Merge with the main repository again.
[gem5.git] / src / cpu / base.cc
index 4dccee0d3c8aa4505a225a03d3550e22917158a6..a0785ac10117857b18564e8b0ab8e9662480899d 100644 (file)
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2002-2005 The Regents of The University of Michigan
+ * Copyright (c) 2011 Regents of the University of California
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  *
  * Authors: Steve Reinhardt
  *          Nathan Binkert
+ *          Rick Strong
  */
 
 #include <iostream>
-#include <string>
 #include <sstream>
+#include <string>
 
-#include "base/cprintf.hh"
+#include "arch/tlb.hh"
 #include "base/loader/symtab.hh"
+#include "base/cprintf.hh"
 #include "base/misc.hh"
 #include "base/output.hh"
+#include "base/trace.hh"
 #include "cpu/base.hh"
 #include "cpu/cpuevent.hh"
-#include "cpu/thread_context.hh"
 #include "cpu/profile.hh"
-#include "sim/sim_exit.hh"
-#include "sim/param.hh"
+#include "cpu/thread_context.hh"
+#include "debug/SyscallVerbose.hh"
+#include "params/BaseCPU.hh"
+#include "sim/full_system.hh"
 #include "sim/process.hh"
 #include "sim/sim_events.hh"
+#include "sim/sim_exit.hh"
 #include "sim/system.hh"
 
-#include "base/trace.hh"
-
 // Hack
 #include "sim/stat_control.hh"
 
@@ -61,13 +65,12 @@ vector<BaseCPU *> BaseCPU::cpuList;
 // been initialized
 int maxThreadsPerCPU = 1;
 
-CPUProgressEvent::CPUProgressEvent(EventQueue *q, Tick ival,
-                                   BaseCPU *_cpu)
-    : Event(q, Event::Progress_Event_Pri), interval(ival),
-      lastNumInst(0), cpu(_cpu)
+CPUProgressEvent::CPUProgressEvent(BaseCPU *_cpu, Tick ival)
+    : Event(Event::Progress_Event_Pri), _interval(ival), lastNumInst(0),
+      cpu(_cpu), _repeatEvent(true)
 {
-    if (interval)
-        schedule(curTick + interval);
+    if (_interval)
+        cpu->schedule(this, curTick() + _interval);
 }
 
 void
@@ -75,129 +78,136 @@ CPUProgressEvent::process()
 {
     Counter temp = cpu->totalInstructions();
 #ifndef NDEBUG
-    double ipc = double(temp - lastNumInst) / (interval / cpu->cycles(1));
+    double ipc = double(temp - lastNumInst) / (_interval / cpu->ticks(1));
 
-    DPRINTFN("%s progress event, instructions committed: %lli, IPC: %0.8d\n",
-             cpu->name(), temp - lastNumInst, ipc);
+    DPRINTFN("%s progress event, total committed:%i, progress insts committed: "
+             "%lli, IPC: %0.8d\n", cpu->name(), temp, temp - lastNumInst,
+             ipc);
     ipc = 0.0;
 #else
-    cprintf("%lli: %s progress event, instructions committed: %lli\n",
-            curTick, cpu->name(), temp - lastNumInst);
+    cprintf("%lli: %s progress event, total committed:%i, progress insts "
+            "committed: %lli\n", curTick(), cpu->name(), temp,
+            temp - lastNumInst);
 #endif
     lastNumInst = temp;
-    schedule(curTick + interval);
+
+    if (_repeatEvent)
+        cpu->schedule(this, curTick() + _interval);
 }
 
 const char *
-CPUProgressEvent::description()
+CPUProgressEvent::description() const
 {
-    return "CPU Progress event";
+    return "CPU Progress";
 }
 
-#if FULL_SYSTEM
-BaseCPU::BaseCPU(Params *p)
-    : MemObject(p->name), clock(p->clock), instCnt(0),
-      params(p), number_of_threads(p->numberOfThreads), system(p->system),
-      phase(p->phase)
-#else
 BaseCPU::BaseCPU(Params *p)
-    : MemObject(p->name), clock(p->clock), params(p),
-      number_of_threads(p->numberOfThreads), system(p->system),
+    : MemObject(p), clock(p->clock), instCnt(0), _cpuId(p->cpu_id),
+      interrupts(p->interrupts),
+      numThreads(p->numThreads), system(p->system),
       phase(p->phase)
-#endif
 {
-//    currentTick = curTick;
-    DPRINTF(FullCPU, "BaseCPU: Creating object, mem address %#x.\n", this);
+//    currentTick = curTick();
+
+    // if Python did not provide a valid ID, do it here
+    if (_cpuId == -1 ) {
+        _cpuId = cpuList.size();
+    }
 
     // add self to global list of CPUs
     cpuList.push_back(this);
 
-    DPRINTF(FullCPU, "BaseCPU: CPU added to cpuList, mem address %#x.\n",
-            this);
+    DPRINTF(SyscallVerbose, "Constructing CPU with id %d\n", _cpuId);
 
-    if (number_of_threads > maxThreadsPerCPU)
-        maxThreadsPerCPU = number_of_threads;
+    if (numThreads > maxThreadsPerCPU)
+        maxThreadsPerCPU = numThreads;
 
     // allocate per-thread instruction-based event queues
-    comInstEventQueue = new EventQueue *[number_of_threads];
-    for (int i = 0; i < number_of_threads; ++i)
-        comInstEventQueue[i] = new EventQueue("instruction-based event queue");
+    comInstEventQueue = new EventQueue *[numThreads];
+    for (ThreadID tid = 0; tid < numThreads; ++tid)
+        comInstEventQueue[tid] =
+            new EventQueue("instruction-based event queue");
 
     //
     // set up instruction-count-based termination events, if any
     //
-    if (p->max_insts_any_thread != 0)
-        for (int i = 0; i < number_of_threads; ++i)
-            schedExitSimLoop("a thread reached the max instruction count",
-                             p->max_insts_any_thread, 0,
-                             comInstEventQueue[i]);
+    if (p->max_insts_any_thread != 0) {
+        const char *cause = "a thread reached the max instruction count";
+        for (ThreadID tid = 0; tid < numThreads; ++tid) {
+            Event *event = new SimLoopExitEvent(cause, 0);
+            comInstEventQueue[tid]->schedule(event, p->max_insts_any_thread);
+        }
+    }
 
     if (p->max_insts_all_threads != 0) {
+        const char *cause = "all threads reached the max instruction count";
+
         // allocate & initialize shared downcounter: each event will
         // decrement this when triggered; simulation will terminate
         // when counter reaches 0
         int *counter = new int;
-        *counter = number_of_threads;
-        for (int i = 0; i < number_of_threads; ++i)
-            new CountedExitEvent(comInstEventQueue[i],
-                "all threads reached the max instruction count",
-                p->max_insts_all_threads, *counter);
+        *counter = numThreads;
+        for (ThreadID tid = 0; tid < numThreads; ++tid) {
+            Event *event = new CountedExitEvent(cause, *counter);
+            comInstEventQueue[tid]->schedule(event, p->max_insts_all_threads);
+        }
     }
 
     // allocate per-thread load-based event queues
-    comLoadEventQueue = new EventQueue *[number_of_threads];
-    for (int i = 0; i < number_of_threads; ++i)
-        comLoadEventQueue[i] = new EventQueue("load-based event queue");
+    comLoadEventQueue = new EventQueue *[numThreads];
+    for (ThreadID tid = 0; tid < numThreads; ++tid)
+        comLoadEventQueue[tid] = new EventQueue("load-based event queue");
 
     //
     // set up instruction-count-based termination events, if any
     //
-    if (p->max_loads_any_thread != 0)
-        for (int i = 0; i < number_of_threads; ++i)
-            schedExitSimLoop("a thread reached the max load count",
-                             p->max_loads_any_thread, 0,
-                             comLoadEventQueue[i]);
+    if (p->max_loads_any_thread != 0) {
+        const char *cause = "a thread reached the max load count";
+        for (ThreadID tid = 0; tid < numThreads; ++tid) {
+            Event *event = new SimLoopExitEvent(cause, 0);
+            comLoadEventQueue[tid]->schedule(event, p->max_loads_any_thread);
+        }
+    }
 
     if (p->max_loads_all_threads != 0) {
+        const char *cause = "all threads reached the max load count";
         // allocate & initialize shared downcounter: each event will
         // decrement this when triggered; simulation will terminate
         // when counter reaches 0
         int *counter = new int;
-        *counter = number_of_threads;
-        for (int i = 0; i < number_of_threads; ++i)
-            new CountedExitEvent(comLoadEventQueue[i],
-                "all threads reached the max load count",
-                p->max_loads_all_threads, *counter);
+        *counter = numThreads;
+        for (ThreadID tid = 0; tid < numThreads; ++tid) {
+            Event *event = new CountedExitEvent(cause, *counter);
+            comLoadEventQueue[tid]->schedule(event, p->max_loads_all_threads);
+        }
     }
 
     functionTracingEnabled = false;
-    if (p->functionTrace) {
-        functionTraceStream = simout.find(csprintf("ftrace.%s", name()));
+    if (p->function_trace) {
+        const string fname = csprintf("ftrace.%s", name());
+        functionTraceStream = simout.find(fname);
+        if (!functionTraceStream)
+            functionTraceStream = simout.create(fname);
+
         currentFunctionStart = currentFunctionEnd = 0;
-        functionEntryTick = p->functionTraceStart;
+        functionEntryTick = p->function_trace_start;
 
-        if (p->functionTraceStart == 0) {
+        if (p->function_trace_start == 0) {
             functionTracingEnabled = true;
         } else {
-            Event *e =
-                new EventWrapper<BaseCPU, &BaseCPU::enableFunctionTrace>(this,
-                                                                         true);
-            e->schedule(p->functionTraceStart);
+            typedef EventWrapper<BaseCPU, &BaseCPU::enableFunctionTrace> wrap;
+            Event *event = new wrap(this, true);
+            schedule(event, p->function_trace_start);
         }
     }
-#if FULL_SYSTEM
-    profileEvent = NULL;
-    if (params->profile)
-        profileEvent = new ProfileEvent(this, params->profile);
-#endif
-}
+    interrupts->setCPU(this);
 
-BaseCPU::Params::Params()
-{
-#if FULL_SYSTEM
-    profile = false;
-#endif
-    checker = NULL;
+    if (FullSystem) {
+        profileEvent = NULL;
+        if (params()->profile)
+            profileEvent = new ProfileEvent(this, params()->profile);
+    }
+    tracer = params()->tracer;
 }
 
 void
@@ -213,22 +223,22 @@ BaseCPU::~BaseCPU()
 void
 BaseCPU::init()
 {
-    if (!params->deferRegistration)
+    if (!params()->defer_registration)
         registerThreadContexts();
 }
 
 void
 BaseCPU::startup()
 {
-#if FULL_SYSTEM
-    if (!params->deferRegistration && profileEvent)
-        profileEvent->schedule(curTick);
-#endif
+    if (FullSystem) {
+        if (!params()->defer_registration && profileEvent)
+            schedule(profileEvent, curTick());
+    }
 
-    if (params->progress_interval) {
-        new CPUProgressEvent(&mainEventQueue,
-                             cycles(params->progress_interval),
-                             this);
+    if (params()->progress_interval) {
+        Tick num_ticks = ticks(params()->progress_interval);
+
+        new CPUProgressEvent(this, num_ticks);
     }
 }
 
@@ -243,6 +253,16 @@ BaseCPU::regStats()
         .desc("number of cpu cycles simulated")
         ;
 
+    numWorkItemsStarted
+        .name(name() + ".numWorkItemsStarted")
+        .desc("number of work items this cpu started")
+        ;
+
+    numWorkItemsCompleted
+        .name(name() + ".numWorkItemsCompleted")
+        .desc("number of work items this cpu completed")
+        ;
+
     int size = threadContexts.size();
     if (size > 1) {
         for (int i = 0; i < size; ++i) {
@@ -252,15 +272,12 @@ BaseCPU::regStats()
         }
     } else if (size == 1)
         threadContexts[0]->regStats(name());
-
-#if FULL_SYSTEM
-#endif
 }
 
 Tick
 BaseCPU::nextCycle()
 {
-    Tick next_tick = curTick - phase + clock - 1;
+    Tick next_tick = curTick() - phase + clock - 1;
     next_tick -= (next_tick % clock);
     next_tick += phase;
     return next_tick;
@@ -270,31 +287,35 @@ Tick
 BaseCPU::nextCycle(Tick begin_tick)
 {
     Tick next_tick = begin_tick;
-    next_tick -= (next_tick % clock);
+    if (next_tick % clock != 0)
+        next_tick = next_tick - (next_tick % clock) + clock;
     next_tick += phase;
 
-    while (next_tick < curTick)
-        next_tick += clock;
-
-    assert(next_tick >= curTick);
+    assert(next_tick >= curTick());
     return next_tick;
 }
 
 void
 BaseCPU::registerThreadContexts()
 {
-    for (int i = 0; i < threadContexts.size(); ++i) {
-        ThreadContext *tc = threadContexts[i];
-
-#if FULL_SYSTEM
-        int id = params->cpu_id;
-        if (id != -1)
-            id += i;
-
-        tc->setCpuId(system->registerThreadContext(tc, id));
-#else
-        tc->setCpuId(tc->getProcessPtr()->registerThreadContext(tc));
-#endif
+    ThreadID size = threadContexts.size();
+    for (ThreadID tid = 0; tid < size; ++tid) {
+        ThreadContext *tc = threadContexts[tid];
+
+        /** This is so that contextId and cpuId match where there is a
+         * 1cpu:1context relationship.  Otherwise, the order of registration
+         * could affect the assignment and cpu 1 could have context id 3, for
+         * example.  We may even want to do something like this for SMT so that
+         * cpu 0 has the lowest thread contexts and cpu N has the highest, but
+         * I'll just do this for now
+         */
+        if (numThreads == 1)
+            tc->setContextId(system->registerThreadContext(tc, _cpuId));
+        else
+            tc->setContextId(system->registerThreadContext(tc));
+
+        if (!FullSystem)
+            tc->getProcessPtr()->assignThreadContext(tc->contextId());
     }
 }
 
@@ -302,9 +323,10 @@ BaseCPU::registerThreadContexts()
 int
 BaseCPU::findContext(ThreadContext *tc)
 {
-    for (int i = 0; i < threadContexts.size(); ++i) {
-        if (tc == threadContexts[i])
-            return i;
+    ThreadID size = threadContexts.size();
+    for (ThreadID tid = 0; tid < size; ++tid) {
+        if (tc == threadContexts[tid])
+            return tid;
     }
     return 0;
 }
@@ -312,11 +334,8 @@ BaseCPU::findContext(ThreadContext *tc)
 void
 BaseCPU::switchOut()
 {
-//    panic("This CPU doesn't support sampling!");
-#if FULL_SYSTEM
     if (profileEvent && profileEvent->scheduled())
-        profileEvent->deschedule();
-#endif
+        deschedule(profileEvent);
 }
 
 void
@@ -324,7 +343,10 @@ BaseCPU::takeOverFrom(BaseCPU *oldCPU, Port *ic, Port *dc)
 {
     assert(threadContexts.size() == oldCPU->threadContexts.size());
 
-    for (int i = 0; i < threadContexts.size(); ++i) {
+    _cpuId = oldCPU->cpuId();
+
+    ThreadID size = threadContexts.size();
+    for (ThreadID i = 0; i < size; ++i) {
         ThreadContext *newTC = threadContexts[i];
         ThreadContext *oldTC = oldCPU->threadContexts[i];
 
@@ -332,106 +354,96 @@ BaseCPU::takeOverFrom(BaseCPU *oldCPU, Port *ic, Port *dc)
 
         CpuEvent::replaceThreadContext(oldTC, newTC);
 
-        assert(newTC->readCpuId() == oldTC->readCpuId());
-#if FULL_SYSTEM
-        system->replaceThreadContext(newTC, newTC->readCpuId());
-#else
-        assert(newTC->getProcessPtr() == oldTC->getProcessPtr());
-        newTC->getProcessPtr()->replaceThreadContext(newTC, newTC->readCpuId());
-#endif
-
-//    TheISA::compareXCs(oldXC, newXC);
+        assert(newTC->contextId() == oldTC->contextId());
+        assert(newTC->threadId() == oldTC->threadId());
+        system->replaceThreadContext(newTC, newTC->contextId());
+
+        /* This code no longer works since the zero register (e.g.,
+         * r31 on Alpha) doesn't necessarily contain zero at this
+         * point.
+           if (DTRACE(Context))
+            ThreadContext::compare(oldTC, newTC);
+        */
+
+        Port  *old_itb_port, *old_dtb_port, *new_itb_port, *new_dtb_port;
+        old_itb_port = oldTC->getITBPtr()->getPort();
+        old_dtb_port = oldTC->getDTBPtr()->getPort();
+        new_itb_port = newTC->getITBPtr()->getPort();
+        new_dtb_port = newTC->getDTBPtr()->getPort();
+
+        // Move over any table walker ports if they exist
+        if (new_itb_port && !new_itb_port->isConnected()) {
+            assert(old_itb_port);
+            Port *peer = old_itb_port->getPeer();;
+            new_itb_port->setPeer(peer);
+            peer->setPeer(new_itb_port);
+        }
+        if (new_dtb_port && !new_dtb_port->isConnected()) {
+            assert(old_dtb_port);
+            Port *peer = old_dtb_port->getPeer();;
+            new_dtb_port->setPeer(peer);
+            peer->setPeer(new_dtb_port);
+        }
     }
 
-#if FULL_SYSTEM
     interrupts = oldCPU->interrupts;
+    interrupts->setCPU(this);
 
-    for (int i = 0; i < threadContexts.size(); ++i)
-        threadContexts[i]->profileClear();
+    if (FullSystem) {
+        for (ThreadID i = 0; i < size; ++i)
+            threadContexts[i]->profileClear();
 
-    // The Sampler must take care of this!
-//    if (profileEvent)
-//        profileEvent->schedule(curTick);
-#endif
+        if (profileEvent)
+            schedule(profileEvent, curTick());
+    }
 
     // Connect new CPU to old CPU's memory only if new CPU isn't
     // connected to anything.  Also connect old CPU's memory to new
     // CPU.
-    Port *peer;
-    if (ic->getPeer() == NULL) {
-        peer = oldCPU->getPort("icache_port")->getPeer();
+    if (!ic->isConnected()) {
+        Port *peer = oldCPU->getPort("icache_port")->getPeer();
         ic->setPeer(peer);
-    } else {
-        peer = ic->getPeer();
+        peer->setPeer(ic);
     }
-    peer->setPeer(ic);
 
-    if (dc->getPeer() == NULL) {
-        peer = oldCPU->getPort("dcache_port")->getPeer();
+    if (!dc->isConnected()) {
+        Port *peer = oldCPU->getPort("dcache_port")->getPeer();
         dc->setPeer(peer);
-    } else {
-        peer = dc->getPeer();
+        peer->setPeer(dc);
     }
-    peer->setPeer(dc);
 }
 
 
-#if FULL_SYSTEM
-BaseCPU::ProfileEvent::ProfileEvent(BaseCPU *_cpu, int _interval)
-    : Event(&mainEventQueue), cpu(_cpu), interval(_interval)
+BaseCPU::ProfileEvent::ProfileEvent(BaseCPU *_cpu, Tick _interval)
+    : cpu(_cpu), interval(_interval)
 { }
 
 void
 BaseCPU::ProfileEvent::process()
 {
-    for (int i = 0, size = cpu->threadContexts.size(); i < size; ++i) {
+    ThreadID size = cpu->threadContexts.size();
+    for (ThreadID i = 0; i < size; ++i) {
         ThreadContext *tc = cpu->threadContexts[i];
         tc->profileSample();
     }
 
-    schedule(curTick + interval);
-}
-
-void
-BaseCPU::post_interrupt(int int_num, int index)
-{
-    interrupts.post(int_num, index);
-}
-
-void
-BaseCPU::clear_interrupt(int int_num, int index)
-{
-    interrupts.clear(int_num, index);
-}
-
-void
-BaseCPU::clear_interrupts()
-{
-    interrupts.clear_all();
-}
-
-uint64_t
-BaseCPU::get_interrupts(int int_num)
-{
-    return interrupts.get_vec(int_num);
+    cpu->schedule(this, curTick() + interval);
 }
 
 void
 BaseCPU::serialize(std::ostream &os)
 {
     SERIALIZE_SCALAR(instCnt);
-    interrupts.serialize(os);
+    interrupts->serialize(os);
 }
 
 void
 BaseCPU::unserialize(Checkpoint *cp, const std::string &section)
 {
     UNSERIALIZE_SCALAR(instCnt);
-    interrupts.unserialize(cp, section);
+    interrupts->unserialize(cp, section);
 }
 
-#endif // FULL_SYSTEM
-
 void
 BaseCPU::traceFunctionsInternal(Addr pc)
 {
@@ -454,10 +466,7 @@ BaseCPU::traceFunctionsInternal(Addr pc)
         }
 
         ccprintf(*functionTraceStream, " (%d)\n%d: %s",
-                 curTick - functionEntryTick, curTick, sym_str);
-        functionEntryTick = curTick;
+                 curTick() - functionEntryTick, curTick(), sym_str);
+        functionEntryTick = curTick();
     }
 }
-
-
-DEFINE_SIM_OBJECT_CLASS_NAME("BaseCPU", BaseCPU)