[JITLink][ELF/AARCH64] Implement Procedure Linkage Table

[sunho]

Summary

This patch implements Procedure Linkage Table (PLT) for ELF/AARCH64. The aarch64 linux calling convention also uses r16 as the intra-procedure-call scratch register same as MachO/ARM64. We can use the same stub sequence for this reason.

Also, BR regiseter doesn't touch X30 register. External function call by BL instruction (touched by CALL26 relocation) will set X30 to the original PC + 4, which is the intended behavior. External function call by B instruction (touched by JUMP26 relocation) doesn't requite to set X30, so the patch will be correct in this case too.

Reference: https://github.com/ARM-software/abi-aa/blob/main/aapcs64/aapcs64.rst#611general-purpose-registers

[sunho]

Rebase

[lhames]

LGTM. Thanks @sunho!

llvm/lib/ExecutionEngine/JITLink/ELF_aarch64.cpp

if (E.getKind() == aarch64::GOTPage21) {
       E.setTarget(GOTEntry);
     } else if (E.getKind() == aarch64::GOTPageOffset12) {
       E.setKind(aarch64::PageOffset12);
       E.setTarget(GOTEntry);
     } else
       llvm_unreachable("Not a GOT edge?");
   }
 
-  bool isExternalBranchEdge(Edge &E) { return false; }
-
-  Symbol &createPLTStub(Symbol &Target) { assert(false && "unimplemetned"); }
+  bool isExternalBranchEdge(Edge &E) {
+    return E.getKind() == aarch64::Branch26 && !E.getTarget().isDefined();
+  }
 
-  void fixPLTEdge(Edge &E, Symbol &Stub) { assert(false && "unimplemetned"); }
+  Symbol &createPLTStub(Symbol &Target) {
+    auto &StubContentBlock = G.createContentBlock(
+        getStubsSection(), getStubBlockContent(), orc::ExecutorAddr(), 1, 0);
+    // Re-use GOT entries for stub targets.
+    auto &GOTEntrySymbol = getGOTEntry(Target);
+    StubContentBlock.addEdge(aarch64::LDRLiteral19, 0, GOTEntrySymbol, 0);
+    return G.addAnonymousSymbol(StubContentBlock, 0, 8, true, false);
+  }
+
+  void fixPLTEdge(Edge &E, Symbol &Stub) {
+    assert(E.getKind() == aarch64::Branch26 && "Not a Branch26 edge?");
+    assert(E.getAddend() == 0 && "Branch26 edge has non-zero addend?");
+    E.setTarget(Stub);
+  }
 
 private:
   Section &getGOTSection() {
     if (!GOTSection)
       GOTSection = &G.createSection("$__GOT", MemProt::Read | MemProt::Exec);
     return *GOTSection;
   }
 
+  Section &getStubsSection() {
+    if (!StubsSection)
+      StubsSection =
+          &G.createSection("$__STUBS", MemProt::Read | MemProt::Exec);
+    return *StubsSection;
+  }
+
   ArrayRef<char> getGOTEntryBlockContent() {
     return {reinterpret_cast<const char *>(NullGOTEntryContent),
             sizeof(NullGOTEntryContent)};
   }
 
+  ArrayRef<char> getStubBlockContent() {
+    return {reinterpret_cast<const char *>(StubContent), sizeof(StubContent)};
+  }
+
   static const uint8_t NullGOTEntryContent[8];
+  static const uint8_t StubContent[8];
   Section *GOTSection = nullptr;
+  Section *StubsSection = nullptr;
 };
 
 const uint8_t PerGraphGOTAndPLTStubsBuilder_ELF_arm64::NullGOTEntryContent[8] =
     {0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00};
+const uint8_t PerGraphGOTAndPLTStubsBuilder_ELF_arm64::StubContent[8] = {
+    0x10, 0x00, 0x00, 0x58, // LDR x16, <literal>
+    0x00, 0x02, 0x1f, 0xd6  // BR  x16
+};
 
 Expected<std::unique_ptr<LinkGraph>>
 createLinkGraphFromELFObject_aarch64(MemoryBufferRef ObjectBuffer) {
   LLVM_DEBUG({
     dbgs() << "Building jitlink graph for new input "
            << ObjectBuffer.getBufferIdentifier() << "...\n";
   });
 

llvm/test/ExecutionEngine/JITLink/AArch64/ELF_aarch64_relocations.s

 # RUN: rm -rf %t && mkdir -p %t
 # RUN: llvm-mc -triple=aarch64-unknown-linux-gnu -relax-relocations=false -position-independent -filetype=obj -o %t/elf_reloc.o %s
 # RUN: llvm-jitlink -noexec \
 # RUN:              -abs external_data=0xdeadbeef \
+# RUN:              -abs external_func=0xcafef00d \
 # RUN:              -check %s %t/elf_reloc.o
 
         .text
 
         .globl  main
         .p2align  2
         .type main,@function
 main:
test_adr_prel:
         .size test_adr_prel, .-test_adr_prel
 
         .globl  test_add_abs_lo12
         .p2align  2
 test_add_abs_lo12:
         add x0, x0, :lo12:named_data
         .size test_add_abs_lo12, .-test_add_abs_lo12
 
+# Check that calls/jumps to external functions trigger the generation of stubs and GOT
+# entries.
+#
+# jitlink-check: decode_operand(test_external_call, 0) = (stub_addr(elf_reloc.o, external_func) - test_external_call)[27:2]
+# jitlink-check: decode_operand(test_external_jump, 0) = (stub_addr(elf_reloc.o, external_func) - test_external_jump)[27:2]
+# jitlink-check: *{8}(got_addr(elf_reloc.o, external_func)) = external_func
+        .globl  test_external_call
+        .p2align  2
+test_external_call:
+        bl   external_func
+        .size test_external_call, .-test_external_call
+
+       .globl  test_external_jump
+        .p2align  2
+test_external_jump:
+        b   external_func
+        .size test_external_jump, .-test_external_jump
+
 # Check R_AARCH64_LDST*_ABS_LO12_NC relocation of a local symbol
 #
 # The immediate value should be the symbol address right shifted according to its instruction bitwidth.
 #
 # jitlink-check: decode_operand(test_ldrb, 2) = named_data[11:0]
 # jitlink-check: decode_operand(test_ldrsb, 2) = (named_data + 0)[11:0]
 # jitlink-check: decode_operand(test_ldrh, 2) = (named_data + 0)[11:1]
 # jitlink-check: decode_operand(test_ldrsh, 2) = (named_data + 0)[11:1]

 #
 # jitlink-check: *{8}local_func_addr_quad = named_func
         .globl  local_func_addr_quad
         .p2align  3
 local_func_addr_quad:
         .xword  named_func
         .size local_func_addr_quad, 8
 
+# Check R_AARCH64_ABS64 relocation of a function pointer to external symbol
+#
+# jitlink-check: *{8}external_func_addr_quad = external_func
+        .globl  external_func_addr_quad
+        .p2align  3
+external_func_addr_quad:
+        .xword  external_func
+        .size external_func_addr_quad, 8
+
 # Check R_AARCH64_ADR_GOT_PAGE / R_AARCH64_LD64_GOT_LO12_NC handling with a
 # reference to an external symbol. Validate both the reference to the GOT entry,
 # and also the content of the GOT entry.
 #
 # For the ADRP :got: instruction we have the 21-bit delta to the 4k page
 # containing the GOT entry for external_data.
 #
 # For the LDR :got_lo12: instruction we have the 12-bit offset of the entry