On Tue, 13 Jul 2021 14:05:33 -0300 Jason Gunthorpe <jgg@xxxxxxxxxx> wrote: > On Tue, Jul 13, 2021 at 06:45:17PM +0200, Halil Pasic wrote: > > > Jason may give it another try to convince us that 0cc00c8d4050 only > > silenced lockdep, but vfio_ap remained prone to deadlocks. To my best > > knowledge using condition variable and a mutex is one of the well known > > ways to implement an rwlock. > > The well known pattern is to use a rwsem. I think you are missing the point. We are discussing whether this qualifies for stable, i.e. if 0cc00c8d4050 is really broken like the patch description says. Using a readers-writers lock (as a primitive) to implement a a readers-writers lock is a fallacy, so I guess you wanted to say that when a readers-writers lock is needed in the kernel the obvious choices are rw_semaphore and/or rwlock_t (depending on the spin). What I wanted to say is using a condition variable and a mutex is not per-see wrong, because one can even implement an readers-writers lock with it. For reference see https://en.wikipedia.org/wiki/Readers%E2%80%93writer_lock > > This: > wait_event_cmd(matrix_mdev->wait_for_kvm, > !matrix_mdev->kvm_busy, > mutex_unlock(&matrix_dev->lock), > mutex_lock(&matrix_dev->lock)); > > > Is not really a rwsem, and is invsible to lockdep. > I agree. But this is not a proof of a problem that qualifies to be fixed using the stable process as documented in https://github.com/torvalds/linux/blob/master/Documentation/process/stable-kernel-rules.rst I'm in favor of rewriting this to use rw_semaphore. I'm not in favor of proclaiming this a fix for stable, because for that you first have to prove that you fix a real problem. I hope we are on the same page. Regards, Halil
