[PATCH v4] watchdog: Fix a watchdog crash in some configurations

[PATCH v4] watchdog: Fix a watchdog crash in some configurations

[john.hubbard]

From: John Hubbard <jhubbard@nvidia.com>

Commit <8fcf2cc768acd845c> in linux-next
introduced a regression in some configurations. Specifically,
with CONFIG_NO_HZ_FULL set, and CONFIG_NO_HZ_FULL_ALL *not* set,
the kernel will crash in lockup_detector_init(), due to a
NULL tick_nohz_full_mask pointer.

This is because the above commit uses tick_nohz_full_mask
(in lockup_detector_init), if CONFIG_NO_HZ_FULL is set, but
tick_nohz_full_mask only gets allocated if either:

    a) CONFIG_NO_HZ_FULL_ALL is set, or

    b) Someone passes in nohz_full=<any_value> on the boot
      args line.

To correct this, change lockup_detector_init so that it does
a runtime check in addition to the ifdef check. This fix is
simpler than my original proposed fix, thanks to Chris Metcalf
for that.

Signed-off-by: John Hubbard <jhubbard@nvidia.com>
---
 kernel/watchdog.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

diff --git a/kernel/watchdog.c b/kernel/watchdog.c
index 40fda2f..be34f7f 100644
--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -921,10 +921,13 @@ void __init lockup_detector_init(void)
 	set_sample_period();
 
 #ifdef CONFIG_NO_HZ_FULL
-	if (!cpumask_empty(tick_nohz_full_mask))
-		pr_info("Disabling watchdog on nohz_full cores by default\n");
-	cpumask_andnot(&watchdog_cpumask, cpu_possible_mask,
-		       tick_nohz_full_mask);
+	if (tick_nohz_full_enabled()) {
+		if (!cpumask_empty(tick_nohz_full_mask))
+			pr_info("Disabling watchdog on nohz_full cores by default\n");
+		cpumask_andnot(&watchdog_cpumask, cpu_possible_mask,
+			       tick_nohz_full_mask);
+	} else
+		cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
 #else
 	cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
 #endif
-- 
2.3.7

Re: [PATCH v4] watchdog: Fix a watchdog crash in some configurations

[Don Zickus]

On Tue, May 05, 2015 at 04:00:24PM -0700, john.hubbard@gmail.com wrote:
> From: John Hubbard <jhubbard@nvidia.com>
> 
> Commit <8fcf2cc768acd845c> in linux-next
> introduced a regression in some configurations. Specifically,
> with CONFIG_NO_HZ_FULL set, and CONFIG_NO_HZ_FULL_ALL *not* set,
> the kernel will crash in lockup_detector_init(), due to a
> NULL tick_nohz_full_mask pointer.
> 
> This is because the above commit uses tick_nohz_full_mask
> (in lockup_detector_init), if CONFIG_NO_HZ_FULL is set, but
> tick_nohz_full_mask only gets allocated if either:
> 
>     a) CONFIG_NO_HZ_FULL_ALL is set, or
> 
>     b) Someone passes in nohz_full=<any_value> on the boot
>       args line.
> 
> To correct this, change lockup_detector_init so that it does
> a runtime check in addition to the ifdef check. This fix is
> simpler than my original proposed fix, thanks to Chris Metcalf
> for that.

Acked-by: Don Zickus <dzickus@redhat.com>

> 
> Signed-off-by: John Hubbard <jhubbard@nvidia.com>
> ---
>  kernel/watchdog.c | 11 +++++++----
>  1 file changed, 7 insertions(+), 4 deletions(-)
> 
> diff --git a/kernel/watchdog.c b/kernel/watchdog.c
> index 40fda2f..be34f7f 100644
> --- a/kernel/watchdog.c
> +++ b/kernel/watchdog.c
> @@ -921,10 +921,13 @@ void __init lockup_detector_init(void)
>  	set_sample_period();
>  
>  #ifdef CONFIG_NO_HZ_FULL
> -	if (!cpumask_empty(tick_nohz_full_mask))
> -		pr_info("Disabling watchdog on nohz_full cores by default\n");
> -	cpumask_andnot(&watchdog_cpumask, cpu_possible_mask,
> -		       tick_nohz_full_mask);
> +	if (tick_nohz_full_enabled()) {
> +		if (!cpumask_empty(tick_nohz_full_mask))
> +			pr_info("Disabling watchdog on nohz_full cores by default\n");
> +		cpumask_andnot(&watchdog_cpumask, cpu_possible_mask,
> +			       tick_nohz_full_mask);
> +	} else
> +		cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
>  #else
>  	cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
>  #endif
> -- 
> 2.3.7
>