Re: [PATCH v4] perf/x86/amd: Don't touch the Host-only bit inside the guest hypervisor

[Dongli Si]

On 28/03/2022 14:03, Peter Zijlstra wrote:
> Better I suppose, but I think the comments can be improved by covering
> the 'why' of things. We can all read the code to see the what of it.

I will add comments to the code to explain 'why'.

> Anyway, doesn't this also affect behaviour? I'm guessing this HO bit is
> only set by perf-record for events it wants to record on the host. But
> by not setting it, we'll also record the activity of the guest.

I think the HO/GO bit can only be set on the host, and should only be set
if SVM is enabled.

When the SVM is disabled, set the HO/GO bit will cause the performance
counters to not work.

Set the HO/GO bit inside the guest will cause the guest emitted
"unchecked MSR access error" warning, can be triggered by running
"perf stat -e instructions:G ls" in the guest, because this will set
the GO bit in the guest, and perf_ctr_virt_mask just mask the HO bit.

My patch does not affect the host, it just fixes the bug in the guest.

> So suppose we create a CPU wide HO event, then it will only count L0
> activity, right? Any L1 (or higher) activite will be invisible.

I don't quite understand your question.

> But with this change on, the L1 HV doesn't provide these same semantics,
> it's guest will be included in that host counter.

I don't think applying this patch will cause L2 guests to be included in
the host counter.

> Or is there additional counter {dis,en}abling on virt enter,exit (resp.)
> to achieve these semantics?

I don't think there is such a counter.

Also, I found that the L1 HV will emitted "unchecked MSR access error"
warning when "perf record" is executed on L2,
because the GO bit of L1 HV is set.

I wrote a new patch to mask the HO/GO bit in the guest, I will send it later.

Regards,
Dongli