Message-ID: <aErcrCKcsi9cpANY@pathway.suse.cz>
Date: Thu, 12 Jun 2025 15:57:00 +0200
From: Petr Mladek <pmladek@...e.com>
To: Doug Anderson <dianders@...omium.org>
Cc: Marcos Paulo de Souza <mpdesouza@...e.com>,
	Steven Rostedt <rostedt@...dmis.org>,
	John Ogness <john.ogness@...utronix.de>,
	Sergey Senozhatsky <senozhatsky@...omium.org>,
	Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
	Jiri Slaby <jirislaby@...nel.org>,
	Jason Wessel <jason.wessel@...driver.com>,
	Daniel Thompson <danielt@...nel.org>,
	Richard Weinberger <richard@....at>,
	Anton Ivanov <anton.ivanov@...bridgegreys.com>,
	Johannes Berg <johannes@...solutions.net>,
	linux-kernel@...r.kernel.org, linux-serial@...r.kernel.org,
	kgdb-bugreport@...ts.sourceforge.net, linux-um@...ts.infradead.org
Subject: Re: [PATCH 4/7] drivers: serial: kgdboc: Check CON_SUSPENDED instead
 of CON_ENABLED

On Tue 2025-06-10 16:18:22, Doug Anderson wrote:
> Hi,
> 
> On Tue, Jun 10, 2025 at 1:03 PM Marcos Paulo de Souza
> <mpdesouza@...e.com> wrote:
> >
> > On Mon, 2025-06-09 at 13:13 -0700, Doug Anderson wrote:
> > > Hi,
> > >
> > > On Fri, Jun 6, 2025 at 7:54 PM Marcos Paulo de Souza
> > > <mpdesouza@...e.com> wrote:
> > > >
> > > > All consoles found on for_each_console are registered, meaning that
> > > > all of
> > > > them are CON_ENABLED. The code tries to find an active console, so
> > > > check if the
> > > > console is not suspended instead.
> > > >
> > > > Signed-off-by: Marcos Paulo de Souza <mpdesouza@...e.com>
> > > > ---
> > > >  drivers/tty/serial/kgdboc.c | 3 ++-
> > > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/tty/serial/kgdboc.c
> > > > b/drivers/tty/serial/kgdboc.c
> > > > index
> > > > 85f6c5a76e0fff556f86f0d45ebc5aadf5b191e8..af6d2208b8ddb82d62f33292b
> > > > 006b2923583a0d2 100644
> > > > --- a/drivers/tty/serial/kgdboc.c
> > > > +++ b/drivers/tty/serial/kgdboc.c
> > > > @@ -577,7 +577,8 @@ static int __init kgdboc_earlycon_init(char
> > > > *opt)
> > > >         console_list_lock();
> > > >         for_each_console(con) {
> > > >                 if (con->write && con->read &&
> > > > -                   (con->flags & (CON_BOOT | CON_ENABLED)) &&
> > > > +                   (con->flags & CON_BOOT) &&
> > > > +                   ((con->flags & CON_SUSPENDED) == 0) &&
> > >
> > > I haven't tried running the code, so I could easily be mistaken,
> > > but...
> > >
> > > ...the above doesn't seem like the correct conversion. The old
> > > expression was:
> > >
> > > (con->flags & (CON_BOOT | CON_ENABLED))
> > >
> > > That would evaluate to non-zero (true) if the console was _either_
> > > "boot" or "enabled".
> > >
> > > The new expression is is:
> > >
> > > (con->flags & CON_BOOT) && ((con->flags & CON_SUSPENDED) == 0)
> > >
> > > That's only true if the console is _both_ "boot" and "not suspended".
> >
> > My idea here was that the users of for_each_console would find the
> > first available console, and by available I would expect them to be
> > usable. In this case, is there any value for kgdboc to use a console
> > that is suspended? Would it work in this case?
> >
> > I never really used kgdboc, but only checking if the console was
> > enabled (which it's always the case here) was something that needed to
> > be fixed.
> >
> > Maybe I'm missing something here as well, so please let me know if I
> > should remove the new check.
> 
> So it's been 5 years since I wrote the code, but reading that I was
> checking for:
> 
>   (con->flags & (CON_BOOT | CON_ENABLED))
> 
> Makes me believe that this was the case when I wrote the code:
> 1. Early boot consoles (earlycon) were not marked as CON_ENABLED but
> were instead marked as CON_BOOT.
> 2. Once consoles became non-early they were moved to CON_ENABLED.
> 
> ...and the code was basically looking for any consoles with a matching
> name that were either boot consoles or normal/enabled consoles.
> 
> Is that a plausible theory? It's also possible that I just was
> confused when I code things up and that I really meant to write:

It is easy to get confused by the register_console() code. And
it has been even worse some years ago.

Anyway, the current code sets CON_ENABLED for all registered
consoles, including CON_BOOT consoles. The flag is cleared only
by console_suspend()[*] or unregister_console().

IMHO, kgdboc_earlycon_init() does not need to care about
console_suspend() because the kernel could not be suspended
during boot. Does this makes sense?

Also it does not need to care about unregister_console() because
the unregistered console won't be listed by for_each_console().

[*] The 1st patch in this patchset updates console_suspend()
    to set CON_SUSPENDED instead of clearing CON_ENABLED.
    It helps to make it clear that it is suspend-related.


Resume:

I would remove the check of CON_ENABLED or CON_SUSPENDED
from kgdboc_earlycon_init() completely.

IMHO, we should keep the check of CON_BOOT because we do not
want to register "normal" console drivers as kgdboc_earlycon_io_ops.
It is later removed by kgdboc_earlycon_deinit(). I guess
that the code is not ready to handle a takeover from normal
to normal (even the same) console driver.

To make it clear, I would use:

	for_each_console(con) {
		if (con->write && con->read &&
		    (con->flags & CON_BOOT) &&
		    (!opt || !opt[0] || strcmp(con->name, opt) == 0))
			break;
	}

And I would do this change as the 1st patch in the patchset
before we change the flag modified by console_suspend().

Best Regards,
Petr

Powered by blists - more mailing lists