Added several calls to API (waitForStatusChange, BPatch_variableExpr

[dyninst.git] / dyninstAPI / src / process.C

/*
 * Copyright (c) 1996 Barton P. Miller
 * 
 * We provide the Paradyn Parallel Performance Tools (below
 * described as Paradyn") on an AS IS basis, and do not warrant its
 * validity or performance.  We reserve the right to update, modify,
 * or discontinue this software at any time.  We shall have no
 * obligation to supply such updates or modifications or any other
 * form of support to you.
 * 
 * This license is for research uses.  For such uses, there is no
 * charge. We define "research use" to mean you may freely use it
 * inside your organization for whatever purposes you see fit. But you
 * may not re-distribute Paradyn or parts of Paradyn, in any form
 * source or binary (including derivatives), electronic or otherwise,
 * to any other organization or entity without our permission.
 * 
 * (for other uses, please contact us at paradyn@cs.wisc.edu)
 * 
 * All warranties, including without limitation, any warranty of
 * merchantability or fitness for a particular purpose, are hereby
 * excluded.
 * 
 * By your use of Paradyn, you understand and agree that we (or any
 * other person or entity with proprietary rights in Paradyn) are
 * under no obligation to provide either maintenance services,
 * update services, notices of latent defects, or correction of
 * defects for Paradyn.
 * 
 * Even if advised of the possibility of such damages, under no
 * circumstances shall we (or any other person or entity with
 * proprietary rights in the software licensed hereunder) be liable
 * to you or any third party for direct, indirect, or consequential
 * damages of any character regardless of type of action, including,
 * without limitation, loss of profits, loss of use, loss of good
 * will, or computer failure or malfunction.  You agree to indemnify
 * us (and any other person or entity with proprietary rights in the
 * software licensed hereunder) for any and all liability it may
 * incur to third parties resulting from your use of Paradyn.
 */

extern "C" {
#ifdef PARADYND_PVM
int pvmputenv (const char *);
int pvmendtask();
#endif
}

#include "util/h/headers.h"
#include "dyninstAPI/src/symtab.h"
#include "dyninstAPI/src/pdThread.h"
#include "dyninstAPI/src/process.h"
#include "dyninstAPI/src/util.h"
#include "dyninstAPI/src/inst.h"
#include "dyninstAPI/src/instP.h"
#include "dyninstAPI/src/dyninstP.h"
#include "dyninstAPI/src/os.h"
#include "paradynd/src/showerror.h"
#include "dyninstAPI/src/dynamiclinking.h"
// #include "paradynd/src/mdld.h"

#ifdef BPATCH_LIBRARY
#include "dyninstAPI/h/BPatch.h"
#else
#include "rtinst/h/rtinst.h"
#include "rtinst/h/trace.h"
#include "paradynd/src/perfStream.h"
#include "paradynd/src/costmetrics.h"
#include "paradynd/src/mdld.h"
#include "paradynd/src/main.h"
#endif

#if defined(SHM_SAMPLING) && defined(MT_THREAD)
extern void generateMTpreamble(char *insn, unsigned &base, process *proc);
#endif

#include "util/h/debugOstream.h"

#ifdef ATTACH_DETACH_DEBUG
debug_ostream attach_cerr(cerr, true);
#else
debug_ostream attach_cerr(cerr, false);
#endif

#ifdef INFERIOR_RPC_DEBUG
debug_ostream inferiorrpc_cerr(cerr, true);
#else
debug_ostream inferiorrpc_cerr(cerr, false);
#endif

#ifdef SHM_SAMPLING_DEBUG
debug_ostream shmsample_cerr(cerr, true);
#else
debug_ostream shmsample_cerr(cerr, false);
#endif

#ifdef FORK_EXEC_DEBUG
debug_ostream forkexec_cerr(cerr, true);
#else
debug_ostream forkexec_cerr(cerr, false);
#endif

#ifdef METRIC_DEBUG
debug_ostream metric_cerr(cerr, true);
#else
debug_ostream metric_cerr(cerr, false);
#endif

#ifdef SIGNAL_DEBUG
debug_ostream signal_cerr(cerr, true);
#else
debug_ostream signal_cerr(cerr, false);
#endif

#ifdef SHAREDOBJ_DEBUG
debug_ostream sharedobj_cerr(cerr, true);
#else
debug_ostream sharedobj_cerr(cerr, false);
#endif

#define FREE_WATERMARK (hp->totalFreeMemAvailable/2)
#define SIZE_WATERMARK 100
static const timeStamp MAX_WAITING_TIME=10.0;
static const timeStamp MAX_DELETING_TIME=2.0;
unsigned inferiorMemAvailable=0;

unsigned activeProcesses; // number of active processes
vector<process*> processVec;
string process::programName;
string process::pdFlavor;
vector<string> process::arg_list;

process *findProcess(int pid) { // make a public static member fn of class process
  unsigned size=processVec.size();
  for (unsigned u=0; u<size; u++)
    if (processVec[u] && processVec[u]->getPid() == pid)
      return processVec[u];
  return NULL;
}

#ifdef SHM_SAMPLING
static unsigned numIntCounters=10000; // rather arbitrary; can we do better?
static unsigned numWallTimers =10000; // rather arbitrary; can we do better?
static unsigned numProcTimers =10000; // rather arbitrary; can we do better?
#endif

bool waitingPeriodIsOver()
{
static timeStamp previous=0;
timeStamp current;
bool waiting=false;

  if (!previous) {
    previous=getCurrentTime(false);
    waiting=true;
  }
  else {
    current=getCurrentTime(false);
    if ( (current-previous) > MAX_WAITING_TIME ) {
      previous=getCurrentTime(false);
      waiting=true;
    }
  }
  return(waiting);
}

Frame::Frame(process *proc) {

    frame_ = pc_ = 0;
    proc->getActiveFrame(&frame_, &pc_);

    uppermostFrame = true;
}

Frame Frame::getPreviousStackFrameInfo(process *proc) const {
   if (frame_ == 0){
      // no prev frame exists; must return frame with 0 pc value otherwise
      // will never break out of loop in walkStack()
      Frame fake_frame(0,0,false);
      return fake_frame; 
   }

   int fp = frame_;
   int rtn = pc_;

   Frame result(0, 0, false);
   if (proc->readDataFromFrame(frame_, &fp, &rtn, uppermostFrame)) {
      result.frame_ = fp;
      result.pc_    = rtn;
   }

   return result;
}

#if !defined(rs6000_ibm_aix4_1)
vector<int> process::getTOCoffsetInfo() const
{
  vector<int> tmp;
  assert(0);
  return tmp; // this is to make the nt compiler happy! - naim
}
#endif

#if !defined(i386_unknown_nt4_0)
// Windows NT has its own version of the walkStack function in pdwinnt.C
// Note: it may not always be possible to do a correct stack walk.
// If it can't do a complete walk, the function should return an empty
// vector, which means that there was an error, and we can't walk the stack.
vector<Address> process::walkStack(bool noPause)
{
  vector<Address> pcs;
  bool needToCont = noPause ? false : (status() == running);

  if (!noPause && !pause()) {
     // pause failed...give up
     cerr << "walkStack: pause failed" << endl;
     return pcs;
  }

  Address sig_addr = 0;
  u_int sig_size = 0;
  if(signal_handler){
      const image *sig_image = (signal_handler->file())->exec();
      if(getBaseAddress(sig_image, sig_addr)){
          sig_addr += signal_handler->getAddress(this);
      } else {
          sig_addr = signal_handler->getAddress(this);
      }
      sig_size = signal_handler->size();
      // printf("signal_handler = %s size = %d addr = 0x%x\n",
      //     (signal_handler->prettyName()).string_of(),sig_size,sig_addr);
  }

  if (pause()) {
    Frame   currentFrame(this);
    Address fpOld = 0;
    while (!currentFrame.isLastFrame()) {
      Address fpNew = currentFrame.getFramePtr();
      if (fpOld >= fpNew) {                             //Not moving up stack
        if (!noPause && needToCont && !continueProc())
          cerr << "walkStack: continueProc failed" << endl;
        vector<Address> ev;                             //Empty vector
        return ev;
      }
      fpOld = fpNew;

      Address next_pc = currentFrame.getPC();
      // printf("currentFrame pc = %d\n",next_pc);
      pcs += next_pc;
      // is this pc in the signal_handler function?
      if(signal_handler && (next_pc >= sig_addr)
          && (next_pc < (sig_addr+sig_size))){
          // check to see if a leaf function was executing when the signal
          // handler was called.  If so, then an extra frame should be added
          // for the leaf function...the call to getPreviousStackFrameInfo
          // will get the function that called the leaf function
          Address leaf_pc = 0;
          if(this->needToAddALeafFrame(currentFrame,leaf_pc)){
              pcs += leaf_pc;
          }
      }
      currentFrame = currentFrame.getPreviousStackFrameInfo(this); 
    }
    pcs += currentFrame.getPC();
  }
  if (!noPause && needToCont) {
     if (!continueProc()){
        cerr << "walkStack: continueProc failed" << endl;
     }
  }  
  return(pcs);
}
#endif

        // disItem is generally unchanged and was previously declared const,
        // however, the heap management occasionally deletes such items
        // so it's now no longer const'd
bool isFreeOK(process *proc, disabledItem &disItem, vector<Address> &pcs) {
  const unsigned disItemPointer = disItem.getPointer();
  const inferiorHeapType disItemHeap = disItem.getHeapType();

#if defined(hppa1_1_hp_hpux)
  if (proc->freeNotOK) return(false);
#endif

  heapItem *ptr=NULL;
  if (!proc->heaps[disItemHeap].heapActive.find(disItemPointer, ptr)) {
    sprintf(errorLine,"Warning: attempt to free not defined heap entry "
                      "%x (pid=%d, heapActive.size()=%d)\n", 
        disItemPointer, proc->getPid(), 
        proc->heaps[disItemHeap].heapActive.size());
    logLine(errorLine);
    //showErrorCallback(67, (const char *)errorLine);
    return(false);
  }
  assert(ptr);

#ifdef FREEDEBUG1
  sprintf(errorLine, "isFreeOK  called on 0x%x\n", ptr->addr);
  logLine(errorLine);
#endif

  // the following can't be const; we're sometimes gonna change it
  vector<unsigVecType> &disItemPoints = disItem.getPointsToCheck(); 

  const unsigned disItemNumPoints = disItemPoints.size();

  for (unsigned int j=0;j<disItemNumPoints;j++) {
    for (unsigned int k=0;k<disItemPoints[j].size();k++) {
      unsigned pointer = disItemPoints[j][k];
#ifdef FREEDEBUG_ON
      if (disItemHeap == dataHeap)
        sprintf(errorLine, "checking DATA pointer 0x%x\n", pointer);
      else
        sprintf(errorLine, "checking TEXT pointer 0x%x\n", pointer);
      logLine(errorLine);
#endif

      const dictionary_hash<unsigned, heapItem*> &heapActivePart =
              proc->splitHeaps ? proc->heaps[textHeap].heapActive :
                                 proc->heaps[dataHeap].heapActive;

      heapItem *np=NULL;
      if (!heapActivePart.find(pointer, np)) { // fills in "np" if found
#ifdef FREEDEBUG1
            sprintf(errorLine, "something freed addr 0x%x from us\n", pointer);
            logLine(errorLine);
#endif
        
        // 
        // This point was deleted already and we don't need it anymore in
        // pointsToCheck
        // 
        const int size=disItemPoints[j].size();
        disItemPoints[j][k] = disItemPoints[j][size-1];
        disItemPoints[j].resize(size-1);

        // need to make sure we check the next item too 
        k--;
      } else {
        // Condition 1
        assert(np);
        if ( (ptr->addr >= np->addr) && 
             (ptr->addr <= (np->addr + np->length)) )
        {

#ifdef FREEDEBUG_ON
          sprintf(errorLine,"*** TEST *** IN isFreeOK: (1) we found 0x%x in our inst. range!\n",ptr->addr);
          logLine(errorLine);
#endif

          return(false);     
        }

        for (unsigned int l=0;l<pcs.size();l++) {
          // Condition 2
          if ((pcs[l] >= ptr->addr) && 
              (pcs[l] <= (ptr->addr + ptr->length))) 
          {

#ifdef FREEDEBUG_ON
    sprintf(errorLine,"*** TEST *** IN isFreeOK: (2) we found 0x%x in our inst. range!\n", ptr->addr);
    logLine(errorLine);
#endif

            return(false);
          }
          // Condition 3
          if ( ((pcs[l] >= np->addr) && (pcs[l] <= (np->addr + np->length))) )
          {

#ifdef FREEDEBUG_ON
    sprintf(errorLine,"*** TEST *** IN isFreeOK: (3) we found PC in our inst. range!\n");
    logLine(errorLine);
#endif

            return(false);     
          }
        }
      }
    }
  }
  return(true);
}

//
// This procedure will try to compact the framented memory available in 
// heapFree. This is an emergency procedure that will be called if we
// are running out of memory to insert instrumentation - naim
//
void inferiorFreeCompact(inferiorHeap *hp)
{
  unsigned size;
  heapItem *np;
  size = hp->heapFree.size();
  unsigned j,i=0;

#ifdef FREEDEBUG
  logLine("***** Trying to compact freed memory...\n");
#endif

  while (i < size) {
    np = hp->heapFree[i];
#ifdef FREEDEBUG1
    sprintf(errorLine,"***** Checking address=%d\n",ALIGN_TO_WORDSIZE(np->addr+np->length));
    logLine(errorLine);
#endif
    for (j=0; j < size; j++) {
      if (i != j) {
        if ( (np->addr+np->length)==(hp->heapFree[j])->addr )
        {
          np->length += (hp->heapFree[j])->length;
          hp->heapFree[j] = hp->heapFree[size-1];
          hp->heapFree.resize(size-1);
          size = hp->heapFree.size();
#ifdef FREEDEBUG1
          sprintf(errorLine,"***** Compacting free memory (%d bytes, i=%d, j=%d, heapFree.size=%d)\n",np->length,i,j,size);
          logLine(errorLine);
#endif
          break;
        }
      }
    }
    if (j == size) i++;
  }

#ifdef FREEDEBUG
  for (i=0;i<hp->disabledList.size();i++) {
    for (j=i+1;j<hp->disabledList.size();j++) {
      if ( (hp->disabledList[i]).getPointer() == 
           (hp->disabledList[j]).getPointer() ) {
        sprintf(errorLine,"***** ERROR: address 0x%x appears more than once\n",(hp->disabledList[j]).getPointer());
        logLine(errorLine);
      }
    }
  }
#endif

#ifdef FREEDEBUG
  logLine("***** Compact memory procedure END...\n");
#endif
}

void inferiorFreeDefered(process *proc, inferiorHeap *hp, bool runOutOfMem)
{
  unsigned int i=0;
  vector<Address> pcs;
  vector<disabledItem> *disList;
  timeStamp initTime, maxDelTime;

  pcs = proc->walkStack();

#if defined(i386_unknown_nt4_0)
  // It may not always be possible to get a correct stack walk.
  // If walkStack fails, it returns an empty vector. In this
  // case, we assume that it is not safe to delete anything
  if (pcs.size() == 0)
    return;
#endif

  // this is a while loop since we don't update i if an item is deleted.
  disList = &hp->disabledList;
  if (runOutOfMem) {
    maxDelTime = MAX_DELETING_TIME*2.0;
    sprintf(errorLine,"Emergency attempt to free memory (pid=%d). Please, wait...\n",proc->getPid());
    logLine(errorLine);
#ifdef FREEDEBUG1
    sprintf(errorLine,"***** disList.size() = %d\n",disList->size());
    logLine(errorLine);
#endif
  }
  else
    maxDelTime = MAX_DELETING_TIME;
  initTime=getCurrentTime(false);
  while ( (i < disList->size()) && 
          ((getCurrentTime(false)-initTime) < maxDelTime) )
  {
    disabledItem &item = (*disList)[i];
    if (isFreeOK(proc,item,pcs)) {
      heapItem *np=NULL;
      unsigned pointer = item.getPointer();
      if (!hp->heapActive.find(pointer,np)) {
        showErrorCallback(96,"");
        return;
      }
      assert(np);

      if (np->status != HEAPallocated) {
        sprintf(errorLine,"Attempt to free already freed heap entry %x\n", pointer);
        logLine(errorLine);
        showErrorCallback(67, (const char *)errorLine); 
        return;
      }
      np->status = HEAPfree;      

      // remove from active list.
      hp->heapActive.undef(pointer);

#ifdef FREEDEBUG_ON
   sprintf(errorLine,"inferiorFreeDefered: deleting 0x%x from heap\n",pointer);
   logLine(errorLine);
#endif

      hp->heapFree += np;
      hp->freed += np->length;

      // updating disabledList
      hp->disabledList[i]=hp->disabledList[hp->disabledList.size()-1];
      hp->disabledList.resize(hp->disabledList.size()-1);
      hp->disabledListTotalMem -= np->length;
      hp->totalFreeMemAvailable += np->length;
      inferiorMemAvailable = hp->totalFreeMemAvailable;
    } else {
      i++;
    }
  }
}

void process::initInferiorHeap(bool initTextHeap)
{
    assert(this->symbols);

    inferiorHeap *hp;
    if (initTextHeap) {
        hp = &this->heaps[textHeap];
    } else {
        hp = &this->heaps[dataHeap];
    }

    heapItem *np = new heapItem;
    if (initTextHeap) {
        bool err;
        np->addr = findInternalAddress("DYNINSTtext", true,err);
        if (err)
          abort();
    } else {
        bool err;
        np->addr = findInternalAddress(INFERIOR_HEAP_BASE, true, err);
        if (err)
          abort();
    }
    np->length = SYN_INST_BUF_SIZE;
    np->status = HEAPfree;

    // make the heap double-word aligned
    Address base = np->addr & ~0x1f;
    Address diff = np->addr - base;
    if (diff) {
      np->addr = base + 32;
      np->length -= (32 - diff);
    }

#ifdef BPATCH_SET_MUTATIONS_ACTIVE
    hp->base = np->addr;
    hp->size = np->length;
#endif /* BPATCH_SET_MUTATIONS_ACTIVE */

    hp->totalFreeMemAvailable = np->length;
    inferiorMemAvailable = hp->totalFreeMemAvailable;

    // need to clear everything here, since this function may be called to 
    // re-init a heap
    hp->heapActive.clear();
    hp->heapFree.resize(0);
    hp->heapFree += np;
    hp->disabledList.resize(0);
    hp->disabledListTotalMem = 0;
    hp->freed = 0;
}

// create a new inferior heap that is a copy of src. This is used when a process
// we are tracing forks.
inferiorHeap::inferiorHeap(const inferiorHeap &src):
    heapActive(addrHash16)
{
    for (unsigned u1 = 0; u1 < src.heapFree.size(); u1++) {
      heapFree += new heapItem(src.heapFree[u1]);
    }

    vector<heapItem *> items = src.heapActive.values();
    for (unsigned u2 = 0; u2 < items.size(); u2++) {
      heapActive[items[u2]->addr] = new heapItem(items[u2]);
    }
    
    for (unsigned u3 = 0; u3 < src.disabledList.size(); u3++) {
      disabledList += src.disabledList[u3];
    }
    disabledListTotalMem = src.disabledListTotalMem;
    totalFreeMemAvailable = src.totalFreeMemAvailable;
    inferiorMemAvailable = totalFreeMemAvailable;
    freed = 0;
}

#ifdef FREEDEBUG
void printHeapFree(process *proc, inferiorHeap *hp, int size)
{
  for (unsigned i=0; i < hp->heapFree.size(); i++) {
    sprintf(errorLine,"***** (pid=%d) i=%d, addr=%d, length=%d, heapFree.size()=%d, size=%d\n",proc->getPid(),i,(hp->heapFree[i])->addr,(hp->heapFree[i])->length,hp->heapFree.size(),size); 
    logLine(errorLine);
  }
}
#endif

//
// This function will return the index corresponding to the next position
// available in heapFree.
//
bool findFreeIndex(inferiorHeap *hp, int size, unsigned *index)
{
  bool foundFree=false;
  unsigned best=0;
  int length=0;
  for (unsigned i=0; i < hp->heapFree.size(); i++) {
    length = (hp->heapFree[i])->length;
    if (length >= size) {
      foundFree = true;
      best=i;
      break;
    }
  }
  if (index) 
    *index = best;
  else
    foundFree=false;
  return(foundFree);
}  

unsigned inferiorMalloc(process *proc, int size, inferiorHeapType type)
{
    inferiorHeap *hp;
    heapItem *np=NULL, *newEntry = NULL;

    assert(size > 0);
    /* round to next cache line size */
    /* 32 bytes on a SPARC */
    size = (size + 0x1f) & ~0x1f; 

    if ((type == textHeap) && (proc->splitHeaps)) {
        hp = &proc->heaps[textHeap];
    } else {
        hp = &proc->heaps[dataHeap];
    }

    bool secondChance=false;
    unsigned foundIndex;
    if (!findFreeIndex(hp,size,&foundIndex)) {

#ifdef FREEDEBUG
      sprintf(errorLine,"==> TEST <== In inferiorMalloc: heap overflow, calling inferiorFreeDefered for a second chance...\n");
      logLine(errorLine);
#endif

      inferiorFreeDefered(proc, hp, true);
      inferiorFreeCompact(hp);
      secondChance=true;
    }

    if (secondChance && !findFreeIndex(hp,size,&foundIndex)) {

#ifdef FREEDEBUG
      printHeapFree(proc,hp,size);
#endif

      sprintf(errorLine, "***** Inferior heap overflow: %d bytes freed, %d bytes requested\n", hp->freed, size);
      logLine(errorLine);
      showErrorCallback(66, (const char *) errorLine);
      P__exit(-1);
      //return(0);
    }

    //
    // We must have found a free position in heapFree
    //

    np = hp->heapFree[foundIndex];
    assert(np);
    if (np->length != size) {
        // divide it up.
        newEntry = new heapItem;
        newEntry->length = np->length - size;
        newEntry->addr = np->addr + size;
        hp->totalFreeMemAvailable -= size;
        inferiorMemAvailable = hp->totalFreeMemAvailable;
        // overwrite the old entry
        hp->heapFree[foundIndex] = newEntry;

        /* now split curr */
        np->length = size;
    } else {
      unsigned i = hp->heapFree.size();
      // copy the last element over this element
      hp->heapFree[foundIndex] = hp->heapFree[i-1];
      hp->heapFree.resize(i-1);
    }

    // mark it used
    np->status = HEAPallocated;

    // onto in use list
    hp->heapActive[np->addr] = np;

    // make sure its a valid pointer.
    assert(np->addr);

    return(np->addr);
}

void inferiorFree(process *proc, unsigned pointer, inferiorHeapType type,
                  const vector<unsigVecType> &pointsToCheck)
{
    inferiorHeapType which = (type == textHeap && proc->splitHeaps) ? textHeap : dataHeap;
    inferiorHeap *hp = &proc->heaps[which];
    heapItem *np=NULL;

    if (!hp->heapActive.find(pointer, np)) {
      showErrorCallback(96,"");
      return;
    }
    assert(np);

    disabledItem newItem(pointer, which, pointsToCheck);

#ifdef FREEDEBUG
    for (unsigned i=0;i<hp->disabledList.size();i++) {
      if (hp->disabledList[i].getPointer() == pointer) {
        sprintf(errorLine,"***** ERROR, pointer 0x%x already defined in disabledList\n",pointer);
        logLine(errorLine);
      }
    }
#endif

    hp->disabledList += newItem;
    hp->disabledListTotalMem += np->length;

#ifdef FREEDEBUG1
    sprintf(errorLine,"==> TEST <== In inferiorFree: disabledList has %d items, %d bytes, and FREE_WATERMARK is %d\n",hp->disabledList.size(),hp->disabledListTotalMem,FREE_WATERMARK);
    logLine(errorLine);
    for (unsigned i=0; i < hp->disabledList.size(); i++) {
        sprintf(errorLine, "   %x is on list\n", hp->disabledList[i].pointer);
        logLine(errorLine);
    }
#endif

    //
    // If the size of the disabled instrumentation list is greater than
    // half of the size of the heapFree list, then we attempt to free
    // all the defered requests. In my opinion, this seems to be a good
    // time to proceed with the defered delete since this is an expensive
    // procedure and should not be executed often - naim 03/19/96
    //
    if (((hp->disabledListTotalMem > FREE_WATERMARK) ||
         (hp->disabledList.size() > SIZE_WATERMARK)) && waitingPeriodIsOver()) 
    {

#ifdef FREEDEBUG
timeStamp t1,t2;
static timeStamp t3=0.0;
static timeStamp totalTime=0.0;
static timeStamp worst=0.0;
static int counter=0;
t1=getCurrentTime(false);
#endif

      inferiorFreeDefered(proc, hp, false);

#ifdef FREEDEBUG
t2=getCurrentTime(false);
if (!t3) t3=t1;
counter++;
totalTime += t2-t1;
if ((t2-t1) > worst) worst=t2-t1;
if ((float)(t2-t1) > 1.0) {
  sprintf(errorLine,">>>> TEST <<<< (pid=%d) inferiorFreeDefered took %5.2f secs, avg=%5.2f, worst=%5.2f, heapFree=%d, heapActive=%d, disabledList=%d, last call=%5.2f\n", proc->getPid(),(float) (t2-t1), (float) (totalTime/counter), (float)worst, hp->heapFree.size(), hp->heapActive.size(), hp->disabledList.size(), (float)(t1-t3));
  logLine(errorLine);
}
t3=t1;
#endif

    }
}


// initializes all DYNINST lib stuff: init the inferior heap and check for 
// required symbols.
// This is only called after we get the first breakpoint (SIGTRAP), because
// the DYNINST lib can be dynamically linked (currently it is only dynamically
// linked on Windows NT)
bool process::initDyninstLib() {

#if defined(i386_unknown_nt4_0)
   /***
     Kludge for Windows NT: we need to call waitProcs here so that
     we can load libdyninstRT when we attach to an already running
     process. The solution to avoid this kludge is to divide 
     attachProcess in two parts: first we attach to a process,
     and later, after libdyninstRT has been loaded,
     we install the call to DYNINSTinit.
    ***/

   // libDyninstRT should already be loaded when we get here,
   // except if the process was created via attach
   if (createdViaAttach) {
     // need to set reachedFirstBreak to false here, so that
     // libdyninstRT gets loaded in waitProcs.
     reachedFirstBreak = false;
     while (!hasLoadedDyninstLib) {
       int status;
       waitProcs(&status);
     }
   }
   assert(hasLoadedDyninstLib);
#endif


  extern vector<sym_data> syms_to_find;
  if (!heapIsOk(syms_to_find))
    return false;

  initInferiorHeap(false);
  if (splitHeaps)
    initInferiorHeap(true);

  return true;
}



//
// Process "normal" (non-attach, non-fork) ctor, for when a new process
// is fired up by paradynd itself.
//

process::process(int iPid, image *iImage, int iTraceLink, int iIoLink
#ifdef SHM_SAMPLING
                 , key_t theShmKey,
                 const vector<fastInferiorHeapMgr::oneHeapStats> &iShmHeapStats
#endif
) :
             baseMap(ipHash), 
             pid(iPid) // needed in fastInferiorHeap ctors below
#ifdef SHM_SAMPLING
             ,previous(0),
             inferiorHeapMgr(theShmKey, iShmHeapStats, iPid),
             theSuperTable(this,
                        iShmHeapStats[0].maxNumElems,
                        iShmHeapStats[1].maxNumElems,
                        iShmHeapStats[2].maxNumElems)
#endif
{
    hasBootstrapped = false;
    save_exitset_ptr = NULL;

    // the next two variables are used only if libdyninstRT is dynamically linked
    hasLoadedDyninstLib = false;
    isLoadingDyninstLib = false;

    reachedFirstBreak = false; // haven't yet seen first trap
    reachedVeryFirstTrap = false;
    createdViaAttach = false;
    needToContinueAfterDYNINSTinit = false;  //Wait for press of "RUN" button

    symbols = iImage;
    mainFunction = NULL; // set in platform dependent function heapIsOk

    status_ = neonatal;
    continueAfterNextStop_ = false;
    deferredContinueProc = false;

#ifndef BPATCH_LIBRARY
    string buffer = string(pid) + string("_") + getHostName();
    rid = resource::newResource(processResource, // parent
                                (void*)this, // handle
                                nullString, // abstraction
                                iImage->name(), // process name
                                0.0, // creation time
                                buffer, // unique name (?)
                                MDL_T_STRING, // mdl type (?)
                                true
                                );
#endif

    parent = NULL;
    bufStart = 0;
    bufEnd = 0;
    inExec = false;

    cumObsCost = 0;
    lastObsCostLow = 0;

    proc_fd = -1;

    trampTableItems = 0;
    memset(trampTable, 0, sizeof(trampTable));
    currentPC_ = 0;
    hasNewPC = false;
    
    dynamiclinking = false;
    dyn = new dynamic_linking;
    shared_objects = 0;
    all_functions = 0;
    all_modules = 0;
    some_modules = 0;
    some_functions = 0;
    waiting_for_resources = false;
    signal_handler = 0;
    execed_ = false;

#ifdef SHM_SAMPLING
#ifdef sparc_sun_sunos4_1_3
   kvmHandle = kvm_open(0, 0, 0, O_RDONLY, 0);
   if (kvmHandle == NULL) {
      perror("could not map child's uarea; kvm_open");
      exit(5);
   }

//   childUareaPtr = tryToMapChildUarea(iPid);
   childUareaPtr = NULL;
#endif
#endif

   splitHeaps = false;

#if defined(rs6000_ibm_aix3_2) || defined(rs6000_ibm_aix4_1)
        // XXXX - move this to a machine dependant place.

        // create a seperate text heap.
        //initInferiorHeap(true);
        splitHeaps = true;
#endif

   traceLink = iTraceLink;
   ioLink = iIoLink;

   RPCs_waiting_for_syscall_to_complete = false;

   // attach to the child process (machine-specific implementation)
   attach(); // error check?
}

//
// Process "attach" ctor, for when paradynd is attaching to an already-existing
// process.
//

process::process(int iPid, image *iSymbols,
                 int afterAttach, // 1 --> pause, 2 --> run, 0 --> leave as is
                 bool &success
#ifdef SHM_SAMPLING
                 , key_t theShmKey,
                 const vector<fastInferiorHeapMgr::oneHeapStats> &iShmHeapStats
#endif
                 ) :
                 baseMap(ipHash),
                 pid(iPid)
#ifdef SHM_SAMPLING
             ,previous(0),
             inferiorHeapMgr(theShmKey, iShmHeapStats, iPid),
             theSuperTable(this,
                        iShmHeapStats[0].maxNumElems,
                        iShmHeapStats[1].maxNumElems,
                        iShmHeapStats[2].maxNumElems)
#endif
{
   RPCs_waiting_for_syscall_to_complete = false;
   save_exitset_ptr = NULL;

   hasBootstrapped = false;
   reachedFirstBreak = true; // the initial trap of program entry was passed long ago...
   reachedVeryFirstTrap = true;
   createdViaAttach = true;

   // the next two variables are used only if libdyninstRT is dynamically linked
   hasLoadedDyninstLib = false;
   isLoadingDyninstLib = false;

   symbols = iSymbols;
   mainFunction = NULL; // set in platform dependent function heapIsOk

   status_ = neonatal;
   continueAfterNextStop_ = false;
   deferredContinueProc = false;

#ifndef BPATCH_LIBRARY
    string buffer = string(pid) + string("_") + getHostName();
    rid = resource::newResource(processResource, // parent
                                (void*)this, // handle
                                nullString, // abstraction
                                symbols->name(),
                                0.0, // creation time
                                buffer, // unique name (?)
                                MDL_T_STRING, // mdl type (?)
                                true
                                );
#endif

    parent = NULL;
    bufStart = 0;
    bufEnd = 0;
    inExec = false;

    cumObsCost = 0;
    lastObsCostLow = 0;

    proc_fd = -1;

    trampTableItems = 0;
    memset(trampTable, 0, sizeof(trampTable));
    currentPC_ = 0;
    hasNewPC = false;
    
    dynamiclinking = false;
    dyn = new dynamic_linking;
    shared_objects = 0;
    all_functions = 0;
    all_modules = 0;
    some_modules = 0;
    some_functions = 0;
    waiting_for_resources = false;
    signal_handler = 0;
    execed_ = false;

    splitHeaps = false;

#if defined(rs6000_ibm_aix3_2) || defined(rs6000_ibm_aix4_1)
        // XXXX - move this to a machine dependant place.

        // create a seperate text heap.
        //initInferiorHeap(true);
        splitHeaps = true;
#endif

   traceLink = -1; // will be set later, when the appl runs DYNINSTinit

   ioLink = -1; // (ARGUABLY) NOT YET IMPLEMENTED...MAYBE WHEN WE ATTACH WE DON'T WANT
                // TO REDIRECT STDIO SO WE CAN LEAVE IT AT -1.

   // Now the actual attach...the moment we've all been waiting for

   attach_cerr << "process attach ctor: about to attach to pid " << getPid() << endl;

   // It is assumed that a call to attach() doesn't affect the running status
   // of the process.  But, unfortunately, some platforms may barf if the
   // running status is anything except paused. (How to deal with this?)
   // Note that solaris in particular seems able to attach even if the process
   // is running.
   if (!attach()) {
      showErrorCallback(26, ""); // unable-to-attach
      success = false;
      return;
   }

#if defined(BPATCH_LIBRARY) && defined(rs6000_ibm_aix4_1)
   wasRunningWhenAttached = false; /* XXX Or should the default be true? */
#else
   wasRunningWhenAttached = isRunning_();
#endif

#if defined(BPATCH_LIBRARY) && defined(rs6000_ibm_aix4_1)
   // We use ptrace of AIX, which stops the process on attach.
   status_ = stopped;
#else
   // Note: we used to pause the program here, but not anymore.
   status_ = running;
#endif

#ifdef i386_unknown_nt4_0 // Except we still pause on NT.
    if (!pause())
        assert(false);
#endif

   if (afterAttach == 0)
      needToContinueAfterDYNINSTinit = wasRunningWhenAttached;
   else if (afterAttach == 1)
      needToContinueAfterDYNINSTinit = false;
   else if (afterAttach == 2)
      needToContinueAfterDYNINSTinit = true;
   else
      assert(false);

   // Does attach() send a SIGTRAP, a la the initial SIGTRAP sent at the
   // end of exec?  It seems that on some platforms it does; on others
   // it doesn't.  Ick.  On solaris, it doesn't.

   // note: we don't call getSharedObjects() yet; that happens once DYNINSTinit
   //       finishes (handleStartProcess)

   // Everything worked
   success = true;
}

//
// Process "fork" ctor, for when a process which is already being monitored by
// paradynd executes the fork syscall.
//

process::process(const process &parentProc, int iPid, int iTrace_fd
#ifdef SHM_SAMPLING
                 ,key_t theShmKey,
                 void *applShmSegPtr,
                 const vector<fastInferiorHeapMgr::oneHeapStats> &iShmHeapStats
#endif
                 ) :
                     baseMap(ipHash) // could change to baseMap(parentProc.baseMap)
#ifdef SHM_SAMPLING
                     ,previous(0),
                     inferiorHeapMgr(parentProc.inferiorHeapMgr, 
                                      applShmSegPtr,
                                      theShmKey, iShmHeapStats, iPid)
                     ,theSuperTable(parentProc.getTable(),this)
#endif
{
    // This is the "fork" ctor

    RPCs_waiting_for_syscall_to_complete = false;
    save_exitset_ptr = NULL;

    hasBootstrapped = false;
       // The child of fork ("this") has yet to run DYNINSTinit.

    // the next two variables are used only if libdyninstRT is dynamically linked
    hasLoadedDyninstLib = false; // TODO: is this the right value?
    isLoadingDyninstLib = false;

    createdViaAttach = parentProc.createdViaAttach;
    wasRunningWhenAttached = true;
    needToContinueAfterDYNINSTinit = true;

    symbols = parentProc.symbols; // shouldn't a reference count also be bumped?
    mainFunction = parentProc.mainFunction;

    traceLink = iTrace_fd;

    ioLink = -1; // when does this get set?

    status_ = neonatal; // is neonatal right?
    continueAfterNextStop_ = false;
    deferredContinueProc = false;

    pid = iPid; 

#ifndef BPATCH_LIBRARY
    string buffer = string(pid) + string("_") + getHostName();
    rid = resource::newResource(processResource, // parent
                                (void*)this, // handle
                                nullString, // abstraction
                                parentProc.symbols->name(),
                                0.0, // creation time
                                buffer, // unique name (?)
                                MDL_T_STRING, // mdl type (?)
                                true
                                );
#endif

    parent = &parentProc;
    
    bufStart = 0;
    bufEnd = 0;

    reachedFirstBreak = true; // initial TRAP has (long since) been reached
    reachedVeryFirstTrap = true;

    splitHeaps = parentProc.splitHeaps;

    heaps[0] = inferiorHeap(parentProc.heaps[0]);
    heaps[1] = inferiorHeap(parentProc.heaps[1]);

    inExec = false;

    cumObsCost = 0;
    lastObsCostLow = 0;

    proc_fd = -1;

    trampTableItems = 0;
    memset(trampTable, 0, sizeof(trampTable));
    currentPC_ = 0;
    hasNewPC = false;

    dynamiclinking = parentProc.dynamiclinking;
    dyn = new dynamic_linking;
    *dyn = *parentProc.dyn;

    shared_objects = 0;

    // make copy of parent's shared_objects vector
    if (parentProc.shared_objects) {
      shared_objects = new vector<shared_object*>;
      for (unsigned u1 = 0; u1 < parentProc.shared_objects->size(); u1++){
        *shared_objects += 
                new shared_object(*(*parentProc.shared_objects)[u1]);
      }
    }

    all_functions = 0;
    if (parentProc.all_functions) {
      all_functions = new vector<function_base *>;
      for (unsigned u2 = 0; u2 < parentProc.all_functions->size(); u2++)
        *all_functions += (*parentProc.all_functions)[u2];
    }

    all_modules = 0;
    if (parentProc.all_modules) {
      all_modules = new vector<module *>;
      for (unsigned u3 = 0; u3 < parentProc.all_modules->size(); u3++)
        *all_modules += (*parentProc.all_modules)[u3];
    }

    some_modules = 0;
    if (parentProc.some_modules) {
      some_modules = new vector<module *>;
      for (unsigned u4 = 0; u4 < parentProc.some_modules->size(); u4++)
        *some_modules += (*parentProc.some_modules)[u4];
    }
    
    some_functions = 0;
    if (parentProc.some_functions) {
      some_functions = new vector<function_base *>;
      for (unsigned u5 = 0; u5 < parentProc.some_functions->size(); u5++)
        *some_functions += (*parentProc.some_functions)[u5];
    }

    waiting_for_resources = false;
    signal_handler = parentProc.signal_handler;
    execed_ = false;

   // threads...
   for (unsigned i=0; i<parentProc.threads.size(); i++) {
     threads += new pdThread(this,parentProc.threads[i]);
   }

#if defined(SHM_SAMPLING) && defined(sparc_sun_sunos4_1_3)
   childUareaPtr = NULL;
#endif

#if defined(SHM_SAMPLING) && defined(MT_THREAD)
   // thread mapping table
   threadMap = new hashTable(parentProc.threadMap);
#endif

   if (!attach()) {     // moved from ::forkProcess
      showErrorCallback(69, "Error in fork: cannot attach to child process");
   }

   status_ = stopped;
      // would neonatal be more appropriate?  Nah, we've reached the first trap
}

#ifdef SHM_SAMPLING
void process::registerInferiorAttachedSegs(void *inferiorAttachedAtPtr) {
   shmsample_cerr << "process pid " << getPid() << ": welcome to register with inferiorAttachedAtPtr=" << inferiorAttachedAtPtr << endl;

   inferiorHeapMgr.registerInferiorAttachedAt(inferiorAttachedAtPtr);
   theSuperTable.setBaseAddrInApplic(0,(intCounter*) inferiorHeapMgr.getSubHeapInApplic(0));
   theSuperTable.setBaseAddrInApplic(1,(tTimer*) inferiorHeapMgr.getSubHeapInApplic(1));
   theSuperTable.setBaseAddrInApplic(2,(tTimer*) inferiorHeapMgr.getSubHeapInApplic(2));
}
#endif


extern bool forkNewProcess(string file, string dir, vector<string> argv, 
                    vector<string>envp, string inputFile, string outputFile,
                    int &traceLink, int &ioLink, 
                    int &pid, int &tid, 
                    int &procHandle, int &thrHandle);

/*
 * Create a new instance of the named process.  Read the symbols and start
 *   the program
 */
process *createProcess(const string File, vector<string> argv, vector<string> envp, const string dir = "")
{
    // prepend the directory (if any) to the file, unless the filename
    // starts with a /
    string file = File;
    if (!file.prefixed_by("/") && dir.length() > 0)
      file = dir + "/" + file;

#if defined(BPATCH_LIBRARY) && !defined(BPATCH_REDIRECT_IO)
    string inputFile;
    string outputFile;
#else
    // check for I/O redirection in arg list.
    string inputFile;
    for (unsigned i1=0; i1<argv.size(); i1++) {
      if (argv[i1] == "<") {
        inputFile = argv[i1+1];
        for (unsigned j=i1+2, k=i1; j<argv.size(); j++, k++)
          argv[k] = argv[j];
        argv.resize(argv.size()-2);
      }
    }
    // TODO -- this assumes no more than 1 of each "<", ">"
    string outputFile;
    for (unsigned i2=0; i2<argv.size(); i2++) {
      if (argv[i2] == ">") {
        outputFile = argv[i2+1];
        for (unsigned j=i2+2, k=i2; j<argv.size(); j++, k++)
          argv[k] = argv[j];
        argv.resize(argv.size()-2);
      }
    }
#endif /* BPATCH_LIBRARY */

    int traceLink;
    int ioLink;
    int pid;
    int tid;
    int procHandle;
    int thrHandle;

    if (!forkNewProcess(file, dir, argv, envp, inputFile, outputFile,
                   traceLink, ioLink, pid, tid, procHandle, thrHandle)) {
      // forkNewProcess is resposible for displaying error messages
      return NULL;
    }

#ifdef BPATCH_LIBRARY
    // Register the pid with the BPatch library (not yet associated with a
    // BPatch_thread object).
    assert(BPatch::bpatch != NULL);
    BPatch::bpatch->registerProvisionalThread(pid);
#endif

#if defined(rs6000_ibm_aix3_2) || defined(rs6000_ibm_aix4_1)
        extern bool establishBaseAddrs(int pid, int &status, bool waitForTrap);
        int status;

        if (!establishBaseAddrs(pid, status, true)) {
            return(NULL);
        }
#endif

#ifndef BPATCH_LIBRARY
// NEW: We bump up batch mode here; the matching bump-down occurs after shared objects
//      are processed (after receiving the SIGSTOP indicating the end of running
//      DYNINSTinit; more specifically, procStopFromDYNINSTinit().
//      Prevents a diabolical w/w deadlock on solaris --ari
tp->resourceBatchMode(true);
#endif /* BPATCH_LIBRARY */

        image *img = image::parseImage(file);
        if (!img) {
            // For better error reporting, two failure return values would be useful
            // One for simple error like because-file-not-because
            // Another for serious errors like found-but-parsing-failed (internal error;
            //    please report to paradyn@cs.wisc.edu)

            string msg = string("Unable to parse image: ") + file;
            showErrorCallback(68, msg.string_of());
            // destroy child process
            OS::osKill(pid);

            return(NULL);
        }

        /* parent */
        statusLine("initializing process data structures");

#ifdef SHM_SAMPLING
        vector<fastInferiorHeapMgr::oneHeapStats> theShmHeapStats(3);
        theShmHeapStats[0].elemNumBytes = sizeof(intCounter);
        theShmHeapStats[0].maxNumElems  = numIntCounters;

        theShmHeapStats[1].elemNumBytes = sizeof(tTimer);
        theShmHeapStats[1].maxNumElems  = numWallTimers;

        theShmHeapStats[2].elemNumBytes = sizeof(tTimer);
        theShmHeapStats[2].maxNumElems  = numProcTimers;
#endif

        process *ret = new process(pid, img, traceLink, ioLink
#ifdef SHM_SAMPLING
                                   , 7000, // shm seg key to try first
                                   theShmHeapStats
#endif
                                   );
           // change this to a ctor that takes in more args

        assert(ret);

        processVec += ret;
        activeProcesses++;

#ifndef BPATCH_LIBRARY
        if (!costMetric::addProcessToAll(ret))
           assert(false);
#endif
        // find the signal handler function
        ret->findSignalHandler(); // should this be in the ctor?

        // initializing vector of threads - thread[0] is really the 
        // same process

#if defined(i386_unknown_nt4_0)
        ret->threads += new pdThread(ret, tid, (handleT)thrHandle);
#else
        ret->threads += new pdThread(ret);
#endif

        // initializing hash table for threads. This table maps threads to
        // positions in the superTable - naim 4/14/97
#if defined(SHM_SAMPLING) && defined(MT_THREAD)
        ret->threadMap = new hashTable(MAX_NUMBER_OF_THREADS,1,0);
#endif

        // we use this flag to solve race condition between inferiorRPC and 
        // continueProc message from paradyn - naim
        ret->deferredContinueProc = false;

        ret->numOfActCounters_is=0;
        ret->numOfActProcTimers_is=0;
        ret->numOfActWallTimers_is=0;

#if defined(rs6000_ibm_aix3_2) || defined(rs6000_ibm_aix4_1)
        // XXXX - this is a hack since establishBaseAddrs needed to wait for
        //    the TRAP signal.
        // We really need to move most of the above code (esp parse image)
        //    to the TRAP signal handler.  The problem is that we don't
        //    know the base addresses until we get the load info via ptrace.
        //    In general it is even harder, since dynamic libs can be loaded
        //    at any time.
        extern int handleSigChild(int pid, int status);

        (void) handleSigChild(pid, status);
#endif
    return ret;

}


void process::DYNINSTinitCompletionCallback(process *theProc,
                                            void *, // user data
                                            unsigned // return value from DYNINSTinit
                                            ) {
   attach_cerr << "Welcome to DYNINSTinitCompletionCallback" << endl;
   theProc->handleCompletionOfDYNINSTinit(true);
}


bool attachProcess(const string &progpath, int pid, int afterAttach
#ifdef BPATCH_LIBRARY
                   , process *&newProcess
#endif
                   ) {
   // implementation of dynRPC::attach() (the igen call)
   // This is meant to be "the other way" to start a process (competes w/ createProcess)

   // progpath gives the full path name of the executable, which we use ONLY to
   // read the symbol table.

   // We try to make progpath optional, since given pid, we should be able to
   // calculate it with a clever enough search of the process' PATH, examining
   // its argv[0], examining its current directory, etc.  /proc gives us this
   // information on solaris...not sure about other platforms...

   // possible values for afterAttach: 1 --> pause, 2 --> run, 0 --> leave as is

   attach_cerr << "welcome to attachProcess for pid " << pid << endl;

   // QUESTION: When we attach to a process, do we want to redirect its stdout/stderr
   //           (like we do when we fork off a new process the 'usual' way)?
   //           My first guess would be no.  -ari
   //           But although we may ignore the io, we still need the trace stream.

   // When we attach to a process, we don't fork...so this routine is much simpler
   // than its "competitor", createProcess() (above).

   // TODO: What about AIX establishBaseAddrs???  Do that now?

   string fullPathToExecutable = process::tryToFindExecutable(progpath, pid);
   if (!fullPathToExecutable.length())
      return false;

#ifndef BPATCH_LIBRARY
   tp->resourceBatchMode(true);
      // matching bump-down occurs in procStopFromDYNINSTinit().
#endif

   image *theImage = image::parseImage(fullPathToExecutable);
   if (theImage == NULL) {
      // two failure return values would be useful here, to differentiate
      // file-not-found vs. catastrophic-parse-error.
      string msg = string("Unable to parse image: ") + fullPathToExecutable;
      showErrorCallback(68, msg.string_of());
      return false; // failure
   }

#ifdef SHM_SAMPLING
   vector<fastInferiorHeapMgr::oneHeapStats> theShmHeapStats(3);
   theShmHeapStats[0].elemNumBytes = sizeof(intCounter);
   theShmHeapStats[0].maxNumElems  = numIntCounters;

   theShmHeapStats[1].elemNumBytes = sizeof(tTimer);
   theShmHeapStats[1].maxNumElems  = numWallTimers;

   theShmHeapStats[2].elemNumBytes = sizeof(tTimer);
   theShmHeapStats[2].maxNumElems  = numProcTimers;
#endif

   // NOTE: the actual attach happens in the process "attach" constructor:
   bool success;
   process *theProc = new process(pid, theImage, afterAttach, success
#ifdef SHM_SAMPLING
                                  ,7000, // shm seg key to try first
                                  theShmHeapStats
#endif                            
                                  );
   assert(theProc);
   if (!success) {
       // XXX Do we need to do something to get rid of theImage, too?
       delete theProc;
       return false;
   }

   // Note: it used to be that the attach ctor called pause()...not anymore...so
   // the process is probably running even as we speak.

   processVec += theProc;
   activeProcesses++;

   theProc->threads += new pdThread(theProc);

#if defined(USES_LIBDYNINSTRT_SO)
   // we now need to dynamically load libdyninstRT.so.1 - naim
   if (theProc->handleStartProcess()) {
     if (!theProc->pause()) {
       logLine("WARNING: pause failed\n");
       assert(0);
     }
     theProc->dlopenDYNINSTlib();
     // this will set isLoadingDyninstLib to true - naim
     if (!theProc->continueProc()) {
       logLine("WARNING: continueProc failed\n");
       assert(0);
     }
     int status;
     while (!theProc->dyninstLibAlreadyLoaded()) {
       theProc->waitProcs(&status);
     }
   }
#endif

   theProc->initDyninstLib();

#ifndef BPATCH_LIBRARY
   if (!costMetric::addProcessToAll(theProc))
      assert(false);
#endif

   // find the signal handler function
   theProc->findSignalHandler(); // shouldn't this be in the ctor?

   // Now force DYNINSTinit() to be invoked, via inferiorRPC.
   string buffer = string("PID=") + string(pid) + ", running DYNINSTinit()...";
   statusLine(buffer.string_of());

#ifdef BPATCH_LIBRARY
   newProcess = theProc;

   vector<AstNode*> the_args(2);

   the_args[0] = new AstNode(AstNode::Constant, (void*)3);
   the_args[1] = new AstNode(AstNode::Constant, (void*)getpid());
#else /* BPATCH_LIBRARY */
   attach_cerr << "calling DYNINSTinit with args:" << endl;

   vector<AstNode*> the_args(3);

#ifdef SHM_SAMPLING
   the_args[0] = new AstNode(AstNode::Constant,
                             (void*)(theProc->getShmKeyUsed()));
   attach_cerr << theProc->getShmKeyUsed() << endl;

   const unsigned shmHeapTotalNumBytes = theProc->getShmHeapTotalNumBytes();
   the_args[1] = new AstNode(AstNode::Constant,
                             (void*)shmHeapTotalNumBytes);
   attach_cerr << shmHeapTotalNumBytes << endl;;
#else
   // 2 dummy args when not shm sampling -- just make sure they're not both -1, which
   // would indicate that we're called from fork
   the_args[0] = new AstNode(AstNode::Constant, (void*)0);
   the_args[1] = new AstNode(AstNode::Constant, (void*)0);
#endif

   /*
      The third argument to DYNINSTinit is our (paradynd's) pid. It is used
      by DYNINSTinit to build the socket path to which it connects to in order
      to get the trace-stream connection.  We make it negative to indicate
      to DYNINSTinit that it's being called from attach (sorry for that little
      kludge...if we didn't have it, we'd probably need to boost DYNINSTinit
      from 3 to 4 parameters).
      
      This socket is set up in controllerMainLoop (perfStream.C).
   */
   the_args[2] = new AstNode(AstNode::Constant, (void*)(-1 * traceConnectInfo));
   attach_cerr << (-1* getpid()) << endl;
#endif /* BPATCH_LIBRARY */

   AstNode *the_ast = new AstNode("DYNINSTinit", the_args);
   for (unsigned j=0;j<the_args.size();j++) removeAst(the_args[j]);

   theProc->postRPCtoDo(the_ast,
                        true, // true --> don't try to update cost yet
                        process::DYNINSTinitCompletionCallback, // callback routine
                        NULL, // user data
                        -1);  // we use -1 if this is not metric definition - naim
      // the rpc will be launched with a call to launchRPCifAppropriate()
      // in the main loop (perfStream.C).
      // DYNINSTinit() ends with a DYNINSTbreakPoint(), so we pick up
      // where we left off in the processing of the forwarded SIGSTOP signal.
      // In other words, there's lots more work to do, but since we can't do it until
      // DYNINSTinit has run, we wait until the SIGSTOP is forwarded.

   // Note: we used to pause() the process while attaching.  Not anymore.
   // The attached process is running even as we speak.  (Though we'll interrupt
   // it pretty soon when the inferior RPC of DYNINSTinit gets launched).

   return true; // successful
}

#ifdef SHM_SAMPLING
bool process::doMajorShmSample(time64 theWallTime) {
   bool result = true; // will be set to false if any processAll() doesn't complete
                       // successfully.

   if (!theSuperTable.doMajorSample  (theWallTime, 0))
      result = false;
      // inferiorProcessTimers used to take in a non-dummy process time as the
      // 2d arg, but it looks like that we need to re-read the process time for
      // each proc timer, at the time of sampling the timer's value, to avoid
      // ugly jagged spikes in histogram (i.e. to avoid incorrect sampled 
      // values).  Come to think of it: the same may have to be done for the 
      // wall time too!!!

   const time64 theProcTime = getInferiorProcessCPUtime();

   // Now sample the observed cost.
   unsigned *costAddr = this->getObsCostLowAddrInParadyndSpace();
   const unsigned theCost = *costAddr; // WARNING: shouldn't we be using a mutex?!

   this->processCost(theCost, theWallTime, theProcTime);

   return result;
}

bool process::doMinorShmSample() {
   // Returns true if the minor sample has successfully completed all outstanding
   // samplings.
   bool result = true; // so far...

   if (!theSuperTable.doMinorSample())
      result = false;

   return result;
}
#endif

extern void removeFromMetricInstances(process *);
extern void disableAllInternalMetrics();

void handleProcessExit(process *proc, int exitStatus) {
  if (proc->status() == exited)
    return;

  proc->Exited(); // updates status line

  --activeProcesses;

#ifndef BPATCH_LIBRARY
  if (activeProcesses == 0)
    disableAllInternalMetrics();
#endif

#ifdef PARADYND_PVM
  if (pvm_running) {
    PDYN_reportSIGCHLD(proc->getPid(), exitStatus);
  }
#endif

  // Perhaps these lines can be un-commented out in the future, but since
  // cleanUpAndExit() does the same thing, and it always gets called
  // (when paradynd detects that paradyn died), it's not really necessary
  // here.  -ari
//  for (unsigned lcv=0; lcv < processVec.size(); lcv++)
//     if (processVec[lcv] == proc) {
//        delete proc; // destructor removes shm segments...
//      processVec[lcv] = NULL;
//     }
}


/*
   process::forkProcess: called when a process forks, to initialize a new
   process object for the child.

   the variable childHasInstrumentation is true if the child process has the 
   instrumentation of the parent. This is the common case.
   On some platforms (AIX) the child does not have any instrumentation because
   the text segment of the child is not a copy of the parent text segment at
   the time of the fork, but a copy of the original text segment of the parent,
   without any instrumentation.
   (actually, childHasInstr is obsoleted by aix's completeTheFork() routine)
*/
process *process::forkProcess(const process *theParent, pid_t childPid,
                      dictionary_hash<instInstance*,instInstance*> &map, // gets filled in
                      int iTrace_fd
#ifdef SHM_SAMPLING
                              ,key_t theKey,
                              void *applAttachedPtr
#endif
                              ) {
#ifdef SHM_SAMPLING
    vector<fastInferiorHeapMgr::oneHeapStats> theShmHeapStats(3);
    theShmHeapStats[0].elemNumBytes = sizeof(intCounter);
    theShmHeapStats[0].maxNumElems  = numIntCounters;
    
    theShmHeapStats[1].elemNumBytes = sizeof(tTimer);
    theShmHeapStats[1].maxNumElems  = numWallTimers;

    theShmHeapStats[2].elemNumBytes = sizeof(tTimer);
    theShmHeapStats[2].maxNumElems  = numProcTimers;
#endif

    forkexec_cerr << "paradynd welcome to process::forkProcess; parent pid=" << theParent->getPid() << "; calling fork ctor now" << endl;

    // Call the "fork" ctor:
    process *ret = new process(*theParent, (int)childPid, iTrace_fd
#ifdef SHM_SAMPLING
                               , theKey,
                               applAttachedPtr,
                               theShmHeapStats
#endif
                               );
    assert(ret);

    forkexec_cerr << "paradynd fork ctor has completed ok...child pid is " << ret->getPid() << endl;

    processVec += ret;
    activeProcesses++;

#ifndef BPATCH_LIBRARY
    if (!costMetric::addProcessToAll(ret))
       assert(false);
#endif

    // We used to do a ret->attach() here...it was moved to the fork ctor, so it's
    // been done already.

    /* all instrumentation on the parent is active on the child */
    /* TODO: what about instrumentation inserted near the fork time??? */
    ret->baseMap = theParent->baseMap; // WHY IS THIS HERE?

    // the following writes to "map", s.t. for each instInstance in the parent
    // process, we have a map to the corresponding one in the child process.
    // that's all this routine does -- it doesn't actually touch
    // any instrumentation (because it doesn't need to -- fork() syscall copied
    // all of the actual instrumentation [but what about AIX and its weird load
    // behavior?])
    copyInstInstances(theParent, ret, map);
         // doesn't copy anything; just writes to "map"

    return ret;
}

#ifdef SHM_SAMPLING
void process::processCost(unsigned obsCostLow,
                          time64 wallTime,
                          time64 processTime) {
   // wallTime and processTime should compare to DYNINSTgetWallTime() and
   // DYNINSTgetCPUtime().

   // check for overflow, add to running total, convert cycles to
   // seconds, and report.
   // Member vrbles of class process: lastObsCostLow and cumObsCost (the latter
   // a 64-bit value).

   // code to handle overflow used to be in rtinst; we borrow it pretty much
   // verbatim. (see rtinst/RTposix.c)
   if (obsCostLow < lastObsCostLow) {
      // we have a wraparound
      cumObsCost += ((unsigned)0xffffffff - lastObsCostLow) + obsCostLow + 1;
   }
   else
      cumObsCost += (obsCostLow - lastObsCostLow);

   lastObsCostLow = obsCostLow;

   extern double cyclesPerSecond; // perfStream.C

   double observedCostSecs = cumObsCost;
   observedCostSecs /= cyclesPerSecond;
//   cerr << "processCost: cyclesPerSecond=" << cyclesPerSecond << "; cum obs cost=" << observedCostSecs << endl;

   // Notice how most of the rest of this is copied from processCost() of metric.C
   // Be sure to keep the two "in sync"!
   timeStamp newSampleTime  = (double)wallTime / 1000000.0; // usec to seconds
   timeStamp newProcessTime = (double)processTime / 1000000.0; // usec to secs

   extern costMetric *totalPredictedCost; // init.C
   extern costMetric *observed_cost; // init.C
   extern costMetric *smooth_obs_cost; // init.C

   const timeStamp lastProcessTime =
                        totalPredictedCost->getLastSampleProcessTime(this);

    // find the portion of uninstrumented time for this interval
    const double unInstTime = ((newProcessTime - lastProcessTime)
                         / (1+currentPredictedCost));
    // update predicted cost
    // note: currentPredictedCost is the same for all processes
    //       this should be changed to be computed on a per process basis
    sampleValue newPredCost = totalPredictedCost->getCumulativeValue(this);
    newPredCost += (float)(currentPredictedCost*unInstTime);

    totalPredictedCost->updateValue(this,newPredCost,
                                    newSampleTime,newProcessTime);
    // update observed cost
    observed_cost->updateValue(this,observedCostSecs,
                               newSampleTime,newProcessTime);

    // update smooth observed cost
    smooth_obs_cost->updateSmoothValue(this,observedCostSecs,
                                       newSampleTime,newProcessTime);
}
#endif

/*
 * Copy data from controller process to the named process.
 */
bool process::writeDataSpace(void *inTracedProcess, int size,
                             const void *inSelf) {
  bool needToCont = false;

  if (status_ == exited)
    return false;

  if (status_ == running) {
    needToCont = true;
    if (! pause())
      return false;
  }

  if (status_ != stopped && status_ != neonatal) {
    showErrorCallback(38, "Internal paradynd error in process::writeDataSpace");
    return false;
  }

  bool res = writeDataSpace_(inTracedProcess, size, inSelf);
  if (!res) {
    string msg = string("System error: unable to write to process data space:")
                   + string(sys_errlist[errno]);
    showErrorCallback(38, msg);
    return false;
  }

  if (needToCont)
    return this->continueProc();
  return true;
}

bool process::readDataSpace(const void *inTracedProcess, int size,
                            void *inSelf, bool displayErrMsg) {
  bool needToCont = false;

  if (status_ == exited)
    return false;

  if (status_ == running) {
    needToCont = true;
    if (! pause()) {
      sprintf(errorLine, "in readDataSpace, status_ = running, but unable to pause()\n");
      logLine(errorLine);
      return false;
    }
  }

  if (status_ != stopped && status_ != neonatal) {
    showErrorCallback(38, "Internal paradynd error in process::readDataSpace");
    sprintf(errorLine, "Internal paradynd error in process::readDataSpace");
    logLine(errorLine);
    return false;
  }

  bool res = readDataSpace_(inTracedProcess, size, inSelf);
  if (!res) {
    if (displayErrMsg) {
      string msg;
      msg=string("System error: unable to read from process data space:")
          + string(sys_errlist[errno]);
      sprintf(errorLine, msg.string_of());
      logLine(errorLine);
      showErrorCallback(38, msg);
    }
    return false;
  }

  if (needToCont) {
    needToCont = this->continueProc();
    if (!needToCont) {
        sprintf(errorLine, "warning : readDataSpace, needToCont FALSE, returning FALSE\n");
        logLine(errorLine);
    }
    return needToCont;
  }
  return true;

}

bool process::writeTextWord(caddr_t inTracedProcess, int data) {
  bool needToCont = false;

  if (status_ == exited)
    return false;

  if (status_ == running) {
    needToCont = true;
    if (! pause())
      return false;
  }

  if (status_ != stopped && status_ != neonatal) {
    string msg = string("Internal paradynd error in process::writeTextWord")
               + string((int)status_);
    showErrorCallback(38, msg);
    //showErrorCallback(38, "Internal paradynd error in process::writeTextWord");
    return false;
  }

#ifdef BPATCH_SET_MUTATIONS_ACTIVE
  if (!isAddrInHeap((Address)inTracedProcess)) {
    if (!saveOriginalInstructions((Address)inTracedProcess, sizeof(int)))
        return false;
    afterMutationList.insertTail((Address)inTracedProcess, sizeof(int), &data);
  }
#endif

  bool res = writeTextWord_(inTracedProcess, data);
  if (!res) {
    string msg = string("System error: unable to write to process text word:")
                   + string(sys_errlist[errno]);
    showErrorCallback(38, msg);
    return false;
  }

  if (needToCont)
    return this->continueProc();
  return true;

}

bool process::writeTextSpace(void *inTracedProcess, int amount, const void *inSelf) {
  bool needToCont = false;

  if (status_ == exited)
    return false;

  if (status_ == running) {
    needToCont = true;
    if (! pause())
      return false;
  }

  if (status_ != stopped && status_ != neonatal) {
    string msg = string("Internal paradynd error in process::writeTextSpace")
               + string((int)status_);
    showErrorCallback(38, msg);
    //showErrorCallback(38, "Internal paradynd error in process::writeTextSpace");
    return false;
  }

#ifdef BPATCH_SET_MUTATIONS_ACTIVE
  if (!isAddrInHeap((Address)inTracedProcess)) {
    if (!saveOriginalInstructions((Address)inTracedProcess, amount))
        return false;
    afterMutationList.insertTail((Address)inTracedProcess, amount, inSelf);
  }
#endif

  bool res = writeTextSpace_(inTracedProcess, amount, inSelf);
  if (!res) {
    string msg = string("System error: unable to write to process text space:")
                   + string(sys_errlist[errno]);
    showErrorCallback(38, msg);
    return false;
  }

  if (needToCont)
    return this->continueProc();
  return true;
}

#ifdef BPATCH_SET_MUTATIONS_ACTIVE
bool process::readTextSpace(const void *inTracedProcess, int amount,
                            const void *inSelf)
{
  bool needToCont = false;

  if (status_ == exited)
    return false;

  if (status_ == running) {
    needToCont = true;
    if (! pause())
      return false;
  }

  if (status_ != stopped && status_ != neonatal) {
    showErrorCallback(38, "Internal paradynd error in process::readTextSpace");
    return false;
  }

  bool res = readTextSpace_((void *)inTracedProcess, amount, inSelf);
  if (!res) {
    string msg;
    msg=string("System error: unable to read from process data space:")
        + string(sys_errlist[errno]);
    showErrorCallback(38, msg);
    return false;
  }

  if (needToCont)
    return this->continueProc();
  return true;

}
#endif /* BPATCH_SET_MUTATIONS_ACTIVE */

bool process::pause() {
  if (status_ == stopped || status_ == neonatal)
    return true;

  if (status_ == exited) {
    sprintf(errorLine, "warn : in process::pause, trying to pause exited process, returning FALSE");
    logLine(errorLine);
    return false;
  }

  if (status_ == running && reachedFirstBreak) {
    bool res = pause_();
    if (!res) {
      sprintf(errorLine, "warn : in process::pause, pause_ unable to pause process");
    logLine(errorLine);
      return false;
    }

    status_ = stopped;
  }
  else {
    // The only remaining combination is: status==running but haven't yet
    // reached first break.  We never want to pause before reaching the
    // first break (trap, actually).  But should we be returning true or false in this
    // case?
  }

  return true;
}

// handleIfDueToSharedObjectMapping: if a trap instruction was caused by
// a dlopen or dlclose event then return true
bool process::handleIfDueToSharedObjectMapping(){

  if(!dyn) { 
    return false;
  }

  vector<shared_object *> *changed_objects = 0;
  u_int change_type = 0;
  bool error_occured = false;
  bool ok = dyn->handleIfDueToSharedObjectMapping(this,&changed_objects,
                                                  change_type,error_occured);

  // if this trap was due to dlopen or dlclose, and if something changed
  // then figure out how it changed and either add or remove shared objects
  if(ok && !error_occured && (change_type != SHAREDOBJECT_NOCHANGE)) {

      // if something was added then call process::addASharedObject with
      // each element in the vector of changed_objects
      if((change_type == SHAREDOBJECT_ADDED) && changed_objects) {
          for(u_int i=0; i < changed_objects->size(); i++) {
             // TODO: currently we aren't handling dlopen because  
             // we don't have the code in place to modify existing metrics
             // This is what we really want to do:
#ifndef BPATCH_LIBRARY
             if (((*changed_objects)[i])->getName() == string(getenv("PARADYN_LIB"))) 
             {
#endif
               if(addASharedObject(*((*changed_objects)[i]))){
                 *shared_objects += (*changed_objects)[i];
                 hasLoadedDyninstLib = true;
                 isLoadingDyninstLib = false;
               } else {
                 logLine("Error after call to addASharedObject\n");
                 delete (*changed_objects)[i];
               }
#ifndef BPATCH_LIBRARY
             } else {
               // for now, just delete shared_objects to avoid memory leeks
               delete (*changed_objects)[i];
             }
#endif
          }
          delete changed_objects;
      } 
      else if((change_type == SHAREDOBJECT_REMOVED) && (changed_objects)) { 
          // TODO: handle this case
          // if something was removed then call process::removeASharedObject
          // with each element in the vector of changed_objects

          // for now, just delete shared_objects to avoid memory leeks
          for(u_int i=0; i < changed_objects->size(); i++){
              delete (*changed_objects)[i];
          }
          delete changed_objects;
      }

      // TODO: add support for adding or removing new code resource once the 
      // process has started running...this means that currently collected
      // metrics may have to have aggregate components added or deleted
      // this should be added to process::addASharedObject and 
      // process::removeASharedObject  
  }
  return ok;
}





//
//  If this process is a dynamic executable, then get all its 
//  shared objects, parse them, and define new instpoints and resources 
//
bool process::handleStartProcess(){

    // get shared objects, parse them, and define new resources 
    // For WindowsNT we don't call getSharedObjects here, instead
    // addASharedObject will be called directly by pdwinnt.C
#if !defined(i386_unknown_nt4_0)
    this->getSharedObjects();
#endif

#ifndef BPATCH_LIBRARY
    if(resource::num_outstanding_creates)
       this->setWaitingForResources();
#endif

    return true;
}

// addASharedObject: This routine is called whenever a new shared object
// has been loaded by the run-time linker
// It processes the image, creates new resources
bool process::addASharedObject(shared_object &new_obj){

    image *img = image::parseImage(new_obj.getName(),new_obj.getBaseAddress());
    if(!img){
        logLine("error parsing image in addASharedObject\n");
        return false;
    }
    new_obj.addImage(img);

    // if the list of all functions and all modules have already been 
    // created for this process, then the functions and modules from this
    // shared object need to be added to those lists 
    if(all_modules){
        *all_modules += *((vector<module *> *)(new_obj.getModules())); 
    }
    if(all_functions){
        vector<function_base *> *normal_funcs = (vector<function_base *> *)
                        (new_obj.getAllFunctions());
        *all_functions += *normal_funcs; 
        normal_funcs = 0;
    }

    // if the signal handler function has not yet been found search for it
    if(!signal_handler){
        signal_handler = img->findOneFunction(SIGNAL_HANDLER);
    }

    // clear the include_funcs flag if this shared object should not be
    // included in the some_functions and some_modules lists
#ifndef BPATCH_LIBRARY
    vector<string> lib_constraints;
    if(mdl_get_lib_constraints(lib_constraints)){
        for(u_int j=0; j < lib_constraints.size(); j++){
           char *where = 0; 
           // if the lib constraint is not of the form "module/function" and
           // if it is contained in the name of this object, then exclude
           // this shared object
           char *obj_name = P_strdup(new_obj.getName().string_of());
           char *lib_name = P_strdup(lib_constraints[j].string_of());
           if(obj_name && lib_name && (where=P_strstr(obj_name, lib_name))){
              new_obj.changeIncludeFuncs(false); 
           }
           if(lib_name) free(lib_name);
           if(obj_name) free(obj_name);
        }
    }

    // This looks a bit wierd at first glance, but apparently is ok -
    //  A shared object has 1 module.  If that module is excluded,
    //  then new_obj.includeFunctions() should return FALSE.  As such,
    //  the some_modules += new_obj.getModules() is OK as long as
    //  shared objects have ONLY 1 module.
    if(new_obj.includeFunctions()){
        if(some_modules) {
            *some_modules += *((vector<module *> *)(new_obj.getModules())); 
        }
        if(some_functions) {
            // gets only functions not excluded by mdl "exclude_node" option
            *some_functions += 
                *((vector<function_base *> *)(new_obj.getSomeFunctions()));
        }
    }
#endif /* BPATCH_LIBRARY */
    return true;
}

// getSharedObjects: This routine is called before main() or on attach
// to an already running process.  It gets and process all shared objects
// that have been mapped into the process's address space
bool process::getSharedObjects() {

    assert(!shared_objects);
    shared_objects = dyn->getSharedObjects(this); 
    if(shared_objects){
        statusLine("parsing shared object files");

#ifndef BPATCH_LIBRARY
        tp->resourceBatchMode(true);
#endif
        // for each element in shared_objects list process the 
        // image file to find new instrumentaiton points
        for(u_int j=0; j < shared_objects->size(); j++){
            //string temp2 = string(i);
            //temp2 += string("th shared obj, addr: ");
            //temp2 += string(((*shared_objects)[i])->getBaseAddress());
            //temp2 += string(" name: ");
            //temp2 += string(((*shared_objects)[i])->getName());
            //temp2 += string("\n");
            // logLine(P_strdup(temp2.string_of()));

            if(!addASharedObject(*((*shared_objects)[j]))){
                logLine("Error after call to addASharedObject\n");
            }
        }

#ifndef BPATCH_LIBRARY
        tp->resourceBatchMode(false);
#endif
        return true;
    }
    // else this a.out does not have a .dynamic section
    dynamiclinking = false;
    return false;
}

// findOneFunction: returns the function associated with func  
// this routine checks both the a.out image and any shared object
// images for this resource.
// Semantics of excluded functions - Once "exclude" works for both
// static and dynamically linked objects, this should return NULL
// if the function being sought is excluded....
#ifndef BPATCH_LIBRARY
function_base *process::findOneFunction(resource *func,resource *mod){
    
    if((!func) || (!mod)) { return 0; }
    if(func->type() != MDL_T_PROCEDURE) { return 0; }
    if(mod->type() != MDL_T_MODULE) { return 0; }

    const vector<string> &f_names = func->names();
    const vector<string> &m_names = mod->names();
    string func_name = f_names[f_names.size() -1]; 
    string mod_name = m_names[m_names.size() -1]; 

    //cerr << "process::findOneFunction called.  function name = " 
    //     << func_name.string_of() << endl;

    // KLUDGE: first search any shared libraries for the module name 
    //  (there is only one module in each shared library, and that 
    //  is the library name)
    if(dynamiclinking && shared_objects){
        for(u_int j=0; j < shared_objects->size(); j++){
            module *next = 0;
            next = ((*shared_objects)[j])->findModule(mod_name,true);
            if(next){
                if(((*shared_objects)[j])->includeFunctions()){ 
                    //cerr << "function found in module " << mod_name.string_of() << endl;
                    return(((*shared_objects)[j])->findOneFunction(func_name,
                                                                   true));
                } 
                else { 
                    //cerr << "function found in module " << mod_name.string()
                    //     << " that module excluded" << endl;
                  return 0;
                } 
            }
        }
    }

    return(symbols->findOneFunction(func_name));
}
#endif /* BPATCH_LIBRARY */

#ifndef BPATCH_LIBRARY
// returns all the functions in the module "mod" that are not excluded by
// exclude_lib or exclude_func
// return 0 on error.
vector<function_base *> *process::getIncludedFunctions(module *mod) {

    if((!mod)) { return 0; }

    //cerr << "process::getIncludedFunctions(" << mod->fileName() << ") called" << endl;

    // KLUDGE: first search any shared libraries for the module name 
    //  (there is only one module in each shared library, and that 
    //  is the library name)
    if(dynamiclinking && shared_objects){
        for(u_int j=0; j < shared_objects->size(); j++){
            module *next = 0;
            next = ((*shared_objects)[j])->findModule(mod->fileName(), true);
            if(next){
                if(((*shared_objects)[j])->includeFunctions()){ 
                    return((vector<function_base *> *)
                           ((*shared_objects)[j])->getSomeFunctions());
                } 
                else { return 0;} 
            }
        }
    }

    // this must be an a.out module so just return the list associated
    // with the module.
    // Now that exclude should work for both static and dynamically
    // linked executables, probably need to either filter excluded
    // files here, or let the module do it and pass the information not
    // to include excluded functions along with this proc call....
    // mcheyney 970927
    return(mod->getIncludedFunctions());
}
#endif /* BPATCH_LIBRARY */


// findOneFunction: returns the function associated with func  
// this routine checks both the a.out image and any shared object
// images for this resource
function_base *process::findOneFunction(const string &func_name){
    // first check a.out for function symbol
    function_base *pdf = symbols->findOneFunction(func_name);
    if(pdf) {
      return pdf;
    }

    // search any shared libraries for the file name 
    if(dynamiclinking && shared_objects){
        for(u_int j=0; j < shared_objects->size(); j++){
            pdf = ((*shared_objects)[j])->findOneFunction(func_name,false);
            if(pdf){
                return(pdf);
            }
    } }
    return(0);
}

// findOneFunctionFromAll: returns the function associated with func  
// this routine checks both the a.out image and any shared object
// images for this resource
function_base *process::findOneFunctionFromAll(const string &func_name){
    
    // first check a.out for function symbol
    function_base *pdf = symbols->findOneFunctionFromAll(func_name);
    if(pdf) return pdf;

    // search any shared libraries for the file name 
    if(dynamiclinking && shared_objects){
        for(u_int j=0; j < shared_objects->size(); j++){
            pdf=((*shared_objects)[j])->findOneFunctionFromAll(func_name,false);
            if(pdf){
                return(pdf);
            }
    } }
    return(0);

    return(0);
}

// findpdFunctionIn: returns the function which contains this address
// This routine checks both the a.out image and any shared object images
// for this function
pd_Function *process::findpdFunctionIn(Address adr) {

    // first check a.out for function symbol
    // findFunctionInInstAndUnInst will search the instrumentable and
    // uninstrumentable functions. We are assuming that "adr" is the
    // entry point of the function, so we will use as the key to search
    // in the dictionary - naim
    pd_Function *pdf = symbols->findFunctionInInstAndUnInst(adr,this);
    if(pdf) return pdf;
    // search any shared libraries for the function 
    if(dynamiclinking && shared_objects){
        for(u_int j=0; j < shared_objects->size(); j++){
          pdf = ((*shared_objects)[j])->findFunctionInInstAndUnInst(adr,this);
          if(pdf){
            return(pdf);          
          }
        }
    }

    if(!all_functions) getAllFunctions();

    // if the function was not found, then see if this addr corresponds
    // to  a function that was relocated in the heap
    if(all_functions){
        for(u_int j=0; j < all_functions->size(); j++){
            Address func_adr = ((*all_functions)[j])->getAddress(this);
            if((adr>=func_adr) && 
                (adr<=(((*all_functions)[j])->size()+func_adr))){
                // yes, this is very bad, but too bad
                return((pd_Function*)((*all_functions)[j]));
            }
        }
    }
    return(0);
}
    

// findFunctionIn: returns the function containing the address "adr"
// this routine checks both the a.out image and any shared object
// images for this resource
function_base *process::findFunctionIn(Address adr){

    // first check a.out for function symbol
    pd_Function *pdf = symbols->findFunctionIn(adr,this);
    if(pdf) return pdf;
    // search any shared libraries for the function 
    if(dynamiclinking && shared_objects){
        for(u_int j=0; j < shared_objects->size(); j++){
            pdf = ((*shared_objects)[j])->findFunctionIn(adr,this);
            if(pdf){
                return(pdf);
            }
    } }

    if(!all_functions) getAllFunctions();

    // if the function was not found, then see if this addr corresponds
    // to  a function that was relocated in the heap
    if(all_functions){
        for(u_int j=0; j < all_functions->size(); j++){
            Address func_adr = ((*all_functions)[j])->getAddress(this);
            if((adr>=func_adr) && 
                (adr<=(((*all_functions)[j])->size()+func_adr))){
                return((*all_functions)[j]);
            }
        }
    }
    return(0);
}
        

        
// findModule: returns the module associated with mod_name 
// this routine checks both the a.out image and any shared object
// images for this resource
// if check_excluded is true it checks to see if the module is excluded
// and if it is it returns 0.  If check_excluded is false it doesn't check
module *process::findModule(const string &mod_name,bool check_excluded){

    // KLUDGE: first search any shared libraries for the module name 
    //  (there is only one module in each shared library, and that 
    //  is the library name)
    if(dynamiclinking && shared_objects){
        for(u_int j=0; j < shared_objects->size(); j++){
            module *next = ((*shared_objects)[j])->findModule(mod_name,
                              check_excluded);
            if(next) {
                return(next);
            }
        }
    }

    // check a.out for function symbol
    //  Note that symbols is data member of type image* (comment says
    //  "information related to the process"....
    return(symbols->findModule(mod_name));
}

// getSymbolInfo:  get symbol info of symbol associated with name n
// this routine starts looking a.out for symbol and then in shared objects
// baseAddr is set to the base address of the object containing the symbol.
// This function appears to return symbol info even if module/function
// is excluded.  In extending excludes to statically linked executable,
// we preserve these semantics....
bool process::getSymbolInfo(const string &name, Symbol &info, Address &baseAddr) const {
    
    // first check a.out for symbol
    if(symbols->symbol_info(name,info))
      return getBaseAddress(symbols, baseAddr);

    // next check shared objects
    if(dynamiclinking && shared_objects) {
      for(u_int j=0; j < shared_objects->size(); j++) {
            if(((*shared_objects)[j])->getSymbolInfo(name,info))
                return getBaseAddress(((*shared_objects)[j])->getImage(), baseAddr); 
      }
    }

    return false;
}


// getAllFunctions: returns a vector of all functions defined in the
// a.out and in the shared objects
// TODO: what to do about duplicate function names?
vector<function_base *> *process::getAllFunctions(){

    // if this list has already been created, return it
    if(all_functions) 
        return all_functions;

    // else create the list of all functions
    all_functions = new vector<function_base *>;
    const vector<function_base *> &blah = 
                    (vector<function_base *> &)(symbols->getAllFunctions());
    *all_functions += blah;

    if(dynamiclinking && shared_objects){
        for(u_int j=0; j < shared_objects->size(); j++){
           vector<function_base *> *funcs = (vector<function_base *> *) 
                        (((*shared_objects)[j])->getAllFunctions());
           if(funcs){
               *all_functions += *funcs; 
           }
        }
    }
    return all_functions;
}
      
// getAllModules: returns a vector of all modules defined in the
// a.out and in the shared objects
// Includes "excluded" modules....
vector<module *> *process::getAllModules(){

    // if the list of all modules has already been created, the return it
    if(all_modules) return all_modules;

    // else create the list of all modules
    all_modules = new vector<module *>;
    *all_modules += *((vector<module *> *)(&(symbols->getAllModules())));

    if(dynamiclinking && shared_objects){
        for(u_int j=0; j < shared_objects->size(); j++){
           vector<module *> *mods = 
                (vector<module *> *)(((*shared_objects)[j])->getModules());
           if(mods) {
               *all_modules += *mods; 
           }
    } } 
    return all_modules;
}

#ifndef BPATCH_LIBRARY
// getIncludedFunctions: returns a vector of all functions defined in the
// a.out and in the shared objects
// TODO: what to do about duplicate function names?
vector<function_base *> *process::getIncludedFunctions(){
    //cerr << "process " << programName << " :: getIncludedFunctions() called"
    //   << endl;
    // if this list has already been created, return it
    if(some_functions) 
        return some_functions;

    // else create the list of all functions
    some_functions = new vector<function_base *>;
    const vector<function_base *> &incl_funcs = 
        (vector<function_base *> &)(symbols->getIncludedFunctions());
        *some_functions += incl_funcs;

    //cerr << " (process::getIncludedFunctions), about to add incl_funcs to some_functions, incl_funcs = " << endl;
    //print_func_vector_by_pretty_name(string(">>>"), &incl_funcs);

    if(dynamiclinking && shared_objects){
        for(u_int j=0; j < shared_objects->size(); j++){
            if(((*shared_objects)[j])->includeFunctions()){
                // kludge: can't assign a vector<derived_class *> to 
                // a vector<base_class *> so recast
                vector<function_base *> *funcs = (vector<function_base *> *)
                        (((*shared_objects)[j])->getSomeFunctions());
                if(funcs) { 
                    *some_functions += (*funcs); 
                } 
            } 
    } } 

    //cerr << " (process::getIncludedFunctions()) about to return fucntion list : ";
    //print_func_vector_by_pretty_name(string("  "), some_functions);

    return some_functions;
}
#endif /* BPATCH_LIBRARY */

// getIncludedModules: returns a vector of all modules defined in the
// a.out and in the shared objects that are included as specified in
// the mdl
vector<module *> *process::getIncludedModules(){

    //cerr << "process::getIncludedModules called" << endl;

    // if the list of all modules has already been created, the return it
    if(some_modules) {
        //cerr << "some_modules already created, returning it:" << endl;
        //print_module_vector_by_short_name(string("  "), (vector<pdmodule*>*)some_modules);
        return some_modules;
    }

    // else create the list of all modules
    some_modules = new vector<module *>;
    *some_modules += *((vector<module *> *)(&(symbols->getIncludedModules())));

    if(dynamiclinking && shared_objects){
        for(u_int j=0; j < shared_objects->size(); j++){
            if(((*shared_objects)[j])->includeFunctions()){
               vector<module *> *mods = (vector<module *> *) 
                        (((*shared_objects)[j])->getModules());
               if(mods) {
                   *some_modules += *mods; 
               }
           }
    } } 

    //cerr << "some_modules newlyu created, returning it:" << endl;
    //print_module_vector_by_short_name(string("  "),
    //    (vector<pdmodule*>*)some_modules);
    return some_modules;
}
      
// getBaseAddress: sets baseAddress to the base address of the 
// image corresponding to which.  It returns true  if image is mapped
// in processes address space, otherwise it returns 0
bool process::getBaseAddress(const image *which,u_int &baseAddress) const {

  if((u_int)(symbols) == (u_int)(which)){
      baseAddress = 0; 
      return true;
  }
  else if (shared_objects) {  
      // find shared object corr. to this image and compute correct address
      for(unsigned j=0; j < shared_objects->size(); j++){ 
          if(((*shared_objects)[j])->isMapped()){
            if(((*shared_objects)[j])->getImageId() == (u_int)which) { 
              baseAddress = ((*shared_objects)[j])->getBaseAddress();
              return true;
          } }
      }
  }
  return false;
}

// findSignalHandler: if signal_handler is 0, then it checks all images
// associtated with this process for the signal handler function.
// Otherwise, the signal handler function has already been found
void process::findSignalHandler(){

    if(SIGNAL_HANDLER == 0) return;
    if(!signal_handler) { 
        // first check a.out for signal handler function
        signal_handler = symbols->findOneFunction(SIGNAL_HANDLER);

        // search any shared libraries for signal handler function
        if(!signal_handler && dynamiclinking && shared_objects) { 
            for(u_int j=0;(j < shared_objects->size()) && !signal_handler; j++){
                signal_handler = 
                      ((*shared_objects)[j])->findOneFunction(SIGNAL_HANDLER,false);
        } }
    }
}


bool process::findInternalSymbol(const string &name, bool warn, internalSym &ret_sym) const {
     // On some platforms, internal symbols may be dynamic linked
     // so we search both the a.out and the shared objects
     Symbol sym;
     Address baseAddr;
     static const string underscore = "_";
     if (getSymbolInfo(name, sym, baseAddr)
         || getSymbolInfo(underscore+name, sym, baseAddr)) {
        ret_sym = internalSym(sym.addr()+baseAddr, name);
        return true;
     }
     if (warn) {
        string msg;
        msg = string("Unable to find symbol: ") + name;
        statusLine(msg.string_of());
        showErrorCallback(28, msg);
     }
     return false;
}

Address process::findInternalAddress(const string &name, bool warn, bool &err) const {
     // On some platforms, internal symbols may be dynamic linked
     // so we search both the a.out and the shared objects
     Symbol sym;
     Address baseAddr;
     static const string underscore = "_";
#if defined(USES_LIBDYNINSTRT_SO)
     // we use "dlopen" because we took out the leading "_"'s from the name
     if (name==string("dlopen")) {
       // if the function is dlopen, we use the address in ld.so.1 directly
       baseAddr = get_dlopen_addr();
       if (baseAddr != NULL) {
         err = false;
         return baseAddr;
       } else {
         err = true;
         return 0;
       }
     }
#endif
     if (getSymbolInfo(name, sym, baseAddr)
         || getSymbolInfo(underscore+name, sym, baseAddr)) {
        err = false;
        return sym.addr()+baseAddr;
     }
     if (warn) {
        string msg;
        msg = string("Unable to find symbol: ") + name;
        statusLine(msg.string_of());
        showErrorCallback(28, msg);
     }
     err = true;
     return 0;
}



bool process::continueProc() {
  if (status_ == exited) return false;

  if (status_ != stopped && status_ != neonatal) {
    showErrorCallback(38, "Internal paradynd error in process::continueProc");
    return false;
  }

  bool res = continueProc_();

  if (!res) {
    showErrorCallback(38, "System error: can't continue process");
    return false;
  }
  status_ = running;
  return true;
}

bool process::detach(const bool paused) {
  if (paused) {
    logLine("detach: pause not implemented\n"); // why not? --ari
  }
  bool res = detach_();
  if (!res) {
    // process may have exited
    return false;
  }
  return true;
}

#ifdef BPATCH_LIBRARY
// XXX Eventually detach() above should go away and this should be
//     renamed detach()
/* process::API_detach: detach from the application, leaving all
   instrumentation place.  Returns true upon success and false upon failure.
   Fails if the application is not stopped when the call is made.  The
   parameter "cont" indicates whether or not the application should be made
   running or not as a consquence of detaching (true indicates that it should
   be run).
*/
bool process::API_detach(const bool cont)
{
  if (status() != neonatal && status() != stopped)
      return false;

  return API_detach_(cont);
}
#endif

/* process::handleExec: called when a process successfully exec's.
   Parse the new image, disable metric instances on the old image, create a
   new (and blank) shm segment.  The process hasn't yet bootstrapped, so we
   mustn't try to enable anything...
*/
void process::handleExec() {
    // NOTE: for shm sampling, the shm segment has been removed, so we
    //       mustn't try to disable any dataReqNodes in the standard way...

   // since the exec syscall has run, we're not ready to enable any m/f pairs or
   // sample anything.
   // So we set hasBootstrapped to false until we run DYNINSTinit again.
   hasBootstrapped = false;

    // all instrumentation that was inserted in this process is gone.
    // set exited here so that the disables won't try to write to process
    status_ = exited; 
   
    // Clean up state from old exec: all dynamic linking stuff, all lists 
    // of functions and modules from old executable

    // can't delete dynamic linking stuff here, because parent process
    // could still have pointers
    dynamiclinking = false;
    dyn = 0; // AHEM.  LEAKED MEMORY!  not if the parent still has a pointer
             // to this dynamic_linking object.
    dyn = new dynamic_linking;
    if(shared_objects){
        for(u_int j=0; j< shared_objects->size(); j++){
            delete (*shared_objects)[j];
        }
        delete shared_objects;
        shared_objects = 0;
    }

    // TODO: when can pdFunction's be deleted???  definitely not here.
    delete some_modules;
    delete some_functions;
    delete all_functions;
    delete all_modules;
    some_modules = 0;
    some_functions = 0;
    all_functions = 0;
    all_modules = 0;
    signal_handler = 0;
    trampTableItems = 0;
    memset(trampTable, 0, sizeof(trampTable));
    baseMap.clear();
    cleanInstFromActivePoints(this);

#if defined(rs6000_ibm_aix3_2) || defined(rs6000_ibm_aix4_1)
    // must call establishBaseAddrs before parsing the new image,
    // but doesn't need to wait for trap, since we already got the trap.
    bool establishBaseAddrs(int pid, int &status, bool waitForTrap);
    int status;
    establishBaseAddrs(getPid(), status, false);
#endif

    image *img = image::parseImage(execFilePath);

    if (!img) {
       // For better error reporting, two failure return values would be useful
       // One for simple error like because-file-not-found
       // Another for serious errors like found-but-parsing-failed (internal error;
       //    please report to paradyn@cs.wisc.edu)

       string msg = string("Unable to parse image: ") + execFilePath;
       showErrorCallback(68, msg.string_of());
       OS::osKill(pid);
          // err..what if we had attached?  Wouldn't a detach be appropriate in this case?
       return;
    }

    // delete proc->symbols ???  No, the image can be shared by more
    // than one process...images and instPoints can not be deleted...TODO
    // add some sort of reference count to these classes so that they can
    // be deleted
    symbols = img; // AHEM!  LEAKED MEMORY!!! ...not if parent has ptr to 
                   // previous image

    // see if new image contains the signal handler function
    this->findSignalHandler();

    // initInferiorHeap can only be called after symbols is set!
#if !defined(USES_LIBDYNINSTRT_SO)
    initInferiorHeap(false);
    if (splitHeaps) {
      initInferiorHeap(true);  // create separate text heap
    }
#endif

    /* update process status */
    reachedFirstBreak = false;
       // we haven't yet seen initial SIGTRAP for this proc (is this right?)
    reachedVeryFirstTrap = false;

    status_ = stopped; // was 'exited'

   // TODO: We should remove (code) items from the where axis, if the exec'd process
   // was the only one who had them.

   // the exec'd process has the same fd's as the pre-exec, so we don't need
   // to re-initialize traceLink or ioLink (is this right???)

   // we don't need to re-attach after an exec (is this right???)
 
#ifdef SHM_SAMPLING
   inferiorHeapMgr.handleExec();
      // reuses the shm seg (paradynd's already attached to it); resets applic-attached-
      // at to NULL.  Quite similar to the (non-fork) ctor, really.

   theSuperTable.handleExec();
#endif

   inExec = false;
   execed_ = true;
}

/* 
   process::cleanUpInstrumentation called when paradynd catch
   a SIGTRAP to find out if there's any previous unfinished instrumentation
   requests 
*/
bool process::cleanUpInstrumentation(bool wasRunning) {
    // Try to process an item off of the waiting list 'instWlist'.
    // If something was found & processed, then true will be returned.
    // Additionally, if true is returned, the process will be continued
    // if 'wasRunning' is true.
    // But if false is returned, then there should be no side effects: noone
    // should be continued, nothing removed from 'instWList', no change
    // to this->status_, and so on (this is important to avoid bugs).

    assert(status_ == stopped); // since we're always called after a SIGTRAP
    Frame frame(this);
    Address pc = frame.getPC();

    // Go thru the instWList to find out the ones to be deleted 
    bool done = false;
    u_int i=0;
    bool found = false;
    while(!done){
        //process *p = (instWList[i])->which_proc;
        if(((instWList[i])->pc_ == pc) && ((instWList[i])->which_proc == this)){
            (instWList[i])->cleanUp(this,pc);
            u_int last = instWList.size()-1;
            delete (instWList[i]);
            instWList[i] = instWList[last];
            instWList.resize(last);
            found = true;
        }
        else {
            i++;
        }
        if(i >= instWList.size()) done = true;
    }
    if(found && wasRunning) continueProc();
    return found;
}

void process::cleanRPCreadyToLaunch(int mid) 
{
  vectorSet<inferiorRPCtoDo> tmpRPCsWaitingToStart;
  while (RPCsWaitingToStart.size() > 0) {
    inferiorRPCtoDo currElem = RPCsWaitingToStart.removeOne();
    if (currElem.mid == mid)
      break;
    else
      tmpRPCsWaitingToStart += currElem;
  }
  // reconstruct RPCsWaitingToStart
  while (tmpRPCsWaitingToStart.size() > 0) {
    RPCsWaitingToStart += tmpRPCsWaitingToStart.removeOne();
  }
}

void process::postRPCtoDo(AstNode *action, bool noCost,
                          void (*callbackFunc)(process *, void *, unsigned),
                          void *userData,
                          int mid) {
   // posts an RPC, but does NOT make any effort to launch it.
   inferiorRPCtoDo theStruct;
   theStruct.action = action;
   theStruct.noCost = noCost;
   theStruct.callbackFunc = callbackFunc;
   theStruct.userData = userData;
   theStruct.mid = mid;

   RPCsWaitingToStart += theStruct;
}

bool process::existsRPCreadyToLaunch() const {
   if (currRunningRPCs.empty() && !RPCsWaitingToStart.empty())
      return true;
   return false;
}

bool process::existsRPCinProgress() const {
   return (!currRunningRPCs.empty());
}

bool process::launchRPCifAppropriate(bool wasRunning, bool finishingSysCall) {
   // asynchronously launches an inferiorRPC iff RPCsWaitingToStart.size() > 0 AND
   // if currRunningRPCs.size()==0 (the latter for safety)

   if (!currRunningRPCs.empty())
      // an RPC is currently executing, so it's not safe to launch a new one.
      return false;

   if (RPCsWaitingToStart.empty())
      // duh, no RPC is waiting to run, so there's nothing to do.
      return false;

   if (status_ == exited)
      return false;

   if (status_ == neonatal)
      // not sure if this should be some kind of error...is the inferior ready
      // to execute inferior RPCs??? For now, we'll allow it.
      ; 

   /* ****************************************************** */

   // Steps to take (on sparc, at least)
   // 1) pause the process and wait for it to stop
   // 2) Get a copy of the registers...store them away
   // 3) create temp trampoline: save, action, restore, trap, illegal
   //    (the illegal is just ot be sure that the trap never returns)
   // 4) set the PC and nPC regs to addr of temp tramp
   // 5) Continue the process...go back to the main loop (SIGTRAP will
   //    eventually be delivered)

   // When SIGTRAP is received,
   // 5) verify that PC is the location of the TRAP instr in the temp tramp
   // 6) free temp trampoline
   // 7) SETREGS to restore all regs, including PC and nPC.
   // 8) continue inferior, if the inferior had been running when we had
   //    paused it in step 1, above.

   if (!finishingSysCall && RPCs_waiting_for_syscall_to_complete) {
      assert(executingSystemCall());
      return false;
   }

   if (!pause()) {
      cerr << "launchRPCifAppropriate failed because pause failed" << endl;
      return false;
   }

   // If we're in the middle of a system call, then it's not safe to launch
   // an inferiorRPC on most platforms.  On some platforms, it's safe, but the
   // actual launching won't take place until the system call finishes.  In such
   // cases it's a good idea to set a breakpoint for when the sys call exits
   // (using /proc PIOCSEXIT), and launch the inferiorRPC then.
   if (!finishingSysCall && executingSystemCall()) {
      if (RPCs_waiting_for_syscall_to_complete) {
         inferiorrpc_cerr << "launchRPCifAppropriate: still waiting for syscall to "
                          << "complete" << endl;
         if (wasRunning) {
            inferiorrpc_cerr << "launchRPC: continuing so syscall may complete" << endl;
            (void)continueProc();
         }
         else
            inferiorrpc_cerr << "launchRPC: sorry not continuing (problem?)" << endl;
         return false;
      }

      // don't do the inferior rpc until the system call finishes.  Determine
      // which system call is in the way, and set a breakpoint at its exit point
      // so we know when it's safe to launch the RPC.  Platform-specific
      // details:

      inferiorrpc_cerr << "launchRPCifAppropriate: within a system call" << endl;

      if (!set_breakpoint_for_syscall_completion()) {
         // sorry, this platform can't set a breakpoint at the system call
         // completion point.  In such a case, we keep polling executingSystemCall()
         // inefficiently.
         if (wasRunning)
            (void)continueProc();

         inferiorrpc_cerr << "launchRPCifAppropriate: couldn't set bkpt for "
                          << "syscall completion; will just poll." << endl;
         return false;
      }

      inferiorrpc_cerr << "launchRPCifAppropriate: set bkpt for syscall completion"
                       << endl;

      // a SIGTRAP will get delivered when the RPC is ready to go.  Until then,
      // mark the rpc as deferred.  Setting this flag prevents us from executing
      // this code too many times.

      RPCs_waiting_for_syscall_to_complete = true;

      if (wasRunning)
         (void)continueProc();

      return false;
   }

   // Okay, we're not in the middle of a system call, so we can fire off the rpc now!
   if (RPCs_waiting_for_syscall_to_complete)
      // not any more
      RPCs_waiting_for_syscall_to_complete = false;

   void *theSavedRegs = getRegisters(); // machine-specific implementation
      // result is allocated via new[]; we'll delete[] it later.
      // return value of NULL indicates total failure.
      // return value of (void *)-1 indicates that the state of the machine isn't quite
      //    ready for an inferiorRPC, and that we should try again 'later'.  In
      //    particular, we must handle the (void *)-1 case very gracefully (i.e., leave
      //    the vrble 'RPCsWaitingToStart' untouched).

   if (theSavedRegs == (void *)-1) {
      // cerr << "launchRPCifAppropriate: deferring" << endl;
      if (wasRunning)
         (void)continueProc();
      return false;
   }

   if (theSavedRegs == NULL) {
      cerr << "launchRPCifAppropriate failed because getRegisters() failed" << endl;
      if (wasRunning)
         (void)continueProc();
      return false;
   }

   if (!wasRunning)
     inferiorrpc_cerr << "NOTE: launchIfAppropriate: wasRunning==false!!" << endl;

   inferiorRPCtoDo todo = RPCsWaitingToStart.removeOne();
      // note: this line should always be below the test for (void*)-1, thus
      // leaving 'RPCsWaitingToStart' alone in that case.

   inferiorRPCinProgress inProgStruct;
   inProgStruct.callbackFunc = todo.callbackFunc;
   inProgStruct.userData = todo.userData;
   inProgStruct.savedRegs = theSavedRegs;
   inProgStruct.wasRunning = wasRunning || finishingSysCall;
      // If finishing up a system call, current state is paused, but we want to
      // set wasRunning to true so that it'll continue when the inferiorRPC
      // completes.  Sorry for the kludge.
   unsigned tempTrampBase = createRPCtempTramp(todo.action,
                                               todo.noCost,
                                               inProgStruct.callbackFunc != NULL,
                                               inProgStruct.breakAddr,
                                               inProgStruct.stopForResultAddr,
                                               inProgStruct.justAfter_stopForResultAddr,
                                               inProgStruct.resultRegister);
      // the last 4 args are written to

   if (tempTrampBase == 0) {
      cerr << "launchRPCifAppropriate failed because createRPCtempTramp failed" << endl;
      if (wasRunning)
         (void)continueProc();
      return false;
   }

   assert(tempTrampBase);

   inProgStruct.firstInstrAddr = tempTrampBase;

   assert(currRunningRPCs.empty()); // since it's unsafe to run > 1 at a time
   currRunningRPCs += inProgStruct;

   inferiorrpc_cerr << "Changing pc and exec.." << endl;

   // change the PC and nPC registers to the addr of the temp tramp
   if (!changePC(tempTrampBase, theSavedRegs)) {
      cerr << "launchRPCifAppropriate failed because changePC() failed" << endl;
      if (wasRunning)
         (void)continueProc();
      return false;
   }

   if (!continueProc()) {
      cerr << "launchRPCifAppropriate: continueProc() failed" << endl;
      return false;
   }

   inferiorrpc_cerr << "inferiorRPC should be running now" << endl;

   return true; // success
}

unsigned process::createRPCtempTramp(AstNode *action,
                                     bool noCost,
                                     bool shouldStopForResult,
                                     unsigned &breakAddr,
                                     unsigned &stopForResultAddr,
                                     unsigned &justAfter_stopForResultAddr,
                                     reg &resultReg) {

   // Returns addr of temp tramp, which was allocated in the inferior heap.
   // You must free it yourself when done.

   // Note how this is, in many ways, a greatly simplified version of
   // addInstFunc().

   // Temp tramp structure: save; code; restore; trap; illegal
   // the illegal is just to make sure that the trap never returns
   // note that we may not need to save and restore anything, since we've
   // already done a GETREGS and we'll restore with a SETREGS, right?

   unsigned char insnBuffer[4096];

   initTramps(); // initializes "regSpace", but only the 1st time it gets called...
   extern registerSpace *regSpace;
   regSpace->resetSpace();

   unsigned count = 0;

   // The following is implemented in an arch-specific source file...
   if (!emitInferiorRPCheader(insnBuffer, count)) {
      // a fancy dialog box is probably called for here...
      cerr << "createRPCtempTramp failed because emitInferiorRPCheader failed." << endl;
      return NULL;
   }

#if defined(SHM_SAMPLING) && defined(MT_THREAD)
   generateMTpreamble((char*)insnBuffer,count,this);
#endif

   resultReg = action->generateCode(this, regSpace,
                                    (char*)insnBuffer,
                                    count, noCost);

   if (!shouldStopForResult) {
      regSpace->freeRegister(resultReg);
   }
   else
      ; // in this case, we'll call freeRegister() the inferior rpc completes

   // Now, the trailer (restore, TRAP, illegal)
   // (the following is implemented in an arch-specific source file...)

   unsigned breakOffset, stopForResultOffset, justAfter_stopForResultOffset;
   if (!emitInferiorRPCtrailer(insnBuffer, count,
                               breakOffset, shouldStopForResult, stopForResultOffset,
                               justAfter_stopForResultOffset)) {
      // last 4 args except shouldStopForResult are modified by the call
      cerr << "createRPCtempTramp failed because emitInferiorRPCtrailer failed." << endl;
      return NULL;
   }

   unsigned tempTrampBase = inferiorMalloc(this, count, textHeap);
   assert(tempTrampBase);


   breakAddr                   = tempTrampBase + breakOffset;
   if (shouldStopForResult) {
      stopForResultAddr           = tempTrampBase + stopForResultOffset;
      justAfter_stopForResultAddr = tempTrampBase + justAfter_stopForResultOffset;
   }
   else {
      stopForResultAddr = justAfter_stopForResultAddr = 0;
   }

   inferiorrpc_cerr << "createRPCtempTramp: temp tramp base=" << (void*)tempTrampBase
                    << ", stopForResultAddr=" << (void*)stopForResultAddr
                    << ", justAfter_stopForResultAddr=" << (void*)justAfter_stopForResultAddr
                    << ", breakAddr=" << (void*)breakAddr
                    << ", count=" << count << " so end addr="
                    << (void*)(tempTrampBase + count - 1) << endl;

#if defined(MT_DEBUG)
   sprintf(errorLine,"********>>>>> tempTrampBase = 0x%x\n",tempTrampBase);
   logLine(errorLine);
#endif

   /* Now, write to the tempTramp, in the inferior addr's data space
      (all tramps are allocated in data space) */
   if (!writeDataSpace((void*)tempTrampBase, count, insnBuffer)) {
      // should put up a nice error dialog window
      cerr << "createRPCtempTramp failed because writeDataSpace failed" << endl;
      return NULL;
   }

   extern int trampBytes; // stats.h
   trampBytes += count;

   return tempTrampBase;
}

bool process::handleTrapIfDueToRPC() {
   // get curr PC register (can assume process is stopped), search for it in
   // 'currRunningRPCs'.  If found, restore regs, do callback, delete tramp, and
   // return true.  Returns false if not processed.

   assert(status_ == stopped); // a TRAP should always stop a process (duh)
   
   if (currRunningRPCs.empty())
      return false; // no chance of a match

   assert(currRunningRPCs.size() == 1);
      // it's unsafe to have > 1 RPCs going on at a time within a single process

   // Okay, time to do a stack trace.
   // If we determine that the PC of any level of the back trace
   // equals the current running RPC's stopForResultAddr or breakAddr,
   // then we have a match.  Note: since all platforms currently
   // inline their trap/ill instruction instead of make a fn call e.g. to
   // DYNINSTbreakPoint(), we can probably get rid of the stack walk.

   int match_type = 0; // 1 --> stop for result, 2 --> really done
   Frame theFrame(this);
   while (true) {
      // do we have a match?
      const int framePC = theFrame.getPC();
      if ((unsigned)framePC == currRunningRPCs[0].breakAddr) {
         // we've got a match!
         match_type = 2;
         break;
      }
      else if (currRunningRPCs[0].callbackFunc != NULL &&
               ((unsigned)framePC == currRunningRPCs[0].stopForResultAddr)) {
         match_type = 1;
         break;
      }

      if (theFrame.isLastFrame())
         // well, we've gone as far as we can, with no match.
         return false;

      // else, backtrace 1 more level
      theFrame = theFrame.getPreviousStackFrameInfo(this);
   }

   assert(match_type == 1 || match_type == 2);

   inferiorRPCinProgress &theStruct = currRunningRPCs[0];

   if (match_type == 1) {
      // We have stopped in order to grab the result.  Grab it and write
      // to theStruct.resultValue, for use when we get match_type==2.

      inferiorrpc_cerr << "Welcome to handleTrapIfDueToRPC match type 1" << endl;

      assert(theStruct.callbackFunc != NULL);
        // must be a callback to ever see this match_type

      const unsigned returnValue = read_inferiorRPC_result_register(theStruct.resultRegister);

      extern registerSpace *regSpace;
      regSpace->freeRegister(theStruct.resultRegister);

      theStruct.resultValue = returnValue;

      // we don't remove the RPC from 'currRunningRPCs', since it's not yet done.
      // we continue the process...but not quite at the PC where we left off, since
      // that will just re-do the trap!  Instead, we need to continue at the location
      // of the next instruction.
      if (!changePC(theStruct.justAfter_stopForResultAddr))
         assert(false);

      if (!continueProc())
         cerr << "RPC getting result: continueProc failed" << endl;

      return true;
   }

   inferiorrpc_cerr << "Welcome to handleTrapIfDueToRPC match type 2" << endl;

   assert(match_type == 2);

   // step 1) restore registers:
   if (!restoreRegisters(theStruct.savedRegs)) {
      cerr << "handleTrapIfDueToRPC failed because restoreRegisters failed" << endl;
      assert(false);
   }
   currRunningRPCs.removeByIndex(0);

   if (currRunningRPCs.empty() && deferredContinueProc) {
     // We have a pending continueProc that we had to delay because
     // there was an inferior RPC in progress at that time, but now
     // we are ready to execute it - naim
     deferredContinueProc=false;
     if (continueProc()) statusLine("application running");
   }

   delete [] theStruct.savedRegs;

   // step 2) delete temp tramp
   vector< vector<unsigned> > pointsToCheck;
      // blank on purpose; deletion is safe to take place even right now
   inferiorFree(this, theStruct.firstInstrAddr, textHeap, pointsToCheck);

   // step 3) continue process, if appropriate
   if (theStruct.wasRunning) {
      inferiorrpc_cerr << "end of rpc -- continuing process, since it had been running" << endl;

      if (!continueProc()) {
         cerr << "RPC completion: continueProc failed" << endl;
      }
   }
   else
      inferiorrpc_cerr << "end of rpc -- leaving process paused, since it wasn't running before" << endl;

   // step 4) invoke user callback, if any
   // note: I feel it's important to do the user callback last, since it
   // may perform arbitrary actions (such as making igen calls) which can lead
   // to re-actions (such as receiving igen calls) that can alter the process
   // (e.g. continuing it).  So clearly we need to restore registers, change the
   // PC, etc. BEFORE any such thing might happen.  Hence we do the callback last.
   // I think the only potential controversy is whether we should do the callback
   // before step 3, above.

   inferiorrpc_cerr << "handleTrapIfDueToRPC match type 2 -- about to do callbackFunc, if any" << endl;

   if (theStruct.callbackFunc)
      theStruct.callbackFunc(this, theStruct.userData, theStruct.resultValue);

   inferiorrpc_cerr << "handleTrapIfDueToRPC match type 2 -- done with callbackFunc, if any" << endl;

   return true;
}

void process::installBootstrapInst() {
   // instrument main to call DYNINSTinit().  Don't use the shm seg for any
   // temp tramp space, since we can't assume that it's been intialized yet.
   // We build an ast saying: "call DYNINSTinit() with args
   // key_base, nbytes, paradynd_pid"

#ifdef BPATCH_LIBRARY
   vector<AstNode *> the_args(2);

   the_args[0] = new AstNode(AstNode::Constant, (void*)1);
   the_args[1] = new AstNode(AstNode::Constant, (void*)getpid());

   AstNode *ast = new AstNode("DYNINSTinit", the_args);
#else
   vector<AstNode *> the_args(3);

   // 2 dummy args when not shm sampling (just don't use -1, which is reserved
   // for fork)
   unsigned numBytes = 0;
   
#ifdef SHM_SAMPLING
   key_t theKey   = getShmKeyUsed();
   numBytes = getShmHeapTotalNumBytes();
#else
   int theKey = 0;
#endif

#ifdef SHM_SAMPLING_DEBUG
   cerr << "paradynd inst.C: about to call DYNINSTinit() with key=" << theKey
        << " and #bytes=" << numBytes << endl;
#endif

   the_args[0] = new AstNode(AstNode::Constant, (void*)theKey);
   the_args[1] = new AstNode(AstNode::Constant, (void*)numBytes);
#ifdef BPATCH_LIBRARY
   // Unused by the dyninstAPI library
   the_args[2] = new AstNode(AstNode::Constant, (void *)0);
#else
   the_args[2] = new AstNode(AstNode::Constant, (void*)traceConnectInfo);
#endif

   AstNode *ast = new AstNode("DYNINSTinit", the_args);
   for (unsigned j=0; j<the_args.size(); j++) {
       removeAst(the_args[j]);
   }
#endif /* BPATCH_LIBRARY */

   function_base *func = getMainFunction();
   assert(func);

   instPoint *func_entry = (instPoint *)func->funcEntry(this);
   addInstFunc(this, func_entry, ast, callPreInsn,
               orderFirstAtPoint,
               true // true --> don't try to have tramp code update the cost
               );
   // returns an "instInstance", which we ignore (but should we?)
   removeAst(ast);
}

void process::installInstrRequests(const vector<instMapping*> &requests) {
   for (unsigned lcv=0; lcv < requests.size(); lcv++) {
      instMapping *req = requests[lcv];

      function_base *func = findOneFunction(req->func);
      if (!func)
         continue;  // probably should have a flag telling us whether errors should
                    // be silently handled or not

      AstNode *ast;
      if (req->where & FUNC_ARG) {
        // ast = new AstNode(req->inst, req->arg);
        ast = new AstNode(req->inst, req->args);
      } else {
        AstNode *tmp = new AstNode(AstNode::Constant, (void*)0);
        ast = new AstNode(req->inst, tmp);
        removeAst(tmp);
      }

      if (req->where & FUNC_EXIT) {
         const vector<instPoint*> func_rets = func->funcExits(this);
         for (unsigned j=0; j < func_rets.size(); j++)
            (void)addInstFunc(this, func_rets[j], ast,
                              callPreInsn, orderLastAtPoint, false);

      }

      if (req->where & FUNC_ENTRY) {
         instPoint *func_entry = (instPoint *)func->funcEntry(this);
         (void)addInstFunc(this, func_entry, ast,
                           callPreInsn, orderLastAtPoint, false);

      }

      if (req->where & FUNC_CALL) {
         vector<instPoint*> func_calls = func->funcCalls(this);
         if (func_calls.size() == 0)
            continue;

         for (unsigned j=0; j < func_calls.size(); j++)
            (void)addInstFunc(this, func_calls[j], ast,
                              callPreInsn, orderLastAtPoint, false);

      }

      removeAst(ast);
   }
}

bool process::extractBootstrapStruct(DYNINST_bootstrapStruct *bs_record) {
   const string vrbleName = "DYNINST_bootstrap_info";
   internalSym sym;

   bool flag = findInternalSymbol(vrbleName, true, sym);
   assert(flag);

   Address symAddr = sym.getAddr();

   if (!readDataSpace((const void*)symAddr, sizeof(*bs_record), bs_record, true)) {
      cerr << "extractBootstrapStruct failed because readDataSpace failed" << endl;
      return false;
   }

   return true;
}

bool process::handleStopDueToExecEntry() {
   // returns true iff we are processing a stop due to the entry point of exec
   // The exec hasn't yet occurred.

   assert(status_ == stopped);

   DYNINST_bootstrapStruct bs_record;
   if (!extractBootstrapStruct(&bs_record))
      assert(false);

   if (bs_record.event != 4)
      return false;

   assert(getPid() == bs_record.pid);

   // for now, we just set aside the following information, to be used after the
   // exec actually happens (we'll get a SIGTRAP for that).
   assert(!inExec);
   inExec = true;
   execFilePath = string(bs_record.path);

   // the process was stopped...let's continue it so we can process the exec...
   assert(status_ == stopped);
   if (!continueProc())
      assert(false);

   // should we set status_ to neonatal now?  Nah, probably having the inExec flag
   // set is good enough...

   // shouldn't we be setting reachedFirstBreak to false???

   return true;
}

int process::procStopFromDYNINSTinit() {
   // possible return values:
   // 0 --> no, the stop wasn't the end of DYNINSTinit
   // 1 --> handled stop at end of DYNINSTinit, leaving program paused
   // 2 --> handled stop at end of DYNINSTinit...which had been invoked via
   //       inferiorRPC...so we've continued the process in order to let the
   //       inferiorRPC get its sigtrap.

   // Note that DYNINSTinit could have been run under several cases:
   // 1) the normal case     (detect by bs_record.event==1 && execed_ == false)
   // 2) called after a fork (detect by bs_record.event==2)
   // 3) called after an exec (detect by bs_record.event==1 and execed_ == true)
   // 4) called for an attach (detect by bs_record.event==3)
   // note that bs_record.event == 4 is reserved for "sending" a tr_exec "record".
   //
   // The exec case is tricky: we must loop thru all component mi's of this process
   // and decide now whether or not to carry them over to the new process.

   // if 0 is returned, there must be no side effects.

   assert(status_ == stopped);

   if (hasBootstrapped)
      return 0;

   DYNINST_bootstrapStruct bs_record;
   if (!extractBootstrapStruct(&bs_record))
      assert(false);

   // Read the structure; if event 0 then it's undefined! (not yet written)
   if (bs_record.event == 0)
      return 0;

   forkexec_cerr << "procStopFromDYNINSTinit pid " << getPid() << "; got rec" << endl;

   assert(bs_record.event == 1 || bs_record.event == 2 || bs_record.event==3);
   assert(bs_record.pid == getPid());

   if (bs_record.event != 3) {
      // we don't want to do this stuff (yet) when DYNINSTinit was run via attach...we
      // want to wait until the inferiorRPC (thru which DYNINSTinit is being run)
      // completes.
      handleCompletionOfDYNINSTinit(false);
      return 1;
   }
   else {
      if (!continueProc())
         assert(false);
      return 2;
   }
}

void process::handleCompletionOfDYNINSTinit(bool fromAttach) {
   // 'event' values: (1) DYNINSTinit was started normally via paradynd
   // or via exec, (2) called from fork, (3) called from attach.

   DYNINST_bootstrapStruct bs_record;
   if (!extractBootstrapStruct(&bs_record))
      assert(false);

   if (!fromAttach) // reset to 0 already if attaching, but other fields (attachedAtPtr) ok
      assert(bs_record.event == 1 || bs_record.event == 2 || bs_record.event==3);

   assert(bs_record.pid == getPid());

   // Note: the process isn't necessarily paused at this moment.  In particular,
   // if we had attached to the process, then it will be running even as we speak.
   // While we're parsing the shared libraries, we should pause.  So do that now.
   bool wasRunning;
   if (needToContinueAfterDYNINSTinit) 
     wasRunning = true;
   else 
     wasRunning = status_ == running;
   (void)pause();

   const bool calledFromFork   = (bs_record.event == 2);
   const bool calledFromExec   = (bs_record.event == 1 && execed_);
   const bool calledFromAttach = fromAttach || bs_record.event == 3;
   if (calledFromAttach)
      assert(createdViaAttach);

#ifdef SHM_SAMPLING
   if (!calledFromFork)
      registerInferiorAttachedSegs(bs_record.appl_attachedAtPtr);
#endif

   if (!calledFromFork)
      getObservedCostAddr();

   // handleStartProcess gets shared objects, so no need to do it again after a fork.
   // (question: do we need to do this after an exec???)
   if (!calledFromFork) {
      string str=string("PID=") + string(bs_record.pid) + ", calling handleStartProcess...";
      statusLine(str.string_of());

#if !defined(USES_LIBDYNINSTRT_SO)
      if (!handleStartProcess()) {
        // reads in shared libraries...can take a while
        logLine("WARNING: handleStartProcess failed\n");
      }
#endif

#ifndef BPATCH_LIBRARY
      // we decrement the batch mode here; it matches the bump-up in createProcess()
      tp->resourceBatchMode(false);
#endif

      str=string("PID=") + string(bs_record.pid) + ", installing default inst...";
      statusLine(str.string_of());

      extern vector<instMapping*> initialRequests; // init.C
      installInstrRequests(initialRequests);

      str=string("PID=") + string(bs_record.pid) + ", propagating mi's...";
      statusLine(str.string_of());

      forkexec_cerr << "procStopFromDYNINSTinit pid " << getPid() << "; about to propagate mi's" << endl;

#ifndef BPATCH_LIBRARY
      if (!calledFromExec) {
         // propagate any metric that is already enabled to the new process.
         // For a forked process, this isn't needed because handleFork() has its own
         // special propagation algorithm (it propagates every aggregate mi having the
         // parent as a component, except for aggregate mi's whose focus is specifically
         // refined to the parent).
         vector<metricDefinitionNode *> MIs = allMIs.values();
         for (unsigned j = 0; j < MIs.size(); j++) {
            MIs[j]->propagateToNewProcess(this);
            // change to a process:: method which takes in the metricDefinitionNode
         }