Re: [PATCH] x86_64: Make the NUMA hash function nodemap allocation dynamic and remove NODEMAPSIZE

[Eric Dumazet]

On Sunday 26 November 2006 21:49, Andi Kleen wrote:
> On Wednesday 15 November 2006 22:48, Amul Shah wrote:
> > This patch removes the statically allocated memory to NUMA node hash map
> > in favor of a dynamically allocated memory to node hash map (it is cache
> > aligned).
> >
> > This patch has the nice side effect in that it allows the hash map to
> > grow for systems with large amounts of memory (256GB - 1TB), but suffer
> > from having small PCI space tacked onto the boot node (which is
> > somewhere between 192MB to 512MB on the ES7000).
> >
> > Signed-off-by: Amul Shah <amul.shah@xxxxxxxxxx>
> >
> > ---
> > Patch applies to 2.6.19-rc4 and has been tested.
> > This patch needs testing on a K8 NUMA platform.
> > Thanks to Eric Dumazet and Andi Kleen for their improvement suggestions.
>
> I had the patch in, but had to drop it again because it makes one of my
> test system triple fault. Haven't done much investigation yet.
>
> BIOS-provided physical RAM map:
>  BIOS-e820: 0000000000000000 - 000000000009fc00 (usable)
>  BIOS-e820: 000000000009fc00 - 00000000000a0000 (reserved)
>  BIOS-e820: 00000000000e6000 - 0000000000100000 (reserved)
>  BIOS-e820: 0000000000100000 - 000000003ef30000 (usable)
>  BIOS-e820: 000000003ef30000 - 000000003ef40000 (ACPI data)
>  BIOS-e820: 000000003ef40000 - 000000003eff0000 (ACPI NVS)
>  BIOS-e820: 000000003eff0000 - 000000003f000000 (reserved)
>  BIOS-e820: 00000000fecf0000 - 00000000fecf1000 (reserved)
>  BIOS-e820: 00000000fed20000 - 00000000feda0000 (reserved)
> end_pfn_map = 1043872
> kernel direct mapping tables up to feda0000 @ 8000-d000
> DMI 2.3 present.
> No NUMA configuration found
> Faking a node at 0000000000000000-000000003ef30000
> <triple fault>
>

Well, I dont have currently an AMD64 test machine so I cannot really help.

With previous implementation, the nimimum shift value was 20 (one megabytes)

If a memnode had a finer range (with chunks not multiple of megabytes), some 
bits of memory could be ignored.

But with your fake node (0-3ef30000), Amul patch may give a shift value of 16.
 Maybe this breaks something in the kernel...

Eric
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/