Re: Unsigned widening casts of binary "not" operations..

[Geert Uytterhoeven]

On Tue, Apr 23, 2013 at 7:37 PM, David Miller <davem@xxxxxxxxxxxxx> wrote:
> From: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
> Date: Tue, 23 Apr 2013 08:42:49 -0700
>
>> An explicit cast fixes it, and shows that you were aware of the issue:
>>
>>    foo &= ~(foo_t)bar;
>>
>> and gcc will generate the right logic. Of course, casts then have
>> their own problems, which your thing avoids (as would just having a
>> "andn" operation in C)
>
> I just want to mention that this is dangerous in different ways, we
> just recently got a patch in the networking that removed such a cast.
> The problem is when the cast narrows, f.e.:
>
>         ~(u8)0
>
> doesn't do what you think it does.  That doesn't evaluate to 0xff.

This is the definition of MAC802154_CHAN_NONE?

We _should_ have noticed this earlier, as old gcc (e.g. 4.1.2) emits a
warning when comparing it to a u8:

net/mac802154/monitor.c: In function âmac802154_monitor_xmitâ:
net/mac802154/monitor.c:49: warning: comparison is always false due to
limited range of data type
net/mac802154/wpan.c: In function âmac802154_wpan_xmitâ:
net/mac802154/wpan.c:323: warning: comparison is always false due to
limited range of data type

Interestingly, none of this is seen in the build logs of the linux-next build
service, which uses gcc 4.2.3, 4.2.4, 4.5.1, and 4.6...

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/