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/pci.h>
45 #include <linux/kref.h>
46 #include <linux/if_infiniband.h>
47 #include <linux/mutex.h>
49 #include <net/neighbour.h>
51 #include <asm/atomic.h>
53 #include <rdma/ib_verbs.h>
54 #include <rdma/ib_pack.h>
55 #include <rdma/ib_sa.h>
60 IPOIB_PACKET_SIZE = 2048,
61 IPOIB_BUF_SIZE = IPOIB_PACKET_SIZE + IB_GRH_BYTES,
65 IPOIB_CM_MTU = 0x10000 - 0x10, /* padding to align header to 16 */
66 IPOIB_CM_BUF_SIZE = IPOIB_CM_MTU + IPOIB_ENCAP_LEN,
67 IPOIB_CM_HEAD_SIZE = IPOIB_CM_BUF_SIZE % PAGE_SIZE,
68 IPOIB_CM_RX_SG = ALIGN(IPOIB_CM_BUF_SIZE, PAGE_SIZE) / PAGE_SIZE,
69 IPOIB_RX_RING_SIZE = 128,
70 IPOIB_TX_RING_SIZE = 64,
71 IPOIB_MAX_QUEUE_SIZE = 8192,
72 IPOIB_MIN_QUEUE_SIZE = 2,
76 IPOIB_MAX_PATH_REC_QUEUE = 3,
77 IPOIB_MAX_MCAST_QUEUE = 3,
79 IPOIB_FLAG_OPER_UP = 0,
80 IPOIB_FLAG_INITIALIZED = 1,
81 IPOIB_FLAG_ADMIN_UP = 2,
82 IPOIB_PKEY_ASSIGNED = 3,
84 IPOIB_FLAG_SUBINTERFACE = 5,
86 IPOIB_STOP_REAPER = 7,
87 IPOIB_MCAST_STARTED = 8,
88 IPOIB_FLAG_NETIF_STOPPED = 9,
89 IPOIB_FLAG_ADMIN_CM = 10,
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];
119 struct ipoib_rx_buf {
124 struct ipoib_tx_buf {
131 struct ipoib_cm_data {
132 __be32 qpn; /* High byte MUST be ignored on receive */
139 struct list_head list;
140 struct net_device *dev;
141 unsigned long jiffies;
148 struct list_head list;
149 struct net_device *dev;
150 struct ipoib_neigh *neigh;
151 struct ipoib_path *path;
152 struct ipoib_tx_buf *tx_ring;
157 struct ib_wc ibwc[IPOIB_NUM_WC];
160 struct ipoib_cm_rx_buf {
162 u64 mapping[IPOIB_CM_RX_SG];
165 struct ipoib_cm_dev_priv {
167 struct ipoib_cm_rx_buf *srq_ring;
169 struct list_head passive_ids;
170 struct work_struct start_task;
171 struct work_struct reap_task;
172 struct work_struct skb_task;
173 struct delayed_work stale_task;
174 struct sk_buff_head skb_queue;
175 struct list_head start_list;
176 struct list_head reap_list;
177 struct ib_wc ibwc[IPOIB_NUM_WC];
178 struct ib_sge rx_sge[IPOIB_CM_RX_SG];
179 struct ib_recv_wr rx_wr;
183 * Device private locking: tx_lock protects members used in TX fast
184 * path (and we use LLTX so upper layers don't do extra locking).
185 * lock protects everything else. lock nests inside of tx_lock (ie
186 * tx_lock must be acquired first if needed).
188 struct ipoib_dev_priv {
191 struct net_device *dev;
195 struct mutex mcast_mutex;
196 struct mutex vlan_mutex;
198 struct rb_root path_tree;
199 struct list_head path_list;
201 struct ipoib_mcast *broadcast;
202 struct list_head multicast_list;
203 struct rb_root multicast_tree;
205 struct delayed_work pkey_task;
206 struct delayed_work mcast_task;
207 struct work_struct flush_task;
208 struct work_struct restart_task;
209 struct delayed_work ah_reap_task;
211 struct ib_device *ca;
220 union ib_gid local_gid;
223 unsigned int admin_mtu;
224 unsigned int mcast_mtu;
226 struct ipoib_rx_buf *rx_ring;
229 struct ipoib_tx_buf *tx_ring;
232 struct ib_sge tx_sge;
233 struct ib_send_wr tx_wr;
235 struct ib_wc ibwc[IPOIB_NUM_WC];
237 struct list_head dead_ahs;
239 struct ib_event_handler event_handler;
241 struct net_device_stats stats;
243 struct net_device *parent;
244 struct list_head child_intfs;
245 struct list_head list;
247 #ifdef CONFIG_INFINIBAND_IPOIB_CM
248 struct ipoib_cm_dev_priv cm;
251 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
252 struct list_head fs_list;
253 struct dentry *mcg_dentry;
254 struct dentry *path_dentry;
259 struct net_device *dev;
261 struct list_head list;
267 struct net_device *dev;
268 struct ib_sa_path_rec pathrec;
270 struct sk_buff_head queue;
272 struct list_head neigh_list;
275 struct ib_sa_query *query;
276 struct completion done;
278 struct rb_node rb_node;
279 struct list_head list;
284 #ifdef CONFIG_INFINIBAND_IPOIB_CM
285 struct ipoib_cm_tx *cm;
288 struct sk_buff_head queue;
290 struct neighbour *neighbour;
292 struct list_head list;
296 * We stash a pointer to our private neighbour information after our
297 * hardware address in neigh->ha. The ALIGN() expression here makes
298 * sure that this pointer is stored aligned so that an unaligned
299 * load is not needed to dereference it.
301 static inline struct ipoib_neigh **to_ipoib_neigh(struct neighbour *neigh)
303 return (void*) neigh + ALIGN(offsetof(struct neighbour, ha) +
304 INFINIBAND_ALEN, sizeof(void *));
307 struct ipoib_neigh *ipoib_neigh_alloc(struct neighbour *neigh);
308 void ipoib_neigh_free(struct net_device *dev, struct ipoib_neigh *neigh);
310 extern struct workqueue_struct *ipoib_workqueue;
314 int ipoib_poll(struct net_device *dev, int *budget);
315 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr);
317 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
318 struct ib_pd *pd, struct ib_ah_attr *attr);
319 void ipoib_free_ah(struct kref *kref);
320 static inline void ipoib_put_ah(struct ipoib_ah *ah)
322 kref_put(&ah->ref, ipoib_free_ah);
325 int ipoib_open(struct net_device *dev);
326 int ipoib_add_pkey_attr(struct net_device *dev);
328 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
329 struct ipoib_ah *address, u32 qpn);
330 void ipoib_reap_ah(struct work_struct *work);
332 void ipoib_flush_paths(struct net_device *dev);
333 struct ipoib_dev_priv *ipoib_intf_alloc(const char *format);
335 int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
336 void ipoib_ib_dev_flush(struct work_struct *work);
337 void ipoib_ib_dev_cleanup(struct net_device *dev);
339 int ipoib_ib_dev_open(struct net_device *dev);
340 int ipoib_ib_dev_up(struct net_device *dev);
341 int ipoib_ib_dev_down(struct net_device *dev, int flush);
342 int ipoib_ib_dev_stop(struct net_device *dev);
344 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
345 void ipoib_dev_cleanup(struct net_device *dev);
347 void ipoib_mcast_join_task(struct work_struct *work);
348 void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb);
350 void ipoib_mcast_restart_task(struct work_struct *work);
351 int ipoib_mcast_start_thread(struct net_device *dev);
352 int ipoib_mcast_stop_thread(struct net_device *dev, int flush);
354 void ipoib_mcast_dev_down(struct net_device *dev);
355 void ipoib_mcast_dev_flush(struct net_device *dev);
357 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
358 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev);
359 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter);
360 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
362 unsigned long *created,
363 unsigned int *queuelen,
364 unsigned int *complete,
365 unsigned int *send_only);
367 struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev);
368 int ipoib_path_iter_next(struct ipoib_path_iter *iter);
369 void ipoib_path_iter_read(struct ipoib_path_iter *iter,
370 struct ipoib_path *path);
373 int ipoib_mcast_attach(struct net_device *dev, u16 mlid,
375 int ipoib_mcast_detach(struct net_device *dev, u16 mlid,
378 int ipoib_init_qp(struct net_device *dev);
379 int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca);
380 void ipoib_transport_dev_cleanup(struct net_device *dev);
382 void ipoib_event(struct ib_event_handler *handler,
383 struct ib_event *record);
385 int ipoib_vlan_add(struct net_device *pdev, unsigned short pkey);
386 int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey);
388 void ipoib_pkey_poll(struct work_struct *work);
389 int ipoib_pkey_dev_delay_open(struct net_device *dev);
391 #ifdef CONFIG_INFINIBAND_IPOIB_CM
393 #define IPOIB_FLAGS_RC 0x80
394 #define IPOIB_FLAGS_UC 0x40
396 /* We don't support UC connections at the moment */
397 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
399 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
401 struct ipoib_dev_priv *priv = netdev_priv(dev);
402 return IPOIB_CM_SUPPORTED(dev->dev_addr) &&
403 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
406 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
408 struct ipoib_dev_priv *priv = netdev_priv(dev);
409 return IPOIB_CM_SUPPORTED(n->ha) &&
410 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
413 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
416 return test_bit(IPOIB_FLAG_OPER_UP, &neigh->cm->flags);
419 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
424 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
429 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx);
430 int ipoib_cm_dev_open(struct net_device *dev);
431 void ipoib_cm_dev_stop(struct net_device *dev);
432 int ipoib_cm_dev_init(struct net_device *dev);
433 int ipoib_cm_add_mode_attr(struct net_device *dev);
434 void ipoib_cm_dev_cleanup(struct net_device *dev);
435 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
436 struct ipoib_neigh *neigh);
437 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx);
438 void ipoib_cm_skb_too_long(struct net_device* dev, struct sk_buff *skb,
440 void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc);
445 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
449 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
455 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
461 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
466 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
471 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx)
477 int ipoib_cm_dev_open(struct net_device *dev)
483 void ipoib_cm_dev_stop(struct net_device *dev)
489 int ipoib_cm_dev_init(struct net_device *dev)
495 void ipoib_cm_dev_cleanup(struct net_device *dev)
501 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
502 struct ipoib_neigh *neigh)
508 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx)
514 int ipoib_cm_add_mode_attr(struct net_device *dev)
519 static inline void ipoib_cm_skb_too_long(struct net_device* dev, struct sk_buff *skb,
522 dev_kfree_skb_any(skb);
525 static inline void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
531 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
532 void ipoib_create_debug_files(struct net_device *dev);
533 void ipoib_delete_debug_files(struct net_device *dev);
534 int ipoib_register_debugfs(void);
535 void ipoib_unregister_debugfs(void);
537 static inline void ipoib_create_debug_files(struct net_device *dev) { }
538 static inline void ipoib_delete_debug_files(struct net_device *dev) { }
539 static inline int ipoib_register_debugfs(void) { return 0; }
540 static inline void ipoib_unregister_debugfs(void) { }
544 #define ipoib_printk(level, priv, format, arg...) \
545 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
546 #define ipoib_warn(priv, format, arg...) \
547 ipoib_printk(KERN_WARNING, priv, format , ## arg)
549 extern int ipoib_sendq_size;
550 extern int ipoib_recvq_size;
552 extern struct ib_sa_client ipoib_sa_client;
554 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
555 extern int ipoib_debug_level;
557 #define ipoib_dbg(priv, format, arg...) \
559 if (ipoib_debug_level > 0) \
560 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
562 #define ipoib_dbg_mcast(priv, format, arg...) \
564 if (mcast_debug_level > 0) \
565 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
567 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
568 #define ipoib_dbg(priv, format, arg...) \
569 do { (void) (priv); } while (0)
570 #define ipoib_dbg_mcast(priv, format, arg...) \
571 do { (void) (priv); } while (0)
572 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
574 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
575 #define ipoib_dbg_data(priv, format, arg...) \
577 if (data_debug_level > 0) \
578 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
580 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
581 #define ipoib_dbg_data(priv, format, arg...) \
582 do { (void) (priv); } while (0)
583 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
586 #define IPOIB_GID_FMT "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:" \
587 "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x"
589 #define IPOIB_GID_RAW_ARG(gid) ((u8 *)(gid))[0], \
606 #define IPOIB_GID_ARG(gid) IPOIB_GID_RAW_ARG((gid).raw)
608 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
610 #endif /* _IPOIB_H */