1Intel Omni-Path (OPA) Virtual Network Interface Controller (VNIC) feature 2supports Ethernet functionality over Omni-Path fabric by encapsulating 3the Ethernet packets between HFI nodes. 4 5Architecture 6============= 7The patterns of exchanges of Omni-Path encapsulated Ethernet packets 8involves one or more virtual Ethernet switches overlaid on the Omni-Path 9fabric topology. A subset of HFI nodes on the Omni-Path fabric are 10permitted to exchange encapsulated Ethernet packets across a particular 11virtual Ethernet switch. The virtual Ethernet switches are logical 12abstractions achieved by configuring the HFI nodes on the fabric for 13header generation and processing. In the simplest configuration all HFI 14nodes across the fabric exchange encapsulated Ethernet packets over a 15single virtual Ethernet switch. A virtual Ethernet switch, is effectively 16an independent Ethernet network. The configuration is performed by an 17Ethernet Manager (EM) which is part of the trusted Fabric Manager (FM) 18application. HFI nodes can have multiple VNICs each connected to a 19different virtual Ethernet switch. The below diagram presents a case 20of two virtual Ethernet switches with two HFI nodes. 21 22 +-------------------+ 23 | Subnet/ | 24 | Ethernet | 25 | Manager | 26 +-------------------+ 27 / / 28 / / 29 / / 30 / / 31+-----------------------------+ +------------------------------+ 32| Virtual Ethernet Switch | | Virtual Ethernet Switch | 33| +---------+ +---------+ | | +---------+ +---------+ | 34| | VPORT | | VPORT | | | | VPORT | | VPORT | | 35+--+---------+----+---------+-+ +-+---------+----+---------+---+ 36 | \ / | 37 | \ / | 38 | \/ | 39 | / \ | 40 | / \ | 41 +-----------+------------+ +-----------+------------+ 42 | VNIC | VNIC | | VNIC | VNIC | 43 +-----------+------------+ +-----------+------------+ 44 | HFI | | HFI | 45 +------------------------+ +------------------------+ 46 47 48The Omni-Path encapsulated Ethernet packet format is as described below. 49 50Bits Field 51------------------------------------ 52Quad Word 0: 530-19 SLID (lower 20 bits) 5420-30 Length (in Quad Words) 5531 BECN bit 5632-51 DLID (lower 20 bits) 5752-56 SC (Service Class) 5857-59 RC (Routing Control) 5960 FECN bit 6061-62 L2 (=10, 16B format) 6163 LT (=1, Link Transfer Head Flit) 62 63Quad Word 1: 640-7 L4 type (=0x78 ETHERNET) 658-11 SLID[23:20] 6612-15 DLID[23:20] 6716-31 PKEY 6832-47 Entropy 6948-63 Reserved 70 71Quad Word 2: 720-15 Reserved 7316-31 L4 header 7432-63 Ethernet Packet 75 76Quad Words 3 to N-1: 770-63 Ethernet packet (pad extended) 78 79Quad Word N (last): 800-23 Ethernet packet (pad extended) 8124-55 ICRC 8256-61 Tail 8362-63 LT (=01, Link Transfer Tail Flit) 84 85Ethernet packet is padded on the transmit side to ensure that the VNIC OPA 86packet is quad word aligned. The 'Tail' field contains the number of bytes 87padded. On the receive side the 'Tail' field is read and the padding is 88removed (along with ICRC, Tail and OPA header) before passing packet up 89the network stack. 90 91The L4 header field contains the virtual Ethernet switch id the VNIC port 92belongs to. On the receive side, this field is used to de-multiplex the 93received VNIC packets to different VNIC ports. 94 95Driver Design 96============== 97Intel OPA VNIC software design is presented in the below diagram. 98OPA VNIC functionality has a HW dependent component and a HW 99independent component. 100 101The support has been added for IB device to allocate and free the RDMA 102netdev devices. The RDMA netdev supports interfacing with the network 103stack thus creating standard network interfaces. OPA_VNIC is an RDMA 104netdev device type. 105 106The HW dependent VNIC functionality is part of the HFI1 driver. It 107implements the verbs to allocate and free the OPA_VNIC RDMA netdev. 108It involves HW resource allocation/management for VNIC functionality. 109It interfaces with the network stack and implements the required 110net_device_ops functions. It expects Omni-Path encapsulated Ethernet 111packets in the transmit path and provides HW access to them. It strips 112the Omni-Path header from the received packets before passing them up 113the network stack. It also implements the RDMA netdev control operations. 114 115The OPA VNIC module implements the HW independent VNIC functionality. 116It consists of two parts. The VNIC Ethernet Management Agent (VEMA) 117registers itself with IB core as an IB client and interfaces with the 118IB MAD stack. It exchanges the management information with the Ethernet 119Manager (EM) and the VNIC netdev. The VNIC netdev part allocates and frees 120the OPA_VNIC RDMA netdev devices. It overrides the net_device_ops functions 121set by HW dependent VNIC driver where required to accommodate any control 122operation. It also handles the encapsulation of Ethernet packets with an 123Omni-Path header in the transmit path. For each VNIC interface, the 124information required for encapsulation is configured by the EM via VEMA MAD 125interface. It also passes any control information to the HW dependent driver 126by invoking the RDMA netdev control operations. 127 128 +-------------------+ +----------------------+ 129 | | | Linux | 130 | IB MAD | | Network | 131 | | | Stack | 132 +-------------------+ +----------------------+ 133 | | | 134 | | | 135 +----------------------------+ | 136 | | | 137 | OPA VNIC Module | | 138 | (OPA VNIC RDMA Netdev | | 139 | & EMA functions) | | 140 | | | 141 +----------------------------+ | 142 | | 143 | | 144 +------------------+ | 145 | IB core | | 146 +------------------+ | 147 | | 148 | | 149 +--------------------------------------------+ 150 | | 151 | HFI1 Driver with VNIC support | 152 | | 153 +--------------------------------------------+ 154