From 8f38a49f7e929fa385b3784725e8a1955db9fc57 Mon Sep 17 00:00:00 2001 From: Bertrand Drouvot Date: Wed, 25 Jun 2025 05:06:40 +0000 Subject: [PATCH v1] Work around Linux kernel bug in do_pages_stat() do_pages_stat() is already handling the input arrays correctly in 32-bit mode, but at the end of the "while (nr_pages)" loop, it incorrectly advances the pages pointer with the wrong word size. Work around is to ensure that pg_numa_query_pages() does not pass more than DO_PAGES_STAT_CHUNK_NR (see do_pages_stat()) pages to do_pages_stat() so that the wrong pointer arithmetic has no effect (as the pages variable is not being used). Linux kernel bug reported here: https://marc.info/?l=linux-mm&m=175077821909222&w=2 --- src/port/pg_numa.c | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+) diff --git a/src/port/pg_numa.c b/src/port/pg_numa.c index 4b487a2a4e8..86921a156b4 100644 --- a/src/port/pg_numa.c +++ b/src/port/pg_numa.c @@ -16,6 +16,7 @@ #include "c.h" #include +#include "miscadmin.h" #include "port/pg_numa.h" /* @@ -46,7 +47,31 @@ pg_numa_init(void) int pg_numa_query_pages(int pid, unsigned long count, void **pages, int *status) { +/* + * Work around Linux kernel bug in 32-bit compat mode: do_pages_stat() has + * incorrect pointer arithmetic for more than DO_PAGES_STAT_CHUNK_NR pages. + */ +#if SIZEOF_SIZE_T == 4 +#define NUMA_QUERY_CHUNK_SIZE 16 /* has to be <= DO_PAGES_STAT_CHUNK_NR + * (do_pages_stat()) */ + for (size_t chunk_start = 0; chunk_start < count; chunk_start += NUMA_QUERY_CHUNK_SIZE) + { + int result; + uint64 chunk_size = Min(NUMA_QUERY_CHUNK_SIZE, count - chunk_start); + + CHECK_FOR_INTERRUPTS(); + + result = numa_move_pages(pid, chunk_size, &pages[chunk_start], NULL, + &status[chunk_start], 0); + + if (result != 0) + return result; + } + + return 0; +#else return numa_move_pages(pid, count, pages, NULL, status, 0); +#endif } int -- 2.34.1