[PATCH] drm/amdgpu: handle gang submit before VMID

[Timur Kristóf]

Can you guys please push this into the next 6.1 RC? This solves a
significant issue with gang submit.

On Fri, 2022-11-18 at 16:30 +0100, Christian König wrote:
> Otherwise it can happen that not all gang members can get a VMID
> assigned and we deadlock.
> 
> Signed-off-by: Christian König <christian.koenig at amd.com>
> Tested-by: Timur Kristóf <timur.kristof at gmail.com>
> Acked-by: Timur Kristóf <timur.kristof at gmail.com>
> Fixes: 68ce8b242242 ("drm/amdgpu: add gang submit backend v2")
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> index cd968e781077..abb99cff8b4b 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> @@ -254,6 +254,9 @@ static struct dma_fence
> *amdgpu_job_dependency(struct drm_sched_job *sched_job,
>                         DRM_ERROR("Error adding fence (%d)\n", r);
>         }
>  
> +       if (!fence && job->gang_submit)
> +               fence = amdgpu_device_switch_gang(ring->adev, job-
> >gang_submit);
> +
>         while (fence == NULL && vm && !job->vmid) {
>                 r = amdgpu_vmid_grab(vm, ring, &job->sync,
>                                      &job->base.s_fence->finished,
> @@ -264,9 +267,6 @@ static struct dma_fence
> *amdgpu_job_dependency(struct drm_sched_job *sched_job,
>                 fence = amdgpu_sync_get_fence(&job->sync);
>         }
>  
> -       if (!fence && job->gang_submit)
> -               fence = amdgpu_device_switch_gang(ring->adev, job-
> >gang_submit);
> -
>         return fence;
>  }
>