Re: [PATCH v2 7/9] Make EFI_LOADER depend on DM and OF_CONTROL

[Tom Rini <trini@konsulko.com>]

[-- Attachment #1: Type: text/plain, Size: 3027 bytes --]

On Thu, Jul 29, 2021 at 01:45:49AM +0200, Heinrich Schuchardt wrote:
> 
> 
> On 7/27/21 12:07 AM, Tom Rini wrote:
> > On Fri, Jul 02, 2021 at 12:36:18PM -0600, Simon Glass wrote:
> > 
> > > This feature should never have been made available when driver model
> > > or devicetree are disabled. Add these as conditions, so that we don't
> > > create even more barriers to migration.
> > > 
> > > Add a note about the substantial size increment associated with this
> > > option.
> > > 
> > > Signed-off-by: Simon Glass <sjg@chromium.org>
> > > ---
> > > 
> > > Changes in v2:
> > > - Split out new patch to make EFI_LOADER depend on DM and OF_CONTROL
> > > - Note the approximate size of this feature in the help
> > > 
> > >   lib/efi_loader/Kconfig | 4 +++-
> > >   1 file changed, 3 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/lib/efi_loader/Kconfig b/lib/efi_loader/Kconfig
> > > index 6242caceb7f..466abfed300 100644
> > > --- a/lib/efi_loader/Kconfig
> > > +++ b/lib/efi_loader/Kconfig
> > > @@ -1,6 +1,6 @@
> > >   config EFI_LOADER
> > >   	bool "Support running UEFI applications"
> > > -	depends on OF_LIBFDT && ( \
> > > +	depends on OF_LIBFDT && DM && OF_CONTROL && ( \
> 
> Didn't Tom eliminate all boards without CONFIG_DM? Shouldn't we get rid
> of this symbol?

No, but I did send out a message about that today as we're very close.
Much closer than I had expected us to be.

> Are there boards using DM and not OF_CONTROL or OF_CONTROL and not DM?

Yes, a few.  It's vexpress_aemv8a_semi, warp (fixed by
https://patchwork.ozlabs.org/project/uboot/patch/20210402180552.1075997-2-pbrobinson@gmail.com/
so false positive), cm_t335, devkit8000, armadillo-800eva, kzm9g and sniper.

> Why are these separate symbols? Isn't this symbol to be eliminated, too?

Simon?

> lib/efi_loader/efi_disk.c is the only place where we maintain duplicate
> code for DM and non-DM. A dependency on CONFIG_BLK (which itself depends
> on CONFIG_DM) would make more sense to me. But only in a patch
> eliminating the non-BLK code.

I just know that off-hand, partition + disk + block has some corner
case, but maybe that corner case is unintentional in terms of usage
today.

> > >   		ARM && (SYS_CPU = arm1136 || \
> > >   			SYS_CPU = arm1176 || \
> > >   			SYS_CPU = armv7   || \
> > > @@ -25,6 +25,8 @@ config EFI_LOADER
> > >   	  will expose the UEFI API to a loaded application, enabling it to
> > >   	  reuse U-Boot's device drivers.
> > > 
> > > +	  For ARM 32-bit, this adds about 90KB to the size of U-Boot.
> > > +
> 
> There is no unit ISO prefix K. Do you mean KiB?
> 
> 90 KiB may be the value today. Will you update it regularly? Otherwise
> don't put a number here.
> 
> I can't see that the effect on size is truly architecture specific. Why
> do you refer to 32bit ARM?
> 
> Such a comment would better fit into a documentation chapter on
> downsizing U-Boot.

Yes, we should probably drop that specific note.

-- 
Tom

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 659 bytes --]