[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [RFC PATCH] numa: enable sparse node numbering
From: |
Nishanth Aravamudan |
Subject: |
Re: [Qemu-devel] [RFC PATCH] numa: enable sparse node numbering |
Date: |
Tue, 24 Jun 2014 09:39:49 -0700 |
User-agent: |
Mutt/1.5.21 (2010-09-15) |
On 24.06.2014 [09:33:48 +0200], Igor Mammedov wrote:
> On Mon, 23 Jun 2014 12:33:10 -0700
> Nishanth Aravamudan <address@hidden> wrote:
>
> > Add a present field to NodeInfo which indicates if a given nodeid was
> > present on the command-line or not. Current code relies on a memory
> > value being passed for a node to indicate presence, which is
> > insufficient in the presence of memoryless nodes or sparse numbering.
> > Adjust the iteration of various NUMA loops to use the maximum known NUMA
> > ID rather than the number of NUMA nodes.
> Why would you need sparse numbering?
We already see this in practice under PowerVM on ppc. This would allow
qemu/KVM to emulate the same behaviors (which often lead to bugs/testing
of underused code-paths in the kernel).
Also, sparse numbering is explicitly allowed by qemu already but doesn't
work at all as expected, at least to me. That is, if I specify various
node IDs, the nodes are allocated linearly in qemu and the NUMA node IDs
in Linux don't correspond to the command-line. Further, the location of
CPUs/memory doesn't correspond to the nodes requested by the user.
> What task exactly are you trying to solve?
I would like to be able to leverage qemu's command-line to reproduce
the NUMA topology seen under PowerVM.
I'll send v3 with a better changelog and a few changes from Hu Tao.
Thanks,
Nish