[ELDK] [PATCH v2 2/2] META-ELDK: u-boot: unbreak target U-Boot build

[Marek Vasut]

On Tuesday, April 22, 2014 at 07:37:04 PM, Gerhard Sittig wrote:
> On Tue, 2014-04-22 at 18:52 +0200, Marek Vasut wrote:
> > On Tuesday, April 22, 2014 at 04:06:27 PM, Gerhard Sittig wrote:
> > > [ ... ]
> > > diff --git a/meta-eldk/recipes-bsp/uboot/u-boot_2014.01.bb
> > > b/meta-eldk/recipes-bsp/uboot/u-boot_2014.01.bb index
> > > 7010cd38347c..b16ceaf0ae59 100644
> > > --- a/meta-eldk/recipes-bsp/uboot/u-boot_2014.01.bb
> > > +++ b/meta-eldk/recipes-bsp/uboot/u-boot_2014.01.bb
> > > @@ -28,6 +28,22 @@ SRC_URI_append_m53evk = " \
> > > 
> > >  			file://0003-ARM-m53evk-Update-default-environment.patch
> > 
> > \
> > 
> > >  			"
> > > 
> > > +# bootloader code might depend on OpenSSL (depending on config
> > > options), +# image handling related user space tools unconditionally
> > > depend on OpenSSL +DEPENDS += " openssl openssl-native "
> > 
> > This is slightly weird, do you really need the non-native OpenSSL ?
> 
> I had assumed that specific configurations of the target
> bootloader do depend on crypto stuff.

Don't assume, test and verify :)

> But I'm ignorant about
> signed/verified/etc boots.  It appears I was wrong.  Will remove
> the non-native openssl dependency.

The tools/mxsimage* depend on native version of openssl, but we do NOT link 
openssl into U-Boot to my knowledge.

> > > +# provide additional make parameters to help build the user space
> > > tools; +# note that the native.bbclass approach does not work here,
> > > the phrase +# HOSTCC="${CCACHE}${HOST_PREFIX}gcc ${HOST_CC_ARCH}"
> > > results in target specs; +# note as well that U-Boot v2014.01 requires
> > > the CFLAGS in the HOSTCC spec, +# passing separate HOST_CFLAGS only
> > > became operational in version v2014.04 +EXTRA_OEMAKE += ' \
> > > +	HOST_TOOLS_ALL=n \
> > > +	HOSTCC="gcc ${BUILD_CFLAGS}" \
> > > +	HOSTLD="ld" \
> > > +	HOSTLDFLAGS="${BUILD_LDFLAGS}" \
> > > +	HOSTSTRIP=true '
> > > +
> > 
> > Why this "HOST_TOOLS_ALL=n" ?
> 
> To reduce the number of make targets that get built.  The
> tools/Makefile will enable all those which depend on the target
> configuration anyway.  This HOST_TOOLS_ALL=n spec just does not
> enable by force what the configuration doesn't require.
> 
> Did I miss something?  Might as well remove this one spec which
> is not strictly related to the build un-break action.

You missed nothing, but then this improvement should also be sent to Yocto ML as 
a separate patch.

Best regards,
Marek Vasut