acpi-cpufreq oddness

acpi-cpufreq oddness

[Dave Jones]

Someone reported to me that modprobe acpi-cpufreq now fails noisily
complaining about invalid arguments.  I was able to reproduce it
on one of my boxes that doesn't have any form of speed scaling at all.
Previously, the modprobe would return with -ENODEV, but now its
getting an -EINVAL from somewhere. (I'm still trying to chase down
exactly where its coming from).

So far in chasing this, all I found is a cosmetic bug,
acpi_cpufreq_target has a..
dprintk("acpi_cpufreq_setpolicy");
in it, which is the only reference to setpolicy in the whole driver.

I don't have handy access to a git tree at the moment, and I've forgotten
the history of the recent changes here.

Alexey/Len, any ideas whats going on here?

		Dave

-- 
http://www.codemonkey.org.uk

RE: acpi-cpufreq oddness

[Pallipadi, Venkatesh]

>-----Original Message-----
>From: cpufreq-bounces@lists.linux.org.uk 
>[mailto:cpufreq-bounces@lists.linux.org.uk] On Behalf Of Dave Jones
>Sent: Friday, August 18, 2006 11:44 AM
>To: Alexey Starikovskiy
>Cc: Brown, Len; linux-acpi@vger.kernel.org; cpufreq@lists.linux.org.uk
>Subject: acpi-cpufreq oddness
>
>Someone reported to me that modprobe acpi-cpufreq now fails noisily
>complaining about invalid arguments.  I was able to reproduce it
>on one of my boxes that doesn't have any form of speed scaling at all.
>Previously, the modprobe would return with -ENODEV, but now its
>getting an -EINVAL from somewhere. (I'm still trying to chase down
>exactly where its coming from).
>
>So far in chasing this, all I found is a cosmetic bug,
>acpi_cpufreq_target has a..
>dprintk("acpi_cpufreq_setpolicy");
>in it, which is the only reference to setpolicy in the whole driver.
>
>I don't have handy access to a git tree at the moment, and 
>I've forgotten
>the history of the recent changes here.
>
>Alexey/Len, any ideas whats going on here?
>
>		Dave
>

Does the error say "Invalid _PSD data" or something similar. Seems to be
a side-effect of earlier software co-ordination patchset. If this is the
error you are seeing, I will send in a quick patch to fix it.

Thanks,
Venki 

Re: acpi-cpufreq oddness

[Dave Jones]

On Fri, Aug 18, 2006 at 11:59:26AM -0700, Pallipadi, Venkatesh wrote:

 > >I don't have handy access to a git tree at the moment, and 
 > >I've forgotten
 > >the history of the recent changes here.
 > >
 > >Alexey/Len, any ideas whats going on here?
 > >
 > >		Dave
 > >
 > 
 > Does the error say "Invalid _PSD data" or something similar. Seems to be
 > a side-effect of earlier software co-ordination patchset. If this is the
 > error you are seeing, I will send in a quick patch to fix it.

No, there's nothing in dmesg.
This is from the perror() in modprobe printing out the human-readable form
of -EINVAL

An strace shows ..

init_module(0x8287a88, 16244, "")       = -1 EINVAL (Invalid argument)
write(2, "FATAL: Error inserting acpi_cpuf"..., 139FATAL: Error inserting acpi_cpufreq (/lib/modules/2.6.17-1.2571.fc6/kernel/arch/i386/kernel/cpu/cpufreq/acpi-cpufreq.ko): Invalid argument


		Dave


-- 
http://www.codemonkey.org.uk

Re: acpi-cpufreq oddness

[Venkatesh Pallipadi]

On Fri, Aug 18, 2006 at 03:04:54PM -0400, Dave Jones wrote:
> On Fri, Aug 18, 2006 at 11:59:26AM -0700, Pallipadi, Venkatesh wrote:
> 
>  > >I don't have handy access to a git tree at the moment, and 
>  > >I've forgotten
>  > >the history of the recent changes here.
>  > >
>  > >Alexey/Len, any ideas whats going on here?
>  > >
>  > >		Dave
>  > >
>  > 
>  > Does the error say "Invalid _PSD data" or something similar. Seems to be
>  > a side-effect of earlier software co-ordination patchset. If this is the
>  > error you are seeing, I will send in a quick patch to fix it.
> 
> No, there's nothing in dmesg.
> This is from the perror() in modprobe printing out the human-readable form
> of -EINVAL
> 
> An strace shows ..
> 
> init_module(0x8287a88, 16244, "")       = -1 EINVAL (Invalid argument)
> write(2, "FATAL: Error inserting acpi_cpuf"..., 139FATAL: Error inserting acpi_cpufreq (/lib/modules/2.6.17-1.2571.fc6/kernel/arch/i386/kernel/cpu/cpufreq/acpi-cpufreq.ko): Invalid argument
> 
> 
> 		Dave
> 
> 
> -- 
> http://www.codemonkey.org.uk

EINVAL is actually coming from early_init call and the return value of that
call should be ignored. Attached patch resolves the issue.

Thanks,
Venki


Ignore the return value of early_init_acpi(), as it can give false error
messages. If there is something really wrong, then register_driver will fail
cleanly with EINVAL later.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>

Index: linux-2.6.18-rc4/arch/i386/kernel/cpu/cpufreq/acpi-cpufreq.c
===================================================================
--- linux-2.6.18-rc4.orig/arch/i386/kernel/cpu/cpufreq/acpi-cpufreq.c
+++ linux-2.6.18-rc4/arch/i386/kernel/cpu/cpufreq/acpi-cpufreq.c
@@ -567,16 +567,11 @@ static struct cpufreq_driver acpi_cpufre
 static int __init
 acpi_cpufreq_init (void)
 {
-	int                     result = 0;
-
 	dprintk("acpi_cpufreq_init\n");
 
-	result = acpi_cpufreq_early_init_acpi();
+	acpi_cpufreq_early_init_acpi();
 
-	if (!result)
- 		result = cpufreq_register_driver(&acpi_cpufreq_driver);
-	
-	return (result);
+ 	return cpufreq_register_driver(&acpi_cpufreq_driver);
 }
 
 

Re: acpi-cpufreq oddness

[Dave Jones]

On Fri, Aug 18, 2006 at 03:04:54PM -0400, Dave Jones wrote:

 > No, there's nothing in dmesg.
 > This is from the perror() in modprobe printing out the human-readable form
 > of -EINVAL
 > 
 > An strace shows ..
 > 
 > init_module(0x8287a88, 16244, "")       = -1 EINVAL (Invalid argument)
 > write(2, "FATAL: Error inserting acpi_cpuf"..., 139FATAL: Error inserting acpi_cpufreq (/lib/modules/2.6.17-1.2571.fc6/kernel/arch/i386/kernel/cpu/cpufreq/acpi-cpufreq.ko): Invalid argument

Ok, after a bit more investigation, the -EINVAL seems to be coming
from acpi_processor_preregister_performance()
Should ..
                if (!performance || !performance[i]) {
                        retval = -EINVAL;
                        continue;
                }

be -ENODEV perhaps ?

		Dave

-- 
http://www.codemonkey.org.uk

Re: acpi-cpufreq oddness

[Dave Jones]

On Fri, Aug 18, 2006 at 01:45:17PM -0700, Venkatesh Pallipadi wrote:

 > EINVAL is actually coming from early_init call and the return value of that
 > call should be ignored. Attached patch resolves the issue.
 
Ok, that'll do the trick :)

The -EINVAL actually came from this bit..

                pr->performance = performance[i];
                cpu_set(i, pr->performance->shared_cpu_map);
                if (acpi_processor_get_psd(pr)) {
                        retval = -EINVAL;
                        continue;
                }

I'll push this on tonight..

		Dave
-- 
http://www.codemonkey.org.uk