lld 18 ELF changes

LLVM 18 will be released. As usual, I maintain lld/ELF and have added some notes to https://github.com/llvm/llvm-project/blob/release/18.x/lld/docs/ReleaseNotes.rst. I've meticulously reviewed nearly all the patches that are not authored by me. I'll delve into some of the key changes.

  • --fat-lto-objects option is added to support LLVM FatLTO. Without --fat-lto-objects, LLD will link LLVM FatLTO objects using the relocatable object file. (D146778 <https://reviews.llvm.org/D146778>_)
  • -Bsymbolic-non-weak is added to directly bind non-weak definitions. (D158322)
  • --lto-validate-all-vtables-have-type-infos, which complements --lto-whole-program-visibility, is added to disable unsafe whole-program devirtualization. --lto-known-safe-vtables=<glob> can be used to mark known-safe vtable symbols. (D155659)
  • --save-temps --lto-emit-asm now derives ELF/asm file names from bitcode file names. ld.lld --save-temps a.o d/b.o -o out will create ELF relocatable files out.lto.a.o/d/out.lto.b.o instead of out1.lto.o/out2.lto.o. (#78835)
  • --no-allow-shlib-undefined now reports errors for DSO referencing non-exported definitions. (#70769)
  • common-page-size can now be larger than the system page-size. (#57618)
  • When call graph profile information is available due to instrumentation or sample PGO, input sections are now sorted using the new cdsort algorithm, better than the previous hfsort algorithm. (D152840)
  • Symbol assignments like a = DEFINED(a) ? a : 0; are now handled. (#65866)
  • OVERLAY now supports optional start address and LMA (#77272)
  • Relocations referencing a symbol defined in /DISCARD/ section now lead to an error. (#69295)
  • For AArch64 MTE, global variable descriptors have been implemented. (D152921)
  • R_AARCH64_GOTPCREL32 is now supported. (#72584)
  • R_LARCH_PCREL20_S2/R_LARCH_ADD6/R_LARCH_CALL36 and extreme code model relocations are now supported.
  • --emit-relocs is now supported for RISC-V linker relaxation. (D159082)
  • Call relaxation respects RVC when mixing +c and -c relocatable files. (#73977)
  • R_RISCV_GOT32_PCREL is now supported. (#72587)
  • R_RISCV_SET_ULEB128/R_RISCV_SUB_ULEB128 relocations are now supported. (#72610) (#77261)
  • RISC-V TLSDESC is now supported. (#79239)
  • SystemZ (s390x) is now supported. (#75643)

Although a substantial feature, the s390x port benefited from Ulrich Weigand's meticulously prepared patch, complete with comprehensive tests from the outset. While I typically provide extensive feedback for such large additions, the patch's exceptional quality minimized the number of comment rounds needed in this instance. I truly appreciate Ulrich taking the time to reply to my remarks on the s390x ABI. His guidance was instrumental.

The RISC-V port has received a few new relocations for .uleb128 label differences and TLSDESC. I'm glad I made these assembler and linker changes in time, allowing LoongArch developers to port the code for LoongArch. LoongArch borrows many designs from RISC-V. If they were to implement these features first, I'd probably have to spend more time on code reviews, and the outcome would be less rewarding since I wouldn't be the original patch author.

R_AARCH64_GOTPCREL32 (G(GDAT(S))+A-P) and R_RISCV_GOT32_PCREL, similar to R_X86_64_GOTPCREL, are new ABI additions used to optimize preemptible _ZTI* symbol references for clang -fexperimental-relative-c++-abi-vtables. The simplification utilizes the optimization originally done for Mach-O, which applies to GOT equivalent global variables (global unnamed, constant, discardable linkage). 

1
2
3
4
5
6
7
8
9
10
11
12
13
        .long   0                               # 0x0
-       .long   (_ZTI1A.rtti_proxy-.L_ZTV1A.local)-8
+       .long   _ZTI1A@GOTPCREL-4
        .long   (_ZN1A3fooEv@PLT-.L_ZTV1A.local)-8

-       .hidden _ZTI1A.rtti_proxy               # @_ZTI1A.rtti_proxy
-       .type   _ZTI1A.rtti_proxy,@object
-       .section        .data.rel.ro._ZTI1A.rtti_proxy,"aGw",@progbits,_ZTI1A.rtti_proxy,comdat
-       .globl  _ZTI1A.rtti_proxy
-       .p2align        3, 0x0
-_ZTI1A.rtti_proxy:
-       .quad   _ZTI1A
-       .size   _ZTI1A.rtti_proxy, 8

To the best of my knowledge, there is no performance-specific change.

Link: lld 17 ELF changes