On 06/04/2021 18:53, Ryan Sharpelletti wrote: > On Fri, Mar 5, 2021 at 2:07 PM Ryan Sharpelletti > <sharpelletti@xxxxxxxxxx> wrote: >> I suspect there is a corner case for io_uring in release 5.8. Did you see any fail? >> Can you please explain where the associated mmdrop/kthread_unuse_mm >> calls for io_sq_thread_acquire_mm in io_init_req(...) are? Look for io_sq_thread_drop_mm(). acquire_mm shouldn't do anything unless it's SQPOLL mode and so executed by sqpoll task. >> Specifically, what would happen if there was an error after calling >> io_sq_thread_acquire_mm (for example, the -EINVAL a few lines after)? Will be cleaned by io_sq_thread(). >> >> From what I can understand, it looks like the kthread_unuse_mm and >> mmput might be handled in the call to io_sq_thread_acquire_mm in >> io_sq_thread. However, I am not seeing where mmdrop might be called. -- Pavel Begunkov
