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 */
136 * Quoting 10.3.1 Queue Pair and EE Context States:
138 * Note, for QPs that are associated with an SRQ, the Consumer should take the
139 * QP through the Error State before invoking a Destroy QP or a Modify QP to the
140 * Reset State. The Consumer may invoke the Destroy QP without first performing
141 * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
142 * Last WQE Reached Event. However, if the Consumer does not wait for the
143 * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
144 * leakage may occur. Therefore, it is good programming practice to tear down a
145 * QP that is associated with an SRQ by using the following process:
147 * - Put the QP in the Error State
148 * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
150 * drain the CQ by invoking the Poll CQ verb and either wait for CQ
151 * to be empty or the number of Poll CQ operations has exceeded
154 * post another WR that completes on the same CQ and wait for this
155 * WR to return as a WC;
156 * - and then invoke a Destroy QP or Reset QP.
158 * We use the second option and wait for a completion on the
159 * rx_drain_qp before destroying QPs attached to our SRQ.
162 enum ipoib_cm_state {
164 IPOIB_CM_RX_ERROR, /* Ignored by stale task */
165 IPOIB_CM_RX_FLUSH /* Last WQE Reached event observed */
171 struct list_head list;
172 struct net_device *dev;
173 unsigned long jiffies;
174 enum ipoib_cm_state state;
181 struct list_head list;
182 struct net_device *dev;
183 struct ipoib_neigh *neigh;
184 struct ipoib_path *path;
185 struct ipoib_tx_buf *tx_ring;
190 struct ib_wc ibwc[IPOIB_NUM_WC];
193 struct ipoib_cm_rx_buf {
195 u64 mapping[IPOIB_CM_RX_SG];
198 struct ipoib_cm_dev_priv {
200 struct ipoib_cm_rx_buf *srq_ring;
202 struct ib_qp *rx_drain_qp; /* generates WR described in 10.3.1 */
203 struct list_head passive_ids; /* state: LIVE */
204 struct list_head rx_error_list; /* state: ERROR */
205 struct list_head rx_flush_list; /* state: FLUSH, drain not started */
206 struct list_head rx_drain_list; /* state: FLUSH, drain started */
207 struct list_head rx_reap_list; /* state: FLUSH, drain done */
208 struct work_struct start_task;
209 struct work_struct reap_task;
210 struct work_struct skb_task;
211 struct work_struct rx_reap_task;
212 struct delayed_work stale_task;
213 struct sk_buff_head skb_queue;
214 struct list_head start_list;
215 struct list_head reap_list;
216 struct ib_wc ibwc[IPOIB_NUM_WC];
217 struct ib_sge rx_sge[IPOIB_CM_RX_SG];
218 struct ib_recv_wr rx_wr;
222 * Device private locking: tx_lock protects members used in TX fast
223 * path (and we use LLTX so upper layers don't do extra locking).
224 * lock protects everything else. lock nests inside of tx_lock (ie
225 * tx_lock must be acquired first if needed).
227 struct ipoib_dev_priv {
230 struct net_device *dev;
234 struct mutex mcast_mutex;
235 struct mutex vlan_mutex;
237 struct rb_root path_tree;
238 struct list_head path_list;
240 struct ipoib_mcast *broadcast;
241 struct list_head multicast_list;
242 struct rb_root multicast_tree;
244 struct delayed_work pkey_poll_task;
245 struct delayed_work mcast_task;
246 struct work_struct flush_task;
247 struct work_struct restart_task;
248 struct delayed_work ah_reap_task;
249 struct work_struct pkey_event_task;
251 struct ib_device *ca;
261 union ib_gid local_gid;
264 unsigned int admin_mtu;
265 unsigned int mcast_mtu;
267 struct ipoib_rx_buf *rx_ring;
270 struct ipoib_tx_buf *tx_ring;
273 struct ib_sge tx_sge;
274 struct ib_send_wr tx_wr;
276 struct ib_wc ibwc[IPOIB_NUM_WC];
278 struct list_head dead_ahs;
280 struct ib_event_handler event_handler;
282 struct net_device_stats stats;
284 struct net_device *parent;
285 struct list_head child_intfs;
286 struct list_head list;
288 #ifdef CONFIG_INFINIBAND_IPOIB_CM
289 struct ipoib_cm_dev_priv cm;
292 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
293 struct list_head fs_list;
294 struct dentry *mcg_dentry;
295 struct dentry *path_dentry;
300 struct net_device *dev;
302 struct list_head list;
308 struct net_device *dev;
309 struct ib_sa_path_rec pathrec;
311 struct sk_buff_head queue;
313 struct list_head neigh_list;
316 struct ib_sa_query *query;
317 struct completion done;
319 struct rb_node rb_node;
320 struct list_head list;
325 #ifdef CONFIG_INFINIBAND_IPOIB_CM
326 struct ipoib_cm_tx *cm;
329 struct sk_buff_head queue;
331 struct neighbour *neighbour;
333 struct list_head list;
337 * We stash a pointer to our private neighbour information after our
338 * hardware address in neigh->ha. The ALIGN() expression here makes
339 * sure that this pointer is stored aligned so that an unaligned
340 * load is not needed to dereference it.
342 static inline struct ipoib_neigh **to_ipoib_neigh(struct neighbour *neigh)
344 return (void*) neigh + ALIGN(offsetof(struct neighbour, ha) +
345 INFINIBAND_ALEN, sizeof(void *));
348 struct ipoib_neigh *ipoib_neigh_alloc(struct neighbour *neigh);
349 void ipoib_neigh_free(struct net_device *dev, struct ipoib_neigh *neigh);
351 extern struct workqueue_struct *ipoib_workqueue;
355 int ipoib_poll(struct net_device *dev, int *budget);
356 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr);
358 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
359 struct ib_pd *pd, struct ib_ah_attr *attr);
360 void ipoib_free_ah(struct kref *kref);
361 static inline void ipoib_put_ah(struct ipoib_ah *ah)
363 kref_put(&ah->ref, ipoib_free_ah);
366 int ipoib_open(struct net_device *dev);
367 int ipoib_add_pkey_attr(struct net_device *dev);
369 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
370 struct ipoib_ah *address, u32 qpn);
371 void ipoib_reap_ah(struct work_struct *work);
373 void ipoib_flush_paths(struct net_device *dev);
374 struct ipoib_dev_priv *ipoib_intf_alloc(const char *format);
376 int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
377 void ipoib_ib_dev_flush(struct work_struct *work);
378 void ipoib_pkey_event(struct work_struct *work);
379 void ipoib_ib_dev_cleanup(struct net_device *dev);
381 int ipoib_ib_dev_open(struct net_device *dev);
382 int ipoib_ib_dev_up(struct net_device *dev);
383 int ipoib_ib_dev_down(struct net_device *dev, int flush);
384 int ipoib_ib_dev_stop(struct net_device *dev, int flush);
386 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
387 void ipoib_dev_cleanup(struct net_device *dev);
389 void ipoib_mcast_join_task(struct work_struct *work);
390 void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb);
392 void ipoib_mcast_restart_task(struct work_struct *work);
393 int ipoib_mcast_start_thread(struct net_device *dev);
394 int ipoib_mcast_stop_thread(struct net_device *dev, int flush);
396 void ipoib_mcast_dev_down(struct net_device *dev);
397 void ipoib_mcast_dev_flush(struct net_device *dev);
399 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
400 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev);
401 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter);
402 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
404 unsigned long *created,
405 unsigned int *queuelen,
406 unsigned int *complete,
407 unsigned int *send_only);
409 struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev);
410 int ipoib_path_iter_next(struct ipoib_path_iter *iter);
411 void ipoib_path_iter_read(struct ipoib_path_iter *iter,
412 struct ipoib_path *path);
415 int ipoib_mcast_attach(struct net_device *dev, u16 mlid,
417 int ipoib_mcast_detach(struct net_device *dev, u16 mlid,
420 int ipoib_init_qp(struct net_device *dev);
421 int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca);
422 void ipoib_transport_dev_cleanup(struct net_device *dev);
424 void ipoib_event(struct ib_event_handler *handler,
425 struct ib_event *record);
427 int ipoib_vlan_add(struct net_device *pdev, unsigned short pkey);
428 int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey);
430 void ipoib_pkey_poll(struct work_struct *work);
431 int ipoib_pkey_dev_delay_open(struct net_device *dev);
432 void ipoib_drain_cq(struct net_device *dev);
434 #ifdef CONFIG_INFINIBAND_IPOIB_CM
436 #define IPOIB_FLAGS_RC 0x80
437 #define IPOIB_FLAGS_UC 0x40
439 /* We don't support UC connections at the moment */
440 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
442 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
444 struct ipoib_dev_priv *priv = netdev_priv(dev);
445 return IPOIB_CM_SUPPORTED(dev->dev_addr) &&
446 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
449 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
451 struct ipoib_dev_priv *priv = netdev_priv(dev);
452 return IPOIB_CM_SUPPORTED(n->ha) &&
453 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
456 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
459 return test_bit(IPOIB_FLAG_OPER_UP, &neigh->cm->flags);
462 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
467 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
472 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx);
473 int ipoib_cm_dev_open(struct net_device *dev);
474 void ipoib_cm_dev_stop(struct net_device *dev);
475 int ipoib_cm_dev_init(struct net_device *dev);
476 int ipoib_cm_add_mode_attr(struct net_device *dev);
477 void ipoib_cm_dev_cleanup(struct net_device *dev);
478 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
479 struct ipoib_neigh *neigh);
480 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx);
481 void ipoib_cm_skb_too_long(struct net_device* dev, struct sk_buff *skb,
483 void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc);
488 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
492 static inline int ipoib_cm_enabled(struct net_device *dev, struct neighbour *n)
498 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
504 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
509 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
514 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx)
520 int ipoib_cm_dev_open(struct net_device *dev)
526 void ipoib_cm_dev_stop(struct net_device *dev)
532 int ipoib_cm_dev_init(struct net_device *dev)
538 void ipoib_cm_dev_cleanup(struct net_device *dev)
544 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
545 struct ipoib_neigh *neigh)
551 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx)
557 int ipoib_cm_add_mode_attr(struct net_device *dev)
562 static inline void ipoib_cm_skb_too_long(struct net_device* dev, struct sk_buff *skb,
565 dev_kfree_skb_any(skb);
568 static inline void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
574 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
575 void ipoib_create_debug_files(struct net_device *dev);
576 void ipoib_delete_debug_files(struct net_device *dev);
577 int ipoib_register_debugfs(void);
578 void ipoib_unregister_debugfs(void);
580 static inline void ipoib_create_debug_files(struct net_device *dev) { }
581 static inline void ipoib_delete_debug_files(struct net_device *dev) { }
582 static inline int ipoib_register_debugfs(void) { return 0; }
583 static inline void ipoib_unregister_debugfs(void) { }
587 #define ipoib_printk(level, priv, format, arg...) \
588 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
589 #define ipoib_warn(priv, format, arg...) \
590 ipoib_printk(KERN_WARNING, priv, format , ## arg)
592 extern int ipoib_sendq_size;
593 extern int ipoib_recvq_size;
595 extern struct ib_sa_client ipoib_sa_client;
597 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
598 extern int ipoib_debug_level;
600 #define ipoib_dbg(priv, format, arg...) \
602 if (ipoib_debug_level > 0) \
603 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
605 #define ipoib_dbg_mcast(priv, format, arg...) \
607 if (mcast_debug_level > 0) \
608 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
610 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
611 #define ipoib_dbg(priv, format, arg...) \
612 do { (void) (priv); } while (0)
613 #define ipoib_dbg_mcast(priv, format, arg...) \
614 do { (void) (priv); } while (0)
615 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
617 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
618 #define ipoib_dbg_data(priv, format, arg...) \
620 if (data_debug_level > 0) \
621 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
623 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
624 #define ipoib_dbg_data(priv, format, arg...) \
625 do { (void) (priv); } while (0)
626 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
629 #define IPOIB_GID_FMT "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:" \
630 "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x"
632 #define IPOIB_GID_RAW_ARG(gid) ((u8 *)(gid))[0], \
649 #define IPOIB_GID_ARG(gid) IPOIB_GID_RAW_ARG((gid).raw)
651 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
653 #endif /* _IPOIB_H */