T30 boot hang with CONFIG_PM_LEGACY=n

T30 boot hang with CONFIG_PM_LEGACY=n

[Len Brown]

My T30 doesn't boot if CONFIG_PM_LEGACY=n
unless "apm=off".

Seems that this build option causes PM_IS_ACTIVE() to be constant 0,
which disables APM's check to see if ACPI is running:

apm_init()
	...
        if (PM_IS_ACTIVE()) {
                printk(KERN_NOTICE "apm: overridden by ACPI.\n");
                apm_info.disabled = 1;
                return -ENODEV;
        }

Apparently when CONFIG_PM_LEGACY was created, CONFIG_APM depended
on it, so apm.c wasn't built.  But that dependency was later removed so it is now possible
to build APM with its check for ACPI  mysteriously disabled -- much to the unhappyness
of my T30.

What's the plan here?  Perhaps that plan should be written down
in kernel/power/Kconfig? 

config PM_LEGACY
        bool "Legacy Power Management API"
        depends on PM
        default y
        ---help---
           Support for pm_register() and friends.

           If unsure, say Y.

thanks,
-Len

Re: T30 boot hang with CONFIG_PM_LEGACY=n

[Dave Jones]

On Wed, Sep 27, 2006 at 12:00:40AM -0400, Len Brown wrote:
 > My T30 doesn't boot if CONFIG_PM_LEGACY=n
 > unless "apm=off".
 > 
 > Seems that this build option causes PM_IS_ACTIVE() to be constant 0,
 > which disables APM's check to see if ACPI is running:
 > 
 > apm_init()
 > 	...
 >         if (PM_IS_ACTIVE()) {
 >                 printk(KERN_NOTICE "apm: overridden by ACPI.\n");
 >                 apm_info.disabled = 1;
 >                 return -ENODEV;
 >         }
 > 
 > Apparently when CONFIG_PM_LEGACY was created, CONFIG_APM depended
 > on it, so apm.c wasn't built.  But that dependency was later removed so it is now possible
 > to build APM with its check for ACPI  mysteriously disabled -- much to the unhappyness
 > of my T30.
 > 
 > What's the plan here?

Good question.  There's no real replacement for pm_active in the non-legacy
config afaik. Perhaps the cleanest alternative is to undeprecate that macro?
The only other option I can think of is doing something like this in apm.c ...

+#ifdef CONFIG_ACPI
-        if (PM_IS_ACTIVE()) {
                 printk(KERN_NOTICE "apm: overridden by ACPI.\n");
                 apm_info.disabled = 1;
                 return -ENODEV;
-         }
+#endif

	Dave

Re: T30 boot hang with CONFIG_PM_LEGACY=n

[Matthew Garrett]

On Wed, Sep 27, 2006 at 12:20:29AM -0400, Dave Jones wrote:

> Good question.  There's no real replacement for pm_active in the non-legacy
> config afaik. Perhaps the cleanest alternative is to undeprecate that macro?
> The only other option I can think of is doing something like this in apm.c ...
> 
> +#ifdef CONFIG_ACPI
> -        if (PM_IS_ACTIVE()) {
>                  printk(KERN_NOTICE "apm: overridden by ACPI.\n");
>                  apm_info.disabled = 1;
>                  return -ENODEV;
> -         }
> +#endif

That's not a good choice - it needs to be a runtime check, not a 
compile-time one. Don't we have an acpi_enabled() function?

-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: T30 boot hang with CONFIG_PM_LEGACY=n

[Dave Jones]

On Wed, Sep 27, 2006 at 03:47:31PM +0100, Matthew Garrett wrote:
 > On Wed, Sep 27, 2006 at 12:20:29AM -0400, Dave Jones wrote:
 > 
 > > Good question.  There's no real replacement for pm_active in the non-legacy
 > > config afaik. Perhaps the cleanest alternative is to undeprecate that macro?
 > > The only other option I can think of is doing something like this in apm.c ...
 > > 
 > > +#ifdef CONFIG_ACPI
 > > -        if (PM_IS_ACTIVE()) {
 > >                  printk(KERN_NOTICE "apm: overridden by ACPI.\n");
 > >                  apm_info.disabled = 1;
 > >                  return -ENODEV;
 > > -         }
 > > +#endif
 > 
 > That's not a good choice - it needs to be a runtime check, not a 
 > compile-time one. Don't we have an acpi_enabled() function?
 
You're right, the above would break if booted with acpi=off.

Hmm.

	Dave