Re: [PATCH] dma-fence: Move signalling tracepoint to before ops detach

["Christian König" <christian.koenig@amd.com>]

On 4/9/26 15:58, Tvrtko Ursulin wrote:
> 
> On 31/03/2026 08:49, Boris Brezillon wrote:
>> On Mon, 30 Mar 2026 14:36:23 +0100
>> Tvrtko Ursulin <tvrtko.ursulin@igalia.com> wrote:
>>
>>> Move the signalling tracepoint to before fence->ops are reset otherwise
>>> tracepoint will dereference a null pointer.
>>
>> I suspect other trace points are impacted too
>> (trace_dma_fence_destroy() is, at the very least).
> 
> Indeed. I wonder why that did not trigger for me, while the one I fix here was an insta-crash...

You need to actually enable the trace points and at least for the destroy one nobody is usually interested in that.

> 
> To fix trace_dma_fence_destroy I think we need a new tracepoint definition ie. move it away from the existing event class - make it just log the context and seqno.
> 
> Anyone has a better idea?

The idea of tracing without accessing fence->ops sounds valid to me.

Alternatively we could call dma_fence_timeline_name() and dma_fence_driver_name() from the tracepoint as well, but that means the tracepoints now require a RCU read side lock.

Regards,
Christian.

> 
> Regards,
> 
> Tvrtko
> 
>>> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
>>> Fixes: 541c8f2468b9 ("dma-buf: detach fence ops on signal v3")
>>> Cc: Christian König <christian.koenig@amd.com>
>>> Cc: Philipp Stanner <phasta@kernel.org>
>>> Cc: Boris Brezillon <boris.brezillon@collabora.com>
>>> Cc: linux-media@vger.kernel.org
>>> Cc: linaro-mm-sig@lists.linaro.org
>>> ---
>>>   drivers/dma-buf/dma-fence.c | 3 ++-
>>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/dma-buf/dma-fence.c b/drivers/dma-buf/dma-fence.c
>>> index 1826ba73094c..1c1eaecaf1b0 100644
>>> --- a/drivers/dma-buf/dma-fence.c
>>> +++ b/drivers/dma-buf/dma-fence.c
>>> @@ -363,6 +363,8 @@ void dma_fence_signal_timestamp_locked(struct dma_fence *fence,
>>>                         &fence->flags)))
>>>           return;
>>>   +    trace_dma_fence_signaled(fence);
>>> +
>>>       /*
>>>        * When neither a release nor a wait operation is specified set the ops
>>>        * pointer to NULL to allow the fence structure to become independent
>>> @@ -377,7 +379,6 @@ void dma_fence_signal_timestamp_locked(struct dma_fence *fence,
>>>         fence->timestamp = timestamp;
>>>       set_bit(DMA_FENCE_FLAG_TIMESTAMP_BIT, &fence->flags);
>>> -    trace_dma_fence_signaled(fence);
>>>         list_for_each_entry_safe(cur, tmp, &cb_list, node) {
>>>           INIT_LIST_HEAD(&cur->node);
>>
>