Merge stever@zizzer:/bk/m5 into isabel.reinhardt.house:/z/stever/bk/m5

--HG--
extra : convert_revision : b0f93bd35d767fd3a520a9fed70a71d40b0056db

18 files changed, 339 insertions, 171 deletions

diff --git a/arch/alpha/ev5.cc b/arch/alpha/ev5.cc
index e4a9a09e3..cc33f6890 100644
--- a/arch/alpha/ev5.cc
+++ b/arch/alpha/ev5.cc
@@ -44,9 +44,9 @@ AlphaISA::swap_palshadow(RegFile *regs, bool use_shadow)
 //  Machine dependent functions
 //
 void
-AlphaISA::init(void *mem, RegFile *regs)
+AlphaISA::initCPU(RegFile *regs)
 {
-    ipr_init(mem, regs);
+    initIPRs(regs);
 }
 
 void
@@ -91,7 +91,7 @@ const int AlphaISA::reg_redir[AlphaISA::NumIntRegs] = {
 //
 //
 void
-AlphaISA::ipr_init(void *mem, RegFile *regs)
+AlphaISA::initIPRs(RegFile *regs)
 {
     uint64_t *ipr = regs->ipr;
 
diff --git a/arch/alpha/fake_syscall.cc b/arch/alpha/fake_syscall.cc
index 584b07801..b2e42daaf 100644
--- a/arch/alpha/fake_syscall.cc
+++ b/arch/alpha/fake_syscall.cc
@@ -511,14 +511,14 @@ getsysinfoFunc(SyscallDesc *desc, int callnum, Process *process,
 
       case OSF::GSI_MAX_CPU: {
           TypedBufferArg<uint32_t> max_cpu(getArg(xc, 1));
-          *max_cpu = process->numCpus;
+          *max_cpu = process->numCpus();
           max_cpu.copyOut(xc->mem);
           return 1;
       }
 
       case OSF::GSI_CPUS_IN_BOX: {
           TypedBufferArg<uint32_t> cpus_in_box(getArg(xc, 1));
-          *cpus_in_box = process->numCpus;
+          *cpus_in_box = process->numCpus();
           cpus_in_box.copyOut(xc->mem);
           return 1;
       }
@@ -534,10 +534,10 @@ getsysinfoFunc(SyscallDesc *desc, int callnum, Process *process,
           TypedBufferArg<OSF::cpu_info> infop(getArg(xc, 1));
 
           infop->current_cpu = 0;
-          infop->cpus_in_box = process->numCpus;
+          infop->cpus_in_box = process->numCpus();
           infop->cpu_type = 57;
-          infop->ncpus = process->numCpus;
-          int cpumask = (1 << process->numCpus) - 1;
+          infop->ncpus = process->numCpus();
+          int cpumask = (1 << process->numCpus()) - 1;
           infop->cpus_present = infop->cpus_running = cpumask;
           infop->cpu_binding = 0;
           infop->cpu_ex_binding = 0;
@@ -743,7 +743,7 @@ tableFunc(SyscallDesc *desc, int callnum, Process *process,
           elp->si_hz = clk_hz;
           elp->si_phz = clk_hz;
           elp->si_boottime = seconds_since_epoch; // seconds since epoch?
-          elp->si_max_procs = process->numCpus;
+          elp->si_max_procs = process->numCpus();
           elp.copyOut(xc->mem);
           return 0;
       }
@@ -1141,20 +1141,20 @@ nxm_task_initFunc(SyscallDesc *desc, int callnum, Process *process,
     cur_addr += sizeof(OSF::nxm_config_info);
     // next comes the per-cpu state vector
     Addr slot_state_addr = cur_addr;
-    int slot_state_size = process->numCpus * sizeof(OSF::nxm_slot_state_t);
+    int slot_state_size = process->numCpus() * sizeof(OSF::nxm_slot_state_t);
     cur_addr += slot_state_size;
     // now the per-RAD state struct (we only support one RAD)
     cur_addr = 0x14000;	// bump up addr for alignment
     Addr rad_state_addr = cur_addr;
     int rad_state_size =
         (sizeof(OSF::nxm_shared)
-         + (process->numCpus-1) * sizeof(OSF::nxm_sched_state));
+         + (process->numCpus()-1) * sizeof(OSF::nxm_sched_state));
     cur_addr += rad_state_size;
 
     // now initialize a config_info struct and copy it out to user space
     TypedBufferArg<OSF::nxm_config_info> config(config_addr);
 
-    config->nxm_nslots_per_rad = process->numCpus;
+    config->nxm_nslots_per_rad = process->numCpus();
     config->nxm_nrads = 1;	// only one RAD in our system!
     config->nxm_slot_state = slot_state_addr;
     config->nxm_rad[0] = rad_state_addr;
@@ -1164,7 +1164,7 @@ nxm_task_initFunc(SyscallDesc *desc, int callnum, Process *process,
     // initialize the slot_state array and copy it out
     TypedBufferArg<OSF::nxm_slot_state_t> slot_state(slot_state_addr,
                                                      slot_state_size);
-    for (int i = 0; i < process->numCpus; ++i) {
+    for (int i = 0; i < process->numCpus(); ++i) {
         // CPU 0 is bound to the calling process; all others are available
         slot_state[i] = (i == 0) ? OSF::NXM_SLOT_BOUND : OSF::NXM_SLOT_AVAIL;
     }
@@ -1180,7 +1180,7 @@ nxm_task_initFunc(SyscallDesc *desc, int callnum, Process *process,
     rad_state->nxm_callback = attrp->nxm_callback;
     rad_state->nxm_version = attrp->nxm_version;
     rad_state->nxm_uniq_offset = attrp->nxm_uniq_offset;
-    for (int i = 0; i < process->numCpus; ++i) {
+    for (int i = 0; i < process->numCpus(); ++i) {
         OSF::nxm_sched_state *ssp = &rad_state->nxm_ss[i];
         ssp->nxm_u.sigmask = 0;
         ssp->nxm_u.sig = 0;
@@ -1250,7 +1250,7 @@ nxm_thread_createFunc(SyscallDesc *desc, int callnum, Process *process,
         abort();
     }
 
-    if (thread_index < 0 | thread_index > process->numCpus) {
+    if (thread_index < 0 | thread_index > process->numCpus()) {
         cerr << "nxm_thread_create: bad thread index " << thread_index
              << endl;
         abort();
@@ -1262,7 +1262,7 @@ nxm_thread_createFunc(SyscallDesc *desc, int callnum, Process *process,
     // back out again.
     int rad_state_size =
         (sizeof(OSF::nxm_shared) +
-         (process->numCpus-1) * sizeof(OSF::nxm_sched_state));
+         (process->numCpus()-1) * sizeof(OSF::nxm_sched_state));
 
     TypedBufferArg<OSF::nxm_shared> rad_state(0x14000,
                                               rad_state_size);
@@ -1294,7 +1294,7 @@ nxm_thread_createFunc(SyscallDesc *desc, int callnum, Process *process,
         rad_state.copyOut(xc->mem);
 
         Addr slot_state_addr = 0x12000 + sizeof(OSF::nxm_config_info);
-        int slot_state_size = process->numCpus * sizeof(OSF::nxm_slot_state_t);
+        int slot_state_size = process->numCpus() * sizeof(OSF::nxm_slot_state_t);
 
         TypedBufferArg<OSF::nxm_slot_state_t> slot_state(slot_state_addr,
                                                          slot_state_size);
@@ -1312,11 +1312,8 @@ nxm_thread_createFunc(SyscallDesc *desc, int callnum, Process *process,
         slot_state.copyOut(xc->mem);
 
         // Find a free simulator execution context.
-        list<ExecContext *> &ecList = process->execContexts;
-        list<ExecContext *>::iterator i = ecList.begin();
-        list<ExecContext *>::iterator end = ecList.end();
-        for (; i != end; ++i) {
-            ExecContext *xc = *i;
+        for (int i = 0; i < process->numCpus(); ++i) {
+            ExecContext *xc = process->execContexts[i];
 
             if (xc->status() == ExecContext::Unallocated) {
                 // inactive context... grab it
diff --git a/base/inifile.cc b/base/inifile.cc
index 7fd2f5568..6baf22850 100644
--- a/base/inifile.cc
+++ b/base/inifile.cc
@@ -217,8 +217,11 @@ bool
 IniFile::Section::add(const std::string &assignment)
 {
     string::size_type offset = assignment.find('=');
-    if (offset == string::npos)  // no '=' found
+    if (offset == string::npos) {
+        // no '=' found
+        cerr << "Can't parse .ini line " << assignment << endl;
         return false;
+    }
 
     // if "+=" rather than just "=" then append value
     bool append = (assignment[offset-1] == '+');
diff --git a/base/remote_gdb.hh b/base/remote_gdb.hh
index d8e09909a..62fd52856 100644
--- a/base/remote_gdb.hh
+++ b/base/remote_gdb.hh
@@ -90,6 +90,8 @@ class RemoteGDB
     RemoteGDB(System *system, ExecContext *context);
     ~RemoteGDB();
 
+    void replaceExecContext(ExecContext *xc) { context = xc; }
+
     void attach(int fd);
     void detach();
     bool isattached();
diff --git a/base/statistics.cc b/base/statistics.cc
index 6593451c4..cde98d861 100644
--- a/base/statistics.cc
+++ b/base/statistics.cc
@@ -846,7 +846,7 @@ DistDisplay(ostream &stream, const string &name, const string &desc,
     for (int i = 0; i < size; ++i) {
         if (flags & nozero && vec[i] == 0.0 ||
             flags & nonan && isnan(vec[i]))
-            return;
+            continue;
 
         _min = i * bucket_size + min;
         _pdf = vec[i] / total * 100.0;
diff --git a/cpu/base_cpu.cc b/cpu/base_cpu.cc
index 90e090d5e..74d2ceada 100644
--- a/cpu/base_cpu.cc
+++ b/cpu/base_cpu.cc
@@ -52,8 +52,8 @@ BaseCPU::BaseCPU(const string &_name, int _number_of_threads,
                  Counter max_insts_all_threads,
                  Counter max_loads_any_thread,
                  Counter max_loads_all_threads,
-                 System *_system, int num, Tick freq)
-    : SimObject(_name), number(num), frequency(freq),
+                 System *_system, Tick freq)
+    : SimObject(_name), frequency(freq),
       number_of_threads(_number_of_threads), system(_system)
 #else
 BaseCPU::BaseCPU(const string &_name, int _number_of_threads,
@@ -120,27 +120,73 @@ BaseCPU::BaseCPU(const string &_name, int _number_of_threads,
                 max_loads_all_threads, *counter);
     }
 
-
 #ifdef FULL_SYSTEM
     memset(interrupts, 0, sizeof(interrupts));
     intstatus = 0;
 #endif
 }
 
+
 void
 BaseCPU::regStats()
 {
-    int size = contexts.size();
+    int size = execContexts.size();
     if (size > 1) {
         for (int i = 0; i < size; ++i) {
             stringstream namestr;
             ccprintf(namestr, "%s.ctx%d", name(), i);
-            contexts[i]->regStats(namestr.str());
+            execContexts[i]->regStats(namestr.str());
         }
     } else if (size == 1)
-        contexts[0]->regStats(name());
+        execContexts[0]->regStats(name());
+}
+
+
+void
+BaseCPU::registerExecContexts()
+{
+    for (int i = 0; i < execContexts.size(); ++i) {
+        ExecContext *xc = execContexts[i];
+        int cpu_id;
+
+#ifdef FULL_SYSTEM
+        cpu_id = system->registerExecContext(xc);
+#else
+        cpu_id = xc->process->registerExecContext(xc);
+#endif
+
+        xc->cpu_id = cpu_id;
+    }
+}
+
+
+void
+BaseCPU::switchOut()
+{
+    // default: do nothing
 }
 
+void
+BaseCPU::takeOverFrom(BaseCPU *oldCPU)
+{
+    assert(execContexts.size() == oldCPU->execContexts.size());
+
+    for (int i = 0; i < execContexts.size(); ++i) {
+        ExecContext *newXC = execContexts[i];
+        ExecContext *oldXC = oldCPU->execContexts[i];
+
+        newXC->takeOverFrom(oldXC);
+        assert(newXC->cpu_id == oldXC->cpu_id);
+#ifdef FULL_SYSTEM
+        system->replaceExecContext(newXC->cpu_id, newXC);
+#else
+        assert(newXC->process == oldXC->process);
+        newXC->process->replaceExecContext(newXC->cpu_id, newXC);
+#endif
+    }
+}
+
+
 #ifdef FULL_SYSTEM
 void
 BaseCPU::post_interrupt(int int_num, int index)
@@ -185,4 +231,10 @@ BaseCPU::clear_interrupts()
 
 #endif // FULL_SYSTEM
 
+//
+// This declaration is not needed now that SamplingCPU provides a
+// BaseCPUBuilder object.
+//
+#if 0
 DEFINE_SIM_OBJECT_CLASS_NAME("BaseCPU", BaseCPU)
+#endif
diff --git a/cpu/base_cpu.hh b/cpu/base_cpu.hh
index e5d9df6de..5a20aaa54 100644
--- a/cpu/base_cpu.hh
+++ b/cpu/base_cpu.hh
@@ -47,7 +47,6 @@ class BaseCPU : public SimObject
 {
 #ifdef FULL_SYSTEM
   protected:
-    int number;
     Tick frequency;
     uint8_t interrupts[NumInterruptLevels];
     uint64_t intstatus;
@@ -71,7 +70,7 @@ class BaseCPU : public SimObject
 #endif
 
   protected:
-    std::vector<ExecContext *> contexts;
+    std::vector<ExecContext *> execContexts;
 
   public:
     virtual void execCtxStatusChg() {}
@@ -82,8 +81,7 @@ class BaseCPU : public SimObject
     BaseCPU(const std::string &_name, int _number_of_threads,
             Counter max_insts_any_thread, Counter max_insts_all_threads,
             Counter max_loads_any_thread, Counter max_loads_all_threads,
-            System *_system,
-            int num, Tick freq);
+            System *_system, Tick freq);
 #else
     BaseCPU(const std::string &_name, int _number_of_threads,
             Counter max_insts_any_thread = 0,
@@ -96,6 +94,16 @@ class BaseCPU : public SimObject
 
     virtual void regStats();
 
+    virtual void registerExecContexts();
+
+    /// Prepare for another CPU to take over execution.  Called by
+    /// takeOverFrom() on its argument.
+    virtual void switchOut();
+
+    /// Take over execution from the given CPU.  Used for warm-up and
+    /// sampling.
+    virtual void takeOverFrom(BaseCPU *);
+
     /**
      *  Number of threads we're actually simulating (<= SMT_MAX_THREADS).
      * This is a constant for the duration of the simulation.
diff --git a/cpu/exec_context.cc b/cpu/exec_context.cc
index 87f7283aa..6c24500cc 100644
--- a/cpu/exec_context.cc
+++ b/cpu/exec_context.cc
@@ -43,42 +43,62 @@ using namespace std;
 #ifdef FULL_SYSTEM
 ExecContext::ExecContext(BaseCPU *_cpu, int _thread_num, System *_sys,
                          AlphaItb *_itb, AlphaDtb *_dtb,
-                         FunctionalMemory *_mem, int _cpu_id)
-    : kernelStats(this, _cpu), cpu(_cpu), thread_num(_thread_num), mem(_mem),
-      itb(_itb), dtb(_dtb), cpu_id(_cpu_id), system(_sys),
-      memCtrl(_sys->memCtrl), physmem(_sys->physmem)
+                         FunctionalMemory *_mem)
+    : kernelStats(this, _cpu), cpu(_cpu), thread_num(_thread_num),
+      cpu_id(-1), mem(_mem), itb(_itb), dtb(_dtb), system(_sys),
+      memCtrl(_sys->memCtrl), physmem(_sys->physmem),
+      func_exe_insn(0), storeCondFailures(0)
 {
     memset(&regs, 0, sizeof(RegFile));
-    _status = Active;
-    func_exe_insn = 0;
-    storeCondFailures = 0;
-    system->registerExecContext(this);
+    setStatus(ExecContext::Unallocated);
 }
 #else
 ExecContext::ExecContext(BaseCPU *_cpu, int _thread_num,
                          Process *_process, int _asid)
-    : cpu(_cpu), thread_num(_thread_num), process(_process), asid (_asid)
+    : cpu(_cpu), thread_num(_thread_num), cpu_id(-1),
+      process(_process), asid (_asid),
+      func_exe_insn(0), storeCondFailures(0)
 {
-
-    // Register with process object. Our 'active' will be set by the
-    // process iff we're the initial context.  Others are reserved for
-    // dynamically created threads.
-    process->registerExecContext(this);
+    setStatus(ExecContext::Unallocated);
 
     mem = process->getMemory();
-
-    func_exe_insn = 0;
-    storeCondFailures = 0;
 }
 
 ExecContext::ExecContext(BaseCPU *_cpu, int _thread_num,
                          FunctionalMemory *_mem, int _asid)
-    : cpu(_cpu), thread_num(_thread_num), process(NULL), mem(_mem),
-      asid(_asid)
+    : cpu(_cpu), thread_num(_thread_num), process(0), mem(_mem), asid(_asid),
+      func_exe_insn(0), storeCondFailures(0)
 {
 }
 #endif
 
+
+void
+ExecContext::takeOverFrom(ExecContext *oldContext)
+{
+    // some things should already be set up
+    assert(mem == oldContext->mem);
+#ifdef FULL_SYSTEM
+    assert(system == oldContext->system);
+#else
+    assert(process == oldContext->process);
+#endif
+
+    // copy over functional state
+    _status = oldContext->_status;
+#ifdef FULL_SYSTEM
+    kernelStats = oldContext->kernelStats;
+#endif
+    regs = oldContext->regs;
+    cpu_id = oldContext->cpu_id;
+    func_exe_insn = oldContext->func_exe_insn;
+
+    storeCondFailures = 0;
+
+    oldContext->_status = ExecContext::Unallocated;
+}
+
+
 void
 ExecContext::setStatus(Status new_status)
 {
diff --git a/cpu/exec_context.hh b/cpu/exec_context.hh
index bd7f6fad9..5c6e84cee 100644
--- a/cpu/exec_context.hh
+++ b/cpu/exec_context.hh
@@ -67,6 +67,11 @@ class ExecContext
 
   public:
     Status status() const { return _status; }
+
+    // Unlike setStatus(), initStatus() has no side effects other than
+    // setting the _status variable.
+    void initStatus(Status init_status) { _status = init_status; }
+
     void setStatus(Status new_status);
 
 #ifdef FULL_SYSTEM
@@ -83,12 +88,15 @@ class ExecContext
     // Index of hardware thread context on the CPU that this represents.
     int thread_num;
 
+    // ID of this context w.r.t. the System or Process object to which
+    // it belongs.  For full-system mode, this is the system CPU ID.
+    int cpu_id;
+
 #ifdef FULL_SYSTEM
 
     FunctionalMemory *mem;
     AlphaItb *itb;
     AlphaDtb *dtb;
-    int cpu_id;
     System *system;
 
     // the following two fields are redundant, since we can always
@@ -124,8 +132,7 @@ class ExecContext
     // constructor: initialize context from given process structure
 #ifdef FULL_SYSTEM
     ExecContext(BaseCPU *_cpu, int _thread_num, System *_system,
-                AlphaItb *_itb, AlphaDtb *_dtb, FunctionalMemory *_dem,
-                int _cpu_id);
+                AlphaItb *_itb, AlphaDtb *_dtb, FunctionalMemory *_dem);
 #else
     ExecContext(BaseCPU *_cpu, int _thread_num, Process *_process, int _asid);
     ExecContext(BaseCPU *_cpu, int _thread_num, FunctionalMemory *_mem,
@@ -133,6 +140,8 @@ class ExecContext
 #endif
     virtual ~ExecContext() {}
 
+    virtual void takeOverFrom(ExecContext *oldContext);
+
     void regStats(const std::string &name);
 
 #ifdef FULL_SYSTEM
@@ -156,7 +165,6 @@ class ExecContext
         return dtb->translate(req, true);
     }
 
-
 #else
     bool validInstAddr(Addr addr)
     { return process->validInstAddr(addr); }
@@ -244,8 +252,8 @@ class ExecContext
         // and all other stores (WH64?).  Unsuccessful Store
         // Conditionals would have returned above, and wouldn't fall
         // through.
-        for (int i = 0; i < system->xcvec.size(); i++){
-            cregs = &system->xcvec[i]->regs.miscRegs;
+        for (int i = 0; i < system->execContexts.size(); i++){
+            cregs = &system->execContexts[i]->regs.miscRegs;
             if ((cregs->lock_addr & ~0xf) == (req->paddr & ~0xf)) {
                 cregs->lock_flag = false;
             }
diff --git a/cpu/simple_cpu/simple_cpu.cc b/cpu/simple_cpu/simple_cpu.cc
index 3c9be172f..c96e31cb0 100644
--- a/cpu/simple_cpu/simple_cpu.cc
+++ b/cpu/simple_cpu/simple_cpu.cc
@@ -103,11 +103,11 @@ SimpleCPU::SimpleCPU(const string &_name,
                      FunctionalMemory *mem,
                      MemInterface *icache_interface,
                      MemInterface *dcache_interface,
-                     int cpu_id, Tick freq)
+                     Tick freq)
     : BaseCPU(_name, /* number_of_threads */ 1,
               max_insts_any_thread, max_insts_all_threads,
               max_loads_any_thread, max_loads_all_threads,
-              _system, cpu_id, freq),
+              _system, freq),
 #else
 SimpleCPU::SimpleCPU(const string &_name, Process *_process,
                      Counter max_insts_any_thread,
@@ -122,61 +122,23 @@ SimpleCPU::SimpleCPU(const string &_name, Process *_process,
 #endif
       tickEvent(this), xc(NULL), cacheCompletionEvent(this)
 {
+    _status = Idle;
 #ifdef FULL_SYSTEM
-    xc = new ExecContext(this, 0, system, itb, dtb, mem, cpu_id);
+    xc = new ExecContext(this, 0, system, itb, dtb, mem);
 
-    _status = Running;
-    if (cpu_id != 0) {
+    TheISA::initCPU(&xc->regs);
 
-       xc->setStatus(ExecContext::Unallocated);
+    IntReg *ipr = xc->regs.ipr;
+    ipr[TheISA::IPR_MCSR] = 0x6;
 
-       //Open a GDB debug session on port (7000 + the cpu_id)
-       (new GDBListener(new RemoteGDB(system, xc), 7000 + cpu_id))->listen();
-
-       AlphaISA::init(system->physmem, &xc->regs);
-
-       fault = Reset_Fault;
-
-       IntReg *ipr = xc->regs.ipr;
-       ipr[TheISA::IPR_MCSR] = 0x6;
-
-       AlphaISA::swap_palshadow(&xc->regs, true);
-
-       xc->regs.pc =
-           ipr[TheISA::IPR_PAL_BASE] + AlphaISA::fault_addr[fault];
-       xc->regs.npc = xc->regs.pc + sizeof(MachInst);
-
-       _status = Idle;
-    }
-    else {
-      system->init(xc);
-
-      // Reset the system
-      //
-      AlphaISA::init(system->physmem, &xc->regs);
-
-      fault = Reset_Fault;
-
-      IntReg *ipr = xc->regs.ipr;
-      ipr[TheISA::IPR_MCSR] = 0x6;
-
-      AlphaISA::swap_palshadow(&xc->regs, true);
-
-      xc->regs.pc = ipr[TheISA::IPR_PAL_BASE] + AlphaISA::fault_addr[fault];
-      xc->regs.npc = xc->regs.pc + sizeof(MachInst);
-
-       _status = Running;
-       tickEvent.schedule(0);
-    }
+    AlphaISA::swap_palshadow(&xc->regs, true);
 
+    fault = Reset_Fault;
+    xc->regs.pc = ipr[TheISA::IPR_PAL_BASE] + AlphaISA::fault_addr[fault];
+    xc->regs.npc = xc->regs.pc + sizeof(MachInst);
 #else
     xc = new ExecContext(this, /* thread_num */ 0, _process, /* asid */ 0);
     fault = No_Fault;
-    if (xc->status() == ExecContext::Active) {
-        _status = Running;
-       tickEvent.schedule(0);
-    } else
-        _status = Idle;
 #endif // !FULL_SYSTEM
 
     icacheInterface = icache_interface;
@@ -193,13 +155,61 @@ SimpleCPU::SimpleCPU(const string &_name, Process *_process,
     lastIcacheStall = 0;
     lastDcacheStall = 0;
 
-    contexts.push_back(xc);
+    execContexts.push_back(xc);
 }
 
 SimpleCPU::~SimpleCPU()
 {
 }
 
+
+void
+SimpleCPU::registerExecContexts()
+{
+    BaseCPU::registerExecContexts();
+
+    // if any of this CPU's ExecContexts are active, mark the CPU as
+    // running and schedule its tick event.
+    for (int i = 0; i < execContexts.size(); ++i) {
+        ExecContext *xc = execContexts[i];
+        if (xc->status() == ExecContext::Active && _status != Running) {
+            _status = Running;
+            // this should only happen at initialization time
+            assert(curTick == 0);
+            tickEvent.schedule(0);
+        }
+    }
+}
+
+
+void
+SimpleCPU::switchOut()
+{
+    _status = SwitchedOut;
+    if (tickEvent.scheduled())
+        tickEvent.squash();
+}
+
+
+void
+SimpleCPU::takeOverFrom(BaseCPU *oldCPU)
+{
+    BaseCPU::takeOverFrom(oldCPU);
+
+    assert(!tickEvent.scheduled());
+
+    // if any of this CPU's ExecContexts are active, mark the CPU as
+    // running and schedule its tick event.
+    for (int i = 0; i < execContexts.size(); ++i) {
+        ExecContext *xc = execContexts[i];
+        if (xc->status() == ExecContext::Active && _status != Running) {
+            _status = Running;
+            tickEvent.schedule(curTick + 1);
+        }
+    }
+}
+
+
 void
 SimpleCPU::regStats()
 {
@@ -488,6 +498,11 @@ SimpleCPU::processCacheCompletion()
         dcacheStallCycles += curTick - lastDcacheStall;
         setStatus(Running);
         break;
+      case SwitchedOut:
+        // If this CPU has been switched out due to sampling/warm-up,
+        // ignore any further status changes (e.g., due to cache
+        // misses outstanding at the time of the switch).
+        return;
       default:
         panic("SimpleCPU::processCacheCompletion: bad state");
         break;
@@ -693,7 +708,6 @@ BEGIN_DECLARE_SIM_OBJECT_PARAMS(SimpleCPU)
     SimObjectParam<AlphaDtb *> dtb;
     SimObjectParam<FunctionalMemory *> mem;
     SimObjectParam<System *> system;
-    Param<int> cpu_id;
     Param<int> mult;
 #else
     SimObjectParam<Process *> workload;
@@ -702,6 +716,8 @@ BEGIN_DECLARE_SIM_OBJECT_PARAMS(SimpleCPU)
     SimObjectParam<BaseMem *> icache;
     SimObjectParam<BaseMem *> dcache;
 
+    Param<bool> defer_registration;
+
 END_DECLARE_SIM_OBJECT_PARAMS(SimpleCPU)
 
 BEGIN_INIT_SIM_OBJECT_PARAMS(SimpleCPU)
@@ -724,39 +740,47 @@ BEGIN_INIT_SIM_OBJECT_PARAMS(SimpleCPU)
     INIT_PARAM(dtb, "Data TLB"),
     INIT_PARAM(mem, "memory"),
     INIT_PARAM(system, "system object"),
-    INIT_PARAM_DFLT(cpu_id, "CPU identification number", 0),
     INIT_PARAM_DFLT(mult, "system clock multiplier", 1),
 #else
     INIT_PARAM(workload, "processes to run"),
 #endif // FULL_SYSTEM
 
     INIT_PARAM_DFLT(icache, "L1 instruction cache object", NULL),
-    INIT_PARAM_DFLT(dcache, "L1 data cache object", NULL)
+    INIT_PARAM_DFLT(dcache, "L1 data cache object", NULL),
+    INIT_PARAM_DFLT(defer_registration, "defer registration with system "
+                    "(for sampling)", false)
 
 END_INIT_SIM_OBJECT_PARAMS(SimpleCPU)
 
 
 CREATE_SIM_OBJECT(SimpleCPU)
 {
+    SimpleCPU *cpu;
 #ifdef FULL_SYSTEM
     if (mult != 1)
         panic("processor clock multiplier must be 1\n");
 
-    return new SimpleCPU(getInstanceName(), system,
-                         max_insts_any_thread, max_insts_all_threads,
-                         max_loads_any_thread, max_loads_all_threads,
-                         itb, dtb, mem,
-                         (icache) ? icache->getInterface() : NULL,
-                         (dcache) ? dcache->getInterface() : NULL,
-                         cpu_id, ticksPerSecond * mult);
+    cpu = new SimpleCPU(getInstanceName(), system,
+                        max_insts_any_thread, max_insts_all_threads,
+                        max_loads_any_thread, max_loads_all_threads,
+                        itb, dtb, mem,
+                        (icache) ? icache->getInterface() : NULL,
+                        (dcache) ? dcache->getInterface() : NULL,
+                        ticksPerSecond * mult);
 #else
 
-    return new SimpleCPU(getInstanceName(), workload,
-                         max_insts_any_thread, max_insts_all_threads,
-                         max_loads_any_thread, max_loads_all_threads,
-                         icache->getInterface(), dcache->getInterface());
+    cpu = new SimpleCPU(getInstanceName(), workload,
+                        max_insts_any_thread, max_insts_all_threads,
+                        max_loads_any_thread, max_loads_all_threads,
+                        icache->getInterface(), dcache->getInterface());
 
 #endif // FULL_SYSTEM
+
+    if (!defer_registration) {
+        cpu->registerExecContexts();
+    }
+
+    return cpu;
 }
 
 REGISTER_SIM_OBJECT("SimpleCPU", SimpleCPU)
diff --git a/cpu/simple_cpu/simple_cpu.hh b/cpu/simple_cpu/simple_cpu.hh
index fd50acfe1..e1b351cab 100644
--- a/cpu/simple_cpu/simple_cpu.hh
+++ b/cpu/simple_cpu/simple_cpu.hh
@@ -92,7 +92,8 @@ class SimpleCPU : public BaseCPU
         Idle,
         IcacheMissStall,
         IcacheMissComplete,
-        DcacheMissStall
+        DcacheMissStall,
+        SwitchedOut
     };
 
   private:
@@ -117,7 +118,7 @@ class SimpleCPU : public BaseCPU
               Counter max_loads_any_thread, Counter max_loads_all_threads,
               AlphaItb *itb, AlphaDtb *dtb, FunctionalMemory *mem,
               MemInterface *icache_interface, MemInterface *dcache_interface,
-              int cpu_id, Tick freq);
+              Tick freq);
 
 #else
 
@@ -135,6 +136,11 @@ class SimpleCPU : public BaseCPU
     // execution context
     ExecContext *xc;
 
+    void registerExecContexts();
+
+    void switchOut();
+    void takeOverFrom(BaseCPU *oldCPU);
+
 #ifdef FULL_SYSTEM
     Addr dbg_vtophys(Addr addr);
 
@@ -171,6 +177,7 @@ class SimpleCPU : public BaseCPU
     CacheCompletionEvent cacheCompletionEvent;
 
     Status status() const { return _status; }
+
     virtual void execCtxStatusChg() {
         if (xc) {
             if (xc->status() == ExecContext::Active)
@@ -182,6 +189,10 @@ class SimpleCPU : public BaseCPU
 
     void setStatus(Status new_status) {
         Status old_status = status();
+
+        // We should never even get here if the CPU has been switched out.
+        assert(old_status != SwitchedOut);
+
         _status = new_status;
 
         switch (status()) {
diff --git a/dev/alpha_console.cc b/dev/alpha_console.cc
index b917d7a96..cd03b3656 100644
--- a/dev/alpha_console.cc
+++ b/dev/alpha_console.cc
@@ -153,7 +153,7 @@ AlphaConsole::write(MemReqPtr req, const uint8_t *data)
         int cpu = val;
         assert(cpu > 0 && "Must not access primary cpu");
 
-        ExecContext *other_xc = req->xc->system->xcvec[cpu];
+        ExecContext *other_xc = req->xc->system->execContexts[cpu];
         other_xc->regs.intRegFile[16] = cpu;
         other_xc->regs.ipr[TheISA::IPR_PALtemp16] = cpu;
         other_xc->regs.intRegFile[0] = cpu;
diff --git a/kern/tru64/tru64_system.cc b/kern/tru64/tru64_system.cc
index 823a00e5a..8b4ff668d 100644
--- a/kern/tru64/tru64_system.cc
+++ b/kern/tru64/tru64_system.cc
@@ -184,24 +184,44 @@ Tru64System::~Tru64System()
 
 }
 
-void
-Tru64System::init(ExecContext *xc)
+int
+Tru64System::registerExecContext(ExecContext *xc)
 {
-    xc->regs = *initRegs;
+    int xcIndex = System::registerExecContext(xc);
+
+    if (xcIndex == 0) {
+        // xc->regs = *initRegs;
+        xc->initStatus(ExecContext::Active);
+    }
+    else {
+        xc->initStatus(ExecContext::Unallocated);
+    }
+
+    RemoteGDB *rgdb = new RemoteGDB(this, xc);
+    GDBListener *gdbl = new GDBListener(rgdb, 7000 + xcIndex);
+    gdbl->listen();
+
+    if (remoteGDB.size() <= xcIndex) {
+        remoteGDB.resize(xcIndex+1);
+    }
+
+    remoteGDB[xcIndex] = rgdb;
+
+    return xcIndex;
+}
 
-    remoteGDB = new RemoteGDB(this, xc);
-    gdbListen = new GDBListener(remoteGDB, 7000);
-    gdbListen->listen();
 
-    // Reset the system
-    //
-    TheISA::init(physmem, &xc->regs);
+void
+Tru64System::replaceExecContext(ExecContext *xc, int xcIndex)
+{
+    System::replaceExecContext(xcIndex, xc);
+    remoteGDB[xcIndex]->replaceExecContext(xc);
 }
 
 bool
 Tru64System::breakpoint()
 {
-    return remoteGDB->trap(ALPHA_KENTRY_IF);
+    return remoteGDB[0]->trap(ALPHA_KENTRY_IF);
 }
 
 BEGIN_DECLARE_SIM_OBJECT_PARAMS(Tru64System)
diff --git a/kern/tru64/tru64_system.hh b/kern/tru64/tru64_system.hh
index f1b8bb532..a67792aaf 100644
--- a/kern/tru64/tru64_system.hh
+++ b/kern/tru64/tru64_system.hh
@@ -29,6 +29,8 @@
 #ifndef __TRU64_SYSTEM_HH__
 #define __TRU64_SYSTEM_HH__
 
+#include <vector>
+
 #include "sim/system.hh"
 #include "targetarch/isa_traits.hh"
 
@@ -48,8 +50,6 @@ class AlphaArguments;
 class Tru64System : public System
 {
   private:
-    ExecContext *xc;
-
     EcoffObject *kernel;
     EcoffObject *console;
 
@@ -74,8 +74,8 @@ class Tru64System : public System
     Addr kernelEntry;
 
   public:
-    RemoteGDB *remoteGDB;
-    GDBListener *gdbListen;
+    std::vector<RemoteGDB *>   remoteGDB;
+    std::vector<GDBListener *> gdbListen;
 
   public:
     Tru64System(const std::string _name,
@@ -88,7 +88,8 @@ class Tru64System : public System
                 const std::string &boot_osflags);
     ~Tru64System();
 
-    void init(ExecContext *xc);
+    int registerExecContext(ExecContext *xc);
+    void replaceExecContext(ExecContext *xc, int xcIndex);
 
     Addr getKernelStart() const { return kernelStart; }
     Addr getKernelEnd() const { return kernelEnd; }
diff --git a/sim/prog.cc b/sim/prog.cc
index 275e1551b..599a0ca9a 100644
--- a/sim/prog.cc
+++ b/sim/prog.cc
@@ -85,8 +85,6 @@ Process::Process(const string &name,
         fd_map[i] = -1;
     }
 
-    numCpus = 0;
-
     num_syscalls = 0;
 
     // other parameters will be initialized when the program is loaded
@@ -136,30 +134,42 @@ Process::openOutputFile(const string &filename)
 }
 
 
-void
-Process::registerExecContext(ExecContext *ec)
+int
+Process::registerExecContext(ExecContext *xc)
 {
-    if (execContexts.empty()) {
+    // add to list
+    int myIndex = execContexts.size();
+    execContexts.push_back(xc);
+
+    if (myIndex == 0) {
         // first exec context for this process... initialize & enable
 
         // copy process's initial regs struct
-        ec->regs = *init_regs;
+        xc->regs = *init_regs;
 
         // mark this context as active
-        ec->setStatus(ExecContext::Active);
+        xc->initStatus(ExecContext::Active);
     }
     else {
-        ec->setStatus(ExecContext::Unallocated);
+        xc->initStatus(ExecContext::Unallocated);
     }
 
-    // add to list
-    execContexts.push_back(ec);
-
-    // increment available CPU count
-    ++numCpus;
+    // return CPU number to caller and increment available CPU count
+    return myIndex;
 }
 
 
+void
+Process::replaceExecContext(int xcIndex, ExecContext *xc)
+{
+    if (xcIndex >= execContexts.size()) {
+        panic("replaceExecContext: bad xcIndex, %d >= %d\n",
+              xcIndex, execContexts.size());
+    }
+
+    execContexts[xcIndex] = xc;
+}
+
 // map simulator fd sim_fd to target fd tgt_fd
 void
 Process::dup_fd(int sim_fd, int tgt_fd)
diff --git a/sim/prog.hh b/sim/prog.hh
index ee4bc59fe..e93dd684c 100644
--- a/sim/prog.hh
+++ b/sim/prog.hh
@@ -36,7 +36,7 @@
 //
 #ifndef FULL_SYSTEM
 
-#include <list>
+#include <vector>
 
 #include "targetarch/isa_traits.hh"
 #include "sim/sim_object.hh"
@@ -55,10 +55,10 @@ class Process : public SimObject
     bool initialContextLoaded;
 
     // execution contexts associated with this process
-    std::list<ExecContext *> execContexts;
-    // number of CPUs assigned to this process: should match number of
-    // contexts in execContexts list
-    unsigned numCpus;
+    std::vector<ExecContext *> execContexts;
+
+    // number of CPUs (esxec contexts, really) assigned to this process.
+    unsigned int numCpus() { return execContexts.size(); }
 
     // record of blocked context
     struct WaitRec
@@ -123,8 +123,12 @@ class Process : public SimObject
     // override of virtual SimObject method: register statistics
     virtual void regStats();
 
-    // register an execution context for this process
-    void registerExecContext(ExecContext *);
+    // register an execution context for this process.
+    // returns xc's cpu number (index into execContexts[])
+    int registerExecContext(ExecContext *xc);
+
+
+    void replaceExecContext(int xcIndex, ExecContext *xc);
 
     // map simulator fd sim_fd to target fd tgt_fd
     void dup_fd(int sim_fd, int tgt_fd);
diff --git a/sim/system.cc b/sim/system.cc
index 903362f0b..af4a4c151 100644
--- a/sim/system.cc
+++ b/sim/system.cc
@@ -56,16 +56,24 @@ System::~System()
 }
 
 
-void
+int
 System::registerExecContext(ExecContext *xc)
 {
-    if (xc->cpu_id >= 12/*MAX_CPUS*/)
-        panic("Too many CPU's\n");
+    int myIndex = execContexts.size();
+    execContexts.push_back(xc);
+    return myIndex;
+}
+
 
-    if (xc->cpu_id >= xcvec.size())
-        xcvec.resize(xc->cpu_id + 1);
+void
+System::replaceExecContext(int xcIndex, ExecContext *xc)
+{
+    if (xcIndex >= execContexts.size()) {
+        panic("replaceExecContext: bad xcIndex, %d >= %d\n",
+              xcIndex, execContexts.size());
+    }
 
-    xcvec[xc->cpu_id] = xc;
+    execContexts[xcIndex] = xc;
 }
 
 
diff --git a/sim/system.hh b/sim/system.hh
index f10b70a4e..081aa3eb7 100644
--- a/sim/system.hh
+++ b/sim/system.hh
@@ -52,16 +52,16 @@ class System : public SimObject
 
     PCEventQueue pcEventQueue;
 
-    std::vector<ExecContext *> xcvec;
-    void registerExecContext(ExecContext *xc);
+    std::vector<ExecContext *> execContexts;
+
+    virtual int registerExecContext(ExecContext *xc);
+    virtual void replaceExecContext(int xcIndex, ExecContext *xc);
 
   public:
     System(const std::string _name, const uint64_t _init_param,
            MemoryController *, PhysicalMemory *);
     ~System();
 
-    virtual void init(ExecContext *xc) = 0;
-
     virtual Addr getKernelStart() const = 0;
     virtual Addr getKernelEnd() const = 0;
     virtual Addr getKernelEntry() const = 0;