1/*
2 * Copyright (c) 2004-2005 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 * Authors: Kevin Lim
29 * Korey Sewell
30 */
31
32#ifndef __CPU_O3_CPU_HH__
33#define __CPU_O3_CPU_HH__
34
35#include <iostream>
36#include <list>
37#include <queue>
38#include <set>
39#include <vector>
40
41#include "arch/isa_traits.hh"
42#include "base/statistics.hh"
43#include "base/timebuf.hh"
44#include "config/full_system.hh"
45#include "cpu/activity.hh"
46#include "cpu/base.hh"
47#include "cpu/simple_thread.hh"
48#include "cpu/o3/comm.hh"
49#include "cpu/o3/cpu_policy.hh"
50#include "cpu/o3/scoreboard.hh"
51#include "cpu/o3/thread_state.hh"
52//#include "cpu/o3/thread_context.hh"
53#include "sim/process.hh"
54
55template <class>
56class Checker;
57class ThreadContext;
58template <class>
59class O3ThreadContext;
60class MemObject;
61class Process;
62
63class BaseO3CPU : public BaseCPU
64{
65 //Stuff that's pretty ISA independent will go here.
66 public:
67 typedef BaseCPU::Params Params;
68
69 BaseO3CPU(Params *params);
70
71 void regStats();
72
73 /** Sets this CPU's ID. */
74 void setCpuId(int id) { cpu_id = id; }
75
76 /** Reads this CPU's ID. */
77 int readCpuId() { return cpu_id; }
78
79 protected:
80 int cpu_id;
81};
82
83/**
84 * FullO3CPU class, has each of the stages (fetch through commit)
85 * within it, as well as all of the time buffers between stages. The
86 * tick() function for the CPU is defined here.
87 */
88template <class Impl>
89class FullO3CPU : public BaseO3CPU
90{
91 public:
92 typedef TheISA::FloatReg FloatReg;
93 typedef TheISA::FloatRegBits FloatRegBits;
94
95 // Typedefs from the Impl here.
96 typedef typename Impl::CPUPol CPUPolicy;
97 typedef typename Impl::Params Params;
98 typedef typename Impl::DynInstPtr DynInstPtr;
99
100 typedef O3ThreadState<Impl> Thread;
101
102 typedef typename std::list<DynInstPtr>::iterator ListIt;
103
104 friend class O3ThreadContext<Impl>;
105
106 public:
107 enum Status {
108 Running,
109 Idle,
110 Halted,
111 Blocked,
112 SwitchedOut
113 };
114
115 /** Overall CPU status. */
116 Status _status;
117
118 /** Per-thread status in CPU, used for SMT. */
119 Status _threadStatus[Impl::MaxThreads];
120
121 private:
122 class TickEvent : public Event
123 {
124 private:
125 /** Pointer to the CPU. */
126 FullO3CPU<Impl> *cpu;
127
128 public:
129 /** Constructs a tick event. */
130 TickEvent(FullO3CPU<Impl> *c);
131
132 /** Processes a tick event, calling tick() on the CPU. */
133 void process();
134 /** Returns the description of the tick event. */
135 const char *description();
136 };
137
138 /** The tick event used for scheduling CPU ticks. */
139 TickEvent tickEvent;
140
141 /** Schedule tick event, regardless of its current state. */
142 void scheduleTickEvent(int delay)
143 {
144 if (tickEvent.squashed())
145 tickEvent.reschedule(curTick + cycles(delay));
146 else if (!tickEvent.scheduled())
147 tickEvent.schedule(curTick + cycles(delay));
148 }
149
150 /** Unschedule tick event, regardless of its current state. */
151 void unscheduleTickEvent()
152 {
153 if (tickEvent.scheduled())
154 tickEvent.squash();
155 }
156
157 class ActivateThreadEvent : public Event
158 {
159 private:
160 /** Number of Thread to Activate */
161 int tid;
162
163 /** Pointer to the CPU. */
164 FullO3CPU<Impl> *cpu;
165
166 public:
167 /** Constructs the event. */
168 ActivateThreadEvent();
169
170 /** Initialize Event */
171 void init(int thread_num, FullO3CPU<Impl> *thread_cpu);
172
173 /** Processes the event, calling activateThread() on the CPU. */
174 void process();
175
176 /** Returns the description of the event. */
177 const char *description();
178 };
179
180 /** Schedule thread to activate , regardless of its current state. */
181 void scheduleActivateThreadEvent(int tid, int delay)
182 {
183 // Schedule thread to activate, regardless of its current state.
184 if (activateThreadEvent[tid].squashed())
185 activateThreadEvent[tid].reschedule(curTick + cycles(delay));
186 else if (!activateThreadEvent[tid].scheduled())
187 activateThreadEvent[tid].schedule(curTick + cycles(delay));
188 }
189
190 /** Unschedule actiavte thread event, regardless of its current state. */
191 void unscheduleActivateThreadEvent(int tid)
192 {
193 if (activateThreadEvent[tid].scheduled())
194 activateThreadEvent[tid].squash();
195 }
196
197 /** The tick event used for scheduling CPU ticks. */
198 ActivateThreadEvent activateThreadEvent[Impl::MaxThreads];
199
200 public:
201 /** Constructs a CPU with the given parameters. */
202 FullO3CPU(Params *params);
203 /** Destructor. */
204 ~FullO3CPU();
205
206 /** Registers statistics. */
207 void fullCPURegStats();
208
209 /** Ticks CPU, calling tick() on each stage, and checking the overall
210 * activity to see if the CPU should deschedule itself.
211 */
212 void tick();
213
214 /** Initialize the CPU */
215 void init();
216
217 /** Returns the Number of Active Threads in the CPU */
218 int numActiveThreads()
219 { return activeThreads.size(); }
220
221 /** Add Thread to Active Threads List */
222 void activateThread(unsigned int tid);
223
224 /** Setup CPU to insert a thread's context */
225 void insertThread(unsigned tid);
226
227 /** Remove all of a thread's context from CPU */
228 void removeThread(unsigned tid);
229
230 /** Count the Total Instructions Committed in the CPU. */
231 virtual Counter totalInstructions() const
232 {
233 Counter total(0);
234
235 for (int i=0; i < thread.size(); i++)
236 total += thread[i]->numInst;
237
238 return total;
239 }
240
241 /** Add Thread to Active Threads List. */
242 void activateContext(int tid, int delay);
243
244 /** Remove Thread from Active Threads List */
245 void suspendContext(int tid);
246
247 /** Remove Thread from Active Threads List &&
248 * Remove Thread Context from CPU.
249 */
250 void deallocateContext(int tid);
251
252 /** Remove Thread from Active Threads List &&
253 * Remove Thread Context from CPU.
254 */
255 void haltContext(int tid);
256
257 /** Activate a Thread When CPU Resources are Available. */
258 void activateWhenReady(int tid);
259
260 /** Add or Remove a Thread Context in the CPU. */
261 void doContextSwitch();
262
263 /** Update The Order In Which We Process Threads. */
264 void updateThreadPriority();
265
266 /** Executes a syscall on this cycle.
267 * ---------------------------------------
268 * Note: this is a virtual function. CPU-Specific
269 * functionality defined in derived classes
270 */
271 virtual void syscall(int tid) { panic("Unimplemented!"); }
272
273 /** Switches out this CPU. */
274 void switchOut(Sampler *sampler);
275
276 /** Signals to this CPU that a stage has completed switching out. */
277 void signalSwitched();
278
279 /** Takes over from another CPU. */
280 void takeOverFrom(BaseCPU *oldCPU);
281
282 /** Get the current instruction sequence number, and increment it. */
283 InstSeqNum getAndIncrementInstSeq()
284 { return globalSeqNum++; }
285
286#if FULL_SYSTEM
287 /** Check if this address is a valid instruction address. */
288 bool validInstAddr(Addr addr) { return true; }
289
290 /** Check if this address is a valid data address. */
291 bool validDataAddr(Addr addr) { return true; }
292
293 /** Get instruction asid. */
294 int getInstAsid(unsigned tid)
295 { return regFile.miscRegs[tid].getInstAsid(); }
296
297 /** Get data asid. */
298 int getDataAsid(unsigned tid)
299 { return regFile.miscRegs[tid].getDataAsid(); }
300#else
301 /** Get instruction asid. */
302 int getInstAsid(unsigned tid)
303 { return thread[tid]->getInstAsid(); }
304
305 /** Get data asid. */
306 int getDataAsid(unsigned tid)
307 { return thread[tid]->getDataAsid(); }
308
309#endif
310
311 /** Register accessors. Index refers to the physical register index. */
312 uint64_t readIntReg(int reg_idx);
313
314 FloatReg readFloatReg(int reg_idx);
315
316 FloatReg readFloatReg(int reg_idx, int width);
317
318 FloatRegBits readFloatRegBits(int reg_idx);
319
320 FloatRegBits readFloatRegBits(int reg_idx, int width);
321
322 void setIntReg(int reg_idx, uint64_t val);
323
324 void setFloatReg(int reg_idx, FloatReg val);
325
326 void setFloatReg(int reg_idx, FloatReg val, int width);
327
328 void setFloatRegBits(int reg_idx, FloatRegBits val);
329
330 void setFloatRegBits(int reg_idx, FloatRegBits val, int width);
331
332 uint64_t readArchIntReg(int reg_idx, unsigned tid);
333
334 float readArchFloatRegSingle(int reg_idx, unsigned tid);
335
336 double readArchFloatRegDouble(int reg_idx, unsigned tid);
337
338 uint64_t readArchFloatRegInt(int reg_idx, unsigned tid);
339
340 /** Architectural register accessors. Looks up in the commit
341 * rename table to obtain the true physical index of the
342 * architected register first, then accesses that physical
343 * register.
344 */
345 void setArchIntReg(int reg_idx, uint64_t val, unsigned tid);
346
347 void setArchFloatRegSingle(int reg_idx, float val, unsigned tid);
348
349 void setArchFloatRegDouble(int reg_idx, double val, unsigned tid);
350
351 void setArchFloatRegInt(int reg_idx, uint64_t val, unsigned tid);
352
353 /** Reads the commit PC of a specific thread. */
354 uint64_t readPC(unsigned tid);
355
356 /** Sets the commit PC of a specific thread. */
357 void setPC(Addr new_PC, unsigned tid);
358
359 /** Reads the next PC of a specific thread. */
360 uint64_t readNextPC(unsigned tid);
361
362 /** Sets the next PC of a specific thread. */
363 void setNextPC(uint64_t val, unsigned tid);
364
365 /** Reads the next NPC of a specific thread. */
366 uint64_t readNextNPC(unsigned tid);
367
368 /** Sets the next NPC of a specific thread. */
369 void setNextNPC(uint64_t val, unsigned tid);
370
371 /** Function to add instruction onto the head of the list of the
372 * instructions. Used when new instructions are fetched.
373 */
374 ListIt addInst(DynInstPtr &inst);
375
376 /** Function to tell the CPU that an instruction has completed. */
377 void instDone(unsigned tid);
378
379 /** Add Instructions to the CPU Remove List*/
380 void addToRemoveList(DynInstPtr &inst);
381
382 /** Remove an instruction from the front end of the list. There's
383 * no restriction on location of the instruction.
384 */
385 void removeFrontInst(DynInstPtr &inst);
386
387 /** Remove all instructions that are not currently in the ROB. */
388 void removeInstsNotInROB(unsigned tid);
389
390 /** Remove all instructions younger than the given sequence number. */
391 void removeInstsUntil(const InstSeqNum &seq_num,unsigned tid);
392
393 /** Removes the instruction pointed to by the iterator. */
394 inline void squashInstIt(const ListIt &instIt, const unsigned &tid);
395
396 /** Cleans up all instructions on the remove list. */
397 void cleanUpRemovedInsts();
398
399 /** Debug function to print all instructions on the list. */
400 void dumpInsts();
401
402 public:
403 /** List of all the instructions in flight. */
404 std::list<DynInstPtr> instList;
405
406 /** List of all the instructions that will be removed at the end of this
407 * cycle.
408 */
409 std::queue<ListIt> removeList;
410
411#ifdef DEBUG
412 /** Debug structure to keep track of the sequence numbers still in
413 * flight.
414 */
415 std::set<InstSeqNum> snList;
416#endif
417
418 /** Records if instructions need to be removed this cycle due to
419 * being retired or squashed.
420 */
421 bool removeInstsThisCycle;
422
423 protected:
424 /** The fetch stage. */
425 typename CPUPolicy::Fetch fetch;
426
427 /** The decode stage. */
428 typename CPUPolicy::Decode decode;
429
430 /** The dispatch stage. */
431 typename CPUPolicy::Rename rename;
432
433 /** The issue/execute/writeback stages. */
434 typename CPUPolicy::IEW iew;
435
436 /** The commit stage. */
437 typename CPUPolicy::Commit commit;
438
439 /** The register file. */
440 typename CPUPolicy::RegFile regFile;
441
442 /** The free list. */
443 typename CPUPolicy::FreeList freeList;
444
445 /** The rename map. */
446 typename CPUPolicy::RenameMap renameMap[Impl::MaxThreads];
447
448 /** The commit rename map. */
449 typename CPUPolicy::RenameMap commitRenameMap[Impl::MaxThreads];
450
451 /** The re-order buffer. */
452 typename CPUPolicy::ROB rob;
453
454 /** Active Threads List */
455 std::list<unsigned> activeThreads;
456
457 /** Integer Register Scoreboard */
458 Scoreboard scoreboard;
459
460 public:
461 /** Enum to give each stage a specific index, so when calling
462 * activateStage() or deactivateStage(), they can specify which stage
463 * is being activated/deactivated.
464 */
465 enum StageIdx {
466 FetchIdx,
467 DecodeIdx,
468 RenameIdx,
469 IEWIdx,
470 CommitIdx,
471 NumStages };
472
473 /** Typedefs from the Impl to get the structs that each of the
474 * time buffers should use.
475 */
476 typedef typename CPUPolicy::TimeStruct TimeStruct;
477
478 typedef typename CPUPolicy::FetchStruct FetchStruct;
479
480 typedef typename CPUPolicy::DecodeStruct DecodeStruct;
481
482 typedef typename CPUPolicy::RenameStruct RenameStruct;
483
484 typedef typename CPUPolicy::IEWStruct IEWStruct;
485
486 /** The main time buffer to do backwards communication. */
487 TimeBuffer<TimeStruct> timeBuffer;
488
489 /** The fetch stage's instruction queue. */
490 TimeBuffer<FetchStruct> fetchQueue;
491
492 /** The decode stage's instruction queue. */
493 TimeBuffer<DecodeStruct> decodeQueue;
494
495 /** The rename stage's instruction queue. */
496 TimeBuffer<RenameStruct> renameQueue;
497
498 /** The IEW stage's instruction queue. */
499 TimeBuffer<IEWStruct> iewQueue;
500
501 private:
502 /** The activity recorder; used to tell if the CPU has any
503 * activity remaining or if it can go to idle and deschedule
504 * itself.
505 */
506 ActivityRecorder activityRec;
507
508 public:
509 /** Records that there was time buffer activity this cycle. */
510 void activityThisCycle() { activityRec.activity(); }
511
512 /** Changes a stage's status to active within the activity recorder. */
513 void activateStage(const StageIdx idx)
514 { activityRec.activateStage(idx); }
515
516 /** Changes a stage's status to inactive within the activity recorder. */
517 void deactivateStage(const StageIdx idx)
518 { activityRec.deactivateStage(idx); }
519
520 /** Wakes the CPU, rescheduling the CPU if it's not already active. */
521 void wakeCPU();
522
523 /** Gets a free thread id. Use if thread ids change across system. */
524 int getFreeTid();
525
526 public:
527 /** Returns a pointer to a thread context. */
528 ThreadContext *tcBase(unsigned tid)
529 {
530 return thread[tid]->getTC();
531 }
532
533 /** The global sequence number counter. */
534 InstSeqNum globalSeqNum;
535
536 /** Pointer to the checker, which can dynamically verify
537 * instruction results at run time. This can be set to NULL if it
538 * is not being used.
539 */
540 Checker<DynInstPtr> *checker;
541
542#if FULL_SYSTEM
543 /** Pointer to the system. */
544 System *system;
545
546 /** Pointer to physical memory. */
547 PhysicalMemory *physmem;
548#endif
549
550 /** Pointer to memory. */
551 MemObject *mem;
552
553 /** Pointer to the sampler */
554 Sampler *sampler;
555
556 /** Counter of how many stages have completed switching out. */
557 int switchCount;
558
559 /** Pointers to all of the threads in the CPU. */
560 std::vector<Thread *> thread;
561
562 /** Pointer to the icache interface. */
563 MemInterface *icacheInterface;
564 /** Pointer to the dcache interface. */
565 MemInterface *dcacheInterface;
566
567 /** Whether or not the CPU should defer its registration. */
568 bool deferRegistration;
569
570 /** Is there a context switch pending? */
571 bool contextSwitch;
572
573 /** Threads Scheduled to Enter CPU */
574 std::list<int> cpuWaitList;
575
576 /** The cycle that the CPU was last running, used for statistics. */
577 Tick lastRunningCycle;
578
579 /** The cycle that the CPU was last activated by a new thread*/
580 Tick lastActivatedCycle;
581
582 /** Number of Threads CPU can process */
583 unsigned numThreads;
584
585 /** Mapping for system thread id to cpu id */
586 std::map<unsigned,unsigned> threadMap;
587
588 /** Available thread ids in the cpu*/
589 std::vector<unsigned> tids;
590
591 /** Stat for total number of times the CPU is descheduled. */
592 Stats::Scalar<> timesIdled;
593 /** Stat for total number of cycles the CPU spends descheduled. */
594 Stats::Scalar<> idleCycles;
595 /** Stat for the number of committed instructions per thread. */
596 Stats::Vector<> committedInsts;
597 /** Stat for the total number of committed instructions. */
598 Stats::Scalar<> totalCommittedInsts;
599 /** Stat for the CPI per thread. */
600 Stats::Formula cpi;
601 /** Stat for the total CPI. */
602 Stats::Formula totalCpi;
603 /** Stat for the IPC per thread. */
604 Stats::Formula ipc;
605 /** Stat for the total IPC. */
606 Stats::Formula totalIpc;
607};
608
609#endif // __CPU_O3_CPU_HH__