Patch Detail
get:
Show a patch.
patch:
Update a patch.
put:
Update a patch.
GET /api/patches/32469/?format=api
{ "id": 32469, "url": "https://patches.dpdk.org/api/patches/32469/?format=api", "web_url": "https://patches.dpdk.org/project/dpdk/patch/c9a798795633a8e4a4bf32f1a57c25bb5ce099fd.1513681966.git.anatoly.burakov@intel.com/", "project": { "id": 1, "url": "https://patches.dpdk.org/api/projects/1/?format=api", "name": "DPDK", "link_name": "dpdk", "list_id": "dev.dpdk.org", "list_email": "dev@dpdk.org", "web_url": "http://core.dpdk.org", "scm_url": "git://dpdk.org/dpdk", "webscm_url": "http://git.dpdk.org/dpdk", "list_archive_url": "https://inbox.dpdk.org/dev", "list_archive_url_format": "https://inbox.dpdk.org/dev/{}", "commit_url_format": "" }, "msgid": "<c9a798795633a8e4a4bf32f1a57c25bb5ce099fd.1513681966.git.anatoly.burakov@intel.com>", "list_archive_url": "https://inbox.dpdk.org/dev/c9a798795633a8e4a4bf32f1a57c25bb5ce099fd.1513681966.git.anatoly.burakov@intel.com", "date": "2017-12-19T11:14:43", "name": "[dpdk-dev,RFC,v2,16/23] eal: enable dynamic memory allocation/free on malloc/free", "commit_ref": null, "pull_url": null, "state": "superseded", "archived": true, "hash": "ccc002acd2b3601635ee373163d6e926ae7ba975", "submitter": { "id": 4, "url": "https://patches.dpdk.org/api/people/4/?format=api", "name": "Anatoly Burakov", "email": "anatoly.burakov@intel.com" }, "delegate": null, "mbox": "https://patches.dpdk.org/project/dpdk/patch/c9a798795633a8e4a4bf32f1a57c25bb5ce099fd.1513681966.git.anatoly.burakov@intel.com/mbox/", "series": [], "comments": "https://patches.dpdk.org/api/patches/32469/comments/", "check": "fail", "checks": "https://patches.dpdk.org/api/patches/32469/checks/", "tags": {}, "related": [], "headers": { "Return-Path": "<dev-bounces@dpdk.org>", "X-Original-To": "patchwork@dpdk.org", "Delivered-To": "patchwork@dpdk.org", "Received": [ "from [92.243.14.124] (localhost [127.0.0.1])\n\tby dpdk.org (Postfix) with ESMTP id 703741B20D;\n\tTue, 19 Dec 2017 12:15:26 +0100 (CET)", "from mga04.intel.com (mga04.intel.com [192.55.52.120])\n\tby dpdk.org (Postfix) with ESMTP id BF8F81B01C\n\tfor <dev@dpdk.org>; Tue, 19 Dec 2017 12:14:56 +0100 (CET)", "from orsmga005.jf.intel.com ([10.7.209.41])\n\tby fmsmga104.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384;\n\t19 Dec 2017 03:14:55 -0800", "from irvmail001.ir.intel.com ([163.33.26.43])\n\tby orsmga005.jf.intel.com with ESMTP; 19 Dec 2017 03:14:53 -0800", "from sivswdev01.ir.intel.com (sivswdev01.ir.intel.com\n\t[10.237.217.45])\n\tby irvmail001.ir.intel.com (8.14.3/8.13.6/MailSET/Hub) with ESMTP id\n\tvBJBEq5l003135; Tue, 19 Dec 2017 11:14:52 GMT", "from sivswdev01.ir.intel.com (localhost [127.0.0.1])\n\tby sivswdev01.ir.intel.com with ESMTP id vBJBEq7I010295;\n\tTue, 19 Dec 2017 11:14:52 GMT", "(from aburakov@localhost)\n\tby sivswdev01.ir.intel.com with LOCAL id vBJBEq8M010291;\n\tTue, 19 Dec 2017 11:14:52 GMT" ], "X-Amp-Result": "SKIPPED(no attachment in message)", "X-Amp-File-Uploaded": "False", "X-ExtLoop1": "1", "X-IronPort-AV": "E=Sophos;i=\"5.45,426,1508828400\"; d=\"scan'208\";a=\"185440369\"", "From": "Anatoly Burakov <anatoly.burakov@intel.com>", "To": "dev@dpdk.org", "Cc": "andras.kovacs@ericsson.com, laszlo.vadkeri@ericsson.com,\n\tkeith.wiles@intel.com, benjamin.walker@intel.com,\n\tbruce.richardson@intel.com, thomas@monjalon.net", "Date": "Tue, 19 Dec 2017 11:14:43 +0000", "Message-Id": "<c9a798795633a8e4a4bf32f1a57c25bb5ce099fd.1513681966.git.anatoly.burakov@intel.com>", "X-Mailer": "git-send-email 1.7.0.7", "In-Reply-To": [ "<cover.1513681966.git.anatoly.burakov@intel.com>", "<cover.1513681966.git.anatoly.burakov@intel.com>" ], "References": [ "<cover.1513681966.git.anatoly.burakov@intel.com>", "<cover.1513681966.git.anatoly.burakov@intel.com>" ], "Subject": "[dpdk-dev] [RFC v2 16/23] eal: enable dynamic memory\n\tallocation/free on malloc/free", "X-BeenThere": "dev@dpdk.org", "X-Mailman-Version": "2.1.15", "Precedence": "list", "List-Id": "DPDK patches and discussions <dev.dpdk.org>", "List-Unsubscribe": "<https://dpdk.org/ml/options/dev>,\n\t<mailto:dev-request@dpdk.org?subject=unsubscribe>", "List-Archive": "<http://dpdk.org/ml/archives/dev/>", "List-Post": "<mailto:dev@dpdk.org>", "List-Help": "<mailto:dev-request@dpdk.org?subject=help>", "List-Subscribe": "<https://dpdk.org/ml/listinfo/dev>,\n\t<mailto:dev-request@dpdk.org?subject=subscribe>", "Errors-To": "dev-bounces@dpdk.org", "Sender": "\"dev\" <dev-bounces@dpdk.org>" }, "content": "This set of changes enables rte_malloc to allocate and free memory\nas needed. The way it works is, first malloc checks if there is\nenough memory already allocated to satisfy user's request. If there\nisn't, we try and allocate more memory. The reverse happens with\nfree - we free an element, check its size (including free element\nmerging due to adjacency) and see if it's bigger than hugepage\nsize and that its start and end span a hugepage or more. Then we\nremove the area from malloc heap (adjusting element lengths where\nappropriate), and deallocate the page.\n\nFor legacy mode, dynamic alloc/free is disabled.\n\nIt is worth noting that memseg lists are being sorted by page size,\nand that we try our best to satisfy user's request. That is, if\nthe user requests an element from a 2MB page memory, we will check\nif we can satisfy that request from existing memory, if not we try\nand allocate more 2MB pages. If that fails and user also specified\na \"size is hint\" flag, we then check other page sizes and try to\nallocate from there. If that fails too, then, depending on flags,\nwe may try allocating from other sockets. In other words, we try\nour best to give the user what they asked for, but going to other\nsockets is last resort - first we try to allocate more memory on\nthe same socket.\n\nSigned-off-by: Anatoly Burakov <anatoly.burakov@intel.com>\n---\n lib/librte_eal/common/eal_common_memzone.c | 19 +-\n lib/librte_eal/common/malloc_elem.c | 96 +++++++++-\n lib/librte_eal/common/malloc_elem.h | 8 +-\n lib/librte_eal/common/malloc_heap.c | 280 +++++++++++++++++++++++++++--\n lib/librte_eal/common/malloc_heap.h | 4 +-\n lib/librte_eal/common/rte_malloc.c | 24 +--\n 6 files changed, 373 insertions(+), 58 deletions(-)", "diff": "diff --git a/lib/librte_eal/common/eal_common_memzone.c b/lib/librte_eal/common/eal_common_memzone.c\nindex f558ac2..c571145 100644\n--- a/lib/librte_eal/common/eal_common_memzone.c\n+++ b/lib/librte_eal/common/eal_common_memzone.c\n@@ -132,7 +132,7 @@ memzone_reserve_aligned_thread_unsafe(const char *name, size_t len,\n \tstruct rte_memzone *mz;\n \tstruct rte_mem_config *mcfg;\n \tsize_t requested_len;\n-\tint socket, i;\n+\tint socket;\n \n \t/* get pointer to global configuration */\n \tmcfg = rte_eal_get_configuration()->mem_config;\n@@ -216,21 +216,8 @@ memzone_reserve_aligned_thread_unsafe(const char *name, size_t len,\n \t\tsocket = socket_id;\n \n \t/* allocate memory on heap */\n-\tvoid *mz_addr = malloc_heap_alloc(&mcfg->malloc_heaps[socket], NULL,\n-\t\t\trequested_len, flags, align, bound);\n-\n-\tif ((mz_addr == NULL) && (socket_id == SOCKET_ID_ANY)) {\n-\t\t/* try other heaps */\n-\t\tfor (i = 0; i < RTE_MAX_NUMA_NODES; i++) {\n-\t\t\tif (socket == i)\n-\t\t\t\tcontinue;\n-\n-\t\t\tmz_addr = malloc_heap_alloc(&mcfg->malloc_heaps[i],\n-\t\t\t\t\tNULL, requested_len, flags, align, bound);\n-\t\t\tif (mz_addr != NULL)\n-\t\t\t\tbreak;\n-\t\t}\n-\t}\n+\tvoid *mz_addr = malloc_heap_alloc(NULL, requested_len, socket, flags,\n+\t\t\talign, bound);\n \n \tif (mz_addr == NULL) {\n \t\trte_errno = ENOMEM;\ndiff --git a/lib/librte_eal/common/malloc_elem.c b/lib/librte_eal/common/malloc_elem.c\nindex ab09b94..48ac604 100644\n--- a/lib/librte_eal/common/malloc_elem.c\n+++ b/lib/librte_eal/common/malloc_elem.c\n@@ -269,8 +269,8 @@ malloc_elem_free_list_insert(struct malloc_elem *elem)\n /*\n * Remove the specified element from its heap's free list.\n */\n-static void\n-elem_free_list_remove(struct malloc_elem *elem)\n+void\n+malloc_elem_free_list_remove(struct malloc_elem *elem)\n {\n \tLIST_REMOVE(elem, free_list);\n }\n@@ -290,7 +290,7 @@ malloc_elem_alloc(struct malloc_elem *elem, size_t size, unsigned align,\n \tconst size_t trailer_size = elem->size - old_elem_size - size -\n \t\tMALLOC_ELEM_OVERHEAD;\n \n-\telem_free_list_remove(elem);\n+\tmalloc_elem_free_list_remove(elem);\n \n \tif (trailer_size > MALLOC_ELEM_OVERHEAD + MIN_DATA_SIZE) {\n \t\t/* split it, too much free space after elem */\n@@ -363,7 +363,7 @@ malloc_elem_join_adjacent_free(struct malloc_elem *elem) {\n \t\terase = RTE_PTR_SUB(elem->next, MALLOC_ELEM_TRAILER_LEN);\n \n \t\t/* remove from free list, join to this one */\n-\t\telem_free_list_remove(elem->next);\n+\t\tmalloc_elem_free_list_remove(elem->next);\n \t\tjoin_elem(elem, elem->next);\n \n \t\t/* erase header and trailer */\n@@ -383,7 +383,7 @@ malloc_elem_join_adjacent_free(struct malloc_elem *elem) {\n \t\terase = RTE_PTR_SUB(elem, MALLOC_ELEM_TRAILER_LEN);\n \n \t\t/* remove from free list, join to this one */\n-\t\telem_free_list_remove(elem->prev);\n+\t\tmalloc_elem_free_list_remove(elem->prev);\n \n \t\tnew_elem = elem->prev;\n \t\tjoin_elem(new_elem, elem);\n@@ -402,7 +402,7 @@ malloc_elem_join_adjacent_free(struct malloc_elem *elem) {\n * blocks either immediately before or immediately after newly freed block\n * are also free, the blocks are merged together.\n */\n-int\n+struct malloc_elem *\n malloc_elem_free(struct malloc_elem *elem)\n {\n \tvoid *ptr;\n@@ -420,7 +420,87 @@ malloc_elem_free(struct malloc_elem *elem)\n \n \tmemset(ptr, 0, data_len);\n \n-\treturn 0;\n+\treturn elem;\n+}\n+\n+/* assume all checks were already done */\n+void\n+malloc_elem_hide_region(struct malloc_elem *elem, void *start, size_t len) {\n+\tsize_t len_before, len_after;\n+\tstruct malloc_elem *prev, *next;\n+\tvoid *end, *elem_end;\n+\n+\tend = RTE_PTR_ADD(start, len);\n+\telem_end = RTE_PTR_ADD(elem, elem->size);\n+\tlen_before = RTE_PTR_DIFF(start, elem);\n+\tlen_after = RTE_PTR_DIFF(elem_end, end);\n+\n+\tprev = elem->prev;\n+\tnext = elem->next;\n+\n+\tif (len_after >= MALLOC_ELEM_OVERHEAD + MIN_DATA_SIZE) {\n+\t\t/* split after */\n+\t\tstruct malloc_elem *split_after = end;\n+\n+\t\tsplit_elem(elem, split_after);\n+\n+\t\tnext = split_after;\n+\n+\t\tmalloc_elem_free_list_insert(split_after);\n+\t} else if (len_after >= MALLOC_ELEM_HEADER_LEN) {\n+\t\tstruct malloc_elem *pad_elem = end;\n+\n+\t\t/* shrink current element */\n+\t\telem->size -= len_after;\n+\t\tmemset(pad_elem, 0, sizeof(*pad_elem));\n+\n+\t\t/* copy next element's data to our pad */\n+\t\tmemcpy(pad_elem, next, sizeof(*pad_elem));\n+\n+\t\t/* pad next element */\n+\t\tnext->state = ELEM_PAD;\n+\t\tnext->pad = len_after;\n+\n+\t\t/* next element is busy, would've been merged otherwise */\n+\t\tpad_elem->pad = len_after;\n+\t\tpad_elem->size += len_after;\n+\t} else if (len_after > 0) {\n+\t\trte_panic(\"Unaligned element, heap is probably corrupt\\n\");\n+\t}\n+\n+\tif (len_before >= MALLOC_ELEM_OVERHEAD + MIN_DATA_SIZE) {\n+\t\t/* split before */\n+\t\tstruct malloc_elem *split_before = start;\n+\n+\t\tsplit_elem(elem, split_before);\n+\n+\t\tprev = elem;\n+\t\telem = split_before;\n+\n+\t\tmalloc_elem_free_list_insert(prev);\n+\t} else if (len_before > 0) {\n+\t\t/*\n+\t\t * unlike with elements after current, here we don't need to\n+\t\t * pad elements, but rather just increase the size of previous\n+\t\t * element, copy the old header and and set up trailer.\n+\t\t */\n+\t\tvoid *trailer = RTE_PTR_ADD(prev,\n+\t\t\t\tprev->size - MALLOC_ELEM_TRAILER_LEN);\n+\t\tstruct malloc_elem *new_elem = start;\n+\n+\t\tmemcpy(new_elem, elem, sizeof(*elem));\n+\t\tnew_elem->size -= len_before;\n+\n+\t\tprev->size += len_before;\n+\t\tset_trailer(prev);\n+\n+\t\telem = new_elem;\n+\n+\t\t/* erase old trailer */\n+\t\tmemset(trailer, 0, MALLOC_ELEM_TRAILER_LEN);\n+\t}\n+\n+\tremove_elem(elem);\n }\n \n /*\n@@ -446,7 +526,7 @@ malloc_elem_resize(struct malloc_elem *elem, size_t size)\n \t/* we now know the element fits, so remove from free list,\n \t * join the two\n \t */\n-\telem_free_list_remove(elem->next);\n+\tmalloc_elem_free_list_remove(elem->next);\n \tjoin_elem(elem, elem->next);\n \n \tif (elem->size - new_size >= MIN_DATA_SIZE + MALLOC_ELEM_OVERHEAD) {\ndiff --git a/lib/librte_eal/common/malloc_elem.h b/lib/librte_eal/common/malloc_elem.h\nindex 330bddc..b47c55e 100644\n--- a/lib/librte_eal/common/malloc_elem.h\n+++ b/lib/librte_eal/common/malloc_elem.h\n@@ -164,7 +164,7 @@ malloc_elem_alloc(struct malloc_elem *elem, size_t size,\n * blocks either immediately before or immediately after newly freed block\n * are also free, the blocks are merged together.\n */\n-int\n+struct malloc_elem *\n malloc_elem_free(struct malloc_elem *elem);\n \n struct malloc_elem *\n@@ -177,6 +177,12 @@ malloc_elem_join_adjacent_free(struct malloc_elem *elem);\n int\n malloc_elem_resize(struct malloc_elem *elem, size_t size);\n \n+void\n+malloc_elem_hide_region(struct malloc_elem *elem, void *start, size_t len);\n+\n+void\n+malloc_elem_free_list_remove(struct malloc_elem *elem);\n+\n /*\n * Given an element size, compute its freelist index.\n */\ndiff --git a/lib/librte_eal/common/malloc_heap.c b/lib/librte_eal/common/malloc_heap.c\nindex 5fa21fe..0d61704 100644\n--- a/lib/librte_eal/common/malloc_heap.c\n+++ b/lib/librte_eal/common/malloc_heap.c\n@@ -49,8 +49,10 @@\n #include <rte_spinlock.h>\n #include <rte_memcpy.h>\n #include <rte_atomic.h>\n+#include <rte_fbarray.h>\n \n #include \"eal_internal_cfg.h\"\n+#include \"eal_memalloc.h\"\n #include \"malloc_elem.h\"\n #include \"malloc_heap.h\"\n \n@@ -151,46 +153,304 @@ find_suitable_element(struct malloc_heap *heap, size_t size,\n * scan fails. Once the new memseg is added, it re-scans and should return\n * the new element after releasing the lock.\n */\n-void *\n-malloc_heap_alloc(struct malloc_heap *heap,\n-\t\tconst char *type __attribute__((unused)), size_t size, unsigned flags,\n-\t\tsize_t align, size_t bound)\n+static void *\n+heap_alloc(struct malloc_heap *heap, const char *type __rte_unused, size_t size,\n+\t\tunsigned flags, size_t align, size_t bound)\n {\n \tstruct malloc_elem *elem;\n \n \tsize = RTE_CACHE_LINE_ROUNDUP(size);\n \talign = RTE_CACHE_LINE_ROUNDUP(align);\n \n-\trte_spinlock_lock(&heap->lock);\n-\n \telem = find_suitable_element(heap, size, flags, align, bound);\n \tif (elem != NULL) {\n \t\telem = malloc_elem_alloc(elem, size, align, bound);\n+\n \t\t/* increase heap's count of allocated elements */\n \t\theap->alloc_count++;\n \t}\n-\trte_spinlock_unlock(&heap->lock);\n \n \treturn elem == NULL ? NULL : (void *)(&elem[1]);\n }\n \n+static void *\n+try_expand_heap(struct malloc_heap *heap, struct rte_memseg_list *msl,\n+\t\tconst char *type, size_t size, int socket, unsigned flags,\n+\t\tsize_t align, size_t bound) {\n+\tstruct malloc_elem *elem;\n+\tstruct rte_memseg **ms;\n+\tsize_t map_len;\n+\tvoid *map_addr;\n+\tint i, n_pages, allocd_pages;\n+\tvoid *ret;\n+\n+\talign = RTE_MAX(align, MALLOC_ELEM_HEADER_LEN);\n+\tmap_len = RTE_ALIGN_CEIL(align + size + MALLOC_ELEM_TRAILER_LEN,\n+\t\t\tmsl->hugepage_sz);\n+\n+\tn_pages = map_len / msl->hugepage_sz;\n+\n+\t/* we can't know in advance how many pages we'll need, so malloc */\n+\tms = malloc(sizeof(*ms) * n_pages);\n+\n+\tallocd_pages = eal_memalloc_alloc_page_bulk(ms, n_pages,\n+\t\t\tmsl->hugepage_sz, socket, true);\n+\n+\t/* make sure we've allocated our pages... */\n+\tif (allocd_pages != n_pages)\n+\t\tgoto free_ms;\n+\n+\tmap_addr = ms[0]->addr;\n+\n+\t/* add newly minted memsegs to malloc heap */\n+\telem = malloc_heap_add_memory(heap, msl, map_addr, map_len);\n+\n+\tRTE_LOG(DEBUG, EAL, \"Heap on socket %d was expanded by %zdMB\\n\",\n+\t\tmsl->socket_id, map_len >> 20ULL);\n+\n+\t/* try once more, as now we have allocated new memory */\n+\tret = heap_alloc(heap, type, size, flags,\n+\t\t\talign == 0 ? 1 : align, bound);\n+\n+\tif (ret == NULL)\n+\t\tgoto free_elem;\n+\n+\tfree(ms);\n+\treturn ret;\n+\n+free_elem:\n+\tmalloc_elem_free_list_remove(elem);\n+\tmalloc_elem_hide_region(elem, map_addr, map_len);\n+\theap->total_size -= map_len;\n+\n+\tRTE_LOG(DEBUG, EAL, \"%s(): couldn't allocate, so shrinking heap on socket %d by %zdMB\\n\",\n+\t\t__func__, socket, map_len >> 20ULL);\n+\n+\tfor (i = 0; i < n_pages; i++) {\n+\t\teal_memalloc_free_page(ms[i]);\n+\t}\n+free_ms:\n+\tfree(ms);\n+\treturn NULL;\n+}\n+\n+static int\n+compare_pagesz(const void *a, const void *b) {\n+\tconst struct rte_memseg_list *msla = a;\n+\tconst struct rte_memseg_list *mslb = b;\n+\n+\tif (msla->hugepage_sz < mslb->hugepage_sz)\n+\t\treturn 1;\n+\tif (msla->hugepage_sz > mslb->hugepage_sz)\n+\t\treturn -1;\n+\treturn 0;\n+}\n+\n+/* this will try lower page sizes first */\n+static void *\n+heap_alloc_on_socket(const char *type, size_t size, int socket,\n+\t\tunsigned flags, size_t align, size_t bound) {\n+\tstruct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;\n+\tstruct malloc_heap *heap = &mcfg->malloc_heaps[socket];\n+\tstruct rte_memseg_list *requested_msls[RTE_MAX_MEMSEG_LISTS];\n+\tstruct rte_memseg_list *other_msls[RTE_MAX_MEMSEG_LISTS];\n+\tint i, n_other_msls = 0, n_requested_msls = 0;\n+\tbool size_hint = (flags & RTE_MEMZONE_SIZE_HINT_ONLY) > 0;\n+\tunsigned size_flags = flags & ~RTE_MEMZONE_SIZE_HINT_ONLY;\n+\tvoid *ret;\n+\n+\trte_spinlock_lock(&(heap->lock));\n+\n+\t/* for legacy mode, try once and with all flags */\n+\tif (internal_config.legacy_mem) {\n+\t\tret = heap_alloc(heap, type, size, flags,\n+\t\t\t\talign == 0 ? 1 : align, bound);\n+\t\tgoto alloc_unlock;\n+\t}\n+\n+\t/*\n+\t * we do not pass the size hint here, because even if allocation fails,\n+\t * we may still be able to allocate memory from appropriate page sizes,\n+\t * we just need to request more memory first.\n+\t */\n+\tret = heap_alloc(heap, type, size, size_flags, align == 0 ? 1 : align,\n+\t\t\tbound);\n+\tif (ret != NULL)\n+\t\tgoto alloc_unlock;\n+\n+\tmemset(requested_msls, 0, sizeof(requested_msls));\n+\tmemset(other_msls, 0, sizeof(other_msls));\n+\n+\t/*\n+\t * go through memseg list and take note of all the page sizes available,\n+\t * and if any of them were specifically requested by the user.\n+\t */\n+\tfor (i = 0; i < RTE_MAX_MEMSEG_LISTS; i++) {\n+\t\tstruct rte_memseg_list *msl = &mcfg->memsegs[i];\n+\n+\t\tif (msl->socket_id != socket)\n+\t\t\tcontinue;\n+\n+\t\tif (msl->base_va == NULL)\n+\t\t\tcontinue;\n+\n+\t\t/* if pages of specific size were requested */\n+\t\tif (size_flags != 0 && check_hugepage_sz(size_flags,\n+\t\t\t\tmsl->hugepage_sz)) {\n+\t\t\trequested_msls[n_requested_msls++] = msl;\n+\t\t} else if (size_flags == 0 || size_hint) {\n+\t\t\tother_msls[n_other_msls++] = msl;\n+\t\t}\n+\t}\n+\n+\t/* sort the lists, smallest first */\n+\tqsort(requested_msls, n_requested_msls, sizeof(requested_msls[0]),\n+\t\t\tcompare_pagesz);\n+\tqsort(other_msls, n_other_msls, sizeof(other_msls[0]),\n+\t\t\tcompare_pagesz);\n+\n+\tfor (i = 0; i < n_requested_msls; i++) {\n+\t\tstruct rte_memseg_list *msl = requested_msls[i];\n+\n+\t\t/*\n+\t\t * do not pass the size hint here, as user expects other page\n+\t\t * sizes first, before resorting to best effort allocation.\n+\t\t */\n+\t\tret = try_expand_heap(heap, msl, type, size, socket, size_flags,\n+\t\t\t\talign, bound);\n+\t\tif (ret != NULL)\n+\t\t\tgoto alloc_unlock;\n+\t}\n+\tif (n_other_msls == 0)\n+\t\tgoto alloc_unlock;\n+\n+\t/* now, try reserving with size hint */\n+\tret = heap_alloc(heap, type, size, flags, align == 0 ? 1 : align,\n+\t\t\tbound);\n+\tif (ret != NULL)\n+\t\tgoto alloc_unlock;\n+\n+\t/*\n+\t * we still couldn't reserve memory, so try expanding heap with other\n+\t * page sizes, if there are any\n+\t */\n+\tfor (i = 0; i < n_other_msls; i++) {\n+\t\tstruct rte_memseg_list *msl = other_msls[i];\n+\n+\t\tret = try_expand_heap(heap, msl, type, size, socket, flags,\n+\t\t\t\talign, bound);\n+\t\tif (ret != NULL)\n+\t\t\tgoto alloc_unlock;\n+\t}\n+alloc_unlock:\n+\trte_spinlock_unlock(&(heap->lock));\n+\treturn ret;\n+}\n+\n+void *\n+malloc_heap_alloc(const char *type, size_t size, int socket_arg, unsigned flags,\n+\t\tsize_t align, size_t bound) {\n+\tint socket, i;\n+\tvoid *ret;\n+\n+\t/* return NULL if size is 0 or alignment is not power-of-2 */\n+\tif (size == 0 || (align && !rte_is_power_of_2(align)))\n+\t\treturn NULL;\n+\n+\tif (!rte_eal_has_hugepages())\n+\t\tsocket_arg = SOCKET_ID_ANY;\n+\n+\tif (socket_arg == SOCKET_ID_ANY)\n+\t\tsocket = malloc_get_numa_socket();\n+\telse\n+\t\tsocket = socket_arg;\n+\n+\t/* Check socket parameter */\n+\tif (socket >= RTE_MAX_NUMA_NODES)\n+\t\treturn NULL;\n+\n+\t// TODO: add warning for alignments bigger than page size if not VFIO\n+\n+\tret = heap_alloc_on_socket(type, size, socket, flags, align, bound);\n+\tif (ret != NULL || socket_arg != SOCKET_ID_ANY)\n+\t\treturn ret;\n+\n+\t/* try other heaps */\n+\tfor (i = 0; i < (int) rte_num_sockets(); i++) {\n+\t\tif (i == socket)\n+\t\t\tcontinue;\n+\t\tret = heap_alloc_on_socket(type, size, socket, flags,\n+\t\t\t\talign, bound);\n+\t\tif (ret != NULL)\n+\t\t\treturn ret;\n+\t}\n+\treturn NULL;\n+}\n+\n int\n malloc_heap_free(struct malloc_elem *elem) {\n \tstruct malloc_heap *heap;\n-\tint ret;\n+\tvoid *start, *aligned_start, *end, *aligned_end;\n+\tsize_t len, aligned_len;\n+\tconst struct rte_memseg_list *msl;\n+\tint n_pages, page_idx, max_page_idx, ret;\n \n \tif (!malloc_elem_cookies_ok(elem) || elem->state != ELEM_BUSY)\n \t\treturn -1;\n \n \t/* elem may be merged with previous element, so keep heap address */\n \theap = elem->heap;\n+\tmsl = elem->msl;\n \n \trte_spinlock_lock(&(heap->lock));\n \n-\tret = malloc_elem_free(elem);\n+\telem = malloc_elem_free(elem);\n \n-\trte_spinlock_unlock(&(heap->lock));\n+\t/* anything after this is a bonus */\n+\tret = 0;\n+\n+\t/* ...of which we can't avail if we are in legacy mode */\n+\tif (internal_config.legacy_mem)\n+\t\tgoto free_unlock;\n+\n+\t/* check if we can free any memory back to the system */\n+\tif (elem->size < msl->hugepage_sz)\n+\t\tgoto free_unlock;\n+\n+\t/* probably, but let's make sure, as we may not be using up full page */\n+\tstart = elem;\n+\tlen = elem->size;\n+\taligned_start = RTE_PTR_ALIGN_CEIL(start, msl->hugepage_sz);\n+\tend = RTE_PTR_ADD(elem, len);\n+\taligned_end = RTE_PTR_ALIGN_FLOOR(end, msl->hugepage_sz);\n \n+\taligned_len = RTE_PTR_DIFF(aligned_end, aligned_start);\n+\n+\t/* can't free anything */\n+\tif (aligned_len < msl->hugepage_sz)\n+\t\tgoto free_unlock;\n+\n+\tmalloc_elem_free_list_remove(elem);\n+\n+\tmalloc_elem_hide_region(elem, (void*) aligned_start, aligned_len);\n+\n+\t/* we don't really care if we fail to deallocate memory */\n+\tn_pages = aligned_len / msl->hugepage_sz;\n+\tpage_idx = RTE_PTR_DIFF(aligned_start, msl->base_va) / msl->hugepage_sz;\n+\tmax_page_idx = page_idx + n_pages;\n+\n+\tfor (; page_idx < max_page_idx; page_idx++) {\n+\t\tstruct rte_memseg *ms;\n+\n+\t\tms = rte_fbarray_get(&msl->memseg_arr, page_idx);\n+\t\teal_memalloc_free_page(ms);\n+\t\theap->total_size -= msl->hugepage_sz;\n+\t}\n+\n+\tRTE_LOG(DEBUG, EAL, \"Heap on socket %d was shrunk by %zdMB\\n\",\n+\t\tmsl->socket_id, aligned_len >> 20ULL);\n+free_unlock:\n+\trte_spinlock_unlock(&(heap->lock));\n \treturn ret;\n }\n \ndiff --git a/lib/librte_eal/common/malloc_heap.h b/lib/librte_eal/common/malloc_heap.h\nindex df04dd8..3fcd14f 100644\n--- a/lib/librte_eal/common/malloc_heap.h\n+++ b/lib/librte_eal/common/malloc_heap.h\n@@ -53,8 +53,8 @@ malloc_get_numa_socket(void)\n }\n \n void *\n-malloc_heap_alloc(struct malloc_heap *heap,\tconst char *type, size_t size,\n-\t\tunsigned flags, size_t align, size_t bound);\n+malloc_heap_alloc(const char *type, size_t size, int socket, unsigned flags,\n+\t\tsize_t align, size_t bound);\n \n int\n malloc_heap_free(struct malloc_elem *elem);\ndiff --git a/lib/librte_eal/common/rte_malloc.c b/lib/librte_eal/common/rte_malloc.c\nindex 92cd7d8..dc3199a 100644\n--- a/lib/librte_eal/common/rte_malloc.c\n+++ b/lib/librte_eal/common/rte_malloc.c\n@@ -68,9 +68,7 @@ void rte_free(void *addr)\n void *\n rte_malloc_socket(const char *type, size_t size, unsigned align, int socket_arg)\n {\n-\tstruct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;\n-\tint socket, i;\n-\tvoid *ret;\n+\tint socket;\n \n \t/* return NULL if size is 0 or alignment is not power-of-2 */\n \tif (size == 0 || (align && !rte_is_power_of_2(align)))\n@@ -88,24 +86,8 @@ rte_malloc_socket(const char *type, size_t size, unsigned align, int socket_arg)\n \tif (socket >= RTE_MAX_NUMA_NODES)\n \t\treturn NULL;\n \n-\tret = malloc_heap_alloc(&mcfg->malloc_heaps[socket], type,\n-\t\t\t\tsize, 0, align == 0 ? 1 : align, 0);\n-\tif (ret != NULL || socket_arg != SOCKET_ID_ANY)\n-\t\treturn ret;\n-\n-\t/* try other heaps */\n-\tfor (i = 0; i < RTE_MAX_NUMA_NODES; i++) {\n-\t\t/* we already tried this one */\n-\t\tif (i == socket)\n-\t\t\tcontinue;\n-\n-\t\tret = malloc_heap_alloc(&mcfg->malloc_heaps[i], type,\n-\t\t\t\t\tsize, 0, align == 0 ? 1 : align, 0);\n-\t\tif (ret != NULL)\n-\t\t\treturn ret;\n-\t}\n-\n-\treturn NULL;\n+\treturn malloc_heap_alloc(type, size, socket_arg, 0,\n+\t\t\talign == 0 ? 1 : align, 0);\n }\n \n /*\n", "prefixes": [ "dpdk-dev", "RFC", "v2", "16/23" ] }