drm/nouveau: implement new VM_BIND uAPI

This commit provides the implementation for the new uapi motivated by the
Vulkan API. It allows user mode drivers (UMDs) to:

1) Initialize a GPU virtual address (VA) space via the new
   DRM_IOCTL_NOUVEAU_VM_INIT ioctl for UMDs to specify the portion of VA
   space managed by the kernel and userspace, respectively.

2) Allocate and free a VA space region as well as bind and unbind memory
   to the GPUs VA space via the new DRM_IOCTL_NOUVEAU_VM_BIND ioctl.
   UMDs can request the named operations to be processed either
   synchronously or asynchronously. It supports DRM syncobjs
   (incl. timelines) as synchronization mechanism. The management of the
   GPU VA mappings is implemented with the DRM GPU VA manager.

3) Execute push buffers with the new DRM_IOCTL_NOUVEAU_EXEC ioctl. The
   execution happens asynchronously. It supports DRM syncobj (incl.
   timelines) as synchronization mechanism. DRM GEM object locking is
   handled with drm_exec.

Both, DRM_IOCTL_NOUVEAU_VM_BIND and DRM_IOCTL_NOUVEAU_EXEC, use the DRM
GPU scheduler for the asynchronous paths.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230804182406.5222-12-dakr@redhat.com

Documentation/gpu/driver-uapi.rst

@@ -13,4 +13,7 @@ drm/nouveau uAPI
 VM_BIND / EXEC uAPI
 -------------------
 
+.. kernel-doc:: drivers/gpu/drm/nouveau/nouveau_exec.c
+    :doc: Overview
+
 .. kernel-doc:: include/uapi/drm/nouveau_drm.h

drivers/gpu/drm/nouveau/Kbuild

@@ -47,6 +47,9 @@ nouveau-y += nouveau_prime.o
 nouveau-y += nouveau_sgdma.o
 nouveau-y += nouveau_ttm.o
 nouveau-y += nouveau_vmm.o
+nouveau-y += nouveau_exec.o
+nouveau-y += nouveau_sched.o
+nouveau-y += nouveau_uvmm.o
 
 # DRM - modesetting
 nouveau-$(CONFIG_DRM_NOUVEAU_BACKLIGHT) += nouveau_backlight.o

drivers/gpu/drm/nouveau/Kconfig

@@ -10,6 +10,8 @@ config DRM_NOUVEAU
 	select DRM_KMS_HELPER
 	select DRM_TTM
 	select DRM_TTM_HELPER
+	select DRM_EXEC
+	select DRM_SCHED
 	select I2C
 	select I2C_ALGOBIT
 	select BACKLIGHT_CLASS_DEVICE if DRM_NOUVEAU_BACKLIGHT

drivers/gpu/drm/nouveau/nouveau_abi16.c

@@ -35,6 +35,7 @@
 #include "nouveau_chan.h"
 #include "nouveau_abi16.h"
 #include "nouveau_vmm.h"
+#include "nouveau_sched.h"
 
 static struct nouveau_abi16 *
 nouveau_abi16(struct drm_file *file_priv)
@@ -125,6 +126,17 @@ nouveau_abi16_chan_fini(struct nouveau_abi16 *abi16,
 {
 	struct nouveau_abi16_ntfy *ntfy, *temp;
 
+	/* When a client exits without waiting for it's queued up jobs to
+	 * finish it might happen that we fault the channel. This is due to
+	 * drm_file_free() calling drm_gem_release() before the postclose()
+	 * callback. Hence, we can't tear down this scheduler entity before
+	 * uvmm mappings are unmapped. Currently, we can't detect this case.
+	 *
+	 * However, this should be rare and harmless, since the channel isn't
+	 * needed anymore.
+	 */
+	nouveau_sched_entity_fini(&chan->sched_entity);
+
 	/* wait for all activity to stop before cleaning up */
 	if (chan->chan)
 		nouveau_channel_idle(chan->chan);
@@ -261,6 +273,13 @@ nouveau_abi16_ioctl_channel_alloc(ABI16_IOCTL_ARGS)
 	if (!drm->channel)
 		return nouveau_abi16_put(abi16, -ENODEV);
 
+	/* If uvmm wasn't initialized until now disable it completely to prevent
+	 * userspace from mixing up UAPIs.
+	 *
+	 * The client lock is already acquired by nouveau_abi16_get().
+	 */
+	__nouveau_cli_disable_uvmm_noinit(cli);
+
 	device = &abi16->device;
 	engine = NV_DEVICE_HOST_RUNLIST_ENGINES_GR;
 
@@ -304,6 +323,11 @@ nouveau_abi16_ioctl_channel_alloc(ABI16_IOCTL_ARGS)
 	if (ret)
 		goto done;
 
+	ret = nouveau_sched_entity_init(&chan->sched_entity, &drm->sched,
+					drm->sched_wq);
+	if (ret)
+		goto done;
+
 	init->channel = chan->chan->chid;
 
 	if (device->info.family >= NV_DEVICE_INFO_V0_TESLA)

drivers/gpu/drm/nouveau/nouveau_abi16.h

@@ -26,6 +26,7 @@ struct nouveau_abi16_chan {
 	struct nouveau_bo *ntfy;
 	struct nouveau_vma *ntfy_vma;
 	struct nvkm_mm  heap;
+	struct nouveau_sched_entity sched_entity;
 };
 
 struct nouveau_abi16 {

drivers/gpu/drm/nouveau/nouveau_bo.c

@@ -199,7 +199,7 @@ 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;
@@ -233,68 +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);
 
@@ -307,18 +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;
 }
 
@@ -332,7 +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);
 
@@ -951,6 +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,
@@ -959,6 +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)

drivers/gpu/drm/nouveau/nouveau_bo.h

@@ -26,6 +26,7 @@ struct nouveau_bo {
 	struct list_head entry;
 	int pbbo_index;
 	bool validate_mapped;
+	bool no_share;
 
 	/* GPU address space is independent of CPU word size */
 	uint64_t offset;
@@ -73,7 +74,7 @@ extern struct ttm_device_funcs nouveau_bo_driver;
 
 void nouveau_bo_move_init(struct nouveau_drm *);
 struct nouveau_bo *nouveau_bo_alloc(struct nouveau_cli *, u64 *size, int *align,
-				    u32 domain, u32 tile_mode, u32 tile_flags);
+				    u32 domain, u32 tile_mode, u32 tile_flags, bool internal);
 int  nouveau_bo_init(struct nouveau_bo *, u64 size, int align, u32 domain,
 		     struct sg_table *sg, struct dma_resv *robj);
 int  nouveau_bo_new(struct nouveau_cli *, u64 size, int align, u32 domain,