Re: [PATCH] kbuild: Enable -Wsometimes-uninitialized

[Masahiro Yamada]

On Wed, May 1, 2019 at 5:54 AM Nathan Chancellor
<natechancellor@xxxxxxxxx> wrote:
>
> On Tue, Apr 30, 2019 at 11:46:44AM +0200, Arnd Bergmann wrote:
> > On Tue, Apr 30, 2019 at 11:33 AM Nathan Chancellor
> > <natechancellor@xxxxxxxxx> wrote:
> > > On Tue, Apr 30, 2019 at 09:16:50AM +0200, Arnd Bergmann wrote:
> > > > On Tue, Apr 30, 2019 at 3:01 AM Nathan Chancellor
> > > > <natechancellor@xxxxxxxxx> wrote:
> > > > >
> > > > > This is Clang's version of GCC's -Wmaybe-uninitialized. Up to this
> > > > > point, it has not been used because -Wuninitialized has been disabled,
> > > > > which also turns off -Wsometimes-uninitialized, meaning that we miss out
> > > > > on finding some bugs [1]. In my experience, it appears to be more
> > > > > accurate than GCC and catch some things that GCC can't.
> > > > >
> > > > > All of these warnings have now been fixed in -next across arm, arm64,
> > > > > and x86_64 defconfig/allyesconfig so this should be enabled for everyone
> > > > > to prevent more from easily creeping in.
> > > > >
> > > > > As of next-20190429:
> > > > >
> > > > > $ git log --oneline --grep="sometimes-uninitialized" | wc -l
> > > > > 45
> > > > >
> > > > > [1]: https://lore.kernel.org/lkml/86649ee4-9794-77a3-502c-f4cd10019c36@xxxxxx/
> > > > >
> > > > > Link: https://github.com/ClangBuiltLinux/linux/issues/381
> > > > > Signed-off-by: Nathan Chancellor <natechancellor@xxxxxxxxx>
> > > > > ---
> > > > >
> > > > > Masahiro, I am not sure how you want to handle merging this with regards
> > > > > to all of the patches floating around in -next but I wanted to send this
> > > > > out to let everyone know this is ready to be turned on.
> > > > >
> > > > > Arnd, are there many remaning -Wsometimes-uninitialized warnings in
> > > > > randconfigs?
> > > >
> > > > No, I don't see any with the patches that I submitted. I haven't checked
> > > > if there are any that still need to get merged into linux-next though.
> > > >
> > > > > diff --git a/scripts/Makefile.extrawarn b/scripts/Makefile.extrawarn
> > > > > index 768306add591..f4332981ea85 100644
> > > > > --- a/scripts/Makefile.extrawarn
> > > > > +++ b/scripts/Makefile.extrawarn
> > > > > @@ -72,5 +72,6 @@ KBUILD_CFLAGS += $(call cc-disable-warning, format)
> > > > >  KBUILD_CFLAGS += $(call cc-disable-warning, sign-compare)
> > > > >  KBUILD_CFLAGS += $(call cc-disable-warning, format-zero-length)
> > > > >  KBUILD_CFLAGS += $(call cc-disable-warning, uninitialized)
> > > > > +KBUILD_CFLAGS += $(call cc-option, -Wsometimes-uninitialized)
> > > > >  endif
> > > > >  endif
> > > >
> > > > This doesn't look right. Shouldn't you remove the line that turns off
> > > > -Wuninitilized
> > > > instead of adding only -Wsometimes-uninitialized?
> > >
> > > Well, there are still some outstanding issues with -Wuninitialized
> > > right? Like with DECLARE_WAIT_QUEUE_HEAD_ONSTACK? I'd rather not
> > > add warnings to the build but if you feel strongly, we could turn it on
> > > then fix them after.
> >
> > Ah, I thought they were all fixed, as I don't see any remaining warnings
> > in my tree. It seems that I never send this workaround for
> > DECLARE_WAIT_QUEUE_HEAD_ONSTACK:
> >
> > diff --git a/include/linux/wait.h b/include/linux/wait.h
> > index 5f3efabc36f4..cbe1ea0fce84 100644
> > --- a/include/linux/wait.h
> > +++ b/include/linux/wait.h
> > @@ -68,8 +68,15 @@ extern void __init_waitqueue_head(struct
> > wait_queue_head *wq_head, const char *n
> >         } while (0)
> >
> >  #ifdef CONFIG_LOCKDEP
> > -# define __WAIT_QUEUE_HEAD_INIT_ONSTACK(name) \
> > -       ({ init_waitqueue_head(&name); name; })
> > +# define __WAIT_QUEUE_HEAD_INIT_ONSTACK(name) {
> >                  \
> > +       .lock           = __SPIN_LOCK_UNLOCKED(name.lock),
> >          \
> > +       .head           = ({
> >          \
> > +               static struct lock_class_key __key;
> >          \
> > +               lockdep_set_class_and_name(&(name).lock, &__key, #
> > name);       \
> > +               (struct list_head){ &(name).head, &(name).head };
> >          \
> > +       }),
> >          \
> > +}
> > +
> >  # define DECLARE_WAIT_QUEUE_HEAD_ONSTACK(name) \
> >         struct wait_queue_head name = __WAIT_QUEUE_HEAD_INIT_ONSTACK(name)
> >  #else
> >
> > Are there any others you see?
> >
> >       Arnd
>
> After applying that diff by hand and another patch that was accepted but
> not in -next (https://lore.kernel.org/lkml/20190325125147.1436083-1-arnd@xxxxxxxx/),
> I see three other warnings:
>
> arm, arm64, and x86_64 allyesconfig:
>
> drivers/net/wireless/rsi/rsi_91x_sdio.c:940:43: error: variable 'data'
> is uninitialized when used here [-Werror,-Wuninitialized]
>
> https://github.com/ClangBuiltLinux/linux/issues/464
>
> x86 allyesconfig:
>
> mm/kasan/common.c:490:40: error: variable 'tag' is uninitialized when
> used here [-Werror,-Wuninitialized]
>
> https://github.com/ClangBuiltLinux/linux/issues/465
>
> drivers/misc/sgi-xp/xpc_partition.c:73:14: error: variable 'buf' is
> uninitialized when used within its own initialization
> [-Werror,-Wuninitialized]
>
> https://github.com/ClangBuiltLinux/linux/issues/466
>
> These shouldn't be difficult to fix then we can turn both uninitialized
> warnings.
>
> Please let me know if you have any input with regards to fixing them, I
> am going to try to look at them later today.
>
> Nathan


I will postpone this patch.


It looks like we agreed to fix warnings first,
then remove $(call cc-disable-warning, uninitialized).





-- 
Best Regards
Masahiro Yamada