atomic.cc revision 2662
1/* 2 * Copyright (c) 2002-2005 The Regents of The University of Michigan 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/utility.hh" 30#include "cpu/exetrace.hh" 31#include "cpu/simple/atomic.hh" 32#include "mem/packet_impl.hh" 33#include "sim/builder.hh" 34 35using namespace std; 36using namespace TheISA; 37 38AtomicSimpleCPU::TickEvent::TickEvent(AtomicSimpleCPU *c) 39 : Event(&mainEventQueue, CPU_Tick_Pri), cpu(c) 40{ 41} 42 43 44void 45AtomicSimpleCPU::TickEvent::process() 46{ 47 cpu->tick(); 48} 49 50const char * 51AtomicSimpleCPU::TickEvent::description() 52{ 53 return "AtomicSimpleCPU tick event"; 54} 55 56 57void 58AtomicSimpleCPU::init() 59{ 60 //Create Memory Ports (conect them up) 61 Port *mem_dport = mem->getPort(""); 62 dcachePort.setPeer(mem_dport); 63 mem_dport->setPeer(&dcachePort); 64 65 Port *mem_iport = mem->getPort(""); 66 icachePort.setPeer(mem_iport); 67 mem_iport->setPeer(&icachePort); 68 69 BaseCPU::init(); 70#if FULL_SYSTEM 71 for (int i = 0; i < execContexts.size(); ++i) { 72 ExecContext *xc = execContexts[i]; 73 74 // initialize CPU, including PC 75 TheISA::initCPU(xc, xc->readCpuId()); 76 } 77#endif 78} 79 80bool 81AtomicSimpleCPU::CpuPort::recvTiming(Packet *pkt) 82{ 83 panic("AtomicSimpleCPU doesn't expect recvAtomic callback!"); 84 return true; 85} 86 87Tick 88AtomicSimpleCPU::CpuPort::recvAtomic(Packet *pkt) 89{ 90 panic("AtomicSimpleCPU doesn't expect recvAtomic callback!"); 91 return curTick; 92} 93 94void 95AtomicSimpleCPU::CpuPort::recvFunctional(Packet *pkt) 96{ 97 panic("AtomicSimpleCPU doesn't expect recvFunctional callback!"); 98} 99 100void 101AtomicSimpleCPU::CpuPort::recvStatusChange(Status status) 102{ 103 if (status == RangeChange) 104 return; 105 106 panic("AtomicSimpleCPU doesn't expect recvStatusChange callback!"); 107} 108 109void 110AtomicSimpleCPU::CpuPort::recvRetry() 111{ 112 panic("AtomicSimpleCPU doesn't expect recvRetry callback!"); 113} 114 115 116AtomicSimpleCPU::AtomicSimpleCPU(Params *p) 117 : BaseSimpleCPU(p), tickEvent(this), 118 width(p->width), simulate_stalls(p->simulate_stalls), 119 icachePort(name() + "-iport", this), dcachePort(name() + "-iport", this) 120{ 121 _status = Idle; 122 123 ifetch_req = new Request(true); 124 ifetch_req->setAsid(0); 125 // @todo fix me and get the real cpu iD!!! 126 ifetch_req->setCpuNum(0); 127 ifetch_req->setSize(sizeof(MachInst)); 128 ifetch_pkt = new Packet(ifetch_req, Packet::ReadReq, Packet::Broadcast); 129 ifetch_pkt->dataStatic(&inst); 130 131 data_read_req = new Request(true); 132 // @todo fix me and get the real cpu iD!!! 133 data_read_req->setCpuNum(0); 134 data_read_req->setAsid(0); 135 data_read_pkt = new Packet(data_read_req, Packet::ReadReq, 136 Packet::Broadcast); 137 data_read_pkt->dataStatic(&dataReg); 138 139 data_write_req = new Request(true); 140 // @todo fix me and get the real cpu iD!!! 141 data_write_req->setCpuNum(0); 142 data_write_req->setAsid(0); 143 data_write_pkt = new Packet(data_write_req, Packet::WriteReq, 144 Packet::Broadcast); 145} 146 147 148AtomicSimpleCPU::~AtomicSimpleCPU() 149{ 150} 151 152void 153AtomicSimpleCPU::serialize(ostream &os) 154{ 155 BaseSimpleCPU::serialize(os); 156 SERIALIZE_ENUM(_status); 157 nameOut(os, csprintf("%s.tickEvent", name())); 158 tickEvent.serialize(os); 159} 160 161void 162AtomicSimpleCPU::unserialize(Checkpoint *cp, const string §ion) 163{ 164 BaseSimpleCPU::unserialize(cp, section); 165 UNSERIALIZE_ENUM(_status); 166 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 167} 168 169void 170AtomicSimpleCPU::switchOut(Sampler *s) 171{ 172 sampler = s; 173 if (status() == Running) { 174 _status = SwitchedOut; 175 176 tickEvent.squash(); 177 } 178 sampler->signalSwitched(); 179} 180 181 182void 183AtomicSimpleCPU::takeOverFrom(BaseCPU *oldCPU) 184{ 185 BaseCPU::takeOverFrom(oldCPU); 186 187 assert(!tickEvent.scheduled()); 188 189 // if any of this CPU's ExecContexts are active, mark the CPU as 190 // running and schedule its tick event. 191 for (int i = 0; i < execContexts.size(); ++i) { 192 ExecContext *xc = execContexts[i]; 193 if (xc->status() == ExecContext::Active && _status != Running) { 194 _status = Running; 195 tickEvent.schedule(curTick); 196 break; 197 } 198 } 199} 200 201 202void 203AtomicSimpleCPU::activateContext(int thread_num, int delay) 204{ 205 assert(thread_num == 0); 206 assert(cpuXC); 207 208 assert(_status == Idle); 209 assert(!tickEvent.scheduled()); 210 211 notIdleFraction++; 212 tickEvent.schedule(curTick + cycles(delay)); 213 _status = Running; 214} 215 216 217void 218AtomicSimpleCPU::suspendContext(int thread_num) 219{ 220 assert(thread_num == 0); 221 assert(cpuXC); 222 223 assert(_status == Running); 224 225 // tick event may not be scheduled if this gets called from inside 226 // an instruction's execution, e.g. "quiesce" 227 if (tickEvent.scheduled()) 228 tickEvent.deschedule(); 229 230 notIdleFraction--; 231 _status = Idle; 232} 233 234 235template <class T> 236Fault 237AtomicSimpleCPU::read(Addr addr, T &data, unsigned flags) 238{ 239 data_read_req->setVaddr(addr); 240 data_read_req->setSize(sizeof(T)); 241 data_read_req->setFlags(flags); 242 data_read_req->setTime(curTick); 243 244 if (traceData) { 245 traceData->setAddr(addr); 246 } 247 248 // translate to physical address 249 Fault fault = cpuXC->translateDataReadReq(data_read_req); 250 251 // Now do the access. 252 if (fault == NoFault) { 253 data_read_pkt->reinitFromRequest(); 254 255 dcache_latency = dcachePort.sendAtomic(data_read_pkt); 256 dcache_access = true; 257 258 assert(data_read_pkt->result == Packet::Success); 259 data = data_read_pkt->get<T>(); 260 261 } 262 263 // This will need a new way to tell if it has a dcache attached. 264 if (data_read_req->getFlags() & UNCACHEABLE) 265 recordEvent("Uncached Read"); 266 267 return fault; 268} 269 270#ifndef DOXYGEN_SHOULD_SKIP_THIS 271 272template 273Fault 274AtomicSimpleCPU::read(Addr addr, uint64_t &data, unsigned flags); 275 276template 277Fault 278AtomicSimpleCPU::read(Addr addr, uint32_t &data, unsigned flags); 279 280template 281Fault 282AtomicSimpleCPU::read(Addr addr, uint16_t &data, unsigned flags); 283 284template 285Fault 286AtomicSimpleCPU::read(Addr addr, uint8_t &data, unsigned flags); 287 288#endif //DOXYGEN_SHOULD_SKIP_THIS 289 290template<> 291Fault 292AtomicSimpleCPU::read(Addr addr, double &data, unsigned flags) 293{ 294 return read(addr, *(uint64_t*)&data, flags); 295} 296 297template<> 298Fault 299AtomicSimpleCPU::read(Addr addr, float &data, unsigned flags) 300{ 301 return read(addr, *(uint32_t*)&data, flags); 302} 303 304 305template<> 306Fault 307AtomicSimpleCPU::read(Addr addr, int32_t &data, unsigned flags) 308{ 309 return read(addr, (uint32_t&)data, flags); 310} 311 312 313template <class T> 314Fault 315AtomicSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res) 316{ 317 data_write_req->setVaddr(addr); 318 data_write_req->setTime(curTick); 319 data_write_req->setSize(sizeof(T)); 320 data_write_req->setFlags(flags); 321 322 if (traceData) { 323 traceData->setAddr(addr); 324 } 325 326 // translate to physical address 327 Fault fault = cpuXC->translateDataWriteReq(data_write_req); 328 329 // Now do the access. 330 if (fault == NoFault) { 331 data = htog(data); 332 data_write_pkt->reinitFromRequest(); 333 data_write_pkt->dataStatic(&data); 334 335 dcache_latency = dcachePort.sendAtomic(data_write_pkt); 336 dcache_access = true; 337 338 assert(data_write_pkt->result == Packet::Success); 339 340 if (res && data_write_req->getFlags() & LOCKED) { 341 *res = data_write_req->getScResult(); 342 } 343 } 344 345 // This will need a new way to tell if it's hooked up to a cache or not. 346 if (data_write_req->getFlags() & UNCACHEABLE) 347 recordEvent("Uncached Write"); 348 349 // If the write needs to have a fault on the access, consider calling 350 // changeStatus() and changing it to "bad addr write" or something. 351 return fault; 352} 353 354 355#ifndef DOXYGEN_SHOULD_SKIP_THIS 356template 357Fault 358AtomicSimpleCPU::write(uint64_t data, Addr addr, 359 unsigned flags, uint64_t *res); 360 361template 362Fault 363AtomicSimpleCPU::write(uint32_t data, Addr addr, 364 unsigned flags, uint64_t *res); 365 366template 367Fault 368AtomicSimpleCPU::write(uint16_t data, Addr addr, 369 unsigned flags, uint64_t *res); 370 371template 372Fault 373AtomicSimpleCPU::write(uint8_t data, Addr addr, 374 unsigned flags, uint64_t *res); 375 376#endif //DOXYGEN_SHOULD_SKIP_THIS 377 378template<> 379Fault 380AtomicSimpleCPU::write(double data, Addr addr, unsigned flags, uint64_t *res) 381{ 382 return write(*(uint64_t*)&data, addr, flags, res); 383} 384 385template<> 386Fault 387AtomicSimpleCPU::write(float data, Addr addr, unsigned flags, uint64_t *res) 388{ 389 return write(*(uint32_t*)&data, addr, flags, res); 390} 391 392 393template<> 394Fault 395AtomicSimpleCPU::write(int32_t data, Addr addr, unsigned flags, uint64_t *res) 396{ 397 return write((uint32_t)data, addr, flags, res); 398} 399 400 401void 402AtomicSimpleCPU::tick() 403{ 404 Tick latency = cycles(1); // instruction takes one cycle by default 405 406 for (int i = 0; i < width; ++i) { 407 numCycles++; 408 409 checkForInterrupts(); 410 411 ifetch_req->resetMin(); 412 Fault fault = setupFetchRequest(ifetch_req); 413 414 if (fault == NoFault) { 415 ifetch_pkt->reinitFromRequest(); 416 417 Tick icache_latency = icachePort.sendAtomic(ifetch_pkt); 418 // ifetch_req is initialized to read the instruction directly 419 // into the CPU object's inst field. 420 421 dcache_access = false; // assume no dcache access 422 preExecute(); 423 fault = curStaticInst->execute(this, traceData); 424 postExecute(); 425 426 if (simulate_stalls) { 427 // This calculation assumes that the icache and dcache 428 // access latencies are always a multiple of the CPU's 429 // cycle time. If not, the next tick event may get 430 // scheduled at a non-integer multiple of the CPU 431 // cycle time. 432 Tick icache_stall = icache_latency - cycles(1); 433 Tick dcache_stall = 434 dcache_access ? dcache_latency - cycles(1) : 0; 435 latency += icache_stall + dcache_stall; 436 } 437 438 } 439 440 advancePC(fault); 441 } 442 443 if (_status != Idle) 444 tickEvent.schedule(curTick + latency); 445} 446 447 448//////////////////////////////////////////////////////////////////////// 449// 450// AtomicSimpleCPU Simulation Object 451// 452BEGIN_DECLARE_SIM_OBJECT_PARAMS(AtomicSimpleCPU) 453 454 Param<Counter> max_insts_any_thread; 455 Param<Counter> max_insts_all_threads; 456 Param<Counter> max_loads_any_thread; 457 Param<Counter> max_loads_all_threads; 458 SimObjectParam<MemObject *> mem; 459 460#if FULL_SYSTEM 461 SimObjectParam<AlphaITB *> itb; 462 SimObjectParam<AlphaDTB *> dtb; 463 SimObjectParam<System *> system; 464 Param<int> cpu_id; 465 Param<Tick> profile; 466#else 467 SimObjectParam<Process *> workload; 468#endif // FULL_SYSTEM 469 470 Param<int> clock; 471 472 Param<bool> defer_registration; 473 Param<int> width; 474 Param<bool> function_trace; 475 Param<Tick> function_trace_start; 476 Param<bool> simulate_stalls; 477 478END_DECLARE_SIM_OBJECT_PARAMS(AtomicSimpleCPU) 479 480BEGIN_INIT_SIM_OBJECT_PARAMS(AtomicSimpleCPU) 481 482 INIT_PARAM(max_insts_any_thread, 483 "terminate when any thread reaches this inst count"), 484 INIT_PARAM(max_insts_all_threads, 485 "terminate when all threads have reached this inst count"), 486 INIT_PARAM(max_loads_any_thread, 487 "terminate when any thread reaches this load count"), 488 INIT_PARAM(max_loads_all_threads, 489 "terminate when all threads have reached this load count"), 490 INIT_PARAM(mem, "memory"), 491 492#if FULL_SYSTEM 493 INIT_PARAM(itb, "Instruction TLB"), 494 INIT_PARAM(dtb, "Data TLB"), 495 INIT_PARAM(system, "system object"), 496 INIT_PARAM(cpu_id, "processor ID"), 497 INIT_PARAM(profile, ""), 498#else 499 INIT_PARAM(workload, "processes to run"), 500#endif // FULL_SYSTEM 501 502 INIT_PARAM(clock, "clock speed"), 503 INIT_PARAM(defer_registration, "defer system registration (for sampling)"), 504 INIT_PARAM(width, "cpu width"), 505 INIT_PARAM(function_trace, "Enable function trace"), 506 INIT_PARAM(function_trace_start, "Cycle to start function trace"), 507 INIT_PARAM(simulate_stalls, "Simulate cache stall cycles") 508 509END_INIT_SIM_OBJECT_PARAMS(AtomicSimpleCPU) 510 511 512CREATE_SIM_OBJECT(AtomicSimpleCPU) 513{ 514 AtomicSimpleCPU::Params *params = new AtomicSimpleCPU::Params(); 515 params->name = getInstanceName(); 516 params->numberOfThreads = 1; 517 params->max_insts_any_thread = max_insts_any_thread; 518 params->max_insts_all_threads = max_insts_all_threads; 519 params->max_loads_any_thread = max_loads_any_thread; 520 params->max_loads_all_threads = max_loads_all_threads; 521 params->deferRegistration = defer_registration; 522 params->clock = clock; 523 params->functionTrace = function_trace; 524 params->functionTraceStart = function_trace_start; 525 params->width = width; 526 params->simulate_stalls = simulate_stalls; 527 params->mem = mem; 528 529#if FULL_SYSTEM 530 params->itb = itb; 531 params->dtb = dtb; 532 params->system = system; 533 params->cpu_id = cpu_id; 534 params->profile = profile; 535#else 536 params->process = workload; 537#endif 538 539 AtomicSimpleCPU *cpu = new AtomicSimpleCPU(params); 540 return cpu; 541} 542 543REGISTER_SIM_OBJECT("AtomicSimpleCPU", AtomicSimpleCPU) 544 545