Since syzbot restarted testing linux-next.git , it is a good chance
to test this patch for unexpected regressions (until you come up with
a good alternative).

On 2018/06/26 23:34, Tetsuo Handa wrote:

Did you get any idea?

On 2018/06/05 3:13, Jens Axboe wrote:

quoted

On 6/4/18 5:19 AM, Tetsuo Handa wrote:

quoted

This problem was already ignored for 8 months. Unless we boost priority,
this problem will be ignored for years. Jens, can we test this patch?

Sorry, it's just that I _really_ hate this patch. We're making up
a weird locking primitive that tracks the process, with a weird
unlock helper that only unlocks if it's the process that is
holding the mutex.

I'll try and think about this a bit, it would be nice if we had
a better alternative than the above.