2 * Copyright (c) 2014 Nicira, Inc.
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:
8 * http://www.apache.org/licenses/LICENSE-2.0
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.
17 /* On non-Linux, these functions are defined inline in ovs-numa.h. */
28 #include <sys/types.h>
34 #include "ovs-thread.h"
35 #include "openvswitch/vlog.h"
37 VLOG_DEFINE_THIS_MODULE(ovs_numa);
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.
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.
50 * NOTE, this module should only be used by the main thread.
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.
56 * TODO: Fix ovs-numa when cpu hotplug is used.
59 #define MAX_NUMA_NODES 128
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. */
68 /* Cpu core on a numa node. */
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. */
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;
85 /* Returns true if 'str' contains all digits. Returns false otherwise. */
87 contain_all_digits(const char *str)
89 return str[strspn(str, "0123456789")] == '\0';
92 /* Discovers all numa nodes and the corresponding cpu cores.
93 * Constructs the 'struct numa_node' and 'struct cpu_core'. */
95 discover_numa_and_core(void)
100 for (i = 0; i < MAX_NUMA_NODES; i++) {
104 /* Constructs the path to node /sys/devices/system/nodeX. */
105 path = xasprintf("/sys/devices/system/node/node%d", i);
108 /* Creates 'struct numa_node' if the 'dir' is non-null. */
110 struct numa_node *n = xzalloc(sizeof *n);
111 struct dirent *subdir;
113 hmap_insert(&all_numa_nodes, &n->hmap_node, hash_int(i, 0));
114 list_init(&n->cores);
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);
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;
133 VLOG_INFO("Discovered %"PRIuSIZE" CPU cores on NUMA node %d",
134 list_size(&n->cores), n->numa_id);
138 if (errno != ENOENT) {
139 VLOG_WARN("opendir(%s) failed (%s)", path,
140 ovs_strerror(errno));
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;
154 /* Gets 'struct cpu_core' by 'core_id'. */
155 static struct cpu_core*
156 get_core_by_core_id(int core_id)
158 struct cpu_core *core = NULL;
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);
169 /* Gets 'struct numa_node' by 'numa_id'. */
170 static struct numa_node*
171 get_numa_by_numa_id(int numa_id)
173 struct numa_node *numa = NULL;
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);
185 /* Extracts the numa node and core info from the 'sysfs'. */
189 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
191 if (ovsthread_once_start(&once)) {
192 discover_numa_and_core();
193 ovsthread_once_done(&once);
198 ovs_numa_numa_id_is_valid(int numa_id)
200 return found_numa_and_core && numa_id < ovs_numa_get_n_numas();
204 ovs_numa_core_id_is_valid(int core_id)
206 return found_numa_and_core && core_id < ovs_numa_get_n_cores();
210 ovs_numa_core_is_pinned(int core_id)
212 struct cpu_core *core = get_core_by_core_id(core_id);
221 /* Returns the number of numa nodes. */
223 ovs_numa_get_n_numas(void)
225 return found_numa_and_core ? hmap_count(&all_numa_nodes)
229 /* Returns the number of cpu cores. */
231 ovs_numa_get_n_cores(void)
233 return found_numa_and_core ? hmap_count(&all_cpu_cores)
237 /* Given 'core_id', returns the corresponding numa node id. Returns
238 * OVS_NUMA_UNSPEC if 'core_id' is invalid. */
240 ovs_numa_get_numa_id(int core_id)
242 struct cpu_core *core = get_core_by_core_id(core_id);
245 return core->numa->numa_id;
248 return OVS_NUMA_UNSPEC;
251 /* Returns the number of cpu cores on numa node. Returns OVS_CORE_UNSPEC
252 * if 'numa_id' is invalid. */
254 ovs_numa_get_n_cores_on_numa(int numa_id)
256 struct numa_node *numa = get_numa_by_numa_id(numa_id);
259 return list_size(&numa->cores);
262 return OVS_CORE_UNSPEC;
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. */
268 ovs_numa_get_n_unpinned_cores_on_numa(int numa_id)
270 struct numa_node *numa = get_numa_by_numa_id(numa_id);
273 struct cpu_core *core;
276 LIST_FOR_EACH(core, list_node, &numa->cores) {
277 if (core->available && !core->pinned) {
284 return OVS_CORE_UNSPEC;
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
291 ovs_numa_try_pin_core_specific(int core_id)
293 struct cpu_core *core = get_core_by_core_id(core_id);
296 if (core->available && !core->pinned) {
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. */
309 ovs_numa_get_unpinned_core_any(void)
311 struct cpu_core *core;
313 HMAP_FOR_EACH(core, hmap_node, &all_cpu_cores) {
314 if (core->available && !core->pinned) {
316 return core->core_id;
320 return OVS_CORE_UNSPEC;
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. */
327 ovs_numa_get_unpinned_core_on_numa(int numa_id)
329 struct numa_node *numa = get_numa_by_numa_id(numa_id);
332 struct cpu_core *core;
334 LIST_FOR_EACH(core, list_node, &numa->cores) {
335 if (core->available && !core->pinned) {
337 return core->core_id;
342 return OVS_CORE_UNSPEC;
345 /* Unpins the core with 'core_id'. */
347 ovs_numa_unpin_core(int core_id)
349 struct cpu_core *core = get_core_by_core_id(core_id);
352 core->pinned = false;
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)
360 struct ovs_numa_dump *dump = NULL;
361 struct numa_node *numa = get_numa_by_numa_id(numa_id);
364 struct cpu_core *core;
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);
371 info->numa_id = numa->numa_id;
372 info->core_id = core->core_id;
373 list_insert(&dump->dump, &info->list_node);
381 ovs_numa_dump_destroy(struct ovs_numa_dump *dump)
383 struct ovs_numa_info *iter, *next;
385 LIST_FOR_EACH_SAFE (iter, next, list_node, &dump->dump) {
386 list_remove(&iter->list_node);
393 /* Reads the cpu mask configuration from 'cmask' and sets the
394 * 'available' of corresponding cores. For unspecified cores,
395 * sets 'available' to false. */
397 ovs_numa_set_cpu_mask(const char *cmask)
402 if (!found_numa_and_core) {
406 /* If no mask specified, resets the 'available' to true for all cores. */
408 struct cpu_core *core;
410 HMAP_FOR_EACH(core, hmap_node, &all_cpu_cores) {
411 core->available = true;
417 for (i = strlen(cmask) - 1; i >= 0; i--) {
418 char hex = toupper(cmask[i]);
421 if (hex >= '0' && hex <= '9') {
423 } else if (hex >= 'A' && hex <= 'F') {
424 bin = hex - 'A' + 10;
427 VLOG_WARN("Invalid cpu mask: %c", cmask[i]);
430 for (j = 0; j < 4; j++) {
431 struct cpu_core *core;
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;
438 if (core_id >= hmap_count(&all_cpu_cores)) {
444 /* For unspecified cores, sets 'available' to false. */
445 while (core_id < hmap_count(&all_cpu_cores)) {
446 struct cpu_core *core;
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;
455 #endif /* __linux__ */