On Thu, 8 May 2008 00:44:06 +0200
Andrea Arcangeli <[EMAIL PROTECTED]> wrote:
> On Wed, May 07, 2008 at 03:31:03PM -0700, Andrew Morton wrote:
> > Nope. We only need to take the global lock before taking *two or more* of
> > the per-vma locks.
> >
> > I really wish I'd thought of that.
>
> I don't see how you can avoid taking the system-wide-global lock
> before every single anon_vma->lock/i_mmap_lock out there without
> mm_lock.
>
> Please note, we can't allow a thread to be in the middle of
> zap_page_range while mmu_notifier_register runs.
>
> vmtruncate takes 1 single lock, the i_mmap_lock of the inode. Not more
> than one lock and we've to still take the global-system-wide lock
> _before_ this single i_mmap_lock and no other lock at all.
>
> Please elaborate, thanks!
umm...
CPU0: CPU1:
spin_lock(a->lock); spin_lock(b->lock);
spin_lock(b->lock); spin_lock(a->lock);
bad.
CPU0: CPU1:
spin_lock(global_lock) spin_lock(global_lock);
spin_lock(a->lock); spin_lock(b->lock);
spin_lock(b->lock); spin_lock(a->lock);
Is OK.
CPU0: CPU1:
spin_lock(global_lock)
spin_lock(a->lock); spin_lock(b->lock);
spin_lock(b->lock); spin_unlock(b->lock);
spin_lock(a->lock);
spin_unlock(a->lock);
also OK.
As long as all code paths which can take two-or-more locks are all covered
by the global lock there is no deadlock scenario. If a thread takes just a
single instance of one of these locks without taking the global_lock then
there is also no deadlock.
Now, if we need to take both anon_vma->lock AND i_mmap_lock in the newly
added mm_lock() thing and we also take both those locks at the same time in
regular code, we're probably screwed.
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
