Re: [PATCH 2/2] x86/msi: Allow writes to registers on the same page as MSI-X table

[Jan Beulich <jbeulich@suse.com>]

On 18.11.2022 14:00, Marek Marczykowski-Górecki wrote:
> On Fri, Nov 18, 2022 at 01:33:39PM +0100, Jan Beulich wrote:
>> On 18.11.2022 13:19, Marek Marczykowski-Górecki wrote:
>>> On Fri, Nov 18, 2022 at 08:20:14AM +0100, Jan Beulich wrote:
>>>> On 17.11.2022 18:31, Marek Marczykowski-Górecki wrote:
>>>>> On Thu, Nov 17, 2022 at 05:34:36PM +0100, Jan Beulich wrote:
>>>>>> Which in turn raises the question: Do you need to handle reads
>>>>>> in the new code in the first place?
>>>>>
>>>>> The page not being mapped is also the reason why I do need to handle
>>>>> reads too.
>>>>
>>>> Just for my own clarity: You mean "not mapped to qemu" here?
>>>
>>> No, to the HVM domain (in p2m). Xen (outside of MSI-X specific code for
>>> HVM) doesn't know where those reads should be from.
>>
>> Hmm, I was expecting them to be mapped r/o to the guest, but perhaps I'm
>> misremembering. Clearly both ept_p2m_type_to_flags() and
>> p2m_type_to_flags() take mmio_ro_ranges into consideration, which is
>> what I was basing my understanding on (without having looked at other
>> places in detail).
> 
> Qemu doesn't map the page (using xc_domain_memory_mapping()) where MSI-X
> table lives. I tried to modify this part, but it resulted in EPT
> violation on write attempt (and my emulation code wasn't called at all).

Well, yes - that leads to the other path I pointed you at, which is
used for both HVM and PV (and which, as per what you say, then looks
to be relevant for PVH only right now).

Jan