Re: [RFC][PATCH] PM: Force GFP_NOIO during suspend/resume (was: Re: [linux-pm] Memory allocations in .suspend became very unreliable)

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Mon, 2010-01-18 at 00:00 +0100, Rafael J. Wysocki wrote:
> On Sunday 17 January 2010, Benjamin Herrenschmidt wrote:
> > On Sun, 2010-01-17 at 14:27 +0100, Rafael J. Wysocki wrote:
> ...
> > However, it's hard to deal with the case of allocations that have
> > already started waiting for IOs. It might be possible to have some VM
> > hook to make them wakeup, re-evaluate the situation and get out of that
> > code path but in any case it would be tricky.
> 
> In the second version of the patch I used an rwsem that made us wait for these
> allocations to complete before we changed gfp_allowed_mask.
> 
> [This is kinda buggy in the version I sent, but I'm going to send an update
> in a minute.]

And nobody screamed due to cache line ping pong caused by this in the
fast path ? :-)

We might want to look at something a bit smarter for that sort of
read-mostly-really-really-mostly construct, though in this case I don't
think RCU is the answer since we are happily scheduling.

I wonder if something per-cpu would do, it's thus the responsibility of
the "writer" to take them all in order for all CPUs.

Cheers,
Ben.