svn commit: r241373 - head/lib/libc/stdlib

[Andrey Chernov]

On 09.10.2012 21:47, Eitan Adler wrote:
> On 9 October 2012 12:33, Andrey Chernov <ache at freebsd.org> wrote:
>> Do you check assembler output for _both_ cases?
> 
> Yes.
...
> http://blog.eitanadler.com/2012/10/reduced-entropy-in-rand-and-random.html

At this URL I see only already known buggy assembler without 'volatile'
keyword (which is fixed by adding 'volatile' in srandomdev()). As I
already mention, adding 'volatile' helps any gcc and clang finally
generated assembler code (checked by cc -S ...).

What happens with LLVM intermediate code, I mean mentioned by David
call void @srand(i32 undef)
is a big question and perhaps clang bug. Please note that we use
'volatile' a lot in the kernel, just 'grep -r volatile /sys'. Some of
that potentially can hit the same (probably) bug. And, in case it is the
bug, it should be fixed in clang.

> volatile is still undefined: see 5.1.2.2.3 and 6.7.2.4 of ISO9899

I don't have ISO9899 nearby, could you directly quote mentioned
sections, please? Do you against 'volatile' usage at all? It seems whole
kernel (see above) contradicts with such point of view.