Merge zizzer:/bk/newmem
[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/physical.hh"
46 #include "sim/builder.hh"
47 #include "sim/eventq.hh"
48 #include "sim/host.hh"
49
50 using namespace std;
51 using namespace TheISA;
52
53 PhysicalMemory::PhysicalMemory(Params *p)
54 : MemObject(p->name), pmemAddr(NULL), port(NULL), lat(p->latency), _params(p)
55 {
56 if (params()->addrRange.size() % TheISA::PageBytes != 0)
57 panic("Memory Size not divisible by page size\n");
58
59 int map_flags = MAP_ANON | MAP_PRIVATE;
60 pmemAddr = (uint8_t *)mmap(NULL, params()->addrRange.size(), PROT_READ | PROT_WRITE,
61 map_flags, -1, 0);
62
63 if (pmemAddr == (void *)MAP_FAILED) {
64 perror("mmap");
65 fatal("Could not mmap!\n");
66 }
67
68 //If requested, initialize all the memory to 0
69 if(params()->zero)
70 memset(pmemAddr, 0, params()->addrRange.size());
71
72 pagePtr = 0;
73 }
74
75 void
76 PhysicalMemory::init()
77 {
78 if (!port)
79 panic("PhysicalMemory not connected to anything!");
80 port->sendStatusChange(Port::RangeChange);
81 }
82
83 PhysicalMemory::~PhysicalMemory()
84 {
85 if (pmemAddr)
86 munmap(pmemAddr, params()->addrRange.size());
87 //Remove memPorts?
88 }
89
90 Addr
91 PhysicalMemory::new_page()
92 {
93 Addr return_addr = pagePtr << LogVMPageSize;
94 return_addr += params()->addrRange.start;
95
96 ++pagePtr;
97 return return_addr;
98 }
99
100 int
101 PhysicalMemory::deviceBlockSize()
102 {
103 //Can accept anysize request
104 return 0;
105 }
106
107 Tick
108 PhysicalMemory::calculateLatency(PacketPtr pkt)
109 {
110 return lat;
111 }
112
113
114
115 // Add load-locked to tracking list. Should only be called if the
116 // operation is a load and the LOCKED flag is set.
117 void
118 PhysicalMemory::trackLoadLocked(Request *req)
119 {
120 Addr paddr = LockedAddr::mask(req->getPaddr());
121
122 // first we check if we already have a locked addr for this
123 // xc. Since each xc only gets one, we just update the
124 // existing record with the new address.
125 list<LockedAddr>::iterator i;
126
127 for (i = lockedAddrList.begin(); i != lockedAddrList.end(); ++i) {
128 if (i->matchesContext(req)) {
129 DPRINTF(LLSC, "Modifying lock record: cpu %d thread %d addr %#x\n",
130 req->getCpuNum(), req->getThreadNum(), paddr);
131 i->addr = paddr;
132 return;
133 }
134 }
135
136 // no record for this xc: need to allocate a new one
137 DPRINTF(LLSC, "Adding lock record: cpu %d thread %d addr %#x\n",
138 req->getCpuNum(), req->getThreadNum(), paddr);
139 lockedAddrList.push_front(LockedAddr(req));
140 }
141
142
143 // Called on *writes* only... both regular stores and
144 // store-conditional operations. Check for conventional stores which
145 // conflict with locked addresses, and for success/failure of store
146 // conditionals.
147 bool
148 PhysicalMemory::checkLockedAddrList(Request *req)
149 {
150 Addr paddr = LockedAddr::mask(req->getPaddr());
151 bool isLocked = req->isLocked();
152
153 // Initialize return value. Non-conditional stores always
154 // succeed. Assume conditional stores will fail until proven
155 // otherwise.
156 bool success = !isLocked;
157
158 // Iterate over list. Note that there could be multiple matching
159 // records, as more than one context could have done a load locked
160 // to this location.
161 list<LockedAddr>::iterator i = lockedAddrList.begin();
162
163 while (i != lockedAddrList.end()) {
164
165 if (i->addr == paddr) {
166 // we have a matching address
167
168 if (isLocked && i->matchesContext(req)) {
169 // it's a store conditional, and as far as the memory
170 // system can tell, the requesting context's lock is
171 // still valid.
172 DPRINTF(LLSC, "StCond success: cpu %d thread %d addr %#x\n",
173 req->getCpuNum(), req->getThreadNum(), paddr);
174 success = true;
175 }
176
177 // Get rid of our record of this lock and advance to next
178 DPRINTF(LLSC, "Erasing lock record: cpu %d thread %d addr %#x\n",
179 i->cpuNum, i->threadNum, paddr);
180 i = lockedAddrList.erase(i);
181 }
182 else {
183 // no match: advance to next record
184 ++i;
185 }
186 }
187
188 if (isLocked) {
189 req->setScResult(success ? 1 : 0);
190 }
191
192 return success;
193 }
194
195 void
196 PhysicalMemory::doFunctionalAccess(PacketPtr pkt)
197 {
198 assert(pkt->getAddr() >= params()->addrRange.start &&
199 pkt->getAddr() + pkt->getSize() <= params()->addrRange.start +
200 params()->addrRange.size());
201
202 if (pkt->isRead()) {
203 if (pkt->req->isLocked()) {
204 trackLoadLocked(pkt->req);
205 }
206 DPRINTF(MemoryAccess, "Performing Read of size %i on address 0x%x\n",
207 pkt->getSize(), pkt->getAddr());
208 memcpy(pkt->getPtr<uint8_t>(),
209 pmemAddr + pkt->getAddr() - params()->addrRange.start,
210 pkt->getSize());
211 }
212 else if (pkt->isWrite()) {
213 if (writeOK(pkt->req)) {
214 DPRINTF(MemoryAccess, "Performing Write of size %i on address 0x%x\n",
215 pkt->getSize(), pkt->getAddr());
216 memcpy(pmemAddr + pkt->getAddr() - params()->addrRange.start,
217 pkt->getPtr<uint8_t>(), pkt->getSize());
218 }
219 }
220 else if (pkt->isInvalidate()) {
221 //upgrade or invalidate
222 pkt->flags |= SATISFIED;
223 }
224 else {
225 panic("unimplemented");
226 }
227
228 pkt->result = Packet::Success;
229 }
230
231 Port *
232 PhysicalMemory::getPort(const std::string &if_name, int idx)
233 {
234 if (if_name == "port" && idx == -1) {
235 if (port != NULL)
236 panic("PhysicalMemory::getPort: additional port requested to memory!");
237 port = new MemoryPort(name() + "-port", this);
238 return port;
239 } else if (if_name == "functional") {
240 /* special port for functional writes at startup. And for memtester */
241 return new MemoryPort(name() + "-funcport", this);
242 } else {
243 panic("PhysicalMemory::getPort: unknown port %s requested", if_name);
244 }
245 }
246
247 void
248 PhysicalMemory::recvStatusChange(Port::Status status)
249 {
250 }
251
252 PhysicalMemory::MemoryPort::MemoryPort(const std::string &_name,
253 PhysicalMemory *_memory)
254 : SimpleTimingPort(_name), memory(_memory)
255 { }
256
257 void
258 PhysicalMemory::MemoryPort::recvStatusChange(Port::Status status)
259 {
260 memory->recvStatusChange(status);
261 }
262
263 void
264 PhysicalMemory::MemoryPort::getDeviceAddressRanges(AddrRangeList &resp,
265 AddrRangeList &snoop)
266 {
267 memory->getAddressRanges(resp, snoop);
268 }
269
270 void
271 PhysicalMemory::getAddressRanges(AddrRangeList &resp, AddrRangeList &snoop)
272 {
273 snoop.clear();
274 resp.clear();
275 resp.push_back(RangeSize(params()->addrRange.start,
276 params()->addrRange.size()));
277 }
278
279 int
280 PhysicalMemory::MemoryPort::deviceBlockSize()
281 {
282 return memory->deviceBlockSize();
283 }
284
285 Tick
286 PhysicalMemory::MemoryPort::recvAtomic(PacketPtr pkt)
287 {
288 memory->doFunctionalAccess(pkt);
289 return memory->calculateLatency(pkt);
290 }
291
292 void
293 PhysicalMemory::MemoryPort::recvFunctional(PacketPtr pkt)
294 {
295 //Since we are overriding the function, make sure to have the impl of the
296 //check or functional accesses here.
297 std::list<std::pair<Tick,PacketPtr> >::iterator i = transmitList.begin();
298 std::list<std::pair<Tick,PacketPtr> >::iterator end = transmitList.end();
299 bool notDone = true;
300
301 while (i != end && notDone) {
302 PacketPtr target = i->second;
303 // If the target contains data, and it overlaps the
304 // probed request, need to update data
305 if (target->intersect(pkt))
306 notDone = fixPacket(pkt, target);
307 i++;
308 }
309
310 // Default implementation of SimpleTimingPort::recvFunctional()
311 // calls recvAtomic() and throws away the latency; we can save a
312 // little here by just not calculating the latency.
313 memory->doFunctionalAccess(pkt);
314 }
315
316 unsigned int
317 PhysicalMemory::drain(Event *de)
318 {
319 int count = port->drain(de);
320 if (count)
321 changeState(Draining);
322 else
323 changeState(Drained);
324 return count;
325 }
326
327 void
328 PhysicalMemory::serialize(ostream &os)
329 {
330 gzFile compressedMem;
331 string filename = name() + ".physmem";
332
333 SERIALIZE_SCALAR(filename);
334
335 // write memory file
336 string thefile = Checkpoint::dir() + "/" + filename.c_str();
337 int fd = creat(thefile.c_str(), 0664);
338 if (fd < 0) {
339 perror("creat");
340 fatal("Can't open physical memory checkpoint file '%s'\n", filename);
341 }
342
343 compressedMem = gzdopen(fd, "wb");
344 if (compressedMem == NULL)
345 fatal("Insufficient memory to allocate compression state for %s\n",
346 filename);
347
348 if (gzwrite(compressedMem, pmemAddr, params()->addrRange.size()) != params()->addrRange.size()) {
349 fatal("Write failed on physical memory checkpoint file '%s'\n",
350 filename);
351 }
352
353 if (gzclose(compressedMem))
354 fatal("Close failed on physical memory checkpoint file '%s'\n",
355 filename);
356 }
357
358 void
359 PhysicalMemory::unserialize(Checkpoint *cp, const string &section)
360 {
361 gzFile compressedMem;
362 long *tempPage;
363 long *pmem_current;
364 uint64_t curSize;
365 uint32_t bytesRead;
366 const int chunkSize = 16384;
367
368
369 string filename;
370
371 UNSERIALIZE_SCALAR(filename);
372
373 filename = cp->cptDir + "/" + filename;
374
375 // mmap memoryfile
376 int fd = open(filename.c_str(), O_RDONLY);
377 if (fd < 0) {
378 perror("open");
379 fatal("Can't open physical memory checkpoint file '%s'", filename);
380 }
381
382 compressedMem = gzdopen(fd, "rb");
383 if (compressedMem == NULL)
384 fatal("Insufficient memory to allocate compression state for %s\n",
385 filename);
386
387 // unmap file that was mmaped in the constructor
388 // This is done here to make sure that gzip and open don't muck with our
389 // nice large space of memory before we reallocate it
390 munmap(pmemAddr, params()->addrRange.size());
391
392 pmemAddr = (uint8_t *)mmap(NULL, params()->addrRange.size(), PROT_READ | PROT_WRITE,
393 MAP_ANON | MAP_PRIVATE, -1, 0);
394
395 if (pmemAddr == (void *)MAP_FAILED) {
396 perror("mmap");
397 fatal("Could not mmap physical memory!\n");
398 }
399
400 curSize = 0;
401 tempPage = (long*)malloc(chunkSize);
402 if (tempPage == NULL)
403 fatal("Unable to malloc memory to read file %s\n", filename);
404
405 /* Only copy bytes that are non-zero, so we don't give the VM system hell */
406 while (curSize < params()->addrRange.size()) {
407 bytesRead = gzread(compressedMem, tempPage, chunkSize);
408 if (bytesRead != chunkSize && bytesRead != params()->addrRange.size() - curSize)
409 fatal("Read failed on physical memory checkpoint file '%s'"
410 " got %d bytes, expected %d or %d bytes\n",
411 filename, bytesRead, chunkSize, params()->addrRange.size()-curSize);
412
413 assert(bytesRead % sizeof(long) == 0);
414
415 for (int x = 0; x < bytesRead/sizeof(long); x++)
416 {
417 if (*(tempPage+x) != 0) {
418 pmem_current = (long*)(pmemAddr + curSize + x * sizeof(long));
419 *pmem_current = *(tempPage+x);
420 }
421 }
422 curSize += bytesRead;
423 }
424
425 free(tempPage);
426
427 if (gzclose(compressedMem))
428 fatal("Close failed on physical memory checkpoint file '%s'\n",
429 filename);
430
431 }
432
433
434 BEGIN_DECLARE_SIM_OBJECT_PARAMS(PhysicalMemory)
435
436 Param<string> file;
437 Param<Range<Addr> > range;
438 Param<Tick> latency;
439 Param<bool> zero;
440
441 END_DECLARE_SIM_OBJECT_PARAMS(PhysicalMemory)
442
443 BEGIN_INIT_SIM_OBJECT_PARAMS(PhysicalMemory)
444
445 INIT_PARAM_DFLT(file, "memory mapped file", ""),
446 INIT_PARAM(range, "Device Address Range"),
447 INIT_PARAM(latency, "Memory access latency"),
448 INIT_PARAM(zero, "Zero initialize memory")
449
450 END_INIT_SIM_OBJECT_PARAMS(PhysicalMemory)
451
452 CREATE_SIM_OBJECT(PhysicalMemory)
453 {
454 PhysicalMemory::Params *p = new PhysicalMemory::Params;
455 p->name = getInstanceName();
456 p->addrRange = range;
457 p->latency = latency;
458 p->zero = zero;
459 return new PhysicalMemory(p);
460 }
461
462 REGISTER_SIM_OBJECT("PhysicalMemory", PhysicalMemory)