linux-kernel - Re: [PATCH] kbuild: add script check for cross compilation utilities

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20190511022458.GA7376@archlinux-i9>
Date:   Fri, 10 May 2019 19:24:58 -0700
From:   Nathan Chancellor <natechancellor@...il.com>
To:     Nick Desaulniers <ndesaulniers@...gle.com>
Cc:     yamada.masahiro@...ionext.com, clang-built-linux@...glegroups.com,
        Michal Marek <michal.lkml@...kovi.net>,
        linux-kbuild@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] kbuild: add script check for cross compilation utilities

Few comments below but nothing major, this seems to work fine as is.

On Thu, May 09, 2019 at 01:19:21PM -0700, 'Nick Desaulniers' via Clang Built Linux wrote:
> When cross compiling via setting CROSS_COMPILE, if the prefixed tools
> are not found, then the host utilities are often instead invoked, and
> produce often difficult to understand errors.  This is most commonly the
> case for developers new to cross compiling the kernel that have yet to
> install the proper cross compilation toolchain. Rather than charge
> headlong into a build that will fail obscurely, check that the tools
> exist before starting to compile, and fail with a friendly error
> message.

This part of the commit message makes it sound like this is a generic
problem when it is actually specific to clang. make will fail on its
own when building with gcc if CROSS_COMPILE is not properly set (since
gcc won't be found).

On a side note, seems kind of odd that clang falls back to the host
tools when a non-host --target argument is used... (how in the world is
that expected to work?)

> 
> Before:
> $ ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf- make CC=clang
> ...
> /usr/bin/as: unrecognized option '-EL'
> clang: error: assembler command failed with exit code 1 (use -v to see
> invocation)
> make[2]: *** [../scripts/Makefile.build:279: scripts/mod/empty.o] Error 1
> make[2]: *** Waiting for unfinished jobs....
> make[1]: *** [/linux/Makefile:1118:
> prepare0] Error 2
> make: *** [Makefile:179: sub-make] Error 2
> 
> After:
> $ ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf- make CC=clang
> $CROSS_COMPILE set to arm-linux-gnueabihf-, but unable to find
> arm-linux-gnueabihf-as.
> Makefile:522: recipe for target 'outputmakefile' failed
> make: *** [outputmakefile] Error 1
> 
> Signed-off-by: Nick Desaulniers <ndesaulniers@...gle.com>

Reviewed-by: Nathan Chancellor <natechancellor@...il.com
Tested-by: Nathan Chancellor <natechancellor@...il.com>

> ---
> Note: this is probably more generally useful, but after a few minutes
> wrestling with Make errors related to "recipe commences before first
> target" and "missing separator," I came to understand my hatred of GNU
> Make. Open to sugguestions for where better to invoke this from the top
> level Makefile.
> 
>  Makefile                      |  1 +
>  scripts/check_crosscompile.sh | 18 ++++++++++++++++++
>  2 files changed, 19 insertions(+)
>  create mode 100755 scripts/check_crosscompile.sh
> 
> diff --git a/Makefile b/Makefile
> index a61a95b6b38f..774339674b59 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -519,6 +519,7 @@ endif
>  
>  ifneq ($(shell $(CC) --version 2>&1 | head -n 1 | grep clang),)
>  ifneq ($(CROSS_COMPILE),)
> +	$(Q)$(CONFIG_SHELL) $(srctree)/scripts/check_crosscompile.sh
>  CLANG_FLAGS	:= --target=$(notdir $(CROSS_COMPILE:%-=%))
>  GCC_TOOLCHAIN_DIR := $(dir $(shell which $(CROSS_COMPILE)elfedit))
>  CLANG_FLAGS	+= --prefix=$(GCC_TOOLCHAIN_DIR)
> diff --git a/scripts/check_crosscompile.sh b/scripts/check_crosscompile.sh
> new file mode 100755
> index 000000000000..f4586fbfee18
> --- /dev/null
> +++ b/scripts/check_crosscompile.sh
> @@ -0,0 +1,18 @@
> +#!/bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# (c) 2019, Nick Desaulniers <ndesaulniers@...gle.com>

I think a space between the comment and function here would look nicer.

> +function check () {
> +  # Remove trailing commands, for example arch/arm/Makefile may add `-EL`.
> +  utility=$(echo ${1} | awk '{print $1;}')

Shellcheck mentions the ${1} should be quoted.

> +  command -v "${utility}" &> /dev/null
> +  if [[ $? != 0 ]]; then

This can be simplified into:

if ! command -v "${utility}" &> /dev/null; then

> +    echo "\$CROSS_COMPILE set to ${CROSS_COMPILE}," \
> +      "but unable to find ${utility}."
> +    exit 1
> +  fi
> +}

Maybe a space here and after utilities?

> +utilities=("${AS}" "${LD}" "${CC}" "${AR}" "${NM}" "${STRIP}" "${OBJCOPY}"
> +  "${OBJDUMP}")

I think this would look a little better with the "${OBJDUMP}" aligned to
the "${AS}" (and maybe split the lines to make them evenly align?)

Another note, this script could in theory be invoked via 'sh' if bash
doesn't exist on a system (see CONFIG_SHELL's definition), where only
POSIX compliant constructs should be used (so no arrays). I don't know
how often this occurs to matter (or if it does in this case) but worth
mentioning.

> +for utility in "${utilities[@]}"; do
> +  check "${utility}"
> +done
> -- 
> 2.21.0.1020.gf2820cf01a-goog