From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752117AbdGIDN7 (ORCPT ); Sat, 8 Jul 2017 23:13:59 -0400 Received: from mail-pf0-f195.google.com ([209.85.192.195]:35664 "EHLO mail-pf0-f195.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751680AbdGIDNz (ORCPT ); Sat, 8 Jul 2017 23:13:55 -0400 From: Nicholas Piggin To: linux-arch@vger.kernel.org Cc: Nicholas Piggin , linux-kbuild@vger.kernel.org, x86@kernel.org, linux-kernel@vger.kernel.org, Nicolas Pitre , Arnd Bergmann , Paul Burton , Linus Torvalds Subject: [RFC PATCH] x86: enable dead code and data elimination (LTO) Date: Sun, 9 Jul 2017 13:13:33 +1000 Message-Id: <20170709031333.29443-1-npiggin@gmail.com> X-Mailer: git-send-email 2.11.0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Allow x86 to select DCDE option under CONFIG_EXPERT to reduce binary size. This is an RFC only for ~4.14/15 kernel. Sending as a single patch to make it easy to review and test for x86, and give other archs a base to look at. This is a _relatively_ simple and low overhead first step to getting some dead code elimination working. I would like to know if people think this is reasonable and wanted. Enabling this option reduces size of x86-64 allnoconfig -Os build significantly: text data bss dec filename 783424 618456 185488 1587368 vmlinux 641277 612536 185488 1439301 vmlinux.dcde 142147 5920 0 148067 difference 18% 1% 0% 9% The improvement with defconfig is much smaller. 10690125 4630536 884736 16205397 vmlinux 10623912 4607432 880640 16111984 vmlinux.dcde 66213 23104 4096 93413 difference 0.6% 0.5% 0.5% 0.6% There are a number of reasons for this. Firstly not everything is built conditionally on a fine grained basis, so the allnoconfig kernel includes lots more code that's never used. But there are also things which pin code, for example BUG and exception tables are marked with KEEP() and reference kernel code which may be otherwise unused. In future, moving to a direct referencing of such tables should allow more size reductions of larger kernel configs. FYI, easiest way to check if you forgot to KEEP a linker table is to look at `readelf -S vmlinux` differences, and to see what is being trimmed, look at nm differences or use --print-gc-sections LD option to see what symbols you're trimming. Linker tables, boot entry, and exception entry tends to require anchoring. --- arch/Kconfig | 15 --------------- arch/x86/Kconfig | 1 + arch/x86/kernel/vmlinux.lds.S | 22 +++++++++++----------- include/asm-generic/vmlinux.lds.h | 11 ++++++++++- init/Kconfig | 27 +++++++++++++++++++++++++++ 5 files changed, 49 insertions(+), 27 deletions(-) diff --git a/arch/Kconfig b/arch/Kconfig index cae0958a2298..00edad06c71f 100644 --- a/arch/Kconfig +++ b/arch/Kconfig @@ -555,21 +555,6 @@ config THIN_ARCHIVES Select this if the architecture wants to use thin archives instead of ld -r to create the built-in.o files. -config LD_DEAD_CODE_DATA_ELIMINATION - bool - help - Select this if the architecture wants to do dead code and - data elimination with the linker by compiling with - -ffunction-sections -fdata-sections and linking with - --gc-sections. - - This requires that the arch annotates or otherwise protects - its external entry points from being discarded. Linker scripts - must also merge .text.*, .data.*, and .bss.* correctly into - output sections. Care must be taken not to pull in unrelated - sections (e.g., '.text.init'). Typically '.' in section names - is used to distinguish them from label names / C identifiers. - config HAVE_ARCH_WITHIN_STACK_FRAMES bool help diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig index 94a18681353d..d885706d7eb9 100644 --- a/arch/x86/Kconfig +++ b/arch/x86/Kconfig @@ -152,6 +152,7 @@ config X86 select HAVE_KPROBES_ON_FTRACE select HAVE_KRETPROBES select HAVE_KVM + select HAVE_LD_DEAD_CODE_DATA_ELIMINATION select HAVE_LIVEPATCH if X86_64 select HAVE_MEMBLOCK select HAVE_MEMBLOCK_NODE_MAP diff --git a/arch/x86/kernel/vmlinux.lds.S b/arch/x86/kernel/vmlinux.lds.S index c8a3b61be0aa..3b5ab910bbd4 100644 --- a/arch/x86/kernel/vmlinux.lds.S +++ b/arch/x86/kernel/vmlinux.lds.S @@ -203,14 +203,14 @@ SECTIONS * See static_cpu_has() for an example. */ .altinstr_aux : AT(ADDR(.altinstr_aux) - LOAD_OFFSET) { - *(.altinstr_aux) + KEEP(*(.altinstr_aux)) } INIT_DATA_SECTION(16) .x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) { __x86_cpu_dev_start = .; - *(.x86_cpu_dev.init) + KEEP(*(.x86_cpu_dev.init)) __x86_cpu_dev_end = .; } @@ -218,7 +218,7 @@ SECTIONS .x86_intel_mid_dev.init : AT(ADDR(.x86_intel_mid_dev.init) - \ LOAD_OFFSET) { __x86_intel_mid_dev_start = .; - *(.x86_intel_mid_dev.init) + KEEP(*(.x86_intel_mid_dev.init)) __x86_intel_mid_dev_end = .; } #endif @@ -232,7 +232,7 @@ SECTIONS . = ALIGN(8); .parainstructions : AT(ADDR(.parainstructions) - LOAD_OFFSET) { __parainstructions = .; - *(.parainstructions) + KEEP(*(.parainstructions)) __parainstructions_end = .; } @@ -244,7 +244,7 @@ SECTIONS . = ALIGN(8); .altinstructions : AT(ADDR(.altinstructions) - LOAD_OFFSET) { __alt_instructions = .; - *(.altinstructions) + KEEP(*(.altinstructions)) __alt_instructions_end = .; } @@ -254,7 +254,7 @@ SECTIONS * get the address and the length of them to patch the kernel safely. */ .altinstr_replacement : AT(ADDR(.altinstr_replacement) - LOAD_OFFSET) { - *(.altinstr_replacement) + KEEP(*(.altinstr_replacement)) } /* @@ -265,14 +265,14 @@ SECTIONS */ .iommu_table : AT(ADDR(.iommu_table) - LOAD_OFFSET) { __iommu_table = .; - *(.iommu_table) + KEEP(*(.iommu_table)) __iommu_table_end = .; } . = ALIGN(8); .apicdrivers : AT(ADDR(.apicdrivers) - LOAD_OFFSET) { __apicdrivers = .; - *(.apicdrivers); + KEEP(*(.apicdrivers)) __apicdrivers_end = .; } @@ -307,7 +307,7 @@ SECTIONS . = ALIGN(PAGE_SIZE); .smp_locks : AT(ADDR(.smp_locks) - LOAD_OFFSET) { __smp_locks = .; - *(.smp_locks) + KEEP(*(.smp_locks)) . = ALIGN(PAGE_SIZE); __smp_locks_end = .; } @@ -323,7 +323,7 @@ SECTIONS .bss : AT(ADDR(.bss) - LOAD_OFFSET) { __bss_start = .; *(.bss..page_aligned) - *(.bss) + *(.bss .bss.[0-9a-zA-Z_]*) . = ALIGN(PAGE_SIZE); __bss_stop = .; } @@ -332,7 +332,7 @@ SECTIONS .brk : AT(ADDR(.brk) - LOAD_OFFSET) { __brk_base = .; . += 64 * 1024; /* 64k alignment slop space */ - *(.brk_reservation) /* areas brk users have reserved */ + KEEP(*(.brk_reservation)) /* areas brk users have reserved */ __brk_limit = .; } diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h index da0be9a8d1de..4ff42789e5e3 100644 --- a/include/asm-generic/vmlinux.lds.h +++ b/include/asm-generic/vmlinux.lds.h @@ -441,12 +441,21 @@ * architectures define .text.foo which is not intended to be pulled in here. * Those enabling LD_DEAD_CODE_DATA_ELIMINATION must ensure they don't have * conflicting section names, and must pull in .text.[0-9a-zA-Z_]* */ +#ifdef CONFIG_LD_DEAD_CODE_DATA_ELIMINATION +#define TEXT_TEXT \ + ALIGN_FUNCTION(); \ + *(.text.hot .text .text.[0-9a-zA-Z_]* .text.fixup .text.unlikely)\ + *(.ref.text) \ + MEM_KEEP(init.text) \ + MEM_KEEP(exit.text) +#else #define TEXT_TEXT \ ALIGN_FUNCTION(); \ *(.text.hot .text .text.fixup .text.unlikely) \ *(.ref.text) \ MEM_KEEP(init.text) \ - MEM_KEEP(exit.text) \ + MEM_KEEP(exit.text) +#endif /* sched.text is aling to function alignment to secure we have same diff --git a/init/Kconfig b/init/Kconfig index 8514b25db21c..9e209593e618 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -1052,6 +1052,33 @@ config CC_OPTIMIZE_FOR_SIZE endchoice +config HAVE_LD_DEAD_CODE_DATA_ELIMINATION + bool + help + This requires that the arch annotates or otherwise protects + its external entry points from being discarded. Linker scripts + must also merge .text.*, .data.*, and .bss.* correctly into + output sections. Care must be taken not to pull in unrelated + sections (e.g., '.text.init'). Typically '.' in section names + is used to distinguish them from label names / C identifiers. + +config LD_DEAD_CODE_DATA_ELIMINATION + bool "Dead code and data elimination (EXPERIMENTAL)" + depends on HAVE_LD_DEAD_CODE_DATA_ELIMINATION + depends on EXPERT + help + Select this if the architecture wants to do dead code and + data elimination with the linker by compiling with + -ffunction-sections -fdata-sections, and linking with + --gc-sections. + + This can reduce on disk and in-memory size of the kernel + code and static data, particularly for small configs and + on small systems. This has the possibility of introducing + silently broken kernel if the required annotations are not + present. This option is not well tested yet, so use at your + own risk. + config SYSCTL bool -- 2.11.0