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>
48 #include <asm/atomic.h>
50 #include <rdma/ib_verbs.h>
51 #include <rdma/ib_pack.h>
52 #include <rdma/ib_sa.h>
59 IPOIB_UD_HEAD_SIZE = IB_GRH_BYTES + IPOIB_ENCAP_LEN,
60 IPOIB_UD_RX_SG = 2, /* max buffer needed for 4K mtu */
62 IPOIB_CM_MTU = 0x10000 - 0x10, /* padding to align header to 16 */
63 IPOIB_CM_BUF_SIZE = IPOIB_CM_MTU + IPOIB_ENCAP_LEN,
64 IPOIB_CM_HEAD_SIZE = IPOIB_CM_BUF_SIZE % PAGE_SIZE,
65 IPOIB_CM_RX_SG = ALIGN(IPOIB_CM_BUF_SIZE, PAGE_SIZE) / PAGE_SIZE,
66 IPOIB_RX_RING_SIZE = 128,
67 IPOIB_TX_RING_SIZE = 64,
68 IPOIB_MAX_QUEUE_SIZE = 8192,
69 IPOIB_MIN_QUEUE_SIZE = 2,
70 IPOIB_CM_MAX_CONN_QP = 4096,
74 IPOIB_MAX_PATH_REC_QUEUE = 3,
75 IPOIB_MAX_MCAST_QUEUE = 3,
77 IPOIB_FLAG_OPER_UP = 0,
78 IPOIB_FLAG_INITIALIZED = 1,
79 IPOIB_FLAG_ADMIN_UP = 2,
80 IPOIB_PKEY_ASSIGNED = 3,
82 IPOIB_FLAG_SUBINTERFACE = 5,
84 IPOIB_STOP_REAPER = 7,
85 IPOIB_MCAST_STARTED = 8,
86 IPOIB_FLAG_ADMIN_CM = 9,
87 IPOIB_FLAG_UMCAST = 10,
90 IPOIB_MAX_BACKOFF_SECONDS = 16,
92 IPOIB_MCAST_FLAG_FOUND = 0, /* used in set_multicast_list */
93 IPOIB_MCAST_FLAG_SENDONLY = 1,
94 IPOIB_MCAST_FLAG_BUSY = 2, /* joining or already joined */
95 IPOIB_MCAST_FLAG_ATTACHED = 3,
98 IPOIB_CM_COPYBREAK = 256,
101 #define IPOIB_OP_RECV (1ul << 31)
102 #ifdef CONFIG_INFINIBAND_IPOIB_CM
103 #define IPOIB_OP_CM (1ul << 30)
105 #define IPOIB_OP_CM (0)
110 struct ipoib_header {
115 struct ipoib_pseudoheader {
116 u8 hwaddr[INFINIBAND_ALEN];
119 /* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
121 struct ib_sa_mcmember_rec mcmember;
122 struct ib_sa_multicast *mc;
125 struct rb_node rb_node;
126 struct list_head list;
128 unsigned long created;
129 unsigned long backoff;
132 unsigned char logcount;
134 struct list_head neigh_list;
136 struct sk_buff_head pkt_queue;
138 struct net_device *dev;
141 struct ipoib_rx_buf {
143 u64 mapping[IPOIB_UD_RX_SG];
146 struct ipoib_tx_buf {
148 u64 mapping[MAX_SKB_FRAGS + 1];
153 struct ipoib_cm_data {
154 __be32 qpn; /* High byte MUST be ignored on receive */
159 * Quoting 10.3.1 Queue Pair and EE Context States:
161 * Note, for QPs that are associated with an SRQ, the Consumer should take the
162 * QP through the Error State before invoking a Destroy QP or a Modify QP to the
163 * Reset State. The Consumer may invoke the Destroy QP without first performing
164 * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
165 * Last WQE Reached Event. However, if the Consumer does not wait for the
166 * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
167 * leakage may occur. Therefore, it is good programming practice to tear down a
168 * QP that is associated with an SRQ by using the following process:
170 * - Put the QP in the Error State
171 * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
173 * drain the CQ by invoking the Poll CQ verb and either wait for CQ
174 * to be empty or the number of Poll CQ operations has exceeded
177 * post another WR that completes on the same CQ and wait for this
178 * WR to return as a WC;
179 * - and then invoke a Destroy QP or Reset QP.
181 * We use the second option and wait for a completion on the
182 * same CQ before destroying QPs attached to our SRQ.
185 enum ipoib_cm_state {
187 IPOIB_CM_RX_ERROR, /* Ignored by stale task */
188 IPOIB_CM_RX_FLUSH /* Last WQE Reached event observed */
194 struct ipoib_cm_rx_buf *rx_ring;
195 struct list_head list;
196 struct net_device *dev;
197 unsigned long jiffies;
198 enum ipoib_cm_state state;
205 struct list_head list;
206 struct net_device *dev;
207 struct ipoib_neigh *neigh;
208 struct ipoib_path *path;
209 struct ipoib_tx_buf *tx_ring;
216 struct ipoib_cm_rx_buf {
218 u64 mapping[IPOIB_CM_RX_SG];
221 struct ipoib_cm_dev_priv {
223 struct ipoib_cm_rx_buf *srq_ring;
225 struct list_head passive_ids; /* state: LIVE */
226 struct list_head rx_error_list; /* state: ERROR */
227 struct list_head rx_flush_list; /* state: FLUSH, drain not started */
228 struct list_head rx_drain_list; /* state: FLUSH, drain started */
229 struct list_head rx_reap_list; /* state: FLUSH, drain done */
230 struct work_struct start_task;
231 struct work_struct reap_task;
232 struct work_struct skb_task;
233 struct work_struct rx_reap_task;
234 struct delayed_work stale_task;
235 struct sk_buff_head skb_queue;
236 struct list_head start_list;
237 struct list_head reap_list;
238 struct ib_wc ibwc[IPOIB_NUM_WC];
239 struct ib_sge rx_sge[IPOIB_CM_RX_SG];
240 struct ib_recv_wr rx_wr;
246 struct ipoib_ethtool_st {
248 u16 max_coalesced_frames;
252 * Device private locking: tx_lock protects members used in TX fast
253 * path (and we use LLTX so upper layers don't do extra locking).
254 * lock protects everything else. lock nests inside of tx_lock (ie
255 * tx_lock must be acquired first if needed).
257 struct ipoib_dev_priv {
260 struct net_device *dev;
262 struct napi_struct napi;
266 struct mutex mcast_mutex;
267 struct mutex vlan_mutex;
269 struct rb_root path_tree;
270 struct list_head path_list;
272 struct ipoib_mcast *broadcast;
273 struct list_head multicast_list;
274 struct rb_root multicast_tree;
276 struct delayed_work pkey_poll_task;
277 struct delayed_work mcast_task;
278 struct work_struct flush_task;
279 struct work_struct restart_task;
280 struct delayed_work ah_reap_task;
281 struct work_struct pkey_event_task;
283 struct ib_device *ca;
289 struct ib_cq *recv_cq;
290 struct ib_cq *send_cq;
294 union ib_gid local_gid;
297 unsigned int admin_mtu;
298 unsigned int mcast_mtu;
299 unsigned int max_ib_mtu;
301 struct ipoib_rx_buf *rx_ring;
304 struct ipoib_tx_buf *tx_ring;
307 struct ib_sge tx_sge[MAX_SKB_FRAGS + 1];
308 struct ib_send_wr tx_wr;
309 unsigned tx_outstanding;
310 struct ib_wc send_wc[MAX_SEND_CQE];
312 struct ib_recv_wr rx_wr;
313 struct ib_sge rx_sge[IPOIB_UD_RX_SG];
315 struct ib_wc ibwc[IPOIB_NUM_WC];
317 struct list_head dead_ahs;
319 struct ib_event_handler event_handler;
321 struct net_device *parent;
322 struct list_head child_intfs;
323 struct list_head list;
325 #ifdef CONFIG_INFINIBAND_IPOIB_CM
326 struct ipoib_cm_dev_priv cm;
329 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
330 struct list_head fs_list;
331 struct dentry *mcg_dentry;
332 struct dentry *path_dentry;
335 struct ipoib_ethtool_st ethtool;
336 struct timer_list poll_timer;
340 struct net_device *dev;
342 struct list_head list;
348 struct net_device *dev;
349 struct ib_sa_path_rec pathrec;
351 struct sk_buff_head queue;
353 struct list_head neigh_list;
356 struct ib_sa_query *query;
357 struct completion done;
359 struct rb_node rb_node;
360 struct list_head list;
365 #ifdef CONFIG_INFINIBAND_IPOIB_CM
366 struct ipoib_cm_tx *cm;
369 struct sk_buff_head queue;
371 struct neighbour *neighbour;
372 struct net_device *dev;
374 struct list_head list;
377 #define IPOIB_UD_MTU(ib_mtu) (ib_mtu - IPOIB_ENCAP_LEN)
378 #define IPOIB_UD_BUF_SIZE(ib_mtu) (ib_mtu + IB_GRH_BYTES)
380 static inline int ipoib_ud_need_sg(unsigned int ib_mtu)
382 return IPOIB_UD_BUF_SIZE(ib_mtu) > PAGE_SIZE;
386 * We stash a pointer to our private neighbour information after our
387 * hardware address in neigh->ha. The ALIGN() expression here makes
388 * sure that this pointer is stored aligned so that an unaligned
389 * load is not needed to dereference it.
391 static inline struct ipoib_neigh **to_ipoib_neigh(struct neighbour *neigh)
393 return (void*) neigh + ALIGN(offsetof(struct neighbour, ha) +
394 INFINIBAND_ALEN, sizeof(void *));
397 struct ipoib_neigh *ipoib_neigh_alloc(struct neighbour *neigh,
398 struct net_device *dev);
399 void ipoib_neigh_free(struct net_device *dev, struct ipoib_neigh *neigh);
401 extern struct workqueue_struct *ipoib_workqueue;
405 int ipoib_poll(struct napi_struct *napi, int budget);
406 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr);
407 void ipoib_send_comp_handler(struct ib_cq *cq, void *dev_ptr);
409 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
410 struct ib_pd *pd, struct ib_ah_attr *attr);
411 void ipoib_free_ah(struct kref *kref);
412 static inline void ipoib_put_ah(struct ipoib_ah *ah)
414 kref_put(&ah->ref, ipoib_free_ah);
417 int ipoib_open(struct net_device *dev);
418 int ipoib_add_pkey_attr(struct net_device *dev);
419 int ipoib_add_umcast_attr(struct net_device *dev);
421 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
422 struct ipoib_ah *address, u32 qpn);
423 void ipoib_reap_ah(struct work_struct *work);
425 void ipoib_flush_paths(struct net_device *dev);
426 struct ipoib_dev_priv *ipoib_intf_alloc(const char *format);
428 int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
429 void ipoib_ib_dev_flush(struct work_struct *work);
430 void ipoib_pkey_event(struct work_struct *work);
431 void ipoib_ib_dev_cleanup(struct net_device *dev);
433 int ipoib_ib_dev_open(struct net_device *dev);
434 int ipoib_ib_dev_up(struct net_device *dev);
435 int ipoib_ib_dev_down(struct net_device *dev, int flush);
436 int ipoib_ib_dev_stop(struct net_device *dev, int flush);
438 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
439 void ipoib_dev_cleanup(struct net_device *dev);
441 void ipoib_mcast_join_task(struct work_struct *work);
442 void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb);
444 void ipoib_mcast_restart_task(struct work_struct *work);
445 int ipoib_mcast_start_thread(struct net_device *dev);
446 int ipoib_mcast_stop_thread(struct net_device *dev, int flush);
448 void ipoib_mcast_dev_down(struct net_device *dev);
449 void ipoib_mcast_dev_flush(struct net_device *dev);
451 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
452 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev);
453 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter);
454 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
456 unsigned long *created,
457 unsigned int *queuelen,
458 unsigned int *complete,
459 unsigned int *send_only);
461 struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev);
462 int ipoib_path_iter_next(struct ipoib_path_iter *iter);
463 void ipoib_path_iter_read(struct ipoib_path_iter *iter,
464 struct ipoib_path *path);
467 int ipoib_mcast_attach(struct net_device *dev, u16 mlid,
469 int ipoib_mcast_detach(struct net_device *dev, u16 mlid,
472 int ipoib_init_qp(struct net_device *dev);
473 int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca);
474 void ipoib_transport_dev_cleanup(struct net_device *dev);
476 void ipoib_event(struct ib_event_handler *handler,
477 struct ib_event *record);
479 int ipoib_vlan_add(struct net_device *pdev, unsigned short pkey);
480 int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey);
482 void ipoib_pkey_poll(struct work_struct *work);
483 int ipoib_pkey_dev_delay_open(struct net_device *dev);
484 void ipoib_drain_cq(struct net_device *dev);
486 void ipoib_set_ethtool_ops(struct net_device *dev);
488 #ifdef CONFIG_INFINIBAND_IPOIB_CM
490 #define IPOIB_FLAGS_RC 0x80
491 #define IPOIB_FLAGS_UC 0x40
493 /* We don't support UC connections at the moment */
494 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
496 extern int ipoib_max_conn_qp;
498 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
500 struct ipoib_dev_priv *priv = netdev_priv(dev);
501 return IPOIB_CM_SUPPORTED(dev->dev_addr) &&
502 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
505 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
507 struct ipoib_dev_priv *priv = netdev_priv(dev);
508 return IPOIB_CM_SUPPORTED(n->ha) &&
509 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
512 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
515 return test_bit(IPOIB_FLAG_OPER_UP, &neigh->cm->flags);
518 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
523 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
528 static inline int ipoib_cm_has_srq(struct net_device *dev)
530 struct ipoib_dev_priv *priv = netdev_priv(dev);
531 return !!priv->cm.srq;
534 static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
536 struct ipoib_dev_priv *priv = netdev_priv(dev);
537 return priv->cm.max_cm_mtu;
540 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx);
541 int ipoib_cm_dev_open(struct net_device *dev);
542 void ipoib_cm_dev_stop(struct net_device *dev);
543 int ipoib_cm_dev_init(struct net_device *dev);
544 int ipoib_cm_add_mode_attr(struct net_device *dev);
545 void ipoib_cm_dev_cleanup(struct net_device *dev);
546 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
547 struct ipoib_neigh *neigh);
548 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx);
549 void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
551 void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc);
552 void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc);
557 #define ipoib_max_conn_qp 0
559 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
563 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
569 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
575 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
580 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
584 static inline int ipoib_cm_has_srq(struct net_device *dev)
589 static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
595 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx)
601 int ipoib_cm_dev_open(struct net_device *dev)
607 void ipoib_cm_dev_stop(struct net_device *dev)
613 int ipoib_cm_dev_init(struct net_device *dev)
619 void ipoib_cm_dev_cleanup(struct net_device *dev)
625 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
626 struct ipoib_neigh *neigh)
632 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx)
638 int ipoib_cm_add_mode_attr(struct net_device *dev)
643 static inline void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
646 dev_kfree_skb_any(skb);
649 static inline void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
653 static inline void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
658 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
659 void ipoib_create_debug_files(struct net_device *dev);
660 void ipoib_delete_debug_files(struct net_device *dev);
661 int ipoib_register_debugfs(void);
662 void ipoib_unregister_debugfs(void);
664 static inline void ipoib_create_debug_files(struct net_device *dev) { }
665 static inline void ipoib_delete_debug_files(struct net_device *dev) { }
666 static inline int ipoib_register_debugfs(void) { return 0; }
667 static inline void ipoib_unregister_debugfs(void) { }
670 #define ipoib_printk(level, priv, format, arg...) \
671 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
672 #define ipoib_warn(priv, format, arg...) \
673 ipoib_printk(KERN_WARNING, priv, format , ## arg)
675 extern int ipoib_sendq_size;
676 extern int ipoib_recvq_size;
678 extern struct ib_sa_client ipoib_sa_client;
680 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
681 extern int ipoib_debug_level;
683 #define ipoib_dbg(priv, format, arg...) \
685 if (ipoib_debug_level > 0) \
686 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
688 #define ipoib_dbg_mcast(priv, format, arg...) \
690 if (mcast_debug_level > 0) \
691 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
693 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
694 #define ipoib_dbg(priv, format, arg...) \
695 do { (void) (priv); } while (0)
696 #define ipoib_dbg_mcast(priv, format, arg...) \
697 do { (void) (priv); } while (0)
698 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
700 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
701 #define ipoib_dbg_data(priv, format, arg...) \
703 if (data_debug_level > 0) \
704 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
706 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
707 #define ipoib_dbg_data(priv, format, arg...) \
708 do { (void) (priv); } while (0)
709 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
712 #define IPOIB_GID_FMT "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:" \
713 "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x"
715 #define IPOIB_GID_RAW_ARG(gid) ((u8 *)(gid))[0], \
732 #define IPOIB_GID_ARG(gid) IPOIB_GID_RAW_ARG((gid).raw)
734 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
736 #endif /* _IPOIB_H */