BAD APICID ?

BAD APICID ?

[Dave Jones]

We got a report from a Fedora user hitting a WARN_ON in the apic setup code.

Here..

1230 #ifdef CONFIG_X86_32
1231         /*
1232          * APIC LDR is initialized.  If logical_apicid mapping was
1233          * initialized during get_smp_config(), make sure it matches the
1234          * actual value.
1235          */
1236         i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1237         WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1238         /* always use the value from LDR */
1239         early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1240                 logical_smp_processor_id();

The full report is at https://bugzilla.redhat.com/show_bug.cgi?id=743930

Report is against our 3.0 builds, but that code look unchanged in 3.1 too,
and even in current Linus head.

Any thoughts on what could cause this ?

	Dave

Re: BAD APICID ?

[Borislav Petkov]

On Mon, Nov 07, 2011 at 12:39:48PM -0500, Dave Jones wrote:
> We got a report from a Fedora user hitting a WARN_ON in the apic setup code.
> 
> Here..
> 
> 1230 #ifdef CONFIG_X86_32
> 1231         /*
> 1232          * APIC LDR is initialized.  If logical_apicid mapping was
> 1233          * initialized during get_smp_config(), make sure it matches the
> 1234          * actual value.
> 1235          */
> 1236         i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
> 1237         WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
> 1238         /* always use the value from LDR */
> 1239         early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
> 1240                 logical_smp_processor_id();
> 
> The full report is at https://bugzilla.redhat.com/show_bug.cgi?id=743930
> 
> Report is against our 3.0 builds, but that code look unchanged in 3.1 too,
> and even in current Linus head.
> 
> Any thoughts on what could cause this ?

Already fixed upstream: 838312be46f3abfbdc175f81c3e54a857994476d

Here's the whole thread: http://lkml.org/lkml/2011/10/14/231

HTH.

-- 
Regards/Gruss,
Boris.

Advanced Micro Devices GmbH
Einsteinring 24, 85609 Dornach
GM: Alberto Bozzo
Reg: Dornach, Landkreis Muenchen
HRB Nr. 43632 WEEE Registernr: 129 19551

Re: BAD APICID ?

[Dave Jones]

On Mon, Nov 07, 2011 at 06:54:02PM +0100, Borislav Petkov wrote:
 > On Mon, Nov 07, 2011 at 12:39:48PM -0500, Dave Jones wrote:
 > > We got a report from a Fedora user hitting a WARN_ON in the apic setup code.
 > > 
 > > Here..
 > > 
 > > 1230 #ifdef CONFIG_X86_32
 > > 1231         /*
 > > 1232          * APIC LDR is initialized.  If logical_apicid mapping was
 > > 1233          * initialized during get_smp_config(), make sure it matches the
 > > 1234          * actual value.
 > > 1235          */
 > > 1236         i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
 > > 1237         WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
 > > 1238         /* always use the value from LDR */
 > > 1239         early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
 > > 1240                 logical_smp_processor_id();
 > > 
 > > The full report is at https://bugzilla.redhat.com/show_bug.cgi?id=743930
 > > 
 > > Report is against our 3.0 builds, but that code look unchanged in 3.1 too,
 > > and even in current Linus head.
 > > 
 > > Any thoughts on what could cause this ?
 > 
 > Already fixed upstream: 838312be46f3abfbdc175f81c3e54a857994476d
 > 
 > Here's the whole thread: http://lkml.org/lkml/2011/10/14/231
 > 
 > HTH.

ah, excellent, thanks.
That seems to have not hit stable, even though it was cc'd. Greg ?

	Dave

Re: BAD APICID ?

[Greg KH]

On Mon, Nov 07, 2011 at 01:09:50PM -0500, Dave Jones wrote:
> On Mon, Nov 07, 2011 at 06:54:02PM +0100, Borislav Petkov wrote:
>  > On Mon, Nov 07, 2011 at 12:39:48PM -0500, Dave Jones wrote:
>  > > We got a report from a Fedora user hitting a WARN_ON in the apic setup code.
>  > > 
>  > > Here..
>  > > 
>  > > 1230 #ifdef CONFIG_X86_32
>  > > 1231         /*
>  > > 1232          * APIC LDR is initialized.  If logical_apicid mapping was
>  > > 1233          * initialized during get_smp_config(), make sure it matches the
>  > > 1234          * actual value.
>  > > 1235          */
>  > > 1236         i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
>  > > 1237         WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
>  > > 1238         /* always use the value from LDR */
>  > > 1239         early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
>  > > 1240                 logical_smp_processor_id();
>  > > 
>  > > The full report is at https://bugzilla.redhat.com/show_bug.cgi?id=743930
>  > > 
>  > > Report is against our 3.0 builds, but that code look unchanged in 3.1 too,
>  > > and even in current Linus head.
>  > > 
>  > > Any thoughts on what could cause this ?
>  > 
>  > Already fixed upstream: 838312be46f3abfbdc175f81c3e54a857994476d
>  > 
>  > Here's the whole thread: http://lkml.org/lkml/2011/10/14/231
>  > 
>  > HTH.
> 
> ah, excellent, thanks.
> That seems to have not hit stable, even though it was cc'd. Greg ?

It's already been queued up for the next stable release, which should be
coming out later this week once the firehose of the 3.2 merge window
slows down.

thanks,

greg k-h