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>
45 #include <rte_common.h>
47 #include <rte_debug.h>
48 #include <rte_memory.h>
49 #include <rte_memzone.h>
50 #include <rte_malloc.h>
51 #include <rte_atomic.h>
52 #include <rte_launch.h>
54 #include <rte_eal_memconfig.h>
55 #include <rte_per_lcore.h>
56 #include <rte_lcore.h>
57 #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_mempool_ops_enqueue_bulk(mp, &obj, 1);
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 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
202 sz->trailer_size = sizeof(struct rte_mempool_objtlr);
204 sz->trailer_size = 0;
207 /* element size is 8 bytes-aligned at least */
208 sz->elt_size = RTE_ALIGN_CEIL(elt_size, sizeof(uint64_t));
210 /* expand trailer to next cache line */
211 if ((flags & MEMPOOL_F_NO_CACHE_ALIGN) == 0) {
212 sz->total_size = sz->header_size + sz->elt_size +
214 sz->trailer_size += ((RTE_MEMPOOL_ALIGN -
215 (sz->total_size & RTE_MEMPOOL_ALIGN_MASK)) &
216 RTE_MEMPOOL_ALIGN_MASK);
220 * increase trailer to add padding between objects in order to
221 * spread them across memory channels/ranks
223 if ((flags & MEMPOOL_F_NO_SPREAD) == 0) {
225 new_size = optimize_object_size(sz->header_size + sz->elt_size +
227 sz->trailer_size = new_size - sz->header_size - sz->elt_size;
230 /* this is the size of an object, including header and trailer */
231 sz->total_size = sz->header_size + sz->elt_size + sz->trailer_size;
233 return sz->total_size;
238 * Calculate maximum amount of memory required to store given number of objects.
241 rte_mempool_xmem_size(uint32_t elt_num, size_t total_elt_sz, uint32_t pg_shift,
242 __rte_unused unsigned int flags)
244 size_t obj_per_page, pg_num, pg_sz;
246 if (total_elt_sz == 0)
250 return total_elt_sz * elt_num;
252 pg_sz = (size_t)1 << pg_shift;
253 obj_per_page = pg_sz / total_elt_sz;
254 if (obj_per_page == 0)
255 return RTE_ALIGN_CEIL(total_elt_sz, pg_sz) * elt_num;
257 pg_num = (elt_num + obj_per_page - 1) / obj_per_page;
258 return pg_num << pg_shift;
262 * Calculate how much memory would be actually required with the
263 * given memory footprint to store required number of elements.
266 rte_mempool_xmem_usage(__rte_unused void *vaddr, uint32_t elt_num,
267 size_t total_elt_sz, const phys_addr_t paddr[], uint32_t pg_num,
268 uint32_t pg_shift, __rte_unused unsigned int flags)
270 uint32_t elt_cnt = 0;
271 phys_addr_t start, end;
273 size_t pg_sz = (size_t)1 << pg_shift;
275 /* if paddr is NULL, assume contiguous memory */
278 end = pg_sz * pg_num;
282 end = paddr[0] + pg_sz;
285 while (elt_cnt < elt_num) {
287 if (end - start >= total_elt_sz) {
288 /* enough contiguous memory, add an object */
289 start += total_elt_sz;
291 } else if (paddr_idx < pg_num) {
292 /* no room to store one obj, add a page */
293 if (end == paddr[paddr_idx]) {
296 start = paddr[paddr_idx];
297 end = paddr[paddr_idx] + pg_sz;
302 /* no more page, return how many elements fit */
303 return -(size_t)elt_cnt;
307 return (size_t)paddr_idx << pg_shift;
310 /* free a memchunk allocated with rte_memzone_reserve() */
312 rte_mempool_memchunk_mz_free(__rte_unused struct rte_mempool_memhdr *memhdr,
315 const struct rte_memzone *mz = opaque;
316 rte_memzone_free(mz);
319 /* Free memory chunks used by a mempool. Objects must be in pool */
321 rte_mempool_free_memchunks(struct rte_mempool *mp)
323 struct rte_mempool_memhdr *memhdr;
326 while (!STAILQ_EMPTY(&mp->elt_list)) {
327 rte_mempool_ops_dequeue_bulk(mp, &elt, 1);
329 STAILQ_REMOVE_HEAD(&mp->elt_list, next);
330 mp->populated_size--;
333 while (!STAILQ_EMPTY(&mp->mem_list)) {
334 memhdr = STAILQ_FIRST(&mp->mem_list);
335 STAILQ_REMOVE_HEAD(&mp->mem_list, next);
336 if (memhdr->free_cb != NULL)
337 memhdr->free_cb(memhdr, memhdr->opaque);
343 /* Add objects in the pool, using a physically contiguous memory
344 * zone. Return the number of objects added, or a negative value
348 rte_mempool_populate_phys(struct rte_mempool *mp, char *vaddr,
349 phys_addr_t paddr, size_t len, rte_mempool_memchunk_free_cb_t *free_cb,
352 unsigned total_elt_sz;
355 struct rte_mempool_memhdr *memhdr;
358 /* create the internal ring if not already done */
359 if ((mp->flags & MEMPOOL_F_POOL_CREATED) == 0) {
360 ret = rte_mempool_ops_alloc(mp);
363 mp->flags |= MEMPOOL_F_POOL_CREATED;
366 /* mempool is already populated */
367 if (mp->populated_size >= mp->size)
370 total_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;
372 /* Detect pool area has sufficient space for elements */
373 if (mp->flags & MEMPOOL_F_CAPA_PHYS_CONTIG) {
374 if (len < total_elt_sz * mp->size) {
375 RTE_LOG(ERR, MEMPOOL,
376 "pool area %" PRIx64 " not enough\n",
382 memhdr = rte_zmalloc("MEMPOOL_MEMHDR", sizeof(*memhdr), 0);
387 memhdr->addr = vaddr;
388 memhdr->phys_addr = paddr;
390 memhdr->free_cb = free_cb;
391 memhdr->opaque = opaque;
393 if (mp->flags & MEMPOOL_F_NO_CACHE_ALIGN)
394 off = RTE_PTR_ALIGN_CEIL(vaddr, 8) - vaddr;
396 off = RTE_PTR_ALIGN_CEIL(vaddr, RTE_CACHE_LINE_SIZE) - vaddr;
398 while (off + total_elt_sz <= len && mp->populated_size < mp->size) {
399 off += mp->header_size;
400 if (paddr == RTE_BAD_PHYS_ADDR)
401 mempool_add_elem(mp, (char *)vaddr + off,
404 mempool_add_elem(mp, (char *)vaddr + off, paddr + off);
405 off += mp->elt_size + mp->trailer_size;
409 /* not enough room to store one object */
413 STAILQ_INSERT_TAIL(&mp->mem_list, memhdr, next);
418 /* Add objects in the pool, using a table of physical pages. Return the
419 * number of objects added, or a negative value on error.
422 rte_mempool_populate_phys_tab(struct rte_mempool *mp, char *vaddr,
423 const phys_addr_t paddr[], uint32_t pg_num, uint32_t pg_shift,
424 rte_mempool_memchunk_free_cb_t *free_cb, void *opaque)
428 size_t pg_sz = (size_t)1 << pg_shift;
430 /* mempool must not be populated */
431 if (mp->nb_mem_chunks != 0)
434 if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG)
435 return rte_mempool_populate_phys(mp, vaddr, RTE_BAD_PHYS_ADDR,
436 pg_num * pg_sz, free_cb, opaque);
438 for (i = 0; i < pg_num && mp->populated_size < mp->size; i += n) {
440 /* populate with the largest group of contiguous pages */
441 for (n = 1; (i + n) < pg_num &&
442 paddr[i + n - 1] + pg_sz == paddr[i + n]; n++)
445 ret = rte_mempool_populate_phys(mp, vaddr + i * pg_sz,
446 paddr[i], n * pg_sz, free_cb, opaque);
448 rte_mempool_free_memchunks(mp);
451 /* no need to call the free callback for next chunks */
458 /* Populate the mempool with a virtual area. Return the number of
459 * objects added, or a negative value on error.
462 rte_mempool_populate_virt(struct rte_mempool *mp, char *addr,
463 size_t len, size_t pg_sz, rte_mempool_memchunk_free_cb_t *free_cb,
467 size_t off, phys_len;
470 /* mempool must not be populated */
471 if (mp->nb_mem_chunks != 0)
473 /* address and len must be page-aligned */
474 if (RTE_PTR_ALIGN_CEIL(addr, pg_sz) != addr)
476 if (RTE_ALIGN_CEIL(len, pg_sz) != len)
479 if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG)
480 return rte_mempool_populate_phys(mp, addr, RTE_BAD_PHYS_ADDR,
481 len, free_cb, opaque);
483 for (off = 0; off + pg_sz <= len &&
484 mp->populated_size < mp->size; off += phys_len) {
486 paddr = rte_mem_virt2phy(addr + off);
487 /* required for xen_dom0 to get the machine address */
488 paddr = rte_mem_phy2mch(-1, paddr);
490 if (paddr == RTE_BAD_PHYS_ADDR && rte_eal_has_hugepages()) {
495 /* populate with the largest group of contiguous pages */
496 for (phys_len = pg_sz; off + phys_len < len; phys_len += pg_sz) {
497 phys_addr_t paddr_tmp;
499 paddr_tmp = rte_mem_virt2phy(addr + off + phys_len);
500 paddr_tmp = rte_mem_phy2mch(-1, paddr_tmp);
502 if (paddr_tmp != paddr + phys_len)
506 ret = rte_mempool_populate_phys(mp, addr + off, paddr,
507 phys_len, free_cb, opaque);
510 /* no need to call the free callback for next chunks */
518 rte_mempool_free_memchunks(mp);
522 /* Default function to populate the mempool: allocate memory in memzones,
523 * and populate them. Return the number of objects added, or a negative
527 rte_mempool_populate_default(struct rte_mempool *mp)
529 unsigned int mz_flags = RTE_MEMZONE_1GB|RTE_MEMZONE_SIZE_HINT_ONLY;
530 char mz_name[RTE_MEMZONE_NAMESIZE];
531 const struct rte_memzone *mz;
532 size_t size, total_elt_sz, align, pg_sz, pg_shift;
535 unsigned int mp_flags;
538 /* mempool must not be populated */
539 if (mp->nb_mem_chunks != 0)
542 /* Get mempool capabilities */
544 ret = rte_mempool_ops_get_capabilities(mp, &mp_flags);
546 RTE_LOG(DEBUG, MEMPOOL, "get_capability not supported for %s\n",
551 /* update mempool capabilities */
552 mp->flags |= mp_flags;
554 if (rte_xen_dom0_supported()) {
555 pg_sz = RTE_PGSIZE_2M;
556 pg_shift = rte_bsf32(pg_sz);
558 } else if (rte_eal_has_hugepages()) {
559 pg_shift = 0; /* not needed, zone is physically contiguous */
561 align = RTE_CACHE_LINE_SIZE;
563 pg_sz = getpagesize();
564 pg_shift = rte_bsf32(pg_sz);
568 total_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;
569 for (mz_id = 0, n = mp->size; n > 0; mz_id++, n -= ret) {
570 size = rte_mempool_xmem_size(n, total_elt_sz, pg_shift,
573 ret = snprintf(mz_name, sizeof(mz_name),
574 RTE_MEMPOOL_MZ_FORMAT "_%d", mp->name, mz_id);
575 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
580 mz = rte_memzone_reserve_aligned(mz_name, size,
581 mp->socket_id, mz_flags, align);
582 /* not enough memory, retry with the biggest zone we have */
584 mz = rte_memzone_reserve_aligned(mz_name, 0,
585 mp->socket_id, mz_flags, align);
591 if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG)
592 paddr = RTE_BAD_PHYS_ADDR;
594 paddr = mz->phys_addr;
596 if (rte_eal_has_hugepages() && !rte_xen_dom0_supported())
597 ret = rte_mempool_populate_phys(mp, mz->addr,
599 rte_mempool_memchunk_mz_free,
600 (void *)(uintptr_t)mz);
602 ret = rte_mempool_populate_virt(mp, mz->addr,
604 rte_mempool_memchunk_mz_free,
605 (void *)(uintptr_t)mz);
607 rte_memzone_free(mz);
615 rte_mempool_free_memchunks(mp);
619 /* return the memory size required for mempool objects in anonymous mem */
621 get_anon_size(const struct rte_mempool *mp)
623 size_t size, total_elt_sz, pg_sz, pg_shift;
625 pg_sz = getpagesize();
626 pg_shift = rte_bsf32(pg_sz);
627 total_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;
628 size = rte_mempool_xmem_size(mp->size, total_elt_sz, pg_shift,
634 /* unmap a memory zone mapped by rte_mempool_populate_anon() */
636 rte_mempool_memchunk_anon_free(struct rte_mempool_memhdr *memhdr,
639 munmap(opaque, get_anon_size(memhdr->mp));
642 /* populate the mempool with an anonymous mapping */
644 rte_mempool_populate_anon(struct rte_mempool *mp)
650 /* mempool is already populated, error */
651 if (!STAILQ_EMPTY(&mp->mem_list)) {
656 /* get chunk of virtually continuous memory */
657 size = get_anon_size(mp);
658 addr = mmap(NULL, size, PROT_READ | PROT_WRITE,
659 MAP_SHARED | MAP_ANONYMOUS, -1, 0);
660 if (addr == MAP_FAILED) {
664 /* can't use MMAP_LOCKED, it does not exist on BSD */
665 if (mlock(addr, size) < 0) {
671 ret = rte_mempool_populate_virt(mp, addr, size, getpagesize(),
672 rte_mempool_memchunk_anon_free, addr);
676 return mp->populated_size;
679 rte_mempool_free_memchunks(mp);
685 rte_mempool_free(struct rte_mempool *mp)
687 struct rte_mempool_list *mempool_list = NULL;
688 struct rte_tailq_entry *te;
693 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
694 rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
695 /* find out tailq entry */
696 TAILQ_FOREACH(te, mempool_list, next) {
697 if (te->data == (void *)mp)
702 TAILQ_REMOVE(mempool_list, te, next);
705 rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
707 rte_mempool_free_memchunks(mp);
708 rte_mempool_ops_free(mp);
709 rte_memzone_free(mp->mz);
713 mempool_cache_init(struct rte_mempool_cache *cache, uint32_t size)
716 cache->flushthresh = CALC_CACHE_FLUSHTHRESH(size);
721 * Create and initialize a cache for objects that are retrieved from and
722 * returned to an underlying mempool. This structure is identical to the
723 * local_cache[lcore_id] pointed to by the mempool structure.
725 struct rte_mempool_cache *
726 rte_mempool_cache_create(uint32_t size, int socket_id)
728 struct rte_mempool_cache *cache;
730 if (size == 0 || size > RTE_MEMPOOL_CACHE_MAX_SIZE) {
735 cache = rte_zmalloc_socket("MEMPOOL_CACHE", sizeof(*cache),
736 RTE_CACHE_LINE_SIZE, socket_id);
738 RTE_LOG(ERR, MEMPOOL, "Cannot allocate mempool cache.\n");
743 mempool_cache_init(cache, size);
749 * Free a cache. It's the responsibility of the user to make sure that any
750 * remaining objects in the cache are flushed to the corresponding
754 rte_mempool_cache_free(struct rte_mempool_cache *cache)
759 /* create an empty mempool */
761 rte_mempool_create_empty(const char *name, unsigned n, unsigned elt_size,
762 unsigned cache_size, unsigned private_data_size,
763 int socket_id, unsigned flags)
765 char mz_name[RTE_MEMZONE_NAMESIZE];
766 struct rte_mempool_list *mempool_list;
767 struct rte_mempool *mp = NULL;
768 struct rte_tailq_entry *te = NULL;
769 const struct rte_memzone *mz = NULL;
771 unsigned int mz_flags = RTE_MEMZONE_1GB|RTE_MEMZONE_SIZE_HINT_ONLY;
772 struct rte_mempool_objsz objsz;
776 /* compilation-time checks */
777 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool) &
778 RTE_CACHE_LINE_MASK) != 0);
779 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_cache) &
780 RTE_CACHE_LINE_MASK) != 0);
781 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
782 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_debug_stats) &
783 RTE_CACHE_LINE_MASK) != 0);
784 RTE_BUILD_BUG_ON((offsetof(struct rte_mempool, stats) &
785 RTE_CACHE_LINE_MASK) != 0);
788 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
790 /* asked cache too big */
791 if (cache_size > RTE_MEMPOOL_CACHE_MAX_SIZE ||
792 CALC_CACHE_FLUSHTHRESH(cache_size) > n) {
797 /* "no cache align" imply "no spread" */
798 if (flags & MEMPOOL_F_NO_CACHE_ALIGN)
799 flags |= MEMPOOL_F_NO_SPREAD;
801 /* calculate mempool object sizes. */
802 if (!rte_mempool_calc_obj_size(elt_size, flags, &objsz)) {
807 rte_rwlock_write_lock(RTE_EAL_MEMPOOL_RWLOCK);
810 * reserve a memory zone for this mempool: private data is
813 private_data_size = (private_data_size +
814 RTE_MEMPOOL_ALIGN_MASK) & (~RTE_MEMPOOL_ALIGN_MASK);
817 /* try to allocate tailq entry */
818 te = rte_zmalloc("MEMPOOL_TAILQ_ENTRY", sizeof(*te), 0);
820 RTE_LOG(ERR, MEMPOOL, "Cannot allocate tailq entry!\n");
824 mempool_size = MEMPOOL_HEADER_SIZE(mp, cache_size);
825 mempool_size += private_data_size;
826 mempool_size = RTE_ALIGN_CEIL(mempool_size, RTE_MEMPOOL_ALIGN);
828 ret = snprintf(mz_name, sizeof(mz_name), RTE_MEMPOOL_MZ_FORMAT, name);
829 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
830 rte_errno = ENAMETOOLONG;
834 mz = rte_memzone_reserve(mz_name, mempool_size, socket_id, mz_flags);
838 /* init the mempool structure */
840 memset(mp, 0, MEMPOOL_HEADER_SIZE(mp, cache_size));
841 ret = snprintf(mp->name, sizeof(mp->name), "%s", name);
842 if (ret < 0 || ret >= (int)sizeof(mp->name)) {
843 rte_errno = ENAMETOOLONG;
849 mp->socket_id = socket_id;
850 mp->elt_size = objsz.elt_size;
851 mp->header_size = objsz.header_size;
852 mp->trailer_size = objsz.trailer_size;
853 /* Size of default caches, zero means disabled. */
854 mp->cache_size = cache_size;
855 mp->private_data_size = private_data_size;
856 STAILQ_INIT(&mp->elt_list);
857 STAILQ_INIT(&mp->mem_list);
860 * local_cache pointer is set even if cache_size is zero.
861 * The local_cache points to just past the elt_pa[] array.
863 mp->local_cache = (struct rte_mempool_cache *)
864 RTE_PTR_ADD(mp, MEMPOOL_HEADER_SIZE(mp, 0));
866 /* Init all default caches. */
867 if (cache_size != 0) {
868 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
869 mempool_cache_init(&mp->local_cache[lcore_id],
875 rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
876 TAILQ_INSERT_TAIL(mempool_list, te, next);
877 rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
878 rte_rwlock_write_unlock(RTE_EAL_MEMPOOL_RWLOCK);
883 rte_rwlock_write_unlock(RTE_EAL_MEMPOOL_RWLOCK);
885 rte_mempool_free(mp);
889 /* create the mempool */
891 rte_mempool_create(const char *name, unsigned n, unsigned elt_size,
892 unsigned cache_size, unsigned private_data_size,
893 rte_mempool_ctor_t *mp_init, void *mp_init_arg,
894 rte_mempool_obj_cb_t *obj_init, void *obj_init_arg,
895 int socket_id, unsigned flags)
898 struct rte_mempool *mp;
900 mp = rte_mempool_create_empty(name, n, elt_size, cache_size,
901 private_data_size, socket_id, flags);
906 * Since we have 4 combinations of the SP/SC/MP/MC examine the flags to
907 * set the correct index into the table of ops structs.
909 if ((flags & MEMPOOL_F_SP_PUT) && (flags & MEMPOOL_F_SC_GET))
910 ret = rte_mempool_set_ops_byname(mp, "ring_sp_sc", NULL);
911 else if (flags & MEMPOOL_F_SP_PUT)
912 ret = rte_mempool_set_ops_byname(mp, "ring_sp_mc", NULL);
913 else if (flags & MEMPOOL_F_SC_GET)
914 ret = rte_mempool_set_ops_byname(mp, "ring_mp_sc", NULL);
916 ret = rte_mempool_set_ops_byname(mp, "ring_mp_mc", NULL);
921 /* call the mempool priv initializer */
923 mp_init(mp, mp_init_arg);
925 if (rte_mempool_populate_default(mp) < 0)
928 /* call the object initializers */
930 rte_mempool_obj_iter(mp, obj_init, obj_init_arg);
935 rte_mempool_free(mp);
940 * Create the mempool over already allocated chunk of memory.
941 * That external memory buffer can consists of physically disjoint pages.
942 * Setting vaddr to NULL, makes mempool to fallback to rte_mempool_create()
946 rte_mempool_xmem_create(const char *name, unsigned n, unsigned elt_size,
947 unsigned cache_size, unsigned private_data_size,
948 rte_mempool_ctor_t *mp_init, void *mp_init_arg,
949 rte_mempool_obj_cb_t *obj_init, void *obj_init_arg,
950 int socket_id, unsigned flags, void *vaddr,
951 const phys_addr_t paddr[], uint32_t pg_num, uint32_t pg_shift)
953 struct rte_mempool *mp = NULL;
956 /* no virtual address supplied, use rte_mempool_create() */
958 return rte_mempool_create(name, n, elt_size, cache_size,
959 private_data_size, mp_init, mp_init_arg,
960 obj_init, obj_init_arg, socket_id, flags);
962 /* check that we have both VA and PA */
968 /* Check that pg_shift parameter is valid. */
969 if (pg_shift > MEMPOOL_PG_SHIFT_MAX) {
974 mp = rte_mempool_create_empty(name, n, elt_size, cache_size,
975 private_data_size, socket_id, flags);
979 /* call the mempool priv initializer */
981 mp_init(mp, mp_init_arg);
983 ret = rte_mempool_populate_phys_tab(mp, vaddr, paddr, pg_num, pg_shift,
985 if (ret < 0 || ret != (int)mp->size)
988 /* call the object initializers */
990 rte_mempool_obj_iter(mp, obj_init, obj_init_arg);
995 rte_mempool_free(mp);
999 /* Return the number of entries in the mempool */
1001 rte_mempool_avail_count(const struct rte_mempool *mp)
1006 count = rte_mempool_ops_get_count(mp);
1008 if (mp->cache_size == 0)
1011 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
1012 count += mp->local_cache[lcore_id].len;
1015 * due to race condition (access to len is not locked), the
1016 * total can be greater than size... so fix the result
1018 if (count > mp->size)
1023 /* return the number of entries allocated from the mempool */
1025 rte_mempool_in_use_count(const struct rte_mempool *mp)
1027 return mp->size - rte_mempool_avail_count(mp);
1030 /* dump the cache status */
1032 rte_mempool_dump_cache(FILE *f, const struct rte_mempool *mp)
1036 unsigned cache_count;
1038 fprintf(f, " internal cache infos:\n");
1039 fprintf(f, " cache_size=%"PRIu32"\n", mp->cache_size);
1041 if (mp->cache_size == 0)
1044 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1045 cache_count = mp->local_cache[lcore_id].len;
1046 fprintf(f, " cache_count[%u]=%"PRIu32"\n",
1047 lcore_id, cache_count);
1048 count += cache_count;
1050 fprintf(f, " total_cache_count=%u\n", count);
1054 #ifndef __INTEL_COMPILER
1055 #pragma GCC diagnostic ignored "-Wcast-qual"
1058 /* check and update cookies or panic (internal) */
1059 void rte_mempool_check_cookies(const struct rte_mempool *mp,
1060 void * const *obj_table_const, unsigned n, int free)
1062 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1063 struct rte_mempool_objhdr *hdr;
1064 struct rte_mempool_objtlr *tlr;
1070 /* Force to drop the "const" attribute. This is done only when
1071 * DEBUG is enabled */
1072 tmp = (void *) obj_table_const;
1078 if (rte_mempool_from_obj(obj) != mp)
1079 rte_panic("MEMPOOL: object is owned by another "
1082 hdr = __mempool_get_header(obj);
1083 cookie = hdr->cookie;
1086 if (cookie != RTE_MEMPOOL_HEADER_COOKIE1) {
1087 RTE_LOG(CRIT, MEMPOOL,
1088 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1089 obj, (const void *) mp, cookie);
1090 rte_panic("MEMPOOL: bad header cookie (put)\n");
1092 hdr->cookie = RTE_MEMPOOL_HEADER_COOKIE2;
1093 } else if (free == 1) {
1094 if (cookie != RTE_MEMPOOL_HEADER_COOKIE2) {
1095 RTE_LOG(CRIT, MEMPOOL,
1096 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1097 obj, (const void *) mp, cookie);
1098 rte_panic("MEMPOOL: bad header cookie (get)\n");
1100 hdr->cookie = RTE_MEMPOOL_HEADER_COOKIE1;
1101 } else if (free == 2) {
1102 if (cookie != RTE_MEMPOOL_HEADER_COOKIE1 &&
1103 cookie != RTE_MEMPOOL_HEADER_COOKIE2) {
1104 RTE_LOG(CRIT, MEMPOOL,
1105 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1106 obj, (const void *) mp, cookie);
1107 rte_panic("MEMPOOL: bad header cookie (audit)\n");
1110 tlr = __mempool_get_trailer(obj);
1111 cookie = tlr->cookie;
1112 if (cookie != RTE_MEMPOOL_TRAILER_COOKIE) {
1113 RTE_LOG(CRIT, MEMPOOL,
1114 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1115 obj, (const void *) mp, cookie);
1116 rte_panic("MEMPOOL: bad trailer cookie\n");
1121 RTE_SET_USED(obj_table_const);
1127 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1129 mempool_obj_audit(struct rte_mempool *mp, __rte_unused void *opaque,
1130 void *obj, __rte_unused unsigned idx)
1132 __mempool_check_cookies(mp, &obj, 1, 2);
1136 mempool_audit_cookies(struct rte_mempool *mp)
1140 num = rte_mempool_obj_iter(mp, mempool_obj_audit, NULL);
1141 if (num != mp->size) {
1142 rte_panic("rte_mempool_obj_iter(mempool=%p, size=%u) "
1143 "iterated only over %u elements\n",
1148 #define mempool_audit_cookies(mp) do {} while(0)
1151 #ifndef __INTEL_COMPILER
1152 #pragma GCC diagnostic error "-Wcast-qual"
1155 /* check cookies before and after objects */
1157 mempool_audit_cache(const struct rte_mempool *mp)
1159 /* check cache size consistency */
1162 if (mp->cache_size == 0)
1165 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1166 const struct rte_mempool_cache *cache;
1167 cache = &mp->local_cache[lcore_id];
1168 if (cache->len > cache->flushthresh) {
1169 RTE_LOG(CRIT, MEMPOOL, "badness on cache[%u]\n",
1171 rte_panic("MEMPOOL: invalid cache len\n");
1176 /* check the consistency of mempool (size, cookies, ...) */
1178 rte_mempool_audit(struct rte_mempool *mp)
1180 mempool_audit_cache(mp);
1181 mempool_audit_cookies(mp);
1183 /* For case where mempool DEBUG is not set, and cache size is 0 */
1187 /* dump the status of the mempool on the console */
1189 rte_mempool_dump(FILE *f, struct rte_mempool *mp)
1191 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1192 struct rte_mempool_debug_stats sum;
1195 struct rte_mempool_memhdr *memhdr;
1196 unsigned common_count;
1197 unsigned cache_count;
1200 RTE_ASSERT(f != NULL);
1201 RTE_ASSERT(mp != NULL);
1203 fprintf(f, "mempool <%s>@%p\n", mp->name, mp);
1204 fprintf(f, " flags=%x\n", mp->flags);
1205 fprintf(f, " pool=%p\n", mp->pool_data);
1206 fprintf(f, " phys_addr=0x%" PRIx64 "\n", mp->mz->phys_addr);
1207 fprintf(f, " nb_mem_chunks=%u\n", mp->nb_mem_chunks);
1208 fprintf(f, " size=%"PRIu32"\n", mp->size);
1209 fprintf(f, " populated_size=%"PRIu32"\n", mp->populated_size);
1210 fprintf(f, " header_size=%"PRIu32"\n", mp->header_size);
1211 fprintf(f, " elt_size=%"PRIu32"\n", mp->elt_size);
1212 fprintf(f, " trailer_size=%"PRIu32"\n", mp->trailer_size);
1213 fprintf(f, " total_obj_size=%"PRIu32"\n",
1214 mp->header_size + mp->elt_size + mp->trailer_size);
1216 fprintf(f, " private_data_size=%"PRIu32"\n", mp->private_data_size);
1218 STAILQ_FOREACH(memhdr, &mp->mem_list, next)
1219 mem_len += memhdr->len;
1221 fprintf(f, " avg bytes/object=%#Lf\n",
1222 (long double)mem_len / mp->size);
1225 cache_count = rte_mempool_dump_cache(f, mp);
1226 common_count = rte_mempool_ops_get_count(mp);
1227 if ((cache_count + common_count) > mp->size)
1228 common_count = mp->size - cache_count;
1229 fprintf(f, " common_pool_count=%u\n", common_count);
1231 /* sum and dump statistics */
1232 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1233 memset(&sum, 0, sizeof(sum));
1234 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1235 sum.put_bulk += mp->stats[lcore_id].put_bulk;
1236 sum.put_objs += mp->stats[lcore_id].put_objs;
1237 sum.get_success_bulk += mp->stats[lcore_id].get_success_bulk;
1238 sum.get_success_objs += mp->stats[lcore_id].get_success_objs;
1239 sum.get_fail_bulk += mp->stats[lcore_id].get_fail_bulk;
1240 sum.get_fail_objs += mp->stats[lcore_id].get_fail_objs;
1242 fprintf(f, " stats:\n");
1243 fprintf(f, " put_bulk=%"PRIu64"\n", sum.put_bulk);
1244 fprintf(f, " put_objs=%"PRIu64"\n", sum.put_objs);
1245 fprintf(f, " get_success_bulk=%"PRIu64"\n", sum.get_success_bulk);
1246 fprintf(f, " get_success_objs=%"PRIu64"\n", sum.get_success_objs);
1247 fprintf(f, " get_fail_bulk=%"PRIu64"\n", sum.get_fail_bulk);
1248 fprintf(f, " get_fail_objs=%"PRIu64"\n", sum.get_fail_objs);
1250 fprintf(f, " no statistics available\n");
1253 rte_mempool_audit(mp);
1256 /* dump the status of all mempools on the console */
1258 rte_mempool_list_dump(FILE *f)
1260 struct rte_mempool *mp = NULL;
1261 struct rte_tailq_entry *te;
1262 struct rte_mempool_list *mempool_list;
1264 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1266 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1268 TAILQ_FOREACH(te, mempool_list, next) {
1269 mp = (struct rte_mempool *) te->data;
1270 rte_mempool_dump(f, mp);
1273 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);
1276 /* search a mempool from its name */
1277 struct rte_mempool *
1278 rte_mempool_lookup(const char *name)
1280 struct rte_mempool *mp = NULL;
1281 struct rte_tailq_entry *te;
1282 struct rte_mempool_list *mempool_list;
1284 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1286 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1288 TAILQ_FOREACH(te, mempool_list, next) {
1289 mp = (struct rte_mempool *) te->data;
1290 if (strncmp(name, mp->name, RTE_MEMPOOL_NAMESIZE) == 0)
1294 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);
1304 void rte_mempool_walk(void (*func)(struct rte_mempool *, void *),
1307 struct rte_tailq_entry *te = NULL;
1308 struct rte_mempool_list *mempool_list;
1311 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1313 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1315 TAILQ_FOREACH_SAFE(te, mempool_list, next, tmp_te) {
1316 (*func)((struct rte_mempool *) te->data, arg);
1319 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);