Re: mmu_notifiers: turn off lockdep around mm_take_all_locks

[Peter Zijlstra <peterz@infradead.org>]

On Tue, 2009-07-07 at 15:37 -0300, Marcelo Tosatti wrote:
> >>>
> >>> Is there any way around this other than completly shutting down lockdep?
> >>>      
> >>
> >> When we created this the promise was that kvm would only do this on a
> >> fresh mm with only a few vmas, has that changed
> >
> > The number of vmas did increase, but not materially.  We do link with  
> > more shared libraries though.
> 
> Yeah, see attached /proc/pid/maps just before the ioctl thats ends up in 
> mmu_notifier_register.
> 
> mm_take_all_locks: file_vma=79 anon_vma=40

Another issue, at about >=256 vmas we'll overflow the preempt count. So
disabling lockdep will only 'fix' this for a short while, until you've
bloated beyond that ;-)

Although you could possibly disable preemption and use
__raw_spin_lock(), that would also side-step the whole lockdep issue,
but it feels like such a horrid hack.

Alternatively we would have to modify the rmap locking, but that would
incur overhead on the regular code paths, so that's probably not worth
the trade-off.

Linus, Ingo, any opinions?