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>
53 #include <linux/inet_lro.h>
57 enum ipoib_flush_level {
66 IPOIB_UD_HEAD_SIZE = IB_GRH_BYTES + IPOIB_ENCAP_LEN,
67 IPOIB_UD_RX_SG = 2, /* max buffer needed for 4K mtu */
69 IPOIB_CM_MTU = 0x10000 - 0x10, /* padding to align header to 16 */
70 IPOIB_CM_BUF_SIZE = IPOIB_CM_MTU + IPOIB_ENCAP_LEN,
71 IPOIB_CM_HEAD_SIZE = IPOIB_CM_BUF_SIZE % PAGE_SIZE,
72 IPOIB_CM_RX_SG = ALIGN(IPOIB_CM_BUF_SIZE, PAGE_SIZE) / PAGE_SIZE,
73 IPOIB_RX_RING_SIZE = 256,
74 IPOIB_TX_RING_SIZE = 128,
75 IPOIB_MAX_QUEUE_SIZE = 8192,
76 IPOIB_MIN_QUEUE_SIZE = 2,
77 IPOIB_CM_MAX_CONN_QP = 4096,
81 IPOIB_MAX_PATH_REC_QUEUE = 3,
82 IPOIB_MAX_MCAST_QUEUE = 3,
84 IPOIB_FLAG_OPER_UP = 0,
85 IPOIB_FLAG_INITIALIZED = 1,
86 IPOIB_FLAG_ADMIN_UP = 2,
87 IPOIB_PKEY_ASSIGNED = 3,
89 IPOIB_FLAG_SUBINTERFACE = 5,
91 IPOIB_STOP_REAPER = 7,
92 IPOIB_FLAG_ADMIN_CM = 9,
93 IPOIB_FLAG_UMCAST = 10,
96 IPOIB_MAX_BACKOFF_SECONDS = 16,
98 IPOIB_MCAST_FLAG_FOUND = 0, /* used in set_multicast_list */
99 IPOIB_MCAST_FLAG_SENDONLY = 1,
100 IPOIB_MCAST_FLAG_BUSY = 2, /* joining or already joined */
101 IPOIB_MCAST_FLAG_ATTACHED = 3,
103 IPOIB_MAX_LRO_DESCRIPTORS = 8,
104 IPOIB_LRO_MAX_AGGR = 64,
107 IPOIB_CM_COPYBREAK = 256,
110 #define IPOIB_OP_RECV (1ul << 31)
111 #ifdef CONFIG_INFINIBAND_IPOIB_CM
112 #define IPOIB_OP_CM (1ul << 30)
114 #define IPOIB_OP_CM (0)
119 struct ipoib_header {
124 struct ipoib_pseudoheader {
125 u8 hwaddr[INFINIBAND_ALEN];
128 /* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
130 struct ib_sa_mcmember_rec mcmember;
131 struct ib_sa_multicast *mc;
134 struct rb_node rb_node;
135 struct list_head list;
137 unsigned long created;
138 unsigned long backoff;
141 unsigned char logcount;
143 struct list_head neigh_list;
145 struct sk_buff_head pkt_queue;
147 struct net_device *dev;
150 struct ipoib_rx_buf {
152 u64 mapping[IPOIB_UD_RX_SG];
155 struct ipoib_tx_buf {
157 u64 mapping[MAX_SKB_FRAGS + 1];
160 struct ipoib_cm_tx_buf {
167 struct ipoib_cm_data {
168 __be32 qpn; /* High byte MUST be ignored on receive */
173 * Quoting 10.3.1 Queue Pair and EE Context States:
175 * Note, for QPs that are associated with an SRQ, the Consumer should take the
176 * QP through the Error State before invoking a Destroy QP or a Modify QP to the
177 * Reset State. The Consumer may invoke the Destroy QP without first performing
178 * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
179 * Last WQE Reached Event. However, if the Consumer does not wait for the
180 * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
181 * leakage may occur. Therefore, it is good programming practice to tear down a
182 * QP that is associated with an SRQ by using the following process:
184 * - Put the QP in the Error State
185 * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
187 * drain the CQ by invoking the Poll CQ verb and either wait for CQ
188 * to be empty or the number of Poll CQ operations has exceeded
191 * post another WR that completes on the same CQ and wait for this
192 * WR to return as a WC;
193 * - and then invoke a Destroy QP or Reset QP.
195 * We use the second option and wait for a completion on the
196 * same CQ before destroying QPs attached to our SRQ.
199 enum ipoib_cm_state {
201 IPOIB_CM_RX_ERROR, /* Ignored by stale task */
202 IPOIB_CM_RX_FLUSH /* Last WQE Reached event observed */
208 struct ipoib_cm_rx_buf *rx_ring;
209 struct list_head list;
210 struct net_device *dev;
211 unsigned long jiffies;
212 enum ipoib_cm_state state;
219 struct list_head list;
220 struct net_device *dev;
221 struct ipoib_neigh *neigh;
222 struct ipoib_path *path;
223 struct ipoib_cm_tx_buf *tx_ring;
230 struct ipoib_cm_rx_buf {
232 u64 mapping[IPOIB_CM_RX_SG];
235 struct ipoib_cm_dev_priv {
237 struct ipoib_cm_rx_buf *srq_ring;
239 struct list_head passive_ids; /* state: LIVE */
240 struct list_head rx_error_list; /* state: ERROR */
241 struct list_head rx_flush_list; /* state: FLUSH, drain not started */
242 struct list_head rx_drain_list; /* state: FLUSH, drain started */
243 struct list_head rx_reap_list; /* state: FLUSH, drain done */
244 struct work_struct start_task;
245 struct work_struct reap_task;
246 struct work_struct skb_task;
247 struct work_struct rx_reap_task;
248 struct delayed_work stale_task;
249 struct sk_buff_head skb_queue;
250 struct list_head start_list;
251 struct list_head reap_list;
252 struct ib_wc ibwc[IPOIB_NUM_WC];
253 struct ib_sge rx_sge[IPOIB_CM_RX_SG];
254 struct ib_recv_wr rx_wr;
260 struct ipoib_ethtool_st {
262 u16 max_coalesced_frames;
266 struct net_lro_mgr lro_mgr;
267 struct net_lro_desc lro_desc[IPOIB_MAX_LRO_DESCRIPTORS];
271 * Device private locking: tx_lock protects members used in TX fast
272 * path (and we use LLTX so upper layers don't do extra locking).
273 * lock protects everything else. lock nests inside of tx_lock (ie
274 * tx_lock must be acquired first if needed).
276 struct ipoib_dev_priv {
279 struct net_device *dev;
281 struct napi_struct napi;
285 struct mutex vlan_mutex;
287 struct rb_root path_tree;
288 struct list_head path_list;
290 struct ipoib_mcast *broadcast;
291 struct list_head multicast_list;
292 struct rb_root multicast_tree;
294 struct delayed_work pkey_poll_task;
295 struct delayed_work mcast_task;
296 struct work_struct flush_light;
297 struct work_struct flush_normal;
298 struct work_struct flush_heavy;
299 struct work_struct restart_task;
300 struct delayed_work ah_reap_task;
302 struct ib_device *ca;
308 struct ib_cq *recv_cq;
309 struct ib_cq *send_cq;
313 union ib_gid local_gid;
316 unsigned int admin_mtu;
317 unsigned int mcast_mtu;
318 unsigned int max_ib_mtu;
320 struct ipoib_rx_buf *rx_ring;
323 struct ipoib_tx_buf *tx_ring;
326 struct ib_sge tx_sge[MAX_SKB_FRAGS + 1];
327 struct ib_send_wr tx_wr;
328 unsigned tx_outstanding;
329 struct ib_wc send_wc[MAX_SEND_CQE];
331 struct ib_recv_wr rx_wr;
332 struct ib_sge rx_sge[IPOIB_UD_RX_SG];
334 struct ib_wc ibwc[IPOIB_NUM_WC];
336 struct list_head dead_ahs;
338 struct ib_event_handler event_handler;
340 struct net_device *parent;
341 struct list_head child_intfs;
342 struct list_head list;
344 #ifdef CONFIG_INFINIBAND_IPOIB_CM
345 struct ipoib_cm_dev_priv cm;
348 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
349 struct list_head fs_list;
350 struct dentry *mcg_dentry;
351 struct dentry *path_dentry;
354 struct ipoib_ethtool_st ethtool;
355 struct timer_list poll_timer;
357 struct ipoib_lro lro;
361 struct net_device *dev;
363 struct list_head list;
369 struct net_device *dev;
370 struct ib_sa_path_rec pathrec;
372 struct sk_buff_head queue;
374 struct list_head neigh_list;
377 struct ib_sa_query *query;
378 struct completion done;
380 struct rb_node rb_node;
381 struct list_head list;
387 #ifdef CONFIG_INFINIBAND_IPOIB_CM
388 struct ipoib_cm_tx *cm;
391 struct sk_buff_head queue;
393 struct neighbour *neighbour;
394 struct net_device *dev;
396 struct list_head list;
399 #define IPOIB_UD_MTU(ib_mtu) (ib_mtu - IPOIB_ENCAP_LEN)
400 #define IPOIB_UD_BUF_SIZE(ib_mtu) (ib_mtu + IB_GRH_BYTES)
402 static inline int ipoib_ud_need_sg(unsigned int ib_mtu)
404 return IPOIB_UD_BUF_SIZE(ib_mtu) > PAGE_SIZE;
408 * We stash a pointer to our private neighbour information after our
409 * hardware address in neigh->ha. The ALIGN() expression here makes
410 * sure that this pointer is stored aligned so that an unaligned
411 * load is not needed to dereference it.
413 static inline struct ipoib_neigh **to_ipoib_neigh(struct neighbour *neigh)
415 return (void*) neigh + ALIGN(offsetof(struct neighbour, ha) +
416 INFINIBAND_ALEN, sizeof(void *));
419 struct ipoib_neigh *ipoib_neigh_alloc(struct neighbour *neigh,
420 struct net_device *dev);
421 void ipoib_neigh_free(struct net_device *dev, struct ipoib_neigh *neigh);
423 extern struct workqueue_struct *ipoib_workqueue;
427 int ipoib_poll(struct napi_struct *napi, int budget);
428 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr);
429 void ipoib_send_comp_handler(struct ib_cq *cq, void *dev_ptr);
431 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
432 struct ib_pd *pd, struct ib_ah_attr *attr);
433 void ipoib_free_ah(struct kref *kref);
434 static inline void ipoib_put_ah(struct ipoib_ah *ah)
436 kref_put(&ah->ref, ipoib_free_ah);
439 int ipoib_open(struct net_device *dev);
440 int ipoib_add_pkey_attr(struct net_device *dev);
441 int ipoib_add_umcast_attr(struct net_device *dev);
443 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
444 struct ipoib_ah *address, u32 qpn);
445 void ipoib_reap_ah(struct work_struct *work);
447 void ipoib_mark_paths_invalid(struct net_device *dev);
448 void ipoib_flush_paths(struct net_device *dev);
449 struct ipoib_dev_priv *ipoib_intf_alloc(const char *format);
451 int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
452 void ipoib_ib_dev_flush_light(struct work_struct *work);
453 void ipoib_ib_dev_flush_normal(struct work_struct *work);
454 void ipoib_ib_dev_flush_heavy(struct work_struct *work);
455 void ipoib_pkey_event(struct work_struct *work);
456 void ipoib_ib_dev_cleanup(struct net_device *dev);
458 int ipoib_ib_dev_open(struct net_device *dev);
459 int ipoib_ib_dev_up(struct net_device *dev);
460 int ipoib_ib_dev_down(struct net_device *dev, int flush);
461 int ipoib_ib_dev_stop(struct net_device *dev, int flush);
463 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
464 void ipoib_dev_cleanup(struct net_device *dev);
466 void ipoib_mcast_join_task(struct work_struct *work);
467 void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb);
469 void ipoib_mcast_restart_task(struct work_struct *work);
470 int ipoib_mcast_start_thread(struct net_device *dev);
471 int ipoib_mcast_stop_thread(struct net_device *dev, int flush);
473 void ipoib_mcast_dev_down(struct net_device *dev);
474 void ipoib_mcast_dev_flush(struct net_device *dev);
476 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
477 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev);
478 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter);
479 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
481 unsigned long *created,
482 unsigned int *queuelen,
483 unsigned int *complete,
484 unsigned int *send_only);
486 struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev);
487 int ipoib_path_iter_next(struct ipoib_path_iter *iter);
488 void ipoib_path_iter_read(struct ipoib_path_iter *iter,
489 struct ipoib_path *path);
492 int ipoib_mcast_attach(struct net_device *dev, u16 mlid,
493 union ib_gid *mgid, int set_qkey);
495 int ipoib_init_qp(struct net_device *dev);
496 int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca);
497 void ipoib_transport_dev_cleanup(struct net_device *dev);
499 void ipoib_event(struct ib_event_handler *handler,
500 struct ib_event *record);
502 int ipoib_vlan_add(struct net_device *pdev, unsigned short pkey);
503 int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey);
505 void ipoib_pkey_poll(struct work_struct *work);
506 int ipoib_pkey_dev_delay_open(struct net_device *dev);
507 void ipoib_drain_cq(struct net_device *dev);
509 void ipoib_set_ethtool_ops(struct net_device *dev);
511 #ifdef CONFIG_INFINIBAND_IPOIB_CM
513 #define IPOIB_FLAGS_RC 0x80
514 #define IPOIB_FLAGS_UC 0x40
516 /* We don't support UC connections at the moment */
517 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
519 extern int ipoib_max_conn_qp;
521 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
523 struct ipoib_dev_priv *priv = netdev_priv(dev);
524 return IPOIB_CM_SUPPORTED(dev->dev_addr) &&
525 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
528 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
530 struct ipoib_dev_priv *priv = netdev_priv(dev);
531 return IPOIB_CM_SUPPORTED(n->ha) &&
532 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
535 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
538 return test_bit(IPOIB_FLAG_OPER_UP, &neigh->cm->flags);
541 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
546 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
551 static inline int ipoib_cm_has_srq(struct net_device *dev)
553 struct ipoib_dev_priv *priv = netdev_priv(dev);
554 return !!priv->cm.srq;
557 static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
559 struct ipoib_dev_priv *priv = netdev_priv(dev);
560 return priv->cm.max_cm_mtu;
563 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx);
564 int ipoib_cm_dev_open(struct net_device *dev);
565 void ipoib_cm_dev_stop(struct net_device *dev);
566 int ipoib_cm_dev_init(struct net_device *dev);
567 int ipoib_cm_add_mode_attr(struct net_device *dev);
568 void ipoib_cm_dev_cleanup(struct net_device *dev);
569 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
570 struct ipoib_neigh *neigh);
571 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx);
572 void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
574 void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc);
575 void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc);
580 #define ipoib_max_conn_qp 0
582 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
586 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
592 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
598 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
603 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
607 static inline int ipoib_cm_has_srq(struct net_device *dev)
612 static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
618 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx)
624 int ipoib_cm_dev_open(struct net_device *dev)
630 void ipoib_cm_dev_stop(struct net_device *dev)
636 int ipoib_cm_dev_init(struct net_device *dev)
642 void ipoib_cm_dev_cleanup(struct net_device *dev)
648 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
649 struct ipoib_neigh *neigh)
655 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx)
661 int ipoib_cm_add_mode_attr(struct net_device *dev)
666 static inline void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
669 dev_kfree_skb_any(skb);
672 static inline void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
676 static inline void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
681 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
682 void ipoib_create_debug_files(struct net_device *dev);
683 void ipoib_delete_debug_files(struct net_device *dev);
684 int ipoib_register_debugfs(void);
685 void ipoib_unregister_debugfs(void);
687 static inline void ipoib_create_debug_files(struct net_device *dev) { }
688 static inline void ipoib_delete_debug_files(struct net_device *dev) { }
689 static inline int ipoib_register_debugfs(void) { return 0; }
690 static inline void ipoib_unregister_debugfs(void) { }
693 #define ipoib_printk(level, priv, format, arg...) \
694 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
695 #define ipoib_warn(priv, format, arg...) \
696 ipoib_printk(KERN_WARNING, priv, format , ## arg)
698 extern int ipoib_sendq_size;
699 extern int ipoib_recvq_size;
701 extern struct ib_sa_client ipoib_sa_client;
703 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
704 extern int ipoib_debug_level;
706 #define ipoib_dbg(priv, format, arg...) \
708 if (ipoib_debug_level > 0) \
709 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
711 #define ipoib_dbg_mcast(priv, format, arg...) \
713 if (mcast_debug_level > 0) \
714 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
716 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
717 #define ipoib_dbg(priv, format, arg...) \
718 do { (void) (priv); } while (0)
719 #define ipoib_dbg_mcast(priv, format, arg...) \
720 do { (void) (priv); } while (0)
721 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
723 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
724 #define ipoib_dbg_data(priv, format, arg...) \
726 if (data_debug_level > 0) \
727 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
729 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
730 #define ipoib_dbg_data(priv, format, arg...) \
731 do { (void) (priv); } while (0)
732 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
735 #define IPOIB_GID_FMT "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:" \
736 "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x"
738 #define IPOIB_GID_RAW_ARG(gid) ((u8 *)(gid))[0], \
755 #define IPOIB_GID_ARG(gid) IPOIB_GID_RAW_ARG((gid).raw)
757 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
759 #endif /* _IPOIB_H */