[BOLT] Add --pad-funcs-before=func:n (llvm#117924)

This complements --pad-funcs, and by using both simultaneously, enables
moving a specific function through the address space without modifying
any code
other than the targeted function (and references to it) by doing
(before+after=constant).

See also: proposed functionality to enable inserting random padding in

https://discourse.llvm.org/t/rfc-lld-feature-for-controlling-for-code-size-dependent-measurement-bias
and llvm#117653

bolt/lib/Core/BinaryEmitter.cpp

@@ -46,13 +46,17 @@ BreakFunctionNames("break-funcs",
   cl::Hidden,
   cl::cat(BoltCategory));
 
-static cl::list<std::string>
-FunctionPadSpec("pad-funcs",
-  cl::CommaSeparated,
-  cl::desc("list of functions to pad with amount of bytes"),
-  cl::value_desc("func1:pad1,func2:pad2,func3:pad3,..."),
-  cl::Hidden,
-  cl::cat(BoltCategory));
+cl::list<std::string>
+    FunctionPadSpec("pad-funcs", cl::CommaSeparated,
+                    cl::desc("list of functions to pad with amount of bytes"),
+                    cl::value_desc("func1:pad1,func2:pad2,func3:pad3,..."),
+                    cl::Hidden, cl::cat(BoltCategory));
+
+cl::list<std::string> FunctionPadBeforeSpec(
+    "pad-funcs-before", cl::CommaSeparated,
+    cl::desc("list of functions to pad with amount of bytes"),
+    cl::value_desc("func1:pad1,func2:pad2,func3:pad3,..."), cl::Hidden,
+    cl::cat(BoltCategory));
 
 static cl::opt<bool> MarkFuncs(
     "mark-funcs",
@@ -70,11 +74,12 @@ X86AlignBranchBoundaryHotOnly("x86-align-branch-boundary-hot-only",
   cl::init(true),
   cl::cat(BoltOptCategory));
 
-size_t padFunction(const BinaryFunction &Function) {
+size_t padFunction(const cl::list<std::string> &Spec,
+                   const BinaryFunction &Function) {
   static std::map<std::string, size_t> FunctionPadding;
 
-  if (FunctionPadding.empty() && !FunctionPadSpec.empty()) {
-    for (std::string &Spec : FunctionPadSpec) {
+  if (FunctionPadding.empty() && !Spec.empty()) {
+    for (const std::string &Spec : Spec) {
       size_t N = Spec.find(':');
       if (N == std::string::npos)
         continue;
@@ -319,6 +324,32 @@ bool BinaryEmitter::emitFunction(BinaryFunction &Function,
     Streamer.emitCodeAlignment(Function.getAlign(), &*BC.STI);
   }
 
+  if (size_t Padding =
+          opts::padFunction(opts::FunctionPadBeforeSpec, Function)) {
+    // Handle padFuncsBefore after the above alignment logic but before
+    // symbol addresses are decided.
+    if (!BC.HasRelocations) {
+      BC.errs() << "BOLT-ERROR: -pad-before-funcs is not supported in "
+                << "non-relocation mode\n";
+      exit(1);
+    }
+
+    // Preserve Function.getMinAlign().
+    if (!isAligned(Function.getMinAlign(), Padding)) {
+      BC.errs() << "BOLT-ERROR: user-requested " << Padding
+                << " padding bytes before function " << Function
+                << " is not a multiple of the minimum function alignment ("
+                << Function.getMinAlign().value() << ").\n";
+      exit(1);
+    }
+
+    LLVM_DEBUG(dbgs() << "BOLT-DEBUG: padding before function " << Function
+                      << " with " << Padding << " bytes\n");
+
+    // Since the padding is not executed, it can be null bytes.
+    Streamer.emitFill(Padding, 0);
+  }
+
   MCContext &Context = Streamer.getContext();
   const MCAsmInfo *MAI = Context.getAsmInfo();
 
@@ -373,7 +404,7 @@ bool BinaryEmitter::emitFunction(BinaryFunction &Function,
   emitFunctionBody(Function, FF, /*EmitCodeOnly=*/false);
 
   // Emit padding if requested.
-  if (size_t Padding = opts::padFunction(Function)) {
+  if (size_t Padding = opts::padFunction(opts::FunctionPadSpec, Function)) {
     LLVM_DEBUG(dbgs() << "BOLT-DEBUG: padding function " << Function << " with "
                       << Padding << " bytes\n");
     Streamer.emitFill(Padding, MAI->getTextAlignFillValue());

bolt/lib/Passes/ReorderFunctions.cpp

@@ -28,7 +28,9 @@ extern cl::OptionCategory BoltOptCategory;
 extern cl::opt<unsigned> Verbosity;
 extern cl::opt<uint32_t> RandomSeed;
 
-extern size_t padFunction(const bolt::BinaryFunction &Function);
+extern size_t padFunction(const cl::list<std::string> &Spec,
+                          const bolt::BinaryFunction &Function);
+extern cl::list<std::string> FunctionPadSpec, FunctionPadBeforeSpec;
 
 extern cl::opt<bolt::ReorderFunctions::ReorderType> ReorderFunctions;
 cl::opt<bolt::ReorderFunctions::ReorderType> ReorderFunctions(
@@ -304,8 +306,12 @@ Error ReorderFunctions::runOnFunctions(BinaryContext &BC) {
                           return false;
                         if (B->isIgnored())
                           return true;
-                        const size_t PadA = opts::padFunction(*A);
-                        const size_t PadB = opts::padFunction(*B);
+                        const size_t PadA =
+                            opts::padFunction(opts::FunctionPadSpec, *A) +
+                            opts::padFunction(opts::FunctionPadBeforeSpec, *A);
+                        const size_t PadB =
+                            opts::padFunction(opts::FunctionPadSpec, *B) +
+                            opts::padFunction(opts::FunctionPadBeforeSpec, *B);
                         if (!PadA || !PadB) {
                           if (PadA)
                             return true;

bolt/test/AArch64/pad-before-funcs.s

@@ -0,0 +1,29 @@
+# Test checks that --pad-before-funcs is working as expected.
+# It should be able to introduce a configurable offset for the _start symbol.
+# It should reject requests which don't obey the code alignment requirement.
+
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %s -o %t.o
+# RUN: %clang %cflags %t.o -o %t.exe -Wl,-q -Wl,--section-start=.text=0x4000
+# RUN: llvm-bolt %t.exe -o %t.bolt.0 --pad-funcs-before=_start:0
+# RUN: llvm-bolt %t.exe -o %t.bolt.4 --pad-funcs-before=_start:4
+# RUN: llvm-bolt %t.exe -o %t.bolt.8 --pad-funcs-before=_start:8
+
+# RUN: not llvm-bolt %t.exe -o %t.bolt.8 --pad-funcs-before=_start:1 2>&1 | FileCheck --check-prefix=CHECK-BAD-ALIGN %s
+
+# CHECK-BAD-ALIGN: user-requested 1 padding bytes before function _start(*2) is not a multiple of the minimum function alignment (4).
+
+# RUN: llvm-objdump --section=.text --disassemble %t.bolt.0 | FileCheck --check-prefix=CHECK-0 %s
+# RUN: llvm-objdump --section=.text --disassemble %t.bolt.4 | FileCheck --check-prefix=CHECK-4 %s
+# RUN: llvm-objdump --section=.text --disassemble %t.bolt.8 | FileCheck --check-prefix=CHECK-8 %s
+
+# Trigger relocation mode in bolt.
+.reloc 0, R_AARCH64_NONE
+
+.section .text
+.globl _start
+
+# CHECK-0: 0000000000400000 <_start>
+# CHECK-4: 0000000000400004 <_start>
+# CHECK-8: 0000000000400008 <_start>
+_start:
+    ret