From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga12.intel.com (mga12.intel.com [192.55.52.136]) by dpdk.org (Postfix) with ESMTP id 22EA15F1D for ; Sat, 3 Mar 2018 14:46:41 +0100 (CET) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from orsmga004.jf.intel.com ([10.7.209.38]) by fmsmga106.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 03 Mar 2018 05:46:37 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.47,418,1515484800"; d="scan'208";a="179582614" Received: from irvmail001.ir.intel.com ([163.33.26.43]) by orsmga004.jf.intel.com with ESMTP; 03 Mar 2018 05:46:33 -0800 Received: from sivswdev01.ir.intel.com (sivswdev01.ir.intel.com [10.237.217.45]) by irvmail001.ir.intel.com (8.14.3/8.13.6/MailSET/Hub) with ESMTP id w23DkWnH012196; Sat, 3 Mar 2018 13:46:32 GMT Received: from sivswdev01.ir.intel.com (localhost [127.0.0.1]) by sivswdev01.ir.intel.com with ESMTP id w23DkWoC023781; Sat, 3 Mar 2018 13:46:32 GMT Received: (from aburakov@localhost) by sivswdev01.ir.intel.com with LOCAL id w23DkWe6023777; Sat, 3 Mar 2018 13:46:32 GMT From: Anatoly Burakov To: dev@dpdk.org Cc: keith.wiles@intel.com, jianfeng.tan@intel.com, andras.kovacs@ericsson.com, laszlo.vadkeri@ericsson.com, benjamin.walker@intel.com, bruce.richardson@intel.com, thomas@monjalon.net, konstantin.ananyev@intel.com, kuralamudhan.ramakrishnan@intel.com, louise.m.daly@intel.com, nelio.laranjeiro@6wind.com, yskoh@mellanox.com, pepperjo@japf.ch, jerin.jacob@caviumnetworks.com, hemant.agrawal@nxp.com, olivier.matz@6wind.com Date: Sat, 3 Mar 2018 13:46:05 +0000 Message-Id: <79bd4a81a4a85908a011a5fffb33b881dd3a93c9.1520083504.git.anatoly.burakov@intel.com> X-Mailer: git-send-email 1.7.0.7 In-Reply-To: References: In-Reply-To: References: Subject: [dpdk-dev] [PATCH 17/41] eal: enable memory hotplug support in rte_malloc X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sat, 03 Mar 2018 13:46:42 -0000 This set of changes enables rte_malloc to allocate and free memory as needed. The way it works is, first malloc checks if there is enough memory already allocated to satisfy user's request. If there isn't, we try and allocate more memory. The reverse happens with free - we free an element, check its size (including free element merging due to adjacency) and see if it's bigger than hugepage size and that its start and end span a hugepage or more. Then we remove the area from malloc heap (adjusting element lengths where appropriate), and deallocate the page. For legacy mode, runtime alloc/free of pages is disabled. It is worth noting that memseg lists are being sorted by page size, and that we try our best to satisfy user's request. That is, if the user requests an element from a 2MB page memory, we will check if we can satisfy that request from existing memory, if not we try and allocate more 2MB pages. If that fails and user also specified a "size is hint" flag, we then check other page sizes and try to allocate from there. If that fails too, then, depending on flags, we may try allocating from other sockets. In other words, we try our best to give the user what they asked for, but going to other sockets is last resort - first we try to allocate more memory on the same socket. Signed-off-by: Anatoly Burakov --- lib/librte_eal/common/eal_common_memzone.c | 23 +- lib/librte_eal/common/malloc_elem.c | 85 ++++++++ lib/librte_eal/common/malloc_elem.h | 3 + lib/librte_eal/common/malloc_heap.c | 332 ++++++++++++++++++++++++++++- lib/librte_eal/common/malloc_heap.h | 4 +- lib/librte_eal/common/rte_malloc.c | 31 +-- 6 files changed, 416 insertions(+), 62 deletions(-) diff --git a/lib/librte_eal/common/eal_common_memzone.c b/lib/librte_eal/common/eal_common_memzone.c index ed36174..718dee8 100644 --- a/lib/librte_eal/common/eal_common_memzone.c +++ b/lib/librte_eal/common/eal_common_memzone.c @@ -103,7 +103,6 @@ memzone_reserve_aligned_thread_unsafe(const char *name, size_t len, struct rte_memzone *mz; struct rte_mem_config *mcfg; size_t requested_len; - int socket, i; /* get pointer to global configuration */ mcfg = rte_eal_get_configuration()->mem_config; @@ -181,27 +180,9 @@ memzone_reserve_aligned_thread_unsafe(const char *name, size_t len, } } - if (socket_id == SOCKET_ID_ANY) - socket = malloc_get_numa_socket(); - else - socket = socket_id; - /* allocate memory on heap */ - void *mz_addr = malloc_heap_alloc(&mcfg->malloc_heaps[socket], NULL, - requested_len, flags, align, bound); - - if ((mz_addr == NULL) && (socket_id == SOCKET_ID_ANY)) { - /* try other heaps */ - for (i = 0; i < RTE_MAX_NUMA_NODES; i++) { - if (socket == i) - continue; - - mz_addr = malloc_heap_alloc(&mcfg->malloc_heaps[i], - NULL, requested_len, flags, align, bound); - if (mz_addr != NULL) - break; - } - } + void *mz_addr = malloc_heap_alloc(NULL, requested_len, socket_id, flags, + align, bound); if (mz_addr == NULL) { rte_errno = ENOMEM; diff --git a/lib/librte_eal/common/malloc_elem.c b/lib/librte_eal/common/malloc_elem.c index 701bffd..eabad66 100644 --- a/lib/librte_eal/common/malloc_elem.c +++ b/lib/librte_eal/common/malloc_elem.c @@ -400,6 +400,91 @@ malloc_elem_free(struct malloc_elem *elem) return elem; } +/* assume all checks were already done */ +void +malloc_elem_hide_region(struct malloc_elem *elem, void *start, size_t len) +{ + size_t len_before, len_after; + struct malloc_elem *prev, *next; + void *end, *elem_end; + + end = RTE_PTR_ADD(start, len); + elem_end = RTE_PTR_ADD(elem, elem->size); + len_before = RTE_PTR_DIFF(start, elem); + len_after = RTE_PTR_DIFF(elem_end, end); + + prev = elem->prev; + next = elem->next; + + if (len_after >= MALLOC_ELEM_OVERHEAD + MIN_DATA_SIZE) { + /* split after */ + struct malloc_elem *split_after = end; + + split_elem(elem, split_after); + + next = split_after; + + malloc_elem_free_list_insert(split_after); + } else if (len_after >= MALLOC_ELEM_HEADER_LEN) { + struct malloc_elem *pad_elem = end; + + /* shrink current element */ + elem->size -= len_after; + memset(pad_elem, 0, sizeof(*pad_elem)); + + /* copy next element's data to our pad */ + memcpy(pad_elem, next, sizeof(*pad_elem)); + + /* pad next element */ + next->state = ELEM_PAD; + next->pad = len_after; + + /* next element is busy, would've been merged otherwise */ + pad_elem->pad = len_after; + pad_elem->size += len_after; + + /* adjust pointers to point to our new pad */ + pad_elem->next->prev = pad_elem; + elem->next = pad_elem; + } else if (len_after > 0) { + rte_panic("Unaligned element, heap is probably corrupt\n"); + } + + if (len_before >= MALLOC_ELEM_OVERHEAD + MIN_DATA_SIZE) { + /* split before */ + struct malloc_elem *split_before = start; + + split_elem(elem, split_before); + + prev = elem; + elem = split_before; + + malloc_elem_free_list_insert(prev); + } else if (len_before > 0) { + /* + * unlike with elements after current, here we don't need to + * pad elements, but rather just increase the size of previous + * element, copy the old header and and set up trailer. + */ + void *trailer = RTE_PTR_ADD(prev, + prev->size - MALLOC_ELEM_TRAILER_LEN); + struct malloc_elem *new_elem = start; + + memcpy(new_elem, elem, sizeof(*elem)); + new_elem->size -= len_before; + + prev->size += len_before; + set_trailer(prev); + + elem = new_elem; + + /* erase old trailer */ + memset(trailer, 0, MALLOC_ELEM_TRAILER_LEN); + } + + remove_elem(elem); +} + /* * attempt to resize a malloc_elem by expanding into any free space * immediately after it in memory. diff --git a/lib/librte_eal/common/malloc_elem.h b/lib/librte_eal/common/malloc_elem.h index 388c16f..6d979d2 100644 --- a/lib/librte_eal/common/malloc_elem.h +++ b/lib/librte_eal/common/malloc_elem.h @@ -152,6 +152,9 @@ int malloc_elem_resize(struct malloc_elem *elem, size_t size); void +malloc_elem_hide_region(struct malloc_elem *elem, void *start, size_t len); + +void malloc_elem_free_list_remove(struct malloc_elem *elem); /* diff --git a/lib/librte_eal/common/malloc_heap.c b/lib/librte_eal/common/malloc_heap.c index 058ad75..87dc9ad 100644 --- a/lib/librte_eal/common/malloc_heap.c +++ b/lib/librte_eal/common/malloc_heap.c @@ -20,8 +20,10 @@ #include #include #include +#include #include "eal_internal_cfg.h" +#include "eal_memalloc.h" #include "malloc_elem.h" #include "malloc_heap.h" @@ -123,48 +125,356 @@ find_suitable_element(struct malloc_heap *heap, size_t size, * scan fails. Once the new memseg is added, it re-scans and should return * the new element after releasing the lock. */ -void * -malloc_heap_alloc(struct malloc_heap *heap, - const char *type __attribute__((unused)), size_t size, unsigned flags, - size_t align, size_t bound) +static void * +heap_alloc(struct malloc_heap *heap, const char *type __rte_unused, size_t size, + unsigned int flags, size_t align, size_t bound) { struct malloc_elem *elem; size = RTE_CACHE_LINE_ROUNDUP(size); align = RTE_CACHE_LINE_ROUNDUP(align); - rte_spinlock_lock(&heap->lock); - elem = find_suitable_element(heap, size, flags, align, bound); if (elem != NULL) { elem = malloc_elem_alloc(elem, size, align, bound); + /* increase heap's count of allocated elements */ heap->alloc_count++; } - rte_spinlock_unlock(&heap->lock); return elem == NULL ? NULL : (void *)(&elem[1]); } +static int +try_expand_heap(struct malloc_heap *heap, uint64_t pg_sz, size_t elt_size, + int socket, unsigned int flags, size_t align, size_t bound) +{ + struct rte_memseg_list *msl; + struct rte_memseg **ms; + struct malloc_elem *elem; + size_t map_len; + int i, n_pages, allocd_pages; + void *ret, *map_addr; + + align = RTE_MAX(align, MALLOC_ELEM_HEADER_LEN); + map_len = RTE_ALIGN_CEIL(align + elt_size + MALLOC_ELEM_TRAILER_LEN, + pg_sz); + + n_pages = map_len / pg_sz; + + /* we can't know in advance how many pages we'll need, so malloc */ + ms = malloc(sizeof(*ms) * n_pages); + + allocd_pages = eal_memalloc_alloc_page_bulk(ms, n_pages, pg_sz, socket, + true); + + /* make sure we've allocated our pages... */ + if (allocd_pages != n_pages) + goto free_ms; + + map_addr = ms[0]->addr; + msl = rte_mem_virt2memseg_list(map_addr); + + /* add newly minted memsegs to malloc heap */ + elem = malloc_heap_add_memory(heap, msl, map_addr, map_len); + + /* try once more, as now we have allocated new memory */ + ret = find_suitable_element(heap, elt_size, flags, align, bound); + + if (ret == NULL) + goto free_elem; + + RTE_LOG(DEBUG, EAL, "Heap on socket %d was expanded by %zdMB\n", + socket, map_len >> 20ULL); + + free(ms); + + return 0; + +free_elem: + malloc_elem_free_list_remove(elem); + malloc_elem_hide_region(elem, map_addr, map_len); + heap->total_size -= map_len; + + for (i = 0; i < n_pages; i++) + eal_memalloc_free_page(ms[i]); +free_ms: + free(ms); + + return -1; +} + +static int +compare_pagesz(const void *a, const void *b) +{ + const struct rte_memseg_list * const*mpa = a; + const struct rte_memseg_list * const*mpb = b; + const struct rte_memseg_list *msla = *mpa; + const struct rte_memseg_list *mslb = *mpb; + uint64_t pg_sz_a = msla->hugepage_sz; + uint64_t pg_sz_b = mslb->hugepage_sz; + + if (pg_sz_a < pg_sz_b) + return -1; + if (pg_sz_a > pg_sz_b) + return 1; + return 0; +} + +static int +alloc_mem_on_socket(size_t size, int socket, unsigned int flags, size_t align, + size_t bound) +{ + struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config; + struct malloc_heap *heap = &mcfg->malloc_heaps[socket]; + struct rte_memseg_list *requested_msls[RTE_MAX_MEMSEG_LISTS]; + struct rte_memseg_list *other_msls[RTE_MAX_MEMSEG_LISTS]; + uint64_t requested_pg_sz[RTE_MAX_MEMSEG_LISTS]; + uint64_t other_pg_sz[RTE_MAX_MEMSEG_LISTS]; + uint64_t prev_pg_sz; + int i, n_other_msls, n_other_pg_sz, n_requested_msls, n_requested_pg_sz; + bool size_hint = (flags & RTE_MEMZONE_SIZE_HINT_ONLY) > 0; + unsigned int size_flags = flags & ~RTE_MEMZONE_SIZE_HINT_ONLY; + void *ret; + + memset(requested_msls, 0, sizeof(requested_msls)); + memset(other_msls, 0, sizeof(other_msls)); + memset(requested_pg_sz, 0, sizeof(requested_pg_sz)); + memset(other_pg_sz, 0, sizeof(other_pg_sz)); + + /* + * go through memseg list and take note of all the page sizes available, + * and if any of them were specifically requested by the user. + */ + n_requested_msls = 0; + n_other_msls = 0; + for (i = 0; i < RTE_MAX_MEMSEG_LISTS; i++) { + struct rte_memseg_list *msl = &mcfg->memsegs[i]; + + if (msl->socket_id != socket) + continue; + + if (msl->base_va == NULL) + continue; + + /* if pages of specific size were requested */ + if (size_flags != 0 && check_hugepage_sz(size_flags, + msl->hugepage_sz)) + requested_msls[n_requested_msls++] = msl; + else if (size_flags == 0 || size_hint) + other_msls[n_other_msls++] = msl; + } + + /* sort the lists, smallest first */ + qsort(requested_msls, n_requested_msls, sizeof(requested_msls[0]), + compare_pagesz); + qsort(other_msls, n_other_msls, sizeof(other_msls[0]), + compare_pagesz); + + /* now, extract page sizes we are supposed to try */ + prev_pg_sz = 0; + n_requested_pg_sz = 0; + for (i = 0; i < n_requested_msls; i++) { + uint64_t pg_sz = requested_msls[i]->hugepage_sz; + + if (prev_pg_sz != pg_sz) { + requested_pg_sz[n_requested_pg_sz++] = pg_sz; + prev_pg_sz = pg_sz; + } + } + prev_pg_sz = 0; + n_other_pg_sz = 0; + for (i = 0; i < n_other_msls; i++) { + uint64_t pg_sz = other_msls[i]->hugepage_sz; + + if (prev_pg_sz != pg_sz) { + other_pg_sz[n_other_pg_sz++] = pg_sz; + prev_pg_sz = pg_sz; + } + } + + /* finally, try allocating memory of specified page sizes, starting from + * the smallest sizes + */ + for (i = 0; i < n_requested_pg_sz; i++) { + uint64_t pg_sz = requested_pg_sz[i]; + + /* + * do not pass the size hint here, as user expects other page + * sizes first, before resorting to best effort allocation. + */ + if (!try_expand_heap(heap, pg_sz, size, socket, size_flags, + align, bound)) + return 0; + } + if (n_other_pg_sz == 0) + return -1; + + /* now, check if we can reserve anything with size hint */ + ret = find_suitable_element(heap, size, flags, align, bound); + if (ret != NULL) + return 0; + + /* + * we still couldn't reserve memory, so try expanding heap with other + * page sizes, if there are any + */ + for (i = 0; i < n_other_pg_sz; i++) { + uint64_t pg_sz = other_pg_sz[i]; + + if (!try_expand_heap(heap, pg_sz, size, socket, flags, + align, bound)) + return 0; + } + return -1; +} + +/* this will try lower page sizes first */ +static void * +heap_alloc_on_socket(const char *type, size_t size, int socket, + unsigned int flags, size_t align, size_t bound) +{ + struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config; + struct malloc_heap *heap = &mcfg->malloc_heaps[socket]; + unsigned int size_flags = flags & ~RTE_MEMZONE_SIZE_HINT_ONLY; + void *ret; + + rte_spinlock_lock(&(heap->lock)); + + align = align == 0 ? 1 : align; + + /* for legacy mode, try once and with all flags */ + if (internal_config.legacy_mem) { + ret = heap_alloc(heap, type, size, flags, align, bound); + goto alloc_unlock; + } + + /* + * we do not pass the size hint here, because even if allocation fails, + * we may still be able to allocate memory from appropriate page sizes, + * we just need to request more memory first. + */ + ret = heap_alloc(heap, type, size, size_flags, align, bound); + if (ret != NULL) + goto alloc_unlock; + + if (!alloc_mem_on_socket(size, socket, flags, align, bound)) { + ret = heap_alloc(heap, type, size, flags, align, bound); + + /* this should have succeeded */ + if (ret == NULL) + rte_panic("Error allocating from heap\n"); + } +alloc_unlock: + rte_spinlock_unlock(&(heap->lock)); + return ret; +} + +void * +malloc_heap_alloc(const char *type, size_t size, int socket_arg, + unsigned int flags, size_t align, size_t bound) +{ + int socket, i; + void *ret; + + /* return NULL if size is 0 or alignment is not power-of-2 */ + if (size == 0 || (align && !rte_is_power_of_2(align))) + return NULL; + + if (!rte_eal_has_hugepages()) + socket_arg = SOCKET_ID_ANY; + + if (socket_arg == SOCKET_ID_ANY) + socket = malloc_get_numa_socket(); + else + socket = socket_arg; + + /* Check socket parameter */ + if (socket >= RTE_MAX_NUMA_NODES) + return NULL; + + ret = heap_alloc_on_socket(type, size, socket, flags, align, bound); + if (ret != NULL || socket_arg != SOCKET_ID_ANY) + return ret; + + /* try other heaps */ + for (i = 0; i < (int) rte_num_sockets(); i++) { + if (i == socket) + continue; + ret = heap_alloc_on_socket(type, size, i, flags, + align, bound); + if (ret != NULL) + return ret; + } + return NULL; +} + int malloc_heap_free(struct malloc_elem *elem) { struct malloc_heap *heap; - struct malloc_elem *ret; + void *start, *aligned_start, *end, *aligned_end; + size_t len, aligned_len; + struct rte_memseg_list *msl; + int n_pages, page_idx, max_page_idx, ret; if (!malloc_elem_cookies_ok(elem) || elem->state != ELEM_BUSY) return -1; /* elem may be merged with previous element, so keep heap address */ heap = elem->heap; + msl = elem->msl; rte_spinlock_lock(&(heap->lock)); - ret = malloc_elem_free(elem); + elem = malloc_elem_free(elem); - rte_spinlock_unlock(&(heap->lock)); + /* anything after this is a bonus */ + ret = 0; + + /* ...of which we can't avail if we are in legacy mode */ + if (internal_config.legacy_mem) + goto free_unlock; + + /* check if we can free any memory back to the system */ + if (elem->size < msl->hugepage_sz) + goto free_unlock; - return ret != NULL ? 0 : -1; + /* probably, but let's make sure, as we may not be using up full page */ + start = elem; + len = elem->size; + aligned_start = RTE_PTR_ALIGN_CEIL(start, msl->hugepage_sz); + end = RTE_PTR_ADD(elem, len); + aligned_end = RTE_PTR_ALIGN_FLOOR(end, msl->hugepage_sz); + + aligned_len = RTE_PTR_DIFF(aligned_end, aligned_start); + + /* can't free anything */ + if (aligned_len < msl->hugepage_sz) + goto free_unlock; + + malloc_elem_free_list_remove(elem); + + malloc_elem_hide_region(elem, (void *) aligned_start, aligned_len); + + /* we don't really care if we fail to deallocate memory */ + n_pages = aligned_len / msl->hugepage_sz; + page_idx = RTE_PTR_DIFF(aligned_start, msl->base_va) / msl->hugepage_sz; + max_page_idx = page_idx + n_pages; + + for (; page_idx < max_page_idx; page_idx++) { + struct rte_memseg *ms; + + ms = rte_fbarray_get(&msl->memseg_arr, page_idx); + eal_memalloc_free_page(ms); + heap->total_size -= msl->hugepage_sz; + } + + RTE_LOG(DEBUG, EAL, "Heap on socket %d was shrunk by %zdMB\n", + msl->socket_id, aligned_len >> 20ULL); +free_unlock: + rte_spinlock_unlock(&(heap->lock)); + return ret; } int diff --git a/lib/librte_eal/common/malloc_heap.h b/lib/librte_eal/common/malloc_heap.h index bb28422..292d578 100644 --- a/lib/librte_eal/common/malloc_heap.h +++ b/lib/librte_eal/common/malloc_heap.h @@ -24,8 +24,8 @@ malloc_get_numa_socket(void) } void * -malloc_heap_alloc(struct malloc_heap *heap, const char *type, size_t size, - unsigned flags, size_t align, size_t bound); +malloc_heap_alloc(const char *type, size_t size, int socket, unsigned int flags, + size_t align, size_t bound); int malloc_heap_free(struct malloc_elem *elem); diff --git a/lib/librte_eal/common/rte_malloc.c b/lib/librte_eal/common/rte_malloc.c index bd7e757..b0fe11c 100644 --- a/lib/librte_eal/common/rte_malloc.c +++ b/lib/librte_eal/common/rte_malloc.c @@ -39,10 +39,6 @@ void rte_free(void *addr) void * rte_malloc_socket(const char *type, size_t size, unsigned align, int socket_arg) { - struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config; - int socket, i; - void *ret; - /* return NULL if size is 0 or alignment is not power-of-2 */ if (size == 0 || (align && !rte_is_power_of_2(align))) return NULL; @@ -50,33 +46,12 @@ rte_malloc_socket(const char *type, size_t size, unsigned align, int socket_arg) if (!rte_eal_has_hugepages()) socket_arg = SOCKET_ID_ANY; - if (socket_arg == SOCKET_ID_ANY) - socket = malloc_get_numa_socket(); - else - socket = socket_arg; - /* Check socket parameter */ - if (socket >= RTE_MAX_NUMA_NODES) + if (socket_arg >= RTE_MAX_NUMA_NODES) return NULL; - ret = malloc_heap_alloc(&mcfg->malloc_heaps[socket], type, - size, 0, align == 0 ? 1 : align, 0); - if (ret != NULL || socket_arg != SOCKET_ID_ANY) - return ret; - - /* try other heaps */ - for (i = 0; i < RTE_MAX_NUMA_NODES; i++) { - /* we already tried this one */ - if (i == socket) - continue; - - ret = malloc_heap_alloc(&mcfg->malloc_heaps[i], type, - size, 0, align == 0 ? 1 : align, 0); - if (ret != NULL) - return ret; - } - - return NULL; + return malloc_heap_alloc(type, size, socket_arg, 0, + align == 0 ? 1 : align, 0); } /* -- 2.7.4