On 1/27/25 14:16, Miklos Szeredi wrote: > On Sat, 25 Jan 2025 at 18:45, Bernd Schubert <bschubert@xxxxxxx> wrote: >> >> The value is read from another task without, while the task that >> had set the value was holding queue->lock. Better use READ_ONCE >> to ensure the compiler cannot optimize the read. > > I do not think this is necessary. The ordering should be ensured by > the io_uring infrastructure, no? Yes, definitely ordered, as it is called only when ent->fuse_req was set before. So yeah, you are right, we can skip this patch. Thanks, Bernd