From: Hui Zhu KCSAN reports a data-race on v6.6 which also exists on mainline: BUG: KCSAN: data-race in zone_page_state_snapshot root/klinux/./include/linux/vmstat.h:232 [inline] BUG: KCSAN: data-race in allow_direct_reclaim+0x150/0x290 root/klinux/mm/vmscan.c:6328 race at unknown origin, with read to 0xffff9e0575410740 of 1 bytes by task 209280 on cpu 15: zone_page_state_snapshot root/klinux/./include/linux/vmstat.h:232 [inline] allow_direct_reclaim+0x150/0x290 root/klinux/mm/vmscan.c:6328 throttle_direct_reclaim+0x1c4/0x480 root/klinux/mm/vmscan.c:6402 try_to_free_pages+0x177/0x590 root/klinux/mm/vmscan.c:6466 __perform_reclaim root/klinux/mm/page_alloc.c:3966 [inline] __alloc_pages_direct_reclaim root/klinux/mm/page_alloc.c:3988 [inline] __alloc_pages_slowpath.constprop.0+0x3b3/0xdf0 root/klinux/mm/page_alloc.c:4395 __alloc_pages_noprof+0x28a/0x2d0 root/klinux/mm/page_alloc.c:4794 alloc_pages_mpol_noprof+0x181/0x470 root/klinux/mm/mempolicy.c:2270 alloc_pages_noprof+0x7a/0xe0 root/klinux/mm/mempolicy.c:2347 io_mem_alloc_compound root/klinux/io_uring/io_uring.c:2848 [inline] io_pages_map+0xb9/0x330 root/klinux/io_uring/io_uring.c:2892 io_allocate_scq_urings+0x231/0x300 root/klinux/io_uring/io_uring.c:3933 io_uring_create+0x29c/0x9c0 root/klinux/io_uring/io_uring.c:4091 io_uring_setup+0xb6/0x100 root/klinux/io_uring/io_uring.c:4203 __do_sys_io_uring_setup root/klinux/io_uring/io_uring.c:4230 [inline] __se_sys_io_uring_setup root/klinux/io_uring/io_uring.c:4224 [inline] __x64_sys_io_uring_setup+0x66/0xe0 root/klinux/io_uring/io_uring.c:4224 x64_sys_call+0x1816/0x1d90 root/klinux/./arch/x86/include/generated/asm/syscalls_64.h:426 do_syscall_x64 root/klinux/arch/x86/entry/common.c:51 [inline] do_syscall_64+0x70/0x2a0 root/klinux/arch/x86/entry/common.c:81 entry_SYSCALL_64_after_hwframe+0x76/0xe0 value changed: 0xdd -> 0xb6 Reported by Kernel Concurrency Sanitizer on: CPU: 15 PID: 209280 Comm: syz-executor.13 Tainted: G S ------- ---- 6.6.89+ #1250 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.3-0-ga6ed6b701f0a-prebuilt.qemu.org 04/01/2014 Found by KCSAN testing on an older kernel; the same race still exists on mainline. zone_page_state_snapshot() sums the per-cpu vm_stat_diff[] deltas of all online CPUs with a plain load, while the allocation and free fast paths update the local delta through this_cpu_cmpxchg() and friends. Those are inline asm, and page_alloc.o is excluded from KCSAN, so the writer is not instrumented and the report is of unknown origin. The race is benign: the function already documents that the result cannot be exactly accurate, and every consumer is a reclaim or compaction heuristic that tolerates an estimate. Annotate the read with data_race(), following commit af1c31acc853 ("mm/vmstat: annotate data race for zone->free_area[order].nr_free"). No functional change intended. Signed-off-by: Hui Zhu --- include/linux/vmstat.h | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h index 5b31d8e7ae405..9df3587a29685 100644 --- a/include/linux/vmstat.h +++ b/include/linux/vmstat.h @@ -236,9 +236,15 @@ static inline unsigned long zone_page_state_snapshot(struct zone *zone, long x = atomic_long_read(&zone->vm_stat[item]); #ifdef CONFIG_SMP + struct per_cpu_zonestat __percpu *pzstats = zone->per_cpu_zonestats; int cpu; + for_each_online_cpu(cpu) - x += per_cpu_ptr(zone->per_cpu_zonestats, cpu)->vm_stat_diff[item]; + /* + * Access to vm_stat_diff is lockless as the result is only + * an estimate. Use data_race to avoid KCSAN warning. + */ + x += data_race(per_cpu_ptr(pzstats, cpu)->vm_stat_diff[item]); if (x < 0) x = 0; -- 2.43.0