On 2024-12-27 6:57 AM, Honggyu Kim wrote:
The fa3bea4e1f82 introduced MPOL_WEIGHTED_INTERLEAVE but it missed adding its counter to "interleave_hit" of numastat, which is located at /sys/devices/system/node/nodeN/ directory. It'd be better to add weighted interleving counter info to the existing "interleave_hit" instead of introducing a new counter "weighted_interleave_hit". Fixes: fa3bea4e1f82 ("mm/mempolicy: introduce MPOL_WEIGHTED_INTERLEAVE for weighted interleaving") Cc: Gregory Price <gourry@xxxxxxxxxx> Cc: Andi Kleen <ak@xxxxxxxxxxxxxxx> Signed-off-by: Honggyu Kim <honggyu.kim@xxxxxx> --- mm/mempolicy.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/mm/mempolicy.c b/mm/mempolicy.c index 04f35659717a..162407fbf2bc 100644 --- a/mm/mempolicy.c +++ b/mm/mempolicy.c @@ -2268,7 +2268,8 @@ struct page *alloc_pages_mpol_noprof(gfp_t gfp, unsigned int order,page = __alloc_pages_noprof(gfp, order, nid, nodemask); - if (unlikely(pol->mode == MPOL_INTERLEAVE) && page) {+ if (unlikely(pol->mode == MPOL_INTERLEAVE || + pol->mode == MPOL_WEIGHTED_INTERLEAVE) && page) { /* skip NUMA_INTERLEAVE_HIT update if numa stats is disabled */ if (static_branch_likely(&vm_numa_stat_key) && page_to_nid(page) == nid) { base-commit: 4bbf9020becbfd8fc2c3da790855b7042fad455b
[+Cc Andrew] Looks good to me, Reviewed-by: Hyeonggon Yoo <hyeonggon.yoo@xxxxxx> We discovered that interleave_hit is not correctly accounted while testing MPOL_WEIGHTED_INTERLEAVE on libnuma and this patch fixes the issue. Tested-by: Yunjeong Mun <yunjeong.mun@xxxxxx> For reference: https://github.com/numactl/numactl/pull/238#issuecomment-2563532910 -- Hyeonggon
