Re: [RFC 2/2] printk: Add more information about the printk caller

[Petr Mladek]

On Thu 2020-09-24 11:17:56, Sergey Senozhatsky wrote:
> On (20/09/23 15:56), Petr Mladek wrote:
> [..]
> >  	/*
> >  	 * To reduce unnecessarily reopening, first check if the descriptor
> > -	 * state and caller ID are correct.
> > +	 * state and caller infromation are correct.
> >  	 */
> > -	d_state = desc_read(desc_ring, id, &desc, NULL, &cid);
> > -	if (d_state != desc_committed || cid != caller_id)
> > +	d_state = desc_read(desc_ring, id, &desc, NULL, &cal);
> > +	if (d_state != desc_committed ||
> > +	    cal.pid != caller->pid ||
> > +	    cal.cpu_ctx != caller->cpu_ctx) {
> 
> You probably might want to factor out ctx check into a static
> inline helper. Since you use this check in several places, and
> we may check more context fields in the future.

Makes sense.

> [..]
> > +/* Information about the process and context that adds the message */
> > +struct printk_caller {
> > +	pid_t pid;	/* thread id */
> > +	u32 cpu_ctx;	/* processor id and interrupt context */
> > +};
> 
> A question. Suppose we have a task which does
> 
> 	CPU0
> 
> 	pr_err(...);
> 
> 	preempt_disable();
> 	pr_err(...);
> 	preempt_enable();
> 
> 	pr_err(...);
> 
> 	rcu_read_lock();
> 	pr_info(...);
> 	rcu_read_unlock();
> 
> Should we distinguish those as 3 different contexts?
> 
> - normal printk
> - printk under disabled preemption (affects scheduling)
> - printk under RCU read side lock (affects RCU grace periods)

Good question. Well, these contexts could not get detected in
PREEMPT_NONE or PREEMPT_VOLUNTARY. Also I wonder where it would
stop. There are several RCU flavors.

I would not distinguish them unless there is a real demand.

Best Regards,
Petr