Re: [PATCH] net/skbuff: silence warnings under memory pressure

[Sergey Senozhatsky]

On (09/06/19 16:55), Petr Mladek wrote:
> > I think we can queue significantly much less irq_work-s from printk().
> > 
> > Petr, Steven, what do you think?
> > 
> > Something like this. Call wake_up_interruptible(), switch to
> > wake_up_klogd() only when called from sched code.
> 
> Replacing irq_work_queue() with wake_up_interruptible() looks
> dangerous to me.
> 
> As a result, all "normal" printk() calls from the scheduler
> code will deadlock. There is almost always a userspace
> logger registered.

I don't see why all printk()-s should deadlock.

A "normal" printk() call will deadlock only when scheduler calls
"normal" printk() under rq or pi locks. But this is illegal anyway,
because console_sem up() calls wake_up_process() - the same function
wake_up_interruptible() calls. IOW "normal" printk() calls from
scheduler end up in scheduler, via console_sem->sched chain. We
already execute wake_up_process()->try_to_wake_up() in printk(),
even when a non-LOGLEVEL_SCHED printk() comes from scheduler.

What am I missing something?

	-ss