From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id F304043E70 for ; Tue, 16 Apr 2024 13:36:31 +0200 (CEST) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id E6F974029E; Tue, 16 Apr 2024 13:36:31 +0200 (CEST) Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) by mails.dpdk.org (Postfix) with ESMTP id 5810540262; Tue, 16 Apr 2024 13:36:30 +0200 (CEST) Received: from mail.maildlp.com (unknown [172.18.186.31]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4VJhh418qTz6J80b; Tue, 16 Apr 2024 19:31:32 +0800 (CST) Received: from frapeml500007.china.huawei.com (unknown [7.182.85.172]) by mail.maildlp.com (Postfix) with ESMTPS id 451F31400F4; Tue, 16 Apr 2024 19:36:28 +0800 (CST) Received: from frapeml500007.china.huawei.com (7.182.85.172) by frapeml500007.china.huawei.com (7.182.85.172) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.1.2507.35; Tue, 16 Apr 2024 13:36:28 +0200 Received: from frapeml500007.china.huawei.com ([7.182.85.172]) by frapeml500007.china.huawei.com ([7.182.85.172]) with mapi id 15.01.2507.035; Tue, 16 Apr 2024 13:36:28 +0200 From: Konstantin Ananyev To: Konstantin Ananyev , =?iso-8859-1?Q?Morten_Br=F8rup?= , David Marchand , "dev@dpdk.org" CC: "thomas@monjalon.net" , "ferruh.yigit@amd.com" , "stable@dpdk.org" , Olivier Matz , Jijiang Liu , "Andrew Rybchenko" , Ferruh Yigit , Kaiwen Deng , "qiming.yang@intel.com" , "yidingx.zhou@intel.com" , Aman Singh , "Yuying Zhang" , Thomas Monjalon , "Jerin Jacob" Subject: RE: [PATCH v2 3/8] mbuf: fix Tx checksum offload examples Thread-Topic: [PATCH v2 3/8] mbuf: fix Tx checksum offload examples Thread-Index: AQHah2gcwh+zxBWPrEOmSW2nVSxrXbFZuegAgAX4BtCAAE6MMIABTCywgAAbf0CAAzI3UIAAC0aQgAAit/CAAAXFUIAF1aYQgAAgzWA= Date: Tue, 16 Apr 2024 11:36:27 +0000 Message-ID: References: <20240405125039.897933-1-david.marchand@redhat.com> <20240405144604.906695-1-david.marchand@redhat.com> <20240405144604.906695-4-david.marchand@redhat.com> <98CBD80474FA8B44BF855DF32C47DC35E9F36C@smartserver.smartshare.dk> <10b564b42f8d4db387f6302701f24ce3@huawei.com> <98CBD80474FA8B44BF855DF32C47DC35E9F381@smartserver.smartshare.dk> <409157f5da3e4c628ca678dd9e2c7957@huawei.com> <98CBD80474FA8B44BF855DF32C47DC35E9F38F@smartserver.smartshare.dk> <52850a78c83445548a0b78bfd04e6f91@huawei.com> <98CBD80474FA8B44BF855DF32C47DC35E9F39F@smartserver.smartshare.dk> <98CBD80474FA8B44BF855DF32C47DC35E9F3A0@smartserver.smartshare.dk> <3f8214e0bcb448338cf2679f753a983d@huawei.com> In-Reply-To: <3f8214e0bcb448338cf2679f753a983d@huawei.com> Accept-Language: en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-originating-ip: [10.206.138.42] Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable MIME-Version: 1.0 X-BeenThere: stable@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: patches for DPDK stable branches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: stable-bounces@dpdk.org > > > > > > > > > > > Mandate use of rte_eth_tx_prepare() in the mbuf Tx > > > checksum > > > > > offload > > > > > > > > > > > examples. > > > > > > > > > > > > > > > > > > > > I strongly disagree with this change! > > > > > > > > > > > > > > > > > > > > It will cause a huge performance degradation for shapin= g > > > > > applications: > > > > > > > > > > > > > > > > > > > > A packet will be processed and finalized at an output o= r > > > > > forwarding > > > > > > > > > pipeline stage, where some other fields might also be > > > written, > > > > > so > > > > > > > > > > zeroing e.g. the out_ip checksum at this stage has low > > > cost > > > > > (no new > > > > > > > > > cache misses). > > > > > > > > > > > > > > > > > > > > Then, the packet might be queued for QoS or similar. > > > > > > > > > > > > > > > > > > > > If rte_eth_tx_prepare() must be called at the egress > > > pipeline > > > > > stage, > > > > > > > > > it has to write to the packet and cause a cache miss per > > > packet, > > > > > > > > > > instead of simply passing on the packet to the NIC > > > hardware. > > > > > > > > > > > > > > > > > > > > It must be possible to finalize the packet at the > > > > > output/forwarding > > > > > > > > > pipeline stage! > > > > > > > > > > > > > > > > > > If you can finalize your packet on output/forwarding, th= en > > > why > > > > > you > > > > > > > > > can't invoke tx_prepare() on the same stage? > > > > > > > > > There seems to be some misunderstanding about what > > > tx_prepare() > > > > > does - > > > > > > > > > in fact it doesn't communicate with HW queue (doesn't upd= ate > > > TXD > > > > > ring, > > > > > > > > > etc.), what it does - just make changes in mbuf itself. > > > > > > > > > Yes, it reads some fields in SW TX queue struct (max numb= er > > > of > > > > > TXDs per > > > > > > > > > packet, etc.), but AFAIK it is safe > > > > > > > > > to call tx_prepare() and tx_burst() from different thread= s. > > > > > > > > > At least on implementations I am aware about. > > > > > > > > > Just checked the docs - it seems not stated explicitly > > > anywhere, > > > > > might > > > > > > > > > be that's why it causing such misunderstanding. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > Also, how is rte_eth_tx_prepare() supposed to work for > > > cloned > > > > > packets > > > > > > > > > egressing on different NIC hardware? > > > > > > > > > > > > > > > > > > If you create a clone of full packet (including L2/L3) > > > headers > > > > > then > > > > > > > > > obviously such construction might not > > > > > > > > > work properly with tx_prepare() over two different NICs. > > > > > > > > > Though In majority of cases you do clone segments with da= ta, > > > > > while at > > > > > > > > > least L2 headers are put into different segments. > > > > > > > > > One simple approach would be to keep L3 header in that > > > separate > > > > > segment. > > > > > > > > > But yes, there is a problem when you'll need to send exac= tly > > > the > > > > > same > > > > > > > > > packet over different NICs. > > > > > > > > > As I remember, for bonding PMD things don't work quite we= ll > > > here > > > > > - you > > > > > > > > > might have a bond over 2 NICs with > > > > > > > > > different tx_prepare() and which one to call might be not > > > clear > > > > > till > > > > > > > > > actual PMD tx_burst() is invoked. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > In theory, it might get even worse if we make this opaq= ue > > > > > instead of > > > > > > > > > transparent and standardized: > > > > > > > > > > One PMD might reset out_ip checksum to 0x0000, and anot= her > > > PMD > > > > > might > > > > > > > > > reset it to 0xFFFF. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > I can only see one solution: > > > > > > > > > > We need to standardize on common minimum requirements f= or > > > how > > > > > to > > > > > > > > > prepare packets for each TX offload. > > > > > > > > > > > > > > > > > > If we can make each and every vendor to agree here - that > > > > > definitely > > > > > > > > > will help to simplify things quite a bit. > > > > > > > > > > > > > > > > An API is more than a function name and parameters. > > > > > > > > It also has preconditions and postconditions. > > > > > > > > > > > > > > > > All major NIC vendors are contributing to DPDK. > > > > > > > > It should be possible to reach consensus for reasonable > > > minimum > > > > > requirements > > > > > > > for offloads. > > > > > > > > Hardware- and driver-specific exceptions can be documented > > > with > > > > > the offload > > > > > > > flag, or with rte_eth_rx/tx_burst(), like the note to > > > > > > > > rte_eth_rx_burst(): > > > > > > > > "Some drivers using vector instructions require that nb_pkt= s > > > is > > > > > divisible by > > > > > > > 4 or 8, depending on the driver implementation." > > > > > > > > > > > > > > If we introduce a rule that everyone supposed to follow and t= hen > > > > > straightway > > > > > > > allow people to have a 'documented exceptions', > > > > > > > for me it means like 'no rule' in practice. > > > > > > > A 'documented exceptions' approach might work if you have 5 > > > > > different PMDs to > > > > > > > support, but not when you have 50+. > > > > > > > No-one would write an app with possible 10 different exceptio= n > > > cases > > > > > in his > > > > > > > head. > > > > > > > Again, with such approach we can forget about backward > > > > > compatibility. > > > > > > > I think we already had this discussion before, my opinion > > > remains > > > > > the same > > > > > > > here - > > > > > > > 'documented exceptions' approach is a way to trouble. > > > > > > > > > > > > The "minimum requirements" should be the lowest common denomina= tor > > > of > > > > > all NICs. > > > > > > Exceptions should be extremely few, for outlier NICs that still > > > want > > > > > to provide an offload and its driver is unable to live up to the > > > > > > minimum requirements. > > > > > > Any exception should require techboard approval. If a NIC/drive= r > > > does > > > > > not support the "minimum requirements" for an offload > > > > > > feature, it is not allowed to claim support for that offload > > > feature, > > > > > or needs to seek approval for an exception. > > > > > > > > > > > > As another option for NICs not supporting the minimum requireme= nts > > > of > > > > > an offload feature, we could introduce offload flags with > > > > > > finer granularity. E.g. one offload flag for "gold standard" TX > > > > > checksum update (where the packet's checksum field can have any > > > > > > value), and another offload flag for "silver standard" TX check= sum > > > > > update (where the packet's checksum field must have a > > > > > > precomputed value). > > > > > > > > > > Actually yes, I was thinking in the same direction - we need some > > > extra > > > > > API to allow user to distinguish. > > > > > Probably we can do something like that: a new API for the ethdev > > > call > > > > > that would take as a parameter > > > > > TX offloads bitmap and in return specify would it need to modify > > > > > contents of packet to support these > > > > > offloads or not. > > > > > Something like: > > > > > int rte_ethdev_tx_offload_pkt_mod_required(unt64_t tx_offloads) > > > > > > > > > > For the majority of the drivers that satisfy these "minimum > > > > > requirements" corresponding devops > > > > > entry will be empty and we'll always return 0, otherwise PMD has = to > > > > > provide a proper devop. > > > > > Then again, it would be up to the user, to determine can he pass > > > same > > > > > packet to 2 different NICs or not. > > > > > > > > > > I suppose it is similar to what you were talking about? > > > > > > > > I was thinking something more simple: > > > > > > > > The NIC exposes its RX and TX offload capabilities to the applicati= on > > > through the rx/tx_offload_capa and other fields in the > > > > rte_eth_dev_info structure returned by rte_eth_dev_info_get(). > > > > > > > > E.g. tx_offload_capa might have the RTE_ETH_TX_OFFLOAD_IPV4_CKSUM f= lag > > > set. > > > > These capability flags (or enums) are mostly undocumented in the co= de, > > > but I guess that the RTE_ETH_TX_OFFLOAD_IPV4_CKSUM > > > > capability means that the NIC is able to update the IPv4 header > > > checksum at egress (on the wire, i.e. without modifying the mbuf or > > > > packet data), and that the application must set RTE_MBUF_F_TX_IP_CK= SUM > > > in the mbufs to utilize this offload. > > > > I would define and document what each capability flag/enum exactly > > > means, the minimum requirements (as defined by the DPDK > > > > community) for the driver to claim support for it, and the > > > requirements for an application to use it. > > > > For the sake of discussion, let's say that > > > RTE_ETH_TX_OFFLOAD_IPV4_CKSUM means "gold standard" TX checksum updat= e > > > capability > > > > (i.e. no requirements to the checksum field in the packet contents)= . > > > > If some NIC requires the checksum field in the packet contents to h= ave > > > a precomputed value, the NIC would not be allowed to claim > > > > the RTE_ETH_TX_OFFLOAD_IPV4_CKSUM capability. > > > > Such a NIC would need to define and document a new capability, e.g. > > > RTE_ETH_TX_OFFLOAD_IPV4_CKSUM_ASSISTED, for the "silver > > > > standard" TX checksum update capability. > > > > In other words: I would encode variations of offload capabilities > > > directly in the capabilities flags. > > > > Then we don't need additional APIs to help interpret those > > > capabilities. > > > > > > I understood your intention with different flags, yes it should work = too > > > I think. > > > The reason I am not very fond of it - it will require to double > > > TX_OFFLOAD flags. > > > > An additional feature flag is only required if a NIC is not conforming = to the "minimum requirements" of an offload feature, and the > > techboard permits introducing a variant of an existing feature. > > There should be very few additional feature flags for variants - except= ions only - or the "minimum requirements" are not broad > > enough to support the majority of NICs. >=20 > Ok, so you suggest to group all existing reqs plus what all current tx_pr= epare() do into "minimum requirements"? > So with current drivers in place we wouldn't need these new flags, but we= 'll reserve such opportunity. > That might work, if there are no contradictory requirements in current PM= Ds, and PMDs maintainers with > less reqs will agree with these 'extra' stuff. Just to check how easy/hard would be to get a consensus, compiled a list of= mbuf changes done by different PMDs in tx_prepare(). See below. Could be not fully correct or complete. PMD maintainers, feel free to updat= e it, if I missed something. >From how it looks to me: if we'll go the way you suggest, then hns3 and virtio will most likely beco= me a 'second class citizens' - will need a special offload flags for them. Plus, either all PMDs that now set tx_prepare()=3DNULL will have to agree t= o require rte_net_intel_cksum_prepare() to be done, or all Intel PMDs and few others = will also be downgraded to 'second class'. PMD: atlantic MOD: rte_net_intel_cksum_prepare() /*for ipv4_hdr->hdr_checksum =3D 0; (tcp|udp)_hdr->cksum=3Drte_ipv(4|6)_ph= dr_cksum(...);*/ PMD: cpfl/idpf MOD: none PMD: em/igb/igc/fm10k/i40e/iavf/ice/ixgbe MOD: rte_net_intel_cksum_prepare() PMD: enic MOD: rte_net_intel_cksum_prepare() PMD: hns3 MOD: rte_net_intel_cksum_prepare() plus some extra: /* * A UDP packet with the same dst_port as VXLAN\VXLAN_GPE\GENEVE wi= ll * be recognized as a tunnel packet in HW. In this case, if UDP CKS= UM * offload is set and the tunnel mask has not been set, the CKSUM w= ill * be wrong since the header length is wrong and driver should comp= lete * the CKSUM to avoid CKSUM error. */ PMD: ionic MOD: none PMD: ngbe MOD: rte_net_intel_cksum_prepare() PMD: qede MOD: none PMD: txgbe MOD: rte_net_intel_cksum_prepare() PMD: virtio: MOD: rte_net_intel_cksum_prepare() plus some extra: - for RTE_MBUF_F_TX_TCP_SEG: virtio_tso_fix_cksum() - for RTE_MBUF_F_TX_VLAN: rte_vlan_insert() PMD: vmxnet3 MOD: rte_net_intel_cksum_prepare() For all other PMDs in our main tree set tx_prepare =3D NULL.