On Sat, Nov 28, 2020 at 10:28:31AM -0800, Linus Torvalds wrote:
> On Fri, Nov 27, 2020 at 11:05 PM Masahiro Yamada <masahiroy@xxxxxxxxxx> wrote:
> >
> > As for the cc1plus cost, I got a similar result.
> >
> > Running scripts/gcc-plugin.sh directly
> > took me 0.5 sec, which is a fourth
> > of the allmodconfig run-time.
> >
> > Actually, I did not know this shell script
> > was so expensive to run...
>
> So it turns out that one reason it's so expensive to run is that it
> does a *lot* more than it claims to do.
>
> It says "we need a c++ compiler that supports the designated
> initializer GNU extension", but then it actually includes a header
> file from hell, rather than just test designated initializers.
>
> This patch makes the cc1plus overhead go down a lot. That said, I'm
> doubtful we really want gcc plugins at all, considering that the only
> real users have all apparently migrated to clang builtin functionality
> instead.
>
> Linus
> scripts/gcc-plugin.sh | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/scripts/gcc-plugin.sh b/scripts/gcc-plugin.sh
> index b79fd0bea838..59db87bff456 100755
> --- a/scripts/gcc-plugin.sh
> +++ b/scripts/gcc-plugin.sh
> @@ -8,8 +8,8 @@ srctree=$(dirname "$0")
> gccplugins_dir=$($* -print-file-name=plugin)
>
> # we need a c++ compiler that supports the designated initializer GNU extension
> +test -e "$gccplugins_dir/include/plugin-version.h" &&
> $HOSTCC -c -x c++ -std=gnu++98 - -fsyntax-only -I $srctree/gcc-plugins -I $gccplugins_dir/include 2>/dev/null <<EOF
> -#include "gcc-common.h"
> class test {
> public:
> int test;
I'm fine dropping this -- I think the need for that portion of the
script's test has evaporated as we've brought the minimum GCC version
up into the neighborhood of "modern".
As for dropping GCC plugins entirely, I'd prefer not -- the big hold-out
for the very paranoid system builders is the randstruct plugin (though
they tend to also use the entropy one too). Clang's version of randstruct
has not gotten unstuck yet.
--
Kees Cook
