[ELF] - Make defsym to work correctly with reserved symbols.

[grimar]

Summary

Fixes PR35744.

Bug happens because of different nature of reserved symbols.

We add them with value -1 and output section as target section:

auto Add = [](StringRef S, int64_t Pos) {
  return addOptionalRegular(S, Out::ElfHeader, Pos, STV_DEFAULT);
};
ElfSym::Etext1 = Add("etext", -1);

Then out code handles -1 as special case:

uint64_t SectionBase::getOffset(uint64_t Offset) const {
...
// For output sections we treat offset -1 as the end of the section.
return Offset == uint64_t(-1) ? OS->Size : Offset;

Though when we handle scripts and particulary expressions like
foo = etext + 1;
this might change special (-1) value and break things.

Patch fixes the issue.

[ruiu]

ELF/LinkerScript.cpp

74–78

In general, when you change existing code, you should update the comment instead of adding sentences to existing comment so that the comment doesn't look patchy. Comments shouldn't reflect the history of code change but just describe why we are doing this.

So, this is just that if an expression is absolute and has no alignment requirement, its value is simply the output value, no?

[grimar]

• Updated.

ELF/LinkerScript.cpp

74–78

Technically yes, but we are unable to use something like following here I think.
(I guess you mean that)

if (isAbsolute() && Alignment == 1)
  return Val;

Because, for example we have in early-assign-symbol.s the following script:

SECTIONS { aaa = ABSOLUTE(foo - 1) + 1; .text  : { *(.text*) } }

And we first evaluate expression and then convert it to absolute:

if (Tok == "ABSOLUTE") {
  Expr Inner = readParenExpr();
  return [=] {
    ExprValue I = Inner();
    I.ForceAbsolute = true;
    return I;
  };
}

Not sure we should change something here.

I rewrote the check though.

[ruiu]

LGTM

ELF/LinkerScript.cpp

 
 uint64_t ExprValue::getSecAddr() const {
   if (Sec)
     return Sec->getOffset(0) + getOutputSectionVA(Sec, Loc);
   return 0;
 }
 
 uint64_t ExprValue::getSectionOffset() const {
   // If the alignment is trivial, we don't have to compute the full
   // value to know the offset. This allows this function to succeed in
   // cases where the output section is not yet known.
-  if (Alignment == 1)
+  if (Alignment == 1 && (!Sec || !Sec->getOutputSection()))
     return Val;
   return getValue() - getSecAddr();
 }
 
 OutputSection *LinkerScript::createOutputSection(StringRef Name,
                                                  StringRef Location) {
   OutputSection *&SecRef = NameToOutputSection[Name];
   OutputSection *Sec;
   if (SecRef && SecRef->Location.empty()) {

test/ELF/defsym-reserved-syms.s

+# REQUIRES: x86
+# RUN: llvm-mc -filetype=obj -triple=x86_64-pc-linux %s -o %t.o
+# RUN: ld.lld -o %t %t.o --defsym=foo2=etext
+# RUN: llvm-readobj -t -s %t | FileCheck %s
+
+## Check 'foo2' value is equal to value of 'etext'.
+# CHECK:     Symbol {
+# CHECK:      Name: foo2
+# CHECK-NEXT:  Value: 0x[[VAL:.*]]
+# CHECK:     Symbol {
+# CHECK:      Name: etext
+# CHECK-NEXT:  Value: 0x[[VAL]]
+
+## Check 'foo2' value set correctly when using
+## reserved symbol 'etext' in expression.
+# RUN: ld.lld -o %t %t.o --defsym=foo2=etext+2
+# RUN: llvm-readobj -t -s %t | FileCheck %s --check-prefix=EXPR
+# EXPR:     Symbol {
+# EXPR:      Name: foo2
+# EXPR-NEXT:  Value: 0x201007
+# EXPR:     Symbol {
+# EXPR:      Name: etext
+# EXPR-NEXT:  Value: 0x201005
+
+.globl foo1
+ foo1 = 0x123
+
+.global _start
+_start:
+  movl $foo2, %edx