SMP broken on pre-ACPI machine.

SMP broken on pre-ACPI machine.

[Dave Jones]

I've been chasing a bug that got filed against the Fedora kernel
a while back:  https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=199052
This is a dual pentium pro from an era before we had ACPI, and
it seems to be falling foul of this test in smpboot.c  ..

    if (!smp_found_config && !acpi_lapic) {
        printk(KERN_NOTICE "SMP motherboard not detected.\n");
        smpboot_clear_io_apic_irqs();
        phys_cpu_present_map = physid_mask_of_physid(0);
        if (APIC_init_uniprocessor())
            printk(KERN_NOTICE "Local APIC not detected."
                       " Using dummy APIC emulation.\n");
        map_cpu_to_logical_apicid();
        cpu_set(0, cpu_sibling_map[0]);
        cpu_set(0, cpu_core_map[0]);
        return;
    }


My initial reaction is that the !acpi_lapic test should be conditional
on some variable that gets set if the ACPI parsing actually succeeded.

Thoughts?

	Dave

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

Re: SMP broken on pre-ACPI machine.

[Sergio Monteiro Basto]

[-- Attachment #1: Type: text/plain, Size: 1243 bytes --]

On Wed, 2006-10-18 at 18:24 -0400, Dave Jones wrote:
> I've been chasing a bug that got filed against the Fedora kernel
> a while back:  https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=199052
> This is a dual pentium pro from an era before we had ACPI, and
> it seems to be falling foul of this test in smpboot.c  ..
> 
>     if (!smp_found_config && !acpi_lapic) {
>         printk(KERN_NOTICE "SMP motherboard not detected.\n");
>         smpboot_clear_io_apic_irqs();
>         phys_cpu_present_map = physid_mask_of_physid(0);
>         if (APIC_init_uniprocessor())
>             printk(KERN_NOTICE "Local APIC not detected."
>                        " Using dummy APIC emulation.\n");
>         map_cpu_to_logical_apicid();
>         cpu_set(0, cpu_sibling_map[0]);
>         cpu_set(0, cpu_core_map[0]);
>         return;
>     }
> 
> 
> My initial reaction is that the !acpi_lapic test should be conditional
> on some variable that gets set if the ACPI parsing actually succeeded.
> 
> Thoughts?
> 
> 	Dave
> 

acpi=off ? this machine should work with APM. BTW, so time ago, this
machine would enter in ACPI blacklist (by the year of bios) and ACPI was
turned off automatically. 

-- 
Sérgio M.B.

[-- Attachment #2: smime.p7s --]
[-- Type: application/x-pkcs7-signature, Size: 2166 bytes --]

Re: SMP broken on pre-ACPI machine.

[Len Brown]

On Wednesday 18 October 2006 18:24, Dave Jones wrote:
> I've been chasing a bug that got filed against the Fedora kernel
> a while back:  https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=199052
> This is a dual pentium pro from an era before we had ACPI, and
> it seems to be falling foul of this test in smpboot.c  ..
> 
>     if (!smp_found_config && !acpi_lapic) {
>         printk(KERN_NOTICE "SMP motherboard not detected.\n");
>         smpboot_clear_io_apic_irqs();
>         phys_cpu_present_map = physid_mask_of_physid(0);
>         if (APIC_init_uniprocessor())
>             printk(KERN_NOTICE "Local APIC not detected."
>                        " Using dummy APIC emulation.\n");
>         map_cpu_to_logical_apicid();
>         cpu_set(0, cpu_sibling_map[0]);
>         cpu_set(0, cpu_core_map[0]);
>         return;
>     }
> 
> 
> My initial reaction is that the !acpi_lapic test should be conditional
> on some variable that gets set if the ACPI parsing actually succeeded.

acpi_lapic isn't related to the problem at hand -- that  smp_found_config is not set.

That said, allowing acpi_lapic=1 to bail out of this check has the sole
function of allowing SMP/PIC configurations.  (smp_found_config
in ACPI mode is set if acpi_lapic and acpi_ioapic are set)
SMP/PIC configurations are not very interesting, except for debugging.
Indeed, MPS prohibits them by mandating an IOAPIC be present for SMP --
but ACPI has no such rule.

-Len

Re: SMP broken on pre-ACPI machine.

[Dave Jones]

On Thu, Oct 19, 2006 at 01:33:40AM -0400, Len Brown wrote:
 > On Wednesday 18 October 2006 18:24, Dave Jones wrote:
 > > I've been chasing a bug that got filed against the Fedora kernel
 > > a while back:  https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=199052
 > > This is a dual pentium pro from an era before we had ACPI, and
 > > it seems to be falling foul of this test in smpboot.c  ..
 > > 
 > >     if (!smp_found_config && !acpi_lapic) {
 > >         printk(KERN_NOTICE "SMP motherboard not detected.\n");
 > > 
 > > My initial reaction is that the !acpi_lapic test should be conditional
 > > on some variable that gets set if the ACPI parsing actually succeeded.
 > 
 > acpi_lapic isn't related to the problem at hand -- that  smp_found_config is not set.

Right, it just seemed odd to me when I was eyeballing this code.

 > That said, allowing acpi_lapic=1 to bail out of this check has the sole
 > function of allowing SMP/PIC configurations.  (smp_found_config
 > in ACPI mode is set if acpi_lapic and acpi_ioapic are set)
 > SMP/PIC configurations are not very interesting, except for debugging.
 > Indeed, MPS prohibits them by mandating an IOAPIC be present for SMP --
 > but ACPI has no such rule.

Why smp_found_config isn't set in that guys configuration is a mystery to me,
as his MPS tables look sane..

MP Table:
#	APIC ID	Version	State		Family	Model	Step	Flags
#	 0	 0x10	 BSP, usable	 6	 2	 1	 0x0381
#	 0	 0x10	 AP, usable	 6	 1	 7	 0xfbff

Hmm, wait, he has unpaired CPUs. I wonder if that's the reason.
I know *some* combinations of PPro's are valid to be paired, but I'll
need to dig out the old docs to be sure.

	Dave

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

Re: SMP broken on pre-ACPI machine.

[Dave Jones]

On Thu, Oct 19, 2006 at 03:16:44PM -0400, Dave Jones wrote:

 > Why smp_found_config isn't set in that guys configuration is a mystery to me,
 > as his MPS tables look sane..
 > 
 > MP Table:
 > #	APIC ID	Version	State		Family	Model	Step	Flags
 > #	 0	 0x10	 BSP, usable	 6	 2	 1	 0x0381
 > #	 0	 0x10	 AP, usable	 6	 1	 7	 0xfbff
 > 
 > Hmm, wait, he has unpaired CPUs. I wonder if that's the reason.
 > I know *some* combinations of PPro's are valid to be paired, but I'll
 > need to dig out the old docs to be sure.
 
Ok, after reading http://www.intel.com/design/archives/processors/pro/docs/242689.htm
I'm more puzzled than ever.  There never was a model 2 Pentium Pro.
Either this BIOS is on crack and putting nonsense in its MPS tables,
or this is a hardware flaw of some sort.

	Dave

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

Re: SMP broken on pre-ACPI machine.

[Alan Cox]

Ar Iau, 2006-10-19 am 16:11 -0400, ysgrifennodd Dave Jones:
> On Thu, Oct 19, 2006 at 03:16:44PM -0400, Dave Jones wrote:
> 
>  > Why smp_found_config isn't set in that guys configuration is a mystery to me,
>  > as his MPS tables look sane..
>  > 
>  > MP Table:
>  > #	APIC ID	Version	State		Family	Model	Step	Flags
>  > #	 0	 0x10	 BSP, usable	 6	 2	 1	 0x0381

Isn't that an "overdrive" ? if so it isn't supposed to be SMP capable

Re: SMP broken on pre-ACPI machine.

[Dave Jones]

On Thu, Oct 19, 2006 at 11:17:19PM +0100, Alan Cox wrote:
 > Ar Iau, 2006-10-19 am 16:11 -0400, ysgrifennodd Dave Jones:
 > > On Thu, Oct 19, 2006 at 03:16:44PM -0400, Dave Jones wrote:
 > > 
 > >  > Why smp_found_config isn't set in that guys configuration is a mystery to me,
 > >  > as his MPS tables look sane..
 > >  > 
 > >  > MP Table:
 > >  > #	APIC ID	Version	State		Family	Model	Step	Flags
 > >  > #	 0	 0x10	 BSP, usable	 6	 2	 1	 0x0381
 > 
 > Isn't that an "overdrive" ? if so it isn't supposed to be SMP capable

I don't think so. The only overdrive that fitted in a socket 8 board
was a PPro->PentiumII thing, which would be model 3 stepping 2 as far
as I can figure out from a lengthy archeology trip through developer.intel.com.

There were also some Socket8->Socket370 convertors, but afaik they just passed
through the family/model/stepping of whatever was plugged into them.

	Dave

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