Sequencer.cc revision 11019:fc1e41e88fd3
1/*
2 * Copyright (c) 1999-2008 Mark D. Hill and David A. Wood
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met: redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer;
9 * redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution;
12 * neither the name of the copyright holders nor the names of its
13 * contributors may be used to endorse or promote products derived from
14 * this software without specific prior written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29#include "arch/x86/ldstflags.hh"
30#include "base/misc.hh"
31#include "base/str.hh"
32#include "cpu/testers/rubytest/RubyTester.hh"
33#include "debug/MemoryAccess.hh"
34#include "debug/ProtocolTrace.hh"
35#include "debug/RubySequencer.hh"
36#include "debug/RubyStats.hh"
37#include "mem/protocol/PrefetchBit.hh"
38#include "mem/protocol/RubyAccessMode.hh"
39#include "mem/ruby/profiler/Profiler.hh"
40#include "mem/ruby/slicc_interface/RubyRequest.hh"
41#include "mem/ruby/system/Sequencer.hh"
42#include "mem/ruby/system/System.hh"
43#include "mem/packet.hh"
44#include "sim/system.hh"
45
46using namespace std;
47
48Sequencer *
49RubySequencerParams::create()
50{
51    return new Sequencer(this);
52}
53
54Sequencer::Sequencer(const Params *p)
55    : RubyPort(p), m_IncompleteTimes(MachineType_NUM), deadlockCheckEvent(this)
56{
57    m_outstanding_count = 0;
58
59    m_instCache_ptr = p->icache;
60    m_dataCache_ptr = p->dcache;
61    m_data_cache_hit_latency = p->dcache_hit_latency;
62    m_inst_cache_hit_latency = p->icache_hit_latency;
63    m_max_outstanding_requests = p->max_outstanding_requests;
64    m_deadlock_threshold = p->deadlock_threshold;
65
66    assert(m_max_outstanding_requests > 0);
67    assert(m_deadlock_threshold > 0);
68    assert(m_instCache_ptr != NULL);
69    assert(m_dataCache_ptr != NULL);
70    assert(m_data_cache_hit_latency > 0);
71    assert(m_inst_cache_hit_latency > 0);
72
73    m_usingNetworkTester = p->using_network_tester;
74}
75
76Sequencer::~Sequencer()
77{
78}
79
80void
81Sequencer::wakeup()
82{
83    assert(drainState() != DrainState::Draining);
84
85    // Check for deadlock of any of the requests
86    Cycles current_time = curCycle();
87
88    // Check across all outstanding requests
89    int total_outstanding = 0;
90
91    RequestTable::iterator read = m_readRequestTable.begin();
92    RequestTable::iterator read_end = m_readRequestTable.end();
93    for (; read != read_end; ++read) {
94        SequencerRequest* request = read->second;
95        if (current_time - request->issue_time < m_deadlock_threshold)
96            continue;
97
98        panic("Possible Deadlock detected. Aborting!\n"
99             "version: %d request.paddr: 0x%x m_readRequestTable: %d "
100             "current time: %u issue_time: %d difference: %d\n", m_version,
101             Address(request->pkt->getAddr()), m_readRequestTable.size(),
102              current_time * clockPeriod(), request->issue_time * clockPeriod(),
103              (current_time * clockPeriod()) - (request->issue_time * clockPeriod()));
104    }
105
106    RequestTable::iterator write = m_writeRequestTable.begin();
107    RequestTable::iterator write_end = m_writeRequestTable.end();
108    for (; write != write_end; ++write) {
109        SequencerRequest* request = write->second;
110        if (current_time - request->issue_time < m_deadlock_threshold)
111            continue;
112
113        panic("Possible Deadlock detected. Aborting!\n"
114             "version: %d request.paddr: 0x%x m_writeRequestTable: %d "
115             "current time: %u issue_time: %d difference: %d\n", m_version,
116             Address(request->pkt->getAddr()), m_writeRequestTable.size(),
117              current_time * clockPeriod(), request->issue_time * clockPeriod(),
118              (current_time * clockPeriod()) - (request->issue_time * clockPeriod()));
119    }
120
121    total_outstanding += m_writeRequestTable.size();
122    total_outstanding += m_readRequestTable.size();
123
124    assert(m_outstanding_count == total_outstanding);
125
126    if (m_outstanding_count > 0) {
127        // If there are still outstanding requests, keep checking
128        schedule(deadlockCheckEvent, clockEdge(m_deadlock_threshold));
129    }
130}
131
132void Sequencer::resetStats()
133{
134    m_latencyHist.reset();
135    m_hitLatencyHist.reset();
136    m_missLatencyHist.reset();
137    for (int i = 0; i < RubyRequestType_NUM; i++) {
138        m_typeLatencyHist[i]->reset();
139        m_hitTypeLatencyHist[i]->reset();
140        m_missTypeLatencyHist[i]->reset();
141        for (int j = 0; j < MachineType_NUM; j++) {
142            m_hitTypeMachLatencyHist[i][j]->reset();
143            m_missTypeMachLatencyHist[i][j]->reset();
144        }
145    }
146
147    for (int i = 0; i < MachineType_NUM; i++) {
148        m_missMachLatencyHist[i]->reset();
149        m_hitMachLatencyHist[i]->reset();
150
151        m_IssueToInitialDelayHist[i]->reset();
152        m_InitialToForwardDelayHist[i]->reset();
153        m_ForwardToFirstResponseDelayHist[i]->reset();
154        m_FirstResponseToCompletionDelayHist[i]->reset();
155
156        m_IncompleteTimes[i] = 0;
157    }
158}
159
160void
161Sequencer::printProgress(ostream& out) const
162{
163#if 0
164    int total_demand = 0;
165    out << "Sequencer Stats Version " << m_version << endl;
166    out << "Current time = " << m_ruby_system->getTime() << endl;
167    out << "---------------" << endl;
168    out << "outstanding requests" << endl;
169
170    out << "proc " << m_Read
171        << " version Requests = " << m_readRequestTable.size() << endl;
172
173    // print the request table
174    RequestTable::iterator read = m_readRequestTable.begin();
175    RequestTable::iterator read_end = m_readRequestTable.end();
176    for (; read != read_end; ++read) {
177        SequencerRequest* request = read->second;
178        out << "\tRequest[ " << i << " ] = " << request->type
179            << " Address " << rkeys[i]
180            << " Posted " << request->issue_time
181            << " PF " << PrefetchBit_No << endl;
182        total_demand++;
183    }
184
185    out << "proc " << m_version
186        << " Write Requests = " << m_writeRequestTable.size << endl;
187
188    // print the request table
189    RequestTable::iterator write = m_writeRequestTable.begin();
190    RequestTable::iterator write_end = m_writeRequestTable.end();
191    for (; write != write_end; ++write) {
192        SequencerRequest* request = write->second;
193        out << "\tRequest[ " << i << " ] = " << request.getType()
194            << " Address " << wkeys[i]
195            << " Posted " << request.getTime()
196            << " PF " << request.getPrefetch() << endl;
197        if (request.getPrefetch() == PrefetchBit_No) {
198            total_demand++;
199        }
200    }
201
202    out << endl;
203
204    out << "Total Number Outstanding: " << m_outstanding_count << endl
205        << "Total Number Demand     : " << total_demand << endl
206        << "Total Number Prefetches : " << m_outstanding_count - total_demand
207        << endl << endl << endl;
208#endif
209}
210
211// Insert the request on the correct request table.  Return true if
212// the entry was already present.
213RequestStatus
214Sequencer::insertRequest(PacketPtr pkt, RubyRequestType request_type)
215{
216    assert(m_outstanding_count ==
217        (m_writeRequestTable.size() + m_readRequestTable.size()));
218
219    // See if we should schedule a deadlock check
220    if (!deadlockCheckEvent.scheduled() &&
221        drainState() != DrainState::Draining) {
222        schedule(deadlockCheckEvent, clockEdge(m_deadlock_threshold));
223    }
224
225    Address line_addr(pkt->getAddr());
226    line_addr.makeLineAddress();
227    // Create a default entry, mapping the address to NULL, the cast is
228    // there to make gcc 4.4 happy
229    RequestTable::value_type default_entry(line_addr,
230                                           (SequencerRequest*) NULL);
231
232    if ((request_type == RubyRequestType_ST) ||
233        (request_type == RubyRequestType_RMW_Read) ||
234        (request_type == RubyRequestType_RMW_Write) ||
235        (request_type == RubyRequestType_Load_Linked) ||
236        (request_type == RubyRequestType_Store_Conditional) ||
237        (request_type == RubyRequestType_Locked_RMW_Read) ||
238        (request_type == RubyRequestType_Locked_RMW_Write) ||
239        (request_type == RubyRequestType_FLUSH)) {
240
241        // Check if there is any outstanding read request for the same
242        // cache line.
243        if (m_readRequestTable.count(line_addr) > 0) {
244            m_store_waiting_on_load++;
245            return RequestStatus_Aliased;
246        }
247
248        pair<RequestTable::iterator, bool> r =
249            m_writeRequestTable.insert(default_entry);
250        if (r.second) {
251            RequestTable::iterator i = r.first;
252            i->second = new SequencerRequest(pkt, request_type, curCycle());
253            m_outstanding_count++;
254        } else {
255          // There is an outstanding write request for the cache line
256          m_store_waiting_on_store++;
257          return RequestStatus_Aliased;
258        }
259    } else {
260        // Check if there is any outstanding write request for the same
261        // cache line.
262        if (m_writeRequestTable.count(line_addr) > 0) {
263            m_load_waiting_on_store++;
264            return RequestStatus_Aliased;
265        }
266
267        pair<RequestTable::iterator, bool> r =
268            m_readRequestTable.insert(default_entry);
269
270        if (r.second) {
271            RequestTable::iterator i = r.first;
272            i->second = new SequencerRequest(pkt, request_type, curCycle());
273            m_outstanding_count++;
274        } else {
275            // There is an outstanding read request for the cache line
276            m_load_waiting_on_load++;
277            return RequestStatus_Aliased;
278        }
279    }
280
281    m_outstandReqHist.sample(m_outstanding_count);
282    assert(m_outstanding_count ==
283        (m_writeRequestTable.size() + m_readRequestTable.size()));
284
285    return RequestStatus_Ready;
286}
287
288void
289Sequencer::markRemoved()
290{
291    m_outstanding_count--;
292    assert(m_outstanding_count ==
293           m_writeRequestTable.size() + m_readRequestTable.size());
294}
295
296void
297Sequencer::removeRequest(SequencerRequest* srequest)
298{
299    assert(m_outstanding_count ==
300           m_writeRequestTable.size() + m_readRequestTable.size());
301
302    Address line_addr(srequest->pkt->getAddr());
303    line_addr.makeLineAddress();
304    if ((srequest->m_type == RubyRequestType_ST) ||
305        (srequest->m_type == RubyRequestType_RMW_Read) ||
306        (srequest->m_type == RubyRequestType_RMW_Write) ||
307        (srequest->m_type == RubyRequestType_Load_Linked) ||
308        (srequest->m_type == RubyRequestType_Store_Conditional) ||
309        (srequest->m_type == RubyRequestType_Locked_RMW_Read) ||
310        (srequest->m_type == RubyRequestType_Locked_RMW_Write)) {
311        m_writeRequestTable.erase(line_addr);
312    } else {
313        m_readRequestTable.erase(line_addr);
314    }
315
316    markRemoved();
317}
318
319void
320Sequencer::invalidateSC(const Address& address)
321{
322    RequestTable::iterator i = m_writeRequestTable.find(address);
323    if (i != m_writeRequestTable.end()) {
324        SequencerRequest* request = i->second;
325        // The controller has lost the coherence permissions, hence the lock
326        // on the cache line maintained by the cache should be cleared.
327        if (request->m_type == RubyRequestType_Store_Conditional) {
328            m_dataCache_ptr->clearLocked(address);
329        }
330    }
331}
332
333bool
334Sequencer::handleLlsc(const Address& address, SequencerRequest* request)
335{
336    //
337    // The success flag indicates whether the LLSC operation was successful.
338    // LL ops will always succeed, but SC may fail if the cache line is no
339    // longer locked.
340    //
341    bool success = true;
342    if (request->m_type == RubyRequestType_Store_Conditional) {
343        if (!m_dataCache_ptr->isLocked(address, m_version)) {
344            //
345            // For failed SC requests, indicate the failure to the cpu by
346            // setting the extra data to zero.
347            //
348            request->pkt->req->setExtraData(0);
349            success = false;
350        } else {
351            //
352            // For successful SC requests, indicate the success to the cpu by
353            // setting the extra data to one.
354            //
355            request->pkt->req->setExtraData(1);
356        }
357        //
358        // Independent of success, all SC operations must clear the lock
359        //
360        m_dataCache_ptr->clearLocked(address);
361    } else if (request->m_type == RubyRequestType_Load_Linked) {
362        //
363        // Note: To fully follow Alpha LLSC semantics, should the LL clear any
364        // previously locked cache lines?
365        //
366        m_dataCache_ptr->setLocked(address, m_version);
367    } else if ((m_dataCache_ptr->isTagPresent(address)) &&
368               (m_dataCache_ptr->isLocked(address, m_version))) {
369        //
370        // Normal writes should clear the locked address
371        //
372        m_dataCache_ptr->clearLocked(address);
373    }
374    return success;
375}
376
377void
378Sequencer::recordMissLatency(const Cycles cycles, const RubyRequestType type,
379                             const MachineType respondingMach,
380                             bool isExternalHit, Cycles issuedTime,
381                             Cycles initialRequestTime,
382                             Cycles forwardRequestTime,
383                             Cycles firstResponseTime, Cycles completionTime)
384{
385    m_latencyHist.sample(cycles);
386    m_typeLatencyHist[type]->sample(cycles);
387
388    if (isExternalHit) {
389        m_missLatencyHist.sample(cycles);
390        m_missTypeLatencyHist[type]->sample(cycles);
391
392        if (respondingMach != MachineType_NUM) {
393            m_missMachLatencyHist[respondingMach]->sample(cycles);
394            m_missTypeMachLatencyHist[type][respondingMach]->sample(cycles);
395
396            if ((issuedTime <= initialRequestTime) &&
397                (initialRequestTime <= forwardRequestTime) &&
398                (forwardRequestTime <= firstResponseTime) &&
399                (firstResponseTime <= completionTime)) {
400
401                m_IssueToInitialDelayHist[respondingMach]->sample(
402                    initialRequestTime - issuedTime);
403                m_InitialToForwardDelayHist[respondingMach]->sample(
404                    forwardRequestTime - initialRequestTime);
405                m_ForwardToFirstResponseDelayHist[respondingMach]->sample(
406                    firstResponseTime - forwardRequestTime);
407                m_FirstResponseToCompletionDelayHist[respondingMach]->sample(
408                    completionTime - firstResponseTime);
409            } else {
410                m_IncompleteTimes[respondingMach]++;
411            }
412        }
413    } else {
414        m_hitLatencyHist.sample(cycles);
415        m_hitTypeLatencyHist[type]->sample(cycles);
416
417        if (respondingMach != MachineType_NUM) {
418            m_hitMachLatencyHist[respondingMach]->sample(cycles);
419            m_hitTypeMachLatencyHist[type][respondingMach]->sample(cycles);
420        }
421    }
422}
423
424void
425Sequencer::writeCallback(const Address& address, DataBlock& data,
426                         const bool externalHit, const MachineType mach,
427                         const Cycles initialRequestTime,
428                         const Cycles forwardRequestTime,
429                         const Cycles firstResponseTime)
430{
431    assert(address == line_address(address));
432    assert(m_writeRequestTable.count(line_address(address)));
433
434    RequestTable::iterator i = m_writeRequestTable.find(address);
435    assert(i != m_writeRequestTable.end());
436    SequencerRequest* request = i->second;
437
438    m_writeRequestTable.erase(i);
439    markRemoved();
440
441    assert((request->m_type == RubyRequestType_ST) ||
442           (request->m_type == RubyRequestType_ATOMIC) ||
443           (request->m_type == RubyRequestType_RMW_Read) ||
444           (request->m_type == RubyRequestType_RMW_Write) ||
445           (request->m_type == RubyRequestType_Load_Linked) ||
446           (request->m_type == RubyRequestType_Store_Conditional) ||
447           (request->m_type == RubyRequestType_Locked_RMW_Read) ||
448           (request->m_type == RubyRequestType_Locked_RMW_Write) ||
449           (request->m_type == RubyRequestType_FLUSH));
450
451    //
452    // For Alpha, properly handle LL, SC, and write requests with respect to
453    // locked cache blocks.
454    //
455    // Not valid for Network_test protocl
456    //
457    bool success = true;
458    if(!m_usingNetworkTester)
459        success = handleLlsc(address, request);
460
461    if (request->m_type == RubyRequestType_Locked_RMW_Read) {
462        m_controller->blockOnQueue(address, m_mandatory_q_ptr);
463    } else if (request->m_type == RubyRequestType_Locked_RMW_Write) {
464        m_controller->unblock(address);
465    }
466
467    hitCallback(request, data, success, mach, externalHit,
468                initialRequestTime, forwardRequestTime, firstResponseTime);
469}
470
471void
472Sequencer::readCallback(const Address& address, DataBlock& data,
473                        bool externalHit, const MachineType mach,
474                        Cycles initialRequestTime,
475                        Cycles forwardRequestTime,
476                        Cycles firstResponseTime)
477{
478    assert(address == line_address(address));
479    assert(m_readRequestTable.count(line_address(address)));
480
481    RequestTable::iterator i = m_readRequestTable.find(address);
482    assert(i != m_readRequestTable.end());
483    SequencerRequest* request = i->second;
484
485    m_readRequestTable.erase(i);
486    markRemoved();
487
488    assert((request->m_type == RubyRequestType_LD) ||
489           (request->m_type == RubyRequestType_IFETCH));
490
491    hitCallback(request, data, true, mach, externalHit,
492                initialRequestTime, forwardRequestTime, firstResponseTime);
493}
494
495void
496Sequencer::hitCallback(SequencerRequest* srequest, DataBlock& data,
497                       bool llscSuccess,
498                       const MachineType mach, const bool externalHit,
499                       const Cycles initialRequestTime,
500                       const Cycles forwardRequestTime,
501                       const Cycles firstResponseTime)
502{
503    PacketPtr pkt = srequest->pkt;
504    Address request_address(pkt->getAddr());
505    Address request_line_address(pkt->getAddr());
506    request_line_address.makeLineAddress();
507    RubyRequestType type = srequest->m_type;
508    Cycles issued_time = srequest->issue_time;
509
510    // Set this cache entry to the most recently used
511    if (type == RubyRequestType_IFETCH) {
512        m_instCache_ptr->setMRU(request_line_address);
513    } else {
514        m_dataCache_ptr->setMRU(request_line_address);
515    }
516
517    assert(curCycle() >= issued_time);
518    Cycles total_latency = curCycle() - issued_time;
519
520    // Profile the latency for all demand accesses.
521    recordMissLatency(total_latency, type, mach, externalHit, issued_time,
522                      initialRequestTime, forwardRequestTime,
523                      firstResponseTime, curCycle());
524
525    DPRINTFR(ProtocolTrace, "%15s %3s %10s%20s %6s>%-6s %s %d cycles\n",
526             curTick(), m_version, "Seq",
527             llscSuccess ? "Done" : "SC_Failed", "", "",
528             request_address, total_latency);
529
530    // update the data unless it is a non-data-carrying flush
531    if (RubySystem::getWarmupEnabled()) {
532        data.setData(pkt->getConstPtr<uint8_t>(),
533                     request_address.getOffset(), pkt->getSize());
534    } else if (!pkt->isFlush()) {
535        if ((type == RubyRequestType_LD) ||
536            (type == RubyRequestType_IFETCH) ||
537            (type == RubyRequestType_RMW_Read) ||
538            (type == RubyRequestType_Locked_RMW_Read) ||
539            (type == RubyRequestType_Load_Linked)) {
540            memcpy(pkt->getPtr<uint8_t>(),
541                   data.getData(request_address.getOffset(), pkt->getSize()),
542                   pkt->getSize());
543            DPRINTF(RubySequencer, "read data %s\n", data);
544        } else {
545            data.setData(pkt->getConstPtr<uint8_t>(),
546                         request_address.getOffset(), pkt->getSize());
547            DPRINTF(RubySequencer, "set data %s\n", data);
548        }
549    }
550
551    // If using the RubyTester, update the RubyTester sender state's
552    // subBlock with the recieved data.  The tester will later access
553    // this state.
554    if (m_usingRubyTester) {
555        DPRINTF(RubySequencer, "hitCallback %s 0x%x using RubyTester\n",
556                pkt->cmdString(), pkt->getAddr());
557        RubyTester::SenderState* testerSenderState =
558            pkt->findNextSenderState<RubyTester::SenderState>();
559        assert(testerSenderState);
560        testerSenderState->subBlock.mergeFrom(data);
561    }
562
563    delete srequest;
564
565    RubySystem *rs = m_ruby_system;
566    if (RubySystem::getWarmupEnabled()) {
567        assert(pkt->req);
568        delete pkt->req;
569        delete pkt;
570        rs->m_cache_recorder->enqueueNextFetchRequest();
571    } else if (RubySystem::getCooldownEnabled()) {
572        delete pkt;
573        rs->m_cache_recorder->enqueueNextFlushRequest();
574    } else {
575        ruby_hit_callback(pkt);
576    }
577}
578
579bool
580Sequencer::empty() const
581{
582    return m_writeRequestTable.empty() && m_readRequestTable.empty();
583}
584
585RequestStatus
586Sequencer::makeRequest(PacketPtr pkt)
587{
588    if (m_outstanding_count >= m_max_outstanding_requests) {
589        return RequestStatus_BufferFull;
590    }
591
592    RubyRequestType primary_type = RubyRequestType_NULL;
593    RubyRequestType secondary_type = RubyRequestType_NULL;
594
595    if (pkt->isLLSC()) {
596        //
597        // Alpha LL/SC instructions need to be handled carefully by the cache
598        // coherence protocol to ensure they follow the proper semantics. In
599        // particular, by identifying the operations as atomic, the protocol
600        // should understand that migratory sharing optimizations should not
601        // be performed (i.e. a load between the LL and SC should not steal
602        // away exclusive permission).
603        //
604        if (pkt->isWrite()) {
605            DPRINTF(RubySequencer, "Issuing SC\n");
606            primary_type = RubyRequestType_Store_Conditional;
607        } else {
608            DPRINTF(RubySequencer, "Issuing LL\n");
609            assert(pkt->isRead());
610            primary_type = RubyRequestType_Load_Linked;
611        }
612        secondary_type = RubyRequestType_ATOMIC;
613    } else if (pkt->req->isLockedRMW()) {
614        //
615        // x86 locked instructions are translated to store cache coherence
616        // requests because these requests should always be treated as read
617        // exclusive operations and should leverage any migratory sharing
618        // optimization built into the protocol.
619        //
620        if (pkt->isWrite()) {
621            DPRINTF(RubySequencer, "Issuing Locked RMW Write\n");
622            primary_type = RubyRequestType_Locked_RMW_Write;
623        } else {
624            DPRINTF(RubySequencer, "Issuing Locked RMW Read\n");
625            assert(pkt->isRead());
626            primary_type = RubyRequestType_Locked_RMW_Read;
627        }
628        secondary_type = RubyRequestType_ST;
629    } else {
630        if (pkt->isRead()) {
631            if (pkt->req->isInstFetch()) {
632                primary_type = secondary_type = RubyRequestType_IFETCH;
633            } else {
634                bool storeCheck = false;
635                // only X86 need the store check
636                if (system->getArch() == Arch::X86ISA) {
637                    uint32_t flags = pkt->req->getFlags();
638                    storeCheck = flags &
639                        (X86ISA::StoreCheck << X86ISA::FlagShift);
640                }
641                if (storeCheck) {
642                    primary_type = RubyRequestType_RMW_Read;
643                    secondary_type = RubyRequestType_ST;
644                } else {
645                    primary_type = secondary_type = RubyRequestType_LD;
646                }
647            }
648        } else if (pkt->isWrite()) {
649            //
650            // Note: M5 packets do not differentiate ST from RMW_Write
651            //
652            primary_type = secondary_type = RubyRequestType_ST;
653        } else if (pkt->isFlush()) {
654          primary_type = secondary_type = RubyRequestType_FLUSH;
655        } else {
656            panic("Unsupported ruby packet type\n");
657        }
658    }
659
660    RequestStatus status = insertRequest(pkt, primary_type);
661    if (status != RequestStatus_Ready)
662        return status;
663
664    issueRequest(pkt, secondary_type);
665
666    // TODO: issue hardware prefetches here
667    return RequestStatus_Issued;
668}
669
670void
671Sequencer::issueRequest(PacketPtr pkt, RubyRequestType secondary_type)
672{
673    assert(pkt != NULL);
674    ContextID proc_id = pkt->req->hasContextId() ?
675        pkt->req->contextId() : InvalidContextID;
676
677    // If valid, copy the pc to the ruby request
678    Addr pc = 0;
679    if (pkt->req->hasPC()) {
680        pc = pkt->req->getPC();
681    }
682
683    // check if the packet has data as for example prefetch and flush
684    // requests do not
685    std::shared_ptr<RubyRequest> msg =
686        std::make_shared<RubyRequest>(clockEdge(), pkt->getAddr(),
687                                      pkt->isFlush() ?
688                                      nullptr : pkt->getPtr<uint8_t>(),
689                                      pkt->getSize(), pc, secondary_type,
690                                      RubyAccessMode_Supervisor, pkt,
691                                      PrefetchBit_No, proc_id);
692
693    DPRINTFR(ProtocolTrace, "%15s %3s %10s%20s %6s>%-6s %s %s\n",
694            curTick(), m_version, "Seq", "Begin", "", "",
695            msg->getPhysicalAddress(),
696            RubyRequestType_to_string(secondary_type));
697
698    // The Sequencer currently assesses instruction and data cache hit latency
699    // for the top-level caches at the beginning of a memory access.
700    // TODO: Eventually, this latency should be moved to represent the actual
701    // cache access latency portion of the memory access. This will require
702    // changing cache controller protocol files to assess the latency on the
703    // access response path.
704    Cycles latency(0);  // Initialize to zero to catch misconfigured latency
705    if (secondary_type == RubyRequestType_IFETCH)
706        latency = m_inst_cache_hit_latency;
707    else
708        latency = m_data_cache_hit_latency;
709
710    // Send the message to the cache controller
711    assert(latency > 0);
712
713    assert(m_mandatory_q_ptr != NULL);
714    m_mandatory_q_ptr->enqueue(msg, latency);
715}
716
717template <class KEY, class VALUE>
718std::ostream &
719operator<<(ostream &out, const m5::hash_map<KEY, VALUE> &map)
720{
721    typename m5::hash_map<KEY, VALUE>::const_iterator i = map.begin();
722    typename m5::hash_map<KEY, VALUE>::const_iterator end = map.end();
723
724    out << "[";
725    for (; i != end; ++i)
726        out << " " << i->first << "=" << i->second;
727    out << " ]";
728
729    return out;
730}
731
732void
733Sequencer::print(ostream& out) const
734{
735    out << "[Sequencer: " << m_version
736        << ", outstanding requests: " << m_outstanding_count
737        << ", read request table: " << m_readRequestTable
738        << ", write request table: " << m_writeRequestTable
739        << "]";
740}
741
742// this can be called from setState whenever coherence permissions are
743// upgraded when invoked, coherence violations will be checked for the
744// given block
745void
746Sequencer::checkCoherence(const Address& addr)
747{
748#ifdef CHECK_COHERENCE
749    m_ruby_system->checkGlobalCoherenceInvariant(addr);
750#endif
751}
752
753void
754Sequencer::recordRequestType(SequencerRequestType requestType) {
755    DPRINTF(RubyStats, "Recorded statistic: %s\n",
756            SequencerRequestType_to_string(requestType));
757}
758
759
760void
761Sequencer::evictionCallback(const Address& address)
762{
763    ruby_eviction_callback(address);
764}
765
766void
767Sequencer::regStats()
768{
769    m_store_waiting_on_load
770        .name(name() + ".store_waiting_on_load")
771        .desc("Number of times a store aliased with a pending load")
772        .flags(Stats::nozero);
773    m_store_waiting_on_store
774        .name(name() + ".store_waiting_on_store")
775        .desc("Number of times a store aliased with a pending store")
776        .flags(Stats::nozero);
777    m_load_waiting_on_load
778        .name(name() + ".load_waiting_on_load")
779        .desc("Number of times a load aliased with a pending load")
780        .flags(Stats::nozero);
781    m_load_waiting_on_store
782        .name(name() + ".load_waiting_on_store")
783        .desc("Number of times a load aliased with a pending store")
784        .flags(Stats::nozero);
785
786    // These statistical variables are not for display.
787    // The profiler will collate these across different
788    // sequencers and display those collated statistics.
789    m_outstandReqHist.init(10);
790    m_latencyHist.init(10);
791    m_hitLatencyHist.init(10);
792    m_missLatencyHist.init(10);
793
794    for (int i = 0; i < RubyRequestType_NUM; i++) {
795        m_typeLatencyHist.push_back(new Stats::Histogram());
796        m_typeLatencyHist[i]->init(10);
797
798        m_hitTypeLatencyHist.push_back(new Stats::Histogram());
799        m_hitTypeLatencyHist[i]->init(10);
800
801        m_missTypeLatencyHist.push_back(new Stats::Histogram());
802        m_missTypeLatencyHist[i]->init(10);
803    }
804
805    for (int i = 0; i < MachineType_NUM; i++) {
806        m_hitMachLatencyHist.push_back(new Stats::Histogram());
807        m_hitMachLatencyHist[i]->init(10);
808
809        m_missMachLatencyHist.push_back(new Stats::Histogram());
810        m_missMachLatencyHist[i]->init(10);
811
812        m_IssueToInitialDelayHist.push_back(new Stats::Histogram());
813        m_IssueToInitialDelayHist[i]->init(10);
814
815        m_InitialToForwardDelayHist.push_back(new Stats::Histogram());
816        m_InitialToForwardDelayHist[i]->init(10);
817
818        m_ForwardToFirstResponseDelayHist.push_back(new Stats::Histogram());
819        m_ForwardToFirstResponseDelayHist[i]->init(10);
820
821        m_FirstResponseToCompletionDelayHist.push_back(new Stats::Histogram());
822        m_FirstResponseToCompletionDelayHist[i]->init(10);
823    }
824
825    for (int i = 0; i < RubyRequestType_NUM; i++) {
826        m_hitTypeMachLatencyHist.push_back(std::vector<Stats::Histogram *>());
827        m_missTypeMachLatencyHist.push_back(std::vector<Stats::Histogram *>());
828
829        for (int j = 0; j < MachineType_NUM; j++) {
830            m_hitTypeMachLatencyHist[i].push_back(new Stats::Histogram());
831            m_hitTypeMachLatencyHist[i][j]->init(10);
832
833            m_missTypeMachLatencyHist[i].push_back(new Stats::Histogram());
834            m_missTypeMachLatencyHist[i][j]->init(10);
835        }
836    }
837}
838