Re: [PATCH] - Fix unmap_vma() bug related to mmu_notifiers

[Robin Holt <holt@sgi.com>]

On Wed, Jan 27, 2010 at 09:18:42PM -0600, Robin Holt wrote:
> On Tue, Jan 26, 2010 at 10:38:53PM +0100, Andrea Arcangeli wrote:
> > On Tue, Jan 26, 2010 at 03:29:04PM -0600, Robin Holt wrote:
> > > On Mon, Jan 25, 2010 at 10:16:15PM +0100, Andrea Arcangeli wrote:
> > > > The old patches are in my ftp area, they should still apply, you
> > > > should concentrate testing with those additional ones applied, then it
> > > > will work for xpmem too ;)
> > > 
> > > Andrea, could you point me at your ftp area?
> > 
> > Sure, this is the very latest version I maintained:
> > 
> > http://www.kernel.org/pub/linux/kernel/people/andrea/patches/v2.6/2.6.26-rc7/mmu-notifier-v18/
> 
> Let me start with what XPMEM currently has.
> 
> We adjusted xpmem so that the mmu_notifier_invalidate_page() callout
> does not need to sleep.  It takes the arguments passed in and adds them
> to a queue for clearing the pages.  We added a seperate kernel thread
> which manages this clearing.
> 
> The mmu_notifier_invalidate_range_end() likewise does not really need
> to sleep either.
> 
> That leaves the mmu_notifier_invalidate_range_start() callout.  This does
> not need to drop the mm_sem.  It does need to be able to sleep waiting
> for the invalidations to complete on the other process.  That other
> process may be on a different SSI connected to the same Numalink fabric.
> 
> I think that with the SRCU patch, we have enough.  Is that true or have
> I missed something?

I wasn't quite complete in my previous email.  Your srcu patch
plus Jack's patch to move the tlb_gather_mmu to after the
mmu_notifier_invalidate_range_start().

Robin