gecko/js/src/jsanalyze.cpp

778 lines
26 KiB
C++

/* -*- Mode: c++; c-basic-offset: 4; tab-width: 40; indent-tabs-mode: nil -*- */
/* vim: set ts=40 sw=4 et tw=99: */
/* ***** BEGIN LICENSE BLOCK *****
* Version: MPL 1.1/GPL 2.0/LGPL 2.1
*
* The contents of this file are subject to the Mozilla Public License Version
* 1.1 (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
* http://www.mozilla.org/MPL/
*
* Software distributed under the License is distributed on an "AS IS" basis,
* WITHOUT WARRANTY OF ANY KIND, either express or implied. See the License
* for the specific language governing rights and limitations under the
* License.
*
* The Original Code is the Mozilla SpiderMonkey bytecode analysis
*
* The Initial Developer of the Original Code is
* Mozilla Foundation
* Portions created by the Initial Developer are Copyright (C) 2010
* the Initial Developer. All Rights Reserved.
*
* Contributor(s):
* Brian Hackett <bhackett@mozilla.com>
*
* Alternatively, the contents of this file may be used under the terms of
* either of the GNU General Public License Version 2 or later (the "GPL"),
* or the GNU Lesser General Public License Version 2.1 or later (the "LGPL"),
* in which case the provisions of the GPL or the LGPL are applicable instead
* of those above. If you wish to allow use of your version of this file only
* under the terms of either the GPL or the LGPL, and not to allow others to
* use your version of this file under the terms of the MPL, indicate your
* decision by deleting the provisions above and replace them with the notice
* and other provisions required by the GPL or the LGPL. If you do not delete
* the provisions above, a recipient may use your version of this file under
* the terms of any one of the MPL, the GPL or the LGPL.
*
* ***** END LICENSE BLOCK ***** */
#include "jsanalyze.h"
#include "jsautooplen.h"
#include "jscompartment.h"
#include "jscntxt.h"
#include "jsinferinlines.h"
namespace js {
namespace analyze {
/////////////////////////////////////////////////////////////////////
// Script
/////////////////////////////////////////////////////////////////////
void
Script::destroy()
{
JS_FinishArenaPool(&pool);
}
/////////////////////////////////////////////////////////////////////
// Bytecode
/////////////////////////////////////////////////////////////////////
bool
Bytecode::mergeDefines(JSContext *cx, Script *script, bool initial,
unsigned newDepth, types::TypeStack *newStack,
uint32 *newArray, unsigned newCount)
{
if (initial) {
/*
* Haven't handled any incoming edges to this bytecode before.
* Define arrays are copy on write, so just reuse the array for this bytecode.
*/
stackDepth = newDepth;
defineArray = newArray;
defineCount = newCount;
#ifdef JS_TYPE_INFERENCE
inStack = newStack;
#endif
return true;
}
#ifdef JS_TYPE_INFERENCE
types::TypeStack::merge(cx, newStack, inStack);
#endif
/*
* This bytecode has multiple incoming edges, intersect the new array with any
* variables known to be defined along other incoming edges.
*/
if (analyzed) {
#ifdef DEBUG
/*
* Once analyzed, a bytecode has its full set of definitions. There are two
* properties we depend on to ensure this. First, bytecode for a function
* is emitted in topological order, and since we analyze bytecodes in the
* order they were emitted we will have seen all incoming jumps except
* for any loop back edges. Second, javascript has structured control flow,
* so loop heads dominate their bodies; the set of variables defined
* on a back edge will be at least as large as at the head of the loop,
* so no intersection or further analysis needs to be done.
*/
JS_ASSERT(stackDepth == newDepth);
for (unsigned i = 0; i < defineCount; i++) {
bool found = false;
for (unsigned j = 0; j < newCount; j++) {
if (newArray[j] == defineArray[i])
found = true;
}
JS_ASSERT(found);
}
#endif
} else {
JS_ASSERT(stackDepth == newDepth);
bool owned = false;
for (unsigned i = 0; i < defineCount; i++) {
bool found = false;
for (unsigned j = 0; j < newCount; j++) {
if (newArray[j] == defineArray[i])
found = true;
}
if (!found) {
/*
* Get a mutable copy of the defines. This can end up making
* several copies for a bytecode if it has many incoming edges
* with progressively smaller sets of defined variables.
*/
if (!owned) {
uint32 *reallocArray = ArenaArray<uint32>(script->pool, defineCount);
if (!reallocArray) {
script->setOOM(cx);
return false;
}
memcpy(reallocArray, defineArray, defineCount * sizeof(uint32));
defineArray = reallocArray;
owned = true;
}
/* Swap with the last element and pop the array. */
defineArray[i--] = defineArray[--defineCount];
}
}
}
return true;
}
/////////////////////////////////////////////////////////////////////
// Analysis
/////////////////////////////////////////////////////////////////////
inline bool
Script::addJump(JSContext *cx, unsigned offset,
unsigned *currentOffset, unsigned *forwardJump,
unsigned stackDepth, types::TypeStack *stack,
uint32 *defineArray, unsigned defineCount)
{
JS_ASSERT(offset < script->length);
Bytecode *&code = codeArray[offset];
bool initial = (code == NULL);
if (initial) {
code = ArenaNew<Bytecode>(pool, this, offset);
if (!code) {
setOOM(cx);
return false;
}
}
if (!code->mergeDefines(cx, this, initial, stackDepth, stack, defineArray, defineCount))
return false;
code->jumpTarget = true;
if (offset < *currentOffset) {
/* Don't follow back edges to bytecode which has already been analyzed. */
if (!code->analyzed) {
if (*forwardJump == 0)
*forwardJump = *currentOffset;
*currentOffset = offset;
}
} else if (offset > *forwardJump) {
*forwardJump = offset;
}
return true;
}
inline void
Script::setLocal(uint32 local, uint32 offset)
{
JS_ASSERT(local < localCount());
JS_ASSERT(offset != LOCAL_CONDITIONALLY_DEFINED);
/*
* It isn't possible to change the point when a variable becomes unconditionally
* defined, or to mark it as unconditionally defined after it has already been
* marked as having a use before def. It *is* possible to mark it as having
* a use before def after marking it as unconditionally defined. In a loop such as:
*
* while ((a = b) != 0) { x = a; }
*
* When walking through the body of this loop, we will first analyze the test
* (which comes after the body in the bytecode stream) as unconditional code,
* and mark a as definitely defined. a is not in the define array when taking
* the loop's back edge, so it is treated as possibly undefined when written to x.
*/
JS_ASSERT(locals[local] == LOCAL_CONDITIONALLY_DEFINED ||
locals[local] == offset || offset == LOCAL_USE_BEFORE_DEF);
locals[local] = offset;
}
static inline ptrdiff_t
GetJumpOffset(jsbytecode *pc, jsbytecode *pc2)
{
uint32 type = JOF_OPTYPE(*pc);
if (JOF_TYPE_IS_EXTENDED_JUMP(type))
return GET_JUMPX_OFFSET(pc2);
return GET_JUMP_OFFSET(pc2);
}
// return whether op bytecodes do not fallthrough (they may do a jump).
static inline bool
BytecodeNoFallThrough(JSOp op)
{
switch (op) {
case JSOP_GOTO:
case JSOP_GOTOX:
case JSOP_DEFAULT:
case JSOP_DEFAULTX:
case JSOP_RETURN:
case JSOP_STOP:
case JSOP_RETRVAL:
case JSOP_THROW:
case JSOP_TABLESWITCH:
case JSOP_TABLESWITCHX:
case JSOP_LOOKUPSWITCH:
case JSOP_LOOKUPSWITCHX:
case JSOP_FILTER:
return true;
case JSOP_GOSUB:
case JSOP_GOSUBX:
// these fall through indirectly, after executing a 'finally'.
return false;
default:
return false;
}
}
/* Untrap a single PC, and retrap it at scope exit. */
struct UntrapOpcode
{
jsbytecode *pc;
bool trap;
UntrapOpcode(JSContext *cx, JSScript *script, jsbytecode *pc)
: pc(pc), trap(JSOp(*pc) == JSOP_TRAP)
{
if (trap)
*pc = JS_GetTrapOpcode(cx, script, pc);
}
~UntrapOpcode()
{
if (trap)
*pc = JSOP_TRAP;
}
};
#ifdef JS_TYPE_INFERENCE
/*
* Information about a currently active static initializer. We keep the stack
* of initializers around during analysis so we can reuse objects when
* processing arrays of arrays or arrays of objects.
*/
struct InitializerInfo
{
/* Object being initialized. */
types::TypeObject *object;
/* Whether this object is an array. */
bool isArray;
/* Any object to use for initializers appearing in the array's elements. */
types::TypeObject *initObject;
/* Whether initObject is an array vs. object. */
bool initArray;
/* Outer initializer, the one this initializer is nested in. */
InitializerInfo *outer;
InitializerInfo() { PodZero(this); }
};
#endif /* JS_TYPE_INFERENCE */
void
Script::init(JSScript *script)
{
this->script = script;
JS_InitArenaPool(&pool, "script_analyze", 256, 8, NULL);
}
void
Script::analyze(JSContext *cx)
{
JS_ASSERT(script && !codeArray && !locals);
unsigned length = script->length;
unsigned nfixed = localCount();
codeArray = ArenaArray<Bytecode*>(pool, length);
locals = ArenaArray<uint32>(pool, nfixed);
if (!codeArray || !locals) {
setOOM(cx);
return;
}
PodZero(codeArray, length);
for (unsigned i = 0; i < nfixed; i++)
locals[i] = LOCAL_CONDITIONALLY_DEFINED;
/*
* Treat locals as having a possible use-before-def if they could be accessed
* by debug code or by eval, or if they could be accessed by an inner script.
*/
if (script->usesEval || cx->compartment->debugMode) {
for (uint32 i = 0; i < nfixed; i++)
setLocal(i, LOCAL_USE_BEFORE_DEF);
}
for (uint32 i = 0; i < script->nClosedVars; i++) {
uint32 slot = script->getClosedVar(i);
if (slot < nfixed)
setLocal(slot, LOCAL_USE_BEFORE_DEF);
}
/*
* If the script is in debug mode, JS_SetFrameReturnValue can be called at
* any safe point.
*/
if (cx->compartment->debugMode)
usesRval = true;
/*
* If we are in the middle of one or more jumps, the offset of the highest
* target jumping over this bytecode. Includes implicit jumps from
* try/catch/finally blocks.
*/
unsigned forwardJump = 0;
/*
* If we are in the middle of a try block, the offset of the highest
* catch/finally/enditer.
*/
unsigned forwardCatch = 0;
#ifdef JS_TYPE_INFERENCE
/* Stack of active initializers. */
InitializerInfo *initializerStack = NULL;
#endif
/* Fill in stack depth and definitions at initial bytecode. */
Bytecode *startcode = ArenaNew<Bytecode>(pool, this, 0);
if (!startcode) {
setOOM(cx);
return;
}
startcode->stackDepth = 0;
codeArray[0] = startcode;
unsigned offset, nextOffset = 0;
while (nextOffset < length) {
offset = nextOffset;
JS_ASSERT(forwardCatch <= forwardJump);
/* Check if the current forward jump/try-block has finished. */
if (forwardJump && forwardJump == offset)
forwardJump = 0;
if (forwardCatch && forwardCatch == offset)
forwardCatch = 0;
Bytecode *code = maybeCode(offset);
jsbytecode *pc = script->code + offset;
UntrapOpcode untrap(cx, script, pc);
JSOp op = (JSOp)*pc;
JS_ASSERT(op < JSOP_LIMIT);
/* Immediate successor of this bytecode. */
unsigned successorOffset = offset + GetBytecodeLength(pc);
/*
* Next bytecode to analyze. This is either the successor, or is an
* earlier bytecode if this bytecode has a loop backedge.
*/
nextOffset = successorOffset;
if (!code) {
/* Haven't found a path by which this bytecode is reachable. */
continue;
}
if (code->analyzed) {
/* No need to reanalyze, see Bytecode::mergeDefines. */
continue;
}
code->analyzed = true;
if (forwardCatch)
code->inTryBlock = true;
unsigned stackDepth = code->stackDepth;
uint32 *defineArray = code->defineArray;
unsigned defineCount = code->defineCount;
if (!forwardJump) {
/*
* There is no jump over this bytecode, nor a containing try block.
* Either this bytecode will definitely be executed, or an exception
* will be thrown which the script does not catch. Either way,
* any variables definitely defined at this bytecode will stay
* defined throughout the rest of the script. We just need to
* remember the offset where the variable became unconditionally
* defined, rather than continue to maintain it in define arrays.
*/
for (unsigned i = 0; i < defineCount; i++) {
uint32 local = defineArray[i];
JS_ASSERT_IF(locals[local] != LOCAL_CONDITIONALLY_DEFINED &&
locals[local] != LOCAL_USE_BEFORE_DEF,
locals[local] <= offset);
if (locals[local] == LOCAL_CONDITIONALLY_DEFINED)
setLocal(local, offset);
}
defineArray = NULL;
defineCount = 0;
}
unsigned nuses = GetUseCount(script, offset);
unsigned ndefs = GetDefCount(script, offset);
JS_ASSERT(stackDepth >= nuses);
stackDepth -= nuses;
stackDepth += ndefs;
types::TypeStack *stack = NULL;
#ifdef JS_TYPE_INFERENCE
stack = code->inStack;
for (unsigned i = 0; i < nuses; i++)
stack = stack->group()->innerStack;
code->pushedArray = ArenaArray<types::TypeStack>(pool, ndefs);
PodZero(code->pushedArray, ndefs);
for (unsigned i = 0; i < ndefs; i++) {
code->pushedArray[i].types.setPool(&pool);
code->pushedArray[i].setInnerStack(stack);
stack = &code->pushedArray[i];
types::InferSpew(types::ISpewOps, "pushed #%u:%05u %u T%u",
id, offset, i, stack->types.id());
}
/* Track the initializer stack and compute new objects for encountered initializers. */
if (op == JSOP_NEWINIT || op == JSOP_NEWARRAY || op == JSOP_NEWOBJECT) {
bool newArray = (op == JSOP_NEWARRAY) || (op == JSOP_NEWINIT && pc[1] == JSProto_Array);
types::TypeObject *object;
if (initializerStack && initializerStack->initObject &&
initializerStack->initArray == newArray) {
object = initializerStack->initObject;
if (newArray)
code->initArray = object;
else
code->initObject = object;
} else {
object = code->getInitObject(cx, newArray);
if (initializerStack && initializerStack->isArray) {
initializerStack->initObject = object;
initializerStack->initArray = newArray;
}
}
InitializerInfo *info = (InitializerInfo *) cx->calloc(sizeof(InitializerInfo));
info->outer = initializerStack;
info->object = object;
info->isArray = newArray;
initializerStack = info;
} else if (op == JSOP_INITELEM || op == JSOP_INITPROP || op == JSOP_INITMETHOD) {
JS_ASSERT(initializerStack);
code->initObject = initializerStack->object;
} else if (op == JSOP_ENDINIT) {
JS_ASSERT(initializerStack);
InitializerInfo *info = initializerStack;
initializerStack = initializerStack->outer;
cx->free(info);
}
#endif /* JS_TYPE_INFERENCE */
switch (op) {
case JSOP_SETRVAL:
case JSOP_POPV:
usesRval = true;
break;
case JSOP_NAME:
case JSOP_CALLNAME:
case JSOP_BINDNAME:
case JSOP_SETNAME:
case JSOP_DELNAME:
case JSOP_INCNAME:
case JSOP_DECNAME:
case JSOP_NAMEINC:
case JSOP_NAMEDEC:
case JSOP_FORNAME:
usesScope = true;
break;
case JSOP_TABLESWITCH:
case JSOP_TABLESWITCHX: {
jsbytecode *pc2 = pc;
unsigned jmplen = (op == JSOP_TABLESWITCH) ? JUMP_OFFSET_LEN : JUMPX_OFFSET_LEN;
unsigned defaultOffset = offset + GetJumpOffset(pc, pc2);
pc2 += jmplen;
jsint low = GET_JUMP_OFFSET(pc2);
pc2 += JUMP_OFFSET_LEN;
jsint high = GET_JUMP_OFFSET(pc2);
pc2 += JUMP_OFFSET_LEN;
if (!addJump(cx, defaultOffset, &nextOffset, &forwardJump,
stackDepth, stack, defineArray, defineCount)) {
return;
}
for (jsint i = low; i <= high; i++) {
unsigned targetOffset = offset + GetJumpOffset(pc, pc2);
if (targetOffset != offset) {
if (!addJump(cx, targetOffset, &nextOffset, &forwardJump,
stackDepth, stack, defineArray, defineCount)) {
return;
}
}
pc2 += jmplen;
}
break;
}
case JSOP_LOOKUPSWITCH:
case JSOP_LOOKUPSWITCHX: {
jsbytecode *pc2 = pc;
unsigned jmplen = (op == JSOP_LOOKUPSWITCH) ? JUMP_OFFSET_LEN : JUMPX_OFFSET_LEN;
unsigned defaultOffset = offset + GetJumpOffset(pc, pc2);
pc2 += jmplen;
unsigned npairs = GET_UINT16(pc2);
pc2 += UINT16_LEN;
if (!addJump(cx, defaultOffset, &nextOffset, &forwardJump,
stackDepth, stack, defineArray, defineCount)) {
return;
}
while (npairs) {
pc2 += INDEX_LEN;
unsigned targetOffset = offset + GetJumpOffset(pc, pc2);
if (!addJump(cx, targetOffset, &nextOffset, &forwardJump,
stackDepth, stack, defineArray, defineCount)) {
return;
}
pc2 += jmplen;
npairs--;
}
break;
}
case JSOP_TRY: {
/*
* Everything between a try and corresponding catch or finally is conditional.
* Note that there is no problem with code which is skipped by a thrown
* exception but is not caught by a later handler in the same function:
* no more code will execute, and it does not matter what is defined.
*/
JSTryNote *tn = script->trynotes()->vector;
JSTryNote *tnlimit = tn + script->trynotes()->length;
for (; tn < tnlimit; tn++) {
unsigned startOffset = script->main - script->code + tn->start;
if (startOffset == offset + 1) {
unsigned catchOffset = startOffset + tn->length;
/* This will overestimate try block code, for multiple catch/finally. */
if (catchOffset > forwardCatch)
forwardCatch = catchOffset;
if (tn->kind != JSTRY_ITER) {
if (!addJump(cx, catchOffset, &nextOffset, &forwardJump,
stackDepth, stack, defineArray, defineCount)) {
return;
}
getCode(catchOffset).exceptionEntry = true;
}
}
}
break;
}
case JSOP_GETLOCAL:
/*
* Watch for uses of variables not known to be defined, and mark
* them as having possible uses before definitions. Ignore GETLOCAL
* followed by a POP, these are generated for, e.g. 'var x;'
*/
if (pc[JSOP_GETLOCAL_LENGTH] != JSOP_POP) {
uint32 local = GET_SLOTNO(pc);
if (local < nfixed && !localDefined(local, offset))
setLocal(local, LOCAL_USE_BEFORE_DEF);
}
break;
case JSOP_CALLLOCAL:
case JSOP_INCLOCAL:
case JSOP_DECLOCAL:
case JSOP_LOCALINC:
case JSOP_LOCALDEC: {
uint32 local = GET_SLOTNO(pc);
if (local < nfixed && !localDefined(local, offset))
setLocal(local, LOCAL_USE_BEFORE_DEF);
break;
}
case JSOP_SETLOCAL:
case JSOP_FORLOCAL: {
uint32 local = GET_SLOTNO(pc);
/*
* The local variable may already have been marked as unconditionally
* defined at a later point in the script, if that definition was in the
* condition for a loop which then jumped back here. In such cases we
* will not treat the variable as ever being defined in the loop body
* (see setLocal).
*/
if (local < nfixed && locals[local] == LOCAL_CONDITIONALLY_DEFINED) {
if (forwardJump) {
/* Add this local to the variables defined after this bytecode. */
uint32 *newArray = ArenaArray<uint32>(pool, defineCount + 1);
if (!newArray) {
setOOM(cx);
return;
}
if (defineCount)
memcpy(newArray, defineArray, defineCount * sizeof(uint32));
defineArray = newArray;
defineArray[defineCount++] = local;
} else {
/* This local is unconditionally defined by this bytecode. */
setLocal(local, offset);
}
}
break;
}
default:
break;
}
uint32 type = JOF_TYPE(js_CodeSpec[op].format);
/* Check basic jump opcodes, which may or may not have a fallthrough. */
if (type == JOF_JUMP || type == JOF_JUMPX) {
/* Some opcodes behave differently on their branching path. */
unsigned newStackDepth = stackDepth;
types::TypeStack *newStack = stack;
switch (op) {
case JSOP_OR:
case JSOP_AND:
case JSOP_ORX:
case JSOP_ANDX:
/*
* OR/AND instructions push the operation result when branching.
* We accounted for this in GetDefCount, so subtract the pushed value
* for the fallthrough case.
*/
stackDepth--;
#ifdef JS_TYPE_INFERENCE
stack = stack->group()->innerStack;
#endif
break;
case JSOP_CASE:
case JSOP_CASEX:
/* Case instructions do not push the lvalue back when branching. */
newStackDepth--;
#ifdef JS_TYPE_INFERENCE
newStack = newStack->group()->innerStack;
#endif
break;
default:;
}
unsigned targetOffset = offset + GetJumpOffset(pc, pc);
if (!addJump(cx, targetOffset, &nextOffset, &forwardJump,
newStackDepth, newStack, defineArray, defineCount)) {
return;
}
}
/* Handle any fallthrough from this opcode. */
if (!BytecodeNoFallThrough(op)) {
JS_ASSERT(successorOffset < script->length);
Bytecode *&nextcode = codeArray[successorOffset];
bool initial = (nextcode == NULL);
if (initial) {
nextcode = ArenaNew<Bytecode>(pool, this, successorOffset);
if (!nextcode) {
setOOM(cx);
return;
}
}
if (!nextcode->mergeDefines(cx, this, initial, stackDepth, stack,
defineArray, defineCount)) {
return;
}
/* Treat the fallthrough of a branch instruction as a jump target. */
if (type == JOF_JUMP || type == JOF_JUMPX)
nextcode->jumpTarget = true;
else
nextcode->fallthrough = true;
}
}
JS_ASSERT(!failed());
JS_ASSERT(forwardJump == 0 && forwardCatch == 0);
#ifdef JS_TYPE_INFERENCE
/* Generate type constraints for the script. */
AnalyzeState state;
state.init(cx, script);
offset = 0;
while (offset < script->length) {
Bytecode *code = maybeCode(offset);
jsbytecode *pc = script->code + offset;
UntrapOpcode untrap(cx, script, pc);
offset += GetBytecodeLength(pc);
if (code && code->analyzed)
analyzeTypes(cx, code, state);
}
state.destroy(cx);
#endif
}
} /* namespace analyze */
} /* namespace js */