Re: [PATCH v2] softlockup: serialized softlockup's log

[lizhe.67@bytedance.com]

On Wed, 22 Nov 2023 14:05:21 -0800 akpm@linux-foundation.org wrote:

>> From: Li Zhe <lizhe.67@bytedance.com>
>> 
>> If multiple CPUs trigger softlockup at the same time with
>> 'softlockup_all_cpu_backtrace=0', the softlockup's logs will appear
>> staggeredly in dmesg, which will affect the viewing of the logs for
>> developer. Since the code path for outputting softlockup logs is not
>> a kernel hotspot and the performance requirements for the code are
>> not strict, locks are used to serialize the softlockup log output to
>> improve the readability of the logs.
>
>Seems reasonable, but...
>
>> --- a/kernel/watchdog.c
>> +++ b/kernel/watchdog.c
>> @@ -28,6 +28,8 @@
>>  #include <linux/kvm_para.h>
>>  
>>  static DEFINE_MUTEX(watchdog_mutex);
>> +/* This lock is used to prevent concurrent actions of softlockup output logs */
>> +static DEFINE_SPINLOCK(watchdog_output_lock);
>
>It would be a little neater to reduce the scope of this - move the
>definition into that little code block in watchdog_timer_fn() where it
>is actually used.

Yes. For this patch, it is more appropriate to put the definition in
watchdog_timer_fn(). It can be moved out in subsequent patches if
necessary. I will send a v3 patch to move it in watchdog_timer_fn().
Thanks for your advice.

>>  #if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HARDLOCKUP_DETECTOR_SPARC64)
>>  # define WATCHDOG_HARDLOCKUP_DEFAULT	1
>> @@ -514,6 +516,7 @@ static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
>>  		/* Start period for the next softlockup warning. */
>>  		update_report_ts();
>>  
>> +		spin_lock(&watchdog_output_lock);
>
>The hrtimer callout function is called from [soft]irq context, yes? 

Yes.

>Doesn't lockdep get upset when we take a spinlock in such a context?

My test results are the same as Doug Anderson's, things seemed OK.

>
>>  		pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
>>  			smp_processor_id(), duration,
>>  			current->comm, task_pid_nr(current));
>> @@ -523,6 +526,7 @@ static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
>>  			show_regs(regs);
>>  		else
>>  			dump_stack();
>> +		spin_unlock(&watchdog_output_lock);
>>  
>>  		if (softlockup_all_cpu_backtrace) {
>>  			trigger_allbutcpu_cpu_backtrace(smp_processor_id());