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_per_lcore.h>
53 #include <rte_lcore.h>
54 #include <rte_branch_prediction.h>
56 #include <rte_errno.h>
57 #include <rte_string_fns.h>
59 #include "rte_mempool.h"
61 TAILQ_HEAD(rte_mempool_list, rte_mempool);
63 /* global list of mempool (used for debug/dump) */
64 static struct rte_mempool_list *mempool_list;
67 * return the greatest common divisor between a and b (fast algorithm)
70 static unsigned get_gcd(unsigned a, unsigned b)
95 * Depending on memory configuration, objects addresses are spreaded
96 * between channels and ranks in RAM: the pool allocator will add
97 * padding between objects. This function return the new size of the
100 static unsigned optimize_object_size(unsigned obj_size)
102 unsigned nrank, nchan;
103 unsigned new_obj_size;
105 /* get number of channels */
106 nchan = rte_memory_get_nchannel();
110 nrank = rte_memory_get_nrank();
114 /* process new object size */
115 new_obj_size = (obj_size + CACHE_LINE_MASK) / CACHE_LINE_SIZE;
116 while (get_gcd(new_obj_size, nrank * nchan) != 1 ||
117 get_gcd(nchan, new_obj_size) != 1)
119 return new_obj_size * CACHE_LINE_SIZE;
122 /* create the mempool */
124 rte_mempool_create(const char *name, unsigned n, unsigned elt_size,
125 unsigned cache_size, unsigned private_data_size,
126 rte_mempool_ctor_t *mp_init, void *mp_init_arg,
127 rte_mempool_obj_ctor_t *obj_init, void *obj_init_arg,
128 int socket_id, unsigned flags)
130 char mz_name[RTE_MEMZONE_NAMESIZE];
131 char rg_name[RTE_RING_NAMESIZE];
132 struct rte_mempool *mp;
134 const struct rte_memzone *mz;
136 int mz_flags = RTE_MEMZONE_1GB|RTE_MEMZONE_SIZE_HINT_ONLY;
138 uint32_t header_size, trailer_size;
139 uint32_t total_elt_size;
143 /* compilation-time checks */
144 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool) &
145 CACHE_LINE_MASK) != 0);
146 #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
147 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_cache) &
148 CACHE_LINE_MASK) != 0);
149 RTE_BUILD_BUG_ON((offsetof(struct rte_mempool, local_cache) &
150 CACHE_LINE_MASK) != 0);
152 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
153 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_debug_stats) &
154 CACHE_LINE_MASK) != 0);
155 RTE_BUILD_BUG_ON((offsetof(struct rte_mempool, stats) &
156 CACHE_LINE_MASK) != 0);
159 /* check that we have an initialised tail queue */
160 if (mempool_list == NULL)
161 if ((mempool_list = RTE_TAILQ_RESERVE("RTE_MEMPOOL", \
162 rte_mempool_list)) == NULL){
163 rte_errno = E_RTE_NO_TAILQ;
167 /* asked cache too big */
168 if (cache_size > RTE_MEMPOOL_CACHE_MAX_SIZE){
173 /* "no cache align" imply "no spread" */
174 if (flags & MEMPOOL_F_NO_CACHE_ALIGN)
175 flags |= MEMPOOL_F_NO_SPREAD;
178 if (flags & MEMPOOL_F_SP_PUT)
179 rg_flags |= RING_F_SP_ENQ;
180 if (flags & MEMPOOL_F_SC_GET)
181 rg_flags |= RING_F_SC_DEQ;
183 /* allocate the ring that will be used to store objects */
184 /* Ring functions will return appropriate errors if we are
185 * running as a secondary process etc., so no checks made
186 * in this function for that condition */
187 rte_snprintf(rg_name, sizeof(rg_name), "MP_%s", name);
188 r = rte_ring_create(rg_name, rte_align32pow2(n+1), socket_id, rg_flags);
193 * In header, we have at least the pointer to the pool, and
194 * optionaly a 64 bits cookie.
197 header_size += sizeof(struct rte_mempool *); /* ptr to pool */
198 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
199 header_size += sizeof(uint64_t); /* cookie */
201 if ((flags & MEMPOOL_F_NO_CACHE_ALIGN) == 0)
202 header_size = (header_size + CACHE_LINE_MASK) & (~CACHE_LINE_MASK);
204 /* trailer contains the cookie in debug mode */
206 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
207 trailer_size += sizeof(uint64_t); /* cookie */
209 /* element size is 8 bytes-aligned at least */
210 elt_size = (elt_size + 7) & (~7);
212 /* expand trailer to next cache line */
213 if ((flags & MEMPOOL_F_NO_CACHE_ALIGN) == 0) {
214 total_elt_size = header_size + elt_size + trailer_size;
215 trailer_size += ((CACHE_LINE_SIZE -
216 (total_elt_size & CACHE_LINE_MASK)) &
221 * increase trailer to add padding between objects in order to
222 * spread them accross memory channels/ranks
224 if ((flags & MEMPOOL_F_NO_SPREAD) == 0) {
226 new_size = optimize_object_size(header_size + elt_size +
228 trailer_size = new_size - header_size - elt_size;
231 /* this is the size of an object, including header and trailer */
232 total_elt_size = header_size + elt_size + trailer_size;
234 /* reserve a memory zone for this mempool: private data is
236 private_data_size = (private_data_size +
237 CACHE_LINE_MASK) & (~CACHE_LINE_MASK);
238 mempool_size = total_elt_size * n +
239 sizeof(struct rte_mempool) + private_data_size;
240 rte_snprintf(mz_name, sizeof(mz_name), "MP_%s", name);
241 mz = rte_memzone_reserve(mz_name, mempool_size, socket_id, mz_flags);
244 * no more memory: in this case we loose previously reserved
245 * space for the as we cannot free it
250 /* init the mempool structure */
252 memset(mp, 0, sizeof(*mp));
253 rte_snprintf(mp->name, sizeof(mp->name), "%s", name);
254 mp->phys_addr = mz->phys_addr;
258 mp->bulk_default = 1;
259 mp->elt_size = elt_size;
260 mp->header_size = header_size;
261 mp->trailer_size = trailer_size;
262 mp->cache_size = cache_size;
263 mp->private_data_size = private_data_size;
265 /* call the initializer */
267 mp_init(mp, mp_init_arg);
269 /* fill the headers and trailers, and add objects in ring */
270 obj = (char *)mp + sizeof(struct rte_mempool) + private_data_size;
271 for (i = 0; i < n; i++) {
272 struct rte_mempool **mpp;
273 obj = (char *)obj + header_size;
275 /* set mempool ptr in header */
276 mpp = __mempool_from_obj(obj);
279 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
280 __mempool_write_header_cookie(obj, 1);
281 __mempool_write_trailer_cookie(obj);
283 /* call the initializer */
285 obj_init(mp, obj_init_arg, obj, i);
287 /* enqueue in ring */
288 rte_ring_sp_enqueue(mp->ring, obj);
289 obj = (char *)obj + elt_size + trailer_size;
292 TAILQ_INSERT_TAIL(mempool_list, mp, next);
296 /* Return the number of entries in the mempool */
298 rte_mempool_count(const struct rte_mempool *mp)
302 count = rte_ring_count(mp->ring);
304 #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
307 if (mp->cache_size == 0)
310 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
311 count += mp->local_cache[lcore_id].len;
316 * due to race condition (access to len is not locked), the
317 * total can be greater than size... so fix the result
319 if (count > mp->size)
324 /* dump the cache status */
326 rte_mempool_dump_cache(const struct rte_mempool *mp)
328 #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
331 unsigned cache_count;
333 printf(" cache infos:\n");
334 printf(" cache_size=%"PRIu32"\n", mp->cache_size);
335 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
336 cache_count = mp->local_cache[lcore_id].len;
337 printf(" cache_count[%u]=%u\n", lcore_id, cache_count);
338 count += cache_count;
340 printf(" total_cache_count=%u\n", count);
344 printf(" cache disabled\n");
349 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
350 /* check cookies before and after objects */
351 #ifndef __INTEL_COMPILER
352 #pragma GCC diagnostic ignored "-Wcast-qual"
355 mempool_audit_cookies(const struct rte_mempool *mp)
359 void * const *obj_table;
361 obj = (char *)mp + sizeof(struct rte_mempool) + mp->private_data_size;
362 for (i = 0; i < mp->size; i++) {
363 obj = (char *)obj + mp->header_size;
365 __mempool_check_cookies(mp, obj_table, 1, 2);
366 obj = (char *)obj + mp->elt_size + mp->trailer_size;
369 #ifndef __INTEL_COMPILER
370 #pragma GCC diagnostic error "-Wcast-qual"
373 #define mempool_audit_cookies(mp) do {} while(0)
376 #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
377 /* check cookies before and after objects */
379 mempool_audit_cache(const struct rte_mempool *mp)
381 /* check cache size consistency */
383 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
384 if (mp->local_cache[lcore_id].len > mp->cache_size) {
385 RTE_LOG(CRIT, MEMPOOL, "badness on cache[%u]\n",
387 rte_panic("MEMPOOL: invalid cache len\n");
392 #define mempool_audit_cache(mp) do {} while(0)
396 /* check the consistency of mempool (size, cookies, ...) */
398 rte_mempool_audit(const struct rte_mempool *mp)
400 mempool_audit_cache(mp);
401 mempool_audit_cookies(mp);
404 /* dump the status of the mempool on the console */
406 rte_mempool_dump(const struct rte_mempool *mp)
408 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
409 struct rte_mempool_debug_stats sum;
412 unsigned common_count;
413 unsigned cache_count;
415 printf("mempool <%s>@%p\n", mp->name, mp);
416 printf(" flags=%x\n", mp->flags);
417 printf(" ring=<%s>@%p\n", mp->ring->name, mp->ring);
418 printf(" size=%"PRIu32"\n", mp->size);
419 printf(" bulk_default=%"PRIu32"\n", mp->bulk_default);
420 printf(" header_size=%"PRIu32"\n", mp->header_size);
421 printf(" elt_size=%"PRIu32"\n", mp->elt_size);
422 printf(" trailer_size=%"PRIu32"\n", mp->trailer_size);
423 printf(" total_obj_size=%"PRIu32"\n",
424 mp->header_size + mp->elt_size + mp->trailer_size);
426 cache_count = rte_mempool_dump_cache(mp);
427 common_count = rte_ring_count(mp->ring);
428 if ((cache_count + common_count) > mp->size)
429 common_count = mp->size - cache_count;
430 printf(" common_pool_count=%u\n", common_count);
432 /* sum and dump statistics */
433 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
434 memset(&sum, 0, sizeof(sum));
435 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
436 sum.put_bulk += mp->stats[lcore_id].put_bulk;
437 sum.put_objs += mp->stats[lcore_id].put_objs;
438 sum.get_success_bulk += mp->stats[lcore_id].get_success_bulk;
439 sum.get_success_objs += mp->stats[lcore_id].get_success_objs;
440 sum.get_fail_bulk += mp->stats[lcore_id].get_fail_bulk;
441 sum.get_fail_objs += mp->stats[lcore_id].get_fail_objs;
444 printf(" put_bulk=%"PRIu64"\n", sum.put_bulk);
445 printf(" put_objs=%"PRIu64"\n", sum.put_objs);
446 printf(" get_success_bulk=%"PRIu64"\n", sum.get_success_bulk);
447 printf(" get_success_objs=%"PRIu64"\n", sum.get_success_objs);
448 printf(" get_fail_bulk=%"PRIu64"\n", sum.get_fail_bulk);
449 printf(" get_fail_objs=%"PRIu64"\n", sum.get_fail_objs);
451 printf(" no statistics available\n");
454 rte_mempool_audit(mp);
457 /* dump the status of all mempools on the console */
459 rte_mempool_list_dump(void)
461 const struct rte_mempool *mp = NULL;
463 TAILQ_FOREACH(mp, mempool_list, next) {
464 rte_mempool_dump(mp);
468 /* search a mempool from its name */
470 rte_mempool_lookup(const char *name)
472 struct rte_mempool *mp = NULL;
474 /* check that we have an initialised tail queue */
475 if (mempool_list == NULL)
476 if ((mempool_list = RTE_TAILQ_RESERVE("RTE_MEMPOOL", \
477 rte_mempool_list)) == NULL){
478 rte_errno = E_RTE_NO_TAILQ;
482 TAILQ_FOREACH(mp, mempool_list, next) {
483 if (strncmp(name, mp->name, RTE_MEMPOOL_NAMESIZE) == 0)