ee1932bec6217176e460d6cc409ede00c05b94fd
[gem5.git] / src / cpu / o3 / fetch.hh
1 /*
2 * Copyright (c) 2010-2012, 2014 ARM Limited
3 * All rights reserved
4 *
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.
13 *
14 * Copyright (c) 2004-2006 The Regents of The University of Michigan
15 * All rights reserved.
16 *
17 * Redistribution and use in source and binary forms, with or without
18 * modification, are permitted provided that the following conditions are
19 * met: redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer;
21 * redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution;
24 * neither the name of the copyright holders nor the names of its
25 * contributors may be used to endorse or promote products derived from
26 * this software without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
29 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
31 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
32 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
33 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
34 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
35 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
36 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
38 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 *
40 * Authors: Kevin Lim
41 * Korey Sewell
42 */
43
44 #ifndef __CPU_O3_FETCH_HH__
45 #define __CPU_O3_FETCH_HH__
46
47 #include "arch/decoder.hh"
48 #include "arch/utility.hh"
49 #include "base/statistics.hh"
50 #include "config/the_isa.hh"
51 #include "cpu/pc_event.hh"
52 #include "cpu/pred/bpred_unit.hh"
53 #include "cpu/timebuf.hh"
54 #include "cpu/translation.hh"
55 #include "enums/FetchPolicy.hh"
56 #include "mem/packet.hh"
57 #include "mem/port.hh"
58 #include "sim/eventq.hh"
59 #include "sim/probe/probe.hh"
60
61 struct DerivO3CPUParams;
62
63 /**
64 * DefaultFetch class handles both single threaded and SMT fetch. Its
65 * width is specified by the parameters; each cycle it tries to fetch
66 * that many instructions. It supports using a branch predictor to
67 * predict direction and targets.
68 * It supports the idling functionality of the CPU by indicating to
69 * the CPU when it is active and inactive.
70 */
71 template <class Impl>
72 class DefaultFetch
73 {
74 public:
75 /** Typedefs from Impl. */
76 typedef typename Impl::CPUPol CPUPol;
77 typedef typename Impl::DynInst DynInst;
78 typedef typename Impl::DynInstPtr DynInstPtr;
79 typedef typename Impl::O3CPU O3CPU;
80
81 /** Typedefs from the CPU policy. */
82 typedef typename CPUPol::FetchStruct FetchStruct;
83 typedef typename CPUPol::TimeStruct TimeStruct;
84
85 /** Typedefs from ISA. */
86 typedef TheISA::MachInst MachInst;
87
88 class FetchTranslation : public BaseTLB::Translation
89 {
90 protected:
91 DefaultFetch<Impl> *fetch;
92
93 public:
94 FetchTranslation(DefaultFetch<Impl> *_fetch)
95 : fetch(_fetch)
96 {}
97
98 void
99 markDelayed()
100 {}
101
102 void
103 finish(const Fault &fault, const RequestPtr &req, ThreadContext *tc,
104 BaseTLB::Mode mode)
105 {
106 assert(mode == BaseTLB::Execute);
107 fetch->finishTranslation(fault, req);
108 delete this;
109 }
110 };
111
112 private:
113 /* Event to delay delivery of a fetch translation result in case of
114 * a fault and the nop to carry the fault cannot be generated
115 * immediately */
116 class FinishTranslationEvent : public Event
117 {
118 private:
119 DefaultFetch<Impl> *fetch;
120 Fault fault;
121 RequestPtr req;
122
123 public:
124 FinishTranslationEvent(DefaultFetch<Impl> *_fetch)
125 : fetch(_fetch), req(nullptr)
126 {}
127
128 void setFault(Fault _fault)
129 {
130 fault = _fault;
131 }
132
133 void setReq(const RequestPtr &_req)
134 {
135 req = _req;
136 }
137
138 /** Process the delayed finish translation */
139 void process()
140 {
141 assert(fetch->numInst < fetch->fetchWidth);
142 fetch->finishTranslation(fault, req);
143 }
144
145 const char *description() const
146 {
147 return "FullO3CPU FetchFinishTranslation";
148 }
149 };
150
151 public:
152 /** Overall fetch status. Used to determine if the CPU can
153 * deschedule itsef due to a lack of activity.
154 */
155 enum FetchStatus {
156 Active,
157 Inactive
158 };
159
160 /** Individual thread status. */
161 enum ThreadStatus {
162 Running,
163 Idle,
164 Squashing,
165 Blocked,
166 Fetching,
167 TrapPending,
168 QuiescePending,
169 ItlbWait,
170 IcacheWaitResponse,
171 IcacheWaitRetry,
172 IcacheAccessComplete,
173 NoGoodAddr
174 };
175
176 private:
177 /** Fetch status. */
178 FetchStatus _status;
179
180 /** Per-thread status. */
181 ThreadStatus fetchStatus[Impl::MaxThreads];
182
183 /** Fetch policy. */
184 FetchPolicy fetchPolicy;
185
186 /** List that has the threads organized by priority. */
187 std::list<ThreadID> priorityList;
188
189 /** Probe points. */
190 ProbePointArg<DynInstPtr> *ppFetch;
191 /** To probe when a fetch request is successfully sent. */
192 ProbePointArg<RequestPtr> *ppFetchRequestSent;
193
194 public:
195 /** DefaultFetch constructor. */
196 DefaultFetch(O3CPU *_cpu, DerivO3CPUParams *params);
197
198 /** Returns the name of fetch. */
199 std::string name() const;
200
201 /** Registers statistics. */
202 void regStats();
203
204 /** Registers probes. */
205 void regProbePoints();
206
207 /** Sets the main backwards communication time buffer pointer. */
208 void setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer);
209
210 /** Sets pointer to list of active threads. */
211 void setActiveThreads(std::list<ThreadID> *at_ptr);
212
213 /** Sets pointer to time buffer used to communicate to the next stage. */
214 void setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr);
215
216 /** Initialize stage. */
217 void startupStage();
218
219 /** Handles retrying the fetch access. */
220 void recvReqRetry();
221
222 /** Processes cache completion event. */
223 void processCacheCompletion(PacketPtr pkt);
224
225 /** Resume after a drain. */
226 void drainResume();
227
228 /** Perform sanity checks after a drain. */
229 void drainSanityCheck() const;
230
231 /** Has the stage drained? */
232 bool isDrained() const;
233
234 /** Takes over from another CPU's thread. */
235 void takeOverFrom();
236
237 /**
238 * Stall the fetch stage after reaching a safe drain point.
239 *
240 * The CPU uses this method to stop fetching instructions from a
241 * thread that has been drained. The drain stall is different from
242 * all other stalls in that it is signaled instantly from the
243 * commit stage (without the normal communication delay) when it
244 * has reached a safe point to drain from.
245 */
246 void drainStall(ThreadID tid);
247
248 /** Tells fetch to wake up from a quiesce instruction. */
249 void wakeFromQuiesce();
250
251 /** For priority-based fetch policies, need to keep update priorityList */
252 void deactivateThread(ThreadID tid);
253 private:
254 /** Reset this pipeline stage */
255 void resetStage();
256
257 /** Changes the status of this stage to active, and indicates this
258 * to the CPU.
259 */
260 inline void switchToActive();
261
262 /** Changes the status of this stage to inactive, and indicates
263 * this to the CPU.
264 */
265 inline void switchToInactive();
266
267 /**
268 * Looks up in the branch predictor to see if the next PC should be
269 * either next PC+=MachInst or a branch target.
270 * @param next_PC Next PC variable passed in by reference. It is
271 * expected to be set to the current PC; it will be updated with what
272 * the next PC will be.
273 * @param next_NPC Used for ISAs which use delay slots.
274 * @return Whether or not a branch was predicted as taken.
275 */
276 bool lookupAndUpdateNextPC(const DynInstPtr &inst, TheISA::PCState &pc);
277
278 /**
279 * Fetches the cache line that contains the fetch PC. Returns any
280 * fault that happened. Puts the data into the class variable
281 * fetchBuffer, which may not hold the entire fetched cache line.
282 * @param vaddr The memory address that is being fetched from.
283 * @param ret_fault The fault reference that will be set to the result of
284 * the icache access.
285 * @param tid Thread id.
286 * @param pc The actual PC of the current instruction.
287 * @return Any fault that occured.
288 */
289 bool fetchCacheLine(Addr vaddr, ThreadID tid, Addr pc);
290 void finishTranslation(const Fault &fault, const RequestPtr &mem_req);
291
292
293 /** Check if an interrupt is pending and that we need to handle
294 */
295 bool
296 checkInterrupt(Addr pc)
297 {
298 return (interruptPending && (THE_ISA != ALPHA_ISA || !(pc & 0x3)));
299 }
300
301 /** Squashes a specific thread and resets the PC. */
302 inline void doSquash(const TheISA::PCState &newPC,
303 const DynInstPtr squashInst, ThreadID tid);
304
305 /** Squashes a specific thread and resets the PC. Also tells the CPU to
306 * remove any instructions between fetch and decode that should be sqaushed.
307 */
308 void squashFromDecode(const TheISA::PCState &newPC,
309 const DynInstPtr squashInst,
310 const InstSeqNum seq_num, ThreadID tid);
311
312 /** Checks if a thread is stalled. */
313 bool checkStall(ThreadID tid) const;
314
315 /** Updates overall fetch stage status; to be called at the end of each
316 * cycle. */
317 FetchStatus updateFetchStatus();
318
319 public:
320 /** Squashes a specific thread and resets the PC. Also tells the CPU to
321 * remove any instructions that are not in the ROB. The source of this
322 * squash should be the commit stage.
323 */
324 void squash(const TheISA::PCState &newPC, const InstSeqNum seq_num,
325 DynInstPtr squashInst, ThreadID tid);
326
327 /** Ticks the fetch stage, processing all inputs signals and fetching
328 * as many instructions as possible.
329 */
330 void tick();
331
332 /** Checks all input signals and updates the status as necessary.
333 * @return: Returns if the status has changed due to input signals.
334 */
335 bool checkSignalsAndUpdate(ThreadID tid);
336
337 /** Does the actual fetching of instructions and passing them on to the
338 * next stage.
339 * @param status_change fetch() sets this variable if there was a status
340 * change (ie switching to IcacheMissStall).
341 */
342 void fetch(bool &status_change);
343
344 /** Align a PC to the start of a fetch buffer block. */
345 Addr fetchBufferAlignPC(Addr addr)
346 {
347 return (addr & ~(fetchBufferMask));
348 }
349
350 /** The decoder. */
351 TheISA::Decoder *decoder[Impl::MaxThreads];
352
353 private:
354 DynInstPtr buildInst(ThreadID tid, StaticInstPtr staticInst,
355 StaticInstPtr curMacroop, TheISA::PCState thisPC,
356 TheISA::PCState nextPC, bool trace);
357
358 /** Returns the appropriate thread to fetch, given the fetch policy. */
359 ThreadID getFetchingThread();
360
361 /** Returns the appropriate thread to fetch using a round robin policy. */
362 ThreadID roundRobin();
363
364 /** Returns the appropriate thread to fetch using the IQ count policy. */
365 ThreadID iqCount();
366
367 /** Returns the appropriate thread to fetch using the LSQ count policy. */
368 ThreadID lsqCount();
369
370 /** Returns the appropriate thread to fetch using the branch count
371 * policy. */
372 ThreadID branchCount();
373
374 /** Pipeline the next I-cache access to the current one. */
375 void pipelineIcacheAccesses(ThreadID tid);
376
377 /** Profile the reasons of fetch stall. */
378 void profileStall(ThreadID tid);
379
380 private:
381 /** Pointer to the O3CPU. */
382 O3CPU *cpu;
383
384 /** Time buffer interface. */
385 TimeBuffer<TimeStruct> *timeBuffer;
386
387 /** Wire to get decode's information from backwards time buffer. */
388 typename TimeBuffer<TimeStruct>::wire fromDecode;
389
390 /** Wire to get rename's information from backwards time buffer. */
391 typename TimeBuffer<TimeStruct>::wire fromRename;
392
393 /** Wire to get iew's information from backwards time buffer. */
394 typename TimeBuffer<TimeStruct>::wire fromIEW;
395
396 /** Wire to get commit's information from backwards time buffer. */
397 typename TimeBuffer<TimeStruct>::wire fromCommit;
398
399 //Might be annoying how this name is different than the queue.
400 /** Wire used to write any information heading to decode. */
401 typename TimeBuffer<FetchStruct>::wire toDecode;
402
403 /** BPredUnit. */
404 BPredUnit *branchPred;
405
406 TheISA::PCState pc[Impl::MaxThreads];
407
408 Addr fetchOffset[Impl::MaxThreads];
409
410 StaticInstPtr macroop[Impl::MaxThreads];
411
412 /** Can the fetch stage redirect from an interrupt on this instruction? */
413 bool delayedCommit[Impl::MaxThreads];
414
415 /** Memory request used to access cache. */
416 RequestPtr memReq[Impl::MaxThreads];
417
418 /** Variable that tracks if fetch has written to the time buffer this
419 * cycle. Used to tell CPU if there is activity this cycle.
420 */
421 bool wroteToTimeBuffer;
422
423 /** Tracks how many instructions has been fetched this cycle. */
424 int numInst;
425
426 /** Source of possible stalls. */
427 struct Stalls {
428 bool decode;
429 bool drain;
430 };
431
432 /** Tracks which stages are telling fetch to stall. */
433 Stalls stalls[Impl::MaxThreads];
434
435 /** Decode to fetch delay. */
436 Cycles decodeToFetchDelay;
437
438 /** Rename to fetch delay. */
439 Cycles renameToFetchDelay;
440
441 /** IEW to fetch delay. */
442 Cycles iewToFetchDelay;
443
444 /** Commit to fetch delay. */
445 Cycles commitToFetchDelay;
446
447 /** The width of fetch in instructions. */
448 unsigned fetchWidth;
449
450 /** The width of decode in instructions. */
451 unsigned decodeWidth;
452
453 /** Is the cache blocked? If so no threads can access it. */
454 bool cacheBlocked;
455
456 /** The packet that is waiting to be retried. */
457 PacketPtr retryPkt;
458
459 /** The thread that is waiting on the cache to tell fetch to retry. */
460 ThreadID retryTid;
461
462 /** Cache block size. */
463 unsigned int cacheBlkSize;
464
465 /** The size of the fetch buffer in bytes. The fetch buffer
466 * itself may be smaller than a cache line.
467 */
468 unsigned fetchBufferSize;
469
470 /** Mask to align a fetch address to a fetch buffer boundary. */
471 Addr fetchBufferMask;
472
473 /** The fetch data that is being fetched and buffered. */
474 uint8_t *fetchBuffer[Impl::MaxThreads];
475
476 /** The PC of the first instruction loaded into the fetch buffer. */
477 Addr fetchBufferPC[Impl::MaxThreads];
478
479 /** The size of the fetch queue in micro-ops */
480 unsigned fetchQueueSize;
481
482 /** Queue of fetched instructions. Per-thread to prevent HoL blocking. */
483 std::deque<DynInstPtr> fetchQueue[Impl::MaxThreads];
484
485 /** Whether or not the fetch buffer data is valid. */
486 bool fetchBufferValid[Impl::MaxThreads];
487
488 /** Size of instructions. */
489 int instSize;
490
491 /** Icache stall statistics. */
492 Counter lastIcacheStall[Impl::MaxThreads];
493
494 /** List of Active Threads */
495 std::list<ThreadID> *activeThreads;
496
497 /** Number of threads. */
498 ThreadID numThreads;
499
500 /** Number of threads that are actively fetching. */
501 ThreadID numFetchingThreads;
502
503 /** Thread ID being fetched. */
504 ThreadID threadFetched;
505
506 /** Checks if there is an interrupt pending. If there is, fetch
507 * must stop once it is not fetching PAL instructions.
508 */
509 bool interruptPending;
510
511 /** Set to true if a pipelined I-cache request should be issued. */
512 bool issuePipelinedIfetch[Impl::MaxThreads];
513
514 /** Event used to delay fault generation of translation faults */
515 FinishTranslationEvent finishTranslationEvent;
516
517 // @todo: Consider making these vectors and tracking on a per thread basis.
518 /** Stat for total number of cycles stalled due to an icache miss. */
519 Stats::Scalar icacheStallCycles;
520 /** Stat for total number of fetched instructions. */
521 Stats::Scalar fetchedInsts;
522 /** Total number of fetched branches. */
523 Stats::Scalar fetchedBranches;
524 /** Stat for total number of predicted branches. */
525 Stats::Scalar predictedBranches;
526 /** Stat for total number of cycles spent fetching. */
527 Stats::Scalar fetchCycles;
528 /** Stat for total number of cycles spent squashing. */
529 Stats::Scalar fetchSquashCycles;
530 /** Stat for total number of cycles spent waiting for translation */
531 Stats::Scalar fetchTlbCycles;
532 /** Stat for total number of cycles spent blocked due to other stages in
533 * the pipeline.
534 */
535 Stats::Scalar fetchIdleCycles;
536 /** Total number of cycles spent blocked. */
537 Stats::Scalar fetchBlockedCycles;
538 /** Total number of cycles spent in any other state. */
539 Stats::Scalar fetchMiscStallCycles;
540 /** Total number of cycles spent in waiting for drains. */
541 Stats::Scalar fetchPendingDrainCycles;
542 /** Total number of stall cycles caused by no active threads to run. */
543 Stats::Scalar fetchNoActiveThreadStallCycles;
544 /** Total number of stall cycles caused by pending traps. */
545 Stats::Scalar fetchPendingTrapStallCycles;
546 /** Total number of stall cycles caused by pending quiesce instructions. */
547 Stats::Scalar fetchPendingQuiesceStallCycles;
548 /** Total number of stall cycles caused by I-cache wait retrys. */
549 Stats::Scalar fetchIcacheWaitRetryStallCycles;
550 /** Stat for total number of fetched cache lines. */
551 Stats::Scalar fetchedCacheLines;
552 /** Total number of outstanding icache accesses that were dropped
553 * due to a squash.
554 */
555 Stats::Scalar fetchIcacheSquashes;
556 /** Total number of outstanding tlb accesses that were dropped
557 * due to a squash.
558 */
559 Stats::Scalar fetchTlbSquashes;
560 /** Distribution of number of instructions fetched each cycle. */
561 Stats::Distribution fetchNisnDist;
562 /** Rate of how often fetch was idle. */
563 Stats::Formula idleRate;
564 /** Number of branch fetches per cycle. */
565 Stats::Formula branchRate;
566 /** Number of instruction fetched per cycle. */
567 Stats::Formula fetchRate;
568 };
569
570 #endif //__CPU_O3_FETCH_HH__