Re: [PATCH] x86/kvm: Show guest system/user cputime in cpustat

[Avi Kivity <avi@redhat.com>]

On 03/12/2010 10:53 AM, Qing He wrote:
>
>> When Qing(CCed) was working on nested VMX in the past, he found PV
>> vmread/vmwrite indeed works well(it would write to the virtual vmcs so vmwrite
>> can also benefit). Though compared to old machine(one our internal patch shows
>> improve more than 5%), NHM get less benefit due to the reduced vmexit cost.
>>
>>      
> One of the hurdles to PVize vmread/vmwrite is the fact that the memory
> layout of physical vmcs remains unknown. Of course it can use the custom
> vmcs layout utilized by nested virtualization, but that looks a little weird,
> since different nested virtualization implementation may create different
> custom layout.
>    

Note we must use a custom layout and cannot depend on the physical 
layout, due to live migration.  The layout becomes an ABI.

> I once used another approach to partially accelerate the vmread/vmwrite
> in nested virtualization case, which also gives good performance gain (around
> 7% on pre-nehalem, based on this, PV vmread/vmwrite had another 7%). That
> is to make a shortcut to handle EXIT_REASON_VM{READ,WRITE}, without
> even turning on the IF.
>    

Interesting.  That means our exit path is inefficient; it seems to imply 
half the time is spent outside the hardware vmexit path.

A quick profile (on non-Nehalem) shows many atomics and calls into the 
lapic, as well as update_cr8_intercept which is sometimes unnecessary; 
these could easily be optimized.

Definitely optimizing the non-paravirt path is preferred to adding more 
paravirtualization.

-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.