(bo, false, false, true);
if (ret && !no_wait_gpu) { - void *sync_obj; - - /* - * Take a reference to the fence and unreserve, - * at this point the buffer should be dead, so - * no new sync objects can be attached. - */ - sync_obj = driver->sync_obj_ref(bo->sync_obj); - - __ttm_bo_unreserve(bo); - spin_unlock(&glob->lru_lock); - - ret = driver->sync_obj_wait(sync_obj, false, interruptible); - driver->sync_obj_unref(&sync_obj); + ret = ttm_bo_unreserve_and_wait(bo, interruptible); if (ret) return ret;
@@ -1513,41 +1570,51 @@ void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo)
EXPORT_SYMBOL(ttm_bo_unmap_virtual);
- int ttm_bo_wait(struct ttm_buffer_object *bo, bool lazy, bool interruptible, bool no_wait) { - struct ttm_bo_driver *driver = bo->bdev->driver; - void *sync_obj; - int ret = 0; - - lockdep_assert_held(&bo->resv->lock.base); + struct reservation_object_list *fobj; + struct reservation_object *resv; + struct fence *excl; + long timeout = 15 * HZ; + int i;
- if (likely(bo->sync_obj == NULL)) - return 0; + resv = bo->resv; + fobj = reservation_object_get_list(resv); + excl = reservation_object_get_excl(resv); + if (excl) { + if (!fence_is_signaled(excl)) { + if (no_wait) + return -EBUSY;
- if (bo->sync_obj) { - if (driver->sync_obj_signaled(bo->sync_obj)) { - driver->sync_obj_unref(&bo->sync_obj); - clear_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags); - return 0; + timeout = fence_wait_timeout(excl, + interruptible, timeout); } + }
- if (no_wait) - return -EBUSY; + for (i = 0; fobj && timeout > 0 && i < fobj->shared_count; ++i) { + struct fence *fence; + fence = rcu_dereference_protected(fobj->shared[i], + reservation_object_held(resv));
- sync_obj = driver->sync_obj_ref(bo->sync_obj); - ret = driver->sync_obj_wait(sync_obj, - lazy, interruptible); + if (!fence_is_signaled(fence)) { + if (no_wait) + return -EBUSY;
- if (likely(ret == 0)) { - clear_bit(TTM_BO_PRIV_FLAG_MOVING, - &bo->priv_flags); - driver->sync_obj_unref(&bo->sync_obj); + timeout = fence_wait_timeout(fence, + interruptible, timeout); } - driver->sync_obj_unref(&sync_obj); } - return ret; + + if (timeout < 0) + return timeout; + + if (timeout == 0) + return -EBUSY; + + reservation_object_add_excl_fence(resv, NULL); + clear_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags); + return 0; } EXPORT_SYMBOL(ttm_bo_wait);
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c index 23db594e55c0..fe806c1ded9e 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_util.c +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c @@ -37,6 +37,7 @@ #include <linux/slab.h> #include <linux/vmalloc.h> #include <linux/module.h> +#include <linux/reservation.h>
void ttm_bo_free_old_node(struct ttm_buffer_object *bo) { @@ -444,8 +445,6 @@ static int ttm_buffer_object_transfer(struct ttm_buffer_object *bo, struct ttm_buffer_object **new_obj) { struct ttm_buffer_object *fbo; - struct ttm_bo_device *bdev = bo->bdev; - struct ttm_bo_driver *driver = bdev->driver; int ret;
fbo = kmalloc(sizeof(*fbo), GFP_KERNEL); @@ -466,10 +465,6 @@ static int ttm_buffer_object_transfer(struct ttm_buffer_object *bo, drm_vma_node_reset(&fbo->vma_node); atomic_set(&fbo->cpu_writers, 0);
- if (bo->sync_obj) - fbo->sync_obj = driver->sync_obj_ref(bo->sync_obj); - else - fbo->sync_obj = NULL; kref_init(&fbo->list_kref); kref_init(&fbo->kref); fbo->destroy = &ttm_transfered_destroy; @@ -642,28 +637,20 @@ void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map) EXPORT_SYMBOL(ttm_bo_kunmap);
int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo, - void *sync_obj, + struct fence *fence, bool evict, bool no_wait_gpu, struct ttm_mem_reg *new_mem) { struct ttm_bo_device *bdev = bo->bdev; - struct ttm_bo_driver *driver = bdev->driver; struct ttm_mem_type_manager *man = &bdev->man[new_mem->mem_type]; struct ttm_mem_reg *old_mem = &bo->mem; int ret; struct ttm_buffer_object *ghost_obj; - void *tmp_obj = NULL;
- if (bo->sync_obj) { - tmp_obj = bo->sync_obj; - bo->sync_obj = NULL; - } - bo->sync_obj = driver->sync_obj_ref(sync_obj); + reservation_object_add_excl_fence(bo->resv, fence); if (evict) { ret = ttm_bo_wait(bo, false, false, false); - if (tmp_obj) - driver->sync_obj_unref(&tmp_obj); if (ret) return ret;
@@ -684,13 +671,13 @@ int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo, */
set_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags); - if (tmp_obj) - driver->sync_obj_unref(&tmp_obj);
ret = ttm_buffer_object_transfer(bo, &ghost_obj); if (ret) return ret;
+ reservation_object_add_excl_fence(ghost_obj->resv, fence); + /** * If we're not moving to fixed memory, the TTM object * needs to stay alive. Otherwhise hang it on the ghost diff --git a/drivers/gpu/drm/ttm/ttm_execbuf_util.c b/drivers/gpu/drm/ttm/ttm_execbuf_util.c index 108730e9147b..adafc0f8ec06 100644 --- a/drivers/gpu/drm/ttm/ttm_execbuf_util.c +++ b/drivers/gpu/drm/ttm/ttm_execbuf_util.c @@ -163,7 +163,7 @@ int ttm_eu_reserve_buffers(struct ww_acquire_ctx *ticket, EXPORT_SYMBOL(ttm_eu_reserve_buffers);
void ttm_eu_fence_buffer_objects(struct ww_acquire_ctx *ticket, - struct list_head *list, void *sync_obj) + struct list_head *list, struct fence *fence) { struct ttm_validate_buffer *entry; struct ttm_buffer_object *bo; @@ -183,18 +183,12 @@ void ttm_eu_fence_buffer_objects(struct ww_acquire_ctx *ticket,
list_for_each_entry(entry, list, head) { bo = entry->bo; - entry->old_sync_obj = bo->sync_obj; - bo->sync_obj = driver->sync_obj_ref(sync_obj); + reservation_object_add_excl_fence(bo->resv, fence); ttm_bo_add_to_lru(bo); __ttm_bo_unreserve(bo); } spin_unlock(&glob->lru_lock); if (ticket) ww_acquire_fini(ticket); - - list_for_each_entry(entry, list, head) { - if (entry->old_sync_obj) - driver->sync_obj_unref(&entry->old_sync_obj); - } } EXPORT_SYMBOL(ttm_eu_fence_buffer_objects); diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c index f15718cc631d..656c88485e14 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c @@ -768,41 +768,6 @@ static int vmw_ttm_fault_reserve_notify(struct ttm_buffer_object *bo) }
/** - * FIXME: We're using the old vmware polling method to sync. - * Do this with fences instead. - */ - -static void *vmw_sync_obj_ref(void *sync_obj) -{ - - return (void *) - vmw_fence_obj_reference((struct vmw_fence_obj *) sync_obj); -} - -static void vmw_sync_obj_unref(void **sync_obj) -{ - vmw_fence_obj_unreference((struct vmw_fence_obj **) sync_obj); -} - -static int vmw_sync_obj_flush(void *sync_obj) -{ - vmw_fence_obj_flush((struct vmw_fence_obj *) sync_obj); - return 0; -} - -static bool vmw_sync_obj_signaled(void *sync_obj) -{ - return vmw_fence_obj_signaled((struct vmw_fence_obj *) sync_obj); -} - -static int vmw_sync_obj_wait(void *sync_obj, bool lazy, bool interruptible) -{ - return vmw_fence_obj_wait((struct vmw_fence_obj *) sync_obj, - lazy, interruptible, - VMW_FENCE_WAIT_TIMEOUT); -} - -/** * vmw_move_notify - TTM move_notify_callback * * @bo: The TTM buffer object about to move. @@ -839,11 +804,6 @@ struct ttm_bo_driver vmw_bo_driver = { .evict_flags = vmw_evict_flags, .move = NULL, .verify_access = vmw_verify_access, - .sync_obj_signaled = vmw_sync_obj_signaled, - .sync_obj_wait = vmw_sync_obj_wait, - .sync_obj_flush = vmw_sync_obj_flush, - .sync_obj_unref = vmw_sync_obj_unref, - .sync_obj_ref = vmw_sync_obj_ref, .move_notify = vmw_move_notify, .swap_notify = vmw_swap_notify, .fault_reserve_notify = &vmw_ttm_fault_reserve_notify, diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c index 6688a6341486..20a1a866ceeb 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c @@ -1419,22 +1419,16 @@ void vmw_fence_single_bo(struct ttm_buffer_object *bo, struct vmw_fence_obj *fence) { struct ttm_bo_device *bdev = bo->bdev; - struct vmw_fence_obj *old_fence_obj; + struct vmw_private *dev_priv = container_of(bdev, struct vmw_private, bdev);
if (fence == NULL) { vmw_execbuf_fence_commands(NULL, dev_priv, &fence, NULL); + reservation_object_add_excl_fence(bo->resv, &fence->base); + fence_put(&fence->base); } else - vmw_fence_obj_reference(fence); - - reservation_object_add_excl_fence(bo->resv, &fence->base); - - old_fence_obj = bo->sync_obj; - bo->sync_obj = fence; - - if (old_fence_obj) - vmw_fence_obj_unreference(&old_fence_obj); + reservation_object_add_excl_fence(bo->resv, &fence->base); }
/** diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h index f34d59b67218..cdc75814a861 100644 --- a/include/drm/ttm/ttm_bo_api.h +++ b/include/drm/ttm/ttm_bo_api.h @@ -163,7 +163,6 @@ struct ttm_tt; * @lru: List head for the lru list. * @ddestroy: List head for the delayed destroy list. * @swap: List head for swap LRU list. - * @sync_obj: Pointer to a synchronization object. * @priv_flags: Flags describing buffer object internal state. * @vma_node: Address space manager node. * @offset: The current GPU offset, which can have different meanings @@ -230,7 +229,6 @@ struct ttm_buffer_object { * Members protected by a bo reservation. */
- void *sync_obj; unsigned long priv_flags;
struct drm_vma_offset_node vma_node; diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h index 0aa6caa59415..71a345ee92d5 100644 --- a/include/drm/ttm/ttm_bo_driver.h +++ b/include/drm/ttm/ttm_bo_driver.h @@ -309,11 +309,6 @@ struct ttm_mem_type_manager { * @move: Callback for a driver to hook in accelerated functions to * move a buffer. * If set to NULL, a potentially slow memcpy() move is used. - * @sync_obj_signaled: See ttm_fence_api.h - * @sync_obj_wait: See ttm_fence_api.h - * @sync_obj_flush: See ttm_fence_api.h - * @sync_obj_unref: See ttm_fence_api.h - * @sync_obj_ref: See ttm_fence_api.h */
struct ttm_bo_driver { @@ -415,23 +410,6 @@ struct ttm_bo_driver { int (*verify_access) (struct ttm_buffer_object *bo, struct file *filp);
- /** - * In case a driver writer dislikes the TTM fence objects, - * the driver writer can replace those with sync objects of - * his / her own. If it turns out that no driver writer is - * using these. I suggest we remove these hooks and plug in - * fences directly. The bo driver needs the following functionality: - * See the corresponding functions in the fence object API - * documentation. - */ - - bool (*sync_obj_signaled) (void *sync_obj); - int (*sync_obj_wait) (void *sync_obj, - bool lazy, bool interruptible); - int (*sync_obj_flush) (void *sync_obj); - void (*sync_obj_unref) (void **sync_obj); - void *(*sync_obj_ref) (void *sync_obj); - /* hook to notify driver about a driver move so it * can do tiling things */ void (*move_notify)(struct ttm_buffer_object *bo, @@ -1031,7 +1009,7 @@ extern void ttm_bo_free_old_node(struct ttm_buffer_object *bo); * ttm_bo_move_accel_cleanup. * * @bo: A pointer to a struct ttm_buffer_object. - * @sync_obj: A sync object that signals when moving is complete. + * @fence: A fence object that signals when moving is complete. * @evict: This is an evict move. Don't return until the buffer is idle. * @no_wait_gpu: Return immediately if the GPU is busy. * @new_mem: struct ttm_mem_reg indicating where to move. @@ -1045,7 +1023,7 @@ extern void ttm_bo_free_old_node(struct ttm_buffer_object *bo); */
extern int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo, - void *sync_obj, + struct fence *fence, bool evict, bool no_wait_gpu, struct ttm_mem_reg *new_mem); /** diff --git a/include/drm/ttm/ttm_execbuf_util.h b/include/drm/ttm/ttm_execbuf_util.h index 8490cb8ee0d8..ff11a424f752 100644 --- a/include/drm/ttm/ttm_execbuf_util.h +++ b/include/drm/ttm/ttm_execbuf_util.h @@ -39,16 +39,11 @@ * * @head: list head for thread-private list. * @bo: refcounted buffer object pointer. - * @reserved: Indicates whether @bo has been reserved for validation. - * @removed: Indicates whether @bo has been removed from lru lists. - * @put_count: Number of outstanding references on bo::list_kref. - * @old_sync_obj: Pointer to a sync object about to be unreferenced */
struct ttm_validate_buffer { struct list_head head; struct ttm_buffer_object *bo; - void *old_sync_obj; };
/** @@ -100,7 +95,7 @@ extern int ttm_eu_reserve_buffers(struct ww_acquire_ctx *ticket, * * @ticket: ww_acquire_ctx from reserve call * @list: thread private list of ttm_validate_buffer structs. - * @sync_obj: The new sync object for the buffers. + * @fence: The new exclusive fence for the buffers. * * This function should be called when command submission is complete, and * it will add a new sync object to bos pointed to by entries on @list. @@ -109,6 +104,7 @@ extern int ttm_eu_reserve_buffers(struct ww_acquire_ctx *ticket, */
extern void ttm_eu_fence_buffer_objects(struct ww_acquire_ctx *ticket, - struct list_head *list, void *sync_obj); + struct list_head *list, + struct fence *fence);
#endif