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,

drivers/gpu/drm/nouveau/nouveau_drm.c

@@ -68,6 +68,9 @@
 #include "nouveau_platform.h"
 #include "nouveau_svm.h"
 #include "nouveau_dmem.h"
+#include "nouveau_exec.h"
+#include "nouveau_uvmm.h"
+#include "nouveau_sched.h"
 
 DECLARE_DYNDBG_CLASSMAP(drm_debug_classes, DD_CLASS_TYPE_DISJOINT_BITS, 0,
 			"DRM_UT_CORE",
@@ -196,6 +199,8 @@ nouveau_cli_fini(struct nouveau_cli *cli)
 	WARN_ON(!list_empty(&cli->worker));
 
 	usif_client_fini(cli);
+	nouveau_uvmm_fini(&cli->uvmm);
+	nouveau_sched_entity_fini(&cli->sched_entity);
 	nouveau_vmm_fini(&cli->svm);
 	nouveau_vmm_fini(&cli->vmm);
 	nvif_mmu_dtor(&cli->mmu);
@@ -301,6 +306,12 @@ nouveau_cli_init(struct nouveau_drm *drm, const char *sname,
 	}
 
 	cli->mem = &mems[ret];
+
+	ret = nouveau_sched_entity_init(&cli->sched_entity, &drm->sched,
+					drm->sched_wq);
+	if (ret)
+		goto done;
+
 	return 0;
 done:
 	if (ret)
@@ -568,10 +579,14 @@ nouveau_drm_device_init(struct drm_device *dev)
 	nvif_parent_ctor(&nouveau_parent, &drm->parent);
 	drm->master.base.object.parent = &drm->parent;
 
-	ret = nouveau_cli_init(drm, "DRM-master", &drm->master);
+	ret = nouveau_sched_init(drm);
 	if (ret)
 		goto fail_alloc;
 
+	ret = nouveau_cli_init(drm, "DRM-master", &drm->master);
+	if (ret)
+		goto fail_sched;
+
 	ret = nouveau_cli_init(drm, "DRM", &drm->client);
 	if (ret)
 		goto fail_master;
@@ -628,7 +643,6 @@ nouveau_drm_device_init(struct drm_device *dev)
 	}
 
 	return 0;
-
 fail_dispinit:
 	nouveau_display_destroy(dev);
 fail_dispctor:
@@ -641,6 +655,8 @@ nouveau_drm_device_init(struct drm_device *dev)
 	nouveau_cli_fini(&drm->client);
 fail_master:
 	nouveau_cli_fini(&drm->master);
+fail_sched:
+	nouveau_sched_fini(drm);
 fail_alloc:
 	nvif_parent_dtor(&drm->parent);
 	kfree(drm);
@@ -692,6 +708,8 @@ nouveau_drm_device_fini(struct drm_device *dev)
 	}
 	mutex_unlock(&drm->clients_lock);
 
+	nouveau_sched_fini(drm);
+
 	nouveau_cli_fini(&drm->client);
 	nouveau_cli_fini(&drm->master);
 	nvif_parent_dtor(&drm->parent);
@@ -1193,6 +1211,9 @@ nouveau_ioctls[] = {
 	DRM_IOCTL_DEF_DRV(NOUVEAU_GEM_CPU_PREP, nouveau_gem_ioctl_cpu_prep, DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(NOUVEAU_GEM_CPU_FINI, nouveau_gem_ioctl_cpu_fini, DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(NOUVEAU_GEM_INFO, nouveau_gem_ioctl_info, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_VM_INIT, nouveau_uvmm_ioctl_vm_init, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_VM_BIND, nouveau_uvmm_ioctl_vm_bind, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_EXEC, nouveau_exec_ioctl_exec, DRM_RENDER_ALLOW),
 };
 
 long
@@ -1240,6 +1261,8 @@ nouveau_driver_fops = {
 static struct drm_driver
 driver_stub = {
 	.driver_features = DRIVER_GEM |
+			   DRIVER_SYNCOBJ | DRIVER_SYNCOBJ_TIMELINE |
+			   DRIVER_GEM_GPUVA |
 			   DRIVER_MODESET |
 			   DRIVER_RENDER,
 	.open = nouveau_drm_open,

drivers/gpu/drm/nouveau/nouveau_drv.h

@@ -10,8 +10,8 @@
 #define DRIVER_DATE		"20120801"
 
 #define DRIVER_MAJOR		1
-#define DRIVER_MINOR		3
-#define DRIVER_PATCHLEVEL	1
+#define DRIVER_MINOR		4
+#define DRIVER_PATCHLEVEL	0
 
 /*
  * 1.1.1:
@@ -63,7 +63,9 @@ struct platform_device;
 
 #include "nouveau_fence.h"
 #include "nouveau_bios.h"
+#include "nouveau_sched.h"
 #include "nouveau_vmm.h"
+#include "nouveau_uvmm.h"
 
 struct nouveau_drm_tile {
 	struct nouveau_fence *fence;
@@ -91,6 +93,10 @@ struct nouveau_cli {
 	struct nvif_mmu mmu;
 	struct nouveau_vmm vmm;
 	struct nouveau_vmm svm;
+	struct nouveau_uvmm uvmm;
+
+	struct nouveau_sched_entity sched_entity;
+
 	const struct nvif_mclass *mem;
 
 	struct list_head head;
@@ -112,15 +118,59 @@ struct nouveau_cli_work {
 	struct dma_fence_cb cb;
 };
 
+static inline struct nouveau_uvmm *
+nouveau_cli_uvmm(struct nouveau_cli *cli)
+{
+	if (!cli || !cli->uvmm.vmm.cli)
+		return NULL;
+
+	return &cli->uvmm;
+}
+
+static inline struct nouveau_uvmm *
+nouveau_cli_uvmm_locked(struct nouveau_cli *cli)
+{
+	struct nouveau_uvmm *uvmm;
+
+	mutex_lock(&cli->mutex);
+	uvmm = nouveau_cli_uvmm(cli);
+	mutex_unlock(&cli->mutex);
+
+	return uvmm;
+}
+
 static inline struct nouveau_vmm *
 nouveau_cli_vmm(struct nouveau_cli *cli)
 {
+	struct nouveau_uvmm *uvmm;
+
+	uvmm = nouveau_cli_uvmm(cli);
+	if (uvmm)
+		return &uvmm->vmm;
+
 	if (cli->svm.cli)
 		return &cli->svm;
 
 	return &cli->vmm;
 }
 
+static inline void
+__nouveau_cli_disable_uvmm_noinit(struct nouveau_cli *cli)
+{
+	struct nouveau_uvmm *uvmm = nouveau_cli_uvmm(cli);
+
+	if (!uvmm)
+		cli->uvmm.disabled = true;
+}
+
+static inline void
+nouveau_cli_disable_uvmm_noinit(struct nouveau_cli *cli)
+{
+	mutex_lock(&cli->mutex);
+	__nouveau_cli_disable_uvmm_noinit(cli);
+	mutex_unlock(&cli->mutex);
+}
+
 void nouveau_cli_work_queue(struct nouveau_cli *, struct dma_fence *,
 			    struct nouveau_cli_work *);
 
@@ -257,6 +307,10 @@ struct nouveau_drm {
 		struct mutex lock;
 		bool component_registered;
 	} audio;
+
+	struct drm_gpu_scheduler sched;
+	struct workqueue_struct *sched_wq;
+
 };
 
 static inline struct nouveau_drm *

drivers/gpu/drm/nouveau/nouveau_exec.h

+/* SPDX-License-Identifier: MIT */
+
+#ifndef __NOUVEAU_EXEC_H__
+#define __NOUVEAU_EXEC_H__
+
+#include <drm/drm_exec.h>
+
+#include "nouveau_drv.h"
+#include "nouveau_sched.h"
+
+struct nouveau_exec_job_args {
+	struct drm_file *file_priv;
+	struct nouveau_sched_entity *sched_entity;
+
+	struct drm_exec exec;
+	struct nouveau_channel *chan;
+
+	struct {
+		struct drm_nouveau_sync *s;
+		u32 count;
+	} in_sync;
+
+	struct {
+		struct drm_nouveau_sync *s;
+		u32 count;
+	} out_sync;
+
+	struct {
+		struct drm_nouveau_exec_push *s;
+		u32 count;
+	} push;
+};
+
+struct nouveau_exec_job {
+	struct nouveau_job base;
+	struct nouveau_fence *fence;
+	struct nouveau_channel *chan;
+
+	struct {
+		struct drm_nouveau_exec_push *s;
+		u32 count;
+	} push;
+};
+
+#define to_nouveau_exec_job(job)		\
+		container_of((job), struct nouveau_exec_job, base)
+
+int nouveau_exec_job_init(struct nouveau_exec_job **job,
+			  struct nouveau_exec_job_args *args);
+
+int nouveau_exec_ioctl_exec(struct drm_device *dev, void __user *data,
+			    struct drm_file *file_priv);
+
+#endif

drivers/gpu/drm/nouveau/nouveau_gem.c

@@ -103,6 +103,7 @@ nouveau_gem_object_open(struct drm_gem_object *gem, struct drm_file *file_priv)
 	struct nouveau_bo *nvbo = nouveau_gem_object(gem);
 	struct nouveau_drm *drm = nouveau_bdev(nvbo->bo.bdev);
 	struct device *dev = drm->dev->dev;
+	struct nouveau_uvmm *uvmm = nouveau_cli_uvmm(cli);
 	struct nouveau_vmm *vmm = nouveau_cli_vmm(cli);
 	struct nouveau_vma *vma;
 	int ret;
@@ -110,6 +111,9 @@ nouveau_gem_object_open(struct drm_gem_object *gem, struct drm_file *file_priv)
 	if (vmm->vmm.object.oclass < NVIF_CLASS_VMM_NV50)
 		return 0;
 
+	if (nvbo->no_share && uvmm && &uvmm->resv != nvbo->bo.base.resv)
+		return -EPERM;
+
 	ret = ttm_bo_reserve(&nvbo->bo, false, false, NULL);
 	if (ret)
 		return ret;
@@ -120,7 +124,11 @@ nouveau_gem_object_open(struct drm_gem_object *gem, struct drm_file *file_priv)
 		goto out;
 	}
 
-	ret = nouveau_vma_new(nvbo, vmm, &vma);
+	/* only create a VMA on binding */
+	if (!nouveau_cli_uvmm(cli))
+		ret = nouveau_vma_new(nvbo, vmm, &vma);
+	else
+		ret = 0;
 	pm_runtime_mark_last_busy(dev);
 	pm_runtime_put_autosuspend(dev);
 out:
@@ -187,6 +195,9 @@ nouveau_gem_object_close(struct drm_gem_object *gem, struct drm_file *file_priv)
 	if (vmm->vmm.object.oclass < NVIF_CLASS_VMM_NV50)
 		return;
 
+	if (nouveau_cli_uvmm(cli))
+		return;
+
 	ret = ttm_bo_reserve(&nvbo->bo, false, false, NULL);
 	if (ret)
 		return;
@@ -209,6 +220,7 @@ const struct drm_gem_object_funcs nouveau_gem_object_funcs = {
 	.free = nouveau_gem_object_del,
 	.open = nouveau_gem_object_open,
 	.close = nouveau_gem_object_close,
+	.export = nouveau_gem_prime_export,
 	.pin = nouveau_gem_prime_pin,
 	.unpin = nouveau_gem_prime_unpin,
 	.get_sg_table = nouveau_gem_prime_get_sg_table,
@@ -224,18 +236,28 @@ nouveau_gem_new(struct nouveau_cli *cli, u64 size, int align, uint32_t domain,
 		struct nouveau_bo **pnvbo)
 {
 	struct nouveau_drm *drm = cli->drm;
+	struct nouveau_uvmm *uvmm = nouveau_cli_uvmm(cli);
+	struct dma_resv *resv = NULL;
 	struct nouveau_bo *nvbo;
 	int ret;
 
+	if (domain & NOUVEAU_GEM_DOMAIN_NO_SHARE) {
+		if (unlikely(!uvmm))
+			return -EINVAL;
+
+		resv = &uvmm->resv;
+	}
+
 	if (!(domain & (NOUVEAU_GEM_DOMAIN_VRAM | NOUVEAU_GEM_DOMAIN_GART)))
 		domain |= NOUVEAU_GEM_DOMAIN_CPU;
 
 	nvbo = nouveau_bo_alloc(cli, &size, &align, domain, tile_mode,
-				tile_flags);
+				tile_flags, false);
 	if (IS_ERR(nvbo))
 		return PTR_ERR(nvbo);
 
 	nvbo->bo.base.funcs = &nouveau_gem_object_funcs;
+	nvbo->no_share = domain & NOUVEAU_GEM_DOMAIN_NO_SHARE;
 
 	/* Initialize the embedded gem-object. We return a single gem-reference
 	 * to the caller, instead of a normal nouveau_bo ttm reference. */
@@ -246,7 +268,14 @@ nouveau_gem_new(struct nouveau_cli *cli, u64 size, int align, uint32_t domain,
 		return ret;
 	}
 
-	ret = nouveau_bo_init(nvbo, size, align, domain, NULL, NULL);
+	if (resv)
+		dma_resv_lock(resv, NULL);
+
+	ret = nouveau_bo_init(nvbo, size, align, domain, NULL, resv);
+
+	if (resv)
+		dma_resv_unlock(resv);
+
 	if (ret)
 		return ret;
 
@@ -279,13 +308,15 @@ nouveau_gem_info(struct drm_file *file_priv, struct drm_gem_object *gem,
 	else
 		rep->domain = NOUVEAU_GEM_DOMAIN_VRAM;
 	rep->offset = nvbo->offset;
-	if (vmm->vmm.object.oclass >= NVIF_CLASS_VMM_NV50) {
+	if (vmm->vmm.object.oclass >= NVIF_CLASS_VMM_NV50 &&
+	    !nouveau_cli_uvmm(cli)) {
 		vma = nouveau_vma_find(nvbo, vmm);
 		if (!vma)
 			return -EINVAL;
 
 		rep->offset = vma->addr;
-	}
+	} else
+		rep->offset = 0;
 
 	rep->size = nvbo->bo.base.size;
 	rep->map_handle = drm_vma_node_offset_addr(&nvbo->bo.base.vma_node);
@@ -310,6 +341,11 @@ nouveau_gem_ioctl_new(struct drm_device *dev, void *data,
 	struct nouveau_bo *nvbo = NULL;
 	int ret = 0;
 
+	/* If uvmm wasn't initialized until now disable it completely to prevent
+	 * userspace from mixing up UAPIs.
+	 */
+	nouveau_cli_disable_uvmm_noinit(cli);
+
 	ret = nouveau_gem_new(cli, req->info.size, req->align,
 			      req->info.domain, req->info.tile_mode,
 			      req->info.tile_flags, &nvbo);
@@ -721,6 +757,9 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
 	if (unlikely(!abi16))
 		return -ENOMEM;
 
+	if (unlikely(nouveau_cli_uvmm(cli)))
+		return -ENOSYS;
+
 	list_for_each_entry(temp, &abi16->channels, head) {
 		if (temp->chan->chid == req->channel) {
 			chan = temp->chan;

drivers/gpu/drm/nouveau/nouveau_gem.h

@@ -37,5 +37,6 @@ extern void nouveau_gem_prime_unpin(struct drm_gem_object *);
 extern struct sg_table *nouveau_gem_prime_get_sg_table(struct drm_gem_object *);
 extern struct drm_gem_object *nouveau_gem_prime_import_sg_table(
 	struct drm_device *, struct dma_buf_attachment *, struct sg_table *);
-
+struct dma_buf *nouveau_gem_prime_export(struct drm_gem_object *gobj,
+					 int flags);
 #endif

drivers/gpu/drm/nouveau/nouveau_mem.h

@@ -35,4 +35,9 @@ int nouveau_mem_vram(struct ttm_resource *, bool contig, u8 page);
 int nouveau_mem_host(struct ttm_resource *, struct ttm_tt *);
 void nouveau_mem_fini(struct nouveau_mem *);
 int nouveau_mem_map(struct nouveau_mem *, struct nvif_vmm *, struct nvif_vma *);
+int
+nouveau_mem_map_fixed(struct nouveau_mem *mem,
+		      struct nvif_vmm *vmm,
+		      u8 kind, u64 addr,
+		      u64 offset, u64 range);
 #endif

drivers/gpu/drm/nouveau/nouveau_prime.c

@@ -50,7 +50,7 @@ struct drm_gem_object *nouveau_gem_prime_import_sg_table(struct drm_device *dev,
 
 	dma_resv_lock(robj, NULL);
 	nvbo = nouveau_bo_alloc(&drm->client, &size, &align,
-				NOUVEAU_GEM_DOMAIN_GART, 0, 0);
+				NOUVEAU_GEM_DOMAIN_GART, 0, 0, true);
 	if (IS_ERR(nvbo)) {
 		obj = ERR_CAST(nvbo);
 		goto unlock;
@@ -102,3 +102,14 @@ void nouveau_gem_prime_unpin(struct drm_gem_object *obj)
 
 	nouveau_bo_unpin(nvbo);
 }
+
+struct dma_buf *nouveau_gem_prime_export(struct drm_gem_object *gobj,
+					 int flags)
+{
+	struct nouveau_bo *nvbo = nouveau_gem_object(gobj);
+
+	if (nvbo->no_share)
+		return ERR_PTR(-EPERM);
+
+	return drm_gem_prime_export(gobj, flags);
+}

drivers/gpu/drm/nouveau/nouveau_sched.c

+// SPDX-License-Identifier: MIT
+
+#include <linux/slab.h>
+#include <drm/gpu_scheduler.h>
+#include <drm/drm_syncobj.h>
+
+#include "nouveau_drv.h"
+#include "nouveau_gem.h"
+#include "nouveau_mem.h"
+#include "nouveau_dma.h"
+#include "nouveau_exec.h"
+#include "nouveau_abi16.h"
+#include "nouveau_sched.h"
+
+/* FIXME
+ *
+ * We want to make sure that jobs currently executing can't be deferred by
+ * other jobs competing for the hardware. Otherwise we might end up with job
+ * timeouts just because of too many clients submitting too many jobs. We don't
+ * want jobs to time out because of system load, but because of the job being
+ * too bulky.
+ *
+ * For now allow for up to 16 concurrent jobs in flight until we know how many
+ * rings the hardware can process in parallel.
+ */
+#define NOUVEAU_SCHED_HW_SUBMISSIONS		16
+#define NOUVEAU_SCHED_JOB_TIMEOUT_MS		10000
+
+int
+nouveau_job_init(struct nouveau_job *job,
+		 struct nouveau_job_args *args)
+{
+	struct nouveau_sched_entity *entity = args->sched_entity;
+	int ret;
+
+	job->file_priv = args->file_priv;
+	job->cli = nouveau_cli(args->file_priv);
+	job->entity = entity;
+
+	job->sync = args->sync;
+	job->resv_usage = args->resv_usage;
+
+	job->ops = args->ops;
+
+	job->in_sync.count = args->in_sync.count;
+	if (job->in_sync.count) {
+		if (job->sync)
+			return -EINVAL;
+
+		job->in_sync.data = kmemdup(args->in_sync.s,
+					 sizeof(*args->in_sync.s) *
+					 args->in_sync.count,
+					 GFP_KERNEL);
+		if (!job->in_sync.data)
+			return -ENOMEM;
+	}
+
+	job->out_sync.count = args->out_sync.count;
+	if (job->out_sync.count) {
+		if (job->sync) {
+			ret = -EINVAL;
+			goto err_free_in_sync;
+		}
+
+		job->out_sync.data = kmemdup(args->out_sync.s,
+					  sizeof(*args->out_sync.s) *
+					  args->out_sync.count,
+					  GFP_KERNEL);
+		if (!job->out_sync.data) {
+			ret = -ENOMEM;
+			goto err_free_in_sync;
+		}
+
+		job->out_sync.objs = kcalloc(job->out_sync.count,
+					     sizeof(*job->out_sync.objs),
+					     GFP_KERNEL);
+		if (!job->out_sync.objs) {
+			ret = -ENOMEM;
+			goto err_free_out_sync;
+		}
+
+		job->out_sync.chains = kcalloc(job->out_sync.count,
+					       sizeof(*job->out_sync.chains),
+					       GFP_KERNEL);
+		if (!job->out_sync.chains) {
+			ret = -ENOMEM;
+			goto err_free_objs;
+		}
+
+	}
+
+	ret = drm_sched_job_init(&job->base, &entity->base, NULL);
+	if (ret)
+		goto err_free_chains;
+
+	job->state = NOUVEAU_JOB_INITIALIZED;
+
+	return 0;
+
+err_free_chains:
+	kfree(job->out_sync.chains);
+err_free_objs:
+	kfree(job->out_sync.objs);
+err_free_out_sync:
+	kfree(job->out_sync.data);
+err_free_in_sync:
+	kfree(job->in_sync.data);
+return ret;
+}
+
+void
+nouveau_job_free(struct nouveau_job *job)
+{
+	kfree(job->in_sync.data);
+	kfree(job->out_sync.data);
+	kfree(job->out_sync.objs);
+	kfree(job->out_sync.chains);
+}
+
+void nouveau_job_fini(struct nouveau_job *job)
+{
+	dma_fence_put(job->done_fence);
+	drm_sched_job_cleanup(&job->base);
+	job->ops->free(job);
+}
+
+static int
+sync_find_fence(struct nouveau_job *job,
+		struct drm_nouveau_sync *sync,
+		struct dma_fence **fence)
+{
+	u32 stype = sync->flags & DRM_NOUVEAU_SYNC_TYPE_MASK;
+	u64 point = 0;
+	int ret;
+
+	if (stype != DRM_NOUVEAU_SYNC_SYNCOBJ &&
+	    stype != DRM_NOUVEAU_SYNC_TIMELINE_SYNCOBJ)
+		return -EOPNOTSUPP;
+
+	if (stype == DRM_NOUVEAU_SYNC_TIMELINE_SYNCOBJ)
+		point = sync->timeline_value;
+
+	ret = drm_syncobj_find_fence(job->file_priv,
+				     sync->handle, point,
+				     sync->flags, fence);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static int
+nouveau_job_add_deps(struct nouveau_job *job)
+{
+	struct dma_fence *in_fence = NULL;
+	int ret, i;
+
+	for (i = 0; i < job->in_sync.count; i++) {
+		struct drm_nouveau_sync *sync = &job->in_sync.data[i];
+
+		ret = sync_find_fence(job, sync, &in_fence);
+		if (ret) {
+			NV_PRINTK(warn, job->cli,
+				  "Failed to find syncobj (-> in): handle=%d\n",
+				  sync->handle);
+			return ret;
+		}
+
+		ret = drm_sched_job_add_dependency(&job->base, in_fence);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+static void
+nouveau_job_fence_attach_cleanup(struct nouveau_job *job)
+{
+	int i;
+
+	for (i = 0; i < job->out_sync.count; i++) {
+		struct drm_syncobj *obj = job->out_sync.objs[i];
+		struct dma_fence_chain *chain = job->out_sync.chains[i];
+
+		if (obj)
+			drm_syncobj_put(obj);
+
+		if (chain)
+			dma_fence_chain_free(chain);
+	}
+}
+
+static int
+nouveau_job_fence_attach_prepare(struct nouveau_job *job)
+{
+	int i, ret;
+
+	for (i = 0; i < job->out_sync.count; i++) {
+		struct drm_nouveau_sync *sync = &job->out_sync.data[i];
+		struct drm_syncobj **pobj = &job->out_sync.objs[i];
+		struct dma_fence_chain **pchain = &job->out_sync.chains[i];
+		u32 stype = sync->flags & DRM_NOUVEAU_SYNC_TYPE_MASK;
+
+		if (stype != DRM_NOUVEAU_SYNC_SYNCOBJ &&
+		    stype != DRM_NOUVEAU_SYNC_TIMELINE_SYNCOBJ) {
+			ret = -EINVAL;
+			goto err_sync_cleanup;
+		}
+
+		*pobj = drm_syncobj_find(job->file_priv, sync->handle);
+		if (!*pobj) {
+			NV_PRINTK(warn, job->cli,
+				  "Failed to find syncobj (-> out): handle=%d\n",
+				  sync->handle);
+			ret = -ENOENT;
+			goto err_sync_cleanup;
+		}
+
+		if (stype == DRM_NOUVEAU_SYNC_TIMELINE_SYNCOBJ) {
+			*pchain = dma_fence_chain_alloc();
+			if (!*pchain) {
+				ret = -ENOMEM;
+				goto err_sync_cleanup;
+			}
+		}
+	}
+
+	return 0;
+
+err_sync_cleanup:
+	nouveau_job_fence_attach_cleanup(job);
+	return ret;
+}
+
+static void
+nouveau_job_fence_attach(struct nouveau_job *job)
+{
+	struct dma_fence *fence = job->done_fence;
+	int i;
+
+	for (i = 0; i < job->out_sync.count; i++) {
+		struct drm_nouveau_sync *sync = &job->out_sync.data[i];
+		struct drm_syncobj **pobj = &job->out_sync.objs[i];
+		struct dma_fence_chain **pchain = &job->out_sync.chains[i];
+		u32 stype = sync->flags & DRM_NOUVEAU_SYNC_TYPE_MASK;
+
+		if (stype == DRM_NOUVEAU_SYNC_TIMELINE_SYNCOBJ) {
+			drm_syncobj_add_point(*pobj, *pchain, fence,
+					      sync->timeline_value);
+		} else {
+			drm_syncobj_replace_fence(*pobj, fence);
+		}
+
+		drm_syncobj_put(*pobj);
+		*pobj = NULL;
+		*pchain = NULL;
+	}
+}
+
+int
+nouveau_job_submit(struct nouveau_job *job)
+{
+	struct nouveau_sched_entity *entity = to_nouveau_sched_entity(job->base.entity);
+	struct dma_fence *done_fence = NULL;
+	int ret;
+
+	ret = nouveau_job_add_deps(job);
+	if (ret)
+		goto err;
+
+	ret = nouveau_job_fence_attach_prepare(job);
+	if (ret)
+		goto err;
+
+	/* Make sure the job appears on the sched_entity's queue in the same
+	 * order as it was submitted.
+	 */
+	mutex_lock(&entity->mutex);
+
+	/* Guarantee we won't fail after the submit() callback returned
+	 * successfully.
+	 */
+	if (job->ops->submit) {
+		ret = job->ops->submit(job);
+		if (ret)
+			goto err_cleanup;
+	}
+
+	drm_sched_job_arm(&job->base);
+	job->done_fence = dma_fence_get(&job->base.s_fence->finished);
+	if (job->sync)
+		done_fence = dma_fence_get(job->done_fence);
+
+	if (job->ops->armed_submit)
+		job->ops->armed_submit(job);
+
+	nouveau_job_fence_attach(job);
+
+	/* Set job state before pushing the job to the scheduler,
+	 * such that we do not overwrite the job state set in run().
+	 */
+	job->state = NOUVEAU_JOB_SUBMIT_SUCCESS;
+
+	drm_sched_entity_push_job(&job->base);
+
+	mutex_unlock(&entity->mutex);
+
+	if (done_fence) {
+		dma_fence_wait(done_fence, true);
+		dma_fence_put(done_fence);
+	}
+
+	return 0;
+
+err_cleanup:
+	mutex_unlock(&entity->mutex);
+	nouveau_job_fence_attach_cleanup(job);
+err:
+	job->state = NOUVEAU_JOB_SUBMIT_FAILED;
+	return ret;
+}
+
+bool
+nouveau_sched_entity_qwork(struct nouveau_sched_entity *entity,
+			   struct work_struct *work)
+{
+	return queue_work(entity->sched_wq, work);
+}
+
+static struct dma_fence *
+nouveau_job_run(struct nouveau_job *job)
+{
+	struct dma_fence *fence;
+
+	fence = job->ops->run(job);
+	if (IS_ERR(fence))
+		job->state = NOUVEAU_JOB_RUN_FAILED;
+	else
+		job->state = NOUVEAU_JOB_RUN_SUCCESS;
+
+	return fence;
+}
+
+static struct dma_fence *
+nouveau_sched_run_job(struct drm_sched_job *sched_job)
+{
+	struct nouveau_job *job = to_nouveau_job(sched_job);
+
+	return nouveau_job_run(job);
+}
+
+static enum drm_gpu_sched_stat
+nouveau_sched_timedout_job(struct drm_sched_job *sched_job)
+{
+	struct nouveau_job *job = to_nouveau_job(sched_job);
+
+	NV_PRINTK(warn, job->cli, "Job timed out.\n");
+
+	if (job->ops->timeout)
+		return job->ops->timeout(job);
+
+	return DRM_GPU_SCHED_STAT_ENODEV;
+}
+
+static void
+nouveau_sched_free_job(struct drm_sched_job *sched_job)
+{
+	struct nouveau_job *job = to_nouveau_job(sched_job);
+
+	nouveau_job_fini(job);
+}
+
+int nouveau_sched_entity_init(struct nouveau_sched_entity *entity,
+			      struct drm_gpu_scheduler *sched,
+			      struct workqueue_struct *sched_wq)
+{
+	mutex_init(&entity->mutex);
+	spin_lock_init(&entity->job.list.lock);
+	INIT_LIST_HEAD(&entity->job.list.head);
+	init_waitqueue_head(&entity->job.wq);
+
+	entity->sched_wq = sched_wq;
+	return drm_sched_entity_init(&entity->base,
+				     DRM_SCHED_PRIORITY_NORMAL,
+				     &sched, 1, NULL);
+}
+
+void
+nouveau_sched_entity_fini(struct nouveau_sched_entity *entity)
+{
+	drm_sched_entity_destroy(&entity->base);
+}
+
+static const struct drm_sched_backend_ops nouveau_sched_ops = {
+	.run_job = nouveau_sched_run_job,
+	.timedout_job = nouveau_sched_timedout_job,
+	.free_job = nouveau_sched_free_job,
+};
+
+int nouveau_sched_init(struct nouveau_drm *drm)
+{
+	struct drm_gpu_scheduler *sched = &drm->sched;
+	long job_hang_limit = msecs_to_jiffies(NOUVEAU_SCHED_JOB_TIMEOUT_MS);
+
+	drm->sched_wq = create_singlethread_workqueue("nouveau_sched_wq");
+	if (!drm->sched_wq)
+		return -ENOMEM;
+
+	return drm_sched_init(sched, &nouveau_sched_ops,
+			      NOUVEAU_SCHED_HW_SUBMISSIONS, 0, job_hang_limit,
+			      NULL, NULL, "nouveau_sched", drm->dev->dev);
+}
+
+void nouveau_sched_fini(struct nouveau_drm *drm)
+{
+	destroy_workqueue(drm->sched_wq);
+	drm_sched_fini(&drm->sched);
+}

drivers/gpu/drm/nouveau/nouveau_sched.h

+/* SPDX-License-Identifier: MIT */
+
+#ifndef NOUVEAU_SCHED_H
+#define NOUVEAU_SCHED_H
+
+#include <linux/types.h>
+
+#include <drm/drm_exec.h>
+#include <drm/gpu_scheduler.h>
+
+#include "nouveau_drv.h"
+
+#define to_nouveau_job(sched_job)		\
+		container_of((sched_job), struct nouveau_job, base)
+
+struct nouveau_job_ops;
+
+enum nouveau_job_state {
+	NOUVEAU_JOB_UNINITIALIZED = 0,
+	NOUVEAU_JOB_INITIALIZED,
+	NOUVEAU_JOB_SUBMIT_SUCCESS,
+	NOUVEAU_JOB_SUBMIT_FAILED,
+	NOUVEAU_JOB_RUN_SUCCESS,
+	NOUVEAU_JOB_RUN_FAILED,
+};
+
+struct nouveau_job_args {
+	struct drm_file *file_priv;
+	struct nouveau_sched_entity *sched_entity;
+
+	enum dma_resv_usage resv_usage;
+	bool sync;
+
+	struct {
+		struct drm_nouveau_sync *s;
+		u32 count;
+	} in_sync;
+
+	struct {
+		struct drm_nouveau_sync *s;
+		u32 count;
+	} out_sync;
+
+	struct nouveau_job_ops *ops;
+};
+
+struct nouveau_job {
+	struct drm_sched_job base;
+
+	enum nouveau_job_state state;
+
+	struct nouveau_sched_entity *entity;
+
+	struct drm_file *file_priv;
+	struct nouveau_cli *cli;
+
+	struct drm_exec exec;
+	enum dma_resv_usage resv_usage;
+	struct dma_fence *done_fence;
+
+	bool sync;
+
+	struct {
+		struct drm_nouveau_sync *data;
+		u32 count;
+	} in_sync;
+
+	struct {
+		struct drm_nouveau_sync *data;
+		struct drm_syncobj **objs;
+		struct dma_fence_chain **chains;
+		u32 count;
+	} out_sync;
+
+	struct nouveau_job_ops {
+		/* If .submit() returns without any error, it is guaranteed that
+		 * armed_submit() is called.
+		 */
+		int (*submit)(struct nouveau_job *);
+		void (*armed_submit)(struct nouveau_job *);
+		struct dma_fence *(*run)(struct nouveau_job *);
+		void (*free)(struct nouveau_job *);
+		enum drm_gpu_sched_stat (*timeout)(struct nouveau_job *);
+	} *ops;
+};
+
+int nouveau_job_ucopy_syncs(struct nouveau_job_args *args,
+			    u32 inc, u64 ins,
+			    u32 outc, u64 outs);
+
+int nouveau_job_init(struct nouveau_job *job,
+		     struct nouveau_job_args *args);
+void nouveau_job_free(struct nouveau_job *job);
+
+int nouveau_job_submit(struct nouveau_job *job);
+void nouveau_job_fini(struct nouveau_job *job);
+
+#define to_nouveau_sched_entity(entity)		\
+		container_of((entity), struct nouveau_sched_entity, base)
+
+struct nouveau_sched_entity {
+	struct drm_sched_entity base;
+	struct mutex mutex;
+
+	struct workqueue_struct *sched_wq;
+
+	struct {
+		struct {
+			struct list_head head;
+			spinlock_t lock;
+		} list;
+		struct wait_queue_head wq;
+	} job;
+};
+
+int nouveau_sched_entity_init(struct nouveau_sched_entity *entity,
+			      struct drm_gpu_scheduler *sched,
+			      struct workqueue_struct *sched_wq);
+void nouveau_sched_entity_fini(struct nouveau_sched_entity *entity);
+
+bool nouveau_sched_entity_qwork(struct nouveau_sched_entity *entity,
+				struct work_struct *work);
+
+int nouveau_sched_init(struct nouveau_drm *drm);
+void nouveau_sched_fini(struct nouveau_drm *drm);
+
+#endif

drivers/gpu/drm/nouveau/nouveau_uvmm.h

+/* SPDX-License-Identifier: MIT */
+
+#ifndef __NOUVEAU_UVMM_H__
+#define __NOUVEAU_UVMM_H__
+
+#include <drm/drm_gpuva_mgr.h>
+
+#include "nouveau_drv.h"
+
+struct nouveau_uvmm {
+	struct nouveau_vmm vmm;
+	struct drm_gpuva_manager umgr;
+	struct maple_tree region_mt;
+	struct mutex mutex;
+	struct dma_resv resv;
+
+	u64 kernel_managed_addr;
+	u64 kernel_managed_size;
+
+	bool disabled;
+};
+
+struct nouveau_uvma_region {
+	struct nouveau_uvmm *uvmm;
+
+	struct {
+		u64 addr;
+		u64 range;
+	} va;
+
+	struct kref kref;
+
+	struct completion complete;
+	bool dirty;
+};
+
+struct nouveau_uvma {
+	struct drm_gpuva va;
+
+	struct nouveau_uvmm *uvmm;
+	struct nouveau_uvma_region *region;
+
+	u8 kind;
+};
+
+struct nouveau_uvmm_bind_job {
+	struct nouveau_job base;
+
+	struct kref kref;
+	struct list_head entry;
+	struct work_struct work;
+	struct completion complete;
+
+	/* struct bind_job_op */
+	struct list_head ops;
+};
+
+struct nouveau_uvmm_bind_job_args {
+	struct drm_file *file_priv;
+	struct nouveau_sched_entity *sched_entity;
+
+	unsigned int flags;
+
+	struct {
+		struct drm_nouveau_sync *s;
+		u32 count;
+	} in_sync;
+
+	struct {
+		struct drm_nouveau_sync *s;
+		u32 count;
+	} out_sync;
+
+	struct {
+		struct drm_nouveau_vm_bind_op *s;
+		u32 count;
+	} op;
+};
+
+#define to_uvmm_bind_job(job) container_of((job), struct nouveau_uvmm_bind_job, base)
+
+#define uvmm_from_mgr(x) container_of((x), struct nouveau_uvmm, umgr)
+#define uvma_from_va(x) container_of((x), struct nouveau_uvma, va)
+
+int nouveau_uvmm_init(struct nouveau_uvmm *uvmm, struct nouveau_cli *cli,
+		      u64 kernel_managed_addr, u64 kernel_managed_size);
+void nouveau_uvmm_fini(struct nouveau_uvmm *uvmm);
+
+void nouveau_uvmm_bo_map_all(struct nouveau_bo *nvbov, struct nouveau_mem *mem);
+void nouveau_uvmm_bo_unmap_all(struct nouveau_bo *nvbo);
+
+int nouveau_uvmm_ioctl_vm_init(struct drm_device *dev, void __user *data,
+			       struct drm_file *file_priv);
+
+int nouveau_uvmm_ioctl_vm_bind(struct drm_device *dev, void __user *data,
+			       struct drm_file *file_priv);
+
+static inline void nouveau_uvmm_lock(struct nouveau_uvmm *uvmm)
+{
+	mutex_lock(&uvmm->mutex);
+}
+
+static inline void nouveau_uvmm_unlock(struct nouveau_uvmm *uvmm)
+{
+	mutex_unlock(&uvmm->mutex);
+}
+
+#endif