39. Memif Poll Mode Driver

Shared memory packet interface (memif) PMD allows for DPDK and any other client using memif (DPDK, VPP, libmemif) to communicate using shared memory. Memif is Linux only.

The created device transmits packets in a raw format. It can be used with Ethernet mode, IP mode, or Punt/Inject. At this moment, only Ethernet mode is supported in DPDK memif implementation.

Memif works in two roles: server and client. Client connects to server over an existing socket. It is also a producer of shared memory file and initializes the shared memory. Each interface can be connected to one peer interface at same time. The peer interface is identified by id parameter. Server creates the socket and listens for any client connection requests. The socket may already exist on the system. Be sure to remove any such sockets, if you are creating a server interface, or you will see an “Address already in use” error. Function rte_pmd_memif_remove(), which removes memif interface, will also remove a listener socket, if it is not being used by any other interface.

The method to enable one or more interfaces is to use the --vdev=net_memif0 option on the DPDK application command line. Each --vdev=net_memif1 option given will create an interface named net_memif0, net_memif1, and so on. Memif uses unix domain socket to transmit control messages. Each memif has a unique id per socket. This id is used to identify peer interface. If you are connecting multiple interfaces using same socket, be sure to specify unique ids id=0, id=1, etc. Note that if you assign a socket to a server interface it becomes a listener socket. Listener socket can not be used by a client interface on same client.

Table 39.1 Memif configuration options
Option Description Default Valid value
id=0 Used to identify peer interface 0 uint32_t
role=server Set memif role client server|client
bsize=1024 Size of single packet buffer 2048 uint16_t
rsize=11 Log2 of ring size. If rsize is 10, actual ring size is 1024 10 1-14
socket=/tmp/memif.sock Socket filename /tmp/memif.sock string len 108
socket-abstract=no Set usage of abstract socket address yes yes|no
owner-uid=1000 Set socket listener owner uid. Only relevant to server with socket-abstract=no unchanged uid_t
owner-gid=1000 Set socket listener owner gid. Only relevant to server with socket-abstract=no unchanged gid_t
mac=01:23:45:ab:cd:ef Mac address 01:ab:23:cd:45:ef  
secret=abc123 Secret is an optional security option, which if specified, must be matched by peer   string len 24
zero-copy=yes Enable/disable zero-copy client mode. Only relevant to client, requires ‘–single-file-segments’ eal argument no yes|no

Connection establishment

In order to create memif connection, two memif interfaces, each in separate process, are needed. One interface in server role and other in client role. It is not possible to connect two interfaces in a single process. Each interface can be connected to one interface at same time, identified by matching id parameter.

Memif driver uses unix domain socket to exchange required information between memif interfaces. Socket file path is specified at interface creation see Memif configuration options table above. If socket is used by server interface, it’s marked as listener socket (in scope of current process) and listens to connection requests from other processes. One socket can be used by multiple interfaces. One process can have client and server interfaces at the same time, provided each role is assigned unique socket.

For detailed information on memif control messages, see: net/memif/memif.h.

Client interface attempts to make a connection on assigned socket. Process listening on this socket will extract the connection request and create a new connected socket (control channel). Then it sends the ‘hello’ message (MEMIF_MSG_TYPE_HELLO), containing configuration boundaries. Client interface adjusts its configuration accordingly, and sends ‘init’ message (MEMIF_MSG_TYPE_INIT). This message among others contains interface id. Driver uses this id to find server interface, and assigns the control channel to this interface. If such interface is found, ‘ack’ message (MEMIF_MSG_TYPE_ACK) is sent. Client interface sends ‘add region’ message (MEMIF_MSG_TYPE_ADD_REGION) for every region allocated. Server responds to each of these messages with ‘ack’ message. Same behavior applies to rings. Client sends ‘add ring’ message (MEMIF_MSG_TYPE_ADD_RING) for every initialized ring. Server again responds to each message with ‘ack’ message. To finalize the connection, client interface sends ‘connect’ message (MEMIF_MSG_TYPE_CONNECT). Upon receiving this message server maps regions to its address space, initializes rings and responds with ‘connected’ message (MEMIF_MSG_TYPE_CONNECTED). Disconnect (MEMIF_MSG_TYPE_DISCONNECT) can be sent by both server and client interfaces at any time, due to driver error or if the interface is being deleted.

Files

  • net/memif/memif.h - control messages definitions
  • net/memif/memif_socket.h
  • net/memif/memif_socket.c

39.1. Shared memory

Shared memory format

Client is producer and server is consumer. Memory regions, are mapped shared memory files, created by memif client and provided to server at connection establishment. Regions contain rings and buffers. Rings and buffers can also be separated into multiple regions. For no-zero-copy, rings and buffers are stored inside single memory region to reduce the number of opened files.

region n (no-zero-copy):

Rings Buffers
C2S rings S2C rings packet buffer 0 . pb ((1 << pmd->run.log2_ring_size)*(c2s + s2c))-1

C2S OR S2C Rings:

ring 0 ring 1 ring num_c2s_rings - 1

ring 0:

ring header (1 << pmd->run.log2_ring_size) * desc

Descriptors are assigned packet buffers in order of rings creation. If we have one ring in each direction and ring size is 1024, then first 1024 buffers will belong to C2S ring and last 1024 will belong to S2C ring. In case of zero-copy, buffers are dequeued and enqueued as needed.

Descriptor format

Quad

Word

6                                                             3 3                             1 1                              
3                                                             2 1                             6 5                             0
0 length region flags
1 metadata offset
  6                                                             3 3                                                              
3                                                             2 1                                                             0

Flags field - flags (Quad Word 0, bits 0:15)

Bits Name Functionality
0 MEMIF_DESC_FLAG_NEXT Is chained buffer. When set, the packet is divided into multiple buffers. May not be contiguous.

Region index - region (Quad Word 0, 16:31)

Index of memory region, the buffer is located in.

Data length - length (Quad Word 0, 32:63)

Length of transmitted/received data.

Data Offset - offset (Quad Word 1, 0:31)

Data start offset from memory region address. .regions[desc->region].addr + desc->offset

Metadata - metadata (Quad Word 1, 32:63)

Buffer metadata.

Files

  • net/memif/memif.h - descriptor and ring definitions
  • net/memif/rte_eth_memif.c - eth_memif_rx() eth_memif_tx()

39.2. Zero-copy client

Zero-copy client can be enabled with memif configuration option ‘zero-copy=yes’. This option is only relevant to client and requires eal argument ‘–single-file-segments’. This limitation is in place, because it is too expensive to identify memseg for each packet buffer, resulting in worse performance than with zero-copy disabled. With single file segments we can calculate offset from the beginning of the file for each packet buffer.

Shared memory format

Region 0 is created by memif driver and contains rings. Client interface exposes DPDK memory (memseg). Instead of using memfd_create() to create new shared file, existing memsegs are used. Server interface functions the same as with zero-copy disabled.

region 0:

Rings
C2S rings S2C rings

region n:

Buffers
memseg

Buffers are dequeued and enqueued as needed. Offset descriptor field is calculated at tx. Only single file segments mode (EAL option –single-file-segments) is supported, as calculating offset from multiple segments is too expensive.

39.2.1. Example: testpmd

In this example we run two instances of testpmd application and transmit packets over memif.

First create server interface:

#./<build_dir>/app/dpdk-testpmd -l 0-1 --proc-type=primary --file-prefix=pmd1 --vdev=net_memif,role=server -- -i

Now create client interface (server must be already running so the client will connect):

#./<build_dir>/app/dpdk-testpmd -l 2-3 --proc-type=primary --file-prefix=pmd2 --vdev=net_memif -- -i

You can also enable zero-copy on client interface:

#./<build_dir>/app/dpdk-testpmd -l 2-3 --proc-type=primary --file-prefix=pmd2 --vdev=net_memif,zero-copy=yes --single-file-segments -- -i

Start forwarding packets:

Client:
    testpmd> start

Server:
    testpmd> start tx_first

Show status:

testpmd> show port stats 0

For more details on testpmd please refer to Testpmd Application User Guide.

39.2.2. Example: testpmd and VPP

For information on how to get and run VPP please see https://wiki.fd.io/view/VPP.

Start VPP in interactive mode (should be by default). Create memif server interface in VPP:

vpp# create interface memif id 0 server no-zero-copy
vpp# set interface state memif0/0 up
vpp# set interface ip address memif0/0 192.168.1.1/24

To see socket filename use show memif command:

vpp# show memif
sockets
 id  listener    filename
  0   yes (1)     /run/vpp/memif.sock
...

Now create memif interface by running testpmd with these command line options:

#./dpdk-testpmd --vdev=net_memif,socket=/run/vpp/memif.sock -- -i

Testpmd should now create memif client interface and try to connect to server. In testpmd set forward option to icmpecho and start forwarding:

testpmd> set fwd icmpecho
testpmd> start

Send ping from VPP:

vpp# ping 192.168.1.2
64 bytes from 192.168.1.2: icmp_seq=2 ttl=254 time=36.2918 ms
64 bytes from 192.168.1.2: icmp_seq=3 ttl=254 time=23.3927 ms
64 bytes from 192.168.1.2: icmp_seq=4 ttl=254 time=24.2975 ms
64 bytes from 192.168.1.2: icmp_seq=5 ttl=254 time=17.7049 ms

39.2.3. Example: testpmd memif loopback

In this example we will create 2 memif ports connected into loopback. The situation is analogous to cross connecting 2 ports of the NIC by cable.

To set the loopback, just use the same socket and id with different roles:

#./dpdk-testpmd --vdev=net_memif0,role=server,id=0 --vdev=net_memif1,role=client,id=0 -- -i

Then start the communication:

testpmd> start tx_first

Finally we can check port stats to see the traffic:

testpmd> show port stats all