Re: [PATCH v1 0/5] KVM in-guest performance monitoring

[Joerg Roedel <joro@8bytes.org>]

On Thu, May 12, 2011 at 05:37:43PM +0300, Avi Kivity wrote:
> On 05/12/2011 05:24 PM, Joerg Roedel wrote:

> One thing we could do is paravirtualize at a lower level, introduce a  
> hypercall for batch MSR reads and writes.  So we can use the existing  
> PMU semantics and code, just optimize the switch.  This is similar to  
> what Xen did with lazy cpu updates, and what kvm did for paravirt  
> pagetable writes.

In general a good idea. For MSR writes this isn't trivial as well. MSR
reads and writes can cause a #GP, this needs to be represented in this
interface. Can we allow that a batched MSR write changes the operation
mode of the vcpu and so on.
Or we limit this interface to the PMU MSRs, then we are pretty much at
a paravirt-pmu.


> I've considered something similar for mmio - use hypercalls for ordinary  
> mmio to avoid calling into the emulator - but virtio uses pio which  
> isn't emulated and we don't have massive consumers of mmio (except  
> perhaps hpet).
>
> (and we can have a cpuid bit that advertises whether we recommend to use  
> this feature for PMU MSRs; if/when we get hardware support, we turn it 
> off)

A cpuid-bit that indicates which pmu is prefered is certainly a good
idea.

	Joerg