Re: [PATCH 4 of 4] Add a trace hypercall to allow tracing from dom0 or domU

[Olaf Hering <olaf@aepfle.de>]

On Fri, Jul 15, George Dunlap wrote:

> On Fri, 2011-07-15 at 16:38 +0100, Olaf Hering wrote:
> > @@ -65,6 +67,7 @@
> >  #define TRC_LOST_RECORDS        (TRC_GEN + 1)
> >  #define TRC_TRACE_WRAP_BUFFER  (TRC_GEN + 2)
> >  #define TRC_TRACE_CPU_CHANGE    (TRC_GEN + 3)
> > +#define TRC_TRACE_GUEST_HYPERCALL (TRC_GEN + 4)
> 
> Looks like perhaps a leftover from an earlier patch? :-)

Yes, thats true. It was a domctrl a few weeks ago.

> > diff -r 6e4aef7b5051 -r f72dcd1b8bbd xen/include/public/xen.h
> > --- a/xen/include/public/xen.h
> > +++ b/xen/include/public/xen.h
> > @@ -94,6 +94,7 @@ DEFINE_XEN_GUEST_HANDLE(xen_pfn_t);
> >  #define __HYPERVISOR_kexec_op             37
> >  #define __HYPERVISOR_tmem_op              38
> >  #define __HYPERVISOR_xc_reserved_op       39 /* reserved for XenClient */
> > +#define __HYPERVISOR_xentrace_op          40
> 
> Seems unnecessary to make a whole hypercall just for this one thing, but
> I'll defer to Keir's judgement on that.  At very least, it should accept
> an "op" command, of which "trace" is just one, so that it's expandable
> in the future without breaking backwards compatibility.

One reason is that trace_hypercall() should not trace itself.

Olaf