Re: [PATCH] kvm: ppc: booke: Restore SPRG3 when entering guest

[Alexander Graf <agraf@suse.de>]

On 17.07.14 18:27, Alexander Graf wrote:
>
> On 17.07.14 18:24, Bharat.Bhushan@freescale.com wrote:
>>
>>> -----Original Message-----
>>> From: Alexander Graf [mailto:agraf@suse.de]
>>> Sent: Thursday, July 17, 2014 9:41 PM
>>> To: Bhushan Bharat-R65777; kvm-ppc@vger.kernel.org
>>> Cc: kvm@vger.kernel.org; Wood Scott-B07421; Yoder Stuart-B08248
>>> Subject: Re: [PATCH] kvm: ppc: booke: Restore SPRG3 when entering guest
>>>
>>>
>>> On 16.07.14 08:02, Bharat Bhushan wrote:
>>>> SPRG3 is guest accessible and SPRG3 can be clobbered by host or
>>>> another guest, So this need to be restored when loading guest state.
>>>>
>>>> Signed-off-by: Bharat Bhushan <Bharat.Bhushan@freescale.com>
>>>> ---
>>>>    arch/powerpc/kvm/booke_interrupts.S | 2 ++
>>>>    1 file changed, 2 insertions(+)
>>>>
>>>> diff --git a/arch/powerpc/kvm/booke_interrupts.S
>>>> b/arch/powerpc/kvm/booke_interrupts.S
>>>> index 2c6deb5ef..0d3403f 100644
>>>> --- a/arch/powerpc/kvm/booke_interrupts.S
>>>> +++ b/arch/powerpc/kvm/booke_interrupts.S
>>>> @@ -459,6 +459,8 @@ lightweight_exit:
>>>>         * written directly to the shared area, so we
>>>>         * need to reload them here with the guest's values.
>>>>         */
>>>> +    PPC_LD(r3, VCPU_SHARED_SPRG3, r5)
>>>> +    mtspr    SPRN_SPRG3, r3
>>> We also need to restore it when resuming the host, no?
>> I do not think host expect some meaningful value when returning from 
>> guest, same true for SPRG4-7.
>> So there seems no reason to save host values and restore them.
>
> Hmm - arch/powerpc/include/asm/reg.h says:
>
>  * All 32-bit:
>  *      - SPRG3 current thread_info pointer
>  *        (virtual on BookE, physical on others)
>
> but I can indeed find no trace of usage anywhere. This at least needs 
> to go into the patch description.

Bah - it obviously is used. It's SPRN_SPRG_THREAD. And it's so 
incredibly important that I have no idea how we could possibly run 
without switching the host value back in very early. And even then our 
interrupt handlers wouldn't work anymore.

This is more complicated :).


Alex