[ELF] Ensure output section is not discarded in addStartEndSymbols()

[andrewng]

Summary

Fixes https://bugs.llvm.org/show_bug.cgi?id=52534.

[andrewng]

This is potentially a "better" alternative to D114103.

[MaskRay]

I agree that this version is better to ensure addr(.preinit_array) = __preinit_array_start.

The R_X86_64_REX_GOTPCRELX issue is related to a limitation of our current framework.
https://maskray.me/blog/2020-12-19-lld-and-gnu-linker-incompatibilities
GNU ld seems to suppress optimization in the case.

lld/test/ELF/linkerscript/preinit-array-empty.test

9

Optional: -T is probably more common and conciser.

lld/ELF/LinkerScript.h

public:
 
   bool shouldKeep(InputSectionBase *s);
   const Defined *assignAddresses();
   void allocateHeaders(std::vector<PhdrEntry *> &phdrs);
   void processSectionCommands();
   void processSymbolAssignments();
   void declareSymbols();
 
+  bool isDiscarded(const OutputSection *sec) const;
+
   // Used to handle INSERT AFTER statements.
   void processInsertCommands();
 
   // SECTIONS command list.
   std::vector<BaseCommand *> sectionCommands;
 
   // PHDRS command list.
   std::vector<PhdrsCommand> phdrsCommands;

lld/ELF/LinkerScript.cpp

if (!(sec->flags & SHF_ALLOC)) {
     dot = savedDot;
   } else if (isTbss) {
     // NOBITS TLS sections are similar. Additionally save the end address.
     ctx->tbssAddr = dot;
     dot = savedDot;
   }
 }
 
-static bool isDiscardable(OutputSection &sec) {
+static bool isDiscardable(const OutputSection &sec) {
   if (sec.name == "/DISCARD/")
     return true;
 
   // We do not want to remove OutputSections with expressions that reference
   // symbols even if the OutputSection is empty. We want to ensure that the
   // expressions can be evaluated and report an error if they cannot.
   if (sec.expressionsUseSymbols)
     return false;
if (auto cmd = dyn_cast<SymbolAssignment>(base))
         continue;
 
     if (!isa<InputSectionDescription>(*base))
       return false;
   }
   return true;
 }
 
+bool LinkerScript::isDiscarded(const OutputSection *sec) const {
+  return hasSectionsCommand && (getFirstInputSection(sec) == nullptr) &&
+         isDiscardable(*sec);
+}
+
 static void maybePropagatePhdrs(OutputSection &sec,
                                 std::vector<StringRef> &phdrs) {
   if (sec.phdrs.empty()) {
     // To match the bfd linker script behaviour, only propagate program
     // headers to sections that are allocated.
     if (sec.flags & SHF_ALLOC)
       sec.phdrs = phdrs;
   } else {

lld/ELF/Writer.cpp

template <class ELFT> void Writer<ELFT>::addStartEndSymbols() {
   //
   // In rare situations, the .text section may not exist. If that's the
   // case, use the image base address as a last resort.
   OutputSection *Default = findSection(".text");
   if (!Default)
     Default = Out::elfHeader;
 
   auto define = [=](StringRef start, StringRef end, OutputSection *os) {
-    if (os) {
+    if (os && !script->isDiscarded(os)) {
       addOptionalRegular(start, os, 0);
       addOptionalRegular(end, os, -1);
     } else {
       addOptionalRegular(start, Default, 0);
       addOptionalRegular(end, Default, 0);
     }
   };
 

lld/test/ELF/linkerscript/preinit-array-empty.test

+# REQUIRES: x86
+# RUN: split-file %s %t
+# RUN: llvm-mc -filetype=obj -triple=x86_64 %t/t.s -o %t.o
+
+## PR52534: https://bugs.llvm.org/show_bug.cgi?id=52534
+## Check case where .preinit_array is discarded.
+## Link should succeed without causing an out of range relocation error.
+# RUN: ld.lld -T %t/discarded.script %t.o -o %t1 --image-base=0x80000000
+# RUN: llvm-readelf -s %t1 | FileCheck --check-prefixes=CHECK,DISCARDED %s
+
+## Check case where .preinit_array is emitted but empty.
+# RUN: ld.lld -T %t/empty.script %t.o -o %t2
+# RUN: llvm-readelf -s %t2 | FileCheck --check-prefixes=CHECK,EMPTY %s
+
+# CHECK:          [[#%x,ADDR:]] 0 NOTYPE  LOCAL  HIDDEN      [[#]] __preinit_array_start
+# CHECK-NEXT:     [[#ADDR]]     0 NOTYPE  LOCAL  HIDDEN      [[#]] __preinit_array_end
+
+# DISCARDED-NEXT: [[#ADDR]]     0 NOTYPE  GLOBAL DEFAULT     [[#]] _start
+
+# EMPTY-NOT:      [[#ADDR]]     0 NOTYPE  GLOBAL DEFAULT     [[#]] _start
+# EMPTY:          [[#ADDR]]     0 NOTYPE  GLOBAL DEFAULT     [[#]] ADDR
+
+#--- t.s
+.global _start
+_start:
+ movq __preinit_array_start@GOTPCREL(%rip),%rax
+ movq __preinit_array_end@GOTPCREL(%rip),%rax
+
+#--- discarded.script
+SECTIONS {
+  .text : { *(.text); }
+  .preinit_array : { *(.preinit_array); }
+}
+
+#--- empty.script
+SECTIONS {
+  .text : { *(.text); }
+  .preinit_array : { ADDR = .; *(.preinit_array); }
+}