Re: [PATCH v4] sched/debug: Use sched_debug_lock to serialize use of cgroup_path[] only

[Waiman Long]

On 4/5/21 8:18 PM, Steven Rostedt wrote:
> On Mon,  5 Apr 2021 19:42:03 -0400
> Waiman Long <longman@xxxxxxxxxx> wrote:
>
> > +/*
> > + * All the print_cpu() callers from sched_debug_show() will be allowed
> > + * to contend for sched_debug_lock and use group_path[] as their SEQ_printf()
> > + * calls will be much faster. However only one print_cpu() caller from
> > + * sysrq_sched_debug_show() which outputs to the console will be allowed
> > + * to use group_path[]. Another parallel console writer will have to use
> > + * a shorter stack buffer instead. Since the console output will be garbled
> > + * anyway, truncation of some cgroup paths shouldn't be a big issue.
> > + */
> > +#define SEQ_printf_task_group_path(m, tg, fmt...)			\
> > +{									\
> > +	unsigned long flags;						\
> > +	int token = m ? TOKEN_NA					\
> > +		      : xchg_acquire(&console_token, TOKEN_NONE);	\
> > +									\
> > +	if (token == TOKEN_NONE) {					\
> > +		char buf[128];						\
> > +		task_group_path(tg, buf, sizeof(buf));			\
> > +		SEQ_printf(m, fmt, buf);				\
> > +	} else {							\
> > +		spin_lock_irqsave(&sched_debug_lock, flags);		\
> > +		task_group_path(tg, group_path, sizeof(group_path));	\
> > +		SEQ_printf(m, fmt, group_path);				\
> > +		spin_unlock_irqrestore(&sched_debug_lock, flags);	\
> > +		if (token == TOKEN_ACQUIRED)				\
> > +			smp_store_release(&console_token, token);	\
> > +	}								\
> >   }
> >   #endif
> And you said my suggestion was complex!
>
> I'll let others review this.
This patch is actually inspired by your suggestion, though it is 
structured differently from your approach. I really want to thank you 
for your valuable feedback.

I realized that printing to a sequence file wasn't really a problem, 
only printing to console can be problematic. That is why I decided to 
allow unlimited use of group_path[] for those users and only one console 
writer is allowed to use it. As for calling touch_nmi_watchdog(), I am 
still thinking where will be the right place to do it, but it can be 
done with a separate patch, if needed.

Cheers,
Longman