IRSymTab: Record _GLOBAL_OFFSET_TABLE_ for ELF x86

In ELF, relocatable files generated for x86-32 and some code models of
x86-64 (medium, large) may reference the special symbol
`_GLOBAL_OFFSET_TABLE_` that is not used in the IR. In an LTO link, if
there is no regular relocatable file referencing the special symbol, the
linker may not define the symbol and lead to a spurious "undefined
symbol" error.

Fix #61101: record that `_GLOBAL_OFFSET_TABLE_` is used in the IR symbol
table.

Note: The `PreservedSymbols` mechanism
(https://reviews.llvm.org/D112595) that just sets `FB_used` is not
applicable.
The `getRuntimeLibcallSymbols` for extracting lazy runtime library
symbols is for symbols that are "always" potentially used, but linkers
don't have the code model information to make a precise decision.

Pull Request: #89463

lld/test/ELF/lto/i386-global-offset-table.ll

@@ -0,0 +1,29 @@
+; REQUIRES: x86
+;; LTO-generated relocatable files may reference _GLOBAL_OFFSET_TABLE_ while
+;; the IR does not mention _GLOBAL_OFFSET_TABLE_.
+;; Test that there is no spurious "undefined symbol" error.
+
+; RUN: rm -rf %t && mkdir %t && cd %t
+; RUN: llvm-as %s -o a.bc
+; RUN: ld.lld -pie a.bc -o a
+; RUN: llvm-nm a | FileCheck %s
+
+; CHECK: d _GLOBAL_OFFSET_TABLE_
+
+target datalayout = "e-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:64-i128:128-f64:32:64-f80:32-n8:16:32-S128"
+target triple = "i386-pc-linux-gnu"
+
+@i = global i32 0
+
+define dso_local void @_start() {
+entry:
+  %0 = load i32, ptr @i
+  %inc = add nsw i32 %0, 1
+  store i32 %inc, ptr @i
+  ret void
+}
+
+!llvm.module.flags = !{!0, !1}
+
+!0 = !{i32 8, !"PIC Level", i32 2}
+!1 = !{i32 7, !"PIE Level", i32 2}

lld/test/ELF/lto/x86-64-global-offset-table.ll

@@ -0,0 +1,71 @@
+; REQUIRES: x86
+;; LTO-generated relocatable files may reference _GLOBAL_OFFSET_TABLE_ while
+;; the IR does not mention _GLOBAL_OFFSET_TABLE_.
+;; Test that there is no spurious "undefined symbol" error.
+
+; RUN: rm -rf %t && split-file %s %t && cd %t
+; RUN: opt -module-summary b.ll -o b.bc
+
+;; Test Thin LTO.
+; RUN: cat a.ll medium.ll | opt -module-summary - -o medium.bc
+; RUN: ld.lld -pie --no-relax medium.bc b.bc -o medium
+; RUN: llvm-objdump -dt medium | FileCheck %s
+
+;; Test regular LTO.
+; RUN: cat a.ll large.ll | llvm-as - -o large.bc
+; RUN: ld.lld -pie large.bc b.bc -o large
+; RUN: llvm-objdump -dt large | FileCheck %s
+
+;; Explicit reference of _GLOBAL_OFFSET_TABLE_ is fine.
+; RUN: cat a.ll medium.ll ref.ll | opt -module-summary - -o ref.bc
+; RUN: ld.lld -pie -u ref ref.bc b.bc -y _GLOBAL_OFFSET_TABLE_ -o ref 2>&1 | FileCheck %s --check-prefix=TRACE
+; RUN: llvm-objdump -dt ref | FileCheck %s
+
+; TRACE:      ref.bc: reference to _GLOBAL_OFFSET_TABLE_
+; TRACE-NEXT: ref.bc: reference to _GLOBAL_OFFSET_TABLE_
+; TRACE-NEXT: <internal>: definition of _GLOBAL_OFFSET_TABLE_
+; TRACE-NEXT: ref.lto.ref.o: reference to _GLOBAL_OFFSET_TABLE_
+
+;; The IR symbol table references _GLOBAL_OFFSET_TABLE_, which causes lld to define the symbol.
+; CHECK: .got.plt       0000000000000000 .hidden _GLOBAL_OFFSET_TABLE_
+; CHECK: movabsq
+
+;--- a.ll
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+@i = external global i32
+
+define dso_local void @_start() {
+entry:
+  %0 = load i32, ptr @i
+  %inc = add nsw i32 %0, 1
+  store i32 %inc, ptr @i
+  ret void
+}
+
+!llvm.module.flags = !{!0, !1, !2, !3}
+
+!0 = !{i32 8, !"PIC Level", i32 2}
+!1 = !{i32 7, !"PIE Level", i32 2}
+!2 = !{i32 1, !"Large Data Threshold", i64 0}
+
+;--- medium.ll
+!3 = !{i32 1, !"Code Model", i32 3}
+
+;--- large.ll
+!3 = !{i32 1, !"Code Model", i32 4}
+
+;--- ref.ll
+@_GLOBAL_OFFSET_TABLE_ = external global [0 x i8]
+
+define dso_local ptr @ref() {
+entry:
+  ret ptr @_GLOBAL_OFFSET_TABLE_
+}
+
+;--- b.ll
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+@i = global i32 0

llvm/lib/Object/ModuleSymbolTable.cpp

@@ -175,6 +175,20 @@ void ModuleSymbolTable::CollectAsmSymbols(
       AsmSymbol(Key, BasicSymbolRef::Flags(Res));
     }
   });
+
+  // In ELF, object code generated for x86-32 and some code models of x86-64 may
+  // reference the special symbol _GLOBAL_OFFSET_TABLE_ that is not used in the
+  // IR. Record it like inline asm symbols.
+  Triple TT(M.getTargetTriple());
+  if (!TT.isOSBinFormatELF() || !TT.isX86())
+    return;
+  auto CM = M.getCodeModel();
+  if (TT.getArch() == Triple::x86 || CM == CodeModel::Medium ||
+      CM == CodeModel::Large) {
+    AsmSymbol("_GLOBAL_OFFSET_TABLE_",
+              BasicSymbolRef::Flags(BasicSymbolRef::SF_Undefined |
+                                    BasicSymbolRef::SF_Global));
+  }
 }
 
 void ModuleSymbolTable::CollectAsmSymvers(

llvm/test/LTO/X86/codemodel-2.ll

@@ -1,5 +1,5 @@
 ; RUN: llvm-as %s -o %t.o
-; RUN: llvm-lto2 run -r %t.o,_start,px %t.o -o %t.s
+; RUN: llvm-lto2 run -r %t.o,_start,px -r %t.o,_GLOBAL_OFFSET_TABLE_, %t.o -o %t.s
 ; RUN: llvm-objdump --no-print-imm-hex -d %t.s.0 | FileCheck %s --check-prefix=CHECK-LARGE
 
 target triple = "x86_64-unknown-linux-gnu"

llvm/test/LTO/X86/codemodel-3.ll

@@ -1,6 +1,7 @@
 ; RUN: llvm-as %s -o %t0.o
 ; RUN: llvm-as < %p/Inputs/codemodel-3.ll > %t1.o
-; RUN: not llvm-lto2 run -r %t0.o,_start,px -r %t1.o,bar,px %t0.o %t1.o -o %t2.s 2>&1 | FileCheck %s 
+; RUN: not llvm-lto2 run -r %t0.o,_start,px -r %t1.o,bar,px -r %t0.o,_GLOBAL_OFFSET_TABLE_, \
+; RUN:   -r %t1.o,_GLOBAL_OFFSET_TABLE_, %t0.o %t1.o -o %t2.s 2>&1 | FileCheck %s 
 
 target triple = "x86_64-unknown-linux-gnu"
 target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"

llvm/test/LTO/X86/largedatathreshold-1.ll

@@ -1,5 +1,5 @@
 ; RUN: llvm-as %s -o %t.o
-; RUN: llvm-lto2 run -r %t.o,_start,px %t.o -o %t.s
+; RUN: llvm-lto2 run -r %t.o,_start,px -r %t.o,_GLOBAL_OFFSET_TABLE_, %t.o -o %t.s
 ; RUN: llvm-objdump -d %t.s.0 | FileCheck %s
 
 target triple = "x86_64-unknown-linux-gnu"

llvm/test/LTO/X86/largedatathreshold-2.ll

@@ -1,5 +1,5 @@
 ; RUN: llvm-as %s -o %t.o
-; RUN: llvm-lto2 run -r %t.o,_start,px %t.o -o %t.s
+; RUN: llvm-lto2 run -r %t.o,_start,px -r %t.o,_GLOBAL_OFFSET_TABLE_, %t.o -o %t.s
 ; RUN: llvm-objdump -d %t.s.0 | FileCheck %s
 
 target triple = "x86_64-unknown-linux-gnu"

llvm/test/LTO/X86/largedatathreshold-3.ll

@@ -1,6 +1,7 @@
 ; RUN: llvm-as %s -o %t0.o
 ; RUN: llvm-as < %p/Inputs/largedatathreshold.ll > %t1.o
-; RUN: not llvm-lto2 run -r %t0.o,_start,px -r %t1.o,bar,px %t0.o %t1.o -o %t2.s 2>&1 | FileCheck %s
+; RUN: not llvm-lto2 run -r %t0.o,_start,px -r %t1.o,bar,px -r %t0.o,_GLOBAL_OFFSET_TABLE_, \
+; RUN:   -r %t1.o,_GLOBAL_OFFSET_TABLE_, %t0.o %t1.o -o %t2.s 2>&1 | FileCheck %s
 
 ; CHECK: 'Large Data Threshold': IDs have conflicting values