Re: [PATCH v2 2/3] x86/hvm: Allow writes to registers on the same page as MSI-X table

[Jan Beulich <jbeulich@xxxxxxxx>]

On 03.04.2023 06:21, Marek Marczykowski-Górecki wrote:
> On Tue, Mar 28, 2023 at 03:03:17PM +0200, Jan Beulich wrote:
>> On 28.03.2023 14:52, Marek Marczykowski-Górecki wrote:
>>> I mean, technically I can probably merge those two handlers together,
>>> but I don't think it will result in nicer code. Especially since the
>>> general direction is to abandon split of MSI-X table access handling
>>> between Xen and QEMU and go with just QEMU doing it, hopefully at some
>>> point not needing msixtbl_mmio_ops anymore (but still needing the one
>>> for adjacent accesses).
>>
>> Hmm, at this point I'm not convinced of this plan. Instead I was hoping
>> that once vPCI properly supports PVH DomU-s, we may also be able to make
>> use of it for HVM, delegating less to qemu rather than more.
> 
> In that case, this code won't be needed anymore, which will also make
> this handler unnecessary.
> 
> Anyway, I tried to merge this handling into existing handlers and the
> resulting patch is slightly bigger, so it doesn't seem to avoid any
> duplication. The only benefit I can think of is avoiding iterating
> msixtbl_list twice (for respective accept callbacks) on each access. Is
> it worth a bit more complicated handlers?

Well, limiting duplication (if possible) is only one aspect. Again
referring to Roger's functionally similar vPCI work, an important (from
my pov) aspect is to avoid consuming another handler slot (via a new
call to hvm_next_io_handler()).

Jan