llvm-6502/lib/CodeGen/SjLjEHPrepare.cpp
Jim Grosbach 8b818d7e98 Move the sjlj exception handling conversions to a back-end pass where they
more properly belong. This allows removing the front-end conditionalized
SJLJ code, and cleans up the generated IR considerably. All of the
infrastructure code (calling _Unwind_SjLj_Register/Unregister, etc) is
added by the SjLjEHPrepare pass.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79250 91177308-0d34-0410-b5e6-96231b3b80d8
2009-08-17 16:41:22 +00:00

528 lines
22 KiB
C++

//===- SjLjEHPass.cpp - Eliminate Invoke & Unwind instructions -----------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This transformation is designed for use by code generators which use SjLj
// based exception handling.
//
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "sjljehprepare"
#include "llvm/Transforms/Scalar.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
#include "llvm/LLVMContext.h"
#include "llvm/Module.h"
#include "llvm/Pass.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Local.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetLowering.h"
#include <set>
using namespace llvm;
STATISTIC(NumInvokes, "Number of invokes replaced");
STATISTIC(NumUnwinds, "Number of unwinds replaced");
STATISTIC(NumSpilled, "Number of registers live across unwind edges");
namespace {
class VISIBILITY_HIDDEN SjLjEHPass : public FunctionPass {
const TargetLowering *TLI;
const Type *FunctionContextTy;
Constant *RegisterFn;
Constant *UnregisterFn;
Constant *ResumeFn;
Constant *BuiltinSetjmpFn;
Constant *FrameAddrFn;
Constant *LSDAAddrFn;
Value *PersonalityFn;
Constant *Selector32Fn;
Constant *Selector64Fn;
Constant *ExceptionFn;
Value *CallSite;
public:
static char ID; // Pass identification, replacement for typeid
explicit SjLjEHPass(const TargetLowering *tli = NULL)
: FunctionPass(&ID), TLI(tli) { }
bool doInitialization(Module &M);
bool runOnFunction(Function &F);
virtual void getAnalysisUsage(AnalysisUsage &AU) const { }
const char *getPassName() const {
return "SJLJ Exception Handling preparation";
}
private:
void markInvokeCallSite(InvokeInst *II, unsigned InvokeNo,
Value *CallSite,
SwitchInst *CatchSwitch);
void splitLiveRangesLiveAcrossInvokes(std::vector<InvokeInst*> &Invokes);
bool insertSjLjEHSupport(Function &F);
};
} // end anonymous namespace
char SjLjEHPass::ID = 0;
// Public Interface To the SjLjEHPass pass.
FunctionPass *llvm::createSjLjEHPass(const TargetLowering *TLI) {
return new SjLjEHPass(TLI);
}
// doInitialization - Make sure that there is a prototype for abort in the
// current module.
bool SjLjEHPass::doInitialization(Module &M) {
// Build the function context structure.
// builtin_setjmp uses a five word jbuf
const Type *VoidPtrTy =
PointerType::getUnqual(Type::getInt8Ty(M.getContext()));
const Type *Int32Ty = Type::getInt32Ty(M.getContext());
FunctionContextTy =
StructType::get(M.getContext(),
VoidPtrTy, // __prev
Int32Ty, // call_site
ArrayType::get(Int32Ty, 4), // __data
VoidPtrTy, // __personality
VoidPtrTy, // __lsda
ArrayType::get(VoidPtrTy, 5), // __jbuf
NULL);
RegisterFn = M.getOrInsertFunction("_Unwind_SjLj_Register",
Type::getVoidTy(M.getContext()),
PointerType::getUnqual(FunctionContextTy),
(Type *)0);
UnregisterFn =
M.getOrInsertFunction("_Unwind_SjLj_Unregister",
Type::getVoidTy(M.getContext()),
PointerType::getUnqual(FunctionContextTy),
(Type *)0);
ResumeFn =
M.getOrInsertFunction("_Unwind_SjLj_Resume",
Type::getVoidTy(M.getContext()),
VoidPtrTy,
(Type *)0);
FrameAddrFn = Intrinsic::getDeclaration(&M, Intrinsic::frameaddress);
BuiltinSetjmpFn = Intrinsic::getDeclaration(&M, Intrinsic::eh_sjlj_setjmp);
LSDAAddrFn = Intrinsic::getDeclaration(&M, Intrinsic::eh_sjlj_lsda);
Selector32Fn = Intrinsic::getDeclaration(&M, Intrinsic::eh_selector_i32);
Selector64Fn = Intrinsic::getDeclaration(&M, Intrinsic::eh_selector_i64);
ExceptionFn = Intrinsic::getDeclaration(&M, Intrinsic::eh_exception);
return true;
}
/// markInvokeCallSite - Insert code to mark the call_site for this invoke
void SjLjEHPass::markInvokeCallSite(InvokeInst *II, unsigned InvokeNo,
Value *CallSite,
SwitchInst *CatchSwitch) {
ConstantInt *CallSiteNoC= ConstantInt::get(Type::getInt32Ty(II->getContext()),
InvokeNo);
// The runtime comes back to the dispatcher with the call_site - 1 in
// the context. Odd, but there it is.
ConstantInt *SwitchValC = ConstantInt::get(Type::getInt32Ty(II->getContext()),
InvokeNo - 1);
// If the unwind edge has phi nodes, split the edge.
if (isa<PHINode>(II->getUnwindDest()->begin())) {
SplitCriticalEdge(II, 1, this);
// If there are any phi nodes left, they must have a single predecessor.
while (PHINode *PN = dyn_cast<PHINode>(II->getUnwindDest()->begin())) {
PN->replaceAllUsesWith(PN->getIncomingValue(0));
PN->eraseFromParent();
}
}
// Insert a store of the invoke num before the invoke and store zero into the
// location afterward.
new StoreInst(CallSiteNoC, CallSite, true, II); // volatile
// Add a switch case to our unwind block.
CatchSwitch->addCase(SwitchValC, II->getUnwindDest());
// We still want this to look like an invoke so we emit the LSDA properly
// FIXME: ??? Or will this cause strangeness with mis-matched IDs like
// when it was in the front end?
}
/// MarkBlocksLiveIn - Insert BB and all of its predescessors into LiveBBs until
/// we reach blocks we've already seen.
static void MarkBlocksLiveIn(BasicBlock *BB, std::set<BasicBlock*> &LiveBBs) {
if (!LiveBBs.insert(BB).second) return; // already been here.
for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI)
MarkBlocksLiveIn(*PI, LiveBBs);
}
// live across unwind edges. Each value that is live across an unwind edge
// we spill into a stack location, guaranteeing that there is nothing live
// across the unwind edge. This process also splits all critical edges
// coming out of invoke's.
void SjLjEHPass::
splitLiveRangesLiveAcrossInvokes(std::vector<InvokeInst*> &Invokes) {
// First step, split all critical edges from invoke instructions.
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
InvokeInst *II = Invokes[i];
SplitCriticalEdge(II, 0, this);
SplitCriticalEdge(II, 1, this);
assert(!isa<PHINode>(II->getNormalDest()) &&
!isa<PHINode>(II->getUnwindDest()) &&
"critical edge splitting left single entry phi nodes?");
}
Function *F = Invokes.back()->getParent()->getParent();
// To avoid having to handle incoming arguments specially, we lower each arg
// to a copy instruction in the entry block. This ensures that the argument
// value itself cannot be live across the entry block.
BasicBlock::iterator AfterAllocaInsertPt = F->begin()->begin();
while (isa<AllocaInst>(AfterAllocaInsertPt) &&
isa<ConstantInt>(cast<AllocaInst>(AfterAllocaInsertPt)->getArraySize()))
++AfterAllocaInsertPt;
for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end();
AI != E; ++AI) {
// This is always a no-op cast because we're casting AI to AI->getType() so
// src and destination types are identical. BitCast is the only possibility.
CastInst *NC = new BitCastInst(
AI, AI->getType(), AI->getName()+".tmp", AfterAllocaInsertPt);
AI->replaceAllUsesWith(NC);
// Normally its is forbidden to replace a CastInst's operand because it
// could cause the opcode to reflect an illegal conversion. However, we're
// replacing it here with the same value it was constructed with to simply
// make NC its user.
NC->setOperand(0, AI);
}
// Finally, scan the code looking for instructions with bad live ranges.
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
// Ignore obvious cases we don't have to handle. In particular, most
// instructions either have no uses or only have a single use inside the
// current block. Ignore them quickly.
Instruction *Inst = II;
if (Inst->use_empty()) continue;
if (Inst->hasOneUse() &&
cast<Instruction>(Inst->use_back())->getParent() == BB &&
!isa<PHINode>(Inst->use_back())) continue;
// If this is an alloca in the entry block, it's not a real register
// value.
if (AllocaInst *AI = dyn_cast<AllocaInst>(Inst))
if (isa<ConstantInt>(AI->getArraySize()) && BB == F->begin())
continue;
// Avoid iterator invalidation by copying users to a temporary vector.
std::vector<Instruction*> Users;
for (Value::use_iterator UI = Inst->use_begin(), E = Inst->use_end();
UI != E; ++UI) {
Instruction *User = cast<Instruction>(*UI);
if (User->getParent() != BB || isa<PHINode>(User))
Users.push_back(User);
}
// Scan all of the uses and see if the live range is live across an unwind
// edge. If we find a use live across an invoke edge, create an alloca
// and spill the value.
std::set<InvokeInst*> InvokesWithStoreInserted;
// Find all of the blocks that this value is live in.
std::set<BasicBlock*> LiveBBs;
LiveBBs.insert(Inst->getParent());
while (!Users.empty()) {
Instruction *U = Users.back();
Users.pop_back();
if (!isa<PHINode>(U)) {
MarkBlocksLiveIn(U->getParent(), LiveBBs);
} else {
// Uses for a PHI node occur in their predecessor block.
PHINode *PN = cast<PHINode>(U);
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingValue(i) == Inst)
MarkBlocksLiveIn(PN->getIncomingBlock(i), LiveBBs);
}
}
// Now that we know all of the blocks that this thing is live in, see if
// it includes any of the unwind locations.
bool NeedsSpill = false;
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
BasicBlock *UnwindBlock = Invokes[i]->getUnwindDest();
if (UnwindBlock != BB && LiveBBs.count(UnwindBlock)) {
NeedsSpill = true;
}
}
// If we decided we need a spill, do it.
if (NeedsSpill) {
++NumSpilled;
DemoteRegToStack(*Inst, true);
}
}
}
bool SjLjEHPass::insertSjLjEHSupport(Function &F) {
std::vector<ReturnInst*> Returns;
std::vector<UnwindInst*> Unwinds;
std::vector<InvokeInst*> Invokes;
// Look through the terminators of the basic blocks to find invokes, returns
// and unwinds
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
if (ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator())) {
// Remember all return instructions in case we insert an invoke into this
// function.
Returns.push_back(RI);
} else if (InvokeInst *II = dyn_cast<InvokeInst>(BB->getTerminator())) {
Invokes.push_back(II);
} else if (UnwindInst *UI = dyn_cast<UnwindInst>(BB->getTerminator())) {
Unwinds.push_back(UI);
}
// If we don't have any invokes or unwinds, there's nothing to do.
if (Unwinds.empty() && Invokes.empty()) return false;
NumInvokes += Invokes.size();
NumUnwinds += Unwinds.size();
if (!Invokes.empty()) {
// We have invokes, so we need to add register/unregister calls to get
// this function onto the global unwind stack.
//
// First thing we need to do is scan the whole function for values that are
// live across unwind edges. Each value that is live across an unwind edge
// we spill into a stack location, guaranteeing that there is nothing live
// across the unwind edge. This process also splits all critical edges
// coming out of invoke's.
splitLiveRangesLiveAcrossInvokes(Invokes);
BasicBlock *EntryBB = F.begin();
// Create an alloca for the incoming jump buffer ptr and the new jump buffer
// that needs to be restored on all exits from the function. This is an
// alloca because the value needs to be added to the global context list.
unsigned Align = 4; // FIXME: Should be a TLI check?
AllocaInst *FunctionContext =
new AllocaInst(FunctionContextTy, 0, Align,
"fcn_context", F.begin()->begin());
Value *Idxs[2];
const Type *Int32Ty = Type::getInt32Ty(F.getContext());
Value *Zero = ConstantInt::get(Int32Ty, 0);
// We need to also keep around a reference to the call_site field
Idxs[0] = Zero;
Idxs[1] = ConstantInt::get(Int32Ty, 1);
CallSite = GetElementPtrInst::Create(FunctionContext, Idxs, Idxs+2,
"call_site",
EntryBB->getTerminator());
// The exception selector comes back in context->data[1]
Idxs[1] = ConstantInt::get(Int32Ty, 2);
Value *FCData = GetElementPtrInst::Create(FunctionContext, Idxs, Idxs+2,
"fc_data",
EntryBB->getTerminator());
Idxs[1] = ConstantInt::get(Int32Ty, 1);
Value *SelectorAddr = GetElementPtrInst::Create(FCData, Idxs, Idxs+2,
"exc_selector_gep",
EntryBB->getTerminator());
// The exception value comes back in context->data[0]
Idxs[1] = Zero;
Value *ExceptionAddr = GetElementPtrInst::Create(FCData, Idxs, Idxs+2,
"exception_gep",
EntryBB->getTerminator());
// Find the eh.selector.* and eh.exception calls. We'll use the first
// ex.selector to determine the right personality function to use. For
// SJLJ, we always use the same personality for the whole function,
// not on a per-selector basis.
// FIXME: That's a bit ugly. Better way?
std::vector<CallInst*> EH_Selectors;
std::vector<CallInst*> EH_Exceptions;
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
// for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
// BasicBlock *Pad = Invokes[0]->getUnwindDest();
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (CI->getCalledFunction() == Selector32Fn ||
CI->getCalledFunction() == Selector64Fn) {
if (!PersonalityFn) PersonalityFn = CI->getOperand(2);
EH_Selectors.push_back(CI);
} else if (CI->getCalledFunction() == ExceptionFn) {
EH_Exceptions.push_back(CI);
}
}
}
}
// The result of the eh.selector call will be replaced with a
// a reference to the selector value returned in the function
// context. We leave the selector itself so the EH analysis later
// can use it.
for (int i = 0, e = EH_Selectors.size(); i < e; ++i) {
CallInst *I = EH_Selectors[i];
Value *SelectorVal = new LoadInst(SelectorAddr, "select_val", true, I);
I->replaceAllUsesWith(SelectorVal);
}
// eh.exception calls are replaced with references to the proper
// location in the context. Unlike eh.selector, the eh.exception
// calls are removed entirely.
for (int i = 0, e = EH_Exceptions.size(); i < e; ++i) {
CallInst *I = EH_Exceptions[i];
// Possible for there to be duplicates, so check to make sure
// the instruction hasn't already been removed.
if (!I->getParent()) continue;
Value *Val = new LoadInst(ExceptionAddr, "exception", true, I);
Val = CastInst::Create(Instruction::IntToPtr, Val,
PointerType::getUnqual(Type::getInt8Ty(F.getContext())),
"", I);
I->replaceAllUsesWith(Val);
I->eraseFromParent();
}
// The entry block changes to have the eh.sjlj.setjmp, with a conditional
// branch to a dispatch block for non-zero returns. If we return normally,
// we're not handling an exception and just register the function context
// and continue.
// Create the dispatch block. The dispatch block is basically a big switch
// statement that goes to all of the invoke landing pads.
BasicBlock *DispatchBlock =
BasicBlock::Create(F.getContext(), "eh.sjlj.setjmp.catch", &F);
// Insert a load in the Catch block, and a switch on its value. By default,
// we go to a block that just does an unwind (which is the correct action
// for a standard call).
BasicBlock *UnwindBlock = BasicBlock::Create(F.getContext(), "unwindbb", &F);
Unwinds.push_back(new UnwindInst(F.getContext(), UnwindBlock));
Value *DispatchLoad = new LoadInst(CallSite, "invoke.num", true,
DispatchBlock);
SwitchInst *DispatchSwitch =
SwitchInst::Create(DispatchLoad, UnwindBlock, Invokes.size(), DispatchBlock);
// Split the entry block to insert the conditional branch for the setjmp.
BasicBlock *ContBlock = EntryBB->splitBasicBlock(EntryBB->getTerminator(),
"eh.sjlj.setjmp.cont");
// Populate the Function Context
// 1. LSDA address
// 2. Personality function address
// 3. jmpbuf (save FP and call eh.sjlj.setjmp)
// LSDA address
Idxs[0] = Zero;
Idxs[1] = ConstantInt::get(Int32Ty, 4);
Value *LSDAFieldPtr =
GetElementPtrInst::Create(FunctionContext, Idxs, Idxs+2,
"lsda_gep",
EntryBB->getTerminator());
Value *LSDA = CallInst::Create(LSDAAddrFn, "lsda_addr",
EntryBB->getTerminator());
new StoreInst(LSDA, LSDAFieldPtr, true, EntryBB->getTerminator());
Idxs[1] = ConstantInt::get(Int32Ty, 3);
Value *PersonalityFieldPtr =
GetElementPtrInst::Create(FunctionContext, Idxs, Idxs+2,
"lsda_gep",
EntryBB->getTerminator());
new StoreInst(PersonalityFn, PersonalityFieldPtr, true,
EntryBB->getTerminator());
// Save the frame pointer.
Idxs[1] = ConstantInt::get(Int32Ty, 5);
Value *FieldPtr
= GetElementPtrInst::Create(FunctionContext, Idxs, Idxs+2,
"jbuf_gep",
EntryBB->getTerminator());
Idxs[1] = ConstantInt::get(Int32Ty, 0);
Value *ElemPtr =
GetElementPtrInst::Create(FieldPtr, Idxs, Idxs+2, "jbuf_fp_gep",
EntryBB->getTerminator());
Value *Val = CallInst::Create(FrameAddrFn,
ConstantInt::get(Int32Ty, 0),
"fp",
EntryBB->getTerminator());
new StoreInst(Val, ElemPtr, true, EntryBB->getTerminator());
// Call the setjmp instrinsic. It fills in the rest of the jmpbuf
Value *SetjmpArg =
CastInst::Create(Instruction::BitCast, FieldPtr,
Type::getInt8Ty(F.getContext())->getPointerTo(), "",
EntryBB->getTerminator());
Value *DispatchVal = CallInst::Create(BuiltinSetjmpFn, SetjmpArg,
"dispatch",
EntryBB->getTerminator());
// check the return value of the setjmp. non-zero goes to dispatcher
Value *IsNormal = new ICmpInst(EntryBB->getTerminator(),
ICmpInst::ICMP_EQ, DispatchVal, Zero,
"notunwind");
// Nuke the uncond branch.
EntryBB->getTerminator()->eraseFromParent();
// Put in a new condbranch in its place.
BranchInst::Create(ContBlock, DispatchBlock, IsNormal, EntryBB);
// Register the function context and make sure it's known to not throw
CallInst *Register =
CallInst::Create(RegisterFn, FunctionContext, "",
ContBlock->getTerminator());
Register->setDoesNotThrow();
// At this point, we are all set up, update the invoke instructions
// to mark their call_site values, and fill in the dispatch switch
// accordingly.
for (unsigned i = 0, e = Invokes.size(); i != e; ++i)
markInvokeCallSite(Invokes[i], i+1, CallSite, DispatchSwitch);
// The front end has likely added calls to _Unwind_Resume. We need
// to find those calls and mark the call_site as -1 immediately prior.
// resume is a noreturn function, so any block that has a call to it
// should end in an 'unreachable' instruction with the call immediately
// prior. That's how we'll search.
// ??? There's got to be a better way. this is fugly.
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
if ((dyn_cast<UnreachableInst>(BB->getTerminator()))) {
BasicBlock::iterator I = BB->getTerminator();
// Check the previous instruction and see if it's a resume call
if (I == BB->begin()) continue;
if (CallInst *CI = dyn_cast<CallInst>(--I)) {
if (CI->getCalledFunction() == ResumeFn) {
Value *NegativeOne = ConstantInt::get(Int32Ty, -1);
new StoreInst(NegativeOne, CallSite, true, I); // volatile
}
}
}
// Replace all unwinds with a branch to the unwind handler.
// ??? Should this ever happen with sjlj exceptions?
for (unsigned i = 0, e = Unwinds.size(); i != e; ++i) {
BranchInst::Create(UnwindBlock, Unwinds[i]);
Unwinds[i]->eraseFromParent();
}
// Finally, for any returns from this function, if this function contains an
// invoke, add a call to unregister the function context.
for (unsigned i = 0, e = Returns.size(); i != e; ++i)
CallInst::Create(UnregisterFn, FunctionContext, "", Returns[i]);
}
return true;
}
bool SjLjEHPass::runOnFunction(Function &F) {
bool Res = insertSjLjEHSupport(F);
return Res;
}