Re: [RFC][PATCHv2 3/8] printk: offload printing from wake_up_klogd_work_func()

[Sergey Senozhatsky]

Hi Petr,

sorry for the delay.

On (03/31/17 16:56), Petr Mladek wrote:
[..]
> > diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
> > index ab6b3b2a68c6..1927b5cb5cbe 100644
> > --- a/kernel/printk/printk.c
> > +++ b/kernel/printk/printk.c
> > @@ -2741,8 +2741,16 @@ static void wake_up_klogd_work_func(struct irq_work *irq_work)
> >  		 * If trylock fails, someone else is doing the printing.
> >  		 * PRINTK_PENDING_OUTPUT bit is cleared by console_unlock().
> >  		 */
> > -		if (console_trylock())
> > -			console_unlock();
> > +		if (printk_kthread_enabled()) {
> > +			wake_up_process(printk_kthread);
> 
> Note that the relation between printk_kthread_enabled()
> and wake_up_process() is racy. The conditions might change
> between these two calls. It looks fine here, well almost.
> 
> The critical point is in vprintk_emit(). It must use the emergency
> mode (call the consoles directly) when it is called from a process
> that started the emergency mode.

hm, we don't guarantee this. printk(), both in threaded and in
emergency modes, can fail to acquire console_sem.

> We could be more relaxed here. IMHO, the only sensitive situation
> is if printk_deferred() is used in the emergency context.
> We might want to use the emergency mode here as well but
> it is not guaranteed.

hm, I don't think any path does

	printk_emergency_begin()
	printk_deferred()
	printk_emergency_end()

and expects logbuf output to be flushed by the time it does
printk_emergency_end(). it's most likely something like this

	printk_emergency_begin()
	printk()
	printk_emergency_end()

the expectations here are more reasonable, but still, no
guarantees are provided (even in non-kthreaded printk mode).

> A solution might be to add one more bit, e.g.
> PRINTK_PENDING_EMERGENCY_OUTPUT. We should force the emergency mode
> here when it is set. It should be cleared together with the normal
> PRINTK_PENDING_OUTPUT.
> 
> Or do you think that this is a corner case that we could
> ignore for now?

hm, I guess we don't really count on irq_work in emergency
situations. but I need more time to think. good questions, Petr.

	-ss