Re: [Xen-users] Xenoprof (0.9.5) in PVOPS

[Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>]

On Wed, May 09, 2012 at 09:21:20AM +0100, Ian Campbell wrote:
> CCing Konrad who I think has a better handle on the current state of
> these patches.
> 
> On Wed, 2012-05-09 at 07:03 +0100, Chulmin Kim wrote:
> > > Subject: Re: [Xen-users] Xenoprof (0.9.5) in PVOPS
> > > 
> > > On Mon, 2012-05-07 at 16:51 +0100, Chulmin Kim wrote:
> > > > Iâm trying to use xenoprof in my Xeon machine with PVOPS (3.1.0
> > > > kernel) & xen 4.1.1.
> > > 
> > > > In the past, I had failed to use the xenoprof in this environment. (1
> > > > year ago)
> > > 
> > > > While baremetal booting shows the counter object in /dev/oprofile/0
> > > > (or 1,2,3) , booting with xen fails to create those objects.
> > > 
> > > > Of course, opcontrol command doesnât work. It only prints out that
> > > > error : counter 0 unavailableâ
> > > 
> > > > Iâve searched the web, but I couldnât get the enough info to resolve
> > > > it.
> > > 
> > > > Is there anyone who knows about this issue?
> > > 
> > > AIUI you need to patch a pvops kernel with a xenoprofile patch. You could 
> > > ask
> > > on the dev list where to find this.
> > 
> > Do you know the reason why the patch is not included in the current xen 
> > version? 
> > As far as I know, the patch was developed 2 years ago. 
> 
> AFAIK the patch was developed against the classic kernel patches and has
> simply been forward ported onto the PV ops kernel. No one has taken the
> time to go through the code, clean it up, ensure it uses all the right
> interfaces and has clean abstractions, split it up into an upstreamable
> patchset etc. Is that something you might be interested in doing?

Michael Petullo (CC-ed here) did a backport which he has posted on:
http://flyn.org/patches/linux-xen-passive-oprofile/linux-3.2-xen-passive-oprofile.patch.gz

which works pretty well. I did use it to instrument the hypervisor and dom0.
Didn't have much luck in figuring out what the HVM guests were doing (it somehow
only gave me IPs that looked like physical addresses).

Anyhow, I took a long look at those patches and the bulk of the changes
were introducing the '--active-domain' and 'passive-domain' functionality in
oprofile. The Xen oprofile backend did not seem that difficult.

So the reaon why the driver has not been upstreamed is simple - have not
had enough time to do it. If you are up to it - I can help along in a mentor
capacity to driver it for upstream acceptance.

> 
> Ian.

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users