Re: [PATCH] signal: fix building with clang

From: Oleg Nesterov
Date: Thu Mar 07 2019 - 10:28:14 EST


On 03/07, Arnd Bergmann wrote:
>
> clang warns about the sigset_t manipulating functions (sigaddset, sigdelset,
> sigisemptyset, ...) because it performs semantic analysis before discarding
> dead code, unlike gcc that does this in the reverse order.
>
> The result is a long list of warnings like:
>
> In file included from arch/arm64/include/asm/ftrace.h:21:
> include/linux/compat.h:489:10: error: array index 3 is past the end of the array (which contains 2 elements) [-Werror,-Warray-bounds]
> case 2: v.sig[3] = (set->sig[1] >> 32); v.sig[2] = set->sig[1];

stupid question... I have no idea if this can work or not, but may be we can just do

--- x/Makefile
+++ x/Makefile
@@ -701,6 +701,7 @@ KBUILD_CPPFLAGS += $(call cc-option,-Qun
KBUILD_CFLAGS += $(call cc-disable-warning, format-invalid-specifier)
KBUILD_CFLAGS += $(call cc-disable-warning, gnu)
KBUILD_CFLAGS += $(call cc-disable-warning, address-of-packed-member)
+KBUILD_CFLAGS += $(call cc-disable-warning, array-bounds)
# Quiet clang warning: comparison of unsigned expression < 0 is always false
KBUILD_CFLAGS += $(call cc-disable-warning, tautological-compare)
# CLANG uses a _MergedGlobals as optimization, but this breaks modpost, as the

?

> As a (rather ugly) workaround,

Yes :/

But I am not going to argue, just a couple of questions.

> I turn the nice switch()/case statements
> into preprocessor conditionals, and where that is not possible, use the
> '%' operator

I can't say what looks worse... to me it would be either use ifdef's or %'s
everywhere in signal.h, with this patch the code doesn't look consistent.
But I won't insist.


> static inline int sigisemptyset(sigset_t *set)
> {
> - switch (_NSIG_WORDS) {
> - case 4:
> - return (set->sig[3] | set->sig[2] |
> - set->sig[1] | set->sig[0]) == 0;
> - case 2:
> - return (set->sig[1] | set->sig[0]) == 0;
> - case 1:
> - return set->sig[0] == 0;
> - default:
> - BUILD_BUG();
> - return 0;
> - }
> +#if _NSIG_WORDS == 4
> + return (set->sig[3] | set->sig[2] |
> + set->sig[1] | set->sig[0]) == 0;
> +#elif _NSIG_WORDS == 2
> + return (set->sig[1] | set->sig[0]) == 0;
> +#elif _NSIG_WORDS == 1
> + return set->sig[0] == 0;
> +#else
> + BUILD_BUG();
> +#endif
> }

Or perhaps we can simply rewrite this and other helpers?

I don't think that, say,

static inline int sigisemptyset(sigset_t *set)
{
for (i = 0; i < ARRAY_SIZE(set->sig); ++i)
set->sig[i] = 0;
}

will make asm worse...

Oleg.