[lug] request for data: cpuinfo on non-x86 cpus

[Hugh Brown]

On Wed, Jul 23, 2008 at 10:15:18AM -0600, Michael J. Hammel wrote:
> On Thu, 2008-07-17 at 17:01 -0600, Zan Lynx wrote:
> > Michael J. Hammel wrote:
> > > You'd think there be some kind of standard, at least within the linux
> > > kernel, for the output from this.  But apparently not.  *Ugh*
> > There are standard ways to get this information, believe it or not.  You
> > might want to use "uname -m".
> > 
> > >From the OSX man page:
> > "The uname utility conforms to IEEE Std 1003.2-1992 (``POSIX.2'')."
> > Example:
> > $ uname -m
> > x86_64
> > $ uname -m
> > armv41
> 
> Interesting, but not necessarily accurate, depending on what you're
> looking for.  My Xeon show "i686", not x86-64 even though the cpuinfo
> flags show "lm".  Probably because I'm running a 32bit kernel on a 64bit
> system.
> 
> > There is also the C function "sysconf" and the shell function "getconf".
> >  These are POSIX too.
> > Example:
> > $ getconf _NPROCESSORS_ONLN
> 
> Lots of info there.  Interesting how many of the values are set to
> 200112 (including the POSIX and POSIX2 versions).  But it's unclear to
> (without deeper digging of the standards) the meaning of many of the
> variables and their values.  
> 
> But in looking at these two I realized I probably have a 32bit kernel
> running on a 64 bit system.  Never needed to run with 64bit so ignored
> the hassle of trying.  The question is:  does /proc/cpuinfo or uname -m
> report the hardware or the kernel or both?  And I wonder if that's
> sufficient to know if the kernel will work with 64bit applications?
> What's worse, I need to figure that out for non-Linux systems eventually
> (though I'm in no hurry to open that can of worms).  So whatever tools I
> use I have to identify both the hardware and the supporting kernel.
> 
> Hmmm.  Generalized remote system identification for a grid is going to
> be a royal pain the buttocks.  I see that now....
> 


When doing network loads, if I accidentally try a 64bit load on a 32-bit system, the net boot kernel complains at me.  So there's got to be a call that checks hardware over kernel.  Have you tried asking on the lkml?

Hugh