[PATCH 09/16] efi_loader: imply FAT, FAT_WRITE

[Mark Kettenis]

> Date: Wed, 1 Apr 2020 09:31:03 +0900
> From: AKASHI Takahiro <takahiro.akashi at linaro.org>
> 
> On Tue, Mar 31, 2020 at 10:20:17AM +0200, Mark Kettenis wrote:
> > > Date: Tue, 31 Mar 2020 16:44:34 +0900
> > > From: AKASHI Takahiro <takahiro.akashi at linaro.org>
> > > 
> > > On Tue, Mar 31, 2020 at 08:44:02AM +0200, Heinrich Schuchardt wrote:
> > > > On  March 31, 2020, 5:28 a.m. UTC Takahiro Akashi wrote:
> > > > > On Fri, Mar 27, 2020 at 06:27:53AM +0100, Heinrich Schuchardt wrote:
> > > > > > The UEFI spec requires support for the FAT file system.
> > > > > >
> > > > > > Signed-off-by: Heinrich Schuchardt <xypron.glpk at gmx.de>
> > > > > > ---
> > > > > >  lib/efi_loader/Kconfig | 2 ++
> > > > > >  1 file changed, 2 insertions(+)
> > > > > >
> > > > > > diff --git a/lib/efi_loader/Kconfig b/lib/efi_loader/Kconfig
> > > > > > index 9890144d41..e10ca05549 100644
> > > > > > --- a/lib/efi_loader/Kconfig
> > > > > > +++ b/lib/efi_loader/Kconfig
> > > > > > @@ -15,6 +15,8 @@ config EFI_LOADER
> > > > > >  	select HAVE_BLOCK_DEVICE
> > > > > >  	select REGEX
> > > > > >  	imply CFB_CONSOLE_ANSI
> > > > > > +	imply FAT
> > > > > > +	imply FAT_WRITE
> > > > >
> > > > > Obviously, this *imply* doesn't enforce enabling FAT.
> > > > > If it is absolutely necessary, another measure should be taken.
> > > > 
> > > > If somebody wants to minimize the U-Boot size it might be necessary to
> > > > do without FAT_WRITE or FAT support.
> > > 
> > > If so, Get/SetVariable won't be supported even in boot time
> > > with your patch applied. It is not practical for almost all users.
> > 
> > I *strongly* disagree with that statement.  Most users don't care
> > about U-Boot providing a full EFI implementation.  They just want to
> > boot their OS.  The basic EFI support in U-Boot is good enough for
> > that and for OpenBSD and some Linux distros on arm/arm64 this is the
> > only bootpath that works.  If adding more code leads to board
> > maintainers disabling EFI support this isn't helpful.
> 
> Okay, so can you please describe the minimum set of functionality
> for you? Without that, the discussion will not be fair.

For OpenBSD/arm and OpenBSD/amr64 we adopted the UEFI interfaces in
u-boot early on, even before most Linux distros did.  As a result our
requirements are very minimal:

Our bootloader uses the folowing boot services:

- HandleProtocol()
- AllocatePages()
- FreePages()
- LocateHandle()
- LocateHandleBuffer()
- LocateProtocol()
- CreateEvent()
- SetTimer()
- WaitForEvent()
- CloseEvent()
- GetMemoryMap()
- SetWatchdogTimer()
- Exit()

runtime services:

- ResetSystem()

and protocols:

- EFI_LOADED_IMAGE_PROTOCOL
- EFI_DEVICE_PATH_PROTOCOL
- EFI_BLOCK_IO_PROTOCOL
- EFI_GRAPHICS_OUTPUT_PROTOCOL
- EFI_RNG_PROTOCOL
- EFI_SIMPLE_NETWORK_PROTOCOL
- EFI_PXE_BASE_CODE_PROTOCOL

Obviously the last four are only used if there is actual hardware to
support these protocols.  But I'd consider them non-optional if such
hardware exists.

Our kernel only uses a couple if runtime services:

- SetVirtualAddressMap()
- GetTime() (optional)
- SetTime() (optional)

Cheers,

Mark