2 * Copyright (c) 2011 ARM Limited
5 * The license below extends only to copyright in the software and shall
6 * not be construed as granting a license to any other intellectual
7 * property including but not limited to intellectual property relating
8 * to a hardware implementation of the functionality of the software
9 * licensed hereunder. You may use the software subject to the license
10 * terms below provided that you ensure that this notice is replicated
11 * unmodified and in its entirety in all distributions of the software,
12 * modified or unmodified, in source code or in binary form.
14 * Copyright (c) 2004-2005 The Regents of The University of Michigan
15 * Copyright (c) 2011 Regents of the University of California
16 * All rights reserved.
18 * Redistribution and use in source and binary forms, with or without
19 * modification, are permitted provided that the following conditions are
20 * met: redistributions of source code must retain the above copyright
21 * notice, this list of conditions and the following disclaimer;
22 * redistributions in binary form must reproduce the above copyright
23 * notice, this list of conditions and the following disclaimer in the
24 * documentation and/or other materials provided with the distribution;
25 * neither the name of the copyright holders nor the names of its
26 * contributors may be used to endorse or promote products derived from
27 * this software without specific prior written permission.
29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
33 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
34 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
35 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
36 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
37 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
38 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
39 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
46 #ifndef __CPU_O3_CPU_HH__
47 #define __CPU_O3_CPU_HH__
55 #include "arch/types.hh"
56 #include "base/statistics.hh"
57 #include "config/the_isa.hh"
58 #include "cpu/o3/comm.hh"
59 #include "cpu/o3/cpu_policy.hh"
60 #include "cpu/o3/scoreboard.hh"
61 #include "cpu/o3/thread_state.hh"
62 #include "cpu/activity.hh"
63 #include "cpu/base.hh"
64 #include "cpu/simple_thread.hh"
65 #include "cpu/timebuf.hh"
66 //#include "cpu/o3/thread_context.hh"
67 #include "params/DerivO3CPU.hh"
68 #include "sim/process.hh"
74 class O3ThreadContext;
82 class BaseO3CPU : public BaseCPU
84 //Stuff that's pretty ISA independent will go here.
86 BaseO3CPU(BaseCPUParams *params);
92 * FullO3CPU class, has each of the stages (fetch through commit)
93 * within it, as well as all of the time buffers between stages. The
94 * tick() function for the CPU is defined here.
97 class FullO3CPU : public BaseO3CPU
100 // Typedefs from the Impl here.
101 typedef typename Impl::CPUPol CPUPolicy;
102 typedef typename Impl::DynInstPtr DynInstPtr;
103 typedef typename Impl::O3CPU O3CPU;
105 typedef O3ThreadState<Impl> ImplState;
106 typedef O3ThreadState<Impl> Thread;
108 typedef typename std::list<DynInstPtr>::iterator ListIt;
110 friend class O3ThreadContext<Impl>;
124 /** Overall CPU status. */
127 /** Per-thread status in CPU, used for SMT. */
128 Status _threadStatus[Impl::MaxThreads];
133 * IcachePort class for instruction fetch.
135 class IcachePort : public CpuPort
138 /** Pointer to fetch. */
139 DefaultFetch<Impl> *fetch;
142 /** Default constructor. */
143 IcachePort(DefaultFetch<Impl> *_fetch, FullO3CPU<Impl>* _cpu)
144 : CpuPort(_fetch->name() + "-iport", _cpu), fetch(_fetch)
149 /** Timing version of receive. Handles setting fetch to the
150 * proper status to start fetching. */
151 virtual bool recvTiming(PacketPtr pkt);
153 /** Handles doing a retry of a failed fetch. */
154 virtual void recvRetry();
158 * DcachePort class for the load/store queue.
160 class DcachePort : public CpuPort
164 /** Pointer to LSQ. */
168 /** Default constructor. */
169 DcachePort(LSQ<Impl> *_lsq, FullO3CPU<Impl>* _cpu)
170 : CpuPort(_lsq->name() + "-dport", _cpu), lsq(_lsq)
175 /** Timing version of receive. Handles writing back and
176 * completing the load or store that has returned from
178 virtual bool recvTiming(PacketPtr pkt);
180 /** Handles doing a retry of the previous send. */
181 virtual void recvRetry();
184 * As this CPU requires snooping to maintain the load store queue
185 * change the behaviour from the base CPU port.
187 * @return true since we have to snoop
189 virtual bool isSnooping() const { return true; }
192 class TickEvent : public Event
195 /** Pointer to the CPU. */
196 FullO3CPU<Impl> *cpu;
199 /** Constructs a tick event. */
200 TickEvent(FullO3CPU<Impl> *c);
202 /** Processes a tick event, calling tick() on the CPU. */
204 /** Returns the description of the tick event. */
205 const char *description() const;
208 /** The tick event used for scheduling CPU ticks. */
211 /** Schedule tick event, regardless of its current state. */
212 void scheduleTickEvent(int delay)
214 if (tickEvent.squashed())
215 reschedule(tickEvent, nextCycle(curTick() + ticks(delay)));
216 else if (!tickEvent.scheduled())
217 schedule(tickEvent, nextCycle(curTick() + ticks(delay)));
220 /** Unschedule tick event, regardless of its current state. */
221 void unscheduleTickEvent()
223 if (tickEvent.scheduled())
227 class ActivateThreadEvent : public Event
230 /** Number of Thread to Activate */
233 /** Pointer to the CPU. */
234 FullO3CPU<Impl> *cpu;
237 /** Constructs the event. */
238 ActivateThreadEvent();
240 /** Initialize Event */
241 void init(int thread_num, FullO3CPU<Impl> *thread_cpu);
243 /** Processes the event, calling activateThread() on the CPU. */
246 /** Returns the description of the event. */
247 const char *description() const;
250 /** Schedule thread to activate , regardless of its current state. */
252 scheduleActivateThreadEvent(ThreadID tid, int delay)
254 // Schedule thread to activate, regardless of its current state.
255 if (activateThreadEvent[tid].squashed())
256 reschedule(activateThreadEvent[tid],
257 nextCycle(curTick() + ticks(delay)));
258 else if (!activateThreadEvent[tid].scheduled()) {
259 Tick when = nextCycle(curTick() + ticks(delay));
261 // Check if the deallocateEvent is also scheduled, and make
262 // sure they do not happen at same time causing a sleep that
263 // is never woken from.
264 if (deallocateContextEvent[tid].scheduled() &&
265 deallocateContextEvent[tid].when() == when) {
269 schedule(activateThreadEvent[tid], when);
273 /** Unschedule actiavte thread event, regardless of its current state. */
275 unscheduleActivateThreadEvent(ThreadID tid)
277 if (activateThreadEvent[tid].scheduled())
278 activateThreadEvent[tid].squash();
281 /** The tick event used for scheduling CPU ticks. */
282 ActivateThreadEvent activateThreadEvent[Impl::MaxThreads];
284 class DeallocateContextEvent : public Event
287 /** Number of Thread to deactivate */
290 /** Should the thread be removed from the CPU? */
293 /** Pointer to the CPU. */
294 FullO3CPU<Impl> *cpu;
297 /** Constructs the event. */
298 DeallocateContextEvent();
300 /** Initialize Event */
301 void init(int thread_num, FullO3CPU<Impl> *thread_cpu);
303 /** Processes the event, calling activateThread() on the CPU. */
306 /** Sets whether the thread should also be removed from the CPU. */
307 void setRemove(bool _remove) { remove = _remove; }
309 /** Returns the description of the event. */
310 const char *description() const;
313 /** Schedule cpu to deallocate thread context.*/
315 scheduleDeallocateContextEvent(ThreadID tid, bool remove, int delay)
317 // Schedule thread to activate, regardless of its current state.
318 if (deallocateContextEvent[tid].squashed())
319 reschedule(deallocateContextEvent[tid],
320 nextCycle(curTick() + ticks(delay)));
321 else if (!deallocateContextEvent[tid].scheduled())
322 schedule(deallocateContextEvent[tid],
323 nextCycle(curTick() + ticks(delay)));
326 /** Unschedule thread deallocation in CPU */
328 unscheduleDeallocateContextEvent(ThreadID tid)
330 if (deallocateContextEvent[tid].scheduled())
331 deallocateContextEvent[tid].squash();
334 /** The tick event used for scheduling CPU ticks. */
335 DeallocateContextEvent deallocateContextEvent[Impl::MaxThreads];
338 /** Constructs a CPU with the given parameters. */
339 FullO3CPU(DerivO3CPUParams *params);
343 /** Registers statistics. */
346 void demapPage(Addr vaddr, uint64_t asn)
348 this->itb->demapPage(vaddr, asn);
349 this->dtb->demapPage(vaddr, asn);
352 void demapInstPage(Addr vaddr, uint64_t asn)
354 this->itb->demapPage(vaddr, asn);
357 void demapDataPage(Addr vaddr, uint64_t asn)
359 this->dtb->demapPage(vaddr, asn);
362 /** Ticks CPU, calling tick() on each stage, and checking the overall
363 * activity to see if the CPU should deschedule itself.
367 /** Initialize the CPU */
370 /** Returns the Number of Active Threads in the CPU */
371 int numActiveThreads()
372 { return activeThreads.size(); }
374 /** Add Thread to Active Threads List */
375 void activateThread(ThreadID tid);
377 /** Remove Thread from Active Threads List */
378 void deactivateThread(ThreadID tid);
380 /** Setup CPU to insert a thread's context */
381 void insertThread(ThreadID tid);
383 /** Remove all of a thread's context from CPU */
384 void removeThread(ThreadID tid);
386 /** Count the Total Instructions Committed in the CPU. */
387 virtual Counter totalInsts() const;
389 /** Count the Total Ops (including micro ops) committed in the CPU. */
390 virtual Counter totalOps() const;
392 /** Add Thread to Active Threads List. */
393 void activateContext(ThreadID tid, int delay);
395 /** Remove Thread from Active Threads List */
396 void suspendContext(ThreadID tid);
398 /** Remove Thread from Active Threads List &&
399 * Possibly Remove Thread Context from CPU.
401 bool scheduleDeallocateContext(ThreadID tid, bool remove, int delay = 1);
403 /** Remove Thread from Active Threads List &&
404 * Remove Thread Context from CPU.
406 void haltContext(ThreadID tid);
408 /** Activate a Thread When CPU Resources are Available. */
409 void activateWhenReady(ThreadID tid);
411 /** Add or Remove a Thread Context in the CPU. */
412 void doContextSwitch();
414 /** Update The Order In Which We Process Threads. */
415 void updateThreadPriority();
417 /** Serialize state. */
418 virtual void serialize(std::ostream &os);
420 /** Unserialize from a checkpoint. */
421 virtual void unserialize(Checkpoint *cp, const std::string §ion);
424 /** Executes a syscall.
425 * @todo: Determine if this needs to be virtual.
427 void syscall(int64_t callnum, ThreadID tid);
429 /** Starts draining the CPU's pipeline of all instructions in
430 * order to stop all memory accesses. */
431 virtual unsigned int drain(Event *drain_event);
433 /** Resumes execution after a drain. */
434 virtual void resume();
436 /** Signals to this CPU that a stage has completed switching out. */
437 void signalDrained();
439 /** Switches out this CPU. */
440 virtual void switchOut();
442 /** Takes over from another CPU. */
443 virtual void takeOverFrom(BaseCPU *oldCPU);
445 /** Get the current instruction sequence number, and increment it. */
446 InstSeqNum getAndIncrementInstSeq()
447 { return globalSeqNum++; }
449 /** Traps to handle given fault. */
450 void trap(Fault fault, ThreadID tid, StaticInstPtr inst);
452 /** HW return from error interrupt. */
453 Fault hwrei(ThreadID tid);
455 bool simPalCheck(int palFunc, ThreadID tid);
457 /** Returns the Fault for any valid interrupt. */
458 Fault getInterrupts();
460 /** Processes any an interrupt fault. */
461 void processInterrupts(Fault interrupt);
463 /** Halts the CPU. */
464 void halt() { panic("Halt not implemented!\n"); }
466 /** Check if this address is a valid instruction address. */
467 bool validInstAddr(Addr addr) { return true; }
469 /** Check if this address is a valid data address. */
470 bool validDataAddr(Addr addr) { return true; }
472 /** Register accessors. Index refers to the physical register index. */
474 /** Reads a miscellaneous register. */
475 TheISA::MiscReg readMiscRegNoEffect(int misc_reg, ThreadID tid);
477 /** Reads a misc. register, including any side effects the read
478 * might have as defined by the architecture.
480 TheISA::MiscReg readMiscReg(int misc_reg, ThreadID tid);
482 /** Sets a miscellaneous register. */
483 void setMiscRegNoEffect(int misc_reg, const TheISA::MiscReg &val,
486 /** Sets a misc. register, including any side effects the write
487 * might have as defined by the architecture.
489 void setMiscReg(int misc_reg, const TheISA::MiscReg &val,
492 uint64_t readIntReg(int reg_idx);
494 TheISA::FloatReg readFloatReg(int reg_idx);
496 TheISA::FloatRegBits readFloatRegBits(int reg_idx);
498 void setIntReg(int reg_idx, uint64_t val);
500 void setFloatReg(int reg_idx, TheISA::FloatReg val);
502 void setFloatRegBits(int reg_idx, TheISA::FloatRegBits val);
504 uint64_t readArchIntReg(int reg_idx, ThreadID tid);
506 float readArchFloatReg(int reg_idx, ThreadID tid);
508 uint64_t readArchFloatRegInt(int reg_idx, ThreadID tid);
510 /** Architectural register accessors. Looks up in the commit
511 * rename table to obtain the true physical index of the
512 * architected register first, then accesses that physical
515 void setArchIntReg(int reg_idx, uint64_t val, ThreadID tid);
517 void setArchFloatReg(int reg_idx, float val, ThreadID tid);
519 void setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid);
521 /** Sets the commit PC state of a specific thread. */
522 void pcState(const TheISA::PCState &newPCState, ThreadID tid);
524 /** Reads the commit PC state of a specific thread. */
525 TheISA::PCState pcState(ThreadID tid);
527 /** Reads the commit PC of a specific thread. */
528 Addr instAddr(ThreadID tid);
530 /** Reads the commit micro PC of a specific thread. */
531 MicroPC microPC(ThreadID tid);
533 /** Reads the next PC of a specific thread. */
534 Addr nextInstAddr(ThreadID tid);
536 /** Initiates a squash of all in-flight instructions for a given
537 * thread. The source of the squash is an external update of
538 * state through the TC.
540 void squashFromTC(ThreadID tid);
542 /** Function to add instruction onto the head of the list of the
543 * instructions. Used when new instructions are fetched.
545 ListIt addInst(DynInstPtr &inst);
547 /** Function to tell the CPU that an instruction has completed. */
548 void instDone(ThreadID tid, DynInstPtr &inst);
550 /** Remove an instruction from the front end of the list. There's
551 * no restriction on location of the instruction.
553 void removeFrontInst(DynInstPtr &inst);
555 /** Remove all instructions that are not currently in the ROB.
556 * There's also an option to not squash delay slot instructions.*/
557 void removeInstsNotInROB(ThreadID tid);
559 /** Remove all instructions younger than the given sequence number. */
560 void removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid);
562 /** Removes the instruction pointed to by the iterator. */
563 inline void squashInstIt(const ListIt &instIt, ThreadID tid);
565 /** Cleans up all instructions on the remove list. */
566 void cleanUpRemovedInsts();
568 /** Debug function to print all instructions on the list. */
573 /** Count of total number of dynamic instructions in flight. */
577 /** List of all the instructions in flight. */
578 std::list<DynInstPtr> instList;
580 /** List of all the instructions that will be removed at the end of this
583 std::queue<ListIt> removeList;
586 /** Debug structure to keep track of the sequence numbers still in
589 std::set<InstSeqNum> snList;
592 /** Records if instructions need to be removed this cycle due to
593 * being retired or squashed.
595 bool removeInstsThisCycle;
598 /** The fetch stage. */
599 typename CPUPolicy::Fetch fetch;
601 /** The decode stage. */
602 typename CPUPolicy::Decode decode;
604 /** The dispatch stage. */
605 typename CPUPolicy::Rename rename;
607 /** The issue/execute/writeback stages. */
608 typename CPUPolicy::IEW iew;
610 /** The commit stage. */
611 typename CPUPolicy::Commit commit;
613 /** The register file. */
614 typename CPUPolicy::RegFile regFile;
616 /** The free list. */
617 typename CPUPolicy::FreeList freeList;
619 /** The rename map. */
620 typename CPUPolicy::RenameMap renameMap[Impl::MaxThreads];
622 /** The commit rename map. */
623 typename CPUPolicy::RenameMap commitRenameMap[Impl::MaxThreads];
625 /** The re-order buffer. */
626 typename CPUPolicy::ROB rob;
628 /** Active Threads List */
629 std::list<ThreadID> activeThreads;
631 /** Integer Register Scoreboard */
632 Scoreboard scoreboard;
634 TheISA::ISA isa[Impl::MaxThreads];
636 /** Instruction port. Note that it has to appear after the fetch stage. */
637 IcachePort icachePort;
639 /** Data port. Note that it has to appear after the iew stages */
640 DcachePort dcachePort;
643 /** Enum to give each stage a specific index, so when calling
644 * activateStage() or deactivateStage(), they can specify which stage
645 * is being activated/deactivated.
655 /** Typedefs from the Impl to get the structs that each of the
656 * time buffers should use.
658 typedef typename CPUPolicy::TimeStruct TimeStruct;
660 typedef typename CPUPolicy::FetchStruct FetchStruct;
662 typedef typename CPUPolicy::DecodeStruct DecodeStruct;
664 typedef typename CPUPolicy::RenameStruct RenameStruct;
666 typedef typename CPUPolicy::IEWStruct IEWStruct;
668 /** The main time buffer to do backwards communication. */
669 TimeBuffer<TimeStruct> timeBuffer;
671 /** The fetch stage's instruction queue. */
672 TimeBuffer<FetchStruct> fetchQueue;
674 /** The decode stage's instruction queue. */
675 TimeBuffer<DecodeStruct> decodeQueue;
677 /** The rename stage's instruction queue. */
678 TimeBuffer<RenameStruct> renameQueue;
680 /** The IEW stage's instruction queue. */
681 TimeBuffer<IEWStruct> iewQueue;
684 /** The activity recorder; used to tell if the CPU has any
685 * activity remaining or if it can go to idle and deschedule
688 ActivityRecorder activityRec;
691 /** Records that there was time buffer activity this cycle. */
692 void activityThisCycle() { activityRec.activity(); }
694 /** Changes a stage's status to active within the activity recorder. */
695 void activateStage(const StageIdx idx)
696 { activityRec.activateStage(idx); }
698 /** Changes a stage's status to inactive within the activity recorder. */
699 void deactivateStage(const StageIdx idx)
700 { activityRec.deactivateStage(idx); }
702 /** Wakes the CPU, rescheduling the CPU if it's not already active. */
705 virtual void wakeup();
707 /** Gets a free thread id. Use if thread ids change across system. */
708 ThreadID getFreeTid();
711 /** Returns a pointer to a thread context. */
715 return thread[tid]->getTC();
718 /** The global sequence number counter. */
719 InstSeqNum globalSeqNum;//[Impl::MaxThreads];
721 /** Pointer to the checker, which can dynamically verify
722 * instruction results at run time. This can be set to NULL if it
725 Checker<Impl> *checker;
727 /** Pointer to the system. */
730 /** Event to call process() on once draining has completed. */
733 /** Counter of how many stages have completed draining. */
736 /** Pointers to all of the threads in the CPU. */
737 std::vector<Thread *> thread;
739 /** Whether or not the CPU should defer its registration. */
740 bool deferRegistration;
742 /** Is there a context switch pending? */
745 /** Threads Scheduled to Enter CPU */
746 std::list<int> cpuWaitList;
748 /** The cycle that the CPU was last running, used for statistics. */
749 Tick lastRunningCycle;
751 /** The cycle that the CPU was last activated by a new thread*/
752 Tick lastActivatedCycle;
754 /** Mapping for system thread id to cpu id */
755 std::map<ThreadID, unsigned> threadMap;
757 /** Available thread ids in the cpu*/
758 std::vector<ThreadID> tids;
760 /** CPU read function, forwards read to LSQ. */
761 Fault read(RequestPtr &req, RequestPtr &sreqLow, RequestPtr &sreqHigh,
762 uint8_t *data, int load_idx)
764 return this->iew.ldstQueue.read(req, sreqLow, sreqHigh,
768 /** CPU write function, forwards write to LSQ. */
769 Fault write(RequestPtr &req, RequestPtr &sreqLow, RequestPtr &sreqHigh,
770 uint8_t *data, int store_idx)
772 return this->iew.ldstQueue.write(req, sreqLow, sreqHigh,
776 /** Used by the fetch unit to get a hold of the instruction port. */
777 virtual CpuPort &getInstPort() { return icachePort; }
779 /** Get the dcache port (used to find block size for translations). */
780 virtual CpuPort &getDataPort() { return dcachePort; }
784 /** Temporary fix for the lock flag, works in the UP case. */
787 /** Stat for total number of times the CPU is descheduled. */
788 Stats::Scalar timesIdled;
789 /** Stat for total number of cycles the CPU spends descheduled. */
790 Stats::Scalar idleCycles;
791 /** Stat for total number of cycles the CPU spends descheduled due to a
792 * quiesce operation or waiting for an interrupt. */
793 Stats::Scalar quiesceCycles;
794 /** Stat for the number of committed instructions per thread. */
795 Stats::Vector committedInsts;
796 /** Stat for the number of committed ops (including micro ops) per thread. */
797 Stats::Vector committedOps;
798 /** Stat for the total number of committed instructions. */
799 Stats::Scalar totalCommittedInsts;
800 /** Stat for the CPI per thread. */
802 /** Stat for the total CPI. */
803 Stats::Formula totalCpi;
804 /** Stat for the IPC per thread. */
806 /** Stat for the total IPC. */
807 Stats::Formula totalIpc;
809 //number of integer register file accesses
810 Stats::Scalar intRegfileReads;
811 Stats::Scalar intRegfileWrites;
812 //number of float register file accesses
813 Stats::Scalar fpRegfileReads;
814 Stats::Scalar fpRegfileWrites;
816 Stats::Scalar miscRegfileReads;
817 Stats::Scalar miscRegfileWrites;
820 #endif // __CPU_O3_CPU_HH__