From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-7.5 required=3.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH,MAILING_LIST_MULTI,SIGNED_OFF_BY,SPF_HELO_NONE,SPF_PASS, URIBL_SBL,URIBL_SBL_A,USER_AGENT_SANE_1 autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 6BFC4C76191 for ; Thu, 18 Jul 2019 09:28:45 +0000 (UTC) Received: from dpdk.org (dpdk.org [92.243.14.124]) by mail.kernel.org (Postfix) with ESMTP id F118120665 for ; Thu, 18 Jul 2019 09:28:44 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org F118120665 Authentication-Results: mail.kernel.org; dmarc=none (p=none dis=none) header.from=solarflare.com Authentication-Results: mail.kernel.org; spf=pass smtp.mailfrom=dev-bounces@dpdk.org Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 872421DBE; Thu, 18 Jul 2019 11:28:43 +0200 (CEST) Received: from dispatch1-us1.ppe-hosted.com (dispatch1-us1.ppe-hosted.com [67.231.154.164]) by dpdk.org (Postfix) with ESMTP id D38C4137C for ; Thu, 18 Jul 2019 11:28:41 +0200 (CEST) X-Virus-Scanned: Proofpoint Essentials engine Received: from webmail.solarflare.com (uk.solarflare.com [193.34.186.16]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-SHA384 (256/256 bits)) (No client certificate requested) by mx1-us1.ppe-hosted.com (PPE Hosted ESMTP Server) with ESMTPS id 4845C780069; Thu, 18 Jul 2019 09:28:40 +0000 (UTC) Received: from [192.168.1.11] (85.187.13.152) by ukex01.SolarFlarecom.com (10.17.10.4) with Microsoft SMTP Server (TLS) id 15.0.1395.4; Thu, 18 Jul 2019 10:28:33 +0100 To: Vamsi Krishna Attunuru , "dev@dpdk.org" CC: "thomas@monjalon.net" , Jerin Jacob Kollanukkaran , "olivier.matz@6wind.com" , "ferruh.yigit@intel.com" , "anatoly.burakov@intel.com" , "Kiran Kumar Kokkilagadda" References: <20190625035700.2953-1-vattunuru@marvell.com> <20190717090408.13717-1-vattunuru@marvell.com> <20190717090408.13717-2-vattunuru@marvell.com> From: Andrew Rybchenko Message-ID: Date: Thu, 18 Jul 2019 12:28:27 +0300 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101 Thunderbird/60.8.0 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset="utf-8"; format=flowed Content-Transfer-Encoding: 8bit Content-Language: en-US X-Originating-IP: [85.187.13.152] X-ClientProxiedBy: ocex03.SolarFlarecom.com (10.20.40.36) To ukex01.SolarFlarecom.com (10.17.10.4) X-TM-AS-Product-Ver: SMEX-12.5.0.1300-8.5.1010-24772.000 X-TM-AS-Result: No-24.342600-8.000000-10 X-TMASE-MatchedRID: j4nUk6F+aLboSitJVour/fZvT2zYoYOwt3aeg7g/usAutoY2UtFqGL+H BMNMGUuYwPb5NDq94CBLqcIPRwhVQCiq3P159fwphuhnXkqpF5O3rIgsEdTKmbxgMf9QE2ebT6Y y0anPBpZyZTZ7pLh4DR+7F/Ev8r3ZBen/uGyyHZZ1BumBmrnKW+Jc6hKWj0C1DfheddyhsqtZnA Nzk+5w6P7HRTil+hePBrbN4AZ6QnPyNWG6Utfmr/go8BKl9ae552mltlE2n8gOOOIzzESoE0bZ7 NFLQepsQORqKBYIytdR4Xyks9lwTTKQuzlt+SDHdhnFihmbnwXgoZy93PpPfC/h9VOvT6AgmMuC 9SVxFE2y0eK+KXVK1Fd4lLBWV1a7MDvZPMbvTD6FU125L7WL70HYlxxUTd1+yWCL+8tLbvbEOb3 Mgo89SEjOsAAPHXqRCyiZPgk444k2Qk9BrE1shJxVZzZr7+O7+TbyofkPZ4q5ZjHyzYrpGu/oWP Ze2PjsjwV/jkOaSQquUfWqF4phwDHqLqoo3JUawY28o+cGA5rJ5SXtoJPLyBlLPW+8b7SakBo2n a6+eDFqXnLG1dXX0psjZZgb5PNsN38xN8Cmv6wdt2/pSR7zwCz0ibuY6L9mVWQnHKxp38i/oixS Pi9jxa8fZYcUEozYETqKC/9i3WIqM+WBa4lkbBcr91Fo5aW9y0Q+dW8+UWRTq//rncsNWb7Itoz LGgGlNuMobdOtMrt/5RoKTr5Jh+ztpCSqSkXKFvAJe/W5wA8ZSo6PM4LsinY6dIju+3Vno8WMkQ Wv6iWhMIDkR/KfwPIAE33ZI208SVdy2uiYUWL6C0ePs7A07QKmARN5PTKc X-TM-AS-User-Approved-Sender: Yes X-TM-AS-User-Blocked-Sender: No X-TMASE-Result: 10--24.342600-8.000000 X-TMASE-Version: SMEX-12.5.0.1300-8.5.1010-24772.000 X-MDID: 1563442121-iIMpPO4f9e9q Subject: Re: [dpdk-dev] [PATCH v7 1/4] mempool: modify mempool populate() to skip objects from page boundaries 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: , Errors-To: dev-bounces@dpdk.org Sender: "dev" On 7/17/19 8:31 PM, Vamsi Krishna Attunuru wrote: >> -----Original Message----- >> From: Andrew Rybchenko >> Sent: Wednesday, July 17, 2019 7:07 PM >> To: Vamsi Krishna Attunuru ; dev@dpdk.org >> Cc: thomas@monjalon.net; Jerin Jacob Kollanukkaran ; >> olivier.matz@6wind.com; ferruh.yigit@intel.com; >> anatoly.burakov@intel.com; Kiran Kumar Kokkilagadda >> >> Subject: Re: [dpdk-dev] [PATCH v7 1/4] mempool: modify mempool >> populate() to skip objects from page boundaries >> >> On 7/17/19 12:04 PM, vattunuru@marvell.com wrote: >>> From: Vamsi Attunuru >>> >>> Currently the phys address of a mempool object populated by the >>> mempool populate default() routine may not be contiguous with in that >> mbuf range. >>> Patch ensures that each object's phys address is contiguous by >>> modifying default behaviour of mempool populate() to prevent objects >>> from being across 2 pages, expect if the size of object is bigger than size of >> page. >>> Since the overhead after this modification will be very minimal >>> considering the hugepage sizes of 512M & 1G, default behaviour is >>> modified except for the object sizes bigger than the page size. >>> >>> Signed-off-by: Vamsi Attunuru >>> Signed-off-by: Kiran Kumar K >> NACK >> >> Looking at MEMPOOL_F_NO_IOVA_CONTIG description I don't understand >> why the patch is necessary at all. So, I'd like to know more. Exact conditions, >> IOVA mode, hugepage sizes, mempool flags and how it is populated. >> > I presume the commit log clarifies the changes in the patch, pls correct me if it's not clear. The requirement is to create mempool of objects that each object's phys address in contiguous with in it's range, having flexibility to create such mempools helpful to the applications like KNI to operate in IOVA=VA mode where KNI kernel mode can safely translate IOVA addresses to PA. As I understand it breaks rte_mempool_populate_default() logic which assumes that page boundaries may be ignored in IOVA=VA mode (see no_pageshift =  ... and above description in the function). Sorry, right now I can't come up with the right fix, but as explained below suggested is wrong. > Regarding the exact conditions driven this approach are, when IOVA mode is set to VA and huge page size of 2MB/512MB used, since KNI application creates mempool without any flags set, mempool populate routine tends to reserve iova contiguous memzones and finally it ends up populating mempool with some objects that might being across two pages(over the page boundary), those mbuf's phys address might not be contiguous and these mempool are not suitable for operating KNI in IOVA=VA mode. > >>> --- >>> lib/librte_mempool/rte_mempool.c | 2 +- >>> lib/librte_mempool/rte_mempool_ops_default.c | 33 >> ++++++++++++++++++++++++++-- >>> 2 files changed, 32 insertions(+), 3 deletions(-) >>> >>> diff --git a/lib/librte_mempool/rte_mempool.c >>> b/lib/librte_mempool/rte_mempool.c >>> index 7260ce0..1c48325 100644 >>> --- a/lib/librte_mempool/rte_mempool.c >>> +++ b/lib/librte_mempool/rte_mempool.c >>> @@ -339,7 +339,7 @@ rte_mempool_populate_iova(struct rte_mempool >> *mp, char *vaddr, >>> i = rte_mempool_ops_populate(mp, mp->size - mp->populated_size, >>> (char *)vaddr + off, >>> (iova == RTE_BAD_IOVA) ? RTE_BAD_IOVA : (iova + off), >>> - len - off, mempool_add_elem, NULL); >>> + len - off, mempool_add_elem, opaque); >> The last argument is the callback opaque value. mempool_add_elem() does >> not use the opaque. But it is incorrect to use it for other purposes and >> require it to be memzone. > To avoid multiple changes in the mempool APIs for adding new variable, opaque has been leveraged here. I think there is no harm in having this approach since it carries the relevant info and quite suitable between the (*mempool_populate_t) calls. Sorry, but it enforces any caller of rte_mempool_populate_iova() (and rte_mempool_populate_virt()) pass memzone as opaque, but by API definition it is free_cb argument. So, if different free_cb is used, different opaque may be required. >>> /* not enough room to store one object */ >>> if (i == 0) { >>> diff --git a/lib/librte_mempool/rte_mempool_ops_default.c >>> b/lib/librte_mempool/rte_mempool_ops_default.c >>> index 4e2bfc8..85da264 100644 >>> --- a/lib/librte_mempool/rte_mempool_ops_default.c >>> +++ b/lib/librte_mempool/rte_mempool_ops_default.c >>> @@ -45,19 +45,48 @@ rte_mempool_op_calc_mem_size_default(const >> struct rte_mempool *mp, >>> return mem_size; >>> } >>> >>> +/* Returns -1 if object falls on a page boundary, else returns 0 */ >>> +static inline int mempool_check_obj_bounds(void *obj, uint64_t >>> +hugepage_sz, size_t elt_sz) { >>> + uintptr_t page_end, elt_addr = (uintptr_t)obj; >>> + uint32_t pg_shift = rte_bsf32(hugepage_sz); >>> + uint64_t page_mask; >>> + >>> + page_mask = ~((1ull << pg_shift) - 1); >>> + page_end = (elt_addr & page_mask) + hugepage_sz; >>> + >>> + if (elt_addr + elt_sz > page_end) >>> + return -1; >>> + >>> + return 0; >>> +} >>> + >>> int >>> rte_mempool_op_populate_default(struct rte_mempool *mp, unsigned >> int max_objs, >>> void *vaddr, rte_iova_t iova, size_t len, >>> rte_mempool_populate_obj_cb_t *obj_cb, void *obj_cb_arg) >>> { >>> - size_t total_elt_sz; >>> - size_t off; >>> + struct rte_memzone *mz = obj_cb_arg; Moreover, the change enforces obj_cb_arg to be memzone, but it is obj_cb argument and obj_cb defines what should be in the obj_cb_arg. >>> + size_t total_elt_sz, off; >> Why two variables are combined into one here? It is unrelated change. >> >>> unsigned int i; >>> void *obj; >>> >>> total_elt_sz = mp->header_size + mp->elt_size + mp->trailer_size; >>> >>> for (off = 0, i = 0; off + total_elt_sz <= len && i < max_objs; >>> i++) { >>> + >>> + /* Skip page boundary check if element is bigger than page */ >>> + if (mz->hugepage_sz >= total_elt_sz) { >>> + if (mempool_check_obj_bounds((char *)vaddr + off, >>> + mz->hugepage_sz, >>> + total_elt_sz) < 0) { >>> + i--; /* Decrement count & skip this obj */ >>> + off += total_elt_sz; >>> + continue; >>> + } >>> + } >>> + >> What I don't like here is that it makes one memory chunk insufficient to >> populate all objects. I.e. we calculated memory chunk size required, but >> skipped some object. May be it is not a problem, but breaks the logic existing >> in the code. >> >>> off += mp->header_size; >>> obj = (char *)vaddr + off; >>> obj_cb(mp, obj_cb_arg, obj,