MemSSA.cpp

Go to the documentation of this file.

//===- MemSSA.cpp -- Base class of pointer analyses------------------//
//
//                     SVF: Static Value-Flow Analysis
//
// Copyright (C) <2013-2017>  <Yulei Sui>
//
 
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU Affero General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
 
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
// GNU Affero General Public License for more details.
 
// You should have received a copy of the GNU Affero General Public License
// along with this program.  If not, see <http://www.gnu.org/licenses/>.
//
//===----------------------------------------------------------------------===//
 
/*
 * MemSSA.cpp
 *
 *  Created on: Dec 14, 2013
 *      Author: Yulei Sui
 */
 
#include "Util/Options.h"
#include "MSSA/MemPartition.h"
#include "MSSA/MemSSA.h"
#include "Graphs/SVFGStat.h"
#include "Graphs/CallGraph.h"
#include "SVFIR/SVFVariables.h"
 
using namespace SVF;
using namespace SVFUtil;
 
double MemSSA::timeOfGeneratingMemRegions = 0;  
double MemSSA::timeOfCreateMUCHI  = 0;  
double MemSSA::timeOfInsertingPHI  = 0; 
double MemSSA::timeOfSSARenaming  = 0;  
 
MemSSA::MemSSA(BVDataPTAImpl* p, bool ptrOnlyMSSA)
{
    pta = p;
    assert((pta->getAnalysisTy()!=PointerAnalysis::Default_PTA)
           && "please specify a pointer analysis");
 
    if (Options::MemPar() == MemPartition::Distinct)
        mrGen = new DistinctMRG(pta, ptrOnlyMSSA);
    else if (Options::MemPar() == MemPartition::IntraDisjoint)
        mrGen = new IntraDisjointMRG(pta, ptrOnlyMSSA);
    else if (Options::MemPar() == MemPartition::InterDisjoint)
        mrGen = new InterDisjointMRG(pta, ptrOnlyMSSA);
    else
        assert(false && "unrecognised memory partition strategy");
 
 
    stat = new MemSSAStat(this);
 
    double mrStart = stat->getClk(true);
    mrGen->generateMRs();
    double mrEnd = stat->getClk(true);
    timeOfGeneratingMemRegions = (mrEnd - mrStart)/TIMEINTERVAL;
}
 
SVFIR* MemSSA::getPAG()
{
    return pta->getPAG();
}
 
void MemSSA::buildMemSSA(const FunObjVar& fun)
{
 
    assert(!isExtCall(&fun) && "we do not build memory ssa for external functions");
 
    DBOUT(DMSSA, outs() << "Building Memory SSA for function " << fun.getName()
          << " \n");
 
    usedRegs.clear();
    reg2BBMap.clear();
 
    double muchiStart = stat->getClk(true);
    createMUCHI(fun);
    double muchiEnd = stat->getClk(true);
    timeOfCreateMUCHI += (muchiEnd - muchiStart)/TIMEINTERVAL;
 
    double phiStart = stat->getClk(true);
    insertPHI(fun);
    double phiEnd = stat->getClk(true);
    timeOfInsertingPHI += (phiEnd - phiStart)/TIMEINTERVAL;
 
    double renameStart = stat->getClk(true);
    SSARename(fun);
    double renameEnd = stat->getClk(true);
    timeOfSSARenaming += (renameEnd - renameStart)/TIMEINTERVAL;
 
}
 
void MemSSA::createMUCHI(const FunObjVar& fun)
{
 
 
    DBOUT(DMSSA,
          outs() << "\t creating mu chi for function " << fun.getName()
          << "\n");
    // 1. create mu/chi
    //  insert a set of mus for memory regions at each load
    //  inset a set of chis for memory regions at each store
 
    // 2. find global names (region name before renaming) of each memory region,
    // collect used mrs in usedRegs, and collect its def basic block in reg2BBMap
    // in the form of mu(r) and r = chi (r)
    // a) mu(r):
    //      if(r \not\in varKills) global = global \cup r
    // b) r = chi(r):
    //      if(r \not\in varKills) global = global \cup r
    //      varKills = varKills \cup r
    //      block(r) = block(r) \cup bb_{chi}
 
    BBList reachableBBs = fun.getReachableBBs();
 
    for (BBList::const_iterator iter = reachableBBs.begin(), eiter = reachableBBs.end();
            iter != eiter; ++iter)
    {
        const SVFBasicBlock* bb = *iter;
        varKills.clear();
        for (const auto& inst: bb->getICFGNodeList())
        {
            if(mrGen->hasSVFStmtList(inst))
            {
                SVFStmtList& pagEdgeList = mrGen->getPAGEdgesFromInst(inst);
                for (SVFStmtList::const_iterator bit = pagEdgeList.begin(),
                        ebit = pagEdgeList.end(); bit != ebit; ++bit)
                {
                    const PAGEdge* inst = *bit;
                    if (const LoadStmt* load = SVFUtil::dyn_cast<LoadStmt>(inst))
                        AddLoadMU(bb, load, mrGen->getLoadMRSet(load));
                    else if (const StoreStmt* store = SVFUtil::dyn_cast<StoreStmt>(inst))
                        AddStoreCHI(bb, store, mrGen->getStoreMRSet(store));
                }
            }
            if (isNonInstricCallSite(inst))
            {
                const CallICFGNode* cs = cast<CallICFGNode>(inst);
                if(mrGen->hasRefMRSet(cs))
                    AddCallSiteMU(cs,mrGen->getCallSiteRefMRSet(cs));
 
                if(mrGen->hasModMRSet(cs))
                    AddCallSiteCHI(cs,mrGen->getCallSiteModMRSet(cs));
            }
        }
    }
 
    // create entry chi for this function including all memory regions
    // initialize them with version 0 and 1 r_1 = chi (r_0)
    for (MRSet::iterator iter = usedRegs.begin(), eiter = usedRegs.end();
            iter != eiter; ++iter)
    {
        const MemRegion* mr = *iter;
        // initialize mem region version and stack for renaming phase
        mr2CounterMap[mr] = 0;
        mr2VerStackMap[mr].clear();
        ENTRYCHI* chi = new ENTRYCHI(&fun, mr);
        chi->setOpVer(newSSAName(mr,chi));
        chi->setResVer(newSSAName(mr,chi));
        funToEntryChiSetMap[&fun].insert(chi);
 
        if(fun.hasReturn())
        {
            RETMU* mu = new RETMU(&fun, mr);
            funToReturnMuSetMap[&fun].insert(mu);
        }
 
    }
 
}
 
/*
 * Insert phi node
 */
void MemSSA::insertPHI(const FunObjVar& fun)
{
 
    DBOUT(DMSSA,
          outs() << "\t insert phi for function " << fun.getName() << "\n");
 
    const Map<const SVFBasicBlock*,Set<const SVFBasicBlock*>>& df = fun.getDomFrontierMap();
    // record whether a phi of mr has already been inserted into the bb.
    BBToMRSetMap bb2MRSetMap;
 
    // start inserting phi node
    for (MRSet::iterator iter = usedRegs.begin(), eiter = usedRegs.end();
            iter != eiter; ++iter)
    {
        const MemRegion* mr = *iter;
 
        BBList bbs = reg2BBMap[mr];
        while (!bbs.empty())
        {
            const SVFBasicBlock* bb = bbs.back();
            bbs.pop_back();
            Map<const SVFBasicBlock*,Set<const SVFBasicBlock*>>::const_iterator it = df.find(bb);
            if(it == df.end())
            {
                writeWrnMsg("bb not in the dominance frontier map??");
                continue;
            }
            const Set<const SVFBasicBlock*>& domSet = it->second;
            for (const SVFBasicBlock* pbb : domSet)
            {
                // if we never insert this phi node before
                if (0 == bb2MRSetMap[pbb].count(mr))
                {
                    bb2MRSetMap[pbb].insert(mr);
                    // insert phi node
                    AddMSSAPHI(pbb,mr);
                    // continue to insert phi in its iterative dominate frontiers
                    bbs.push_back(pbb);
                }
            }
        }
    }
 
}
 
void MemSSA::SSARename(const FunObjVar& fun)
{
 
    DBOUT(DMSSA,
          outs() << "\t ssa rename for function " << fun.getName() << "\n");
 
    SSARenameBB(*fun.getEntryBlock());
}
 
void MemSSA::SSARenameBB(const SVFBasicBlock& bb)
{
 
    // record which mem region needs to pop stack
    MRVector memRegs;
 
    // rename phi result op
    // for each r = phi (...)
    //      rewrite r as new name
    if (hasPHISet(&bb))
        RenamePhiRes(getPHISet(&bb),memRegs);
 
 
    // process mu and chi
    // for each mu(r)
    //      rewrite r with top mrver of stack(r)
    // for each r = chi(r')
    //      rewrite r' with top mrver of stack(r)
    //      rewrite r with new name
 
    for (const auto& pNode: bb.getICFGNodeList())
    {
        if(mrGen->hasSVFStmtList(pNode))
        {
            SVFStmtList& pagEdgeList = mrGen->getPAGEdgesFromInst(pNode);
            for(SVFStmtList::const_iterator bit = pagEdgeList.begin(), ebit= pagEdgeList.end();
                    bit!=ebit; ++bit)
            {
                const PAGEdge* inst = *bit;
                if (const LoadStmt* load = SVFUtil::dyn_cast<LoadStmt>(inst))
                    RenameMuSet(getMUSet(load));
 
                else if (const StoreStmt* store = SVFUtil::dyn_cast<StoreStmt>(inst))
                    RenameChiSet(getCHISet(store),memRegs);
 
            }
        }
        if (isNonInstricCallSite(pNode))
        {
            const CallICFGNode* cs = cast<CallICFGNode>(pNode);
            if(mrGen->hasRefMRSet(cs))
                RenameMuSet(getMUSet(cs));
 
            if(mrGen->hasModMRSet(cs))
                RenameChiSet(getCHISet(cs),memRegs);
        }
        else if(isRetInstNode(pNode))
        {
            const FunObjVar* fun = bb.getParent();
            RenameMuSet(getReturnMuSet(fun));
        }
    }
 
 
    // fill phi operands of succ basic blocks
    for (const SVFBasicBlock* succ : bb.getSuccessors())
    {
        u32_t pos = bb.getBBPredecessorPos(succ);
        if (hasPHISet(succ))
            RenamePhiOps(getPHISet(succ),pos,memRegs);
    }
 
    // for succ basic block in dominator tree
    const FunObjVar* fun = bb.getParent();
    const Map<const SVFBasicBlock*,Set<const SVFBasicBlock*>>& dtBBsMap = fun->getDomTreeMap();
    Map<const SVFBasicBlock*,Set<const SVFBasicBlock*>>::const_iterator mapIter = dtBBsMap.find(&bb);
    if (mapIter != dtBBsMap.end())
    {
        const Set<const SVFBasicBlock*>& dtBBs = mapIter->second;
        for (const SVFBasicBlock* dtbb : dtBBs)
        {
            SSARenameBB(*dtbb);
        }
    }
    // for each r = chi(..), and r = phi(..)
    //      pop ver stack(r)
    while (!memRegs.empty())
    {
        const MemRegion* mr = memRegs.back();
        memRegs.pop_back();
        mr2VerStackMap[mr].pop_back();
    }
 
}
 
MRVer* MemSSA::newSSAName(const MemRegion* mr, MSSADEF* def)
{
    assert(0 != mr2CounterMap.count(mr)
           && "did not find initial version in map? ");
    assert(0 != mr2VerStackMap.count(mr)
           && "did not find initial stack in map? ");
 
    MRVERSION version = mr2CounterMap[mr];
    mr2CounterMap[mr] = version + 1;
    auto mrVer = std::make_unique<MRVer>(mr, version, def);
    auto mrVerPtr = mrVer.get();
    mr2VerStackMap[mr].push_back(mrVerPtr);
    usedMRVers.push_back(std::move(mrVer));
    return mrVerPtr;
}
 
void MemSSA::destroy()
{
 
    for (LoadToMUSetMap::iterator iter = load2MuSetMap.begin(), eiter =
                load2MuSetMap.end(); iter != eiter; ++iter)
    {
        for (MUSet::iterator it = iter->second.begin(), eit =
                    iter->second.end(); it != eit; ++it)
        {
            delete *it;
        }
    }
 
    for (StoreToChiSetMap::iterator iter = store2ChiSetMap.begin(), eiter =
                store2ChiSetMap.end(); iter != eiter; ++iter)
    {
        for (CHISet::iterator it = iter->second.begin(), eit =
                    iter->second.end(); it != eit; ++it)
        {
            delete *it;
        }
    }
 
    for (CallSiteToMUSetMap::iterator iter = callsiteToMuSetMap.begin(),
            eiter = callsiteToMuSetMap.end(); iter != eiter; ++iter)
    {
        for (MUSet::iterator it = iter->second.begin(), eit =
                    iter->second.end(); it != eit; ++it)
        {
            delete *it;
        }
    }
 
    for (CallSiteToCHISetMap::iterator iter = callsiteToChiSetMap.begin(),
            eiter = callsiteToChiSetMap.end(); iter != eiter; ++iter)
    {
        for (CHISet::iterator it = iter->second.begin(), eit =
                    iter->second.end(); it != eit; ++it)
        {
            delete *it;
        }
    }
 
    for (FunToEntryChiSetMap::iterator iter = funToEntryChiSetMap.begin(),
            eiter = funToEntryChiSetMap.end(); iter != eiter; ++iter)
    {
        for (CHISet::iterator it = iter->second.begin(), eit =
                    iter->second.end(); it != eit; ++it)
        {
            delete *it;
        }
    }
 
    for (FunToReturnMuSetMap::iterator iter = funToReturnMuSetMap.begin(),
            eiter = funToReturnMuSetMap.end(); iter != eiter; ++iter)
    {
        for (MUSet::iterator it = iter->second.begin(), eit =
                    iter->second.end(); it != eit; ++it)
        {
            delete *it;
        }
    }
 
    for (BBToPhiSetMap::iterator iter = bb2PhiSetMap.begin(), eiter =
                bb2PhiSetMap.end(); iter != eiter; ++iter)
    {
        for (PHISet::iterator it = iter->second.begin(), eit =
                    iter->second.end(); it != eit; ++it)
        {
            delete *it;
        }
    }
 
    delete mrGen;
    mrGen = nullptr;
    delete stat;
    stat = nullptr;
    pta = nullptr;
}
 
void MemSSA::performStat()
{
    if(pta->printStat())
        stat->performStat();
}
 
u32_t MemSSA::getLoadMuNum() const
{
    u32_t num = 0;
    LoadToMUSetMap::const_iterator it = load2MuSetMap.begin();
    LoadToMUSetMap::const_iterator eit = load2MuSetMap.end();
    for (; it != eit; it++)
    {
        const MUSet & muSet = it->second;
        num+= muSet.size();
    }
    return num;
}
 
 
u32_t MemSSA::getStoreChiNum() const
{
    u32_t num = 0;
    StoreToChiSetMap::const_iterator it = store2ChiSetMap.begin();
    StoreToChiSetMap::const_iterator eit = store2ChiSetMap.end();
    for (; it != eit; it++)
    {
        const CHISet& chiSet = it->second;
        num += chiSet.size();
    }
    return num;
}
 
 
u32_t MemSSA::getFunEntryChiNum() const
{
    u32_t num = 0;
    FunToEntryChiSetMap::const_iterator it = funToEntryChiSetMap.begin();
    FunToEntryChiSetMap::const_iterator eit = funToEntryChiSetMap.end();
    for (; it != eit; it++)
    {
        const CHISet& chiSet = it->second;
        num += chiSet.size();
    }
    return num;
}
 
 
u32_t MemSSA::getFunRetMuNum() const
{
    u32_t num = 0;
    FunToReturnMuSetMap::const_iterator it = funToReturnMuSetMap.begin();
    FunToReturnMuSetMap::const_iterator eit = funToReturnMuSetMap.end();
    for (; it != eit; it++)
    {
        const MUSet & muSet = it->second;
        num+= muSet.size();
    }
    return num;
}
 
 
u32_t MemSSA::getCallSiteMuNum() const
{
    u32_t num = 0;
    CallSiteToMUSetMap::const_iterator it = callsiteToMuSetMap.begin();
    CallSiteToMUSetMap::const_iterator eit = callsiteToMuSetMap.end();
    for (; it != eit; it++)
    {
        const MUSet & muSet = it->second;
        num+= muSet.size();
    }
    return num;
}
 
 
u32_t MemSSA::getCallSiteChiNum() const
{
    u32_t num = 0;
    CallSiteToCHISetMap::const_iterator it = callsiteToChiSetMap.begin();
    CallSiteToCHISetMap::const_iterator eit = callsiteToChiSetMap.end();
    for (; it != eit; it++)
    {
        const CHISet & chiSet = it->second;
        num+= chiSet.size();
    }
    return num;
}
 
 
u32_t MemSSA::getBBPhiNum() const
{
    u32_t num = 0;
    BBToPhiSetMap::const_iterator it = bb2PhiSetMap.begin();
    BBToPhiSetMap::const_iterator eit = bb2PhiSetMap.end();
    for (; it != eit; it++)
    {
        const PHISet & phiSet = it->second;
        num+= phiSet.size();
    }
    return num;
}
 
void MemSSA::dumpMSSA(OutStream& Out)
{
 
    CallGraph* svfirCallGraph = PAG::getPAG()->getCallGraph();
    for (const auto& item: *svfirCallGraph)
    {
        const FunObjVar* fun = item.second->getFunction();
        if(Options::MSSAFun()!="" && Options::MSSAFun()!=fun->getName())
            continue;
 
        Out << "==========FUNCTION: " << fun->getName() << "==========\n";
        // dump function entry chi nodes
        if (hasFuncEntryChi(fun))
        {
            CHISet & entry_chis = getFuncEntryChiSet(fun);
            for (CHISet::iterator chi_it = entry_chis.begin(); chi_it != entry_chis.end(); chi_it++)
            {
                (*chi_it)->dump();
            }
        }
 
        for (FunObjVar::const_bb_iterator bit = fun->begin(), ebit = fun->end();
                bit != ebit; ++bit)
        {
            const SVFBasicBlock* bb = bit->second;
            Out << bb->getName() << "\n";
            PHISet& phiSet = getPHISet(bb);
            for(PHISet::iterator pi = phiSet.begin(), epi = phiSet.end(); pi !=epi; ++pi)
            {
                (*pi)->dump();
            }
 
            bool last_is_chi = false;
            for (const auto& inst: bb->getICFGNodeList())
            {
                bool isAppCall = isNonInstricCallSite(inst) && !isExtCall(inst);
                if (isAppCall || isHeapAllocExtCall(inst))
                {
                    const CallICFGNode* cs = cast<CallICFGNode>(inst);
                    if(hasMU(cs))
                    {
                        if (!last_is_chi)
                        {
                            Out << "\n";
                        }
                        for (MUSet::iterator mit = getMUSet(cs).begin(), emit = getMUSet(cs).end();
                                mit != emit; ++mit)
                        {
                            (*mit)->dump();
                        }
                    }
 
                    Out << inst->toString() << "\n";
 
                    if(hasCHI(cs))
                    {
                        for (CHISet::iterator cit = getCHISet(cs).begin(), ecit = getCHISet(cs).end();
                                cit != ecit; ++cit)
                        {
                            (*cit)->dump();
                        }
                        Out << "\n";
                        last_is_chi = true;
                    }
                    else
                        last_is_chi = false;
                }
                else
                {
                    bool dump_preamble = false;
                    SVFStmtList& pagEdgeList = mrGen->getPAGEdgesFromInst(inst);
                    for(SVFStmtList::const_iterator bit = pagEdgeList.begin(), ebit= pagEdgeList.end();
                            bit!=ebit; ++bit)
                    {
                        const PAGEdge* edge = *bit;
                        if (const LoadStmt* load = SVFUtil::dyn_cast<LoadStmt>(edge))
                        {
                            MUSet& muSet = getMUSet(load);
                            for(MUSet::iterator it = muSet.begin(), eit = muSet.end(); it!=eit; ++it)
                            {
                                if (!dump_preamble && !last_is_chi)
                                {
                                    Out << "\n";
                                    dump_preamble = true;
                                }
                                (*it)->dump();
                            }
                        }
                    }
 
                    Out << inst->toString() << "\n";
 
                    bool has_chi = false;
                    for(SVFStmtList::const_iterator bit = pagEdgeList.begin(), ebit= pagEdgeList.end();
                            bit!=ebit; ++bit)
                    {
                        const PAGEdge* edge = *bit;
                        if (const StoreStmt* store = SVFUtil::dyn_cast<StoreStmt>(edge))
                        {
                            CHISet& chiSet = getCHISet(store);
                            for(CHISet::iterator it = chiSet.begin(), eit = chiSet.end(); it!=eit; ++it)
                            {
                                has_chi = true;
                                (*it)->dump();
                            }
                        }
                    }
                    if (has_chi)
                    {
                        Out << "\n";
                        last_is_chi = true;
                    }
                    else
                        last_is_chi = false;
                }
            }
        }
 
        // dump return mu nodes
        if (hasReturnMu(fun))
        {
            MUSet & return_mus = getReturnMuSet(fun);
            for (MUSet::iterator mu_it = return_mus.begin(); mu_it != return_mus.end(); mu_it++)
            {
                (*mu_it)->dump();
            }
        }
    }
}