diff options
Diffstat (limited to 'drivers/gpu/drm/ttm')
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_bo.c | 280 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_bo_manager.c | 9 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_bo_util.c | 28 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_bo_vm.c | 3 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_execbuf_util.c | 146 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_object.c | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_page_alloc_dma.c | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_tt.c | 13 |
8 files changed, 191 insertions, 292 deletions
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index 3da89d5dab6..a11969acfea 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...) @@ -53,12 +54,13 @@ static struct attribute ttm_bo_count = { .mode = S_IRUGO }; -static inline int ttm_mem_type_from_flags(uint32_t flags, uint32_t *mem_type) +static inline int ttm_mem_type_from_place(const struct ttm_place *place, + uint32_t *mem_type) { int i; for (i = 0; i <= TTM_PL_PRIV5; i++) - if (flags & (1 << i)) { + if (place->flags & (1 << i)) { *mem_type = i; return 0; } @@ -89,12 +91,12 @@ static void ttm_bo_mem_space_debug(struct ttm_buffer_object *bo, bo, bo->mem.num_pages, bo->mem.size >> 10, bo->mem.size >> 20); for (i = 0; i < placement->num_placement; i++) { - ret = ttm_mem_type_from_flags(placement->placement[i], + ret = ttm_mem_type_from_place(&placement->placement[i], &mem_type); if (ret) return; pr_err(" placement[%d]=0x%08X (%d)\n", - i, placement->placement[i], mem_type); + i, placement->placement[i].flags, mem_type); ttm_mem_type_debug(bo->bdev, mem_type); } } @@ -141,7 +143,6 @@ 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)); @@ -402,36 +403,48 @@ 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; spin_lock(&glob->lru_lock); ret = __ttm_bo_reserve(bo, false, true, false, NULL); - 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); - - spin_unlock(&glob->lru_lock); - ttm_bo_cleanup_memtype_use(bo); + if (!ret) { + if (!ttm_bo_wait(bo, false, false, true)) { + 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; + } else + ttm_bo_flush_all_fences(bo); /* * Make NO_EVICT bos immediately available to @@ -450,10 +463,6 @@ 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); } @@ -474,44 +483,26 @@ 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; - spin_lock(&bdev->fence_lock); 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); - spin_unlock(&bdev->fence_lock); - - __ttm_bo_unreserve(bo); + long lret; + ww_mutex_unlock(&bo->resv->lock); spin_unlock(&glob->lru_lock); - ret = driver->sync_obj_wait(sync_obj, false, interruptible); - driver->sync_obj_unref(&sync_obj); - if (ret) - return ret; + lret = reservation_object_wait_timeout_rcu(bo->resv, + true, + interruptible, + 30 * HZ); - /* - * 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) - return ret; + if (lret < 0) + return lret; + else if (lret == 0) + return -EBUSY; spin_lock(&glob->lru_lock); ret = __ttm_bo_reserve(bo, false, true, false, NULL); @@ -528,8 +519,14 @@ static int ttm_bo_cleanup_refs_and_unlock(struct ttm_buffer_object *bo, spin_unlock(&glob->lru_lock); return 0; } - } else - spin_unlock(&bdev->fence_lock); + + /* + * remove sync_obj with ttm_bo_wait, the wait should be + * finished, and no new wait object should have been added. + */ + ret = ttm_bo_wait(bo, false, false, true); + WARN_ON(ret); + } if (ret || unlikely(list_empty(&bo->ddestroy))) { __ttm_bo_unreserve(bo); @@ -667,9 +664,7 @@ 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) { @@ -685,8 +680,6 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible, evict_mem.bus.io_reserved_vm = false; evict_mem.bus.io_reserved_count = 0; - placement.fpfn = 0; - placement.lpfn = 0; placement.num_placement = 0; placement.num_busy_placement = 0; bdev->driver->evict_flags(bo, &placement); @@ -774,7 +767,7 @@ EXPORT_SYMBOL(ttm_bo_mem_put); */ static int ttm_bo_mem_force_space(struct ttm_buffer_object *bo, uint32_t mem_type, - struct ttm_placement *placement, + const struct ttm_place *place, struct ttm_mem_reg *mem, bool interruptible, bool no_wait_gpu) @@ -784,7 +777,7 @@ static int ttm_bo_mem_force_space(struct ttm_buffer_object *bo, int ret; do { - ret = (*man->func->get_node)(man, bo, placement, 0, mem); + ret = (*man->func->get_node)(man, bo, place, mem); if (unlikely(ret != 0)) return ret; if (mem->mm_node) @@ -827,18 +820,18 @@ static uint32_t ttm_bo_select_caching(struct ttm_mem_type_manager *man, static bool ttm_bo_mt_compatible(struct ttm_mem_type_manager *man, uint32_t mem_type, - uint32_t proposed_placement, + const struct ttm_place *place, uint32_t *masked_placement) { uint32_t cur_flags = ttm_bo_type_flags(mem_type); - if ((cur_flags & proposed_placement & TTM_PL_MASK_MEM) == 0) + if ((cur_flags & place->flags & TTM_PL_MASK_MEM) == 0) return false; - if ((proposed_placement & man->available_caching) == 0) + if ((place->flags & man->available_caching) == 0) return false; - cur_flags |= (proposed_placement & man->available_caching); + cur_flags |= (place->flags & man->available_caching); *masked_placement = cur_flags; return true; @@ -869,15 +862,14 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo, mem->mm_node = NULL; for (i = 0; i < placement->num_placement; ++i) { - ret = ttm_mem_type_from_flags(placement->placement[i], - &mem_type); + const struct ttm_place *place = &placement->placement[i]; + + ret = ttm_mem_type_from_place(place, &mem_type); if (ret) return ret; man = &bdev->man[mem_type]; - type_ok = ttm_bo_mt_compatible(man, - mem_type, - placement->placement[i], + type_ok = ttm_bo_mt_compatible(man, mem_type, place, &cur_flags); if (!type_ok) @@ -889,7 +881,7 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo, * Use the access and other non-mapping-related flag bits from * the memory placement flags to the current flags */ - ttm_flag_masked(&cur_flags, placement->placement[i], + ttm_flag_masked(&cur_flags, place->flags, ~TTM_PL_MASK_MEMTYPE); if (mem_type == TTM_PL_SYSTEM) @@ -897,8 +889,7 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo, if (man->has_type && man->use_type) { type_found = true; - ret = (*man->func->get_node)(man, bo, placement, - cur_flags, mem); + ret = (*man->func->get_node)(man, bo, place, mem); if (unlikely(ret)) return ret; } @@ -916,17 +907,15 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo, return -EINVAL; for (i = 0; i < placement->num_busy_placement; ++i) { - ret = ttm_mem_type_from_flags(placement->busy_placement[i], - &mem_type); + const struct ttm_place *place = &placement->busy_placement[i]; + + ret = ttm_mem_type_from_place(place, &mem_type); if (ret) return ret; man = &bdev->man[mem_type]; if (!man->has_type) continue; - if (!ttm_bo_mt_compatible(man, - mem_type, - placement->busy_placement[i], - &cur_flags)) + if (!ttm_bo_mt_compatible(man, mem_type, place, &cur_flags)) continue; cur_flags = ttm_bo_select_caching(man, bo->mem.placement, @@ -935,7 +924,7 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo, * Use the access and other non-mapping-related flag bits from * the memory placement flags to the current flags */ - ttm_flag_masked(&cur_flags, placement->busy_placement[i], + ttm_flag_masked(&cur_flags, place->flags, ~TTM_PL_MASK_MEMTYPE); if (mem_type == TTM_PL_SYSTEM) { @@ -945,7 +934,7 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo, return 0; } - ret = ttm_bo_mem_force_space(bo, mem_type, placement, mem, + ret = ttm_bo_mem_force_space(bo, mem_type, place, mem, interruptible, no_wait_gpu); if (ret == 0 && mem->mm_node) { mem->placement = cur_flags; @@ -966,7 +955,6 @@ 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); @@ -975,9 +963,7 @@ 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; @@ -1006,20 +992,27 @@ static bool ttm_bo_mem_compat(struct ttm_placement *placement, { int i; - if (mem->mm_node && placement->lpfn != 0 && - (mem->start < placement->fpfn || - mem->start + mem->num_pages > placement->lpfn)) - return false; - for (i = 0; i < placement->num_placement; i++) { - *new_flags = placement->placement[i]; + const struct ttm_place *heap = &placement->placement[i]; + if (mem->mm_node && heap->lpfn != 0 && + (mem->start < heap->fpfn || + mem->start + mem->num_pages > heap->lpfn)) + continue; + + *new_flags = heap->flags; if ((*new_flags & mem->placement & TTM_PL_MASK_CACHING) && (*new_flags & mem->placement & TTM_PL_MASK_MEM)) return true; } for (i = 0; i < placement->num_busy_placement; i++) { - *new_flags = placement->busy_placement[i]; + const struct ttm_place *heap = &placement->busy_placement[i]; + if (mem->mm_node && heap->lpfn != 0 && + (mem->start < heap->fpfn || + mem->start + mem->num_pages > heap->lpfn)) + continue; + + *new_flags = heap->flags; if ((*new_flags & mem->placement & TTM_PL_MASK_CACHING) && (*new_flags & mem->placement & TTM_PL_MASK_MEM)) return true; @@ -1037,11 +1030,6 @@ int ttm_bo_validate(struct ttm_buffer_object *bo, uint32_t new_flags; lockdep_assert_held(&bo->resv->lock.base); - /* Check that range is valid */ - if (placement->lpfn || placement->fpfn) - if (placement->fpfn > placement->lpfn || - (placement->lpfn - placement->fpfn) < bo->num_pages) - return -EINVAL; /* * Check whether we need to move buffer. */ @@ -1070,15 +1058,6 @@ int ttm_bo_validate(struct ttm_buffer_object *bo, } EXPORT_SYMBOL(ttm_bo_validate); -int ttm_bo_check_placement(struct ttm_buffer_object *bo, - struct ttm_placement *placement) -{ - BUG_ON((placement->fpfn || placement->lpfn) && - (bo->mem.num_pages > (placement->lpfn - placement->fpfn))); - - return 0; -} - int ttm_bo_init(struct ttm_bo_device *bdev, struct ttm_buffer_object *bo, unsigned long size, @@ -1147,15 +1126,12 @@ int ttm_bo_init(struct ttm_bo_device *bdev, atomic_inc(&bo->glob->bo_count); drm_vma_node_reset(&bo->vma_node); - ret = ttm_bo_check_placement(bo, placement); - /* * For ttm_bo_type_device buffers, allocate * address space from the device. */ - if (likely(!ret) && - (bo->type == ttm_bo_type_device || - bo->type == ttm_bo_type_sg)) + if (bo->type == ttm_bo_type_device || + bo->type == ttm_bo_type_sg) ret = drm_vma_offset_add(&bdev->vma_manager, &bo->vma_node, bo->mem.num_pages); @@ -1477,7 +1453,6 @@ 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); @@ -1530,65 +1505,56 @@ 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; - struct ttm_bo_device *bdev = bo->bdev; - void *sync_obj; - int ret = 0; - - if (likely(bo->sync_obj == NULL)) - return 0; + struct reservation_object_list *fobj; + struct reservation_object *resv; + struct fence *excl; + long timeout = 15 * HZ; + int i; - while (bo->sync_obj) { + 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 (driver->sync_obj_signaled(bo->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(&tmp_obj); - spin_lock(&bdev->fence_lock); - continue; + 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); - 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; - 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); + if (!fence_is_signaled(fence)) { + if (no_wait) + return -EBUSY; + + timeout = fence_wait_timeout(fence, + interruptible, timeout); } } + + 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); int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait) { - struct ttm_bo_device *bdev = bo->bdev; int ret = 0; /* @@ -1598,9 +1564,7 @@ int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait) ret = ttm_bo_reserve(bo, true, no_wait, false, NULL); 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); @@ -1657,9 +1621,7 @@ 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; diff --git a/drivers/gpu/drm/ttm/ttm_bo_manager.c b/drivers/gpu/drm/ttm/ttm_bo_manager.c index 9e103a4875c..964387fc5c8 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_manager.c +++ b/drivers/gpu/drm/ttm/ttm_bo_manager.c @@ -49,8 +49,7 @@ struct ttm_range_manager { static int ttm_bo_man_get_node(struct ttm_mem_type_manager *man, struct ttm_buffer_object *bo, - struct ttm_placement *placement, - uint32_t flags, + const struct ttm_place *place, struct ttm_mem_reg *mem) { struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv; @@ -60,7 +59,7 @@ static int ttm_bo_man_get_node(struct ttm_mem_type_manager *man, unsigned long lpfn; int ret; - lpfn = placement->lpfn; + lpfn = place->lpfn; if (!lpfn) lpfn = man->size; @@ -68,13 +67,13 @@ static int ttm_bo_man_get_node(struct ttm_mem_type_manager *man, if (!node) return -ENOMEM; - if (flags & TTM_PL_FLAG_TOPDOWN) + if (place->flags & TTM_PL_FLAG_TOPDOWN) aflags = DRM_MM_CREATE_TOP; spin_lock(&rman->lock); ret = drm_mm_insert_node_in_range_generic(mm, node, mem->num_pages, mem->page_alignment, 0, - placement->fpfn, lpfn, + place->fpfn, lpfn, DRM_MM_SEARCH_BEST, aflags); spin_unlock(&rman->lock); diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c index 30e5d90cb7b..824af90cbe3 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); @@ -466,12 +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); - 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; @@ -644,30 +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; - spin_lock(&bdev->fence_lock); - 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); - spin_unlock(&bdev->fence_lock); - if (tmp_obj) - driver->sync_obj_unref(&tmp_obj); if (ret) return ret; @@ -688,14 +671,13 @@ 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); 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_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c index 0ce48e5a9cb..d05437f219e 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_vm.c +++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c @@ -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; } diff --git a/drivers/gpu/drm/ttm/ttm_execbuf_util.c b/drivers/gpu/drm/ttm/ttm_execbuf_util.c index e8dac875852..adafc0f8ec0 100644 --- a/drivers/gpu/drm/ttm/ttm_execbuf_util.c +++ b/drivers/gpu/drm/ttm/ttm_execbuf_util.c @@ -32,20 +32,12 @@ #include <linux/sched.h> #include <linux/module.h> -static void ttm_eu_backoff_reservation_locked(struct list_head *list) +static void ttm_eu_backoff_reservation_reverse(struct list_head *list, + struct ttm_validate_buffer *entry) { - struct ttm_validate_buffer *entry; - - list_for_each_entry(entry, list, head) { + list_for_each_entry_continue_reverse(entry, list, head) { struct ttm_buffer_object *bo = entry->bo; - if (!entry->reserved) - continue; - entry->reserved = false; - if (entry->removed) { - ttm_bo_add_to_lru(bo); - entry->removed = false; - } __ttm_bo_unreserve(bo); } } @@ -56,27 +48,9 @@ static void ttm_eu_del_from_lru_locked(struct list_head *list) list_for_each_entry(entry, list, head) { struct ttm_buffer_object *bo = entry->bo; - if (!entry->reserved) - continue; - - if (!entry->removed) { - entry->put_count = ttm_bo_del_from_lru(bo); - entry->removed = true; - } - } -} - -static void ttm_eu_list_ref_sub(struct list_head *list) -{ - struct ttm_validate_buffer *entry; - - list_for_each_entry(entry, list, head) { - struct ttm_buffer_object *bo = entry->bo; + unsigned put_count = ttm_bo_del_from_lru(bo); - if (entry->put_count) { - ttm_bo_list_ref_sub(bo, entry->put_count, true); - entry->put_count = 0; - } + ttm_bo_list_ref_sub(bo, put_count, true); } } @@ -91,11 +65,18 @@ void ttm_eu_backoff_reservation(struct ww_acquire_ctx *ticket, entry = list_first_entry(list, struct ttm_validate_buffer, head); glob = entry->bo->glob; + spin_lock(&glob->lru_lock); - ttm_eu_backoff_reservation_locked(list); + list_for_each_entry(entry, list, head) { + struct ttm_buffer_object *bo = entry->bo; + + ttm_bo_add_to_lru(bo); + __ttm_bo_unreserve(bo); + } + spin_unlock(&glob->lru_lock); + if (ticket) ww_acquire_fini(ticket); - spin_unlock(&glob->lru_lock); } EXPORT_SYMBOL(ttm_eu_backoff_reservation); @@ -112,7 +93,7 @@ EXPORT_SYMBOL(ttm_eu_backoff_reservation); */ int ttm_eu_reserve_buffers(struct ww_acquire_ctx *ticket, - struct list_head *list) + struct list_head *list, bool intr) { struct ttm_bo_global *glob; struct ttm_validate_buffer *entry; @@ -121,60 +102,55 @@ int ttm_eu_reserve_buffers(struct ww_acquire_ctx *ticket, if (list_empty(list)) return 0; - list_for_each_entry(entry, list, head) { - entry->reserved = false; - entry->put_count = 0; - entry->removed = false; - } - entry = list_first_entry(list, struct ttm_validate_buffer, head); glob = entry->bo->glob; if (ticket) ww_acquire_init(ticket, &reservation_ww_class); -retry: + list_for_each_entry(entry, list, head) { struct ttm_buffer_object *bo = entry->bo; - /* already slowpath reserved? */ - if (entry->reserved) + ret = __ttm_bo_reserve(bo, intr, (ticket == NULL), true, + ticket); + if (!ret && unlikely(atomic_read(&bo->cpu_writers) > 0)) { + __ttm_bo_unreserve(bo); + + ret = -EBUSY; + } + + if (!ret) continue; - ret = __ttm_bo_reserve(bo, true, (ticket == NULL), true, - ticket); + /* uh oh, we lost out, drop every reservation and try + * to only reserve this buffer, then start over if + * this succeeds. + */ + ttm_eu_backoff_reservation_reverse(list, entry); - if (ret == -EDEADLK) { - /* uh oh, we lost out, drop every reservation and try - * to only reserve this buffer, then start over if - * this succeeds. - */ - BUG_ON(ticket == NULL); - spin_lock(&glob->lru_lock); - ttm_eu_backoff_reservation_locked(list); - spin_unlock(&glob->lru_lock); - ttm_eu_list_ref_sub(list); + if (ret == -EDEADLK && intr) { ret = ww_mutex_lock_slow_interruptible(&bo->resv->lock, ticket); - if (unlikely(ret != 0)) { - if (ret == -EINTR) - ret = -ERESTARTSYS; - goto err_fini; - } + } else if (ret == -EDEADLK) { + ww_mutex_lock_slow(&bo->resv->lock, ticket); + ret = 0; + } - entry->reserved = true; - if (unlikely(atomic_read(&bo->cpu_writers) > 0)) { - ret = -EBUSY; - goto err; + if (unlikely(ret != 0)) { + if (ret == -EINTR) + ret = -ERESTARTSYS; + if (ticket) { + ww_acquire_done(ticket); + ww_acquire_fini(ticket); } - goto retry; - } else if (ret) - goto err; - - entry->reserved = true; - if (unlikely(atomic_read(&bo->cpu_writers) > 0)) { - ret = -EBUSY; - goto err; + return ret; } + + /* move this item to the front of the list, + * forces correct iteration of the loop without keeping track + */ + list_del(&entry->head); + list_add(&entry->head, list); } if (ticket) @@ -182,25 +158,12 @@ retry: spin_lock(&glob->lru_lock); ttm_eu_del_from_lru_locked(list); spin_unlock(&glob->lru_lock); - ttm_eu_list_ref_sub(list); return 0; - -err: - spin_lock(&glob->lru_lock); - ttm_eu_backoff_reservation_locked(list); - spin_unlock(&glob->lru_lock); - ttm_eu_list_ref_sub(list); -err_fini: - if (ticket) { - ww_acquire_done(ticket); - ww_acquire_fini(ticket); - } - return ret; } 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; @@ -217,24 +180,15 @@ 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; - 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); - entry->reserved = false; } - spin_unlock(&bdev->fence_lock); 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/ttm/ttm_object.c b/drivers/gpu/drm/ttm/ttm_object.c index d2a05335278..12c87110db3 100644 --- a/drivers/gpu/drm/ttm/ttm_object.c +++ b/drivers/gpu/drm/ttm/ttm_object.c @@ -695,7 +695,7 @@ int ttm_prime_handle_to_fd(struct ttm_object_file *tfile, } dma_buf = dma_buf_export(prime, &tdev->ops, - prime->size, flags); + prime->size, flags, NULL); if (IS_ERR(dma_buf)) { ret = PTR_ERR(dma_buf); ttm_mem_global_free(tdev->mem_glob, diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c index ca65df14476..c96db433f8a 100644 --- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c +++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c @@ -848,6 +848,7 @@ static int ttm_dma_pool_get_pages(struct dma_pool *pool, if (count) { d_page = list_first_entry(&pool->free_list, struct dma_page, page_list); ttm->pages[index] = d_page->p; + ttm_dma->cpu_address[index] = d_page->vaddr; ttm_dma->dma_address[index] = d_page->dma; list_move_tail(&d_page->page_list, &ttm_dma->pages_list); r = 0; @@ -979,6 +980,7 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev) INIT_LIST_HEAD(&ttm_dma->pages_list); for (i = 0; i < ttm->num_pages; i++) { ttm->pages[i] = NULL; + ttm_dma->cpu_address[i] = 0; ttm_dma->dma_address[i] = 0; } diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c index 75f31909004..bf080abc86d 100644 --- a/drivers/gpu/drm/ttm/ttm_tt.c +++ b/drivers/gpu/drm/ttm/ttm_tt.c @@ -55,9 +55,12 @@ static void ttm_tt_alloc_page_directory(struct ttm_tt *ttm) static void ttm_dma_tt_alloc_page_directory(struct ttm_dma_tt *ttm) { - ttm->ttm.pages = drm_calloc_large(ttm->ttm.num_pages, sizeof(void*)); - ttm->dma_address = drm_calloc_large(ttm->ttm.num_pages, - sizeof(*ttm->dma_address)); + ttm->ttm.pages = drm_calloc_large(ttm->ttm.num_pages, + sizeof(*ttm->ttm.pages) + + sizeof(*ttm->dma_address) + + sizeof(*ttm->cpu_address)); + ttm->cpu_address = (void *) (ttm->ttm.pages + ttm->ttm.num_pages); + ttm->dma_address = (void *) (ttm->cpu_address + ttm->ttm.num_pages); } #ifdef CONFIG_X86 @@ -228,7 +231,7 @@ int ttm_dma_tt_init(struct ttm_dma_tt *ttm_dma, struct ttm_bo_device *bdev, INIT_LIST_HEAD(&ttm_dma->pages_list); ttm_dma_tt_alloc_page_directory(ttm_dma); - if (!ttm->pages || !ttm_dma->dma_address) { + if (!ttm->pages) { ttm_tt_destroy(ttm); pr_err("Failed allocating page table\n"); return -ENOMEM; @@ -243,7 +246,7 @@ void ttm_dma_tt_fini(struct ttm_dma_tt *ttm_dma) drm_free_large(ttm->pages); ttm->pages = NULL; - drm_free_large(ttm_dma->dma_address); + ttm_dma->cpu_address = NULL; ttm_dma->dma_address = NULL; } EXPORT_SYMBOL(ttm_dma_tt_fini); |