Messages in this thread | | | Date | Thu, 20 Mar 2014 22:19:13 -0400 | From | Steven Rostedt <> | Subject | Re: [RFC PATCH] cifs: Fix possible deadlock with cifs and work queues |
| |
On Thu, 20 Mar 2014 19:53:46 -0400 Jeff Layton <jlayton@redhat.com> wrote:
> Wait...why does the work running on CPU1 end up blocked on down_read()? > Is it really getting stuck on the down_write you mention? >
rwsems are fair locks. Readers will not block on a reader lock unless there's a writer waiting. That's the key. As soon as a writer blocks on a lock that is held by a reader (or multiple readers), new readers coming in will also block to let the writer get a chance. Otherwise, it is a unfair lock and the readers can starve the writer.
But people tend to forget that a waiting writer causes readers to block on each other, and if the reader locks can deadlock each other, they will deadlock with a writer waiting.
-- Steve
| |