diff options
Diffstat (limited to 'drivers/gpu')
42 files changed, 1870 insertions, 656 deletions
diff --git a/drivers/gpu/drm/drm_gem.c b/drivers/gpu/drm/drm_gem.c index 186d62eb063..396e60ce811 100644 --- a/drivers/gpu/drm/drm_gem.c +++ b/drivers/gpu/drm/drm_gem.c @@ -285,6 +285,94 @@ again: } EXPORT_SYMBOL(drm_gem_handle_create); + +/** + * drm_gem_free_mmap_offset - release a fake mmap offset for an object + * @obj: obj in question + * + * This routine frees fake offsets allocated by drm_gem_create_mmap_offset(). + */ +void +drm_gem_free_mmap_offset(struct drm_gem_object *obj) +{ + struct drm_device *dev = obj->dev; + struct drm_gem_mm *mm = dev->mm_private; + struct drm_map_list *list = &obj->map_list; + + drm_ht_remove_item(&mm->offset_hash, &list->hash); + drm_mm_put_block(list->file_offset_node); + kfree(list->map); + list->map = NULL; +} +EXPORT_SYMBOL(drm_gem_free_mmap_offset); + +/** + * drm_gem_create_mmap_offset - create a fake mmap offset for an object + * @obj: obj in question + * + * GEM memory mapping works by handing back to userspace a fake mmap offset + * it can use in a subsequent mmap(2) call. The DRM core code then looks + * up the object based on the offset and sets up the various memory mapping + * structures. + * + * This routine allocates and attaches a fake offset for @obj. + */ +int +drm_gem_create_mmap_offset(struct drm_gem_object *obj) +{ + struct drm_device *dev = obj->dev; + struct drm_gem_mm *mm = dev->mm_private; + struct drm_map_list *list; + struct drm_local_map *map; + int ret = 0; + + /* Set the object up for mmap'ing */ + list = &obj->map_list; + list->map = kzalloc(sizeof(struct drm_map_list), GFP_KERNEL); + if (!list->map) + return -ENOMEM; + + map = list->map; + map->type = _DRM_GEM; + map->size = obj->size; + map->handle = obj; + + /* Get a DRM GEM mmap offset allocated... */ + list->file_offset_node = drm_mm_search_free(&mm->offset_manager, + obj->size / PAGE_SIZE, 0, 0); + + if (!list->file_offset_node) { + DRM_ERROR("failed to allocate offset for bo %d\n", obj->name); + ret = -ENOSPC; + goto out_free_list; + } + + list->file_offset_node = drm_mm_get_block(list->file_offset_node, + obj->size / PAGE_SIZE, 0); + if (!list->file_offset_node) { + ret = -ENOMEM; + goto out_free_list; + } + + list->hash.key = list->file_offset_node->start; + ret = drm_ht_insert_item(&mm->offset_hash, &list->hash); + if (ret) { + DRM_ERROR("failed to add to map hash\n"); + goto out_free_mm; + } + + return 0; + +out_free_mm: + drm_mm_put_block(list->file_offset_node); +out_free_list: + kfree(list->map); + list->map = NULL; + + return ret; +} +EXPORT_SYMBOL(drm_gem_create_mmap_offset); + /** Returns a reference to the object named by the handle. */ struct drm_gem_object * drm_gem_object_lookup(struct drm_device *dev, struct drm_file *filp, diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index a546a71fb06..ee59f31316e 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -1265,74 +1265,6 @@ out: } /** - * i915_gem_create_mmap_offset - create a fake mmap offset for an object - * @obj: obj in question - * - * GEM memory mapping works by handing back to userspace a fake mmap offset - * it can use in a subsequent mmap(2) call. The DRM core code then looks - * up the object based on the offset and sets up the various memory mapping - * structures. - * - * This routine allocates and attaches a fake offset for @obj. - */ -static int -i915_gem_create_mmap_offset(struct drm_i915_gem_object *obj) -{ - struct drm_device *dev = obj->base.dev; - struct drm_gem_mm *mm = dev->mm_private; - struct drm_map_list *list; - struct drm_local_map *map; - int ret = 0; - - /* Set the object up for mmap'ing */ - list = &obj->base.map_list; - list->map = kzalloc(sizeof(struct drm_map_list), GFP_KERNEL); - if (!list->map) - return -ENOMEM; - - map = list->map; - map->type = _DRM_GEM; - map->size = obj->base.size; - map->handle = obj; - - /* Get a DRM GEM mmap offset allocated... */ - list->file_offset_node = drm_mm_search_free(&mm->offset_manager, - obj->base.size / PAGE_SIZE, - 0, 0); - if (!list->file_offset_node) { - DRM_ERROR("failed to allocate offset for bo %d\n", - obj->base.name); - ret = -ENOSPC; - goto out_free_list; - } - - list->file_offset_node = drm_mm_get_block(list->file_offset_node, - obj->base.size / PAGE_SIZE, - 0); - if (!list->file_offset_node) { - ret = -ENOMEM; - goto out_free_list; - } - - list->hash.key = list->file_offset_node->start; - ret = drm_ht_insert_item(&mm->offset_hash, &list->hash); - if (ret) { - DRM_ERROR("failed to add to map hash\n"); - goto out_free_mm; - } - - return 0; - -out_free_mm: - drm_mm_put_block(list->file_offset_node); -out_free_list: - kfree(list->map); - list->map = NULL; - - return ret; -} - -/** * i915_gem_release_mmap - remove physical page mappings * @obj: obj in question * @@ -1360,19 +1292,6 @@ i915_gem_release_mmap(struct drm_i915_gem_object *obj) obj->fault_mappable = false; } -static void -i915_gem_free_mmap_offset(struct drm_i915_gem_object *obj) -{ - struct drm_device *dev = obj->base.dev; - struct drm_gem_mm *mm = dev->mm_private; - struct drm_map_list *list = &obj->base.map_list; - - drm_ht_remove_item(&mm->offset_hash, &list->hash); - drm_mm_put_block(list->file_offset_node); - kfree(list->map); - list->map = NULL; -} - static uint32_t i915_gem_get_gtt_size(struct drm_device *dev, uint32_t size, int tiling_mode) { @@ -1485,7 +1404,7 @@ i915_gem_mmap_gtt(struct drm_file *file, } if (!obj->base.map_list.map) { - ret = i915_gem_create_mmap_offset(obj); + ret = drm_gem_create_mmap_offset(&obj->base); if (ret) goto out; } @@ -3752,7 +3671,7 @@ static void i915_gem_free_object_tail(struct drm_i915_gem_object *obj) trace_i915_gem_object_destroy(obj); if (obj->base.map_list.map) - i915_gem_free_mmap_offset(obj); + drm_gem_free_mmap_offset(&obj->base); drm_gem_object_release(&obj->base); i915_gem_info_remove_obj(dev_priv, obj->base.size); diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c index 7226f419e17..424dff5d0ab 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -1104,7 +1104,8 @@ 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_bo_wait(&nvbo->bo, false, false, false, + TTM_USAGE_READWRITE); 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 5f0bc57fdaa..322bf62a064 100644 --- a/drivers/gpu/drm/nouveau/nouveau_gem.c +++ b/drivers/gpu/drm/nouveau/nouveau_gem.c @@ -589,7 +589,8 @@ 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); + ret = ttm_bo_wait(&nvbo->bo, false, false, false, + TTM_USAGE_READWRITE); spin_unlock(&nvbo->bo.bdev->fence_lock); if (ret) { NV_ERROR(dev, "reloc wait_idle failed: %d\n", ret); @@ -825,7 +826,7 @@ 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); + ret = ttm_bo_wait(&nvbo->bo, true, true, no_wait, TTM_USAGE_READWRITE); spin_unlock(&nvbo->bo.bdev->fence_lock); drm_gem_object_unreference_unlocked(gem); return ret; diff --git a/drivers/gpu/drm/radeon/evergreen.c b/drivers/gpu/drm/radeon/evergreen.c index dc0a5b56c81..a72dbb3e133 100644 --- a/drivers/gpu/drm/radeon/evergreen.c +++ b/drivers/gpu/drm/radeon/evergreen.c @@ -935,6 +935,9 @@ int evergreen_pcie_gart_enable(struct radeon_device *rdev) WREG32(VM_CONTEXT1_CNTL, 0); evergreen_pcie_gart_tlb_flush(rdev); + DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", + (unsigned)(rdev->mc.gtt_size >> 20), + (unsigned long long)rdev->gart.table_addr); rdev->gart.ready = true; return 0; } diff --git a/drivers/gpu/drm/radeon/ni.c b/drivers/gpu/drm/radeon/ni.c index cbf57d75d92..bf4fce7c43f 100644 --- a/drivers/gpu/drm/radeon/ni.c +++ b/drivers/gpu/drm/radeon/ni.c @@ -999,6 +999,9 @@ int cayman_pcie_gart_enable(struct radeon_device *rdev) WREG32(VM_CONTEXT1_CNTL, 0); cayman_pcie_gart_tlb_flush(rdev); + DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", + (unsigned)(rdev->mc.gtt_size >> 20), + (unsigned long long)rdev->gart.table_addr); rdev->gart.ready = true; return 0; } diff --git a/drivers/gpu/drm/radeon/r100.c b/drivers/gpu/drm/radeon/r100.c index f2204cb1ccd..574f2c7c6dd 100644 --- a/drivers/gpu/drm/radeon/r100.c +++ b/drivers/gpu/drm/radeon/r100.c @@ -513,6 +513,9 @@ int r100_pci_gart_enable(struct radeon_device *rdev) tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN; WREG32(RADEON_AIC_CNTL, tmp); r100_pci_gart_tlb_flush(rdev); + DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", + (unsigned)(rdev->mc.gtt_size >> 20), + (unsigned long long)rdev->gart.table_addr); rdev->gart.ready = true; return 0; } diff --git a/drivers/gpu/drm/radeon/r300.c b/drivers/gpu/drm/radeon/r300.c index 55a7f190027..33f2b68c680 100644 --- a/drivers/gpu/drm/radeon/r300.c +++ b/drivers/gpu/drm/radeon/r300.c @@ -144,8 +144,9 @@ int rv370_pcie_gart_enable(struct radeon_device *rdev) tmp |= RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD; WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp); rv370_pcie_gart_tlb_flush(rdev); - DRM_INFO("PCIE GART of %uM enabled (table at 0x%08X).\n", - (unsigned)(rdev->mc.gtt_size >> 20), table_addr); + DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", + (unsigned)(rdev->mc.gtt_size >> 20), + (unsigned long long)table_addr); rdev->gart.ready = true; return 0; } diff --git a/drivers/gpu/drm/radeon/r600.c b/drivers/gpu/drm/radeon/r600.c index aa5571b73aa..334aee6eab7 100644 --- a/drivers/gpu/drm/radeon/r600.c +++ b/drivers/gpu/drm/radeon/r600.c @@ -993,6 +993,9 @@ int r600_pcie_gart_enable(struct radeon_device *rdev) WREG32(VM_CONTEXT0_CNTL + (i * 4), 0); r600_pcie_gart_tlb_flush(rdev); + DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", + (unsigned)(rdev->mc.gtt_size >> 20), + (unsigned long long)rdev->gart.table_addr); rdev->gart.ready = true; return 0; } diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h index 32807baf55e..0040d28816f 100644 --- a/drivers/gpu/drm/radeon/radeon.h +++ b/drivers/gpu/drm/radeon/radeon.h @@ -1143,6 +1143,8 @@ int radeon_gem_set_tiling_ioctl(struct drm_device *dev, void *data, struct drm_file *filp); int radeon_gem_get_tiling_ioctl(struct drm_device *dev, void *data, struct drm_file *filp); +int radeon_gem_wait_ioctl(struct drm_device *dev, void *data, + struct drm_file *filp); /* VRAM scratch page for HDP bug */ struct r700_vram_scratch { diff --git a/drivers/gpu/drm/radeon/radeon_cs.c b/drivers/gpu/drm/radeon/radeon_cs.c index fae00c0d75a..f0b9066abc5 100644 --- a/drivers/gpu/drm/radeon/radeon_cs.c +++ b/drivers/gpu/drm/radeon/radeon_cs.c @@ -80,6 +80,10 @@ 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; + if (r->read_domains) + p->relocs[i].lobj.tv.usage |= TTM_USAGE_READ; + if (r->write_domain) + p->relocs[i].lobj.tv.usage |= TTM_USAGE_WRITE; 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_drv.c b/drivers/gpu/drm/radeon/radeon_drv.c index e71d2ed7fa1..bd187e097e7 100644 --- a/drivers/gpu/drm/radeon/radeon_drv.c +++ b/drivers/gpu/drm/radeon/radeon_drv.c @@ -52,9 +52,10 @@ * 2.9.0 - r600 tiling (s3tc,rgtc) working, SET_PREDICATION packet 3 on r600 + eg, backend query * 2.10.0 - fusion 2D tiling * 2.11.0 - backend map, initial compute support for the CS checker + * 2.12.0 - DRM_RADEON_GEM_WAIT ioctl */ #define KMS_DRIVER_MAJOR 2 -#define KMS_DRIVER_MINOR 11 +#define KMS_DRIVER_MINOR 12 #define KMS_DRIVER_PATCHLEVEL 0 int radeon_driver_load_kms(struct drm_device *dev, unsigned long flags); int radeon_driver_unload_kms(struct drm_device *dev); diff --git a/drivers/gpu/drm/radeon/radeon_gart.c b/drivers/gpu/drm/radeon/radeon_gart.c index a533f52fd16..fdc3a9a54bf 100644 --- a/drivers/gpu/drm/radeon/radeon_gart.c +++ b/drivers/gpu/drm/radeon/radeon_gart.c @@ -142,7 +142,7 @@ void radeon_gart_unbind(struct radeon_device *rdev, unsigned offset, u64 page_base; if (!rdev->gart.ready) { - WARN(1, "trying to unbind memory to unitialized GART !\n"); + WARN(1, "trying to unbind memory from uninitialized GART !\n"); return; } t = offset / RADEON_GPU_PAGE_SIZE; @@ -174,7 +174,7 @@ int radeon_gart_bind(struct radeon_device *rdev, unsigned offset, int i, j; if (!rdev->gart.ready) { - WARN(1, "trying to bind memory to unitialized GART !\n"); + WARN(1, "trying to bind memory to uninitialized GART !\n"); return -EINVAL; } t = offset / RADEON_GPU_PAGE_SIZE; diff --git a/drivers/gpu/drm/radeon/radeon_gem.c b/drivers/gpu/drm/radeon/radeon_gem.c index aa1ca2dea42..2edc2a40d4d 100644 --- a/drivers/gpu/drm/radeon/radeon_gem.c +++ b/drivers/gpu/drm/radeon/radeon_gem.c @@ -122,7 +122,7 @@ int radeon_gem_set_domain(struct drm_gem_object *gobj, } if (domain == RADEON_GEM_DOMAIN_CPU) { /* Asking for cpu access wait for object idle */ - r = radeon_bo_wait(robj, NULL, false); + r = radeon_bo_wait(robj, NULL, false, TTM_USAGE_READWRITE); if (r) { printk(KERN_ERR "Failed to wait for object !\n"); return r; @@ -273,7 +273,7 @@ int radeon_gem_busy_ioctl(struct drm_device *dev, void *data, return -ENOENT; } robj = gem_to_radeon_bo(gobj); - r = radeon_bo_wait(robj, &cur_placement, true); + r = radeon_bo_wait(robj, &cur_placement, true, TTM_USAGE_READWRITE); switch (cur_placement) { case TTM_PL_VRAM: args->domain = RADEON_GEM_DOMAIN_VRAM; @@ -303,7 +303,7 @@ int radeon_gem_wait_idle_ioctl(struct drm_device *dev, void *data, return -ENOENT; } robj = gem_to_radeon_bo(gobj); - r = radeon_bo_wait(robj, NULL, false); + r = radeon_bo_wait(robj, NULL, false, TTM_USAGE_READWRITE); /* callback hw specific functions if any */ if (robj->rdev->asic->ioctl_wait_idle) robj->rdev->asic->ioctl_wait_idle(robj->rdev, robj); @@ -311,6 +311,36 @@ int radeon_gem_wait_idle_ioctl(struct drm_device *dev, void *data, return r; } +int radeon_gem_wait_ioctl(struct drm_device *dev, void *data, + struct drm_file *filp) +{ + struct drm_radeon_gem_wait *args = data; + struct drm_gem_object *gobj; + struct radeon_bo *robj; + bool no_wait = (args->flags & RADEON_GEM_NO_WAIT) != 0; + enum ttm_buffer_usage usage = 0; + int r; + + if (args->flags & RADEON_GEM_USAGE_READ) + usage |= TTM_USAGE_READ; + if (args->flags & RADEON_GEM_USAGE_WRITE) + usage |= TTM_USAGE_WRITE; + if (!usage) + usage = TTM_USAGE_READWRITE; + + gobj = drm_gem_object_lookup(dev, filp, args->handle); + if (gobj == NULL) { + return -ENOENT; + } + robj = gem_to_radeon_bo(gobj); + r = radeon_bo_wait(robj, NULL, no_wait, usage); + /* callback hw specific functions if any */ + if (!no_wait && robj->rdev->asic->ioctl_wait_idle) + robj->rdev->asic->ioctl_wait_idle(robj->rdev, robj); + drm_gem_object_unreference_unlocked(gobj); + return r; +} + int radeon_gem_set_tiling_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) { diff --git a/drivers/gpu/drm/radeon/radeon_kms.c b/drivers/gpu/drm/radeon/radeon_kms.c index be2c1224e68..a749c262663 100644 --- a/drivers/gpu/drm/radeon/radeon_kms.c +++ b/drivers/gpu/drm/radeon/radeon_kms.c @@ -451,5 +451,6 @@ struct drm_ioctl_desc radeon_ioctls_kms[] = { DRM_IOCTL_DEF_DRV(RADEON_GEM_SET_TILING, radeon_gem_set_tiling_ioctl, DRM_AUTH|DRM_UNLOCKED), DRM_IOCTL_DEF_DRV(RADEON_GEM_GET_TILING, radeon_gem_get_tiling_ioctl, DRM_AUTH|DRM_UNLOCKED), DRM_IOCTL_DEF_DRV(RADEON_GEM_BUSY, radeon_gem_busy_ioctl, DRM_AUTH|DRM_UNLOCKED), + DRM_IOCTL_DEF_DRV(RADEON_GEM_WAIT, radeon_gem_wait_ioctl, DRM_AUTH|DRM_UNLOCKED), }; int radeon_max_kms_ioctl = DRM_ARRAY_SIZE(radeon_ioctls_kms); diff --git a/drivers/gpu/drm/radeon/radeon_object.h b/drivers/gpu/drm/radeon/radeon_object.h index ede6c13628f..a057a8e5a6e 100644 --- a/drivers/gpu/drm/radeon/radeon_object.h +++ b/drivers/gpu/drm/radeon/radeon_object.h @@ -119,7 +119,7 @@ static inline u64 radeon_bo_mmap_offset(struct radeon_bo *bo) } static inline int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type, - bool no_wait) + bool no_wait, enum ttm_buffer_usage usage) { int r; @@ -130,7 +130,7 @@ static inline int radeon_bo_wait(struct radeon_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, usage); spin_unlock(&bo->tbo.bdev->fence_lock); ttm_bo_unreserve(&bo->tbo); return r; diff --git a/drivers/gpu/drm/radeon/rs400.c b/drivers/gpu/drm/radeon/rs400.c index aa6a66eeb4e..89a6e1ecea8 100644 --- a/drivers/gpu/drm/radeon/rs400.c +++ b/drivers/gpu/drm/radeon/rs400.c @@ -182,6 +182,9 @@ int rs400_gart_enable(struct radeon_device *rdev) /* Enable gart */ WREG32_MC(RS480_AGP_ADDRESS_SPACE_SIZE, (RS480_GART_EN | size_reg)); rs400_gart_tlb_flush(rdev); + DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", + (unsigned)(rdev->mc.gtt_size >> 20), + (unsigned long long)rdev->gart.table_addr); rdev->gart.ready = true; return 0; } diff --git a/drivers/gpu/drm/radeon/rs600.c b/drivers/gpu/drm/radeon/rs600.c index 4b5d0e6974a..9320dd6404f 100644 --- a/drivers/gpu/drm/radeon/rs600.c +++ b/drivers/gpu/drm/radeon/rs600.c @@ -484,6 +484,9 @@ static int rs600_gart_enable(struct radeon_device *rdev) tmp = RREG32_MC(R_000009_MC_CNTL1); WREG32_MC(R_000009_MC_CNTL1, (tmp | S_000009_ENABLE_PAGE_TABLES(1))); rs600_gart_tlb_flush(rdev); + DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", + (unsigned)(rdev->mc.gtt_size >> 20), + (unsigned long long)rdev->gart.table_addr); rdev->gart.ready = true; return 0; } diff --git a/drivers/gpu/drm/radeon/rv770.c b/drivers/gpu/drm/radeon/rv770.c index 4720d000d44..80928f9ff80 100644 --- a/drivers/gpu/drm/radeon/rv770.c +++ b/drivers/gpu/drm/radeon/rv770.c @@ -161,6 +161,9 @@ int rv770_pcie_gart_enable(struct radeon_device *rdev) WREG32(VM_CONTEXT0_CNTL + (i * 4), 0); r600_pcie_gart_tlb_flush(rdev); + DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", + (unsigned)(rdev->mc.gtt_size >> 20), + (unsigned long long)rdev->gart.table_addr); rdev->gart.ready = true; return 0; } diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index a4d38d85909..b824d9bdd87 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -498,7 +498,7 @@ 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); + (void) ttm_bo_wait(bo, false, false, true, TTM_USAGE_READWRITE); if (!bo->sync_obj) { spin_lock(&glob->lru_lock); @@ -566,7 +566,8 @@ 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); + ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu, + TTM_USAGE_READWRITE); spin_unlock(&bdev->fence_lock); if (unlikely(ret != 0)) @@ -725,7 +726,8 @@ 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); + ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu, + TTM_USAGE_READWRITE); spin_unlock(&bdev->fence_lock); if (unlikely(ret != 0)) { @@ -1072,7 +1074,8 @@ 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); + ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu, + TTM_USAGE_READWRITE); spin_unlock(&bdev->fence_lock); if (ret) return ret; @@ -1692,34 +1695,83 @@ 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) + bool lazy, bool interruptible, bool no_wait, + enum ttm_buffer_usage usage) { 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; - if (likely(bo->sync_obj == NULL)) + 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)) return 0; - while (bo->sync_obj) { + while (*bo_sync_obj) { - 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); + if (driver->sync_obj_signaled(*bo_sync_obj, bo->sync_obj_arg)) { + ttm_bo_unref_sync_obj_locked(bo, *bo_sync_obj, NULL); 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, @@ -1730,16 +1782,9 @@ 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)) { - 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); + ttm_bo_unref_sync_obj_locked(bo, *bo_sync_obj, &sync_obj); } else { spin_unlock(&bdev->fence_lock); driver->sync_obj_unref(&sync_obj); @@ -1763,7 +1808,7 @@ 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); + ret = ttm_bo_wait(bo, false, true, no_wait, TTM_USAGE_READWRITE); spin_unlock(&bdev->fence_lock); if (likely(ret == 0)) atomic_inc(&bo->cpu_writers); @@ -1837,7 +1882,7 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink) */ spin_lock(&bo->bdev->fence_lock); - ret = ttm_bo_wait(bo, false, false, false); + ret = ttm_bo_wait(bo, false, false, false, TTM_USAGE_READWRITE); 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 ae3c6f5dd2b..6135f58169c 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_util.c +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c @@ -436,6 +436,8 @@ 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; @@ -618,20 +620,30 @@ 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; + void *tmp_obj = NULL, *tmp_obj_read = NULL, *tmp_obj_write = NULL; spin_lock(&bdev->fence_lock); - if (bo->sync_obj) { + if (bo->sync_obj) tmp_obj = bo->sync_obj; - bo->sync_obj = NULL; - } + 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 = 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); + ret = ttm_bo_wait(bo, false, false, false, + TTM_USAGE_READWRITE); 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; @@ -655,6 +667,10 @@ 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 221b924aceb..ff1e26f4b09 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); + ret = ttm_bo_wait(bo, false, true, false, TTM_USAGE_READWRITE); 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 3832fe10b4d..36d111a8823 100644 --- a/drivers/gpu/drm/ttm/ttm_execbuf_util.c +++ b/drivers/gpu/drm/ttm/ttm_execbuf_util.c @@ -221,8 +221,18 @@ 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; @@ -231,8 +241,15 @@ 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/Makefile b/drivers/gpu/drm/vmwgfx/Makefile index c9281a1b1d3..7d8e9d5d498 100644 --- a/drivers/gpu/drm/vmwgfx/Makefile +++ b/drivers/gpu/drm/vmwgfx/Makefile @@ -4,6 +4,7 @@ ccflags-y := -Iinclude/drm vmwgfx-y := vmwgfx_execbuf.o vmwgfx_gmr.o vmwgfx_kms.o vmwgfx_drv.o \ vmwgfx_fb.o vmwgfx_ioctl.o vmwgfx_resource.o vmwgfx_buffer.o \ vmwgfx_fifo.o vmwgfx_irq.o vmwgfx_ldu.o vmwgfx_ttm_glue.o \ - vmwgfx_overlay.o vmwgfx_fence.o vmwgfx_gmrid_manager.o + vmwgfx_overlay.o vmwgfx_marker.o vmwgfx_gmrid_manager.o \ + vmwgfx_fence.o obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o diff --git a/drivers/gpu/drm/vmwgfx/svga_reg.h b/drivers/gpu/drm/vmwgfx/svga_reg.h index 1b96c2ec07d..ec5aad9b6ed 100644 --- a/drivers/gpu/drm/vmwgfx/svga_reg.h +++ b/drivers/gpu/drm/vmwgfx/svga_reg.h @@ -39,6 +39,15 @@ #define PCI_DEVICE_ID_VMWARE_SVGA2 0x0405 /* + * SVGA_REG_ENABLE bit definitions. + */ +#define SVGA_REG_ENABLE_DISABLE 0 +#define SVGA_REG_ENABLE_ENABLE 1 +#define SVGA_REG_ENABLE_HIDE 2 +#define SVGA_REG_ENABLE_ENABLE_HIDE (SVGA_REG_ENABLE_ENABLE |\ + SVGA_REG_ENABLE_HIDE) + +/* * Legal values for the SVGA_REG_CURSOR_ON register in old-fashioned * cursor bypass mode. This is still supported, but no new guest * drivers should use it. @@ -158,7 +167,9 @@ enum { SVGA_REG_GMR_MAX_DESCRIPTOR_LENGTH = 44, SVGA_REG_TRACES = 45, /* Enable trace-based updates even when FIFO is on */ - SVGA_REG_TOP = 46, /* Must be 1 more than the last register */ + SVGA_REG_GMRS_MAX_PAGES = 46, /* Maximum number of 4KB pages for all GMRs */ + SVGA_REG_MEMORY_SIZE = 47, /* Total dedicated device memory excluding FIFO */ + SVGA_REG_TOP = 48, /* Must be 1 more than the last register */ SVGA_PALETTE_BASE = 1024, /* Base of SVGA color map */ /* Next 768 (== 256*3) registers exist for colormap */ @@ -370,6 +381,15 @@ struct SVGASignedPoint { * Note the holes in the bitfield. Missing bits have been deprecated, * and must not be reused. Those capabilities will never be reported * by new versions of the SVGA device. + * + * SVGA_CAP_GMR2 -- + * Provides asynchronous commands to define and remap guest memory + * regions. Adds device registers SVGA_REG_GMRS_MAX_PAGES and + * SVGA_REG_MEMORY_SIZE. + * + * SVGA_CAP_SCREEN_OBJECT_2 -- + * Allow screen object support, and require backing stores from the + * guest for each screen object. */ #define SVGA_CAP_NONE 0x00000000 @@ -387,6 +407,8 @@ struct SVGASignedPoint { #define SVGA_CAP_DISPLAY_TOPOLOGY 0x00080000 // Legacy multi-monitor support #define SVGA_CAP_GMR 0x00100000 #define SVGA_CAP_TRACES 0x00200000 +#define SVGA_CAP_GMR2 0x00400000 +#define SVGA_CAP_SCREEN_OBJECT_2 0x00800000 /* @@ -885,6 +907,8 @@ typedef enum { SVGA_CMD_BLIT_SCREEN_TO_GMRFB = 38, SVGA_CMD_ANNOTATION_FILL = 39, SVGA_CMD_ANNOTATION_COPY = 40, + SVGA_CMD_DEFINE_GMR2 = 41, + SVGA_CMD_REMAP_GMR2 = 42, SVGA_CMD_MAX } SVGAFifoCmdId; @@ -1343,4 +1367,74 @@ struct { uint32 srcScreenId; } SVGAFifoCmdAnnotationCopy; + +/* + * SVGA_CMD_DEFINE_GMR2 -- + * + * Define guest memory region v2. See the description of GMRs above. + * + * Availability: + * SVGA_CAP_GMR2 + */ + +typedef +struct { + uint32 gmrId; + uint32 numPages; +} +SVGAFifoCmdDefineGMR2; + + +/* + * SVGA_CMD_REMAP_GMR2 -- + * + * Remap guest memory region v2. See the description of GMRs above. + * + * This command allows guest to modify a portion of an existing GMR by + * invalidating it or reassigning it to different guest physical pages. + * The pages are identified by physical page number (PPN). The pages + * are assumed to be pinned and valid for DMA operations. + * + * Description of command flags: + * + * SVGA_REMAP_GMR2_VIA_GMR: If enabled, references a PPN list in a GMR. + * The PPN list must not overlap with the remap region (this can be + * handled trivially by referencing a separate GMR). If flag is + * disabled, PPN list is appended to SVGARemapGMR command. + * + * SVGA_REMAP_GMR2_PPN64: If set, PPN list is in PPN64 format, otherwise + * it is in PPN32 format. + * + * SVGA_REMAP_GMR2_SINGLE_PPN: If set, PPN list contains a single entry. + * A single PPN can be used to invalidate a portion of a GMR or + * map it to to a single guest scratch page. + * + * Availability: + * SVGA_CAP_GMR2 + */ + +typedef enum { + SVGA_REMAP_GMR2_PPN32 = 0, + SVGA_REMAP_GMR2_VIA_GMR = (1 << 0), + SVGA_REMAP_GMR2_PPN64 = (1 << 1), + SVGA_REMAP_GMR2_SINGLE_PPN = (1 << 2), +} SVGARemapGMR2Flags; + +typedef +struct { + uint32 gmrId; + SVGARemapGMR2Flags flags; + uint32 offsetPages; /* offset in pages to begin remap */ + uint32 numPages; /* number of pages to remap */ + /* + * Followed by additional data depending on SVGARemapGMR2Flags. + * + * If flag SVGA_REMAP_GMR2_VIA_GMR is set, single SVGAGuestPtr follows. + * Otherwise an array of page descriptors in PPN32 or PPN64 format + * (according to flag SVGA_REMAP_GMR2_PPN64) follows. If flag + * SVGA_REMAP_GMR2_SINGLE_PPN is set, array contains a single entry. + */ +} +SVGAFifoCmdRemapGMR2; + #endif diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c index 87e43e0733b..5d665ce8cbe 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c @@ -274,39 +274,39 @@ static int vmw_ttm_fault_reserve_notify(struct ttm_buffer_object *bo) static void *vmw_sync_obj_ref(void *sync_obj) { - return sync_obj; + + return (void *) + vmw_fence_obj_reference((struct vmw_fence_obj *) sync_obj); } static void vmw_sync_obj_unref(void **sync_obj) { - *sync_obj = NULL; + vmw_fence_obj_unreference((struct vmw_fence_obj **) sync_obj); } static int vmw_sync_obj_flush(void *sync_obj, void *sync_arg) { - struct vmw_private *dev_priv = (struct vmw_private *)sync_arg; - - mutex_lock(&dev_priv->hw_mutex); - vmw_write(dev_priv, SVGA_REG_SYNC, SVGA_SYNC_GENERIC); - mutex_unlock(&dev_priv->hw_mutex); + vmw_fence_obj_flush((struct vmw_fence_obj *) sync_obj); return 0; } static bool vmw_sync_obj_signaled(void *sync_obj, void *sync_arg) { - struct vmw_private *dev_priv = (struct vmw_private *)sync_arg; - uint32_t sequence = (unsigned long) sync_obj; + unsigned long flags = (unsigned long) sync_arg; + return vmw_fence_obj_signaled((struct vmw_fence_obj *) sync_obj, + (uint32_t) flags); - return vmw_fence_signaled(dev_priv, sequence); } static int vmw_sync_obj_wait(void *sync_obj, void *sync_arg, bool lazy, bool interruptible) { - struct vmw_private *dev_priv = (struct vmw_private *)sync_arg; - uint32_t sequence = (unsigned long) sync_obj; + unsigned long flags = (unsigned long) sync_arg; - return vmw_wait_fence(dev_priv, false, sequence, false, 3*HZ); + return vmw_fence_obj_wait((struct vmw_fence_obj *) sync_obj, + (uint32_t) flags, + lazy, interruptible, + VMW_FENCE_WAIT_TIMEOUT); } struct ttm_bo_driver vmw_bo_driver = { diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c index 96949b93d92..d4829cbf326 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c @@ -82,16 +82,18 @@ #define DRM_IOCTL_VMW_EXECBUF \ DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_EXECBUF, \ struct drm_vmw_execbuf_arg) -#define DRM_IOCTL_VMW_FIFO_DEBUG \ - DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_FIFO_DEBUG, \ - struct drm_vmw_fifo_debug_arg) +#define DRM_IOCTL_VMW_GET_3D_CAP \ + DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_GET_3D_CAP, \ + struct drm_vmw_get_3d_cap_arg) #define DRM_IOCTL_VMW_FENCE_WAIT \ DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_FENCE_WAIT, \ struct drm_vmw_fence_wait_arg) -#define DRM_IOCTL_VMW_UPDATE_LAYOUT \ - DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_UPDATE_LAYOUT, \ - struct drm_vmw_update_layout_arg) - +#define DRM_IOCTL_VMW_FENCE_SIGNALED \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_FENCE_SIGNALED, \ + struct drm_vmw_fence_signaled_arg) +#define DRM_IOCTL_VMW_FENCE_UNREF \ + DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_FENCE_UNREF, \ + struct drm_vmw_fence_arg) /** * The core DRM version of this macro doesn't account for @@ -135,12 +137,15 @@ static struct drm_ioctl_desc vmw_ioctls[] = { DRM_AUTH | DRM_UNLOCKED), VMW_IOCTL_DEF(VMW_EXECBUF, vmw_execbuf_ioctl, DRM_AUTH | DRM_UNLOCKED), - VMW_IOCTL_DEF(VMW_FIFO_DEBUG, vmw_fifo_debug_ioctl, - DRM_AUTH | DRM_ROOT_ONLY | DRM_MASTER | DRM_UNLOCKED), - VMW_IOCTL_DEF(VMW_FENCE_WAIT, vmw_fence_wait_ioctl, + VMW_IOCTL_DEF(VMW_FENCE_WAIT, vmw_fence_obj_wait_ioctl, + DRM_AUTH | DRM_UNLOCKED), + VMW_IOCTL_DEF(VMW_FENCE_SIGNALED, + vmw_fence_obj_signaled_ioctl, + DRM_AUTH | DRM_UNLOCKED), + VMW_IOCTL_DEF(VMW_FENCE_UNREF, vmw_fence_obj_unref_ioctl, + DRM_AUTH | DRM_UNLOCKED), + VMW_IOCTL_DEF(VMW_GET_3D_CAP, vmw_get_cap_3d_ioctl, DRM_AUTH | DRM_UNLOCKED), - VMW_IOCTL_DEF(VMW_UPDATE_LAYOUT, vmw_kms_update_layout_ioctl, - DRM_MASTER | DRM_CONTROL_ALLOW | DRM_UNLOCKED) }; static struct pci_device_id vmw_pci_id_list[] = { @@ -189,6 +194,10 @@ static void vmw_print_capabilities(uint32_t capabilities) DRM_INFO(" GMR.\n"); if (capabilities & SVGA_CAP_TRACES) DRM_INFO(" Traces.\n"); + if (capabilities & SVGA_CAP_GMR2) + DRM_INFO(" GMR2.\n"); + if (capabilities & SVGA_CAP_SCREEN_OBJECT_2) + DRM_INFO(" Screen Object 2.\n"); } static int vmw_request_device(struct vmw_private *dev_priv) @@ -200,16 +209,25 @@ static int vmw_request_device(struct vmw_private *dev_priv) DRM_ERROR("Unable to initialize FIFO.\n"); return ret; } + vmw_fence_fifo_up(dev_priv->fman); return 0; } static void vmw_release_device(struct vmw_private *dev_priv) { + vmw_fence_fifo_down(dev_priv->fman); vmw_fifo_release(dev_priv, &dev_priv->fifo); } -int vmw_3d_resource_inc(struct vmw_private *dev_priv) +/** + * Increase the 3d resource refcount. + * If the count was prevously zero, initialize the fifo, switching to svga + * mode. Note that the master holds a ref as well, and may request an + * explicit switch to svga mode if fb is not running, using @unhide_svga. + */ +int vmw_3d_resource_inc(struct vmw_private *dev_priv, + bool unhide_svga) { int ret = 0; @@ -218,19 +236,42 @@ int vmw_3d_resource_inc(struct vmw_private *dev_priv) ret = vmw_request_device(dev_priv); if (unlikely(ret != 0)) --dev_priv->num_3d_resources; + } else if (unhide_svga) { + mutex_lock(&dev_priv->hw_mutex); + vmw_write(dev_priv, SVGA_REG_ENABLE, + vmw_read(dev_priv, SVGA_REG_ENABLE) & + ~SVGA_REG_ENABLE_HIDE); + mutex_unlock(&dev_priv->hw_mutex); } + mutex_unlock(&dev_priv->release_mutex); return ret; } - -void vmw_3d_resource_dec(struct vmw_private *dev_priv) +/** + * Decrease the 3d resource refcount. + * If the count reaches zero, disable the fifo, switching to vga mode. + * Note that the master holds a refcount as well, and may request an + * explicit switch to vga mode when it releases its refcount to account + * for the situation of an X server vt switch to VGA with 3d resources + * active. + */ +void vmw_3d_resource_dec(struct vmw_private *dev_priv, + bool hide_svga) { int32_t n3d; mutex_lock(&dev_priv->release_mutex); if (unlikely(--dev_priv->num_3d_resources == 0)) vmw_release_device(dev_priv); + else if (hide_svga) { + mutex_lock(&dev_priv->hw_mutex); + vmw_write(dev_priv, SVGA_REG_ENABLE, + vmw_read(dev_priv, SVGA_REG_ENABLE) | + SVGA_REG_ENABLE_HIDE); + mutex_unlock(&dev_priv->hw_mutex); + } + n3d = (int32_t) dev_priv->num_3d_resources; mutex_unlock(&dev_priv->release_mutex); @@ -252,7 +293,7 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) dev_priv->dev = dev; dev_priv->vmw_chipset = chipset; - dev_priv->last_read_sequence = (uint32_t) -100; + dev_priv->last_read_seqno = (uint32_t) -100; mutex_init(&dev_priv->hw_mutex); mutex_init(&dev_priv->cmdbuf_mutex); mutex_init(&dev_priv->release_mutex); @@ -263,7 +304,7 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) mutex_init(&dev_priv->init_mutex); init_waitqueue_head(&dev_priv->fence_queue); init_waitqueue_head(&dev_priv->fifo_queue); - atomic_set(&dev_priv->fence_queue_waiters, 0); + dev_priv->fence_queue_waiters = 0; atomic_set(&dev_priv->fifo_queue_waiters, 0); dev_priv->io_start = pci_resource_start(dev->pdev, 0); @@ -292,6 +333,12 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) dev_priv->max_gmr_ids = vmw_read(dev_priv, SVGA_REG_GMR_MAX_IDS); } + if (dev_priv->capabilities & SVGA_CAP_GMR2) { + dev_priv->max_gmr_pages = + vmw_read(dev_priv, SVGA_REG_GMRS_MAX_PAGES); + dev_priv->memory_size = + vmw_read(dev_priv, SVGA_REG_MEMORY_SIZE); + } dev_priv->vram_size = vmw_read(dev_priv, SVGA_REG_VRAM_SIZE); dev_priv->mmio_size = vmw_read(dev_priv, SVGA_REG_MEM_SIZE); @@ -308,6 +355,12 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) DRM_INFO("Max GMR descriptors is %u\n", (unsigned)dev_priv->max_gmr_descriptors); } + if (dev_priv->capabilities & SVGA_CAP_GMR2) { + DRM_INFO("Max number of GMR pages is %u\n", + (unsigned)dev_priv->max_gmr_pages); + DRM_INFO("Max dedicated hypervisor graphics memory is %u\n", + (unsigned)dev_priv->memory_size); + } DRM_INFO("VRAM at 0x%08x size is %u kiB\n", dev_priv->vram_start, dev_priv->vram_size / 1024); DRM_INFO("MMIO at 0x%08x size is %u kiB\n", @@ -394,12 +447,16 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset) goto out_no_device; } } + + dev_priv->fman = vmw_fence_manager_init(dev_priv); + if (unlikely(dev_priv->fman == NULL)) + goto out_no_fman; ret = vmw_kms_init(dev_priv); if (unlikely(ret != 0)) goto out_no_kms; vmw_overlay_init(dev_priv); if (dev_priv->enable_fb) { - ret = vmw_3d_resource_inc(dev_priv); + ret = vmw_3d_resource_inc(dev_priv, false); if (unlikely(ret != 0)) goto out_no_fifo; vmw_kms_save_vga(dev_priv); @@ -429,12 +486,14 @@ out_no_irq: if (dev_priv->enable_fb) { vmw_fb_close(dev_priv); vmw_kms_restore_vga(dev_priv); - vmw_3d_resource_dec(dev_priv); + vmw_3d_resource_dec(dev_priv, false); } out_no_fifo: vmw_overlay_close(dev_priv); vmw_kms_close(dev_priv); out_no_kms: + vmw_fence_manager_takedown(dev_priv->fman); +out_no_fman: if (dev_priv->stealth) pci_release_region(dev->pdev, 2); else @@ -467,15 +526,18 @@ static int vmw_driver_unload(struct drm_device *dev) unregister_pm_notifier(&dev_priv->pm_nb); + if (dev_priv->ctx.cmd_bounce) + vfree(dev_priv->ctx.cmd_bounce); if (dev_priv->capabilities & SVGA_CAP_IRQMASK) drm_irq_uninstall(dev_priv->dev); if (dev_priv->enable_fb) { vmw_fb_close(dev_priv); vmw_kms_restore_vga(dev_priv); - vmw_3d_resource_dec(dev_priv); + vmw_3d_resource_dec(dev_priv, false); } vmw_kms_close(dev_priv); vmw_overlay_close(dev_priv); + vmw_fence_manager_takedown(dev_priv->fman); if (dev_priv->stealth) pci_release_region(dev->pdev, 2); else @@ -646,7 +708,7 @@ static int vmw_master_set(struct drm_device *dev, int ret = 0; if (!dev_priv->enable_fb) { - ret = vmw_3d_resource_inc(dev_priv); + ret = vmw_3d_resource_inc(dev_priv, true); if (unlikely(ret != 0)) return ret; vmw_kms_save_vga(dev_priv); @@ -688,7 +750,7 @@ out_no_active_lock: vmw_write(dev_priv, SVGA_REG_TRACES, 1); mutex_unlock(&dev_priv->hw_mutex); vmw_kms_restore_vga(dev_priv); - vmw_3d_resource_dec(dev_priv); + vmw_3d_resource_dec(dev_priv, true); } return ret; } @@ -726,7 +788,7 @@ static void vmw_master_drop(struct drm_device *dev, vmw_write(dev_priv, SVGA_REG_TRACES, 1); mutex_unlock(&dev_priv->hw_mutex); vmw_kms_restore_vga(dev_priv); - vmw_3d_resource_dec(dev_priv); + vmw_3d_resource_dec(dev_priv, true); } dev_priv->active_master = &dev_priv->fbdev_master; @@ -835,7 +897,7 @@ static int vmw_pm_prepare(struct device *kdev) */ dev_priv->suspended = true; if (dev_priv->enable_fb) - vmw_3d_resource_dec(dev_priv); + vmw_3d_resource_dec(dev_priv, true); if (dev_priv->num_3d_resources != 0) { @@ -843,7 +905,7 @@ static int vmw_pm_prepare(struct device *kdev) "while 3D resources are active.\n"); if (dev_priv->enable_fb) - vmw_3d_resource_inc(dev_priv); + vmw_3d_resource_inc(dev_priv, true); dev_priv->suspended = false; return -EBUSY; } @@ -862,7 +924,7 @@ static void vmw_pm_complete(struct device *kdev) * start fifo. */ if (dev_priv->enable_fb) - vmw_3d_resource_inc(dev_priv); + vmw_3d_resource_inc(dev_priv, false); dev_priv->suspended = false; } diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h index 10fc01f69c4..564a8158211 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h @@ -38,20 +38,27 @@ #include "ttm/ttm_lock.h" #include "ttm/ttm_execbuf_util.h" #include "ttm/ttm_module.h" +#include "vmwgfx_fence.h" -#define VMWGFX_DRIVER_DATE "20100927" -#define VMWGFX_DRIVER_MAJOR 1 -#define VMWGFX_DRIVER_MINOR 4 +#define VMWGFX_DRIVER_DATE "20110901" +#define VMWGFX_DRIVER_MAJOR 2 +#define VMWGFX_DRIVER_MINOR 0 #define VMWGFX_DRIVER_PATCHLEVEL 0 #define VMWGFX_FILE_PAGE_OFFSET 0x00100000 #define VMWGFX_FIFO_STATIC_SIZE (1024*1024) #define VMWGFX_MAX_RELOCATIONS 2048 -#define VMWGFX_MAX_GMRS 2048 +#define VMWGFX_MAX_VALIDATIONS 2048 #define VMWGFX_MAX_DISPLAYS 16 +#define VMWGFX_CMD_BOUNCE_INIT_SIZE 32768 #define VMW_PL_GMR TTM_PL_PRIV0 #define VMW_PL_FLAG_GMR TTM_PL_FLAG_PRIV0 +#define VMW_RES_CONTEXT ttm_driver_type0 +#define VMW_RES_SURFACE ttm_driver_type1 +#define VMW_RES_STREAM ttm_driver_type2 +#define VMW_RES_FENCE ttm_driver_type3 + struct vmw_fpriv { struct drm_master *locked_master; struct ttm_object_file *tfile; @@ -74,7 +81,7 @@ struct vmw_resource { bool avail; void (*hw_destroy) (struct vmw_resource *res); void (*res_free) (struct vmw_resource *res); - + bool on_validate_list; /* TODO is a generic snooper needed? */ #if 0 void (*snoop)(struct vmw_resource *res, @@ -104,7 +111,7 @@ struct vmw_surface { struct vmw_cursor_snooper snooper; }; -struct vmw_fence_queue { +struct vmw_marker_queue { struct list_head head; struct timespec lag; struct timespec lag_time; @@ -115,16 +122,12 @@ struct vmw_fifo_state { unsigned long reserved_size; __le32 *dynamic_buffer; __le32 *static_buffer; - __le32 *last_buffer; - uint32_t last_data_size; - uint32_t last_buffer_size; - bool last_buffer_add; unsigned long static_buffer_size; bool using_bounce_buffer; uint32_t capabilities; struct mutex fifo_mutex; struct rw_semaphore rwsem; - struct vmw_fence_queue fence_queue; + struct vmw_marker_queue marker_queue; }; struct vmw_relocation { @@ -143,8 +146,12 @@ struct vmw_sw_context{ struct list_head validate_nodes; struct vmw_relocation relocs[VMWGFX_MAX_RELOCATIONS]; uint32_t cur_reloc; - struct ttm_validate_buffer val_bufs[VMWGFX_MAX_GMRS]; + struct ttm_validate_buffer val_bufs[VMWGFX_MAX_VALIDATIONS]; uint32_t cur_val_buf; + uint32_t *cmd_bounce; + uint32_t cmd_bounce_size; + struct vmw_resource *resources[VMWGFX_MAX_VALIDATIONS]; + uint32_t num_ref_resources; }; struct vmw_legacy_display; @@ -185,6 +192,8 @@ struct vmw_private { uint32_t capabilities; uint32_t max_gmr_descriptors; uint32_t max_gmr_ids; + uint32_t max_gmr_pages; + uint32_t memory_size; bool has_gmr; struct mutex hw_mutex; @@ -195,12 +204,7 @@ struct vmw_private { struct vmw_vga_topology_state vga_save[VMWGFX_MAX_DISPLAYS]; uint32_t vga_width; uint32_t vga_height; - uint32_t vga_depth; uint32_t vga_bpp; - uint32_t vga_pseudo; - uint32_t vga_red_mask; - uint32_t vga_green_mask; - uint32_t vga_blue_mask; uint32_t vga_bpl; uint32_t vga_pitchlock; @@ -240,13 +244,14 @@ struct vmw_private { * Fencing and IRQs. */ - atomic_t fence_seq; + atomic_t marker_seq; wait_queue_head_t fence_queue; wait_queue_head_t fifo_queue; - atomic_t fence_queue_waiters; + int fence_queue_waiters; /* Protected by hw_mutex */ atomic_t fifo_queue_waiters; - uint32_t last_read_sequence; + uint32_t last_read_seqno; spinlock_t irq_lock; + struct vmw_fence_manager *fman; /* * Device state @@ -319,8 +324,8 @@ static inline uint32_t vmw_read(struct vmw_private *dev_priv, return val; } -int vmw_3d_resource_inc(struct vmw_private *dev_priv); -void vmw_3d_resource_dec(struct vmw_private *dev_priv); +int vmw_3d_resource_inc(struct vmw_private *dev_priv, bool unhide_svga); +void vmw_3d_resource_dec(struct vmw_private *dev_priv, bool hide_svga); /** * GMR utilities - vmwgfx_gmr.c @@ -345,7 +350,8 @@ extern int vmw_context_define_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); extern int vmw_context_check(struct vmw_private *dev_priv, struct ttm_object_file *tfile, - int id); + int id, + struct vmw_resource **p_res); extern void vmw_surface_res_free(struct vmw_resource *res); extern int vmw_surface_init(struct vmw_private *dev_priv, struct vmw_surface *srf, @@ -398,7 +404,7 @@ extern int vmw_user_stream_lookup(struct vmw_private *dev_priv, extern int vmw_getparam_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); -extern int vmw_fifo_debug_ioctl(struct drm_device *dev, void *data, +extern int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); /** @@ -412,9 +418,8 @@ extern void vmw_fifo_release(struct vmw_private *dev_priv, extern void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes); extern void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes); extern int vmw_fifo_send_fence(struct vmw_private *dev_priv, - uint32_t *sequence); + uint32_t *seqno); extern void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason); -extern int vmw_fifo_mmap(struct file *filp, struct vm_area_struct *vma); extern bool vmw_fifo_have_3d(struct vmw_private *dev_priv); extern bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv); @@ -450,39 +455,38 @@ extern int vmw_execbuf_ioctl(struct drm_device *dev, void *data, */ extern irqreturn_t vmw_irq_handler(DRM_IRQ_ARGS); -extern int vmw_wait_fence(struct vmw_private *dev_priv, bool lazy, - uint32_t sequence, bool interruptible, - unsigned long timeout); +extern int vmw_wait_seqno(struct vmw_private *dev_priv, bool lazy, + uint32_t seqno, bool interruptible, + unsigned long timeout); extern void vmw_irq_preinstall(struct drm_device *dev); extern int vmw_irq_postinstall(struct drm_device *dev); extern void vmw_irq_uninstall(struct drm_device *dev); -extern bool vmw_fence_signaled(struct vmw_private *dev_priv, - uint32_t sequence); -extern int vmw_fence_wait_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv); +extern bool vmw_seqno_passed(struct vmw_private *dev_priv, + uint32_t seqno); extern int vmw_fallback_wait(struct vmw_private *dev_priv, bool lazy, bool fifo_idle, - uint32_t sequence, + uint32_t seqno, bool interruptible, unsigned long timeout); -extern void vmw_update_sequence(struct vmw_private *dev_priv, +extern void vmw_update_seqno(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo_state); - +extern void vmw_seqno_waiter_add(struct vmw_private *dev_priv); +extern void vmw_seqno_waiter_remove(struct vmw_private *dev_priv); /** - * Rudimentary fence objects currently used only for throttling - - * vmwgfx_fence.c + * Rudimentary fence-like objects currently used only for throttling - + * vmwgfx_marker.c */ -extern void vmw_fence_queue_init(struct vmw_fence_queue *queue); -extern void vmw_fence_queue_takedown(struct vmw_fence_queue *queue); -extern int vmw_fence_push(struct vmw_fence_queue *queue, - uint32_t sequence); -extern int vmw_fence_pull(struct vmw_fence_queue *queue, - uint32_t signaled_sequence); +extern void vmw_marker_queue_init(struct vmw_marker_queue *queue); +extern void vmw_marker_queue_takedown(struct vmw_marker_queue *queue); +extern int vmw_marker_push(struct vmw_marker_queue *queue, + uint32_t seqno); +extern int vmw_marker_pull(struct vmw_marker_queue *queue, + uint32_t signaled_seqno); extern int vmw_wait_lag(struct vmw_private *dev_priv, - struct vmw_fence_queue *queue, uint32_t us); + struct vmw_marker_queue *queue, uint32_t us); /** * Kernel framebuffer - vmwgfx_fb.c @@ -508,11 +512,9 @@ void vmw_kms_cursor_snoop(struct vmw_surface *srf, struct ttm_object_file *tfile, struct ttm_buffer_object *bo, SVGA3dCmdHeader *header); -void vmw_kms_write_svga(struct vmw_private *vmw_priv, - unsigned width, unsigned height, unsigned pitch, - unsigned bbp, unsigned depth); -int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv); +int vmw_kms_write_svga(struct vmw_private *vmw_priv, + unsigned width, unsigned height, unsigned pitch, + unsigned bpp, unsigned depth); void vmw_kms_idle_workqueues(struct vmw_master *vmaster); bool vmw_kms_validate_mode_vram(struct vmw_private *dev_priv, uint32_t pitch, @@ -576,4 +578,8 @@ static inline struct vmw_dma_buffer *vmw_dmabuf_reference(struct vmw_dma_buffer return NULL; } +static inline struct ttm_mem_global *vmw_mem_glob(struct vmw_private *dev_priv) +{ + return (struct ttm_mem_global *) dev_priv->mem_global_ref.object; +} #endif diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c index 41b95ed6dbc..fa26e647f48 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c @@ -44,10 +44,36 @@ static int vmw_cmd_ok(struct vmw_private *dev_priv, return 0; } + +static int vmw_resource_to_validate_list(struct vmw_sw_context *sw_context, + struct vmw_resource **p_res) +{ + int ret = 0; + struct vmw_resource *res = *p_res; + + if (!res->on_validate_list) { + if (sw_context->num_ref_resources >= VMWGFX_MAX_VALIDATIONS) { + DRM_ERROR("Too many resources referenced in " + "command stream.\n"); + ret = -ENOMEM; + goto out; + } + sw_context->resources[sw_context->num_ref_resources++] = res; + res->on_validate_list = true; + return 0; + } + +out: + vmw_resource_unreference(p_res); + return ret; +} + static int vmw_cmd_cid_check(struct vmw_private *dev_priv, struct vmw_sw_context *sw_context, SVGA3dCmdHeader *header) { + struct vmw_resource *ctx; + struct vmw_cid_cmd { SVGA3dCmdHeader header; __le32 cid; @@ -58,7 +84,8 @@ static int vmw_cmd_cid_check(struct vmw_private *dev_priv, if (likely(sw_context->cid_valid && cmd->cid == sw_context->last_cid)) return 0; - ret = vmw_context_check(dev_priv, sw_context->tfile, cmd->cid); + ret = vmw_context_check(dev_priv, sw_context->tfile, cmd->cid, + &ctx); if (unlikely(ret != 0)) { DRM_ERROR("Could not find or use context %u\n", (unsigned) cmd->cid); @@ -67,39 +94,43 @@ static int vmw_cmd_cid_check(struct vmw_private *dev_priv, sw_context->last_cid = cmd->cid; sw_context->cid_valid = true; - - return 0; + return vmw_resource_to_validate_list(sw_context, &ctx); } static int vmw_cmd_sid_check(struct vmw_private *dev_priv, struct vmw_sw_context *sw_context, uint32_t *sid) { + struct vmw_surface *srf; + int ret; + struct vmw_resource *res; + if (*sid == SVGA3D_INVALID_ID) return 0; - if (unlikely((!sw_context->sid_valid || - *sid != sw_context->last_sid))) { - int real_id; - int ret = vmw_surface_check(dev_priv, sw_context->tfile, - *sid, &real_id); - - if (unlikely(ret != 0)) { - DRM_ERROR("Could ot find or use surface 0x%08x " - "address 0x%08lx\n", - (unsigned int) *sid, - (unsigned long) sid); - return ret; - } - - sw_context->last_sid = *sid; - sw_context->sid_valid = true; - *sid = real_id; - sw_context->sid_translation = real_id; - } else + if (likely((sw_context->sid_valid && + *sid == sw_context->last_sid))) { *sid = sw_context->sid_translation; + return 0; + } - return 0; + ret = vmw_user_surface_lookup_handle(dev_priv, sw_context->tfile, + *sid, &srf); + if (unlikely(ret != 0)) { + DRM_ERROR("Could ot find or use surface 0x%08x " + "address 0x%08lx\n", + (unsigned int) *sid, + (unsigned long) sid); + return ret; + } + + sw_context->last_sid = *sid; + sw_context->sid_valid = true; + sw_context->sid_translation = srf->res.id; + *sid = sw_context->sid_translation; + + res = &srf->res; + return vmw_resource_to_validate_list(sw_context, &res); } @@ -213,7 +244,7 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv, reloc->location = ptr; cur_validate_node = vmw_dmabuf_validate_node(bo, sw_context->cur_val_buf); - if (unlikely(cur_validate_node >= VMWGFX_MAX_GMRS)) { + if (unlikely(cur_validate_node >= VMWGFX_MAX_VALIDATIONS)) { DRM_ERROR("Max number of DMA buffers per submission" " exceeded.\n"); ret = -EINVAL; @@ -224,7 +255,8 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv, if (unlikely(cur_validate_node == sw_context->cur_val_buf)) { val_buf = &sw_context->val_bufs[cur_validate_node]; val_buf->bo = ttm_bo_reference(bo); - val_buf->new_sync_obj_arg = (void *) dev_priv; + val_buf->usage = TTM_USAGE_READWRITE; + val_buf->new_sync_obj_arg = (void *) DRM_VMW_FENCE_FLAG_EXEC; list_add_tail(&val_buf->head, &sw_context->validate_nodes); ++sw_context->cur_val_buf; } @@ -289,7 +321,6 @@ static int vmw_cmd_wait_query(struct vmw_private *dev_priv, return 0; } - static int vmw_cmd_dma(struct vmw_private *dev_priv, struct vmw_sw_context *sw_context, SVGA3dCmdHeader *header) @@ -302,6 +333,7 @@ static int vmw_cmd_dma(struct vmw_private *dev_priv, SVGA3dCmdSurfaceDMA dma; } *cmd; int ret; + struct vmw_resource *res; cmd = container_of(header, struct vmw_dma_cmd, header); ret = vmw_translate_guest_ptr(dev_priv, sw_context, @@ -318,17 +350,16 @@ static int vmw_cmd_dma(struct vmw_private *dev_priv, goto out_no_reloc; } - /** + /* * Patch command stream with device SID. */ - cmd->dma.host.sid = srf->res.id; vmw_kms_cursor_snoop(srf, sw_context->tfile, bo, header); - /** - * FIXME: May deadlock here when called from the - * command parsing code. - */ - vmw_surface_unreference(&srf); + + vmw_dmabuf_unreference(&vmw_bo); + + res = &srf->res; + return vmw_resource_to_validate_list(sw_context, &res); out_no_reloc: vmw_dmabuf_unreference(&vmw_bo); @@ -500,8 +531,9 @@ out_err: static int vmw_cmd_check_all(struct vmw_private *dev_priv, struct vmw_sw_context *sw_context, - void *buf, uint32_t size) + uint32_t size) { + void *buf = sw_context->cmd_bounce; int32_t cur_size = size; int ret; @@ -550,7 +582,11 @@ static void vmw_apply_relocations(struct vmw_sw_context *sw_context) static void vmw_clear_validations(struct vmw_sw_context *sw_context) { struct ttm_validate_buffer *entry, *next; + uint32_t i = sw_context->num_ref_resources; + /* + * Drop references to DMA buffers held during command submission. + */ list_for_each_entry_safe(entry, next, &sw_context->validate_nodes, head) { list_del(&entry->head); @@ -559,6 +595,14 @@ static void vmw_clear_validations(struct vmw_sw_context *sw_context) sw_context->cur_val_buf--; } BUG_ON(sw_context->cur_val_buf != 0); + + /* + * Drop references to resources held during command submission. + */ + while (i-- > 0) { + sw_context->resources[i]->on_validate_list = false; + vmw_resource_unreference(&sw_context->resources[i]); + } } static int vmw_validate_single_buffer(struct vmw_private *dev_priv, @@ -602,6 +646,79 @@ static int vmw_validate_buffers(struct vmw_private *dev_priv, return 0; } +static int vmw_resize_cmd_bounce(struct vmw_sw_context *sw_context, + uint32_t size) +{ + if (likely(sw_context->cmd_bounce_size >= size)) + return 0; + + if (sw_context->cmd_bounce_size == 0) + sw_context->cmd_bounce_size = VMWGFX_CMD_BOUNCE_INIT_SIZE; + + while (sw_context->cmd_bounce_size < size) { + sw_context->cmd_bounce_size = + PAGE_ALIGN(sw_context->cmd_bounce_size + + (sw_context->cmd_bounce_size >> 1)); + } + + if (sw_context->cmd_bounce != NULL) + vfree(sw_context->cmd_bounce); + + sw_context->cmd_bounce = vmalloc(sw_context->cmd_bounce_size); + + if (sw_context->cmd_bounce == NULL) { + DRM_ERROR("Failed to allocate command bounce buffer.\n"); + sw_context->cmd_bounce_size = 0; + return -ENOMEM; + } + + return 0; +} + +/** + * vmw_execbuf_fence_commands - create and submit a command stream fence + * + * Creates a fence object and submits a command stream marker. + * If this fails for some reason, We sync the fifo and return NULL. + * It is then safe to fence buffers with a NULL pointer. + */ + +int vmw_execbuf_fence_commands(struct drm_file *file_priv, + struct vmw_private *dev_priv, + struct vmw_fence_obj **p_fence, + uint32_t *p_handle) +{ + uint32_t sequence; + int ret; + bool synced = false; + + + ret = vmw_fifo_send_fence(dev_priv, &sequence); + if (unlikely(ret != 0)) { + DRM_ERROR("Fence submission error. Syncing.\n"); + synced = true; + } + + if (p_handle != NULL) + ret = vmw_user_fence_create(file_priv, dev_priv->fman, + sequence, + DRM_VMW_FENCE_FLAG_EXEC, + p_fence, p_handle); + else + ret = vmw_fence_create(dev_priv->fman, sequence, + DRM_VMW_FENCE_FLAG_EXEC, + p_fence); + + if (unlikely(ret != 0 && !synced)) { + (void) vmw_fallback_wait(dev_priv, false, false, + sequence, false, + VMW_FENCE_WAIT_TIMEOUT); + *p_fence = NULL; + } + + return 0; +} + int vmw_execbuf_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { @@ -612,9 +729,24 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data, int ret; void *user_cmd; void *cmd; - uint32_t sequence; struct vmw_sw_context *sw_context = &dev_priv->ctx; struct vmw_master *vmaster = vmw_master(file_priv->master); + struct vmw_fence_obj *fence; + uint32_t handle; + + /* + * This will allow us to extend the ioctl argument while + * maintaining backwards compatibility: + * We take different code paths depending on the value of + * arg->version. + */ + + if (unlikely(arg->version != DRM_VMW_EXECBUF_VERSION)) { + DRM_ERROR("Incorrect execbuf version.\n"); + DRM_ERROR("You're running outdated experimental " + "vmwgfx user-space drivers."); + return -EINVAL; + } ret = ttm_read_lock(&vmaster->lock, true); if (unlikely(ret != 0)) @@ -626,20 +758,18 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data, goto out_no_cmd_mutex; } - cmd = vmw_fifo_reserve(dev_priv, arg->command_size); - if (unlikely(cmd == NULL)) { - DRM_ERROR("Failed reserving fifo space for commands.\n"); - ret = -ENOMEM; + ret = vmw_resize_cmd_bounce(sw_context, arg->command_size); + if (unlikely(ret != 0)) goto out_unlock; - } user_cmd = (void __user *)(unsigned long)arg->commands; - ret = copy_from_user(cmd, user_cmd, arg->command_size); + ret = copy_from_user(sw_context->cmd_bounce, + user_cmd, arg->command_size); if (unlikely(ret != 0)) { ret = -EFAULT; DRM_ERROR("Failed copying commands.\n"); - goto out_commit; + goto out_unlock; } sw_context->tfile = vmw_fpriv(file_priv)->tfile; @@ -647,12 +777,14 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data, sw_context->sid_valid = false; sw_context->cur_reloc = 0; sw_context->cur_val_buf = 0; + sw_context->num_ref_resources = 0; INIT_LIST_HEAD(&sw_context->validate_nodes); - ret = vmw_cmd_check_all(dev_priv, sw_context, cmd, arg->command_size); + ret = vmw_cmd_check_all(dev_priv, sw_context, arg->command_size); if (unlikely(ret != 0)) goto out_err; + ret = ttm_eu_reserve_buffers(&sw_context->validate_nodes); if (unlikely(ret != 0)) goto out_err; @@ -664,53 +796,86 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data, vmw_apply_relocations(sw_context); if (arg->throttle_us) { - ret = vmw_wait_lag(dev_priv, &dev_priv->fifo.fence_queue, + ret = vmw_wait_lag(dev_priv, &dev_priv->fifo.marker_queue, arg->throttle_us); if (unlikely(ret != 0)) - goto out_err; + goto out_throttle; } - vmw_fifo_commit(dev_priv, arg->command_size); - - ret = vmw_fifo_send_fence(dev_priv, &sequence); + cmd = vmw_fifo_reserve(dev_priv, arg->command_size); + if (unlikely(cmd == NULL)) { + DRM_ERROR("Failed reserving fifo space for commands.\n"); + ret = -ENOMEM; + goto out_err; + } - ttm_eu_fence_buffer_objects(&sw_context->validate_nodes, - (void *)(unsigned long) sequence); - vmw_clear_validations(sw_context); - mutex_unlock(&dev_priv->cmdbuf_mutex); + memcpy(cmd, sw_context->cmd_bounce, arg->command_size); + vmw_fifo_commit(dev_priv, arg->command_size); + user_fence_rep = (struct drm_vmw_fence_rep __user *) + (unsigned long)arg->fence_rep; + ret = vmw_execbuf_fence_commands(file_priv, dev_priv, + &fence, + (user_fence_rep) ? &handle : NULL); /* * This error is harmless, because if fence submission fails, - * vmw_fifo_send_fence will sync. + * vmw_fifo_send_fence will sync. The error will be propagated to + * user-space in @fence_rep */ if (ret != 0) DRM_ERROR("Fence submission error. Syncing.\n"); - fence_rep.error = ret; - fence_rep.fence_seq = (uint64_t) sequence; - fence_rep.pad64 = 0; + ttm_eu_fence_buffer_objects(&sw_context->validate_nodes, + (void *) fence); - user_fence_rep = (struct drm_vmw_fence_rep __user *) - (unsigned long)arg->fence_rep; + vmw_clear_validations(sw_context); + mutex_unlock(&dev_priv->cmdbuf_mutex); - /* - * copy_to_user errors will be detected by user space not - * seeing fence_rep::error filled in. - */ + if (user_fence_rep) { + fence_rep.error = ret; + fence_rep.handle = handle; + fence_rep.seqno = fence->seqno; + vmw_update_seqno(dev_priv, &dev_priv->fifo); + fence_rep.passed_seqno = dev_priv->last_read_seqno; + + /* + * copy_to_user errors will be detected by user space not + * seeing fence_rep::error filled in. Typically + * user-space would have pre-set that member to -EFAULT. + */ + ret = copy_to_user(user_fence_rep, &fence_rep, + sizeof(fence_rep)); + + /* + * User-space lost the fence object. We need to sync + * and unreference the handle. + */ + if (unlikely(ret != 0) && (fence_rep.error == 0)) { + BUG_ON(fence == NULL); + + ttm_ref_object_base_unref(vmw_fpriv(file_priv)->tfile, + handle, TTM_REF_USAGE); + DRM_ERROR("Fence copy error. Syncing.\n"); + (void) vmw_fence_obj_wait(fence, + fence->signal_mask, + false, false, + VMW_FENCE_WAIT_TIMEOUT); + } + } - ret = copy_to_user(user_fence_rep, &fence_rep, sizeof(fence_rep)); + if (likely(fence != NULL)) + vmw_fence_obj_unreference(&fence); vmw_kms_cursor_post_execbuf(dev_priv); ttm_read_unlock(&vmaster->lock); return 0; out_err: vmw_free_relocations(sw_context); +out_throttle: ttm_eu_backoff_reservation(&sw_context->validate_nodes); vmw_clear_validations(sw_context); -out_commit: - vmw_fifo_commit(dev_priv, 0); out_unlock: mutex_unlock(&dev_priv->cmdbuf_mutex); out_no_cmd_mutex: diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c index bfab60c938a..b1888e801e2 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c @@ -158,10 +158,14 @@ static int vmw_fb_set_par(struct fb_info *info) { struct vmw_fb_par *par = info->par; struct vmw_private *vmw_priv = par->vmw_priv; + int ret; + + ret = vmw_kms_write_svga(vmw_priv, info->var.xres, info->var.yres, + info->fix.line_length, + par->bpp, par->depth); + if (ret) + return ret; - vmw_kms_write_svga(vmw_priv, info->var.xres, info->var.yres, - info->fix.line_length, - par->bpp, par->depth); if (vmw_priv->capabilities & SVGA_CAP_DISPLAY_TOPOLOGY) { /* TODO check if pitch and offset changes */ vmw_write(vmw_priv, SVGA_REG_NUM_GUEST_DISPLAYS, 1); @@ -405,14 +409,14 @@ int vmw_fb_init(struct vmw_private *vmw_priv) struct fb_info *info; unsigned initial_width, initial_height; unsigned fb_width, fb_height; - unsigned fb_bbp, fb_depth, fb_offset, fb_pitch, fb_size; + unsigned fb_bpp, fb_depth, fb_offset, fb_pitch, fb_size; int ret; /* XXX These shouldn't be hardcoded. */ initial_width = 800; initial_height = 600; - fb_bbp = 32; + fb_bpp = 32; fb_depth = 24; /* XXX As shouldn't these be as well. */ @@ -422,7 +426,7 @@ int vmw_fb_init(struct vmw_private *vmw_priv) initial_width = min(fb_width, initial_width); initial_height = min(fb_height, initial_height); - fb_pitch = fb_width * fb_bbp / 8; + fb_pitch = fb_width * fb_bpp / 8; fb_size = fb_pitch * fb_height; fb_offset = vmw_read(vmw_priv, SVGA_REG_FB_OFFSET); @@ -437,7 +441,7 @@ int vmw_fb_init(struct vmw_private *vmw_priv) par = info->par; par->vmw_priv = vmw_priv; par->depth = fb_depth; - par->bpp = fb_bbp; + par->bpp = fb_bpp; par->vmalloc = NULL; par->max_width = fb_width; par->max_height = fb_height; diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c index 61eacc1b5ca..5065a140fdf 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c @@ -1,6 +1,6 @@ /************************************************************************** * - * Copyright (C) 2010 VMware, Inc., Palo Alto, CA., USA + * Copyright © 2011 VMware, Inc., Palo Alto, CA., USA * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a @@ -25,149 +25,595 @@ * **************************************************************************/ - +#include "drmP.h" #include "vmwgfx_drv.h" -struct vmw_fence { - struct list_head head; - uint32_t sequence; - struct timespec submitted; +#define VMW_FENCE_WRAP (1 << 31) + +struct vmw_fence_manager { + int num_fence_objects; + struct vmw_private *dev_priv; + spinlock_t lock; + u32 next_seqno; + struct list_head fence_list; + struct work_struct work; + u32 user_fence_size; + u32 fence_size; + bool fifo_down; + struct list_head cleanup_list; }; -void vmw_fence_queue_init(struct vmw_fence_queue *queue) +struct vmw_user_fence { + struct ttm_base_object base; + struct vmw_fence_obj fence; +}; + +/** + * vmw_fence_destroy_locked + * + */ + +static void vmw_fence_obj_destroy_locked(struct kref *kref) { - INIT_LIST_HEAD(&queue->head); - queue->lag = ns_to_timespec(0); - getrawmonotonic(&queue->lag_time); - spin_lock_init(&queue->lock); + struct vmw_fence_obj *fence = + container_of(kref, struct vmw_fence_obj, kref); + + struct vmw_fence_manager *fman = fence->fman; + unsigned int num_fences; + + list_del_init(&fence->head); + num_fences = --fman->num_fence_objects; + spin_unlock_irq(&fman->lock); + if (fence->destroy) + fence->destroy(fence); + else + kfree(fence); + + spin_lock_irq(&fman->lock); } -void vmw_fence_queue_takedown(struct vmw_fence_queue *queue) + +/** + * Execute signal actions on fences recently signaled. + * This is done from a workqueue so we don't have to execute + * signal actions from atomic context. + */ + +static void vmw_fence_work_func(struct work_struct *work) { - struct vmw_fence *fence, *next; + struct vmw_fence_manager *fman = + container_of(work, struct vmw_fence_manager, work); + struct list_head list; + struct vmw_fence_action *action, *next_action; - spin_lock(&queue->lock); - list_for_each_entry_safe(fence, next, &queue->head, head) { - kfree(fence); - } - spin_unlock(&queue->lock); + do { + INIT_LIST_HEAD(&list); + spin_lock_irq(&fman->lock); + list_splice_init(&fman->cleanup_list, &list); + spin_unlock_irq(&fman->lock); + + if (list_empty(&list)) + return; + + /* + * At this point, only we should be able to manipulate the + * list heads of the actions we have on the private list. + */ + + list_for_each_entry_safe(action, next_action, &list, head) { + list_del_init(&action->head); + action->cleanup(action); + } + } while (1); } -int vmw_fence_push(struct vmw_fence_queue *queue, - uint32_t sequence) +struct vmw_fence_manager *vmw_fence_manager_init(struct vmw_private *dev_priv) { - struct vmw_fence *fence = kmalloc(sizeof(*fence), GFP_KERNEL); + struct vmw_fence_manager *fman = kzalloc(sizeof(*fman), GFP_KERNEL); - if (unlikely(!fence)) - return -ENOMEM; + if (unlikely(fman == NULL)) + return NULL; - fence->sequence = sequence; - getrawmonotonic(&fence->submitted); - spin_lock(&queue->lock); - list_add_tail(&fence->head, &queue->head); - spin_unlock(&queue->lock); + fman->dev_priv = dev_priv; + spin_lock_init(&fman->lock); + INIT_LIST_HEAD(&fman->fence_list); + INIT_LIST_HEAD(&fman->cleanup_list); + INIT_WORK(&fman->work, &vmw_fence_work_func); + fman->fifo_down = true; + fman->user_fence_size = ttm_round_pot(sizeof(struct vmw_user_fence)); + fman->fence_size = ttm_round_pot(sizeof(struct vmw_fence_obj)); - return 0; + return fman; } -int vmw_fence_pull(struct vmw_fence_queue *queue, - uint32_t signaled_sequence) +void vmw_fence_manager_takedown(struct vmw_fence_manager *fman) { - struct vmw_fence *fence, *next; - struct timespec now; - bool updated = false; + unsigned long irq_flags; + bool lists_empty; + + (void) cancel_work_sync(&fman->work); - spin_lock(&queue->lock); - getrawmonotonic(&now); + spin_lock_irqsave(&fman->lock, irq_flags); + lists_empty = list_empty(&fman->fence_list) && + list_empty(&fman->cleanup_list); + spin_unlock_irqrestore(&fman->lock, irq_flags); - if (list_empty(&queue->head)) { - queue->lag = ns_to_timespec(0); - queue->lag_time = now; - updated = true; + BUG_ON(!lists_empty); + kfree(fman); +} + +static int vmw_fence_obj_init(struct vmw_fence_manager *fman, + struct vmw_fence_obj *fence, + u32 seqno, + uint32_t mask, + void (*destroy) (struct vmw_fence_obj *fence)) +{ + unsigned long irq_flags; + unsigned int num_fences; + int ret = 0; + + fence->seqno = seqno; + INIT_LIST_HEAD(&fence->seq_passed_actions); + fence->fman = fman; + fence->signaled = 0; + fence->signal_mask = mask; + kref_init(&fence->kref); + fence->destroy = destroy; + init_waitqueue_head(&fence->queue); + + spin_lock_irqsave(&fman->lock, irq_flags); + if (unlikely(fman->fifo_down)) { + ret = -EBUSY; goto out_unlock; } + list_add_tail(&fence->head, &fman->fence_list); + num_fences = ++fman->num_fence_objects; - list_for_each_entry_safe(fence, next, &queue->head, head) { - if (signaled_sequence - fence->sequence > (1 << 30)) - continue; +out_unlock: + spin_unlock_irqrestore(&fman->lock, irq_flags); + return ret; - queue->lag = timespec_sub(now, fence->submitted); - queue->lag_time = now; - updated = true; - list_del(&fence->head); - kfree(fence); +} + +struct vmw_fence_obj *vmw_fence_obj_reference(struct vmw_fence_obj *fence) +{ + kref_get(&fence->kref); + return fence; +} + +/** + * vmw_fence_obj_unreference + * + * Note that this function may not be entered with disabled irqs since + * it may re-enable them in the destroy function. + * + */ +void vmw_fence_obj_unreference(struct vmw_fence_obj **fence_p) +{ + struct vmw_fence_obj *fence = *fence_p; + struct vmw_fence_manager *fman = fence->fman; + + *fence_p = NULL; + spin_lock_irq(&fman->lock); + BUG_ON(atomic_read(&fence->kref.refcount) == 0); + kref_put(&fence->kref, vmw_fence_obj_destroy_locked); + spin_unlock_irq(&fman->lock); +} + +void vmw_fences_perform_actions(struct vmw_fence_manager *fman, + struct list_head *list) +{ + struct vmw_fence_action *action, *next_action; + + list_for_each_entry_safe(action, next_action, list, head) { + list_del_init(&action->head); + if (action->seq_passed != NULL) + action->seq_passed(action); + + /* + * Add the cleanup action to the cleanup list so that + * it will be performed by a worker task. + */ + + if (action->cleanup != NULL) + list_add_tail(&action->head, &fman->cleanup_list); } +} -out_unlock: - spin_unlock(&queue->lock); +void vmw_fences_update(struct vmw_fence_manager *fman, u32 seqno) +{ + unsigned long flags; + struct vmw_fence_obj *fence, *next_fence; + struct list_head action_list; + + spin_lock_irqsave(&fman->lock, flags); + list_for_each_entry_safe(fence, next_fence, &fman->fence_list, head) { + if (seqno - fence->seqno < VMW_FENCE_WRAP) { + list_del_init(&fence->head); + fence->signaled |= DRM_VMW_FENCE_FLAG_EXEC; + INIT_LIST_HEAD(&action_list); + list_splice_init(&fence->seq_passed_actions, + &action_list); + vmw_fences_perform_actions(fman, &action_list); + wake_up_all(&fence->queue); + } - return (updated) ? 0 : -EBUSY; + } + if (!list_empty(&fman->cleanup_list)) + (void) schedule_work(&fman->work); + spin_unlock_irqrestore(&fman->lock, flags); } -static struct timespec vmw_timespec_add(struct timespec t1, - struct timespec t2) + +bool vmw_fence_obj_signaled(struct vmw_fence_obj *fence, + uint32_t flags) { - t1.tv_sec += t2.tv_sec; - t1.tv_nsec += t2.tv_nsec; - if (t1.tv_nsec >= 1000000000L) { - t1.tv_sec += 1; - t1.tv_nsec -= 1000000000L; + struct vmw_fence_manager *fman = fence->fman; + unsigned long irq_flags; + uint32_t signaled; + + spin_lock_irqsave(&fman->lock, irq_flags); + signaled = fence->signaled; + spin_unlock_irqrestore(&fman->lock, irq_flags); + + flags &= fence->signal_mask; + if ((signaled & flags) == flags) + return 1; + + if ((signaled & DRM_VMW_FENCE_FLAG_EXEC) == 0) { + struct vmw_private *dev_priv = fman->dev_priv; + __le32 __iomem *fifo_mem = dev_priv->mmio_virt; + u32 seqno; + + seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE); + vmw_fences_update(fman, seqno); } - return t1; + spin_lock_irqsave(&fman->lock, irq_flags); + signaled = fence->signaled; + spin_unlock_irqrestore(&fman->lock, irq_flags); + + return ((signaled & flags) == flags); } -static struct timespec vmw_fifo_lag(struct vmw_fence_queue *queue) +int vmw_fence_obj_wait(struct vmw_fence_obj *fence, + uint32_t flags, bool lazy, + bool interruptible, unsigned long timeout) { - struct timespec now; + struct vmw_private *dev_priv = fence->fman->dev_priv; + long ret; + + if (likely(vmw_fence_obj_signaled(fence, flags))) + return 0; + + vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC); + vmw_seqno_waiter_add(dev_priv); + + if (interruptible) + ret = wait_event_interruptible_timeout + (fence->queue, + vmw_fence_obj_signaled(fence, flags), + timeout); + else + ret = wait_event_timeout + (fence->queue, + vmw_fence_obj_signaled(fence, flags), + timeout); + + vmw_seqno_waiter_remove(dev_priv); - spin_lock(&queue->lock); - getrawmonotonic(&now); - queue->lag = vmw_timespec_add(queue->lag, - timespec_sub(now, queue->lag_time)); - queue->lag_time = now; - spin_unlock(&queue->lock); - return queue->lag; + if (unlikely(ret == 0)) + ret = -EBUSY; + else if (likely(ret > 0)) + ret = 0; + + return ret; } +void vmw_fence_obj_flush(struct vmw_fence_obj *fence) +{ + struct vmw_private *dev_priv = fence->fman->dev_priv; + + vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC); +} -static bool vmw_lag_lt(struct vmw_fence_queue *queue, - uint32_t us) +static void vmw_fence_destroy(struct vmw_fence_obj *fence) { - struct timespec lag, cond; + struct vmw_fence_manager *fman = fence->fman; - cond = ns_to_timespec((s64) us * 1000); - lag = vmw_fifo_lag(queue); - return (timespec_compare(&lag, &cond) < 1); + kfree(fence); + /* + * Free kernel space accounting. + */ + ttm_mem_global_free(vmw_mem_glob(fman->dev_priv), + fman->fence_size); } -int vmw_wait_lag(struct vmw_private *dev_priv, - struct vmw_fence_queue *queue, uint32_t us) +int vmw_fence_create(struct vmw_fence_manager *fman, + uint32_t seqno, + uint32_t mask, + struct vmw_fence_obj **p_fence) { - struct vmw_fence *fence; - uint32_t sequence; + struct ttm_mem_global *mem_glob = vmw_mem_glob(fman->dev_priv); + struct vmw_fence_obj *fence; int ret; - while (!vmw_lag_lt(queue, us)) { - spin_lock(&queue->lock); - if (list_empty(&queue->head)) - sequence = atomic_read(&dev_priv->fence_seq); - else { - fence = list_first_entry(&queue->head, - struct vmw_fence, head); - sequence = fence->sequence; + ret = ttm_mem_global_alloc(mem_glob, fman->fence_size, + false, false); + if (unlikely(ret != 0)) + return ret; + + fence = kzalloc(sizeof(*fence), GFP_KERNEL); + if (unlikely(fence == NULL)) { + ret = -ENOMEM; + goto out_no_object; + } + + ret = vmw_fence_obj_init(fman, fence, seqno, mask, + vmw_fence_destroy); + if (unlikely(ret != 0)) + goto out_err_init; + + *p_fence = fence; + return 0; + +out_err_init: + kfree(fence); +out_no_object: + ttm_mem_global_free(mem_glob, fman->fence_size); + return ret; +} + + +static void vmw_user_fence_destroy(struct vmw_fence_obj *fence) +{ + struct vmw_user_fence *ufence = + container_of(fence, struct vmw_user_fence, fence); + struct vmw_fence_manager *fman = fence->fman; + + kfree(ufence); + /* + * Free kernel space accounting. + */ + ttm_mem_global_free(vmw_mem_glob(fman->dev_priv), + fman->user_fence_size); +} + +static void vmw_user_fence_base_release(struct ttm_base_object **p_base) +{ + struct ttm_base_object *base = *p_base; + struct vmw_user_fence *ufence = + container_of(base, struct vmw_user_fence, base); + struct vmw_fence_obj *fence = &ufence->fence; + + *p_base = NULL; + vmw_fence_obj_unreference(&fence); +} + +int vmw_user_fence_create(struct drm_file *file_priv, + struct vmw_fence_manager *fman, + uint32_t seqno, + uint32_t mask, + struct vmw_fence_obj **p_fence, + uint32_t *p_handle) +{ + struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; + struct vmw_user_fence *ufence; + struct vmw_fence_obj *tmp; + struct ttm_mem_global *mem_glob = vmw_mem_glob(fman->dev_priv); + int ret; + + /* + * Kernel memory space accounting, since this object may + * be created by a user-space request. + */ + + ret = ttm_mem_global_alloc(mem_glob, fman->user_fence_size, + false, false); + if (unlikely(ret != 0)) + return ret; + + ufence = kzalloc(sizeof(*ufence), GFP_KERNEL); + if (unlikely(ufence == NULL)) { + ret = -ENOMEM; + goto out_no_object; + } + + ret = vmw_fence_obj_init(fman, &ufence->fence, seqno, + mask, vmw_user_fence_destroy); + if (unlikely(ret != 0)) { + kfree(ufence); + goto out_no_object; + } + + /* + * The base object holds a reference which is freed in + * vmw_user_fence_base_release. + */ + tmp = vmw_fence_obj_reference(&ufence->fence); + ret = ttm_base_object_init(tfile, &ufence->base, false, + VMW_RES_FENCE, + &vmw_user_fence_base_release, NULL); + + + if (unlikely(ret != 0)) { + /* + * Free the base object's reference + */ + vmw_fence_obj_unreference(&tmp); + goto out_err; + } + + *p_fence = &ufence->fence; + *p_handle = ufence->base.hash.key; + + return 0; +out_err: + tmp = &ufence->fence; + vmw_fence_obj_unreference(&tmp); +out_no_object: + ttm_mem_global_free(mem_glob, fman->user_fence_size); + return ret; +} + + +/** + * vmw_fence_fifo_down - signal all unsignaled fence objects. + */ + +void vmw_fence_fifo_down(struct vmw_fence_manager *fman) +{ + unsigned long irq_flags; + struct list_head action_list; + int ret; + + /* + * The list may be altered while we traverse it, so always + * restart when we've released the fman->lock. + */ + + spin_lock_irqsave(&fman->lock, irq_flags); + fman->fifo_down = true; + while (!list_empty(&fman->fence_list)) { + struct vmw_fence_obj *fence = + list_entry(fman->fence_list.prev, struct vmw_fence_obj, + head); + kref_get(&fence->kref); + spin_unlock_irq(&fman->lock); + + ret = vmw_fence_obj_wait(fence, fence->signal_mask, + false, false, + VMW_FENCE_WAIT_TIMEOUT); + + if (unlikely(ret != 0)) { + list_del_init(&fence->head); + fence->signaled |= DRM_VMW_FENCE_FLAG_EXEC; + INIT_LIST_HEAD(&action_list); + list_splice_init(&fence->seq_passed_actions, + &action_list); + vmw_fences_perform_actions(fman, &action_list); + wake_up_all(&fence->queue); } - spin_unlock(&queue->lock); - ret = vmw_wait_fence(dev_priv, false, sequence, true, - 3*HZ); + spin_lock_irq(&fman->lock); - if (unlikely(ret != 0)) - return ret; + BUG_ON(!list_empty(&fence->head)); + kref_put(&fence->kref, vmw_fence_obj_destroy_locked); + } + spin_unlock_irqrestore(&fman->lock, irq_flags); +} + +void vmw_fence_fifo_up(struct vmw_fence_manager *fman) +{ + unsigned long irq_flags; + + spin_lock_irqsave(&fman->lock, irq_flags); + fman->fifo_down = false; + spin_unlock_irqrestore(&fman->lock, irq_flags); +} + + +int vmw_fence_obj_wait_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_vmw_fence_wait_arg *arg = + (struct drm_vmw_fence_wait_arg *)data; + unsigned long timeout; + struct ttm_base_object *base; + struct vmw_fence_obj *fence; + struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; + int ret; + uint64_t wait_timeout = ((uint64_t)arg->timeout_us * HZ); + + /* + * 64-bit division not present on 32-bit systems, so do an + * approximation. (Divide by 1000000). + */ + + wait_timeout = (wait_timeout >> 20) + (wait_timeout >> 24) - + (wait_timeout >> 26); + + if (!arg->cookie_valid) { + arg->cookie_valid = 1; + arg->kernel_cookie = jiffies + wait_timeout; + } + + base = ttm_base_object_lookup(tfile, arg->handle); + if (unlikely(base == NULL)) { + printk(KERN_ERR "Wait invalid fence object handle " + "0x%08lx.\n", + (unsigned long)arg->handle); + return -EINVAL; + } + + fence = &(container_of(base, struct vmw_user_fence, base)->fence); + + timeout = jiffies; + if (time_after_eq(timeout, (unsigned long)arg->kernel_cookie)) { + ret = ((vmw_fence_obj_signaled(fence, arg->flags)) ? + 0 : -EBUSY); + goto out; + } + + timeout = (unsigned long)arg->kernel_cookie - timeout; + + ret = vmw_fence_obj_wait(fence, arg->flags, arg->lazy, true, timeout); + +out: + ttm_base_object_unref(&base); + + /* + * Optionally unref the fence object. + */ + + if (ret == 0 && (arg->wait_options & DRM_VMW_WAIT_OPTION_UNREF)) + return ttm_ref_object_base_unref(tfile, arg->handle, + TTM_REF_USAGE); + return ret; +} + +int vmw_fence_obj_signaled_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_vmw_fence_signaled_arg *arg = + (struct drm_vmw_fence_signaled_arg *) data; + struct ttm_base_object *base; + struct vmw_fence_obj *fence; + struct vmw_fence_manager *fman; + struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile; + struct vmw_private *dev_priv = vmw_priv(dev); - (void) vmw_fence_pull(queue, sequence); + base = ttm_base_object_lookup(tfile, arg->handle); + if (unlikely(base == NULL)) { + printk(KERN_ERR "Fence signaled invalid fence object handle " + "0x%08lx.\n", + (unsigned long)arg->handle); + return -EINVAL; } + + fence = &(container_of(base, struct vmw_user_fence, base)->fence); + fman = fence->fman; + + arg->signaled = vmw_fence_obj_signaled(fence, arg->flags); + spin_lock_irq(&fman->lock); + + arg->signaled_flags = fence->signaled; + arg->passed_seqno = dev_priv->last_read_seqno; + spin_unlock_irq(&fman->lock); + + ttm_base_object_unref(&base); + return 0; } +int vmw_fence_obj_unref_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_vmw_fence_arg *arg = + (struct drm_vmw_fence_arg *) data; + + return ttm_ref_object_base_unref(vmw_fpriv(file_priv)->tfile, + arg->handle, + TTM_REF_USAGE); +} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.h b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.h new file mode 100644 index 00000000000..93074064aaf --- /dev/null +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.h @@ -0,0 +1,105 @@ +/************************************************************************** + * + * Copyright © 2011 VMware, Inc., Palo Alto, CA., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef _VMWGFX_FENCE_H_ + +#define VMW_FENCE_WAIT_TIMEOUT (5*HZ) + +struct vmw_private; + +struct vmw_fence_manager; + +/** + * + * + */ +struct vmw_fence_action { + struct list_head head; + void (*seq_passed) (struct vmw_fence_action *action); + void (*cleanup) (struct vmw_fence_action *action); +}; + +struct vmw_fence_obj { + struct kref kref; + u32 seqno; + + struct vmw_fence_manager *fman; + struct list_head head; + uint32_t signaled; + uint32_t signal_mask; + struct list_head seq_passed_actions; + void (*destroy)(struct vmw_fence_obj *fence); + wait_queue_head_t queue; +}; + +extern struct vmw_fence_manager * +vmw_fence_manager_init(struct vmw_private *dev_priv); + +extern void vmw_fence_manager_takedown(struct vmw_fence_manager *fman); + +extern void vmw_fence_obj_unreference(struct vmw_fence_obj **fence_p); + +extern struct vmw_fence_obj * +vmw_fence_obj_reference(struct vmw_fence_obj *fence); + +extern void vmw_fences_update(struct vmw_fence_manager *fman, + u32 sequence); + +extern bool vmw_fence_obj_signaled(struct vmw_fence_obj *fence, + uint32_t flags); + +extern int vmw_fence_obj_wait(struct vmw_fence_obj *fence, uint32_t flags, + bool lazy, + bool interruptible, unsigned long timeout); + +extern void vmw_fence_obj_flush(struct vmw_fence_obj *fence); + +extern int vmw_fence_create(struct vmw_fence_manager *fman, + uint32_t seqno, + uint32_t mask, + struct vmw_fence_obj **p_fence); + +extern int vmw_user_fence_create(struct drm_file *file_priv, + struct vmw_fence_manager *fman, + uint32_t sequence, + uint32_t mask, + struct vmw_fence_obj **p_fence, + uint32_t *p_handle); + +extern void vmw_fence_fifo_up(struct vmw_fence_manager *fman); + +extern void vmw_fence_fifo_down(struct vmw_fence_manager *fman); + +extern int vmw_fence_obj_wait_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); + +extern int vmw_fence_obj_signaled_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); + +extern int vmw_fence_obj_unref_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +#endif /* _VMWGFX_FENCE_H_ */ diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c index 635c0ffee7f..3ba9cac579e 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c @@ -72,22 +72,12 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo) uint32_t max; uint32_t min; uint32_t dummy; - int ret; fifo->static_buffer_size = VMWGFX_FIFO_STATIC_SIZE; fifo->static_buffer = vmalloc(fifo->static_buffer_size); if (unlikely(fifo->static_buffer == NULL)) return -ENOMEM; - fifo->last_buffer_size = VMWGFX_FIFO_STATIC_SIZE; - fifo->last_data_size = 0; - fifo->last_buffer_add = false; - fifo->last_buffer = vmalloc(fifo->last_buffer_size); - if (unlikely(fifo->last_buffer == NULL)) { - ret = -ENOMEM; - goto out_err; - } - fifo->dynamic_buffer = NULL; fifo->reserved_size = 0; fifo->using_bounce_buffer = false; @@ -137,14 +127,10 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo) (unsigned int) min, (unsigned int) fifo->capabilities); - atomic_set(&dev_priv->fence_seq, dev_priv->last_read_sequence); - iowrite32(dev_priv->last_read_sequence, fifo_mem + SVGA_FIFO_FENCE); - vmw_fence_queue_init(&fifo->fence_queue); + atomic_set(&dev_priv->marker_seq, dev_priv->last_read_seqno); + iowrite32(dev_priv->last_read_seqno, fifo_mem + SVGA_FIFO_FENCE); + vmw_marker_queue_init(&fifo->marker_queue); return vmw_fifo_send_fence(dev_priv, &dummy); -out_err: - vfree(fifo->static_buffer); - fifo->static_buffer = NULL; - return ret; } void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason) @@ -170,7 +156,7 @@ void vmw_fifo_release(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo) while (vmw_read(dev_priv, SVGA_REG_BUSY) != 0) vmw_write(dev_priv, SVGA_REG_SYNC, SVGA_SYNC_GENERIC); - dev_priv->last_read_sequence = ioread32(fifo_mem + SVGA_FIFO_FENCE); + dev_priv->last_read_seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE); vmw_write(dev_priv, SVGA_REG_CONFIG_DONE, dev_priv->config_done_state); @@ -180,12 +166,7 @@ void vmw_fifo_release(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo) dev_priv->traces_state); mutex_unlock(&dev_priv->hw_mutex); - vmw_fence_queue_takedown(&fifo->fence_queue); - - if (likely(fifo->last_buffer != NULL)) { - vfree(fifo->last_buffer); - fifo->last_buffer = NULL; - } + vmw_marker_queue_takedown(&fifo->marker_queue); if (likely(fifo->static_buffer != NULL)) { vfree(fifo->static_buffer); @@ -466,7 +447,7 @@ void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes) mutex_unlock(&fifo_state->fifo_mutex); } -int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *sequence) +int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *seqno) { struct vmw_fifo_state *fifo_state = &dev_priv->fifo; struct svga_fifo_cmd_fence *cmd_fence; @@ -476,16 +457,16 @@ int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *sequence) fm = vmw_fifo_reserve(dev_priv, bytes); if (unlikely(fm == NULL)) { - *sequence = atomic_read(&dev_priv->fence_seq); + *seqno = atomic_read(&dev_priv->marker_seq); ret = -ENOMEM; - (void)vmw_fallback_wait(dev_priv, false, true, *sequence, + (void)vmw_fallback_wait(dev_priv, false, true, *seqno, false, 3*HZ); goto out_err; } do { - *sequence = atomic_add_return(1, &dev_priv->fence_seq); - } while (*sequence == 0); + *seqno = atomic_add_return(1, &dev_priv->marker_seq); + } while (*seqno == 0); if (!(fifo_state->capabilities & SVGA_FIFO_CAP_FENCE)) { @@ -502,61 +483,11 @@ int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *sequence) cmd_fence = (struct svga_fifo_cmd_fence *) ((unsigned long)fm + sizeof(__le32)); - iowrite32(*sequence, &cmd_fence->fence); - fifo_state->last_buffer_add = true; + iowrite32(*seqno, &cmd_fence->fence); vmw_fifo_commit(dev_priv, bytes); - fifo_state->last_buffer_add = false; - (void) vmw_fence_push(&fifo_state->fence_queue, *sequence); - vmw_update_sequence(dev_priv, fifo_state); + (void) vmw_marker_push(&fifo_state->marker_queue, *seqno); + vmw_update_seqno(dev_priv, fifo_state); out_err: return ret; } - -/** - * Map the first page of the FIFO read-only to user-space. - */ - -static int vmw_fifo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) -{ - int ret; - unsigned long address = (unsigned long)vmf->virtual_address; - - if (address != vma->vm_start) - return VM_FAULT_SIGBUS; - - ret = vm_insert_pfn(vma, address, vma->vm_pgoff); - if (likely(ret == -EBUSY || ret == 0)) - return VM_FAULT_NOPAGE; - else if (ret == -ENOMEM) - return VM_FAULT_OOM; - - return VM_FAULT_SIGBUS; -} - -static struct vm_operations_struct vmw_fifo_vm_ops = { - .fault = vmw_fifo_vm_fault, - .open = NULL, - .close = NULL -}; - -int vmw_fifo_mmap(struct file *filp, struct vm_area_struct *vma) -{ - struct drm_file *file_priv; - struct vmw_private *dev_priv; - - file_priv = filp->private_data; - dev_priv = vmw_priv(file_priv->minor->dev); - - if (vma->vm_pgoff != (dev_priv->mmio_start >> PAGE_SHIFT) || - (vma->vm_end - vma->vm_start) != PAGE_SIZE) - return -EINVAL; - - vma->vm_flags &= ~(VM_WRITE | VM_MAYWRITE); - vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_SHARED; - vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); - vma->vm_page_prot = ttm_io_prot(TTM_PL_FLAG_UNCACHED, - vma->vm_page_prot); - vma->vm_ops = &vmw_fifo_vm_ops; - return 0; -} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c b/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c index de0c5948521..f4e7763a769 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c @@ -1,6 +1,6 @@ /************************************************************************** * - * Copyright © 2009 VMware, Inc., Palo Alto, CA., USA + * Copyright © 2009-2011 VMware, Inc., Palo Alto, CA., USA * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a @@ -29,6 +29,77 @@ #include "drmP.h" #include "ttm/ttm_bo_driver.h" +#define VMW_PPN_SIZE sizeof(unsigned long) + +static int vmw_gmr2_bind(struct vmw_private *dev_priv, + struct page *pages[], + unsigned long num_pages, + int gmr_id) +{ + SVGAFifoCmdDefineGMR2 define_cmd; + SVGAFifoCmdRemapGMR2 remap_cmd; + uint32_t define_size = sizeof(define_cmd) + 4; + uint32_t remap_size = VMW_PPN_SIZE * num_pages + sizeof(remap_cmd) + 4; + uint32_t *cmd; + uint32_t *cmd_orig; + uint32_t i; + + cmd_orig = cmd = vmw_fifo_reserve(dev_priv, define_size + remap_size); + if (unlikely(cmd == NULL)) + return -ENOMEM; + + define_cmd.gmrId = gmr_id; + define_cmd.numPages = num_pages; + + remap_cmd.gmrId = gmr_id; + remap_cmd.flags = (VMW_PPN_SIZE > sizeof(*cmd)) ? + SVGA_REMAP_GMR2_PPN64 : SVGA_REMAP_GMR2_PPN32; + remap_cmd.offsetPages = 0; + remap_cmd.numPages = num_pages; + + *cmd++ = SVGA_CMD_DEFINE_GMR2; + memcpy(cmd, &define_cmd, sizeof(define_cmd)); + cmd += sizeof(define_cmd) / sizeof(uint32); + + *cmd++ = SVGA_CMD_REMAP_GMR2; + memcpy(cmd, &remap_cmd, sizeof(remap_cmd)); + cmd += sizeof(remap_cmd) / sizeof(uint32); + + for (i = 0; i < num_pages; ++i) { + if (VMW_PPN_SIZE > 4) + *cmd = page_to_pfn(*pages++); + else + *((uint64_t *)cmd) = page_to_pfn(*pages++); + + cmd += VMW_PPN_SIZE / sizeof(*cmd); + } + + vmw_fifo_commit(dev_priv, define_size + remap_size); + + return 0; +} + +static void vmw_gmr2_unbind(struct vmw_private *dev_priv, + int gmr_id) +{ + SVGAFifoCmdDefineGMR2 define_cmd; + uint32_t define_size = sizeof(define_cmd) + 4; + uint32_t *cmd; + + cmd = vmw_fifo_reserve(dev_priv, define_size); + if (unlikely(cmd == NULL)) { + DRM_ERROR("GMR2 unbind failed.\n"); + return; + } + define_cmd.gmrId = gmr_id; + define_cmd.numPages = 0; + + *cmd++ = SVGA_CMD_DEFINE_GMR2; + memcpy(cmd, &define_cmd, sizeof(define_cmd)); + + vmw_fifo_commit(dev_priv, define_size); +} + /** * FIXME: Adjust to the ttm lowmem / highmem storage to minimize * the number of used descriptors. @@ -170,6 +241,9 @@ int vmw_gmr_bind(struct vmw_private *dev_priv, struct list_head desc_pages; int ret; + if (likely(dev_priv->capabilities & SVGA_CAP_GMR2)) + return vmw_gmr2_bind(dev_priv, pages, num_pages, gmr_id); + if (unlikely(!(dev_priv->capabilities & SVGA_CAP_GMR))) return -EINVAL; @@ -192,6 +266,11 @@ int vmw_gmr_bind(struct vmw_private *dev_priv, void vmw_gmr_unbind(struct vmw_private *dev_priv, int gmr_id) { + if (likely(dev_priv->capabilities & SVGA_CAP_GMR2)) { + vmw_gmr2_unbind(dev_priv, gmr_id); + return; + } + mutex_lock(&dev_priv->hw_mutex); vmw_write(dev_priv, SVGA_REG_GMR_ID, gmr_id); wmb(); diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c b/drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c index ac6e0d1bd62..5f717152cff 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_gmrid_manager.c @@ -40,6 +40,8 @@ struct vmwgfx_gmrid_man { spinlock_t lock; struct ida gmr_ida; uint32_t max_gmr_ids; + uint32_t max_gmr_pages; + uint32_t used_gmr_pages; }; static int vmw_gmrid_man_get_node(struct ttm_mem_type_manager *man, @@ -49,33 +51,50 @@ static int vmw_gmrid_man_get_node(struct ttm_mem_type_manager *man, { struct vmwgfx_gmrid_man *gman = (struct vmwgfx_gmrid_man *)man->priv; - int ret; + int ret = 0; int id; mem->mm_node = NULL; - do { - if (unlikely(ida_pre_get(&gman->gmr_ida, GFP_KERNEL) == 0)) - return -ENOMEM; + spin_lock(&gman->lock); + + if (gman->max_gmr_pages > 0) { + gman->used_gmr_pages += bo->num_pages; + if (unlikely(gman->used_gmr_pages > gman->max_gmr_pages)) + goto out_err_locked; + } + do { + spin_unlock(&gman->lock); + if (unlikely(ida_pre_get(&gman->gmr_ida, GFP_KERNEL) == 0)) { + ret = -ENOMEM; + goto out_err; + } spin_lock(&gman->lock); - ret = ida_get_new(&gman->gmr_ida, &id); + ret = ida_get_new(&gman->gmr_ida, &id); if (unlikely(ret == 0 && id >= gman->max_gmr_ids)) { ida_remove(&gman->gmr_ida, id); - spin_unlock(&gman->lock); - return 0; + ret = 0; + goto out_err_locked; } - - spin_unlock(&gman->lock); - } while (ret == -EAGAIN); if (likely(ret == 0)) { mem->mm_node = gman; mem->start = id; - } + mem->num_pages = bo->num_pages; + } else + goto out_err_locked; + + spin_unlock(&gman->lock); + return 0; +out_err: + spin_lock(&gman->lock); +out_err_locked: + gman->used_gmr_pages -= bo->num_pages; + spin_unlock(&gman->lock); return ret; } @@ -88,6 +107,7 @@ static void vmw_gmrid_man_put_node(struct ttm_mem_type_manager *man, if (mem->mm_node) { spin_lock(&gman->lock); ida_remove(&gman->gmr_ida, mem->start); + gman->used_gmr_pages -= mem->num_pages; spin_unlock(&gman->lock); mem->mm_node = NULL; } @@ -96,6 +116,8 @@ static void vmw_gmrid_man_put_node(struct ttm_mem_type_manager *man, static int vmw_gmrid_man_init(struct ttm_mem_type_manager *man, unsigned long p_size) { + struct vmw_private *dev_priv = + container_of(man->bdev, struct vmw_private, bdev); struct vmwgfx_gmrid_man *gman = kzalloc(sizeof(*gman), GFP_KERNEL); @@ -103,6 +125,8 @@ static int vmw_gmrid_man_init(struct ttm_mem_type_manager *man, return -ENOMEM; spin_lock_init(&gman->lock); + gman->max_gmr_pages = dev_priv->max_gmr_pages; + gman->used_gmr_pages = 0; ida_init(&gman->gmr_ida); gman->max_gmr_ids = p_size; man->priv = (void *) gman; diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c index 570d57775a5..5ecf9666064 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c @@ -45,9 +45,6 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data, case DRM_VMW_PARAM_3D: param->value = vmw_fifo_have_3d(dev_priv) ? 1 : 0; break; - case DRM_VMW_PARAM_FIFO_OFFSET: - param->value = dev_priv->mmio_start; - break; case DRM_VMW_PARAM_HW_CAPS: param->value = dev_priv->capabilities; break; @@ -57,6 +54,13 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data, case DRM_VMW_PARAM_MAX_FB_SIZE: param->value = dev_priv->vram_size; break; + case DRM_VMW_PARAM_FIFO_HW_VERSION: + { + __le32 __iomem *fifo_mem = dev_priv->mmio_virt; + + param->value = ioread32(fifo_mem + SVGA_FIFO_3D_HWVERSION); + break; + } default: DRM_ERROR("Illegal vmwgfx get param request: %d\n", param->param); @@ -66,25 +70,43 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data, return 0; } -int vmw_fifo_debug_ioctl(struct drm_device *dev, void *data, + +int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { + struct drm_vmw_get_3d_cap_arg *arg = + (struct drm_vmw_get_3d_cap_arg *) data; struct vmw_private *dev_priv = vmw_priv(dev); - struct vmw_fifo_state *fifo_state = &dev_priv->fifo; - struct drm_vmw_fifo_debug_arg *arg = - (struct drm_vmw_fifo_debug_arg *)data; - __le32 __user *buffer = (__le32 __user *) - (unsigned long)arg->debug_buffer; + uint32_t size; + __le32 __iomem *fifo_mem; + void __user *buffer = (void __user *)((unsigned long)(arg->buffer)); + void *bounce; + int ret; - if (unlikely(fifo_state->last_buffer == NULL)) + if (unlikely(arg->pad64 != 0)) { + DRM_ERROR("Illegal GET_3D_CAP argument.\n"); return -EINVAL; + } - if (arg->debug_buffer_size < fifo_state->last_data_size) { - arg->used_size = arg->debug_buffer_size; - arg->did_not_fit = 1; - } else { - arg->used_size = fifo_state->last_data_size; - arg->did_not_fit = 0; + size = (SVGA_FIFO_3D_CAPS_LAST - SVGA_FIFO_3D_CAPS + 1) << 2; + + if (arg->max_size < size) + size = arg->max_size; + + bounce = vmalloc(size); + if (unlikely(bounce == NULL)) { + DRM_ERROR("Failed to allocate bounce buffer for 3D caps.\n"); + return -ENOMEM; } - return copy_to_user(buffer, fifo_state->last_buffer, arg->used_size); + + fifo_mem = dev_priv->mmio_virt; + memcpy_fromio(bounce, &fifo_mem[SVGA_FIFO_3D_CAPS], size); + + ret = copy_to_user(buffer, bounce, size); + vfree(bounce); + + if (unlikely(ret != 0)) + DRM_ERROR("Failed to report 3D caps info.\n"); + + return ret; } diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c index e92298a6a38..a005292a890 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c @@ -40,8 +40,13 @@ irqreturn_t vmw_irq_handler(DRM_IRQ_ARGS) status = inl(dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); spin_unlock(&dev_priv->irq_lock); - if (status & SVGA_IRQFLAG_ANY_FENCE) + if (status & SVGA_IRQFLAG_ANY_FENCE) { + __le32 __iomem *fifo_mem = dev_priv->mmio_virt; + uint32_t seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE); + + vmw_fences_update(dev_priv->fman, seqno); wake_up_all(&dev_priv->fence_queue); + } if (status & SVGA_IRQFLAG_FIFO_PROGRESS) wake_up_all(&dev_priv->fifo_queue); @@ -53,7 +58,7 @@ irqreturn_t vmw_irq_handler(DRM_IRQ_ARGS) return IRQ_NONE; } -static bool vmw_fifo_idle(struct vmw_private *dev_priv, uint32_t sequence) +static bool vmw_fifo_idle(struct vmw_private *dev_priv, uint32_t seqno) { uint32_t busy; @@ -64,43 +69,43 @@ static bool vmw_fifo_idle(struct vmw_private *dev_priv, uint32_t sequence) return (busy == 0); } -void vmw_update_sequence(struct vmw_private *dev_priv, +void vmw_update_seqno(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo_state) { __le32 __iomem *fifo_mem = dev_priv->mmio_virt; + uint32_t seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE); - uint32_t sequence = ioread32(fifo_mem + SVGA_FIFO_FENCE); - - if (dev_priv->last_read_sequence != sequence) { - dev_priv->last_read_sequence = sequence; - vmw_fence_pull(&fifo_state->fence_queue, sequence); + if (dev_priv->last_read_seqno != seqno) { + dev_priv->last_read_seqno = seqno; + vmw_marker_pull(&fifo_state->marker_queue, seqno); + vmw_fences_update(dev_priv->fman, seqno); } } -bool vmw_fence_signaled(struct vmw_private *dev_priv, - uint32_t sequence) +bool vmw_seqno_passed(struct vmw_private *dev_priv, + uint32_t seqno) { struct vmw_fifo_state *fifo_state; bool ret; - if (likely(dev_priv->last_read_sequence - sequence < VMW_FENCE_WRAP)) + if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP)) return true; fifo_state = &dev_priv->fifo; - vmw_update_sequence(dev_priv, fifo_state); - if (likely(dev_priv->last_read_sequence - sequence < VMW_FENCE_WRAP)) + vmw_update_seqno(dev_priv, fifo_state); + if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP)) return true; if (!(fifo_state->capabilities & SVGA_FIFO_CAP_FENCE) && - vmw_fifo_idle(dev_priv, sequence)) + vmw_fifo_idle(dev_priv, seqno)) return true; /** - * Then check if the sequence is higher than what we've actually + * Then check if the seqno is higher than what we've actually * emitted. Then the fence is stale and signaled. */ - ret = ((atomic_read(&dev_priv->fence_seq) - sequence) + ret = ((atomic_read(&dev_priv->marker_seq) - seqno) > VMW_FENCE_WRAP); return ret; @@ -109,7 +114,7 @@ bool vmw_fence_signaled(struct vmw_private *dev_priv, int vmw_fallback_wait(struct vmw_private *dev_priv, bool lazy, bool fifo_idle, - uint32_t sequence, + uint32_t seqno, bool interruptible, unsigned long timeout) { @@ -123,7 +128,7 @@ int vmw_fallback_wait(struct vmw_private *dev_priv, DEFINE_WAIT(__wait); wait_condition = (fifo_idle) ? &vmw_fifo_idle : - &vmw_fence_signaled; + &vmw_seqno_passed; /** * Block command submission while waiting for idle. @@ -131,14 +136,14 @@ int vmw_fallback_wait(struct vmw_private *dev_priv, if (fifo_idle) down_read(&fifo_state->rwsem); - signal_seq = atomic_read(&dev_priv->fence_seq); + signal_seq = atomic_read(&dev_priv->marker_seq); ret = 0; for (;;) { prepare_to_wait(&dev_priv->fence_queue, &__wait, (interruptible) ? TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE); - if (wait_condition(dev_priv, sequence)) + if (wait_condition(dev_priv, seqno)) break; if (time_after_eq(jiffies, end_jiffies)) { DRM_ERROR("SVGA device lockup.\n"); @@ -175,68 +180,81 @@ int vmw_fallback_wait(struct vmw_private *dev_priv, return ret; } -int vmw_wait_fence(struct vmw_private *dev_priv, - bool lazy, uint32_t sequence, - bool interruptible, unsigned long timeout) +void vmw_seqno_waiter_add(struct vmw_private *dev_priv) +{ + mutex_lock(&dev_priv->hw_mutex); + if (dev_priv->fence_queue_waiters++ == 0) { + unsigned long irq_flags; + + spin_lock_irqsave(&dev_priv->irq_lock, irq_flags); + outl(SVGA_IRQFLAG_ANY_FENCE, + dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); + vmw_write(dev_priv, SVGA_REG_IRQMASK, + vmw_read(dev_priv, SVGA_REG_IRQMASK) | + SVGA_IRQFLAG_ANY_FENCE); + spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags); + } + mutex_unlock(&dev_priv->hw_mutex); +} + +void vmw_seqno_waiter_remove(struct vmw_private *dev_priv) +{ + mutex_lock(&dev_priv->hw_mutex); + if (--dev_priv->fence_queue_waiters == 0) { + unsigned long irq_flags; + + spin_lock_irqsave(&dev_priv->irq_lock, irq_flags); + vmw_write(dev_priv, SVGA_REG_IRQMASK, + vmw_read(dev_priv, SVGA_REG_IRQMASK) & + ~SVGA_IRQFLAG_ANY_FENCE); + spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags); + } + mutex_unlock(&dev_priv->hw_mutex); +} + +int vmw_wait_seqno(struct vmw_private *dev_priv, + bool lazy, uint32_t seqno, + bool interruptible, unsigned long timeout) { long ret; - unsigned long irq_flags; struct vmw_fifo_state *fifo = &dev_priv->fifo; - if (likely(dev_priv->last_read_sequence - sequence < VMW_FENCE_WRAP)) + if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP)) return 0; - if (likely(vmw_fence_signaled(dev_priv, sequence))) + if (likely(vmw_seqno_passed(dev_priv, seqno))) return 0; vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC); if (!(fifo->capabilities & SVGA_FIFO_CAP_FENCE)) - return vmw_fallback_wait(dev_priv, lazy, true, sequence, + return vmw_fallback_wait(dev_priv, lazy, true, seqno, interruptible, timeout); if (!(dev_priv->capabilities & SVGA_CAP_IRQMASK)) - return vmw_fallback_wait(dev_priv, lazy, false, sequence, + return vmw_fallback_wait(dev_priv, lazy, false, seqno, interruptible, timeout); - mutex_lock(&dev_priv->hw_mutex); - if (atomic_add_return(1, &dev_priv->fence_queue_waiters) > 0) { - spin_lock_irqsave(&dev_priv->irq_lock, irq_flags); - outl(SVGA_IRQFLAG_ANY_FENCE, - dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); - vmw_write(dev_priv, SVGA_REG_IRQMASK, - vmw_read(dev_priv, SVGA_REG_IRQMASK) | - SVGA_IRQFLAG_ANY_FENCE); - spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags); - } - mutex_unlock(&dev_priv->hw_mutex); + vmw_seqno_waiter_add(dev_priv); if (interruptible) ret = wait_event_interruptible_timeout (dev_priv->fence_queue, - vmw_fence_signaled(dev_priv, sequence), + vmw_seqno_passed(dev_priv, seqno), timeout); else ret = wait_event_timeout (dev_priv->fence_queue, - vmw_fence_signaled(dev_priv, sequence), + vmw_seqno_passed(dev_priv, seqno), timeout); + vmw_seqno_waiter_remove(dev_priv); + if (unlikely(ret == 0)) ret = -EBUSY; else if (likely(ret > 0)) ret = 0; - mutex_lock(&dev_priv->hw_mutex); - if (atomic_dec_and_test(&dev_priv->fence_queue_waiters)) { - spin_lock_irqsave(&dev_priv->irq_lock, irq_flags); - vmw_write(dev_priv, SVGA_REG_IRQMASK, - vmw_read(dev_priv, SVGA_REG_IRQMASK) & - ~SVGA_IRQFLAG_ANY_FENCE); - spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags); - } - mutex_unlock(&dev_priv->hw_mutex); - return ret; } @@ -273,25 +291,3 @@ void vmw_irq_uninstall(struct drm_device *dev) status = inl(dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); outl(status, dev_priv->io_start + VMWGFX_IRQSTATUS_PORT); } - -#define VMW_FENCE_WAIT_TIMEOUT 3*HZ; - -int vmw_fence_wait_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv) -{ - struct drm_vmw_fence_wait_arg *arg = - (struct drm_vmw_fence_wait_arg *)data; - unsigned long timeout; - - if (!arg->cookie_valid) { - arg->cookie_valid = 1; - arg->kernel_cookie = jiffies + VMW_FENCE_WAIT_TIMEOUT; - } - - timeout = jiffies; - if (time_after_eq(timeout, (unsigned long)arg->kernel_cookie)) - return -EBUSY; - - timeout = (unsigned long)arg->kernel_cookie - timeout; - return vmw_wait_fence(vmw_priv(dev), true, arg->sequence, true, timeout); -} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c index dfe32e62bd9..1a4c84cecca 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c @@ -567,6 +567,9 @@ static int vmw_kms_new_framebuffer_surface(struct vmw_private *dev_priv, case 15: format = SVGA3D_A1R5G5B5; break; + case 8: + format = SVGA3D_LUMINANCE8; + break; default: DRM_ERROR("Invalid color depth: %d\n", mode_cmd->depth); return -EINVAL; @@ -987,9 +990,9 @@ out: return ret; } -void vmw_kms_write_svga(struct vmw_private *vmw_priv, +int vmw_kms_write_svga(struct vmw_private *vmw_priv, unsigned width, unsigned height, unsigned pitch, - unsigned bbp, unsigned depth) + unsigned bpp, unsigned depth) { if (vmw_priv->capabilities & SVGA_CAP_PITCHLOCK) vmw_write(vmw_priv, SVGA_REG_PITCHLOCK, pitch); @@ -997,11 +1000,15 @@ void vmw_kms_write_svga(struct vmw_private *vmw_priv, iowrite32(pitch, vmw_priv->mmio_virt + SVGA_FIFO_PITCHLOCK); vmw_write(vmw_priv, SVGA_REG_WIDTH, width); vmw_write(vmw_priv, SVGA_REG_HEIGHT, height); - vmw_write(vmw_priv, SVGA_REG_BITS_PER_PIXEL, bbp); - vmw_write(vmw_priv, SVGA_REG_DEPTH, depth); - vmw_write(vmw_priv, SVGA_REG_RED_MASK, 0x00ff0000); - vmw_write(vmw_priv, SVGA_REG_GREEN_MASK, 0x0000ff00); - vmw_write(vmw_priv, SVGA_REG_BLUE_MASK, 0x000000ff); + vmw_write(vmw_priv, SVGA_REG_BITS_PER_PIXEL, bpp); + + if (vmw_read(vmw_priv, SVGA_REG_DEPTH) != depth) { + DRM_ERROR("Invalid depth %u for %u bpp, host expects %u\n", + depth, bpp, vmw_read(vmw_priv, SVGA_REG_DEPTH)); + return -EINVAL; + } + + return 0; } int vmw_kms_save_vga(struct vmw_private *vmw_priv) @@ -1011,12 +1018,7 @@ int vmw_kms_save_vga(struct vmw_private *vmw_priv) vmw_priv->vga_width = vmw_read(vmw_priv, SVGA_REG_WIDTH); vmw_priv->vga_height = vmw_read(vmw_priv, SVGA_REG_HEIGHT); - vmw_priv->vga_depth = vmw_read(vmw_priv, SVGA_REG_DEPTH); vmw_priv->vga_bpp = vmw_read(vmw_priv, SVGA_REG_BITS_PER_PIXEL); - vmw_priv->vga_pseudo = vmw_read(vmw_priv, SVGA_REG_PSEUDOCOLOR); - vmw_priv->vga_red_mask = vmw_read(vmw_priv, SVGA_REG_RED_MASK); - vmw_priv->vga_blue_mask = vmw_read(vmw_priv, SVGA_REG_BLUE_MASK); - vmw_priv->vga_green_mask = vmw_read(vmw_priv, SVGA_REG_GREEN_MASK); if (vmw_priv->capabilities & SVGA_CAP_PITCHLOCK) vmw_priv->vga_pitchlock = vmw_read(vmw_priv, SVGA_REG_PITCHLOCK); @@ -1065,12 +1067,7 @@ int vmw_kms_restore_vga(struct vmw_private *vmw_priv) vmw_write(vmw_priv, SVGA_REG_WIDTH, vmw_priv->vga_width); vmw_write(vmw_priv, SVGA_REG_HEIGHT, vmw_priv->vga_height); - vmw_write(vmw_priv, SVGA_REG_DEPTH, vmw_priv->vga_depth); vmw_write(vmw_priv, SVGA_REG_BITS_PER_PIXEL, vmw_priv->vga_bpp); - vmw_write(vmw_priv, SVGA_REG_PSEUDOCOLOR, vmw_priv->vga_pseudo); - vmw_write(vmw_priv, SVGA_REG_RED_MASK, vmw_priv->vga_red_mask); - vmw_write(vmw_priv, SVGA_REG_GREEN_MASK, vmw_priv->vga_green_mask); - vmw_write(vmw_priv, SVGA_REG_BLUE_MASK, vmw_priv->vga_blue_mask); if (vmw_priv->capabilities & SVGA_CAP_PITCHLOCK) vmw_write(vmw_priv, SVGA_REG_PITCHLOCK, vmw_priv->vga_pitchlock); @@ -1095,52 +1092,6 @@ int vmw_kms_restore_vga(struct vmw_private *vmw_priv) return 0; } -int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv) -{ - struct vmw_private *dev_priv = vmw_priv(dev); - struct drm_vmw_update_layout_arg *arg = - (struct drm_vmw_update_layout_arg *)data; - struct vmw_master *vmaster = vmw_master(file_priv->master); - void __user *user_rects; - struct drm_vmw_rect *rects; - unsigned rects_size; - int ret; - - ret = ttm_read_lock(&vmaster->lock, true); - if (unlikely(ret != 0)) - return ret; - - if (!arg->num_outputs) { - struct drm_vmw_rect def_rect = {0, 0, 800, 600}; - vmw_kms_ldu_update_layout(dev_priv, 1, &def_rect); - goto out_unlock; - } - - rects_size = arg->num_outputs * sizeof(struct drm_vmw_rect); - rects = kzalloc(rects_size, GFP_KERNEL); - if (unlikely(!rects)) { - ret = -ENOMEM; - goto out_unlock; - } - - user_rects = (void __user *)(unsigned long)arg->rects; - ret = copy_from_user(rects, user_rects, rects_size); - if (unlikely(ret != 0)) { - DRM_ERROR("Failed to get rects.\n"); - ret = -EFAULT; - goto out_free; - } - - vmw_kms_ldu_update_layout(dev_priv, arg->num_outputs, rects); - -out_free: - kfree(rects); -out_unlock: - ttm_read_unlock(&vmaster->lock); - return ret; -} - bool vmw_kms_validate_mode_vram(struct vmw_private *dev_priv, uint32_t pitch, uint32_t height) diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c index b3a2cd5118d..7e1901c4f06 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c @@ -83,6 +83,15 @@ static void vmw_ldu_crtc_gamma_set(struct drm_crtc *crtc, u16 *r, u16 *g, u16 *b, uint32_t start, uint32_t size) { + struct vmw_private *dev_priv = vmw_priv(crtc->dev); + int i; + + for (i = 0; i < size; i++) { + DRM_DEBUG("%d r/g/b = 0x%04x / 0x%04x / 0x%04x\n", i, r[i], g[i], b[i]); + vmw_write(dev_priv, SVGA_PALETTE_BASE + i * 3 + 0, r[i] >> 8); + vmw_write(dev_priv, SVGA_PALETTE_BASE + i * 3 + 1, g[i] >> 8); + vmw_write(dev_priv, SVGA_PALETTE_BASE + i * 3 + 2, b[i] >> 8); + } } static void vmw_ldu_crtc_destroy(struct drm_crtc *crtc) @@ -114,10 +123,8 @@ static int vmw_ldu_commit_list(struct vmw_private *dev_priv) return 0; fb = entry->base.crtc.fb; - vmw_kms_write_svga(dev_priv, w, h, fb->pitch, - fb->bits_per_pixel, fb->depth); - - return 0; + return vmw_kms_write_svga(dev_priv, w, h, fb->pitch, + fb->bits_per_pixel, fb->depth); } if (!list_empty(&lds->active)) { @@ -265,9 +272,7 @@ static int vmw_ldu_crtc_set_config(struct drm_mode_set *set) vmw_ldu_del_active(dev_priv, ldu); - vmw_ldu_commit_list(dev_priv); - - return 0; + return vmw_ldu_commit_list(dev_priv); } @@ -292,9 +297,7 @@ static int vmw_ldu_crtc_set_config(struct drm_mode_set *set) vmw_ldu_add_active(dev_priv, ldu, vfb); - vmw_ldu_commit_list(dev_priv); - - return 0; + return vmw_ldu_commit_list(dev_priv); } static struct drm_crtc_funcs vmw_legacy_crtc_funcs = { @@ -340,9 +343,16 @@ static enum drm_connector_status vmw_ldu_connector_detect(struct drm_connector *connector, bool force) { - if (vmw_connector_to_ldu(connector)->pref_active) - return connector_status_connected; - return connector_status_disconnected; + uint32_t num_displays; + struct drm_device *dev = connector->dev; + struct vmw_private *dev_priv = vmw_priv(dev); + + mutex_lock(&dev_priv->hw_mutex); + num_displays = vmw_read(dev_priv, SVGA_REG_NUM_DISPLAYS); + mutex_unlock(&dev_priv->hw_mutex); + + return ((vmw_connector_to_ldu(connector)->base.unit < num_displays) ? + connector_status_connected : connector_status_disconnected); } static const struct drm_display_mode vmw_ldu_connector_builtin[] = { @@ -540,6 +550,8 @@ static int vmw_ldu_init(struct vmw_private *dev_priv, unsigned unit) drm_crtc_init(dev, crtc, &vmw_legacy_crtc_funcs); + drm_mode_crtc_set_gamma_size(crtc, 256); + drm_connector_attach_property(connector, dev->mode_config.dirty_info_property, 1); diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_marker.c b/drivers/gpu/drm/vmwgfx/vmwgfx_marker.c new file mode 100644 index 00000000000..8a8725c2716 --- /dev/null +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_marker.c @@ -0,0 +1,171 @@ +/************************************************************************** + * + * Copyright (C) 2010 VMware, Inc., Palo Alto, CA., USA + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "vmwgfx_drv.h" + +struct vmw_marker { + struct list_head head; + uint32_t seqno; + struct timespec submitted; +}; + +void vmw_marker_queue_init(struct vmw_marker_queue *queue) +{ + INIT_LIST_HEAD(&queue->head); + queue->lag = ns_to_timespec(0); + getrawmonotonic(&queue->lag_time); + spin_lock_init(&queue->lock); +} + +void vmw_marker_queue_takedown(struct vmw_marker_queue *queue) +{ + struct vmw_marker *marker, *next; + + spin_lock(&queue->lock); + list_for_each_entry_safe(marker, next, &queue->head, head) { + kfree(marker); + } + spin_unlock(&queue->lock); +} + +int vmw_marker_push(struct vmw_marker_queue *queue, + uint32_t seqno) +{ + struct vmw_marker *marker = kmalloc(sizeof(*marker), GFP_KERNEL); + + if (unlikely(!marker)) + return -ENOMEM; + + marker->seqno = seqno; + getrawmonotonic(&marker->submitted); + spin_lock(&queue->lock); + list_add_tail(&marker->head, &queue->head); + spin_unlock(&queue->lock); + + return 0; +} + +int vmw_marker_pull(struct vmw_marker_queue *queue, + uint32_t signaled_seqno) +{ + struct vmw_marker *marker, *next; + struct timespec now; + bool updated = false; + + spin_lock(&queue->lock); + getrawmonotonic(&now); + + if (list_empty(&queue->head)) { + queue->lag = ns_to_timespec(0); + queue->lag_time = now; + updated = true; + goto out_unlock; + } + + list_for_each_entry_safe(marker, next, &queue->head, head) { + if (signaled_seqno - marker->seqno > (1 << 30)) + continue; + + queue->lag = timespec_sub(now, marker->submitted); + queue->lag_time = now; + updated = true; + list_del(&marker->head); + kfree(marker); + } + +out_unlock: + spin_unlock(&queue->lock); + + return (updated) ? 0 : -EBUSY; +} + +static struct timespec vmw_timespec_add(struct timespec t1, + struct timespec t2) +{ + t1.tv_sec += t2.tv_sec; + t1.tv_nsec += t2.tv_nsec; + if (t1.tv_nsec >= 1000000000L) { + t1.tv_sec += 1; + t1.tv_nsec -= 1000000000L; + } + + return t1; +} + +static struct timespec vmw_fifo_lag(struct vmw_marker_queue *queue) +{ + struct timespec now; + + spin_lock(&queue->lock); + getrawmonotonic(&now); + queue->lag = vmw_timespec_add(queue->lag, + timespec_sub(now, queue->lag_time)); + queue->lag_time = now; + spin_unlock(&queue->lock); + return queue->lag; +} + + +static bool vmw_lag_lt(struct vmw_marker_queue *queue, + uint32_t us) +{ + struct timespec lag, cond; + + cond = ns_to_timespec((s64) us * 1000); + lag = vmw_fifo_lag(queue); + return (timespec_compare(&lag, &cond) < 1); +} + +int vmw_wait_lag(struct vmw_private *dev_priv, + struct vmw_marker_queue *queue, uint32_t us) +{ + struct vmw_marker *marker; + uint32_t seqno; + int ret; + + while (!vmw_lag_lt(queue, us)) { + spin_lock(&queue->lock); + if (list_empty(&queue->head)) + seqno = atomic_read(&dev_priv->marker_seq); + else { + marker = list_first_entry(&queue->head, + struct vmw_marker, head); + seqno = marker->seqno; + } + spin_unlock(&queue->lock); + + ret = vmw_wait_seqno(dev_priv, false, seqno, true, + 3*HZ); + + if (unlikely(ret != 0)) + return ret; + + (void) vmw_marker_pull(queue, seqno); + } + return 0; +} diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c index bfe1bcce7f8..c1b6ffd4ce7 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c @@ -31,10 +31,6 @@ #include "ttm/ttm_placement.h" #include "drmP.h" -#define VMW_RES_CONTEXT ttm_driver_type0 -#define VMW_RES_SURFACE ttm_driver_type1 -#define VMW_RES_STREAM ttm_driver_type2 - struct vmw_user_context { struct ttm_base_object base; struct vmw_resource res; @@ -211,7 +207,7 @@ static void vmw_hw_context_destroy(struct vmw_resource *res) cmd->body.cid = cpu_to_le32(res->id); vmw_fifo_commit(dev_priv, sizeof(*cmd)); - vmw_3d_resource_dec(dev_priv); + vmw_3d_resource_dec(dev_priv, false); } static int vmw_context_init(struct vmw_private *dev_priv, @@ -248,7 +244,7 @@ static int vmw_context_init(struct vmw_private *dev_priv, cmd->body.cid = cpu_to_le32(res->id); vmw_fifo_commit(dev_priv, sizeof(*cmd)); - (void) vmw_3d_resource_inc(dev_priv); + (void) vmw_3d_resource_inc(dev_priv, false); vmw_resource_activate(res, vmw_hw_context_destroy); return 0; } @@ -364,7 +360,8 @@ out_err: int vmw_context_check(struct vmw_private *dev_priv, struct ttm_object_file *tfile, - int id) + int id, + struct vmw_resource **p_res) { struct vmw_resource *res; int ret = 0; @@ -376,6 +373,8 @@ int vmw_context_check(struct vmw_private *dev_priv, container_of(res, struct vmw_user_context, res); if (ctx->base.tfile != tfile && !ctx->base.shareable) ret = -EPERM; + if (p_res) + *p_res = vmw_resource_reference(res); } else ret = -EINVAL; read_unlock(&dev_priv->resource_lock); @@ -408,7 +407,7 @@ static void vmw_hw_surface_destroy(struct vmw_resource *res) cmd->body.sid = cpu_to_le32(res->id); vmw_fifo_commit(dev_priv, sizeof(*cmd)); - vmw_3d_resource_dec(dev_priv); + vmw_3d_resource_dec(dev_priv, false); } void vmw_surface_res_free(struct vmw_resource *res) @@ -476,7 +475,7 @@ int vmw_surface_init(struct vmw_private *dev_priv, } vmw_fifo_commit(dev_priv, submit_size); - (void) vmw_3d_resource_inc(dev_priv); + (void) vmw_3d_resource_inc(dev_priv, false); vmw_resource_activate(res, vmw_hw_surface_destroy); return 0; } diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c index 1e8eedd901e..d3c11f5184f 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c @@ -34,9 +34,8 @@ int vmw_mmap(struct file *filp, struct vm_area_struct *vma) struct vmw_private *dev_priv; if (unlikely(vma->vm_pgoff < VMWGFX_FILE_PAGE_OFFSET)) { - if (vmw_fifo_mmap(filp, vma) == 0) - return 0; - return drm_mmap(filp, vma); + DRM_ERROR("Illegal attempt to mmap old fifo space.\n"); + return -EINVAL; } file_priv = filp->private_data; |