Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Fri, 14 Jul 2023 11:41:27 +0200
From:   Petr Mladek <pmladek@suse.com>
To:     Sergey Senozhatsky <senozhatsky@chromium.org>
Cc:     John Ogness <john.ogness@linutronix.de>,
        Steven Rostedt <rostedt@goodmis.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH printk v2 2/5] printk: Add NMI safety to
 console_flush_on_panic() and console_unblank()
Message-ID: <ZLEYR-eaZTWXO2Ld@alley>
References: <20230710134524.25232-1-john.ogness@linutronix.de>
 <20230710134524.25232-3-john.ogness@linutronix.de>
 <ZK14p-ocWuuHkSAQ@alley>
 <878rbkrg16.fsf@jogness.linutronix.de>
 <ZLANiK_1YoBu1YpU@alley>
 <20230714040049.GA81525@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230714040049.GA81525@google.com>
Precedence: bulk

On Fri 2023-07-14 13:00:49, Sergey Senozhatsky wrote:
> On (23/07/13 16:43), Petr Mladek wrote:
> > 
> > Simple removal of console_trylock() in console_flush_on_panic() would
> > cause that other CPUs might still be able to take it and race.
> > The problem is avoided by checking panic_in_progress() in console_lock()
> > and console_trylock(). They will never succeed on non-panic CPUs.
> > 
> 
> In theory, we also can have non-panic CPU in console_flush_all(),
> which should let panic CPU to take over the next time it checks
> abandon_console_lock_in_panic() (other_cpu_in_panic() after 5/5),
> but it may not happen immediately. I wonder if we somehow can/want
> to "wait" in console_flush_on_panic() for non-panic CPU handover?

Good point. It might actually be any console_lock() owner,
including printk() on other CPU.

I think that we might need to add some wait() as we did in the last
attempt, see the commit b87f02307d3cfbda76852 ("printk: Wait for
the global console lock when the system is going down").

Anyway, it will be more important after introducing the kthreads.
There is a non-trivial chance that they would block the lock.
They might be busy when handling a message printed right before
the panic() was called. At least, this is what I saw in the last
attempt to introduce the kthreads.

But maybe, it will be somehow hidden in the new atomic lock.
It might be passed to a printk context with a higher priority
and it uses some wait internally, see the waiting in the patch
https://lore.kernel.org/all/20230302195618.156940-7-john.ogness@linutronix.de/

Anyway, this patch does not make it worse. It just ignores the
potential console_lock owner in console_flush_on_panic() another
way.

Best Regards,
Petr