Multiplex Proxy Ethernet Devices

The xeth Linux module provides the xeth-mux network device driver that
uses encapsulation to multiplex its virtual devices that proxy for ports
and functions of an on-board or remote switch chip.

Modes

The mux may operate in these modes: switch-less, unlinked-switch,
linked-switch, and switch-driver.

Figure 1: switch-less mode

The daemon models a chip in development or provides a network simulation for
protocol development. Each frame is encapsulated and forwarded through a raw
socket connected to the mux (bold line.) The daemon and mux are also connected
by a SEQPACKET socket for side band protocol messages
(dashed line.)

Figure 2: unlinked-switch mode

The daemon has vfio-pci control (dotted line) of the switch chip.

Figure 3: linked-switch mode

The daemon has vfio-pci control of the switch chip which forwards
encapsulated frames through the mux lower device links.

Figure 4: switch-driver mode

The switch kernel device driver controls the chip and forwards encapsulated
frames through the mux like the above links.

Exceptions

In switch-less and unlinked-switch modes, all received frames are encapsulated
and marked before forwarding through a raw socket to the mux. In linked-switch
mode the switch chip may be unable to encapsulate and forward some frames like
ARP requests, TTL=1; so the daemon also forwards these through a raw socket to
the mux.

The mux transmit vector distinguishes the marked encapsulation to demux and
receive through the associated proxy device (e.g. vlan priority 7.)

The mux recognizes switch-less mode by it not having any lower links and
forwards all proxy transmit frames with unmarked encapsulation to the
daemon's raw socket through its own receive handler. In any other mode the
mux encapsulates transmit frames before sending through a hash identified
lower link.

Platform Interface

The xeth module includes mux and port platform drivers that when probed
through the matching compatibility entries, instantiate the respective mux
and port network devices with the properties given by the firmware though
device tree and ACPI.

type	property	default
	mux
string	compatible	"xeth,mux"
string	name	"xeth-mux"
u8	encap	0 (vlan)
u16	ports	32
u8[6]	link0-mac-address	none
u8[6]	link1-mac-address	none
u8[6]	link2-mac-address	none
u8[6]	link3-mac-address	none
u8[6]	link4-mac-address	none
u8[6]	link5-mac-address	none
u8[6]	link6-mac-address	none
u8[6]	link7-mac-address	none
string[]	link-akas	none
u8	txqs	1
u8	rxqs	1
string[]	priv-flags	none
string[]	stat-names	none
u8[2]	qsfp-i2c-addrs	{ 0x50, 0x51 }
	port
string	compatible	"xeth,port"
string	name	"xeth%"
string	mux	"xeth-mux"
u8[6]	mac-address	random
u8	qsfp-bus	none
u8	txqs	1
u8	rxqs	1

Admin

In addition to platform instantiated devices, an administrator may create
a mux through rtnl with iproute2.

ip link add [NAME] [link DEV] type xeth-mux [encap ENCAP]

The mode is implied from the link parameter: if given, the mux is in either
linked-switch or switch-driver mode; otherwise it's in switch-less or
unlinked-switch modes.

The admin may also create xeth proxy devices with these iproute2 commands.

ip link add [NAME] link XETH_MUX type xeth-port [xid XID]

ip link add [NAME] link XETH_MUX type xeth-loopback [channel ID]

ip link add [NAME] link XETH_PORT type xeth-lag
ip link set ANOTHER_XETH_PORT master XETH_LAG

ip link add [NAME.VLAN] link XETH_PORT_OR_LAG type xeth-vlan [vid VID]

ip link add [NAME] link XETH_PORT_LAG_OR_VLAN type xeth-bridge
ip link set ANOTHER_XETH_PORT_LAG_OR_VLAN master XETH_BRIDGE

L2 Forwarding Offload

The switch offloads forwarding for all devices by default. Each device may be configured to defer to Linux forwarding with:

ethtool -K XETH_DEV l2-fwd-offload off

side band protocol

On admin-up, the mux opens it's @NAME socket for a side band protocol with a
switch daemon through a SEQPACKET connection. For switch-driver mode, the
protocol messages are designed to be distinguishable from through traffic so
the protocol is actually encoded in-band.

The mux provides the following information to the switch daemon or driver:

• proxy interfaces
• FIB
• neighbor entries

The switch daemon or driver provides this to the xeth-mux:

• port carrier state
• port ethtool stats
• proxy link stats

Either daemon or switch driver initiate this protocol with a DUMP_IFINFO
request.

SWITCH              xeth-mux
   | DUMP_IFINFO ------>|
   |<---------- XIFINFO |
   |        ...         |
   |<------------ BREAK |
   | DUMP_FIBINFO ----->|
   |<--------- XFIBINFO |
   |        ...         |
   |<------------ BREAK |
   | SPEED ------------>|
   | CARRIER ---------->|
   | LINK_STAT -------->|
   | ETHTOOL_STAT ----->|
   |        ...         |

In the above diagram, XIFINFO refers to the extended interface info for each
proxy device: IFINFO, ETHTOOL_FLAGS, IFA, IFA6, and UPPER devices.

The XFIBINFO refers to this extended forwarding info of each network:
FIBENTRY, FIB6ENTRY, NEIGH_UPDATE, NETNS_ADD, NETNS_DEL.

After the BREAK reply to the DUMP_FIBINFO request, the mux continues sending
FIB and interface updates while the daemon or switch driver relays
negotiated port speed, carrier state, and periodic stats.

See dkms/xeth_uapi.h for message definitions.