Re: Runtime trouble with commit dbd952127d (seccomp: introduce writer locking)

[Guenter Roeck <linux@roeck-us.net>]

On Mon, Aug 11, 2014 at 12:51:22PM -0700, Kees Cook wrote:
> On Mon, Aug 11, 2014 at 7:11 AM, Guenter Roeck <linux@roeck-us.net> wrote:
> > On 08/11/2014 04:48 AM, Oleg Nesterov wrote:
> >>
> >> On 08/10, Linus Torvalds wrote:
> >>>
> >>>
> >>> So that should just be converted to assert_spin_is_locked().
> >>
> >>
> >> I still think that lockdep_assert_held() is better. Unlike
> >> assert_spin_locked() it checks that this lock is held by us, and this
> >> is what we want in this case.
> >>
> >
> > assert_spin_locked maps to "BUG_ON(!raw_spin_is_locked(x))"
> > which it seems is exactly what the current code is doing.
> > I submitted a patch to make that change to use assert_spin_locked.
> > Presumably the author had a reason for using BUG_ON and not
> > lockdep_assert_held(), ie to perform the checks all the time
> > and not just while debugging. For me this was the safe change
> > to make. Anything else should, in my opinion, come from the
> > original author who introduced the code.
> 
> Thanks for the patch! Yeah, that's a weird case; I think we need some
> documentation in the header file about the UP vs SMP logic when using
> spin_is_locked(). I note that all other stuff gets hidden behind the
> _up and _smp headers.
> 
Guess one has to know what to look for.

Documentation/scsi/ChangeLog.megaraid gives a hint, as do the comments
next to WARN_ON_SMP.

Not that I knew before last night :-).

> I don't prefer lockdep_assert_held(), though, since I want lock
> failures to hit BUG. I'll apply the patch and ask James to pull it.
> 
Thanks a lot!

Guenter