Re: [PATCH] ArmVirtPkg: remove QemuVideoDxe from ArmVirtQemu and ArmVirtQemuKernel

[Ard Biesheuvel <ard.biesheuvel@linaro.org>]

On 24 August 2017 at 13:02, Leif Lindholm <leif.lindholm@linaro.org> wrote:
> On Wed, Aug 23, 2017 at 04:00:57PM +0100, Leif Lindholm wrote:
>> On Wed, Aug 23, 2017 at 03:36:37PM +0200, Laszlo Ersek wrote:
>> > On 08/23/17 15:17, Ard Biesheuvel wrote:
>> > >>>> (My R-b stands; these are comments for a possible followup patch.)
>> > >>>>
>> > >>>> Please see:
>> > >>>>
>> > >>>> - commit 84a75f70e903 ("OvmfPkg/QemuVideoDxe: enable ARM builds",
>> > >>>> 2015-02-23),
>> > >>>>
>> > >>>> - commit 05a537945872 ("OvmfPkg/QemuVideoDxe: Helper functions for
>> > >>>> unaligned port I/O.", 2017-04-07)?
>> > >>>>
>> > >>>> In my opinion, we should now revert parts of these commits, in one
>> > >>>> followup patch:
>> > >>>>
>> > >>>> - from the first commit, we should revert only the "VALID_ARCHITECTURES"
>> > >>>> comment change (the rest is built upon by the second commit, and should
>> > >>>> be preserved)
>> > >>>>
>> > >>>> - from the second commit, we should revert the addition of [Sources.ARM,
>> > >>>> Sources.AARCH64].
>> > >>>>
>> > >>>> This boils down to removing ARM and AARCH64 references from the
>> > >>>> QemuVideoDxe.inf file. If you agree, could you please submit such a
>> > >>>> followup patch?
>> > >>>
>> > >>> Sure, but pending the graphical GRUB discussion.
>> > >>
>> > >> So, after looking at the GRUB code, I am leaning towards agreeing that
>> > >> this is actually not a problem at all ... probably. The efi_gop driver
>> > >> does a Blt() of the entire screen from an off-screen buffer for all
>> > >> updates _unless_ it fails to allocate that off-screen buffer.
>> > >>
>> > >> So, basically, if you run out of memory at that point, it will try to
>> > >> preserve a way to get messages out about that. I will send a question
>> > >> out to grub-devel regarding this behaviour.
>> > >>
>> > >> However, looking at the specification, a question remains over how
>> > >> software can determine whether direct FB access is possible. I mean, a
>> > >> value of 0 seems like a decent hint, but the spec says nothing on the
>> > >> topic.
>> > >>
>> > >
>> > > It will assume the FB is accessible unless the pixel format is PixelBltOnly
>> > >
>> >
>> > Correct, the UEFI spec (v2.7) says in "12.9 Graphics Output Protocol":
>> >
>> >   PixelBltOnly  This mode does not support a physical frame buffer.
>> >
>> > and
>> >
>> >   PixelFormat   Enumeration that defines the physical format of the
>> >                 pixel. A value of PixelBltOnly implies that a linear
>> >                 frame buffer is not available for this mode.
>>
>> -ETOOMANYLEVELSOFINDIRECTION
>> Right, so that's a non-issue.
>>
>> Hopefully, that means other operating systems can also deal with the
>> lack.
>>
>> > IIRC simply recognizing and accepting this enum constant was the point
>> > of Alex's patch.
>>
>> Ah, yes, that makes sense now.
>> It's a bit surprising things work without this patch, but GRUBs
>> fallback seems to match anyway.
>
> To clarify:
> Reviewed-by: Leif Lindholm <leif.lindholm@linaro.org>

Pushed, thanks.