1 /* Copyright (c) 2007, 2009-2017. The SimGrid Team. All rights reserved. */
3 /* This program is free software; you can redistribute it and/or modify it
4 * under the terms of the license (GNU LGPL) which comes with this package. */
6 /* Shared allocations are handled through shared memory segments.
7 * Associated data and metadata are used as follows:
10 * `allocs' dict ---- -.
11 * ---------- shared_data_t shared_metadata_t / | | |
12 * .->| <name> | ---> -------------------- <--. ----------------- | | | |
13 * | ---------- | fd of <name> | | | size of mmap | --| | | |
14 * | | count (2) | |-- | data | \ | | |
15 * `----------------- | <name> | | ----------------- ---- |
16 * -------------------- | ^ |
18 * | | `allocs_metadata' dict |
19 * | | ---------------------- |
20 * | `-- | <addr of mmap #1> |<-'
21 * | .-- | <addr of mmap #2> |<-.
22 * | | ---------------------- |
28 * | shared_metadata_t / | |
29 * | ----------------- | | |
30 * | | size of mmap | --| | |
32 * ----------------- | | |
39 #include "private.hpp"
43 #include <sys/types.h>
53 #define MAP_ANONYMOUS MAP_ANON
57 #define MAP_POPULATE 0
60 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_shared, smpi, "Logging specific to SMPI (shared memory macros)");
62 #define PTR_STRLEN (2 + 2 * sizeof(void*) + 1)
65 /** Some location in the source code
67 * This information is used by SMPI_SHARED_MALLOC to allocate some shared memory for all simulated processes.
70 class smpi_source_location {
72 smpi_source_location(const char* filename, int line)
73 : filename(xbt_strdup(filename)), filename_length(strlen(filename)), line(line)
77 /** Pointer to a static string containing the file name */
78 char* filename = nullptr;
79 int filename_length = 0;
82 bool operator==(smpi_source_location const& that) const
84 return filename_length == that.filename_length && line == that.line &&
85 std::memcmp(filename, that.filename, filename_length) == 0;
87 bool operator!=(smpi_source_location const& that) const { return !(*this == that); }
93 template <> class hash<smpi_source_location> {
95 typedef smpi_source_location argument_type;
96 typedef std::size_t result_type;
97 result_type operator()(smpi_source_location const& loc) const
99 return xbt_str_hash_ext(loc.filename, loc.filename_length) ^
100 xbt_str_hash_ext((const char*)&loc.line, sizeof(loc.line));
112 std::unordered_map<smpi_source_location, shared_data_t> allocs;
113 typedef std::unordered_map<smpi_source_location, shared_data_t>::value_type shared_data_key_type;
117 std::vector<std::pair<int, int>> private_blocks;
118 shared_data_key_type* data;
121 std::map<void*, shared_metadata_t> allocs_metadata;
122 xbt_dict_t calls = nullptr; /* Allocated on first use */
124 static int smpi_shared_malloc_bogusfile = -1;
125 static unsigned long smpi_shared_malloc_blocksize = 1UL << 20;
130 void smpi_shared_destroy()
133 allocs_metadata.clear();
134 xbt_dict_free(&calls);
137 static size_t shm_size(int fd) {
140 if(fstat(fd, &st) < 0) {
141 xbt_die("Could not stat fd %d: %s", fd, strerror(errno));
143 return static_cast<size_t>(st.st_size);
147 static void* shm_map(int fd, size_t size, shared_data_key_type* data) {
148 char loc[PTR_STRLEN];
149 shared_metadata_t meta;
151 if(size > shm_size(fd) && (ftruncate(fd, static_cast<off_t>(size)) < 0)) {
152 xbt_die("Could not truncate fd %d to %zu: %s", fd, size, strerror(errno));
155 void* mem = mmap(nullptr, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
156 if(mem == MAP_FAILED) {
158 "Failed to map fd %d with size %zu: %s\n"
159 "If you are running a lot of ranks, you may be exceeding the amount of mappings allowed per process.\n"
160 "On Linux systems, change this value with sudo sysctl -w vm.max_map_count=newvalue (default value: 65536)\n"
161 "Please see http://simgrid.gforge.inria.fr/simgrid/latest/doc/html/options.html#options_virt for more info.",
162 fd, size, strerror(errno));
164 snprintf(loc, PTR_STRLEN, "%p", mem);
167 allocs_metadata[mem] = meta;
168 XBT_DEBUG("MMAP %zu to %p", size, mem);
172 void *smpi_shared_malloc_local(size_t size, const char *file, int line)
175 smpi_source_location loc(file, line);
176 auto res = allocs.insert(std::make_pair(loc, shared_data_t()));
177 auto data = res.first;
179 // The insertion did not take place.
180 // Generate a shared memory name from the address of the shared_data:
181 char shmname[32]; // cannot be longer than PSHMNAMLEN = 31 on Mac OS X (shm_open raises ENAMETOOLONG otherwise)
182 snprintf(shmname, 31, "/shmalloc%p", &*data);
183 int fd = shm_open(shmname, O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
186 xbt_die("Please cleanup /dev/shm/%s", shmname);
188 xbt_die("An unhandled error occurred while opening %s. shm_open: %s", shmname, strerror(errno));
190 data->second.fd = fd;
191 data->second.count = 1;
192 mem = shm_map(fd, size, &*data);
193 if (shm_unlink(shmname) < 0) {
194 XBT_WARN("Could not early unlink %s. shm_unlink: %s", shmname, strerror(errno));
196 XBT_DEBUG("Mapping %s at %p through %d", shmname, mem, fd);
198 mem = shm_map(data->second.fd, size, &*data);
199 data->second.count++;
201 XBT_DEBUG("Shared malloc %zu in %p (metadata at %p)", size, mem, &*data);
205 // Align functions, from http://stackoverflow.com/questions/4840410/how-to-align-a-pointer-in-c
206 #define PAGE_SIZE 0x1000
207 #define ALIGN_UP(n, align) (((n) + (align)-1) & -(align))
208 #define ALIGN_DOWN(n, align) ((n) & -(align))
210 void *smpi_shared_malloc_global__(size_t size, const char *file, int line, int *shared_block_offsets, int nb_shared_blocks) {
212 xbt_assert(smpi_shared_malloc_blocksize % PAGE_SIZE == 0, "The block size of shared malloc should be a multiple of the page size.");
213 /* First reserve memory area */
214 mem = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
216 xbt_assert(mem != MAP_FAILED, "Failed to allocate %luMiB of memory. Run \"sysctl vm.overcommit_memory=1\" as root "
217 "to allow big allocations.\n",
218 (unsigned long)(size >> 20));
220 /* Create bogus file if not done already */
221 if (smpi_shared_malloc_bogusfile == -1) {
222 /* Create a fd to a new file on disk, make it smpi_shared_malloc_blocksize big, and unlink it.
223 * It still exists in memory but not in the file system (thus it cannot be leaked). */
224 smpi_shared_malloc_blocksize = static_cast<unsigned long>(xbt_cfg_get_double("smpi/shared-malloc-blocksize"));
225 XBT_DEBUG("global shared allocation. Blocksize %lu", smpi_shared_malloc_blocksize);
226 char* name = xbt_strdup("/tmp/simgrid-shmalloc-XXXXXX");
227 smpi_shared_malloc_bogusfile = mkstemp(name);
230 char* dumb = (char*)calloc(1, smpi_shared_malloc_blocksize);
231 ssize_t err = write(smpi_shared_malloc_bogusfile, dumb, smpi_shared_malloc_blocksize);
233 xbt_die("Could not write bogus file for shared malloc");
237 /* Map the bogus file in place of the anonymous memory */
238 for(int i_block = 0; i_block < nb_shared_blocks; i_block ++) {
239 int start_offset = shared_block_offsets[2*i_block];
240 int stop_offset = shared_block_offsets[2*i_block+1];
241 xbt_assert(0 <= start_offset, "start_offset (%d) should be greater than 0", start_offset);
242 xbt_assert(start_offset < stop_offset, "start_offset (%d) should be lower than stop offset (%d)", start_offset, stop_offset);
243 xbt_assert(stop_offset <= size, "stop_offset (%d) should be lower than size (%lu)", stop_offset, size);
244 if(i_block < nb_shared_blocks-1)
245 xbt_assert(stop_offset < shared_block_offsets[2*i_block+2],
246 "stop_offset (%d) should be lower than its successor start offset (%d)", stop_offset, shared_block_offsets[2*i_block+2]);
247 // fprintf(stderr, "shared block 0x%x - 0x%x\n", start_offset, stop_offset);
248 int start_block_offset = ALIGN_UP(start_offset, smpi_shared_malloc_blocksize);
249 int stop_block_offset = ALIGN_DOWN(stop_offset, smpi_shared_malloc_blocksize);
251 for (i = start_block_offset / smpi_shared_malloc_blocksize; i < stop_block_offset / smpi_shared_malloc_blocksize; i++) {
252 // fprintf(stderr, "\tmmap:for 0x%x - 0x%x\n", i*smpi_shared_malloc_blocksize, smpi_shared_malloc_blocksize);
253 void* pos = (void*)((unsigned long)mem + i * smpi_shared_malloc_blocksize);
254 void* res = mmap(pos, smpi_shared_malloc_blocksize, PROT_READ | PROT_WRITE, MAP_FIXED | MAP_SHARED | MAP_POPULATE,
255 smpi_shared_malloc_bogusfile, 0);
256 xbt_assert(res == pos, "Could not map folded virtual memory (%s). Do you perhaps need to increase the "
257 "size of the mapped file using --cfg=smpi/shared-malloc-blocksize=newvalue (default 1048576) ?"
258 "You can also try using the sysctl vm.max_map_count",
261 int low_page_start_offset = ALIGN_UP(start_offset, PAGE_SIZE);
262 int low_page_stop_offset = start_block_offset < ALIGN_DOWN(stop_offset, PAGE_SIZE) ? start_block_offset : ALIGN_DOWN(stop_offset, PAGE_SIZE);
263 if(low_page_start_offset < low_page_stop_offset) {
264 // fprintf(stderr, "\tmmap:low 0x%x - 0x%x\n", low_page_start_offset, low_page_stop_offset-low_page_start_offset);
265 void* pos = (void*)((unsigned long)mem + low_page_start_offset);
266 void* res = mmap(pos, low_page_stop_offset-low_page_start_offset, PROT_READ | PROT_WRITE, MAP_FIXED | MAP_SHARED | MAP_POPULATE,
267 smpi_shared_malloc_bogusfile, 0);
268 xbt_assert(res == pos, "Could not map folded virtual memory (%s). Do you perhaps need to increase the "
269 "size of the mapped file using --cfg=smpi/shared-malloc-blocksize=newvalue (default 1048576) ?"
270 "You can also try using the sysctl vm.max_map_count",
273 if(low_page_stop_offset <= stop_block_offset) {
274 int high_page_stop_offset = stop_offset == size ? size : ALIGN_DOWN(stop_offset, PAGE_SIZE);
275 if(high_page_stop_offset > stop_block_offset) {
276 // fprintf(stderr, "\tmmap:high 0x%x - 0x%x\n", stop_block_offset, high_page_stop_offset-stop_block_offset);
277 void* pos = (void*)((unsigned long)mem + stop_block_offset);
278 void* res = mmap(pos, high_page_stop_offset-stop_block_offset, PROT_READ | PROT_WRITE, MAP_FIXED | MAP_SHARED | MAP_POPULATE,
279 smpi_shared_malloc_bogusfile, 0);
280 xbt_assert(res == pos, "Could not map folded virtual memory (%s). Do you perhaps need to increase the "
281 "size of the mapped file using --cfg=smpi/shared-malloc-blocksize=newvalue (default 1048576) ?"
282 "You can also try using the sysctl vm.max_map_count",
288 shared_metadata_t newmeta;
289 //register metadata for memcpy avoidance
290 shared_data_key_type* data = (shared_data_key_type*)xbt_malloc(sizeof(shared_data_key_type));
291 data->second.fd = -1;
292 data->second.count = 1;
295 if(shared_block_offsets[0] > 0) {
296 newmeta.private_blocks.push_back(std::make_pair(0, shared_block_offsets[0]));
299 for(i_block = 0; i_block < nb_shared_blocks-1; i_block ++) {
300 newmeta.private_blocks.push_back(std::make_pair(shared_block_offsets[2*i_block+1], shared_block_offsets[2*i_block+2]));
302 if(shared_block_offsets[2*i_block+1] < size) {
303 newmeta.private_blocks.push_back(std::make_pair(shared_block_offsets[2*i_block+1], size));
305 allocs_metadata[mem] = newmeta;
311 * When nb_shared_blocks == -1, default behavior of smpi_shared_malloc: everything is shared.
312 * Otherwise, only the blocks described by shared_block_offsets are shared.
313 * This array contains the offsets (in bytes) of the block to share.
314 * Even indices are the start offsets (included), odd indices are the stop offsets (excluded).
315 * For instance, if shared_block_offsets == {27, 42}, then the elements mem[27], mem[28], ..., mem[41] are shared. The others are not.
317 void *smpi_shared_malloc_global(size_t size, const char *file, int line, int *shared_block_offsets=NULL, int nb_shared_blocks=-1) {
318 int tmp_shared_block_offsets[2];
319 if(nb_shared_blocks == -1) {
320 nb_shared_blocks = 1;
321 shared_block_offsets = tmp_shared_block_offsets;
322 shared_block_offsets[0] = 0;
323 shared_block_offsets[1] = size;
325 return smpi_shared_malloc_global__(size, file, line, shared_block_offsets, nb_shared_blocks);
328 void *smpi_shared_malloc(size_t size, const char *file, int line) {
330 if (size > 0 && smpi_cfg_shared_malloc == shmalloc_local) {
331 mem = smpi_shared_malloc_local(size, file, line);
332 } else if (smpi_cfg_shared_malloc == shmalloc_global) {
333 mem = smpi_shared_malloc_global(size, file, line);
335 mem = xbt_malloc(size);
336 XBT_DEBUG("Classic malloc %zu in %p", size, mem);
341 int smpi_is_shared(void* ptr, std::vector<std::pair<int, int>> &private_blocks, int *offset){
342 private_blocks.clear(); // being paranoid
343 if (allocs_metadata.empty())
345 if ( smpi_cfg_shared_malloc == shmalloc_local || smpi_cfg_shared_malloc == shmalloc_global) {
346 auto low = allocs_metadata.lower_bound(ptr);
347 if (low->first==ptr) {
348 private_blocks = low->second.private_blocks;
352 if (low == allocs_metadata.begin())
355 if (ptr < (char*)low->first + low->second.size) {
356 xbt_assert(ptr > (char*)low->first, "Oops, there seems to be a bug in the shared memory metadata.");
357 *offset = ((uint8_t*)ptr) - ((uint8_t*) low->first);
358 private_blocks = low->second.private_blocks;
367 std::vector<std::pair<int, int>> shift_and_frame_private_blocks(const std::vector<std::pair<int, int>> vec, int offset, int buff_size) {
368 std::vector<std::pair<int, int>> result;
369 for(auto block: vec) {
370 auto new_block = std::make_pair(std::min(std::max(0, block.first-offset), buff_size),
371 std::min(std::max(0, block.second-offset), buff_size));
372 if(new_block.second > 0 && new_block.first < buff_size)
373 result.push_back(new_block);
378 std::vector<std::pair<int, int>> merge_private_blocks(std::vector<std::pair<int, int>> src, std::vector<std::pair<int, int>> dst) {
379 std::vector<std::pair<int, int>> result;
380 unsigned i_src=0, i_dst=0;
381 while(i_src < src.size() && i_dst < dst.size()) {
382 std::pair<int, int> block;
383 if(src[i_src].second <= dst[i_dst].first) {
386 else if(dst[i_dst].second <= src[i_src].first) {
389 else { // src.second > dst.first && dst.second > src.first → the blocks are overlapping
390 block = std::make_pair(std::max(src[i_src].first, dst[i_dst].first),
391 std::min(src[i_src].second, dst[i_dst].second));
392 result.push_back(block);
393 if(src[i_src].second < dst[i_dst].second)
402 void smpi_shared_free(void *ptr)
404 if (smpi_cfg_shared_malloc == shmalloc_local) {
405 char loc[PTR_STRLEN];
406 snprintf(loc, PTR_STRLEN, "%p", ptr);
407 auto meta = allocs_metadata.find(ptr);
408 if (meta == allocs_metadata.end()) {
409 XBT_WARN("Cannot free: %p was not shared-allocated by SMPI - maybe its size was 0?", ptr);
412 shared_data_t* data = &meta->second.data->second;
413 if (munmap(ptr, meta->second.size) < 0) {
414 XBT_WARN("Unmapping of fd %d failed: %s", data->fd, strerror(errno));
417 if (data->count <= 0) {
419 allocs.erase(allocs.find(meta->second.data->first));
420 allocs_metadata.erase(ptr);
421 XBT_DEBUG("Shared free - with removal - of %p", ptr);
423 XBT_DEBUG("Shared free - no removal - of %p, count = %d", ptr, data->count);
426 } else if (smpi_cfg_shared_malloc == shmalloc_global) {
427 auto meta = allocs_metadata.find(ptr);
428 if (meta != allocs_metadata.end()){
429 meta->second.data->second.count--;
430 if(meta->second.data->second.count==0)
431 xbt_free(meta->second.data);
434 munmap(ptr, 0); // the POSIX says that I should not give 0 as a length, but it seems to work OK
436 XBT_DEBUG("Classic free of %p", ptr);
442 int smpi_shared_known_call(const char* func, const char* input)
444 char* loc = bprintf("%s:%s", func, input);
447 if (calls==nullptr) {
448 calls = xbt_dict_new_homogeneous(nullptr);
451 xbt_dict_get(calls, loc); /* Succeed or throw */
457 if (ex.category != not_found_error)
467 void* smpi_shared_get_call(const char* func, const char* input) {
468 char* loc = bprintf("%s:%s", func, input);
470 if (calls == nullptr)
471 calls = xbt_dict_new_homogeneous(nullptr);
472 void* data = xbt_dict_get(calls, loc);
477 void* smpi_shared_set_call(const char* func, const char* input, void* data) {
478 char* loc = bprintf("%s:%s", func, input);
480 if (calls == nullptr)
481 calls = xbt_dict_new_homogeneous(nullptr);
482 xbt_dict_set(calls, loc, data, nullptr);