From: Will Deacon <will.deacon@arm.com> To: Sami Tolvanen <samitolvanen@google.com> Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>, Nick Desaulniers <ndesaulniers@google.com>, Peter Zijlstra <peterz@infradead.org>, Alex Matveev <alxmtvv@gmail.com>, Andi Kleen <ak@linux.intel.com>, Ard Biesheuvel <ard.biesheuvel@linaro.org>, Greg Hackmann <ghackmann@google.com>, Kees Cook <keescook@chromium.org>, linux-arm-kernel@lists.infradead.org, Linux Kbuild mailing list <linux-kbuild@vger.kernel.org>, LKML <linux-kernel@vger.kernel.org>, Mark Rutland <mark.rutland@arm.com>, Masahiro Yamada <yamada.masahiro@socionext.com>, Maxim Kuvyrkov <maxim.kuvyrkov@linaro.org>, Michal Marek <michal.lkml@markovi.net>, Yury Norov <ynorov@caviumnetworks.com>, Matthias Kaehlcke <mka@chromium.org>, Alexander Potapenko <glider@google.com>, Stephen Hines <srhines@google.com>, Pirama Arumuga Nainar <pirama@google.com>, Manoj Gupta <manojgupta@google.com> Subject: Re: [PATCH v2 18/18] arm64: select ARCH_SUPPORTS_LTO_CLANG Date: Mon, 20 Nov 2017 18:05:55 +0000 [thread overview] Message-ID: <20171120180554.GM32488@arm.com> (raw) In-Reply-To: <20171116201701.GA143965@samitolvanen.mtv.corp.google.com> Hi Sami, On Thu, Nov 16, 2017 at 12:17:01PM -0800, Sami Tolvanen wrote: > On Thu, Nov 16, 2017 at 11:13:07AM -0800, Paul E. McKenney wrote: > > Ah, if "this patch set" meant "adding LTO", I stand corrected and I > > apologize for my confusion. > > Again, I'm not proposing for LTO to be enabled by default. These patches > just make it possible to enable it. Are you saying the possibility > that a future compiler update breaks something is a blocker even for > experimental features? My opinion here is that it's a blocker for LTO on arm64, as the symptoms aren't different to building with a compiler that has subtle code generation issues. If you *really* want to support LTO, we could consider giving acquire semantics to READ_ONCE for LTO builds, but that means trading RCU read-side performance for whatever gains are provided by LTO. Please don't confuse this with a reluctance to support clang; I'm keen to see that supported, so let's focus on that for the moment (although I don't see a good reason to support old clang builds with known issues). > > I agree that we need LTO/PGO to be housebroken from an LKMM viewpoint > > before it is enabled. > > Can you elaborate what's needed from clang before this can move > forward? For example, if you have specific test cases in mind, we can > always work on including them in the LLVM test suite. This is a thorny issue, but RCU (specifically rcu_dereference but probably also some READ_ONCEs) relies on being able to utilise syntactic dependency chains to order local accesses to shared variables. Paul has spent a fair amount of time working to define these dependency chains here: https://wg21.link/P0190 Although the current direction of the C++ committee is to prefer that dependencies are explicitly "marked", this is not deemed to be acceptable for the kernel (in other words, everything is always considered "marked"). If a compiler is capable of breaking unmarked dependency chains, then we will run into subtle concurrency issues on arm64 because the hardware is also capable of reordering independent accesses. My understanding is that LTO makes these sort of optimisations far more likely because whole-program analysis can enable value prediction, which can convert address dependencies into control dependcies; the latter not ordering reads on arm64. Will
WARNING: multiple messages have this Message-ID (diff)
From: will.deacon@arm.com (Will Deacon) To: linux-arm-kernel@lists.infradead.org Subject: [PATCH v2 18/18] arm64: select ARCH_SUPPORTS_LTO_CLANG Date: Mon, 20 Nov 2017 18:05:55 +0000 [thread overview] Message-ID: <20171120180554.GM32488@arm.com> (raw) In-Reply-To: <20171116201701.GA143965@samitolvanen.mtv.corp.google.com> Hi Sami, On Thu, Nov 16, 2017 at 12:17:01PM -0800, Sami Tolvanen wrote: > On Thu, Nov 16, 2017 at 11:13:07AM -0800, Paul E. McKenney wrote: > > Ah, if "this patch set" meant "adding LTO", I stand corrected and I > > apologize for my confusion. > > Again, I'm not proposing for LTO to be enabled by default. These patches > just make it possible to enable it. Are you saying the possibility > that a future compiler update breaks something is a blocker even for > experimental features? My opinion here is that it's a blocker for LTO on arm64, as the symptoms aren't different to building with a compiler that has subtle code generation issues. If you *really* want to support LTO, we could consider giving acquire semantics to READ_ONCE for LTO builds, but that means trading RCU read-side performance for whatever gains are provided by LTO. Please don't confuse this with a reluctance to support clang; I'm keen to see that supported, so let's focus on that for the moment (although I don't see a good reason to support old clang builds with known issues). > > I agree that we need LTO/PGO to be housebroken from an LKMM viewpoint > > before it is enabled. > > Can you elaborate what's needed from clang before this can move > forward? For example, if you have specific test cases in mind, we can > always work on including them in the LLVM test suite. This is a thorny issue, but RCU (specifically rcu_dereference but probably also some READ_ONCEs) relies on being able to utilise syntactic dependency chains to order local accesses to shared variables. Paul has spent a fair amount of time working to define these dependency chains here: https://wg21.link/P0190 Although the current direction of the C++ committee is to prefer that dependencies are explicitly "marked", this is not deemed to be acceptable for the kernel (in other words, everything is always considered "marked"). If a compiler is capable of breaking unmarked dependency chains, then we will run into subtle concurrency issues on arm64 because the hardware is also capable of reordering independent accesses. My understanding is that LTO makes these sort of optimisations far more likely because whole-program analysis can enable value prediction, which can convert address dependencies into control dependcies; the latter not ordering reads on arm64. Will
next prev parent reply other threads:[~2017-11-20 18:05 UTC|newest] Thread overview: 226+ messages / expand[flat|nested] mbox.gz Atom feed top 2017-11-15 21:34 [PATCH v2 00/18] Add support for clang LTO Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-15 21:34 ` [PATCH v2 01/18] kbuild: add ld-name macro and support for GNU gold Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-15 21:53 ` Kees Cook 2017-11-15 21:53 ` Kees Cook 2017-11-15 21:34 ` [PATCH v2 02/18] kbuild: fix LD_DEAD_CODE_DATA_ELIMINATION with " Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-15 21:34 ` [PATCH v2 03/18] kbuild: move gcc-version.sh to cc-version.sh and add clang support Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-15 21:48 ` Kees Cook 2017-11-15 21:48 ` Kees Cook 2017-11-15 22:06 ` Sami Tolvanen 2017-11-15 22:06 ` Sami Tolvanen 2017-11-15 21:34 ` [PATCH v2 04/18] arm64: use -mno-implicit-float instead of -mgeneral-regs-only Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-15 21:34 ` [PATCH v2 05/18] arm64: fix -m for GNU gold Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-16 10:55 ` Yury Norov 2017-11-16 10:55 ` Yury Norov 2017-11-16 16:55 ` Nick Desaulniers 2017-11-16 16:55 ` Nick Desaulniers 2017-11-15 21:34 ` [PATCH v2 06/18] arm64: kvm: use -fno-jump-tables with clang Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-16 11:46 ` Will Deacon 2017-11-16 11:46 ` Will Deacon 2017-11-16 16:25 ` Sami Tolvanen 2017-11-16 16:25 ` Sami Tolvanen 2017-11-20 14:41 ` Mark Rutland 2017-11-20 14:41 ` Mark Rutland 2017-11-20 14:43 ` Mark Rutland 2017-11-20 14:43 ` Mark Rutland 2017-11-20 14:47 ` Ard Biesheuvel 2017-11-20 14:47 ` Ard Biesheuvel 2017-11-20 14:47 ` Ard Biesheuvel 2017-11-15 21:34 ` [PATCH v2 07/18] arm64: keep .altinstructions and .altinstr_replacement Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-15 21:34 ` [PATCH v2 08/18] arm64: don't disable ADR_PREL_PG_HI21* with ARM64_ERRATUM_843419 Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-15 22:29 ` Ard Biesheuvel 2017-11-15 22:29 ` Ard Biesheuvel 2017-11-15 22:29 ` Ard Biesheuvel 2017-11-16 11:44 ` Will Deacon 2017-11-16 11:44 ` Will Deacon 2017-11-16 11:44 ` Will Deacon 2017-11-16 16:32 ` Sami Tolvanen 2017-11-16 16:32 ` Sami Tolvanen 2017-11-16 16:32 ` Sami Tolvanen 2017-11-16 16:34 ` Ard Biesheuvel 2017-11-16 16:34 ` Ard Biesheuvel 2017-11-16 16:34 ` Ard Biesheuvel 2017-11-16 21:37 ` Sami Tolvanen 2017-11-16 21:37 ` Sami Tolvanen 2017-11-16 21:37 ` Sami Tolvanen 2017-11-16 22:14 ` Ard Biesheuvel 2017-11-16 22:14 ` Ard Biesheuvel 2017-11-16 22:14 ` Ard Biesheuvel 2017-11-16 22:25 ` Ard Biesheuvel 2017-11-16 22:25 ` Ard Biesheuvel 2017-11-16 22:25 ` Ard Biesheuvel 2017-11-16 23:09 ` Sami Tolvanen 2017-11-16 23:09 ` Sami Tolvanen 2017-11-16 23:09 ` Sami Tolvanen 2017-11-16 23:20 ` Ard Biesheuvel 2017-11-16 23:20 ` Ard Biesheuvel 2017-11-16 23:20 ` Ard Biesheuvel 2017-11-16 23:50 ` Sami Tolvanen 2017-11-16 23:50 ` Sami Tolvanen 2017-11-16 23:50 ` Sami Tolvanen 2017-11-17 9:54 ` Ard Biesheuvel 2017-11-17 9:54 ` Ard Biesheuvel 2017-11-17 9:54 ` Ard Biesheuvel 2017-11-17 18:49 ` Sami Tolvanen 2017-11-17 18:49 ` Sami Tolvanen 2017-11-17 18:49 ` Sami Tolvanen 2017-11-15 21:34 ` [PATCH v2 09/18] arm64: explicitly pass --no-fix-cortex-a53-843419 to GNU gold Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-16 11:47 ` Will Deacon 2017-11-16 11:47 ` Will Deacon 2017-11-16 16:35 ` Sami Tolvanen 2017-11-16 16:35 ` Sami Tolvanen 2017-11-20 14:47 ` Mark Rutland 2017-11-20 14:47 ` Mark Rutland 2017-11-20 20:35 ` Sami Tolvanen 2017-11-20 20:35 ` Sami Tolvanen 2017-11-21 11:15 ` Mark Rutland 2017-11-21 11:15 ` Mark Rutland 2017-11-15 21:34 ` [PATCH v2 10/18] arm64: add a workaround for GNU gold with ARM64_MODULE_PLTS Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-16 11:50 ` Will Deacon 2017-11-16 11:50 ` Will Deacon 2017-11-16 16:41 ` Sami Tolvanen 2017-11-16 16:41 ` Sami Tolvanen 2017-11-16 18:47 ` Will Deacon 2017-11-16 18:47 ` Will Deacon 2017-11-15 21:34 ` [PATCH v2 11/18] arm64: make mrs_s and msr_s macros work with LTO Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-16 11:54 ` Will Deacon 2017-11-16 11:54 ` Will Deacon 2017-11-16 13:07 ` Yury Norov 2017-11-16 13:07 ` Yury Norov 2017-11-16 13:55 ` Robin Murphy 2017-11-16 13:55 ` Robin Murphy 2017-11-16 21:29 ` Yury Norov 2017-11-16 21:29 ` Yury Norov 2017-11-16 22:54 ` Alex Matveev 2017-11-16 22:54 ` Alex Matveev 2017-12-04 17:34 ` Nick Desaulniers 2017-12-04 17:34 ` Nick Desaulniers 2017-11-16 13:56 ` Segher Boessenkool 2017-11-16 13:56 ` Segher Boessenkool 2017-11-16 16:46 ` Sami Tolvanen 2017-11-16 16:46 ` Sami Tolvanen 2017-11-16 17:01 ` Segher Boessenkool 2017-11-16 17:01 ` Segher Boessenkool 2017-11-16 17:11 ` Sami Tolvanen 2017-11-16 17:11 ` Sami Tolvanen 2017-11-16 16:43 ` Sami Tolvanen 2017-11-16 16:43 ` Sami Tolvanen 2017-11-16 16:44 ` Nick Desaulniers 2017-11-16 16:44 ` Nick Desaulniers 2017-11-16 18:14 ` Alex Matveev 2017-11-16 18:14 ` Alex Matveev 2017-11-15 21:34 ` [PATCH v2 12/18] kbuild: add support for clang LTO Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-15 22:06 ` Kees Cook 2017-11-15 22:06 ` Kees Cook 2017-11-18 3:21 ` [v2,12/18] " Nicholas Piggin 2017-11-18 3:21 ` Nicholas Piggin 2017-11-20 20:21 ` Sami Tolvanen 2017-11-20 20:21 ` Sami Tolvanen 2017-11-21 1:01 ` Nicholas Piggin 2017-11-21 1:01 ` Nicholas Piggin 2017-11-29 23:30 ` Sami Tolvanen 2017-11-29 23:30 ` Sami Tolvanen 2017-11-15 21:34 ` [PATCH v2 13/18] kbuild: fix dynamic ftrace with " Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-15 21:34 ` [PATCH v2 14/18] scripts/mod: disable LTO for empty.c Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-15 21:34 ` [PATCH v2 15/18] efi/libstub: disable LTO Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-15 21:34 ` [PATCH v2 16/18] arm64: crypto: disable LTO for aes-ce-cipher.c Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-20 15:20 ` Mark Rutland 2017-11-20 15:20 ` Mark Rutland 2017-11-20 15:25 ` Ard Biesheuvel 2017-11-20 15:25 ` Ard Biesheuvel 2017-11-20 15:25 ` Ard Biesheuvel 2017-11-20 21:01 ` Sami Tolvanen 2017-11-20 21:01 ` Sami Tolvanen 2017-11-20 21:01 ` Sami Tolvanen 2017-11-21 11:47 ` Mark Rutland 2017-11-21 11:47 ` Mark Rutland 2017-11-21 11:47 ` Mark Rutland 2017-11-20 20:51 ` Sami Tolvanen 2017-11-20 20:51 ` Sami Tolvanen 2017-11-20 21:29 ` Alex Matveev 2017-11-20 21:29 ` Alex Matveev 2017-11-20 21:31 ` Ard Biesheuvel 2017-11-20 21:31 ` Ard Biesheuvel 2017-11-20 21:31 ` Ard Biesheuvel 2017-11-20 22:03 ` Alex Matveev 2017-11-20 22:03 ` Alex Matveev 2017-11-20 22:03 ` Alex Matveev 2017-11-15 21:34 ` [PATCH v2 17/18] arm64: disable RANDOMIZE_MODULE_REGION_FULL with LTO_CLANG Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-15 21:34 ` [PATCH v2 18/18] arm64: select ARCH_SUPPORTS_LTO_CLANG Sami Tolvanen 2017-11-15 21:34 ` Sami Tolvanen 2017-11-16 11:58 ` Will Deacon 2017-11-16 11:58 ` Will Deacon 2017-11-16 16:17 ` Sami Tolvanen 2017-11-16 16:17 ` Sami Tolvanen 2017-11-16 16:30 ` Peter Zijlstra 2017-11-16 16:30 ` Peter Zijlstra 2017-11-16 16:50 ` Nick Desaulniers 2017-11-16 16:50 ` Nick Desaulniers 2017-11-16 16:59 ` Peter Zijlstra 2017-11-16 16:59 ` Peter Zijlstra 2017-11-16 17:16 ` Nick Desaulniers 2017-11-16 17:16 ` Nick Desaulniers 2017-11-16 17:34 ` Peter Zijlstra 2017-11-16 17:34 ` Peter Zijlstra 2017-11-16 17:48 ` Paul E. McKenney 2017-11-16 17:48 ` Paul E. McKenney 2017-11-16 18:16 ` Nick Desaulniers 2017-11-16 18:16 ` Nick Desaulniers 2017-11-16 18:39 ` Paul E. McKenney 2017-11-16 18:39 ` Paul E. McKenney 2017-11-16 18:45 ` Will Deacon 2017-11-16 18:45 ` Will Deacon 2017-11-16 19:13 ` Paul E. McKenney 2017-11-16 19:13 ` Paul E. McKenney 2017-11-16 20:17 ` Sami Tolvanen 2017-11-16 20:17 ` Sami Tolvanen 2017-11-20 18:05 ` Will Deacon [this message] 2017-11-20 18:05 ` Will Deacon 2017-11-20 19:25 ` Peter Zijlstra 2017-11-20 19:25 ` Peter Zijlstra 2017-11-20 19:28 ` Peter Zijlstra 2017-11-20 19:28 ` Peter Zijlstra 2017-11-20 20:52 ` Paul E. McKenney 2017-11-20 20:52 ` Paul E. McKenney 2017-11-20 19:32 ` Peter Zijlstra 2017-11-20 19:32 ` Peter Zijlstra 2017-11-20 20:53 ` Paul E. McKenney 2017-11-20 20:53 ` Paul E. McKenney 2017-11-21 17:23 ` David Laight 2017-11-21 17:23 ` David Laight 2017-11-21 17:23 ` David Laight 2017-11-21 18:51 ` Paul E. McKenney 2017-11-21 18:51 ` Paul E. McKenney 2017-11-21 18:51 ` Paul E. McKenney 2017-11-23 13:42 ` Alexander Potapenko 2017-11-23 13:42 ` Alexander Potapenko 2017-11-24 7:52 ` Dmitry Vyukov 2017-11-24 7:52 ` Dmitry Vyukov 2017-11-16 20:53 ` [PATCH v2 00/18] Add support for clang LTO Yury Norov 2017-11-16 20:53 ` Yury Norov 2017-11-16 21:38 ` Sami Tolvanen 2017-11-16 21:38 ` Sami Tolvanen 2017-11-20 15:21 ` Mark Rutland 2017-11-20 15:21 ` Mark Rutland 2017-11-20 21:04 ` Sami Tolvanen 2017-11-20 21:04 ` Sami Tolvanen 2017-11-21 11:53 ` Mark Rutland 2017-11-21 11:53 ` Mark Rutland
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=20171120180554.GM32488@arm.com \ --to=will.deacon@arm.com \ --cc=ak@linux.intel.com \ --cc=alxmtvv@gmail.com \ --cc=ard.biesheuvel@linaro.org \ --cc=ghackmann@google.com \ --cc=glider@google.com \ --cc=keescook@chromium.org \ --cc=linux-arm-kernel@lists.infradead.org \ --cc=linux-kbuild@vger.kernel.org \ --cc=linux-kernel@vger.kernel.org \ --cc=manojgupta@google.com \ --cc=mark.rutland@arm.com \ --cc=maxim.kuvyrkov@linaro.org \ --cc=michal.lkml@markovi.net \ --cc=mka@chromium.org \ --cc=ndesaulniers@google.com \ --cc=paulmck@linux.vnet.ibm.com \ --cc=peterz@infradead.org \ --cc=pirama@google.com \ --cc=samitolvanen@google.com \ --cc=srhines@google.com \ --cc=yamada.masahiro@socionext.com \ --cc=ynorov@caviumnetworks.com \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.