On Tue, Oct 03, 2017 at 07:47:20PM +0300, Andrey Ryabinin wrote:
> Commit 3a321d2a3dde separated NUMA counters from zone counters, but
> the NUMA_INTERLEAVE_HIT call site wasn't updated to use the new interface.
> So alloc_page_interleave() actually increments NR_ZONE_INACTIVE_FILE
> instead of NUMA_INTERLEAVE_HIT.
>
> Fix this by using __inc_numa_state() interface to increment
> NUMA_INTERLEAVE_HIT.
>
> Fixes: 3a321d2a3dde ("mm: change the call sites of numa statistics items")
> Signed-off-by: Andrey Ryabinin <aryabinin@xxxxxxxxxxxxx>
> ---
> mm/mempolicy.c | 9 +++++++--
> 1 file changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/mm/mempolicy.c b/mm/mempolicy.c
> index 006ba625c0b8..3a18f0a091c4 100644
> --- a/mm/mempolicy.c
> +++ b/mm/mempolicy.c
> @@ -1920,8 +1920,13 @@ static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
> struct page *page;
>
> page = __alloc_pages(gfp, order, nid);
> - if (page && page_to_nid(page) == nid)
> - inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
> + if (page && page_to_nid(page) == nid) {
> + unsigned long flags;
> +
> + local_irq_save(flags);
> + __inc_numa_state(page_zone(page), NUMA_INTERLEAVE_HIT);
> + local_irq_restore(flags);
> + }
> return page;
> }
alloc_page_interleave is only called from !irq contexts and the requirements
for __inc_numa_state should only require interrupt disabling if that
particular counter can be updated from interrupt context. Disabling
preemption should be sufficient for NUMA_INTERLEAVE_HIT and would be cheaper.
--
Mel Gorman
SUSE Labs
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
