From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
        id S1756210AbdKCSHI (ORCPT <rfc822;w@1wt.eu>);
        Fri, 3 Nov 2017 14:07:08 -0400
Received: from mail-wm0-f47.google.com ([74.125.82.47]:55208 "EHLO
        mail-wm0-f47.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
        with ESMTP id S1756087AbdKCSHG (ORCPT
        <rfc822;linux-kernel@vger.kernel.org>);
        Fri, 3 Nov 2017 14:07:06 -0400
X-Google-Smtp-Source: ABhQp+ScV698Up+UADzrnlzGbif1fs1n5oblRWv9pnrYYFe077ua0Ptlsqc0w++o8DlW/cfNh6OCOKw5KWWvqAihvqg=
MIME-Version: 1.0
In-Reply-To: <20171103175156.f44o7kuxgzym3pbt@lakrids.cambridge.arm.com>
References: <20171103171203.107569-1-samitolvanen@google.com> <20171103175156.f44o7kuxgzym3pbt@lakrids.cambridge.arm.com>
From: Nick Desaulniers <ndesaulniers@google.com>
Date: Fri, 3 Nov 2017 11:07:04 -0700
Message-ID: <CAKwvOdkTSXuxiFVUPXmefnDM6uV3BLDEJ+suO0BgfZxC6+Srbw@mail.gmail.com>
Subject: Re: [PATCH 00/15] Add support for clang LTO
To: Mark Rutland <mark.rutland@arm.com>
Cc: Sami Tolvanen <samitolvanen@google.com>,
        LKML <linux-kernel@vger.kernel.org>,
        linux-arm-kernel@lists.infradead.org,
        Matthias Kaehlcke <mka@chromium.org>,
        Kees Cook <keescook@chromium.org>,
        Greg Hackmann <ghackmann@google.com>, George Burgess <gbiv@google.com>,
        Stephen Hines <srhines@google.com>,
        Pirama Arumuga Nainar <pirama@google.com>,
        Manoj Gupta <manojgupta@google.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org

On Fri, Nov 3, 2017 at 10:51 AM, Mark Rutland <mark.rutland@arm.com> wrote:
> I had to create an aarch64-linux-gnu-clang wrapper, too. I'm not sure if
> there's build system help to avoid needing that?

Gah!  So a BIG difference with Clang vs GCC for cross compiling is
that Clang by default ships with all backends enabled, and uses a
`-target <triple>` CFLAG to determine the arch to cross compile for,
while GCC is configured at compile time to support one back end, IIUC.
Clang _can_ be built with all other back ends disabled but one, but I
have a problem with this approach:  distributions of Clang are rarely
configured this way (all backends enabled by default), and it seems we
just found this past week that Clang if configured with just one
backend, it will silently ignore -target flags for other backends and
generate code for the configured backend (this led to:
https://lkml.org/lkml/2017/11/2/892).  I consider this a bug in Clang,
so I just filed: https://bugs.llvm.org/show_bug.cgi?id=35196

>   CC      arch/arm64/crypto/aes-ce-cipher.o
> fatal error: error in backend: Do not know how to split the result of this operator!

Yep, we've seen this.  It was the FINAL bug in Clang for compiling the
kernel for arm64.  It was recently fixed upstream in llvm by
gbiv@google.com, but so recent that you'll need to either compiler
Clang from source from ToT or work around it like we have in Android
with:
https://android.googlesource.com/kernel/msm/+/9093342a0186dad05095b70f1806938310ace6e7

aka patch 7 in this series.

From mboxrd@z Thu Jan  1 00:00:00 1970
From: ndesaulniers@google.com (Nick Desaulniers)
Date: Fri, 3 Nov 2017 11:07:04 -0700
Subject: [PATCH 00/15] Add support for clang LTO
In-Reply-To: <20171103175156.f44o7kuxgzym3pbt@lakrids.cambridge.arm.com>
References: <20171103171203.107569-1-samitolvanen@google.com>
 <20171103175156.f44o7kuxgzym3pbt@lakrids.cambridge.arm.com>
Message-ID: <CAKwvOdkTSXuxiFVUPXmefnDM6uV3BLDEJ+suO0BgfZxC6+Srbw@mail.gmail.com>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

On Fri, Nov 3, 2017 at 10:51 AM, Mark Rutland <mark.rutland@arm.com> wrote:
> I had to create an aarch64-linux-gnu-clang wrapper, too. I'm not sure if
> there's build system help to avoid needing that?

Gah!  So a BIG difference with Clang vs GCC for cross compiling is
that Clang by default ships with all backends enabled, and uses a
`-target <triple>` CFLAG to determine the arch to cross compile for,
while GCC is configured at compile time to support one back end, IIUC.
Clang _can_ be built with all other back ends disabled but one, but I
have a problem with this approach:  distributions of Clang are rarely
configured this way (all backends enabled by default), and it seems we
just found this past week that Clang if configured with just one
backend, it will silently ignore -target flags for other backends and
generate code for the configured backend (this led to:
https://lkml.org/lkml/2017/11/2/892).  I consider this a bug in Clang,
so I just filed: https://bugs.llvm.org/show_bug.cgi?id=35196

>   CC      arch/arm64/crypto/aes-ce-cipher.o
> fatal error: error in backend: Do not know how to split the result of this operator!

Yep, we've seen this.  It was the FINAL bug in Clang for compiling the
kernel for arm64.  It was recently fixed upstream in llvm by
gbiv at google.com, but so recent that you'll need to either compiler
Clang from source from ToT or work around it like we have in Android
with:
https://android.googlesource.com/kernel/msm/+/9093342a0186dad05095b70f1806938310ace6e7

aka patch 7 in this series.