Re: [RFC PATCH 2/2] drm/amdgpu: disable gpu load balancer for vcn jobs

[Nirmoy <nirmodas@xxxxxxx>]

On 3/11/20 7:03 PM, Christian König wrote:
> Am 11.03.20 um 18:18 schrieb Nirmoy Das:
> > VCN HW  doesn't support dynamic load balance on multiple
> > instances for a context. This modifies the entity's sched_list
> > to a sched_list consist of only one drm scheduler.
> >
> > Signed-off-by: Nirmoy Das <nirmoy.das@xxxxxxx>
> > ---
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c  |  1 +
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 25 +++++++++++++++++++++++++
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h |  2 ++
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_job.c |  1 +
> >   4 files changed, 29 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c 
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> > index 8304d0c87899..00032093d8a9 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> > @@ -1257,6 +1257,7 @@ static int amdgpu_cs_submit(struct 
> > amdgpu_cs_parser *p,
> >       priority = job->base.s_priority;
> >       drm_sched_entity_push_job(&job->base, entity);
> >   +    amdgpu_ctx_limit_load_balance(entity);
> >       amdgpu_vm_move_to_lru_tail(p->adev, &fpriv->vm);
> >         ttm_eu_fence_buffer_objects(&p->ticket, &p->validated, 
> > p->fence);
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c 
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
> > index fa575bdc03c8..57b49188306d 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
> > @@ -139,6 +139,7 @@ static int amdgpu_ctx_init_entity(struct 
> > amdgpu_ctx *ctx, const u32 hw_ip, const
> >       if (r)
> >           goto error_free_entity;
> >   +    entity->hw_ip = hw_ip;
> >       ctx->entities[hw_ip][ring] = entity;
> >       return 0;
> >   @@ -559,6 +560,30 @@ void amdgpu_ctx_priority_override(struct 
> > amdgpu_ctx *ctx,
> >       }
> >   }
> >   +static void limit_vcn_load_balance(struct amdgpu_ctx_entity *centity)
> > +{
> > +    struct drm_gpu_scheduler **scheds = &centity->entity.rq->sched;
> > +
> > +    if (drm_sched_entity_num_jobs(&centity->entity) == 1)
>
> That check doesn't work correctly, the job might actually already be 
> processed when we hit here.


Okay now I know what Andrey meant. I will resend a updated patch.


Thanks,

Nirmoy


> > + drm_sched_entity_modify_sched(&centity->entity, scheds, 1);
>
> Just always update the scheduler here.
>
> > +
> > +}
> > +
> > +void amdgpu_ctx_limit_load_balance(struct drm_sched_entity *entity)
> > +{
> > +    struct amdgpu_ctx_entity *centity = to_amdgpu_ctx_entity(entity);
> > +
> > +    if (!centity)
> > +        return;
>
> That check looks superfluous to me.
>
> > +
> > +    switch (centity->hw_ip) {
>
> Better get the ring from entity->rq->sched instead.
>
> > +    case AMDGPU_HW_IP_VCN_DEC:
> > +    case AMDGPU_HW_IP_VCN_ENC:
>
> Maybe better to make that a flag in the ring functions, but this way 
> works as well.
>
> Regards,
> Christian.
>
> > +        limit_vcn_load_balance(centity);
> > +    }
> > +
> > +}
> > +
> >   int amdgpu_ctx_wait_prev_fence(struct amdgpu_ctx *ctx,
> >                      struct drm_sched_entity *entity)
> >   {
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h 
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h
> > index de490f183af2..d52d8d562d77 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h
> > @@ -33,6 +33,7 @@ struct amdgpu_fpriv;
> >     struct amdgpu_ctx_entity {
> >       uint64_t        sequence;
> > +    uint32_t        hw_ip;
> >       struct drm_sched_entity    entity;
> >       struct dma_fence    *fences[];
> >   };
> > @@ -90,5 +91,6 @@ void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr);
> >     void amdgpu_ctx_init_sched(struct amdgpu_device *adev);
> >   +void amdgpu_ctx_limit_load_balance(struct drm_sched_entity *entity);
> >     #endif
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c 
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> > index 4981e443a884..955d12bc89ae 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> > @@ -154,6 +154,7 @@ int amdgpu_job_submit(struct amdgpu_job *job, 
> > struct drm_sched_entity *entity,
> >       amdgpu_job_free_resources(job);
> >       priority = job->base.s_priority;
> >       drm_sched_entity_push_job(&job->base, entity);
> > +    amdgpu_ctx_limit_load_balance(entity);
> >         return 0;
> >   }
_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx