On Tue, Jun 18, 2024 at 03:51:44PM +0200, Petr Mladek wrote:
> On Tue 2024-06-18 07:54:50, Tony Lindgren wrote:
> > We are now matching and updating the preferred console, not adding it.
> > Let's update the naming accordingly to avoid confusion.
> >
> > --- a/drivers/tty/serial/serial_base_bus.c
> > +++ b/drivers/tty/serial/serial_base_bus.c
> > @@ -304,7 +305,7 @@ int serial_base_add_preferred_console(struct uart_driver *drv,
>
> I was curious whether this patch renamed everything. And it seems
> that it did not rename serial_base_add_preferred_console().
Oops, will update the naming for that too.
> > const char *port_match __free(kfree) = NULL;
> > int ret;
> >
> > - ret = serial_base_add_prefcon(drv->dev_name, port->line);
> > + ret = serial_base_match_and_update_prefcon(drv->dev_name, port->line);
> > if (ret)
> > return ret;
> >
>
> Honestly, I do not understand what are all these layers for.
> Especially, serial_base_match_and_update_prefcon() looks suspicious:
>
> static int serial_base_match_and_update_prefcon(const char *name, int idx)
> {
> const char *char_match __free(kfree) = NULL;
> const char *nmbr_match __free(kfree) = NULL;
> int ret;
>
> /* Handle ttyS specific options */
> if (strstarts(name, "ttyS")) {
> /* No name, just a number */
> nmbr_match = kasprintf(GFP_KERNEL, "%i", idx);
> if (!nmbr_match)
> return -ENODEV;
>
> ret = serial_base_match_and_update_one_prefcon(nmbr_match, name, idx);
> if (ret)
> return ret;
>
> /* Sparc ttya and ttyb */
> ret = serial_base_add_sparc_console(name, idx);
> if (ret)
> return ret;
> }
>
> /* Handle the traditional character device name style console=ttyS0 */
> char_match = kasprintf(GFP_KERNEL, "%s%i", name, idx);
> if (!char_match)
> return -ENOMEM;
>
> return serial_base_match_and_update_one_prefcon(char_match, name, idx);
> }
>
> It seems to try whether c->devname matches a number "X", or "ttySX".
> It even tries the sparc-specific transformations in
> serial_base_add_sparc_console()
>
> But this is the original format which does _not_ include ":".
> It never will be stored in c->devname and will never match.
Good catch, this won't do anything now with console_setup()
checking for ":" for deferred consoles. So we should revert commit
a0f32e2dd998 ("serial: core: Handle serial console options").
> I think that it has been the case even before this patchset.
For the earlier case, I tested things with serial handling removed
from console_setup() to let the serial layer handle the quirks.
With the new handling, we could just eventually defer the serial
consoles in console_setup(), and let the serial core do the quirk
handling. No immediate need for that though, that would be just
longer term clean-up.
> I think that we should remove these layers and check just
> the "DEVNAME:X.Y" format, aka "%s:%d.%d" [*].
Yes let's revert the quirk handling.
> [*] It would be nice to use the same printf format "%s:%d.%d"
> in both serial_base_device_init() and also in the functions
> matching the devname to make it clear that these are
> the same names. Heh, I just guess that these are the same
> names.
I'll do a separate clean-up patch for that, and that can then
be used for the match() too eventually.
Regards,
Tony
