On 26.07.21 18:54, Mantas Mikulėnas wrote: > On Mon, Jul 26, 2021 at 12:45 PM Jan Kiszka <jan.kiszka@xxxxxxxxxxx> wrote: >> >> On 26.07.21 11:40, Jan Kiszka wrote: >>> On 26.07.21 11:19, Mantas Mikulėnas wrote: >>>> Hello, >>>> >>>> I have a Dell Inspiron 15-5547 laptop, with systemd configured to set >>>> the watchdog to a 2-minute expiry (due to reasons): >>>> >>>> # /etc/systemd/system.conf >>>> [Manager] >>>> RuntimeWatchdogSec=2min >>>> >>>> So far this setting has worked without problems (including kernels >>>> 5.12.15 and 5.13.1); however, with kernel 5.13.4 the system inevitably >>>> reboots after a few minutes of uptime. >>>> >>>> I have tracked the issue down to commit 5e65819a006e "watchdog: >>>> iTCO_wdt: Account for rebooting on second timeout" in the 5.13.x >>>> branch (commit cb011044e34c upstream). There are no unexpected reboots >>>> when running 5.13.4 with this commit reverted. >>>> >>>> Indeed with the original 5.13.4 kernel, `wdctl` always reports >>>> "Timeleft:" counting down from 60 seconds (sometimes very nearly >>>> reaching 0), even though "Timeout" is still reported to be 120. >>>> >>>> (systemd pokes the watchdog as part of its main loop, trying to so >>>> approximately "between 1/4 and 1/2" of the configured interval. >>>> According to wdctl these pings usually happen every 35-50 seconds but >>>> sometimes nearly at the 60-second mark, and thanks to the kernel now >>>> also dividing the requested expiry by /2 which systemd is unaware of, >>>> sometimes this ends up being a *very* close race to 0.) >>>> >>>> This is a Haswell-era machine (i7-4510U) and seems to have a "version >>>> 0" watchdog: >>>> >>>> Jul 26 11:34:04 archlinux kernel: Linux version 5.13.4-arch2-1 >>>> (linux@archlinux) (gcc (GCC) 11.1.0, GNU ld (GNU Binutils) 2.36.1) #1 >>>> SMP PREEMPT Thu, 22 Jul 2021 20:46:28 +0000 >>>> Jul 26 11:34:14 frost kernel: iTCO_vendor_support: vendor-support=0 >>>> Jul 26 11:34:14 frost kernel: iTCO_wdt iTCO_wdt.3.auto: Found a Lynx >>>> Point_LP TCO device (Version=2, TCOBASE=0x1860) >>>> Jul 26 11:34:14 frost systemd[1]: Using hardware watchdog 'iTCO_wdt', >>>> version 0, device /dev/watchdog >>>> Jul 26 11:34:14 frost systemd[1]: Set hardware watchdog to 2min. >>>> Jul 26 11:34:14 frost kernel: iTCO_wdt iTCO_wdt.3.auto: initialized. >>>> heartbeat=30 sec (nowayout=0) >>>> >>> >>> Could you printk SMI_EN(p) in iTCO_wdt_set_timeout() >>> (drivers/watchdog/iTCO_wdt.c)? This is where we decide whether SMIs are >>> working, thus the countdown will only run once. Apparently, something is >>> wrong with the detection on this system. >>> >> >> Wait, found it: >> >> diff --git a/drivers/watchdog/iTCO_wdt.c b/drivers/watchdog/iTCO_wdt.c >> index b3f604669e2c..643c6c2d0b72 100644 >> --- a/drivers/watchdog/iTCO_wdt.c >> +++ b/drivers/watchdog/iTCO_wdt.c >> @@ -362,7 +362,7 @@ static int iTCO_wdt_set_timeout(struct watchdog_device *wd_dev, unsigned int t) >> * Otherwise, the BIOS generally reboots when the SMI triggers. >> */ >> if (p->smi_res && >> - (SMI_EN(p) & (TCO_EN | GBL_SMI_EN)) != (TCO_EN | GBL_SMI_EN)) >> + (inl(SMI_EN(p)) & (TCO_EN | GBL_SMI_EN)) != (TCO_EN | GBL_SMI_EN)) >> tmrval /= 2; >> >> /* from the specs: */ > > Rebuilt with this and it fixes the issue, thanks. > Thanks for confirming! Please also reply with a "Tested-by: ..." on the patch I sent earlier. Jan -- Siemens AG, T RDA IOT Corporate Competence Center Embedded Linux
