Re: [Invitation] bi-weekly guest_memfd upstream call on 2024-11-14

[Alexey Kardashevskiy <aik@amd.com>]

On 24/12/24 22:27, David Hildenbrand wrote:
> On 24.12.24 05:21, Alexey Kardashevskiy wrote:
>> On 14/11/24 13:27, Chao Gao wrote:
>>>>> With in-place conversion, QEMU can map shared memory and supply the 
>>>>> virtual
>>>>> address to VFIO to set up DMA mappings. From this perspective, 
>>>>> in-place
>>>>> conversion doesn't change or require any changes to the way QEMU 
>>>>> interacts
>>>>> with VFIO. So, the key for device assignment remains updating DMA 
>>>>> mappings
>>>>> accordingly during shared/private conversions. It seems that 
>>>>> whether in-place
>>>>> conversion is in use (i.e., whether shared memory is managed by 
>>>>> guest_memfd or
>>>>> not) doesn't require big changes to that proposal. Not sure if 
>>>>> anyone thinks
>>>>> otherwise. We want to align with you on the direction for device 
>>>>> assignment
>>>>> support for guest_memfd.
>>>>> (I set aside the idea of letting KVM manage the IOMMU page table in 
>>>>> the above
>>>>>     analysis because we probably won't get that support in the near 
>>>>> future)
>>>>
>>>> Right. So devices would also only be to access "shared" memory.
>>>
>>> Yes, this is the situation without TDX-Connect support. Even when 
>>> TDX-Connect
>>> comes into play, devices will initially be attached in shared mode 
>>> and later
>>> converted to private mode. From this perspective, TDX-Connect will be 
>>> built on
>>> this shared device assignment proposal.
>>>
>>>>
>>>>>
>>>>> Could you please add this topic to the agenda?
>>>>
>>>> Will do. But I'm afraid the agenda for tomorrow is pretty packed, so 
>>>> we might
>>>> not get to talk about it in more detail before the meeting in 2 weeks.
>>>
>>> Understood. is there any QEMU patch available for in-place 
>>> conversion? we would
>>> like to play with it and also do some experiments w/ assigned 
>>> devices. This
>>> might help us identify more potential issues for discussion.
>>
>>
>> Have you found out if there are patches, somewhere? I am interested too.
> 
> I remember that so far only Kernel patches are available [1], I assume 
> because Google focuses on other user space than QEMU. So I suspect the 
> QEMU integration is still TBD.
> 
> 
> [1] https://lkml.kernel.org/r/20241213164811.2006197-1-tabba@google.com

thanks for confirming. I saw that but could not spot the in-place 
conversion there. And I had to re-watch how pKVM actually works :)

-- 
Alexey