[Intel-gfx] [PATCH 3/4] dma-buf: further relax reservation_object_add_shared_fence

Chris Wilson chris at chris-wilson.co.uk
Wed Aug 7 16:10:12 UTC 2019


Quoting Christian König (2019-08-07 14:53:11)
> Other cores don't busy wait any more and we removed the last user of checking
> the seqno for changes. Drop updating the number for shared fences altogether.
> 
> Signed-off-by: Christian König <christian.koenig at amd.com>
Reviewed-by: Chris Wilson <chris at chris-wilson.co.uk>

> ---
>  drivers/dma-buf/reservation.c                    | 6 ------
>  drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 7 +------
>  2 files changed, 1 insertion(+), 12 deletions(-)
> 
> diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c
> index 8fcaddffd5d4..90bc6ef03598 100644
> --- a/drivers/dma-buf/reservation.c
> +++ b/drivers/dma-buf/reservation.c
> @@ -237,9 +237,6 @@ void reservation_object_add_shared_fence(struct reservation_object *obj,
>         fobj = reservation_object_get_list(obj);
>         count = fobj->shared_count;
>  
> -       preempt_disable();
> -       write_seqcount_begin(&obj->seq);
> -
>         for (i = 0; i < count; ++i) {
>  
>                 old = rcu_dereference_protected(fobj->shared[i],
> @@ -257,9 +254,6 @@ void reservation_object_add_shared_fence(struct reservation_object *obj,
>         RCU_INIT_POINTER(fobj->shared[i], fence);
>         /* pointer update must be visible before we extend the shared_count */
>         smp_store_mb(fobj->shared_count, count);
> -
> -       write_seqcount_end(&obj->seq);
> -       preempt_enable();
>         dma_fence_put(old);
>  }
>  EXPORT_SYMBOL(reservation_object_add_shared_fence);
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
> index fe062b76ec91..a4640ddc24d1 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
> @@ -251,12 +251,7 @@ static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo,
>         new->shared_max = old->shared_max;
>         new->shared_count = k;
>  
> -       /* Install the new fence list, seqcount provides the barriers */
> -       preempt_disable();
> -       write_seqcount_begin(&resv->seq);
> -       RCU_INIT_POINTER(resv->fence, new);
> -       write_seqcount_end(&resv->seq);
> -       preempt_enable();
> +       rcu_assign_pointer(resv->fence, new);

but you'll probably want a local ack for amdgpu/
-Chris


More information about the Intel-gfx mailing list