[External] Re: [PATCH 1/2] watchdog: refactor watchdog_hld functionality

[Ian Rogers]

On Fri, Sep 5, 2025 at 4:57 PM Doug Anderson <dianders at chromium.org> wrote:
>
> Hi,
>
> On Wed, Sep 3, 2025 at 4:56 AM yunhui cui <cuiyunhui at bytedance.com> wrote:
> > I’ve read your linked content, which details the pros and cons of perf
> > watchdog and buddy watchdog.
> > I think everyone will agree on choosing one as the default.
> > It seems there’s no kernel/watchdog entry in MAINTAINERS—what’s next
> > for these two approaches?
>
> I guess to start, someone (you?) should send some patches to the list.
> Maybe one patch to make buddy the default and one to change the
> description of the "perf" lockup detector say that its usage is
> discouraged, that it might be removed, that people should use the
> "buddy" detector instead, and that if there's a reason someone needs
> the "perf" detector instead of the buddy one then they should make
> some loud noises.
>
> You'd want to CC folks who were involved in recent watchdog changes
> and make sure to CC Andrew (akpm). If folks react positive and Andrew
> agrees then he'll likely land the the patches and we'll have made
> forward progress. :-)

+1

There are also things like /proc/sys/kernel/nmi_watchdog being used to
enable/disable the hard lookup detector. If we could move that to a
unique file so that perf is less confused in places like:
https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/perf/util/util.c#n70
ie. perf shouldn't warn about the NMI watchdog being enabled and
taking a perf event when it doesn't.

Thanks,
Ian