On Sat, Mar 12, 2016 at 06:50:26PM -0500, Joshua Hunt wrote:
> While working on a script to restore all sysctl params before a series of
> tests I found that writing any value into the
> /proc/sys/kernel/{nmi_watchdog,soft_watchdog,watchdog,watchdog_thresh}
> causes them to call proc_watchdog_update(). Not only that, but when I
> wrote to these proc files in a loop I could easily trigger a soft lockup.
>
> [ 955.756196] NMI watchdog: enabled on all CPUs, permanently consumes one
> hw-PMU counter.
> [ 955.765994] NMI watchdog: enabled on all CPUs, permanently consumes one
> hw-PMU counter.
> [ 955.774619] NMI watchdog: enabled on all CPUs, permanently consumes one
> hw-PMU counter.
> [ 955.783182] NMI watchdog: enabled on all CPUs, permanently consumes one
> hw-PMU counter.
> [ 959.788319] NMI watchdog: BUG: soft lockup - CPU#4 stuck for 30s!
> [swapper/4:0]
> [ 959.788325] NMI watchdog: BUG: soft lockup - CPU#5 stuck for 30s!
> [swapper/5:0]
>
> There doesn't appear to be a reason for doing this work other every time a
> write occurs, so only do the work when the values change.
Hi Josh,
Thanks for the patch. I have no objections to it, but Uli and myself were
interested in the reason for the softlockups. Uli is going to provide a
test patch to see if his theory is correct. That way we fix the underlying
issue and then apply your patch on top. Make sense?
Cheers,
Don
>
> Signed-off-by: Josh Hunt <joh...@akamai.com>
> ---
> kernel/watchdog.c | 9 ++++++++-
> 1 file changed, 8 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/watchdog.c b/kernel/watchdog.c
> index b3ace6e..9acb29f 100644
> --- a/kernel/watchdog.c
> +++ b/kernel/watchdog.c
> @@ -923,6 +923,9 @@ static int proc_watchdog_common(int which, struct
> ctl_table *table, int write,
> * both lockup detectors are disabled if proc_watchdog_update()
> * returns an error.
> */
> + if (old == new)
> + goto out;
> +
> err = proc_watchdog_update();
> }
> out:
> @@ -967,7 +970,7 @@ int proc_soft_watchdog(struct ctl_table *table, int write,
> int proc_watchdog_thresh(struct ctl_table *table, int write,
> void __user *buffer, size_t *lenp, loff_t *ppos)
> {
> - int err, old;
> + int err, old, new;
>
> get_online_cpus();
> mutex_lock(&watchdog_proc_mutex);
> @@ -987,6 +990,10 @@ int proc_watchdog_thresh(struct ctl_table *table, int
> write,
> /*
> * Update the sample period. Restore on failure.
> */
> + new = ACCESS_ONCE(watchdog_thresh);
> + if (old == new)
> + goto out;
> +
> set_sample_period();
> err = proc_watchdog_update();
> if (err) {
> --
> 1.7.9.5
>
