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,
244 size_t obj_per_page, pg_num, pg_sz;
247 mask = MEMPOOL_F_CAPA_BLK_ALIGNED_OBJECTS | MEMPOOL_F_CAPA_PHYS_CONTIG;
248 if ((flags & mask) == mask)
249 /* alignment need one additional object */
252 if (total_elt_sz == 0)
256 return total_elt_sz * elt_num;
258 pg_sz = (size_t)1 << pg_shift;
259 obj_per_page = pg_sz / total_elt_sz;
260 if (obj_per_page == 0)
261 return RTE_ALIGN_CEIL(total_elt_sz, pg_sz) * elt_num;
263 pg_num = (elt_num + obj_per_page - 1) / obj_per_page;
264 return pg_num << pg_shift;
268 * Calculate how much memory would be actually required with the
269 * given memory footprint to store required number of elements.
272 rte_mempool_xmem_usage(__rte_unused void *vaddr, uint32_t elt_num,
273 size_t total_elt_sz, const phys_addr_t paddr[], uint32_t pg_num,
274 uint32_t pg_shift, unsigned int flags)
276 uint32_t elt_cnt = 0;
277 phys_addr_t start, end;
279 size_t pg_sz = (size_t)1 << pg_shift;
282 mask = MEMPOOL_F_CAPA_BLK_ALIGNED_OBJECTS | MEMPOOL_F_CAPA_PHYS_CONTIG;
283 if ((flags & mask) == mask)
284 /* alignment need one additional object */
287 /* if paddr is NULL, assume contiguous memory */
290 end = pg_sz * pg_num;
294 end = paddr[0] + pg_sz;
297 while (elt_cnt < elt_num) {
299 if (end - start >= total_elt_sz) {
300 /* enough contiguous memory, add an object */
301 start += total_elt_sz;
303 } else if (paddr_idx < pg_num) {
304 /* no room to store one obj, add a page */
305 if (end == paddr[paddr_idx]) {
308 start = paddr[paddr_idx];
309 end = paddr[paddr_idx] + pg_sz;
314 /* no more page, return how many elements fit */
315 return -(size_t)elt_cnt;
319 return (size_t)paddr_idx << pg_shift;
322 /* free a memchunk allocated with rte_memzone_reserve() */
324 rte_mempool_memchunk_mz_free(__rte_unused struct rte_mempool_memhdr *memhdr,
327 const struct rte_memzone *mz = opaque;
328 rte_memzone_free(mz);
331 /* Free memory chunks used by a mempool. Objects must be in pool */
333 rte_mempool_free_memchunks(struct rte_mempool *mp)
335 struct rte_mempool_memhdr *memhdr;
338 while (!STAILQ_EMPTY(&mp->elt_list)) {
339 rte_mempool_ops_dequeue_bulk(mp, &elt, 1);
341 STAILQ_REMOVE_HEAD(&mp->elt_list, next);
342 mp->populated_size--;
345 while (!STAILQ_EMPTY(&mp->mem_list)) {
346 memhdr = STAILQ_FIRST(&mp->mem_list);
347 STAILQ_REMOVE_HEAD(&mp->mem_list, next);
348 if (memhdr->free_cb != NULL)
349 memhdr->free_cb(memhdr, memhdr->opaque);
355 /* Add objects in the pool, using a physically contiguous memory
356 * zone. Return the number of objects added, or a negative value
360 rte_mempool_populate_phys(struct rte_mempool *mp, char *vaddr,
361 phys_addr_t paddr, size_t len, rte_mempool_memchunk_free_cb_t *free_cb,
364 unsigned total_elt_sz;
367 struct rte_mempool_memhdr *memhdr;
370 /* create the internal ring if not already done */
371 if ((mp->flags & MEMPOOL_F_POOL_CREATED) == 0) {
372 ret = rte_mempool_ops_alloc(mp);
375 mp->flags |= MEMPOOL_F_POOL_CREATED;
378 /* mempool is already populated */
379 if (mp->populated_size >= mp->size)
382 total_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;
384 /* Detect pool area has sufficient space for elements */
385 if (mp->flags & MEMPOOL_F_CAPA_PHYS_CONTIG) {
386 if (len < total_elt_sz * mp->size) {
387 RTE_LOG(ERR, MEMPOOL,
388 "pool area %" PRIx64 " not enough\n",
394 memhdr = rte_zmalloc("MEMPOOL_MEMHDR", sizeof(*memhdr), 0);
399 memhdr->addr = vaddr;
400 memhdr->phys_addr = paddr;
402 memhdr->free_cb = free_cb;
403 memhdr->opaque = opaque;
405 if (mp->flags & MEMPOOL_F_CAPA_BLK_ALIGNED_OBJECTS)
406 /* align object start address to a multiple of total_elt_sz */
407 off = total_elt_sz - ((uintptr_t)vaddr % total_elt_sz);
408 else if (mp->flags & MEMPOOL_F_NO_CACHE_ALIGN)
409 off = RTE_PTR_ALIGN_CEIL(vaddr, 8) - vaddr;
411 off = RTE_PTR_ALIGN_CEIL(vaddr, RTE_CACHE_LINE_SIZE) - vaddr;
413 while (off + total_elt_sz <= len && mp->populated_size < mp->size) {
414 off += mp->header_size;
415 if (paddr == RTE_BAD_PHYS_ADDR)
416 mempool_add_elem(mp, (char *)vaddr + off,
419 mempool_add_elem(mp, (char *)vaddr + off, paddr + off);
420 off += mp->elt_size + mp->trailer_size;
424 /* not enough room to store one object */
428 STAILQ_INSERT_TAIL(&mp->mem_list, memhdr, next);
433 /* Add objects in the pool, using a table of physical pages. Return the
434 * number of objects added, or a negative value on error.
437 rte_mempool_populate_phys_tab(struct rte_mempool *mp, char *vaddr,
438 const phys_addr_t paddr[], uint32_t pg_num, uint32_t pg_shift,
439 rte_mempool_memchunk_free_cb_t *free_cb, void *opaque)
443 size_t pg_sz = (size_t)1 << pg_shift;
445 /* mempool must not be populated */
446 if (mp->nb_mem_chunks != 0)
449 if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG)
450 return rte_mempool_populate_phys(mp, vaddr, RTE_BAD_PHYS_ADDR,
451 pg_num * pg_sz, free_cb, opaque);
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 + n - 1] + 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 if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG)
495 return rte_mempool_populate_phys(mp, addr, RTE_BAD_PHYS_ADDR,
496 len, free_cb, opaque);
498 for (off = 0; off + pg_sz <= len &&
499 mp->populated_size < mp->size; off += phys_len) {
501 paddr = rte_mem_virt2phy(addr + off);
502 /* required for xen_dom0 to get the machine address */
503 paddr = rte_mem_phy2mch(-1, paddr);
505 if (paddr == RTE_BAD_PHYS_ADDR && rte_eal_has_hugepages()) {
510 /* populate with the largest group of contiguous pages */
511 for (phys_len = pg_sz; off + phys_len < len; phys_len += pg_sz) {
512 phys_addr_t paddr_tmp;
514 paddr_tmp = rte_mem_virt2phy(addr + off + phys_len);
515 paddr_tmp = rte_mem_phy2mch(-1, paddr_tmp);
517 if (paddr_tmp != paddr + phys_len)
521 ret = rte_mempool_populate_phys(mp, addr + off, paddr,
522 phys_len, free_cb, opaque);
525 /* no need to call the free callback for next chunks */
533 rte_mempool_free_memchunks(mp);
537 /* Default function to populate the mempool: allocate memory in memzones,
538 * and populate them. Return the number of objects added, or a negative
542 rte_mempool_populate_default(struct rte_mempool *mp)
544 unsigned int mz_flags = RTE_MEMZONE_1GB|RTE_MEMZONE_SIZE_HINT_ONLY;
545 char mz_name[RTE_MEMZONE_NAMESIZE];
546 const struct rte_memzone *mz;
547 size_t size, total_elt_sz, align, pg_sz, pg_shift;
550 unsigned int mp_flags;
553 /* mempool must not be populated */
554 if (mp->nb_mem_chunks != 0)
557 /* Get mempool capabilities */
559 ret = rte_mempool_ops_get_capabilities(mp, &mp_flags);
561 RTE_LOG(DEBUG, MEMPOOL, "get_capability not supported for %s\n",
566 /* update mempool capabilities */
567 mp->flags |= mp_flags;
569 if (rte_xen_dom0_supported()) {
570 pg_sz = RTE_PGSIZE_2M;
571 pg_shift = rte_bsf32(pg_sz);
573 } else if (rte_eal_has_hugepages()) {
574 pg_shift = 0; /* not needed, zone is physically contiguous */
576 align = RTE_CACHE_LINE_SIZE;
578 pg_sz = getpagesize();
579 pg_shift = rte_bsf32(pg_sz);
583 total_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;
584 for (mz_id = 0, n = mp->size; n > 0; mz_id++, n -= ret) {
585 size = rte_mempool_xmem_size(n, total_elt_sz, pg_shift,
588 ret = snprintf(mz_name, sizeof(mz_name),
589 RTE_MEMPOOL_MZ_FORMAT "_%d", mp->name, mz_id);
590 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
595 mz = rte_memzone_reserve_aligned(mz_name, size,
596 mp->socket_id, mz_flags, align);
597 /* not enough memory, retry with the biggest zone we have */
599 mz = rte_memzone_reserve_aligned(mz_name, 0,
600 mp->socket_id, mz_flags, align);
606 if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG)
607 paddr = RTE_BAD_PHYS_ADDR;
609 paddr = mz->phys_addr;
611 if (rte_eal_has_hugepages() && !rte_xen_dom0_supported())
612 ret = rte_mempool_populate_phys(mp, mz->addr,
614 rte_mempool_memchunk_mz_free,
615 (void *)(uintptr_t)mz);
617 ret = rte_mempool_populate_virt(mp, mz->addr,
619 rte_mempool_memchunk_mz_free,
620 (void *)(uintptr_t)mz);
622 rte_memzone_free(mz);
630 rte_mempool_free_memchunks(mp);
634 /* return the memory size required for mempool objects in anonymous mem */
636 get_anon_size(const struct rte_mempool *mp)
638 size_t size, total_elt_sz, pg_sz, pg_shift;
640 pg_sz = getpagesize();
641 pg_shift = rte_bsf32(pg_sz);
642 total_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;
643 size = rte_mempool_xmem_size(mp->size, total_elt_sz, pg_shift,
649 /* unmap a memory zone mapped by rte_mempool_populate_anon() */
651 rte_mempool_memchunk_anon_free(struct rte_mempool_memhdr *memhdr,
654 munmap(opaque, get_anon_size(memhdr->mp));
657 /* populate the mempool with an anonymous mapping */
659 rte_mempool_populate_anon(struct rte_mempool *mp)
665 /* mempool is already populated, error */
666 if (!STAILQ_EMPTY(&mp->mem_list)) {
671 /* get chunk of virtually continuous memory */
672 size = get_anon_size(mp);
673 addr = mmap(NULL, size, PROT_READ | PROT_WRITE,
674 MAP_SHARED | MAP_ANONYMOUS, -1, 0);
675 if (addr == MAP_FAILED) {
679 /* can't use MMAP_LOCKED, it does not exist on BSD */
680 if (mlock(addr, size) < 0) {
686 ret = rte_mempool_populate_virt(mp, addr, size, getpagesize(),
687 rte_mempool_memchunk_anon_free, addr);
691 return mp->populated_size;
694 rte_mempool_free_memchunks(mp);
700 rte_mempool_free(struct rte_mempool *mp)
702 struct rte_mempool_list *mempool_list = NULL;
703 struct rte_tailq_entry *te;
708 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
709 rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
710 /* find out tailq entry */
711 TAILQ_FOREACH(te, mempool_list, next) {
712 if (te->data == (void *)mp)
717 TAILQ_REMOVE(mempool_list, te, next);
720 rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
722 rte_mempool_free_memchunks(mp);
723 rte_mempool_ops_free(mp);
724 rte_memzone_free(mp->mz);
728 mempool_cache_init(struct rte_mempool_cache *cache, uint32_t size)
731 cache->flushthresh = CALC_CACHE_FLUSHTHRESH(size);
736 * Create and initialize a cache for objects that are retrieved from and
737 * returned to an underlying mempool. This structure is identical to the
738 * local_cache[lcore_id] pointed to by the mempool structure.
740 struct rte_mempool_cache *
741 rte_mempool_cache_create(uint32_t size, int socket_id)
743 struct rte_mempool_cache *cache;
745 if (size == 0 || size > RTE_MEMPOOL_CACHE_MAX_SIZE) {
750 cache = rte_zmalloc_socket("MEMPOOL_CACHE", sizeof(*cache),
751 RTE_CACHE_LINE_SIZE, socket_id);
753 RTE_LOG(ERR, MEMPOOL, "Cannot allocate mempool cache.\n");
758 mempool_cache_init(cache, size);
764 * Free a cache. It's the responsibility of the user to make sure that any
765 * remaining objects in the cache are flushed to the corresponding
769 rte_mempool_cache_free(struct rte_mempool_cache *cache)
774 /* create an empty mempool */
776 rte_mempool_create_empty(const char *name, unsigned n, unsigned elt_size,
777 unsigned cache_size, unsigned private_data_size,
778 int socket_id, unsigned flags)
780 char mz_name[RTE_MEMZONE_NAMESIZE];
781 struct rte_mempool_list *mempool_list;
782 struct rte_mempool *mp = NULL;
783 struct rte_tailq_entry *te = NULL;
784 const struct rte_memzone *mz = NULL;
786 unsigned int mz_flags = RTE_MEMZONE_1GB|RTE_MEMZONE_SIZE_HINT_ONLY;
787 struct rte_mempool_objsz objsz;
791 /* compilation-time checks */
792 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool) &
793 RTE_CACHE_LINE_MASK) != 0);
794 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_cache) &
795 RTE_CACHE_LINE_MASK) != 0);
796 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
797 RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_debug_stats) &
798 RTE_CACHE_LINE_MASK) != 0);
799 RTE_BUILD_BUG_ON((offsetof(struct rte_mempool, stats) &
800 RTE_CACHE_LINE_MASK) != 0);
803 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
805 /* asked cache too big */
806 if (cache_size > RTE_MEMPOOL_CACHE_MAX_SIZE ||
807 CALC_CACHE_FLUSHTHRESH(cache_size) > n) {
812 /* "no cache align" imply "no spread" */
813 if (flags & MEMPOOL_F_NO_CACHE_ALIGN)
814 flags |= MEMPOOL_F_NO_SPREAD;
816 /* calculate mempool object sizes. */
817 if (!rte_mempool_calc_obj_size(elt_size, flags, &objsz)) {
822 rte_rwlock_write_lock(RTE_EAL_MEMPOOL_RWLOCK);
825 * reserve a memory zone for this mempool: private data is
828 private_data_size = (private_data_size +
829 RTE_MEMPOOL_ALIGN_MASK) & (~RTE_MEMPOOL_ALIGN_MASK);
832 /* try to allocate tailq entry */
833 te = rte_zmalloc("MEMPOOL_TAILQ_ENTRY", sizeof(*te), 0);
835 RTE_LOG(ERR, MEMPOOL, "Cannot allocate tailq entry!\n");
839 mempool_size = MEMPOOL_HEADER_SIZE(mp, cache_size);
840 mempool_size += private_data_size;
841 mempool_size = RTE_ALIGN_CEIL(mempool_size, RTE_MEMPOOL_ALIGN);
843 ret = snprintf(mz_name, sizeof(mz_name), RTE_MEMPOOL_MZ_FORMAT, name);
844 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
845 rte_errno = ENAMETOOLONG;
849 mz = rte_memzone_reserve(mz_name, mempool_size, socket_id, mz_flags);
853 /* init the mempool structure */
855 memset(mp, 0, MEMPOOL_HEADER_SIZE(mp, cache_size));
856 ret = snprintf(mp->name, sizeof(mp->name), "%s", name);
857 if (ret < 0 || ret >= (int)sizeof(mp->name)) {
858 rte_errno = ENAMETOOLONG;
864 mp->socket_id = socket_id;
865 mp->elt_size = objsz.elt_size;
866 mp->header_size = objsz.header_size;
867 mp->trailer_size = objsz.trailer_size;
868 /* Size of default caches, zero means disabled. */
869 mp->cache_size = cache_size;
870 mp->private_data_size = private_data_size;
871 STAILQ_INIT(&mp->elt_list);
872 STAILQ_INIT(&mp->mem_list);
875 * local_cache pointer is set even if cache_size is zero.
876 * The local_cache points to just past the elt_pa[] array.
878 mp->local_cache = (struct rte_mempool_cache *)
879 RTE_PTR_ADD(mp, MEMPOOL_HEADER_SIZE(mp, 0));
881 /* Init all default caches. */
882 if (cache_size != 0) {
883 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
884 mempool_cache_init(&mp->local_cache[lcore_id],
890 rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
891 TAILQ_INSERT_TAIL(mempool_list, te, next);
892 rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
893 rte_rwlock_write_unlock(RTE_EAL_MEMPOOL_RWLOCK);
898 rte_rwlock_write_unlock(RTE_EAL_MEMPOOL_RWLOCK);
900 rte_mempool_free(mp);
904 /* create the mempool */
906 rte_mempool_create(const char *name, unsigned n, unsigned elt_size,
907 unsigned cache_size, unsigned private_data_size,
908 rte_mempool_ctor_t *mp_init, void *mp_init_arg,
909 rte_mempool_obj_cb_t *obj_init, void *obj_init_arg,
910 int socket_id, unsigned flags)
913 struct rte_mempool *mp;
915 mp = rte_mempool_create_empty(name, n, elt_size, cache_size,
916 private_data_size, socket_id, flags);
921 * Since we have 4 combinations of the SP/SC/MP/MC examine the flags to
922 * set the correct index into the table of ops structs.
924 if ((flags & MEMPOOL_F_SP_PUT) && (flags & MEMPOOL_F_SC_GET))
925 ret = rte_mempool_set_ops_byname(mp, "ring_sp_sc", NULL);
926 else if (flags & MEMPOOL_F_SP_PUT)
927 ret = rte_mempool_set_ops_byname(mp, "ring_sp_mc", NULL);
928 else if (flags & MEMPOOL_F_SC_GET)
929 ret = rte_mempool_set_ops_byname(mp, "ring_mp_sc", NULL);
931 ret = rte_mempool_set_ops_byname(mp, "ring_mp_mc", NULL);
936 /* call the mempool priv initializer */
938 mp_init(mp, mp_init_arg);
940 if (rte_mempool_populate_default(mp) < 0)
943 /* call the object initializers */
945 rte_mempool_obj_iter(mp, obj_init, obj_init_arg);
950 rte_mempool_free(mp);
955 * Create the mempool over already allocated chunk of memory.
956 * That external memory buffer can consists of physically disjoint pages.
957 * Setting vaddr to NULL, makes mempool to fallback to rte_mempool_create()
961 rte_mempool_xmem_create(const char *name, unsigned n, unsigned elt_size,
962 unsigned cache_size, unsigned private_data_size,
963 rte_mempool_ctor_t *mp_init, void *mp_init_arg,
964 rte_mempool_obj_cb_t *obj_init, void *obj_init_arg,
965 int socket_id, unsigned flags, void *vaddr,
966 const phys_addr_t paddr[], uint32_t pg_num, uint32_t pg_shift)
968 struct rte_mempool *mp = NULL;
971 /* no virtual address supplied, use rte_mempool_create() */
973 return rte_mempool_create(name, n, elt_size, cache_size,
974 private_data_size, mp_init, mp_init_arg,
975 obj_init, obj_init_arg, socket_id, flags);
977 /* check that we have both VA and PA */
983 /* Check that pg_shift parameter is valid. */
984 if (pg_shift > MEMPOOL_PG_SHIFT_MAX) {
989 mp = rte_mempool_create_empty(name, n, elt_size, cache_size,
990 private_data_size, socket_id, flags);
994 /* call the mempool priv initializer */
996 mp_init(mp, mp_init_arg);
998 ret = rte_mempool_populate_phys_tab(mp, vaddr, paddr, pg_num, pg_shift,
1000 if (ret < 0 || ret != (int)mp->size)
1003 /* call the object initializers */
1005 rte_mempool_obj_iter(mp, obj_init, obj_init_arg);
1010 rte_mempool_free(mp);
1014 /* Return the number of entries in the mempool */
1016 rte_mempool_avail_count(const struct rte_mempool *mp)
1021 count = rte_mempool_ops_get_count(mp);
1023 if (mp->cache_size == 0)
1026 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
1027 count += mp->local_cache[lcore_id].len;
1030 * due to race condition (access to len is not locked), the
1031 * total can be greater than size... so fix the result
1033 if (count > mp->size)
1038 /* return the number of entries allocated from the mempool */
1040 rte_mempool_in_use_count(const struct rte_mempool *mp)
1042 return mp->size - rte_mempool_avail_count(mp);
1045 /* dump the cache status */
1047 rte_mempool_dump_cache(FILE *f, const struct rte_mempool *mp)
1051 unsigned cache_count;
1053 fprintf(f, " internal cache infos:\n");
1054 fprintf(f, " cache_size=%"PRIu32"\n", mp->cache_size);
1056 if (mp->cache_size == 0)
1059 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1060 cache_count = mp->local_cache[lcore_id].len;
1061 fprintf(f, " cache_count[%u]=%"PRIu32"\n",
1062 lcore_id, cache_count);
1063 count += cache_count;
1065 fprintf(f, " total_cache_count=%u\n", count);
1069 #ifndef __INTEL_COMPILER
1070 #pragma GCC diagnostic ignored "-Wcast-qual"
1073 /* check and update cookies or panic (internal) */
1074 void rte_mempool_check_cookies(const struct rte_mempool *mp,
1075 void * const *obj_table_const, unsigned n, int free)
1077 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1078 struct rte_mempool_objhdr *hdr;
1079 struct rte_mempool_objtlr *tlr;
1085 /* Force to drop the "const" attribute. This is done only when
1086 * DEBUG is enabled */
1087 tmp = (void *) obj_table_const;
1093 if (rte_mempool_from_obj(obj) != mp)
1094 rte_panic("MEMPOOL: object is owned by another "
1097 hdr = __mempool_get_header(obj);
1098 cookie = hdr->cookie;
1101 if (cookie != RTE_MEMPOOL_HEADER_COOKIE1) {
1102 RTE_LOG(CRIT, MEMPOOL,
1103 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1104 obj, (const void *) mp, cookie);
1105 rte_panic("MEMPOOL: bad header cookie (put)\n");
1107 hdr->cookie = RTE_MEMPOOL_HEADER_COOKIE2;
1108 } else if (free == 1) {
1109 if (cookie != RTE_MEMPOOL_HEADER_COOKIE2) {
1110 RTE_LOG(CRIT, MEMPOOL,
1111 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1112 obj, (const void *) mp, cookie);
1113 rte_panic("MEMPOOL: bad header cookie (get)\n");
1115 hdr->cookie = RTE_MEMPOOL_HEADER_COOKIE1;
1116 } else if (free == 2) {
1117 if (cookie != RTE_MEMPOOL_HEADER_COOKIE1 &&
1118 cookie != RTE_MEMPOOL_HEADER_COOKIE2) {
1119 RTE_LOG(CRIT, MEMPOOL,
1120 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1121 obj, (const void *) mp, cookie);
1122 rte_panic("MEMPOOL: bad header cookie (audit)\n");
1125 tlr = __mempool_get_trailer(obj);
1126 cookie = tlr->cookie;
1127 if (cookie != RTE_MEMPOOL_TRAILER_COOKIE) {
1128 RTE_LOG(CRIT, MEMPOOL,
1129 "obj=%p, mempool=%p, cookie=%" PRIx64 "\n",
1130 obj, (const void *) mp, cookie);
1131 rte_panic("MEMPOOL: bad trailer cookie\n");
1136 RTE_SET_USED(obj_table_const);
1142 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1144 mempool_obj_audit(struct rte_mempool *mp, __rte_unused void *opaque,
1145 void *obj, __rte_unused unsigned idx)
1147 __mempool_check_cookies(mp, &obj, 1, 2);
1151 mempool_audit_cookies(struct rte_mempool *mp)
1155 num = rte_mempool_obj_iter(mp, mempool_obj_audit, NULL);
1156 if (num != mp->size) {
1157 rte_panic("rte_mempool_obj_iter(mempool=%p, size=%u) "
1158 "iterated only over %u elements\n",
1163 #define mempool_audit_cookies(mp) do {} while(0)
1166 #ifndef __INTEL_COMPILER
1167 #pragma GCC diagnostic error "-Wcast-qual"
1170 /* check cookies before and after objects */
1172 mempool_audit_cache(const struct rte_mempool *mp)
1174 /* check cache size consistency */
1177 if (mp->cache_size == 0)
1180 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1181 const struct rte_mempool_cache *cache;
1182 cache = &mp->local_cache[lcore_id];
1183 if (cache->len > cache->flushthresh) {
1184 RTE_LOG(CRIT, MEMPOOL, "badness on cache[%u]\n",
1186 rte_panic("MEMPOOL: invalid cache len\n");
1191 /* check the consistency of mempool (size, cookies, ...) */
1193 rte_mempool_audit(struct rte_mempool *mp)
1195 mempool_audit_cache(mp);
1196 mempool_audit_cookies(mp);
1198 /* For case where mempool DEBUG is not set, and cache size is 0 */
1202 /* dump the status of the mempool on the console */
1204 rte_mempool_dump(FILE *f, struct rte_mempool *mp)
1206 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1207 struct rte_mempool_debug_stats sum;
1210 struct rte_mempool_memhdr *memhdr;
1211 unsigned common_count;
1212 unsigned cache_count;
1215 RTE_ASSERT(f != NULL);
1216 RTE_ASSERT(mp != NULL);
1218 fprintf(f, "mempool <%s>@%p\n", mp->name, mp);
1219 fprintf(f, " flags=%x\n", mp->flags);
1220 fprintf(f, " pool=%p\n", mp->pool_data);
1221 fprintf(f, " phys_addr=0x%" PRIx64 "\n", mp->mz->phys_addr);
1222 fprintf(f, " nb_mem_chunks=%u\n", mp->nb_mem_chunks);
1223 fprintf(f, " size=%"PRIu32"\n", mp->size);
1224 fprintf(f, " populated_size=%"PRIu32"\n", mp->populated_size);
1225 fprintf(f, " header_size=%"PRIu32"\n", mp->header_size);
1226 fprintf(f, " elt_size=%"PRIu32"\n", mp->elt_size);
1227 fprintf(f, " trailer_size=%"PRIu32"\n", mp->trailer_size);
1228 fprintf(f, " total_obj_size=%"PRIu32"\n",
1229 mp->header_size + mp->elt_size + mp->trailer_size);
1231 fprintf(f, " private_data_size=%"PRIu32"\n", mp->private_data_size);
1233 STAILQ_FOREACH(memhdr, &mp->mem_list, next)
1234 mem_len += memhdr->len;
1236 fprintf(f, " avg bytes/object=%#Lf\n",
1237 (long double)mem_len / mp->size);
1240 cache_count = rte_mempool_dump_cache(f, mp);
1241 common_count = rte_mempool_ops_get_count(mp);
1242 if ((cache_count + common_count) > mp->size)
1243 common_count = mp->size - cache_count;
1244 fprintf(f, " common_pool_count=%u\n", common_count);
1246 /* sum and dump statistics */
1247 #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
1248 memset(&sum, 0, sizeof(sum));
1249 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1250 sum.put_bulk += mp->stats[lcore_id].put_bulk;
1251 sum.put_objs += mp->stats[lcore_id].put_objs;
1252 sum.get_success_bulk += mp->stats[lcore_id].get_success_bulk;
1253 sum.get_success_objs += mp->stats[lcore_id].get_success_objs;
1254 sum.get_fail_bulk += mp->stats[lcore_id].get_fail_bulk;
1255 sum.get_fail_objs += mp->stats[lcore_id].get_fail_objs;
1257 fprintf(f, " stats:\n");
1258 fprintf(f, " put_bulk=%"PRIu64"\n", sum.put_bulk);
1259 fprintf(f, " put_objs=%"PRIu64"\n", sum.put_objs);
1260 fprintf(f, " get_success_bulk=%"PRIu64"\n", sum.get_success_bulk);
1261 fprintf(f, " get_success_objs=%"PRIu64"\n", sum.get_success_objs);
1262 fprintf(f, " get_fail_bulk=%"PRIu64"\n", sum.get_fail_bulk);
1263 fprintf(f, " get_fail_objs=%"PRIu64"\n", sum.get_fail_objs);
1265 fprintf(f, " no statistics available\n");
1268 rte_mempool_audit(mp);
1271 /* dump the status of all mempools on the console */
1273 rte_mempool_list_dump(FILE *f)
1275 struct rte_mempool *mp = NULL;
1276 struct rte_tailq_entry *te;
1277 struct rte_mempool_list *mempool_list;
1279 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1281 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1283 TAILQ_FOREACH(te, mempool_list, next) {
1284 mp = (struct rte_mempool *) te->data;
1285 rte_mempool_dump(f, mp);
1288 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);
1291 /* search a mempool from its name */
1292 struct rte_mempool *
1293 rte_mempool_lookup(const char *name)
1295 struct rte_mempool *mp = NULL;
1296 struct rte_tailq_entry *te;
1297 struct rte_mempool_list *mempool_list;
1299 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1301 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1303 TAILQ_FOREACH(te, mempool_list, next) {
1304 mp = (struct rte_mempool *) te->data;
1305 if (strncmp(name, mp->name, RTE_MEMPOOL_NAMESIZE) == 0)
1309 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);
1319 void rte_mempool_walk(void (*func)(struct rte_mempool *, void *),
1322 struct rte_tailq_entry *te = NULL;
1323 struct rte_mempool_list *mempool_list;
1326 mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
1328 rte_rwlock_read_lock(RTE_EAL_MEMPOOL_RWLOCK);
1330 TAILQ_FOREACH_SAFE(te, mempool_list, next, tmp_te) {
1331 (*func)((struct rte_mempool *) te->data, arg);
1334 rte_rwlock_read_unlock(RTE_EAL_MEMPOOL_RWLOCK);