[llvm] [BOLT] Preserve branches past function boundaries (PR #218400)

[via llvm-commits]

llvmorg-github-actions[bot] wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-bolt

Author: Adam Bzowski (adam-bzowski-arm)

<details>
<summary>Changes</summary>



---
Full diff: https://github.com/llvm/llvm-project/pull/218400.diff


3 Files Affected:

- (modified) bolt/lib/Core/BinaryFunction.cpp (+5-9) 
- (modified) bolt/test/AArch64/invalid-code-padding.s (+2-2) 
- (added) bolt/test/AArch64/jump-past-end.s (+43) 


``````````diff
diff --git a/bolt/lib/Core/BinaryFunction.cpp b/bolt/lib/Core/BinaryFunction.cpp
index a81fa2f45c206..6daef15949d65 100644
--- a/bolt/lib/Core/BinaryFunction.cpp
+++ b/bolt/lib/Core/BinaryFunction.cpp
@@ -1452,16 +1452,13 @@ Error BinaryFunction::disassemble() {
                 TargetAddress < getAddress() + getMaxSize() &&
                 !(BC.isAArch64() &&
                   BC.handleAArch64Veneer(TargetAddress, /*MatchOnly*/ true))) {
-              // Result of __builtin_unreachable().
+              // This may be an effect of source-level undefined behavior, such
+              // as __builtin_unreachable(). It may also be valid code at a
+              // temporary label omitted from the symbol table by the assembler.
+              // Since BOLT cannot distinguish these cases, preserve the branch.
               errs() << "BOLT-WARNING: jump past end detected at 0x"
                      << Twine::utohexstr(AbsoluteInstrAddr) << " in function "
-                     << *this << " : replacing with nop.\n";
-              BC.MIB->createNoop(Instruction);
-              if (IsCondBranch) {
-                // Register branch offset for profile validation.
-                IgnoredBranches.emplace_back(Offset, Offset + Size);
-              }
-              goto add_instruction;
+                     << *this << ".\n";
             }
             // May update Instruction and IsCall
             TargetSymbol = handleExternalReference(Instruction, Size, Offset,
@@ -1539,7 +1536,6 @@ Error BinaryFunction::disassemble() {
       }
     }
 
-add_instruction:
     if (!getDWARFUnits().empty()) {
       SmallVector<DebugLineTableRowRef, 1> Rows;
       for (const auto &[_, Unit] : getDWARFUnits()) {
diff --git a/bolt/test/AArch64/invalid-code-padding.s b/bolt/test/AArch64/invalid-code-padding.s
index 4706e600621ab..2b2b53427d2ff 100644
--- a/bolt/test/AArch64/invalid-code-padding.s
+++ b/bolt/test/AArch64/invalid-code-padding.s
@@ -9,12 +9,12 @@
   .type foo, %function
 foo:
   cmp x0, x1
-  b.eq .Ltmp1
+  b.eq .Ltmp2
   adrp x1, jmptbl
   add x1, x1, :lo12:jmptbl
   ldrsw x2, [x1, x2, lsl #2]
   br x2
-  b .Ltmp1
+  b .Ltmp2
 .Ltmp2:
   add x0, x0, x1
   ret
diff --git a/bolt/test/AArch64/jump-past-end.s b/bolt/test/AArch64/jump-past-end.s
new file mode 100644
index 0000000000000..fe9cdc6330c3b
--- /dev/null
+++ b/bolt/test/AArch64/jump-past-end.s
@@ -0,0 +1,43 @@
+## Check that BOLT preserves a branch to code immediately past the size of a
+## function. Such code can result from a temporary assembler symbol that is
+## not retained in the linked binary.
+
+# REQUIRES: system-linux, target=aarch64{{.*}}
+
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-linux %s -o %t.o
+# RUN: %clang %cflags %t.o -o %t.exe -nostdlib -Wl,-q,-e,_start
+# RUN: llvm-bolt %t.exe -o %t.bolt --relocs=1 --lite=0 2>&1 \
+# RUN:   | FileCheck %s --check-prefix=WARNING
+# RUN: llvm-objdump -d --no-show-raw-insn -j .text %t.bolt \
+# RUN:   | FileCheck %s --check-prefix=DISASM
+
+# WARNING: BOLT-WARNING: jump past end detected
+
+# DISASM: <f>:
+# DISASM-NEXT: {{.*}}cbnz x0,
+# DISASM-NEXT: {{.*}}ret
+
+  .text
+  .global _start
+  .type _start, %function
+_start:
+  b f
+  .size _start, .-_start
+
+  .global f
+  .type f, %function
+f:
+  cbnz x0, .Lf_helper
+  ret
+  .size f, .-f
+
+  .type .Lf_helper, %function
+.Lf_helper:
+  mov x1, #42
+  ret
+  .size .Lf_helper, .-.Lf_helper
+
+  .type next, %function
+next:
+  ret
+  .size next, .-next

``````````

</details>


https://github.com/llvm/llvm-project/pull/218400