Kernel performance measurement tool

Amit Kucheria amit.kucheria at canonical.com
Fri Sep 11 10:09:46 UTC 2009


On 09 Sep 10, Tim Gardner wrote:
> Andy Whitcroft wrote:
> > On Thu, Sep 10, 2009 at 04:39:58PM +0100, Scott James Remnant wrote:
> >> On Thu, 2009-09-10 at 09:11 -0600, Tim Gardner wrote:
> >>
> >>> Scott - 2.6.31 contains a new performance measurement tool that is
> >>> maintained within the upstream kernel git repository. By virtue of its
> >>> location, I assume that this tool is very sensitive to the kernel ABI.
> >>> How would you suggest we package this executable? Place it in
> >>> /lib/modules/`uname -r` somewhere? I don't think the default install
> >>> location from the Makefile is sensible with respect to providing a
> >>> unique instance of the executable for each kernel ABI.
> >>>
> >> /lib/kernel/$(ver)/tool maybe?
> >>
> >> With a shell script in /usr/bin that picks the right one?
> > 
> > That sounds pretty sensible to me.  We could have a single script which
> > was sharable for that too.
> > 
> > -apw
> 
> The /usr/bin/perf script should go in a package other then the kernel
> package. Otherwise it'll get removed or updated every time a linux-image
> package is swizzled.
> 
> Whats a kernel related package where we could drop this script? Does
> initramfs-tools make sense? udev? It hardly seems worthwhile creating a
> whole new package just for this 2 line script.

Since it is so tied in to the kernel, IMHO it should stay in the kernel
package.

No point it putting it into non-obvious packages just to save
it being reinstalled on every kernel upgrade.

-- 
----------------------------------------------------------------------
Amit Kucheria, Kernel Engineer || amit.kucheria at canonical.com
----------------------------------------------------------------------




More information about the kernel-team mailing list