Re: [PATCH 0/2] Relax restrictions on setting CONFIG_NUMA on x86

[Mel Gorman <mel@csn.ul.ie>]

On (22/01/08 14:33), Andi Kleen didst pronounce:
> 
> > Without SRAT support, a compile-error occurs because ACPI table parsing
> > functions are only available in x86-64. This patch also adds no-op stubs
> > and prints a warning message. What likely needs to be done is sharing
> > the table parsing functions between 32 and 64 bit if they are
> > compatible.
> 
> I'm a little confused by your patch.
> 
> i386 already has srat parsing code (just written in a horrible hackish way); 
> but it exists arch/x86/kernel/srat_32.c
> 

Yes, I spotted that. Enabling it required a Kconfig change or two and
enabling BOOT_IOREMAP. It then crashes early in boot on a call to strlen()
so I went with the stubs and SRAT disabled for the moment.

> That one tended to explode on Opteron, but apparently worked on some 
> Summit boxes.
> 
> You're saying you want to remove that code and replace it based on something
> based on the drivers/acpi/numa.c parsing code?

I made the assumption that they were basically the same. That is obviously
wrong from what you say below.

> While that's in theory
> a worthy goal it will not actually help all that much because numa.c only
> does some high level parsing, but nothing of the actual low level work
> of setting things up.
> 

Ok, understood. When I next revisit this, I'll look at making ACPI_SRAT
and BOOT_IOREMAP work on normal machines and see what happens. Thanks.

-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab