lib/ovs-numa.c

   1 /*
   2  * Copyright (c) 2014 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 /* On non-Linux, these functions are defined inline in ovs-numa.h. */
  18 #ifdef __linux__
  19
  20 #include <config.h>
  21 #include "ovs-numa.h"
  22
  23 #include <ctype.h>
  24 #include <dirent.h>
  25 #include <errno.h>
  26 #include <stddef.h>
  27 #include <string.h>
  28 #include <sys/types.h>
  29 #include <unistd.h>
  30
  31 #include "hash.h"
  32 #include "hmap.h"
  33 #include "list.h"
  34 #include "ovs-thread.h"
  35 #include "openvswitch/vlog.h"
  36
  37 VLOG_DEFINE_THIS_MODULE(ovs_numa);
  38
  39 /* ovs-numa module
  40  * ===============
  41  *
  42  * This module stores the affinity information of numa nodes and cpu cores.
  43  * It also provides functions to bookkeep the pin of threads on cpu cores.
  44  *
  45  * It is assumed that the numa node ids and cpu core ids all start from 0 and
  46  * range continuously.  So, for example, if 'ovs_numa_get_n_cores()' returns N,
  47  * user can assume core ids from 0 to N-1 are all valid and there is a
  48  * 'struct cpu_core' for each id.
  49  *
  50  * NOTE, this module should only be used by the main thread.
  51  *
  52  * NOTE, the assumption above will fail when cpu hotplug is used.  In that
  53  * case ovs-numa will not function correctly.  For now, add a TODO entry
  54  * for addressing it in the future.
  55  *
  56  * TODO: Fix ovs-numa when cpu hotplug is used.
  57  */
  58
  59 #define MAX_NUMA_NODES 128
  60
  61 /* numa node. */
  62 struct numa_node {
  63     struct hmap_node hmap_node;     /* In the 'all_numa_nodes'. */
  64     struct ovs_list cores;          /* List of cpu cores on the numa node. */
  65     int numa_id;                    /* numa node id. */
  66 };
  67
  68 /* Cpu core on a numa node. */
  69 struct cpu_core {
  70     struct hmap_node hmap_node;/* In the 'all_cpu_cores'. */
  71     struct ovs_list list_node; /* In 'numa_node->cores' list. */
  72     struct numa_node *numa;    /* numa node containing the core. */
  73     int core_id;               /* Core id. */
  74     bool available;            /* If the core can be pinned. */
  75     bool pinned;               /* If a thread has been pinned to the core. */
  76 };
  77
  78 /* Contains all 'struct numa_node's. */
  79 static struct hmap all_numa_nodes = HMAP_INITIALIZER(&all_numa_nodes);
  80 /* Contains all 'struct cpu_core's. */
  81 static struct hmap all_cpu_cores = HMAP_INITIALIZER(&all_cpu_cores);
  82 /* True if numa node and core info are correctly extracted. */
  83 static bool found_numa_and_core;
  84
  85 /* Returns true if 'str' contains all digits.  Returns false otherwise. */
  86 static bool
  87 contain_all_digits(const char *str)
  88 {
  89     return str[strspn(str, "0123456789")] == '\0';
  90 }
  91
  92 /* Discovers all numa nodes and the corresponding cpu cores.
  93  * Constructs the 'struct numa_node' and 'struct cpu_core'. */
  94 static void
  95 discover_numa_and_core(void)
  96 {
  97     int n_cpus = 0;
  98     int i;
  99
 100     for (i = 0; i < MAX_NUMA_NODES; i++) {
 101         DIR *dir;
 102         char* path;
 103
 104         /* Constructs the path to node /sys/devices/system/nodeX. */
 105         path = xasprintf("/sys/devices/system/node/node%d", i);
 106         dir = opendir(path);
 107
 108         /* Creates 'struct numa_node' if the 'dir' is non-null. */
 109         if (dir) {
 110             struct numa_node *n = xzalloc(sizeof *n);
 111             struct dirent *subdir;
 112
 113             hmap_insert(&all_numa_nodes, &n->hmap_node, hash_int(i, 0));
 114             list_init(&n->cores);
 115             n->numa_id = i;
 116
 117             while ((subdir = readdir(dir)) != NULL) {
 118                 if (!strncmp(subdir->d_name, "cpu", 3)
 119                     && contain_all_digits(subdir->d_name + 3)){
 120                     struct cpu_core *c = xzalloc(sizeof *c);
 121                     uint32_t core_id;
 122
 123                     core_id = strtoul(subdir->d_name + 3, NULL, 10);
 124                     hmap_insert(&all_cpu_cores, &c->hmap_node,
 125                                 hash_int(core_id, 0));
 126                     list_insert(&n->cores, &c->list_node);
 127                     c->core_id = core_id;
 128                     c->numa = n;
 129                     c->available = true;
 130                     n_cpus++;
 131                 }
 132             }
 133             VLOG_INFO("Discovered %"PRIuSIZE" CPU cores on NUMA node %d",
 134                       list_size(&n->cores), n->numa_id);
 135             free(path);
 136             closedir(dir);
 137         } else {
 138             if (errno != ENOENT) {
 139                 VLOG_WARN("opendir(%s) failed (%s)", path,
 140                           ovs_strerror(errno));
 141             }
 142             free(path);
 143             break;
 144         }
 145     }
 146
 147     VLOG_INFO("Discovered %"PRIuSIZE" NUMA nodes and %d CPU cores",
 148                hmap_count(&all_numa_nodes), n_cpus);
 149     if (hmap_count(&all_numa_nodes) && hmap_count(&all_cpu_cores)) {
 150         found_numa_and_core = true;
 151     }
 152 }
 153
 154 /* Gets 'struct cpu_core' by 'core_id'. */
 155 static struct cpu_core*
 156 get_core_by_core_id(int core_id)
 157 {
 158     struct cpu_core *core = NULL;
 159
 160     if (ovs_numa_core_id_is_valid(core_id)) {
 161         core = CONTAINER_OF(hmap_first_with_hash(&all_cpu_cores,
 162                                                  hash_int(core_id, 0)),
 163                             struct cpu_core, hmap_node);
 164     }
 165
 166     return core;
 167 }
 168
 169 /* Gets 'struct numa_node' by 'numa_id'. */
 170 static struct numa_node*
 171 get_numa_by_numa_id(int numa_id)
 172 {
 173     struct numa_node *numa = NULL;
 174
 175     if (ovs_numa_numa_id_is_valid(numa_id)) {
 176         numa = CONTAINER_OF(hmap_first_with_hash(&all_numa_nodes,
 177                                                  hash_int(numa_id, 0)),
 178                             struct numa_node, hmap_node);
 179     }
 180
 181     return numa;
 182 }
 183
 184 \f
 185 /* Extracts the numa node and core info from the 'sysfs'. */
 186 void
 187 ovs_numa_init(void)
 188 {
 189     static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
 190
 191     if (ovsthread_once_start(&once)) {
 192         discover_numa_and_core();
 193         ovsthread_once_done(&once);
 194     }
 195 }
 196
 197 bool
 198 ovs_numa_numa_id_is_valid(int numa_id)
 199 {
 200     return found_numa_and_core && numa_id < ovs_numa_get_n_numas();
 201 }
 202
 203 bool
 204 ovs_numa_core_id_is_valid(int core_id)
 205 {
 206     return found_numa_and_core && core_id < ovs_numa_get_n_cores();
 207 }
 208
 209 bool
 210 ovs_numa_core_is_pinned(int core_id)
 211 {
 212     struct cpu_core *core = get_core_by_core_id(core_id);
 213
 214     if (core) {
 215         return core->pinned;
 216     }
 217
 218     return false;
 219 }
 220
 221 /* Returns the number of numa nodes. */
 222 int
 223 ovs_numa_get_n_numas(void)
 224 {
 225     return found_numa_and_core ? hmap_count(&all_numa_nodes)
 226                                : OVS_NUMA_UNSPEC;
 227 }
 228
 229 /* Returns the number of cpu cores. */
 230 int
 231 ovs_numa_get_n_cores(void)
 232 {
 233     return found_numa_and_core ? hmap_count(&all_cpu_cores)
 234                                : OVS_CORE_UNSPEC;
 235 }
 236
 237 /* Given 'core_id', returns the corresponding numa node id.  Returns
 238  * OVS_NUMA_UNSPEC if 'core_id' is invalid. */
 239 int
 240 ovs_numa_get_numa_id(int core_id)
 241 {
 242     struct cpu_core *core = get_core_by_core_id(core_id);
 243
 244     if (core) {
 245         return core->numa->numa_id;
 246     }
 247
 248     return OVS_NUMA_UNSPEC;
 249 }
 250
 251 /* Returns the number of cpu cores on numa node.  Returns OVS_CORE_UNSPEC
 252  * if 'numa_id' is invalid. */
 253 int
 254 ovs_numa_get_n_cores_on_numa(int numa_id)
 255 {
 256     struct numa_node *numa = get_numa_by_numa_id(numa_id);
 257
 258     if (numa) {
 259         return list_size(&numa->cores);
 260     }
 261
 262     return OVS_CORE_UNSPEC;
 263 }
 264
 265 /* Returns the number of cpu cores that are available and unpinned
 266  * on numa node.  Returns OVS_CORE_UNSPEC if 'numa_id' is invalid. */
 267 int
 268 ovs_numa_get_n_unpinned_cores_on_numa(int numa_id)
 269 {
 270     struct numa_node *numa = get_numa_by_numa_id(numa_id);
 271
 272     if (numa) {
 273         struct cpu_core *core;
 274         int count = 0;
 275
 276         LIST_FOR_EACH(core, list_node, &numa->cores) {
 277             if (core->available && !core->pinned) {
 278                 count++;
 279             }
 280         }
 281         return count;
 282     }
 283
 284     return OVS_CORE_UNSPEC;
 285 }
 286
 287 /* Given 'core_id', tries to pin that core.  Returns true, if succeeds.
 288  * False, if the core has already been pinned, or if it is invalid or
 289  * not available. */
 290 bool
 291 ovs_numa_try_pin_core_specific(int core_id)
 292 {
 293     struct cpu_core *core = get_core_by_core_id(core_id);
 294
 295     if (core) {
 296         if (core->available && !core->pinned) {
 297             core->pinned = true;
 298             return true;
 299         }
 300     }
 301
 302     return false;
 303 }
 304
 305 /* Searches through all cores for an unpinned and available core.  Returns
 306  * the 'core_id' if found and sets the 'core->pinned' to true.  Otherwise,
 307  * returns OVS_CORE_UNSPEC. */
 308 int
 309 ovs_numa_get_unpinned_core_any(void)
 310 {
 311     struct cpu_core *core;
 312
 313     HMAP_FOR_EACH(core, hmap_node, &all_cpu_cores) {
 314         if (core->available && !core->pinned) {
 315             core->pinned = true;
 316             return core->core_id;
 317         }
 318     }
 319
 320     return OVS_CORE_UNSPEC;
 321 }
 322
 323 /* Searches through all cores on numa node with 'numa_id' for an
 324  * unpinned and available core.  Returns the core_id if found and
 325  * sets the 'core->pinned' to true.  Otherwise, returns OVS_CORE_UNSPEC. */
 326 int
 327 ovs_numa_get_unpinned_core_on_numa(int numa_id)
 328 {
 329     struct numa_node *numa = get_numa_by_numa_id(numa_id);
 330
 331     if (numa) {
 332         struct cpu_core *core;
 333
 334         LIST_FOR_EACH(core, list_node, &numa->cores) {
 335             if (core->available && !core->pinned) {
 336                 core->pinned = true;
 337                 return core->core_id;
 338             }
 339         }
 340     }
 341
 342     return OVS_CORE_UNSPEC;
 343 }
 344
 345 /* Unpins the core with 'core_id'. */
 346 void
 347 ovs_numa_unpin_core(int core_id)
 348 {
 349     struct cpu_core *core = get_core_by_core_id(core_id);
 350
 351     if (core) {
 352         core->pinned = false;
 353     }
 354 }
 355
 356 /* Given the 'numa_id', returns dump of all cores on the numa node. */
 357 struct ovs_numa_dump *
 358 ovs_numa_dump_cores_on_numa(int numa_id)
 359 {
 360     struct ovs_numa_dump *dump = NULL;
 361     struct numa_node *numa = get_numa_by_numa_id(numa_id);
 362
 363     if (numa) {
 364         struct cpu_core *core;
 365
 366         dump = xmalloc(sizeof *dump);
 367         list_init(&dump->dump);
 368         LIST_FOR_EACH(core, list_node, &numa->cores) {
 369             struct ovs_numa_info *info = xmalloc(sizeof *info);
 370
 371             info->numa_id = numa->numa_id;
 372             info->core_id = core->core_id;
 373             list_insert(&dump->dump, &info->list_node);
 374         }
 375     }
 376
 377     return dump;
 378 }
 379
 380 void
 381 ovs_numa_dump_destroy(struct ovs_numa_dump *dump)
 382 {
 383     struct ovs_numa_info *iter, *next;
 384
 385     LIST_FOR_EACH_SAFE (iter, next, list_node, &dump->dump) {
 386         list_remove(&iter->list_node);
 387         free(iter);
 388     }
 389
 390     free(dump);
 391 }
 392
 393 /* Reads the cpu mask configuration from 'cmask' and sets the
 394  * 'available' of corresponding cores.  For unspecified cores,
 395  * sets 'available' to false. */
 396 void
 397 ovs_numa_set_cpu_mask(const char *cmask)
 398 {
 399     int core_id = 0;
 400     int i;
 401
 402     if (!found_numa_and_core) {
 403         return;
 404     }
 405
 406     /* If no mask specified, resets the 'available' to true for all cores. */
 407     if (!cmask) {
 408         struct cpu_core *core;
 409
 410         HMAP_FOR_EACH(core, hmap_node, &all_cpu_cores) {
 411             core->available = true;
 412         }
 413
 414         return;
 415     }
 416
 417     for (i = strlen(cmask) - 1; i >= 0; i--) {
 418         char hex = toupper(cmask[i]);
 419         int bin, j;
 420
 421         if (hex >= '0' && hex <= '9') {
 422             bin = hex - '0';
 423         } else if (hex >= 'A' && hex <= 'F') {
 424             bin = hex - 'A' + 10;
 425         } else {
 426             bin = 0;
 427             VLOG_WARN("Invalid cpu mask: %c", cmask[i]);
 428         }
 429
 430         for (j = 0; j < 4; j++) {
 431             struct cpu_core *core;
 432
 433             core = CONTAINER_OF(hmap_first_with_hash(&all_cpu_cores,
 434                                                      hash_int(core_id++, 0)),
 435                                 struct cpu_core, hmap_node);
 436             core->available = (bin >> j) & 0x1;
 437
 438             if (core_id >= hmap_count(&all_cpu_cores)) {
 439                 return;
 440             }
 441         }
 442     }
 443
 444     /* For unspecified cores, sets 'available' to false.  */
 445     while (core_id < hmap_count(&all_cpu_cores)) {
 446         struct cpu_core *core;
 447
 448         core = CONTAINER_OF(hmap_first_with_hash(&all_cpu_cores,
 449                                                  hash_int(core_id++, 0)),
 450                             struct cpu_core, hmap_node);
 451         core->available = false;
 452     }
 453 }
 454
 455 #endif /* __linux__ */