Re: [PATCH] mm/vmstat: annotate data race for per-cpu pageset fields

["Vlastimil Babka (SUSE)" <vbabka@kernel.org>]

On 8/27/26 9:05 AM, Hui Zhu wrote:
> From: Hui Zhu <zhuhui@kylinos.cn>
> 
> zoneinfo_show_print() reads pcp->count, pcp->high, pcp->batch,
> pcp->high_min, pcp->high_max and the per-cpu stat_threshold while
> holding only zone->lock, which does not synchronize these fields.
> The writers are the page allocation and free fast paths under
> pcp->lock, decay_pcp_high() which updates pcp->high without any
> lock, pageset_update() which writes batch/high_min/high_max with
> WRITE_ONCE(), and refresh_zone_stat_thresholds() which writes
> stat_threshold locklessly.
> 
> The race is benign: the values are only printed to /proc/zoneinfo,
> they are naturally aligned integers, and pageset_update() already
> documents that users of batch/high_min/high_max must cope with the
> fields changing asynchronously.  Annotate the reads with data_race(),
> following commit af1c31acc853 ("mm/vmstat: annotate data race for
> zone->free_area[order].nr_free").
> 
> Found by KCSAN testing on an older kernel; the same race still
> exists on mainline.  No functional change intended.
> 
> BUG: KCSAN: data-race in zoneinfo_show_print+0x355/0x520 root/klinux/mm/vmstat.c:1774
> race at unknown origin, with read to 0xffff8e1835410808 of 4 bytes by task 22653 on cpu 12:
> zoneinfo_show_print+0x355/0x520 root/klinux/mm/vmstat.c:1774
> walk_zones_in_node root/klinux/mm/vmstat.c:1496 [inline]
> zoneinfo_show+0x41/0x70 root/klinux/mm/vmstat.c:1806
> seq_read_iter+0x30c/0x970 root/klinux/fs/seq_file.c:230
> proc_reg_read_iter+0x10c/0x170 root/klinux/fs/proc/inode.c:305
> copy_splice_read+0x2a1/0x4e0 root/klinux/fs/splice.c:365
> do_splice_read root/klinux/fs/splice.c:985 [inline]
> do_splice_read+0x139/0x1a0 root/klinux/fs/splice.c:959
> splice_direct_to_actor+0x16b/0x540 root/klinux/fs/splice.c:1089
> do_splice_direct_actor root/klinux/fs/splice.c:1207 [inline]
> do_splice_direct+0x10a/0x180 root/klinux/fs/splice.c:1233
> do_sendfile+0x6ea/0x7e0 root/klinux/fs/read_write.c:1363
> __do_sys_sendfile64 root/klinux/fs/read_write.c:1424 [inline]
> __se_sys_sendfile64 root/klinux/fs/read_write.c:1410 [inline]
> __x64_sys_sendfile64+0x117/0x130 root/klinux/fs/read_write.c:1410
> x64_sys_call+0x1cc7/0x1ee0 root/klinux/./arch/x86/include/generated/asm/syscalls_64.h:41
> do_syscall_x64 root/klinux/arch/x86/entry/common.c:46 [inline]
> do_syscall_64+0x75/0x2c0 root/klinux/arch/x86/entry/common.c:76
> entry_SYSCALL_64_after_hwframe+0x76/0xe0
> value changed: 0x000001b2 -> 0x000001b1
> Reported by Kernel Concurrency Sanitizer on:
> CPU: 12 PID: 22653 Comm: syz-executor.12 Not tainted 6.6.140+ #672
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.3-0-ga6ed6b701f0a-prebuilt.qemu.org 04/01/2014
> 
> Signed-off-by: Hui Zhu <zhuhui@kylinos.cn>

Acked-by: Vlastimil Babka (SUSE) <vbabka@kernel.org>

> ---
>  mm/vmstat.c | 17 +++++++++++------
>  1 file changed, 11 insertions(+), 6 deletions(-)
> 
> diff --git a/mm/vmstat.c b/mm/vmstat.c
> index cb57714539fb..a3e809c57f29 100644
> --- a/mm/vmstat.c
> +++ b/mm/vmstat.c
> @@ -1837,6 +1837,11 @@ static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
>  		struct per_cpu_zonestat __maybe_unused *pzstats;
>  
>  		pcp = per_cpu_ptr(zone->per_cpu_pageset, i);
> +		/*
> +		 * Access to the per-cpu pageset fields is lockless as they
> +		 * are used only for printing purposes. Use data_race to
> +		 * avoid KCSAN warning.
> +		 */
>  		seq_printf(m,
>  			   "\n    cpu: %i"
>  			   "\n              count:    %i"
> @@ -1845,15 +1850,15 @@ static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
>  			   "\n              high_min: %i"
>  			   "\n              high_max: %i",
>  			   i,
> -			   pcp->count,
> -			   pcp->high,
> -			   pcp->batch,
> -			   pcp->high_min,
> -			   pcp->high_max);
> +			   data_race(pcp->count),
> +			   data_race(pcp->high),
> +			   data_race(pcp->batch),
> +			   data_race(pcp->high_min),
> +			   data_race(pcp->high_max));
>  #ifdef CONFIG_SMP
>  		pzstats = per_cpu_ptr(zone->per_cpu_zonestats, i);
>  		seq_printf(m, "\n  vm stats threshold: %d",
> -				pzstats->stat_threshold);
> +				data_race(pzstats->stat_threshold));
>  #endif
>  	}
>  	seq_printf(m,