答复: [PATCH][v2] watchdog/hardlockup: reassign last_timestamp when enable nmi event

From: Li,Rongqing
Date: Thu Nov 07 2019 - 06:10:11 EST


Ping

Thanks

-Li

> -----邮件原件-----
> 发件人: linux-kernel-owner@xxxxxxxxxxxxxxx
> [mailto:linux-kernel-owner@xxxxxxxxxxxxxxx] 代表 Li RongQing
> 发送时间: 2019年10月15日 14:34
> 收件人: akpm@xxxxxxxxxxxxxxxxxxxx; sergey.senozhatsky.work@xxxxxxxxx;
> linux-kernel@xxxxxxxxxxxxxxx; tglx@xxxxxxxxxxxxx; Kan.liang@xxxxxxxxx
> 主题: [PATCH][v2] watchdog/hardlockup: reassign last_timestamp when
> enable nmi event
>
> last_timestamp is not initialized and is zero after boot, or stop to forward when
> nmi watchdog is disabled; and false positives still is possible when restart NMI
> timer after stopping 120 seconds
>
> so reassign last_timestamp always when enable nmi event
>
> Fixes: 7edaeb6841df ("kernel/watchdog: Prevent false positives with turbo
> modes")
> Signed-off-by: Li RongQing <lirongqing@xxxxxxxxx>
> Signed-off-by: Zhang Yu <zhangyu31@xxxxxxxxx>
> ---
>
> v1-->v2: make it be able to be compiled on no
> v1-->CONFIG_HARDLOCKUP_CHECK_TIMESTAMP platform
>
> kernel/watchdog_hld.c | 18 +++++++++++++++++-
> 1 file changed, 17 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/watchdog_hld.c b/kernel/watchdog_hld.c index
> 247bf0b1582c..f14d18280387 100644
> --- a/kernel/watchdog_hld.c
> +++ b/kernel/watchdog_hld.c
> @@ -91,11 +91,24 @@ static bool watchdog_check_timestamp(void)
> __this_cpu_write(last_timestamp, now);
> return true;
> }
> +
> +static void watchdog_touch_timestamp(int cpu) {
> +
> + ktime_t now = ktime_get_mono_fast_ns();
> +
> + per_cpu(last_timestamp, cpu) = now;
> +}
> #else
> static inline bool watchdog_check_timestamp(void) {
> return true;
> }
> +
> +static void watchdog_touch_timestamp(int cpu) {
> +
> +}
> #endif
>
> static struct perf_event_attr wd_hw_attr = { @@ -196,6 +209,7 @@ void
> hardlockup_detector_perf_enable(void)
> if (!atomic_fetch_inc(&watchdog_cpus))
> pr_info("Enabled. Permanently consumes one hw-PMU counter.\n");
>
> + watchdog_touch_timestamp(smp_processor_id());
> perf_event_enable(this_cpu_read(watchdog_ev));
> }
>
> @@ -274,8 +288,10 @@ void __init hardlockup_detector_perf_restart(void)
> for_each_online_cpu(cpu) {
> struct perf_event *event = per_cpu(watchdog_ev, cpu);
>
> - if (event)
> + if (event) {
> + watchdog_touch_timestamp(cpu);
> perf_event_enable(event);
> + }
> }
> }
>
> --
> 2.16.2