Re: [PATCH] mm/hugetlb: account for allowed nodes when gathering surplus pages

On Sep 9, 2026, at 15:46, Huaisheng Ye [off-list ref] wrote:

Hugetlb reservations are accounted globally, but hugetlb_acct_memory()
also verifies that the current cpuset and MPOL_BIND policy contain enough
free huge pages to add a new reservation.

gather_surplus_pages() calculates its allocation shortfall from the global
free and reserved counters. If the global pool has enough free pages, but
those pages reside outside the nodes allowed by the task, it allocates no
surplus pages. The subsequent allowed_mems_nr() check then rejects the
reservation and mmap() fails with ENOMEM, even when
nr_overcommit_hugepages permits allocating surplus pages on the allowed
nodes.

Calculate both the global shortfall and the shortfall within the allowed
nodes, and allocate the larger of the two. Include surplus pages allocated
outside hugetlb_lock in both calculations when rechecking after reacquiring
the lock. These pages are constrained by alloc_nodemask, so they satisfy
both shortages.

Easy way to reproduce this issue with 2+ NUMA nodes system:

 # echo 0 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepage
 # echo 3 > /sys/devices/system/node/node0/hugepages/hugepages-2048kB/nr_hugepages
 # echo 1 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_overcommit_hugepages
 # cd tools/testing/selftests/mm
 # numactl --membind=1 ./hugetlb-mmap 2 21
 TAP version 13
 # [INFO] detected hugetlb page size: 2048 KiB
 # [INFO] detected hugetlb page size: 1048576 KiB
 # 2048 kB hugepages
 1..2
 # Mapping 2 Mbytes
 Bail out! mmap: Cannot allocate memory (12)
 # Planned tests != run tests (2 != 0)
 # Totals: pass:0 fail:0 xfail:0 xpass:0 skip:0 error:0

This fixes hugetlb mappings when, for example, a task runs with
MPOL_BIND on Node 1 while the existing free huge pages are on Node 0.

Similar issue also could be found in ltp if the free pages of global pool
reside outside the nodes allowed by the application.
 # cd ltp/testcases/kernel/mem/hugetlb/hugemmap/
 # numactl --cpunodebind=0 --membind=1 ./hugemmap10

Signed-off-by: Huaisheng Ye <redacted>

Acked-by: Muchun Song <muchun.song@linux.dev>

Thanks.