[PATCH] [20/82] i386: Panic the system when a NUMA kernel doesn't run on IBM NUMA

[PATCH] [20/82] i386: Panic the system when a NUMA kernel doesn't run on IBM NUMA

[Andi Kleen]

It has been broken forever anywhere else and is not too useful
anyways so best to disable it.

Signed-off-by: Andi Kleen <ak@suse.de>

---
 arch/i386/Kconfig       |    3 +++
 arch/i386/kernel/srat.c |    8 ++++++++
 2 files changed, 11 insertions(+)

Index: linux/arch/i386/kernel/srat.c
===================================================================
--- linux.orig/arch/i386/kernel/srat.c
+++ linux/arch/i386/kernel/srat.c
@@ -312,6 +312,14 @@ int __init get_memcfg_from_srat(void)
 	int tables = 0;
 	int i = 0;
 
+	extern int use_cyclone;
+	if (use_cyclone == 0) {
+		/* Make sure user sees something */
+		static const char s[] __initdata = "Not an IBM x440/NUMAQ. Don't use i386 CONFIG_NUMA anywhere else.";
+		early_printk(s);
+		panic(s);
+	}
+
 	if (ACPI_FAILURE(acpi_find_root_pointer(ACPI_PHYSICAL_ADDRESSING,
 						rsdp_address))) {
 		printk("%s: System description tables not found\n",
Index: linux/arch/i386/Kconfig
===================================================================
--- linux.orig/arch/i386/Kconfig
+++ linux/arch/i386/Kconfig
@@ -523,6 +523,9 @@ config NUMA
 	depends on SMP && HIGHMEM64G && (X86_NUMAQ || X86_GENERICARCH || (X86_SUMMIT && ACPI))
 	default n if X86_PC
 	default y if (X86_NUMAQ || X86_SUMMIT)
+	help
+		NUMA support. Note this only works on IBM x440 or IBM NUMAQ.
+		Don't try to use it anywhere else.
 
 comment "NUMA (Summit) requires SMP, 64GB highmem support, ACPI"
 	depends on X86_SUMMIT && (!HIGHMEM64G || !ACPI)

Re: [PATCH] [20/82] i386: Panic the system when a NUMA kernel doesn't run on IBM NUMA

[Dave Jones]

On Sat, Jun 24, 2006 at 02:19:28AM +0200, Andi Kleen wrote:


 > +	extern int use_cyclone;
 > +	if (use_cyclone == 0) {
 > +		/* Make sure user sees something */
 > +		static const char s[] __initdata = "Not an IBM x440/NUMAQ. Don't use i386 CONFIG_NUMA anywhere else.";
 > +		early_printk(s);
 > +		panic(s);
 > +	}

non-IBM Machines do still boot with that enabled though don't they?
Turning previously working configurations into panic()'s seems a bit odd,
even with a warning.

		Dave

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

Re: [PATCH] [20/82] i386: Panic the system when a NUMA kernel doesn't run on IBM NUMA

[Andi Kleen]

On Saturday 24 June 2006 02:38, Dave Jones wrote:
> On Sat, Jun 24, 2006 at 02:19:28AM +0200, Andi Kleen wrote:
> 
> 
>  > +	extern int use_cyclone;
>  > +	if (use_cyclone == 0) {
>  > +		/* Make sure user sees something */
>  > +		static const char s[] __initdata = "Not an IBM x440/NUMAQ. Don't use i386 CONFIG_NUMA anywhere else.";
>  > +		early_printk(s);
>  > +		panic(s);
>  > +	}
> 
> non-IBM Machines do still boot with that enabled though don't they?


No they don't - as they likely didn't do before. e.g. Opterons generally
break and that brings the point across clearer.

The rationale is that CONFIG_NUMA is very rarely used on i386 (even on summit)
and always does bitrot quickly. It also doesn't work at all on a wide
range of machines.

I'm sure someone will bring up now an example where their non Summit 
machine booted with CONFIG_NUMA, but they were just extremly lucky
and unlikely to be for very long.

-Andi

Re: [PATCH] [20/82] i386: Panic the system when a NUMA kernel doesn't run on IBM NUMA

[Keith Mannthey]

On 6/23/06, Andi Kleen <ak@suse.de> wrote:
> On Saturday 24 June 2006 02:38, Dave Jones wrote:
> > On Sat, Jun 24, 2006 at 02:19:28AM +0200, Andi Kleen wrote:
> >
> >
> >  > +  extern int use_cyclone;
> >  > +  if (use_cyclone == 0) {
> >  > +          /* Make sure user sees something */
> >  > +          static const char s[] __initdata = "Not an IBM x440/NUMAQ. Don't use i386 CONFIG_NUMA anywhere else.";
> >  > +          early_printk(s);
> >  > +          panic(s);
> >  > +  }
> >
> > non-IBM Machines do still boot with that enabled though don't they?
>
>
> No they don't - as they likely didn't do before. e.g. Opterons generally
> break and that brings the point across clearer.
>
> The rationale is that CONFIG_NUMA is very rarely used on i386 (even on summit)
> and always does bitrot quickly. It also doesn't work at all on a wide
> range of machines.

I agree that i386 CONFIG_NUMA is only ment to boot on small subset of
hw but there is litte motivation to boot a numa kernel on a non-numa
box.  I am supprised that no one has enabled i386 AMD NUMA (the one
numa box that regular people have access to).

> I'm sure someone will bring up now an example where their non Summit
> machine booted with CONFIG_NUMA, but they were just extremly lucky
> and unlikely to be for very long.

Current Summit HW  (x460) dosen't use/have a cyclone.  There are
patches submitted to this list to support it's i386 NUMA boot.

Thanks,
  Keith

Re: [PATCH] [20/82] i386: Panic the system when a NUMA kernel doesn't run on IBM NUMA

[Andi Kleen]

>
> I agree that i386 CONFIG_NUMA is only ment to boot on small subset of
> hw but there is litte motivation to boot a numa kernel on a non-numa
> box.  I am supprised that no one has enabled i386 AMD NUMA (the one
> numa box that regular people have access to).

I never bothered because 32bit NUMA with lowmem only on node 0 
isn't very useful. Using it with 64bit makes much more sense.

> > I'm sure someone will bring up now an example where their non Summit
> > machine booted with CONFIG_NUMA, but they were just extremly lucky
> > and unlikely to be for very long.
>
> Current Summit HW  (x460) dosen't use/have a cyclone.  There are
> patches submitted to this list to support it's i386 NUMA boot.

Hmm, that's a good point. When the detection doesn't work
reliable it's no good.

Ok, Linus please don't apply that patch for now then.

-Andi