4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5 * Copyright(c) 2016 6WIND S.A.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * * Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * * Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in
16 * the documentation and/or other materials provided with the
18 * * Neither the name of Intel Corporation nor the names of its
19 * contributors may be used to endorse or promote products derived
20 * from this software without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
42 #include <sys/queue.h>
44 #include <rte_common.h>
46 #include <rte_debug.h>
47 #include <rte_memory.h>
48 #include <rte_memzone.h>
49 #include <rte_malloc.h>
50 #include <rte_atomic.h>
51 #include <rte_launch.h>
53 #include <rte_eal_memconfig.h>
54 #include <rte_per_lcore.h>
55 #include <rte_lcore.h>
56 #include <rte_branch_prediction.h>
58 #include <rte_errno.h>
59 #include <rte_string_fns.h>
60 #include <rte_spinlock.h>
62 #include "rte_mempool.h"
64 TAILQ_HEAD(rte_mempool_list, rte_tailq_entry);
66 static struct rte_tailq_elem rte_mempool_tailq = {
67 .name = "RTE_MEMPOOL",
69 EAL_REGISTER_TAILQ(rte_mempool_tailq)
71 #define CACHE_FLUSHTHRESH_MULTIPLIER 1.5
72 #define CALC_CACHE_FLUSHTHRESH(c) \
73 ((typeof(c))((c) * CACHE_FLUSHTHRESH_MULTIPLIER))
76 * return the greatest common divisor between a and b (fast algorithm)
79 static unsigned get_gcd(unsigned a, unsigned b)
104 * Depending on memory configuration, objects addresses are spread
105 * between channels and ranks in RAM: the pool allocator will add
106 * padding between objects. This function return the new size of the
109 static unsigned optimize_object_size(unsigned obj_size)
111 unsigned nrank, nchan;
112 unsigned new_obj_size;
114 /* get number of channels */
115 nchan = rte_memory_get_nchannel();
119 nrank = rte_memory_get_nrank();
123 /* process new object size */
124 new_obj_size = (obj_size + RTE_MEMPOOL_ALIGN_MASK) / RTE_MEMPOOL_ALIGN;
125 while (get_gcd(new_obj_size, nrank * nchan) != 1)
127 return new_obj_size * RTE_MEMPOOL_ALIGN;
131 mempool_add_elem(struct rte_mempool *mp, void *obj, phys_addr_t physaddr)
133 struct rte_mempool_objhdr *hdr;
134 struct rte_mempool_objtlr *tlr __rte_unused;
136 /* set mempool ptr in header */
137 hdr = RTE_PTR_SUB(obj, sizeof(*hdr));
139 hdr->physaddr = physaddr;
140 STAILQ_INSERT_TAIL(&mp->elt_list, hdr, next);
141 mp->populated_size++;
143 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
144 hdr->cookie = RTE_MEMPOOL_HEADER_COOKIE2;
145 tlr = __mempool_get_trailer(obj);
146 tlr->cookie = RTE_MEMPOOL_TRAILER_COOKIE;
149 /* enqueue in ring */
150 rte_ring_sp_enqueue(mp->ring, obj);
153 /* call obj_cb() for each mempool element */
155 rte_mempool_obj_iter(struct rte_mempool *mp,
156 rte_mempool_obj_cb_t *obj_cb, void *obj_cb_arg)
158 struct rte_mempool_objhdr *hdr;
162 STAILQ_FOREACH(hdr, &mp->elt_list, next) {
163 obj = (char *)hdr + sizeof(*hdr);
164 obj_cb(mp, obj_cb_arg, obj, n);
171 /* call mem_cb() for each mempool memory chunk */
173 rte_mempool_mem_iter(struct rte_mempool *mp,
174 rte_mempool_mem_cb_t *mem_cb, void *mem_cb_arg)
176 struct rte_mempool_memhdr *hdr;
179 STAILQ_FOREACH(hdr, &mp->mem_list, next) {
180 mem_cb(mp, mem_cb_arg, hdr, n);
187 /* get the header, trailer and total size of a mempool element. */
189 rte_mempool_calc_obj_size(uint32_t elt_size, uint32_t flags,
190 struct rte_mempool_objsz *sz)
192 struct rte_mempool_objsz lsz;
194 sz = (sz != NULL) ? sz : &lsz;
196 sz->header_size = sizeof(struct rte_mempool_objhdr);
197 if ((flags & MEMPOOL_F_NO_CACHE_ALIGN) == 0)
198 sz->header_size = RTE_ALIGN_CEIL(sz->header_size,
201 sz->trailer_size = sizeof(struct rte_mempool_objtlr);
203 /* element size is 8 bytes-aligned at least */
204 sz->elt_size = RTE_ALIGN_CEIL(elt_size, sizeof(uint64_t));
206 /* expand trailer to next cache line */
207 if ((flags & MEMPOOL_F_NO_CACHE_ALIGN) == 0) {
208 sz->total_size = sz->header_size + sz->elt_size +
210 sz->trailer_size += ((RTE_MEMPOOL_ALIGN -
211 (sz->total_size & RTE_MEMPOOL_ALIGN_MASK)) &
212 RTE_MEMPOOL_ALIGN_MASK);
216 * increase trailer to add padding between objects in order to
217 * spread them across memory channels/ranks
219 if ((flags & MEMPOOL_F_NO_SPREAD) == 0) {
221 new_size = optimize_object_size(sz->header_size + sz->elt_size +
223 sz->trailer_size = new_size - sz->header_size - sz->elt_size;
226 /* this is the size of an object, including header and trailer */
227 sz->total_size = sz->header_size + sz->elt_size + sz->trailer_size;
229 return sz->total_size;
234 * Calculate maximum amount of memory required to store given number of objects.
237 rte_mempool_xmem_size(uint32_t elt_num, size_t total_elt_sz, uint32_t pg_shift)
239 size_t obj_per_page, pg_num, pg_sz;
242 return total_elt_sz * elt_num;
244 pg_sz = (size_t)1 << pg_shift;
245 obj_per_page = pg_sz / total_elt_sz;
246 if (obj_per_page == 0)
247 return RTE_ALIGN_CEIL(total_elt_sz, pg_sz) * elt_num;
249 pg_num = (elt_num + obj_per_page - 1) / obj_per_page;
250 return pg_num << pg_shift;
254 * Calculate how much memory would be actually required with the
255 * given memory footprint to store required number of elements.
258 rte_mempool_xmem_usage(__rte_unused void *vaddr, uint32_t elt_num,
259 size_t total_elt_sz, const phys_addr_t paddr[], uint32_t pg_num,
262 uint32_t elt_cnt = 0;
263 phys_addr_t start, end;
265 size_t pg_sz = (size_t)1 << pg_shift;
267 /* if paddr is NULL, assume contiguous memory */
270 end = pg_sz * pg_num;
274 end = paddr[0] + pg_sz;
277 while (elt_cnt < elt_num) {
279 if (end - start >= total_elt_sz) {
280 /* enough contiguous memory, add an object */
281 start += total_elt_sz;
283 } else if (paddr_idx < pg_num) {
284 /* no room to store one obj, add a page */
285 if (end == paddr[paddr_idx]) {
288 start = paddr[paddr_idx];
289 end = paddr[paddr_idx] + pg_sz;
294 /* no more page, return how many elements fit */
295 return -(size_t)elt_cnt;
299 return (size_t)paddr_idx << pg_shift;
302 #ifndef RTE_LIBRTE_XEN_DOM0
303 /* stub if DOM0 support not configured */
305 rte_dom0_mempool_create(const char *name __rte_unused,
306 unsigned n __rte_unused,
307 unsigned elt_size __rte_unused,
308 unsigned cache_size __rte_unused,
309 unsigned private_data_size __rte_unused,
310 rte_mempool_ctor_t *mp_init __rte_unused,
311 void *mp_init_arg __rte_unused,
312 rte_mempool_obj_ctor_t *obj_init __rte_unused,
313 void *obj_init_arg __rte_unused,
314 int socket_id __rte_unused,
315 unsigned flags __rte_unused)
322 /* create the internal ring */
324 rte_mempool_ring_create(struct rte_mempool *mp)
327 char rg_name[RTE_RING_NAMESIZE];
330 snprintf(rg_name, sizeof(rg_name), RTE_MEMPOOL_MZ_FORMAT, mp->name);
333 if (mp->flags & MEMPOOL_F_SP_PUT)
334 rg_flags |= RING_F_SP_ENQ;
335 if (mp->flags & MEMPOOL_F_SC_GET)
336 rg_flags |= RING_F_SC_DEQ;
338 /* Allocate the ring that will be used to store objects.
339 * Ring functions will return appropriate errors if we are
340 * running as a secondary process etc., so no checks made
341 * in this function for that condition.
343 r = rte_ring_create(rg_name, rte_align32pow2(mp->size + 1),
344 mp->socket_id, rg_flags);
352 /* free a memchunk allocated with rte_memzone_reserve() */
354 rte_mempool_memchunk_mz_free(__rte_unused struct rte_mempool_memhdr *memhdr,
357 const struct rte_memzone *mz = opaque;
358 rte_memzone_free(mz);
361 /* Free memory chunks used by a mempool. Objects must be in pool */
363 rte_mempool_free_memchunks(struct rte_mempool *mp)
365 struct rte_mempool_memhdr *memhdr;
368 while (!STAILQ_EMPTY(&mp->elt_list)) {
369 rte_ring_sc_dequeue(mp->ring, &elt);
371 STAILQ_REMOVE_HEAD(&mp->elt_list, next);
372 mp->populated_size--;
375 while (!STAILQ_EMPTY(&mp->mem_list)) {
376 memhdr = STAILQ_FIRST(&mp->mem_list);
377 STAILQ_REMOVE_HEAD(&mp->mem_list, next);
378 if (memhdr->free_cb != NULL)
379 memhdr->free_cb(memhdr, memhdr->opaque);
385 /* Add objects in the pool, using a physically contiguous memory
386 * zone. Return the number of objects added, or a negative value
390 rte_mempool_populate_phys(struct rte_mempool *mp, char *vaddr,
391 phys_addr_t paddr, size_t len, rte_mempool_memchunk_free_cb_t *free_cb,
394 unsigned total_elt_sz;
397 struct rte_mempool_memhdr *memhdr;
399 /* mempool is already populated */
400 if (mp->populated_size >= mp->size)
403 total_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;
405 memhdr = rte_zmalloc("MEMPOOL_MEMHDR", sizeof(*memhdr), 0);
410 memhdr->addr = vaddr;
411 memhdr->phys_addr = paddr;
413 memhdr->free_cb = free_cb;
414 memhdr->opaque = opaque;
416 if (mp->flags & MEMPOOL_F_NO_CACHE_ALIGN)
417 off = RTE_PTR_ALIGN_CEIL(vaddr, 8) - vaddr;
419 off = RTE_PTR_ALIGN_CEIL(vaddr, RTE_CACHE_LINE_SIZE) - vaddr;
421 while (off + total_elt_sz <= len && mp->populated_size < mp->size) {
422 off += mp->header_size;
423 mempool_add_elem(mp, (char *)vaddr + off, paddr + off);
424 off += mp->elt_size + mp->trailer_size;
428 /* not enough room to store one object */
432 STAILQ_INSERT_TAIL(&mp->mem_list, memhdr, next);
437 /* Add objects in the pool, using a table of physical pages. Return the
438 * number of objects added, or a negative value on error.
441 rte_mempool_populate_phys_tab(struct rte_mempool *mp, char *vaddr,
442 const phys_addr_t paddr[], uint32_t pg_num, uint32_t pg_shift,
443 rte_mempool_memchunk_free_cb_t *free_cb, void *opaque)
447 size_t pg_sz = (size_t)1 << pg_shift;
449 /* mempool must not be populated */
450 if (mp->nb_mem_chunks != 0)
453 for (i = 0; i < pg_num && mp->populated_size < mp->size; i += n) {
455 /* populate with the largest group of contiguous pages */
456 for (n = 1; (i + n) < pg_num &&
457 paddr[i] + pg_sz == paddr[i+n]; n++)
460 ret = rte_mempool_populate_phys(mp, vaddr + i * pg_sz,
461 paddr[i], n * pg_sz, free_cb, opaque);
463 rte_mempool_free_memchunks(mp);
466 /* no need to call the free callback for next chunks */
473 /* Populate the mempool with a virtual area. Return the number of
474 * objects added, or a negative value on error.
477 rte_mempool_populate_virt(struct rte_mempool *mp, char *addr,
478 size_t len, size_t pg_sz, rte_mempool_memchunk_free_cb_t *free_cb,
482 size_t off, phys_len;
485 /* mempool must not be populated */
486 if (mp->nb_mem_chunks != 0)
488 /* address and len must be page-aligned */
489 if (RTE_PTR_ALIGN_CEIL(addr, pg_sz) != addr)
491 if (RTE_ALIGN_CEIL(len, pg_sz) != len)
494 for (off = 0; off + pg_sz <= len &&
495 mp->populated_size < mp->size; off += phys_len) {
497 paddr = rte_mem_virt2phy(addr + off);
498 if (paddr == RTE_BAD_PHYS_ADDR) {
503 /* populate with the largest group of contiguous pages */
504 for (phys_len = pg_sz; off + phys_len < len; phys_len += pg_sz) {
505 phys_addr_t paddr_tmp;
507 paddr_tmp = rte_mem_virt2phy(addr + off + phys_len);
508 paddr_tmp = rte_mem_phy2mch(-1, paddr_tmp);
510 if (paddr_tmp != paddr + phys_len)
514 ret = rte_mempool_populate_phys(mp, addr + off, paddr,
515 phys_len, free_cb, opaque);
518 /* no need to call the free callback for next chunks */
526 rte_mempool_free_memchunks(mp);
530 /* Default function to populate the mempool: allocate memory in memzones,
531 * and populate them. Return the number of objects added, or a negative
535 rte_mempool_populate_default(struct rte_mempool *mp)
537 int mz_flags = RTE_MEMZONE_1GB|RTE_MEMZONE_SIZE_HINT_ONLY;
538 char mz_name[RTE_MEMZONE_NAMESIZE];
539 const struct rte_memzone *mz;
540 size_t size, total_elt_sz, align, pg_sz, pg_shift;
544 /* mempool must not be populated */
545 if (mp->nb_mem_chunks != 0)
548 if (rte_eal_has_hugepages()) {
549 pg_shift = 0; /* not needed, zone is physically contiguous */
551 align = RTE_CACHE_LINE_SIZE;
553 pg_sz = getpagesize();
554 pg_shift = rte_bsf32(pg_sz);
558 total_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;
559 for (mz_id = 0, n = mp->size; n > 0; mz_id++, n -= ret) {
560 size = rte_mempool_xmem_size(n, total_elt_sz, pg_shift);
562 ret = snprintf(mz_name, sizeof(mz_name),
563 RTE_MEMPOOL_MZ_FORMAT "_%d", mp->name, mz_id);
564 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
569 mz = rte_memzone_reserve_aligned(mz_name, size,
570 mp->socket_id, mz_flags, align);
571 /* not enough memory, retry with the biggest zone we have */
573 mz = rte_memzone_reserve_aligned(mz_name, 0,
574 mp->socket_id, mz_flags, align);
580 if (rte_eal_has_hugepages())
581 ret = rte_mempool_populate_phys(mp, mz->addr,
582 mz->phys_addr, mz->len,
583 rte_mempool_memchunk_mz_free,
584 (void *)(uintptr_t)mz);
586 ret = rte_mempool_populate_virt(mp, mz->addr,
588 rte_mempool_memchunk_mz_free,
589 (void *)(uintptr_t)mz);
597 rte_mempool_free_memchunks(mp);
603 rte_mempool_free(struct rte_mempool *mp)
605 struct rte_mempool_list *mempool_list = NULL;
606 struct rte_tailq_entry *te;
611 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
612 rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
613 /* find out tailq entry */
614 TAILQ_FOREACH(te, mempool_list, next) {
615 if (te->data == (void *)mp)
620 TAILQ_REMOVE(mempool_list, te, next);
623 rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
625 rte_mempool_free_memchunks(mp);
626 rte_ring_free(mp->ring);
627 rte_memzone_free(mp->mz);
630 /* create an empty mempool */
631 static struct rte_mempool *
632 rte_mempool_create_empty(const char *name, unsigned n, unsigned elt_size,
633 unsigned cache_size, unsigned private_data_size,
634 int socket_id, unsigned flags)
636 char mz_name[RTE_MEMZONE_NAMESIZE];
637 struct rte_mempool_list *mempool_list;
638 struct rte_mempool *mp = NULL;
639 struct rte_tailq_entry *te = NULL;
640 const struct rte_memzone *mz = NULL;
642 int mz_flags = RTE_MEMZONE_1GB|RTE_MEMZONE_SIZE_HINT_ONLY;
643 struct rte_mempool_objsz objsz;
645 /* compilation-time checks */
646 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool) &
647 RTE_CACHE_LINE_MASK) != 0);
648 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_cache) &
649 RTE_CACHE_LINE_MASK) != 0);
650 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
651 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_debug_stats) &
652 RTE_CACHE_LINE_MASK) != 0);
653 RTE_BUILD_BUG_ON((offsetof(struct rte_mempool, stats) &
654 RTE_CACHE_LINE_MASK) != 0);
657 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
659 /* asked cache too big */
660 if (cache_size > RTE_MEMPOOL_CACHE_MAX_SIZE ||
661 CALC_CACHE_FLUSHTHRESH(cache_size) > n) {
666 /* "no cache align" imply "no spread" */
667 if (flags & MEMPOOL_F_NO_CACHE_ALIGN)
668 flags |= MEMPOOL_F_NO_SPREAD;
670 /* calculate mempool object sizes. */
671 if (!rte_mempool_calc_obj_size(elt_size, flags, &objsz)) {
676 rte_rwlock_write_lock(RTE_EAL_MEMPOOL_RWLOCK);
679 * reserve a memory zone for this mempool: private data is
682 private_data_size = (private_data_size +
683 RTE_MEMPOOL_ALIGN_MASK) & (~RTE_MEMPOOL_ALIGN_MASK);
686 /* try to allocate tailq entry */
687 te = rte_zmalloc("MEMPOOL_TAILQ_ENTRY", sizeof(*te), 0);
689 RTE_LOG(ERR, MEMPOOL, "Cannot allocate tailq entry!\n");
693 mempool_size = MEMPOOL_HEADER_SIZE(mp, cache_size);
694 mempool_size += private_data_size;
695 mempool_size = RTE_ALIGN_CEIL(mempool_size, RTE_MEMPOOL_ALIGN);
697 snprintf(mz_name, sizeof(mz_name), RTE_MEMPOOL_MZ_FORMAT, name);
699 mz = rte_memzone_reserve(mz_name, mempool_size, socket_id, mz_flags);
703 /* init the mempool structure */
705 memset(mp, 0, sizeof(*mp));
706 snprintf(mp->name, sizeof(mp->name), "%s", name);
708 mp->socket_id = socket_id;
711 mp->socket_id = socket_id;
712 mp->elt_size = objsz.elt_size;
713 mp->header_size = objsz.header_size;
714 mp->trailer_size = objsz.trailer_size;
715 mp->cache_size = cache_size;
716 mp->cache_flushthresh = CALC_CACHE_FLUSHTHRESH(cache_size);
717 mp->private_data_size = private_data_size;
718 STAILQ_INIT(&mp->elt_list);
719 STAILQ_INIT(&mp->mem_list);
721 if (rte_mempool_ring_create(mp) < 0)
725 * local_cache pointer is set even if cache_size is zero.
726 * The local_cache points to just past the elt_pa[] array.
728 mp->local_cache = (struct rte_mempool_cache *)
729 RTE_PTR_ADD(mp, MEMPOOL_HEADER_SIZE(mp, 0));
732 rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
733 TAILQ_INSERT_TAIL(mempool_list, te, next);
734 rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
735 rte_rwlock_write_unlock(RTE_EAL_MEMPOOL_RWLOCK);
740 rte_rwlock_write_unlock(RTE_EAL_MEMPOOL_RWLOCK);
742 rte_mempool_free(mp);
746 /* create the mempool */
748 rte_mempool_create(const char *name, unsigned n, unsigned elt_size,
749 unsigned cache_size, unsigned private_data_size,
750 rte_mempool_ctor_t *mp_init, void *mp_init_arg,
751 rte_mempool_obj_cb_t *obj_init, void *obj_init_arg,
752 int socket_id, unsigned flags)
754 struct rte_mempool *mp;
756 if (rte_xen_dom0_supported())
757 return rte_dom0_mempool_create(name, n, elt_size,
758 cache_size, private_data_size,
759 mp_init, mp_init_arg,
760 obj_init, obj_init_arg,
763 mp = rte_mempool_create_empty(name, n, elt_size, cache_size,
764 private_data_size, socket_id, flags);
768 /* call the mempool priv initializer */
770 mp_init(mp, mp_init_arg);
772 if (rte_mempool_populate_default(mp) < 0)
775 /* call the object initializers */
777 rte_mempool_obj_iter(mp, obj_init, obj_init_arg);
782 rte_mempool_free(mp);
787 * Create the mempool over already allocated chunk of memory.
788 * That external memory buffer can consists of physically disjoint pages.
789 * Setting vaddr to NULL, makes mempool to fallback to original behaviour
790 * and allocate space for mempool and it's elements as one big chunk of
791 * physically continuos memory.
794 rte_mempool_xmem_create(const char *name, unsigned n, unsigned elt_size,
795 unsigned cache_size, unsigned private_data_size,
796 rte_mempool_ctor_t *mp_init, void *mp_init_arg,
797 rte_mempool_obj_cb_t *obj_init, void *obj_init_arg,
798 int socket_id, unsigned flags, void *vaddr,
799 const phys_addr_t paddr[], uint32_t pg_num, uint32_t pg_shift)
801 struct rte_mempool *mp = NULL;
804 /* no virtual address supplied, use rte_mempool_create() */
806 return rte_mempool_create(name, n, elt_size, cache_size,
807 private_data_size, mp_init, mp_init_arg,
808 obj_init, obj_init_arg, socket_id, flags);
810 /* check that we have both VA and PA */
816 /* Check that pg_shift parameter is valid. */
817 if (pg_shift > MEMPOOL_PG_SHIFT_MAX) {
822 mp = rte_mempool_create_empty(name, n, elt_size, cache_size,
823 private_data_size, socket_id, flags);
827 /* call the mempool priv initializer */
829 mp_init(mp, mp_init_arg);
831 ret = rte_mempool_populate_phys_tab(mp, vaddr, paddr, pg_num, pg_shift,
833 if (ret < 0 || ret != (int)mp->size)
836 /* call the object initializers */
838 rte_mempool_obj_iter(mp, obj_init, obj_init_arg);
843 rte_mempool_free(mp);
847 /* Return the number of entries in the mempool */
849 rte_mempool_count(const struct rte_mempool *mp)
854 count = rte_ring_count(mp->ring);
856 if (mp->cache_size == 0)
859 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
860 count += mp->local_cache[lcore_id].len;
863 * due to race condition (access to len is not locked), the
864 * total can be greater than size... so fix the result
866 if (count > mp->size)
871 /* dump the cache status */
873 rte_mempool_dump_cache(FILE *f, const struct rte_mempool *mp)
877 unsigned cache_count;
879 fprintf(f, " cache infos:\n");
880 fprintf(f, " cache_size=%"PRIu32"\n", mp->cache_size);
882 if (mp->cache_size == 0)
885 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
886 cache_count = mp->local_cache[lcore_id].len;
887 fprintf(f, " cache_count[%u]=%u\n", lcore_id, cache_count);
888 count += cache_count;
890 fprintf(f, " total_cache_count=%u\n", count);
894 #ifndef __INTEL_COMPILER
895 #pragma GCC diagnostic ignored "-Wcast-qual"
898 /* check and update cookies or panic (internal) */
899 void rte_mempool_check_cookies(const struct rte_mempool *mp,
900 void * const *obj_table_const, unsigned n, int free)
902 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
903 struct rte_mempool_objhdr *hdr;
904 struct rte_mempool_objtlr *tlr;
910 /* Force to drop the "const" attribute. This is done only when
911 * DEBUG is enabled */
912 tmp = (void *) obj_table_const;
913 obj_table = (void **) tmp;
918 if (rte_mempool_from_obj(obj) != mp)
919 rte_panic("MEMPOOL: object is owned by another "
922 hdr = __mempool_get_header(obj);
923 cookie = hdr->cookie;
926 if (cookie != RTE_MEMPOOL_HEADER_COOKIE1) {
927 rte_log_set_history(0);
928 RTE_LOG(CRIT, MEMPOOL,
929 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
930 obj, (const void *) mp, cookie);
931 rte_panic("MEMPOOL: bad header cookie (put)\n");
933 hdr->cookie = RTE_MEMPOOL_HEADER_COOKIE2;
934 } else if (free == 1) {
935 if (cookie != RTE_MEMPOOL_HEADER_COOKIE2) {
936 rte_log_set_history(0);
937 RTE_LOG(CRIT, MEMPOOL,
938 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
939 obj, (const void *) mp, cookie);
940 rte_panic("MEMPOOL: bad header cookie (get)\n");
942 hdr->cookie = RTE_MEMPOOL_HEADER_COOKIE1;
943 } else if (free == 2) {
944 if (cookie != RTE_MEMPOOL_HEADER_COOKIE1 &&
945 cookie != RTE_MEMPOOL_HEADER_COOKIE2) {
946 rte_log_set_history(0);
947 RTE_LOG(CRIT, MEMPOOL,
948 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
949 obj, (const void *) mp, cookie);
950 rte_panic("MEMPOOL: bad header cookie (audit)\n");
953 tlr = __mempool_get_trailer(obj);
954 cookie = tlr->cookie;
955 if (cookie != RTE_MEMPOOL_TRAILER_COOKIE) {
956 rte_log_set_history(0);
957 RTE_LOG(CRIT, MEMPOOL,
958 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
959 obj, (const void *) mp, cookie);
960 rte_panic("MEMPOOL: bad trailer cookie\n");
965 RTE_SET_USED(obj_table_const);
971 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
973 mempool_obj_audit(struct rte_mempool *mp, __rte_unused void *opaque,
974 void *obj, __rte_unused unsigned idx)
976 __mempool_check_cookies(mp, &obj, 1, 2);
980 mempool_audit_cookies(struct rte_mempool *mp)
984 num = rte_mempool_obj_iter(mp, mempool_obj_audit, NULL);
985 if (num != mp->size) {
986 rte_panic("rte_mempool_obj_iter(mempool=%p, size=%u) "
987 "iterated only over %u elements\n",
992 #define mempool_audit_cookies(mp) do {} while(0)
995 #ifndef __INTEL_COMPILER
996 #pragma GCC diagnostic error "-Wcast-qual"
999 /* check cookies before and after objects */
1001 mempool_audit_cache(const struct rte_mempool *mp)
1003 /* check cache size consistency */
1006 if (mp->cache_size == 0)
1009 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1010 if (mp->local_cache[lcore_id].len > mp->cache_flushthresh) {
1011 RTE_LOG(CRIT, MEMPOOL, "badness on cache[%u]\n",
1013 rte_panic("MEMPOOL: invalid cache len\n");
1018 /* check the consistency of mempool (size, cookies, ...) */
1020 rte_mempool_audit(struct rte_mempool *mp)
1022 mempool_audit_cache(mp);
1023 mempool_audit_cookies(mp);
1025 /* For case where mempool DEBUG is not set, and cache size is 0 */
1029 /* dump the status of the mempool on the console */
1031 rte_mempool_dump(FILE *f, struct rte_mempool *mp)
1033 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1034 struct rte_mempool_debug_stats sum;
1037 struct rte_mempool_memhdr *memhdr;
1038 unsigned common_count;
1039 unsigned cache_count;
1042 RTE_ASSERT(f != NULL);
1043 RTE_ASSERT(mp != NULL);
1045 fprintf(f, "mempool <%s>@%p\n", mp->name, mp);
1046 fprintf(f, " flags=%x\n", mp->flags);
1047 fprintf(f, " ring=<%s>@%p\n", mp->ring->name, mp->ring);
1048 fprintf(f, " phys_addr=0x%" PRIx64 "\n", mp->mz->phys_addr);
1049 fprintf(f, " nb_mem_chunks=%u\n", mp->nb_mem_chunks);
1050 fprintf(f, " size=%"PRIu32"\n", mp->size);
1051 fprintf(f, " populated_size=%"PRIu32"\n", mp->populated_size);
1052 fprintf(f, " header_size=%"PRIu32"\n", mp->header_size);
1053 fprintf(f, " elt_size=%"PRIu32"\n", mp->elt_size);
1054 fprintf(f, " trailer_size=%"PRIu32"\n", mp->trailer_size);
1055 fprintf(f, " total_obj_size=%"PRIu32"\n",
1056 mp->header_size + mp->elt_size + mp->trailer_size);
1058 fprintf(f, " private_data_size=%"PRIu32"\n", mp->private_data_size);
1060 STAILQ_FOREACH(memhdr, &mp->mem_list, next)
1061 mem_len += memhdr->len;
1063 fprintf(f, " avg bytes/object=%#Lf\n",
1064 (long double)mem_len / mp->size);
1067 cache_count = rte_mempool_dump_cache(f, mp);
1068 common_count = rte_ring_count(mp->ring);
1069 if ((cache_count + common_count) > mp->size)
1070 common_count = mp->size - cache_count;
1071 fprintf(f, " common_pool_count=%u\n", common_count);
1073 /* sum and dump statistics */
1074 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1075 memset(&sum, 0, sizeof(sum));
1076 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1077 sum.put_bulk += mp->stats[lcore_id].put_bulk;
1078 sum.put_objs += mp->stats[lcore_id].put_objs;
1079 sum.get_success_bulk += mp->stats[lcore_id].get_success_bulk;
1080 sum.get_success_objs += mp->stats[lcore_id].get_success_objs;
1081 sum.get_fail_bulk += mp->stats[lcore_id].get_fail_bulk;
1082 sum.get_fail_objs += mp->stats[lcore_id].get_fail_objs;
1084 fprintf(f, " stats:\n");
1085 fprintf(f, " put_bulk=%"PRIu64"\n", sum.put_bulk);
1086 fprintf(f, " put_objs=%"PRIu64"\n", sum.put_objs);
1087 fprintf(f, " get_success_bulk=%"PRIu64"\n", sum.get_success_bulk);
1088 fprintf(f, " get_success_objs=%"PRIu64"\n", sum.get_success_objs);
1089 fprintf(f, " get_fail_bulk=%"PRIu64"\n", sum.get_fail_bulk);
1090 fprintf(f, " get_fail_objs=%"PRIu64"\n", sum.get_fail_objs);
1092 fprintf(f, " no statistics available\n");
1095 rte_mempool_audit(mp);
1098 /* dump the status of all mempools on the console */
1100 rte_mempool_list_dump(FILE *f)
1102 struct rte_mempool *mp = NULL;
1103 struct rte_tailq_entry *te;
1104 struct rte_mempool_list *mempool_list;
1106 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1108 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1110 TAILQ_FOREACH(te, mempool_list, next) {
1111 mp = (struct rte_mempool *) te->data;
1112 rte_mempool_dump(f, mp);
1115 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);
1118 /* search a mempool from its name */
1119 struct rte_mempool *
1120 rte_mempool_lookup(const char *name)
1122 struct rte_mempool *mp = NULL;
1123 struct rte_tailq_entry *te;
1124 struct rte_mempool_list *mempool_list;
1126 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1128 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1130 TAILQ_FOREACH(te, mempool_list, next) {
1131 mp = (struct rte_mempool *) te->data;
1132 if (strncmp(name, mp->name, RTE_MEMPOOL_NAMESIZE) == 0)
1136 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);
1146 void rte_mempool_walk(void (*func)(struct rte_mempool *, void *),
1149 struct rte_tailq_entry *te = NULL;
1150 struct rte_mempool_list *mempool_list;
1152 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1154 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1156 TAILQ_FOREACH(te, mempool_list, next) {
1157 (*func)((struct rte_mempool *) te->data, arg);
1160 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);