Re: [Qemu-devel] [PATCH v5 0/6] i386: expose floppy-related objects in SSDT

[Laszlo Ersek <lersek@redhat.com>]

On 01/13/16 16:49, Roman Kagan wrote:
> On Wed, Jan 13, 2016 at 03:36:18PM +0100, Laszlo Ersek wrote:
>> On 12/30/15 21:11, Roman Kagan wrote:
>>> Windows on UEFI systems is only capable of detecting the presence and
>>> the type of floppy drives via corresponding ACPI objects.
>>
>> I'm late to the party, but please allow me a question:
>>
>> how did you figure out that UEFI Windows requires this?
>>
>> In general, what the ACPI specification says is at best a "guideline"
>> for Windows. So how did you prove this was a requirement for Windows?
> 
> Well, my statement above that Windows on UEFI can detect floppies *only*
> via ACPI is probably a bit stronger than I can actually prove but
> 
> - Windows on OVMF didn't see floppies before the patch, while Linux did
>   (by querying CMOS)

Indeed.

https://bugzilla.redhat.com/show_bug.cgi?id=1212317

> 
> - a number of sources on the internet hinted that Windows needed ACPI
>   assistance for that, e.g.:
> 
>   https://www.reactos.org/wiki/UEFI#Floppy
>   https://social.technet.microsoft.com/Forums/windows/en-US/f17db175-d146-4518-b2e9-c12a15031222/legacy-floppy-compatibility-with-uefi-boot?forum=w7itprohardware
>   https://social.technet.microsoft.com/Forums/windows/en-US/e91ec27b-0c2d-44a3-b949-e77fa810a4c0/windows-7-uefi-fdd-how-to?forum=w7itprohardware

Right, I found these. (The last two anyway.)

I also found technet comments that plainly stated "it would never work".

(Under your first link, I can read as well: "There have been reports
that Windows does not properly support motherboard floppy controllers
when booting from UEFI. The cause is not definitively known though a
couple of pieces of data have emerged, one pointing to an issue with ACPI".)

> - the links mentioned the need in _FDE object but indicated it only
>   allowed for successful enumeration of floppies, not the actual access;
>   I proved that experimentally
> 
> - the ACPI spec stated that _FDE went in concert with _FDI so I tried it
>   and it worked out

Thank you for confirming.

So, improving Windows compat in QEMU remains trial-and-error-based, and
occasionally reverse-engineering-based. Deplorable.

> Voila.  Besides, I later discovered that a similar research had been
> carried out for Parallels proprietary hypervisor, with a similar
> outcome.

That is, large amounts of work are being duplicated between participants
in this industry segment, because Microsoft doesn't give a flying fsck
about documenting their exact platform requirements. (The fact that _FDE
and _FDI are described in the ACPI spec means exactly zilch, because
Microsoft have ignored e.g. DataTableRegion from the same spec, since
ACPI 2.0 -- the year 2000.) I'm quite sure this obscurity is
intentional, and meant to spread uncertainty and waste competitors'
resources.

Whereas they are having a field day whenever they look at open source
components in a hybrid virt stack.

Nothing to see here, move along.

Thanks again.
Laszlo