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>
59 #include <rte_errno.h>
60 #include <rte_string_fns.h>
61 #include <rte_spinlock.h>
63 #include "rte_mempool.h"
65 TAILQ_HEAD(rte_mempool_list, rte_tailq_entry);
67 static struct rte_tailq_elem rte_mempool_tailq = {
68 .name = "RTE_MEMPOOL",
70 EAL_REGISTER_TAILQ(rte_mempool_tailq)
72 #define CACHE_FLUSHTHRESH_MULTIPLIER 1.5
73 #define CALC_CACHE_FLUSHTHRESH(c) \
74 ((typeof(c))((c) * CACHE_FLUSHTHRESH_MULTIPLIER))
77 * return the greatest common divisor between a and b (fast algorithm)
80 static unsigned get_gcd(unsigned a, unsigned b)
105 * Depending on memory configuration, objects addresses are spread
106 * between channels and ranks in RAM: the pool allocator will add
107 * padding between objects. This function return the new size of the
110 static unsigned optimize_object_size(unsigned obj_size)
112 unsigned nrank, nchan;
113 unsigned new_obj_size;
115 /* get number of channels */
116 nchan = rte_memory_get_nchannel();
120 nrank = rte_memory_get_nrank();
124 /* process new object size */
125 new_obj_size = (obj_size + RTE_MEMPOOL_ALIGN_MASK) / RTE_MEMPOOL_ALIGN;
126 while (get_gcd(new_obj_size, nrank * nchan) != 1)
128 return new_obj_size * RTE_MEMPOOL_ALIGN;
132 mempool_add_elem(struct rte_mempool *mp, void *obj, phys_addr_t physaddr)
134 struct rte_mempool_objhdr *hdr;
135 struct rte_mempool_objtlr *tlr __rte_unused;
137 /* set mempool ptr in header */
138 hdr = RTE_PTR_SUB(obj, sizeof(*hdr));
140 hdr->physaddr = physaddr;
141 STAILQ_INSERT_TAIL(&mp->elt_list, hdr, next);
142 mp->populated_size++;
144 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
145 hdr->cookie = RTE_MEMPOOL_HEADER_COOKIE2;
146 tlr = __mempool_get_trailer(obj);
147 tlr->cookie = RTE_MEMPOOL_TRAILER_COOKIE;
150 /* enqueue in ring */
151 rte_mempool_ops_enqueue_bulk(mp, &obj, 1);
154 /* call obj_cb() for each mempool element */
156 rte_mempool_obj_iter(struct rte_mempool *mp,
157 rte_mempool_obj_cb_t *obj_cb, void *obj_cb_arg)
159 struct rte_mempool_objhdr *hdr;
163 STAILQ_FOREACH(hdr, &mp->elt_list, next) {
164 obj = (char *)hdr + sizeof(*hdr);
165 obj_cb(mp, obj_cb_arg, obj, n);
172 /* call mem_cb() for each mempool memory chunk */
174 rte_mempool_mem_iter(struct rte_mempool *mp,
175 rte_mempool_mem_cb_t *mem_cb, void *mem_cb_arg)
177 struct rte_mempool_memhdr *hdr;
180 STAILQ_FOREACH(hdr, &mp->mem_list, next) {
181 mem_cb(mp, mem_cb_arg, hdr, n);
188 /* get the header, trailer and total size of a mempool element. */
190 rte_mempool_calc_obj_size(uint32_t elt_size, uint32_t flags,
191 struct rte_mempool_objsz *sz)
193 struct rte_mempool_objsz lsz;
195 sz = (sz != NULL) ? sz : &lsz;
197 sz->header_size = sizeof(struct rte_mempool_objhdr);
198 if ((flags & MEMPOOL_F_NO_CACHE_ALIGN) == 0)
199 sz->header_size = RTE_ALIGN_CEIL(sz->header_size,
202 sz->trailer_size = sizeof(struct rte_mempool_objtlr);
204 /* element size is 8 bytes-aligned at least */
205 sz->elt_size = RTE_ALIGN_CEIL(elt_size, sizeof(uint64_t));
207 /* expand trailer to next cache line */
208 if ((flags & MEMPOOL_F_NO_CACHE_ALIGN) == 0) {
209 sz->total_size = sz->header_size + sz->elt_size +
211 sz->trailer_size += ((RTE_MEMPOOL_ALIGN -
212 (sz->total_size & RTE_MEMPOOL_ALIGN_MASK)) &
213 RTE_MEMPOOL_ALIGN_MASK);
217 * increase trailer to add padding between objects in order to
218 * spread them across memory channels/ranks
220 if ((flags & MEMPOOL_F_NO_SPREAD) == 0) {
222 new_size = optimize_object_size(sz->header_size + sz->elt_size +
224 sz->trailer_size = new_size - sz->header_size - sz->elt_size;
227 /* this is the size of an object, including header and trailer */
228 sz->total_size = sz->header_size + sz->elt_size + sz->trailer_size;
230 return sz->total_size;
235 * Calculate maximum amount of memory required to store given number of objects.
238 rte_mempool_xmem_size(uint32_t elt_num, size_t total_elt_sz, uint32_t pg_shift)
240 size_t obj_per_page, pg_num, pg_sz;
242 if (total_elt_sz == 0)
246 return total_elt_sz * elt_num;
248 pg_sz = (size_t)1 << pg_shift;
249 obj_per_page = pg_sz / total_elt_sz;
250 if (obj_per_page == 0)
251 return RTE_ALIGN_CEIL(total_elt_sz, pg_sz) * elt_num;
253 pg_num = (elt_num + obj_per_page - 1) / obj_per_page;
254 return pg_num << pg_shift;
258 * Calculate how much memory would be actually required with the
259 * given memory footprint to store required number of elements.
262 rte_mempool_xmem_usage(__rte_unused void *vaddr, uint32_t elt_num,
263 size_t total_elt_sz, const phys_addr_t paddr[], uint32_t pg_num,
266 uint32_t elt_cnt = 0;
267 phys_addr_t start, end;
269 size_t pg_sz = (size_t)1 << pg_shift;
271 /* if paddr is NULL, assume contiguous memory */
274 end = pg_sz * pg_num;
278 end = paddr[0] + pg_sz;
281 while (elt_cnt < elt_num) {
283 if (end - start >= total_elt_sz) {
284 /* enough contiguous memory, add an object */
285 start += total_elt_sz;
287 } else if (paddr_idx < pg_num) {
288 /* no room to store one obj, add a page */
289 if (end == paddr[paddr_idx]) {
292 start = paddr[paddr_idx];
293 end = paddr[paddr_idx] + pg_sz;
298 /* no more page, return how many elements fit */
299 return -(size_t)elt_cnt;
303 return (size_t)paddr_idx << pg_shift;
306 /* free a memchunk allocated with rte_memzone_reserve() */
308 rte_mempool_memchunk_mz_free(__rte_unused struct rte_mempool_memhdr *memhdr,
311 const struct rte_memzone *mz = opaque;
312 rte_memzone_free(mz);
315 /* Free memory chunks used by a mempool. Objects must be in pool */
317 rte_mempool_free_memchunks(struct rte_mempool *mp)
319 struct rte_mempool_memhdr *memhdr;
322 while (!STAILQ_EMPTY(&mp->elt_list)) {
323 rte_mempool_ops_dequeue_bulk(mp, &elt, 1);
325 STAILQ_REMOVE_HEAD(&mp->elt_list, next);
326 mp->populated_size--;
329 while (!STAILQ_EMPTY(&mp->mem_list)) {
330 memhdr = STAILQ_FIRST(&mp->mem_list);
331 STAILQ_REMOVE_HEAD(&mp->mem_list, next);
332 if (memhdr->free_cb != NULL)
333 memhdr->free_cb(memhdr, memhdr->opaque);
339 /* Add objects in the pool, using a physically contiguous memory
340 * zone. Return the number of objects added, or a negative value
344 rte_mempool_populate_phys(struct rte_mempool *mp, char *vaddr,
345 phys_addr_t paddr, size_t len, rte_mempool_memchunk_free_cb_t *free_cb,
348 unsigned total_elt_sz;
351 struct rte_mempool_memhdr *memhdr;
354 /* create the internal ring if not already done */
355 if ((mp->flags & MEMPOOL_F_POOL_CREATED) == 0) {
356 ret = rte_mempool_ops_alloc(mp);
359 mp->flags |= MEMPOOL_F_POOL_CREATED;
362 /* mempool is already populated */
363 if (mp->populated_size >= mp->size)
366 total_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;
368 memhdr = rte_zmalloc("MEMPOOL_MEMHDR", sizeof(*memhdr), 0);
373 memhdr->addr = vaddr;
374 memhdr->phys_addr = paddr;
376 memhdr->free_cb = free_cb;
377 memhdr->opaque = opaque;
379 if (mp->flags & MEMPOOL_F_NO_CACHE_ALIGN)
380 off = RTE_PTR_ALIGN_CEIL(vaddr, 8) - vaddr;
382 off = RTE_PTR_ALIGN_CEIL(vaddr, RTE_CACHE_LINE_SIZE) - vaddr;
384 while (off + total_elt_sz <= len && mp->populated_size < mp->size) {
385 off += mp->header_size;
386 if (paddr == RTE_BAD_PHYS_ADDR)
387 mempool_add_elem(mp, (char *)vaddr + off,
390 mempool_add_elem(mp, (char *)vaddr + off, paddr + off);
391 off += mp->elt_size + mp->trailer_size;
395 /* not enough room to store one object */
399 STAILQ_INSERT_TAIL(&mp->mem_list, memhdr, next);
404 /* Add objects in the pool, using a table of physical pages. Return the
405 * number of objects added, or a negative value on error.
408 rte_mempool_populate_phys_tab(struct rte_mempool *mp, char *vaddr,
409 const phys_addr_t paddr[], uint32_t pg_num, uint32_t pg_shift,
410 rte_mempool_memchunk_free_cb_t *free_cb, void *opaque)
414 size_t pg_sz = (size_t)1 << pg_shift;
416 /* mempool must not be populated */
417 if (mp->nb_mem_chunks != 0)
420 if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG)
421 return rte_mempool_populate_phys(mp, vaddr, RTE_BAD_PHYS_ADDR,
422 pg_num * pg_sz, free_cb, opaque);
424 for (i = 0; i < pg_num && mp->populated_size < mp->size; i += n) {
426 /* populate with the largest group of contiguous pages */
427 for (n = 1; (i + n) < pg_num &&
428 paddr[i] + pg_sz == paddr[i+n]; n++)
431 ret = rte_mempool_populate_phys(mp, vaddr + i * pg_sz,
432 paddr[i], n * pg_sz, free_cb, opaque);
434 rte_mempool_free_memchunks(mp);
437 /* no need to call the free callback for next chunks */
444 /* Populate the mempool with a virtual area. Return the number of
445 * objects added, or a negative value on error.
448 rte_mempool_populate_virt(struct rte_mempool *mp, char *addr,
449 size_t len, size_t pg_sz, rte_mempool_memchunk_free_cb_t *free_cb,
453 size_t off, phys_len;
456 /* mempool must not be populated */
457 if (mp->nb_mem_chunks != 0)
459 /* address and len must be page-aligned */
460 if (RTE_PTR_ALIGN_CEIL(addr, pg_sz) != addr)
462 if (RTE_ALIGN_CEIL(len, pg_sz) != len)
465 if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG)
466 return rte_mempool_populate_phys(mp, addr, RTE_BAD_PHYS_ADDR,
467 len, free_cb, opaque);
469 for (off = 0; off + pg_sz <= len &&
470 mp->populated_size < mp->size; off += phys_len) {
472 paddr = rte_mem_virt2phy(addr + off);
473 /* required for xen_dom0 to get the machine address */
474 paddr = rte_mem_phy2mch(-1, paddr);
476 if (paddr == RTE_BAD_PHYS_ADDR) {
481 /* populate with the largest group of contiguous pages */
482 for (phys_len = pg_sz; off + phys_len < len; phys_len += pg_sz) {
483 phys_addr_t paddr_tmp;
485 paddr_tmp = rte_mem_virt2phy(addr + off + phys_len);
486 paddr_tmp = rte_mem_phy2mch(-1, paddr_tmp);
488 if (paddr_tmp != paddr + phys_len)
492 ret = rte_mempool_populate_phys(mp, addr + off, paddr,
493 phys_len, free_cb, opaque);
496 /* no need to call the free callback for next chunks */
504 rte_mempool_free_memchunks(mp);
508 /* Default function to populate the mempool: allocate memory in memzones,
509 * and populate them. Return the number of objects added, or a negative
513 rte_mempool_populate_default(struct rte_mempool *mp)
515 int mz_flags = RTE_MEMZONE_1GB|RTE_MEMZONE_SIZE_HINT_ONLY;
516 char mz_name[RTE_MEMZONE_NAMESIZE];
517 const struct rte_memzone *mz;
518 size_t size, total_elt_sz, align, pg_sz, pg_shift;
523 /* mempool must not be populated */
524 if (mp->nb_mem_chunks != 0)
527 if (rte_eal_has_hugepages()) {
528 pg_shift = 0; /* not needed, zone is physically contiguous */
530 align = RTE_CACHE_LINE_SIZE;
532 pg_sz = getpagesize();
533 pg_shift = rte_bsf32(pg_sz);
537 total_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;
538 for (mz_id = 0, n = mp->size; n > 0; mz_id++, n -= ret) {
539 size = rte_mempool_xmem_size(n, total_elt_sz, pg_shift);
541 ret = snprintf(mz_name, sizeof(mz_name),
542 RTE_MEMPOOL_MZ_FORMAT "_%d", mp->name, mz_id);
543 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
548 mz = rte_memzone_reserve_aligned(mz_name, size,
549 mp->socket_id, mz_flags, align);
550 /* not enough memory, retry with the biggest zone we have */
552 mz = rte_memzone_reserve_aligned(mz_name, 0,
553 mp->socket_id, mz_flags, align);
559 if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG)
560 paddr = RTE_BAD_PHYS_ADDR;
562 paddr = mz->phys_addr;
564 if (rte_eal_has_hugepages() && !rte_xen_dom0_supported())
565 ret = rte_mempool_populate_phys(mp, mz->addr,
567 rte_mempool_memchunk_mz_free,
568 (void *)(uintptr_t)mz);
570 ret = rte_mempool_populate_virt(mp, mz->addr,
572 rte_mempool_memchunk_mz_free,
573 (void *)(uintptr_t)mz);
581 rte_mempool_free_memchunks(mp);
585 /* return the memory size required for mempool objects in anonymous mem */
587 get_anon_size(const struct rte_mempool *mp)
589 size_t size, total_elt_sz, pg_sz, pg_shift;
591 pg_sz = getpagesize();
592 pg_shift = rte_bsf32(pg_sz);
593 total_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;
594 size = rte_mempool_xmem_size(mp->size, total_elt_sz, pg_shift);
599 /* unmap a memory zone mapped by rte_mempool_populate_anon() */
601 rte_mempool_memchunk_anon_free(struct rte_mempool_memhdr *memhdr,
604 munmap(opaque, get_anon_size(memhdr->mp));
607 /* populate the mempool with an anonymous mapping */
609 rte_mempool_populate_anon(struct rte_mempool *mp)
615 /* mempool is already populated, error */
616 if (!STAILQ_EMPTY(&mp->mem_list)) {
621 /* get chunk of virtually continuous memory */
622 size = get_anon_size(mp);
623 addr = mmap(NULL, size, PROT_READ | PROT_WRITE,
624 MAP_SHARED | MAP_ANONYMOUS, -1, 0);
625 if (addr == MAP_FAILED) {
629 /* can't use MMAP_LOCKED, it does not exist on BSD */
630 if (mlock(addr, size) < 0) {
636 ret = rte_mempool_populate_virt(mp, addr, size, getpagesize(),
637 rte_mempool_memchunk_anon_free, addr);
641 return mp->populated_size;
644 rte_mempool_free_memchunks(mp);
650 rte_mempool_free(struct rte_mempool *mp)
652 struct rte_mempool_list *mempool_list = NULL;
653 struct rte_tailq_entry *te;
658 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
659 rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
660 /* find out tailq entry */
661 TAILQ_FOREACH(te, mempool_list, next) {
662 if (te->data == (void *)mp)
667 TAILQ_REMOVE(mempool_list, te, next);
670 rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
672 rte_mempool_free_memchunks(mp);
673 rte_mempool_ops_free(mp);
674 rte_memzone_free(mp->mz);
677 /* create an empty mempool */
679 rte_mempool_create_empty(const char *name, unsigned n, unsigned elt_size,
680 unsigned cache_size, unsigned private_data_size,
681 int socket_id, unsigned flags)
683 char mz_name[RTE_MEMZONE_NAMESIZE];
684 struct rte_mempool_list *mempool_list;
685 struct rte_mempool *mp = NULL;
686 struct rte_tailq_entry *te = NULL;
687 const struct rte_memzone *mz = NULL;
689 int mz_flags = RTE_MEMZONE_1GB|RTE_MEMZONE_SIZE_HINT_ONLY;
690 struct rte_mempool_objsz objsz;
693 /* compilation-time checks */
694 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool) &
695 RTE_CACHE_LINE_MASK) != 0);
696 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_cache) &
697 RTE_CACHE_LINE_MASK) != 0);
698 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
699 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_debug_stats) &
700 RTE_CACHE_LINE_MASK) != 0);
701 RTE_BUILD_BUG_ON((offsetof(struct rte_mempool, stats) &
702 RTE_CACHE_LINE_MASK) != 0);
705 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
707 /* asked cache too big */
708 if (cache_size > RTE_MEMPOOL_CACHE_MAX_SIZE ||
709 CALC_CACHE_FLUSHTHRESH(cache_size) > n) {
714 /* "no cache align" imply "no spread" */
715 if (flags & MEMPOOL_F_NO_CACHE_ALIGN)
716 flags |= MEMPOOL_F_NO_SPREAD;
718 /* calculate mempool object sizes. */
719 if (!rte_mempool_calc_obj_size(elt_size, flags, &objsz)) {
724 rte_rwlock_write_lock(RTE_EAL_MEMPOOL_RWLOCK);
727 * reserve a memory zone for this mempool: private data is
730 private_data_size = (private_data_size +
731 RTE_MEMPOOL_ALIGN_MASK) & (~RTE_MEMPOOL_ALIGN_MASK);
734 /* try to allocate tailq entry */
735 te = rte_zmalloc("MEMPOOL_TAILQ_ENTRY", sizeof(*te), 0);
737 RTE_LOG(ERR, MEMPOOL, "Cannot allocate tailq entry!\n");
741 mempool_size = MEMPOOL_HEADER_SIZE(mp, cache_size);
742 mempool_size += private_data_size;
743 mempool_size = RTE_ALIGN_CEIL(mempool_size, RTE_MEMPOOL_ALIGN);
745 ret = snprintf(mz_name, sizeof(mz_name), RTE_MEMPOOL_MZ_FORMAT, name);
746 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
747 rte_errno = ENAMETOOLONG;
751 mz = rte_memzone_reserve(mz_name, mempool_size, socket_id, mz_flags);
755 /* init the mempool structure */
757 memset(mp, 0, MEMPOOL_HEADER_SIZE(mp, cache_size));
758 ret = snprintf(mp->name, sizeof(mp->name), "%s", name);
759 if (ret < 0 || ret >= (int)sizeof(mp->name)) {
760 rte_errno = ENAMETOOLONG;
764 mp->socket_id = socket_id;
767 mp->socket_id = socket_id;
768 mp->elt_size = objsz.elt_size;
769 mp->header_size = objsz.header_size;
770 mp->trailer_size = objsz.trailer_size;
771 mp->cache_size = cache_size;
772 mp->cache_flushthresh = CALC_CACHE_FLUSHTHRESH(cache_size);
773 mp->private_data_size = private_data_size;
774 STAILQ_INIT(&mp->elt_list);
775 STAILQ_INIT(&mp->mem_list);
778 * local_cache pointer is set even if cache_size is zero.
779 * The local_cache points to just past the elt_pa[] array.
781 mp->local_cache = (struct rte_mempool_cache *)
782 RTE_PTR_ADD(mp, MEMPOOL_HEADER_SIZE(mp, 0));
786 rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
787 TAILQ_INSERT_TAIL(mempool_list, te, next);
788 rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
789 rte_rwlock_write_unlock(RTE_EAL_MEMPOOL_RWLOCK);
794 rte_rwlock_write_unlock(RTE_EAL_MEMPOOL_RWLOCK);
796 rte_mempool_free(mp);
800 /* create the mempool */
802 rte_mempool_create(const char *name, unsigned n, unsigned elt_size,
803 unsigned cache_size, unsigned private_data_size,
804 rte_mempool_ctor_t *mp_init, void *mp_init_arg,
805 rte_mempool_obj_cb_t *obj_init, void *obj_init_arg,
806 int socket_id, unsigned flags)
808 struct rte_mempool *mp;
810 mp = rte_mempool_create_empty(name, n, elt_size, cache_size,
811 private_data_size, socket_id, flags);
816 * Since we have 4 combinations of the SP/SC/MP/MC examine the flags to
817 * set the correct index into the table of ops structs.
819 if (flags & (MEMPOOL_F_SP_PUT | MEMPOOL_F_SC_GET))
820 rte_mempool_set_ops_byname(mp, "ring_sp_sc", NULL);
821 else if (flags & MEMPOOL_F_SP_PUT)
822 rte_mempool_set_ops_byname(mp, "ring_sp_mc", NULL);
823 else if (flags & MEMPOOL_F_SC_GET)
824 rte_mempool_set_ops_byname(mp, "ring_mp_sc", NULL);
826 rte_mempool_set_ops_byname(mp, "ring_mp_mc", NULL);
828 /* call the mempool priv initializer */
830 mp_init(mp, mp_init_arg);
832 if (rte_mempool_populate_default(mp) < 0)
835 /* call the object initializers */
837 rte_mempool_obj_iter(mp, obj_init, obj_init_arg);
842 rte_mempool_free(mp);
847 * Create the mempool over already allocated chunk of memory.
848 * That external memory buffer can consists of physically disjoint pages.
849 * Setting vaddr to NULL, makes mempool to fallback to original behaviour
850 * and allocate space for mempool and it's elements as one big chunk of
851 * physically continuos memory.
854 rte_mempool_xmem_create(const char *name, unsigned n, unsigned elt_size,
855 unsigned cache_size, unsigned private_data_size,
856 rte_mempool_ctor_t *mp_init, void *mp_init_arg,
857 rte_mempool_obj_cb_t *obj_init, void *obj_init_arg,
858 int socket_id, unsigned flags, void *vaddr,
859 const phys_addr_t paddr[], uint32_t pg_num, uint32_t pg_shift)
861 struct rte_mempool *mp = NULL;
864 /* no virtual address supplied, use rte_mempool_create() */
866 return rte_mempool_create(name, n, elt_size, cache_size,
867 private_data_size, mp_init, mp_init_arg,
868 obj_init, obj_init_arg, socket_id, flags);
870 /* check that we have both VA and PA */
876 /* Check that pg_shift parameter is valid. */
877 if (pg_shift > MEMPOOL_PG_SHIFT_MAX) {
882 mp = rte_mempool_create_empty(name, n, elt_size, cache_size,
883 private_data_size, socket_id, flags);
887 /* call the mempool priv initializer */
889 mp_init(mp, mp_init_arg);
891 ret = rte_mempool_populate_phys_tab(mp, vaddr, paddr, pg_num, pg_shift,
893 if (ret < 0 || ret != (int)mp->size)
896 /* call the object initializers */
898 rte_mempool_obj_iter(mp, obj_init, obj_init_arg);
903 rte_mempool_free(mp);
907 /* Return the number of entries in the mempool */
909 rte_mempool_count(const struct rte_mempool *mp)
914 count = rte_mempool_ops_get_count(mp);
916 if (mp->cache_size == 0)
919 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
920 count += mp->local_cache[lcore_id].len;
923 * due to race condition (access to len is not locked), the
924 * total can be greater than size... so fix the result
926 if (count > mp->size)
931 /* dump the cache status */
933 rte_mempool_dump_cache(FILE *f, const struct rte_mempool *mp)
937 unsigned cache_count;
939 fprintf(f, " cache infos:\n");
940 fprintf(f, " cache_size=%"PRIu32"\n", mp->cache_size);
942 if (mp->cache_size == 0)
945 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
946 cache_count = mp->local_cache[lcore_id].len;
947 fprintf(f, " cache_count[%u]=%u\n", lcore_id, cache_count);
948 count += cache_count;
950 fprintf(f, " total_cache_count=%u\n", count);
954 #ifndef __INTEL_COMPILER
955 #pragma GCC diagnostic ignored "-Wcast-qual"
958 /* check and update cookies or panic (internal) */
959 void rte_mempool_check_cookies(const struct rte_mempool *mp,
960 void * const *obj_table_const, unsigned n, int free)
962 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
963 struct rte_mempool_objhdr *hdr;
964 struct rte_mempool_objtlr *tlr;
970 /* Force to drop the "const" attribute. This is done only when
971 * DEBUG is enabled */
972 tmp = (void *) obj_table_const;
973 obj_table = (void **) tmp;
978 if (rte_mempool_from_obj(obj) != mp)
979 rte_panic("MEMPOOL: object is owned by another "
982 hdr = __mempool_get_header(obj);
983 cookie = hdr->cookie;
986 if (cookie != RTE_MEMPOOL_HEADER_COOKIE1) {
987 RTE_LOG(CRIT, MEMPOOL,
988 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
989 obj, (const void *) mp, cookie);
990 rte_panic("MEMPOOL: bad header cookie (put)\n");
992 hdr->cookie = RTE_MEMPOOL_HEADER_COOKIE2;
993 } else if (free == 1) {
994 if (cookie != RTE_MEMPOOL_HEADER_COOKIE2) {
995 RTE_LOG(CRIT, MEMPOOL,
996 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
997 obj, (const void *) mp, cookie);
998 rte_panic("MEMPOOL: bad header cookie (get)\n");
1000 hdr->cookie = RTE_MEMPOOL_HEADER_COOKIE1;
1001 } else if (free == 2) {
1002 if (cookie != RTE_MEMPOOL_HEADER_COOKIE1 &&
1003 cookie != RTE_MEMPOOL_HEADER_COOKIE2) {
1004 RTE_LOG(CRIT, MEMPOOL,
1005 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1006 obj, (const void *) mp, cookie);
1007 rte_panic("MEMPOOL: bad header cookie (audit)\n");
1010 tlr = __mempool_get_trailer(obj);
1011 cookie = tlr->cookie;
1012 if (cookie != RTE_MEMPOOL_TRAILER_COOKIE) {
1013 RTE_LOG(CRIT, MEMPOOL,
1014 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1015 obj, (const void *) mp, cookie);
1016 rte_panic("MEMPOOL: bad trailer cookie\n");
1021 RTE_SET_USED(obj_table_const);
1027 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1029 mempool_obj_audit(struct rte_mempool *mp, __rte_unused void *opaque,
1030 void *obj, __rte_unused unsigned idx)
1032 __mempool_check_cookies(mp, &obj, 1, 2);
1036 mempool_audit_cookies(struct rte_mempool *mp)
1040 num = rte_mempool_obj_iter(mp, mempool_obj_audit, NULL);
1041 if (num != mp->size) {
1042 rte_panic("rte_mempool_obj_iter(mempool=%p, size=%u) "
1043 "iterated only over %u elements\n",
1048 #define mempool_audit_cookies(mp) do {} while(0)
1051 #ifndef __INTEL_COMPILER
1052 #pragma GCC diagnostic error "-Wcast-qual"
1055 /* check cookies before and after objects */
1057 mempool_audit_cache(const struct rte_mempool *mp)
1059 /* check cache size consistency */
1062 if (mp->cache_size == 0)
1065 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1066 if (mp->local_cache[lcore_id].len > mp->cache_flushthresh) {
1067 RTE_LOG(CRIT, MEMPOOL, "badness on cache[%u]\n",
1069 rte_panic("MEMPOOL: invalid cache len\n");
1074 /* check the consistency of mempool (size, cookies, ...) */
1076 rte_mempool_audit(struct rte_mempool *mp)
1078 mempool_audit_cache(mp);
1079 mempool_audit_cookies(mp);
1081 /* For case where mempool DEBUG is not set, and cache size is 0 */
1085 /* dump the status of the mempool on the console */
1087 rte_mempool_dump(FILE *f, struct rte_mempool *mp)
1089 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1090 struct rte_mempool_debug_stats sum;
1093 struct rte_mempool_memhdr *memhdr;
1094 unsigned common_count;
1095 unsigned cache_count;
1098 RTE_ASSERT(f != NULL);
1099 RTE_ASSERT(mp != NULL);
1101 fprintf(f, "mempool <%s>@%p\n", mp->name, mp);
1102 fprintf(f, " flags=%x\n", mp->flags);
1103 fprintf(f, " pool=%p\n", mp->pool_data);
1104 fprintf(f, " phys_addr=0x%" PRIx64 "\n", mp->mz->phys_addr);
1105 fprintf(f, " nb_mem_chunks=%u\n", mp->nb_mem_chunks);
1106 fprintf(f, " size=%"PRIu32"\n", mp->size);
1107 fprintf(f, " populated_size=%"PRIu32"\n", mp->populated_size);
1108 fprintf(f, " header_size=%"PRIu32"\n", mp->header_size);
1109 fprintf(f, " elt_size=%"PRIu32"\n", mp->elt_size);
1110 fprintf(f, " trailer_size=%"PRIu32"\n", mp->trailer_size);
1111 fprintf(f, " total_obj_size=%"PRIu32"\n",
1112 mp->header_size + mp->elt_size + mp->trailer_size);
1114 fprintf(f, " private_data_size=%"PRIu32"\n", mp->private_data_size);
1116 STAILQ_FOREACH(memhdr, &mp->mem_list, next)
1117 mem_len += memhdr->len;
1119 fprintf(f, " avg bytes/object=%#Lf\n",
1120 (long double)mem_len / mp->size);
1123 cache_count = rte_mempool_dump_cache(f, mp);
1124 common_count = rte_mempool_ops_get_count(mp);
1125 if ((cache_count + common_count) > mp->size)
1126 common_count = mp->size - cache_count;
1127 fprintf(f, " common_pool_count=%u\n", common_count);
1129 /* sum and dump statistics */
1130 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1131 memset(&sum, 0, sizeof(sum));
1132 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1133 sum.put_bulk += mp->stats[lcore_id].put_bulk;
1134 sum.put_objs += mp->stats[lcore_id].put_objs;
1135 sum.get_success_bulk += mp->stats[lcore_id].get_success_bulk;
1136 sum.get_success_objs += mp->stats[lcore_id].get_success_objs;
1137 sum.get_fail_bulk += mp->stats[lcore_id].get_fail_bulk;
1138 sum.get_fail_objs += mp->stats[lcore_id].get_fail_objs;
1140 fprintf(f, " stats:\n");
1141 fprintf(f, " put_bulk=%"PRIu64"\n", sum.put_bulk);
1142 fprintf(f, " put_objs=%"PRIu64"\n", sum.put_objs);
1143 fprintf(f, " get_success_bulk=%"PRIu64"\n", sum.get_success_bulk);
1144 fprintf(f, " get_success_objs=%"PRIu64"\n", sum.get_success_objs);
1145 fprintf(f, " get_fail_bulk=%"PRIu64"\n", sum.get_fail_bulk);
1146 fprintf(f, " get_fail_objs=%"PRIu64"\n", sum.get_fail_objs);
1148 fprintf(f, " no statistics available\n");
1151 rte_mempool_audit(mp);
1154 /* dump the status of all mempools on the console */
1156 rte_mempool_list_dump(FILE *f)
1158 struct rte_mempool *mp = NULL;
1159 struct rte_tailq_entry *te;
1160 struct rte_mempool_list *mempool_list;
1162 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1164 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1166 TAILQ_FOREACH(te, mempool_list, next) {
1167 mp = (struct rte_mempool *) te->data;
1168 rte_mempool_dump(f, mp);
1171 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);
1174 /* search a mempool from its name */
1175 struct rte_mempool *
1176 rte_mempool_lookup(const char *name)
1178 struct rte_mempool *mp = NULL;
1179 struct rte_tailq_entry *te;
1180 struct rte_mempool_list *mempool_list;
1182 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1184 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1186 TAILQ_FOREACH(te, mempool_list, next) {
1187 mp = (struct rte_mempool *) te->data;
1188 if (strncmp(name, mp->name, RTE_MEMPOOL_NAMESIZE) == 0)
1192 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);
1202 void rte_mempool_walk(void (*func)(struct rte_mempool *, void *),
1205 struct rte_tailq_entry *te = NULL;
1206 struct rte_mempool_list *mempool_list;
1208 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1210 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1212 TAILQ_FOREACH(te, mempool_list, next) {
1213 (*func)((struct rte_mempool *) te->data, arg);
1216 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);