[eoib] Add Ethernet over Infiniband (EoIB) driver

EoIB is a fairly simple protocol in which raw Ethernet frames (excluding the CRC) are encapsulated within Infiniband Unreliable Datagrams, with a four-byte fixed EoIB header (which conveys no actual information). The Ethernet broadcast domain is provided by a multicast group, similar to the IPoIB IPv4 multicast group. The mapping from Ethernet MAC addresses to Infiniband address vectors is achieved by snooping incoming traffic and building a peer cache which can then be used to map a MAC address into a port GID. The address vector is completed using a path record lookup, as for IPoIB. Note that this requires every packet to include a GRH. Add basic support for EoIB devices. This driver is substantially derived from the IPoIB driver. There is currently no mechanism for automatically creating EoIB devices. Signed-off-by: Michael Brown <mcb30@ipxe.org>
2016-03-09 00:26:56 +00:00 · 2016-03-09 00:26:56 +00:00 · 9154d7a65c
parent 5bcaa1e4d4
commit 9154d7a65c
3 changed files with 815 additions and 0 deletions
--- a/src/drivers/net/eoib.c
+++ b/src/drivers/net/eoib.c
@ -0,0 +1,754 @@
+/*
+ * Copyright (C) 2016 Michael Brown <mbrown@fensystems.co.uk>.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of the
+ * License, or any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
+ * 02110-1301, USA.
+ *
+ * You can also choose to distribute this program under the terms of
+ * the Unmodified Binary Distribution Licence (as given in the file
+ * COPYING.UBDL), provided that you have satisfied its requirements.
+ */
+
+FILE_LICENCE ( GPL2_OR_LATER_OR_UBDL );
+
+#include <stdio.h>
+#include <string.h>
+#include <errno.h>
+#include <ipxe/errortab.h>
+#include <ipxe/malloc.h>
+#include <ipxe/iobuf.h>
+#include <ipxe/if_ether.h>
+#include <ipxe/netdevice.h>
+#include <ipxe/ethernet.h>
+#include <ipxe/infiniband.h>
+#include <ipxe/ib_mcast.h>
+#include <ipxe/ib_pathrec.h>
+#include <ipxe/eoib.h>
+
+/** @file
+ *
+ * Ethernet over Infiniband
+ *
+ */
+
+/** Number of EoIB send work queue entries */
+#define EOIB_NUM_SEND_WQES 8
+
+/** Number of EoIB receive work queue entries */
+#define EOIB_NUM_RECV_WQES 4
+
+/** Number of EoIB completion queue entries */
+#define EOIB_NUM_CQES 16
+
+/** Link status for "broadcast join in progress" */
+#define EINPROGRESS_JOINING __einfo_error ( EINFO_EINPROGRESS_JOINING )
+#define EINFO_EINPROGRESS_JOINING __einfo_uniqify \
+	( EINFO_EINPROGRESS, 0x01, "Joining" )
+
+/** Human-readable message for the link status */
+struct errortab eoib_errors[] __errortab = {
+	__einfo_errortab ( EINFO_EINPROGRESS_JOINING ),
+};
+
+/** List of EoIB devices */
+static LIST_HEAD ( eoib_devices );
+
+static struct net_device_operations eoib_operations;
+
+/****************************************************************************
+ *
+ * EoIB peer cache
+ *
+ ****************************************************************************
+ */
+
+/** An EoIB peer cache entry */
+struct eoib_peer {
+	/** List of EoIB peer cache entries */
+	struct list_head list;
+	/** Ethernet MAC */
+	uint8_t mac[ETH_ALEN];
+	/** Infiniband address vector */
+	struct ib_address_vector av;
+};
+
+/**
+ * Find EoIB peer cache entry
+ *
+ * @v eoib		EoIB device
+ * @v mac		Ethernet MAC
+ * @ret peer		EoIB peer, or NULL if not found
+ */
+static struct eoib_peer * eoib_find_peer ( struct eoib_device *eoib,
+					   const uint8_t *mac ) {
+	struct eoib_peer *peer;
+
+	/* Find peer cache entry */
+	list_for_each_entry ( peer, &eoib->peers, list ) {
+		if ( memcmp ( mac, peer->mac, sizeof ( peer->mac ) ) == 0 ) {
+			/* Move peer to start of list */
+			list_del ( &peer->list );
+			list_add ( &peer->list, &eoib->peers );
+			return peer;
+		}
+	}
+
+	return NULL;
+}
+
+/**
+ * Create EoIB peer cache entry
+ *
+ * @v eoib		EoIB device
+ * @v mac		Ethernet MAC
+ * @ret peer		EoIB peer, or NULL on error
+ */
+static struct eoib_peer * eoib_create_peer ( struct eoib_device *eoib,
+					     const uint8_t *mac ) {
+	struct eoib_peer *peer;
+
+	/* Allocate and initialise peer cache entry */
+	peer = zalloc ( sizeof ( *peer ) );
+	if ( peer ) {
+		memcpy ( peer->mac, mac, sizeof ( peer->mac ) );
+		list_add ( &peer->list, &eoib->peers );
+	}
+	return peer;
+}
+
+/**
+ * Flush EoIB peer cache
+ *
+ * @v eoib		EoIB device
+ */
+static void eoib_flush_peers ( struct eoib_device *eoib ) {
+	struct eoib_peer *peer;
+	struct eoib_peer *tmp;
+
+	list_for_each_entry_safe ( peer, tmp, &eoib->peers, list ) {
+		list_del ( &peer->list );
+		free ( peer );
+	}
+}
+
+/**
+ * Discard some entries from the peer cache
+ *
+ * @ret discarded	Number of cached items discarded
+ */
+static unsigned int eoib_discard ( void ) {
+	struct net_device *netdev;
+	struct eoib_device *eoib;
+	struct eoib_peer *peer;
+	unsigned int discarded = 0;
+
+	/* Try to discard one cache entry for each EoIB device */
+	for_each_netdev ( netdev ) {
+
+		/* Skip non-EoIB devices */
+		if ( netdev->op != &eoib_operations )
+			continue;
+		eoib = netdev->priv;
+
+		/* Discard least recently used cache entry (if any) */
+		list_for_each_entry_reverse ( peer, &eoib->peers, list ) {
+			list_del ( &peer->list );
+			free ( peer );
+			discarded++;
+			break;
+		}
+	}
+
+	return discarded;
+}
+
+/** EoIB cache discarder */
+struct cache_discarder eoib_discarder __cache_discarder ( CACHE_EXPENSIVE ) = {
+	.discard = eoib_discard,
+};
+
+/**
+ * Find destination address vector
+ *
+ * @v eoib		EoIB device
+ * @v mac		Ethernet MAC
+ * @ret av		Address vector, or NULL to send as broadcast
+ */
+static struct ib_address_vector * eoib_tx_av ( struct eoib_device *eoib,
+					       const uint8_t *mac ) {
+	struct ib_device *ibdev = eoib->ibdev;
+	struct eoib_peer *peer;
+	int rc;
+
+	/* If this is a broadcast or multicast MAC address, then send
+	 * this packet as a broadcast.
+	 */
+	if ( is_multicast_ether_addr ( mac ) ) {
+		DBGCP ( eoib, "EoIB %s %s TX multicast\n",
+			eoib->name, eth_ntoa ( mac ) );
+		return NULL;
+	}
+
+	/* If we have no peer cache entry, then create one and send
+	 * this packet as a broadcast.
+	 */
+	peer = eoib_find_peer ( eoib, mac );
+	if ( ! peer ) {
+		DBGC ( eoib, "EoIB %s %s TX unknown\n",
+		       eoib->name, eth_ntoa ( mac ) );
+		eoib_create_peer ( eoib, mac );
+		return NULL;
+	}
+
+	/* If we have not yet recorded a received GID and QPN for this
+	 * peer cache entry, then send this packet as a broadcast.
+	 */
+	if ( ! peer->av.gid_present ) {
+		DBGCP ( eoib, "EoIB %s %s TX not yet recorded\n",
+			eoib->name, eth_ntoa ( mac ) );
+		return NULL;
+	}
+
+	/* If we have not yet resolved a path to this peer, then send
+	 * this packet as a broadcast.
+	 */
+	if ( ( rc = ib_resolve_path ( ibdev, &peer->av ) ) != 0 ) {
+		DBGCP ( eoib, "EoIB %s %s TX not yet resolved\n",
+			eoib->name, eth_ntoa ( mac ) );
+		return NULL;
+	}
+
+	/* Force use of GRH even for local destinations */
+	peer->av.gid_present = 1;
+
+	/* We have a fully resolved peer: send this packet as a
+	 * unicast.
+	 */
+	DBGCP ( eoib, "EoIB %s %s TX " IB_GID_FMT " QPN %#lx\n", eoib->name,
+		eth_ntoa ( mac ), IB_GID_ARGS ( &peer->av.gid ), peer->av.qpn );
+	return &peer->av;
+}
+
+/**
+ * Record source address vector
+ *
+ * @v eoib		EoIB device
+ * @v mac		Ethernet MAC
+ * @v lid		Infiniband LID
+ */
+static void eoib_rx_av ( struct eoib_device *eoib, const uint8_t *mac,
+			 const struct ib_address_vector *av ) {
+	const union ib_gid *gid = &av->gid;
+	unsigned long qpn = av->qpn;
+	struct eoib_peer *peer;
+
+	/* Sanity checks */
+	if ( ! av->gid_present ) {
+		DBGC ( eoib, "EoIB %s %s RX with no GID\n",
+		       eoib->name, eth_ntoa ( mac ) );
+		return;
+	}
+
+	/* Find peer cache entry (if any) */
+	peer = eoib_find_peer ( eoib, mac );
+	if ( ! peer ) {
+		DBGCP ( eoib, "EoIB %s %s RX " IB_GID_FMT " (ignored)\n",
+			eoib->name, eth_ntoa ( mac ), IB_GID_ARGS ( gid ) );
+		return;
+	}
+
+	/* Do nothing if peer cache entry is complete and correct */
+	if ( ( peer->av.lid == av->lid ) && ( peer->av.qpn == qpn ) ) {
+		DBGCP ( eoib, "EoIB %s %s RX unchanged\n",
+			eoib->name, eth_ntoa ( mac ) );
+		return;
+	}
+
+	/* Update peer cache entry */
+	peer->av.qpn = qpn;
+	peer->av.qkey = eoib->broadcast.qkey;
+	peer->av.gid_present = 1;
+	memcpy ( &peer->av.gid, gid, sizeof ( peer->av.gid ) );
+	DBGC ( eoib, "EoIB %s %s RX " IB_GID_FMT " QPN %#lx\n", eoib->name,
+	       eth_ntoa ( mac ), IB_GID_ARGS ( &peer->av.gid ), peer->av.qpn );
+}
+
+/****************************************************************************
+ *
+ * EoIB network device
+ *
+ ****************************************************************************
+ */
+
+/**
+ * Transmit packet via EoIB network device
+ *
+ * @v netdev		Network device
+ * @v iobuf		I/O buffer
+ * @ret rc		Return status code
+ */
+static int eoib_transmit ( struct net_device *netdev,
+			   struct io_buffer *iobuf ) {
+	struct eoib_device *eoib = netdev->priv;
+	struct eoib_header *eoib_hdr;
+	struct ethhdr *ethhdr;
+	struct ib_address_vector *av;
+	size_t zlen;
+
+	/* Sanity checks */
+	assert ( iob_len ( iobuf ) >= sizeof ( *ethhdr ) );
+	assert ( iob_headroom ( iobuf ) >= sizeof ( *eoib_hdr ) );
+
+	/* Look up destination address vector */
+	ethhdr = iobuf->data;
+	av = eoib_tx_av ( eoib, ethhdr->h_dest );
+
+	/* Prepend EoIB header */
+	eoib_hdr = iob_push ( iobuf, sizeof ( *eoib_hdr ) );
+	eoib_hdr->magic = htons ( EOIB_MAGIC );
+	eoib_hdr->reserved = 0;
+
+	/* Pad buffer to minimum Ethernet frame size */
+	zlen = ( sizeof ( *eoib_hdr ) + ETH_ZLEN );
+	assert ( zlen <= IOB_ZLEN );
+	if ( iob_len ( iobuf ) < zlen )
+		iob_pad ( iobuf, zlen );
+
+	/* If we have no unicast address then send as a broadcast */
+	if ( ! av )
+		av = &eoib->broadcast;
+
+	/* Post send work queue entry */
+	return ib_post_send ( eoib->ibdev, eoib->qp, av, iobuf );
+}
+
+/**
+ * Handle EoIB send completion
+ *
+ * @v ibdev		Infiniband device
+ * @v qp		Queue pair
+ * @v iobuf		I/O buffer
+ * @v rc		Completion status code
+ */
+static void eoib_complete_send ( struct ib_device *ibdev __unused,
+				 struct ib_queue_pair *qp,
+				 struct io_buffer *iobuf, int rc ) {
+	struct eoib_device *eoib = ib_qp_get_ownerdata ( qp );
+
+	netdev_tx_complete_err ( eoib->netdev, iobuf, rc );
+}
+
+/**
+ * Handle EoIB receive completion
+ *
+ * @v ibdev		Infiniband device
+ * @v qp		Queue pair
+ * @v dest		Destination address vector, or NULL
+ * @v source		Source address vector, or NULL
+ * @v iobuf		I/O buffer
+ * @v rc		Completion status code
+ */
+static void eoib_complete_recv ( struct ib_device *ibdev __unused,
+				 struct ib_queue_pair *qp,
+				 struct ib_address_vector *dest __unused,
+				 struct ib_address_vector *source,
+				 struct io_buffer *iobuf, int rc ) {
+	struct eoib_device *eoib = ib_qp_get_ownerdata ( qp );
+	struct net_device *netdev = eoib->netdev;
+	struct eoib_header *eoib_hdr;
+	struct ethhdr *ethhdr;
+
+	/* Record errors */
+	if ( rc != 0 ) {
+		netdev_rx_err ( netdev, iobuf, rc );
+		return;
+	}
+
+	/* Sanity check */
+	if ( iob_len ( iobuf ) < ( sizeof ( *eoib_hdr ) + sizeof ( *ethhdr ) )){
+		DBGC ( eoib, "EoIB %s received packet too short to "
+		       "contain EoIB and Ethernet headers\n", eoib->name );
+		DBGC_HD ( eoib, iobuf->data, iob_len ( iobuf ) );
+		netdev_rx_err ( netdev, iobuf, -EIO );
+		return;
+	}
+	if ( ! source ) {
+		DBGC ( eoib, "EoIB %s received packet without address "
+		       "vector\n", eoib->name );
+		netdev_rx_err ( netdev, iobuf, -ENOTTY );
+		return;
+	}
+
+	/* Strip EoIB header */
+	iob_pull ( iobuf, sizeof ( *eoib_hdr ) );
+
+	/* Update neighbour cache entry, if any */
+	ethhdr = iobuf->data;
+	eoib_rx_av ( eoib, ethhdr->h_source, source );
+
+	/* Hand off to network layer */
+	netdev_rx ( netdev, iobuf );
+}
+
+/** EoIB completion operations */
+static struct ib_completion_queue_operations eoib_cq_op = {
+	.complete_send = eoib_complete_send,
+	.complete_recv = eoib_complete_recv,
+};
+
+/** EoIB queue pair operations */
+static struct ib_queue_pair_operations eoib_qp_op = {
+	.alloc_iob = alloc_iob,
+};
+
+/**
+ * Poll EoIB network device
+ *
+ * @v netdev		Network device
+ */
+static void eoib_poll ( struct net_device *netdev ) {
+	struct eoib_device *eoib = netdev->priv;
+	struct ib_device *ibdev = eoib->ibdev;
+
+	/* Poll Infiniband device */
+	ib_poll_eq ( ibdev );
+
+	/* Poll the retry timers (required for EoIB multicast join) */
+	retry_poll();
+}
+
+/**
+ * Handle EoIB broadcast multicast group join completion
+ *
+ * @v membership	Multicast group membership
+ * @v rc		Status code
+ */
+static void eoib_join_complete ( struct ib_mc_membership *membership, int rc ) {
+	struct eoib_device *eoib =
+		container_of ( membership, struct eoib_device, membership );
+
+	/* Record join status as link status */
+	netdev_link_err ( eoib->netdev, rc );
+}
+
+/**
+ * Join EoIB broadcast multicast group
+ *
+ * @v eoib		EoIB device
+ * @ret rc		Return status code
+ */
+static int eoib_join_broadcast_group ( struct eoib_device *eoib ) {
+	int rc;
+
+	/* Join multicast group */
+	if ( ( rc = ib_mcast_join ( eoib->ibdev, eoib->qp,
+				    &eoib->membership, &eoib->broadcast, 0,
+				    eoib_join_complete ) ) != 0 ) {
+		DBGC ( eoib, "EoIB %s could not join broadcast group: %s\n",
+		       eoib->name, strerror ( rc ) );
+		return rc;
+	}
+
+	return 0;
+}
+
+/**
+ * Leave EoIB broadcast multicast group
+ *
+ * @v eoib		EoIB device
+ */
+static void eoib_leave_broadcast_group ( struct eoib_device *eoib ) {
+
+	/* Leave multicast group */
+	ib_mcast_leave ( eoib->ibdev, eoib->qp, &eoib->membership );
+}
+
+/**
+ * Handle link status change
+ *
+ * @v eoib		EoIB device
+ */
+static void eoib_link_state_changed ( struct eoib_device *eoib ) {
+	struct net_device *netdev = eoib->netdev;
+	struct ib_device *ibdev = eoib->ibdev;
+	int rc;
+
+	/* Leave existing broadcast group */
+	if ( eoib->qp )
+		eoib_leave_broadcast_group ( eoib );
+
+	/* Update broadcast GID based on potentially-new partition key */
+	eoib->broadcast.gid.words[2] = htons ( ibdev->pkey | IB_PKEY_FULL );
+
+	/* Set net device link state to reflect Infiniband link state */
+	rc = ib_link_rc ( ibdev );
+	netdev_link_err ( netdev, ( rc ? rc : -EINPROGRESS_JOINING ) );
+
+	/* Join new broadcast group */
+	if ( ib_is_open ( ibdev ) && ib_link_ok ( ibdev ) && eoib->qp &&
+	     ( ( rc = eoib_join_broadcast_group ( eoib ) ) != 0 ) ) {
+		DBGC ( eoib, "EoIB %s could not rejoin broadcast group: "
+		       "%s\n", eoib->name, strerror ( rc ) );
+		netdev_link_err ( netdev, rc );
+		return;
+	}
+}
+
+/**
+ * Open EoIB network device
+ *
+ * @v netdev		Network device
+ * @ret rc		Return status code
+ */
+static int eoib_open ( struct net_device *netdev ) {
+	struct eoib_device *eoib = netdev->priv;
+	struct ib_device *ibdev = eoib->ibdev;
+	int rc;
+
+	/* Open IB device */
+	if ( ( rc = ib_open ( ibdev ) ) != 0 ) {
+		DBGC ( eoib, "EoIB %s could not open %s: %s\n",
+		       eoib->name, ibdev->name, strerror ( rc ) );
+		goto err_ib_open;
+	}
+
+	/* Allocate completion queue */
+	eoib->cq = ib_create_cq ( ibdev, EOIB_NUM_CQES, &eoib_cq_op );
+	if ( ! eoib->cq ) {
+		DBGC ( eoib, "EoIB %s could not allocate completion queue\n",
+		       eoib->name );
+		rc = -ENOMEM;
+		goto err_create_cq;
+	}
+
+	/* Allocate queue pair */
+	eoib->qp = ib_create_qp ( ibdev, IB_QPT_UD, EOIB_NUM_SEND_WQES,
+				   eoib->cq, EOIB_NUM_RECV_WQES, eoib->cq,
+				  &eoib_qp_op, netdev->name );
+	if ( ! eoib->qp ) {
+		DBGC ( eoib, "EoIB %s could not allocate queue pair\n",
+		       eoib->name );
+		rc = -ENOMEM;
+		goto err_create_qp;
+	}
+	ib_qp_set_ownerdata ( eoib->qp, eoib );
+
+	/* Fill receive rings */
+	ib_refill_recv ( ibdev, eoib->qp );
+
+	/* Fake a link status change to join the broadcast group */
+	eoib_link_state_changed ( eoib );
+
+	return 0;
+
+	ib_destroy_qp ( ibdev, eoib->qp );
+	eoib->qp = NULL;
+ err_create_qp:
+	ib_destroy_cq ( ibdev, eoib->cq );
+	eoib->cq = NULL;
+ err_create_cq:
+	ib_close ( ibdev );
+ err_ib_open:
+	return rc;
+}
+
+/**
+ * Close EoIB network device
+ *
+ * @v netdev		Network device
+ */
+static void eoib_close ( struct net_device *netdev ) {
+	struct eoib_device *eoib = netdev->priv;
+	struct ib_device *ibdev = eoib->ibdev;
+
+	/* Flush peer cache */
+	eoib_flush_peers ( eoib );
+
+	/* Leave broadcast group */
+	eoib_leave_broadcast_group ( eoib );
+
+	/* Tear down the queues */
+	ib_destroy_qp ( ibdev, eoib->qp );
+	eoib->qp = NULL;
+	ib_destroy_cq ( ibdev, eoib->cq );
+	eoib->cq = NULL;
+
+	/* Close IB device */
+	ib_close ( ibdev );
+}
+
+/** EoIB network device operations */
+static struct net_device_operations eoib_operations = {
+	.open		= eoib_open,
+	.close		= eoib_close,
+	.transmit	= eoib_transmit,
+	.poll		= eoib_poll,
+};
+
+/**
+ * Create EoIB device
+ *
+ * @v ibdev		Infiniband device
+ * @v hw_addr		Ethernet MAC
+ * @v broadcast		Broadcast address vector
+ * @v name		Interface name (or NULL to use default)
+ * @ret rc		Return status code
+ */
+int eoib_create ( struct ib_device *ibdev, const uint8_t *hw_addr,
+		  struct ib_address_vector *broadcast, const char *name ) {
+	struct net_device *netdev;
+	struct eoib_device *eoib;
+	int rc;
+
+	/* Allocate network device */
+	netdev = alloc_etherdev ( sizeof ( *eoib ) );
+	if ( ! netdev ) {
+		rc = -ENOMEM;
+		goto err_alloc;
+	}
+	netdev_init ( netdev, &eoib_operations );
+	eoib = netdev->priv;
+	netdev->dev = ibdev->dev;
+	eoib->netdev = netdev;
+	eoib->ibdev = ibdev_get ( ibdev );
+	memcpy ( &eoib->broadcast, broadcast, sizeof ( eoib->broadcast ) );
+	INIT_LIST_HEAD ( &eoib->peers );
+
+	/* Set MAC address */
+	memcpy ( netdev->hw_addr, hw_addr, ETH_ALEN );
+
+	/* Set interface name, if applicable */
+	if ( name )
+		snprintf ( netdev->name, sizeof ( netdev->name ), "%s", name );
+	eoib->name = netdev->name;
+
+	/* Add to list of EoIB devices */
+	list_add_tail ( &eoib->list, &eoib_devices );
+
+	/* Register network device */
+	if ( ( rc = register_netdev ( netdev ) ) != 0 )
+		goto err_register;
+
+	DBGC ( eoib, "EoIB %s created for %s MAC %s\n",
+	       eoib->name, ibdev->name, eth_ntoa ( hw_addr ) );
+	DBGC ( eoib, "EoIB %s broadcast GID " IB_GID_FMT "\n",
+	       eoib->name, IB_GID_ARGS ( &broadcast->gid ) );
+	return 0;
+
+	unregister_netdev ( netdev );
+ err_register:
+	list_del ( &eoib->list );
+	ibdev_put ( ibdev );
+	netdev_nullify ( netdev );
+	netdev_put ( netdev );
+ err_alloc:
+	return rc;
+}
+
+/**
+ * Find EoIB device
+ *
+ * @v ibdev		Infiniband device
+ * @v hw_addr		Original Ethernet MAC
+ * @ret eoib		EoIB device
+ */
+struct eoib_device * eoib_find ( struct ib_device *ibdev,
+				 const uint8_t *hw_addr ) {
+	struct eoib_device *eoib;
+
+	list_for_each_entry ( eoib, &eoib_devices, list ) {
+		if ( ( eoib->ibdev == ibdev ) &&
+		     ( memcmp ( eoib->netdev->hw_addr, hw_addr,
+				ETH_ALEN ) == 0 ) )
+			return eoib;
+	}
+	return NULL;
+}
+
+/**
+ * Remove EoIB device
+ *
+ * @v eoib		EoIB device
+ */
+void eoib_destroy ( struct eoib_device *eoib ) {
+	struct net_device *netdev = eoib->netdev;
+
+	/* Unregister network device */
+	unregister_netdev ( netdev );
+
+	/* Remove from list of network devices */
+	list_del ( &eoib->list );
+
+	/* Drop reference to Infiniband device */
+	ibdev_put ( eoib->ibdev );
+
+	/* Free network device */
+	DBGC ( eoib, "EoIB %s destroyed\n", eoib->name );
+	netdev_nullify ( netdev );
+	netdev_put ( netdev );
+}
+
+/**
+ * Probe EoIB device
+ *
+ * @v ibdev		Infiniband device
+ * @ret rc		Return status code
+ */
+static int eoib_probe ( struct ib_device *ibdev __unused ) {
+
+	/* EoIB devices are not created automatically */
+	return 0;
+}
+
+/**
+ * Handle device or link status change
+ *
+ * @v ibdev		Infiniband device
+ */
+static void eoib_notify ( struct ib_device *ibdev ) {
+	struct eoib_device *eoib;
+
+	/* Handle link status change for any attached EoIB devices */
+	list_for_each_entry ( eoib, &eoib_devices, list ) {
+		if ( eoib->ibdev != ibdev )
+			continue;
+		eoib_link_state_changed ( eoib );
+	}
+}
+
+/**
+ * Remove EoIB device
+ *
+ * @v ibdev		Infiniband device
+ */
+static void eoib_remove ( struct ib_device *ibdev ) {
+	struct eoib_device *eoib;
+	struct eoib_device *tmp;
+
+	/* Remove any attached EoIB devices */
+	list_for_each_entry_safe ( eoib, tmp, &eoib_devices, list ) {
+		if ( eoib->ibdev != ibdev )
+			continue;
+		eoib_destroy ( eoib );
+	}
+}
+
+/** EoIB driver */
+struct ib_driver eoib_driver __ib_driver = {
+	.name = "EoIB",
+	.probe = eoib_probe,
+	.notify = eoib_notify,
+	.remove = eoib_remove,
+};
--- a/src/include/ipxe/eoib.h
+++ b/src/include/ipxe/eoib.h
@ -0,0 +1,60 @@
+#ifndef _IPXE_EOIB_H
+#define _IPXE_EOIB_H
+
+/** @file
+ *
+ * Ethernet over Infiniband
+ *
+ */
+
+FILE_LICENCE ( GPL2_OR_LATER_OR_UBDL );
+
+#include <stdint.h>
+#include <byteswap.h>
+#include <ipxe/netdevice.h>
+#include <ipxe/infiniband.h>
+#include <ipxe/ib_mcast.h>
+
+/** An EoIB header */
+struct eoib_header {
+	/** Signature */
+	uint16_t magic;
+	/** Reserved */
+	uint16_t reserved;
+} __attribute__ (( packed ));
+
+/** EoIB magic signature */
+#define EOIB_MAGIC 0x8919
+
+/** An EoIB device */
+struct eoib_device {
+	/** Name */
+	const char *name;
+	/** Network device */
+	struct net_device *netdev;
+	/** Underlying Infiniband device */
+	struct ib_device *ibdev;
+	/** List of EoIB devices */
+	struct list_head list;
+	/** Broadcast address */
+	struct ib_address_vector broadcast;
+
+	/** Completion queue */
+	struct ib_completion_queue *cq;
+	/** Queue pair */
+	struct ib_queue_pair *qp;
+	/** Broadcast group membership */
+	struct ib_mc_membership membership;
+
+	/** Peer cache */
+	struct list_head peers;
+};
+
+extern int eoib_create ( struct ib_device *ibdev, const uint8_t *hw_addr,
+			 struct ib_address_vector *broadcast,
+			 const char *name );
+extern struct eoib_device * eoib_find ( struct ib_device *ibdev,
+					const uint8_t *hw_addr );
+extern void eoib_destroy ( struct eoib_device *eoib );
+
+#endif /* _IPXE_EOIB_H */
--- a/src/include/ipxe/errfile.h
+++ b/src/include/ipxe/errfile.h
@ -185,6 +185,7 @@ FILE_LICENCE ( GPL2_OR_LATER_OR_UBDL );
 #define ERRFILE_intelxvf	     ( ERRFILE_DRIVER | 0x00790000 )
 #define ERRFILE_smsc95xx	     ( ERRFILE_DRIVER | 0x007a0000 )
 #define ERRFILE_acm		     ( ERRFILE_DRIVER | 0x007b0000 )
+#define ERRFILE_eoib		     ( ERRFILE_DRIVER | 0x007c0000 )

 #define ERRFILE_aoe			( ERRFILE_NET | 0x00000000 )
 #define ERRFILE_arp			( ERRFILE_NET | 0x00010000 )