/*
+ * Copyright (c) 2011-2012 ARM Limited
+ * All rights reserved
+ *
+ * The license below extends only to copyright in the software and shall
+ * not be construed as granting a license to any other intellectual
+ * property including but not limited to intellectual property relating
+ * to a hardware implementation of the functionality of the software
+ * licensed hereunder. You may use the software subject to the license
+ * terms below provided that you ensure that this notice is replicated
+ * unmodified and in its entirety in all distributions of the software,
+ * modified or unmodified, in source code or in binary form.
+ *
* Copyright (c) 2002-2005 The Regents of The University of Michigan
+ * Copyright (c) 2011 Regents of the University of California
+ * Copyright (c) 2013 Advanced Micro Devices, Inc.
+ * Copyright (c) 2013 Mark D. Hill and David A. Wood
* 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/checker/cpu.hh"
#include "cpu/base.hh"
#include "cpu/cpuevent.hh"
-#include "cpu/thread_context.hh"
#include "cpu/profile.hh"
+#include "cpu/thread_context.hh"
+#include "debug/Mwait.hh"
+#include "debug/SyscallVerbose.hh"
+#include "mem/page_table.hh"
#include "params/BaseCPU.hh"
-#include "sim/sim_exit.hh"
+#include "sim/full_system.hh"
#include "sim/process.hh"
#include "sim/sim_events.hh"
+#include "sim/sim_exit.hh"
#include "sim/system.hh"
// Hack
cpu(_cpu), _repeatEvent(true)
{
if (_interval)
- cpu->schedule(this, curTick + _interval);
+ cpu->schedule(this, curTick() + _interval);
}
void
CPUProgressEvent::process()
{
- Counter temp = cpu->totalInstructions();
+ Counter temp = cpu->totalOps();
#ifndef NDEBUG
- double ipc = double(temp - lastNumInst) / (_interval / cpu->ticks(1));
+ double ipc = double(temp - lastNumInst) / (_interval / cpu->clockPeriod());
DPRINTFN("%s progress event, total committed:%i, progress insts committed: "
"%lli, IPC: %0.8d\n", cpu->name(), temp, temp - lastNumInst,
ipc = 0.0;
#else
cprintf("%lli: %s progress event, total committed:%i, progress insts "
- "committed: %lli\n", curTick, cpu->name(), temp,
+ "committed: %lli\n", curTick(), cpu->name(), temp,
temp - lastNumInst);
#endif
lastNumInst = temp;
if (_repeatEvent)
- cpu->schedule(this, curTick + _interval);
+ cpu->schedule(this, curTick() + _interval);
}
const char *
return "CPU Progress";
}
-#if FULL_SYSTEM
-BaseCPU::BaseCPU(Params *p)
- : MemObject(p), clock(p->clock), instCnt(0), _cpuId(p->cpu_id),
- interrupts(p->interrupts),
- numThreads(p->numThreads), system(p->system),
- phase(p->phase)
-#else
-BaseCPU::BaseCPU(Params *p)
- : MemObject(p), clock(p->clock), _cpuId(p->cpu_id),
+BaseCPU::BaseCPU(Params *p, bool is_checker)
+ : MemObject(p), instCnt(0), _cpuId(p->cpu_id), _socketId(p->socket_id),
+ _instMasterId(p->system->getMasterId(name() + ".inst")),
+ _dataMasterId(p->system->getMasterId(name() + ".data")),
+ _taskId(ContextSwitchTaskId::Unknown), _pid(Request::invldPid),
+ _switchedOut(p->switched_out), _cacheLineSize(p->system->cacheLineSize()),
+ interrupts(p->interrupts), profileEvent(NULL),
numThreads(p->numThreads), system(p->system),
- phase(p->phase)
-#endif
+ functionTraceStream(nullptr), currentFunctionStart(0),
+ currentFunctionEnd(0), functionEntryTick(0),
+ addressMonitor()
{
-// 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(SyscallVerbose, "Constructing CPU with id %d\n", _cpuId);
+ DPRINTF(SyscallVerbose, "Constructing CPU with id %d, socket id %d\n",
+ _cpuId, _socketId);
if (numThreads > maxThreadsPerCPU)
maxThreadsPerCPU = numThreads;
//
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);
- }
+ for (ThreadID tid = 0; tid < numThreads; ++tid)
+ scheduleInstStop(tid, p->max_insts_any_thread, cause);
+ }
+
+ // Set up instruction-count-based termination events for SimPoints
+ // Typically, there are more than one action points.
+ // Simulation.py is responsible to take the necessary actions upon
+ // exitting the simulation loop.
+ if (!p->simpoint_start_insts.empty()) {
+ const char *cause = "simpoint starting point found";
+ for (size_t i = 0; i < p->simpoint_start_insts.size(); ++i)
+ scheduleInstStop(0, p->simpoint_start_insts[i], cause);
}
if (p->max_insts_all_threads != 0) {
*counter = numThreads;
for (ThreadID tid = 0; tid < numThreads; ++tid) {
Event *event = new CountedExitEvent(cause, *counter);
- comInstEventQueue[tid]->schedule(event, p->max_insts_any_thread);
+ comInstEventQueue[tid]->schedule(event, p->max_insts_all_threads);
}
}
//
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);
- }
+ for (ThreadID tid = 0; tid < numThreads; ++tid)
+ scheduleLoadStop(tid, p->max_loads_any_thread, cause);
}
if (p->max_loads_all_threads != 0) {
functionTracingEnabled = false;
if (p->function_trace) {
- functionTraceStream = simout.find(csprintf("ftrace.%s", name()));
+ const string fname = csprintf("ftrace.%s", name());
+ functionTraceStream = simout.find(fname);
+ if (!functionTraceStream)
+ functionTraceStream = simout.create(fname);
+
currentFunctionStart = currentFunctionEnd = 0;
functionEntryTick = p->function_trace_start;
schedule(event, p->function_trace_start);
}
}
-#if FULL_SYSTEM
- interrupts->setCPU(this);
- profileEvent = NULL;
- if (params()->profile)
- profileEvent = new ProfileEvent(this, params()->profile);
-#endif
+ // The interrupts should always be present unless this CPU is
+ // switched in later or in case it is a checker CPU
+ if (!params()->switched_out && !is_checker) {
+ if (interrupts) {
+ interrupts->setCPU(this);
+ } else {
+ fatal("CPU %s has no interrupt controller.\n"
+ "Ensure createInterruptController() is called.\n", name());
+ }
+ }
+
+ if (FullSystem) {
+ if (params()->profile)
+ profileEvent = new ProfileEvent(this, params()->profile);
+ }
tracer = params()->tracer;
+
+ if (params()->isa.size() != numThreads) {
+ fatal("Number of ISAs (%i) assigned to the CPU does not equal number "
+ "of threads (%i).\n", params()->isa.size(), numThreads);
+ }
}
void
BaseCPU::~BaseCPU()
{
+ delete profileEvent;
+ delete[] comLoadEventQueue;
+ delete[] comInstEventQueue;
+}
+
+void
+BaseCPU::armMonitor(Addr address)
+{
+ addressMonitor.armed = true;
+ addressMonitor.vAddr = address;
+ addressMonitor.pAddr = 0x0;
+ DPRINTF(Mwait,"Armed monitor (vAddr=0x%lx)\n", address);
+}
+
+bool
+BaseCPU::mwait(PacketPtr pkt)
+{
+ if(addressMonitor.gotWakeup == false) {
+ int block_size = cacheLineSize();
+ uint64_t mask = ~((uint64_t)(block_size - 1));
+
+ assert(pkt->req->hasPaddr());
+ addressMonitor.pAddr = pkt->getAddr() & mask;
+ addressMonitor.waiting = true;
+
+ DPRINTF(Mwait,"mwait called (vAddr=0x%lx, line's paddr=0x%lx)\n",
+ addressMonitor.vAddr, addressMonitor.pAddr);
+ return true;
+ } else {
+ addressMonitor.gotWakeup = false;
+ return false;
+ }
+}
+
+void
+BaseCPU::mwaitAtomic(ThreadContext *tc, TheISA::TLB *dtb)
+{
+ Request req;
+ Addr addr = addressMonitor.vAddr;
+ int block_size = cacheLineSize();
+ uint64_t mask = ~((uint64_t)(block_size - 1));
+ int size = block_size;
+
+ //The address of the next line if it crosses a cache line boundary.
+ Addr secondAddr = roundDown(addr + size - 1, block_size);
+
+ if (secondAddr > addr)
+ size = secondAddr - addr;
+
+ req.setVirt(0, addr, size, 0x0, dataMasterId(), tc->instAddr());
+
+ // translate to physical address
+ Fault fault = dtb->translateAtomic(&req, tc, BaseTLB::Read);
+ assert(fault == NoFault);
+
+ addressMonitor.pAddr = req.getPaddr() & mask;
+ addressMonitor.waiting = true;
+
+ DPRINTF(Mwait,"mwait called (vAddr=0x%lx, line's paddr=0x%lx)\n",
+ addressMonitor.vAddr, addressMonitor.pAddr);
}
void
BaseCPU::init()
{
- if (!params()->defer_registration)
+ if (!params()->switched_out) {
registerThreadContexts();
+
+ verifyMemoryMode();
+ }
}
void
BaseCPU::startup()
{
-#if FULL_SYSTEM
- if (!params()->defer_registration && profileEvent)
- schedule(profileEvent, curTick);
-#endif
+ if (FullSystem) {
+ if (!params()->switched_out && profileEvent)
+ schedule(profileEvent, curTick());
+ }
if (params()->progress_interval) {
- Tick num_ticks = ticks(params()->progress_interval);
-
- Event *event;
- event = new CPUProgressEvent(this, num_ticks);
+ new CPUProgressEvent(this, params()->progress_interval);
}
}
+ProbePoints::PMUUPtr
+BaseCPU::pmuProbePoint(const char *name)
+{
+ ProbePoints::PMUUPtr ptr;
+ ptr.reset(new ProbePoints::PMU(getProbeManager(), name));
+
+ return ptr;
+}
+
+void
+BaseCPU::regProbePoints()
+{
+ ppCycles = pmuProbePoint("Cycles");
+
+ ppRetiredInsts = pmuProbePoint("RetiredInsts");
+ ppRetiredLoads = pmuProbePoint("RetiredLoads");
+ ppRetiredStores = pmuProbePoint("RetiredStores");
+ ppRetiredBranches = pmuProbePoint("RetiredBranches");
+}
+
+void
+BaseCPU::probeInstCommit(const StaticInstPtr &inst)
+{
+ if (!inst->isMicroop() || inst->isLastMicroop())
+ ppRetiredInsts->notify(1);
+
+
+ if (inst->isLoad())
+ ppRetiredLoads->notify(1);
+
+ if (inst->isStore())
+ ppRetiredStores->notify(1);
+
+ if (inst->isControl())
+ ppRetiredBranches->notify(1);
+}
void
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) {
}
} else if (size == 1)
threadContexts[0]->regStats(name());
-
-#if FULL_SYSTEM
-#endif
}
-Tick
-BaseCPU::nextCycle()
+BaseMasterPort &
+BaseCPU::getMasterPort(const string &if_name, PortID idx)
{
- Tick next_tick = curTick - phase + clock - 1;
- next_tick -= (next_tick % clock);
- next_tick += phase;
- return next_tick;
-}
-
-Tick
-BaseCPU::nextCycle(Tick begin_tick)
-{
- Tick next_tick = begin_tick;
- if (next_tick % clock != 0)
- next_tick = next_tick - (next_tick % clock) + clock;
- next_tick += phase;
-
- assert(next_tick >= curTick);
- return next_tick;
+ // Get the right port based on name. This applies to all the
+ // subclasses of the base CPU and relies on their implementation
+ // of getDataPort and getInstPort. In all cases there methods
+ // return a MasterPort pointer.
+ if (if_name == "dcache_port")
+ return getDataPort();
+ else if (if_name == "icache_port")
+ return getInstPort();
+ else
+ return MemObject::getMasterPort(if_name, idx);
}
void
tc->setContextId(system->registerThreadContext(tc, _cpuId));
else
tc->setContextId(system->registerThreadContext(tc));
-#if !FULL_SYSTEM
- tc->getProcessPtr()->assignThreadContext(tc->contextId());
-#endif
+
+ if (!FullSystem)
+ tc->getProcessPtr()->assignThreadContext(tc->contextId());
}
}
void
BaseCPU::switchOut()
{
-// panic("This CPU doesn't support sampling!");
-#if FULL_SYSTEM
+ assert(!_switchedOut);
+ _switchedOut = true;
if (profileEvent && profileEvent->scheduled())
deschedule(profileEvent);
-#endif
+
+ // Flush all TLBs in the CPU to avoid having stale translations if
+ // it gets switched in later.
+ flushTLBs();
}
void
-BaseCPU::takeOverFrom(BaseCPU *oldCPU, Port *ic, Port *dc)
+BaseCPU::takeOverFrom(BaseCPU *oldCPU)
{
assert(threadContexts.size() == oldCPU->threadContexts.size());
-
- _cpuId = oldCPU->cpuId();
+ assert(_cpuId == oldCPU->cpuId());
+ assert(_switchedOut);
+ assert(oldCPU != this);
+ _pid = oldCPU->getPid();
+ _taskId = oldCPU->taskId();
+ _switchedOut = false;
ThreadID size = threadContexts.size();
for (ThreadID i = 0; i < size; ++i) {
if (DTRACE(Context))
ThreadContext::compare(oldTC, newTC);
*/
+
+ BaseMasterPort *old_itb_port = oldTC->getITBPtr()->getMasterPort();
+ BaseMasterPort *old_dtb_port = oldTC->getDTBPtr()->getMasterPort();
+ BaseMasterPort *new_itb_port = newTC->getITBPtr()->getMasterPort();
+ BaseMasterPort *new_dtb_port = newTC->getDTBPtr()->getMasterPort();
+
+ // Move over any table walker ports if they exist
+ if (new_itb_port) {
+ assert(!new_itb_port->isConnected());
+ assert(old_itb_port);
+ assert(old_itb_port->isConnected());
+ BaseSlavePort &slavePort = old_itb_port->getSlavePort();
+ old_itb_port->unbind();
+ new_itb_port->bind(slavePort);
+ }
+ if (new_dtb_port) {
+ assert(!new_dtb_port->isConnected());
+ assert(old_dtb_port);
+ assert(old_dtb_port->isConnected());
+ BaseSlavePort &slavePort = old_dtb_port->getSlavePort();
+ old_dtb_port->unbind();
+ new_dtb_port->bind(slavePort);
+ }
+ newTC->getITBPtr()->takeOverFrom(oldTC->getITBPtr());
+ newTC->getDTBPtr()->takeOverFrom(oldTC->getDTBPtr());
+
+ // Checker whether or not we have to transfer CheckerCPU
+ // objects over in the switch
+ CheckerCPU *oldChecker = oldTC->getCheckerCpuPtr();
+ CheckerCPU *newChecker = newTC->getCheckerCpuPtr();
+ if (oldChecker && newChecker) {
+ BaseMasterPort *old_checker_itb_port =
+ oldChecker->getITBPtr()->getMasterPort();
+ BaseMasterPort *old_checker_dtb_port =
+ oldChecker->getDTBPtr()->getMasterPort();
+ BaseMasterPort *new_checker_itb_port =
+ newChecker->getITBPtr()->getMasterPort();
+ BaseMasterPort *new_checker_dtb_port =
+ newChecker->getDTBPtr()->getMasterPort();
+
+ newChecker->getITBPtr()->takeOverFrom(oldChecker->getITBPtr());
+ newChecker->getDTBPtr()->takeOverFrom(oldChecker->getDTBPtr());
+
+ // Move over any table walker ports if they exist for checker
+ if (new_checker_itb_port) {
+ assert(!new_checker_itb_port->isConnected());
+ assert(old_checker_itb_port);
+ assert(old_checker_itb_port->isConnected());
+ BaseSlavePort &slavePort =
+ old_checker_itb_port->getSlavePort();
+ old_checker_itb_port->unbind();
+ new_checker_itb_port->bind(slavePort);
+ }
+ if (new_checker_dtb_port) {
+ assert(!new_checker_dtb_port->isConnected());
+ assert(old_checker_dtb_port);
+ assert(old_checker_dtb_port->isConnected());
+ BaseSlavePort &slavePort =
+ old_checker_dtb_port->getSlavePort();
+ old_checker_dtb_port->unbind();
+ new_checker_dtb_port->bind(slavePort);
+ }
+ }
}
-#if FULL_SYSTEM
interrupts = oldCPU->interrupts;
interrupts->setCPU(this);
+ oldCPU->interrupts = NULL;
- for (ThreadID i = 0; i < size; ++i)
- threadContexts[i]->profileClear();
-
- if (profileEvent)
- schedule(profileEvent, curTick);
-#endif
+ if (FullSystem) {
+ for (ThreadID i = 0; i < size; ++i)
+ threadContexts[i]->profileClear();
- // 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.
- if (!ic->isConnected()) {
- Port *peer = oldCPU->getPort("icache_port")->getPeer();
- ic->setPeer(peer);
- peer->setPeer(ic);
+ if (profileEvent)
+ schedule(profileEvent, curTick());
}
- if (!dc->isConnected()) {
- Port *peer = oldCPU->getPort("dcache_port")->getPeer();
- dc->setPeer(peer);
- peer->setPeer(dc);
+ // All CPUs have an instruction and a data port, and the new CPU's
+ // ports are dangling while the old CPU has its ports connected
+ // already. Unbind the old CPU and then bind the ports of the one
+ // we are switching to.
+ assert(!getInstPort().isConnected());
+ assert(oldCPU->getInstPort().isConnected());
+ BaseSlavePort &inst_peer_port = oldCPU->getInstPort().getSlavePort();
+ oldCPU->getInstPort().unbind();
+ getInstPort().bind(inst_peer_port);
+
+ assert(!getDataPort().isConnected());
+ assert(oldCPU->getDataPort().isConnected());
+ BaseSlavePort &data_peer_port = oldCPU->getDataPort().getSlavePort();
+ oldCPU->getDataPort().unbind();
+ getDataPort().bind(data_peer_port);
+}
+
+void
+BaseCPU::flushTLBs()
+{
+ for (ThreadID i = 0; i < threadContexts.size(); ++i) {
+ ThreadContext &tc(*threadContexts[i]);
+ CheckerCPU *checker(tc.getCheckerCpuPtr());
+
+ tc.getITBPtr()->flushAll();
+ tc.getDTBPtr()->flushAll();
+ if (checker) {
+ checker->getITBPtr()->flushAll();
+ checker->getDTBPtr()->flushAll();
+ }
}
}
-#if FULL_SYSTEM
BaseCPU::ProfileEvent::ProfileEvent(BaseCPU *_cpu, Tick _interval)
: cpu(_cpu), interval(_interval)
{ }
tc->profileSample();
}
- cpu->schedule(this, curTick + interval);
+ cpu->schedule(this, curTick() + interval);
}
void
BaseCPU::serialize(std::ostream &os)
{
SERIALIZE_SCALAR(instCnt);
- interrupts->serialize(os);
+
+ if (!_switchedOut) {
+ /* Unlike _pid, _taskId is not serialized, as they are dynamically
+ * assigned unique ids that are only meaningful for the duration of
+ * a specific run. We will need to serialize the entire taskMap in
+ * system. */
+ SERIALIZE_SCALAR(_pid);
+
+ interrupts->serialize(os);
+
+ // Serialize the threads, this is done by the CPU implementation.
+ for (ThreadID i = 0; i < numThreads; ++i) {
+ nameOut(os, csprintf("%s.xc.%i", name(), i));
+ serializeThread(os, i);
+ }
+ }
}
void
BaseCPU::unserialize(Checkpoint *cp, const std::string §ion)
{
UNSERIALIZE_SCALAR(instCnt);
- interrupts->unserialize(cp, section);
+
+ if (!_switchedOut) {
+ UNSERIALIZE_SCALAR(_pid);
+ interrupts->unserialize(cp, section);
+
+ // Unserialize the threads, this is done by the CPU implementation.
+ for (ThreadID i = 0; i < numThreads; ++i)
+ unserializeThread(cp, csprintf("%s.xc.%i", section, i), i);
+ }
+}
+
+void
+BaseCPU::scheduleInstStop(ThreadID tid, Counter insts, const char *cause)
+{
+ const Tick now(comInstEventQueue[tid]->getCurTick());
+ Event *event(new LocalSimLoopExitEvent(cause, 0));
+
+ comInstEventQueue[tid]->schedule(event, now + insts);
+}
+
+AddressMonitor::AddressMonitor() {
+ armed = false;
+ waiting = false;
+ gotWakeup = false;
+}
+
+bool AddressMonitor::doMonitor(PacketPtr pkt) {
+ assert(pkt->req->hasPaddr());
+ if(armed && waiting) {
+ if(pAddr == pkt->getAddr()) {
+ DPRINTF(Mwait,"pAddr=0x%lx invalidated: waking up core\n",
+ pkt->getAddr());
+ waiting = false;
+ return true;
+ }
+ }
+ return false;
+}
+
+void
+BaseCPU::scheduleLoadStop(ThreadID tid, Counter loads, const char *cause)
+{
+ const Tick now(comLoadEventQueue[tid]->getCurTick());
+ Event *event(new LocalSimLoopExitEvent(cause, 0));
+
+ comLoadEventQueue[tid]->schedule(event, now + loads);
}
-#endif // FULL_SYSTEM
void
BaseCPU::traceFunctionsInternal(Addr pc)
}
ccprintf(*functionTraceStream, " (%d)\n%d: %s",
- curTick - functionEntryTick, curTick, sym_str);
- functionEntryTick = curTick;
+ curTick() - functionEntryTick, curTick(), sym_str);
+ functionEntryTick = curTick();
}
}