drm/nouveau: implement 1:1 scheduler - entity relationship

Recent patches to the DRM scheduler [1][2] allow for a variable number
of run-queues and add support for (shared) workqueues rather than
dedicated kthreads per scheduler. This allows us to create a 1:1
relationship between a GPU scheduler and a scheduler entity, in order to
properly support firmware schedulers being able to handle an arbitrary
amount of dynamically allocated command ring buffers. This perfectly
matches Nouveau's needs, hence make use of it.

Topology wise we create one scheduler instance per client (handling
VM_BIND jobs) and one scheduler instance per channel (handling EXEC
jobs).

All channel scheduler instances share a workqueue, but every client
scheduler instance has a dedicated workqueue. The latter is required to
ensure that for VM_BIND job's free_job() work and run_job() work can
always run concurrently and hence, free_job() work can never stall
run_job() work. For EXEC jobs we don't have this requirement, since EXEC
job's free_job() does not require to take any locks which indirectly or
directly are held for allocations elsewhere.

[1] https://lore.kernel.org/all/8f53f7ef-7621-4f0b-bdef-d8d20bc497ff@redhat.com/T/
[2] https://lore.kernel.org/all/20231031032439.1558703-1-matthew.brost@intel.com/T/

Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231114002728.3491-1-dakr@redhat.com

1 files changed, 21 insertions, 48 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_uvmm.c b/drivers/gpu/drm/nouveau/nouveau_uvmm.c
index 90e6732ee9b5..46493fcb82fd 100644
--- a/drivers/gpu/drm/nouveau/nouveau_uvmm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_uvmm.c
@@ -963,6 +963,12 @@ nouveau_uvmm_bind_job_free(struct kref *kref)
 {
 	struct nouveau_uvmm_bind_job *job =
 		container_of(kref, struct nouveau_uvmm_bind_job, kref);
+	struct bind_job_op *op, *next;
+
+	list_for_each_op_safe(op, next, &job->ops) {
+		list_del(&op->entry);
+		kfree(op);
+	}
 
 	nouveau_job_free(&job->base);
 	kfree(job);
@@ -1004,14 +1010,16 @@ bind_validate_op(struct nouveau_job *job,
 static void
 bind_validate_map_sparse(struct nouveau_job *job, u64 addr, u64 range)
 {
-	struct nouveau_uvmm_bind_job *bind_job;
-	struct nouveau_sched_entity *entity = job->entity;
+	struct nouveau_sched *sched = job->sched;
+	struct nouveau_job *__job;
 	struct bind_job_op *op;
 	u64 end = addr + range;
 
 again:
-	spin_lock(&entity->job.list.lock);
-	list_for_each_entry(bind_job, &entity->job.list.head, entry) {
+	spin_lock(&sched->job.list.lock);
+	list_for_each_entry(__job, &sched->job.list.head, entry) {
+		struct nouveau_uvmm_bind_job *bind_job = to_uvmm_bind_job(__job);
+
 		list_for_each_op(op, &bind_job->ops) {
 			if (op->op == OP_UNMAP) {
 				u64 op_addr = op->va.addr;
@@ -1019,7 +1027,7 @@ again:
 
 				if (!(end <= op_addr || addr >= op_end)) {
 					nouveau_uvmm_bind_job_get(bind_job);
-					spin_unlock(&entity->job.list.lock);
+					spin_unlock(&sched->job.list.lock);
 					wait_for_completion(&bind_job->complete);
 					nouveau_uvmm_bind_job_put(bind_job);
 					goto again;
@@ -1027,7 +1035,7 @@ again:
 			}
 		}
 	}
-	spin_unlock(&entity->job.list.lock);
+	spin_unlock(&sched->job.list.lock);
 }
 
 static int
@@ -1183,7 +1191,6 @@ nouveau_uvmm_bind_job_submit(struct nouveau_job *job,
 {
 	struct nouveau_uvmm *uvmm = nouveau_cli_uvmm(job->cli);
 	struct nouveau_uvmm_bind_job *bind_job = to_uvmm_bind_job(job);
-	struct nouveau_sched_entity *entity = job->entity;
 	struct drm_exec *exec = &vme->exec;
 	struct bind_job_op *op;
 	int ret;
@@ -1380,10 +1387,6 @@ nouveau_uvmm_bind_job_submit(struct nouveau_job *job,
 	}
 	nouveau_uvmm_unlock(uvmm);
 
-	spin_lock(&entity->job.list.lock);
-	list_add(&bind_job->entry, &entity->job.list.head);
-	spin_unlock(&entity->job.list.lock);
-
 	return 0;
 
 unwind_continue:
@@ -1466,14 +1469,11 @@ out:
 }
 
 static void
-nouveau_uvmm_bind_job_free_work_fn(struct work_struct *work)
+nouveau_uvmm_bind_job_cleanup(struct nouveau_job *job)
 {
-	struct nouveau_uvmm_bind_job *bind_job =
-		container_of(work, struct nouveau_uvmm_bind_job, work);
-	struct nouveau_job *job = &bind_job->base;
+	struct nouveau_uvmm_bind_job *bind_job = to_uvmm_bind_job(job);
 	struct nouveau_uvmm *uvmm = nouveau_cli_uvmm(job->cli);
-	struct nouveau_sched_entity *entity = job->entity;
-	struct bind_job_op *op, *next;
+	struct bind_job_op *op;
 
 	list_for_each_op(op, &bind_job->ops) {
 		struct drm_gem_object *obj = op->gem.obj;
@@ -1525,38 +1525,17 @@ nouveau_uvmm_bind_job_free_work_fn(struct work_struct *work)
 			drm_gem_object_put(obj);
 	}
 
-	spin_lock(&entity->job.list.lock);
-	list_del(&bind_job->entry);
-	spin_unlock(&entity->job.list.lock);
-
+	nouveau_job_done(job);
 	complete_all(&bind_job->complete);
-	wake_up(&entity->job.wq);
-
-	/* Remove and free ops after removing the bind job from the job list to
-	 * avoid races against bind_validate_map_sparse().
-	 */
-	list_for_each_op_safe(op, next, &bind_job->ops) {
-		list_del(&op->entry);
-		kfree(op);
-	}
 
 	nouveau_uvmm_bind_job_put(bind_job);
 }
 
-static void
-nouveau_uvmm_bind_job_free_qwork(struct nouveau_job *job)
-{
-	struct nouveau_uvmm_bind_job *bind_job = to_uvmm_bind_job(job);
-	struct nouveau_sched_entity *entity = job->entity;
-
-	nouveau_sched_entity_qwork(entity, &bind_job->work);
-}
-
 static struct nouveau_job_ops nouveau_bind_job_ops = {
 	.submit = nouveau_uvmm_bind_job_submit,
 	.armed_submit = nouveau_uvmm_bind_job_armed_submit,
 	.run = nouveau_uvmm_bind_job_run,
-	.free = nouveau_uvmm_bind_job_free_qwork,
+	.free = nouveau_uvmm_bind_job_cleanup,
 };
 
 static int
@@ -1617,7 +1596,6 @@ nouveau_uvmm_bind_job_init(struct nouveau_uvmm_bind_job **pjob,
 		return ret;
 
 	INIT_LIST_HEAD(&job->ops);
-	INIT_LIST_HEAD(&job->entry);
 
 	for (i = 0; i < __args->op.count; i++) {
 		ret = bind_job_op_from_uop(&op, &__args->op.s[i]);
@@ -1628,9 +1606,8 @@ nouveau_uvmm_bind_job_init(struct nouveau_uvmm_bind_job **pjob,
 	}
 
 	init_completion(&job->complete);
-	INIT_WORK(&job->work, nouveau_uvmm_bind_job_free_work_fn);
 
-	args.sched_entity = __args->sched_entity;
+	args.sched = __args->sched;
 	args.file_priv = __args->file_priv;
 
 	args.in_sync.count = __args->in_sync.count;
@@ -1758,7 +1735,7 @@ nouveau_uvmm_ioctl_vm_bind(struct drm_device *dev,
 	if (ret)
 		return ret;
 
-	args.sched_entity = &cli->sched_entity;
+	args.sched = &cli->sched;
 	args.file_priv = file_priv;
 
 	ret = nouveau_uvmm_vm_bind(&args);
@@ -1910,12 +1887,8 @@ nouveau_uvmm_fini(struct nouveau_uvmm *uvmm)
 	MA_STATE(mas, &uvmm->region_mt, 0, 0);
 	struct nouveau_uvma_region *reg;
 	struct nouveau_cli *cli = uvmm->vmm.cli;
-	struct nouveau_sched_entity *entity = &cli->sched_entity;
 	struct drm_gpuva *va, *next;
 
-	rmb(); /* for list_empty to work without lock */
-	wait_event(entity->job.wq, list_empty(&entity->job.list.head));
-
 	nouveau_uvmm_lock(uvmm);
 	drm_gpuvm_for_each_va_safe(va, next, &uvmm->base) {
 		struct nouveau_uvma *uvma = uvma_from_va(va);