[PATCH 1/2] dma-buf: keep only not signaled fence in reservation_object_add_shared_replace v2

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Quoting Christian König (2017-10-30 14:59:03)
> From: Christian König <christian.koenig at amd.com>
> 
> The amdgpu issue to also need signaled fences in the reservation objects
> should be fixed by now.
> 
> Optimize the list by keeping only the not signaled yet fences around.
> 
> v2: temporary put the signaled fences at the end of the new container
> 
> Signed-off-by: Christian König <christian.koenig at amd.com>
> ---
>  drivers/dma-buf/reservation.c | 36 ++++++++++++++++++++++++++----------
>  1 file changed, 26 insertions(+), 10 deletions(-)
> 
> diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c
> index b44d9d7db347..6fc794576997 100644
> --- a/drivers/dma-buf/reservation.c
> +++ b/drivers/dma-buf/reservation.c
> @@ -145,8 +145,8 @@ reservation_object_add_shared_replace(struct reservation_object *obj,
>                                       struct reservation_object_list *fobj,
>                                       struct dma_fence *fence)
>  {
> -       unsigned i;
>         struct dma_fence *old_fence = NULL;
> +       unsigned i, j, k;
>  
>         dma_fence_get(fence);
>  
> @@ -162,9 +162,7 @@ reservation_object_add_shared_replace(struct reservation_object *obj,
>          * references from the old struct are carried over to
>          * the new.
>          */
> -       fobj->shared_count = old->shared_count;
> -
> -       for (i = 0; i < old->shared_count; ++i) {
> +       for (i = 0, j = 0, k = fobj->shared_max; i < old->shared_count; ++i) {
>                 struct dma_fence *check;
>  
>                 check = rcu_dereference_protected(old->shared[i],
> @@ -172,10 +170,18 @@ reservation_object_add_shared_replace(struct reservation_object *obj,
>  
>                 if (!old_fence && check->context == fence->context) {
>                         old_fence = check;
> -                       RCU_INIT_POINTER(fobj->shared[i], fence);
> -               } else
> -                       RCU_INIT_POINTER(fobj->shared[i], check);
> +                       RCU_INIT_POINTER(fobj->shared[j++], fence);
> +               } else if (!dma_fence_is_signaled(check)) {
> +                       RCU_INIT_POINTER(fobj->shared[j++], check);
> +               } else {
> +                       /*
> +                        * Temporary save the signaled fences at the end of the
> +                        * new container
> +                        */
> +                       RCU_INIT_POINTER(fobj->shared[--k], check);
> +               }
>         }
> +       fobj->shared_count = j;
>         if (!old_fence) {
>                 RCU_INIT_POINTER(fobj->shared[fobj->shared_count], fence);
>                 fobj->shared_count++;
> @@ -192,10 +198,20 @@ reservation_object_add_shared_replace(struct reservation_object *obj,
>         write_seqcount_end(&obj->seq);
>         preempt_enable();
>  
> -       if (old)
> -               kfree_rcu(old, rcu);
> -
>         dma_fence_put(old_fence);

We don't need to keep old_fence (local var) around any more in this
scheme. By construction, the new shared_list is large enough to hold all
the old_fences+1, so you can use a loop like

j = 0;
k = fobj->shared_max;
RCU_INIT_POINTER(fobj->shared[j++], fence);
for (i = 0; i < old->shared_count; ++i) {
	struct dma_fence *check;

	check = rcu_dereference_protected(old->shared[i],
					  reservation_object_held(obj));

	if (check->context == fence->context ||
	    dma_fence_is_signaled(check))
		RCU_INIT_POINTER(fobj->shared[--k], check);
	else
		RCU_INIT_POINTER(fobj->shared[j++], check);
}
fobj->shared_count = j;

> +
> +       if (!old)
> +               return;
> +
> +       /* Drop the references to the signaled fences */
> +       for (i = k; i < fobj->shared_max; ++i) {
> +               struct dma_fence *f;
> +
> +               f = rcu_dereference_protected(fobj->shared[i],
> +                                             reservation_object_held(obj));
> +               dma_fence_put(f);
> +       }
> +       kfree_rcu(old, rcu);
>  }

Fwiw, this fixes a big problem with the lack of pruning of the
reservation_object, so \o/
-Chris


[Index of Archives]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux