Merge remote-tracking branch 'origin/main' into vplan-induction-resume-values

Author: fhahn

.ci/generate_test_report.py

@@ -309,10 +309,8 @@ def test_report_size_limit(self):
 # If include failures is False, total number of test will be reported but their names
 # and output will not be.
 def _generate_report(title, junit_objects, size_limit=1024 * 1024, list_failures=True):
-    style = None
-
     if not junit_objects:
-        return ("", style)
+        return ("", "success")
 
     failures = {}
     tests_run = 0
@@ -403,22 +401,23 @@ def generate_report(title, junit_files):
 
     report, style = generate_report(args.title, args.junit_files)
 
-    p = subprocess.Popen(
-        [
-            "buildkite-agent",
-            "annotate",
-            "--context",
-            args.context,
-            "--style",
-            style,
-        ],
-        stdin=subprocess.PIPE,
-        stderr=subprocess.PIPE,
-        universal_newlines=True,
-    )
+    if report:
+        p = subprocess.Popen(
+            [
+                "buildkite-agent",
+                "annotate",
+                "--context",
+                args.context,
+                "--style",
+                style,
+            ],
+            stdin=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            universal_newlines=True,
+        )
 
-    # The report can be larger than the buffer for command arguments so we send
-    # it over stdin instead.
-    _, err = p.communicate(input=report)
-    if p.returncode:
-        raise RuntimeError(f"Failed to send report to buildkite-agent:\n{err}")
+        # The report can be larger than the buffer for command arguments so we send
+        # it over stdin instead.
+        _, err = p.communicate(input=report)
+        if p.returncode:
+            raise RuntimeError(f"Failed to send report to buildkite-agent:\n{err}")

.github/new-issues-labeler.yml

@@ -27,3 +27,6 @@
 
 'bolt':
   - '/\bbolt(?!\-)\b/i'
+
+'infra:commit-access-request':
+  - '/Request Commit Access/'

.github/workflows/libcxx-build-and-test.yaml

@@ -45,12 +45,11 @@ env:
   LLVM_SYMBOLIZER_PATH: "/usr/bin/llvm-symbolizer-19"
   CLANG_CRASH_DIAGNOSTICS_DIR: "crash_diagnostics"
 
-
 jobs:
   stage1:
     if: github.repository_owner == 'llvm'
-    runs-on: libcxx-runners-set
-    container: ghcr.io/libcxx/actions-builder:testing-2024-09-21
+    runs-on: libcxx-self-hosted-linux
+    container: ghcr.io/llvm/libcxx-linux-builder:0fd6f684b9c84c32d6cbfd9742402e788b2879f1
     continue-on-error: false
     strategy:
       fail-fast: false
@@ -86,8 +85,8 @@ jobs:
             **/crash_diagnostics/*
   stage2:
     if: github.repository_owner == 'llvm'
-    runs-on: libcxx-runners-set
-    container: ghcr.io/libcxx/actions-builder:testing-2024-09-21
+    runs-on: libcxx-self-hosted-linux
+    container: ghcr.io/llvm/libcxx-linux-builder:0fd6f684b9c84c32d6cbfd9742402e788b2879f1
     needs: [ stage1 ]
     continue-on-error: false
     strategy:
@@ -161,21 +160,21 @@ jobs:
           'generic-static',
           'bootstrapping-build'
         ]
-        machine: [ 'libcxx-runners-set' ]
+        machine: [ 'libcxx-self-hosted-linux' ]
         include:
         - config: 'generic-cxx26'
-          machine: libcxx-runners-set
+          machine: libcxx-self-hosted-linux
         - config: 'generic-asan'
-          machine: libcxx-runners-set
+          machine: libcxx-self-hosted-linux
         - config: 'generic-tsan'
-          machine: libcxx-runners-set
+          machine: libcxx-self-hosted-linux
         - config: 'generic-ubsan'
-          machine: libcxx-runners-set
+          machine: libcxx-self-hosted-linux
         # Use a larger machine for MSAN to avoid timeout and memory allocation issues.
         - config: 'generic-msan'
-          machine: libcxx-runners-set
+          machine: libcxx-self-hosted-linux
     runs-on: ${{ matrix.machine }}
-    container: ghcr.io/libcxx/actions-builder:testing-2024-09-21
+    container: ghcr.io/llvm/libcxx-linux-builder:0fd6f684b9c84c32d6cbfd9742402e788b2879f1
     steps:
       - uses: actions/checkout@v4
       - name: ${{ matrix.config }}

bolt/include/bolt/Core/BinarySection.h

@@ -87,6 +87,7 @@ class BinarySection {
                                    // been renamed)
   uint64_t OutputAddress{0};       // Section address for the rewritten binary.
   uint64_t OutputSize{0};          // Section size in the rewritten binary.
+                                   // Can exceed OutputContents with padding.
   uint64_t OutputFileOffset{0};    // File offset in the rewritten binary file.
   StringRef OutputContents;        // Rewritten section contents.
   const uint64_t SectionNumber;    // Order in which the section was created.
@@ -474,6 +475,11 @@ class BinarySection {
   /// Use name \p SectionName for the section during the emission.
   void emitAsData(MCStreamer &Streamer, const Twine &SectionName) const;
 
+  /// Write finalized contents of the section. If OutputSize exceeds the size of
+  /// the OutputContents, append zero padding to the stream and return the
+  /// number of byte written which should match the OutputSize.
+  uint64_t write(raw_ostream &OS) const;
+
   using SymbolResolverFuncTy = llvm::function_ref<uint64_t(const MCSymbol *)>;
 
   /// Flush all pending relocations to patch original contents of sections
@@ -497,6 +503,9 @@ class BinarySection {
     IsFinalized = true;
   }
 
+  /// When writing section contents, add \p PaddingSize zero bytes at the end.
+  void addPadding(uint64_t PaddingSize) { OutputSize += PaddingSize; }
+
   /// Reorder the contents of this section according to /p Order.  If
   /// /p Inplace is true, the entire contents of the section is reordered,
   /// otherwise the new contents contain only the reordered data.

bolt/lib/Core/BinaryEmitter.cpp

@@ -416,17 +416,6 @@ void BinaryEmitter::emitFunctionBody(BinaryFunction &BF, FunctionFragment &FF,
     BF.duplicateConstantIslands();
   }
 
-  if (!FF.empty() && FF.front()->isLandingPad()) {
-    assert(!FF.front()->isEntryPoint() &&
-           "Landing pad cannot be entry point of function");
-    // If the first block of the fragment is a landing pad, it's offset from the
-    // start of the area that the corresponding LSDA describes is zero. In this
-    // case, the call site entries in that LSDA have 0 as offset to the landing
-    // pad, which the runtime interprets as "no handler". To prevent this,
-    // insert some padding.
-    Streamer.emitBytes(BC.MIB->getTrapFillValue());
-  }
-
   // Track the first emitted instruction with debug info.
   bool FirstInstr = true;
   for (BinaryBasicBlock *const BB : FF) {
@@ -906,17 +895,6 @@ void BinaryEmitter::emitLSDA(BinaryFunction &BF, const FunctionFragment &FF) {
   if (Sites.empty())
     return;
 
-  // Calculate callsite table size. Size of each callsite entry is:
-  //
-  //  sizeof(start) + sizeof(length) + sizeof(LP) + sizeof(uleb128(action))
-  //
-  // or
-  //
-  //  sizeof(dwarf::DW_EH_PE_data4) * 3 + sizeof(uleb128(action))
-  uint64_t CallSiteTableLength = llvm::size(Sites) * 4 * 3;
-  for (const auto &FragmentCallSite : Sites)
-    CallSiteTableLength += getULEB128Size(FragmentCallSite.second.Action);
-
   Streamer.switchSection(BC.MOFI->getLSDASection());
 
   const unsigned TTypeEncoding = BF.getLSDATypeEncoding();
@@ -937,74 +915,77 @@ void BinaryEmitter::emitLSDA(BinaryFunction &BF, const FunctionFragment &FF) {
   // Emit the LSDA header.
 
   // If LPStart is omitted, then the start of the FDE is used as a base for
-  // landing pad displacements. Then if a cold fragment starts with
-  // a landing pad, this means that the first landing pad offset will be 0.
-  // As a result, the exception handling runtime will ignore this landing pad
-  // because zero offset denotes the absence of a landing pad.
-  // For this reason, when the binary has fixed starting address we emit LPStart
-  // as 0 and output the absolute value of the landing pad in the table.
+  // landing pad displacements. Then, if a cold fragment starts with a landing
+  // pad, this means that the first landing pad offset will be 0. However, C++
+  // runtime treats 0 as if there is no landing pad present, thus we *must* emit
+  // non-zero offsets for all valid LPs.
   //
-  // If the base address can change, we cannot use absolute addresses for
-  // landing pads (at least not without runtime relocations). Hence, we fall
-  // back to emitting landing pads relative to the FDE start.
-  // As we are emitting label differences, we have to guarantee both labels are
-  // defined in the same section and hence cannot place the landing pad into a
-  // cold fragment when the corresponding call site is in the hot fragment.
-  // Because of this issue and the previously described issue of possible
-  // zero-offset landing pad we have to place landing pads in the same section
-  // as the corresponding invokes for shared objects.
+  // As a solution, for fixed-address binaries we set LPStart to 0, and for
+  // position-independent binaries we set LP start to FDE start minus one byte
+  // for FDEs that start with a landing pad.
+  const bool NeedsLPAdjustment = !FF.empty() && FF.front()->isLandingPad();
   std::function<void(const MCSymbol *)> emitLandingPad;
   if (BC.HasFixedLoadAddress) {
     Streamer.emitIntValue(dwarf::DW_EH_PE_udata4, 1); // LPStart format
     Streamer.emitIntValue(0, 4);                      // LPStart
     emitLandingPad = [&](const MCSymbol *LPSymbol) {
-      if (!LPSymbol)
-        Streamer.emitIntValue(0, 4);
-      else
+      if (LPSymbol)
         Streamer.emitSymbolValue(LPSymbol, 4);
+      else
+        Streamer.emitIntValue(0, 4);
     };
   } else {
-    Streamer.emitIntValue(dwarf::DW_EH_PE_omit, 1); // LPStart format
+    if (NeedsLPAdjustment) {
+      // Use relative LPStart format and emit LPStart as [SymbolStart - 1].
+      Streamer.emitIntValue(dwarf::DW_EH_PE_pcrel | dwarf::DW_EH_PE_sdata4, 1);
+      MCSymbol *DotSymbol = BC.Ctx->createTempSymbol("LPBase");
+      Streamer.emitLabel(DotSymbol);
+
+      const MCExpr *LPStartExpr = MCBinaryExpr::createSub(
+          MCSymbolRefExpr::create(StartSymbol, *BC.Ctx),
+          MCSymbolRefExpr::create(DotSymbol, *BC.Ctx), *BC.Ctx);
+      LPStartExpr = MCBinaryExpr::createSub(
+          LPStartExpr, MCConstantExpr::create(1, *BC.Ctx), *BC.Ctx);
+      Streamer.emitValue(LPStartExpr, 4);
+    } else {
+      // DW_EH_PE_omit means FDE start (StartSymbol) will be used as LPStart.
+      Streamer.emitIntValue(dwarf::DW_EH_PE_omit, 1);
+    }
     emitLandingPad = [&](const MCSymbol *LPSymbol) {
-      if (!LPSymbol)
+      if (LPSymbol) {
+        const MCExpr *LPOffsetExpr = MCBinaryExpr::createSub(
+            MCSymbolRefExpr::create(LPSymbol, *BC.Ctx),
+            MCSymbolRefExpr::create(StartSymbol, *BC.Ctx), *BC.Ctx);
+        if (NeedsLPAdjustment)
+          LPOffsetExpr = MCBinaryExpr::createAdd(
+              LPOffsetExpr, MCConstantExpr::create(1, *BC.Ctx), *BC.Ctx);
+        Streamer.emitValue(LPOffsetExpr, 4);
+      } else {
         Streamer.emitIntValue(0, 4);
-      else
-        Streamer.emitAbsoluteSymbolDiff(LPSymbol, StartSymbol, 4);
+      }
     };
   }
 
   Streamer.emitIntValue(TTypeEncoding, 1); // TType format
 
-  // See the comment in EHStreamer::emitExceptionTable() on to use
-  // uleb128 encoding (which can use variable number of bytes to encode the same
-  // value) to ensure type info table is properly aligned at 4 bytes without
-  // iteratively fixing sizes of the tables.
-  unsigned CallSiteTableLengthSize = getULEB128Size(CallSiteTableLength);
-  unsigned TTypeBaseOffset =
-      sizeof(int8_t) +                 // Call site format
-      CallSiteTableLengthSize +        // Call site table length size
-      CallSiteTableLength +            // Call site table length
-      BF.getLSDAActionTable().size() + // Actions table size
-      BF.getLSDATypeTable().size() * TTypeEncodingSize; // Types table size
-  unsigned TTypeBaseOffsetSize = getULEB128Size(TTypeBaseOffset);
-  unsigned TotalSize = sizeof(int8_t) +      // LPStart format
-                       sizeof(int8_t) +      // TType format
-                       TTypeBaseOffsetSize + // TType base offset size
-                       TTypeBaseOffset;      // TType base offset
-  unsigned SizeAlign = (4 - TotalSize) & 3;
-
-  if (TTypeEncoding != dwarf::DW_EH_PE_omit)
-    // Account for any extra padding that will be added to the call site table
-    // length.
-    Streamer.emitULEB128IntValue(TTypeBaseOffset,
-                                 /*PadTo=*/TTypeBaseOffsetSize + SizeAlign);
+  MCSymbol *TTBaseLabel = nullptr;
+  if (TTypeEncoding != dwarf::DW_EH_PE_omit) {
+    TTBaseLabel = BC.Ctx->createTempSymbol("TTBase");
+    MCSymbol *TTBaseRefLabel = BC.Ctx->createTempSymbol("TTBaseRef");
+    Streamer.emitAbsoluteSymbolDiffAsULEB128(TTBaseLabel, TTBaseRefLabel);
+    Streamer.emitLabel(TTBaseRefLabel);
+  }
 
   // Emit the landing pad call site table. We use signed data4 since we can emit
   // a landing pad in a different part of the split function that could appear
   // earlier in the address space than LPStart.
   Streamer.emitIntValue(dwarf::DW_EH_PE_sdata4, 1);
-  Streamer.emitULEB128IntValue(CallSiteTableLength);
 
+  MCSymbol *CSTStartLabel = BC.Ctx->createTempSymbol("CSTStart");
+  MCSymbol *CSTEndLabel = BC.Ctx->createTempSymbol("CSTEnd");
+  Streamer.emitAbsoluteSymbolDiffAsULEB128(CSTEndLabel, CSTStartLabel);
+
+  Streamer.emitLabel(CSTStartLabel);
   for (const auto &FragmentCallSite : Sites) {
     const BinaryFunction::CallSite &CallSite = FragmentCallSite.second;
     const MCSymbol *BeginLabel = CallSite.Start;
@@ -1020,6 +1001,7 @@ void BinaryEmitter::emitLSDA(BinaryFunction &BF, const FunctionFragment &FF) {
     emitLandingPad(CallSite.LP);
     Streamer.emitULEB128IntValue(CallSite.Action);
   }
+  Streamer.emitLabel(CSTEndLabel);
 
   // Write out action, type, and type index tables at the end.
   //
@@ -1038,6 +1020,8 @@ void BinaryEmitter::emitLSDA(BinaryFunction &BF, const FunctionFragment &FF) {
   assert(TypeTable.size() == BF.getLSDATypeTable().size() &&
          "indirect type table size mismatch");
 
+  Streamer.emitValueToAlignment(Align(TTypeAlignment));
+
   for (int Index = TypeTable.size() - 1; Index >= 0; --Index) {
     const uint64_t TypeAddress = TypeTable[Index];
     switch (TTypeEncoding & 0x70) {
@@ -1063,6 +1047,10 @@ void BinaryEmitter::emitLSDA(BinaryFunction &BF, const FunctionFragment &FF) {
     }
     }
   }
+
+  if (TTypeEncoding != dwarf::DW_EH_PE_omit)
+    Streamer.emitLabel(TTBaseLabel);
+
   for (uint8_t const &Byte : BF.getLSDATypeIndexTable())
     Streamer.emitIntValue(Byte, 1);
 }

bolt/lib/Core/BinarySection.cpp

@@ -142,6 +142,15 @@ void BinarySection::emitAsData(MCStreamer &Streamer,
     Streamer.emitLabel(BC.Ctx->getOrCreateSymbol("__hot_data_end"));
 }
 
+uint64_t BinarySection::write(raw_ostream &OS) const {
+  const uint64_t NumValidContentBytes =
+      std::min<uint64_t>(getOutputContents().size(), getOutputSize());
+  OS.write(getOutputContents().data(), NumValidContentBytes);
+  if (getOutputSize() > NumValidContentBytes)
+    OS.write_zeros(getOutputSize() - NumValidContentBytes);
+  return getOutputSize();
+}
+
 void BinarySection::flushPendingRelocations(raw_pwrite_stream &OS,
                                             SymbolResolverFuncTy Resolver) {
   if (PendingRelocations.empty() && Patches.empty())