Quoting Christian König (2017-10-30 14:59:04)
> 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 handling by replacing a signaled fence when adding a new
> shared one.
>
> Signed-off-by: Christian König <christian.koenig at amd.com>
> ---
> drivers/dma-buf/reservation.c | 18 +++++++++++++++---
> 1 file changed, 15 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c
> index 6fc794576997..a3928ce9f311 100644
> --- a/drivers/dma-buf/reservation.c
> +++ b/drivers/dma-buf/reservation.c
> @@ -104,7 +104,8 @@ reservation_object_add_shared_inplace(struct reservation_object *obj,
> struct reservation_object_list *fobj,
> struct dma_fence *fence)
> {
> - u32 i;
> + struct dma_fence *signaled = NULL;
> + u32 i, signaled_idx;
>
> dma_fence_get(fence);
>
> @@ -126,17 +127,28 @@ reservation_object_add_shared_inplace(struct reservation_object *obj,
> dma_fence_put(old_fence);
> return;
> }
> +
> + if (!signaled && dma_fence_is_signaled(old_fence)) {
> + signaled = old_fence;
> + signaled_idx = i;
> + }
How much do we care about only keeping one fence per-ctx here? You could
rearrange this to break on old_fence->context == fence->context ||
dma_fence_is_signaled(old_fence) then everyone can use the final block.
-Chris
