LLD, the LLVM linker, is a mature and fast linker supporting multiplebinary formats (ELF, Mach-O, PE/COFF, WebAssembly). Designed as astandalone program, the code base relies heavily on global state, makingit less than ideal for library integration. As outlined in RFC:Revisiting LLD-as-a-library design, two main hurdles exist:Fatal errors: they exit the process without returning control to thecaller.Global variable conflicts: shared global variables do not allow twoconcurrent invocation.I understand that calling a linker API could be convenient,especially when you want to avoid shipping anothe ...继续阅读 (3)

LLVM's C++ API doesn't offer a stability guarantee. This meansfunction signatures can change or be removed between versions, forcingprojects to adapt.On the other hand, LLVM has an extensive API surface. When a librarylikellvm/lib/Yrelies functionality from another library,the API is often exported in header files underllvm/include/llvm/X/, even if it is not intended to beuser-facing.To be compatible with multiple LLVM versions, many projects rely on#ifdirectives based on theLLVM_VERSION_MAJORmacro. This post explores the specific techniques used by ccls to ensurecompatibility with LLVM versio ...继续阅读 (3)

Aftermigrating fromVim to Emacsas my primary C++ editor in 2015, I switched from Vimto Neovim for miscellaneous non-C++ tasks as it is more convenient in aterminal. Customizing the editor with a language you are comfortablewith is important. I found myself increasingly drawn to Neovim'sterminal-based simplicity for various non-C++ tasks. Recently, I'verefined my Neovim setup to the point where I can confidently migrate myentire C++ workflow away from Emacs. This post explores the keyimprovements I've made to achieve this seamless transition.Key mappingI've implemented custom functions that sim ...继续阅读 (5)

My C++ language server ccls implements a semantic highlightingfeatures through the non-standard messages$ccls/publishSemanticHighlightand$ccls/publishSkippedRanges, supported by both emacs-cclsand vscode-ccls.For years, my primary editor was Emacs. In 2018, I created emacs-ccls(a fork of emacs-cquery) and enjoyed its rainbow semantic highlighting.My setup also relied heavily on two features:Bolding and underlining variables of static duration storageItalicizing member functions and variables12(setqccls-sem-highlight-method 'font-lock)(ccls-use-default-rainbow-sem-highlight)My elisp skills ...继续阅读 (9)

LLVM 19 will soon be released. This post provides a summary of mycontributions in this release cycle to record my learning progress.LLVM MC: Integratedassembler improvements in LLVM 19lld/ELF:lld 19 ELFchangesAArch64[AArch64]Implement -fno-plt for SelectionDAG/GlobalISel[AArch64]Support optional constant offset for constraint "S"[AArch64AsmParser]Allow branch target symbol to have a shift/extend modifier name[MC,AArch64]Create mapping symbols with non-unique namesARM[ARM,MC]Support FDPIC relocations. See MMU-less systems andFDPICRISC-V[RISCV]Support constraint "s"[MC,RISCV]Create mapping symbo ...继续阅读 (31)

LLVM 19 will be released. As usual, I maintain lld/ELF and have addedsome notes to https://github.com/llvm/llvm-project/blob/release/19.x/lld/docs/ReleaseNotes.rst.I've meticulously reviewed nearly all the patches that are not authoredby me. I'll delve into some of the key changes.EI_OSABIin the output is now inferred from inputobject files. (#97144)--compress-sections={none,zlib,zstd}[:level]is added to compress matched output sections without theSHF_ALLOCflag. (#84855) (#90567)The default compression level for zlib is now independent of linkeroptimization level (Z_BEST_SPEED).zstd ...继续阅读 (21)

In object files, certain code patterns incorporate data directlywithin the code or transit between instruction sets. This practice canpose challenges for disassemblers as data might be mistakenlyinterpreted as code, leading to nonsensical output. In addition, codefrom instruction set A might be disassembled as instruction set B. Toaddress the issues, some architectures define mapping symbols todescribe state transition. Let's explore this concept using an AArch32code example:1234567891011121314151617181920212223.textadr r0, .LJTI0_0vldr d0, .LCPI0_0bl thumb_callee.LBB0_1:nop.LBB0_2:nop. ...继续阅读 (32)

The output ofld.lld -vincludes a message "compatiblewith GNU linkers" to address detectionmechanism used by GNU Libtool. This problem is described by Softwarecompatibility and our own "User-Agent" problem.The latestm4/libtool.m4continues to rely on aGNUcheck.12345678910[AC_CACHE_CHECK([if the linker ($LD) is GNU ld], lt_cv_prog_gnu_ld,[# I'd rather use --version here, but apparently some GNU lds only accept -v.case `$LD -v 2>&1 ...继续阅读 (41)

Within the LLVM project, MC is a library responsible for handlingassembly, disassembly, and object file formats. [Intro to the LLVM MCProject], which was written back in 2010, remains a good source tounderstand the high-level structures.In the latest release cycle, substantial effort has been dedicated torefining MC's internal representation for improved performance andreadability. These changes havedecreased compiletime significantly. This blog post will delve into the details,providing insights into the specific changes.MCAssemblerandMCAsmLayoutMCAssemblermanages assembler states (includings ...继续阅读 (37)

GNU ld's output section layout is determined by a linker script,which can be either internal (default) or external (specified with-Tor-dT). Within the linker script,SECTIONScommands define how input sections are mapped intooutput sections.Input sections not explicitly placed bySECTIONScommands are termed "orphansections".Orphan sections are sections present in the input files which are notexplicitly placed into the output file by the linker script. The linkerwill still copy these sections into the output file by either finding,or creating a suitable output section in which to place the orphane ...继续阅读 (32)

The ELF object file format is adopted by many UNIX-like operatingsystems. Understanding its specifications can be an interestingjourney.The format consists of the generic specification, processor-specificspecifications, and OS-specific specifications. Three key documentsoften surface when searching for the generic specification:Tool Interface Standard (TIS) Portable Formats Specification,version 1.2onhttps://refspecs.linuxfoundation.org/SystemV Application Binary Interface - DRAFT - 10 June 2013onwww.sco.comOracle Solaris Linkers and Libraries GuideThe TIS specification breaks ELF into the gen ...继续阅读 (46)

The Linux kernel is written in C, but it also leverages extensionsprovided by GCC. In 2022, it moved from GCC/Clang-std=gnu89to-std=gnu11. This articleexplores my notes on how these GNU extensions are utilized within thekernel.Statement expressionsStatementexpressionsare commonly used in macros.LocallabelsSome macros use this extension to restart a for loop in a macro'sreplacement list.123456789101112131415161718192021222324// include/linux/wait.h#define___wait_event(wq_head, condition, state, exclusive, ret, cmd) \({ ...继续阅读 (44)

tl;dr Clang 19 will remove the-mrelax-alldefault at-O0, significantly decreasing the text section size forx86.Span-dependent instructionsIn assembly languages, some instructions with an immediate operandcan be encoded in two (or more) forms with different sizes. On x86-64, aJMP/JCC (jumps) can be encoded either in 2 bytes with a 8-bit relativeoffset or 6 bytes with a 32-bit relative offset. The short form ispreferred because it takes less space. However, when the target of thejump is too far away, the long form must be used.1234ja foo # jump near if above, 77ja foo # jump short if above, ...继续阅读 (50)

QOI, the Quite OK Image format,has been gaining in popularity. Chris Wellons offers a great analysis.QOI's key advantages is its simplicity. Being a byte-oriented formatwithout entropy encoding, it can be further compressed with generic datacompression programs like lz4, xz, and zstd. PNG, on the other hand,uses DEFLATE compression internally and is typically resistant tofurther compression. By applying a stronger compression algorithm on QOIoutput, you can often achieve a smaller file size compared to PNG.Lasse Collin has shared some effective options for compressinguncompressed BMP/TIFF file ...继续阅读 (42)

QOI, the Quite OK Image format,has been gaining in popularity. Chris Wellons offers a great analysis.QOI's key advantages is its simplicity. Being a byte-oriented formatwithout entropy encoding, it can be further compressed with genericcodecs like lz4, xz, and zstd. PNG, on the other hand, uses DEFLATEcompression internally and is typically resistant to furthercompression. By applying a stronger compression algorithm on QOI output,you can often achieve a smaller file size compared to PNG.Lasse Collin has shared some effective options for compressinguncompressed BMP/TIFF files. I tested them on ...继续阅读 (44)

ELF's design emphasizes naturalsize and alignment guidelines for its control structures. Whileensured efficient processing in the old days, this can lead to largerfile sizes. I propose "Light ELF" (EV_LIGHT, version 2) – awhimsical exploration inspired by Light Elves of Tolkien's legendarium(who had seen the light of the Two Trees in Valinor).In a light ELF file, thee_versionmember of the ELFheader is set to 2.EV_CURRENTremains 1 for backwardcompatibility.123#defineEV_NONE 0#defineEV_CURRENT 1#defineEV_LIGHT 2When linking a program, traditional ELF (version 1) and light ELF(version 2) files ca ...继续阅读 (39)

ELF's design emphasizes natural sizeand alignment guidelines for its control structures. However, thisapproach has substantial size drawbacks.In a release build of llvm-project(-O3 -ffunction-sections -fdata-sections, the sectionheader tables occupy 13.4% of the.ofile size.I propose an alternative section header table format that is signaledbye_shentsize == 0in the ELF header.e_shentsize == sizeof(Elf64_Shdr)(or the 32-bitcounterpart) selects the traditional section header table format.nshdrdenotes the number of sections (includingSHN_UNDEF). The compact section header table (located ate_shoff ...继续阅读 (44)

In ISO C++ standards, [basic.start.term] specifies that:Constructed objects ([dcl.init]) with static storage duration aredestroyed and functions registered with std::atexit are called as partof a call to std::exit ([support.start.term]). The call to std::exit issequenced before the destructions and the registered functions. [Note1: Returning from main invokes std​::​exit ([basic.start.main]). — endnote]For example, consider the following code:1structA{ ~A(); } a;The destructor for object a will be registered for execution atprogram termination.In the Itanium C++ ABI, object construct ...继续阅读 (44)

ELF's design emphasizes natural size and alignment guidelines for itscontrol structures. This principle, outlined in Proceedings of theSummer 1990 USENIX Conference,ELF: An Object File to MitigateMischievous Misoneism, promotes ease of random access forstructures like program headers, section headers, and symbols.All data structures that the object file format defines follow the"natural" size and alignment guidelines for the relevant class. Ifnecessary, data structures contain explicit padding to ensure 4-bytealignment for 4-byte objects, to force structure sizes to a multiple offour, etc. Dat ...继续阅读 (48)

UNDER CONSTRUCTION still working on experimentsELF's design emphasizes natural size and alignment guidelines for itscontrol structures. This principle, outlined in Proceedings of theSummer 1990 USENIX Conference,ELF: An Object File to MitigateMischievous Misoneism, promotes ease of random access forstructures like program headers, section headers, and symbols.All data structures that the object file format defines follow the"natural" size and alignment guidelines for the relevant class. Ifnecessary, data structures contain explicit padding to ensure 4-bytealignment for 4-byte objects, to force ...继续阅读 (51)

LLVM 18 will soon be relased. This post provides a summary of mycontributions in this release cycle to record my learning progress.LLVM binary utility maintenance, e.g.adopted llvm-readobjstyle ObjectFile specific dumpers[llvm-objdump][X86] Add @plt symbols for .plt.got[llvm-readobj] Printfor relocation target with an empty namecompiler-rt maintenance, e.g.changed lsan to work with high-entropy ASLR for x86-64 Linuxremoved crypt and crypt_rinterceptorsto work with glibcimplementedinterceptorsfor glibc 2.38__isoc23_strtoland__isoc23_scanffamily functionstsan: Respect !nosanitize metadata and re ...继续阅读 (52)

UNDER CONSTRUCTIONThis article describes some ABI and toolchain notes about systemswithout a Memory Management Unit (MMU), with a focus on FDPIC and thefuture RISC-V FDPIC ABI.Linux binfmt loadersfs/Kconfig.binfmtdefines a few loaders.BINFMT_ELFdefaults to y and depends onMMU.BINFMT_ELF_FDPICdefaults to y whenBINFMT_ELFis not selected. A few architecture supportBINFMT_ELF_FDPICfor NOMMU. ARM supports FDPIC even with aMMU.BINFMT_FLATis provided for a few architectures.BINFMT_AOUT, removed in 2022, had been supported foralpha/arm/x86-32.BFLTuClinux uses an object file format called Binary Flat f ...继续阅读 (53)

UNDER CONSTRUCTIONThis article describes some ABI and toolchain notes about systemswithout a Memory Management Unit (MMU), with a focus on FDPIC and thefuture RISC-V FDPIC ABI.Linux binfmt loadersfs/Kconfig.binfmtdefines a few loaders.BINFMT_ELFdefaults to y and depends onMMU.BINFMT_ELF_FDPICdefaults to y whenBINFMT_ELFis not selected. A few architecture supportBINFMT_ELF_FDPICfor NOMMU. ARM supports FDPIC even with aMMU.BINFMT_FLATis provided for a few architectures.BINFMT_AOUT, removed in 2022, had been supported foralpha/arm/x86-32.BFLTuClinux uses an object file format called Binary Flat f ...继续阅读 (57)

LLVM 18 will be released. As usual, I maintain lld/ELF and have addedsome notes to https://github.com/llvm/llvm-project/blob/release/18.x/lld/docs/ReleaseNotes.rst.Here I will elaborate on some changes.--fat-lto-objectsoption is added to support LLVMFatLTO. Without--fat-lto-objects, LLD will link LLVMFatLTO objects using the relocatable object file.(D146778_)-Bsymbolic-non-weakis added to directly bind non-weakdefinitions.(D158322_)--lto-validate-all-vtables-have-type-infos, whichcomplements--lto-whole-program-visibility, is added todisable unsafe whole-program devirtualization.--lto-known-saf ...继续阅读 (49)

This article describes some notes about z/Architecturewith a focus on the ELF ABI and ELF linkers. An lld/ELF patch sparked mymotivation to write this post.z/Architectureis a mainframe computer architecture supporting 24-bit, 31-bit, and64-bit addressing modes. It is the latest generation in a lineagestretching back to the 1964 with IBM System/360 (32-bit general purposeregisters and 24-bit addressing). This lineage includes System/370(1970), System/370 Extended Architecture (1983), Enterprise SystemsArchitecture/370 (1988), and Enterprise Systems Architecture/390 (1990).For a deeper dive into ...继续阅读 (77)

GCC's aarch64 port, from the beginning (2012), supports the constraint"S" to reference a symbol or label. This is nice, as you can create anartificial reference for linker garbage collection, defining sections tohold symbol addresses, or even enabling more creative applications.12345678namespacens {externinta[4]; }voidfun();voidfoo(){label:asm(".pushsection .xxx,\"aw\"; .dc.a %0; .popsection"::"S"(&ns;::a[3]));asm(".reloc ., BFD_RELOC_NONE, %0"::"S"(fun));asm("// %0"::"S"(&&label;));}Using the genericrormconstraint in suchcases would instruct GCC to generate instructions to ...继续阅读 (59)

Key metrics for code coverage include:function coverage: determines whether each function beenexecuted.line coverage (aka statement coverage): determines whether everyline has been executed.branch coverage: ensures that both the true and false branches ofeach conditional statement or the conditional of each loop statementbeen evaluated.Condition coverage offers a more fine-grained evaluation of branchcoverage. It requires that each individual boolean subexpression(condition) within a compound expression be evaluated to both true andfalse. For example, in the boolean expressionif (a>0 && f(b) & ...继续阅读 (64)

I'll make this post more polished.Patch musl per Re: Draftriscv64 TLSDESC implementation123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263diff --git c/arch/riscv64/reloc.h w/arch/riscv64/reloc.hindex 1ca13811..7c7c0611 100644--- c/arch/riscv64/reloc.h+++ w/arch/riscv64/reloc.h@@ -17,6 +17,7 @@#define REL_DTPMOD R_RISCV_TLS_DTPMOD64#define REL_DTPOFF R_RISCV_TLS_DTPREL64#define REL_TPOFF R_RISCV_TLS_TPREL64+#define REL_TLSDESC R_RISCV_TLSDESC#define CRTJMP(pc,sp) __asm__ __volatile__( \"mv sp, %1 ; jr %0" : : ...继续阅读 (57)

My journey with the LLVM project began with a deep dive into theworld of lld and binary utilities. Countless hours were spent unravelingthe intricacies of object file formats and shaping LLVM's relevantcomponents. Though my interests have since broadened, object fileformats remain a personal fascination, often drawing me into discussionsaround potential changes within LLVM.This article compares several prominent object file formats, drawingupon my experience and insights.At the heart of each format lies the representation of essentialcomponents like symbols, sections, and relocations. For each ...继续阅读 (51)

一如既往，主要在工具链领域耕耘。llvm-projectI made 700+ commits this year. Many are clean-up commits or fixup forothers' work. I hope that I can do more useful work next year.Enabled--features=layering_checkfor Bazel's llvm andclang projectsImplementedLLVM_ENABLE_REVERSE_ITERATIONforStringMapAddedllvm::xxh3_64bitsand adopted it inllvm/clang/lldMade AArch64 changes to msan and dfsanMade various improvements to the clang driver, including-###exit code, auxiliary files, errors forunsupported target-specific options,-fsanitize=kcfi,and XRayMade various sanitizer changesMade-fsanitize=functionwork for C and non-x8 ...继续阅读 (79)