mce_64.c: mce_cpu_quirks being ignored

mce_64.c: mce_cpu_quirks being ignored

[Max Asbock]

A recent change to mce_init in -mm does the following:

@@ -462,7 +463,7 @@ static void mce_init(void *dummy)
 		wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff);
 
 	for (i = 0; i < banks; i++) {
-		wrmsrl(MSR_IA32_MC0_CTL+4*i, bank[i]);
+		wrmsrl(MSR_IA32_MC0_CTL+4*i, ~0UL);
 		wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0);
 	}
....
 
This change appears to ignore the fact that mce_cpu_quirks() clears a
bit in bank[4] for certain CPUs, as in:

static void __cpuinit mce_cpu_quirks(struct cpuinfo_x86 *c)
{
        /* This should be disabled by the BIOS, but isn't always */
        if (c->x86_vendor == X86_VENDOR_AMD) {
                if(c->x86 == 15)
                        /* disable GART TBL walk error reporting, which trips off
                           incorrectly with the IOMMU & 3ware & Cerberus. */
                        clear_bit(10, &bank[4]);
....

Is turning off that bit still needed?

Max

Re: mce_64.c: mce_cpu_quirks being ignored

[Andi Kleen]

Max Asbock wrote:



> static void __cpuinit mce_cpu_quirks(struct cpuinfo_x86 *c)
> {
>         /* This should be disabled by the BIOS, but isn't always */
>         if (c->x86_vendor == X86_VENDOR_AMD) {
>                 if(c->x86 == 15)
>                         /* disable GART TBL walk error reporting, which trips off
>                            incorrectly with the IOMMU & 3ware & Cerberus. */
>                         clear_bit(10, &bank[4]);
> ....
> 
> Is turning off that bit still needed?

Yes it is. Also referencing the bank is needed for other reasons anyways,
otherwise it would ignore the user sysfs choice on reconfiguration.

Venki, I did a patch for dynamic banks anyways, but haven't submitted it yet.

-Andi

Re: mce_64.c: mce_cpu_quirks being ignored

[Venki Pallipadi]

On Thu, Jun 19, 2008 at 02:16:17PM -0700, Andi Kleen wrote:
> Max Asbock wrote:
> 
> 
> 
> > static void __cpuinit mce_cpu_quirks(struct cpuinfo_x86 *c)
> > {
> >         /* This should be disabled by the BIOS, but isn't always */
> >         if (c->x86_vendor == X86_VENDOR_AMD) {
> >                 if(c->x86 == 15)
> >                         /* disable GART TBL walk error reporting, which trips off
> >                            incorrectly with the IOMMU & 3ware & Cerberus. */
> >                         clear_bit(10, &bank[4]);
> > ....
> >
> > Is turning off that bit still needed?
> 
> Yes it is. Also referencing the bank is needed for other reasons anyways,
> otherwise it would ignore the user sysfs choice on reconfiguration.
> 
> Venki, I did a patch for dynamic banks anyways, but haven't submitted it yet.
> 

Yes. That particular quirk getting ignored was a bug. Below patch fixes
the bug, until we have the dynamic banks support.

sysfs choice configuration should not have any issues with the earlier patch
as we look for NR_SYSFS_BANKS in do_machine_check().

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

---
 arch/x86/kernel/cpu/mcheck/mce_64.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

Index: linux-2.6/arch/x86/kernel/cpu/mcheck/mce_64.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/cpu/mcheck/mce_64.c	2008-06-24 16:01:23.000000000 -0700
+++ linux-2.6/arch/x86/kernel/cpu/mcheck/mce_64.c	2008-06-24 16:48:01.000000000 -0700
@@ -463,7 +463,11 @@ static void mce_init(void *dummy)
 		wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff);
 
 	for (i = 0; i < banks; i++) {
-		wrmsrl(MSR_IA32_MC0_CTL+4*i, ~0UL);
+		if (i < NR_SYSFS_BANKS)
+			wrmsrl(MSR_IA32_MC0_CTL+4*i, bank[i]);
+		else
+			wrmsrl(MSR_IA32_MC0_CTL+4*i, ~0UL);
+
 		wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0);
 	}
 }

Re: mce_64.c: mce_cpu_quirks being ignored

[Ingo Molnar]

* Venki Pallipadi <venkatesh.pallipadi@intel.com> wrote:

> > Venki, I did a patch for dynamic banks anyways, but haven't 
> > submitted it yet.
> 
> Yes. That particular quirk getting ignored was a bug. Below patch 
> fixes the bug, until we have the dynamic banks support.
> 
> sysfs choice configuration should not have any issues with the earlier 
> patch as we look for NR_SYSFS_BANKS in do_machine_check().
> 
> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>

applied to tip/x86/mce - thanks Venki.

	Ingo

Re: mce_64.c: mce_cpu_quirks being ignored

[Andi Kleen]

Ingo Molnar wrote:
> * Venki Pallipadi <venkatesh.pallipadi@intel.com> wrote:
> 
>>> Venki, I did a patch for dynamic banks anyways, but haven't 
>>> submitted it yet.
>> Yes. That particular quirk getting ignored was a bug. Below patch 
>> fixes the bug, until we have the dynamic banks support.
>>
>> sysfs choice configuration should not have any issues with the earlier 
>> patch as we look for NR_SYSFS_BANKS in do_machine_check().
>>
>> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
> 
> applied to tip/x86/mce - thanks Venki.

The dynamic banks patch is posted
http://marc.info/?l=linux-kernel&m=121493114604815&w=2
, but it turned out it requires
a sysfs infrastructure change for the auto-generated attributes.
I did that, but the patch is rather large
(http://marc.info/?l=linux-kernel&m=121493094104469&w=2)

Greg said he would queue the change, so dynamic banks could
be added (replacing Venki's two patches), but it would need
to be submitted after the infrastructure change.

-Andi