Since 4.12.0 kernel, iTCO_wdt watchdog driver reboot too soon on my Intel hardwares (i.e Dell M610, Thinkpad X1 carbon, ASUSTeK H81I-PLUS/i3-4160). With a 4.11.0 my watchdog timeout was set to 30s. About halt way (15s), the watchdog daemon (systemd here) reset the timeout. With 4.12.[0123], when you set it to 30s, it reboots at 15s. Snap. Rollback to 4.11 fix the issue. Set the timeout to a higher value, like 60s fix too.
I confirm this problem for a Lenovo T440s Laptop and a Desktop with Asus H87-Plus mainboard. Commit 1fccb73011ea8a5fa0c6d357c33fa29c695139ea seems like an obvious suspect.
I can reproduce exactly this issue on an ASUS UX31A with kernel version 4.12.3 after upgrading from 4.11.9. The described work-around works for me as well.
Bug confirmed. I reverted patch 1fccb73011ea8a5fa0c6d357c33fa29c695139ea.
Any chance that this fix be submitted to stable/linux-4.12.y, stable/linux-4.13.y ?
Could you please test if the turn_SMI_watchdog_clear_off module option works for you (e.g. turn_SMI_watchdog_clear_off=99)?
(In reply to Paolo Bonzini from comment #5) Sorry for the late reply. No, turn_SMI_watchdog_clear_off=99 does not fix the issue on my Dell Latitude E7470. Kernel with patch from this bug applied reports: [ 13.593418] iTCO_vendor_support: vendor-support=0 [ 13.594447] iTCO_wdt: Intel TCO WatchDog Timer Driver v1.11 [ 13.594577] iTCO_wdt: Found a Intel PCH TCO device (Version=4, TCOBASE=0x0400) [ 13.596893] iTCO_wdt: initialized. heartbeat=30 sec (nowayout=0)
FTR, applied in stable/linux-4.13.y: ff04be02de1b watchdog: Revert "iTCO_wdt: all versions count down twice" Not in stable/linux-4.12.y yet.
Closing, as the bug is fixed. The fix never went to stable/linux-4.12.y because it was already EoL at that time.