Re: Not emulated registers on BOOKE_HV (GS-mode)

[Alexander Graf <agraf@suse.de>]

On 05/16/2012 11:02 AM, Sethi Varun-B16395 wrote:
>
>> -----Original Message-----
>> From: kvm-ppc-owner@vger.kernel.org [mailto:kvm-ppc-
>> owner@vger.kernel.org] On Behalf Of Bhushan Bharat-R65777
>> Sent: Wednesday, May 16, 2012 12:58 PM
>> To: Alexander Graf
>> Cc: Wood Scott-B07421; Yoder Stuart-B08248; kvm-ppc@vger.kernel.org
>> Subject: Not emulated registers on BOOKE_HV (GS-mode)
>>
>> Hi Alex,
>>
>> There is below comment in arch/powerpc/kvm/booke_emulate.c
>>
>> /*
>>   * NOTE: some of these registers are not emulated on BOOKE_HV (GS-mode).
>>   * Their backing store is in real registers, and these functions
>>   * will return the wrong result if called for them in another context
>>   * (such as debugging).
>>   */
>>
>> "some of these registers are not emulated on BOOKE_HV (GS-mode)"
>>   1) Is not that mtspr()/mfspr() for "not emulated" registers should
>> follow EMULATE_FAIL path? So should be ifdef out for BOOKE_HV? Otherwise
>> the emulation code execute.
>>   2) Or These are not emulated because the GS mode have direct access to
>> these registers, Right? So no trap?
>>
> For BOOKE_HV mtspr/mfspr would get mapped to hardware maintained guest shadow copies.
> For example guest access to dear would get mapped to gdear and wouldn't trap
>
>> "and these functions will return the wrong result if called for them in
>> another context (such as debugging)."
>>   1) So do you mean that guest is not supposed to access these registers
>> in normal scenario but the debugger (some command on gdb in guest) can
>> access these register? then does it make sense to treat mtspr() as nop
>> and mfspr returns 0/undefined?
>>
> Actually, in a normal scenario (For BOOKE_HV) guest (MSR[GS] = 1) would be able to directly access these registers
> without hypervisor intervention. However I am not sure under what condition would the guest access generate a trap, all
> guest accesses would be with MSR[GS]=1.

Yup. E500mc would just never reach that code - and even if it did, it 
simply wouldn't do anything useful.

What did you guys need this change for? I'm sure there's some incentive 
behind it, right? The only one I can see right now would be reduced code 
size.


Alex