Build failures with CC=arm-linux-gnueabihf-gcc-11

[Michael Olbrich]

On Thu, Nov 18, 2021 at 03:32:04PM +0000, Ian Abbott wrote:
> Hi,
> 
> My Debian "testing" build system updated the default version of
> arm-linux-gnueabihf-gcc from 10.x to 11.x.  Building barebox in thumb-2 mode
> for an ARMv7 (it's a SoCFPGA) with this compiler fails with assembler errors
> such as:
> 
>   CC      common/state/state_variables.o
> {standard input}: Assembler messages:
> {standard input}:1535: Error: selected processor does not support `rev
> r3,r3' in Thumb mode
> {standard input}:1576: Error: selected processor does not support `rev
> r3,r3' in Thumb mode
> 
> Building in verbose mode (V=1) shows that "make" has selected the compile
> flags `-march=armv5t -Wa,-march=armv7-a` instead of the usual
> `-march=armv7-a`.
> 
> Looking a little deeper, all of the following lines in "arch/arm/Makefile"
> are failing to select the first option and defaulting to the second option
> (if any):

It probably is added from here:

arch-$(CONFIG_CPU_32v7)		:=-D__LINUX_ARM_ARCH__=7 $(call cc-option,-march=armv7-a,-march=armv5t -Wa$(comma)-march=armv7-a)

The check for -march=armv7-a fails for some reason so the fallback is used.

I've seen this when CC is a wrapper and already adds -mcpu=... to the
commandline. In that case gcc now complains about the combination and the
check fails.

Michael

> arch-$(CONFIG_CPU_64v8)		:= -D__LINUX_ARM_ARCH__=8 $(call
> cc-option,-march=armv8-a)
> arch-$(CONFIG_CPU_32v7)		:=-D__LINUX_ARM_ARCH__=7 $(call
> cc-option,-march=armv7-a,-march=armv5t -Wa$(comma)-march=armv7-a)
> arch-$(CONFIG_CPU_32v6)		:=-D__LINUX_ARM_ARCH__=6 $(call
> cc-option,-march=armv6,-march=armv5t -Wa$(comma)-march=armv6)
> arch-$(CONFIG_CPU_32v5)		:=-D__LINUX_ARM_ARCH__=5 $(call
> cc-option,-march=armv5te,-march=armv4t)
> 
> The error from the compiler is:
> 
> cc1: error: ‘-mfloat-abi=hard’: selected architecture lacks an FPU
> 
> (Note: the error is only shown after tweaking the `try-run` command in
> "scripts/Kbuild.include" so that it does not send error output to
> "/dev/null".)
> 
> That error seems to be a change in behaviour between GCC 10.x and GCC 11.x.
> 
> At the point in the Makefile that the error occurs, the `-msoft-float`
> option has not been added to `KBUILD_CPPFLAGS` yet. That is done further
> down the Makefile except for `CONFIG_CPU_V8`.  I guess it needs to be added
> somewhere before problematic lines in the Makefile.  I'm not sure if
> `-msoft-float` is also needed for 64-bit ARMv8, but the error suggests it is
> required there too.

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |