On 8/18/21 7:49 AM, Hans de Goede wrote:
Hi,
On 8/18/21 4:13 PM, Guenter Roeck wrote:
On 8/18/21 2:57 AM, Jose Noguera wrote:
Hello all!
I’m Jose, working in Red Hat on Fedora IoT and RHEL For Edge related projects.
While trying to add the feature of figuring out whether the current boot was triggered or not by a hardware watchdog using wdctl, we’ve found an odd behaviour in the drivers definition that we would like to check with you.
Looking specifically for the flag WDIOF_CARDRESET, we can find 40 files that contain it:
$ grep -rl CARDRESET drivers/watchdog/ | wc -l
40
but only 19 of them have the flag advertised in the options field of the watchdog_info struct returned by the WDIOC_GETSUPPORT ioctl.
This leads to wdctl not showing WDIOF_CARDRESET for drivers like i6300esb, even when the board had been reset this way:
$ sudo wdctl
Device: /dev/watchdog0
Identity: i6300ESB timer [version 0]
Timeout: 30 seconds
Pre-timeout: 0 seconds
FLAG DESCRIPTION STATUS BOOT-STATUS
KEEPALIVEPING Keep alive ping reply 1 0
MAGICCLOSE Supports magic close char 0 0
SETTIMEOUT Set timeout (in seconds) 0 0
Working with Hans (in CC), we added a little patch (BugZilla ticket: 1993983) on wdctl and this would be the output of the command when it was a card reset triggered boot:
$ sudo wdctl
Device: /dev/watchdog0
Identity: i6300ESB timer [version 0]
Thank you all for your time,
Jose
Timeout: 30 seconds
Pre-timeout: 0 seconds
FLAG DESCRIPTION STATUS BOOT-STATUS
CARDRESET Card previously reset the CPU 1 1
KEEPALIVEPING Keep alive ping reply 1 0
MAGICCLOSE Supports magic close char 0 0
SETTIMEOUT Set timeout (in seconds) 0 0
So our question is, may we know what is intended to be present in ident.options? What should the API call WDIOC_GETSUPPORT return in the options field?
Search for WDIOC_GETSUPPORT in Documentation/watchdog/watchdog-api.rst.
I don't see any ambiguity there. Patches welcome.
Ok, so drivers which may set CARDRESET in their GETBOOTSTATUS reply, but
don't advertise this in their GETSUPPORT watchdog_info.options reply
are buggy and should be fixed, got it. Thanks.
I've made a note about fixing this in a possible-kernel-projects document
which I keep for when people who are interested in kernel development
ask me for projects.
With an add-on: If the driver in question is an old-style driver, anyone
affected by the problem should really convert the driver to support
the watchdog subsystem.
Thanks,
Guenter
Regards,
Hans
