On Wed, 21 Sep 2016, David Madore wrote: > On Tue, Sep 20, 2016 at 03:50:09PM +0300, Mika Westerberg wrote: > > Does the machine have WDAT ACPI table (see /sys/firmware/acpi/tables/*)? > > If it does, you can try the new WDAT watchdog driver instead [1]. It > > still uses the same hardware, though but via set of instructions > > provided by the BIOS that should work (given the vendor has tested > > it on Windows). > > > > [1] http://www.mail-archive.com/linux-kernel@vger.kernel.org/msg1230607.html > > Thanks for pointing this out. My motherboard's BIOS does not have > this ACPI table, unfortunately, but it's at least good to know that > some do, and take the hardware watchdog seriously.
The ones that take the hardware watchdog seriously will command the power supply to do a power cycle when it triggers, which pretty much cuts power to everything that is not hanging off the +5VSB (standby power) line. Let's just say that SSDs don't like it, at all. Avoid at *all* *costs*. I have been using the kernel's software watchdog on most systems because of that: it just soft-reboots, which is good enough almost every time and doesn't mess with the SSDs. The proper fix is to have two levels of watchdogs, a soft reboot on time T for the first level, and a power cycle on time 5T (to give the BIOS enough of a time window to reset the second level watchdog during a soft reboot). -- Henrique Holschuh
