To decouple the dma_fence_ops lifetime from dma_fences lifetime RCU
support was added to said function, coupled with using the signaled bit
to detect whether the fence_ops might be gone already.

When implementing that a wrong string was set as a default return
parameter, indicating that every driver whose fence is already signalled
must be detached, which is frankly wrong.

Reported-by: Danilo Krummrich <[email protected]>
Fixes: 506aa8b02a8d ("dma-fence: Add safe access helpers and document the 
rules")
Signed-off-by: Philipp Stanner <[email protected]>
---
When this was merged, it sadly slipped by me. I think this entire RCU
mechanism was / is an overengineered idea.

If we look at who actually uses dma_fence_driver_name() and
dma_fence_timeline_name() – functions from which the largest share of
the fence_ops vs. fence lifetime issue stems from – we discover that
there is a single user:

i915.

Isn't that driver even deprecated?

I think the better thing to do is: remove these functions alltogether,
or at least deprecate them. Then the only lifetime issue left so solve
is the callback functions.

P.
---
 drivers/dma-buf/dma-fence.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/dma-buf/dma-fence.c b/drivers/dma-buf/dma-fence.c
index 3f78c56b58dc..1875a0abebd3 100644
--- a/drivers/dma-buf/dma-fence.c
+++ b/drivers/dma-buf/dma-fence.c
@@ -1111,7 +1111,7 @@ const char __rcu *dma_fence_driver_name(struct dma_fence 
*fence)
        if (!test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags))
                return fence->ops->get_driver_name(fence);
        else
-               return "detached-driver";
+               return "driver-whose-fence-is-already-signalled";
 }
 EXPORT_SYMBOL(dma_fence_driver_name);
 
-- 
2.49.0

Reply via email to