Re: [PATCH v3 2/4] KVM: arm64: GICv4.1: Try to save hw pending state in save_pending_tables

[Shenming Lu]

On 2021/3/12 17:02, Marc Zyngier wrote:
> On Thu, 11 Mar 2021 12:31:48 +0000,
> Shenming Lu <lushenming@xxxxxxxxxx> wrote:
>>
>> On 2021/3/11 17:09, Marc Zyngier wrote:
> 
>>> I have asked that question in the past: is it actually safe to remap
>>> the vPEs and expect them to be runnable
>>
>> In my opinion, logically it can work, but there might be problems like the
>> one below that I didn't notice...
> 
> One thing is that you will have lost interrupts in the meantime
> (assuming your devices are still alive). How will you make up for
> that?

I think that devices should be paused for (not only) saving interrupt states,
and in fact, that's exactly what such as VFIO devices do...

> 
>>
>>>
>>> Also, the current code assumes that VMAPP.PTZ can be advertised if a
>>> VPT is mapped for the first time. Clearly, it is unlikely that the VPT
>>> will be only populated with 0s, so you'll end up with state corruption
>>> on the first remap.
>>
>> Oh, thanks for pointing it out.
>> And if we always signal PTZ when alloc = 1, does it mean that we
>> can't remap the vPE when the VPT is not empty, thus there is no
>> chance to get the VLPI state?  Could we just assume that the VPT is
>> not empty when first mapping the vPE?
> 
> I think we should drop the setting of PTZ altogether. It is a silly
> micro-optimisation, and if the HW can't parse the VPT efficiently when
> it is empty, then the HW is pretty bad, PTZ or not.

agree :-)

Thanks,
Shenming

> 
> Thanks,
> 
> 	M.
>