2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 * $Id: ipoib.h 1358 2004-12-17 22:00:11Z roland $
40 #include <linux/list.h>
41 #include <linux/skbuff.h>
42 #include <linux/netdevice.h>
43 #include <linux/workqueue.h>
44 #include <linux/kref.h>
45 #include <linux/if_infiniband.h>
46 #include <linux/mutex.h>
48 #include <net/neighbour.h>
50 #include <asm/atomic.h>
52 #include <rdma/ib_verbs.h>
53 #include <rdma/ib_pack.h>
54 #include <rdma/ib_sa.h>
59 IPOIB_PACKET_SIZE = 2048,
60 IPOIB_BUF_SIZE = IPOIB_PACKET_SIZE + IB_GRH_BYTES,
64 IPOIB_CM_MTU = 0x10000 - 0x10, /* padding to align header to 16 */
65 IPOIB_CM_BUF_SIZE = IPOIB_CM_MTU + IPOIB_ENCAP_LEN,
66 IPOIB_CM_HEAD_SIZE = IPOIB_CM_BUF_SIZE % PAGE_SIZE,
67 IPOIB_CM_RX_SG = ALIGN(IPOIB_CM_BUF_SIZE, PAGE_SIZE) / PAGE_SIZE,
68 IPOIB_RX_RING_SIZE = 128,
69 IPOIB_TX_RING_SIZE = 64,
70 IPOIB_MAX_QUEUE_SIZE = 8192,
71 IPOIB_MIN_QUEUE_SIZE = 2,
72 IPOIB_CM_MAX_CONN_QP = 4096,
76 IPOIB_MAX_PATH_REC_QUEUE = 3,
77 IPOIB_MAX_MCAST_QUEUE = 3,
79 IPOIB_FLAG_OPER_UP = 0,
80 IPOIB_FLAG_INITIALIZED = 1,
81 IPOIB_FLAG_ADMIN_UP = 2,
82 IPOIB_PKEY_ASSIGNED = 3,
84 IPOIB_FLAG_SUBINTERFACE = 5,
86 IPOIB_STOP_REAPER = 7,
87 IPOIB_MCAST_STARTED = 8,
88 IPOIB_FLAG_ADMIN_CM = 9,
89 IPOIB_FLAG_UMCAST = 10,
92 IPOIB_MAX_BACKOFF_SECONDS = 16,
94 IPOIB_MCAST_FLAG_FOUND = 0, /* used in set_multicast_list */
95 IPOIB_MCAST_FLAG_SENDONLY = 1,
96 IPOIB_MCAST_FLAG_BUSY = 2, /* joining or already joined */
97 IPOIB_MCAST_FLAG_ATTACHED = 3,
100 #define IPOIB_OP_RECV (1ul << 31)
101 #ifdef CONFIG_INFINIBAND_IPOIB_CM
102 #define IPOIB_OP_CM (1ul << 30)
104 #define IPOIB_OP_CM (0)
109 struct ipoib_header {
114 struct ipoib_pseudoheader {
115 u8 hwaddr[INFINIBAND_ALEN];
118 /* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
120 struct ib_sa_mcmember_rec mcmember;
121 struct ib_sa_multicast *mc;
124 struct rb_node rb_node;
125 struct list_head list;
127 unsigned long created;
128 unsigned long backoff;
131 unsigned char logcount;
133 struct list_head neigh_list;
135 struct sk_buff_head pkt_queue;
137 struct net_device *dev;
140 struct ipoib_rx_buf {
145 struct ipoib_tx_buf {
147 u64 mapping[MAX_SKB_FRAGS + 1];
152 struct ipoib_cm_data {
153 __be32 qpn; /* High byte MUST be ignored on receive */
158 * Quoting 10.3.1 Queue Pair and EE Context States:
160 * Note, for QPs that are associated with an SRQ, the Consumer should take the
161 * QP through the Error State before invoking a Destroy QP or a Modify QP to the
162 * Reset State. The Consumer may invoke the Destroy QP without first performing
163 * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
164 * Last WQE Reached Event. However, if the Consumer does not wait for the
165 * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
166 * leakage may occur. Therefore, it is good programming practice to tear down a
167 * QP that is associated with an SRQ by using the following process:
169 * - Put the QP in the Error State
170 * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
172 * drain the CQ by invoking the Poll CQ verb and either wait for CQ
173 * to be empty or the number of Poll CQ operations has exceeded
176 * post another WR that completes on the same CQ and wait for this
177 * WR to return as a WC;
178 * - and then invoke a Destroy QP or Reset QP.
180 * We use the second option and wait for a completion on the
181 * same CQ before destroying QPs attached to our SRQ.
184 enum ipoib_cm_state {
186 IPOIB_CM_RX_ERROR, /* Ignored by stale task */
187 IPOIB_CM_RX_FLUSH /* Last WQE Reached event observed */
193 struct ipoib_cm_rx_buf *rx_ring;
194 struct list_head list;
195 struct net_device *dev;
196 unsigned long jiffies;
197 enum ipoib_cm_state state;
204 struct list_head list;
205 struct net_device *dev;
206 struct ipoib_neigh *neigh;
207 struct ipoib_path *path;
208 struct ipoib_tx_buf *tx_ring;
215 struct ipoib_cm_rx_buf {
217 u64 mapping[IPOIB_CM_RX_SG];
220 struct ipoib_cm_dev_priv {
222 struct ipoib_cm_rx_buf *srq_ring;
224 struct list_head passive_ids; /* state: LIVE */
225 struct list_head rx_error_list; /* state: ERROR */
226 struct list_head rx_flush_list; /* state: FLUSH, drain not started */
227 struct list_head rx_drain_list; /* state: FLUSH, drain started */
228 struct list_head rx_reap_list; /* state: FLUSH, drain done */
229 struct work_struct start_task;
230 struct work_struct reap_task;
231 struct work_struct skb_task;
232 struct work_struct rx_reap_task;
233 struct delayed_work stale_task;
234 struct sk_buff_head skb_queue;
235 struct list_head start_list;
236 struct list_head reap_list;
237 struct ib_wc ibwc[IPOIB_NUM_WC];
238 struct ib_sge rx_sge[IPOIB_CM_RX_SG];
239 struct ib_recv_wr rx_wr;
245 struct ipoib_ethtool_st {
247 u16 max_coalesced_frames;
251 * Device private locking: tx_lock protects members used in TX fast
252 * path (and we use LLTX so upper layers don't do extra locking).
253 * lock protects everything else. lock nests inside of tx_lock (ie
254 * tx_lock must be acquired first if needed).
256 struct ipoib_dev_priv {
259 struct net_device *dev;
261 struct napi_struct napi;
265 struct mutex mcast_mutex;
266 struct mutex vlan_mutex;
268 struct rb_root path_tree;
269 struct list_head path_list;
271 struct ipoib_mcast *broadcast;
272 struct list_head multicast_list;
273 struct rb_root multicast_tree;
275 struct delayed_work pkey_poll_task;
276 struct delayed_work mcast_task;
277 struct work_struct flush_task;
278 struct work_struct restart_task;
279 struct delayed_work ah_reap_task;
280 struct work_struct pkey_event_task;
282 struct ib_device *ca;
292 union ib_gid local_gid;
295 unsigned int admin_mtu;
296 unsigned int mcast_mtu;
298 struct ipoib_rx_buf *rx_ring;
301 struct ipoib_tx_buf *tx_ring;
304 struct ib_sge tx_sge[MAX_SKB_FRAGS + 1];
305 struct ib_send_wr tx_wr;
306 unsigned tx_outstanding;
308 struct ib_wc ibwc[IPOIB_NUM_WC];
310 struct list_head dead_ahs;
312 struct ib_event_handler event_handler;
314 struct net_device *parent;
315 struct list_head child_intfs;
316 struct list_head list;
318 #ifdef CONFIG_INFINIBAND_IPOIB_CM
319 struct ipoib_cm_dev_priv cm;
322 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
323 struct list_head fs_list;
324 struct dentry *mcg_dentry;
325 struct dentry *path_dentry;
328 struct ipoib_ethtool_st ethtool;
332 struct net_device *dev;
334 struct list_head list;
340 struct net_device *dev;
341 struct ib_sa_path_rec pathrec;
343 struct sk_buff_head queue;
345 struct list_head neigh_list;
348 struct ib_sa_query *query;
349 struct completion done;
351 struct rb_node rb_node;
352 struct list_head list;
357 #ifdef CONFIG_INFINIBAND_IPOIB_CM
358 struct ipoib_cm_tx *cm;
361 struct sk_buff_head queue;
363 struct neighbour *neighbour;
364 struct net_device *dev;
366 struct list_head list;
370 * We stash a pointer to our private neighbour information after our
371 * hardware address in neigh->ha. The ALIGN() expression here makes
372 * sure that this pointer is stored aligned so that an unaligned
373 * load is not needed to dereference it.
375 static inline struct ipoib_neigh **to_ipoib_neigh(struct neighbour *neigh)
377 return (void*) neigh + ALIGN(offsetof(struct neighbour, ha) +
378 INFINIBAND_ALEN, sizeof(void *));
381 struct ipoib_neigh *ipoib_neigh_alloc(struct neighbour *neigh,
382 struct net_device *dev);
383 void ipoib_neigh_free(struct net_device *dev, struct ipoib_neigh *neigh);
385 extern struct workqueue_struct *ipoib_workqueue;
389 int ipoib_poll(struct napi_struct *napi, int budget);
390 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr);
392 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
393 struct ib_pd *pd, struct ib_ah_attr *attr);
394 void ipoib_free_ah(struct kref *kref);
395 static inline void ipoib_put_ah(struct ipoib_ah *ah)
397 kref_put(&ah->ref, ipoib_free_ah);
400 int ipoib_open(struct net_device *dev);
401 int ipoib_add_pkey_attr(struct net_device *dev);
402 int ipoib_add_umcast_attr(struct net_device *dev);
404 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
405 struct ipoib_ah *address, u32 qpn);
406 void ipoib_reap_ah(struct work_struct *work);
408 void ipoib_flush_paths(struct net_device *dev);
409 struct ipoib_dev_priv *ipoib_intf_alloc(const char *format);
411 int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
412 void ipoib_ib_dev_flush(struct work_struct *work);
413 void ipoib_pkey_event(struct work_struct *work);
414 void ipoib_ib_dev_cleanup(struct net_device *dev);
416 int ipoib_ib_dev_open(struct net_device *dev);
417 int ipoib_ib_dev_up(struct net_device *dev);
418 int ipoib_ib_dev_down(struct net_device *dev, int flush);
419 int ipoib_ib_dev_stop(struct net_device *dev, int flush);
421 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
422 void ipoib_dev_cleanup(struct net_device *dev);
424 void ipoib_mcast_join_task(struct work_struct *work);
425 void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb);
427 void ipoib_mcast_restart_task(struct work_struct *work);
428 int ipoib_mcast_start_thread(struct net_device *dev);
429 int ipoib_mcast_stop_thread(struct net_device *dev, int flush);
431 void ipoib_mcast_dev_down(struct net_device *dev);
432 void ipoib_mcast_dev_flush(struct net_device *dev);
434 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
435 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev);
436 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter);
437 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
439 unsigned long *created,
440 unsigned int *queuelen,
441 unsigned int *complete,
442 unsigned int *send_only);
444 struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev);
445 int ipoib_path_iter_next(struct ipoib_path_iter *iter);
446 void ipoib_path_iter_read(struct ipoib_path_iter *iter,
447 struct ipoib_path *path);
450 int ipoib_mcast_attach(struct net_device *dev, u16 mlid,
452 int ipoib_mcast_detach(struct net_device *dev, u16 mlid,
455 int ipoib_init_qp(struct net_device *dev);
456 int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca);
457 void ipoib_transport_dev_cleanup(struct net_device *dev);
459 void ipoib_event(struct ib_event_handler *handler,
460 struct ib_event *record);
462 int ipoib_vlan_add(struct net_device *pdev, unsigned short pkey);
463 int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey);
465 void ipoib_pkey_poll(struct work_struct *work);
466 int ipoib_pkey_dev_delay_open(struct net_device *dev);
467 void ipoib_drain_cq(struct net_device *dev);
469 void ipoib_set_ethtool_ops(struct net_device *dev);
471 #ifdef CONFIG_INFINIBAND_IPOIB_CM
473 #define IPOIB_FLAGS_RC 0x80
474 #define IPOIB_FLAGS_UC 0x40
476 /* We don't support UC connections at the moment */
477 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
479 extern int ipoib_max_conn_qp;
481 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
483 struct ipoib_dev_priv *priv = netdev_priv(dev);
484 return IPOIB_CM_SUPPORTED(dev->dev_addr) &&
485 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
488 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
490 struct ipoib_dev_priv *priv = netdev_priv(dev);
491 return IPOIB_CM_SUPPORTED(n->ha) &&
492 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
495 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
498 return test_bit(IPOIB_FLAG_OPER_UP, &neigh->cm->flags);
501 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
506 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
511 static inline int ipoib_cm_has_srq(struct net_device *dev)
513 struct ipoib_dev_priv *priv = netdev_priv(dev);
514 return !!priv->cm.srq;
517 static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
519 struct ipoib_dev_priv *priv = netdev_priv(dev);
520 return priv->cm.max_cm_mtu;
523 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx);
524 int ipoib_cm_dev_open(struct net_device *dev);
525 void ipoib_cm_dev_stop(struct net_device *dev);
526 int ipoib_cm_dev_init(struct net_device *dev);
527 int ipoib_cm_add_mode_attr(struct net_device *dev);
528 void ipoib_cm_dev_cleanup(struct net_device *dev);
529 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
530 struct ipoib_neigh *neigh);
531 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx);
532 void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
534 void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc);
535 void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc);
540 #define ipoib_max_conn_qp 0
542 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
546 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
552 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
558 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
563 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
567 static inline int ipoib_cm_has_srq(struct net_device *dev)
572 static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
578 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx)
584 int ipoib_cm_dev_open(struct net_device *dev)
590 void ipoib_cm_dev_stop(struct net_device *dev)
596 int ipoib_cm_dev_init(struct net_device *dev)
602 void ipoib_cm_dev_cleanup(struct net_device *dev)
608 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
609 struct ipoib_neigh *neigh)
615 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx)
621 int ipoib_cm_add_mode_attr(struct net_device *dev)
626 static inline void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
629 dev_kfree_skb_any(skb);
632 static inline void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
636 static inline void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
641 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
642 void ipoib_create_debug_files(struct net_device *dev);
643 void ipoib_delete_debug_files(struct net_device *dev);
644 int ipoib_register_debugfs(void);
645 void ipoib_unregister_debugfs(void);
647 static inline void ipoib_create_debug_files(struct net_device *dev) { }
648 static inline void ipoib_delete_debug_files(struct net_device *dev) { }
649 static inline int ipoib_register_debugfs(void) { return 0; }
650 static inline void ipoib_unregister_debugfs(void) { }
654 #define ipoib_printk(level, priv, format, arg...) \
655 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
656 #define ipoib_warn(priv, format, arg...) \
657 ipoib_printk(KERN_WARNING, priv, format , ## arg)
659 extern int ipoib_sendq_size;
660 extern int ipoib_recvq_size;
662 extern struct ib_sa_client ipoib_sa_client;
664 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
665 extern int ipoib_debug_level;
667 #define ipoib_dbg(priv, format, arg...) \
669 if (ipoib_debug_level > 0) \
670 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
672 #define ipoib_dbg_mcast(priv, format, arg...) \
674 if (mcast_debug_level > 0) \
675 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
677 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
678 #define ipoib_dbg(priv, format, arg...) \
679 do { (void) (priv); } while (0)
680 #define ipoib_dbg_mcast(priv, format, arg...) \
681 do { (void) (priv); } while (0)
682 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
684 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
685 #define ipoib_dbg_data(priv, format, arg...) \
687 if (data_debug_level > 0) \
688 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
690 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
691 #define ipoib_dbg_data(priv, format, arg...) \
692 do { (void) (priv); } while (0)
693 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
696 #define IPOIB_GID_FMT "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:" \
697 "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x"
699 #define IPOIB_GID_RAW_ARG(gid) ((u8 *)(gid))[0], \
716 #define IPOIB_GID_ARG(gid) IPOIB_GID_RAW_ARG((gid).raw)
718 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
720 #endif /* _IPOIB_H */