Re: [PATCH v5 06/14] vpci/header: implement guest BAR register handlers

[Jan Beulich <jbeulich@xxxxxxxx>]

On 03.02.2022 13:36, Oleksandr Andrushchenko wrote:
> Hi, Bertrand!
> 
> On 26.11.21 14:19, Oleksandr Andrushchenko wrote:
>> Hi, Bertrand!
>>
>> On 25.11.21 18:28, Bertrand Marquis wrote:
>>> Hi Oleksandr,
>>>
>>>> On 25 Nov 2021, at 11:02, Oleksandr Andrushchenko <andr2000@xxxxxxxxx> 
>>>> wrote:
>>>>
>>>> From: Oleksandr Andrushchenko <oleksandr_andrushchenko@xxxxxxxx>
>>>>
>>>> Add relevant vpci register handlers when assigning PCI device to a domain
>>>> and remove those when de-assigning. This allows having different
>>>> handlers for different domains, e.g. hwdom and other guests.
>>>>
>>>> Emulate guest BAR register values: this allows creating a guest view
>>>> of the registers and emulates size and properties probe as it is done
>>>> during PCI device enumeration by the guest.
>>>>
>>>> ROM BAR is only handled for the hardware domain and for guest domains
>>>> there is a stub: at the moment PCI expansion ROM handling is supported
>>>> for x86 only and it might not be used by other architectures without
>>>> emulating x86. Other use-cases may include using that expansion ROM before
>>>> Xen boots, hence no emulation is needed in Xen itself. Or when a guest
>>>> wants to use the ROM code which seems to be rare.
>>> In the generic code, bars for ioports are actually skipped (check code 
>>> before
>>> in header.c, in case of ioports there is a continue) and no handler is 
>>> registered for them.
>>> The consequence will be that a guest will access hardware when reading 
>>> those BARs.
>> Yes, this seems to be a valid point
> So, with the approach we have developed these days we will ignore all writes
> and return ~0 for reads for all unhandled ops, e.g. those which do not have 
> explicit
> register handlers employed. Thus, this case will fall into unhandled clause.

Except that I guess BARs are special in that reads may not return ~0,
or else the low bits carry a meaning we don't want to convey. Unused
BARs need to be hard-wired to 0, I think.

Jan