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 "mem/ruby/system/Sequencer.hh" 30 31#include "arch/x86/ldstflags.hh" 32#include "base/logging.hh" 33#include "base/str.hh" 34#include "cpu/testers/rubytest/RubyTester.hh" 35#include "debug/MemoryAccess.hh" 36#include "debug/ProtocolTrace.hh" 37#include "debug/RubySequencer.hh" 38#include "debug/RubyStats.hh" 39#include "mem/packet.hh" 40#include "mem/ruby/profiler/Profiler.hh" 41#include "mem/ruby/protocol/PrefetchBit.hh" 42#include "mem/ruby/protocol/RubyAccessMode.hh" 43#include "mem/ruby/slicc_interface/RubyRequest.hh" 44#include "mem/ruby/system/RubySystem.hh" 45#include "sim/system.hh" 46 47using namespace std; 48 49Sequencer * 50RubySequencerParams::create() 51{ 52 return new Sequencer(this); 53} 54 55Sequencer::Sequencer(const Params *p) 56 : RubyPort(p), m_IncompleteTimes(MachineType_NUM), 57 deadlockCheckEvent([this]{ wakeup(); }, "Sequencer deadlock check") 58{ 59 m_outstanding_count = 0; 60 61 m_instCache_ptr = p->icache; 62 m_dataCache_ptr = p->dcache; 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 72 m_runningGarnetStandalone = p->garnet_standalone; 73} 74 75Sequencer::~Sequencer() 76{ 77} 78 79void 80Sequencer::wakeup() 81{ 82 assert(drainState() != DrainState::Draining); 83 84 // Check for deadlock of any of the requests 85 Cycles current_time = curCycle(); 86 87 // Check across all outstanding requests 88 int total_outstanding = 0; 89 90 RequestTable::iterator read = m_readRequestTable.begin(); 91 RequestTable::iterator read_end = m_readRequestTable.end(); 92 for (; read != read_end; ++read) { 93 SequencerRequest* request = read->second; 94 if (current_time - request->issue_time < m_deadlock_threshold) 95 continue; 96 97 panic("Possible Deadlock detected. Aborting!\n" 98 "version: %d request.paddr: 0x%x m_readRequestTable: %d " 99 "current time: %u issue_time: %d difference: %d\n", m_version, 100 request->pkt->getAddr(), m_readRequestTable.size(), 101 current_time * clockPeriod(), request->issue_time * clockPeriod(), 102 (current_time * clockPeriod()) - (request->issue_time * clockPeriod())); 103 } 104 105 RequestTable::iterator write = m_writeRequestTable.begin(); 106 RequestTable::iterator write_end = m_writeRequestTable.end(); 107 for (; write != write_end; ++write) { 108 SequencerRequest* request = write->second; 109 if (current_time - request->issue_time < m_deadlock_threshold) 110 continue; 111 112 panic("Possible Deadlock detected. Aborting!\n" 113 "version: %d request.paddr: 0x%x m_writeRequestTable: %d " 114 "current time: %u issue_time: %d difference: %d\n", m_version, 115 request->pkt->getAddr(), m_writeRequestTable.size(), 116 current_time * clockPeriod(), request->issue_time * clockPeriod(), 117 (current_time * clockPeriod()) - (request->issue_time * clockPeriod())); 118 } 119 120 total_outstanding += m_writeRequestTable.size(); 121 total_outstanding += m_readRequestTable.size(); 122 123 assert(m_outstanding_count == total_outstanding); 124 125 if (m_outstanding_count > 0) { 126 // If there are still outstanding requests, keep checking 127 schedule(deadlockCheckEvent, clockEdge(m_deadlock_threshold)); 128 } 129} 130 131void Sequencer::resetStats() 132{ 133 m_latencyHist.reset(); 134 m_hitLatencyHist.reset(); 135 m_missLatencyHist.reset(); 136 for (int i = 0; i < RubyRequestType_NUM; i++) { 137 m_typeLatencyHist[i]->reset(); 138 m_hitTypeLatencyHist[i]->reset(); 139 m_missTypeLatencyHist[i]->reset(); 140 for (int j = 0; j < MachineType_NUM; j++) { 141 m_hitTypeMachLatencyHist[i][j]->reset(); 142 m_missTypeMachLatencyHist[i][j]->reset(); 143 } 144 } 145 146 for (int i = 0; i < MachineType_NUM; i++) { 147 m_missMachLatencyHist[i]->reset(); 148 m_hitMachLatencyHist[i]->reset(); 149 150 m_IssueToInitialDelayHist[i]->reset(); 151 m_InitialToForwardDelayHist[i]->reset(); 152 m_ForwardToFirstResponseDelayHist[i]->reset(); 153 m_FirstResponseToCompletionDelayHist[i]->reset(); 154 155 m_IncompleteTimes[i] = 0; 156 } 157} 158 159// Insert the request on the correct request table. Return true if 160// the entry was already present. 161RequestStatus 162Sequencer::insertRequest(PacketPtr pkt, RubyRequestType request_type) 163{ 164 assert(m_outstanding_count == 165 (m_writeRequestTable.size() + m_readRequestTable.size())); 166 167 // See if we should schedule a deadlock check 168 if (!deadlockCheckEvent.scheduled() && 169 drainState() != DrainState::Draining) { 170 schedule(deadlockCheckEvent, clockEdge(m_deadlock_threshold)); 171 } 172 173 Addr line_addr = makeLineAddress(pkt->getAddr()); 174 175 // Check if the line is blocked for a Locked_RMW 176 if (m_controller->isBlocked(line_addr) && 177 (request_type != RubyRequestType_Locked_RMW_Write)) { 178 // Return that this request's cache line address aliases with 179 // a prior request that locked the cache line. The request cannot 180 // proceed until the cache line is unlocked by a Locked_RMW_Write 181 return RequestStatus_Aliased; 182 } 183 184 // Create a default entry, mapping the address to NULL, the cast is 185 // there to make gcc 4.4 happy 186 RequestTable::value_type default_entry(line_addr, 187 (SequencerRequest*) NULL); 188 189 if ((request_type == RubyRequestType_ST) || 190 (request_type == RubyRequestType_RMW_Read) || 191 (request_type == RubyRequestType_RMW_Write) || 192 (request_type == RubyRequestType_Load_Linked) || 193 (request_type == RubyRequestType_Store_Conditional) || 194 (request_type == RubyRequestType_Locked_RMW_Read) || 195 (request_type == RubyRequestType_Locked_RMW_Write) || 196 (request_type == RubyRequestType_FLUSH)) { 197 198 // Check if there is any outstanding read request for the same 199 // cache line. 200 if (m_readRequestTable.count(line_addr) > 0) { 201 m_store_waiting_on_load++; 202 return RequestStatus_Aliased; 203 } 204 205 pair<RequestTable::iterator, bool> r = 206 m_writeRequestTable.insert(default_entry); 207 if (r.second) { 208 RequestTable::iterator i = r.first; 209 i->second = new SequencerRequest(pkt, request_type, curCycle()); 210 m_outstanding_count++; 211 } else { 212 // There is an outstanding write request for the cache line 213 m_store_waiting_on_store++; 214 return RequestStatus_Aliased; 215 } 216 } else { 217 // Check if there is any outstanding write request for the same 218 // cache line. 219 if (m_writeRequestTable.count(line_addr) > 0) { 220 m_load_waiting_on_store++; 221 return RequestStatus_Aliased; 222 } 223 224 pair<RequestTable::iterator, bool> r = 225 m_readRequestTable.insert(default_entry); 226 227 if (r.second) { 228 RequestTable::iterator i = r.first; 229 i->second = new SequencerRequest(pkt, request_type, curCycle()); 230 m_outstanding_count++; 231 } else { 232 // There is an outstanding read request for the cache line 233 m_load_waiting_on_load++; 234 return RequestStatus_Aliased; 235 } 236 } 237 238 m_outstandReqHist.sample(m_outstanding_count); 239 assert(m_outstanding_count == 240 (m_writeRequestTable.size() + m_readRequestTable.size())); 241 242 return RequestStatus_Ready; 243} 244 245void 246Sequencer::markRemoved() 247{ 248 m_outstanding_count--; 249 assert(m_outstanding_count == 250 m_writeRequestTable.size() + m_readRequestTable.size()); 251} 252 253void 254Sequencer::invalidateSC(Addr address) 255{ 256 AbstractCacheEntry *e = m_dataCache_ptr->lookup(address); 257 // The controller has lost the coherence permissions, hence the lock 258 // on the cache line maintained by the cache should be cleared. 259 if (e && e->isLocked(m_version)) { 260 e->clearLocked(); 261 } 262} 263 264bool 265Sequencer::handleLlsc(Addr address, SequencerRequest* request) 266{ 267 AbstractCacheEntry *e = m_dataCache_ptr->lookup(address); 268 if (!e) 269 return true; 270 271 // The success flag indicates whether the LLSC operation was successful. 272 // LL ops will always succeed, but SC may fail if the cache line is no 273 // longer locked. 274 bool success = true; 275 if (request->m_type == RubyRequestType_Store_Conditional) { 276 if (!e->isLocked(m_version)) { 277 // 278 // For failed SC requests, indicate the failure to the cpu by 279 // setting the extra data to zero. 280 // 281 request->pkt->req->setExtraData(0); 282 success = false; 283 } else { 284 // 285 // For successful SC requests, indicate the success to the cpu by 286 // setting the extra data to one. 287 // 288 request->pkt->req->setExtraData(1); 289 } 290 // 291 // Independent of success, all SC operations must clear the lock 292 // 293 e->clearLocked(); 294 } else if (request->m_type == RubyRequestType_Load_Linked) { 295 // 296 // Note: To fully follow Alpha LLSC semantics, should the LL clear any 297 // previously locked cache lines? 298 // 299 e->setLocked(m_version); 300 } else if (e->isLocked(m_version)) { 301 // 302 // Normal writes should clear the locked address 303 // 304 e->clearLocked(); 305 } 306 return success; 307} 308 309void 310Sequencer::recordMissLatency(const Cycles cycles, const RubyRequestType type, 311 const MachineType respondingMach, 312 bool isExternalHit, Cycles issuedTime, 313 Cycles initialRequestTime, 314 Cycles forwardRequestTime, 315 Cycles firstResponseTime, Cycles completionTime) 316{ 317 m_latencyHist.sample(cycles); 318 m_typeLatencyHist[type]->sample(cycles); 319 320 if (isExternalHit) { 321 m_missLatencyHist.sample(cycles); 322 m_missTypeLatencyHist[type]->sample(cycles); 323 324 if (respondingMach != MachineType_NUM) { 325 m_missMachLatencyHist[respondingMach]->sample(cycles); 326 m_missTypeMachLatencyHist[type][respondingMach]->sample(cycles); 327 328 if ((issuedTime <= initialRequestTime) && 329 (initialRequestTime <= forwardRequestTime) && 330 (forwardRequestTime <= firstResponseTime) && 331 (firstResponseTime <= completionTime)) { 332 333 m_IssueToInitialDelayHist[respondingMach]->sample( 334 initialRequestTime - issuedTime); 335 m_InitialToForwardDelayHist[respondingMach]->sample( 336 forwardRequestTime - initialRequestTime); 337 m_ForwardToFirstResponseDelayHist[respondingMach]->sample( 338 firstResponseTime - forwardRequestTime); 339 m_FirstResponseToCompletionDelayHist[respondingMach]->sample( 340 completionTime - firstResponseTime); 341 } else { 342 m_IncompleteTimes[respondingMach]++; 343 } 344 } 345 } else { 346 m_hitLatencyHist.sample(cycles); 347 m_hitTypeLatencyHist[type]->sample(cycles); 348 349 if (respondingMach != MachineType_NUM) { 350 m_hitMachLatencyHist[respondingMach]->sample(cycles); 351 m_hitTypeMachLatencyHist[type][respondingMach]->sample(cycles); 352 } 353 } 354} 355 356void 357Sequencer::writeCallback(Addr address, DataBlock& data, 358 const bool externalHit, const MachineType mach, 359 const Cycles initialRequestTime, 360 const Cycles forwardRequestTime, 361 const Cycles firstResponseTime) 362{ 363 assert(address == makeLineAddress(address)); 364 assert(m_writeRequestTable.count(makeLineAddress(address))); 365 366 RequestTable::iterator i = m_writeRequestTable.find(address); 367 assert(i != m_writeRequestTable.end()); 368 SequencerRequest* request = i->second; 369 370 m_writeRequestTable.erase(i); 371 markRemoved(); 372 373 assert((request->m_type == RubyRequestType_ST) || 374 (request->m_type == RubyRequestType_ATOMIC) || 375 (request->m_type == RubyRequestType_RMW_Read) || 376 (request->m_type == RubyRequestType_RMW_Write) || 377 (request->m_type == RubyRequestType_Load_Linked) || 378 (request->m_type == RubyRequestType_Store_Conditional) || 379 (request->m_type == RubyRequestType_Locked_RMW_Read) || 380 (request->m_type == RubyRequestType_Locked_RMW_Write) || 381 (request->m_type == RubyRequestType_FLUSH)); 382 383 // 384 // For Alpha, properly handle LL, SC, and write requests with respect to 385 // locked cache blocks. 386 // 387 // Not valid for Garnet_standalone protocl 388 // 389 bool success = true; 390 if (!m_runningGarnetStandalone) 391 success = handleLlsc(address, request); 392 393 // Handle SLICC block_on behavior for Locked_RMW accesses. NOTE: the 394 // address variable here is assumed to be a line address, so when 395 // blocking buffers, must check line addresses. 396 if (request->m_type == RubyRequestType_Locked_RMW_Read) { 397 // blockOnQueue blocks all first-level cache controller queues 398 // waiting on memory accesses for the specified address that go to 399 // the specified queue. In this case, a Locked_RMW_Write must go to 400 // the mandatory_q before unblocking the first-level controller. 401 // This will block standard loads, stores, ifetches, etc. 402 m_controller->blockOnQueue(address, m_mandatory_q_ptr); 403 } else if (request->m_type == RubyRequestType_Locked_RMW_Write) { 404 m_controller->unblock(address); 405 } 406 407 hitCallback(request, data, success, mach, externalHit, 408 initialRequestTime, forwardRequestTime, firstResponseTime); 409} 410 411void 412Sequencer::readCallback(Addr address, DataBlock& data, 413 bool externalHit, const MachineType mach, 414 Cycles initialRequestTime, 415 Cycles forwardRequestTime, 416 Cycles firstResponseTime) 417{ 418 assert(address == makeLineAddress(address)); 419 assert(m_readRequestTable.count(makeLineAddress(address))); 420 421 RequestTable::iterator i = m_readRequestTable.find(address); 422 assert(i != m_readRequestTable.end()); 423 SequencerRequest* request = i->second; 424 425 m_readRequestTable.erase(i); 426 markRemoved(); 427 428 assert((request->m_type == RubyRequestType_LD) || 429 (request->m_type == RubyRequestType_IFETCH)); 430 431 hitCallback(request, data, true, mach, externalHit, 432 initialRequestTime, forwardRequestTime, firstResponseTime); 433} 434 435void 436Sequencer::hitCallback(SequencerRequest* srequest, DataBlock& data, 437 bool llscSuccess, 438 const MachineType mach, const bool externalHit, 439 const Cycles initialRequestTime, 440 const Cycles forwardRequestTime, 441 const Cycles firstResponseTime) 442{ 443 warn_once("Replacement policy updates recently became the responsibility " 444 "of SLICC state machines. Make sure to setMRU() near callbacks " 445 "in .sm files!"); 446 447 PacketPtr pkt = srequest->pkt; 448 Addr request_address(pkt->getAddr()); 449 RubyRequestType type = srequest->m_type; 450 Cycles issued_time = srequest->issue_time; 451 452 assert(curCycle() >= issued_time); 453 Cycles total_latency = curCycle() - issued_time; 454 455 // Profile the latency for all demand accesses. 456 recordMissLatency(total_latency, type, mach, externalHit, issued_time, 457 initialRequestTime, forwardRequestTime, 458 firstResponseTime, curCycle()); 459 460 DPRINTFR(ProtocolTrace, "%15s %3s %10s%20s %6s>%-6s %#x %d cycles\n", 461 curTick(), m_version, "Seq", 462 llscSuccess ? "Done" : "SC_Failed", "", "", 463 printAddress(request_address), total_latency); 464 465 // update the data unless it is a non-data-carrying flush 466 if (RubySystem::getWarmupEnabled()) { 467 data.setData(pkt->getConstPtr<uint8_t>(), 468 getOffset(request_address), pkt->getSize()); 469 } else if (!pkt->isFlush()) { 470 if ((type == RubyRequestType_LD) || 471 (type == RubyRequestType_IFETCH) || 472 (type == RubyRequestType_RMW_Read) || 473 (type == RubyRequestType_Locked_RMW_Read) || 474 (type == RubyRequestType_Load_Linked)) { 475 pkt->setData( 476 data.getData(getOffset(request_address), pkt->getSize())); 477 DPRINTF(RubySequencer, "read data %s\n", data); 478 } else if (pkt->req->isSwap()) { 479 std::vector<uint8_t> overwrite_val(pkt->getSize()); 480 pkt->writeData(&overwrite_val[0]); 481 pkt->setData( 482 data.getData(getOffset(request_address), pkt->getSize())); 483 data.setData(&overwrite_val[0], 484 getOffset(request_address), pkt->getSize()); 485 DPRINTF(RubySequencer, "swap data %s\n", data); 486 } else if (type != RubyRequestType_Store_Conditional || llscSuccess) { 487 // Types of stores set the actual data here, apart from 488 // failed Store Conditional requests 489 data.setData(pkt->getConstPtr<uint8_t>(), 490 getOffset(request_address), pkt->getSize()); 491 DPRINTF(RubySequencer, "set data %s\n", data); 492 } 493 } 494 495 // If using the RubyTester, update the RubyTester sender state's 496 // subBlock with the recieved data. The tester will later access 497 // this state. 498 if (m_usingRubyTester) { 499 DPRINTF(RubySequencer, "hitCallback %s 0x%x using RubyTester\n", 500 pkt->cmdString(), pkt->getAddr()); 501 RubyTester::SenderState* testerSenderState = 502 pkt->findNextSenderState<RubyTester::SenderState>(); 503 assert(testerSenderState); 504 testerSenderState->subBlock.mergeFrom(data); 505 } 506 507 delete srequest; 508 509 RubySystem *rs = m_ruby_system; 510 if (RubySystem::getWarmupEnabled()) { 511 assert(pkt->req); 512 delete pkt; 513 rs->m_cache_recorder->enqueueNextFetchRequest(); 514 } else if (RubySystem::getCooldownEnabled()) { 515 delete pkt; 516 rs->m_cache_recorder->enqueueNextFlushRequest(); 517 } else { 518 ruby_hit_callback(pkt); 519 testDrainComplete(); 520 } 521} 522 523bool 524Sequencer::empty() const 525{ 526 return m_writeRequestTable.empty() && m_readRequestTable.empty(); 527} 528 529RequestStatus 530Sequencer::makeRequest(PacketPtr pkt) 531{ 532 if (m_outstanding_count >= m_max_outstanding_requests) { 533 return RequestStatus_BufferFull; 534 } 535 536 RubyRequestType primary_type = RubyRequestType_NULL; 537 RubyRequestType secondary_type = RubyRequestType_NULL; 538 539 if (pkt->isLLSC()) { 540 // 541 // Alpha LL/SC instructions need to be handled carefully by the cache 542 // coherence protocol to ensure they follow the proper semantics. In 543 // particular, by identifying the operations as atomic, the protocol 544 // should understand that migratory sharing optimizations should not 545 // be performed (i.e. a load between the LL and SC should not steal 546 // away exclusive permission). 547 // 548 if (pkt->isWrite()) { 549 DPRINTF(RubySequencer, "Issuing SC\n"); 550 primary_type = RubyRequestType_Store_Conditional; 551 } else { 552 DPRINTF(RubySequencer, "Issuing LL\n"); 553 assert(pkt->isRead()); 554 primary_type = RubyRequestType_Load_Linked; 555 } 556 secondary_type = RubyRequestType_ATOMIC; 557 } else if (pkt->req->isLockedRMW()) { 558 // 559 // x86 locked instructions are translated to store cache coherence 560 // requests because these requests should always be treated as read 561 // exclusive operations and should leverage any migratory sharing 562 // optimization built into the protocol. 563 // 564 if (pkt->isWrite()) { 565 DPRINTF(RubySequencer, "Issuing Locked RMW Write\n"); 566 primary_type = RubyRequestType_Locked_RMW_Write; 567 } else { 568 DPRINTF(RubySequencer, "Issuing Locked RMW Read\n"); 569 assert(pkt->isRead()); 570 primary_type = RubyRequestType_Locked_RMW_Read; 571 } 572 secondary_type = RubyRequestType_ST; 573 } else { 574 // 575 // To support SwapReq, we need to check isWrite() first: a SwapReq 576 // should always be treated like a write, but since a SwapReq implies 577 // both isWrite() and isRead() are true, check isWrite() first here. 578 // 579 if (pkt->isWrite()) { 580 // 581 // Note: M5 packets do not differentiate ST from RMW_Write 582 // 583 primary_type = secondary_type = RubyRequestType_ST; 584 } else if (pkt->isRead()) { 585 if (pkt->req->isInstFetch()) { 586 primary_type = secondary_type = RubyRequestType_IFETCH; 587 } else { 588 bool storeCheck = false; 589 // only X86 need the store check 590 if (system->getArch() == Arch::X86ISA) { 591 uint32_t flags = pkt->req->getFlags(); 592 storeCheck = flags & 593 (X86ISA::StoreCheck << X86ISA::FlagShift); 594 } 595 if (storeCheck) { 596 primary_type = RubyRequestType_RMW_Read; 597 secondary_type = RubyRequestType_ST; 598 } else { 599 primary_type = secondary_type = RubyRequestType_LD; 600 } 601 } 602 } else if (pkt->isFlush()) { 603 primary_type = secondary_type = RubyRequestType_FLUSH; 604 } else { 605 panic("Unsupported ruby packet type\n"); 606 } 607 } 608 609 RequestStatus status = insertRequest(pkt, primary_type); 610 if (status != RequestStatus_Ready) 611 return status; 612 613 issueRequest(pkt, secondary_type); 614 615 // TODO: issue hardware prefetches here 616 return RequestStatus_Issued; 617} 618 619void 620Sequencer::issueRequest(PacketPtr pkt, RubyRequestType secondary_type) 621{ 622 assert(pkt != NULL); 623 ContextID proc_id = pkt->req->hasContextId() ? 624 pkt->req->contextId() : InvalidContextID; 625 626 ContextID core_id = coreId(); 627 628 // If valid, copy the pc to the ruby request 629 Addr pc = 0; 630 if (pkt->req->hasPC()) { 631 pc = pkt->req->getPC(); 632 } 633 634 // check if the packet has data as for example prefetch and flush 635 // requests do not 636 std::shared_ptr<RubyRequest> msg = 637 std::make_shared<RubyRequest>(clockEdge(), pkt->getAddr(), 638 pkt->isFlush() ? 639 nullptr : pkt->getPtr<uint8_t>(), 640 pkt->getSize(), pc, secondary_type, 641 RubyAccessMode_Supervisor, pkt, 642 PrefetchBit_No, proc_id, core_id); 643 644 DPRINTFR(ProtocolTrace, "%15s %3s %10s%20s %6s>%-6s %#x %s\n", 645 curTick(), m_version, "Seq", "Begin", "", "", 646 printAddress(msg->getPhysicalAddress()), 647 RubyRequestType_to_string(secondary_type)); 648 649 Tick latency = cyclesToTicks( 650 m_controller->mandatoryQueueLatency(secondary_type)); 651 assert(latency > 0); 652 653 assert(m_mandatory_q_ptr != NULL); 654 m_mandatory_q_ptr->enqueue(msg, clockEdge(), latency); 655} 656 657template <class KEY, class VALUE> 658std::ostream & 659operator<<(ostream &out, const std::unordered_map<KEY, VALUE> &map) 660{ 661 auto i = map.begin(); 662 auto end = map.end(); 663 664 out << "["; 665 for (; i != end; ++i) 666 out << " " << i->first << "=" << i->second; 667 out << " ]"; 668 669 return out; 670} 671 672void 673Sequencer::print(ostream& out) const 674{ 675 out << "[Sequencer: " << m_version 676 << ", outstanding requests: " << m_outstanding_count 677 << ", read request table: " << m_readRequestTable 678 << ", write request table: " << m_writeRequestTable 679 << "]"; 680} 681 682// this can be called from setState whenever coherence permissions are 683// upgraded when invoked, coherence violations will be checked for the 684// given block 685void 686Sequencer::checkCoherence(Addr addr) 687{ 688} 689 690void 691Sequencer::recordRequestType(SequencerRequestType requestType) { 692 DPRINTF(RubyStats, "Recorded statistic: %s\n", 693 SequencerRequestType_to_string(requestType)); 694} 695 696 697void 698Sequencer::evictionCallback(Addr address) 699{ 700 ruby_eviction_callback(address); 701} 702 703void 704Sequencer::regStats() 705{ 706 RubyPort::regStats(); 707 708 m_store_waiting_on_load 709 .name(name() + ".store_waiting_on_load") 710 .desc("Number of times a store aliased with a pending load") 711 .flags(Stats::nozero); 712 m_store_waiting_on_store 713 .name(name() + ".store_waiting_on_store") 714 .desc("Number of times a store aliased with a pending store") 715 .flags(Stats::nozero); 716 m_load_waiting_on_load 717 .name(name() + ".load_waiting_on_load") 718 .desc("Number of times a load aliased with a pending load") 719 .flags(Stats::nozero); 720 m_load_waiting_on_store 721 .name(name() + ".load_waiting_on_store") 722 .desc("Number of times a load aliased with a pending store") 723 .flags(Stats::nozero); 724 725 // These statistical variables are not for display. 726 // The profiler will collate these across different 727 // sequencers and display those collated statistics. 728 m_outstandReqHist.init(10); 729 m_latencyHist.init(10); 730 m_hitLatencyHist.init(10); 731 m_missLatencyHist.init(10); 732 733 for (int i = 0; i < RubyRequestType_NUM; i++) { 734 m_typeLatencyHist.push_back(new Stats::Histogram()); 735 m_typeLatencyHist[i]->init(10); 736 737 m_hitTypeLatencyHist.push_back(new Stats::Histogram()); 738 m_hitTypeLatencyHist[i]->init(10); 739 740 m_missTypeLatencyHist.push_back(new Stats::Histogram()); 741 m_missTypeLatencyHist[i]->init(10); 742 } 743 744 for (int i = 0; i < MachineType_NUM; i++) { 745 m_hitMachLatencyHist.push_back(new Stats::Histogram()); 746 m_hitMachLatencyHist[i]->init(10); 747 748 m_missMachLatencyHist.push_back(new Stats::Histogram()); 749 m_missMachLatencyHist[i]->init(10); 750 751 m_IssueToInitialDelayHist.push_back(new Stats::Histogram()); 752 m_IssueToInitialDelayHist[i]->init(10); 753 754 m_InitialToForwardDelayHist.push_back(new Stats::Histogram()); 755 m_InitialToForwardDelayHist[i]->init(10); 756 757 m_ForwardToFirstResponseDelayHist.push_back(new Stats::Histogram()); 758 m_ForwardToFirstResponseDelayHist[i]->init(10); 759 760 m_FirstResponseToCompletionDelayHist.push_back(new Stats::Histogram()); 761 m_FirstResponseToCompletionDelayHist[i]->init(10); 762 } 763 764 for (int i = 0; i < RubyRequestType_NUM; i++) { 765 m_hitTypeMachLatencyHist.push_back(std::vector<Stats::Histogram *>()); 766 m_missTypeMachLatencyHist.push_back(std::vector<Stats::Histogram *>()); 767 768 for (int j = 0; j < MachineType_NUM; j++) { 769 m_hitTypeMachLatencyHist[i].push_back(new Stats::Histogram()); 770 m_hitTypeMachLatencyHist[i][j]->init(10); 771 772 m_missTypeMachLatencyHist[i].push_back(new Stats::Histogram()); 773 m_missTypeMachLatencyHist[i][j]->init(10); 774 } 775 } 776} 777