Re: [PATCH v4 14/17] x86/VPMU: Save VPMU state for PV guests during context switch

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

On 02/04/2014 06:38 AM, Jan Beulich wrote:
>>>> On 21.01.14 at 20:08, Boris Ostrovsky <boris.ostrovsky@oracle.com> wrote:
>> Save VPMU state during context switch for both HVM and PV guests unless we
>> are in PMU privileged mode (i.e. dom0 is doing all profiling) and the
>> switched
>> out domain is not the control domain. The latter condition is needed because
>> me may have just turned the privileged PMU mode on and thus need to save
>> last domain.
> While this is understandable, ...
>
>> --- a/xen/arch/x86/domain.c
>> +++ b/xen/arch/x86/domain.c
>> @@ -1444,17 +1444,16 @@ void context_switch(struct vcpu *prev, struct vcpu *next)
>>       }
>>   
>>       if (prev != next)
>> -        update_runstate_area(prev);
>> -
>> -    if ( is_hvm_vcpu(prev) )
>>       {
>> -        if (prev != next)
>> +        update_runstate_area(prev);
>> +        if ( !(vpmu_mode & XENPMU_MODE_PRIV) ||
>> +             !is_control_domain(prev->domain) )
>>               vpmu_save(prev);
> ... I'd really like you to investigate ways to achieve the same effect
> without this extra second condition added to the context switch path.
> E.g. by synchronously issuing a save on all affected vCPU-s when
> privileged mode gets turned on.

Yes, I should do something like that.

-boris