[COFF] Align ARM64 range extension thunks at instruction boundary

[TomTan]

Summary

RangeExtensionThunkARM64 is created for out-of-range branches on Windows ARM64 because branch instructions has limited bits to encode target address. Currently, RangeExtensionThunkARM64 is appended to its referencing COFF section from object file at link time without any alignment requirement, so if size of the preceding COFF section is not aligned to instruction boundary (4 bytes), RangeExtensionThunkARM64 will emit thunk instructions at unaligned address which is never a valid branch target on ARM64, and usually trigger invalid instruction exception when branching to it.

This PR fixes it by requiring such thunks to align at 4 bytes.

[mstorsjo]

LGTM, but I think we should do the same for RangeExtensionThunkARM as well (with alignment 2).

lld/COFF/Chunks.h

public:
   size_t getSize() const override;
   void writeTo(uint8_t *buf) const override;
 
   Defined *target;
 };
 
 class RangeExtensionThunkARM64 : public NonSectionChunk {
 public:
-  explicit RangeExtensionThunkARM64(Defined *t) : target(t) {}
+  explicit RangeExtensionThunkARM64(Defined *t) : target(t) { setAlignment(4); }
   size_t getSize() const override;
   void writeTo(uint8_t *buf) const override;
 
   Defined *target;
 };
 
 // Windows-specific.
 // See comments for DefinedLocalImport class.

lld/test/COFF/arm64-thunks.s

 // REQUIRES: aarch64
 // RUN: llvm-mc -filetype=obj -triple=aarch64-windows %s -o %t.obj
 // RUN: lld-link -entry:main -subsystem:console %t.obj -out:%t.exe -verbose 2>&1 | FileCheck -check-prefix=VERBOSE %s
 // RUN: llvm-objdump -d %t.exe | FileCheck -check-prefix=DISASM %s
 
-// VERBOSE: Added 1 thunks with margin {{.*}} in 1 passes
+// VERBOSE: Added 2 thunks with margin {{.*}} in 1 passes
 
     .globl main
     .globl func1
+    .globl func2
     .text
 main:
     tbz w0, #0, func1
     ret
     .section .text$a, "xr"
     .space 0x8000
     .section .text$b, "xr"
 func1:
+    tbz w0, #0, func2
+    ret
+    .space 1
+    .section .text$c, "xr"
+    .space 0x8000
+    .section .text$d, "xr"
+    .align 2
+func2:
     ret
 
 // DISASM: 0000000140001000 .text:
 // DISASM: 140001000:      40 00 00 36     tbz     w0, #0, #8 <.text+0x8>
 // DISASM: 140001004:      c0 03 5f d6     ret
 // DISASM: 140001008:      50 00 00 90     adrp    x16, #32768
 // DISASM: 14000100c:      10 52 00 91     add     x16, x16, #20
 // DISASM: 140001010:      00 02 1f d6     br      x16
 
-// DISASM: 140009014:      c0 03 5f d6     ret
+// DISASM: 140009014:      60 00 00 36     tbz     w0, #0, #12 <.text+0x8020>
+// DISASM: 140009018:      c0 03 5f d6     ret
+
+// DISASM: 140009020:      50 00 00 90     adrp    x16, #32768
+// DISASM: 140009024:      10 b2 00 91     add     x16, x16, #44
+// DISASM: 140009028:      00 02 1f d6     br      x16
+
+// DISASM: 14001102c:      c0 03 5f d6     ret