Based on kernel version 4.3. Page generated on 2015-11-02 12:49 EST.
1 IP OVER INFINIBAND 2 3 The ib_ipoib driver is an implementation of the IP over InfiniBand 4 protocol as specified by RFC 4391 and 4392, issued by the IETF ipoib 5 working group. It is a "native" implementation in the sense of 6 setting the interface type to ARPHRD_INFINIBAND and the hardware 7 address length to 20 (earlier proprietary implementations 8 masqueraded to the kernel as ethernet interfaces). 9 10 Partitions and P_Keys 11 12 When the IPoIB driver is loaded, it creates one interface for each 13 port using the P_Key at index 0. To create an interface with a 14 different P_Key, write the desired P_Key into the main interface's 15 /sys/class/net/<intf name>/create_child file. For example: 16 17 echo 0x8001 > /sys/class/net/ib0/create_child 18 19 This will create an interface named ib0.8001 with P_Key 0x8001. To 20 remove a subinterface, use the "delete_child" file: 21 22 echo 0x8001 > /sys/class/net/ib0/delete_child 23 24 The P_Key for any interface is given by the "pkey" file, and the 25 main interface for a subinterface is in "parent." 26 27 Child interface create/delete can also be done using IPoIB's 28 rtnl_link_ops, where childs created using either way behave the same. 29 30 Datagram vs Connected modes 31 32 The IPoIB driver supports two modes of operation: datagram and 33 connected. The mode is set and read through an interface's 34 /sys/class/net/<intf name>/mode file. 35 36 In datagram mode, the IB UD (Unreliable Datagram) transport is used 37 and so the interface MTU has is equal to the IB L2 MTU minus the 38 IPoIB encapsulation header (4 bytes). For example, in a typical IB 39 fabric with a 2K MTU, the IPoIB MTU will be 2048 - 4 = 2044 bytes. 40 41 In connected mode, the IB RC (Reliable Connected) transport is used. 42 Connected mode takes advantage of the connected nature of the IB 43 transport and allows an MTU up to the maximal IP packet size of 64K, 44 which reduces the number of IP packets needed for handling large UDP 45 datagrams, TCP segments, etc and increases the performance for large 46 messages. 47 48 In connected mode, the interface's UD QP is still used for multicast 49 and communication with peers that don't support connected mode. In 50 this case, RX emulation of ICMP PMTU packets is used to cause the 51 networking stack to use the smaller UD MTU for these neighbours. 52 53 Stateless offloads 54 55 If the IB HW supports IPoIB stateless offloads, IPoIB advertises 56 TCP/IP checksum and/or Large Send (LSO) offloading capability to the 57 network stack. 58 59 Large Receive (LRO) offloading is also implemented and may be turned 60 on/off using ethtool calls. Currently LRO is supported only for 61 checksum offload capable devices. 62 63 Stateless offloads are supported only in datagram mode. 64 65 Interrupt moderation 66 67 If the underlying IB device supports CQ event moderation, one can 68 use ethtool to set interrupt mitigation parameters and thus reduce 69 the overhead incurred by handling interrupts. The main code path of 70 IPoIB doesn't use events for TX completion signaling so only RX 71 moderation is supported. 72 73 Debugging Information 74 75 By compiling the IPoIB driver with CONFIG_INFINIBAND_IPOIB_DEBUG set 76 to 'y', tracing messages are compiled into the driver. They are 77 turned on by setting the module parameters debug_level and 78 mcast_debug_level to 1. These parameters can be controlled at 79 runtime through files in /sys/module/ib_ipoib/. 80 81 CONFIG_INFINIBAND_IPOIB_DEBUG also enables files in the debugfs 82 virtual filesystem. By mounting this filesystem, for example with 83 84 mount -t debugfs none /sys/kernel/debug 85 86 it is possible to get statistics about multicast groups from the 87 files /sys/kernel/debug/ipoib/ib0_mcg and so on. 88 89 The performance impact of this option is negligible, so it 90 is safe to enable this option with debug_level set to 0 for normal 91 operation. 92 93 CONFIG_INFINIBAND_IPOIB_DEBUG_DATA enables even more debug output in 94 the data path when data_debug_level is set to 1. However, even with 95 the output disabled, enabling this configuration option will affect 96 performance, because it adds tests to the fast path. 97 98 References 99 100 Transmission of IP over InfiniBand (IPoIB) (RFC 4391) 101 http://ietf.org/rfc/rfc4391.txt 102 IP over InfiniBand (IPoIB) Architecture (RFC 4392) 103 http://ietf.org/rfc/rfc4392.txt 104 IP over InfiniBand: Connected Mode (RFC 4755) 105 http://ietf.org/rfc/rfc4755.txt