Re: [PATCH] printk/nmi: Prevent deadlock when serializing NMI backtraces

[Petr Mladek]

On Fri 2018-05-18 15:38:20, Sergey Senozhatsky wrote:
> On (05/18/18 11:07), Sergey Senozhatsky wrote:
> > 
> > if (this_cpu_read(printk_context) & PRINTK_SAFE_CONTEXT_MASK) ||
> > 	raw_spin_is_locked(&logbuf_lock)
> > 
> > just to check per-CPU `printk_context' first and only afterwards
> > access the global `logbuf_lock'. printk_nmi_enter() happens on
> > every CPU, so maybe we can avoid some overhead by checking the
> > local per-CPU data first.
> 
> Nah, may be it won't. This, probably, would have been the case if we
> had continue to call console drivers from printk_safe section [at least].
> CPUs don't spend that much time in printk_safe sections.

Yeah, I do not think that the more complicated code is worth it.

There is really minimal chance to hit printk_safe context on the given
CPU. The eventual win would be negligible to the cost of printk().
In case of trigger_all_cpu_backtrace(), most CPUs spend a lot
of time waiting for the spinlock in nmi_cpu_backtrace() anyway.

Thanks for the ack.

Best Regards,
Petr