Re: [edk2-devel] Intel NUC platform firmware -- no serial I/O support?

["Laszlo Ersek" <lersek@redhat.com>]

On 04/08/22 11:48, Gerd Hoffmann wrote:
> 
>   Hi,
>  
>> And It Just Works (TM), with the following two commands in the UEFI
>> shell (after I copied the binaries to the USB stick, alongside the UEFI
>> Shell binary I built earlier):
>>
>>> Shell> fs0:
>>> FS0:\> cd efi\boot
>>> FS0:\efi\boot\> load SerialDxe.efi
>>> Image 'FS0:\EFI\BOOT\SerialDxe.efi' loaded at 2C801000 - Success
>>> FS0:\efi\boot\> load TerminalDxe.efi
>>> Image 'FS0:\EFI\BOOT\TerminalDxe.efi' loaded at 2C7FB000 - Success
>>> FS0:\efi\boot\>
>>
>> At this point, the UEFI console is properly multiplexed to both serial
>> and HDMI+USB.
> 
> Neat.  /me makes a mental note that one can load drivers like this.
> Can this be automated to run on each boot?  With a startup.nsh script?

Yes, that's what I did:
<https://listman.redhat.com/archives/edk2-devel-archive/2022-April/047910.html>.

Even better would have been to copy the SerialDxe and TerminalDxe
drivers to the hard disk's EFI system partition, and then register them
as Driver#### / DriverOrder. I tried that with "bcfg driver addp", and
then validated with "bcfg driver dump -v" -- however, alas, this system
seems to ignore Driver* in platform BDS.

Regarding the startup.nsh script: the contents are like this:

load hd0c0b:\efi\boot\SerialDxe.efi
load hd0c0b:\efi\boot\TerminalDxe.efi

The "hd0c0b:" filesystem identifier is the "consistent naming" kind
(printed by "map -c"), as opposed to FS0: and friends. Without an FS
identifier, the script wouldn't work (the shell wouldn't know on what FS
to look for the pathname \efi\boot\SerialDxe.efi, because, at startup,
there would not be a "current" filesystem). And given that I had to
specify the filesystem, the "consistent" naming looked better. The UEFI
Shell spec explains what the consistent naming rules are -- in brief, as
long as I plug the stick in the same USB port, the FS identifier will work.

Thanks,
Laszlo