Source code

Revision control

Other Tools

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*-
* vim: set ts=8 sts=2 et sw=2 tw=80:
* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#ifndef vm_SharedStencil_h
#define vm_SharedStencil_h
#include "mozilla/Assertions.h" // MOZ_ASSERT, MOZ_CRASH
#include "mozilla/Atomics.h" // mozilla::{Atomic, SequentiallyConsistent}
#include "mozilla/CheckedInt.h" // mozilla::CheckedInt
#include "mozilla/HashFunctions.h" // mozilla::HahNumber, mozilla::HashBytes
#include "mozilla/HashTable.h" // mozilla::HashSet
#include "mozilla/MemoryReporting.h" // mozilla::MallocSizeOf
#include "mozilla/RefPtr.h" // RefPtr
#include "mozilla/Span.h" // mozilla::Span
#include <stddef.h> // size_t
#include <stdint.h> // uint8_t, uint16_t, uint32_t
#include "frontend/SourceNotes.h" // js::SrcNote
#include "frontend/TypedIndex.h" // js::frontend::TypedIndex
#include "js/AllocPolicy.h" // js::SystemAllocPolicy
#include "js/TypeDecls.h" // JSContext,jsbytecode
#include "js/UniquePtr.h" // js::UniquePtr
#include "util/EnumFlags.h" // js::EnumFlags
#include "util/TrailingArray.h" // js::TrailingArray
#include "vm/StencilEnums.h" // js::{TryNoteKind,ImmutableScriptFlagsEnum,MutableScriptFlagsEnum}
//
// Data structures shared between Stencil and the VM.
//
namespace js {
namespace frontend {
class StencilXDR;
} // namespace frontend
// Index into gcthings array.
class GCThingIndexType;
class GCThingIndex : public frontend::TypedIndex<GCThingIndexType> {
// Delegate constructors;
using Base = frontend::TypedIndex<GCThingIndexType>;
using Base::Base;
public:
static constexpr GCThingIndex outermostScopeIndex() {
return GCThingIndex(0);
}
static constexpr GCThingIndex invalid() { return GCThingIndex(UINT32_MAX); }
GCThingIndex next() const { return GCThingIndex(index + 1); }
};
/*
* Exception handling record.
*/
struct TryNote {
uint32_t kind_; /* one of TryNoteKind */
uint32_t stackDepth; /* stack depth upon exception handler entry */
uint32_t start; /* start of the try statement or loop relative
to script->code() */
uint32_t length; /* length of the try statement or loop */
TryNote(uint32_t kind, uint32_t stackDepth, uint32_t start, uint32_t length)
: kind_(kind), stackDepth(stackDepth), start(start), length(length) {}
TryNote() = default;
TryNoteKind kind() const { return TryNoteKind(kind_); }
bool isLoop() const {
switch (kind()) {
case TryNoteKind::Loop:
case TryNoteKind::ForIn:
case TryNoteKind::ForOf:
return true;
case TryNoteKind::Catch:
case TryNoteKind::Finally:
case TryNoteKind::ForOfIterClose:
case TryNoteKind::Destructuring:
return false;
}
MOZ_CRASH("Unexpected try note kind");
}
};
// A block scope has a range in bytecode: it is entered at some offset, and left
// at some later offset. Scopes can be nested. Given an offset, the
// ScopeNote containing that offset whose with the highest start value
// indicates the block scope. The block scope list is sorted by increasing
// start value.
//
// It is possible to leave a scope nonlocally, for example via a "break"
// statement, so there may be short bytecode ranges in a block scope in which we
// are popping the block chain in preparation for a goto. These exits are also
// nested with respect to outer scopes. The scopes in these exits are indicated
// by the "index" field, just like any other block. If a nonlocal exit pops the
// last block scope, the index will be NoScopeIndex.
//
struct ScopeNote {
// Sentinel index for no Scope.
static constexpr GCThingIndex NoScopeIndex = GCThingIndex::invalid();
// Sentinel index for no ScopeNote.
static const uint32_t NoScopeNoteIndex = UINT32_MAX;
// Index of the js::Scope in the script's gcthings array, or NoScopeIndex if
// there is no block scope in this range.
GCThingIndex index;
// Bytecode offset at which this scope starts relative to script->code().
uint32_t start = 0;
// Length of bytecode span this scope covers.
uint32_t length = 0;
// Index of parent block scope in notes, or NoScopeNoteIndex.
uint32_t parent = 0;
};
// Range of characters in scriptSource which contains a script's source,
// that is, the range used by the Parser to produce a script.
//
// For most functions the fields point to the following locations.
//
// function * foo(a, b) { return a + b; }
// ^ ^ ^
// | | |
// | sourceStart sourceEnd
// | |
// toStringStart toStringEnd
//
// For the special case of class constructors, the spec requires us to use an
// alternate definition of toStringStart / toStringEnd.
//
// class C { constructor() { this.field = 42; } }
// ^ ^ ^ ^
// | | | |
// | sourceStart sourceEnd |
// | |
// toStringStart toStringEnd
//
// Implicit class constructors use the following definitions.
//
// class C { someMethod() { } }
// ^ ^
// | |
// sourceStart sourceEnd
// | |
// toStringStart toStringEnd
//
// Field initializer lambdas are internal details of the engine, but we still
// provide a sensible definition of these values.
//
// class C { static field = 1 }
// class C { field = 1 }
// class C { somefield }
// ^ ^
// | |
// sourceStart sourceEnd
//
// The non-static private class methods (including getters and setters) ALSO
// create a hidden initializer lambda in addition to the method itself. These
// lambdas are not exposed directly to script.
//
// class C { #field() { } }
// class C { get #field() { } }
// class C { async #field() { } }
// class C { * #field() { } }
// ^ ^
// | |
// sourceStart sourceEnd
//
// NOTE: These are counted in Code Units from the start of the script source.
//
// Also included in the SourceExtent is the line and column numbers of the
// sourceStart position. Compilation options may specify the initial line and
// column number.
//
// NOTE: Column number may saturate and must not be used as unique identifier.
struct SourceExtent {
SourceExtent() = default;
SourceExtent(uint32_t sourceStart, uint32_t sourceEnd, uint32_t toStringStart,
uint32_t toStringEnd, uint32_t lineno, uint32_t column)
: sourceStart(sourceStart),
sourceEnd(sourceEnd),
toStringStart(toStringStart),
toStringEnd(toStringEnd),
lineno(lineno),
column(column) {}
static SourceExtent makeGlobalExtent(uint32_t len) {
return SourceExtent(0, len, 0, len, 1, 0);
}
static SourceExtent makeGlobalExtent(uint32_t len, uint32_t lineno,
uint32_t column) {
return SourceExtent(0, len, 0, len, lineno, column);
}
uint32_t sourceStart = 0;
uint32_t sourceEnd = 0;
uint32_t toStringStart = 0;
uint32_t toStringEnd = 0;
// Line and column of |sourceStart_| position.
uint32_t lineno = 1; // 1-indexed.
uint32_t column = 0; // Count of Code Points
};
class ImmutableScriptFlags : public EnumFlags<ImmutableScriptFlagsEnum> {
public:
ImmutableScriptFlags() = default;
explicit ImmutableScriptFlags(FieldType rawFlags) : EnumFlags(rawFlags) {}
operator FieldType() const { return flags_; }
};
class MutableScriptFlags : public EnumFlags<MutableScriptFlagsEnum> {
public:
MutableScriptFlags() = default;
MutableScriptFlags& operator&=(const FieldType rhs) {
flags_ &= rhs;
return *this;
}
MutableScriptFlags& operator|=(const FieldType rhs) {
flags_ |= rhs;
return *this;
}
operator FieldType() const { return flags_; }
};
// [SMDOC] JSScript data layout (immutable)
//
// ImmutableScriptData stores variable-length script data that may be shared
// between scripts with the same bytecode, even across different GC Zones.
// Abstractly this structure consists of multiple (optional) arrays that are
// exposed as mozilla::Span<T>. These arrays exist in a single heap allocation.
//
// Under the hood, ImmutableScriptData is a fixed-size header class followed
// the various array bodies interleaved with metadata to compactly encode the
// bounds. These arrays have varying requirements for alignment, performance,
// and jit-friendliness which leads to the complex indexing system below.
//
// Note: The '----' separators are for readability only.
//
// ----
// <ImmutableScriptData itself>
// ----
// (REQUIRED) Flags structure
// (REQUIRED) Array of jsbytecode constituting code()
// (REQUIRED) Array of SrcNote constituting notes()
// ----
// (OPTIONAL) Array of uint32_t optional-offsets
// optArrayOffset:
// ----
// L0:
// (OPTIONAL) Array of uint32_t constituting resumeOffsets()
// L1:
// (OPTIONAL) Array of ScopeNote constituting scopeNotes()
// L2:
// (OPTIONAL) Array of TryNote constituting tryNotes()
// L3:
// ----
//
// NOTE: The notes() array must have been null-padded such that
// flags/code/notes together have uint32_t alignment.
//
// The labels shown are recorded as byte-offsets relative to 'this'. This is to
// reduce memory as well as make ImmutableScriptData easier to share across
// processes.
//
// The L0/L1/L2/L3 labels indicate the start and end of the optional arrays.
// Some of these labels may refer to the same location if the array between
// them is empty. Each unique label position has an offset stored in the
// optional-offsets table. Note that we also avoid entries for labels that
// match 'optArrayOffset'. This saves memory when arrays are empty.
//
// The flags() data indicates (for each optional array) which entry from the
// optional-offsets table marks the *end* of array. The array starts where the
// previous array ends (with the first array beginning at 'optArrayOffset').
// The optional-offset table is addressed at negative indices from
// 'optArrayOffset'.
//
// In general, the length of each array is computed from subtracting the start
// offset of the array from the start offset of the subsequent array. The
// notable exception is that bytecode length is stored explicitly.
class alignas(uint32_t) ImmutableScriptData final : public TrailingArray {
private:
Offset optArrayOffset_ = 0;
// Length of bytecode
uint32_t codeLength_ = 0;
public:
// Offset of main entry point from code, after predef'ing prologue.
uint32_t mainOffset = 0;
// Fixed frame slots.
uint32_t nfixed = 0;
// Slots plus maximum stack depth.
uint32_t nslots = 0;
// Index into the gcthings array of the body scope.
GCThingIndex bodyScopeIndex;
// Number of IC entries to allocate in JitScript for Baseline ICs.
uint32_t numICEntries = 0;
// ES6 function length.
uint16_t funLength = 0;
// NOTE: The raw bytes of this structure are used for hashing so use explicit
// padding values as needed for predicatable results across compilers.
uint16_t padding = 0;
private:
struct Flags {
uint8_t resumeOffsetsEndIndex : 2;
uint8_t scopeNotesEndIndex : 2;
uint8_t tryNotesEndIndex : 2;
uint8_t _unused : 2;
};
static_assert(sizeof(Flags) == sizeof(uint8_t),
"Structure packing is broken");
// Offsets (in bytes) from 'this' to each component array. The delta between
// each offset and the next offset is the size of each array and is defined
// even if an array is empty.
Offset flagOffset() const { return offsetOfCode() - sizeof(Flags); }
Offset codeOffset() const { return offsetOfCode(); }
Offset noteOffset() const { return offsetOfCode() + codeLength_; }
Offset optionalOffsetsOffset() const {
// Determine the location to beginning of optional-offsets array by looking
// at index for try-notes.
//
// optionalOffsetsOffset():
// (OPTIONAL) tryNotesEndOffset
// (OPTIONAL) scopeNotesEndOffset
// (OPTIONAL) resumeOffsetsEndOffset
// optArrayOffset_:
// ....
unsigned numOffsets = flags().tryNotesEndIndex;
MOZ_ASSERT(numOffsets >= flags().scopeNotesEndIndex);
MOZ_ASSERT(numOffsets >= flags().resumeOffsetsEndIndex);
return optArrayOffset_ - (numOffsets * sizeof(Offset));
}
Offset resumeOffsetsOffset() const { return optArrayOffset_; }
Offset scopeNotesOffset() const {
return getOptionalOffset(flags().resumeOffsetsEndIndex);
}
Offset tryNotesOffset() const {
return getOptionalOffset(flags().scopeNotesEndIndex);
}
Offset endOffset() const {
return getOptionalOffset(flags().tryNotesEndIndex);
}
void initOptionalArrays(Offset* cursor, uint32_t numResumeOffsets,
uint32_t numScopeNotes, uint32_t numTryNotes);
// Initialize to GC-safe state
ImmutableScriptData(uint32_t codeLength, uint32_t noteLength,
uint32_t numResumeOffsets, uint32_t numScopeNotes,
uint32_t numTryNotes);
void setOptionalOffset(int index, Offset offset) {
MOZ_ASSERT(index > 0);
MOZ_ASSERT(offset != optArrayOffset_, "Do not store implicit offset");
offsetToPointer<Offset>(optArrayOffset_)[-index] = offset;
}
Offset getOptionalOffset(int index) const {
// The index 0 represents (implicitly) the offset 'optArrayOffset_'.
if (index == 0) {
return optArrayOffset_;
}
ImmutableScriptData* this_ = const_cast<ImmutableScriptData*>(this);
return this_->offsetToPointer<Offset>(optArrayOffset_)[-index];
}
public:
static js::UniquePtr<ImmutableScriptData> new_(
JSContext* cx, uint32_t mainOffset, uint32_t nfixed, uint32_t nslots,
GCThingIndex bodyScopeIndex, uint32_t numICEntries, bool isFunction,
uint16_t funLength, mozilla::Span<const jsbytecode> code,
mozilla::Span<const SrcNote> notes,
mozilla::Span<const uint32_t> resumeOffsets,
mozilla::Span<const ScopeNote> scopeNotes,
mozilla::Span<const TryNote> tryNotes);
static js::UniquePtr<ImmutableScriptData> new_(
JSContext* cx, uint32_t codeLength, uint32_t noteLength,
uint32_t numResumeOffsets, uint32_t numScopeNotes, uint32_t numTryNotes);
static js::UniquePtr<ImmutableScriptData> new_(JSContext* cx,
uint32_t totalSize);
#ifdef DEBUG
// Validate the content, after XDR decoding.
void validate(uint32_t totalSize);
#endif
private:
static mozilla::CheckedInt<uint32_t> sizeFor(uint32_t codeLength,
uint32_t noteLength,
uint32_t numResumeOffsets,
uint32_t numScopeNotes,
uint32_t numTryNotes);
public:
// The code() and note() arrays together maintain an target alignment by
// padding the source notes with null. This allows arrays with stricter
// alignment requirements to follow them.
static constexpr size_t CodeNoteAlign = sizeof(uint32_t);
// Compute number of null notes to pad out source notes with.
static uint32_t ComputeNotePadding(uint32_t codeLength, uint32_t noteLength) {
uint32_t flagLength = sizeof(Flags);
uint32_t nullLength =
CodeNoteAlign - (flagLength + codeLength + noteLength) % CodeNoteAlign;
// The source notes must have at least one null-terminator.
MOZ_ASSERT(nullLength >= 1);
return nullLength;
}
// Span over all raw bytes in this struct and its trailing arrays.
mozilla::Span<const uint8_t> immutableData() const {
size_t allocSize = endOffset();
return mozilla::Span{reinterpret_cast<const uint8_t*>(this), allocSize};
}
private:
Flags& flagsRef() { return *offsetToPointer<Flags>(flagOffset()); }
const Flags& flags() const {
return const_cast<ImmutableScriptData*>(this)->flagsRef();
}
public:
uint32_t codeLength() const { return codeLength_; }
jsbytecode* code() { return offsetToPointer<jsbytecode>(codeOffset()); }
mozilla::Span<jsbytecode> codeSpan() { return {code(), codeLength()}; }
uint32_t noteLength() const {
return numElements<SrcNote>(noteOffset(), optionalOffsetsOffset());
}
SrcNote* notes() { return offsetToPointer<SrcNote>(noteOffset()); }
mozilla::Span<SrcNote> notesSpan() { return {notes(), noteLength()}; }
mozilla::Span<uint32_t> resumeOffsets() {
return mozilla::Span{offsetToPointer<uint32_t>(resumeOffsetsOffset()),
offsetToPointer<uint32_t>(scopeNotesOffset())};
}
mozilla::Span<ScopeNote> scopeNotes() {
return mozilla::Span{offsetToPointer<ScopeNote>(scopeNotesOffset()),
offsetToPointer<ScopeNote>(tryNotesOffset())};
}
mozilla::Span<TryNote> tryNotes() {
return mozilla::Span{offsetToPointer<TryNote>(tryNotesOffset()),
offsetToPointer<TryNote>(endOffset())};
}
// Expose offsets to the JITs.
static constexpr size_t offsetOfCode() {
return sizeof(ImmutableScriptData) + sizeof(Flags);
}
static constexpr size_t offsetOfResumeOffsetsOffset() {
// Resume-offsets are the first optional array if they exist. Locate the
// array with the 'optArrayOffset_' field.
static_assert(sizeof(Offset) == sizeof(uint32_t),
"JIT expect Offset to be uint32_t");
return offsetof(ImmutableScriptData, optArrayOffset_);
}
static constexpr size_t offsetOfNfixed() {
return offsetof(ImmutableScriptData, nfixed);
}
static constexpr size_t offsetOfNslots() {
return offsetof(ImmutableScriptData, nslots);
}
static constexpr size_t offsetOfFunLength() {
return offsetof(ImmutableScriptData, funLength);
}
// ImmutableScriptData has trailing data so isn't copyable or movable.
ImmutableScriptData(const ImmutableScriptData&) = delete;
ImmutableScriptData& operator=(const ImmutableScriptData&) = delete;
};
// Wrapper type for ImmutableScriptData to allow sharing across a JSRuntime.
//
// Note: This is distinct from ImmutableScriptData because it contains a mutable
// ref-count while the ImmutableScriptData may live in read-only memory.
//
// Note: This is *not* directly inlined into the SharedImmutableScriptDataTable
// because scripts point directly to object and table resizing moves
// entries. This allows for fast finalization by decrementing the
// ref-count directly without doing a hash-table lookup.
class SharedImmutableScriptData {
// This class is reference counted as follows: each pointer from a JSScript
// counts as one reference plus there may be one reference from the shared
// script data table.
mozilla::Atomic<uint32_t, mozilla::SequentiallyConsistent> refCount_ = {};
js::UniquePtr<ImmutableScriptData> isd_ = nullptr;
// End of fields.
friend class ::JSScript;
friend class js::frontend::StencilXDR;
public:
SharedImmutableScriptData() = default;
// Hash over the contents of SharedImmutableScriptData and its
// ImmutableScriptData.
struct Hasher;
uint32_t refCount() const { return refCount_; }
void AddRef() { refCount_++; }
void Release() {
MOZ_ASSERT(refCount_ != 0);
uint32_t remain = --refCount_;
if (remain == 0) {
isd_ = nullptr;
js_free(this);
}
}
static constexpr size_t offsetOfISD() {
return offsetof(SharedImmutableScriptData, isd_);
}
private:
static SharedImmutableScriptData* create(JSContext* cx);
public:
static SharedImmutableScriptData* createWith(
JSContext* cx, js::UniquePtr<ImmutableScriptData>&& isd);
size_t sizeOfIncludingThis(mozilla::MallocSizeOf mallocSizeOf) {
return mallocSizeOf(this) + mallocSizeOf(isd_.get());
}
// SharedImmutableScriptData has trailing data so isn't copyable or movable.
SharedImmutableScriptData(const SharedImmutableScriptData&) = delete;
SharedImmutableScriptData& operator=(const SharedImmutableScriptData&) =
delete;
static bool shareScriptData(JSContext* cx,
RefPtr<SharedImmutableScriptData>& sisd);
size_t immutableDataLength() const { return isd_->immutableData().Length(); }
uint32_t nfixed() const { return isd_->nfixed; }
};
// Matches SharedImmutableScriptData objects that have the same atoms as well as
// contain the same bytes in their ImmutableScriptData.
struct SharedImmutableScriptData::Hasher {
using Lookup = RefPtr<SharedImmutableScriptData>;
static mozilla::HashNumber hash(const Lookup& l) {
mozilla::Span<const uint8_t> immutableData = l->isd_->immutableData();
return mozilla::HashBytes(immutableData.data(), immutableData.size());
}
static bool match(SharedImmutableScriptData* entry, const Lookup& lookup) {
return (entry->isd_->immutableData() == lookup->isd_->immutableData());
}
};
using SharedImmutableScriptDataTable =
mozilla::HashSet<SharedImmutableScriptData*,
SharedImmutableScriptData::Hasher, SystemAllocPolicy>;
struct MemberInitializers {
static constexpr uint32_t MaxInitializers = INT32_MAX;
#ifdef DEBUG
bool valid = false;
#endif
// This struct will eventually have a vector of constant values for optimizing
// field initializers.
uint32_t numMemberInitializers = 0;
explicit MemberInitializers(uint32_t numMemberInitializers)
:
#ifdef DEBUG
valid(true),
#endif
numMemberInitializers(numMemberInitializers) {
}
static MemberInitializers Invalid() { return MemberInitializers(); }
// Singleton to use for class constructors that do not have to initialize any
// fields. This is used when we elide the trivial data but still need a valid
// set to stop scope walking.
static const MemberInitializers& Empty() {
static const MemberInitializers zeroInitializers(0);
return zeroInitializers;
}
uint32_t serialize() const { return numMemberInitializers; }
private:
MemberInitializers() = default;
};
} // namespace js
#endif /* vm_SharedStencil_h */