[PATCH] drm/scheduler: fix setting the priorty for entities

[Dieter@xxxxxxxxxxxxx (Dieter Nützel)]

Am 03.08.2018 13:09, schrieb Christian König:
> Am 03.08.2018 um 03:08 schrieb Dieter Nützel:
>> Hello Christian, AMD guys,
>> 
>> this one _together_ with these series
>> [PATCH 1/7] drm/amdgpu: use new scheduler load balancing for VMs
>> https://lists.freedesktop.org/archives/amd-gfx/2018-August/024802.html
>> 
>> on top of
>> amd-staging-drm-next 53d5f1e4a6d9
>> 
>> freeze whole system (Intel Xeon X3470, RX580) during _first_ mouse 
>> move.
>> Same for sddm login or first move in KDE Plasma 5.
>> NO logs so far. - Expected?
> 
> Not even remotely, can you double check which patch from the "[PATCH
> 1/7] drm/amdgpu: use new scheduler load balancing for VMs" series is
> causing the issue?

Ups,

_both_ 'series' on top of

bf1fd52b0632 (origin/amd-staging-drm-next) drm/amdgpu: move gem 
definitions into amdgpu_gem header

works without a hitch.

But I have new (latest) µcode from openSUSE Tumbleweed.
kernel-firmware-20180730-35.1.src.rpm

Tested-by: Dieter Nützel <Dieter at nuetzel-hh.de>

Sorry about the noise.
Dieter


> Thanks,
> Christian.
> 
>> 
>> Greetings,
>> Dieter
>> 
>> Am 01.08.2018 16:27, schrieb Christian König:
>>> Since we now deal with multiple rq we need to update all of them, not
>>> just the current one.
>>> 
>>> Signed-off-by: Christian König <christian.koenig at amd.com>
>>> ---
>>>  drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c   |  3 +--
>>> Â drivers/gpu/drm/scheduler/gpu_scheduler.c | 36 
>>> ++++++++++++++++++++-----------
>>>  include/drm/gpu_scheduler.h               |  5 ++---
>>> Â 3 files changed, 26 insertions(+), 18 deletions(-)
>>> 
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>>> index df6965761046..9fcc14e2dfcf 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>>> @@ -407,12 +407,11 @@ void amdgpu_ctx_priority_override(struct 
>>> amdgpu_ctx *ctx,
>>> Â Â Â Â  for (i = 0; i < adev->num_rings; i++) {
>>> Â Â Â Â Â Â Â Â  ring = adev->rings[i];
>>> Â Â Â Â Â Â Â Â  entity = &ctx->rings[i].entity;
>>> -Â Â Â Â Â Â Â  rq = &ring->sched.sched_rq[ctx_prio];
>>> 
>>> Â Â Â Â Â Â Â Â  if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
>>> Â Â Â Â Â Â Â Â Â Â Â Â  continue;
>>> 
>>> -Â Â Â Â Â Â Â  drm_sched_entity_set_rq(entity, rq);
>>> +Â Â Â Â Â Â Â  drm_sched_entity_set_priority(entity, ctx_prio);
>>> Â Â Â Â  }
>>> Â }
>>> 
>>> diff --git a/drivers/gpu/drm/scheduler/gpu_scheduler.c
>>> b/drivers/gpu/drm/scheduler/gpu_scheduler.c
>>> index 05dc6ecd4003..85908c7f913e 100644
>>> --- a/drivers/gpu/drm/scheduler/gpu_scheduler.c
>>> +++ b/drivers/gpu/drm/scheduler/gpu_scheduler.c
>>> @@ -419,29 +419,39 @@ static void drm_sched_entity_clear_dep(struct
>>> dma_fence *f, struct dma_fence_cb
>>> Â }
>>> 
>>> Â /**
>>> - * drm_sched_entity_set_rq - Sets the run queue for an entity
>>> + * drm_sched_entity_set_rq_priority - helper for 
>>> drm_sched_entity_set_priority
>>> + */
>>> +static void drm_sched_entity_set_rq_priority(struct drm_sched_rq 
>>> **rq,
>>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  enum drm_sched_priority priority)
>>> +{
>>> +Â Â Â  *rq = &(*rq)->sched->sched_rq[priority];
>>> +}
>>> +
>>> +/**
>>> + * drm_sched_entity_set_priority - Sets priority of the entity
>>> Â  *
>>> Â  * @entity: scheduler entity
>>> - * @rq: scheduler run queue
>>> + * @priority: scheduler priority
>>> Â  *
>>> - * Sets the run queue for an entity and removes the entity from the 
>>> previous
>>> - * run queue in which was present.
>>> + * Update the priority of runqueus used for the entity.
>>> Â  */
>>> -void drm_sched_entity_set_rq(struct drm_sched_entity *entity,
>>> -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  struct drm_sched_rq *rq)
>>> +void drm_sched_entity_set_priority(struct drm_sched_entity *entity,
>>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  enum drm_sched_priority priority)
>>> Â {
>>> -Â Â Â  if (entity->rq == rq)
>>> -Â Â Â Â Â Â Â  return;
>>> -
>>> -Â Â Â  BUG_ON(!rq);
>>> +Â Â Â  unsigned int i;
>>> 
>>> Â Â Â Â  spin_lock(&entity->rq_lock);
>>> +
>>> +Â Â Â  for (i = 0; i < entity->num_rq_list; ++i)
>>> + drm_sched_entity_set_rq_priority(&entity->rq_list[i], priority);
>>> +
>>> Â Â Â Â  drm_sched_rq_remove_entity(entity->rq, entity);
>>> -Â Â Â  entity->rq = rq;
>>> -Â Â Â  drm_sched_rq_add_entity(rq, entity);
>>> +Â Â Â  drm_sched_entity_set_rq_priority(&entity->rq, priority);
>>> +Â Â Â  drm_sched_rq_add_entity(entity->rq, entity);
>>> +
>>> Â Â Â Â  spin_unlock(&entity->rq_lock);
>>> Â }
>>> -EXPORT_SYMBOL(drm_sched_entity_set_rq);
>>> +EXPORT_SYMBOL(drm_sched_entity_set_priority);
>>> 
>>> Â /**
>>> Â  * drm_sched_dependency_optimized
>>> diff --git a/include/drm/gpu_scheduler.h 
>>> b/include/drm/gpu_scheduler.h
>>> index 0c4cfe689d4c..22c0f88f7d8f 100644
>>> --- a/include/drm/gpu_scheduler.h
>>> +++ b/include/drm/gpu_scheduler.h
>>> @@ -298,9 +298,8 @@ void drm_sched_entity_fini(struct 
>>> drm_sched_entity *entity);
>>> Â void drm_sched_entity_destroy(struct drm_sched_entity *entity);
>>> Â void drm_sched_entity_push_job(struct drm_sched_job *sched_job,
>>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  struct drm_sched_entity *entity);
>>> -void drm_sched_entity_set_rq(struct drm_sched_entity *entity,
>>> -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  struct drm_sched_rq *rq);
>>> -
>>> +void drm_sched_entity_set_priority(struct drm_sched_entity *entity,
>>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  enum drm_sched_priority priority);
>>> Â struct drm_sched_fence *drm_sched_fence_create(
>>> Â Â Â Â  struct drm_sched_entity *s_entity, void *owner);
>>> Â void drm_sched_fence_scheduled(struct drm_sched_fence *fence);