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,
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 #define IPOIB_OP_RECV (1ul << 31)
99 #ifdef CONFIG_INFINIBAND_IPOIB_CM
100 #define IPOIB_CM_OP_SRQ (1ul << 30)
102 #define IPOIB_CM_OP_SRQ (0)
107 struct ipoib_header {
112 struct ipoib_pseudoheader {
113 u8 hwaddr[INFINIBAND_ALEN];
118 struct ipoib_rx_buf {
123 struct ipoib_tx_buf {
130 struct ipoib_cm_data {
131 __be32 qpn; /* High byte MUST be ignored on receive */
138 struct list_head list;
139 struct net_device *dev;
140 unsigned long jiffies;
147 struct list_head list;
148 struct net_device *dev;
149 struct ipoib_neigh *neigh;
150 struct ipoib_path *path;
151 struct ipoib_tx_buf *tx_ring;
156 struct ib_wc ibwc[IPOIB_NUM_WC];
159 struct ipoib_cm_rx_buf {
161 u64 mapping[IPOIB_CM_RX_SG];
164 struct ipoib_cm_dev_priv {
166 struct ipoib_cm_rx_buf *srq_ring;
168 struct list_head passive_ids;
169 struct work_struct start_task;
170 struct work_struct reap_task;
171 struct work_struct skb_task;
172 struct delayed_work stale_task;
173 struct sk_buff_head skb_queue;
174 struct list_head start_list;
175 struct list_head reap_list;
176 struct ib_wc ibwc[IPOIB_NUM_WC];
177 struct ib_sge rx_sge[IPOIB_CM_RX_SG];
178 struct ib_recv_wr rx_wr;
182 * Device private locking: tx_lock protects members used in TX fast
183 * path (and we use LLTX so upper layers don't do extra locking).
184 * lock protects everything else. lock nests inside of tx_lock (ie
185 * tx_lock must be acquired first if needed).
187 struct ipoib_dev_priv {
190 struct net_device *dev;
194 struct mutex mcast_mutex;
195 struct mutex vlan_mutex;
197 struct rb_root path_tree;
198 struct list_head path_list;
200 struct ipoib_mcast *broadcast;
201 struct list_head multicast_list;
202 struct rb_root multicast_tree;
204 struct delayed_work pkey_task;
205 struct delayed_work mcast_task;
206 struct work_struct flush_task;
207 struct work_struct restart_task;
208 struct delayed_work ah_reap_task;
210 struct ib_device *ca;
219 union ib_gid local_gid;
222 unsigned int admin_mtu;
223 unsigned int mcast_mtu;
225 struct ipoib_rx_buf *rx_ring;
228 struct ipoib_tx_buf *tx_ring;
231 struct ib_sge tx_sge;
232 struct ib_send_wr tx_wr;
234 struct ib_wc ibwc[IPOIB_NUM_WC];
236 struct list_head dead_ahs;
238 struct ib_event_handler event_handler;
240 struct net_device_stats stats;
242 struct net_device *parent;
243 struct list_head child_intfs;
244 struct list_head list;
246 #ifdef CONFIG_INFINIBAND_IPOIB_CM
247 struct ipoib_cm_dev_priv cm;
250 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
251 struct list_head fs_list;
252 struct dentry *mcg_dentry;
253 struct dentry *path_dentry;
258 struct net_device *dev;
260 struct list_head list;
266 struct net_device *dev;
267 struct ib_sa_path_rec pathrec;
269 struct sk_buff_head queue;
271 struct list_head neigh_list;
274 struct ib_sa_query *query;
275 struct completion done;
277 struct rb_node rb_node;
278 struct list_head list;
283 #ifdef CONFIG_INFINIBAND_IPOIB_CM
284 struct ipoib_cm_tx *cm;
287 struct sk_buff_head queue;
289 struct neighbour *neighbour;
291 struct list_head list;
295 * We stash a pointer to our private neighbour information after our
296 * hardware address in neigh->ha. The ALIGN() expression here makes
297 * sure that this pointer is stored aligned so that an unaligned
298 * load is not needed to dereference it.
300 static inline struct ipoib_neigh **to_ipoib_neigh(struct neighbour *neigh)
302 return (void*) neigh + ALIGN(offsetof(struct neighbour, ha) +
303 INFINIBAND_ALEN, sizeof(void *));
306 struct ipoib_neigh *ipoib_neigh_alloc(struct neighbour *neigh);
307 void ipoib_neigh_free(struct net_device *dev, struct ipoib_neigh *neigh);
309 extern struct workqueue_struct *ipoib_workqueue;
313 int ipoib_poll(struct net_device *dev, int *budget);
314 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr);
316 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
317 struct ib_pd *pd, struct ib_ah_attr *attr);
318 void ipoib_free_ah(struct kref *kref);
319 static inline void ipoib_put_ah(struct ipoib_ah *ah)
321 kref_put(&ah->ref, ipoib_free_ah);
324 int ipoib_open(struct net_device *dev);
325 int ipoib_add_pkey_attr(struct net_device *dev);
327 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
328 struct ipoib_ah *address, u32 qpn);
329 void ipoib_reap_ah(struct work_struct *work);
331 void ipoib_flush_paths(struct net_device *dev);
332 struct ipoib_dev_priv *ipoib_intf_alloc(const char *format);
334 int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
335 void ipoib_ib_dev_flush(struct work_struct *work);
336 void ipoib_ib_dev_cleanup(struct net_device *dev);
338 int ipoib_ib_dev_open(struct net_device *dev);
339 int ipoib_ib_dev_up(struct net_device *dev);
340 int ipoib_ib_dev_down(struct net_device *dev, int flush);
341 int ipoib_ib_dev_stop(struct net_device *dev);
343 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
344 void ipoib_dev_cleanup(struct net_device *dev);
346 void ipoib_mcast_join_task(struct work_struct *work);
347 void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb);
349 void ipoib_mcast_restart_task(struct work_struct *work);
350 int ipoib_mcast_start_thread(struct net_device *dev);
351 int ipoib_mcast_stop_thread(struct net_device *dev, int flush);
353 void ipoib_mcast_dev_down(struct net_device *dev);
354 void ipoib_mcast_dev_flush(struct net_device *dev);
356 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
357 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev);
358 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter);
359 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
361 unsigned long *created,
362 unsigned int *queuelen,
363 unsigned int *complete,
364 unsigned int *send_only);
366 struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev);
367 int ipoib_path_iter_next(struct ipoib_path_iter *iter);
368 void ipoib_path_iter_read(struct ipoib_path_iter *iter,
369 struct ipoib_path *path);
372 int ipoib_mcast_attach(struct net_device *dev, u16 mlid,
374 int ipoib_mcast_detach(struct net_device *dev, u16 mlid,
377 int ipoib_init_qp(struct net_device *dev);
378 int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca);
379 void ipoib_transport_dev_cleanup(struct net_device *dev);
381 void ipoib_event(struct ib_event_handler *handler,
382 struct ib_event *record);
384 int ipoib_vlan_add(struct net_device *pdev, unsigned short pkey);
385 int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey);
387 void ipoib_pkey_poll(struct work_struct *work);
388 int ipoib_pkey_dev_delay_open(struct net_device *dev);
390 #ifdef CONFIG_INFINIBAND_IPOIB_CM
392 #define IPOIB_FLAGS_RC 0x80
393 #define IPOIB_FLAGS_UC 0x40
395 /* We don't support UC connections at the moment */
396 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
398 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
400 struct ipoib_dev_priv *priv = netdev_priv(dev);
401 return IPOIB_CM_SUPPORTED(dev->dev_addr) &&
402 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
405 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
407 struct ipoib_dev_priv *priv = netdev_priv(dev);
408 return IPOIB_CM_SUPPORTED(n->ha) &&
409 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
412 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
415 return test_bit(IPOIB_FLAG_OPER_UP, &neigh->cm->flags);
418 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
423 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
428 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx);
429 int ipoib_cm_dev_open(struct net_device *dev);
430 void ipoib_cm_dev_stop(struct net_device *dev);
431 int ipoib_cm_dev_init(struct net_device *dev);
432 int ipoib_cm_add_mode_attr(struct net_device *dev);
433 void ipoib_cm_dev_cleanup(struct net_device *dev);
434 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
435 struct ipoib_neigh *neigh);
436 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx);
437 void ipoib_cm_skb_too_long(struct net_device* dev, struct sk_buff *skb,
439 void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc);
444 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
448 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
454 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
460 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
465 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
470 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx)
476 int ipoib_cm_dev_open(struct net_device *dev)
482 void ipoib_cm_dev_stop(struct net_device *dev)
488 int ipoib_cm_dev_init(struct net_device *dev)
494 void ipoib_cm_dev_cleanup(struct net_device *dev)
500 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
501 struct ipoib_neigh *neigh)
507 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx)
513 int ipoib_cm_add_mode_attr(struct net_device *dev)
518 static inline void ipoib_cm_skb_too_long(struct net_device* dev, struct sk_buff *skb,
521 dev_kfree_skb_any(skb);
524 static inline void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
530 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
531 void ipoib_create_debug_files(struct net_device *dev);
532 void ipoib_delete_debug_files(struct net_device *dev);
533 int ipoib_register_debugfs(void);
534 void ipoib_unregister_debugfs(void);
536 static inline void ipoib_create_debug_files(struct net_device *dev) { }
537 static inline void ipoib_delete_debug_files(struct net_device *dev) { }
538 static inline int ipoib_register_debugfs(void) { return 0; }
539 static inline void ipoib_unregister_debugfs(void) { }
543 #define ipoib_printk(level, priv, format, arg...) \
544 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
545 #define ipoib_warn(priv, format, arg...) \
546 ipoib_printk(KERN_WARNING, priv, format , ## arg)
548 extern int ipoib_sendq_size;
549 extern int ipoib_recvq_size;
551 extern struct ib_sa_client ipoib_sa_client;
553 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
554 extern int ipoib_debug_level;
556 #define ipoib_dbg(priv, format, arg...) \
558 if (ipoib_debug_level > 0) \
559 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
561 #define ipoib_dbg_mcast(priv, format, arg...) \
563 if (mcast_debug_level > 0) \
564 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
566 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
567 #define ipoib_dbg(priv, format, arg...) \
568 do { (void) (priv); } while (0)
569 #define ipoib_dbg_mcast(priv, format, arg...) \
570 do { (void) (priv); } while (0)
571 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
573 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
574 #define ipoib_dbg_data(priv, format, arg...) \
576 if (data_debug_level > 0) \
577 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
579 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
580 #define ipoib_dbg_data(priv, format, arg...) \
581 do { (void) (priv); } while (0)
582 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
585 #define IPOIB_GID_FMT "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:" \
586 "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x"
588 #define IPOIB_GID_RAW_ARG(gid) ((u8 *)(gid))[0], \
605 #define IPOIB_GID_ARG(gid) IPOIB_GID_RAW_ARG((gid).raw)
607 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
609 #endif /* _IPOIB_H */