[llvm-branch-commits] [lld] release/18.x: [ELF] Eliminate symbols demoted due to /DISCARD/ discarded sections (#85167) (PR #85266)

[via llvm-branch-commits]

https://github.com/llvmbot created https://github.com/llvm/llvm-project/pull/85266

Backport 8fe3e70e810b409dce36f6d415e86f0f9b1cf22d

Requested by: @nikic

>From 122ba9f100705213774cff2038db953ff8174d91 Mon Sep 17 00:00:00 2001
From: Fangrui Song <i at maskray.me>
Date: Thu, 14 Mar 2024 09:51:27 -0700
Subject: [PATCH] [ELF] Eliminate symbols demoted due to /DISCARD/ discarded
 sections (#85167)

#69295 demoted Defined symbols relative to discarded sections.
If such a symbol is unreferenced, the desired behavior is to
eliminate it from .symtab just like --gc-sections discarded
definitions.
Linux kernel's CONFIG_DEBUG_FORCE_WEAK_PER_CPU=y configuration expects
that the unreferenced `unused` is not emitted to .symtab
(https://github.com/ClangBuiltLinux/linux/issues/2006).

For relocations referencing demoted symbols, the symbol index restores
to 0 like older lld (`R_X86_64_64 0` in `discard-section.s`).

Fix #85048

(cherry picked from commit 8fe3e70e810b409dce36f6d415e86f0f9b1cf22d)
---
 lld/ELF/Writer.cpp                          |  3 +++
 lld/test/ELF/linkerscript/discard-section.s | 25 +++++++++++++--------
 2 files changed, 19 insertions(+), 9 deletions(-)

diff --git a/lld/ELF/Writer.cpp b/lld/ELF/Writer.cpp
index 6df43a34be013a..8a08b0fcc90dbc 100644
--- a/lld/ELF/Writer.cpp
+++ b/lld/ELF/Writer.cpp
@@ -261,6 +261,9 @@ static void demoteDefined(Defined &sym, DenseMap<SectionBase *, size_t> &map) {
   Undefined(sym.file, sym.getName(), binding, sym.stOther, sym.type,
             /*discardedSecIdx=*/map.lookup(sym.section))
       .overwrite(sym);
+  // Eliminate from the symbol table, otherwise we would leave an undefined
+  // symbol if the symbol is unreferenced in the absence of GC.
+  sym.isUsedInRegularObj = false;
 }
 
 // If all references to a DSO happen to be weak, the DSO is not added to
diff --git a/lld/test/ELF/linkerscript/discard-section.s b/lld/test/ELF/linkerscript/discard-section.s
index 24f3b2b73e991f..0bbebac59bb345 100644
--- a/lld/test/ELF/linkerscript/discard-section.s
+++ b/lld/test/ELF/linkerscript/discard-section.s
@@ -9,6 +9,9 @@
 # RUN: ld.lld -r -T a.lds a.o b.o -o a.ro 2>&1 | FileCheck %s --check-prefix=WARNING --implicit-check-not=warning:
 # RUN: llvm-readelf -r -s a.ro | FileCheck %s --check-prefix=RELOC
 
+# RUN: ld.lld -r --gc-sections -T a.lds a.o b.o -o a.gc.ro --no-fatal-warnings
+# RUN: llvm-readelf -r -s a.gc.ro | FileCheck %s --check-prefix=RELOC-GC
+
 # LOCAL:      error: relocation refers to a discarded section: .aaa
 # LOCAL-NEXT: >>> defined in a.o
 # LOCAL-NEXT: >>> referenced by a.o:(.bbb+0x0)
@@ -32,16 +35,18 @@
 # WARNING:      warning: relocation refers to a discarded section: .aaa
 # WARNING-NEXT: >>> referenced by a.o:(.rela.bbb+0x0)
 
+## GNU ld reports "defined in discarded secion" errors even in -r mode.
+## We set the symbol index to 0.
 # RELOC:      Relocation section '.rela.bbb' at offset {{.*}} contains 1 entries:
 # RELOC-NEXT:     Offset             Info             Type               Symbol's Value  Symbol's Name + Addend
 # RELOC-NEXT: 0000000000000000  0000000000000000 R_X86_64_NONE                             0
 # RELOC-EMPTY:
 # RELOC-NEXT: Relocation section '.rela.data' at offset {{.*}} contains 4 entries:
 # RELOC-NEXT:     Offset             Info             Type               Symbol's Value  Symbol's Name + Addend
-# RELOC-NEXT: 0000000000000000  0000000500000001 R_X86_64_64            0000000000000000 global + 0
-# RELOC-NEXT: 0000000000000008  0000000700000001 R_X86_64_64            0000000000000000 weak + 0
-# RELOC-NEXT: 0000000000000010  0000000600000001 R_X86_64_64            0000000000000000 weakref1 + 0
-# RELOC-NEXT: 0000000000000018  0000000800000001 R_X86_64_64            0000000000000000 weakref2 + 0
+# RELOC-NEXT: 0000000000000000  0000000000000001 R_X86_64_64                             0
+# RELOC-NEXT: 0000000000000008  0000000000000001 R_X86_64_64                             0
+# RELOC-NEXT: 0000000000000010  0000000000000001 R_X86_64_64                             0
+# RELOC-NEXT: 0000000000000018  0000000000000001 R_X86_64_64                             0
 
 # RELOC:      Num:    Value          Size Type    Bind   Vis      Ndx Name
 # RELOC-NEXT:   0: 0000000000000000     0 NOTYPE  LOCAL  DEFAULT  UND
@@ -49,23 +54,25 @@
 # RELOC-NEXT:   2: 0000000000000000     0 SECTION LOCAL  DEFAULT    2 .bbb
 # RELOC-NEXT:   3: 0000000000000000     0 SECTION LOCAL  DEFAULT    4 .data
 # RELOC-NEXT:   4: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT    1 _start
-# RELOC-NEXT:   5: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT   UND global
-# RELOC-NEXT:   6: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT   UND weakref1
-# RELOC-NEXT:   7: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT   UND weak
-# RELOC-NEXT:   8: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT   UND weakref2
 # RELOC-EMPTY:
 
+# RELOC-GC:   There are no relocations in this file.
+
 #--- a.s
 .globl _start
 _start:
 
 .section .aaa,"a"
-.globl global, weakref1
+.globl global, weakref1, unused
 .weak weak, weakref2
 global:
 weak:
 weakref1:
 weakref2:
+## Eliminate `unused` just like GC discarded definitions.
+## Linux kernel's CONFIG_DEBUG_FORCE_WEAK_PER_CPU=y configuration expects
+## that the unreferenced `unused` is not emitted to .symtab.
+unused:
   .quad 0
 
 .section .bbb,"aw"