1 files changed, 140 insertions, 79 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index c2ec91cc845d..19cab37ac69c 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -199,12 +199,12 @@ nouveau_bo_fixup_align(struct nouveau_bo *nvbo, int *align, u64 *size)
 
 struct nouveau_bo *
 nouveau_bo_alloc(struct nouveau_cli *cli, u64 *size, int *align, u32 domain,
-		 u32 tile_mode, u32 tile_flags)
+		 u32 tile_mode, u32 tile_flags, bool internal)
 {
 	struct nouveau_drm *drm = cli->drm;
 	struct nouveau_bo *nvbo;
 	struct nvif_mmu *mmu = &cli->mmu;
-	struct nvif_vmm *vmm = cli->svm.cli ? &cli->svm.vmm : &cli->vmm.vmm;
+	struct nvif_vmm *vmm = &nouveau_cli_vmm(cli)->vmm;
 	int i, pi = -1;
 
 	if (!*size) {
@@ -215,6 +215,7 @@ nouveau_bo_alloc(struct nouveau_cli *cli, u64 *size, int *align, u32 domain,
 	nvbo = kzalloc(sizeof(struct nouveau_bo), GFP_KERNEL);
 	if (!nvbo)
 		return ERR_PTR(-ENOMEM);
+
 	INIT_LIST_HEAD(&nvbo->head);
 	INIT_LIST_HEAD(&nvbo->entry);
 	INIT_LIST_HEAD(&nvbo->vma_list);
@@ -232,68 +233,103 @@ nouveau_bo_alloc(struct nouveau_cli *cli, u64 *size, int *align, u32 domain,
 			nvbo->force_coherent = true;
 	}
 
-	if (cli->device.info.family >= NV_DEVICE_INFO_V0_FERMI) {
-		nvbo->kind = (tile_flags & 0x0000ff00) >> 8;
-		if (!nvif_mmu_kind_valid(mmu, nvbo->kind)) {
-			kfree(nvbo);
-			return ERR_PTR(-EINVAL);
+	nvbo->contig = !(tile_flags & NOUVEAU_GEM_TILE_NONCONTIG);
+	if (!nouveau_cli_uvmm(cli) || internal) {
+		/* for BO noVM allocs, don't assign kinds */
+		if (cli->device.info.family >= NV_DEVICE_INFO_V0_FERMI) {
+			nvbo->kind = (tile_flags & 0x0000ff00) >> 8;
+			if (!nvif_mmu_kind_valid(mmu, nvbo->kind)) {
+				kfree(nvbo);
+				return ERR_PTR(-EINVAL);
+			}
+
+			nvbo->comp = mmu->kind[nvbo->kind] != nvbo->kind;
+		} else if (cli->device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
+			nvbo->kind = (tile_flags & 0x00007f00) >> 8;
+			nvbo->comp = (tile_flags & 0x00030000) >> 16;
+			if (!nvif_mmu_kind_valid(mmu, nvbo->kind)) {
+				kfree(nvbo);
+				return ERR_PTR(-EINVAL);
+			}
+		} else {
+			nvbo->zeta = (tile_flags & 0x00000007);
 		}
+		nvbo->mode = tile_mode;
+
+		/* Determine the desirable target GPU page size for the buffer. */
+		for (i = 0; i < vmm->page_nr; i++) {
+			/* Because we cannot currently allow VMM maps to fail
+			 * during buffer migration, we need to determine page
+			 * size for the buffer up-front, and pre-allocate its
+			 * page tables.
+			 *
+			 * Skip page sizes that can't support needed domains.
+			 */
+			if (cli->device.info.family > NV_DEVICE_INFO_V0_CURIE &&
+			    (domain & NOUVEAU_GEM_DOMAIN_VRAM) && !vmm->page[i].vram)
+				continue;
+			if ((domain & NOUVEAU_GEM_DOMAIN_GART) &&
+			    (!vmm->page[i].host || vmm->page[i].shift > PAGE_SHIFT))
+				continue;
 
-		nvbo->comp = mmu->kind[nvbo->kind] != nvbo->kind;
-	} else
-	if (cli->device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
-		nvbo->kind = (tile_flags & 0x00007f00) >> 8;
-		nvbo->comp = (tile_flags & 0x00030000) >> 16;
-		if (!nvif_mmu_kind_valid(mmu, nvbo->kind)) {
+			/* Select this page size if it's the first that supports
+			 * the potential memory domains, or when it's compatible
+			 * with the requested compression settings.
+			 */
+			if (pi < 0 || !nvbo->comp || vmm->page[i].comp)
+				pi = i;
+
+			/* Stop once the buffer is larger than the current page size. */
+			if (*size >= 1ULL << vmm->page[i].shift)
+				break;
+		}
+
+		if (WARN_ON(pi < 0)) {
 			kfree(nvbo);
 			return ERR_PTR(-EINVAL);
 		}
-	} else {
-		nvbo->zeta = (tile_flags & 0x00000007);
-	}
-	nvbo->mode = tile_mode;
-	nvbo->contig = !(tile_flags & NOUVEAU_GEM_TILE_NONCONTIG);
-
-	/* Determine the desirable target GPU page size for the buffer. */
-	for (i = 0; i < vmm->page_nr; i++) {
-		/* Because we cannot currently allow VMM maps to fail
-		 * during buffer migration, we need to determine page
-		 * size for the buffer up-front, and pre-allocate its
-		 * page tables.
-		 *
-		 * Skip page sizes that can't support needed domains.
-		 */
-		if (cli->device.info.family > NV_DEVICE_INFO_V0_CURIE &&
-		    (domain & NOUVEAU_GEM_DOMAIN_VRAM) && !vmm->page[i].vram)
-			continue;
-		if ((domain & NOUVEAU_GEM_DOMAIN_GART) &&
-		    (!vmm->page[i].host || vmm->page[i].shift > PAGE_SHIFT))
-			continue;
 
-		/* Select this page size if it's the first that supports
-		 * the potential memory domains, or when it's compatible
-		 * with the requested compression settings.
-		 */
-		if (pi < 0 || !nvbo->comp || vmm->page[i].comp)
-			pi = i;
-
-		/* Stop once the buffer is larger than the current page size. */
-		if (*size >= 1ULL << vmm->page[i].shift)
-			break;
-	}
+		/* Disable compression if suitable settings couldn't be found. */
+		if (nvbo->comp && !vmm->page[pi].comp) {
+			if (mmu->object.oclass >= NVIF_CLASS_MMU_GF100)
+				nvbo->kind = mmu->kind[nvbo->kind];
+			nvbo->comp = 0;
+		}
+		nvbo->page = vmm->page[pi].shift;
+	} else {
+		/* reject other tile flags when in VM mode. */
+		if (tile_mode)
+			return ERR_PTR(-EINVAL);
+		if (tile_flags & ~NOUVEAU_GEM_TILE_NONCONTIG)
+			return ERR_PTR(-EINVAL);
 
-	if (WARN_ON(pi < 0)) {
-		kfree(nvbo);
-		return ERR_PTR(-EINVAL);
-	}
+		/* Determine the desirable target GPU page size for the buffer. */
+		for (i = 0; i < vmm->page_nr; i++) {
+			/* Because we cannot currently allow VMM maps to fail
+			 * during buffer migration, we need to determine page
+			 * size for the buffer up-front, and pre-allocate its
+			 * page tables.
+			 *
+			 * Skip page sizes that can't support needed domains.
+			 */
+			if ((domain & NOUVEAU_GEM_DOMAIN_VRAM) && !vmm->page[i].vram)
+				continue;
+			if ((domain & NOUVEAU_GEM_DOMAIN_GART) &&
+			    (!vmm->page[i].host || vmm->page[i].shift > PAGE_SHIFT))
+				continue;
 
-	/* Disable compression if suitable settings couldn't be found. */
-	if (nvbo->comp && !vmm->page[pi].comp) {
-		if (mmu->object.oclass >= NVIF_CLASS_MMU_GF100)
-			nvbo->kind = mmu->kind[nvbo->kind];
-		nvbo->comp = 0;
+			if (pi < 0)
+				pi = i;
+			/* Stop once the buffer is larger than the current page size. */
+			if (*size >= 1ULL << vmm->page[i].shift)
+				break;
+		}
+		if (WARN_ON(pi < 0)) {
+			kfree(nvbo);
+			return ERR_PTR(-EINVAL);
+		}
+		nvbo->page = vmm->page[pi].shift;
 	}
-	nvbo->page = vmm->page[pi].shift;
 
 	nouveau_bo_fixup_align(nvbo, align, size);
 
@@ -306,18 +342,26 @@ nouveau_bo_init(struct nouveau_bo *nvbo, u64 size, int align, u32 domain,
 {
 	int type = sg ? ttm_bo_type_sg : ttm_bo_type_device;
 	int ret;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false,
+		.resv = robj,
+	};
 
 	nouveau_bo_placement_set(nvbo, domain, 0);
 	INIT_LIST_HEAD(&nvbo->io_reserve_lru);
 
-	ret = ttm_bo_init_validate(nvbo->bo.bdev, &nvbo->bo, type,
-				   &nvbo->placement, align >> PAGE_SHIFT, false,
+	ret = ttm_bo_init_reserved(nvbo->bo.bdev, &nvbo->bo, type,
+				   &nvbo->placement, align >> PAGE_SHIFT, &ctx,
 				   sg, robj, nouveau_bo_del_ttm);
 	if (ret) {
 		/* ttm will call nouveau_bo_del_ttm if it fails.. */
 		return ret;
 	}
 
+	if (!robj)
+		ttm_bo_unreserve(&nvbo->bo);
+
 	return 0;
 }
 
@@ -331,7 +375,7 @@ nouveau_bo_new(struct nouveau_cli *cli, u64 size, int align,
 	int ret;
 
 	nvbo = nouveau_bo_alloc(cli, &size, &align, domain, tile_mode,
-				tile_flags);
+				tile_flags, true);
 	if (IS_ERR(nvbo))
 		return PTR_ERR(nvbo);
 
@@ -339,6 +383,11 @@ nouveau_bo_new(struct nouveau_cli *cli, u64 size, int align,
 	dma_resv_init(&nvbo->bo.base._resv);
 	drm_vma_node_reset(&nvbo->bo.base.vma_node);
 
+	/* This must be called before ttm_bo_init_reserved(). Subsequent
+	 * bo_move() callbacks might already iterate the GEMs GPUVA list.
+	 */
+	drm_gem_gpuva_init(&nvbo->bo.base);
+
 	ret = nouveau_bo_init(nvbo, size, align, domain, sg, robj);
 	if (ret)
 		return ret;
@@ -817,29 +866,39 @@ nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict,
 		mutex_lock(&cli->mutex);
 	else
 		mutex_lock_nested(&cli->mutex, SINGLE_DEPTH_NESTING);
+
 	ret = nouveau_fence_sync(nouveau_bo(bo), chan, true, ctx->interruptible);
-	if (ret == 0) {
-		ret = drm->ttm.move(chan, bo, bo->resource, new_reg);
-		if (ret == 0) {
-			ret = nouveau_fence_new(chan, false, &fence);
-			if (ret == 0) {
-				/* TODO: figure out a better solution here
-				 *
-				 * wait on the fence here explicitly as going through
-				 * ttm_bo_move_accel_cleanup somehow doesn't seem to do it.
-				 *
-				 * Without this the operation can timeout and we'll fallback to a
-				 * software copy, which might take several minutes to finish.
-				 */
-				nouveau_fence_wait(fence, false, false);
-				ret = ttm_bo_move_accel_cleanup(bo,
-								&fence->base,
-								evict, false,
-								new_reg);
-				nouveau_fence_unref(&fence);
-			}
-		}
+	if (ret)
+		goto out_unlock;
+
+	ret = drm->ttm.move(chan, bo, bo->resource, new_reg);
+	if (ret)
+		goto out_unlock;
+
+	ret = nouveau_fence_new(&fence);
+	if (ret)
+		goto out_unlock;
+
+	ret = nouveau_fence_emit(fence, chan);
+	if (ret) {
+		nouveau_fence_unref(&fence);
+		goto out_unlock;
 	}
+
+	/* TODO: figure out a better solution here
+	 *
+	 * wait on the fence here explicitly as going through
+	 * ttm_bo_move_accel_cleanup somehow doesn't seem to do it.
+	 *
+	 * Without this the operation can timeout and we'll fallback to a
+	 * software copy, which might take several minutes to finish.
+	 */
+	nouveau_fence_wait(fence, false, false);
+	ret = ttm_bo_move_accel_cleanup(bo, &fence->base, evict, false,
+					new_reg);
+	nouveau_fence_unref(&fence);
+
+out_unlock:
 	mutex_unlock(&cli->mutex);
 	return ret;
 }
@@ -935,6 +994,7 @@ static void nouveau_bo_move_ntfy(struct ttm_buffer_object *bo,
 		list_for_each_entry(vma, &nvbo->vma_list, head) {
 			nouveau_vma_map(vma, mem);
 		}
+		nouveau_uvmm_bo_map_all(nvbo, mem);
 	} else {
 		list_for_each_entry(vma, &nvbo->vma_list, head) {
 			ret = dma_resv_wait_timeout(bo->base.resv,
@@ -943,6 +1003,7 @@ static void nouveau_bo_move_ntfy(struct ttm_buffer_object *bo,
 			WARN_ON(ret <= 0);
 			nouveau_vma_unmap(vma);
 		}
+		nouveau_uvmm_bo_unmap_all(nvbo);
 	}
 
 	if (new_reg)