Re: [PATCH 2/2] x86/video: do not assume a video mode to be unconditionally present

[Jan Beulich <jbeulich@xxxxxxxx>]

On 19.04.2024 09:42, Roger Pau Monné wrote:
> On Tue, Apr 02, 2024 at 11:49:20AM +0200, Jan Beulich wrote:
>> On 28.03.2024 16:35, Roger Pau Monne wrote:
>>> There's no reason to assume VGA text mode 3 to be unconditionally available.
>>> With the addition of booting Xen itself in PVH mode there's a boot path that
>>> explicitly short-circuits all the real-mode logic, including the VGA 
>>> detection.
>>>
>>> Leave the default user selected mode as text mode 3 in boot_vid_mode, but do
>>> not populate boot_vid_info with any default settings.  It will either be
>>> populated by the real-mode video detection code, or left zeroed in case
>>> real-mode code is skipped.
>>>
>>> Note that only PVH skips the real-mode portion of the boot trampoline,
>>> otherwise the only way to skip it is to set `no-real-mode` on the command 
>>> line,
>>> and the description for the option already notes that VGA would be disabled 
>>> as
>>> a result of skipping real-mode bootstrap.
>>>
>>> This fixes Xen incorrectly reporting:
>>>
>>> (XEN) Video information:
>>> (XEN)  VGA is text mode 80x25, font 8x16
>>>
>>> When booted as a PVH guest.
>>
>> And what effect does this have on a bare-metal boot with no-real-mode in use?
>> The default on x86 hardware still is that in the absence of other 
>> information,
>> a VGA of some kind can be assumed to be there. Yes, there are headless
>> systems, but better assume VGA is there when there's not than the other way
>> around.
> 
> But that contradicts the text of the 'no-real-mode' option, which
> explicitly notes:
> 
> "Do not execute real-mode bootstrap code when booting Xen. This option
> should not be used except for debugging. It will effectively disable
> the vga option, which relies on real mode to set the video mode."

Well. Even without setting a video mode, _some_ mode is set (by firmware) as
long as there is a VGA. In the absence of a "vga=" option iirc we'd retrieve
that setting, unless we're not allowed to by "no-real-mode". In which case,
as indicated, we may still be better off guessing a basic mode than kind of
suggesting the absence of any VGA.

Besides, when booting from EFI vga= is respected despite not going through
real mode. So the quoted text isn't quite right anyway.

Jan

>> What I would have expected is for the PVH boot path to clear boot_vid_info.
> 
> Well, my intention was to fix both PVH and also make the
> implementation of the 'no-real-mode' option consistent with the
> documentation.
> 
> Thanks, Roger.