Merge zizzer.eecs.umich.edu:/bk/newmem
[gem5.git] / src / cpu / memtest / memtest.cc
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 * Authors: Erik Hallnor
29 * Steve Reinhardt
30 */
31
32 // FIX ME: make trackBlkAddr use blocksize from actual cache, not hard coded
33
34 #include <iomanip>
35 #include <set>
36 #include <string>
37 #include <vector>
38
39 #include "base/misc.hh"
40 #include "base/statistics.hh"
41 //#include "cpu/simple_thread.hh"
42 #include "cpu/memtest/memtest.hh"
43 //#include "mem/cache/base_cache.hh"
44 //#include "mem/physical.hh"
45 #include "sim/builder.hh"
46 #include "sim/sim_events.hh"
47 #include "sim/stats.hh"
48 #include "mem/packet.hh"
49 #include "mem/request.hh"
50 #include "mem/port.hh"
51 #include "mem/mem_object.hh"
52
53 using namespace std;
54
55 int TESTER_ALLOCATOR=0;
56
57 bool
58 MemTest::CpuPort::recvTiming(Packet *pkt)
59 {
60 memtest->completeRequest(pkt);
61 return true;
62 }
63
64 Tick
65 MemTest::CpuPort::recvAtomic(Packet *pkt)
66 {
67 panic("MemTest doesn't expect recvAtomic callback!");
68 return curTick;
69 }
70
71 void
72 MemTest::CpuPort::recvFunctional(Packet *pkt)
73 {
74 //Do nothing if we see one come through
75 if (curTick != 0)//Supress warning durring initialization
76 warn("Functional Writes not implemented in MemTester\n");
77 //Need to find any response values that intersect and update
78 return;
79 }
80
81 void
82 MemTest::CpuPort::recvStatusChange(Status status)
83 {
84 if (status == RangeChange)
85 return;
86
87 panic("MemTest doesn't expect recvStatusChange callback!");
88 }
89
90 void
91 MemTest::CpuPort::recvRetry()
92 {
93 memtest->doRetry();
94 }
95
96 void
97 MemTest::sendPkt(Packet *pkt) {
98 if (atomic) {
99 cachePort.sendAtomic(pkt);
100 pkt->makeAtomicResponse();
101 completeRequest(pkt);
102 }
103 else if (!cachePort.sendTiming(pkt)) {
104 accessRetry = true;
105 retryPkt = pkt;
106 }
107
108 }
109
110 MemTest::MemTest(const string &name,
111 // MemInterface *_cache_interface,
112 // PhysicalMemory *main_mem,
113 // PhysicalMemory *check_mem,
114 unsigned _memorySize,
115 unsigned _percentReads,
116 // unsigned _percentCopies,
117 unsigned _percentUncacheable,
118 unsigned _progressInterval,
119 unsigned _percentSourceUnaligned,
120 unsigned _percentDestUnaligned,
121 Addr _traceAddr,
122 Counter _max_loads,
123 bool _atomic)
124 : MemObject(name),
125 tickEvent(this),
126 cachePort("test", this),
127 funcPort("functional", this),
128 retryPkt(NULL),
129 // mainMem(main_mem),
130 // checkMem(check_mem),
131 size(_memorySize),
132 percentReads(_percentReads),
133 // percentCopies(_percentCopies),
134 percentUncacheable(_percentUncacheable),
135 progressInterval(_progressInterval),
136 nextProgressMessage(_progressInterval),
137 percentSourceUnaligned(_percentSourceUnaligned),
138 percentDestUnaligned(percentDestUnaligned),
139 maxLoads(_max_loads),
140 atomic(_atomic)
141 {
142 vector<string> cmd;
143 cmd.push_back("/bin/ls");
144 vector<string> null_vec;
145 // thread = new SimpleThread(NULL, 0, NULL, 0, mainMem);
146 curTick = 0;
147
148 // Needs to be masked off once we know the block size.
149 traceBlockAddr = _traceAddr;
150 baseAddr1 = 0x100000;
151 baseAddr2 = 0x400000;
152 uncacheAddr = 0x800000;
153
154 // set up counters
155 noResponseCycles = 0;
156 numReads = 0;
157 tickEvent.schedule(0);
158
159 id = TESTER_ALLOCATOR++;
160 if (TESTER_ALLOCATOR > 8)
161 panic("False sharing memtester only allows up to 8 testers");
162
163 accessRetry = false;
164 }
165
166 Port *
167 MemTest::getPort(const std::string &if_name, int idx)
168 {
169 if (if_name == "functional")
170 return &funcPort;
171 else if (if_name == "test")
172 return &cachePort;
173 else
174 panic("No Such Port\n");
175 }
176
177 void
178 MemTest::init()
179 {
180 // By the time init() is called, the ports should be hooked up.
181 blockSize = cachePort.peerBlockSize();
182 blockAddrMask = blockSize - 1;
183 traceBlockAddr = blockAddr(traceBlockAddr);
184
185 // set up intial memory contents here
186
187 cachePort.memsetBlob(baseAddr1, 1, size);
188 funcPort.memsetBlob(baseAddr1, 1, size);
189 cachePort.memsetBlob(baseAddr2, 2, size);
190 funcPort.memsetBlob(baseAddr2, 2, size);
191 cachePort.memsetBlob(uncacheAddr, 3, size);
192 funcPort.memsetBlob(uncacheAddr, 3, size);
193 }
194
195 static void
196 printData(ostream &os, uint8_t *data, int nbytes)
197 {
198 os << hex << setfill('0');
199 // assume little-endian: print bytes from highest address to lowest
200 for (uint8_t *dp = data + nbytes - 1; dp >= data; --dp) {
201 os << setw(2) << (unsigned)*dp;
202 }
203 os << dec;
204 }
205
206 void
207 MemTest::completeRequest(Packet *pkt)
208 {
209 MemTestSenderState *state =
210 dynamic_cast<MemTestSenderState *>(pkt->senderState);
211
212 uint8_t *data = state->data;
213 uint8_t *pkt_data = pkt->getPtr<uint8_t>();
214 Request *req = pkt->req;
215
216 //Remove the address from the list of outstanding
217 std::set<unsigned>::iterator removeAddr = outstandingAddrs.find(req->getPaddr());
218 assert(removeAddr != outstandingAddrs.end());
219 outstandingAddrs.erase(removeAddr);
220
221 switch (pkt->cmd) {
222 case Packet::ReadResp:
223
224 if (memcmp(pkt_data, data, pkt->getSize()) != 0) {
225 cerr << name() << ": on read of 0x" << hex << req->getPaddr()
226 << " (0x" << hex << blockAddr(req->getPaddr()) << ")"
227 << "@ cycle " << dec << curTick
228 << ", cache returns 0x";
229 printData(cerr, pkt_data, pkt->getSize());
230 cerr << ", expected 0x";
231 printData(cerr, data, pkt->getSize());
232 cerr << endl;
233 fatal("");
234 }
235
236 numReads++;
237 numReadsStat++;
238
239 if (numReads == nextProgressMessage) {
240 ccprintf(cerr, "%s: completed %d read accesses @%d\n",
241 name(), numReads, curTick);
242 nextProgressMessage += progressInterval;
243 }
244
245 if (numReads >= maxLoads)
246 exitSimLoop("Maximum number of loads reached!");
247 break;
248
249 case Packet::WriteResp:
250 numWritesStat++;
251 break;
252 /*
253 case Copy:
254 //Also remove dest from outstanding list
255 removeAddr = outstandingAddrs.find(req->dest);
256 assert(removeAddr != outstandingAddrs.end());
257 outstandingAddrs.erase(removeAddr);
258 numCopiesStat++;
259 break;
260 */
261 default:
262 panic("invalid command");
263 }
264
265 if (blockAddr(req->getPaddr()) == traceBlockAddr) {
266 cerr << name() << ": completed "
267 << (pkt->isWrite() ? "write" : "read")
268 << " access of "
269 << dec << pkt->getSize() << " bytes at address 0x"
270 << hex << req->getPaddr()
271 << " (0x" << hex << blockAddr(req->getPaddr()) << ")"
272 << ", value = 0x";
273 printData(cerr, pkt_data, pkt->getSize());
274 cerr << " @ cycle " << dec << curTick;
275
276 cerr << endl;
277 }
278
279 noResponseCycles = 0;
280 delete state;
281 delete [] data;
282 delete pkt->req;
283 delete pkt;
284 }
285
286 void
287 MemTest::regStats()
288 {
289 using namespace Stats;
290
291 numReadsStat
292 .name(name() + ".num_reads")
293 .desc("number of read accesses completed")
294 ;
295
296 numWritesStat
297 .name(name() + ".num_writes")
298 .desc("number of write accesses completed")
299 ;
300
301 numCopiesStat
302 .name(name() + ".num_copies")
303 .desc("number of copy accesses completed")
304 ;
305 }
306
307 void
308 MemTest::tick()
309 {
310 if (!tickEvent.scheduled())
311 tickEvent.schedule(curTick + cycles(1));
312
313 if (++noResponseCycles >= 500000) {
314 cerr << name() << ": deadlocked at cycle " << curTick << endl;
315 fatal("");
316 }
317
318 if (accessRetry) {
319 return;
320 }
321
322 //make new request
323 unsigned cmd = random() % 100;
324 unsigned offset = random() % size;
325 unsigned base = random() % 2;
326 uint64_t data = random();
327 unsigned access_size = random() % 4;
328 unsigned cacheable = random() % 100;
329
330 //If we aren't doing copies, use id as offset, and do a false sharing
331 //mem tester
332 //We can eliminate the lower bits of the offset, and then use the id
333 //to offset within the blks
334 offset &= ~63; //Not the low order bits
335 offset += id;
336 access_size = 0;
337
338 Request *req = new Request();
339 uint32_t flags = 0;
340 Addr paddr;
341
342 if (cacheable < percentUncacheable) {
343 flags |= UNCACHEABLE;
344 paddr = uncacheAddr + offset;
345 } else {
346 paddr = ((base) ? baseAddr1 : baseAddr2) + offset;
347 }
348 //bool probe = (random() % 2 == 1) && !req->isUncacheable();
349 bool probe = false;
350
351 paddr &= ~((1 << access_size) - 1);
352 req->setPhys(paddr, 1 << access_size, flags);
353 req->setThreadContext(id,0);
354
355 uint8_t *result = new uint8_t[8];
356
357 if (cmd < percentReads) {
358 // read
359
360 //For now we only allow one outstanding request per addreess per tester
361 //This means we assume CPU does write forwarding to reads that alias something
362 //in the cpu store buffer.
363 if (outstandingAddrs.find(paddr) != outstandingAddrs.end()) {
364 delete result;
365 delete req;
366 return;
367 }
368 else outstandingAddrs.insert(paddr);
369
370 // ***** NOTE FOR RON: I'm not sure how to access checkMem. - Kevin
371 funcPort.readBlob(req->getPaddr(), result, req->getSize());
372
373 if (blockAddr(paddr) == traceBlockAddr) {
374 cerr << name()
375 << ": initiating read "
376 << ((probe) ? "probe of " : "access of ")
377 << dec << req->getSize() << " bytes from addr 0x"
378 << hex << paddr
379 << " (0x" << hex << blockAddr(paddr) << ")"
380 << " at cycle "
381 << dec << curTick << endl;
382 }
383
384 Packet *pkt = new Packet(req, Packet::ReadReq, Packet::Broadcast);
385 pkt->dataDynamicArray(new uint8_t[req->getSize()]);
386 MemTestSenderState *state = new MemTestSenderState(result);
387 pkt->senderState = state;
388
389 if (probe) {
390 cachePort.sendFunctional(pkt);
391 completeRequest(pkt);
392 } else {
393 // req->completionEvent = new MemCompleteEvent(req, result, this);
394 sendPkt(pkt);
395 }
396 } else {
397 // write
398
399 //For now we only allow one outstanding request per addreess per tester
400 //This means we assume CPU does write forwarding to reads that alias something
401 //in the cpu store buffer.
402 if (outstandingAddrs.find(paddr) != outstandingAddrs.end()) {
403 delete [] result;
404 delete req;
405 return;
406 }
407
408 else outstandingAddrs.insert(paddr);
409
410 /*
411 if (blockAddr(req->getPaddr()) == traceBlockAddr) {
412 cerr << name() << ": initiating write "
413 << ((probe)?"probe of ":"access of ")
414 << dec << req->getSize() << " bytes (value = 0x";
415 printData(cerr, data_pkt->getPtr(), req->getSize());
416 cerr << ") to addr 0x"
417 << hex << req->getPaddr()
418 << " (0x" << hex << blockAddr(req->getPaddr()) << ")"
419 << " at cycle "
420 << dec << curTick << endl;
421 }
422 */
423 Packet *pkt = new Packet(req, Packet::WriteReq, Packet::Broadcast);
424 uint8_t *pkt_data = new uint8_t[req->getSize()];
425 pkt->dataDynamicArray(pkt_data);
426 memcpy(pkt_data, &data, req->getSize());
427 MemTestSenderState *state = new MemTestSenderState(result);
428 pkt->senderState = state;
429
430 funcPort.writeBlob(req->getPaddr(), pkt_data, req->getSize());
431
432 if (probe) {
433 cachePort.sendFunctional(pkt);
434 completeRequest(pkt);
435 } else {
436 // req->completionEvent = new MemCompleteEvent(req, NULL, this);
437 sendPkt(pkt);
438 }
439 }
440 /* else {
441 // copy
442 unsigned source_align = random() % 100;
443 unsigned dest_align = random() % 100;
444 unsigned offset2 = random() % size;
445
446 Addr source = ((base) ? baseAddr1 : baseAddr2) + offset;
447 Addr dest = ((base) ? baseAddr2 : baseAddr1) + offset2;
448 if (outstandingAddrs.find(source) != outstandingAddrs.end()) return;
449 else outstandingAddrs.insert(source);
450 if (outstandingAddrs.find(dest) != outstandingAddrs.end()) return;
451 else outstandingAddrs.insert(dest);
452
453 if (source_align >= percentSourceUnaligned) {
454 source = blockAddr(source);
455 }
456 if (dest_align >= percentDestUnaligned) {
457 dest = blockAddr(dest);
458 }
459 req->cmd = Copy;
460 req->flags &= ~UNCACHEABLE;
461 req->paddr = source;
462 req->dest = dest;
463 delete [] req->data;
464 req->data = new uint8_t[blockSize];
465 req->size = blockSize;
466 if (source == traceBlockAddr || dest == traceBlockAddr) {
467 cerr << name()
468 << ": initiating copy of "
469 << dec << req->size << " bytes from addr 0x"
470 << hex << source
471 << " (0x" << hex << blockAddr(source) << ")"
472 << " to addr 0x"
473 << hex << dest
474 << " (0x" << hex << blockAddr(dest) << ")"
475 << " at cycle "
476 << dec << curTick << endl;
477 }*
478 cacheInterface->access(req);
479 uint8_t result[blockSize];
480 checkMem->access(Read, source, &result, blockSize);
481 checkMem->access(Write, dest, &result, blockSize);
482 }
483 */
484 }
485
486 void
487 MemTest::doRetry()
488 {
489 if (cachePort.sendTiming(retryPkt)) {
490 accessRetry = false;
491 retryPkt = NULL;
492 }
493 }
494
495 BEGIN_DECLARE_SIM_OBJECT_PARAMS(MemTest)
496
497 // SimObjectParam<BaseCache *> cache;
498 // SimObjectParam<PhysicalMemory *> main_mem;
499 // SimObjectParam<PhysicalMemory *> check_mem;
500 Param<unsigned> memory_size;
501 Param<unsigned> percent_reads;
502 // Param<unsigned> percent_copies;
503 Param<unsigned> percent_uncacheable;
504 Param<unsigned> progress_interval;
505 Param<unsigned> percent_source_unaligned;
506 Param<unsigned> percent_dest_unaligned;
507 Param<Addr> trace_addr;
508 Param<Counter> max_loads;
509 Param<bool> atomic;
510
511 END_DECLARE_SIM_OBJECT_PARAMS(MemTest)
512
513
514 BEGIN_INIT_SIM_OBJECT_PARAMS(MemTest)
515
516 // INIT_PARAM(cache, "L1 cache"),
517 // INIT_PARAM(main_mem, "hierarchical memory"),
518 // INIT_PARAM(check_mem, "check memory"),
519 INIT_PARAM(memory_size, "memory size"),
520 INIT_PARAM(percent_reads, "target read percentage"),
521 // INIT_PARAM(percent_copies, "target copy percentage"),
522 INIT_PARAM(percent_uncacheable, "target uncacheable percentage"),
523 INIT_PARAM(progress_interval, "progress report interval (in accesses)"),
524 INIT_PARAM(percent_source_unaligned,
525 "percent of copy source address that are unaligned"),
526 INIT_PARAM(percent_dest_unaligned,
527 "percent of copy dest address that are unaligned"),
528 INIT_PARAM(trace_addr, "address to trace"),
529 INIT_PARAM(max_loads, "terminate when we have reached this load count"),
530 INIT_PARAM(atomic, "Is the tester testing atomic mode (or timing)")
531
532 END_INIT_SIM_OBJECT_PARAMS(MemTest)
533
534
535 CREATE_SIM_OBJECT(MemTest)
536 {
537 return new MemTest(getInstanceName(), /*cache->getInterface(),*/ /*main_mem,*/
538 /*check_mem,*/ memory_size, percent_reads, /*percent_copies,*/
539 percent_uncacheable, progress_interval,
540 percent_source_unaligned, percent_dest_unaligned,
541 trace_addr, max_loads, atomic);
542 }
543
544 REGISTER_SIM_OBJECT("MemTest", MemTest)