Re: [PATCH] scripts: check-function-names: propagate nm failures

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] scripts: check-function-names: buffering full vmlinux.o nm output in shell
--

commit 329efcc5adb68fe0ddbe8ccd5ee716b893c4188e
Author: Hamdy Abou El Anein <hamdy.aea@protonmail.com>

scripts: check-function-names: propagate nm failures

This commit captures the output of nm in a shell variable before piping it
to awk and grep so that check-function-names.sh exits with an error if nm
fails on the input object file.

> diff --git a/scripts/check-function-names.sh b/scripts/check-function-names.sh
> index 94883e690627d..f07a0e1cf3394 100755
> --- a/scripts/check-function-names.sh
> +++ b/scripts/check-function-names.sh
> @@ -13,7 +13,8 @@ if [ ! -f "$objfile" ]; then
>  	exit 1
>  fi
>  
> -bad_symbols=$(${NM:-nm} -p "$objfile" | awk '$2 ~ /^[TtWw]$/ {print $3}' |
> +symbols=$(${NM:-nm} -p "$objfile") || exit 1
> +bad_symbols=$(printf '%s\n' "$symbols" | awk '$2 ~ /^[TtWw]$/ {print $3}' |
>  	      grep -E '^(startup|exit|split|unlikely|hot|unknown)(\.|$)')

[Severity: Medium]
Does capturing the entire nm output into symbols and expanding it into
printf introduce a build-time performance and memory regression when linking
vmlinux.o?

In scripts/Makefile.vmlinux_o, rule_ld_vmlinux.o runs this script on
vmlinux.o in the serial path of the kernel link:

scripts/Makefile.vmlinux_o:
cmd_check_function_names = $(srctree)/scripts/check-function-names.sh $@

define rule_ld_vmlinux.o
	$(call cmd_and_savecmd,ld_vmlinux.o)
	$(call cmd,gen_objtooldep)
	$(call cmd,check_function_names)
endef

Because vmlinux.o contains the full built-in kernel symbol table (spanning
many megabytes and hundreds of thousands of symbols, especially on
allmodconfig or allyesconfig builds), previously nm -p streamed directly
into awk in parallel and command substitution only buffered the matching
bad_symbols output (which is empty in the normal case).

With this change, the shell waits for nm to finish while buffering the
entire multi-megabyte symbol table in the symbols variable, and then forks a
subshell that expands and copies the entire string as an argument to printf
before piping it to awk, working against the serial link optimization in
commit c2970d8b52d1 ("kbuild: do not sort nm output where the order is
irrelevant").

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261103105829.10511-1-info@daylightlinux.ch?part=1