On Wed 2017-06-14 18:11:28, Sergey Senozhatsky wrote:
> On (06/13/17 14:54), Petr Mladek wrote:
>
> > This patch modifies the code that enables the configured consoles.
> > It sets the CON_CONSDEV flag also when we register the first
> > console. It causes that one of the registered consoles will
> > always have CON_CONSDEV flag set.
>
> hm.... my impression was that we shouldn't set CON_CONSDEV if the
> console has no ->device. but then, once again, unregister_console()
> does not care and register_console() cares only in one place. so I'm
> a bit in doubt.
Great catch! It helped me to get even better picture.
If I get it correctly, the console driver for /dev/console is
found in tty_lookup_driver() by console_device(). Where
console_device() skips console drivers without any device.
CON_CONSDEV is basically handled only in register_console()
and unregister_console(). The driver with this flag is kept at
the beginning of the console_drivers list. It causes that
console_device() will return the console with this flag.
Also it is in sync with the commit cd3a1b8562d28490b33 ("printk: don't
prefer unsuited consoles on registration"). It added the check
for existing device into the fallback code in console_register()
so that console_device() might find a valid one.
BTW: The flag CON_CONSDEV first appeared in Linux 2.1.92.
It replaced CON_FIRST. Documentation/serial-console.txt
was modified so that the last console on the command line
will be used for /dev/console instead of the first one.
In each case, it seems that CON_CONSDEV is closely related
to an existing device and does not make much sense without it.
Heh, the handling in unregister_console() looks buggy. Well,
it probably does not break anything. The function does not change
the order of drivers and console_device() searches
the entire list until it finds a console with device.
Anyway, it would make sense to clean it as well.
Thanks a lot for review.
Best Regards,
Petr
