Re: replay log: Re: [PATCH printk v4 38/39] printk: relieve console_lock of list synchronization duties

From: Petr Mladek <pmladek@suse.com>
To: John Ogness <john.ogness@linutronix.de>
Cc: Sergey Senozhatsky <senozhatsky@chromium.org>,
	Steven Rostedt <rostedt@goodmis.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	linux-kernel@vger.kernel.org
Subject: Re: replay log: Re: [PATCH printk v4 38/39] printk: relieve console_lock of list synchronization duties
Date: Tue, 15 Nov 2022 19:36:06 +0100	[thread overview]
Message-ID: <Y3PcFvz1BXrDcyPm@alley> (raw)
In-Reply-To: <87r0y4m955.fsf@jogness.linutronix.de>

On Tue 2022-11-15 18:21:34, John Ogness wrote:
> On 2022-11-15, Petr Mladek <pmladek@suse.com> wrote:
> >>> --- a/kernel/printk/printk.c
> >>> +++ b/kernel/printk/printk.c
> >>> @@ -3334,6 +3330,11 @@ void register_console(struct console *newcon)
> >>>  		 * boot console that is the furthest behind.
> >>>  		 */
> >>>  		if (bootcon_registered && !keep_bootcon) {
> >>> +			/*
> >>> +			 * Hold the console_lock to guarantee safe access to
> >>> +			 * console->seq.
> >>> +			 */
> >>> +			console_lock();
> >>>  			for_each_console(con) {
> >>>  				if ((con->flags & CON_BOOT) &&
> >>>  				    (con->flags & CON_ENABLED) &&
> >>> @@ -3341,6 +3342,7 @@ void register_console(struct console *newcon)
> >>>  					newcon->seq = con->seq;
> >>>  				}
> >>>  			}
> >>> +			console_unlock();
> >
> > So, without the above two hunks:
> >
> > Reviewed-by: Petr Mladek <pmladek@suse.com>
> 
> Note that we actually need those hunks to guarantee a consistent @seq
> value. The console_lock is the only synchronization mechanism available
> to read console->seq.

Yes, we need a solution. But it does not need to be in this patch.

This patch removes console_lock() on some locations. But this
particular code was called without console_lock() even before
this patch.

Note that the regression was added in the 3rd patch that moved
this code outside console_lock().

Maybe, the easiest solution would be to do in the 3rd patch [*]:

	} else {
		/* Begin with next message. */
		newcon->seq = prb_next_seq(prb);
		/*
		 * Try hard to show the pending messages on boot consoles.
		 * so that the new console does not start too late.
		 */
		pr_flush();
	}

It should behave as good and as bad as the original code.

[*] Or move the code and add this change before the 3rd patch
    to keep this questionable solution separated and avoid
    the regression.

Best Regards,
Petr