Re: [RFC][PATCH] printk: Fixup the nmi printk mess

[Petr Mladek <pmladek@suse.cz>]

On Wed 2015-06-10 17:29:17, Peter Zijlstra wrote:
> On Wed, Jun 10, 2015 at 04:31:55PM +0200, Petr Mladek wrote:
> > If another NMI comes at this point, it will start filling the buffer
> > from the beginning. If it is fast enough, it might override the text
> > that we print above.
> 
> How so? If the cmpxchg succeeded and len == 0, we flushed everything and
> are done with it, if another NMI comes in and 'overwrites' it, that's
> fine, right?

Shame on me. Somehow I thought that there was xchg() and not cmpxchg().


> > > +static int vprintk_nmi(const char *fmt, va_list args)
> > > +{
> > > +	struct nmi_seq_buf *s = this_cpu_ptr(&nmi_print_seq);
> > > +	unsigned int len = seq_buf_used(&s->seq);
> > > +
> > > +	irq_work_queue(&s->work);
> > > +	seq_buf_vprintf(&s->seq, fmt, args);
>
> No, everything is strictly per cpu.

I do not know why but I expected that irq_work could get proceed on
any CPU. You are right. They are proceed on the same one.

Similar with the other mail. Sigh, I was too fast yesterday. I am
sorry for the noise.

Best Regards,
Petr