[Coverage][WebAssembly] Add initial support for WebAssembly/WASI (llv…

…m#111332)

Currently, WebAssembly/WASI target does not provide direct support for
code coverage.
This patch set fixes several issues to unlock the feature. The main
changes are:

1. Port `compiler-rt/lib/profile` to WebAssembly/WASI.
2. Adjust profile metadata sections for Wasm object file format.
- [CodeGen] Emit `__llvm_covmap` and `__llvm_covfun` as custom sections
instead of data segments.
    - [lld] Align the interval space of custom sections at link time.
- [llvm-cov] Copy misaligned custom section data if the start address is
not aligned.
    - [llvm-cov] Read `__llvm_prf_names` from data segments
3. [clang] Link with profile runtime libraries if requested

See each commit message for more details and rationale.
This is part of the effort to add code coverage support in Wasm target
of Swift toolchain.

clang/lib/Driver/ToolChains/WebAssembly.cpp

@@ -163,6 +163,8 @@ void wasm::Linker::ConstructJob(Compilation &C, const JobAction &JA,
     AddRunTimeLibs(ToolChain, ToolChain.getDriver(), CmdArgs, Args);
   }
 
+  ToolChain.addProfileRTLibs(Args, CmdArgs);
+
   CmdArgs.push_back("-o");
   CmdArgs.push_back(Output.getFilename());
 

compiler-rt/cmake/Modules/AllSupportedArchDefs.cmake

@@ -77,7 +77,7 @@ set(ALL_HWASAN_SUPPORTED_ARCH ${X86_64} ${ARM64} ${RISCV64})
 set(ALL_MEMPROF_SUPPORTED_ARCH ${X86_64})
 set(ALL_PROFILE_SUPPORTED_ARCH ${X86} ${X86_64} ${ARM32} ${ARM64} ${PPC32} ${PPC64}
     ${MIPS32} ${MIPS64} ${S390X} ${SPARC} ${SPARCV9} ${HEXAGON}
-    ${RISCV32} ${RISCV64} ${LOONGARCH64})
+    ${RISCV32} ${RISCV64} ${LOONGARCH64} ${WASM32})
 set(ALL_CTX_PROFILE_SUPPORTED_ARCH ${X86_64})
 if (OS_NAME MATCHES "FreeBSD")
   set(ALL_TSAN_SUPPORTED_ARCH ${X86_64} ${MIPS64} ${ARM64})

compiler-rt/cmake/config-ix.cmake

@@ -822,7 +822,7 @@ else()
 endif()
 
 if (PROFILE_SUPPORTED_ARCH AND NOT LLVM_USE_SANITIZER AND
-    OS_NAME MATCHES "Darwin|Linux|FreeBSD|Windows|Android|Fuchsia|SunOS|NetBSD|AIX")
+    OS_NAME MATCHES "Darwin|Linux|FreeBSD|Windows|Android|Fuchsia|SunOS|NetBSD|AIX|WASI")
   set(COMPILER_RT_HAS_PROFILE TRUE)
 else()
   set(COMPILER_RT_HAS_PROFILE FALSE)

compiler-rt/lib/profile/CMakeLists.txt

@@ -38,6 +38,17 @@ int main() {
 
 " COMPILER_RT_TARGET_HAS_FCNTL_LCK)
 
+CHECK_CXX_SOURCE_COMPILES("
+#include <sys/file.h>
+
+int fd;
+int main() {
+  flock(fd, LOCK_EX);
+  return 0;
+}
+
+" COMPILER_RT_TARGET_HAS_FLOCK)
+
 CHECK_CXX_SOURCE_COMPILES("
 #include <sys/utsname.h>
 int main() {
@@ -93,6 +104,13 @@ if(FUCHSIA OR UNIX)
      -Wno-pedantic)
 endif()
 
+if(CMAKE_SYSTEM_NAME STREQUAL "WASI")
+  set(EXTRA_FLAGS
+      ${EXTRA_FLAGS}
+      -D_WASI_EMULATED_MMAN
+      -D_WASI_EMULATED_GETPID)
+endif()
+
 if(COMPILER_RT_TARGET_HAS_ATOMICS)
  set(EXTRA_FLAGS
      ${EXTRA_FLAGS}
@@ -105,6 +123,12 @@ if(COMPILER_RT_TARGET_HAS_FCNTL_LCK)
      -DCOMPILER_RT_HAS_FCNTL_LCK=1)
 endif()
 
+if(COMPILER_RT_TARGET_HAS_FLOCK)
+  set(EXTRA_FLAGS
+      ${EXTRA_FLAGS}
+      -DCOMPILER_RT_HAS_FLOCK=1)
+endif()
+
 if(COMPILER_RT_TARGET_HAS_UNAME)
  set(EXTRA_FLAGS
      ${EXTRA_FLAGS}

compiler-rt/lib/profile/GCDAProfiling.c

@@ -584,7 +584,7 @@ void llvm_reset_counters(void) {
   }
 }
 
-#if !defined(_WIN32)
+#if !defined(_WIN32) && !defined(__wasm__)
 COMPILER_RT_VISIBILITY
 pid_t __gcov_fork() {
   pid_t parent_pid = getpid();

compiler-rt/lib/profile/InstrProfilingPlatformLinux.c

@@ -6,11 +6,11 @@
 |*
 \*===----------------------------------------------------------------------===*/
 
-#if defined(__linux__) || defined(__FreeBSD__) || defined(__Fuchsia__) || \
-    (defined(__sun__) && defined(__svr4__)) || defined(__NetBSD__) || \
-    defined(_AIX)
+#if defined(__linux__) || defined(__FreeBSD__) || defined(__Fuchsia__) ||      \
+    (defined(__sun__) && defined(__svr4__)) || defined(__NetBSD__) ||          \
+    defined(_AIX) || defined(__wasm__)
 
-#if !defined(_AIX)
+#if !defined(_AIX) && !defined(__wasm__)
 #include <elf.h>
 #include <link.h>
 #endif

compiler-rt/lib/profile/InstrProfilingPlatformOther.c

@@ -8,7 +8,8 @@
 
 #if !defined(__APPLE__) && !defined(__linux__) && !defined(__FreeBSD__) &&     \
     !defined(__Fuchsia__) && !(defined(__sun__) && defined(__svr4__)) &&       \
-    !defined(__NetBSD__) && !defined(_WIN32) && !defined(_AIX)
+    !defined(__NetBSD__) && !defined(_WIN32) && !defined(_AIX) &&              \
+    !defined(__wasm__)
 
 #include <stdlib.h>
 #include <stdio.h>

compiler-rt/lib/profile/InstrProfilingPort.h

@@ -54,7 +54,7 @@
 #endif
 
 #define COMPILER_RT_MAX_HOSTLEN 128
-#ifdef __ORBIS__
+#if defined(__ORBIS__) || defined(__wasi__)
 #define COMPILER_RT_GETHOSTNAME(Name, Len) ((void)(Name), (void)(Len), (-1))
 #else
 #define COMPILER_RT_GETHOSTNAME(Name, Len) lprofGetHostName(Name, Len)

compiler-rt/lib/profile/InstrProfilingUtil.c

@@ -152,9 +152,11 @@ COMPILER_RT_VISIBILITY int lprofLockFd(int fd) {
     }
   }
   return 0;
-#else
+#elif defined(COMPILER_RT_HAS_FLOCK)
   flock(fd, LOCK_EX);
   return 0;
+#else
+  return 0;
 #endif
 }
 
@@ -177,9 +179,11 @@ COMPILER_RT_VISIBILITY int lprofUnlockFd(int fd) {
     }
   }
   return 0;
-#else
+#elif defined(COMPILER_RT_HAS_FLOCK)
   flock(fd, LOCK_UN);
   return 0;
+#else
+  return 0;
 #endif
 }
 
@@ -353,8 +357,8 @@ COMPILER_RT_VISIBILITY void lprofRestoreSigKill(void) {
 
 COMPILER_RT_VISIBILITY int lprofReleaseMemoryPagesToOS(uintptr_t Begin,
                                                        uintptr_t End) {
-#if defined(__ve__)
-  // VE doesn't support madvise.
+#if defined(__ve__) || defined(__wasi__)
+  // VE and WASI doesn't support madvise.
   return 0;
 #else
   size_t PageSize = getpagesize();

lld/test/wasm/custom-section-align.s

@@ -0,0 +1,31 @@
+# RUN: llvm-mc -filetype=obj -triple=wasm32-unknown-unknown -o %t.o %s
+# RUN: wasm-ld --no-entry %t.o -o %t.wasm
+# RUN: obj2yaml %t.wasm | FileCheck %s
+
+# Check that "__llvm_covfun" custom section is aligned to 8 bytes.
+
+        .section        .custom_section.__llvm_covfun,"GR",@,__covrec_A
+        .int32  1
+        .int8   2
+# pad   .int8   0
+#       .int8   0
+#       .int8   0
+
+        .section        .custom_section.__llvm_covfun,"GR",@,__covrec_B
+        .int32  3
+
+# CHECK:      - Type:            CUSTOM
+# CHECK-NEXT:   Name:            __llvm_covfun
+# CHECK-NEXT:   Payload:         '010000000200000003000000'
+
+# Check that regular custom sections are not aligned.
+        .section        .custom_section.foo,"GR",@,foo_A
+        .int32  1
+        .int8   2
+
+        .section        .custom_section.foo,"GR",@,foo_B
+        .int32  3
+
+# CHECK:      - Type:            CUSTOM
+# CHECK-NEXT:   Name:            foo
+# CHECK-NEXT:   Payload:         '010000000203000000'

lld/wasm/InputChunks.h

@@ -177,8 +177,9 @@ class MergeInputChunk : public InputChunk {
     inputSectionOffset = seg.SectionOffset;
   }
 
-  MergeInputChunk(const WasmSection &s, ObjFile *f)
-      : InputChunk(f, Merge, s.Name, 0, llvm::wasm::WASM_SEG_FLAG_STRINGS) {
+  MergeInputChunk(const WasmSection &s, ObjFile *f, uint32_t alignment)
+      : InputChunk(f, Merge, s.Name, alignment,
+                   llvm::wasm::WASM_SEG_FLAG_STRINGS) {
     assert(s.Type == llvm::wasm::WASM_SEC_CUSTOM);
     comdat = s.Comdat;
     rawData = s.Content;
@@ -234,6 +235,7 @@ class SyntheticMergedChunk : public InputChunk {
 
   void addMergeChunk(MergeInputChunk *ms) {
     comdat = ms->getComdat();
+    alignment = std::max(alignment, ms->alignment);
     ms->parent = this;
     chunks.push_back(ms);
   }
@@ -337,8 +339,8 @@ class SyntheticFunction : public InputFunction {
 // Represents a single Wasm Section within an input file.
 class InputSection : public InputChunk {
 public:
-  InputSection(const WasmSection &s, ObjFile *f)
-      : InputChunk(f, InputChunk::Section, s.Name),
+  InputSection(const WasmSection &s, ObjFile *f, uint32_t alignment)
+      : InputChunk(f, InputChunk::Section, s.Name, alignment),
         tombstoneValue(getTombstoneForSection(s.Name)), section(s) {
     assert(section.Type == llvm::wasm::WASM_SEC_CUSTOM);
     comdat = section.Comdat;

lld/wasm/InputFiles.cpp

@@ -18,6 +18,7 @@
 #include "llvm/BinaryFormat/Wasm.h"
 #include "llvm/Object/Binary.h"
 #include "llvm/Object/Wasm.h"
+#include "llvm/ProfileData/InstrProf.h"
 #include "llvm/Support/Path.h"
 #include "llvm/Support/TarWriter.h"
 #include "llvm/Support/raw_ostream.h"
@@ -451,6 +452,18 @@ void SharedFile::parse() {
   }
 }
 
+// Returns the alignment for a custom section. This is used to concatenate
+// custom sections with the same name into a single custom section.
+static uint32_t getCustomSectionAlignment(const WasmSection &sec) {
+  // TODO: Add a section attribute for alignment in the linking spec.
+  if (sec.Name == getInstrProfSectionName(IPSK_covfun, Triple::Wasm) ||
+      sec.Name == getInstrProfSectionName(IPSK_covmap, Triple::Wasm)) {
+    // llvm-cov assumes that coverage metadata sections are 8-byte aligned.
+    return 8;
+  }
+  return 1;
+}
+
 WasmFileBase::WasmFileBase(Kind k, MemoryBufferRef m) : InputFile(k, m) {
   // Parse a memory buffer as a wasm file.
   LLVM_DEBUG(dbgs() << "Reading object: " << toString(this) << "\n");
@@ -520,10 +533,11 @@ void ObjFile::parse(bool ignoreComdats) {
       dataSection = &section;
     } else if (section.Type == WASM_SEC_CUSTOM) {
       InputChunk *customSec;
+      uint32_t alignment = getCustomSectionAlignment(section);
       if (shouldMerge(section))
-        customSec = make<MergeInputChunk>(section, this);
+        customSec = make<MergeInputChunk>(section, this, alignment);
       else
-        customSec = make<InputSection>(section, this);
+        customSec = make<InputSection>(section, this, alignment);
       customSec->discarded = isExcludedByComdat(customSec);
       customSections.emplace_back(customSec);
       customSections.back()->setRelocations(section.Relocations);

lld/wasm/OutputSections.cpp

@@ -244,6 +244,7 @@ void CustomSection::finalizeContents() {
 
   for (InputChunk *section : inputSections) {
     assert(!section->discarded);
+    payloadSize = alignTo(payloadSize, section->alignment);
     section->outSecOff = payloadSize;
     payloadSize += section->getSize();
   }

llvm/include/llvm/ProfileData/Coverage/CoverageMappingReader.h

@@ -180,6 +180,7 @@ class BinaryCoverageReader : public CoverageMappingReader {
   };
 
   using FuncRecordsStorage = std::unique_ptr<MemoryBuffer>;
+  using CoverageMapCopyStorage = std::unique_ptr<MemoryBuffer>;
 
 private:
   std::vector<std::string> Filenames;
@@ -195,9 +196,16 @@ class BinaryCoverageReader : public CoverageMappingReader {
   // D69471, which can split up function records into multiple sections on ELF.
   FuncRecordsStorage FuncRecords;
 
+  // Used to tie the lifetimes of an optional copy of the coverage mapping data
+  // to the lifetime of this BinaryCoverageReader instance. Needed to support
+  // Wasm object format, which might require realignment of section contents.
+  CoverageMapCopyStorage CoverageMapCopy;
+
   BinaryCoverageReader(std::unique_ptr<InstrProfSymtab> Symtab,
-                       FuncRecordsStorage &&FuncRecords)
-      : ProfileNames(std::move(Symtab)), FuncRecords(std::move(FuncRecords)) {}
+                       FuncRecordsStorage &&FuncRecords,
+                       CoverageMapCopyStorage &&CoverageMapCopy)
+      : ProfileNames(std::move(Symtab)), FuncRecords(std::move(FuncRecords)),
+        CoverageMapCopy(std::move(CoverageMapCopy)) {}
 
 public:
   BinaryCoverageReader(const BinaryCoverageReader &) = delete;
@@ -212,6 +220,7 @@ class BinaryCoverageReader : public CoverageMappingReader {
   static Expected<std::unique_ptr<BinaryCoverageReader>>
   createCoverageReaderFromBuffer(
       StringRef Coverage, FuncRecordsStorage &&FuncRecords,
+      CoverageMapCopyStorage &&CoverageMap,
       std::unique_ptr<InstrProfSymtab> ProfileNamesPtr, uint8_t BytesInAddress,
       llvm::endianness Endian, StringRef CompilationDir = "");
 

llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp

@@ -2171,7 +2171,11 @@ MCSection *TargetLoweringObjectFileWasm::getExplicitSectionGlobal(
   // This could be avoided if all data segements (the wasm sense) were
   // represented as their own sections (in the llvm sense).
   // TODO(sbc): https://github.com/WebAssembly/tool-conventions/issues/138
-  if (Name == ".llvmcmd" || Name == ".llvmbc")
+  if (Name == getInstrProfSectionName(IPSK_covmap, Triple::Wasm,
+                                      /*AddSegmentInfo=*/false) ||
+      Name == getInstrProfSectionName(IPSK_covfun, Triple::Wasm,
+                                      /*AddSegmentInfo=*/false) ||
+      Name == ".llvmbc" || Name == ".llvmcmd")
     Kind = SectionKind::getMetadata();
 
   StringRef Group = "";

llvm/lib/MC/MCContext.cpp

@@ -757,6 +757,11 @@ MCSectionWasm *MCContext::getWasmSection(const Twine &Section, SectionKind K,
   if (!Group.isTriviallyEmpty() && !Group.str().empty()) {
     GroupSym = cast<MCSymbolWasm>(getOrCreateSymbol(Group));
     GroupSym->setComdat(true);
+    if (K.isMetadata() && !GroupSym->getType().has_value()) {
+      // Comdat group symbol associated with a custom section is a section
+      // symbol (not a data symbol).
+      GroupSym->setType(wasm::WASM_SYMBOL_TYPE_SECTION);
+    }
   }
 
   return getWasmSection(Section, K, Flags, GroupSym, UniqueID);