Re: [PATCH 02/28] drm/ttm: stop pruning fences after wait

Am 30.11.21 um 10:02 schrieb Daniel Vetter:

On Mon, Nov 29, 2021 at 01:06:33PM +0100, Christian König wrote:

quoted

This is just abusing internals of the dma_resv object.

Signed-off-by: Christian König <christian.koenig@amd.com>

Yeah I think if we want this back we could do a _locked version of the
wait, which prunes internally.

Reviewed-by: Daniel Vetter <redacted>

Btw I wonder, should we put the ttm_bo_wait wrapper on the chopping block
in gpu/todo.rst? It's really just complications I think in most cases. And
it would be nice if ttm has the same errno semantics for these as everyone
else, I always get very confused about this stuff ...

I've already done that quite a bit, e.g. removed most of the users.

What's left are the cases where we wait or test signaling inside of TTM 
and I think I can get rid of that with the LRU rework.

So yeah, already in the pipeline.

Regards,
Christian.

Cheers, Daniel

quoted

---
  drivers/gpu/drm/ttm/ttm_bo.c | 1 -
  1 file changed, 1 deletion(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index e4a20a3a5d16..fc124457ba2f 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c

@@ -1086,7 +1086,6 @@ int ttm_bo_wait(struct ttm_buffer_object *bo,
  	if (timeout == 0)
  		return -EBUSY;
  
-	dma_resv_add_excl_fence(bo->base.resv, NULL);
  	return 0;
  }
  EXPORT_SYMBOL(ttm_bo_wait);

-- 
2.25.1