Merge vm1.(none):/home/stever/bk/newmem-head
[gem5.git] / src / mem / physical.cc
1 /*
2 * Copyright (c) 2001-2005 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: Ron Dreslinski
29 * Ali Saidi
30 */
31
32 #include <sys/types.h>
33 #include <sys/mman.h>
34 #include <errno.h>
35 #include <fcntl.h>
36 #include <unistd.h>
37 #include <zlib.h>
38
39 #include <iostream>
40 #include <string>
41
42 #include "arch/isa_traits.hh"
43 #include "base/misc.hh"
44 #include "config/full_system.hh"
45 #include "mem/packet_access.hh"
46 #include "mem/physical.hh"
47 #include "sim/builder.hh"
48 #include "sim/eventq.hh"
49 #include "sim/host.hh"
50
51 using namespace std;
52 using namespace TheISA;
53
54 PhysicalMemory::PhysicalMemory(Params *p)
55 : MemObject(p->name), pmemAddr(NULL), lat(p->latency), _params(p)
56 {
57 if (params()->addrRange.size() % TheISA::PageBytes != 0)
58 panic("Memory Size not divisible by page size\n");
59
60 int map_flags = MAP_ANON | MAP_PRIVATE;
61 pmemAddr =
62 (uint8_t *)mmap(NULL, params()->addrRange.size(),
63 PROT_READ | PROT_WRITE, map_flags, -1, 0);
64
65 if (pmemAddr == (void *)MAP_FAILED) {
66 perror("mmap");
67 fatal("Could not mmap!\n");
68 }
69
70 //If requested, initialize all the memory to 0
71 if(params()->zero)
72 memset(pmemAddr, 0, params()->addrRange.size());
73
74 pagePtr = 0;
75 }
76
77 void
78 PhysicalMemory::init()
79 {
80 if (ports.size() == 0) {
81 fatal("PhysicalMemory object %s is unconnected!", name());
82 }
83
84 for (PortIterator pi = ports.begin(); pi != ports.end(); ++pi) {
85 if (*pi)
86 (*pi)->sendStatusChange(Port::RangeChange);
87 }
88 }
89
90 PhysicalMemory::~PhysicalMemory()
91 {
92 if (pmemAddr)
93 munmap((char*)pmemAddr, params()->addrRange.size());
94 //Remove memPorts?
95 }
96
97 Addr
98 PhysicalMemory::new_page()
99 {
100 Addr return_addr = pagePtr << LogVMPageSize;
101 return_addr += start();
102
103 ++pagePtr;
104 return return_addr;
105 }
106
107 int
108 PhysicalMemory::deviceBlockSize()
109 {
110 //Can accept anysize request
111 return 0;
112 }
113
114 Tick
115 PhysicalMemory::calculateLatency(PacketPtr pkt)
116 {
117 return lat;
118 }
119
120
121
122 // Add load-locked to tracking list. Should only be called if the
123 // operation is a load and the LOCKED flag is set.
124 void
125 PhysicalMemory::trackLoadLocked(PacketPtr pkt)
126 {
127 Request *req = pkt->req;
128 Addr paddr = LockedAddr::mask(req->getPaddr());
129
130 // first we check if we already have a locked addr for this
131 // xc. Since each xc only gets one, we just update the
132 // existing record with the new address.
133 list<LockedAddr>::iterator i;
134
135 for (i = lockedAddrList.begin(); i != lockedAddrList.end(); ++i) {
136 if (i->matchesContext(req)) {
137 DPRINTF(LLSC, "Modifying lock record: cpu %d thread %d addr %#x\n",
138 req->getCpuNum(), req->getThreadNum(), paddr);
139 i->addr = paddr;
140 return;
141 }
142 }
143
144 // no record for this xc: need to allocate a new one
145 DPRINTF(LLSC, "Adding lock record: cpu %d thread %d addr %#x\n",
146 req->getCpuNum(), req->getThreadNum(), paddr);
147 lockedAddrList.push_front(LockedAddr(req));
148 }
149
150
151 // Called on *writes* only... both regular stores and
152 // store-conditional operations. Check for conventional stores which
153 // conflict with locked addresses, and for success/failure of store
154 // conditionals.
155 bool
156 PhysicalMemory::checkLockedAddrList(PacketPtr pkt)
157 {
158 Request *req = pkt->req;
159 Addr paddr = LockedAddr::mask(req->getPaddr());
160 bool isLocked = pkt->isLocked();
161
162 // Initialize return value. Non-conditional stores always
163 // succeed. Assume conditional stores will fail until proven
164 // otherwise.
165 bool success = !isLocked;
166
167 // Iterate over list. Note that there could be multiple matching
168 // records, as more than one context could have done a load locked
169 // to this location.
170 list<LockedAddr>::iterator i = lockedAddrList.begin();
171
172 while (i != lockedAddrList.end()) {
173
174 if (i->addr == paddr) {
175 // we have a matching address
176
177 if (isLocked && i->matchesContext(req)) {
178 // it's a store conditional, and as far as the memory
179 // system can tell, the requesting context's lock is
180 // still valid.
181 DPRINTF(LLSC, "StCond success: cpu %d thread %d addr %#x\n",
182 req->getCpuNum(), req->getThreadNum(), paddr);
183 success = true;
184 }
185
186 // Get rid of our record of this lock and advance to next
187 DPRINTF(LLSC, "Erasing lock record: cpu %d thread %d addr %#x\n",
188 i->cpuNum, i->threadNum, paddr);
189 i = lockedAddrList.erase(i);
190 }
191 else {
192 // no match: advance to next record
193 ++i;
194 }
195 }
196
197 if (isLocked) {
198 req->setExtraData(success ? 1 : 0);
199 }
200
201 return success;
202 }
203
204
205 #if TRACING_ON
206
207 #define CASE(A, T) \
208 case sizeof(T): \
209 DPRINTF(MemoryAccess, A " of size %i on address 0x%x data 0x%x\n", \
210 pkt->getSize(), pkt->getAddr(), pkt->get<T>()); \
211 break
212
213
214 #define TRACE_PACKET(A) \
215 do { \
216 switch (pkt->getSize()) { \
217 CASE(A, uint64_t); \
218 CASE(A, uint32_t); \
219 CASE(A, uint16_t); \
220 CASE(A, uint8_t); \
221 default: \
222 DPRINTF(MemoryAccess, A " of size %i on address 0x%x\n", \
223 pkt->getSize(), pkt->getAddr()); \
224 } \
225 } while (0)
226
227 #else
228
229 #define TRACE_PACKET(A)
230
231 #endif
232
233 Tick
234 PhysicalMemory::doAtomicAccess(PacketPtr pkt)
235 {
236 assert(pkt->getAddr() >= start() &&
237 pkt->getAddr() + pkt->getSize() <= start() + size());
238
239 if (pkt->memInhibitAsserted()) {
240 DPRINTF(MemoryAccess, "mem inhibited on 0x%x: not responding\n",
241 pkt->getAddr());
242 return 0;
243 }
244
245 uint8_t *hostAddr = pmemAddr + pkt->getAddr() - start();
246
247 if (pkt->cmd == MemCmd::SwapReq) {
248 IntReg overwrite_val;
249 bool overwrite_mem;
250 uint64_t condition_val64;
251 uint32_t condition_val32;
252
253 assert(sizeof(IntReg) >= pkt->getSize());
254
255 overwrite_mem = true;
256 // keep a copy of our possible write value, and copy what is at the
257 // memory address into the packet
258 std::memcpy(&overwrite_val, pkt->getPtr<uint8_t>(), pkt->getSize());
259 std::memcpy(pkt->getPtr<uint8_t>(), hostAddr, pkt->getSize());
260
261 if (pkt->req->isCondSwap()) {
262 if (pkt->getSize() == sizeof(uint64_t)) {
263 condition_val64 = pkt->req->getExtraData();
264 overwrite_mem = !std::memcmp(&condition_val64, hostAddr,
265 sizeof(uint64_t));
266 } else if (pkt->getSize() == sizeof(uint32_t)) {
267 condition_val32 = (uint32_t)pkt->req->getExtraData();
268 overwrite_mem = !std::memcmp(&condition_val32, hostAddr,
269 sizeof(uint32_t));
270 } else
271 panic("Invalid size for conditional read/write\n");
272 }
273
274 if (overwrite_mem)
275 std::memcpy(hostAddr, &overwrite_val, pkt->getSize());
276
277 TRACE_PACKET("Read/Write");
278 } else if (pkt->isRead()) {
279 assert(!pkt->isWrite());
280 if (pkt->isLocked()) {
281 trackLoadLocked(pkt);
282 }
283 memcpy(pkt->getPtr<uint8_t>(), hostAddr, pkt->getSize());
284 TRACE_PACKET("Read");
285 } else if (pkt->isWrite()) {
286 if (writeOK(pkt)) {
287 memcpy(hostAddr, pkt->getPtr<uint8_t>(), pkt->getSize());
288 TRACE_PACKET("Write");
289 }
290 } else if (pkt->isInvalidate()) {
291 //upgrade or invalidate
292 if (pkt->needsResponse()) {
293 pkt->makeAtomicResponse();
294 }
295 } else {
296 panic("unimplemented");
297 }
298
299 if (pkt->needsResponse()) {
300 pkt->makeAtomicResponse();
301 }
302 return calculateLatency(pkt);
303 }
304
305
306 void
307 PhysicalMemory::doFunctionalAccess(PacketPtr pkt)
308 {
309 assert(pkt->getAddr() >= start() &&
310 pkt->getAddr() + pkt->getSize() <= start() + size());
311
312 uint8_t *hostAddr = pmemAddr + pkt->getAddr() - start();
313
314 if (pkt->cmd == MemCmd::ReadReq) {
315 memcpy(pkt->getPtr<uint8_t>(), hostAddr, pkt->getSize());
316 TRACE_PACKET("Read");
317 } else if (pkt->cmd == MemCmd::WriteReq) {
318 memcpy(hostAddr, pkt->getPtr<uint8_t>(), pkt->getSize());
319 TRACE_PACKET("Write");
320 } else {
321 panic("PhysicalMemory: unimplemented functional command %s",
322 pkt->cmdString());
323 }
324
325 pkt->result = Packet::Success;
326 }
327
328
329 Port *
330 PhysicalMemory::getPort(const std::string &if_name, int idx)
331 {
332 // Accept request for "functional" port for backwards compatibility
333 // with places where this function is called from C++. I'd prefer
334 // to move all these into Python someday.
335 if (if_name == "functional") {
336 return new MemoryPort(csprintf("%s-functional", name()), this);
337 }
338
339 if (if_name != "port") {
340 panic("PhysicalMemory::getPort: unknown port %s requested", if_name);
341 }
342
343 if (idx >= ports.size()) {
344 ports.resize(idx+1);
345 }
346
347 if (ports[idx] != NULL) {
348 panic("PhysicalMemory::getPort: port %d already assigned", idx);
349 }
350
351 MemoryPort *port =
352 new MemoryPort(csprintf("%s-port%d", name(), idx), this);
353
354 ports[idx] = port;
355 return port;
356 }
357
358
359 void
360 PhysicalMemory::recvStatusChange(Port::Status status)
361 {
362 }
363
364 PhysicalMemory::MemoryPort::MemoryPort(const std::string &_name,
365 PhysicalMemory *_memory)
366 : SimpleTimingPort(_name), memory(_memory)
367 { }
368
369 void
370 PhysicalMemory::MemoryPort::recvStatusChange(Port::Status status)
371 {
372 memory->recvStatusChange(status);
373 }
374
375 void
376 PhysicalMemory::MemoryPort::getDeviceAddressRanges(AddrRangeList &resp,
377 bool &snoop)
378 {
379 memory->getAddressRanges(resp, snoop);
380 }
381
382 void
383 PhysicalMemory::getAddressRanges(AddrRangeList &resp, bool &snoop)
384 {
385 snoop = false;
386 resp.clear();
387 resp.push_back(RangeSize(start(), params()->addrRange.size()));
388 }
389
390 int
391 PhysicalMemory::MemoryPort::deviceBlockSize()
392 {
393 return memory->deviceBlockSize();
394 }
395
396 Tick
397 PhysicalMemory::MemoryPort::recvAtomic(PacketPtr pkt)
398 {
399 return memory->doAtomicAccess(pkt);
400 }
401
402 void
403 PhysicalMemory::MemoryPort::recvFunctional(PacketPtr pkt)
404 {
405 checkFunctional(pkt);
406
407 // Default implementation of SimpleTimingPort::recvFunctional()
408 // calls recvAtomic() and throws away the latency; we can save a
409 // little here by just not calculating the latency.
410 memory->doFunctionalAccess(pkt);
411 }
412
413 unsigned int
414 PhysicalMemory::drain(Event *de)
415 {
416 int count = 0;
417 for (PortIterator pi = ports.begin(); pi != ports.end(); ++pi) {
418 count += (*pi)->drain(de);
419 }
420
421 if (count)
422 changeState(Draining);
423 else
424 changeState(Drained);
425 return count;
426 }
427
428 void
429 PhysicalMemory::serialize(ostream &os)
430 {
431 gzFile compressedMem;
432 string filename = name() + ".physmem";
433
434 SERIALIZE_SCALAR(filename);
435
436 // write memory file
437 string thefile = Checkpoint::dir() + "/" + filename.c_str();
438 int fd = creat(thefile.c_str(), 0664);
439 if (fd < 0) {
440 perror("creat");
441 fatal("Can't open physical memory checkpoint file '%s'\n", filename);
442 }
443
444 compressedMem = gzdopen(fd, "wb");
445 if (compressedMem == NULL)
446 fatal("Insufficient memory to allocate compression state for %s\n",
447 filename);
448
449 if (gzwrite(compressedMem, pmemAddr, params()->addrRange.size()) != params()->addrRange.size()) {
450 fatal("Write failed on physical memory checkpoint file '%s'\n",
451 filename);
452 }
453
454 if (gzclose(compressedMem))
455 fatal("Close failed on physical memory checkpoint file '%s'\n",
456 filename);
457 }
458
459 void
460 PhysicalMemory::unserialize(Checkpoint *cp, const string &section)
461 {
462 gzFile compressedMem;
463 long *tempPage;
464 long *pmem_current;
465 uint64_t curSize;
466 uint32_t bytesRead;
467 const int chunkSize = 16384;
468
469
470 string filename;
471
472 UNSERIALIZE_SCALAR(filename);
473
474 filename = cp->cptDir + "/" + filename;
475
476 // mmap memoryfile
477 int fd = open(filename.c_str(), O_RDONLY);
478 if (fd < 0) {
479 perror("open");
480 fatal("Can't open physical memory checkpoint file '%s'", filename);
481 }
482
483 compressedMem = gzdopen(fd, "rb");
484 if (compressedMem == NULL)
485 fatal("Insufficient memory to allocate compression state for %s\n",
486 filename);
487
488 // unmap file that was mmaped in the constructor
489 // This is done here to make sure that gzip and open don't muck with our
490 // nice large space of memory before we reallocate it
491 munmap((char*)pmemAddr, params()->addrRange.size());
492
493 pmemAddr = (uint8_t *)mmap(NULL, params()->addrRange.size(), PROT_READ | PROT_WRITE,
494 MAP_ANON | MAP_PRIVATE, -1, 0);
495
496 if (pmemAddr == (void *)MAP_FAILED) {
497 perror("mmap");
498 fatal("Could not mmap physical memory!\n");
499 }
500
501 curSize = 0;
502 tempPage = (long*)malloc(chunkSize);
503 if (tempPage == NULL)
504 fatal("Unable to malloc memory to read file %s\n", filename);
505
506 /* Only copy bytes that are non-zero, so we don't give the VM system hell */
507 while (curSize < params()->addrRange.size()) {
508 bytesRead = gzread(compressedMem, tempPage, chunkSize);
509 if (bytesRead != chunkSize && bytesRead != params()->addrRange.size() - curSize)
510 fatal("Read failed on physical memory checkpoint file '%s'"
511 " got %d bytes, expected %d or %d bytes\n",
512 filename, bytesRead, chunkSize, params()->addrRange.size()-curSize);
513
514 assert(bytesRead % sizeof(long) == 0);
515
516 for (int x = 0; x < bytesRead/sizeof(long); x++)
517 {
518 if (*(tempPage+x) != 0) {
519 pmem_current = (long*)(pmemAddr + curSize + x * sizeof(long));
520 *pmem_current = *(tempPage+x);
521 }
522 }
523 curSize += bytesRead;
524 }
525
526 free(tempPage);
527
528 if (gzclose(compressedMem))
529 fatal("Close failed on physical memory checkpoint file '%s'\n",
530 filename);
531
532 }
533
534
535 BEGIN_DECLARE_SIM_OBJECT_PARAMS(PhysicalMemory)
536
537 Param<string> file;
538 Param<Range<Addr> > range;
539 Param<Tick> latency;
540 Param<bool> zero;
541
542 END_DECLARE_SIM_OBJECT_PARAMS(PhysicalMemory)
543
544 BEGIN_INIT_SIM_OBJECT_PARAMS(PhysicalMemory)
545
546 INIT_PARAM_DFLT(file, "memory mapped file", ""),
547 INIT_PARAM(range, "Device Address Range"),
548 INIT_PARAM(latency, "Memory access latency"),
549 INIT_PARAM(zero, "Zero initialize memory")
550
551 END_INIT_SIM_OBJECT_PARAMS(PhysicalMemory)
552
553 CREATE_SIM_OBJECT(PhysicalMemory)
554 {
555 PhysicalMemory::Params *p = new PhysicalMemory::Params;
556 p->name = getInstanceName();
557 p->addrRange = range;
558 p->latency = latency;
559 p->zero = zero;
560 return new PhysicalMemory(p);
561 }
562
563 REGISTER_SIM_OBJECT("PhysicalMemory", PhysicalMemory)