[systemd-devel] [PATCH 1/1] graphics: Do not require nonstandard console control protocol

[Kay Sievers]

On Wed, Jan 15, 2014 at 4:01 PM, Tom Gundersen <teg at jklm.no> wrote:
> On Wed, Jan 15, 2014 at 3:55 PM, Joonas Lahtinen
> <joonas.lahtinen at linux.intel.com> wrote:
>> On 15.01.2014 13:34, Tom Gundersen wrote:
>>>
>>> On Wed, Jan 15, 2014 at 12:08 PM, Joonas Lahtinen
>>> <joonas.lahtinen at linux.intel.com> wrote:
>>>>
>>>> Console Control Protocol is a nonstandard UEFI interface and lack of
>>>> it shall not be treated as an error.
>>>
>>> Ah, great! I was struggling to make the splash work on tianocore for
>>> the longest time, and this fixes it indeed.
>>>
>>> Applied and pushed.
>>
>>
>> Great that it helped. The same goes for other implementations too.
>>
>>
>>> Only minor issue left now is that we need to explicitly clear the
>>> screen before writing in console mode (as can be seen if you boot into
>>> tianocore and press 'P' followed by any other keys to switch between
>>> the modes). Care to take a look at that?
>>
>>
>> I think it is more by design than a bug, as now the graphics mode switch
>> function does nothing. There's no telling the UEFI implementation if you
>> want to draw text or graphics. I've seen multiple implementations in some
>> Windows bootloader leaves some text on screen behind logo, and others where
>> only cursor is seen, and in rare case the background is completely black.
>> It's just not specified by standard (or at least not implemented widely), if
>> the text output should cause screen to clear.
>>
>> Fbcon behaves pretty much the same way, the graphics get overwritten by the
>> text, and the text continues from where it left off, independent of each
>> other. So it's up to the application to enforce screen clearing if they want
>> to.
>>
>> One simple solution to ensure uniform behaviour across implementations could
>> be to do explicit clear on the screen and home the cursor to upper left
>> always when a text mode is requested. And vice versa, clear screen when
>> graphics mode is requested.
>>
>> Was this the problem you meant?
>
> Yeah, that's what I had in mind. I guess it only is a "problem" in the
> debug case, but still uniformity would be good.

Added the call. Seems to work fine here.

Thanks,
Kay