[WebAssembly] Allow absolute symbols in the linking section (symbol table) (llvm#67493)

Fixes a crash in `-Wl,-emit-relocs` where the linker was not able to
write linker-synthetic absolute symbols to the symbol table.

This change adds a new symbol flag (`WASM_SYMBOL_ABS`), which means that
the symbol's offset is absolute and not relative to a given segment.
Such symbols include `__stack_low` and `__stack_low`.

Note that wasm object files never contains such symbols, only binaries
linked with `-Wl,-emit-relocs`.

Fixes: llvm#67111

Author: sbc100

lld/test/wasm/emit-relocs.s

@@ -17,6 +17,8 @@ _start:
   drop
   i32.const foo
   drop
+  i32.const __stack_low
+  drop
   end_function
 
 .section .bss.data,"",@
@@ -54,3 +56,9 @@ foo:
 # CHECK-NEXT:         Name:            ret32
 # CHECK-NEXT:         Flags:           [ VISIBILITY_HIDDEN ]
 # CHECK-NEXT:         Function:        1
+# CHECK-NEXT:       - Index:           2
+# CHECK-NEXT:         Kind:            DATA
+# CHECK-NEXT:         Name:            __stack_low
+# CHECK-NEXT:         Flags:           [ VISIBILITY_HIDDEN, ABSOLUTE ]
+# CHECK-NEXT:         Offset:          1040
+# CHECK-NEXT:         Size:            0

lld/wasm/SymbolTable.cpp

@@ -233,7 +233,8 @@ DefinedData *SymbolTable::addOptionalDataSymbol(StringRef name,
   else if (!s || s->isDefined())
     return nullptr;
   LLVM_DEBUG(dbgs() << "addOptionalDataSymbol: " << name << "\n");
-  auto *rtn = replaceSymbol<DefinedData>(s, name, WASM_SYMBOL_VISIBILITY_HIDDEN);
+  auto *rtn = replaceSymbol<DefinedData>(
+      s, name, WASM_SYMBOL_VISIBILITY_HIDDEN | WASM_SYMBOL_ABSOLUTE);
   rtn->setVA(value);
   rtn->referenced = true;
   return rtn;
@@ -243,7 +244,8 @@ DefinedData *SymbolTable::addSyntheticDataSymbol(StringRef name,
                                                  uint32_t flags) {
   LLVM_DEBUG(dbgs() << "addSyntheticDataSymbol: " << name << "\n");
   assert(!find(name));
-  return replaceSymbol<DefinedData>(insertName(name).first, name, flags);
+  return replaceSymbol<DefinedData>(insertName(name).first, name,
+                                    flags | WASM_SYMBOL_ABSOLUTE);
 }
 
 DefinedGlobal *SymbolTable::addSyntheticGlobal(StringRef name, uint32_t flags,

lld/wasm/SyntheticSections.cpp

@@ -665,9 +665,14 @@ void LinkingSection::writeBody() {
       } else if (isa<DataSymbol>(sym)) {
         writeStr(sub.os, sym->getName(), "sym name");
         if (auto *dataSym = dyn_cast<DefinedData>(sym)) {
-          writeUleb128(sub.os, dataSym->getOutputSegmentIndex(), "index");
-          writeUleb128(sub.os, dataSym->getOutputSegmentOffset(),
-                       "data offset");
+          if (dataSym->segment) {
+            writeUleb128(sub.os, dataSym->getOutputSegmentIndex(), "index");
+            writeUleb128(sub.os, dataSym->getOutputSegmentOffset(),
+                         "data offset");
+          } else {
+            writeUleb128(sub.os, 0, "index");
+            writeUleb128(sub.os, dataSym->getVA(), "data offset");
+          }
           writeUleb128(sub.os, dataSym->getSize(), "data size");
         }
       } else {

llvm/include/llvm/BinaryFormat/Wasm.h

@@ -412,6 +412,7 @@ const unsigned WASM_SYMBOL_EXPORTED = 0x20;
 const unsigned WASM_SYMBOL_EXPLICIT_NAME = 0x40;
 const unsigned WASM_SYMBOL_NO_STRIP = 0x80;
 const unsigned WASM_SYMBOL_TLS = 0x100;
+const unsigned WASM_SYMBOL_ABSOLUTE = 0x200;
 
 #define WASM_RELOC(name, value) name = value,
 

llvm/lib/Object/WasmObjectFile.cpp

@@ -723,17 +723,21 @@ Error WasmObjectFile::parseLinkingSectionSymtab(ReadContext &Ctx) {
       Info.Name = readString(Ctx);
       if (IsDefined) {
         auto Index = readVaruint32(Ctx);
-        if (Index >= DataSegments.size())
-          return make_error<GenericBinaryError>("invalid data segment index",
-                                                object_error::parse_failed);
         auto Offset = readVaruint64(Ctx);
         auto Size = readVaruint64(Ctx);
-        size_t SegmentSize = DataSegments[Index].Data.Content.size();
-        if (Offset > SegmentSize)
-          return make_error<GenericBinaryError>(
-              "invalid data symbol offset: `" + Info.Name + "` (offset: " +
-                  Twine(Offset) + " segment size: " + Twine(SegmentSize) + ")",
-              object_error::parse_failed);
+        if (!(Info.Flags & wasm::WASM_SYMBOL_ABSOLUTE)) {
+          if (static_cast<size_t>(Index) >= DataSegments.size())
+            return make_error<GenericBinaryError>(
+                "invalid data segment index: " + Twine(Index),
+                object_error::parse_failed);
+          size_t SegmentSize = DataSegments[Index].Data.Content.size();
+          if (Offset > SegmentSize)
+            return make_error<GenericBinaryError>(
+                "invalid data symbol offset: `" + Info.Name +
+                    "` (offset: " + Twine(Offset) +
+                    " segment size: " + Twine(SegmentSize) + ")",
+                object_error::parse_failed);
+        }
         Info.DataRef = wasm::WasmDataReference{Index, Offset, Size};
       }
       break;

llvm/lib/ObjectYAML/WasmYAML.cpp

@@ -518,7 +518,9 @@ void MappingTraits<WasmYAML::SymbolInfo>::mapping(IO &IO,
     IO.mapRequired("Tag", Info.ElementIndex);
   } else if (Info.Kind == wasm::WASM_SYMBOL_TYPE_DATA) {
     if ((Info.Flags & wasm::WASM_SYMBOL_UNDEFINED) == 0) {
-      IO.mapRequired("Segment", Info.DataRef.Segment);
+      if ((Info.Flags & wasm::WASM_SYMBOL_ABSOLUTE) == 0) {
+        IO.mapRequired("Segment", Info.DataRef.Segment);
+      }
       IO.mapOptional("Offset", Info.DataRef.Offset, 0u);
       IO.mapRequired("Size", Info.DataRef.Size);
     }
@@ -573,6 +575,7 @@ void ScalarBitSetTraits<WasmYAML::SymbolFlags>::bitset(
   BCaseMask(EXPLICIT_NAME, EXPLICIT_NAME);
   BCaseMask(NO_STRIP, NO_STRIP);
   BCaseMask(TLS, TLS);
+  BCaseMask(ABSOLUTE, ABSOLUTE);
 #undef BCaseMask
 }