Make use of the scheduler's submission limit and scheduler job's submission unit count to account for the actual size of a job, such that we fill up the ring efficiently. Signed-off-by: Danilo Krummrich <dakr@xxxxxxxxxx> --- This patch is based on Matt's scheduler work [1] and [2]. [1] https://lore.kernel.org/dri-devel/20230919050155.2647172-1-matthew.brost@xxxxxxxxx/ [2] https://lore.kernel.org/nouveau/20230924224555.15595-1-dakr@xxxxxxxxxx/T/#u --- drivers/gpu/drm/nouveau/nouveau_abi16.c | 3 ++- drivers/gpu/drm/nouveau/nouveau_drm.c | 2 +- drivers/gpu/drm/nouveau/nouveau_exec.c | 3 ++- drivers/gpu/drm/nouveau/nouveau_sched.c | 9 ++++----- drivers/gpu/drm/nouveau/nouveau_sched.h | 3 ++- 5 files changed, 11 insertions(+), 9 deletions(-) diff --git a/drivers/gpu/drm/nouveau/nouveau_abi16.c b/drivers/gpu/drm/nouveau/nouveau_abi16.c index f8e59cfb1d34..207945700c94 100644 --- a/drivers/gpu/drm/nouveau/nouveau_abi16.c +++ b/drivers/gpu/drm/nouveau/nouveau_abi16.c @@ -316,7 +316,8 @@ nouveau_abi16_ioctl_channel_alloc(ABI16_IOCTL_ARGS) if (ret) goto done; - ret = nouveau_sched_init(&chan->sched, drm, drm->sched_wq); + ret = nouveau_sched_init(&chan->sched, drm, drm->sched_wq, + chan->chan->dma.ib_max); if (ret) goto done; diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c b/drivers/gpu/drm/nouveau/nouveau_drm.c index dbe1f08ac540..2b695c521821 100644 --- a/drivers/gpu/drm/nouveau/nouveau_drm.c +++ b/drivers/gpu/drm/nouveau/nouveau_drm.c @@ -317,7 +317,7 @@ nouveau_cli_init(struct nouveau_drm *drm, const char *sname, * locks which indirectly or directly are held for allocations * elsewhere. */ - ret = nouveau_sched_init(&cli->sched, drm, NULL); + ret = nouveau_sched_init(&cli->sched, drm, NULL, 1); if (ret) goto done; diff --git a/drivers/gpu/drm/nouveau/nouveau_exec.c b/drivers/gpu/drm/nouveau/nouveau_exec.c index 5a70fcf571f7..5d1533b413d3 100644 --- a/drivers/gpu/drm/nouveau/nouveau_exec.c +++ b/drivers/gpu/drm/nouveau/nouveau_exec.c @@ -253,10 +253,11 @@ nouveau_exec_job_init(struct nouveau_exec_job **pjob, } } + args.file_priv = __args->file_priv; job->chan = __args->chan; args.sched = __args->sched; - args.file_priv = __args->file_priv; + args.submission_units = job->push.count; args.in_sync.count = __args->in_sync.count; args.in_sync.s = __args->in_sync.s; diff --git a/drivers/gpu/drm/nouveau/nouveau_sched.c b/drivers/gpu/drm/nouveau/nouveau_sched.c index 0dbd0ea6d02e..6b32fd4eefb0 100644 --- a/drivers/gpu/drm/nouveau/nouveau_sched.c +++ b/drivers/gpu/drm/nouveau/nouveau_sched.c @@ -12,7 +12,6 @@ #include "nouveau_abi16.h" #include "nouveau_sched.h" -#define NOUVEAU_SCHED_HW_SUBMISSIONS 1 #define NOUVEAU_SCHED_JOB_TIMEOUT_MS 10000 int @@ -77,10 +76,10 @@ nouveau_job_init(struct nouveau_job *job, ret = -ENOMEM; goto err_free_objs; } - } - ret = drm_sched_job_init(&job->base, &sched->entity, 1, NULL); + ret = drm_sched_job_init(&job->base, &sched->entity, + args->submission_units, NULL); if (ret) goto err_free_chains; @@ -387,7 +386,7 @@ static const struct drm_sched_backend_ops nouveau_sched_ops = { }; int nouveau_sched_init(struct nouveau_sched *sched, struct nouveau_drm *drm, - struct workqueue_struct *wq) + struct workqueue_struct *wq, u32 submission_limit) { struct drm_gpu_scheduler *drm_sched = &sched->base; struct drm_sched_entity *entity = &sched->entity; @@ -404,7 +403,7 @@ int nouveau_sched_init(struct nouveau_sched *sched, struct nouveau_drm *drm, } ret = drm_sched_init(drm_sched, &nouveau_sched_ops, wq, - NOUVEAU_SCHED_HW_SUBMISSIONS, 0, job_hang_limit, + submission_limit, 0, job_hang_limit, NULL, NULL, "nouveau_sched", DRM_SCHED_POLICY_SINGLE_ENTITY, drm->dev->dev); if (ret) diff --git a/drivers/gpu/drm/nouveau/nouveau_sched.h b/drivers/gpu/drm/nouveau/nouveau_sched.h index 026f33d9b70c..c1d603759914 100644 --- a/drivers/gpu/drm/nouveau/nouveau_sched.h +++ b/drivers/gpu/drm/nouveau/nouveau_sched.h @@ -27,6 +27,7 @@ enum nouveau_job_state { struct nouveau_job_args { struct drm_file *file_priv; struct nouveau_sched *sched; + u32 submission_units; enum dma_resv_usage resv_usage; bool sync; @@ -112,7 +113,7 @@ struct nouveau_sched { }; int nouveau_sched_init(struct nouveau_sched *sched, struct nouveau_drm *drm, - struct workqueue_struct *wq); + struct workqueue_struct *wq, u32 submission_limit); void nouveau_sched_fini(struct nouveau_sched *sched); #endif -- 2.41.0