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,
75 IPOIB_MAX_PATH_REC_QUEUE = 3,
76 IPOIB_MAX_MCAST_QUEUE = 3,
78 IPOIB_FLAG_OPER_UP = 0,
79 IPOIB_FLAG_INITIALIZED = 1,
80 IPOIB_FLAG_ADMIN_UP = 2,
81 IPOIB_PKEY_ASSIGNED = 3,
83 IPOIB_FLAG_SUBINTERFACE = 5,
85 IPOIB_STOP_REAPER = 7,
86 IPOIB_MCAST_STARTED = 8,
87 IPOIB_FLAG_NETIF_STOPPED = 9,
88 IPOIB_FLAG_ADMIN_CM = 10,
89 IPOIB_FLAG_UMCAST = 11,
91 IPOIB_MAX_BACKOFF_SECONDS = 16,
93 IPOIB_MCAST_FLAG_FOUND = 0, /* used in set_multicast_list */
94 IPOIB_MCAST_FLAG_SENDONLY = 1,
95 IPOIB_MCAST_FLAG_BUSY = 2, /* joining or already joined */
96 IPOIB_MCAST_FLAG_ATTACHED = 3,
99 #define IPOIB_OP_RECV (1ul << 31)
100 #ifdef CONFIG_INFINIBAND_IPOIB_CM
101 #define IPOIB_CM_OP_SRQ (1ul << 30)
103 #define IPOIB_CM_OP_SRQ (0)
108 struct ipoib_header {
113 struct ipoib_pseudoheader {
114 u8 hwaddr[INFINIBAND_ALEN];
117 /* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
119 struct ib_sa_mcmember_rec mcmember;
120 struct ib_sa_multicast *mc;
123 struct rb_node rb_node;
124 struct list_head list;
126 unsigned long created;
127 unsigned long backoff;
130 unsigned char logcount;
132 struct list_head neigh_list;
134 struct sk_buff_head pkt_queue;
136 struct net_device *dev;
139 struct ipoib_rx_buf {
144 struct ipoib_tx_buf {
151 struct ipoib_cm_data {
152 __be32 qpn; /* High byte MUST be ignored on receive */
157 * Quoting 10.3.1 Queue Pair and EE Context States:
159 * Note, for QPs that are associated with an SRQ, the Consumer should take the
160 * QP through the Error State before invoking a Destroy QP or a Modify QP to the
161 * Reset State. The Consumer may invoke the Destroy QP without first performing
162 * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
163 * Last WQE Reached Event. However, if the Consumer does not wait for the
164 * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
165 * leakage may occur. Therefore, it is good programming practice to tear down a
166 * QP that is associated with an SRQ by using the following process:
168 * - Put the QP in the Error State
169 * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
171 * drain the CQ by invoking the Poll CQ verb and either wait for CQ
172 * to be empty or the number of Poll CQ operations has exceeded
175 * post another WR that completes on the same CQ and wait for this
176 * WR to return as a WC;
177 * - and then invoke a Destroy QP or Reset QP.
179 * We use the second option and wait for a completion on the
180 * same CQ before destroying QPs attached to our SRQ.
183 enum ipoib_cm_state {
185 IPOIB_CM_RX_ERROR, /* Ignored by stale task */
186 IPOIB_CM_RX_FLUSH /* Last WQE Reached event observed */
192 struct list_head list;
193 struct net_device *dev;
194 unsigned long jiffies;
195 enum ipoib_cm_state state;
202 struct list_head list;
203 struct net_device *dev;
204 struct ipoib_neigh *neigh;
205 struct ipoib_path *path;
206 struct ipoib_tx_buf *tx_ring;
211 struct ib_wc ibwc[IPOIB_NUM_WC];
214 struct ipoib_cm_rx_buf {
216 u64 mapping[IPOIB_CM_RX_SG];
219 struct ipoib_cm_dev_priv {
221 struct ipoib_cm_rx_buf *srq_ring;
223 struct list_head passive_ids; /* state: LIVE */
224 struct list_head rx_error_list; /* state: ERROR */
225 struct list_head rx_flush_list; /* state: FLUSH, drain not started */
226 struct list_head rx_drain_list; /* state: FLUSH, drain started */
227 struct list_head rx_reap_list; /* state: FLUSH, drain done */
228 struct work_struct start_task;
229 struct work_struct reap_task;
230 struct work_struct skb_task;
231 struct work_struct rx_reap_task;
232 struct delayed_work stale_task;
233 struct sk_buff_head skb_queue;
234 struct list_head start_list;
235 struct list_head reap_list;
236 struct ib_wc ibwc[IPOIB_NUM_WC];
237 struct ib_sge rx_sge[IPOIB_CM_RX_SG];
238 struct ib_recv_wr rx_wr;
242 * Device private locking: tx_lock protects members used in TX fast
243 * path (and we use LLTX so upper layers don't do extra locking).
244 * lock protects everything else. lock nests inside of tx_lock (ie
245 * tx_lock must be acquired first if needed).
247 struct ipoib_dev_priv {
250 struct net_device *dev;
254 struct mutex mcast_mutex;
255 struct mutex vlan_mutex;
257 struct rb_root path_tree;
258 struct list_head path_list;
260 struct ipoib_mcast *broadcast;
261 struct list_head multicast_list;
262 struct rb_root multicast_tree;
264 struct delayed_work pkey_poll_task;
265 struct delayed_work mcast_task;
266 struct work_struct flush_task;
267 struct work_struct restart_task;
268 struct delayed_work ah_reap_task;
269 struct work_struct pkey_event_task;
271 struct ib_device *ca;
281 union ib_gid local_gid;
284 unsigned int admin_mtu;
285 unsigned int mcast_mtu;
287 struct ipoib_rx_buf *rx_ring;
290 struct ipoib_tx_buf *tx_ring;
293 struct ib_sge tx_sge;
294 struct ib_send_wr tx_wr;
296 struct ib_wc ibwc[IPOIB_NUM_WC];
298 struct list_head dead_ahs;
300 struct ib_event_handler event_handler;
302 struct net_device_stats stats;
304 struct net_device *parent;
305 struct list_head child_intfs;
306 struct list_head list;
308 #ifdef CONFIG_INFINIBAND_IPOIB_CM
309 struct ipoib_cm_dev_priv cm;
312 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
313 struct list_head fs_list;
314 struct dentry *mcg_dentry;
315 struct dentry *path_dentry;
320 struct net_device *dev;
322 struct list_head list;
328 struct net_device *dev;
329 struct ib_sa_path_rec pathrec;
331 struct sk_buff_head queue;
333 struct list_head neigh_list;
336 struct ib_sa_query *query;
337 struct completion done;
339 struct rb_node rb_node;
340 struct list_head list;
345 #ifdef CONFIG_INFINIBAND_IPOIB_CM
346 struct ipoib_cm_tx *cm;
349 struct sk_buff_head queue;
351 struct neighbour *neighbour;
353 struct list_head list;
357 * We stash a pointer to our private neighbour information after our
358 * hardware address in neigh->ha. The ALIGN() expression here makes
359 * sure that this pointer is stored aligned so that an unaligned
360 * load is not needed to dereference it.
362 static inline struct ipoib_neigh **to_ipoib_neigh(struct neighbour *neigh)
364 return (void*) neigh + ALIGN(offsetof(struct neighbour, ha) +
365 INFINIBAND_ALEN, sizeof(void *));
368 struct ipoib_neigh *ipoib_neigh_alloc(struct neighbour *neigh);
369 void ipoib_neigh_free(struct net_device *dev, struct ipoib_neigh *neigh);
371 extern struct workqueue_struct *ipoib_workqueue;
375 int ipoib_poll(struct net_device *dev, int *budget);
376 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr);
378 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
379 struct ib_pd *pd, struct ib_ah_attr *attr);
380 void ipoib_free_ah(struct kref *kref);
381 static inline void ipoib_put_ah(struct ipoib_ah *ah)
383 kref_put(&ah->ref, ipoib_free_ah);
386 int ipoib_open(struct net_device *dev);
387 int ipoib_add_pkey_attr(struct net_device *dev);
388 int ipoib_add_umcast_attr(struct net_device *dev);
390 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
391 struct ipoib_ah *address, u32 qpn);
392 void ipoib_reap_ah(struct work_struct *work);
394 void ipoib_flush_paths(struct net_device *dev);
395 struct ipoib_dev_priv *ipoib_intf_alloc(const char *format);
397 int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
398 void ipoib_ib_dev_flush(struct work_struct *work);
399 void ipoib_pkey_event(struct work_struct *work);
400 void ipoib_ib_dev_cleanup(struct net_device *dev);
402 int ipoib_ib_dev_open(struct net_device *dev);
403 int ipoib_ib_dev_up(struct net_device *dev);
404 int ipoib_ib_dev_down(struct net_device *dev, int flush);
405 int ipoib_ib_dev_stop(struct net_device *dev, int flush);
407 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
408 void ipoib_dev_cleanup(struct net_device *dev);
410 void ipoib_mcast_join_task(struct work_struct *work);
411 void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb);
413 void ipoib_mcast_restart_task(struct work_struct *work);
414 int ipoib_mcast_start_thread(struct net_device *dev);
415 int ipoib_mcast_stop_thread(struct net_device *dev, int flush);
417 void ipoib_mcast_dev_down(struct net_device *dev);
418 void ipoib_mcast_dev_flush(struct net_device *dev);
420 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
421 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev);
422 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter);
423 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
425 unsigned long *created,
426 unsigned int *queuelen,
427 unsigned int *complete,
428 unsigned int *send_only);
430 struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev);
431 int ipoib_path_iter_next(struct ipoib_path_iter *iter);
432 void ipoib_path_iter_read(struct ipoib_path_iter *iter,
433 struct ipoib_path *path);
436 int ipoib_mcast_attach(struct net_device *dev, u16 mlid,
438 int ipoib_mcast_detach(struct net_device *dev, u16 mlid,
441 int ipoib_init_qp(struct net_device *dev);
442 int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca);
443 void ipoib_transport_dev_cleanup(struct net_device *dev);
445 void ipoib_event(struct ib_event_handler *handler,
446 struct ib_event *record);
448 int ipoib_vlan_add(struct net_device *pdev, unsigned short pkey);
449 int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey);
451 void ipoib_pkey_poll(struct work_struct *work);
452 int ipoib_pkey_dev_delay_open(struct net_device *dev);
453 void ipoib_drain_cq(struct net_device *dev);
455 #ifdef CONFIG_INFINIBAND_IPOIB_CM
457 #define IPOIB_FLAGS_RC 0x80
458 #define IPOIB_FLAGS_UC 0x40
460 /* We don't support UC connections at the moment */
461 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
463 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
465 struct ipoib_dev_priv *priv = netdev_priv(dev);
466 return IPOIB_CM_SUPPORTED(dev->dev_addr) &&
467 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
470 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
472 struct ipoib_dev_priv *priv = netdev_priv(dev);
473 return IPOIB_CM_SUPPORTED(n->ha) &&
474 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
477 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
480 return test_bit(IPOIB_FLAG_OPER_UP, &neigh->cm->flags);
483 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
488 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
493 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx);
494 int ipoib_cm_dev_open(struct net_device *dev);
495 void ipoib_cm_dev_stop(struct net_device *dev);
496 int ipoib_cm_dev_init(struct net_device *dev);
497 int ipoib_cm_add_mode_attr(struct net_device *dev);
498 void ipoib_cm_dev_cleanup(struct net_device *dev);
499 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
500 struct ipoib_neigh *neigh);
501 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx);
502 void ipoib_cm_skb_too_long(struct net_device* dev, struct sk_buff *skb,
504 void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc);
509 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
513 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
519 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
525 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
530 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
535 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)
547 void ipoib_cm_dev_stop(struct net_device *dev)
553 int ipoib_cm_dev_init(struct net_device *dev)
559 void ipoib_cm_dev_cleanup(struct net_device *dev)
565 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
566 struct ipoib_neigh *neigh)
572 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx)
578 int ipoib_cm_add_mode_attr(struct net_device *dev)
583 static inline void ipoib_cm_skb_too_long(struct net_device* dev, struct sk_buff *skb,
586 dev_kfree_skb_any(skb);
589 static inline void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
595 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
596 void ipoib_create_debug_files(struct net_device *dev);
597 void ipoib_delete_debug_files(struct net_device *dev);
598 int ipoib_register_debugfs(void);
599 void ipoib_unregister_debugfs(void);
601 static inline void ipoib_create_debug_files(struct net_device *dev) { }
602 static inline void ipoib_delete_debug_files(struct net_device *dev) { }
603 static inline int ipoib_register_debugfs(void) { return 0; }
604 static inline void ipoib_unregister_debugfs(void) { }
608 #define ipoib_printk(level, priv, format, arg...) \
609 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
610 #define ipoib_warn(priv, format, arg...) \
611 ipoib_printk(KERN_WARNING, priv, format , ## arg)
613 extern int ipoib_sendq_size;
614 extern int ipoib_recvq_size;
616 extern struct ib_sa_client ipoib_sa_client;
618 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
619 extern int ipoib_debug_level;
621 #define ipoib_dbg(priv, format, arg...) \
623 if (ipoib_debug_level > 0) \
624 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
626 #define ipoib_dbg_mcast(priv, format, arg...) \
628 if (mcast_debug_level > 0) \
629 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
631 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
632 #define ipoib_dbg(priv, format, arg...) \
633 do { (void) (priv); } while (0)
634 #define ipoib_dbg_mcast(priv, format, arg...) \
635 do { (void) (priv); } while (0)
636 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
638 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
639 #define ipoib_dbg_data(priv, format, arg...) \
641 if (data_debug_level > 0) \
642 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
644 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
645 #define ipoib_dbg_data(priv, format, arg...) \
646 do { (void) (priv); } while (0)
647 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
650 #define IPOIB_GID_FMT "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:" \
651 "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x"
653 #define IPOIB_GID_RAW_ARG(gid) ((u8 *)(gid))[0], \
670 #define IPOIB_GID_ARG(gid) IPOIB_GID_RAW_ARG((gid).raw)
672 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
674 #endif /* _IPOIB_H */