[PATCH v5 02/18] watchdog/perf: More properly prevent false positives with turbo modes

Petr Mladek pmladek at suse.com
Tue May 23 19:35:02 AEST 2023

On Fri 2023-05-19 10:18:26, Douglas Anderson wrote:
> Currently, in the watchdog_overflow_callback() we first check to see
> if the watchdog had been touched and _then_ we handle the workaround
> for turbo mode. This order should be reversed.
> Specifically, "touching" the hardlockup detector's watchdog should
> avoid lockups being detected for one period that should be roughly the
> same regardless of whether we're running turbo or not. That means that
> we should do the extra accounting for turbo _before_ we look at (and
> clear) the global indicating that we've been touched.

The ideal solution would be to reset the turbo-mode-related
variables when the watchdog is touched. And keep checking
watchdog_nmi_touch first.

But this ordering change should be good enough. It causes that
we always check watchdog_nmi_touch when the turbo-more-related
variables are already reset.

> NOTE: this fix is made based on code inspection. I am not aware of any
> reports where the old code would have generated false positives. That
> being said, this order seems more correct and also makes it easier
> down the line to share code with the "buddy" hardlockup detector.
> Fixes: 7edaeb6841df ("kernel/watchdog: Prevent false positives with turbo modes")
> Signed-off-by: Douglas Anderson <dianders at chromium.org>

Reviewed-by: Petr Mladek <pmladek at suse.com>

Best Regards,

More information about the Linuxppc-dev mailing list