Re: [RFC][PATCHv3 2/5] printk: introduce printing kernel thread

[Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>]

On (07/03/17 15:34), Steven Rostedt wrote:
> > +#define PRINTK_FLOOD_DEFAULT_DELAY	10
> > +
> >  int printk_delay_msec __read_mostly;
> >  
> > +static inline void __printk_delay(int m)
> > +{
> > +	while (m--) {
> > +		mdelay(1);
> > +		touch_nmi_watchdog();
> > +	}
> > +}
> > +
> >  static inline void printk_delay(void)
> >  {
> > -	if (unlikely(printk_delay_msec)) {
> > -		int m = printk_delay_msec;
> > +	unsigned long flags;
> > +	u64 console_seen = 0, console_to_see;
> >  
> > -		while (m--) {
> > -			mdelay(1);
> > -			touch_nmi_watchdog();
> > -		}
> > +	if (printk_delay_msec) {
> > +		__printk_delay(printk_delay_msec);
> > +		return;
> > +	}
> > +
> 
> This had better be an option, and not default.

yes.

> And what happens if the printk caller happens to preempt the one
> doing the writes to consoles?

in short - we just burn CPU cycles. that case is broken.

that's mostly the reason behind PRINTK_FLOOD_DEFAULT_DELAY being quite
small.

one can simply do

	console_lock();
	printk();
	printk();
	....
	printk();
	console_unlock();

and trigger a useless throttling. a needed one in general case,
but useless in the given circumstances.

not sure if we can properly throttle printk in all of the cases.
we know that console_sem is locked, but we don't know what for.
is CPU that owns the console_sem is now in console_unlock() or
somewhere in fbcon, or anywhere else. we probably need not to
throttle printk() if we know that console_sem is already locked
by this_cpu and we simply call printk either from IRQ that
preempted console_unlock() on this_cpu or recursive printk from
console_unlock()... and so on.

	-ss