Patch Detail
get:
Show a patch.
patch:
Update a patch.
put:
Update a patch.
GET /api/patches/62231/?format=api
{ "id": 62231, "url": "http://patchwork.dpdk.org/api/patches/62231/?format=api", "web_url": "http://patchwork.dpdk.org/project/dpdk/patch/20191030143619.4007-6-olivier.matz@6wind.com/", "project": { "id": 1, "url": "http://patchwork.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": "<20191030143619.4007-6-olivier.matz@6wind.com>", "list_archive_url": "https://inbox.dpdk.org/dev/20191030143619.4007-6-olivier.matz@6wind.com", "date": "2019-10-30T14:36:18", "name": "[v2,5/6] mempool: prevent objects from being across pages", "commit_ref": null, "pull_url": null, "state": "superseded", "archived": true, "hash": "d144d3e1fce8459488ecc117ab4146593cc86a3a", "submitter": { "id": 8, "url": "http://patchwork.dpdk.org/api/people/8/?format=api", "name": "Olivier Matz", "email": "olivier.matz@6wind.com" }, "delegate": null, "mbox": "http://patchwork.dpdk.org/project/dpdk/patch/20191030143619.4007-6-olivier.matz@6wind.com/mbox/", "series": [ { "id": 7161, "url": "http://patchwork.dpdk.org/api/series/7161/?format=api", "web_url": "http://patchwork.dpdk.org/project/dpdk/list/?series=7161", "date": "2019-10-30T14:36:15", "name": "mempool: avoid objects allocations across pages", "version": 2, "mbox": "http://patchwork.dpdk.org/series/7161/mbox/" } ], "comments": "http://patchwork.dpdk.org/api/patches/62231/comments/", "check": "fail", "checks": "http://patchwork.dpdk.org/api/patches/62231/checks/", "tags": {}, "related": [], "headers": { "Return-Path": "<dev-bounces@dpdk.org>", "X-Original-To": "patchwork@inbox.dpdk.org", "Delivered-To": "patchwork@inbox.dpdk.org", "Received": [ "from dpdk.org (dpdk.org [92.243.14.124])\n\tby inbox.dpdk.org (Postfix) with ESMTP id CE972A00BE;\n\tWed, 30 Oct 2019 15:37:58 +0100 (CET)", "from [92.243.14.124] (localhost [127.0.0.1])\n\tby dpdk.org (Postfix) with ESMTP id 33A2B1C065;\n\tWed, 30 Oct 2019 15:37:12 +0100 (CET)", "from proxy.6wind.com (host.76.145.23.62.rev.coltfrance.com\n [62.23.145.76]) by dpdk.org (Postfix) with ESMTP id A5E891BFD4\n for <dev@dpdk.org>; Wed, 30 Oct 2019 15:36:57 +0100 (CET)", "from glumotte.dev.6wind.com. (unknown [10.16.0.195])\n by proxy.6wind.com (Postfix) with ESMTP id E17A9338A4D;\n Wed, 30 Oct 2019 15:36:55 +0100 (CET)" ], "From": "Olivier Matz <olivier.matz@6wind.com>", "To": "dev@dpdk.org", "Cc": "Anatoly Burakov <anatoly.burakov@intel.com>,\n Andrew Rybchenko <arybchenko@solarflare.com>,\n Ferruh Yigit <ferruh.yigit@linux.intel.com>,\n \"Giridharan, Ganesan\" <ggiridharan@rbbn.com>,\n Jerin Jacob Kollanukkaran <jerinj@marvell.com>,\n \"Kiran Kumar Kokkilagadda\" <kirankumark@marvell.com>,\n Stephen Hemminger <sthemmin@microsoft.com>,\n Thomas Monjalon <thomas@monjalon.net>,\n Vamsi Krishna Attunuru <vattunuru@marvell.com>", "Date": "Wed, 30 Oct 2019 15:36:18 +0100", "Message-Id": "<20191030143619.4007-6-olivier.matz@6wind.com>", "X-Mailer": "git-send-email 2.20.1", "In-Reply-To": "<20191030143619.4007-1-olivier.matz@6wind.com>", "References": "<20190719133845.32432-1-olivier.matz@6wind.com>\n <20191030143619.4007-1-olivier.matz@6wind.com>", "MIME-Version": "1.0", "Content-Transfer-Encoding": "8bit", "Subject": "[dpdk-dev] [PATCH v2 5/6] mempool: prevent objects from being\n\tacross pages", "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://mails.dpdk.org/options/dev>,\n <mailto:dev-request@dpdk.org?subject=unsubscribe>", "List-Archive": "<http://mails.dpdk.org/archives/dev/>", "List-Post": "<mailto:dev@dpdk.org>", "List-Help": "<mailto:dev-request@dpdk.org?subject=help>", "List-Subscribe": "<https://mails.dpdk.org/listinfo/dev>,\n <mailto:dev-request@dpdk.org?subject=subscribe>", "Errors-To": "dev-bounces@dpdk.org", "Sender": "\"dev\" <dev-bounces@dpdk.org>" }, "content": "When populating a mempool, ensure that objects are not located across\nseveral pages, except if user did not request iova contiguous objects.\n\nSigned-off-by: Vamsi Krishna Attunuru <vattunuru@marvell.com>\nSigned-off-by: Olivier Matz <olivier.matz@6wind.com>\n---\n drivers/mempool/octeontx2/Makefile | 3 +\n drivers/mempool/octeontx2/meson.build | 3 +\n drivers/mempool/octeontx2/otx2_mempool_ops.c | 119 ++++++++++++++++---\n lib/librte_mempool/rte_mempool.c | 23 ++--\n lib/librte_mempool/rte_mempool_ops_default.c | 32 ++++-\n 5 files changed, 147 insertions(+), 33 deletions(-)", "diff": "diff --git a/drivers/mempool/octeontx2/Makefile b/drivers/mempool/octeontx2/Makefile\nindex 87cce22c6..d781cbfc6 100644\n--- a/drivers/mempool/octeontx2/Makefile\n+++ b/drivers/mempool/octeontx2/Makefile\n@@ -27,6 +27,9 @@ EXPORT_MAP := rte_mempool_octeontx2_version.map\n \n LIBABIVER := 1\n \n+# for rte_mempool_get_page_size\n+CFLAGS += -DALLOW_EXPERIMENTAL_API\n+\n #\n # all source are stored in SRCS-y\n #\ndiff --git a/drivers/mempool/octeontx2/meson.build b/drivers/mempool/octeontx2/meson.build\nindex 9fde40f0e..28f9634da 100644\n--- a/drivers/mempool/octeontx2/meson.build\n+++ b/drivers/mempool/octeontx2/meson.build\n@@ -21,3 +21,6 @@ foreach flag: extra_flags\n endforeach\n \n deps += ['eal', 'mbuf', 'kvargs', 'bus_pci', 'common_octeontx2', 'mempool']\n+\n+# for rte_mempool_get_page_size\n+allow_experimental_apis = true\ndiff --git a/drivers/mempool/octeontx2/otx2_mempool_ops.c b/drivers/mempool/octeontx2/otx2_mempool_ops.c\nindex d769575f4..47117aec6 100644\n--- a/drivers/mempool/octeontx2/otx2_mempool_ops.c\n+++ b/drivers/mempool/octeontx2/otx2_mempool_ops.c\n@@ -713,12 +713,76 @@ static ssize_t\n otx2_npa_calc_mem_size(const struct rte_mempool *mp, uint32_t obj_num,\n \t\t uint32_t pg_shift, size_t *min_chunk_size, size_t *align)\n {\n-\t/*\n-\t * Simply need space for one more object to be able to\n-\t * fulfill alignment requirements.\n-\t */\n-\treturn rte_mempool_op_calc_mem_size_default(mp, obj_num + 1, pg_shift,\n-\t\t\t\t\t\t min_chunk_size, align);\n+\tsize_t total_elt_sz;\n+\tsize_t obj_per_page, pg_sz, objs_in_last_page;\n+\tsize_t mem_size;\n+\n+\t/* derived from rte_mempool_op_calc_mem_size_default() */\n+\n+\ttotal_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;\n+\n+\tif (total_elt_sz == 0) {\n+\t\tmem_size = 0;\n+\t} else if (pg_shift == 0) {\n+\t\t/* one object margin to fix alignment */\n+\t\tmem_size = total_elt_sz * (obj_num + 1);\n+\t} else {\n+\t\tpg_sz = (size_t)1 << pg_shift;\n+\t\tobj_per_page = pg_sz / total_elt_sz;\n+\n+\t\t/* we need to keep one object to fix alignment */\n+\t\tif (obj_per_page > 0)\n+\t\t\tobj_per_page--;\n+\n+\t\tif (obj_per_page == 0) {\n+\t\t\t/*\n+\t\t\t * Note that if object size is bigger than page size,\n+\t\t\t * then it is assumed that pages are grouped in subsets\n+\t\t\t * of physically continuous pages big enough to store\n+\t\t\t * at least one object.\n+\t\t\t */\n+\t\t\tmem_size = RTE_ALIGN_CEIL(2 * total_elt_sz,\n+\t\t\t\t\t\tpg_sz) * obj_num;\n+\t\t} else {\n+\t\t\t/* In the best case, the allocator will return a\n+\t\t\t * page-aligned address. For example, with 5 objs,\n+\t\t\t * the required space is as below:\n+\t\t\t * | page0 | page1 | page2 (last) |\n+\t\t\t * |obj0 |obj1 |xxx|obj2 |obj3 |xxx|obj4|\n+\t\t\t * <------------- mem_size ------------->\n+\t\t\t */\n+\t\t\tobjs_in_last_page = ((obj_num - 1) % obj_per_page) + 1;\n+\t\t\t/* room required for the last page */\n+\t\t\tmem_size = objs_in_last_page * total_elt_sz;\n+\t\t\t/* room required for other pages */\n+\t\t\tmem_size += ((obj_num - objs_in_last_page) /\n+\t\t\t\tobj_per_page) << pg_shift;\n+\n+\t\t\t/* In the worst case, the allocator returns a\n+\t\t\t * non-aligned pointer, wasting up to\n+\t\t\t * total_elt_sz. Add a margin for that.\n+\t\t\t */\n+\t\t\t mem_size += total_elt_sz - 1;\n+\t\t}\n+\t}\n+\n+\t*min_chunk_size = total_elt_sz * 2;\n+\t*align = RTE_CACHE_LINE_SIZE;\n+\n+\treturn mem_size;\n+}\n+\n+/* Returns -1 if object crosses a page boundary, else returns 0 */\n+static int\n+check_obj_bounds(char *obj, size_t pg_sz, size_t elt_sz)\n+{\n+\tif (pg_sz == 0)\n+\t\treturn 0;\n+\tif (elt_sz > pg_sz)\n+\t\treturn 0;\n+\tif (RTE_PTR_ALIGN(obj, pg_sz) != RTE_PTR_ALIGN(obj + elt_sz - 1, pg_sz))\n+\t\treturn -1;\n+\treturn 0;\n }\n \n static int\n@@ -726,8 +790,12 @@ otx2_npa_populate(struct rte_mempool *mp, unsigned int max_objs, void *vaddr,\n \t\t rte_iova_t iova, size_t len,\n \t\t rte_mempool_populate_obj_cb_t *obj_cb, void *obj_cb_arg)\n {\n-\tsize_t total_elt_sz;\n+\tchar *va = vaddr;\n+\tsize_t total_elt_sz, pg_sz;\n \tsize_t off;\n+\tunsigned int i;\n+\tvoid *obj;\n+\tint ret;\n \n \tif (iova == RTE_BAD_IOVA)\n \t\treturn -EINVAL;\n@@ -735,22 +803,45 @@ otx2_npa_populate(struct rte_mempool *mp, unsigned int max_objs, void *vaddr,\n \ttotal_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;\n \n \t/* Align object start address to a multiple of total_elt_sz */\n-\toff = total_elt_sz - ((uintptr_t)vaddr % total_elt_sz);\n+\toff = total_elt_sz - (((uintptr_t)(va - 1) % total_elt_sz) + 1);\n \n \tif (len < off)\n \t\treturn -EINVAL;\n \n-\tvaddr = (char *)vaddr + off;\n-\tiova += off;\n-\tlen -= off;\n \n-\tnpa_lf_aura_op_range_set(mp->pool_id, iova, iova + len);\n+\tnpa_lf_aura_op_range_set(mp->pool_id, iova + off, iova + len - off);\n \n \tif (npa_lf_aura_range_update_check(mp->pool_id) < 0)\n \t\treturn -EBUSY;\n \n-\treturn rte_mempool_op_populate_default(mp, max_objs, vaddr, iova, len,\n-\t\t\t\t\t obj_cb, obj_cb_arg);\n+\t/* the following is derived from rte_mempool_op_populate_default() */\n+\n+\tret = rte_mempool_get_page_size(mp, &pg_sz);\n+\tif (ret < 0)\n+\t\treturn ret;\n+\n+\tfor (i = 0; i < max_objs; i++) {\n+\t\t/* avoid objects to cross page boundaries, and align\n+\t\t * offset to a multiple of total_elt_sz.\n+\t\t */\n+\t\tif (check_obj_bounds(va + off, pg_sz, total_elt_sz) < 0) {\n+\t\t\toff += RTE_PTR_ALIGN_CEIL(va + off, pg_sz) - (va + off);\n+\t\t\toff += total_elt_sz - (((uintptr_t)(va + off - 1) %\n+\t\t\t\t\t\ttotal_elt_sz) + 1);\n+\t\t}\n+\n+\t\tif (off + total_elt_sz > len)\n+\t\t\tbreak;\n+\n+\t\toff += mp->header_size;\n+\t\tobj = va + off;\n+\t\tobj_cb(mp, obj_cb_arg, obj,\n+\t\t (iova == RTE_BAD_IOVA) ? RTE_BAD_IOVA : (iova + off));\n+\t\trte_mempool_ops_enqueue_bulk(mp, &obj, 1);\n+\t\toff += mp->elt_size + mp->trailer_size;\n+\t}\n+\n+\treturn i;\n }\n \n static struct rte_mempool_ops otx2_npa_ops = {\ndiff --git a/lib/librte_mempool/rte_mempool.c b/lib/librte_mempool/rte_mempool.c\nindex 758c5410b..d3db9273d 100644\n--- a/lib/librte_mempool/rte_mempool.c\n+++ b/lib/librte_mempool/rte_mempool.c\n@@ -431,8 +431,6 @@ rte_mempool_get_page_size(struct rte_mempool *mp, size_t *pg_sz)\n \n \tif (!need_iova_contig_obj)\n \t\t*pg_sz = 0;\n-\telse if (!alloc_in_ext_mem && rte_eal_iova_mode() == RTE_IOVA_VA)\n-\t\t*pg_sz = 0;\n \telse if (rte_eal_has_hugepages() || alloc_in_ext_mem)\n \t\t*pg_sz = get_min_page_size(mp->socket_id);\n \telse\n@@ -481,17 +479,15 @@ rte_mempool_populate_default(struct rte_mempool *mp)\n \t * then just set page shift and page size to 0, because the user has\n \t * indicated that there's no need to care about anything.\n \t *\n-\t * if we do need contiguous objects, there is also an option to reserve\n-\t * the entire mempool memory as one contiguous block of memory, in\n-\t * which case the page shift and alignment wouldn't matter as well.\n+\t * if we do need contiguous objects (if a mempool driver has its\n+\t * own calc_size() method returning min_chunk_size = mem_size),\n+\t * there is also an option to reserve the entire mempool memory\n+\t * as one contiguous block of memory.\n \t *\n \t * if we require contiguous objects, but not necessarily the entire\n-\t * mempool reserved space to be contiguous, then there are two options.\n-\t *\n-\t * if our IO addresses are virtual, not actual physical (IOVA as VA\n-\t * case), then no page shift needed - our memory allocation will give us\n-\t * contiguous IO memory as far as the hardware is concerned, so\n-\t * act as if we're getting contiguous memory.\n+\t * mempool reserved space to be contiguous, pg_sz will be != 0,\n+\t * and the default ops->populate() will take care of not placing\n+\t * objects across pages.\n \t *\n \t * if our IO addresses are physical, we may get memory from bigger\n \t * pages, or we might get memory from smaller pages, and how much of it\n@@ -504,11 +500,6 @@ rte_mempool_populate_default(struct rte_mempool *mp)\n \t *\n \t * If we fail to get enough contiguous memory, then we'll go and\n \t * reserve space in smaller chunks.\n-\t *\n-\t * We also have to take into account the fact that memory that we're\n-\t * going to allocate from can belong to an externally allocated memory\n-\t * area, in which case the assumption of IOVA as VA mode being\n-\t * synonymous with IOVA contiguousness will not hold.\n \t */\n \n \tneed_iova_contig_obj = !(mp->flags & MEMPOOL_F_NO_IOVA_CONTIG);\ndiff --git a/lib/librte_mempool/rte_mempool_ops_default.c b/lib/librte_mempool/rte_mempool_ops_default.c\nindex f6aea7662..e5cd4600f 100644\n--- a/lib/librte_mempool/rte_mempool_ops_default.c\n+++ b/lib/librte_mempool/rte_mempool_ops_default.c\n@@ -61,21 +61,47 @@ rte_mempool_op_calc_mem_size_default(const struct rte_mempool *mp,\n \treturn mem_size;\n }\n \n+/* Returns -1 if object crosses a page boundary, else returns 0 */\n+static int\n+check_obj_bounds(char *obj, size_t pg_sz, size_t elt_sz)\n+{\n+\tif (pg_sz == 0)\n+\t\treturn 0;\n+\tif (elt_sz > pg_sz)\n+\t\treturn 0;\n+\tif (RTE_PTR_ALIGN(obj, pg_sz) != RTE_PTR_ALIGN(obj + elt_sz - 1, pg_sz))\n+\t\treturn -1;\n+\treturn 0;\n+}\n+\n int\n rte_mempool_op_populate_default(struct rte_mempool *mp, unsigned int max_objs,\n \t\tvoid *vaddr, rte_iova_t iova, size_t len,\n \t\trte_mempool_populate_obj_cb_t *obj_cb, void *obj_cb_arg)\n {\n-\tsize_t total_elt_sz;\n+\tchar *va = vaddr;\n+\tsize_t total_elt_sz, pg_sz;\n \tsize_t off;\n \tunsigned int i;\n \tvoid *obj;\n+\tint ret;\n+\n+\tret = rte_mempool_get_page_size(mp, &pg_sz);\n+\tif (ret < 0)\n+\t\treturn ret;\n \n \ttotal_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size;\n \n-\tfor (off = 0, i = 0; off + total_elt_sz <= len && i < max_objs; i++) {\n+\tfor (off = 0, i = 0; i < max_objs; i++) {\n+\t\t/* avoid objects to cross page boundaries */\n+\t\tif (check_obj_bounds(va + off, pg_sz, total_elt_sz) < 0)\n+\t\t\toff += RTE_PTR_ALIGN_CEIL(va + off, pg_sz) - (va + off);\n+\n+\t\tif (off + total_elt_sz > len)\n+\t\t\tbreak;\n+\n \t\toff += mp->header_size;\n-\t\tobj = (char *)vaddr + off;\n+\t\tobj = va + off;\n \t\tobj_cb(mp, obj_cb_arg, obj,\n \t\t (iova == RTE_BAD_IOVA) ? RTE_BAD_IOVA : (iova + off));\n \t\trte_mempool_ops_enqueue_bulk(mp, &obj, 1);\n", "prefixes": [ "v2", "5/6" ] }