!6 [Backport] [BOLT] [AArch64] Handle data at the beginning of a function when disassembling and building CFG.

From: @xiongzhou4 
Reviewed-by: @li-yancheng 
Signed-off-by: @li-yancheng

0001-Backport-Handle-data-at-the-beginning-of-a-function.patch

@@ -0,0 +1,109 @@
+From 963d19d86cca708957055f3072cbe8473c8e52e3 Mon Sep 17 00:00:00 2001
+From: xiongzhou4 <xiongzhou4@huawei.com>
+Date: Thu, 15 Jun 2023 20:34:42 +0800
+Subject: [PATCH] [Backport] [BOLT] [AArch64] Handle data at the beginning of a
+ function when disassembling and building CFG.
+
+---
+ bolt/src/BinaryFunction.cpp          | 11 +++++++----
+ bolt/src/BinaryFunction.h            |  9 +++++++++
+ bolt/src/Exceptions.cpp              |  2 +-
+ bolt/test/AArch64/data-at-0-offset.c | 17 +++++++++++++++++
+ 4 files changed, 34 insertions(+), 5 deletions(-)
+ create mode 100644 bolt/test/AArch64/data-at-0-offset.c
+
+diff --git a/bolt/src/BinaryFunction.cpp b/bolt/src/BinaryFunction.cpp
+index 9414b83ad..93de63d3f 100644
+--- a/bolt/src/BinaryFunction.cpp
++++ b/bolt/src/BinaryFunction.cpp
+@@ -1445,6 +1445,9 @@ add_instruction:
+     addInstruction(Offset, std::move(Instruction));
+   }
+ 
++  if (uint64_t Offset = getFirstInstructionOffset())
++    Labels[Offset] = BC.Ctx->createNamedTempSymbol();
++
+   clearList(Relocations);
+ 
+   if (!IsSimple) {
+@@ -1987,7 +1990,7 @@ bool BinaryFunction::buildCFG(MCPlusBuilder::AllocatorIdTy AllocatorId) {
+     return false;
+ 
+   assert(BasicBlocks.empty() && "basic block list should be empty");
+-  assert((Labels.find(0) != Labels.end()) &&
++  assert((Labels.find(getFirstInstructionOffset()) != Labels.end()) &&
+          "first instruction should always have a label");
+ 
+   // Create basic blocks in the original layout order:
+@@ -2087,9 +2090,9 @@ bool BinaryFunction::buildCFG(MCPlusBuilder::AllocatorIdTy AllocatorId) {
+         updateOffset(LastInstrOffset);
+       }
+     }
+-    if (Offset == 0) {
+-      // Add associated CFI pseudos in the first offset (0)
+-      addCFIPlaceholders(0, InsertBB);
++    if (Offset == getFirstInstructionOffset()) {
++      // Add associated CFI pseudos in the first offset.
++      addCFIPlaceholders(Offset, InsertBB);
+     }
+ 
+     const bool IsBlockEnd = MIB->isTerminator(Instr);
+diff --git a/bolt/src/BinaryFunction.h b/bolt/src/BinaryFunction.h
+index a824f3d58..3b435cfb3 100644
+--- a/bolt/src/BinaryFunction.h
++++ b/bolt/src/BinaryFunction.h
+@@ -967,6 +967,15 @@ public:
+     return const_cast<BinaryFunction *>(this)->getInstructionAtOffset(Offset);
+   }
+ 
++  /// Return offset for the first instruction. If there is data at the
++  /// beginning of a function then offset of the first instruction could
++  /// be different from 0.
++  uint64_t getFirstInstructionOffset() const {
++    if (Instructions.empty())
++      return 0;
++    return Instructions.begin()->first;
++  }
++
+   /// Return jump table that covers a given \p Address in memory.
+   JumpTable *getJumpTableContainingAddress(uint64_t Address) {
+     auto JTI = JumpTables.upper_bound(Address);
+diff --git a/bolt/src/Exceptions.cpp b/bolt/src/Exceptions.cpp
+index f73054aa8..0a64e3fca 100644
+--- a/bolt/src/Exceptions.cpp
++++ b/bolt/src/Exceptions.cpp
+@@ -499,7 +499,7 @@ bool CFIReaderWriter::fillCFIInfoFor(BinaryFunction &Function) const {
+   Optional<uint64_t> LSDA = CurFDE.getLSDAAddress();
+   Function.setLSDAAddress(LSDA ? *LSDA : 0);
+ 
+-  uint64_t Offset = 0;
++  uint64_t Offset = Function.getFirstInstructionOffset();
+   uint64_t CodeAlignment = CurFDE.getLinkedCIE()->getCodeAlignmentFactor();
+   uint64_t DataAlignment = CurFDE.getLinkedCIE()->getDataAlignmentFactor();
+   if (CurFDE.getLinkedCIE()->getPersonalityAddress()) {
+diff --git a/bolt/test/AArch64/data-at-0-offset.c b/bolt/test/AArch64/data-at-0-offset.c
+new file mode 100644
+index 000000000..e0c689a19
+--- /dev/null
++++ b/bolt/test/AArch64/data-at-0-offset.c
+@@ -0,0 +1,17 @@
++// RUN: %clang %cflags -O2 -fPIE -Wl,-q -pie  %s -o %t.exe
++// RUN: llvm-bolt %t.exe -o %t.bolt 2>&1 | FileCheck %s
++// CHECK-NOT: BOLT-WARNING: unable to disassemble instruction at offset
++
++void extra_space() {
++  asm volatile(".rept 256\n"
++               "    .byte 0xff\n"
++               ".endr\n");
++  return;
++}
++
++int main(int argc, char **argv) {
++  void (*fn)(void);
++  fn = extra_space + 256;
++  fn();
++  return 0;
++}
+-- 
+2.33.0
+

llvm-bolt.spec

@@ -1,13 +1,15 @@
 Name:           llvm-bolt
 Version:        0
-Release:        1.20211016.gitb72f753
+Release:        2.20211016.gitb72f753
 Summary:        BOLT is a post-link optimizer developed to speed up large applications
 License:        Apache 2.0
 URL:            https://github.com/facebookincubator/BOLT
 Source0:        %{name}-%{version}.tar.xz
 
 BuildRequires:  gcc gcc-c++ cmake ninja-build libstdc++-static chrpath ncurses-devel zlib-devel
-Requires:	glibc zlib ncurses-libs libstdc++ libgcc
+Requires:       glibc zlib ncurses-libs libstdc++ libgcc
+
+Patch1:         0001-Backport-Handle-data-at-the-beginning-of-a-function.patch
 
 %description
 BOLT is a post-link optimizer developed to speed up large applications.
@@ -15,7 +17,7 @@ It achieves the improvements by optimizing application's code layout based
 on execution profile gathered by sampling profiler, such as Linux perf tool.
 
 %prep
-%setup -q
+%autosetup -p1
 mkdir -p _build
 cd _build
 %{__cmake} -G Ninja ../llvm -DCMAKE_BUILD_TYPE=RelWithDebInfo -DLLVM_ENABLE_ASSERTIONS=ON \
@@ -54,6 +56,12 @@ chrpath -d %{_bolt_install_dir}/bin/merge-fdata
 %attr(-,root,root) %{_bindir}/perf2bolt
 
 %changelog
+* Thu Jun 15 2023 Xiong Zhou <xiongzhou4@huawei.com> 0-2.20211016.gitb72f753
+- Type:backport
+- ID:NA
+- SUG:NA
+- DESC: Handle data at the beginning of a function when disassembling and building CFG.
+
 * Mon Dec 19 2022 liyancheng <412998149@qq.com> 0-1.20211016.gitb72f753
 - Type:fix
 - ID:NA