Re: 2.6.16-mm1

Re: 2.6.16-mm1

[Russell King]

On Thu, Mar 23, 2006 at 03:31:45PM +0100, Michal Piotrowski wrote:
> Hi,
> 
> On 23/03/06, Andrew Morton <akpm@osdl.org> wrote:
> >
> > ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.16/2.6.16-mm1/
> >
> 
> Something went wrong with serial code.
> 
> make
> [..]
>   CC      init/version.o
>   LD      init/built-in.o
>   LD      .tmp_vmlinux1
> drivers/built-in.o: In function
> `serial_pnp_probe':/usr/src/linux-mm/drivers/serial/8250_pnp.c:435:
> undefined reference to `serial8250_register_port'
> drivers/built-in.o: In function
> `serial_pnp_remove':/usr/src/linux-mm/drivers/serial/8250_pnp.c:447:
> undefined reference to `serial8250_unregister_port'
> drivers/built-in.o: In function
> `pciserial_suspend_ports':/usr/src/linux-mm/drivers/serial/8250_pci.c:1690:
> undefined reference to `serial8250_suspend_port'
> drivers/built-in.o: In function
> `pciserial_resume_ports':/usr/src/linux-mm/drivers/serial/8250_pci.c:1706:
> undefined reference to `serial8250_resume_port'
> drivers/built-in.o: In function
> `pciserial_remove_ports':/usr/src/linux-mm/drivers/serial/8250_pci.c:1665:
> undefined reference to `serial8250_unregister_port'
> drivers/built-in.o: In function
> `pciserial_init_ports':/usr/src/linux-mm/drivers/serial/8250_pci.c:1640:
> undefined reference to `serial8250_register_port'
> make[1]: *** [.tmp_vmlinux1] B??d 1
> make: *** [_all] B??d 2
> 
> Here is config http://www.stardust.webpages.pl/files/mm/2.6.16-mm1/mm-config

CONFIG_SERIAL_8250=m
CONFIG_SERIAL_8250_PCI=y
CONFIG_SERIAL_8250_PNP=y

That's an illegal configuration.

+config SERIAL_8250_PCI
+       tristate "8250/16550 PCI device support" if EMBEDDED
+       depends on SERIAL_8250 && PCI
+       default y

if SERIAL_8250 is 'm' there's no way that SERIAL_8250_PCI should be 'y'.
Maybe it's a bug in kconfig - seems like it.  Let's try some experiments:

config SYM_Y
        bool
        default y

config SYM_M
        tristate
        default m

config SYM_D
        tristate 'SYM_D'
        depends on SYM_M && SYM_Y

With this I'm offered:

SYM_D (SYM_D) [N/m] (NEW)

so Kconfig thinks the only legal values for SYM_D are 'n' and 'm', and
the default is 'n'.

If I add a default line to SYM_D thusly:

config SYM_D
        tristate 'SYM_D'
        depends on SYM_M && SYM_Y
        default y

I'm now offered:

SYM_D (SYM_D) [M/n] (NEW)

Okay, so the default is now 'm', but the legal values are still only 'n'
and 'm'.  I can only select 'm' or 'n', and this is what I end up with in
the config file.  Now, if I remove the prompt text:

config SYM_D
        tristate
        depends on SYM_M && SYM_Y
        default y

and hey presto, suddenly 'y' becomes a legal value.

CONFIG_SYM_Y=y
CONFIG_SYM_M=m
CONFIG_SYM_D=y

So it would seem to be a Kconfig bug.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core

Re: 2.6.16-mm1

[Michal Piotrowski]

Hi Russell,

On 23/03/06, Russell King <rmk+lkml@arm.linux.org.uk> wrote:
> CONFIG_SERIAL_8250=m
> CONFIG_SERIAL_8250_PCI=y
> CONFIG_SERIAL_8250_PNP=y
>
> That's an illegal configuration.
>
[snip]
>
> So it would seem to be a Kconfig bug.

Thanks for explanation of the problem

Everything is ok, when I state CONFIG_SERIAL_8250=y in kernel config.

Regards,
Michal

--
Michal K. K. Piotrowski
LTG - Linux Testers Group
(http://www.stardust.webpages.pl/ltg/wiki/)

Re: 2.6.16-mm1

[Roman Zippel]

Hi,

On Thu, 23 Mar 2006, Russell King wrote:

> Okay, so the default is now 'm', but the legal values are still only 'n'
> and 'm'.  I can only select 'm' or 'n', and this is what I end up with in
> the config file.  Now, if I remove the prompt text:
> 
> config SYM_D
>         tristate
>         depends on SYM_M && SYM_Y
>         default y
> 
> and hey presto, suddenly 'y' becomes a legal value.
> 
> CONFIG_SYM_Y=y
> CONFIG_SYM_M=m
> CONFIG_SYM_D=y
> 
> So it would seem to be a Kconfig bug.

No, it's not a bug, that's really the correct behaviour. It has its roots 
in the cml1 converter, where statements like this:

if [ "$CONFIG_FOO" = "y" ]; then
  define_tristate CONFIG_BAR y
fi

would become:

config BAR
	default y
	depends on FOO=y

The basic idea is that the dependency only enables the default and the 
default sets the symbol to whatever you want.

I thought about the other behaviour, but at that time the if syntax or 
select didn't exists yet, so it really wasn't a problem yet and I decided 
to keep it closer to cml1 instead.

In the meantime the Kconfig syntax has grown and this is not the first 
time this has come up. I'm not completely opposed against changing the 
behaviour, but I don't want to do it just for fun either. It would require 
at least to grep through the current Kconfig rules to check whether 
something depends on the current behaviour.

As alternative you can change the default to "SYM_M" (basically repeating 
the dependency without the booleans).

bye, Roman

Re: 2.6.16-mm1

[Russell King]

On Fri, Mar 24, 2006 at 12:28:27PM +0100, Roman Zippel wrote:
> Hi,
> 
> On Thu, 23 Mar 2006, Russell King wrote:
> 
> > Okay, so the default is now 'm', but the legal values are still only 'n'
> > and 'm'.  I can only select 'm' or 'n', and this is what I end up with in
> > the config file.  Now, if I remove the prompt text:
> > 
> > config SYM_D
> >         tristate
> >         depends on SYM_M && SYM_Y
> >         default y
> > 
> > and hey presto, suddenly 'y' becomes a legal value.
> > 
> > CONFIG_SYM_Y=y
> > CONFIG_SYM_M=m
> > CONFIG_SYM_D=y
> > 
> > So it would seem to be a Kconfig bug.
> 
> No, it's not a bug, that's really the correct behaviour. It has its roots 
> in the cml1 converter, where statements like this:
> 
> if [ "$CONFIG_FOO" = "y" ]; then
>   define_tristate CONFIG_BAR y
> fi
> 
> would become:
> 
> config BAR
> 	default y
> 	depends on FOO=y

Okay, so going to the exact problem case, the behaviour we require is:

SERIAL_8250	PCI	EMBEDDED	gives SERIAL_8250_PCI
	n	X	X		n
	X	n	X		n
	m	y	n		m
	y	y	n		y
	m	y	y		user selects 'm' or 'n'
	y	y	y		user selects 'y', 'm' or 'n'

the correct way to tell Kconfig to give us that is:

+config SERIAL_8250_PCI
+       tristate "8250/16550 PCI device support" if EMBEDDED
+       depends on SERIAL_8250 && PCI
+       default SERIAL_8250
+       help
+         This builds standard PCI serial support. You may be able to
+         disable this feature if you only need legacy serial support.
+         Saves about 9K.

?

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core

Re: 2.6.16-mm1

[Roman Zippel]

Hi,

On Fri, 24 Mar 2006, Russell King wrote:

> the correct way to tell Kconfig to give us that is:
> 
> +config SERIAL_8250_PCI
> +       tristate "8250/16550 PCI device support" if EMBEDDED
> +       depends on SERIAL_8250 && PCI
> +       default SERIAL_8250
> +       help
> +         This builds standard PCI serial support. You may be able to
> +         disable this feature if you only need legacy serial support.
> +         Saves about 9K.
> 
> ?

Yes, this should do it.

bye, Roman