[PATCH v6 12/12] test: efi: boot: Add a test for the efi bootmeth

[Simon Glass]

Hi Heinrich,

On Tue, 15 Oct 2024 at 05:36, Heinrich Schuchardt <xypron.glpk at gmx.de> wrote:
>
> On 15.10.24 12:19, Mark Kettenis wrote:
> >> Date: Mon, 14 Oct 2024 15:11:50 -0600
> >> From: Tom Rini <trini at konsulko.com>
> >>
> >> On Mon, Oct 14, 2024 at 01:13:41PM -0600, Simon Glass wrote:
> >>
> >> [snip]
> >>> Or perhaps just have a way to turn it off? I first sent this patch
> >>> last November. It is just wrong to generate output like this which we
> >>> don't want. There isn't even a test for it, so just add a way to
> >>> disable it, and be done!
> >>
> >> I don't know that it's unwanted. As I'm trying to get Heinrich to
> >> explain, _why_ does efi_setup_console_size need to exist, and do what it
> >> does? This isn't the case of color-coding the output of tests as they
> >> happen, there must be a reason we care about knowing the console size.
> >> At that point we can figure out if the right answer is:
> >> - Don't generate that check on serial ports, it's somewhere between
> >>    misleading to wrong.
> >> - text-based tests just need to expect and skip it because there's a
> >>    good reason to need to know the console size and not just assume
> >>    80x24.
> >> - Something else we won't know until it's clearly explained why we do
> >>    this.
> >
> > Sadly this is a misfeature of UEFI.  The EFI_SIMPLE_TEXT_OUTPUT_PROTOCOL
> > is required if console devices are supported by the UEFI
> > implementation, and is built around the concept of "text mode" with a
> > fixed number of character rows and columns.  That pretty much assumes
> > there is some sort of terminal emulator on the other end.  And this is
> > pretty much incompatible with anything that just want to log serial
> > output.
> >
> > I think it is possible to do better though.  Instead of calling
> > efi_setup_console_size() in efi_init_obj_list(), we could postpone
> > this until the application makes a call that requires us to know the
> > size.  This would mean that a simple EFI application that just uses
> > EFI_SIMPLE_TEXT_OUTPUT_PROTOCOL.OutputString() (such as the Hello
> > world EFI application or the OpenBSD EFI bootloader) wouldn't have to
> > do the size query.
>
> The implementation of EFI_SIMPLE_TEXT_OUTPUT_PROTOCOL.OutputString()
> needs to know the output size to update the cursor position in
>
>    EFI_SIMPLE_TEXT_OUTPUT_PROTOCOL.MODE.CursorColumn
>    EFI_SIMPLE_TEXT_OUTPUT_PROTOCOL.MODE.CursorRow
>
> The EFI_SIMPLE_TEXT_OUTPUT_PROTOCOL is passed to the entry point of any
> EFI application via the system table.
>
> We could delay the invocation of efi_setup_console_size() to the launch
> of the first EFI application (efi_start_image()).

That's a start, but I think the right answer here is to use my flag
and in efi_setup_console_size() don't call query_console_serial() if
the flag is set. Then the default (80x25) will be returned, which is
fine for the test which is the subject of this series.

Regards,
Simon