Re: [PATCH v1 2/2] perf/core: Fake regs for leaked kernel samples

[peterz]

On Fri, Aug 07, 2020 at 02:24:30PM +0800, Jin, Yao wrote:
> Hi Peter,
> 
> On 8/6/2020 7:00 PM, peterz@xxxxxxxxxxxxx wrote:
> > On Thu, Aug 06, 2020 at 11:18:27AM +0200, peterz@xxxxxxxxxxxxx wrote:
> > 
> > > Suppose we have nested virt:
> > > 
> > > 	L0-hv
> > > 	|
> > > 	G0/L1-hv
> > > 	   |
> > > 	   G1
> > > 
> > > And we're running in G0, then:
> > > 
> > >   - 'exclude_hv' would exclude L0 events
> > >   - 'exclude_host' would ... exclude L1-hv events?
> > >   - 'exclude_guest' would ... exclude G1 events?
> > 
> > So in arch/x86/events/intel/core.c we have:
> > 
> > static inline void intel_set_masks(struct perf_event *event, int idx)
> > {
> > 	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
> > 
> > 	if (event->attr.exclude_host)
> > 		__set_bit(idx, (unsigned long *)&cpuc->intel_ctrl_guest_mask);
> > 	if (event->attr.exclude_guest)
> > 		__set_bit(idx, (unsigned long *)&cpuc->intel_ctrl_host_mask);
> > 	if (event_is_checkpointed(event))
> > 		__set_bit(idx, (unsigned long *)&cpuc->intel_cp_status);
> > }
> > 
> 
> exclude_host is now set by guest (pmc_reprogram_counter,
> arch/x86/kvm/pmu.c). When enabling the event, we can check exclude_host to
> know if it's a guest.
> 
> Otherwise we may need more flags in event->attr to indicate the status.
> 
> > which is, afaict, just plain wrong. Should that not be something like:
> > 
> > 	if (!event->attr.exclude_host)
> > 		__set_bit(idx, (unsigned long *)&cpuc->intel_ctrl_host_mask);
> > 	if (!event->attr.exclude_guest)
> > 		__set_bit(idx, (unsigned long *)&cpuc->intel_ctrl_guest_mask);
> > 
> > 
> 
> How can we know it's guest or host even if exclude_host is set in guest?

I'm not following you, consider:

	xh	xg	h	g	h'	g'
	0	0	0	0	1	1
	0	1	1	0	1	0
	1	0	0	1	0	1
	1	1	1	1	0	0


So the 0,0 and 1,1 cases get flipped. I have a suspicion, but this
_really_ should have fat comments all over :-(

What a sodding trainwreck..