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 shared_data_key_type* data;
120 std::map<void*, shared_metadata_t> allocs_metadata;
121 xbt_dict_t calls = nullptr; /* Allocated on first use */
123 static int smpi_shared_malloc_bogusfile = -1;
124 static unsigned long smpi_shared_malloc_blocksize = 1UL << 20;
129 void smpi_shared_destroy()
132 allocs_metadata.clear();
133 xbt_dict_free(&calls);
136 static size_t shm_size(int fd) {
139 if(fstat(fd, &st) < 0) {
140 xbt_die("Could not stat fd %d: %s", fd, strerror(errno));
142 return static_cast<size_t>(st.st_size);
146 static void* shm_map(int fd, size_t size, shared_data_key_type* data) {
147 char loc[PTR_STRLEN];
148 shared_metadata_t meta;
150 if(size > shm_size(fd) && (ftruncate(fd, static_cast<off_t>(size)) < 0)) {
151 xbt_die("Could not truncate fd %d to %zu: %s", fd, size, strerror(errno));
154 void* mem = mmap(nullptr, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
155 if(mem == MAP_FAILED) {
157 "Failed to map fd %d with size %zu: %s\n"
158 "If you are running a lot of ranks, you may be exceeding the amount of mappings allowed per process.\n"
159 "On Linux systems, change this value with sudo sysctl -w vm.max_map_count=newvalue (default value: 65536)\n"
160 "Please see http://simgrid.gforge.inria.fr/simgrid/latest/doc/html/options.html#options_virt for more info.",
161 fd, size, strerror(errno));
163 snprintf(loc, PTR_STRLEN, "%p", mem);
166 allocs_metadata[mem] = meta;
167 XBT_DEBUG("MMAP %zu to %p", size, mem);
171 void *smpi_shared_malloc_local(size_t size, const char *file, int line)
174 smpi_source_location loc(file, line);
175 auto res = allocs.insert(std::make_pair(loc, shared_data_t()));
176 auto data = res.first;
178 // The insertion did not take place.
179 // Generate a shared memory name from the address of the shared_data:
180 char shmname[32]; // cannot be longer than PSHMNAMLEN = 31 on Mac OS X (shm_open raises ENAMETOOLONG otherwise)
181 snprintf(shmname, 31, "/shmalloc%p", &*data);
182 int fd = shm_open(shmname, O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
185 xbt_die("Please cleanup /dev/shm/%s", shmname);
187 xbt_die("An unhandled error occurred while opening %s. shm_open: %s", shmname, strerror(errno));
189 data->second.fd = fd;
190 data->second.count = 1;
191 mem = shm_map(fd, size, &*data);
192 if (shm_unlink(shmname) < 0) {
193 XBT_WARN("Could not early unlink %s. shm_unlink: %s", shmname, strerror(errno));
195 XBT_DEBUG("Mapping %s at %p through %d", shmname, mem, fd);
197 mem = shm_map(data->second.fd, size, &*data);
198 data->second.count++;
200 XBT_DEBUG("Shared malloc %zu in %p (metadata at %p)", size, mem, &*data);
204 // Align functions, from http://stackoverflow.com/questions/4840410/how-to-align-a-pointer-in-c
205 #define PAGE_SIZE 0x1000
206 #define ALIGN_UP(n, align) (((n) + (align)-1) & -(align))
207 #define ALIGN_DOWN(n, align) ((n) & -(align))
209 void *smpi_shared_malloc_global__(size_t size, const char *file, int line, int *shared_block_offsets, int nb_shared_blocks) {
211 xbt_assert(smpi_shared_malloc_blocksize % PAGE_SIZE == 0, "The block size of shared malloc should be a multiple of the page size.");
212 /* First reserve memory area */
213 mem = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
215 xbt_assert(mem != MAP_FAILED, "Failed to allocate %luMiB of memory. Run \"sysctl vm.overcommit_memory=1\" as root "
216 "to allow big allocations.\n",
217 (unsigned long)(size >> 20));
219 /* Create bogus file if not done already */
220 if (smpi_shared_malloc_bogusfile == -1) {
221 /* Create a fd to a new file on disk, make it smpi_shared_malloc_blocksize big, and unlink it.
222 * It still exists in memory but not in the file system (thus it cannot be leaked). */
223 smpi_shared_malloc_blocksize = static_cast<unsigned long>(xbt_cfg_get_double("smpi/shared-malloc-blocksize"));
224 XBT_DEBUG("global shared allocation. Blocksize %lu", smpi_shared_malloc_blocksize);
225 char* name = xbt_strdup("/tmp/simgrid-shmalloc-XXXXXX");
226 smpi_shared_malloc_bogusfile = mkstemp(name);
229 char* dumb = (char*)calloc(1, smpi_shared_malloc_blocksize);
230 ssize_t err = write(smpi_shared_malloc_bogusfile, dumb, smpi_shared_malloc_blocksize);
232 xbt_die("Could not write bogus file for shared malloc");
236 /* Map the bogus file in place of the anonymous memory */
237 for(int i_block = 0; i_block < nb_shared_blocks; i_block ++) {
238 int start_offset = shared_block_offsets[2*i_block];
239 int stop_offset = shared_block_offsets[2*i_block+1];
240 xbt_assert(0 <= start_offset, "start_offset (%d) should be greater than 0", start_offset);
241 xbt_assert(start_offset < stop_offset, "start_offset (%d) should be lower than stop offset (%d)", start_offset, stop_offset);
242 xbt_assert(stop_offset <= size, "stop_offset (%d) should be lower than size (%lu)", stop_offset, size);
243 // fprintf(stderr, "shared block 0x%x - 0x%x\n", start_offset, stop_offset);
244 int start_block_offset = ALIGN_UP(start_offset, smpi_shared_malloc_blocksize);
245 int stop_block_offset = ALIGN_DOWN(stop_offset, smpi_shared_malloc_blocksize);
247 for (i = start_block_offset / smpi_shared_malloc_blocksize; i < stop_block_offset / smpi_shared_malloc_blocksize; i++) {
248 // fprintf(stderr, "\tmmap:for 0x%x - 0x%x\n", i*smpi_shared_malloc_blocksize, smpi_shared_malloc_blocksize);
249 void* pos = (void*)((unsigned long)mem + i * smpi_shared_malloc_blocksize);
250 void* res = mmap(pos, smpi_shared_malloc_blocksize, PROT_READ | PROT_WRITE, MAP_FIXED | MAP_SHARED | MAP_POPULATE,
251 smpi_shared_malloc_bogusfile, 0);
252 xbt_assert(res == pos, "Could not map folded virtual memory (%s). Do you perhaps need to increase the "
253 "size of the mapped file using --cfg=smpi/shared-malloc-blocksize=newvalue (default 1048576) ?"
254 "You can also try using the sysctl vm.max_map_count",
257 int low_page_start_offset = ALIGN_UP(start_offset, PAGE_SIZE);
258 int low_page_stop_offset = start_block_offset < ALIGN_DOWN(stop_offset, PAGE_SIZE) ? start_block_offset : ALIGN_DOWN(stop_offset, PAGE_SIZE);
259 if(low_page_start_offset < low_page_stop_offset) {
260 // fprintf(stderr, "\tmmap:low 0x%x - 0x%x\n", low_page_start_offset, low_page_stop_offset-low_page_start_offset);
261 void* pos = (void*)((unsigned long)mem + low_page_start_offset);
262 void* res = mmap(pos, low_page_stop_offset-low_page_start_offset, PROT_READ | PROT_WRITE, MAP_FIXED | MAP_SHARED | MAP_POPULATE,
263 smpi_shared_malloc_bogusfile, 0);
264 xbt_assert(res == pos, "Could not map folded virtual memory (%s). Do you perhaps need to increase the "
265 "size of the mapped file using --cfg=smpi/shared-malloc-blocksize=newvalue (default 1048576) ?"
266 "You can also try using the sysctl vm.max_map_count",
269 if(low_page_stop_offset <= stop_block_offset) {
270 int high_page_stop_offset = stop_offset == size ? size : ALIGN_DOWN(stop_offset, PAGE_SIZE);
271 if(high_page_stop_offset > stop_block_offset) {
272 // fprintf(stderr, "\tmmap:high 0x%x - 0x%x\n", stop_block_offset, high_page_stop_offset-stop_block_offset);
273 void* pos = (void*)((unsigned long)mem + stop_block_offset);
274 void* res = mmap(pos, high_page_stop_offset-stop_block_offset, PROT_READ | PROT_WRITE, MAP_FIXED | MAP_SHARED | MAP_POPULATE,
275 smpi_shared_malloc_bogusfile, 0);
276 xbt_assert(res == pos, "Could not map folded virtual memory (%s). Do you perhaps need to increase the "
277 "size of the mapped file using --cfg=smpi/shared-malloc-blocksize=newvalue (default 1048576) ?"
278 "You can also try using the sysctl vm.max_map_count",
284 if(nb_shared_blocks == 1 && shared_block_offsets[0] == 0 && shared_block_offsets[1] == size) {
285 shared_metadata_t newmeta;
286 //register metadata for memcpy avoidance
287 shared_data_key_type* data = (shared_data_key_type*)xbt_malloc(sizeof(shared_data_key_type));
288 data->second.fd = -1;
289 data->second.count = 1;
292 allocs_metadata[mem] = newmeta;
299 * When nb_shared_blocks == -1, default behavior of smpi_shared_malloc: everything is shared.
300 * Otherwise, only the blocks described by shared_block_offsets are shared.
301 * This array contains the offsets (in bytes) of the block to share.
302 * Even indices are the start offsets (included), odd indices are the stop offsets (excluded).
303 * For instance, if shared_block_offsets == {27, 42}, then the elements mem[27], mem[28], ..., mem[41] are shared. The others are not.
305 void *smpi_shared_malloc_global(size_t size, const char *file, int line, int *shared_block_offsets=NULL, int nb_shared_blocks=-1) {
306 int tmp_shared_block_offsets[2];
307 if(nb_shared_blocks == -1) {
308 nb_shared_blocks = 1;
309 shared_block_offsets = tmp_shared_block_offsets;
310 shared_block_offsets[0] = 0;
311 shared_block_offsets[1] = size;
313 return smpi_shared_malloc_global__(size, file, line, shared_block_offsets, nb_shared_blocks);
316 void *smpi_shared_malloc(size_t size, const char *file, int line) {
318 if (size > 0 && smpi_cfg_shared_malloc == shmalloc_local) {
319 mem = smpi_shared_malloc_local(size, file, line);
320 } else if (smpi_cfg_shared_malloc == shmalloc_global) {
321 mem = smpi_shared_malloc_global(size, file, line);
323 mem = xbt_malloc(size);
324 XBT_DEBUG("Classic malloc %zu in %p", size, mem);
329 int smpi_is_shared(void* ptr){
330 if (allocs_metadata.empty())
332 if ( smpi_cfg_shared_malloc == shmalloc_local || smpi_cfg_shared_malloc == shmalloc_global) {
333 auto low = allocs_metadata.lower_bound(ptr);
336 if (low == allocs_metadata.begin())
339 if (ptr < (char*)low->first + low->second.size)
347 void smpi_shared_free(void *ptr)
349 if (smpi_cfg_shared_malloc == shmalloc_local) {
350 char loc[PTR_STRLEN];
351 snprintf(loc, PTR_STRLEN, "%p", ptr);
352 auto meta = allocs_metadata.find(ptr);
353 if (meta == allocs_metadata.end()) {
354 XBT_WARN("Cannot free: %p was not shared-allocated by SMPI - maybe its size was 0?", ptr);
357 shared_data_t* data = &meta->second.data->second;
358 if (munmap(ptr, meta->second.size) < 0) {
359 XBT_WARN("Unmapping of fd %d failed: %s", data->fd, strerror(errno));
362 if (data->count <= 0) {
364 allocs.erase(allocs.find(meta->second.data->first));
365 allocs_metadata.erase(ptr);
366 XBT_DEBUG("Shared free - with removal - of %p", ptr);
368 XBT_DEBUG("Shared free - no removal - of %p, count = %d", ptr, data->count);
371 } else if (smpi_cfg_shared_malloc == shmalloc_global) {
372 auto meta = allocs_metadata.find(ptr);
373 if (meta != allocs_metadata.end()){
374 meta->second.data->second.count--;
375 if(meta->second.data->second.count==0)
376 xbt_free(meta->second.data);
379 munmap(ptr, 0); // the POSIX says that I should not give 0 as a length, but it seems to work OK
381 XBT_DEBUG("Classic free of %p", ptr);
387 int smpi_shared_known_call(const char* func, const char* input)
389 char* loc = bprintf("%s:%s", func, input);
392 if (calls==nullptr) {
393 calls = xbt_dict_new_homogeneous(nullptr);
396 xbt_dict_get(calls, loc); /* Succeed or throw */
402 if (ex.category != not_found_error)
412 void* smpi_shared_get_call(const char* func, const char* input) {
413 char* loc = bprintf("%s:%s", func, input);
415 if (calls == nullptr)
416 calls = xbt_dict_new_homogeneous(nullptr);
417 void* data = xbt_dict_get(calls, loc);
422 void* smpi_shared_set_call(const char* func, const char* input, void* data) {
423 char* loc = bprintf("%s:%s", func, input);
425 if (calls == nullptr)
426 calls = xbt_dict_new_homogeneous(nullptr);
427 xbt_dict_set(calls, loc, data, nullptr);