Re: [PATCH] KVM: PPC: Not optimizing MSR_CE and MSR_DE with paravirt.

[Alexander Graf <agraf@suse.de>]

On 05/16/2012 02:27 PM, Sethi Varun-B16395 wrote:
>
>> -----Original Message-----
>> From: kvm-ppc-owner@vger.kernel.org [mailto:kvm-ppc-
>> owner@vger.kernel.org] On Behalf Of Alexander Graf
>> Sent: Wednesday, May 16, 2012 2:42 PM
>> To: Bhushan Bharat-R65777
>> Cc: kvm-ppc@vger.kernel.org
>> Subject: Re: [PATCH] KVM: PPC: Not optimizing MSR_CE and MSR_DE with
>> paravirt.
>>
>> On 05/16/2012 10:42 AM, Bhushan Bharat-R65777 wrote:
>>>> -----Original Message-----
>>>> From: kvm-ppc-owner@vger.kernel.org
>>>> [mailto:kvm-ppc-owner@vger.kernel.org] On Behalf Of Alexander Graf
>>>> Sent: Tuesday, May 15, 2012 8:01 PM
>>>> To: Bhushan Bharat-R65777
>>>> Cc: kvm-ppc@vger.kernel.org; Bhushan Bharat-R65777
>>>> Subject: Re: [PATCH] KVM: PPC: Not optimizing MSR_CE and MSR_DE with
>> paravirt.
>>>> On 05/15/2012 09:37 AM, Bharat Bhushan wrote:
>>>>> From: Bhushan Bharat-R65777<R65777@freescale.com>
>>>>>
>>>>> If there is pending critical or machine check interrupt then guest
>>>>> would like to capture it when guest enable MSR.CE and MSR_ME
>> respectively.
>>>>> Also as mostly MSR_CE and MSR_ME are updated with rfi/rfci/rfmii
>>>>> which anyway traps so removing the the paravirt optimization for
>>>>> MSR.CE and MSR.ME.
>>>> It's only not safe for e500mc and above, right?
>>> For e500mc and above the paravirt emulation code will not come into
>> picture, And critical and machine check interrupt will happen only if MSR
>> have corresponding bits set. So they are already safe. Is not it?
>>
>> Yup, though it might be worth documenting the fact with a few #ifdef's,
>> in case anyone wants to run PR KVM on e500mc ever.
>>
> Synchronous error report machine checks don't depend on MSR[ME] in case of e500mc.
>
>>>> E500v2 and book3s should be
>>>> fine.
>>> And with this patch e500v2 will be fine? Not sure of book3s :).
>> Well, e500v2 has edge triggered MCs, no? MSR.CE is unsafe however, as
>> criticals are basically the same as externals.
>>
> In case of e500v2 machine check would be reported if MSR[ME], but if MSR[ME] is not set
> the core would enter a check stop state. Yes, machine checks on e500v2 are edge triggered.
> Why is MSR[CE] unsafe?

MSR=0
* critical interrupt comes in *
mtmsr(MSR_CE)
--> MSR = MSR_CE
* interrupt should be delivered, but host doesn't get notified that 
MSR_CE is changing *

However, we completely ignore critical interrupts in KVM these days, no?


Alex