Skip to contentSkip to conversation

CONFIG_K8_NUMA x86_64 no-memory node bug

DISCUSSION John Blackwood 2 participants 3 messages Feb 1, 2006 – Feb 7, 2006

lore
Share & tools
lore
#0 lore raw

I would like to mention a bug with the x86_64 CONFIG_K8_NUMA support.

We have a 4 CPU AMD Opteron (processor 846 -- no dual core) system that boots up with a 2.6.15.2-based kernel with NUMA enabled if all the numa nodes are populated with memory modules.

If we then pull out the memory module for the 3rd CPU/node, then the kernel will no longer boot.

In this configuration, after the grub 'boot' command is entered, no output is seen, and the system appears to be hung.

While this is admittedly a 'degraded' configuration, it would be nice if the kernel could handle having a middle numbered node without memory.

I believe that removing the 4th CPU's memory module so that only the last CPU/node is without memory, then the system boots up fine.

It seems that having a middle node without memory is what causes this problem to occur.

Thanks for reading this.

#1 lore raw
Quoted 12 lines · John Blackwood

I would like to mention a bug with the x86_64 CONFIG_K8_NUMA support.

We have a 4 CPU AMD Opteron (processor 846 -- no dual core) system that boots up with a 2.6.15.2-based kernel with NUMA enabled if all the numa nodes are populated with memory modules.

If we then pull out the memory module for the 3rd CPU/node, then the kernel will no longer boot.

In this configuration, after the grub 'boot' command is entered, no output is seen, and the system appears to be hung.

Quoted 8 lines

While this is admittedly a 'degraded' configuration, it would be nice if the kernel could handle having a middle numbered node without memory.

I believe that removing the 4th CPU's memory module so that only the last CPU/node is without memory, then the system boots up fine.

It seems that having a middle node without memory is what causes this problem to occur.

Does it boot with numa=noacpi?

If yes then I likely already fixed the bug. But send boot log of the failure anyways.

-Andi

#2 lore raw
In reply to John Blackwood

John,

Can you please double check if the problem is still there in the latest 2.6.16-rc2-git* kernel? I fixed a couple of SRAT issues in there and at least one change could have fixed your problem too.

Thanks,
-Andi