Re: [PATCH 0/2] selftests: harness: refactor __constructor_order

[Kees Cook <keescook@chromium.org>]

On Fri, May 17, 2024 at 08:45:04PM +0900, Masahiro Yamada wrote:
> 
> This series refactors __constructor_order because
> __constructor_order_last() is unneeded.
> 
> BTW, the comments in kselftest_harness.h was confusing to me.
> 
> As far as I tested, all arches executed constructors in the forward
> order.
> 
> [test code]
> 
>   #include <stdio.h>
> 
>   static int x;
> 
>   static void __attribute__((constructor)) increment(void)
>   {
>            x += 1;
>   }
> 
>   static void __attribute__((constructor)) multiply(void)
>   {
>           x *= 2;
>   }
> 
>   int main(void)
>   {
>           printf("foo = %d\n", x);
>           return 0;
>   }
> 
> It should print 2 for forward order systems, 1 for reverse order systems.
> 
> I executed it on some archtes by using QEMU. I always got 2.

IIRC, and it was a long time ago now, it was actually a difference
between libc implementations where I encountered the problem. Maybe
glibc vs Bionic?

-Kees

-- 
Kees Cook