timing.hh (12085:de78ea63e0ca) timing.hh (12749:223c83ed9979)
1/*
2 * Copyright (c) 2012-2013,2015 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) 2002-2005 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: Steve Reinhardt
41 */
42
43#ifndef __CPU_SIMPLE_TIMING_HH__
44#define __CPU_SIMPLE_TIMING_HH__
45
46#include "cpu/simple/base.hh"
47#include "cpu/simple/exec_context.hh"
48#include "cpu/translation.hh"
49#include "params/TimingSimpleCPU.hh"
50
51class TimingSimpleCPU : public BaseSimpleCPU
52{
53 public:
54
55 TimingSimpleCPU(TimingSimpleCPUParams * params);
56 virtual ~TimingSimpleCPU();
57
58 void init() override;
59
60 private:
61
62 /*
63 * If an access needs to be broken into fragments, currently at most two,
64 * the the following two classes are used as the sender state of the
65 * packets so the CPU can keep track of everything. In the main packet
66 * sender state, there's an array with a spot for each fragment. If a
67 * fragment has already been accepted by the CPU, aka isn't waiting for
68 * a retry, it's pointer is NULL. After each fragment has successfully
69 * been processed, the "outstanding" counter is decremented. Once the
70 * count is zero, the entire larger access is complete.
71 */
72 class SplitMainSenderState : public Packet::SenderState
73 {
74 public:
75 int outstanding;
76 PacketPtr fragments[2];
77
78 int
79 getPendingFragment()
80 {
81 if (fragments[0]) {
82 return 0;
83 } else if (fragments[1]) {
84 return 1;
85 } else {
86 return -1;
87 }
88 }
89 };
90
91 class SplitFragmentSenderState : public Packet::SenderState
92 {
93 public:
94 SplitFragmentSenderState(PacketPtr _bigPkt, int _index) :
95 bigPkt(_bigPkt), index(_index)
96 {}
97 PacketPtr bigPkt;
98 int index;
99
100 void
101 clearFromParent()
102 {
103 SplitMainSenderState * main_send_state =
104 dynamic_cast<SplitMainSenderState *>(bigPkt->senderState);
105 main_send_state->fragments[index] = NULL;
106 }
107 };
108
109 class FetchTranslation : public BaseTLB::Translation
110 {
111 protected:
112 TimingSimpleCPU *cpu;
113
114 public:
115 FetchTranslation(TimingSimpleCPU *_cpu)
116 : cpu(_cpu)
117 {}
118
119 void
120 markDelayed()
121 {
122 assert(cpu->_status == BaseSimpleCPU::Running);
123 cpu->_status = ITBWaitResponse;
124 }
125
126 void
1/*
2 * Copyright (c) 2012-2013,2015 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) 2002-2005 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: Steve Reinhardt
41 */
42
43#ifndef __CPU_SIMPLE_TIMING_HH__
44#define __CPU_SIMPLE_TIMING_HH__
45
46#include "cpu/simple/base.hh"
47#include "cpu/simple/exec_context.hh"
48#include "cpu/translation.hh"
49#include "params/TimingSimpleCPU.hh"
50
51class TimingSimpleCPU : public BaseSimpleCPU
52{
53 public:
54
55 TimingSimpleCPU(TimingSimpleCPUParams * params);
56 virtual ~TimingSimpleCPU();
57
58 void init() override;
59
60 private:
61
62 /*
63 * If an access needs to be broken into fragments, currently at most two,
64 * the the following two classes are used as the sender state of the
65 * packets so the CPU can keep track of everything. In the main packet
66 * sender state, there's an array with a spot for each fragment. If a
67 * fragment has already been accepted by the CPU, aka isn't waiting for
68 * a retry, it's pointer is NULL. After each fragment has successfully
69 * been processed, the "outstanding" counter is decremented. Once the
70 * count is zero, the entire larger access is complete.
71 */
72 class SplitMainSenderState : public Packet::SenderState
73 {
74 public:
75 int outstanding;
76 PacketPtr fragments[2];
77
78 int
79 getPendingFragment()
80 {
81 if (fragments[0]) {
82 return 0;
83 } else if (fragments[1]) {
84 return 1;
85 } else {
86 return -1;
87 }
88 }
89 };
90
91 class SplitFragmentSenderState : public Packet::SenderState
92 {
93 public:
94 SplitFragmentSenderState(PacketPtr _bigPkt, int _index) :
95 bigPkt(_bigPkt), index(_index)
96 {}
97 PacketPtr bigPkt;
98 int index;
99
100 void
101 clearFromParent()
102 {
103 SplitMainSenderState * main_send_state =
104 dynamic_cast<SplitMainSenderState *>(bigPkt->senderState);
105 main_send_state->fragments[index] = NULL;
106 }
107 };
108
109 class FetchTranslation : public BaseTLB::Translation
110 {
111 protected:
112 TimingSimpleCPU *cpu;
113
114 public:
115 FetchTranslation(TimingSimpleCPU *_cpu)
116 : cpu(_cpu)
117 {}
118
119 void
120 markDelayed()
121 {
122 assert(cpu->_status == BaseSimpleCPU::Running);
123 cpu->_status = ITBWaitResponse;
124 }
125
126 void
127 finish(const Fault &fault, RequestPtr req, ThreadContext *tc,
127 finish(const Fault &fault, const RequestPtr &req, ThreadContext *tc,
128 BaseTLB::Mode mode)
129 {
130 cpu->sendFetch(fault, req, tc);
131 }
132 };
133 FetchTranslation fetchTranslation;
134
135 void threadSnoop(PacketPtr pkt, ThreadID sender);
128 BaseTLB::Mode mode)
129 {
130 cpu->sendFetch(fault, req, tc);
131 }
132 };
133 FetchTranslation fetchTranslation;
134
135 void threadSnoop(PacketPtr pkt, ThreadID sender);
136 void sendData(RequestPtr req, uint8_t *data, uint64_t *res, bool read);
137 void sendSplitData(RequestPtr req1, RequestPtr req2, RequestPtr req,
136 void sendData(const RequestPtr &req,
137 uint8_t *data, uint64_t *res, bool read);
138 void sendSplitData(const RequestPtr &req1, const RequestPtr &req2,
139 const RequestPtr &req,
138 uint8_t *data, bool read);
139
140 void translationFault(const Fault &fault);
141
140 uint8_t *data, bool read);
141
142 void translationFault(const Fault &fault);
143
142 PacketPtr buildPacket(RequestPtr req, bool read);
144 PacketPtr buildPacket(const RequestPtr &req, bool read);
143 void buildSplitPacket(PacketPtr &pkt1, PacketPtr &pkt2,
145 void buildSplitPacket(PacketPtr &pkt1, PacketPtr &pkt2,
144 RequestPtr req1, RequestPtr req2, RequestPtr req,
146 const RequestPtr &req1, const RequestPtr &req2,
147 const RequestPtr &req,
145 uint8_t *data, bool read);
146
147 bool handleReadPacket(PacketPtr pkt);
148 // This function always implicitly uses dcache_pkt.
149 bool handleWritePacket();
150
151 /**
152 * A TimingCPUPort overrides the default behaviour of the
153 * recvTiming and recvRetry and implements events for the
154 * scheduling of handling of incoming packets in the following
155 * cycle.
156 */
157 class TimingCPUPort : public MasterPort
158 {
159 public:
160
161 TimingCPUPort(const std::string& _name, TimingSimpleCPU* _cpu)
162 : MasterPort(_name, _cpu), cpu(_cpu),
163 retryRespEvent([this]{ sendRetryResp(); }, name())
164 { }
165
166 protected:
167
168 TimingSimpleCPU* cpu;
169
170 struct TickEvent : public Event
171 {
172 PacketPtr pkt;
173 TimingSimpleCPU *cpu;
174
175 TickEvent(TimingSimpleCPU *_cpu) : pkt(NULL), cpu(_cpu) {}
176 const char *description() const { return "Timing CPU tick"; }
177 void schedule(PacketPtr _pkt, Tick t);
178 };
179
180 EventFunctionWrapper retryRespEvent;
181 };
182
183 class IcachePort : public TimingCPUPort
184 {
185 public:
186
187 IcachePort(TimingSimpleCPU *_cpu)
188 : TimingCPUPort(_cpu->name() + ".icache_port", _cpu),
189 tickEvent(_cpu)
190 { }
191
192 protected:
193
194 virtual bool recvTimingResp(PacketPtr pkt);
195
196 virtual void recvReqRetry();
197
198 struct ITickEvent : public TickEvent
199 {
200
201 ITickEvent(TimingSimpleCPU *_cpu)
202 : TickEvent(_cpu) {}
203 void process();
204 const char *description() const { return "Timing CPU icache tick"; }
205 };
206
207 ITickEvent tickEvent;
208
209 };
210
211 class DcachePort : public TimingCPUPort
212 {
213 public:
214
215 DcachePort(TimingSimpleCPU *_cpu)
216 : TimingCPUPort(_cpu->name() + ".dcache_port", _cpu),
217 tickEvent(_cpu)
218 {
219 cacheBlockMask = ~(cpu->cacheLineSize() - 1);
220 }
221
222 Addr cacheBlockMask;
223 protected:
224
225 /** Snoop a coherence request, we need to check if this causes
226 * a wakeup event on a cpu that is monitoring an address
227 */
228 virtual void recvTimingSnoopReq(PacketPtr pkt);
229 virtual void recvFunctionalSnoop(PacketPtr pkt);
230
231 virtual bool recvTimingResp(PacketPtr pkt);
232
233 virtual void recvReqRetry();
234
235 virtual bool isSnooping() const {
236 return true;
237 }
238
239 struct DTickEvent : public TickEvent
240 {
241 DTickEvent(TimingSimpleCPU *_cpu)
242 : TickEvent(_cpu) {}
243 void process();
244 const char *description() const { return "Timing CPU dcache tick"; }
245 };
246
247 DTickEvent tickEvent;
248
249 };
250
251 void updateCycleCounts();
252
253 IcachePort icachePort;
254 DcachePort dcachePort;
255
256 PacketPtr ifetch_pkt;
257 PacketPtr dcache_pkt;
258
259 Cycles previousCycle;
260
261 protected:
262
263 /** Return a reference to the data port. */
264 MasterPort &getDataPort() override { return dcachePort; }
265
266 /** Return a reference to the instruction port. */
267 MasterPort &getInstPort() override { return icachePort; }
268
269 public:
270
271 DrainState drain() override;
272 void drainResume() override;
273
274 void switchOut() override;
275 void takeOverFrom(BaseCPU *oldCPU) override;
276
277 void verifyMemoryMode() const override;
278
279 void activateContext(ThreadID thread_num) override;
280 void suspendContext(ThreadID thread_num) override;
281
282 Fault readMem(Addr addr, uint8_t *data, unsigned size,
283 Request::Flags flags) override;
284
285 Fault initiateMemRead(Addr addr, unsigned size,
286 Request::Flags flags) override;
287
288 Fault writeMem(uint8_t *data, unsigned size,
289 Addr addr, Request::Flags flags, uint64_t *res) override;
290
291 void fetch();
148 uint8_t *data, bool read);
149
150 bool handleReadPacket(PacketPtr pkt);
151 // This function always implicitly uses dcache_pkt.
152 bool handleWritePacket();
153
154 /**
155 * A TimingCPUPort overrides the default behaviour of the
156 * recvTiming and recvRetry and implements events for the
157 * scheduling of handling of incoming packets in the following
158 * cycle.
159 */
160 class TimingCPUPort : public MasterPort
161 {
162 public:
163
164 TimingCPUPort(const std::string& _name, TimingSimpleCPU* _cpu)
165 : MasterPort(_name, _cpu), cpu(_cpu),
166 retryRespEvent([this]{ sendRetryResp(); }, name())
167 { }
168
169 protected:
170
171 TimingSimpleCPU* cpu;
172
173 struct TickEvent : public Event
174 {
175 PacketPtr pkt;
176 TimingSimpleCPU *cpu;
177
178 TickEvent(TimingSimpleCPU *_cpu) : pkt(NULL), cpu(_cpu) {}
179 const char *description() const { return "Timing CPU tick"; }
180 void schedule(PacketPtr _pkt, Tick t);
181 };
182
183 EventFunctionWrapper retryRespEvent;
184 };
185
186 class IcachePort : public TimingCPUPort
187 {
188 public:
189
190 IcachePort(TimingSimpleCPU *_cpu)
191 : TimingCPUPort(_cpu->name() + ".icache_port", _cpu),
192 tickEvent(_cpu)
193 { }
194
195 protected:
196
197 virtual bool recvTimingResp(PacketPtr pkt);
198
199 virtual void recvReqRetry();
200
201 struct ITickEvent : public TickEvent
202 {
203
204 ITickEvent(TimingSimpleCPU *_cpu)
205 : TickEvent(_cpu) {}
206 void process();
207 const char *description() const { return "Timing CPU icache tick"; }
208 };
209
210 ITickEvent tickEvent;
211
212 };
213
214 class DcachePort : public TimingCPUPort
215 {
216 public:
217
218 DcachePort(TimingSimpleCPU *_cpu)
219 : TimingCPUPort(_cpu->name() + ".dcache_port", _cpu),
220 tickEvent(_cpu)
221 {
222 cacheBlockMask = ~(cpu->cacheLineSize() - 1);
223 }
224
225 Addr cacheBlockMask;
226 protected:
227
228 /** Snoop a coherence request, we need to check if this causes
229 * a wakeup event on a cpu that is monitoring an address
230 */
231 virtual void recvTimingSnoopReq(PacketPtr pkt);
232 virtual void recvFunctionalSnoop(PacketPtr pkt);
233
234 virtual bool recvTimingResp(PacketPtr pkt);
235
236 virtual void recvReqRetry();
237
238 virtual bool isSnooping() const {
239 return true;
240 }
241
242 struct DTickEvent : public TickEvent
243 {
244 DTickEvent(TimingSimpleCPU *_cpu)
245 : TickEvent(_cpu) {}
246 void process();
247 const char *description() const { return "Timing CPU dcache tick"; }
248 };
249
250 DTickEvent tickEvent;
251
252 };
253
254 void updateCycleCounts();
255
256 IcachePort icachePort;
257 DcachePort dcachePort;
258
259 PacketPtr ifetch_pkt;
260 PacketPtr dcache_pkt;
261
262 Cycles previousCycle;
263
264 protected:
265
266 /** Return a reference to the data port. */
267 MasterPort &getDataPort() override { return dcachePort; }
268
269 /** Return a reference to the instruction port. */
270 MasterPort &getInstPort() override { return icachePort; }
271
272 public:
273
274 DrainState drain() override;
275 void drainResume() override;
276
277 void switchOut() override;
278 void takeOverFrom(BaseCPU *oldCPU) override;
279
280 void verifyMemoryMode() const override;
281
282 void activateContext(ThreadID thread_num) override;
283 void suspendContext(ThreadID thread_num) override;
284
285 Fault readMem(Addr addr, uint8_t *data, unsigned size,
286 Request::Flags flags) override;
287
288 Fault initiateMemRead(Addr addr, unsigned size,
289 Request::Flags flags) override;
290
291 Fault writeMem(uint8_t *data, unsigned size,
292 Addr addr, Request::Flags flags, uint64_t *res) override;
293
294 void fetch();
292 void sendFetch(const Fault &fault, RequestPtr req, ThreadContext *tc);
295 void sendFetch(const Fault &fault,
296 const RequestPtr &req, ThreadContext *tc);
293 void completeIfetch(PacketPtr );
294 void completeDataAccess(PacketPtr pkt);
295 void advanceInst(const Fault &fault);
296
297 /** This function is used by the page table walker to determine if it could
298 * translate the a pending request or if the underlying request has been
299 * squashed. This always returns false for the simple timing CPU as it never
300 * executes any instructions speculatively.
301 * @ return Is the current instruction squashed?
302 */
303 bool isSquashed() const { return false; }
304
305 /**
306 * Print state of address in memory system via PrintReq (for
307 * debugging).
308 */
309 void printAddr(Addr a);
310
311 /**
312 * Finish a DTB translation.
313 * @param state The DTB translation state.
314 */
315 void finishTranslation(WholeTranslationState *state);
316
317 private:
318
319 EventFunctionWrapper fetchEvent;
320
321 struct IprEvent : Event {
322 Packet *pkt;
323 TimingSimpleCPU *cpu;
324 IprEvent(Packet *_pkt, TimingSimpleCPU *_cpu, Tick t);
325 virtual void process();
326 virtual const char *description() const;
327 };
328
329 /**
330 * Check if a system is in a drained state.
331 *
332 * We need to drain if:
333 * <ul>
334 * <li>We are in the middle of a microcode sequence as some CPUs
335 * (e.g., HW accelerated CPUs) can't be started in the middle
336 * of a gem5 microcode sequence.
337 *
338 * <li>Stay at PC is true.
339 *
340 * <li>A fetch event is scheduled. Normally this would never be the
341 * case with microPC() == 0, but right after a context is
342 * activated it can happen.
343 * </ul>
344 */
345 bool isDrained() {
346 SimpleExecContext& t_info = *threadInfo[curThread];
347 SimpleThread* thread = t_info.thread;
348
349 return thread->microPC() == 0 && !t_info.stayAtPC &&
350 !fetchEvent.scheduled();
351 }
352
353 /**
354 * Try to complete a drain request.
355 *
356 * @returns true if the CPU is drained, false otherwise.
357 */
358 bool tryCompleteDrain();
359};
360
361#endif // __CPU_SIMPLE_TIMING_HH__
297 void completeIfetch(PacketPtr );
298 void completeDataAccess(PacketPtr pkt);
299 void advanceInst(const Fault &fault);
300
301 /** This function is used by the page table walker to determine if it could
302 * translate the a pending request or if the underlying request has been
303 * squashed. This always returns false for the simple timing CPU as it never
304 * executes any instructions speculatively.
305 * @ return Is the current instruction squashed?
306 */
307 bool isSquashed() const { return false; }
308
309 /**
310 * Print state of address in memory system via PrintReq (for
311 * debugging).
312 */
313 void printAddr(Addr a);
314
315 /**
316 * Finish a DTB translation.
317 * @param state The DTB translation state.
318 */
319 void finishTranslation(WholeTranslationState *state);
320
321 private:
322
323 EventFunctionWrapper fetchEvent;
324
325 struct IprEvent : Event {
326 Packet *pkt;
327 TimingSimpleCPU *cpu;
328 IprEvent(Packet *_pkt, TimingSimpleCPU *_cpu, Tick t);
329 virtual void process();
330 virtual const char *description() const;
331 };
332
333 /**
334 * Check if a system is in a drained state.
335 *
336 * We need to drain if:
337 * <ul>
338 * <li>We are in the middle of a microcode sequence as some CPUs
339 * (e.g., HW accelerated CPUs) can't be started in the middle
340 * of a gem5 microcode sequence.
341 *
342 * <li>Stay at PC is true.
343 *
344 * <li>A fetch event is scheduled. Normally this would never be the
345 * case with microPC() == 0, but right after a context is
346 * activated it can happen.
347 * </ul>
348 */
349 bool isDrained() {
350 SimpleExecContext& t_info = *threadInfo[curThread];
351 SimpleThread* thread = t_info.thread;
352
353 return thread->microPC() == 0 && !t_info.stayAtPC &&
354 !fetchEvent.scheduled();
355 }
356
357 /**
358 * Try to complete a drain request.
359 *
360 * @returns true if the CPU is drained, false otherwise.
361 */
362 bool tryCompleteDrain();
363};
364
365#endif // __CPU_SIMPLE_TIMING_HH__