diff options
| author | Thomas Gleixner <tglx@linutronix.de> | 2024-04-09 13:29:12 +0300 |
|---|---|---|
| committer | Ingo Molnar <mingo@kernel.org> | 2024-04-10 11:13:42 +0300 |
| commit | f87cbcb345d059f0377b4fa0ba1b766a17fc3710 (patch) | |
| tree | 5593baf8c37a2d77828e4f26c237eb2127009a7e /include/asm-generic/hardirq.h | |
| parent | 6d029c25b71f2de2838a6f093ce0fa0e69336154 (diff) | |
| download | linux-f87cbcb345d059f0377b4fa0ba1b766a17fc3710.tar.xz | |
timekeeping: Use READ/WRITE_ONCE() for tick_do_timer_cpu
tick_do_timer_cpu is used lockless to check which CPU needs to take care
of the per tick timekeeping duty. This is done to avoid a thundering
herd problem on jiffies_lock.
The read and writes are not annotated so KCSAN complains about data races:
BUG: KCSAN: data-race in tick_nohz_idle_stop_tick / tick_nohz_next_event
write to 0xffffffff8a2bda30 of 4 bytes by task 0 on cpu 26:
tick_nohz_idle_stop_tick+0x3b1/0x4a0
do_idle+0x1e3/0x250
read to 0xffffffff8a2bda30 of 4 bytes by task 0 on cpu 16:
tick_nohz_next_event+0xe7/0x1e0
tick_nohz_get_sleep_length+0xa7/0xe0
menu_select+0x82/0xb90
cpuidle_select+0x44/0x60
do_idle+0x1c2/0x250
value changed: 0x0000001a -> 0xffffffff
Annotate them with READ/WRITE_ONCE() to document the intentional data race.
Reported-by: Mirsad Todorovac <mirsad.todorovac@alu.unizg.hr>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Sean Anderson <sean.anderson@seco.com>
Link: https://lore.kernel.org/r/87cyqy7rt3.ffs@tglx
Diffstat (limited to 'include/asm-generic/hardirq.h')
0 files changed, 0 insertions, 0 deletions