Sequencer.cc revision 11321:02e930db812d
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/RubySystem.hh" 42#include "mem/ruby/system/Sequencer.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 m_coreId = p->coreid; // for tracking the two CorePair sequencers 67 assert(m_max_outstanding_requests > 0); 68 assert(m_deadlock_threshold > 0); 69 assert(m_instCache_ptr != NULL); 70 assert(m_dataCache_ptr != NULL); 71 assert(m_data_cache_hit_latency > 0); 72 assert(m_inst_cache_hit_latency > 0); 73 74 m_usingNetworkTester = p->using_network_tester; 75} 76 77Sequencer::~Sequencer() 78{ 79} 80 81void 82Sequencer::wakeup() 83{ 84 assert(drainState() != DrainState::Draining); 85 86 // Check for deadlock of any of the requests 87 Cycles current_time = curCycle(); 88 89 // Check across all outstanding requests 90 int total_outstanding = 0; 91 92 RequestTable::iterator read = m_readRequestTable.begin(); 93 RequestTable::iterator read_end = m_readRequestTable.end(); 94 for (; read != read_end; ++read) { 95 SequencerRequest* request = read->second; 96 if (current_time - request->issue_time < m_deadlock_threshold) 97 continue; 98 99 panic("Possible Deadlock detected. Aborting!\n" 100 "version: %d request.paddr: 0x%x m_readRequestTable: %d " 101 "current time: %u issue_time: %d difference: %d\n", m_version, 102 request->pkt->getAddr(), m_readRequestTable.size(), 103 current_time * clockPeriod(), request->issue_time * clockPeriod(), 104 (current_time * clockPeriod()) - (request->issue_time * clockPeriod())); 105 } 106 107 RequestTable::iterator write = m_writeRequestTable.begin(); 108 RequestTable::iterator write_end = m_writeRequestTable.end(); 109 for (; write != write_end; ++write) { 110 SequencerRequest* request = write->second; 111 if (current_time - request->issue_time < m_deadlock_threshold) 112 continue; 113 114 panic("Possible Deadlock detected. Aborting!\n" 115 "version: %d request.paddr: 0x%x m_writeRequestTable: %d " 116 "current time: %u issue_time: %d difference: %d\n", m_version, 117 request->pkt->getAddr(), m_writeRequestTable.size(), 118 current_time * clockPeriod(), request->issue_time * clockPeriod(), 119 (current_time * clockPeriod()) - (request->issue_time * clockPeriod())); 120 } 121 122 total_outstanding += m_writeRequestTable.size(); 123 total_outstanding += m_readRequestTable.size(); 124 125 assert(m_outstanding_count == total_outstanding); 126 127 if (m_outstanding_count > 0) { 128 // If there are still outstanding requests, keep checking 129 schedule(deadlockCheckEvent, clockEdge(m_deadlock_threshold)); 130 } 131} 132 133void Sequencer::resetStats() 134{ 135 m_latencyHist.reset(); 136 m_hitLatencyHist.reset(); 137 m_missLatencyHist.reset(); 138 for (int i = 0; i < RubyRequestType_NUM; i++) { 139 m_typeLatencyHist[i]->reset(); 140 m_hitTypeLatencyHist[i]->reset(); 141 m_missTypeLatencyHist[i]->reset(); 142 for (int j = 0; j < MachineType_NUM; j++) { 143 m_hitTypeMachLatencyHist[i][j]->reset(); 144 m_missTypeMachLatencyHist[i][j]->reset(); 145 } 146 } 147 148 for (int i = 0; i < MachineType_NUM; i++) { 149 m_missMachLatencyHist[i]->reset(); 150 m_hitMachLatencyHist[i]->reset(); 151 152 m_IssueToInitialDelayHist[i]->reset(); 153 m_InitialToForwardDelayHist[i]->reset(); 154 m_ForwardToFirstResponseDelayHist[i]->reset(); 155 m_FirstResponseToCompletionDelayHist[i]->reset(); 156 157 m_IncompleteTimes[i] = 0; 158 } 159} 160 161// Insert the request on the correct request table. Return true if 162// the entry was already present. 163RequestStatus 164Sequencer::insertRequest(PacketPtr pkt, RubyRequestType request_type) 165{ 166 assert(m_outstanding_count == 167 (m_writeRequestTable.size() + m_readRequestTable.size())); 168 169 // See if we should schedule a deadlock check 170 if (!deadlockCheckEvent.scheduled() && 171 drainState() != DrainState::Draining) { 172 schedule(deadlockCheckEvent, clockEdge(m_deadlock_threshold)); 173 } 174 175 Addr line_addr = makeLineAddress(pkt->getAddr()); 176 // Create a default entry, mapping the address to NULL, the cast is 177 // there to make gcc 4.4 happy 178 RequestTable::value_type default_entry(line_addr, 179 (SequencerRequest*) NULL); 180 181 if ((request_type == RubyRequestType_ST) || 182 (request_type == RubyRequestType_RMW_Read) || 183 (request_type == RubyRequestType_RMW_Write) || 184 (request_type == RubyRequestType_Load_Linked) || 185 (request_type == RubyRequestType_Store_Conditional) || 186 (request_type == RubyRequestType_Locked_RMW_Read) || 187 (request_type == RubyRequestType_Locked_RMW_Write) || 188 (request_type == RubyRequestType_FLUSH)) { 189 190 // Check if there is any outstanding read request for the same 191 // cache line. 192 if (m_readRequestTable.count(line_addr) > 0) { 193 m_store_waiting_on_load++; 194 return RequestStatus_Aliased; 195 } 196 197 pair<RequestTable::iterator, bool> r = 198 m_writeRequestTable.insert(default_entry); 199 if (r.second) { 200 RequestTable::iterator i = r.first; 201 i->second = new SequencerRequest(pkt, request_type, curCycle()); 202 m_outstanding_count++; 203 } else { 204 // There is an outstanding write request for the cache line 205 m_store_waiting_on_store++; 206 return RequestStatus_Aliased; 207 } 208 } else { 209 // Check if there is any outstanding write request for the same 210 // cache line. 211 if (m_writeRequestTable.count(line_addr) > 0) { 212 m_load_waiting_on_store++; 213 return RequestStatus_Aliased; 214 } 215 216 pair<RequestTable::iterator, bool> r = 217 m_readRequestTable.insert(default_entry); 218 219 if (r.second) { 220 RequestTable::iterator i = r.first; 221 i->second = new SequencerRequest(pkt, request_type, curCycle()); 222 m_outstanding_count++; 223 } else { 224 // There is an outstanding read request for the cache line 225 m_load_waiting_on_load++; 226 return RequestStatus_Aliased; 227 } 228 } 229 230 m_outstandReqHist.sample(m_outstanding_count); 231 assert(m_outstanding_count == 232 (m_writeRequestTable.size() + m_readRequestTable.size())); 233 234 return RequestStatus_Ready; 235} 236 237void 238Sequencer::markRemoved() 239{ 240 m_outstanding_count--; 241 assert(m_outstanding_count == 242 m_writeRequestTable.size() + m_readRequestTable.size()); 243} 244 245void 246Sequencer::invalidateSC(Addr address) 247{ 248 AbstractCacheEntry *e = m_dataCache_ptr->lookup(address); 249 // The controller has lost the coherence permissions, hence the lock 250 // on the cache line maintained by the cache should be cleared. 251 if (e && e->isLocked(m_version)) { 252 e->clearLocked(); 253 } 254} 255 256bool 257Sequencer::handleLlsc(Addr address, SequencerRequest* request) 258{ 259 AbstractCacheEntry *e = m_dataCache_ptr->lookup(address); 260 if (!e) 261 return true; 262 263 // The success flag indicates whether the LLSC operation was successful. 264 // LL ops will always succeed, but SC may fail if the cache line is no 265 // longer locked. 266 bool success = true; 267 if (request->m_type == RubyRequestType_Store_Conditional) { 268 if (!e->isLocked(m_version)) { 269 // 270 // For failed SC requests, indicate the failure to the cpu by 271 // setting the extra data to zero. 272 // 273 request->pkt->req->setExtraData(0); 274 success = false; 275 } else { 276 // 277 // For successful SC requests, indicate the success to the cpu by 278 // setting the extra data to one. 279 // 280 request->pkt->req->setExtraData(1); 281 } 282 // 283 // Independent of success, all SC operations must clear the lock 284 // 285 e->clearLocked(); 286 } else if (request->m_type == RubyRequestType_Load_Linked) { 287 // 288 // Note: To fully follow Alpha LLSC semantics, should the LL clear any 289 // previously locked cache lines? 290 // 291 e->setLocked(m_version); 292 } else if (e->isLocked(m_version)) { 293 // 294 // Normal writes should clear the locked address 295 // 296 e->clearLocked(); 297 } 298 return success; 299} 300 301void 302Sequencer::recordMissLatency(const Cycles cycles, const RubyRequestType type, 303 const MachineType respondingMach, 304 bool isExternalHit, Cycles issuedTime, 305 Cycles initialRequestTime, 306 Cycles forwardRequestTime, 307 Cycles firstResponseTime, Cycles completionTime) 308{ 309 m_latencyHist.sample(cycles); 310 m_typeLatencyHist[type]->sample(cycles); 311 312 if (isExternalHit) { 313 m_missLatencyHist.sample(cycles); 314 m_missTypeLatencyHist[type]->sample(cycles); 315 316 if (respondingMach != MachineType_NUM) { 317 m_missMachLatencyHist[respondingMach]->sample(cycles); 318 m_missTypeMachLatencyHist[type][respondingMach]->sample(cycles); 319 320 if ((issuedTime <= initialRequestTime) && 321 (initialRequestTime <= forwardRequestTime) && 322 (forwardRequestTime <= firstResponseTime) && 323 (firstResponseTime <= completionTime)) { 324 325 m_IssueToInitialDelayHist[respondingMach]->sample( 326 initialRequestTime - issuedTime); 327 m_InitialToForwardDelayHist[respondingMach]->sample( 328 forwardRequestTime - initialRequestTime); 329 m_ForwardToFirstResponseDelayHist[respondingMach]->sample( 330 firstResponseTime - forwardRequestTime); 331 m_FirstResponseToCompletionDelayHist[respondingMach]->sample( 332 completionTime - firstResponseTime); 333 } else { 334 m_IncompleteTimes[respondingMach]++; 335 } 336 } 337 } else { 338 m_hitLatencyHist.sample(cycles); 339 m_hitTypeLatencyHist[type]->sample(cycles); 340 341 if (respondingMach != MachineType_NUM) { 342 m_hitMachLatencyHist[respondingMach]->sample(cycles); 343 m_hitTypeMachLatencyHist[type][respondingMach]->sample(cycles); 344 } 345 } 346} 347 348void 349Sequencer::writeCallback(Addr address, DataBlock& data, 350 const bool externalHit, const MachineType mach, 351 const Cycles initialRequestTime, 352 const Cycles forwardRequestTime, 353 const Cycles firstResponseTime) 354{ 355 assert(address == makeLineAddress(address)); 356 assert(m_writeRequestTable.count(makeLineAddress(address))); 357 358 RequestTable::iterator i = m_writeRequestTable.find(address); 359 assert(i != m_writeRequestTable.end()); 360 SequencerRequest* request = i->second; 361 362 m_writeRequestTable.erase(i); 363 markRemoved(); 364 365 assert((request->m_type == RubyRequestType_ST) || 366 (request->m_type == RubyRequestType_ATOMIC) || 367 (request->m_type == RubyRequestType_RMW_Read) || 368 (request->m_type == RubyRequestType_RMW_Write) || 369 (request->m_type == RubyRequestType_Load_Linked) || 370 (request->m_type == RubyRequestType_Store_Conditional) || 371 (request->m_type == RubyRequestType_Locked_RMW_Read) || 372 (request->m_type == RubyRequestType_Locked_RMW_Write) || 373 (request->m_type == RubyRequestType_FLUSH)); 374 375 // 376 // For Alpha, properly handle LL, SC, and write requests with respect to 377 // locked cache blocks. 378 // 379 // Not valid for Network_test protocl 380 // 381 bool success = true; 382 if (!m_usingNetworkTester) 383 success = handleLlsc(address, request); 384 385 if (request->m_type == RubyRequestType_Locked_RMW_Read) { 386 m_controller->blockOnQueue(address, m_mandatory_q_ptr); 387 } else if (request->m_type == RubyRequestType_Locked_RMW_Write) { 388 m_controller->unblock(address); 389 } 390 391 hitCallback(request, data, success, mach, externalHit, 392 initialRequestTime, forwardRequestTime, firstResponseTime); 393} 394 395void 396Sequencer::readCallback(Addr address, DataBlock& data, 397 bool externalHit, const MachineType mach, 398 Cycles initialRequestTime, 399 Cycles forwardRequestTime, 400 Cycles firstResponseTime) 401{ 402 assert(address == makeLineAddress(address)); 403 assert(m_readRequestTable.count(makeLineAddress(address))); 404 405 RequestTable::iterator i = m_readRequestTable.find(address); 406 assert(i != m_readRequestTable.end()); 407 SequencerRequest* request = i->second; 408 409 m_readRequestTable.erase(i); 410 markRemoved(); 411 412 assert((request->m_type == RubyRequestType_LD) || 413 (request->m_type == RubyRequestType_IFETCH)); 414 415 hitCallback(request, data, true, mach, externalHit, 416 initialRequestTime, forwardRequestTime, firstResponseTime); 417} 418 419void 420Sequencer::hitCallback(SequencerRequest* srequest, DataBlock& data, 421 bool llscSuccess, 422 const MachineType mach, const bool externalHit, 423 const Cycles initialRequestTime, 424 const Cycles forwardRequestTime, 425 const Cycles firstResponseTime) 426{ 427 warn_once("Replacement policy updates recently became the responsibility " 428 "of SLICC state machines. Make sure to setMRU() near callbacks " 429 "in .sm files!"); 430 431 PacketPtr pkt = srequest->pkt; 432 Addr request_address(pkt->getAddr()); 433 RubyRequestType type = srequest->m_type; 434 Cycles issued_time = srequest->issue_time; 435 436 assert(curCycle() >= issued_time); 437 Cycles total_latency = curCycle() - issued_time; 438 439 // Profile the latency for all demand accesses. 440 recordMissLatency(total_latency, type, mach, externalHit, issued_time, 441 initialRequestTime, forwardRequestTime, 442 firstResponseTime, curCycle()); 443 444 DPRINTFR(ProtocolTrace, "%15s %3s %10s%20s %6s>%-6s %#x %d cycles\n", 445 curTick(), m_version, "Seq", 446 llscSuccess ? "Done" : "SC_Failed", "", "", 447 printAddress(request_address), total_latency); 448 449 // update the data unless it is a non-data-carrying flush 450 if (RubySystem::getWarmupEnabled()) { 451 data.setData(pkt->getConstPtr<uint8_t>(), 452 getOffset(request_address), pkt->getSize()); 453 } else if (!pkt->isFlush()) { 454 if ((type == RubyRequestType_LD) || 455 (type == RubyRequestType_IFETCH) || 456 (type == RubyRequestType_RMW_Read) || 457 (type == RubyRequestType_Locked_RMW_Read) || 458 (type == RubyRequestType_Load_Linked)) { 459 memcpy(pkt->getPtr<uint8_t>(), 460 data.getData(getOffset(request_address), pkt->getSize()), 461 pkt->getSize()); 462 DPRINTF(RubySequencer, "read data %s\n", data); 463 } else { 464 data.setData(pkt->getConstPtr<uint8_t>(), 465 getOffset(request_address), pkt->getSize()); 466 DPRINTF(RubySequencer, "set data %s\n", data); 467 } 468 } 469 470 // If using the RubyTester, update the RubyTester sender state's 471 // subBlock with the recieved data. The tester will later access 472 // this state. 473 if (m_usingRubyTester) { 474 DPRINTF(RubySequencer, "hitCallback %s 0x%x using RubyTester\n", 475 pkt->cmdString(), pkt->getAddr()); 476 RubyTester::SenderState* testerSenderState = 477 pkt->findNextSenderState<RubyTester::SenderState>(); 478 assert(testerSenderState); 479 testerSenderState->subBlock.mergeFrom(data); 480 } 481 482 delete srequest; 483 484 RubySystem *rs = m_ruby_system; 485 if (RubySystem::getWarmupEnabled()) { 486 assert(pkt->req); 487 delete pkt->req; 488 delete pkt; 489 rs->m_cache_recorder->enqueueNextFetchRequest(); 490 } else if (RubySystem::getCooldownEnabled()) { 491 delete pkt; 492 rs->m_cache_recorder->enqueueNextFlushRequest(); 493 } else { 494 ruby_hit_callback(pkt); 495 testDrainComplete(); 496 } 497} 498 499bool 500Sequencer::empty() const 501{ 502 return m_writeRequestTable.empty() && m_readRequestTable.empty(); 503} 504 505RequestStatus 506Sequencer::makeRequest(PacketPtr pkt) 507{ 508 if (m_outstanding_count >= m_max_outstanding_requests) { 509 return RequestStatus_BufferFull; 510 } 511 512 RubyRequestType primary_type = RubyRequestType_NULL; 513 RubyRequestType secondary_type = RubyRequestType_NULL; 514 515 if (pkt->isLLSC()) { 516 // 517 // Alpha LL/SC instructions need to be handled carefully by the cache 518 // coherence protocol to ensure they follow the proper semantics. In 519 // particular, by identifying the operations as atomic, the protocol 520 // should understand that migratory sharing optimizations should not 521 // be performed (i.e. a load between the LL and SC should not steal 522 // away exclusive permission). 523 // 524 if (pkt->isWrite()) { 525 DPRINTF(RubySequencer, "Issuing SC\n"); 526 primary_type = RubyRequestType_Store_Conditional; 527 } else { 528 DPRINTF(RubySequencer, "Issuing LL\n"); 529 assert(pkt->isRead()); 530 primary_type = RubyRequestType_Load_Linked; 531 } 532 secondary_type = RubyRequestType_ATOMIC; 533 } else if (pkt->req->isLockedRMW()) { 534 // 535 // x86 locked instructions are translated to store cache coherence 536 // requests because these requests should always be treated as read 537 // exclusive operations and should leverage any migratory sharing 538 // optimization built into the protocol. 539 // 540 if (pkt->isWrite()) { 541 DPRINTF(RubySequencer, "Issuing Locked RMW Write\n"); 542 primary_type = RubyRequestType_Locked_RMW_Write; 543 } else { 544 DPRINTF(RubySequencer, "Issuing Locked RMW Read\n"); 545 assert(pkt->isRead()); 546 primary_type = RubyRequestType_Locked_RMW_Read; 547 } 548 secondary_type = RubyRequestType_ST; 549 } else { 550 if (pkt->isRead()) { 551 if (pkt->req->isInstFetch()) { 552 primary_type = secondary_type = RubyRequestType_IFETCH; 553 } else { 554 bool storeCheck = false; 555 // only X86 need the store check 556 if (system->getArch() == Arch::X86ISA) { 557 uint32_t flags = pkt->req->getFlags(); 558 storeCheck = flags & 559 (X86ISA::StoreCheck << X86ISA::FlagShift); 560 } 561 if (storeCheck) { 562 primary_type = RubyRequestType_RMW_Read; 563 secondary_type = RubyRequestType_ST; 564 } else { 565 primary_type = secondary_type = RubyRequestType_LD; 566 } 567 } 568 } else if (pkt->isWrite()) { 569 // 570 // Note: M5 packets do not differentiate ST from RMW_Write 571 // 572 primary_type = secondary_type = RubyRequestType_ST; 573 } else if (pkt->isFlush()) { 574 primary_type = secondary_type = RubyRequestType_FLUSH; 575 } else { 576 panic("Unsupported ruby packet type\n"); 577 } 578 } 579 580 RequestStatus status = insertRequest(pkt, primary_type); 581 if (status != RequestStatus_Ready) 582 return status; 583 584 issueRequest(pkt, secondary_type); 585 586 // TODO: issue hardware prefetches here 587 return RequestStatus_Issued; 588} 589 590void 591Sequencer::issueRequest(PacketPtr pkt, RubyRequestType secondary_type) 592{ 593 assert(pkt != NULL); 594 ContextID proc_id = pkt->req->hasContextId() ? 595 pkt->req->contextId() : InvalidContextID; 596 597 ContextID core_id = coreId(); 598 599 // If valid, copy the pc to the ruby request 600 Addr pc = 0; 601 if (pkt->req->hasPC()) { 602 pc = pkt->req->getPC(); 603 } 604 605 // check if the packet has data as for example prefetch and flush 606 // requests do not 607 std::shared_ptr<RubyRequest> msg = 608 std::make_shared<RubyRequest>(clockEdge(), pkt->getAddr(), 609 pkt->isFlush() ? 610 nullptr : pkt->getPtr<uint8_t>(), 611 pkt->getSize(), pc, secondary_type, 612 RubyAccessMode_Supervisor, pkt, 613 PrefetchBit_No, proc_id, core_id); 614 615 DPRINTFR(ProtocolTrace, "%15s %3s %10s%20s %6s>%-6s %#x %s\n", 616 curTick(), m_version, "Seq", "Begin", "", "", 617 printAddress(msg->getPhysicalAddress()), 618 RubyRequestType_to_string(secondary_type)); 619 620 // The Sequencer currently assesses instruction and data cache hit latency 621 // for the top-level caches at the beginning of a memory access. 622 // TODO: Eventually, this latency should be moved to represent the actual 623 // cache access latency portion of the memory access. This will require 624 // changing cache controller protocol files to assess the latency on the 625 // access response path. 626 Cycles latency(0); // Initialize to zero to catch misconfigured latency 627 if (secondary_type == RubyRequestType_IFETCH) 628 latency = m_inst_cache_hit_latency; 629 else 630 latency = m_data_cache_hit_latency; 631 632 // Send the message to the cache controller 633 assert(latency > 0); 634 635 assert(m_mandatory_q_ptr != NULL); 636 m_mandatory_q_ptr->enqueue(msg, clockEdge(), cyclesToTicks(latency)); 637} 638 639template <class KEY, class VALUE> 640std::ostream & 641operator<<(ostream &out, const std::unordered_map<KEY, VALUE> &map) 642{ 643 auto i = map.begin(); 644 auto end = map.end(); 645 646 out << "["; 647 for (; i != end; ++i) 648 out << " " << i->first << "=" << i->second; 649 out << " ]"; 650 651 return out; 652} 653 654void 655Sequencer::print(ostream& out) const 656{ 657 out << "[Sequencer: " << m_version 658 << ", outstanding requests: " << m_outstanding_count 659 << ", read request table: " << m_readRequestTable 660 << ", write request table: " << m_writeRequestTable 661 << "]"; 662} 663 664// this can be called from setState whenever coherence permissions are 665// upgraded when invoked, coherence violations will be checked for the 666// given block 667void 668Sequencer::checkCoherence(Addr addr) 669{ 670#ifdef CHECK_COHERENCE 671 m_ruby_system->checkGlobalCoherenceInvariant(addr); 672#endif 673} 674 675void 676Sequencer::recordRequestType(SequencerRequestType requestType) { 677 DPRINTF(RubyStats, "Recorded statistic: %s\n", 678 SequencerRequestType_to_string(requestType)); 679} 680 681 682void 683Sequencer::evictionCallback(Addr address) 684{ 685 ruby_eviction_callback(address); 686} 687 688void 689Sequencer::regStats() 690{ 691 m_store_waiting_on_load 692 .name(name() + ".store_waiting_on_load") 693 .desc("Number of times a store aliased with a pending load") 694 .flags(Stats::nozero); 695 m_store_waiting_on_store 696 .name(name() + ".store_waiting_on_store") 697 .desc("Number of times a store aliased with a pending store") 698 .flags(Stats::nozero); 699 m_load_waiting_on_load 700 .name(name() + ".load_waiting_on_load") 701 .desc("Number of times a load aliased with a pending load") 702 .flags(Stats::nozero); 703 m_load_waiting_on_store 704 .name(name() + ".load_waiting_on_store") 705 .desc("Number of times a load aliased with a pending store") 706 .flags(Stats::nozero); 707 708 // These statistical variables are not for display. 709 // The profiler will collate these across different 710 // sequencers and display those collated statistics. 711 m_outstandReqHist.init(10); 712 m_latencyHist.init(10); 713 m_hitLatencyHist.init(10); 714 m_missLatencyHist.init(10); 715 716 for (int i = 0; i < RubyRequestType_NUM; i++) { 717 m_typeLatencyHist.push_back(new Stats::Histogram()); 718 m_typeLatencyHist[i]->init(10); 719 720 m_hitTypeLatencyHist.push_back(new Stats::Histogram()); 721 m_hitTypeLatencyHist[i]->init(10); 722 723 m_missTypeLatencyHist.push_back(new Stats::Histogram()); 724 m_missTypeLatencyHist[i]->init(10); 725 } 726 727 for (int i = 0; i < MachineType_NUM; i++) { 728 m_hitMachLatencyHist.push_back(new Stats::Histogram()); 729 m_hitMachLatencyHist[i]->init(10); 730 731 m_missMachLatencyHist.push_back(new Stats::Histogram()); 732 m_missMachLatencyHist[i]->init(10); 733 734 m_IssueToInitialDelayHist.push_back(new Stats::Histogram()); 735 m_IssueToInitialDelayHist[i]->init(10); 736 737 m_InitialToForwardDelayHist.push_back(new Stats::Histogram()); 738 m_InitialToForwardDelayHist[i]->init(10); 739 740 m_ForwardToFirstResponseDelayHist.push_back(new Stats::Histogram()); 741 m_ForwardToFirstResponseDelayHist[i]->init(10); 742 743 m_FirstResponseToCompletionDelayHist.push_back(new Stats::Histogram()); 744 m_FirstResponseToCompletionDelayHist[i]->init(10); 745 } 746 747 for (int i = 0; i < RubyRequestType_NUM; i++) { 748 m_hitTypeMachLatencyHist.push_back(std::vector<Stats::Histogram *>()); 749 m_missTypeMachLatencyHist.push_back(std::vector<Stats::Histogram *>()); 750 751 for (int j = 0; j < MachineType_NUM; j++) { 752 m_hitTypeMachLatencyHist[i].push_back(new Stats::Histogram()); 753 m_hitTypeMachLatencyHist[i][j]->init(10); 754 755 m_missTypeMachLatencyHist[i].push_back(new Stats::Histogram()); 756 m_missTypeMachLatencyHist[i][j]->init(10); 757 } 758 } 759} 760