Re: [RFC][PATCH] PM: disable nonboot cpus before suspending devices

[Andrew Morton <akpm@linux-foundation.org>]

On Mon, 1 Feb 2010 16:30:04 +0100 "Rafael J. Wysocki" <rjw@sisk.pl> wrote:

> On Monday 01 February 2010, Sebastian Ott wrote:
> > Hi Rafael,
> > 
> > since you didn't like the idea of calling the driver callbacks with just
> > one cpu enabled, we gave your patch: "MM / PM: Force GFP_NOIO during
> > suspend/hibernation and resume" a try and i can confirm that this
> > fixes the issue on s390.
> 
> Great, thanks for testing!
> 
> > Will this go in 2.6.33/stable?
> 
> That depends on Andrew, actually.
> 
> Andrew, what do you think of the patch at:
> http://patchwork.kernel.org/patch/74740/mbox/ ?
> 
> It helps people and I don't see any major drawbacks of it.
> 

Seems sane.  A couple of minor things:

- the names mm_force_noio_allocations() and mm_allow_io_allocations()
  are a bit sucky.  Asymmetrical.

- the functions don't nest: if someone calls
  mm_force_noio_allocations() twice in succession then the kernel is
  all mucked up.  Why not:

gfp_t mm_set_gfp_mask(gfp_t mask)
{
	gfp_t ret = gfp_allowed_mask;

	gfp_allowed_mask = mask;
	return ret;
}

which is of course racy :)  Could add a local spinlock if really worried.

All your current callers can easily save the old value in a local.