[MCJIT] Make MCJIT honor symbol visibility settings when populating the global

symbol table.

Patch by Anthony Pesch. Thanks Anthony!



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220245 91177308-0d34-0410-b5e6-96231b3b80d8

include/llvm/Object/ELFObjectFile.h

@@ -177,6 +177,19 @@ protected:
     return DRI;
   }
 
+  bool isExportedToOtherDSO(const Elf_Sym *ESym) const {
+    unsigned char Binding = ESym->getBinding();
+    unsigned char Visibility = ESym->getVisibility();
+
+    if (Binding != ELF::STB_GLOBAL && Binding != ELF::STB_WEAK)
+      return false;
+
+    if (Visibility != ELF::STV_DEFAULT && Visibility != ELF::STV_PROTECTED)
+      return false;
+
+    return true;
+  }
+
   // This flag is used for classof, to distinguish ELFObjectFile from
   // its subclass. If more subclasses will be created, this flag will
   // have to become an enum.
@@ -373,6 +386,9 @@ uint32_t ELFObjectFile<ELFT>::getSymbolFlags(DataRefImpl Symb) const {
       EF.getSymbolTableIndex(ESym) == ELF::SHN_COMMON)
     Result |= SymbolRef::SF_Common;
 
+  if (isExportedToOtherDSO(ESym))
+    Result |= SymbolRef::SF_Exported;
+
   return Result;
 }
 

include/llvm/Object/ELFTypes.h

@@ -158,7 +158,7 @@ struct Elf_Sym_Base<ELFType<TargetEndianness, MaxAlign, false> > {
   Elf_Addr st_value;      // Value or address associated with the symbol
   Elf_Word st_size;       // Size of the symbol
   unsigned char st_info;  // Symbol's type and binding attributes
-  unsigned char st_other; // Must be zero; reserved
+  unsigned char st_other; // Visibility in the lower 2 bits, the rest is zero
   Elf_Half st_shndx;      // Which section (header table index) it's defined in
 };
 
@@ -167,7 +167,7 @@ struct Elf_Sym_Base<ELFType<TargetEndianness, MaxAlign, true> > {
   LLVM_ELF_IMPORT_TYPES(TargetEndianness, MaxAlign, true)
   Elf_Word st_name;       // Symbol name (index into string table)
   unsigned char st_info;  // Symbol's type and binding attributes
-  unsigned char st_other; // Must be zero; reserved
+  unsigned char st_other; // Visibility in the lower 2 bits, the rest is zero
   Elf_Half st_shndx;      // Which section (header table index) it's defined in
   Elf_Addr st_value;      // Value or address associated with the symbol
   Elf_Xword st_size;      // Size of the symbol
@@ -176,16 +176,21 @@ struct Elf_Sym_Base<ELFType<TargetEndianness, MaxAlign, true> > {
 template <class ELFT>
 struct Elf_Sym_Impl : Elf_Sym_Base<ELFT> {
   using Elf_Sym_Base<ELFT>::st_info;
+  using Elf_Sym_Base<ELFT>::st_other;
 
   // These accessors and mutators correspond to the ELF32_ST_BIND,
   // ELF32_ST_TYPE, and ELF32_ST_INFO macros defined in the ELF specification:
   unsigned char getBinding() const { return st_info >> 4; }
   unsigned char getType() const { return st_info & 0x0f; }
+  unsigned char getVisibility() const { return st_other & 0x3; }
   void setBinding(unsigned char b) { setBindingAndType(b, getType()); }
   void setType(unsigned char t) { setBindingAndType(getBinding(), t); }
   void setBindingAndType(unsigned char b, unsigned char t) {
     st_info = (b << 4) + (t & 0x0f);
   }
+  void setVisibility(unsigned char v) {
+    st_other = (st_other & ~0x3) | (v & 0x3);
+  }
 };
 
 /// Elf_Versym: This is the structure of entries in the SHT_GNU_versym section

include/llvm/Object/SymbolicFile.h

@@ -87,9 +87,10 @@ public:
     SF_Absolute = 1U << 3,       // Absolute symbol
     SF_Common = 1U << 4,         // Symbol has common linkage
     SF_Indirect = 1U << 5,       // Symbol is an alias to another symbol
-    SF_FormatSpecific = 1U << 6, // Specific to the object file format
+    SF_Exported = 1U << 6,       // Symbol is visible to other DSOs
+    SF_FormatSpecific = 1U << 7, // Specific to the object file format
                                  // (e.g. section symbols)
-    SF_Thumb = 1U << 7           // Thumb symbol in a 32-bit ARM binary
+    SF_Thumb = 1U << 8,          // Thumb symbol in a 32-bit ARM binary
   };
 
   BasicSymbolRef() : OwningObject(nullptr) { }

include/llvm/Support/ELF.h

@@ -1458,25 +1458,29 @@ struct Elf32_Sym {
   Elf32_Addr    st_value; // Value or address associated with the symbol
   Elf32_Word    st_size;  // Size of the symbol
   unsigned char st_info;  // Symbol's type and binding attributes
-  unsigned char st_other; // Must be zero; reserved
+  unsigned char st_other; // Visibility in the lower 2 bits, the rest is zero
   Elf32_Half    st_shndx; // Which section (header table index) it's defined in
 
   // These accessors and mutators correspond to the ELF32_ST_BIND,
   // ELF32_ST_TYPE, and ELF32_ST_INFO macros defined in the ELF specification:
   unsigned char getBinding() const { return st_info >> 4; }
   unsigned char getType() const { return st_info & 0x0f; }
+  unsigned char getVisibility() const { return st_other & 0x3; }
   void setBinding(unsigned char b) { setBindingAndType(b, getType()); }
   void setType(unsigned char t) { setBindingAndType(getBinding(), t); }
   void setBindingAndType(unsigned char b, unsigned char t) {
     st_info = (b << 4) + (t & 0x0f);
   }
+  void setVisibility(unsigned char v) {
+    st_other = (st_other & ~0x3) | (v & 0x3);
+  }
 };
 
 // Symbol table entries for ELF64.
 struct Elf64_Sym {
   Elf64_Word      st_name;  // Symbol name (index into string table)
   unsigned char   st_info;  // Symbol's type and binding attributes
-  unsigned char   st_other; // Must be zero; reserved
+  unsigned char   st_other; // Visibility in the lower 2 bits, the rest is zero
   Elf64_Half      st_shndx; // Which section (header tbl index) it's defined in
   Elf64_Addr      st_value; // Value or address associated with the symbol
   Elf64_Xword     st_size;  // Size of the symbol
@@ -1485,11 +1489,15 @@ struct Elf64_Sym {
   // symbol table entries.
   unsigned char getBinding() const { return st_info >> 4; }
   unsigned char getType() const { return st_info & 0x0f; }
+  unsigned char getVisibility() const { return st_other & 0x3; }
   void setBinding(unsigned char b) { setBindingAndType(b, getType()); }
   void setType(unsigned char t) { setBindingAndType(getBinding(), t); }
   void setBindingAndType(unsigned char b, unsigned char t) {
     st_info = (b << 4) + (t & 0x0f);
   }
+  void setVisibility(unsigned char v) {
+    st_other = (st_other & ~0x3) | (v & 0x3);
+  }
 };
 
 // The size (in bytes) of symbol table entries.

lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp

@@ -205,10 +205,14 @@ RuntimeDyldImpl::loadObject(std::unique_ptr<ObjectImage> Obj) {
         bool IsCode = SI->isText();
         unsigned SectionID =
             findOrEmitSection(*Obj, *SI, IsCode, LocalSections);
+        // Add the symbol to the local symbol table for this module.
         LocalSymbols[Name.data()] = SymbolLoc(SectionID, SectOffset);
         DEBUG(dbgs() << "\tOffset: " << format("%p", (uintptr_t)SectOffset)
                      << " flags: " << Flags << " SID: " << SectionID);
-        GlobalSymbolTable[Name] = SymbolLoc(SectionID, SectOffset);
+        // If exported, add to the global symbol table for other modules to also link in.
+        if (Flags & SymbolRef::SF_Exported) {
+          GlobalSymbolTable[Name] = SymbolLoc(SectionID, SectOffset);
+        }
       }
     }
     DEBUG(dbgs() << "\tType: " << SymType << " Name: " << Name << "\n");

lib/Object/MachOObjectFile.cpp

@@ -454,6 +454,9 @@ uint32_t MachOObjectFile::getSymbolFlags(DataRefImpl DRI) const {
       if (Value && Value != UnknownAddressOrSize)
         Result |= SymbolRef::SF_Common;
     }
+
+    if (!(MachOType & MachO::N_PEXT))
+      Result |= SymbolRef::SF_Exported;
   }
 
   if (MachOFlags & (MachO::N_WEAK_REF | MachO::N_WEAK_DEF))

test/ExecutionEngine/RuntimeDyld/AArch64/MachO_ARM64_relocations.s

@@ -18,6 +18,7 @@ _foo:
 # and therefore ignored.
 # rtdyld-check:  decode_operand(br1, 0)[25:0] = (_foo - br1)[27:2]
 _test_branch_reloc:
+    .globl  br1
 br1:
     b _foo
     ret
@@ -32,8 +33,10 @@ br1:
     .globl  _test_adrp_ldr
     .align  2
 _test_adrp_ldr:
+    .globl  adrp1
 adrp1:
     adrp x0, _ptr@PAGE
+    .globl  ldr1
 ldr1:
     ldr  x0, [x0, _ptr@PAGEOFF]
     ret
@@ -49,8 +52,10 @@ ldr1:
     .globl  _test_adrp_ldr
     .align  2
 _test_got_adrp_ldr:
+    .globl  adrp2
 adrp2:
     adrp x0, _ptr@GOTPAGE
+    .globl  ldr2
 ldr2:
     ldr  x0, [x0, _ptr@GOTPAGEOFF]
     ret

test/ExecutionEngine/RuntimeDyld/ARM/MachO_ARM_PIC_relocations.s

@@ -8,12 +8,15 @@
 bar:
 # Check lower 16-bits of section difference relocation
 # rtdyld-check: decode_operand(insn1, 1) = (foo$non_lazy_ptr-(nextPC+8))[15:0]
+        .globl  insn1
 insn1:
         movw    r0, :lower16:(foo$non_lazy_ptr-(nextPC+8))
 # Check upper 16-bits of section difference relocation
 # rtdyld-check: decode_operand(insn2, 2) = (foo$non_lazy_ptr-(nextPC+8))[31:16]
+        .globl  insn2
 insn2:
         movt    r0, :upper16:(foo$non_lazy_ptr-(nextPC+8))
+        .globl  nextPC
 nextPC:
         add     r1, r0, r0
 
@@ -25,6 +28,7 @@ nextPC:
 # rtdyld-check: *{4}(stub_addr(foo.o, __text, baz) + 4) = baz
 #
 # rtdyld-check: decode_operand(insn3, 0) = stub_addr(foo.o, __text, baz) - (insn3 + 8)
+        .globl  insn3
 insn3:
         bl      baz
 
@@ -44,6 +48,7 @@ insn4:
 # rtdyld-check: *{4}foo$non_lazy_ptr = foo
         .section	__DATA,__nl_symbol_ptr,non_lazy_symbol_pointers
   	.align	2
+        .globl  foo$non_lazy_ptr
 foo$non_lazy_ptr:
 	.indirect_symbol	foo
 	.long	0

test/ExecutionEngine/RuntimeDyld/X86/MachO_i386_DynNoPIC_relocations.s

@@ -6,16 +6,19 @@
 	.align	4, 0x90
 bar:
 	calll	tmp0$pb
+	.globl  tmp0$pb
 tmp0$pb:
 	popl	%eax
 # Test section difference relocation to non-lazy ptr section.
 # rtdyld-check: decode_operand(inst1, 4) = x$non_lazy_ptr - tmp0$pb
+	.globl  inst1
 inst1:
 	movl	x$non_lazy_ptr-tmp0$pb(%eax), %eax
         movl    (%eax), %ebx
 
 # Test VANILLA relocation to jump table.
 # rtdyld-check: decode_operand(inst2, 0) = bling$stub - next_pc(inst2)
+	.globl  inst2
 inst2:
         calll	bling$stub
         addl    %ebx, %eax
@@ -27,11 +30,13 @@ inst3:
 	retl
 
 	.section	__IMPORT,__jump_table,symbol_stubs,pure_instructions+self_modifying_code,5
+	.globl  bling$stub
 bling$stub:
 	.indirect_symbol	bling
 	.ascii	"\364\364\364\364\364"
 
 	.section	__IMPORT,__pointers,non_lazy_symbol_pointers
+	.globl  x$non_lazy_ptr
 x$non_lazy_ptr:
 	.indirect_symbol	x
 	.long	0

test/ExecutionEngine/RuntimeDyld/X86/MachO_x86-64_PIC_relocations.s

@@ -12,11 +12,13 @@ foo:
 main:
 # Test PC-rel branch.
 # rtdyld-check: decode_operand(insn1, 0) = foo - next_pc(insn1)
+	.globl  insn1
 insn1:
         callq	foo
 
 # Test PC-rel signed.
 # rtdyld-check: decode_operand(insn2, 4) = x - next_pc(insn2)
+	.globl  insn2
 insn2:
 	movl	x(%rip), %eax
 
@@ -25,6 +27,7 @@ insn2:
 # references the correct GOT entry address:
 # rtdyld-check: *{8}(stub_addr(test_x86-64.o, __text, y)) = y
 # rtdyld-check: decode_operand(insn3, 4) = stub_addr(test_x86-64.o, __text, y) - next_pc(insn3)
+	.globl  insn3
 insn3:
         movq	y@GOTPCREL(%rip), %rax