> > > This patch resets preferred_zone to an allowed zone in the slowpath if
> > > the allocation context is constrained by current's cpuset.
> >
> > Well, preferred_zone has meaning. If it's not possible to allocate from
> > that zone in the current cpuset context, it's not really preferred. Why
> > not set it in the fast path so there isn't a useless call to
> > get_page_from_freelist()?
> >
>
> It may be the preferred zone even if it isn't allowed by current's cpuset
> such as if the allocation is __GFP_WAIT or the task has been oom killed
> and has the TIF_MEMDIE bit set, so the preferred zone in the fastpath is
> accurate in these cases. In the slowpath, the former is protected by
> checking for ALLOC_CPUSET and the latter is usually only set after the
> page allocator has looped at least once and triggered the oom killer to be
> killed.
>
> I didn't want to add a branch to test for these possibilities in the
> fastpath, however, since preferred_zone isn't of critical importance until
> it's used in the slowpath (ignoring the statistical usage).
I'm glad to you are keeping fastpath concern. However you don't need
nodemask-and in this case. Because zonelist->zref[0] is always in nodemask.
Please see policy_zonelist(). So, you can just replace nodemask with cpuset_mems_allowed.
This is not only simple, but also improve a consisteny of mempolicy.
---
mm/page_alloc.c | 3 ++-
1 files changed, 2 insertions(+), 1 deletions(-)
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 07a6544..876de04 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -2146,7 +2146,8 @@ __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
get_mems_allowed();
/* The preferred zone is used for statistics later */
- first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
+ first_zones_zonelist(zonelist, high_zoneidx,
+ &cpuset_current_mems_allowed, &preferred_zone);
if (!preferred_zone) {
put_mems_allowed();
return NULL;
--
1.6.5.2
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxxx For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>