This is the mail archive of the
mailing list for the glibc project.
Re: [PATCH] Use uname not sysctl to get the kernel revision
- From: Andi Kleen <ak at suse dot de>
- To: Benjamin Herrenschmidt <benh at kernel dot crashing dot org>
- Cc: Steve Munroe <sjmunroe at us dot ibm dot com>, Theodore Tso <tytso at mit dot edu>, libc-alpha at sourceware dot org, linux-kernel at vger dot kernel dot org, akpm at osdl dot org, Arjan van de Ven <arjan at infradead dot org>, Ulrich Drepper <drepper at redhat dot com>, "Eric W. Biederman" <ebiederm at xmission dot com>, "Randy.Dunlap" <rdunlap at xenotime dot net>
- Date: Fri, 14 Jul 2006 21:09:34 +0200
- Subject: Re: [PATCH] Use uname not sysctl to get the kernel revision
- References: <OF1343C031.500862D1-ON862571A9.00817A27-862571A9.00826BED@us.ibm.com> <email@example.com>
On Friday 14 July 2006 20:49, Benjamin Herrenschmidt wrote:
> > We will need an implementation that will fall back to sys_sysctl for older
> > kernels. This is already common practice in glibc. I don't really
> > understand the performance concern because it seems to me that
> > _is_smp_system() is only called once per process.
> > But isn't this the kind of thing that the Aux Vector is for? I like vDSO
> > too, but I think it is best deployed for information of a more dynamic
> > nature and performance sensitive.
> For a simple "is_smp" kind of flag, I would tend to agree with the
> above... for more complex NUMA topology and/or cache characteristics,
> which is quite a bigger amount of information, I'm not sure it's worth
> copying all of that data on every process exec (and making the initial
> AT_ parsing slower). Especially since very few processes actually care
> about those.
I've actually spent some thought on that recently. The motivation
came from someone who wanted the number of CPUs in a fast way
to tune AMD64 memcpy etc. better.
My proposal was to supply four new count:
number of cores, number of siblings, number of sockets, number of nodes
These all fit easily in 16bit so it would be 2 new entries in the
aux vector (128 bit total). Shouldn't be much overhead to write this.
If you need more exact topology you can probably eat the overhead
of parsing /proc/cpuinfo or read it from sysfs (or just use libnuma
which supplies most of this in an easy way)
Doing it in a vDSO would be in theory ok for me too, except that x86-64
doesn't have one so far. Even in vDSO I wouldn't add much more than this
(like bitmaps and similar) because otherwise cpu/node hotplug could be racy.
Also I'm reluctant to redo /proc/cpuinfo and /sys for this.