Re: [Xen-ia64-devel] [PATCH 0/7 TAKE 2] xenoprof for xen/ia64

[Isaku Yamahata <yamahata@xxxxxxxxxxxxx>]

supplement:
I expected that the xenoprof implementation would be revised
in the future.
The perfmon project is pushing arch-generic perfmon API
(both in-kernerl and user-kernel) which also replace x86
nmi interface.
When their patches are included in vanilla kernel and the xen
base kernel becomes the one which includes the new perfmon,
xenoprof implementation would have to be revised. And it would
be a good time to clean up current xenoprof/ia64 and to define
an arch-generic perfmon hypercall interface.

thanks.

On Tue, Nov 28, 2006 at 06:13:54PM +0900, Isaku Yamahata wrote:
> 
> Hi Alex.
> Thank you for your review.
> 
> I updated the patches and attached the tar balls which also
> includes missing patches in xen-ia64-unstable.hg.
> Please find it.
> 
> On Mon, Nov 27, 2006 at 03:06:27PM -0700, Alex Williamson wrote:
> 
> > +       if ( !softirq_pending(smp_processor_id()) ) {
> > +               if (!can_do_pal_halt)
> > +                       safe_halt();
> > +               else
> > +                       cpu_relax();
> > +       }
> > 
> > Is this logic backwards?  It's opposite the kernel.
> 
> Ouch! fixed.
> 
> 
> > --- a/xen/include/public/arch-ia64.h    Thu Oct 26 14:27:25 2006 +0900
> > +++ b/xen/include/public/arch-ia64.h    Wed Nov 22 20:41:08 2006 +0900
> > @@ -388,6 +388,9 @@ DEFINE_XEN_GUEST_HANDLE(vcpu_guest_conte
> >  #define IA64_DOM0VP_tlb_untrack_page    10
> > 
> > 
> > +/* xen perfmon */
> > +#define IA64_DOM0VP_perfmon             11
> > +
> > 
> > We don't have IA64_DOM0VP_tlb_untrack_page in the upstream tree, should
> > this be IA64_DOM0VP op #8? (or are we missing a patch upstream)
> 
> I changed it to 8.
> 
> 
> > ---- 12593_12361c7cc046_xenoprof_ia64_linux_side.patch
> > 
> > Having two different definitions for __perfmon_init/exit() for the
> > non-Xen code path is confusing.  Consider accessing it through a macro
> > in the Xen path (ex. xen_permon_init).  Then __perfmon_init/exit() could
> > be static for the non-Xen case (not that I'm a fan of adding more
> > #ifdefs).  Perhaps we need a #define LINUX_STATIC (?)
> 
> I hope I refined it.
> 
> 
> -- 
> yamahata

-- 
yamahata

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel