Re: [PATCH] gnttab: lock the left grant table earlier in __gnttab_unmap_common()

[Matt Wilson <msw@linux.com>]

On Mon, Nov 11, 2013 at 10:53:23AM +0000, David Vrabel wrote:
> On 11/11/13 02:07, Matt Wilson wrote:
> > From: Matt Wilson <msw@amazon.com>
> > 
> > Luckily today maptrack_limit never shrinks. But if at some point in
> > the future this were to change, checking maptrack_limit without
> > holding the grant table spinlock would no longer be safe.
> 
> I don't think we should extend region the grant table lock protects if
> it's not needed since we know this lock is heavily contended.

I've been looking into the grant table lock contention. Moving this up
didn't degrade performance in the highly contended test scenario I've
set up.

This change makes more sense when put in the context of some locking
changes I've been experimenting with.

> Also, doesn't the 'l' and 'r' prefixes mean 'local' and 'remote' not
> 'left' and 'right'?

Oops. As an aside, I've never liked the "remote" designation. The
common use of "foreign" also makes this confusing. Every time I draw a
dom0 <-> domU diagram on a whiteboard dom0 is on the left, leading me
to this mistake. ;-)

I'll update the description.

--msw