[PATCH v2 01/16] Makefile: detect HOST_ARCH properly when CROSS_COMPILE is multi-word

[Simon Glass]

Hi Jerome,

On Mon, 2 Sept 2024 at 09:21, Jerome Forissier
<jerome.forissier at linaro.org> wrote:
>
>
>
> On 9/1/24 22:09, Simon Glass wrote:
> > Hi Jerome,
> >
> > On Fri, 30 Aug 2024 at 04:17, Jerome Forissier
> > <jerome.forissier at linaro.org> wrote:
> >>
> >> When CROSS_COMPILE contains multiple words, HOST_ARCH is not properly
> >> detected and the sandbox build fail. It typically happens when using
> >> ccache. For example:
> >>
> >>  $ make sandbox_defconfig
> >>  $ make CROSS_COMPILE="ccache x86_64-linux-gnu-" \
> >>         CC="ccache x86_64-linux-gnu-gcc"
> >>  [...]
> >>  In file included from boot/bootmeth_efi.c:16:
> >>  include/efi_default_filename.h:33:2: error: #error Unsupported UEFI architecture
> >>     33 | #error Unsupported UEFI architecture
> >>        |  ^~~~~
> >>
> >> A similar error occurs when the build is done via buildman and
> >> ~/.buildman contains:
> >>
> >>   [toolchain-wrapper]
> >>   wrapper = ccache
> >>
> >> Fix the issue by considering only the last word in $(CROSS_COMPILE).
> >>
> >> Signed-off-by: Jerome Forissier <jerome.forissier at linaro.org>
> >> Reviewed-by: Peter Robinson <pbrobinson at gmail.com>
> >> Reviewed-by: Ilias Apalodimas <ilias.apalodimas at linaro.org>
> >> ---
> >>  Makefile | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > This patch[1] seems to fix the problem you mention, for the CC part at least.
>
> No, it fails with buildman:
>
> $ git log --oneline -n 1
> 454b2b2e6ed (HEAD) efi: Use the same filename for all sandbox builds
> jerome at builder:~/work/u-boot ((454b2b2e6ed...))$ ./tools/buildman/buildman sandbox_noinst
> Building current source for 1 boards (1 thread, 24 jobs per thread)
>    sandbox:  +   sandbox_noinst
> +arch/sandbox/lib/reloc_sandbox_efi.c:10:15: error: operator '==' has no left operand
> +   10 | #if HOST_ARCH == HOST_ARCH_X86_64
> +      |               ^~
> [...]

Hmm, I can't really dig into that...

>
> > For the CROSS_COMPILE bit, why is it valid to use a cross compiler on sandbox?
>
> Why not? "CROSS_COMPILE=x86_64-linux-gnu-" should result in x86_64-linux-gnu-gcc
> being used which is equivalent to gcc on a x86_64 system.

Because it isn't a cross compiler, is it?

Why is CC= not enough? After all, sandbox never cross-compiles.

>
> >>
> >> diff --git a/Makefile b/Makefile
> >> index f90e48f58a..dc7bdd7942 100644
> >> --- a/Makefile
> >> +++ b/Makefile
> >> @@ -21,7 +21,7 @@ include include/host_arch.h
> >>  ifeq ("", "$(CROSS_COMPILE)")
> >>    MK_ARCH="${shell uname -m}"
> >>  else
> >> -  MK_ARCH="${shell echo $(CROSS_COMPILE) | sed -n 's/^[[:space:]]*\([^\/]*\/\)*\([^-]*\)-[^[:space:]]*/\2/p'}"
> >> +  MK_ARCH="${shell echo ${lastword $(CROSS_COMPILE)} | sed -n 's/^[[:space:]]*\([^\/]*\/\)*\([^-]*\)-[^[:space:]]*/\2/p'}"
> >>  endif
> >>  unexport HOST_ARCH
> >>  ifeq ("x86_64", $(MK_ARCH))
> >> --
> >> 2.40.1
> >>
> >
> > Regards,
> > Simon
> >
> > [1] https://patchwork.ozlabs.org/project/uboot/patch/20240815202424.766778-9-sjg@chromium.org/

Regards,
Simon