Re: [PATCH] kernel.h: Disable -Wreturn-stack-address for _THIS_IP_

From: Nick Desaulniers
Date: Mon Jul 30 2018 - 17:15:37 EST


On Mon, Jul 30, 2018 at 1:01 PM Nathan Chancellor
<natechancellor@xxxxxxxxx> wrote:
>
> On Mon, Jul 30, 2018 at 12:48:06PM -0700, Nick Desaulniers wrote:
> > On Mon, Jul 30, 2018 at 11:39 AM Nathan Chancellor
> > <natechancellor@xxxxxxxxx> wrote:
> > >
> > > On Mon, Jul 30, 2018 at 10:06:20AM -0700, Nick Desaulniers wrote:
> > > > Starting with Clang-7.0, _THIS_IP_ generates -Wreturn-stack-address
> > > > warnings for almost every translation unit. In general, I'd prefer to
> > > > leave this on (returning the address of a stack allocated variable is in
> > > > general a bad idea) and disable it only at whitelisted call sites.
> > > >
> > > > We can't do something like:
> > > > #pragma clang diagnostic push
> > > > #pragma clang diagnostic ignored "-Wreturn-stack-address"
> > > > <code>
> > > > #pragma clang diagnostic pop
> > > >
> > > > in a GNU Statement Expression or macro, hence we use _Pragma, which is
> > > > its raison d'Ãtre: https://gcc.gnu.org/onlinedocs/cpp/Pragmas.html
> > > >
> > > > Cc: stable@xxxxxxxxxxxxxxx # 4.17, 4.14, 4.9, 4.4
> > > > Signed-off-by: Nick Desaulniers <ndesaulniers@xxxxxxxxxx>
> > > > ---
> > > > include/linux/kernel.h | 10 +++++++++-
> > > > 1 file changed, 9 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/include/linux/kernel.h b/include/linux/kernel.h
> > > > index 941dc0a5a877..5906f5727f90 100644
> > > > --- a/include/linux/kernel.h
> > > > +++ b/include/linux/kernel.h
> > > > @@ -168,7 +168,15 @@
> > > >
> > > >
> > > > #define _RET_IP_ (unsigned long)__builtin_return_address(0)
> > > > -#define _THIS_IP_ ({ __label__ __here; __here: (unsigned long)&&__here; })
> > > > +#define _THIS_IP_ ( \
> > > > +{ \
> > > > + _Pragma("clang diagnostic push") \
> > > > + _Pragma("clang diagnostic ignored \"-Wreturn-stack-address\"") \
> > > > + __label__ __here; \
> > > > +__here: (unsigned long)&&__here; \
> > > > + _Pragma("clang diagnostic pop") \
> > > > +} \
> > > > +)
> > > >
> > > > #ifdef CONFIG_LBDAF
> > > > # include <asm/div64.h>
> > > > --
> > > > 2.18.0.233.g985f88cf7e-goog
> > > >
> > >
> > > This generates a ton of warnings with GCC:
> > >
> > > In file included from ./include/linux/spinlock.h:58,
> > > from ./include/linux/mmzone.h:8,
> > > from ./include/linux/gfp.h:6,
> > > from ./include/linux/slab.h:15,
> > > from ./include/linux/crypto.h:24,
> > > from arch/x86/kernel/asm-offsets.c:9:
> > > ./include/linux/bottom_half.h: In function âlocal_bh_disableâ:
> > > ./include/linux/bottom_half.h:19: error: ignoring #pragma clang diagnostic [-Werror=unknown-pragmas]
> > > __local_bh_disable_ip(_THIS_IP_, SOFTIRQ_DISABLE_OFFSET);
> > >
> > > ./include/linux/bottom_half.h:19: error: ignoring #pragma clang diagnostic [-Werror=unknown-pragmas]
> > > ./include/linux/bottom_half.h:19: error: ignoring #pragma clang diagnostic [-Werror=unknown-pragmas]
> > > ./include/linux/bottom_half.h: In function âlocal_bh_enableâ:
> > > ./include/linux/bottom_half.h:32: error: ignoring #pragma clang diagnostic [-Werror=unknown-pragmas]
> > > __local_bh_enable_ip(_THIS_IP_, SOFTIRQ_DISABLE_OFFSET);
> > >
> > > ./include/linux/bottom_half.h:32: error: ignoring #pragma clang diagnostic [-Werror=unknown-pragmas]
> > > ./include/linux/bottom_half.h:32: error: ignoring #pragma clang diagnostic [-Werror=unknown-pragmas]
> > > cc1: all warnings being treated as errors
> > > make[1]: *** [Kbuild:56: arch/x86/kernel/asm-offsets.s] Error 1
> > > make: *** [Makefile:1081: prepare0] Error 2
> >
> > Ah, good catch. I thought I had tested locally with gcc-8, but it
> > seems that it was likely only godbolt. I do see the errors locally
> > when building with gcc-8.
> >
> > >
> > > A proper solution is probably going to involve what was done for the
> > > -Wattribute-alias warnings from GCC 8 in commits 8793bb7f4a9d ("kbuild:
> > > add macro for controlling warnings to linux/compiler.h") and
> > > bee20031772a ("disable -Wattribute-alias warning for SYSCALL_DEFINEx()")
> > >
> > > I'll take a look at it in a bit unless someone beats me to it.
> >
> > I'll fix this up, triple check with gcc-8, and attribute you in the
> > Suggested-by tag. Thank you for verifying.
> > --
> > Thanks,
> > ~Nick Desaulniers
>
> I forgot to mention there is a similar macro that will need this same
> fix in arch/arm64/include/asm/processor.h (current_text_addr), if you
> want to tackle it in v2. Do CC me on the next series so that I can test,
> thank you for doing this!

!!!

$ grep -R current_text_addr

shows many definitions of current_text_addr per arch/ that are all
essentially the same, and very very few uses of it. I prefer to
consolidate current_text_addr() in a follow up series, since a common
definition with per-arch overrides will have to get a lot of sign
offs.

Though, for arm64, a simple:
-#define current_text_addr() ({ __label__ _l; _l: &&_l;})
+#define current_text_addr() _THIS_IP_

does the trick, it would be better to have a single shared definition.

--
Thanks,
~Nick Desaulniers