Re: [RFC][PATCHv2 2/8] printk: introduce printing kernel thread

From: Pavel Machek <pavel@ucw.cz>
To: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>,
	Steven Rostedt <rostedt@goodmis.org>,
	Petr Mladek <pmladek@suse.com>, Jan Kara <jack@suse.cz>,
	Andrew Morton <akpm@linux-foundation.org>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Peter Zijlstra <peterz@infradead.org>,
	"Rafael J . Wysocki" <rjw@rjwysocki.net>,
	Eric Biederman <ebiederm@xmission.com>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Jiri Slaby <jslaby@suse.com>, Len Brown <len.brown@intel.com>,
	linux-kernel@vger.kernel.org
Subject: Re: [RFC][PATCHv2 2/8] printk: introduce printing kernel thread
Date: Fri, 7 Apr 2017 14:06:42 +0200	[thread overview]
Message-ID: <20170407120642.GB4756@amd> (raw)
In-Reply-To: <20170407081546.GD1091@jagdpanzerIV.localdomain>

[-- Attachment #1: Type: text/plain, Size: 2178 bytes --]

Hi!

> On (04/07/17 09:21), Pavel Machek wrote:
> > > spin_dump() and trigger_all_cpu_backtrace() result in a bunch of
> > > additional printk()-s so CPU0 has even more job to do in console_unlock(),
> > > while it still holds the contended spin_lock. and so on; there are
> > > many other examples.
> > > 
> > > so should we declare a "we can spend only 2 seconds in direct printk()
> > > and then must offload printing" rule? I don't think it's much better
> > > than a simpler "we always offload, as long as we think it's safe".
> > 
> > I believe we should do the 2 seconds rule. It allows us to print "some
> > messages delayed" message, so at least whoever is trying to debug the
> > crash will have the hints that he needs to look at the printk system.
> 
> do you mean panic()? in panic() we call console_flush_on_panic(),
> which immediately outputs all pending logbuf messages. printk()
> offloading does not happen there.

Not panic(). I have seen many crashes where we had printk(KERN_ERR)
and then hard hang. And the printk() was really important for debugging.

> > "we always offload, as long as we think it's safe" rule does not
> > really work, as printk() can not detect if it is safe or not.
> 
> but "2 seconds" rule has that "as long as we think it's safe" string
> attached as well. just because we do offloading. which is sometimes
> un-safe. so regardless the timeout value (0 seconds or 2 seconds) we
> still need some sort of a hint from the path that issues printk()
> because that path (panic, kexec, sysrq, etc.) knows for sure when
> things are abnormal. printk() is pretty clueless in this regard.
> /* well, I still think that EMERG loglevel thing is not completely
>  broken. */

Well, at least with my solution you know there are messages that were
not printed.

Yes, you'd still want to switch printk_now() for stuff like
panic(). But if you get it wrong (and you will), at least you will see
the "something is missing here" message in the log. 

								Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 181 bytes --]