From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-db5eur01on0108.outbound.protection.outlook.com ([104.47.2.108]:22782 "EHLO EUR01-DB5-obe.outbound.protection.outlook.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S933320AbcKWIZ2 (ORCPT ); Wed, 23 Nov 2016 03:25:28 -0500 Subject: Re: [PATCH] kasan: turn off -fsanitize-address-use-after-scope for now To: Arnd Bergmann References: <20161122225424.3739294-1-arnd@arndb.de> CC: =?UTF-8?Q?Martin_Li=c5=a1ka?= , , Michal Marek , Alexander Potapenko , Dmitry Vyukov , , , From: Andrey Ryabinin Message-ID: Date: Wed, 23 Nov 2016 11:25:49 +0300 MIME-Version: 1.0 In-Reply-To: <20161122225424.3739294-1-arnd@arndb.de> Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 8bit Sender: stable-owner@vger.kernel.org List-ID: On 11/23/2016 01:54 AM, Arnd Bergmann wrote: > In the upcoming gcc-7 release, the -fsanitize=kernel-address option > implies -fsanitize-address-use-after-scope, which relies on the > definition of two global functions, causing many link errors if they > are not defined, e.g: > > arch/x86/built-in.o: In function `x86_pmu_handle_irq': > (.text+0x88e6): undefined reference to `__asan_unpoison_stack_memory' > arch/x86/built-in.o: In function `x86_pmu_handle_irq': > (.text+0x8ad7): undefined reference to `__asan_poison_stack_memory' > kernel/built-in.o: In function `perf_tp_event': > (.text+0x225472): undefined reference to `__asan_unpoison_stack_memory' > kernel/built-in.o: In function `perf_tp_event': > (.text+0x22583a): undefined reference to `__asan_unpoison_stack_memory' > kernel/built-in.o: In function `perf_tp_event': > (.text+0x2258ae): undefined reference to `__asan_poison_stack_memory' > kernel/built-in.o: In function `perf_event_aux_event': > > I think we really want to define those two functions so we can make > use of a helpful feature, but as I have no idea what they are supposed > to do, I'd suggest to turn the option off on existing kernels to allow > building with gcc-7 and kasan. > > For some reason, the problem showed up in only a few randconfig builds, > but it is easy to reproduce using an x86-64 tinyconfig build with > CONFIG_KASAN=y. > > If we decide to take this approach, we probably want to do the same > change on all stable kernels that support KASAN, i.e. v4.0 or higher. > We have implementation for this feature: "[PATCH] kasan: support use-after-scope detection" - http://lkml.kernel.org/r/<1479226045-145148-1-git-send-email-dvyukov@google.com> and given how simple it is I'd suggest to just backport implementation. BTW, we also need "[PATCH] kasan: update kasan_global for gcc 7" - http://lkml.kernel.org/r/<1479219743-28682-1-git-send-email-dvyukov@google.com> in v4.0+ stable. > Link: https://gcc.gnu.org/viewcvs/gcc?limit_changes=0&view=revision&revision=241896 > Cc: Martin Liška > Cc: Andrey Ryabinin > Cc: stable@vger.kernel.org > Signed-off-by: Arnd Bergmann > --- > scripts/Makefile.kasan | 2 ++ > 1 file changed, 2 insertions(+) > > diff --git a/scripts/Makefile.kasan b/scripts/Makefile.kasan > index 37323b0df374..0e68fef09f76 100644 > --- a/scripts/Makefile.kasan > +++ b/scripts/Makefile.kasan > @@ -29,3 +29,5 @@ else > endif > endif > endif > + > +CFLAGS_KASAN += $(call cc-option, -fno-sanitize-address-use-after-scope) >