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 memhdr = rte_zmalloc("MEMPOOL_MEMHDR", sizeof(*memhdr), 0);
377 memhdr->addr = vaddr;
378 memhdr->phys_addr = paddr;
380 memhdr->free_cb = free_cb;
381 memhdr->opaque = opaque;
383 if (mp->flags & MEMPOOL_F_NO_CACHE_ALIGN)
384 off = RTE_PTR_ALIGN_CEIL(vaddr, 8) - vaddr;
386 off = RTE_PTR_ALIGN_CEIL(vaddr, RTE_CACHE_LINE_SIZE) - vaddr;
388 while (off + total_elt_sz <= len && mp->populated_size < mp->size) {
389 off += mp->header_size;
390 if (paddr == RTE_BAD_PHYS_ADDR)
391 mempool_add_elem(mp, (char *)vaddr + off,
394 mempool_add_elem(mp, (char *)vaddr + off, paddr + off);
395 off += mp->elt_size + mp->trailer_size;
399 /* not enough room to store one object */
403 STAILQ_INSERT_TAIL(&mp->mem_list, memhdr, next);
408 /* Add objects in the pool, using a table of physical pages. Return the
409 * number of objects added, or a negative value on error.
412 rte_mempool_populate_phys_tab(struct rte_mempool *mp, char *vaddr,
413 const phys_addr_t paddr[], uint32_t pg_num, uint32_t pg_shift,
414 rte_mempool_memchunk_free_cb_t *free_cb, void *opaque)
418 size_t pg_sz = (size_t)1 << pg_shift;
420 /* mempool must not be populated */
421 if (mp->nb_mem_chunks != 0)
424 if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG)
425 return rte_mempool_populate_phys(mp, vaddr, RTE_BAD_PHYS_ADDR,
426 pg_num * pg_sz, free_cb, opaque);
428 for (i = 0; i < pg_num && mp->populated_size < mp->size; i += n) {
430 /* populate with the largest group of contiguous pages */
431 for (n = 1; (i + n) < pg_num &&
432 paddr[i + n - 1] + pg_sz == paddr[i + n]; n++)
435 ret = rte_mempool_populate_phys(mp, vaddr + i * pg_sz,
436 paddr[i], n * pg_sz, free_cb, opaque);
438 rte_mempool_free_memchunks(mp);
441 /* no need to call the free callback for next chunks */
448 /* Populate the mempool with a virtual area. Return the number of
449 * objects added, or a negative value on error.
452 rte_mempool_populate_virt(struct rte_mempool *mp, char *addr,
453 size_t len, size_t pg_sz, rte_mempool_memchunk_free_cb_t *free_cb,
457 size_t off, phys_len;
460 /* mempool must not be populated */
461 if (mp->nb_mem_chunks != 0)
463 /* address and len must be page-aligned */
464 if (RTE_PTR_ALIGN_CEIL(addr, pg_sz) != addr)
466 if (RTE_ALIGN_CEIL(len, pg_sz) != len)
469 if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG)
470 return rte_mempool_populate_phys(mp, addr, RTE_BAD_PHYS_ADDR,
471 len, free_cb, opaque);
473 for (off = 0; off + pg_sz <= len &&
474 mp->populated_size < mp->size; off += phys_len) {
476 paddr = rte_mem_virt2phy(addr + off);
477 /* required for xen_dom0 to get the machine address */
478 paddr = rte_mem_phy2mch(-1, paddr);
480 if (paddr == RTE_BAD_PHYS_ADDR && rte_eal_has_hugepages()) {
485 /* populate with the largest group of contiguous pages */
486 for (phys_len = pg_sz; off + phys_len < len; phys_len += pg_sz) {
487 phys_addr_t paddr_tmp;
489 paddr_tmp = rte_mem_virt2phy(addr + off + phys_len);
490 paddr_tmp = rte_mem_phy2mch(-1, paddr_tmp);
492 if (paddr_tmp != paddr + phys_len)
496 ret = rte_mempool_populate_phys(mp, addr + off, paddr,
497 phys_len, free_cb, opaque);
500 /* no need to call the free callback for next chunks */
508 rte_mempool_free_memchunks(mp);
512 /* Default function to populate the mempool: allocate memory in memzones,
513 * and populate them. Return the number of objects added, or a negative
517 rte_mempool_populate_default(struct rte_mempool *mp)
519 unsigned int mz_flags = RTE_MEMZONE_1GB|RTE_MEMZONE_SIZE_HINT_ONLY;
520 char mz_name[RTE_MEMZONE_NAMESIZE];
521 const struct rte_memzone *mz;
522 size_t size, total_elt_sz, align, pg_sz, pg_shift;
525 unsigned int mp_flags;
528 /* mempool must not be populated */
529 if (mp->nb_mem_chunks != 0)
532 /* Get mempool capabilities */
534 ret = rte_mempool_ops_get_capabilities(mp, &mp_flags);
536 RTE_LOG(DEBUG, MEMPOOL, "get_capability not supported for %s\n",
541 /* update mempool capabilities */
542 mp->flags |= mp_flags;
544 if (rte_xen_dom0_supported()) {
545 pg_sz = RTE_PGSIZE_2M;
546 pg_shift = rte_bsf32(pg_sz);
548 } else 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,
563 ret = snprintf(mz_name, sizeof(mz_name),
564 RTE_MEMPOOL_MZ_FORMAT "_%d", mp->name, mz_id);
565 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
570 mz = rte_memzone_reserve_aligned(mz_name, size,
571 mp->socket_id, mz_flags, align);
572 /* not enough memory, retry with the biggest zone we have */
574 mz = rte_memzone_reserve_aligned(mz_name, 0,
575 mp->socket_id, mz_flags, align);
581 if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG)
582 paddr = RTE_BAD_PHYS_ADDR;
584 paddr = mz->phys_addr;
586 if (rte_eal_has_hugepages() && !rte_xen_dom0_supported())
587 ret = rte_mempool_populate_phys(mp, mz->addr,
589 rte_mempool_memchunk_mz_free,
590 (void *)(uintptr_t)mz);
592 ret = rte_mempool_populate_virt(mp, mz->addr,
594 rte_mempool_memchunk_mz_free,
595 (void *)(uintptr_t)mz);
597 rte_memzone_free(mz);
605 rte_mempool_free_memchunks(mp);
609 /* return the memory size required for mempool objects in anonymous mem */
611 get_anon_size(const struct rte_mempool *mp)
613 size_t size, total_elt_sz, pg_sz, pg_shift;
615 pg_sz = getpagesize();
616 pg_shift = rte_bsf32(pg_sz);
617 total_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;
618 size = rte_mempool_xmem_size(mp->size, total_elt_sz, pg_shift,
624 /* unmap a memory zone mapped by rte_mempool_populate_anon() */
626 rte_mempool_memchunk_anon_free(struct rte_mempool_memhdr *memhdr,
629 munmap(opaque, get_anon_size(memhdr->mp));
632 /* populate the mempool with an anonymous mapping */
634 rte_mempool_populate_anon(struct rte_mempool *mp)
640 /* mempool is already populated, error */
641 if (!STAILQ_EMPTY(&mp->mem_list)) {
646 /* get chunk of virtually continuous memory */
647 size = get_anon_size(mp);
648 addr = mmap(NULL, size, PROT_READ | PROT_WRITE,
649 MAP_SHARED | MAP_ANONYMOUS, -1, 0);
650 if (addr == MAP_FAILED) {
654 /* can't use MMAP_LOCKED, it does not exist on BSD */
655 if (mlock(addr, size) < 0) {
661 ret = rte_mempool_populate_virt(mp, addr, size, getpagesize(),
662 rte_mempool_memchunk_anon_free, addr);
666 return mp->populated_size;
669 rte_mempool_free_memchunks(mp);
675 rte_mempool_free(struct rte_mempool *mp)
677 struct rte_mempool_list *mempool_list = NULL;
678 struct rte_tailq_entry *te;
683 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
684 rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
685 /* find out tailq entry */
686 TAILQ_FOREACH(te, mempool_list, next) {
687 if (te->data == (void *)mp)
692 TAILQ_REMOVE(mempool_list, te, next);
695 rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
697 rte_mempool_free_memchunks(mp);
698 rte_mempool_ops_free(mp);
699 rte_memzone_free(mp->mz);
703 mempool_cache_init(struct rte_mempool_cache *cache, uint32_t size)
706 cache->flushthresh = CALC_CACHE_FLUSHTHRESH(size);
711 * Create and initialize a cache for objects that are retrieved from and
712 * returned to an underlying mempool. This structure is identical to the
713 * local_cache[lcore_id] pointed to by the mempool structure.
715 struct rte_mempool_cache *
716 rte_mempool_cache_create(uint32_t size, int socket_id)
718 struct rte_mempool_cache *cache;
720 if (size == 0 || size > RTE_MEMPOOL_CACHE_MAX_SIZE) {
725 cache = rte_zmalloc_socket("MEMPOOL_CACHE", sizeof(*cache),
726 RTE_CACHE_LINE_SIZE, socket_id);
728 RTE_LOG(ERR, MEMPOOL, "Cannot allocate mempool cache.\n");
733 mempool_cache_init(cache, size);
739 * Free a cache. It's the responsibility of the user to make sure that any
740 * remaining objects in the cache are flushed to the corresponding
744 rte_mempool_cache_free(struct rte_mempool_cache *cache)
749 /* create an empty mempool */
751 rte_mempool_create_empty(const char *name, unsigned n, unsigned elt_size,
752 unsigned cache_size, unsigned private_data_size,
753 int socket_id, unsigned flags)
755 char mz_name[RTE_MEMZONE_NAMESIZE];
756 struct rte_mempool_list *mempool_list;
757 struct rte_mempool *mp = NULL;
758 struct rte_tailq_entry *te = NULL;
759 const struct rte_memzone *mz = NULL;
761 unsigned int mz_flags = RTE_MEMZONE_1GB|RTE_MEMZONE_SIZE_HINT_ONLY;
762 struct rte_mempool_objsz objsz;
766 /* compilation-time checks */
767 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool) &
768 RTE_CACHE_LINE_MASK) != 0);
769 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_cache) &
770 RTE_CACHE_LINE_MASK) != 0);
771 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
772 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_debug_stats) &
773 RTE_CACHE_LINE_MASK) != 0);
774 RTE_BUILD_BUG_ON((offsetof(struct rte_mempool, stats) &
775 RTE_CACHE_LINE_MASK) != 0);
778 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
780 /* asked cache too big */
781 if (cache_size > RTE_MEMPOOL_CACHE_MAX_SIZE ||
782 CALC_CACHE_FLUSHTHRESH(cache_size) > n) {
787 /* "no cache align" imply "no spread" */
788 if (flags & MEMPOOL_F_NO_CACHE_ALIGN)
789 flags |= MEMPOOL_F_NO_SPREAD;
791 /* calculate mempool object sizes. */
792 if (!rte_mempool_calc_obj_size(elt_size, flags, &objsz)) {
797 rte_rwlock_write_lock(RTE_EAL_MEMPOOL_RWLOCK);
800 * reserve a memory zone for this mempool: private data is
803 private_data_size = (private_data_size +
804 RTE_MEMPOOL_ALIGN_MASK) & (~RTE_MEMPOOL_ALIGN_MASK);
807 /* try to allocate tailq entry */
808 te = rte_zmalloc("MEMPOOL_TAILQ_ENTRY", sizeof(*te), 0);
810 RTE_LOG(ERR, MEMPOOL, "Cannot allocate tailq entry!\n");
814 mempool_size = MEMPOOL_HEADER_SIZE(mp, cache_size);
815 mempool_size += private_data_size;
816 mempool_size = RTE_ALIGN_CEIL(mempool_size, RTE_MEMPOOL_ALIGN);
818 ret = snprintf(mz_name, sizeof(mz_name), RTE_MEMPOOL_MZ_FORMAT, name);
819 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
820 rte_errno = ENAMETOOLONG;
824 mz = rte_memzone_reserve(mz_name, mempool_size, socket_id, mz_flags);
828 /* init the mempool structure */
830 memset(mp, 0, MEMPOOL_HEADER_SIZE(mp, cache_size));
831 ret = snprintf(mp->name, sizeof(mp->name), "%s", name);
832 if (ret < 0 || ret >= (int)sizeof(mp->name)) {
833 rte_errno = ENAMETOOLONG;
839 mp->socket_id = socket_id;
840 mp->elt_size = objsz.elt_size;
841 mp->header_size = objsz.header_size;
842 mp->trailer_size = objsz.trailer_size;
843 /* Size of default caches, zero means disabled. */
844 mp->cache_size = cache_size;
845 mp->private_data_size = private_data_size;
846 STAILQ_INIT(&mp->elt_list);
847 STAILQ_INIT(&mp->mem_list);
850 * local_cache pointer is set even if cache_size is zero.
851 * The local_cache points to just past the elt_pa[] array.
853 mp->local_cache = (struct rte_mempool_cache *)
854 RTE_PTR_ADD(mp, MEMPOOL_HEADER_SIZE(mp, 0));
856 /* Init all default caches. */
857 if (cache_size != 0) {
858 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
859 mempool_cache_init(&mp->local_cache[lcore_id],
865 rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
866 TAILQ_INSERT_TAIL(mempool_list, te, next);
867 rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
868 rte_rwlock_write_unlock(RTE_EAL_MEMPOOL_RWLOCK);
873 rte_rwlock_write_unlock(RTE_EAL_MEMPOOL_RWLOCK);
875 rte_mempool_free(mp);
879 /* create the mempool */
881 rte_mempool_create(const char *name, unsigned n, unsigned elt_size,
882 unsigned cache_size, unsigned private_data_size,
883 rte_mempool_ctor_t *mp_init, void *mp_init_arg,
884 rte_mempool_obj_cb_t *obj_init, void *obj_init_arg,
885 int socket_id, unsigned flags)
888 struct rte_mempool *mp;
890 mp = rte_mempool_create_empty(name, n, elt_size, cache_size,
891 private_data_size, socket_id, flags);
896 * Since we have 4 combinations of the SP/SC/MP/MC examine the flags to
897 * set the correct index into the table of ops structs.
899 if ((flags & MEMPOOL_F_SP_PUT) && (flags & MEMPOOL_F_SC_GET))
900 ret = rte_mempool_set_ops_byname(mp, "ring_sp_sc", NULL);
901 else if (flags & MEMPOOL_F_SP_PUT)
902 ret = rte_mempool_set_ops_byname(mp, "ring_sp_mc", NULL);
903 else if (flags & MEMPOOL_F_SC_GET)
904 ret = rte_mempool_set_ops_byname(mp, "ring_mp_sc", NULL);
906 ret = rte_mempool_set_ops_byname(mp, "ring_mp_mc", NULL);
911 /* call the mempool priv initializer */
913 mp_init(mp, mp_init_arg);
915 if (rte_mempool_populate_default(mp) < 0)
918 /* call the object initializers */
920 rte_mempool_obj_iter(mp, obj_init, obj_init_arg);
925 rte_mempool_free(mp);
930 * Create the mempool over already allocated chunk of memory.
931 * That external memory buffer can consists of physically disjoint pages.
932 * Setting vaddr to NULL, makes mempool to fallback to rte_mempool_create()
936 rte_mempool_xmem_create(const char *name, unsigned n, unsigned elt_size,
937 unsigned cache_size, unsigned private_data_size,
938 rte_mempool_ctor_t *mp_init, void *mp_init_arg,
939 rte_mempool_obj_cb_t *obj_init, void *obj_init_arg,
940 int socket_id, unsigned flags, void *vaddr,
941 const phys_addr_t paddr[], uint32_t pg_num, uint32_t pg_shift)
943 struct rte_mempool *mp = NULL;
946 /* no virtual address supplied, use rte_mempool_create() */
948 return rte_mempool_create(name, n, elt_size, cache_size,
949 private_data_size, mp_init, mp_init_arg,
950 obj_init, obj_init_arg, socket_id, flags);
952 /* check that we have both VA and PA */
958 /* Check that pg_shift parameter is valid. */
959 if (pg_shift > MEMPOOL_PG_SHIFT_MAX) {
964 mp = rte_mempool_create_empty(name, n, elt_size, cache_size,
965 private_data_size, socket_id, flags);
969 /* call the mempool priv initializer */
971 mp_init(mp, mp_init_arg);
973 ret = rte_mempool_populate_phys_tab(mp, vaddr, paddr, pg_num, pg_shift,
975 if (ret < 0 || ret != (int)mp->size)
978 /* call the object initializers */
980 rte_mempool_obj_iter(mp, obj_init, obj_init_arg);
985 rte_mempool_free(mp);
989 /* Return the number of entries in the mempool */
991 rte_mempool_avail_count(const struct rte_mempool *mp)
996 count = rte_mempool_ops_get_count(mp);
998 if (mp->cache_size == 0)
1001 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
1002 count += mp->local_cache[lcore_id].len;
1005 * due to race condition (access to len is not locked), the
1006 * total can be greater than size... so fix the result
1008 if (count > mp->size)
1013 /* return the number of entries allocated from the mempool */
1015 rte_mempool_in_use_count(const struct rte_mempool *mp)
1017 return mp->size - rte_mempool_avail_count(mp);
1020 /* dump the cache status */
1022 rte_mempool_dump_cache(FILE *f, const struct rte_mempool *mp)
1026 unsigned cache_count;
1028 fprintf(f, " internal cache infos:\n");
1029 fprintf(f, " cache_size=%"PRIu32"\n", mp->cache_size);
1031 if (mp->cache_size == 0)
1034 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1035 cache_count = mp->local_cache[lcore_id].len;
1036 fprintf(f, " cache_count[%u]=%"PRIu32"\n",
1037 lcore_id, cache_count);
1038 count += cache_count;
1040 fprintf(f, " total_cache_count=%u\n", count);
1044 #ifndef __INTEL_COMPILER
1045 #pragma GCC diagnostic ignored "-Wcast-qual"
1048 /* check and update cookies or panic (internal) */
1049 void rte_mempool_check_cookies(const struct rte_mempool *mp,
1050 void * const *obj_table_const, unsigned n, int free)
1052 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1053 struct rte_mempool_objhdr *hdr;
1054 struct rte_mempool_objtlr *tlr;
1060 /* Force to drop the "const" attribute. This is done only when
1061 * DEBUG is enabled */
1062 tmp = (void *) obj_table_const;
1068 if (rte_mempool_from_obj(obj) != mp)
1069 rte_panic("MEMPOOL: object is owned by another "
1072 hdr = __mempool_get_header(obj);
1073 cookie = hdr->cookie;
1076 if (cookie != RTE_MEMPOOL_HEADER_COOKIE1) {
1077 RTE_LOG(CRIT, MEMPOOL,
1078 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1079 obj, (const void *) mp, cookie);
1080 rte_panic("MEMPOOL: bad header cookie (put)\n");
1082 hdr->cookie = RTE_MEMPOOL_HEADER_COOKIE2;
1083 } else if (free == 1) {
1084 if (cookie != RTE_MEMPOOL_HEADER_COOKIE2) {
1085 RTE_LOG(CRIT, MEMPOOL,
1086 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1087 obj, (const void *) mp, cookie);
1088 rte_panic("MEMPOOL: bad header cookie (get)\n");
1090 hdr->cookie = RTE_MEMPOOL_HEADER_COOKIE1;
1091 } else if (free == 2) {
1092 if (cookie != RTE_MEMPOOL_HEADER_COOKIE1 &&
1093 cookie != RTE_MEMPOOL_HEADER_COOKIE2) {
1094 RTE_LOG(CRIT, MEMPOOL,
1095 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1096 obj, (const void *) mp, cookie);
1097 rte_panic("MEMPOOL: bad header cookie (audit)\n");
1100 tlr = __mempool_get_trailer(obj);
1101 cookie = tlr->cookie;
1102 if (cookie != RTE_MEMPOOL_TRAILER_COOKIE) {
1103 RTE_LOG(CRIT, MEMPOOL,
1104 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1105 obj, (const void *) mp, cookie);
1106 rte_panic("MEMPOOL: bad trailer cookie\n");
1111 RTE_SET_USED(obj_table_const);
1117 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1119 mempool_obj_audit(struct rte_mempool *mp, __rte_unused void *opaque,
1120 void *obj, __rte_unused unsigned idx)
1122 __mempool_check_cookies(mp, &obj, 1, 2);
1126 mempool_audit_cookies(struct rte_mempool *mp)
1130 num = rte_mempool_obj_iter(mp, mempool_obj_audit, NULL);
1131 if (num != mp->size) {
1132 rte_panic("rte_mempool_obj_iter(mempool=%p, size=%u) "
1133 "iterated only over %u elements\n",
1138 #define mempool_audit_cookies(mp) do {} while(0)
1141 #ifndef __INTEL_COMPILER
1142 #pragma GCC diagnostic error "-Wcast-qual"
1145 /* check cookies before and after objects */
1147 mempool_audit_cache(const struct rte_mempool *mp)
1149 /* check cache size consistency */
1152 if (mp->cache_size == 0)
1155 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1156 const struct rte_mempool_cache *cache;
1157 cache = &mp->local_cache[lcore_id];
1158 if (cache->len > cache->flushthresh) {
1159 RTE_LOG(CRIT, MEMPOOL, "badness on cache[%u]\n",
1161 rte_panic("MEMPOOL: invalid cache len\n");
1166 /* check the consistency of mempool (size, cookies, ...) */
1168 rte_mempool_audit(struct rte_mempool *mp)
1170 mempool_audit_cache(mp);
1171 mempool_audit_cookies(mp);
1173 /* For case where mempool DEBUG is not set, and cache size is 0 */
1177 /* dump the status of the mempool on the console */
1179 rte_mempool_dump(FILE *f, struct rte_mempool *mp)
1181 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1182 struct rte_mempool_debug_stats sum;
1185 struct rte_mempool_memhdr *memhdr;
1186 unsigned common_count;
1187 unsigned cache_count;
1190 RTE_ASSERT(f != NULL);
1191 RTE_ASSERT(mp != NULL);
1193 fprintf(f, "mempool <%s>@%p\n", mp->name, mp);
1194 fprintf(f, " flags=%x\n", mp->flags);
1195 fprintf(f, " pool=%p\n", mp->pool_data);
1196 fprintf(f, " phys_addr=0x%" PRIx64 "\n", mp->mz->phys_addr);
1197 fprintf(f, " nb_mem_chunks=%u\n", mp->nb_mem_chunks);
1198 fprintf(f, " size=%"PRIu32"\n", mp->size);
1199 fprintf(f, " populated_size=%"PRIu32"\n", mp->populated_size);
1200 fprintf(f, " header_size=%"PRIu32"\n", mp->header_size);
1201 fprintf(f, " elt_size=%"PRIu32"\n", mp->elt_size);
1202 fprintf(f, " trailer_size=%"PRIu32"\n", mp->trailer_size);
1203 fprintf(f, " total_obj_size=%"PRIu32"\n",
1204 mp->header_size + mp->elt_size + mp->trailer_size);
1206 fprintf(f, " private_data_size=%"PRIu32"\n", mp->private_data_size);
1208 STAILQ_FOREACH(memhdr, &mp->mem_list, next)
1209 mem_len += memhdr->len;
1211 fprintf(f, " avg bytes/object=%#Lf\n",
1212 (long double)mem_len / mp->size);
1215 cache_count = rte_mempool_dump_cache(f, mp);
1216 common_count = rte_mempool_ops_get_count(mp);
1217 if ((cache_count + common_count) > mp->size)
1218 common_count = mp->size - cache_count;
1219 fprintf(f, " common_pool_count=%u\n", common_count);
1221 /* sum and dump statistics */
1222 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1223 memset(&sum, 0, sizeof(sum));
1224 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1225 sum.put_bulk += mp->stats[lcore_id].put_bulk;
1226 sum.put_objs += mp->stats[lcore_id].put_objs;
1227 sum.get_success_bulk += mp->stats[lcore_id].get_success_bulk;
1228 sum.get_success_objs += mp->stats[lcore_id].get_success_objs;
1229 sum.get_fail_bulk += mp->stats[lcore_id].get_fail_bulk;
1230 sum.get_fail_objs += mp->stats[lcore_id].get_fail_objs;
1232 fprintf(f, " stats:\n");
1233 fprintf(f, " put_bulk=%"PRIu64"\n", sum.put_bulk);
1234 fprintf(f, " put_objs=%"PRIu64"\n", sum.put_objs);
1235 fprintf(f, " get_success_bulk=%"PRIu64"\n", sum.get_success_bulk);
1236 fprintf(f, " get_success_objs=%"PRIu64"\n", sum.get_success_objs);
1237 fprintf(f, " get_fail_bulk=%"PRIu64"\n", sum.get_fail_bulk);
1238 fprintf(f, " get_fail_objs=%"PRIu64"\n", sum.get_fail_objs);
1240 fprintf(f, " no statistics available\n");
1243 rte_mempool_audit(mp);
1246 /* dump the status of all mempools on the console */
1248 rte_mempool_list_dump(FILE *f)
1250 struct rte_mempool *mp = NULL;
1251 struct rte_tailq_entry *te;
1252 struct rte_mempool_list *mempool_list;
1254 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1256 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1258 TAILQ_FOREACH(te, mempool_list, next) {
1259 mp = (struct rte_mempool *) te->data;
1260 rte_mempool_dump(f, mp);
1263 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);
1266 /* search a mempool from its name */
1267 struct rte_mempool *
1268 rte_mempool_lookup(const char *name)
1270 struct rte_mempool *mp = NULL;
1271 struct rte_tailq_entry *te;
1272 struct rte_mempool_list *mempool_list;
1274 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1276 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1278 TAILQ_FOREACH(te, mempool_list, next) {
1279 mp = (struct rte_mempool *) te->data;
1280 if (strncmp(name, mp->name, RTE_MEMPOOL_NAMESIZE) == 0)
1284 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);
1294 void rte_mempool_walk(void (*func)(struct rte_mempool *, void *),
1297 struct rte_tailq_entry *te = NULL;
1298 struct rte_mempool_list *mempool_list;
1301 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1303 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1305 TAILQ_FOREACH_SAFE(te, mempool_list, next, tmp_te) {
1306 (*func)((struct rte_mempool *) te->data, arg);
1309 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);