Re: [PATCH] epoll: Fix spurious lockdep warnings.

[Nelson Elhage <nelhage@ksplice.com>]

Sure -- it's quite simple if you've worked with epoll before. This
cut-down version is even simpler than the previous one I had, and I'd
be happy to add it to the commit message.

--------------------8<--------------------
#include <sys/epoll.h>

int main(void) {
    int e1, e2;
    struct epoll_event evt = {
        .events = EPOLLIN
    };

    e1 = epoll_create1(0);
    e2 = epoll_create1(0);
    epoll_ctl(e1, EPOLL_CTL_ADD, e2, &evt);
    return 0;
}
--------------------8<--------------------

- Nelson

On Sun, Jul 31, 2011 at 05:06:10PM +0200, Paul Bolle wrote:
> On Sat, 2011-07-30 at 18:30 -0400, Nelson Elhage wrote:
> >  I've tested this on a synthetic epoll test case, that just adds e1 to
> >  e2 and then does an epoll_wait(). I verified that it caused lockdep
> >  problems on 3.0 and that this patch fixed it, but I haven't done more
> >  extensive testing.
> 
> I was unable to come up with such a test case myself. Could you perhaps
> share it?
> 
> (Maybe that test case could even be added tot the commit message. I seem
> to remember an earlier commit that you were involved with which had a
> test case added. That helped me understand eventpoll's interface - at
> least enough to pinpoint the problem. Looking at a test case is much
> easier than grepping through a program like systemd. Issues in non-test
> case programs tend to increase the bug hunting challenge: one is faced
> with an issue in an interface one hasn't used before triggered by a
> program one hasn't studied before.)
> 
> > Paul, are you able to test systemd against this?
> 
> I hope to do so shortly (ie, in the next 24 hours).
> 
> 
> Paul Bolle
>