[PATCH 0 of 5] RFC: Handling of sparse, memoryless, and cpuless nodes

[PATCH 0 of 5] RFC: Handling of sparse, memoryless, and cpuless nodes

[Tim Pepper]

In response to a bug report where 'numactl --hardware' was segfaulting
with distro 2.0.3 version I've dug into the current state of
libnuma/numactl after not having looked at it in quite some time.  The
following series of patches resulted and hopefully clarifies things a bit
for the next person who comes along and accounts for the possibility of
sparsely numbered, memoryless and cpuless nodes in 'numactl --hardware'.
There are certainly some things in the patches that aren't quite optimal
and other things I saw that I didn't try to tackle (eg: there is a lot
of seeming inconsistency around the different "all" node pointers),
but the updated code works for me and I was trying to limit changes to
not start driving in the direction of a v3 library (or hwloc?).

The first 4 patches are relatively minor cleanups.

The bulk is in patch 5 which then deals with better handling sparse node
numbering and fixes parse_numbers() in particular which was experiencing
a buffer overflow in the face of this topology:

available: 2 nodes (0,3)
node 0 cpus: none
node 0 size: 0 MB
node 0 free: 0 MB
node 3 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
node 3 size: 15360 MB
node 3 free: 12524 MB
node distances:
node 0 3
 0: 10 40
 3: 40 10

I've tested the changes on "interesting" topology hardware to which I
have access, but would greatly appreciate if others might do similar.

-- 
Tim Pepper  <lnxninja@linux.vnet.ibm.com>
IBM Linux Technology Center

Re: [PATCH 0 of 5] RFC: Handling of sparse, memoryless, and cpuless nodes

[Cliff Wickman]

Thanks, Tim.
(Yes, I'm still here.  Just been on vacation since you sent your patches.)

I tested your patches with the usual test suite, on an 8-node ia64 and
on an 8-node x86_64 (uv).

Now available at ftp://oss.sgi.com/www/projects/libnuma/download/
as numactl-2.0.6-rc1.tar.gz

-Cliff

On Wed, Sep 22, 2010 at 11:03:47AM -0700, Tim Pepper wrote:
> In response to a bug report where 'numactl --hardware' was segfaulting
> with distro 2.0.3 version I've dug into the current state of
> libnuma/numactl after not having looked at it in quite some time.  The
> following series of patches resulted and hopefully clarifies things a bit
> for the next person who comes along and accounts for the possibility of
> sparsely numbered, memoryless and cpuless nodes in 'numactl --hardware'.
> There are certainly some things in the patches that aren't quite optimal
> and other things I saw that I didn't try to tackle (eg: there is a lot
> of seeming inconsistency around the different "all" node pointers),
> but the updated code works for me and I was trying to limit changes to
> not start driving in the direction of a v3 library (or hwloc?).
> 
> The first 4 patches are relatively minor cleanups.
> 
> The bulk is in patch 5 which then deals with better handling sparse node
> numbering and fixes parse_numbers() in particular which was experiencing
> a buffer overflow in the face of this topology:
> 
> available: 2 nodes (0,3)
> node 0 cpus: none
> node 0 size: 0 MB
> node 0 free: 0 MB
> node 3 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
> node 3 size: 15360 MB
> node 3 free: 12524 MB
> node distances:
> node 0 3
>  0: 10 40
>  3: 40 10
> 
> I've tested the changes on "interesting" topology hardware to which I
> have access, but would greatly appreciate if others might do similar.
> 
> -- 
> Tim Pepper  <lnxninja@linux.vnet.ibm.com>
> IBM Linux Technology Center
> --
> To unsubscribe from this list: send the line "unsubscribe linux-numa" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

-- 
Cliff Wickman
SGI
cpw@sgi.com
(651) 683-3824