drm/amdgpu: set compute queue priority at mqd_init
Subsystems (4)
PATCH v3 1/4 drm/amdgpu: set compute queue priority at mqd_init
We were changing compute ring priority while rings were being used before every job submission which is not recommended. This patch sets compute queue priority at mqd initialization for gfx8, gfx9 and gfx10.
Policy: make queue 0 of each pipe as high priority compute queue
High/normal priority compute sched lists are generated from set of high/normal priority compute queues. At context creation, entity of compute queue get a sched list from high or normal priority depending on ctx->priority
--- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 4 -- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 59 +++++++++++++++++++++--- drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 8 ++++ drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h | 15 +++++- drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 6 --- drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 1 + drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 19 ++++++++ drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 23 +++++++-- drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 20 ++++++++ 9 files changed, 134 insertions(+), 21 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +0 -4
@@ -1205,7 +1205,6 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,struct amdgpu_fpriv *fpriv = p->filp->driver_priv;struct drm_sched_entity *entity = p->entity;enum drm_sched_priority priority; -struct amdgpu_ring *ring;struct amdgpu_bo_list_entry *e;struct amdgpu_job *job;uint64_t seq;
@@ -1258,9 +1257,6 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, priority = job->base.s_priority;drm_sched_entity_push_job (&job->base, entity); - ring =to_amdgpu_ring (entity->rq->sched); -amdgpu_ring_priority_get (ring, priority); -amdgpu_vm_move_to_lru_tail (p->adev, &fpriv->vm);ttm_eu_fence_buffer_objects (&p->ticket, &p->validated, p->fence);
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +52 -7
92 lines not on this page · raw message
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +8 -0
@@ -192,6 +192,14 @@ static bool amdgpu_gfx_is_multipipe_capable(struct amdgpu_device *adev)return adev->gfx.mec.num_mec >1 ; } +bool amdgpu_gfx_is_high_priority_compute_queue (struct amdgpu_device *adev, +int queue) +{ +/* Policy: make queue 0 of each pipe as high priority compute queue */ +return (queue ==0 ); + +} +void amdgpu_gfx_compute_queue_acquire (struct amdgpu_device *adev) {int i, queue, pipe, mec;
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h +14 -1
@@ -40,6 +40,16 @@#define AMDGPU_MAX_GFX_QUEUES KGD_MAX_QUEUES#define AMDGPU_MAX_COMPUTE_QUEUES KGD_MAX_QUEUES +enum gfx_pipe_priority { +AMDGPU_GFX_PIPE_PRIO_LOW =0 , +AMDGPU_GFX_PIPE_PRIO_NORMAL , +AMDGPU_GFX_PIPE_PRIO_HIGH , +AMDGPU_GFX_PIPE_PRIO_MAX + +}; + +#define AMDGPU_GFX_QUEUE_PRIORITY_MINIMUM 0 +#define AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM 15struct amdgpu_mec {struct amdgpu_bo *hpd_eop_obj;
@@ -280,8 +290,9 @@ struct amdgpu_gfx {uint32_t num_gfx_sched;unsigned num_gfx_rings;struct amdgpu_ring compute_ring[AMDGPU_MAX_COMPUTE_RINGS ]; +struct drm_gpu_scheduler **compute_prio_sched[AMDGPU_GFX_PIPE_PRIO_MAX ];struct drm_gpu_scheduler *compute_sched[AMDGPU_MAX_COMPUTE_RINGS ]; -uint32_t num_compute_sched; +uint32_t num_compute_sched[DRM_SCHED_PRIORITY_MAX ];unsigned num_compute_rings;struct amdgpu_irq_src eop_irq;struct amdgpu_irq_src priv_reg_irq;
@@ -363,6 +374,8 @@ void amdgpu_gfx_bit_to_mec_queue(struct amdgpu_device *adev, int bit, int *mec, int *pipe, int *queue);bool amdgpu_gfx_is_mec_queue_enabled (struct amdgpu_device *adev,int mec,int pipe,int queue); +bool amdgpu_gfx_is_high_priority_compute_queue (struct amdgpu_device *adev, +int queue);int amdgpu_gfx_me_queue_to_bit (struct amdgpu_device *adev,int me,int pipe,int queue); void amdgpu_gfx_bit_to_me_queue(struct amdgpu_device *adev, int bit,
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +0 -6
@@ -117,12 +117,10 @@ void amdgpu_job_free_resources(struct amdgpu_job *job)static void amdgpu_job_free_cb (struct drm_sched_job *s_job) { -struct amdgpu_ring *ring =to_amdgpu_ring (s_job->sched);struct amdgpu_job *job =to_amdgpu_job (s_job);drm_sched_job_cleanup (s_job); -amdgpu_ring_priority_put (ring, s_job->s_priority);dma_fence_put (job->fence);amdgpu_sync_free (&job->sync);amdgpu_sync_free (&job->sched_sync);
@@ -143,7 +141,6 @@ int amdgpu_job_submit(struct amdgpu_job *job, struct drm_sched_entity *entity, void *owner, struct dma_fence **f) { enum drm_sched_priority priority; -struct amdgpu_ring *ring;int r;if (!f)
@@ -158,9 +155,6 @@ int amdgpu_job_submit(struct amdgpu_job *job, struct drm_sched_entity *entity, priority = job->base.s_priority;drm_sched_entity_push_job (&job->base, entity); - ring =to_amdgpu_ring (entity->rq->sched); -amdgpu_ring_priority_get (ring, priority); -return 0 ; }
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h +1 -0
@@ -222,6 +222,7 @@ struct amdgpu_ring {struct mutex priority_mutex;/* protected by priority_mutex */ int priority; +bool high_priority;#if defined(CONFIG_DEBUG_FS )struct dentry *ent;
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +19 -0
@@ -3211,6 +3211,22 @@ static int gfx_v10_0_cp_async_gfx_ring_resume(struct amdgpu_device *adev)return r; } +static void gfx_v10_0_compute_mqd_set_priority (struct amdgpu_ring *ring,struct v10_compute_mqd *mqd) +{ +struct amdgpu_device *adev = ring->adev; + +if (ring->funcs->type ==AMDGPU_RING_TYPE_COMPUTE ) { +if (amdgpu_gfx_is_high_priority_compute_queue (adev, ring->queue)) { + mqd->cp_hqd_pipe_priority =AMDGPU_GFX_PIPE_PRIO_HIGH ; + ring->high_priority = true; + mqd->cp_hqd_queue_priority = +AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM ; + } +else + ring->high_priority = false; + } +} +static int gfx_v10_0_compute_mqd_init (struct amdgpu_ring *ring) {struct amdgpu_device *adev = ring->adev;
@@ -3336,6 +3352,9 @@ static int gfx_v10_0_compute_mqd_init(struct amdgpu_ring *ring) tmp =REG_SET_FIELD (tmp,CP_HQD_IB_CONTROL ,MIN_IB_AVAIL_SIZE ,3 ); mqd->cp_hqd_ib_control = tmp; +/* set static priority for a queue/ring */ +gfx_v10_0_compute_mqd_set_priority (ring, mqd); +/* map_queues packet doesn't need activate the queue, * so only kiq need set this field. */
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +20 -3
@@ -4430,6 +4430,22 @@ static int gfx_v8_0_deactivate_hqd(struct amdgpu_device *adev, u32 req)return r; } +static void gfx_v8_0_mqd_set_priority (struct amdgpu_ring *ring,struct vi_mqd *mqd) +{ +struct amdgpu_device *adev = ring->adev; + +if (ring->funcs->type ==AMDGPU_RING_TYPE_COMPUTE ) { +if (amdgpu_gfx_is_high_priority_compute_queue (adev, ring->queue)) { + mqd->cp_hqd_pipe_priority =AMDGPU_GFX_PIPE_PRIO_HIGH ; + ring->high_priority = true; + mqd->cp_hqd_queue_priority = +AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM ; + } +else + ring->high_priority = false; + } +} +static int gfx_v8_0_mqd_init (struct amdgpu_ring *ring) {struct amdgpu_device *adev = ring->adev;
@@ -4553,9 +4569,6 @@ static int gfx_v8_0_mqd_init(struct amdgpu_ring *ring)/* defaults */ mqd->cp_hqd_eop_rptr =RREG32 (mmCP_HQD_EOP_RPTR); mqd->cp_hqd_eop_wptr =RREG32 (mmCP_HQD_EOP_WPTR); - mqd->cp_hqd_pipe_priority =RREG32 (mmCP_HQD_PIPE_PRIORITY); - mqd->cp_hqd_queue_priority =RREG32 (mmCP_HQD_QUEUE_PRIORITY); - mqd->cp_hqd_quantum =RREG32 (mmCP_HQD_QUANTUM); mqd->cp_hqd_ctx_save_base_addr_lo =RREG32 (mmCP_HQD_CTX_SAVE_BASE_ADDR_LO); mqd->cp_hqd_ctx_save_base_addr_hi =RREG32 (mmCP_HQD_CTX_SAVE_BASE_ADDR_HI); mqd->cp_hqd_cntl_stack_offset =RREG32 (mmCP_HQD_CNTL_STACK_OFFSET);
@@ -4567,6 +4580,10 @@ static int gfx_v8_0_mqd_init(struct amdgpu_ring *ring) mqd->cp_hqd_eop_wptr_mem =RREG32 (mmCP_HQD_EOP_WPTR_MEM); mqd->cp_hqd_eop_dones =RREG32 (mmCP_HQD_EOP_DONES); +/* set static priority for a queue/ring */ +gfx_v8_0_mqd_set_priority (ring, mqd); + mqd->cp_hqd_quantum =RREG32 (mmCP_HQD_QUANTUM); +/* map_queues packet doesn't need activate the queue, * so only kiq need set this field. */
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +20 -0
@@ -3314,6 +3314,22 @@ static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring)WREG32_SOC15_RLC (GC ,0 , mmRLC_CP_SCHEDULERS, tmp); } +static void gfx_v9_0_mqd_set_priority (struct amdgpu_ring *ring,struct v9_mqd *mqd) +{ +struct amdgpu_device *adev = ring->adev; + +if (ring->funcs->type ==AMDGPU_RING_TYPE_COMPUTE ) { +if (amdgpu_gfx_is_high_priority_compute_queue (adev, ring->queue)) { + mqd->cp_hqd_pipe_priority =AMDGPU_GFX_PIPE_PRIO_HIGH ; + ring->high_priority = true; + mqd->cp_hqd_queue_priority = +AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM ; + } +else + ring->high_priority = false; + } +} +static int gfx_v9_0_mqd_init (struct amdgpu_ring *ring) {struct amdgpu_device *adev = ring->adev;
@@ -3450,6 +3466,10 @@ static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring) tmp =REG_SET_FIELD (tmp,CP_HQD_IB_CONTROL ,MIN_IB_AVAIL_SIZE ,3 ); mqd->cp_hqd_ib_control = tmp; +/* set static priority for a queue/ring */ +gfx_v9_0_mqd_set_priority (ring, mqd); + mqd->cp_hqd_quantum =RREG32 (mmCP_HQD_QUANTUM); +/* map_queues packet doesn't need activate the queue, * so only kiq need set this field. */
--
2.25.0
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH v2 2/4 drm/scheduler: implement a function to modify sched list
implement drm_sched_entity_modify_sched() which can modify existing sched_list with a different one. This is going to be helpful when userspace changes priority of a ctx/entity then driver can switch to corresponding hw shced list for that priority
--- drivers/gpu/drm/scheduler/sched_entity.c | 19 +++++++++++++++++++ include/drm/gpu_scheduler.h | 4 ++++ 2 files changed, 23 insertions(+)
drivers/gpu/drm/scheduler/sched_entity.c +19 -0
@@ -83,6 +83,25 @@ int drm_sched_entity_init(struct drm_sched_entity *entity, }EXPORT_SYMBOL (drm_sched_entity_init); +/** +* drm_sched_entity_modify_sched - Modify sched of an entity +* +* @entity: scheduler entity to init +* @sched_list: the list of new drm scheds which will replace +* existing entity->sched_list +* @num_sched_list: number of drm sched in sched_list +*/ +void drm_sched_entity_modify_sched (struct drm_sched_entity *entity, +struct drm_gpu_scheduler **sched_list, +unsigned int num_sched_list) +{ +WARN_ON (!num_sched_list || !sched_list); + + entity->sched_list = sched_list; + entity->num_sched_list = num_sched_list; +} +EXPORT_SYMBOL (drm_sched_entity_modify_sched); +/** * drm_sched_entity_is_idle - Check if entity is idle *
include/drm/gpu_scheduler.h +4 -0
@@ -297,6 +297,10 @@ void drm_sched_fini(struct drm_gpu_scheduler *sched);int drm_sched_job_init (struct drm_sched_job *job,struct drm_sched_entity *entity,void *owner); +void drm_sched_entity_modify_sched (struct drm_sched_entity *entity, +struct drm_gpu_scheduler **sched_list, +unsigned int num_sched_list); +void drm_sched_job_cleanup (struct drm_sched_job *job);void drm_sched_wakeup (struct drm_gpu_scheduler *sched);void drm_sched_stop (struct drm_gpu_scheduler *sched,struct drm_sched_job *bad);
--
2.25.0
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH v3 3/4 drm/amdgpu: change hw sched list on ctx priority override
In reply to Nirmoy DasSwitch to appropriate sched list for an entity on priority override.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 30 +++++++++++++++++++++---- 1 file changed, 26 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +26 -4
@@ -522,6 +522,30 @@ struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,return fence; } +static void amdgpu_ctx_set_entity_priority (struct amdgpu_ctx *ctx, +struct amdgpu_ctx_entity *aentity, +int hw_ip,enum drm_sched_priority priority) +{ +struct amdgpu_device *adev = ctx->adev; +enum gfx_pipe_priority compute_priority; +struct drm_gpu_scheduler **scheds =NULL ; +unsigned num_scheds =0 ; + +switch (hw_ip) { +case AMDGPU_HW_IP_COMPUTE : + compute_priority = +amdgpu_ctx_sched_prio_to_compute_prio (priority); + scheds = adev->gfx.compute_prio_sched[compute_priority]; + num_scheds = adev->gfx.num_compute_sched[compute_priority]; +break ; +default : +return ; + } + +drm_sched_entity_modify_sched (&aentity->entity, scheds, num_scheds); +drm_sched_entity_set_priority (&aentity->entity, priority); +} +void amdgpu_ctx_priority_override (struct amdgpu_ctx *ctx,enum drm_sched_priority priority) {
@@ -534,13 +558,11 @@ void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx, ctx->init_priority : ctx->override_priority;for (i =0 ; i <AMDGPU_HW_IP_NUM ; ++i) {for (j =0 ; j < amdgpu_ctx_num_entities[i]; ++j) { -struct drm_sched_entity *entity; -if (!ctx->entities[i][j])continue ; - entity = &ctx->entities[i][j]->entity; -drm_sched_entity_set_priority (entity, ctx_prio); +amdgpu_ctx_set_entity_priority (ctx, ctx->entities[i][j], + i, ctx_prio); } } }
--
2.25.0
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 4/4 drm/amdgpu: remove unused functions
In reply to Nirmoy Dasamdgpu statically set priority for compute queues at initialization so remove all the functions responsible changing compute queue priority dynamically
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c | 70 ---------------- drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 7 -- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 99 ---------------------- drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 100 ----------------------- 4 files changed, 276 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c +0 -70
@@ -150,76 +150,6 @@ void amdgpu_ring_undo(struct amdgpu_ring *ring) ring->funcs->end_use (ring); } -/** -* amdgpu_ring_priority_put - restore a ring's priority -* -* @ring: amdgpu_ring structure holding the information -* @priority: target priority -* -* Release a request for executing at @priority -*/ -void amdgpu_ring_priority_put (struct amdgpu_ring *ring, -enum drm_sched_priority priority) -{ -int i; - -if (!ring->funcs->set_priority) -return ; - -if (atomic_dec_return (&ring->num_jobs[priority]) >0 ) -return ; - -/* no need to restore if the job is already at the lowest priority */ -if (priority ==DRM_SCHED_PRIORITY_NORMAL ) -return ; - -mutex_lock (&ring->priority_mutex); -/* something higher prio is executing, no need to decay */ -if (ring->priority > priority) - goto out_unlock; - -/* decay priority to the next level with a job available */ -for (i = priority; i >=DRM_SCHED_PRIORITY_MIN ; i--) { -if (i ==DRM_SCHED_PRIORITY_NORMAL - ||atomic_read (&ring->num_jobs[i])) { - ring->priority = i; - ring->funcs->set_priority (ring, i); -break ; - } - } - -out_unlock: -mutex_unlock (&ring->priority_mutex); -} - -/** -* amdgpu_ring_priority_get - change the ring's priority -* -* @ring: amdgpu_ring structure holding the information -* @priority: target priority -* -* Request a ring's priority to be raised to @priority (refcounted). -*/ -void amdgpu_ring_priority_get (struct amdgpu_ring *ring, -enum drm_sched_priority priority) -{ -if (!ring->funcs->set_priority) -return ; - -if (atomic_inc_return (&ring->num_jobs[priority]) <=0 ) -return ; - -mutex_lock (&ring->priority_mutex); -if (priority <= ring->priority) - goto out_unlock; - - ring->priority = priority; - ring->funcs->set_priority (ring, priority); - -out_unlock: -mutex_unlock (&ring->priority_mutex); -} -/** * amdgpu_ring_init - init driver ring struct. *
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h +0 -7
@@ -167,9 +167,6 @@ struct amdgpu_ring_funcs { uint32_t reg0, uint32_t reg1, uint32_t ref, uint32_t mask);void (*emit_tmz)(struct amdgpu_ring *ring,bool start); -/* priority functions */ -void (*set_priority) (struct amdgpu_ring *ring, -enum drm_sched_priority priority);/* Try to soft recover the ring to make the fence signal */ void (*soft_recovery)(struct amdgpu_ring *ring,unsigned vmid);int (*preempt_ib)(struct amdgpu_ring *ring);
@@ -259,10 +256,6 @@ void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count);void amdgpu_ring_generic_pad_ib (struct amdgpu_ring *ring,struct amdgpu_ib *ib);void amdgpu_ring_commit (struct amdgpu_ring *ring);void amdgpu_ring_undo (struct amdgpu_ring *ring); -void amdgpu_ring_priority_get (struct amdgpu_ring *ring, -enum drm_sched_priority priority); -void amdgpu_ring_priority_put (struct amdgpu_ring *ring, -enum drm_sched_priority priority);int amdgpu_ring_init (struct amdgpu_device *adev,struct amdgpu_ring *ring,unsigned ring_size,struct amdgpu_irq_src *irq_src,unsigned irq_type);
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +0 -99
111 lines not on this page · raw message
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +0 -100
112 lines not on this page · raw message
Signature · 6 lines
2.25.0 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx