[LLD][ELF] Skip non-SHF_ALLOC sections when checking max VA and max VA difference in relaxOnce()

[Enna1]

For non-SHF_ALLOC sections, sh_addr is set to 0.
Skip sections without SHF_ALLOC flag, so minVA will not be set to 0 with non-SHF_ALLOC sections, and the size of non-SHF_ALLOC sections will not contribute to maxVA.

[llvmbot]

@llvm/pr-subscribers-lld

@llvm/pr-subscribers-lld-elf

Author: Mingjie Xu (Enna1)

Changes

For non-SHF_ALLOC sections, sh_addr is set to 0.
Skip sections without SHF_ALLOC flag, so minVA will not be set to 0 with non-SHF_ALLOC sections, and the size of non-SHF_ALLOC sections will not contribute to maxVA.

---

Full diff: https://github.com/llvm/llvm-project/pull/145863.diff

1 Files Affected:

• (modified) lld/ELF/Arch/X86_64.cpp (+2)

diff --git a/lld/ELF/Arch/X86_64.cpp b/lld/ELF/Arch/X86_64.cpp
index 163505102d0ec..488f4803b2cb4 100644
--- a/lld/ELF/Arch/X86_64.cpp
+++ b/lld/ELF/Arch/X86_64.cpp
@@ -320,6 +320,8 @@ bool X86_64::deleteFallThruJmpInsn(InputSection &is, InputFile *file,
 bool X86_64::relaxOnce(int pass) const {
   uint64_t minVA = UINT64_MAX, maxVA = 0;
   for (OutputSection *osec : ctx.outputSections) {
+    if (!(osec->flags & SHF_ALLOC))
+      continue;
     minVA = std::min(minVA, osec->addr);
     maxVA = std::max(maxVA, osec->addr + osec->size);
   }

[MaskRay]

This is correct but can you edit /x86-64-gotpc-relax-too-far.s to make this testable?

[MaskRay]

.

[Enna1]

Thanks for the review!

This is correct but can you edit /x86-64-gotpc-relax-too-far.s to make this testable?

Tried but didn't make it...

The max VA and max VA difference checks are for early return.
The real decision for not relaxing R_X86_64_(REX_)GOTPCRELX is made by checking !isInt<32>(getRelocTargetVA).
Even if max VA or max VA difference in -pie/-shared is >= 2^31 (say the size of .strtab section is 2^31), isInt<32>(getRelocTargetVA) still can be true and R_X86_64_(REX_)GOTPCRELX will be relaxed.
I thinks current implementation doesn't have correctnees issue, but this change can remove some redundant isInt<32>(getRelocTargetVA) checks, so I didn't come up how to extend /x86-64-gotpc-relax-too-far.s to test this.

Do you have any suggestions? Thanks!

[MaskRay]

You are right that the SHF_ALLOC condition just skips some check and does not change the behavior. Then why does it matter? Do you find a scenario where having the condition improves performance?

[Enna1]

I'm looking for strategies that alleviate relocation overflow pressure. While trying to understand 9d6ec28 and f3c4dae, I found the max VA difference takes non-SHF_ALLOC sections into account, which is a bit confusing.

Do you find a scenario where having the condition improves performance?

I will test with internal cases to see if there is any performance improvement.

[Enna1]

Tested a internal large binary.
The size of .debug_info section and .debug_ranges section > 2^31.
With non-SHF_ALLOC sections, maxVA is 0x8ab15c0c (the size of .debug_info section).
Without non-SHF_ALLOC sections, maxVA is 0x3f42d818 (the end VA of .bss section).

readelf output:

Section Headers:
  [Nr] Name              Type            Address          Off    Size   ES Flg Lk Inf Al
  [ 0]                   NULL            0000000000000000 000000 000000 00      0   0  0
  [ 1] .interp           PROGBITS        00000000002002e0 0002e0 000035 00   A  0   0  1
  [ 2] .note.ABI-tag     NOTE            0000000000200318 000318 000020 00   A  0   0  4
  [ 3] .note.gnu.build-id NOTE            0000000000200338 000338 000024 00   A  0   0  4
  [ 4] .dynsym           DYNSYM          0000000000200360 000360 1b54900 18   A  8   1  8
  [ 5] .gnu.version      VERSYM          0000000001d54c60 1b54c60 2470c0 02   A  4   0  2
  [ 6] .gnu.version_r    VERNEED         0000000001f9bd20 1d9bd20 000510 00   A  8  14  4
  [ 7] .gnu.hash         GNU_HASH        0000000001f9c230 1d9c230 7af7b8 00   A  4   0  8
  [ 8] .dynstr           STRTAB          000000000274b9e8 254b9e8 d86753a 00   A  0   0  1
  [ 9] .rela.dyn         RELA            000000000ffb2f28 fdb2f28 280680 18   A  4   0  8
  [10] .rela.plt         RELA            00000000102335a8 100335a8 008c58 18  AI  4  33  8
  [11] .rodata           PROGBITS        000000001023d000 1003d000 211a18e 00 AMS  0   0 4096
  [12] .gcc_except_table PROGBITS        0000000012357190 12157190 1b1d73c 00   A  0   0  4
  [13] xxx               PROGBITS        0000000013e748d0 13c748d0 2fbf20 00   A  0   0 16
  [14] .eh_frame_hdr     PROGBITS        00000000141707f0 13f707f0 73e204 00   A  0   0  4
  [15] .eh_frame         PROGBITS        00000000148ae9f8 146ae9f8 327fe64 00   A  0   0  8
  [16] .text             PROGBITS        0000000017b2f900 1792e900 24cf90a0 00  AX  0   0 256
  [17] .init             PROGBITS        000000003c8289a0 3c6279a0 00001c 00  AX  0   0  4
  [18] .fini             PROGBITS        000000003c8289bc 3c6279bc 000009 00  AX  0   0  4
  [19] xxxx              PROGBITS        000000003c8289c6 3c6279c6 0001c0 00  AX  0   0  2
  [20] .plt              PROGBITS        000000003c828b90 3c627b90 005da0 00  AX  0   0 16
  [21] .tdata            PROGBITS        000000003c82f940 3c62d940 000b68 00 WAT  0   0 32
  [22] .tbss             NOBITS          000000003c8304b0 3c62e4a8 17e2a9 00 WAT  0   0 16
  [23] .fini_array       FINI_ARRAY      000000003c8304a8 3c62e4a8 000010 00  WA  0   0  8
  [24] .init_array       INIT_ARRAY      000000003c8304b8 3c62e4b8 00c520 00  WA  0   0  8
  [25] .data.rel.ro      PROGBITS        000000003c83c9e0 3c63a9e0 15afba0 00  WA  0   0 16
  [26] .dynamic          DYNAMIC         000000003ddec580 3dbea580 000460 10  WA  8   0  8
  [27] .got              PROGBITS        000000003ddec9e0 3dbea9e0 000848 00  WA  0   0  8
  [28] .bss.rel.ro       NOBITS          000000003dded228 3dbeb228 000018 00  WA  0   0  8
  [29] .relro_padding    NOBITS          000000003dded240 3dbeb228 000dc0 00  WA  0   0  1
  [30] .data             PROGBITS        000000003ddee240 3dbeb240 113c00 00  WA  0   0 16
  [31] .tm_clone_table   PROGBITS        000000003df01e40 3dcfee40 000000 00  WA  0   0  8
  [32] xxxxx             PROGBITS        000000003df01e40 3dcfee40 000060 00  WA  0   0 16
  [33] .got.plt          PROGBITS        000000003df01ea0 3dcfeea0 002ee0 00  WA  0   0  8
  [34] .bss              NOBITS          000000003df04d80 3dd01d80 1528a98 00 WAo  0   0 128
  [35] .comment          PROGBITS        0000000000000000 3dd01d80 000302 01  MS  0   0  1
  [36] .debug_aranges    PROGBITS        0000000000000000 3dd02090 000220 00      0   0 16
  [37] .debug_info       PROGBITS        0000000000000000 3dd022b0 8ab15c0c 00      0   0  1
  [38] .debug_abbrev     PROGBITS        0000000000000000 c8817ebc 620275 00      0   0  1
  [39] .debug_line       PROGBITS        0000000000000000 c8e38131 4311b0ab 00      0   0  1
  [40] .debug_str        PROGBITS        0000000000000000 10bf531dc 366716bd 01  MS  0   0  1
  [41] .debug_ranges     PROGBITS        0000000000000000 1425c48a0 8857f890 00      0   0 16
  [42] .debug_addr       PROGBITS        0000000000000000 1cab44130 2bf15390 00      0   0  1
  [43] .GCC.command.line PROGBITS        0000000000000000 1f6a594c0 2db0cf8 01  MS  0   0  1
  [44] .debug_loc        PROGBITS        0000000000000000 1f980a1b8 d81682 00      0   0  1
  [45] .debug_frame      PROGBITS        0000000000000000 1fa58b840 000060 00      0   0  8
  [46] .note.stapsdt     NOTE            0000000000000000 1fa58b8a0 00057c 00      0   0  4
  [47] .debug_macinfo    PROGBITS        0000000000000000 1fa58be1c fb2ea4 00      0   0  1
  [48] .gdb_index        PROGBITS        0000000000000000 1fb53ecc0 3142da3e 00      0   0  1
  [49] .symtab           SYMTAB          0000000000000000 22c96c700 28dba88 18     51 591036  8
  [50] .shstrtab         STRTAB          0000000000000000 22f248188 00022e 00      0   0  1
  [51] .strtab           STRTAB          0000000000000000 22f2483b6 ebe91ca 00      0   0  1

Added llvm::TimeTraceScope timeScope("relaxOnce"); in bool X86_64::relaxOnce(int pass) function.
Time trace shows skipping non-SHF_ALLOC sections omits the relaxOnce() cost, ~141ms in this case.

Attached time-trace files:
base.time-trace.json
test.time-trace.json