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
38 #include <linux/list.h>
39 #include <linux/skbuff.h>
40 #include <linux/netdevice.h>
41 #include <linux/workqueue.h>
42 #include <linux/kref.h>
43 #include <linux/if_infiniband.h>
44 #include <linux/mutex.h>
46 #include <net/neighbour.h>
47 #include <net/sch_generic.h>
49 #include <linux/atomic.h>
51 #include <rdma/ib_verbs.h>
52 #include <rdma/ib_pack.h>
53 #include <rdma/ib_sa.h>
54 #include <linux/sched.h>
58 enum ipoib_flush_level {
67 IPOIB_UD_HEAD_SIZE = IB_GRH_BYTES + IPOIB_ENCAP_LEN,
68 IPOIB_UD_RX_SG = 2, /* max buffer needed for 4K mtu */
70 IPOIB_CM_MTU = 0x10000 - 0x10, /* padding to align header to 16 */
71 IPOIB_CM_BUF_SIZE = IPOIB_CM_MTU + IPOIB_ENCAP_LEN,
72 IPOIB_CM_HEAD_SIZE = IPOIB_CM_BUF_SIZE % PAGE_SIZE,
73 IPOIB_CM_RX_SG = ALIGN(IPOIB_CM_BUF_SIZE, PAGE_SIZE) / PAGE_SIZE,
74 IPOIB_RX_RING_SIZE = 256,
75 IPOIB_TX_RING_SIZE = 128,
76 IPOIB_MAX_QUEUE_SIZE = 8192,
77 IPOIB_MIN_QUEUE_SIZE = 2,
78 IPOIB_CM_MAX_CONN_QP = 4096,
82 IPOIB_MAX_PATH_REC_QUEUE = 3,
83 IPOIB_MAX_MCAST_QUEUE = 3,
85 IPOIB_FLAG_OPER_UP = 0,
86 IPOIB_FLAG_INITIALIZED = 1,
87 IPOIB_FLAG_ADMIN_UP = 2,
88 IPOIB_PKEY_ASSIGNED = 3,
90 IPOIB_FLAG_SUBINTERFACE = 5,
92 IPOIB_STOP_REAPER = 7,
93 IPOIB_FLAG_ADMIN_CM = 9,
94 IPOIB_FLAG_UMCAST = 10,
95 IPOIB_STOP_NEIGH_GC = 11,
96 IPOIB_NEIGH_TBL_FLUSH = 12,
98 IPOIB_MAX_BACKOFF_SECONDS = 16,
100 IPOIB_MCAST_FLAG_FOUND = 0, /* used in set_multicast_list */
101 IPOIB_MCAST_FLAG_SENDONLY = 1,
102 IPOIB_MCAST_FLAG_BUSY = 2, /* joining or already joined */
103 IPOIB_MCAST_FLAG_ATTACHED = 3,
106 IPOIB_CM_COPYBREAK = 256,
109 IPOIB_LEGACY_CHILD = 1,
110 IPOIB_RTNL_CHILD = 2,
113 #define IPOIB_OP_RECV (1ul << 31)
114 #ifdef CONFIG_INFINIBAND_IPOIB_CM
115 #define IPOIB_OP_CM (1ul << 30)
117 #define IPOIB_OP_CM (0)
122 struct ipoib_header {
128 struct qdisc_skb_cb qdisc_cb;
129 u8 hwaddr[INFINIBAND_ALEN];
132 /* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
134 struct ib_sa_mcmember_rec mcmember;
135 struct ib_sa_multicast *mc;
138 struct rb_node rb_node;
139 struct list_head list;
141 unsigned long created;
142 unsigned long backoff;
145 unsigned char logcount;
147 struct list_head neigh_list;
149 struct sk_buff_head pkt_queue;
151 struct net_device *dev;
154 struct ipoib_rx_buf {
156 u64 mapping[IPOIB_UD_RX_SG];
159 struct ipoib_tx_buf {
161 u64 mapping[MAX_SKB_FRAGS + 1];
164 struct ipoib_cm_tx_buf {
171 struct ipoib_cm_data {
172 __be32 qpn; /* High byte MUST be ignored on receive */
177 * Quoting 10.3.1 Queue Pair and EE Context States:
179 * Note, for QPs that are associated with an SRQ, the Consumer should take the
180 * QP through the Error State before invoking a Destroy QP or a Modify QP to the
181 * Reset State. The Consumer may invoke the Destroy QP without first performing
182 * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
183 * Last WQE Reached Event. However, if the Consumer does not wait for the
184 * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
185 * leakage may occur. Therefore, it is good programming practice to tear down a
186 * QP that is associated with an SRQ by using the following process:
188 * - Put the QP in the Error State
189 * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
191 * drain the CQ by invoking the Poll CQ verb and either wait for CQ
192 * to be empty or the number of Poll CQ operations has exceeded
195 * post another WR that completes on the same CQ and wait for this
196 * WR to return as a WC;
197 * - and then invoke a Destroy QP or Reset QP.
199 * We use the second option and wait for a completion on the
200 * same CQ before destroying QPs attached to our SRQ.
203 enum ipoib_cm_state {
205 IPOIB_CM_RX_ERROR, /* Ignored by stale task */
206 IPOIB_CM_RX_FLUSH /* Last WQE Reached event observed */
212 struct ipoib_cm_rx_buf *rx_ring;
213 struct list_head list;
214 struct net_device *dev;
215 unsigned long jiffies;
216 enum ipoib_cm_state state;
223 struct list_head list;
224 struct net_device *dev;
225 struct ipoib_neigh *neigh;
226 struct ipoib_path *path;
227 struct ipoib_cm_tx_buf *tx_ring;
234 struct ipoib_cm_rx_buf {
236 u64 mapping[IPOIB_CM_RX_SG];
239 struct ipoib_cm_dev_priv {
241 struct ipoib_cm_rx_buf *srq_ring;
243 struct list_head passive_ids; /* state: LIVE */
244 struct list_head rx_error_list; /* state: ERROR */
245 struct list_head rx_flush_list; /* state: FLUSH, drain not started */
246 struct list_head rx_drain_list; /* state: FLUSH, drain started */
247 struct list_head rx_reap_list; /* state: FLUSH, drain done */
248 struct work_struct start_task;
249 struct work_struct reap_task;
250 struct work_struct skb_task;
251 struct work_struct rx_reap_task;
252 struct delayed_work stale_task;
253 struct sk_buff_head skb_queue;
254 struct list_head start_list;
255 struct list_head reap_list;
256 struct ib_wc ibwc[IPOIB_NUM_WC];
257 struct ib_sge rx_sge[IPOIB_CM_RX_SG];
258 struct ib_recv_wr rx_wr;
264 struct ipoib_ethtool_st {
266 u16 max_coalesced_frames;
269 struct ipoib_neigh_hash {
270 struct ipoib_neigh __rcu **buckets;
276 struct ipoib_neigh_table {
277 struct ipoib_neigh_hash __rcu *htbl;
280 struct completion flushed;
284 * Device private locking: network stack tx_lock protects members used
285 * in TX fast path, lock protects everything else. lock nests inside
286 * of tx_lock (ie tx_lock must be acquired first if needed).
288 struct ipoib_dev_priv {
291 struct net_device *dev;
293 struct napi_struct napi;
297 struct mutex vlan_mutex;
299 struct rb_root path_tree;
300 struct list_head path_list;
302 struct ipoib_neigh_table ntbl;
304 struct ipoib_mcast *broadcast;
305 struct list_head multicast_list;
306 struct rb_root multicast_tree;
308 struct delayed_work pkey_poll_task;
309 struct delayed_work mcast_task;
310 struct work_struct carrier_on_task;
311 struct work_struct flush_light;
312 struct work_struct flush_normal;
313 struct work_struct flush_heavy;
314 struct work_struct restart_task;
315 struct delayed_work ah_reap_task;
316 struct delayed_work neigh_reap_task;
317 struct ib_device *ca;
323 struct ib_cq *recv_cq;
324 struct ib_cq *send_cq;
328 union ib_gid local_gid;
331 unsigned int admin_mtu;
332 unsigned int mcast_mtu;
333 unsigned int max_ib_mtu;
335 struct ipoib_rx_buf *rx_ring;
337 struct ipoib_tx_buf *tx_ring;
340 struct ib_sge tx_sge[MAX_SKB_FRAGS + 1];
341 struct ib_send_wr tx_wr;
342 unsigned tx_outstanding;
343 struct ib_wc send_wc[MAX_SEND_CQE];
345 struct ib_recv_wr rx_wr;
346 struct ib_sge rx_sge[IPOIB_UD_RX_SG];
348 struct ib_wc ibwc[IPOIB_NUM_WC];
350 struct list_head dead_ahs;
352 struct ib_event_handler event_handler;
354 struct net_device *parent;
355 struct list_head child_intfs;
356 struct list_head list;
359 #ifdef CONFIG_INFINIBAND_IPOIB_CM
360 struct ipoib_cm_dev_priv cm;
363 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
364 struct list_head fs_list;
365 struct dentry *mcg_dentry;
366 struct dentry *path_dentry;
369 struct ipoib_ethtool_st ethtool;
370 struct timer_list poll_timer;
374 struct net_device *dev;
376 struct list_head list;
382 struct net_device *dev;
383 struct ib_sa_path_rec pathrec;
385 struct sk_buff_head queue;
387 struct list_head neigh_list;
390 struct ib_sa_query *query;
391 struct completion done;
393 struct rb_node rb_node;
394 struct list_head list;
400 #ifdef CONFIG_INFINIBAND_IPOIB_CM
401 struct ipoib_cm_tx *cm;
403 u8 daddr[INFINIBAND_ALEN];
404 struct sk_buff_head queue;
406 struct net_device *dev;
408 struct list_head list;
409 struct ipoib_neigh __rcu *hnext;
415 #define IPOIB_UD_MTU(ib_mtu) (ib_mtu - IPOIB_ENCAP_LEN)
416 #define IPOIB_UD_BUF_SIZE(ib_mtu) (ib_mtu + IB_GRH_BYTES)
418 static inline int ipoib_ud_need_sg(unsigned int ib_mtu)
420 return IPOIB_UD_BUF_SIZE(ib_mtu) > PAGE_SIZE;
423 void ipoib_neigh_dtor(struct ipoib_neigh *neigh);
424 static inline void ipoib_neigh_put(struct ipoib_neigh *neigh)
426 if (atomic_dec_and_test(&neigh->refcnt))
427 ipoib_neigh_dtor(neigh);
429 struct ipoib_neigh *ipoib_neigh_get(struct net_device *dev, u8 *daddr);
430 struct ipoib_neigh *ipoib_neigh_alloc(u8 *daddr,
431 struct net_device *dev);
432 void ipoib_neigh_free(struct ipoib_neigh *neigh);
433 void ipoib_del_neighs_by_gid(struct net_device *dev, u8 *gid);
435 extern struct workqueue_struct *ipoib_workqueue;
439 int ipoib_poll(struct napi_struct *napi, int budget);
440 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr);
441 void ipoib_send_comp_handler(struct ib_cq *cq, void *dev_ptr);
443 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
444 struct ib_pd *pd, struct ib_ah_attr *attr);
445 void ipoib_free_ah(struct kref *kref);
446 static inline void ipoib_put_ah(struct ipoib_ah *ah)
448 kref_put(&ah->ref, ipoib_free_ah);
450 int ipoib_open(struct net_device *dev);
451 int ipoib_add_pkey_attr(struct net_device *dev);
452 int ipoib_add_umcast_attr(struct net_device *dev);
454 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
455 struct ipoib_ah *address, u32 qpn);
456 void ipoib_reap_ah(struct work_struct *work);
458 void ipoib_mark_paths_invalid(struct net_device *dev);
459 void ipoib_flush_paths(struct net_device *dev);
460 struct ipoib_dev_priv *ipoib_intf_alloc(const char *format);
462 int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
463 void ipoib_ib_dev_flush_light(struct work_struct *work);
464 void ipoib_ib_dev_flush_normal(struct work_struct *work);
465 void ipoib_ib_dev_flush_heavy(struct work_struct *work);
466 void ipoib_pkey_event(struct work_struct *work);
467 void ipoib_ib_dev_cleanup(struct net_device *dev);
469 int ipoib_ib_dev_open(struct net_device *dev);
470 int ipoib_ib_dev_up(struct net_device *dev);
471 int ipoib_ib_dev_down(struct net_device *dev, int flush);
472 int ipoib_ib_dev_stop(struct net_device *dev, int flush);
474 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
475 void ipoib_dev_cleanup(struct net_device *dev);
477 void ipoib_mcast_join_task(struct work_struct *work);
478 void ipoib_mcast_carrier_on_task(struct work_struct *work);
479 void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb);
481 void ipoib_mcast_restart_task(struct work_struct *work);
482 int ipoib_mcast_start_thread(struct net_device *dev);
483 int ipoib_mcast_stop_thread(struct net_device *dev, int flush);
485 void ipoib_mcast_dev_down(struct net_device *dev);
486 void ipoib_mcast_dev_flush(struct net_device *dev);
488 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
489 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev);
490 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter);
491 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
493 unsigned long *created,
494 unsigned int *queuelen,
495 unsigned int *complete,
496 unsigned int *send_only);
498 struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev);
499 int ipoib_path_iter_next(struct ipoib_path_iter *iter);
500 void ipoib_path_iter_read(struct ipoib_path_iter *iter,
501 struct ipoib_path *path);
504 int ipoib_mcast_attach(struct net_device *dev, u16 mlid,
505 union ib_gid *mgid, int set_qkey);
507 int ipoib_init_qp(struct net_device *dev);
508 int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca);
509 void ipoib_transport_dev_cleanup(struct net_device *dev);
511 void ipoib_event(struct ib_event_handler *handler,
512 struct ib_event *record);
514 int ipoib_vlan_add(struct net_device *pdev, unsigned short pkey);
515 int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey);
517 int __ipoib_vlan_add(struct ipoib_dev_priv *ppriv, struct ipoib_dev_priv *priv,
518 u16 pkey, int child_type);
520 int __init ipoib_netlink_init(void);
521 void __exit ipoib_netlink_fini(void);
523 void ipoib_setup(struct net_device *dev);
525 void ipoib_pkey_poll(struct work_struct *work);
526 int ipoib_pkey_dev_delay_open(struct net_device *dev);
527 void ipoib_drain_cq(struct net_device *dev);
529 void ipoib_set_ethtool_ops(struct net_device *dev);
530 int ipoib_set_dev_features(struct ipoib_dev_priv *priv, struct ib_device *hca);
532 #ifdef CONFIG_INFINIBAND_IPOIB_CM
534 #define IPOIB_FLAGS_RC 0x80
535 #define IPOIB_FLAGS_UC 0x40
537 /* We don't support UC connections at the moment */
538 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
540 extern int ipoib_max_conn_qp;
542 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
544 struct ipoib_dev_priv *priv = netdev_priv(dev);
545 return IPOIB_CM_SUPPORTED(dev->dev_addr) &&
546 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
549 static inline int ipoib_cm_enabled(struct net_device *dev, u8 *hwaddr)
551 struct ipoib_dev_priv *priv = netdev_priv(dev);
552 return IPOIB_CM_SUPPORTED(hwaddr) &&
553 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
556 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
559 return test_bit(IPOIB_FLAG_OPER_UP, &neigh->cm->flags);
562 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
567 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
572 static inline int ipoib_cm_has_srq(struct net_device *dev)
574 struct ipoib_dev_priv *priv = netdev_priv(dev);
575 return !!priv->cm.srq;
578 static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
580 struct ipoib_dev_priv *priv = netdev_priv(dev);
581 return priv->cm.max_cm_mtu;
584 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx);
585 int ipoib_cm_dev_open(struct net_device *dev);
586 void ipoib_cm_dev_stop(struct net_device *dev);
587 int ipoib_cm_dev_init(struct net_device *dev);
588 int ipoib_cm_add_mode_attr(struct net_device *dev);
589 void ipoib_cm_dev_cleanup(struct net_device *dev);
590 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
591 struct ipoib_neigh *neigh);
592 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx);
593 void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
595 void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc);
596 void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc);
601 #define ipoib_max_conn_qp 0
603 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
607 static inline int ipoib_cm_enabled(struct net_device *dev, u8 *hwaddr)
613 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
619 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
624 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
628 static inline int ipoib_cm_has_srq(struct net_device *dev)
633 static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
639 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx)
645 int ipoib_cm_dev_open(struct net_device *dev)
651 void ipoib_cm_dev_stop(struct net_device *dev)
657 int ipoib_cm_dev_init(struct net_device *dev)
663 void ipoib_cm_dev_cleanup(struct net_device *dev)
669 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
670 struct ipoib_neigh *neigh)
676 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx)
682 int ipoib_cm_add_mode_attr(struct net_device *dev)
687 static inline void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
690 dev_kfree_skb_any(skb);
693 static inline void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
697 static inline void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
702 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
703 void ipoib_create_debug_files(struct net_device *dev);
704 void ipoib_delete_debug_files(struct net_device *dev);
705 int ipoib_register_debugfs(void);
706 void ipoib_unregister_debugfs(void);
708 static inline void ipoib_create_debug_files(struct net_device *dev) { }
709 static inline void ipoib_delete_debug_files(struct net_device *dev) { }
710 static inline int ipoib_register_debugfs(void) { return 0; }
711 static inline void ipoib_unregister_debugfs(void) { }
714 #define ipoib_printk(level, priv, format, arg...) \
715 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
716 #define ipoib_warn(priv, format, arg...) \
717 ipoib_printk(KERN_WARNING, priv, format , ## arg)
719 extern int ipoib_sendq_size;
720 extern int ipoib_recvq_size;
722 extern struct ib_sa_client ipoib_sa_client;
724 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
725 extern int ipoib_debug_level;
727 #define ipoib_dbg(priv, format, arg...) \
729 if (ipoib_debug_level > 0) \
730 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
732 #define ipoib_dbg_mcast(priv, format, arg...) \
734 if (mcast_debug_level > 0) \
735 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
737 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
738 #define ipoib_dbg(priv, format, arg...) \
739 do { (void) (priv); } while (0)
740 #define ipoib_dbg_mcast(priv, format, arg...) \
741 do { (void) (priv); } while (0)
742 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
744 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
745 #define ipoib_dbg_data(priv, format, arg...) \
747 if (data_debug_level > 0) \
748 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
750 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
751 #define ipoib_dbg_data(priv, format, arg...) \
752 do { (void) (priv); } while (0)
753 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
755 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
757 #endif /* _IPOIB_H */