4 * Copyright(c) 2010-2012 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
41 #include <sys/queue.h>
43 #include <rte_common.h>
45 #include <rte_debug.h>
46 #include <rte_memory.h>
47 #include <rte_memzone.h>
48 #include <rte_atomic.h>
49 #include <rte_launch.h>
50 #include <rte_tailq.h>
52 #include <rte_eal_memconfig.h>
53 #include <rte_per_lcore.h>
54 #include <rte_lcore.h>
55 #include <rte_branch_prediction.h>
57 #include <rte_errno.h>
58 #include <rte_string_fns.h>
60 #include "rte_mempool.h"
62 TAILQ_HEAD(rte_mempool_list, rte_mempool);
66 * return the greatest common divisor between a and b (fast algorithm)
69 static unsigned get_gcd(unsigned a, unsigned b)
94 * Depending on memory configuration, objects addresses are spreaded
95 * between channels and ranks in RAM: the pool allocator will add
96 * padding between objects. This function return the new size of the
99 static unsigned optimize_object_size(unsigned obj_size)
101 unsigned nrank, nchan;
102 unsigned new_obj_size;
104 /* get number of channels */
105 nchan = rte_memory_get_nchannel();
109 nrank = rte_memory_get_nrank();
113 /* process new object size */
114 new_obj_size = (obj_size + CACHE_LINE_MASK) / CACHE_LINE_SIZE;
115 while (get_gcd(new_obj_size, nrank * nchan) != 1 ||
116 get_gcd(nchan, new_obj_size) != 1)
118 return new_obj_size * CACHE_LINE_SIZE;
121 /* create the mempool */
123 rte_mempool_create(const char *name, unsigned n, unsigned elt_size,
124 unsigned cache_size, unsigned private_data_size,
125 rte_mempool_ctor_t *mp_init, void *mp_init_arg,
126 rte_mempool_obj_ctor_t *obj_init, void *obj_init_arg,
127 int socket_id, unsigned flags)
129 char mz_name[RTE_MEMZONE_NAMESIZE];
130 char rg_name[RTE_RING_NAMESIZE];
131 struct rte_mempool *mp = NULL;
133 const struct rte_memzone *mz;
135 int mz_flags = RTE_MEMZONE_1GB|RTE_MEMZONE_SIZE_HINT_ONLY;
137 uint32_t header_size, trailer_size;
138 uint32_t total_elt_size;
142 /* compilation-time checks */
143 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool) &
144 CACHE_LINE_MASK) != 0);
145 #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
146 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_cache) &
147 CACHE_LINE_MASK) != 0);
148 RTE_BUILD_BUG_ON((offsetof(struct rte_mempool, local_cache) &
149 CACHE_LINE_MASK) != 0);
151 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
152 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_debug_stats) &
153 CACHE_LINE_MASK) != 0);
154 RTE_BUILD_BUG_ON((offsetof(struct rte_mempool, stats) &
155 CACHE_LINE_MASK) != 0);
158 /* check that we have an initialised tail queue */
159 if (RTE_TAILQ_LOOKUP_BY_IDX(RTE_TAILQ_MEMPOOL, rte_mempool_list) == NULL) {
160 rte_errno = E_RTE_NO_TAILQ;
164 /* asked cache too big */
165 if (cache_size > RTE_MEMPOOL_CACHE_MAX_SIZE){
170 /* "no cache align" imply "no spread" */
171 if (flags & MEMPOOL_F_NO_CACHE_ALIGN)
172 flags |= MEMPOOL_F_NO_SPREAD;
175 if (flags & MEMPOOL_F_SP_PUT)
176 rg_flags |= RING_F_SP_ENQ;
177 if (flags & MEMPOOL_F_SC_GET)
178 rg_flags |= RING_F_SC_DEQ;
180 /* allocate the ring that will be used to store objects */
181 /* Ring functions will return appropriate errors if we are
182 * running as a secondary process etc., so no checks made
183 * in this function for that condition */
184 rte_snprintf(rg_name, sizeof(rg_name), "MP_%s", name);
185 r = rte_ring_create(rg_name, rte_align32pow2(n+1), socket_id, rg_flags);
190 * In header, we have at least the pointer to the pool, and
191 * optionaly a 64 bits cookie.
194 header_size += sizeof(struct rte_mempool *); /* ptr to pool */
195 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
196 header_size += sizeof(uint64_t); /* cookie */
198 if ((flags & MEMPOOL_F_NO_CACHE_ALIGN) == 0)
199 header_size = (header_size + CACHE_LINE_MASK) & (~CACHE_LINE_MASK);
201 /* trailer contains the cookie in debug mode */
203 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
204 trailer_size += sizeof(uint64_t); /* cookie */
206 /* element size is 8 bytes-aligned at least */
207 elt_size = (elt_size + 7) & (~7);
209 /* expand trailer to next cache line */
210 if ((flags & MEMPOOL_F_NO_CACHE_ALIGN) == 0) {
211 total_elt_size = header_size + elt_size + trailer_size;
212 trailer_size += ((CACHE_LINE_SIZE -
213 (total_elt_size & CACHE_LINE_MASK)) &
218 * increase trailer to add padding between objects in order to
219 * spread them accross memory channels/ranks
221 if ((flags & MEMPOOL_F_NO_SPREAD) == 0) {
223 new_size = optimize_object_size(header_size + elt_size +
225 trailer_size = new_size - header_size - elt_size;
228 /* this is the size of an object, including header and trailer */
229 total_elt_size = header_size + elt_size + trailer_size;
231 /* reserve a memory zone for this mempool: private data is
233 private_data_size = (private_data_size +
234 CACHE_LINE_MASK) & (~CACHE_LINE_MASK);
235 mempool_size = total_elt_size * n +
236 sizeof(struct rte_mempool) + private_data_size;
237 rte_snprintf(mz_name, sizeof(mz_name), "MP_%s", name);
238 mz = rte_memzone_reserve(mz_name, mempool_size, socket_id, mz_flags);
241 * no more memory: in this case we loose previously reserved
242 * space for the as we cannot free it
247 /* init the mempool structure */
249 memset(mp, 0, sizeof(*mp));
250 rte_snprintf(mp->name, sizeof(mp->name), "%s", name);
251 mp->phys_addr = mz->phys_addr;
255 mp->bulk_default = 1;
256 mp->elt_size = elt_size;
257 mp->header_size = header_size;
258 mp->trailer_size = trailer_size;
259 mp->cache_size = cache_size;
260 mp->private_data_size = private_data_size;
262 /* call the initializer */
264 mp_init(mp, mp_init_arg);
266 /* fill the headers and trailers, and add objects in ring */
267 obj = (char *)mp + sizeof(struct rte_mempool) + private_data_size;
268 for (i = 0; i < n; i++) {
269 struct rte_mempool **mpp;
270 obj = (char *)obj + header_size;
272 /* set mempool ptr in header */
273 mpp = __mempool_from_obj(obj);
276 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
277 __mempool_write_header_cookie(obj, 1);
278 __mempool_write_trailer_cookie(obj);
280 /* call the initializer */
282 obj_init(mp, obj_init_arg, obj, i);
284 /* enqueue in ring */
285 rte_ring_sp_enqueue(mp->ring, obj);
286 obj = (char *)obj + elt_size + trailer_size;
289 RTE_EAL_TAILQ_INSERT_TAIL(RTE_TAILQ_MEMPOOL, rte_mempool_list, mp);
294 /* Return the number of entries in the mempool */
296 rte_mempool_count(const struct rte_mempool *mp)
300 count = rte_ring_count(mp->ring);
302 #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
305 if (mp->cache_size == 0)
308 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
309 count += mp->local_cache[lcore_id].len;
314 * due to race condition (access to len is not locked), the
315 * total can be greater than size... so fix the result
317 if (count > mp->size)
322 /* dump the cache status */
324 rte_mempool_dump_cache(const struct rte_mempool *mp)
326 #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
329 unsigned cache_count;
331 printf(" cache infos:\n");
332 printf(" cache_size=%"PRIu32"\n", mp->cache_size);
333 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
334 cache_count = mp->local_cache[lcore_id].len;
335 printf(" cache_count[%u]=%u\n", lcore_id, cache_count);
336 count += cache_count;
338 printf(" total_cache_count=%u\n", count);
342 printf(" cache disabled\n");
347 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
348 /* check cookies before and after objects */
349 #ifndef __INTEL_COMPILER
350 #pragma GCC diagnostic ignored "-Wcast-qual"
353 mempool_audit_cookies(const struct rte_mempool *mp)
357 void * const *obj_table;
359 obj = (char *)mp + sizeof(struct rte_mempool) + mp->private_data_size;
360 for (i = 0; i < mp->size; i++) {
361 obj = (char *)obj + mp->header_size;
363 __mempool_check_cookies(mp, obj_table, 1, 2);
364 obj = (char *)obj + mp->elt_size + mp->trailer_size;
367 #ifndef __INTEL_COMPILER
368 #pragma GCC diagnostic error "-Wcast-qual"
371 #define mempool_audit_cookies(mp) do {} while(0)
374 #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
375 /* check cookies before and after objects */
377 mempool_audit_cache(const struct rte_mempool *mp)
379 /* check cache size consistency */
381 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
382 if (mp->local_cache[lcore_id].len > mp->cache_size) {
383 RTE_LOG(CRIT, MEMPOOL, "badness on cache[%u]\n",
385 rte_panic("MEMPOOL: invalid cache len\n");
390 #define mempool_audit_cache(mp) do {} while(0)
394 /* check the consistency of mempool (size, cookies, ...) */
396 rte_mempool_audit(const struct rte_mempool *mp)
398 mempool_audit_cache(mp);
399 mempool_audit_cookies(mp);
402 /* dump the status of the mempool on the console */
404 rte_mempool_dump(const struct rte_mempool *mp)
406 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
407 struct rte_mempool_debug_stats sum;
410 unsigned common_count;
411 unsigned cache_count;
413 printf("mempool <%s>@%p\n", mp->name, mp);
414 printf(" flags=%x\n", mp->flags);
415 printf(" ring=<%s>@%p\n", mp->ring->name, mp->ring);
416 printf(" size=%"PRIu32"\n", mp->size);
417 printf(" bulk_default=%"PRIu32"\n", mp->bulk_default);
418 printf(" header_size=%"PRIu32"\n", mp->header_size);
419 printf(" elt_size=%"PRIu32"\n", mp->elt_size);
420 printf(" trailer_size=%"PRIu32"\n", mp->trailer_size);
421 printf(" total_obj_size=%"PRIu32"\n",
422 mp->header_size + mp->elt_size + mp->trailer_size);
424 cache_count = rte_mempool_dump_cache(mp);
425 common_count = rte_ring_count(mp->ring);
426 if ((cache_count + common_count) > mp->size)
427 common_count = mp->size - cache_count;
428 printf(" common_pool_count=%u\n", common_count);
430 /* sum and dump statistics */
431 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
432 memset(&sum, 0, sizeof(sum));
433 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
434 sum.put_bulk += mp->stats[lcore_id].put_bulk;
435 sum.put_objs += mp->stats[lcore_id].put_objs;
436 sum.get_success_bulk += mp->stats[lcore_id].get_success_bulk;
437 sum.get_success_objs += mp->stats[lcore_id].get_success_objs;
438 sum.get_fail_bulk += mp->stats[lcore_id].get_fail_bulk;
439 sum.get_fail_objs += mp->stats[lcore_id].get_fail_objs;
442 printf(" put_bulk=%"PRIu64"\n", sum.put_bulk);
443 printf(" put_objs=%"PRIu64"\n", sum.put_objs);
444 printf(" get_success_bulk=%"PRIu64"\n", sum.get_success_bulk);
445 printf(" get_success_objs=%"PRIu64"\n", sum.get_success_objs);
446 printf(" get_fail_bulk=%"PRIu64"\n", sum.get_fail_bulk);
447 printf(" get_fail_objs=%"PRIu64"\n", sum.get_fail_objs);
449 printf(" no statistics available\n");
452 rte_mempool_audit(mp);
455 /* dump the status of all mempools on the console */
457 rte_mempool_list_dump(void)
459 const struct rte_mempool *mp = NULL;
460 struct rte_mempool_list *mempool_list;
463 RTE_TAILQ_LOOKUP_BY_IDX(RTE_TAILQ_MEMPOOL, rte_mempool_list)) == NULL) {
464 rte_errno = E_RTE_NO_TAILQ;
468 TAILQ_FOREACH(mp, mempool_list, next) {
469 rte_mempool_dump(mp);
473 /* search a mempool from its name */
475 rte_mempool_lookup(const char *name)
477 struct rte_mempool *mp = NULL;
478 struct rte_mempool_list *mempool_list;
481 RTE_TAILQ_LOOKUP_BY_IDX(RTE_TAILQ_MEMPOOL, rte_mempool_list)) == NULL) {
482 rte_errno = E_RTE_NO_TAILQ;
486 TAILQ_FOREACH(mp, mempool_list, next) {
487 if (strncmp(name, mp->name, RTE_MEMPOOL_NAMESIZE) == 0)