Loading...
1/*
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.
5 *
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:
11 *
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
14 * conditions are met:
15 *
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
18 * disclaimer.
19 *
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.
24 *
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
32 * SOFTWARE.
33 */
34
35#ifndef _IPOIB_H
36#define _IPOIB_H
37
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>
45
46#include <net/neighbour.h>
47#include <net/sch_generic.h>
48
49#include <linux/atomic.h>
50
51#include <rdma/ib_verbs.h>
52#include <rdma/ib_pack.h>
53#include <rdma/ib_sa.h>
54#include <linux/sched.h>
55/* constants */
56
57enum ipoib_flush_level {
58 IPOIB_FLUSH_LIGHT,
59 IPOIB_FLUSH_NORMAL,
60 IPOIB_FLUSH_HEAVY
61};
62
63enum {
64 IPOIB_ENCAP_LEN = 4,
65 IPOIB_PSEUDO_LEN = 20,
66 IPOIB_HARD_LEN = IPOIB_ENCAP_LEN + IPOIB_PSEUDO_LEN,
67
68 IPOIB_UD_HEAD_SIZE = IB_GRH_BYTES + IPOIB_ENCAP_LEN,
69 IPOIB_UD_RX_SG = 2, /* max buffer needed for 4K mtu */
70
71 IPOIB_CM_MTU = 0x10000 - 0x10, /* padding to align header to 16 */
72 IPOIB_CM_BUF_SIZE = IPOIB_CM_MTU + IPOIB_ENCAP_LEN,
73 IPOIB_CM_HEAD_SIZE = IPOIB_CM_BUF_SIZE % PAGE_SIZE,
74 IPOIB_CM_RX_SG = ALIGN(IPOIB_CM_BUF_SIZE, PAGE_SIZE) / PAGE_SIZE,
75 IPOIB_RX_RING_SIZE = 256,
76 IPOIB_TX_RING_SIZE = 128,
77 IPOIB_MAX_QUEUE_SIZE = 8192,
78 IPOIB_MIN_QUEUE_SIZE = 2,
79 IPOIB_CM_MAX_CONN_QP = 4096,
80
81 IPOIB_NUM_WC = 4,
82
83 IPOIB_MAX_PATH_REC_QUEUE = 3,
84 IPOIB_MAX_MCAST_QUEUE = 64,
85
86 IPOIB_FLAG_OPER_UP = 0,
87 IPOIB_FLAG_INITIALIZED = 1,
88 IPOIB_FLAG_ADMIN_UP = 2,
89 IPOIB_PKEY_ASSIGNED = 3,
90 IPOIB_FLAG_SUBINTERFACE = 5,
91 IPOIB_STOP_REAPER = 7,
92 IPOIB_FLAG_ADMIN_CM = 9,
93 IPOIB_FLAG_UMCAST = 10,
94 IPOIB_NEIGH_TBL_FLUSH = 12,
95 IPOIB_FLAG_DEV_ADDR_SET = 13,
96 IPOIB_FLAG_DEV_ADDR_CTRL = 14,
97
98 IPOIB_MAX_BACKOFF_SECONDS = 16,
99
100 IPOIB_MCAST_FLAG_FOUND = 0, /* used in set_multicast_list */
101 IPOIB_MCAST_FLAG_SENDONLY = 1,
102 /*
103 * For IPOIB_MCAST_FLAG_BUSY
104 * When set, in flight join and mcast->mc is unreliable
105 * When clear and mcast->mc IS_ERR_OR_NULL, need to restart or
106 * haven't started yet
107 * When clear and mcast->mc is valid pointer, join was successful
108 */
109 IPOIB_MCAST_FLAG_BUSY = 2,
110 IPOIB_MCAST_FLAG_ATTACHED = 3,
111
112 MAX_SEND_CQE = 64,
113 IPOIB_CM_COPYBREAK = 256,
114
115 IPOIB_NON_CHILD = 0,
116 IPOIB_LEGACY_CHILD = 1,
117 IPOIB_RTNL_CHILD = 2,
118};
119
120#define IPOIB_OP_RECV (1ul << 31)
121#ifdef CONFIG_INFINIBAND_IPOIB_CM
122#define IPOIB_OP_CM (1ul << 30)
123#else
124#define IPOIB_OP_CM (0)
125#endif
126
127#define IPOIB_QPN_MASK ((__force u32) cpu_to_be32(0xFFFFFF))
128
129/* structs */
130
131struct ipoib_header {
132 __be16 proto;
133 u16 reserved;
134};
135
136struct ipoib_pseudo_header {
137 u8 hwaddr[INFINIBAND_ALEN];
138};
139
140static inline void skb_add_pseudo_hdr(struct sk_buff *skb)
141{
142 char *data = skb_push(skb, IPOIB_PSEUDO_LEN);
143
144 /*
145 * only the ipoib header is present now, make room for a dummy
146 * pseudo header and set skb field accordingly
147 */
148 memset(data, 0, IPOIB_PSEUDO_LEN);
149 skb_reset_mac_header(skb);
150 skb_pull(skb, IPOIB_HARD_LEN);
151}
152
153static inline struct ipoib_dev_priv *ipoib_priv(const struct net_device *dev)
154{
155 struct rdma_netdev *rn = netdev_priv(dev);
156
157 return rn->clnt_priv;
158}
159
160/* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
161struct ipoib_mcast {
162 struct ib_sa_mcmember_rec mcmember;
163 struct ib_sa_multicast *mc;
164 struct ipoib_ah *ah;
165
166 struct rb_node rb_node;
167 struct list_head list;
168
169 unsigned long created;
170 unsigned long backoff;
171 unsigned long delay_until;
172
173 unsigned long flags;
174 unsigned char logcount;
175
176 struct list_head neigh_list;
177
178 struct sk_buff_head pkt_queue;
179
180 struct net_device *dev;
181 struct completion done;
182};
183
184struct ipoib_rx_buf {
185 struct sk_buff *skb;
186 u64 mapping[IPOIB_UD_RX_SG];
187};
188
189struct ipoib_tx_buf {
190 struct sk_buff *skb;
191 u64 mapping[MAX_SKB_FRAGS + 1];
192};
193
194struct ib_cm_id;
195
196struct ipoib_cm_data {
197 __be32 qpn; /* High byte MUST be ignored on receive */
198 __be32 mtu;
199};
200
201/*
202 * Quoting 10.3.1 Queue Pair and EE Context States:
203 *
204 * Note, for QPs that are associated with an SRQ, the Consumer should take the
205 * QP through the Error State before invoking a Destroy QP or a Modify QP to the
206 * Reset State. The Consumer may invoke the Destroy QP without first performing
207 * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
208 * Last WQE Reached Event. However, if the Consumer does not wait for the
209 * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
210 * leakage may occur. Therefore, it is good programming practice to tear down a
211 * QP that is associated with an SRQ by using the following process:
212 *
213 * - Put the QP in the Error State
214 * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
215 * - either:
216 * drain the CQ by invoking the Poll CQ verb and either wait for CQ
217 * to be empty or the number of Poll CQ operations has exceeded
218 * CQ capacity size;
219 * - or
220 * post another WR that completes on the same CQ and wait for this
221 * WR to return as a WC;
222 * - and then invoke a Destroy QP or Reset QP.
223 *
224 * We use the second option and wait for a completion on the
225 * same CQ before destroying QPs attached to our SRQ.
226 */
227
228enum ipoib_cm_state {
229 IPOIB_CM_RX_LIVE,
230 IPOIB_CM_RX_ERROR, /* Ignored by stale task */
231 IPOIB_CM_RX_FLUSH /* Last WQE Reached event observed */
232};
233
234struct ipoib_cm_rx {
235 struct ib_cm_id *id;
236 struct ib_qp *qp;
237 struct ipoib_cm_rx_buf *rx_ring;
238 struct list_head list;
239 struct net_device *dev;
240 unsigned long jiffies;
241 enum ipoib_cm_state state;
242 int recv_count;
243};
244
245struct ipoib_cm_tx {
246 struct ib_cm_id *id;
247 struct ib_qp *qp;
248 struct list_head list;
249 struct net_device *dev;
250 struct ipoib_neigh *neigh;
251 struct ipoib_tx_buf *tx_ring;
252 unsigned int tx_head;
253 unsigned int tx_tail;
254 unsigned long flags;
255 u32 mtu;
256 unsigned int max_send_sge;
257};
258
259struct ipoib_cm_rx_buf {
260 struct sk_buff *skb;
261 u64 mapping[IPOIB_CM_RX_SG];
262};
263
264struct ipoib_cm_dev_priv {
265 struct ib_srq *srq;
266 struct ipoib_cm_rx_buf *srq_ring;
267 struct ib_cm_id *id;
268 struct list_head passive_ids; /* state: LIVE */
269 struct list_head rx_error_list; /* state: ERROR */
270 struct list_head rx_flush_list; /* state: FLUSH, drain not started */
271 struct list_head rx_drain_list; /* state: FLUSH, drain started */
272 struct list_head rx_reap_list; /* state: FLUSH, drain done */
273 struct work_struct start_task;
274 struct work_struct reap_task;
275 struct work_struct skb_task;
276 struct work_struct rx_reap_task;
277 struct delayed_work stale_task;
278 struct sk_buff_head skb_queue;
279 struct list_head start_list;
280 struct list_head reap_list;
281 struct ib_wc ibwc[IPOIB_NUM_WC];
282 struct ib_sge rx_sge[IPOIB_CM_RX_SG];
283 struct ib_recv_wr rx_wr;
284 int nonsrq_conn_qp;
285 int max_cm_mtu;
286 int num_frags;
287};
288
289struct ipoib_ethtool_st {
290 u16 coalesce_usecs;
291 u16 max_coalesced_frames;
292};
293
294struct ipoib_neigh_table;
295
296struct ipoib_neigh_hash {
297 struct ipoib_neigh_table *ntbl;
298 struct ipoib_neigh __rcu **buckets;
299 struct rcu_head rcu;
300 u32 mask;
301 u32 size;
302};
303
304struct ipoib_neigh_table {
305 struct ipoib_neigh_hash __rcu *htbl;
306 atomic_t entries;
307 struct completion flushed;
308 struct completion deleted;
309};
310
311struct ipoib_qp_state_validate {
312 struct work_struct work;
313 struct ipoib_dev_priv *priv;
314};
315
316/*
317 * Device private locking: network stack tx_lock protects members used
318 * in TX fast path, lock protects everything else. lock nests inside
319 * of tx_lock (ie tx_lock must be acquired first if needed).
320 */
321struct ipoib_dev_priv {
322 spinlock_t lock;
323
324 struct net_device *dev;
325 void (*next_priv_destructor)(struct net_device *dev);
326
327 struct napi_struct send_napi;
328 struct napi_struct recv_napi;
329
330 unsigned long flags;
331
332 /*
333 * This protects access to the child_intfs list.
334 * To READ from child_intfs the RTNL or vlan_rwsem read side must be
335 * held. To WRITE RTNL and the vlan_rwsem write side must be held (in
336 * that order) This lock exists because we have a few contexts where
337 * we need the child_intfs, but do not want to grab the RTNL.
338 */
339 struct rw_semaphore vlan_rwsem;
340 struct mutex mcast_mutex;
341
342 struct rb_root path_tree;
343 struct list_head path_list;
344
345 struct ipoib_neigh_table ntbl;
346
347 struct ipoib_mcast *broadcast;
348 struct list_head multicast_list;
349 struct rb_root multicast_tree;
350
351 struct workqueue_struct *wq;
352 struct delayed_work mcast_task;
353 struct work_struct carrier_on_task;
354 struct work_struct reschedule_napi_work;
355 struct work_struct flush_light;
356 struct work_struct flush_normal;
357 struct work_struct flush_heavy;
358 struct work_struct restart_task;
359 struct work_struct tx_timeout_work;
360 struct delayed_work ah_reap_task;
361 struct delayed_work neigh_reap_task;
362 struct ib_device *ca;
363 u8 port;
364 u16 pkey;
365 u16 pkey_index;
366 struct ib_pd *pd;
367 struct ib_cq *recv_cq;
368 struct ib_cq *send_cq;
369 struct ib_qp *qp;
370 u32 qkey;
371
372 union ib_gid local_gid;
373 u32 local_lid;
374
375 unsigned int admin_mtu;
376 unsigned int mcast_mtu;
377 unsigned int max_ib_mtu;
378
379 struct ipoib_rx_buf *rx_ring;
380
381 struct ipoib_tx_buf *tx_ring;
382 /* cyclic ring variables for managing tx_ring, for UD only */
383 unsigned int tx_head;
384 unsigned int tx_tail;
385 /* cyclic ring variables for counting overall outstanding send WRs */
386 unsigned int global_tx_head;
387 unsigned int global_tx_tail;
388 struct ib_sge tx_sge[MAX_SKB_FRAGS + 1];
389 struct ib_ud_wr tx_wr;
390 struct ib_wc send_wc[MAX_SEND_CQE];
391
392 struct ib_recv_wr rx_wr;
393 struct ib_sge rx_sge[IPOIB_UD_RX_SG];
394
395 struct ib_wc ibwc[IPOIB_NUM_WC];
396
397 struct list_head dead_ahs;
398
399 struct ib_event_handler event_handler;
400
401 struct net_device *parent;
402 struct list_head child_intfs;
403 struct list_head list;
404 int child_type;
405
406#ifdef CONFIG_INFINIBAND_IPOIB_CM
407 struct ipoib_cm_dev_priv cm;
408#endif
409
410#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
411 struct list_head fs_list;
412 struct dentry *mcg_dentry;
413 struct dentry *path_dentry;
414#endif
415 u64 hca_caps;
416 u64 kernel_caps;
417 struct ipoib_ethtool_st ethtool;
418 unsigned int max_send_sge;
419 const struct net_device_ops *rn_ops;
420};
421
422struct ipoib_ah {
423 struct net_device *dev;
424 struct ib_ah *ah;
425 struct list_head list;
426 struct kref ref;
427 unsigned int last_send;
428 int valid;
429};
430
431struct ipoib_path {
432 struct net_device *dev;
433 struct sa_path_rec pathrec;
434 struct ipoib_ah *ah;
435 struct sk_buff_head queue;
436
437 struct list_head neigh_list;
438
439 int query_id;
440 struct ib_sa_query *query;
441 struct completion done;
442
443 struct rb_node rb_node;
444 struct list_head list;
445};
446
447struct ipoib_neigh {
448 struct ipoib_ah *ah;
449#ifdef CONFIG_INFINIBAND_IPOIB_CM
450 struct ipoib_cm_tx *cm;
451#endif
452 u8 daddr[INFINIBAND_ALEN];
453 struct sk_buff_head queue;
454
455 struct net_device *dev;
456
457 struct list_head list;
458 struct ipoib_neigh __rcu *hnext;
459 struct rcu_head rcu;
460 refcount_t refcnt;
461 unsigned long alive;
462};
463
464#define IPOIB_UD_MTU(ib_mtu) (ib_mtu - IPOIB_ENCAP_LEN)
465#define IPOIB_UD_BUF_SIZE(ib_mtu) (ib_mtu + IB_GRH_BYTES)
466
467void ipoib_neigh_dtor(struct ipoib_neigh *neigh);
468static inline void ipoib_neigh_put(struct ipoib_neigh *neigh)
469{
470 if (refcount_dec_and_test(&neigh->refcnt))
471 ipoib_neigh_dtor(neigh);
472}
473struct ipoib_neigh *ipoib_neigh_get(struct net_device *dev, u8 *daddr);
474struct ipoib_neigh *ipoib_neigh_alloc(u8 *daddr,
475 struct net_device *dev);
476void ipoib_neigh_free(struct ipoib_neigh *neigh);
477void ipoib_del_neighs_by_gid(struct net_device *dev, u8 *gid);
478
479extern struct workqueue_struct *ipoib_workqueue;
480
481/* functions */
482
483int ipoib_rx_poll(struct napi_struct *napi, int budget);
484int ipoib_tx_poll(struct napi_struct *napi, int budget);
485void ipoib_ib_rx_completion(struct ib_cq *cq, void *ctx_ptr);
486void ipoib_ib_tx_completion(struct ib_cq *cq, void *ctx_ptr);
487
488struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
489 struct ib_pd *pd, struct rdma_ah_attr *attr);
490void ipoib_free_ah(struct kref *kref);
491static inline void ipoib_put_ah(struct ipoib_ah *ah)
492{
493 kref_put(&ah->ref, ipoib_free_ah);
494}
495int ipoib_open(struct net_device *dev);
496void ipoib_intf_free(struct net_device *dev);
497int ipoib_add_pkey_attr(struct net_device *dev);
498int ipoib_add_umcast_attr(struct net_device *dev);
499
500int ipoib_send(struct net_device *dev, struct sk_buff *skb,
501 struct ib_ah *address, u32 dqpn);
502void ipoib_reap_ah(struct work_struct *work);
503
504void ipoib_napi_schedule_work(struct work_struct *work);
505struct ipoib_path *__path_find(struct net_device *dev, void *gid);
506void ipoib_mark_paths_invalid(struct net_device *dev);
507void ipoib_flush_paths(struct net_device *dev);
508struct net_device *ipoib_intf_alloc(struct ib_device *hca, u32 port,
509 const char *format);
510int ipoib_intf_init(struct ib_device *hca, u32 port, const char *format,
511 struct net_device *dev);
512void ipoib_ib_tx_timer_func(struct timer_list *t);
513void ipoib_ib_dev_flush_light(struct work_struct *work);
514void ipoib_ib_dev_flush_normal(struct work_struct *work);
515void ipoib_ib_dev_flush_heavy(struct work_struct *work);
516void ipoib_ib_tx_timeout_work(struct work_struct *work);
517void ipoib_pkey_event(struct work_struct *work);
518void ipoib_ib_dev_cleanup(struct net_device *dev);
519
520int ipoib_ib_dev_open_default(struct net_device *dev);
521int ipoib_ib_dev_open(struct net_device *dev);
522void ipoib_ib_dev_stop(struct net_device *dev);
523void ipoib_ib_dev_up(struct net_device *dev);
524void ipoib_ib_dev_down(struct net_device *dev);
525int ipoib_ib_dev_stop_default(struct net_device *dev);
526void ipoib_pkey_dev_check_presence(struct net_device *dev);
527
528void ipoib_mcast_join_task(struct work_struct *work);
529void ipoib_mcast_carrier_on_task(struct work_struct *work);
530void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb);
531
532void ipoib_mcast_restart_task(struct work_struct *work);
533void ipoib_mcast_start_thread(struct net_device *dev);
534void ipoib_mcast_stop_thread(struct net_device *dev);
535
536void ipoib_mcast_dev_down(struct net_device *dev);
537void ipoib_mcast_dev_flush(struct net_device *dev);
538
539int ipoib_dma_map_tx(struct ib_device *ca, struct ipoib_tx_buf *tx_req);
540void ipoib_dma_unmap_tx(struct ipoib_dev_priv *priv,
541 struct ipoib_tx_buf *tx_req);
542
543struct rtnl_link_ops *ipoib_get_link_ops(void);
544
545static inline void ipoib_build_sge(struct ipoib_dev_priv *priv,
546 struct ipoib_tx_buf *tx_req)
547{
548 int i, off;
549 struct sk_buff *skb = tx_req->skb;
550 skb_frag_t *frags = skb_shinfo(skb)->frags;
551 int nr_frags = skb_shinfo(skb)->nr_frags;
552 u64 *mapping = tx_req->mapping;
553
554 if (skb_headlen(skb)) {
555 priv->tx_sge[0].addr = mapping[0];
556 priv->tx_sge[0].length = skb_headlen(skb);
557 off = 1;
558 } else
559 off = 0;
560
561 for (i = 0; i < nr_frags; ++i) {
562 priv->tx_sge[i + off].addr = mapping[i + off];
563 priv->tx_sge[i + off].length = skb_frag_size(&frags[i]);
564 }
565 priv->tx_wr.wr.num_sge = nr_frags + off;
566}
567
568#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
569struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev);
570int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter);
571void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
572 union ib_gid *gid,
573 unsigned long *created,
574 unsigned int *queuelen,
575 unsigned int *complete,
576 unsigned int *send_only);
577
578struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev);
579int ipoib_path_iter_next(struct ipoib_path_iter *iter);
580void ipoib_path_iter_read(struct ipoib_path_iter *iter,
581 struct ipoib_path *path);
582#endif
583
584int ipoib_mcast_attach(struct net_device *dev, struct ib_device *hca,
585 union ib_gid *mgid, u16 mlid, int set_qkey, u32 qkey);
586int ipoib_mcast_detach(struct net_device *dev, struct ib_device *hca,
587 union ib_gid *mgid, u16 mlid);
588void ipoib_mcast_remove_list(struct list_head *remove_list);
589void ipoib_check_and_add_mcast_sendonly(struct ipoib_dev_priv *priv, u8 *mgid,
590 struct list_head *remove_list);
591
592int ipoib_init_qp(struct net_device *dev);
593int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca);
594void ipoib_transport_dev_cleanup(struct net_device *dev);
595
596void ipoib_event(struct ib_event_handler *handler,
597 struct ib_event *record);
598
599int ipoib_vlan_add(struct net_device *pdev, unsigned short pkey);
600int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey);
601
602int __ipoib_vlan_add(struct ipoib_dev_priv *ppriv, struct ipoib_dev_priv *priv,
603 u16 pkey, int child_type);
604
605int __init ipoib_netlink_init(void);
606void __exit ipoib_netlink_fini(void);
607
608void ipoib_set_umcast(struct net_device *ndev, int umcast_val);
609int ipoib_set_mode(struct net_device *dev, const char *buf);
610
611void ipoib_setup_common(struct net_device *dev);
612
613void ipoib_pkey_open(struct ipoib_dev_priv *priv);
614void ipoib_drain_cq(struct net_device *dev);
615
616void ipoib_set_ethtool_ops(struct net_device *dev);
617
618#define IPOIB_FLAGS_RC 0x80
619#define IPOIB_FLAGS_UC 0x40
620
621/* We don't support UC connections at the moment */
622#define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
623
624#ifdef CONFIG_INFINIBAND_IPOIB_CM
625
626extern int ipoib_max_conn_qp;
627
628static inline int ipoib_cm_admin_enabled(struct net_device *dev)
629{
630 struct ipoib_dev_priv *priv = ipoib_priv(dev);
631 return IPOIB_CM_SUPPORTED(dev->dev_addr) &&
632 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
633}
634
635static inline int ipoib_cm_enabled(struct net_device *dev, u8 *hwaddr)
636{
637 struct ipoib_dev_priv *priv = ipoib_priv(dev);
638 return IPOIB_CM_SUPPORTED(hwaddr) &&
639 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
640}
641
642static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
643
644{
645 return test_bit(IPOIB_FLAG_OPER_UP, &neigh->cm->flags);
646}
647
648static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
649{
650 return neigh->cm;
651}
652
653static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
654{
655 neigh->cm = tx;
656}
657
658static inline int ipoib_cm_has_srq(struct net_device *dev)
659{
660 struct ipoib_dev_priv *priv = ipoib_priv(dev);
661 return !!priv->cm.srq;
662}
663
664static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
665{
666 struct ipoib_dev_priv *priv = ipoib_priv(dev);
667 return priv->cm.max_cm_mtu;
668}
669
670void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx);
671int ipoib_cm_dev_open(struct net_device *dev);
672void ipoib_cm_dev_stop(struct net_device *dev);
673int ipoib_cm_dev_init(struct net_device *dev);
674int ipoib_cm_add_mode_attr(struct net_device *dev);
675void ipoib_cm_dev_cleanup(struct net_device *dev);
676struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
677 struct ipoib_neigh *neigh);
678void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx);
679void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
680 unsigned int mtu);
681void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc);
682void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc);
683#else
684
685#define ipoib_max_conn_qp 0
686
687static inline int ipoib_cm_admin_enabled(struct net_device *dev)
688{
689 return 0;
690}
691static inline int ipoib_cm_enabled(struct net_device *dev, u8 *hwaddr)
692
693{
694 return 0;
695}
696
697static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
698
699{
700 return 0;
701}
702
703static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
704{
705 return NULL;
706}
707
708static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
709{
710}
711
712static inline int ipoib_cm_has_srq(struct net_device *dev)
713{
714 return 0;
715}
716
717static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
718{
719 return 0;
720}
721
722static inline
723void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx)
724{
725 return;
726}
727
728static inline
729int ipoib_cm_dev_open(struct net_device *dev)
730{
731 return 0;
732}
733
734static inline
735void ipoib_cm_dev_stop(struct net_device *dev)
736{
737 return;
738}
739
740static inline
741int ipoib_cm_dev_init(struct net_device *dev)
742{
743 return -EOPNOTSUPP;
744}
745
746static inline
747void ipoib_cm_dev_cleanup(struct net_device *dev)
748{
749 return;
750}
751
752static inline
753struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
754 struct ipoib_neigh *neigh)
755{
756 return NULL;
757}
758
759static inline
760void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx)
761{
762 return;
763}
764
765static inline
766int ipoib_cm_add_mode_attr(struct net_device *dev)
767{
768 return 0;
769}
770
771static inline void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
772 unsigned int mtu)
773{
774 dev_kfree_skb_any(skb);
775}
776
777static inline void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
778{
779}
780
781static inline void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
782{
783}
784#endif
785
786#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
787void ipoib_create_debug_files(struct net_device *dev);
788void ipoib_delete_debug_files(struct net_device *dev);
789void ipoib_register_debugfs(void);
790void ipoib_unregister_debugfs(void);
791#else
792static inline void ipoib_create_debug_files(struct net_device *dev) { }
793static inline void ipoib_delete_debug_files(struct net_device *dev) { }
794static inline void ipoib_register_debugfs(void) { }
795static inline void ipoib_unregister_debugfs(void) { }
796#endif
797
798#define ipoib_printk(level, priv, format, arg...) \
799 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
800#define ipoib_warn(priv, format, arg...) \
801do { \
802 static DEFINE_RATELIMIT_STATE(_rs, \
803 10 * HZ /*10 seconds */, \
804 100); \
805 if (__ratelimit(&_rs)) \
806 ipoib_printk(KERN_WARNING, priv, format , ## arg);\
807} while (0)
808
809extern int ipoib_sendq_size;
810extern int ipoib_recvq_size;
811
812extern struct ib_sa_client ipoib_sa_client;
813
814#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
815extern int ipoib_debug_level;
816
817#define ipoib_dbg(priv, format, arg...) \
818 do { \
819 if (ipoib_debug_level > 0) \
820 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
821 } while (0)
822#define ipoib_dbg_mcast(priv, format, arg...) \
823 do { \
824 if (mcast_debug_level > 0) \
825 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
826 } while (0)
827#else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
828#define ipoib_dbg(priv, format, arg...) \
829 do { (void) (priv); } while (0)
830#define ipoib_dbg_mcast(priv, format, arg...) \
831 do { (void) (priv); } while (0)
832#endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
833
834#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
835#define ipoib_dbg_data(priv, format, arg...) \
836 do { \
837 if (data_debug_level > 0) \
838 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
839 } while (0)
840#else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
841#define ipoib_dbg_data(priv, format, arg...) \
842 do { (void) (priv); } while (0)
843#endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
844
845#define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
846
847#endif /* _IPOIB_H */
1/*
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.
5 *
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:
11 *
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
14 * conditions are met:
15 *
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
18 * disclaimer.
19 *
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.
24 *
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
32 * SOFTWARE.
33 */
34
35#ifndef _IPOIB_H
36#define _IPOIB_H
37
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>
45
46#include <net/neighbour.h>
47#include <net/sch_generic.h>
48
49#include <linux/atomic.h>
50
51#include <rdma/ib_verbs.h>
52#include <rdma/ib_pack.h>
53#include <rdma/ib_sa.h>
54#include <linux/sched.h>
55/* constants */
56
57enum ipoib_flush_level {
58 IPOIB_FLUSH_LIGHT,
59 IPOIB_FLUSH_NORMAL,
60 IPOIB_FLUSH_HEAVY
61};
62
63enum {
64 IPOIB_ENCAP_LEN = 4,
65 IPOIB_PSEUDO_LEN = 20,
66 IPOIB_HARD_LEN = IPOIB_ENCAP_LEN + IPOIB_PSEUDO_LEN,
67
68 IPOIB_UD_HEAD_SIZE = IB_GRH_BYTES + IPOIB_ENCAP_LEN,
69 IPOIB_UD_RX_SG = 2, /* max buffer needed for 4K mtu */
70
71 IPOIB_CM_MTU = 0x10000 - 0x10, /* padding to align header to 16 */
72 IPOIB_CM_BUF_SIZE = IPOIB_CM_MTU + IPOIB_ENCAP_LEN,
73 IPOIB_CM_HEAD_SIZE = IPOIB_CM_BUF_SIZE % PAGE_SIZE,
74 IPOIB_CM_RX_SG = ALIGN(IPOIB_CM_BUF_SIZE, PAGE_SIZE) / PAGE_SIZE,
75 IPOIB_RX_RING_SIZE = 256,
76 IPOIB_TX_RING_SIZE = 128,
77 IPOIB_MAX_QUEUE_SIZE = 8192,
78 IPOIB_MIN_QUEUE_SIZE = 2,
79 IPOIB_CM_MAX_CONN_QP = 4096,
80
81 IPOIB_NUM_WC = 4,
82
83 IPOIB_MAX_PATH_REC_QUEUE = 3,
84 IPOIB_MAX_MCAST_QUEUE = 64,
85
86 IPOIB_FLAG_OPER_UP = 0,
87 IPOIB_FLAG_INITIALIZED = 1,
88 IPOIB_FLAG_ADMIN_UP = 2,
89 IPOIB_PKEY_ASSIGNED = 3,
90 IPOIB_FLAG_SUBINTERFACE = 5,
91 IPOIB_STOP_REAPER = 7,
92 IPOIB_FLAG_ADMIN_CM = 9,
93 IPOIB_FLAG_UMCAST = 10,
94 IPOIB_NEIGH_TBL_FLUSH = 12,
95 IPOIB_FLAG_DEV_ADDR_SET = 13,
96 IPOIB_FLAG_DEV_ADDR_CTRL = 14,
97
98 IPOIB_MAX_BACKOFF_SECONDS = 16,
99
100 IPOIB_MCAST_FLAG_FOUND = 0, /* used in set_multicast_list */
101 IPOIB_MCAST_FLAG_SENDONLY = 1,
102 /*
103 * For IPOIB_MCAST_FLAG_BUSY
104 * When set, in flight join and mcast->mc is unreliable
105 * When clear and mcast->mc IS_ERR_OR_NULL, need to restart or
106 * haven't started yet
107 * When clear and mcast->mc is valid pointer, join was successful
108 */
109 IPOIB_MCAST_FLAG_BUSY = 2,
110 IPOIB_MCAST_FLAG_ATTACHED = 3,
111
112 MAX_SEND_CQE = 64,
113 IPOIB_CM_COPYBREAK = 256,
114
115 IPOIB_NON_CHILD = 0,
116 IPOIB_LEGACY_CHILD = 1,
117 IPOIB_RTNL_CHILD = 2,
118};
119
120#define IPOIB_OP_RECV (1ul << 31)
121#ifdef CONFIG_INFINIBAND_IPOIB_CM
122#define IPOIB_OP_CM (1ul << 30)
123#else
124#define IPOIB_OP_CM (0)
125#endif
126
127#define IPOIB_QPN_MASK ((__force u32) cpu_to_be32(0xFFFFFF))
128
129/* structs */
130
131struct ipoib_header {
132 __be16 proto;
133 u16 reserved;
134};
135
136struct ipoib_pseudo_header {
137 u8 hwaddr[INFINIBAND_ALEN];
138};
139
140static inline void skb_add_pseudo_hdr(struct sk_buff *skb)
141{
142 char *data = skb_push(skb, IPOIB_PSEUDO_LEN);
143
144 /*
145 * only the ipoib header is present now, make room for a dummy
146 * pseudo header and set skb field accordingly
147 */
148 memset(data, 0, IPOIB_PSEUDO_LEN);
149 skb_reset_mac_header(skb);
150 skb_pull(skb, IPOIB_HARD_LEN);
151}
152
153static inline struct ipoib_dev_priv *ipoib_priv(const struct net_device *dev)
154{
155 struct rdma_netdev *rn = netdev_priv(dev);
156
157 return rn->clnt_priv;
158}
159
160/* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
161struct ipoib_mcast {
162 struct ib_sa_mcmember_rec mcmember;
163 struct ib_sa_multicast *mc;
164 struct ipoib_ah *ah;
165
166 struct rb_node rb_node;
167 struct list_head list;
168
169 unsigned long created;
170 unsigned long backoff;
171 unsigned long delay_until;
172
173 unsigned long flags;
174 unsigned char logcount;
175
176 struct list_head neigh_list;
177
178 struct sk_buff_head pkt_queue;
179
180 struct net_device *dev;
181 struct completion done;
182};
183
184struct ipoib_rx_buf {
185 struct sk_buff *skb;
186 u64 mapping[IPOIB_UD_RX_SG];
187};
188
189struct ipoib_tx_buf {
190 struct sk_buff *skb;
191 u64 mapping[MAX_SKB_FRAGS + 1];
192};
193
194struct ib_cm_id;
195
196struct ipoib_cm_data {
197 __be32 qpn; /* High byte MUST be ignored on receive */
198 __be32 mtu;
199};
200
201/*
202 * Quoting 10.3.1 Queue Pair and EE Context States:
203 *
204 * Note, for QPs that are associated with an SRQ, the Consumer should take the
205 * QP through the Error State before invoking a Destroy QP or a Modify QP to the
206 * Reset State. The Consumer may invoke the Destroy QP without first performing
207 * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
208 * Last WQE Reached Event. However, if the Consumer does not wait for the
209 * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
210 * leakage may occur. Therefore, it is good programming practice to tear down a
211 * QP that is associated with an SRQ by using the following process:
212 *
213 * - Put the QP in the Error State
214 * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
215 * - either:
216 * drain the CQ by invoking the Poll CQ verb and either wait for CQ
217 * to be empty or the number of Poll CQ operations has exceeded
218 * CQ capacity size;
219 * - or
220 * post another WR that completes on the same CQ and wait for this
221 * WR to return as a WC;
222 * - and then invoke a Destroy QP or Reset QP.
223 *
224 * We use the second option and wait for a completion on the
225 * same CQ before destroying QPs attached to our SRQ.
226 */
227
228enum ipoib_cm_state {
229 IPOIB_CM_RX_LIVE,
230 IPOIB_CM_RX_ERROR, /* Ignored by stale task */
231 IPOIB_CM_RX_FLUSH /* Last WQE Reached event observed */
232};
233
234struct ipoib_cm_rx {
235 struct ib_cm_id *id;
236 struct ib_qp *qp;
237 struct ipoib_cm_rx_buf *rx_ring;
238 struct list_head list;
239 struct net_device *dev;
240 unsigned long jiffies;
241 enum ipoib_cm_state state;
242 int recv_count;
243};
244
245struct ipoib_cm_tx {
246 struct ib_cm_id *id;
247 struct ib_qp *qp;
248 struct list_head list;
249 struct net_device *dev;
250 struct ipoib_neigh *neigh;
251 struct ipoib_tx_buf *tx_ring;
252 unsigned int tx_head;
253 unsigned int tx_tail;
254 unsigned long flags;
255 u32 mtu;
256 unsigned int max_send_sge;
257};
258
259struct ipoib_cm_rx_buf {
260 struct sk_buff *skb;
261 u64 mapping[IPOIB_CM_RX_SG];
262};
263
264struct ipoib_cm_dev_priv {
265 struct ib_srq *srq;
266 struct ipoib_cm_rx_buf *srq_ring;
267 struct ib_cm_id *id;
268 struct list_head passive_ids; /* state: LIVE */
269 struct list_head rx_error_list; /* state: ERROR */
270 struct list_head rx_flush_list; /* state: FLUSH, drain not started */
271 struct list_head rx_drain_list; /* state: FLUSH, drain started */
272 struct list_head rx_reap_list; /* state: FLUSH, drain done */
273 struct work_struct start_task;
274 struct work_struct reap_task;
275 struct work_struct skb_task;
276 struct work_struct rx_reap_task;
277 struct delayed_work stale_task;
278 struct sk_buff_head skb_queue;
279 struct list_head start_list;
280 struct list_head reap_list;
281 struct ib_wc ibwc[IPOIB_NUM_WC];
282 struct ib_sge rx_sge[IPOIB_CM_RX_SG];
283 struct ib_recv_wr rx_wr;
284 int nonsrq_conn_qp;
285 int max_cm_mtu;
286 int num_frags;
287};
288
289struct ipoib_ethtool_st {
290 u16 coalesce_usecs;
291 u16 max_coalesced_frames;
292};
293
294struct ipoib_neigh_table;
295
296struct ipoib_neigh_hash {
297 struct ipoib_neigh_table *ntbl;
298 struct ipoib_neigh __rcu **buckets;
299 struct rcu_head rcu;
300 u32 mask;
301 u32 size;
302};
303
304struct ipoib_neigh_table {
305 struct ipoib_neigh_hash __rcu *htbl;
306 atomic_t entries;
307 struct completion flushed;
308 struct completion deleted;
309};
310
311struct ipoib_qp_state_validate {
312 struct work_struct work;
313 struct ipoib_dev_priv *priv;
314};
315
316/*
317 * Device private locking: network stack tx_lock protects members used
318 * in TX fast path, lock protects everything else. lock nests inside
319 * of tx_lock (ie tx_lock must be acquired first if needed).
320 */
321struct ipoib_dev_priv {
322 spinlock_t lock;
323
324 struct net_device *dev;
325 void (*next_priv_destructor)(struct net_device *dev);
326
327 struct napi_struct send_napi;
328 struct napi_struct recv_napi;
329
330 unsigned long flags;
331
332 /*
333 * This protects access to the child_intfs list.
334 * To READ from child_intfs the RTNL or vlan_rwsem read side must be
335 * held. To WRITE RTNL and the vlan_rwsem write side must be held (in
336 * that order) This lock exists because we have a few contexts where
337 * we need the child_intfs, but do not want to grab the RTNL.
338 */
339 struct rw_semaphore vlan_rwsem;
340 struct mutex mcast_mutex;
341
342 struct rb_root path_tree;
343 struct list_head path_list;
344
345 struct ipoib_neigh_table ntbl;
346
347 struct ipoib_mcast *broadcast;
348 struct list_head multicast_list;
349 struct rb_root multicast_tree;
350
351 struct workqueue_struct *wq;
352 struct delayed_work mcast_task;
353 struct work_struct carrier_on_task;
354 struct work_struct flush_light;
355 struct work_struct flush_normal;
356 struct work_struct flush_heavy;
357 struct work_struct restart_task;
358 struct delayed_work ah_reap_task;
359 struct delayed_work neigh_reap_task;
360 struct ib_device *ca;
361 u8 port;
362 u16 pkey;
363 u16 pkey_index;
364 struct ib_pd *pd;
365 struct ib_cq *recv_cq;
366 struct ib_cq *send_cq;
367 struct ib_qp *qp;
368 u32 qkey;
369
370 union ib_gid local_gid;
371 u32 local_lid;
372
373 unsigned int admin_mtu;
374 unsigned int mcast_mtu;
375 unsigned int max_ib_mtu;
376
377 struct ipoib_rx_buf *rx_ring;
378
379 struct ipoib_tx_buf *tx_ring;
380 unsigned int tx_head;
381 unsigned int tx_tail;
382 struct ib_sge tx_sge[MAX_SKB_FRAGS + 1];
383 struct ib_ud_wr tx_wr;
384 struct ib_wc send_wc[MAX_SEND_CQE];
385
386 struct ib_recv_wr rx_wr;
387 struct ib_sge rx_sge[IPOIB_UD_RX_SG];
388
389 struct ib_wc ibwc[IPOIB_NUM_WC];
390
391 struct list_head dead_ahs;
392
393 struct ib_event_handler event_handler;
394
395 struct net_device *parent;
396 struct list_head child_intfs;
397 struct list_head list;
398 int child_type;
399
400#ifdef CONFIG_INFINIBAND_IPOIB_CM
401 struct ipoib_cm_dev_priv cm;
402#endif
403
404#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
405 struct list_head fs_list;
406 struct dentry *mcg_dentry;
407 struct dentry *path_dentry;
408#endif
409 u64 hca_caps;
410 struct ipoib_ethtool_st ethtool;
411 unsigned int max_send_sge;
412 bool sm_fullmember_sendonly_support;
413 const struct net_device_ops *rn_ops;
414};
415
416struct ipoib_ah {
417 struct net_device *dev;
418 struct ib_ah *ah;
419 struct list_head list;
420 struct kref ref;
421 unsigned int last_send;
422 int valid;
423};
424
425struct ipoib_path {
426 struct net_device *dev;
427 struct sa_path_rec pathrec;
428 struct ipoib_ah *ah;
429 struct sk_buff_head queue;
430
431 struct list_head neigh_list;
432
433 int query_id;
434 struct ib_sa_query *query;
435 struct completion done;
436
437 struct rb_node rb_node;
438 struct list_head list;
439};
440
441struct ipoib_neigh {
442 struct ipoib_ah *ah;
443#ifdef CONFIG_INFINIBAND_IPOIB_CM
444 struct ipoib_cm_tx *cm;
445#endif
446 u8 daddr[INFINIBAND_ALEN];
447 struct sk_buff_head queue;
448
449 struct net_device *dev;
450
451 struct list_head list;
452 struct ipoib_neigh __rcu *hnext;
453 struct rcu_head rcu;
454 atomic_t refcnt;
455 unsigned long alive;
456};
457
458#define IPOIB_UD_MTU(ib_mtu) (ib_mtu - IPOIB_ENCAP_LEN)
459#define IPOIB_UD_BUF_SIZE(ib_mtu) (ib_mtu + IB_GRH_BYTES)
460
461void ipoib_neigh_dtor(struct ipoib_neigh *neigh);
462static inline void ipoib_neigh_put(struct ipoib_neigh *neigh)
463{
464 if (atomic_dec_and_test(&neigh->refcnt))
465 ipoib_neigh_dtor(neigh);
466}
467struct ipoib_neigh *ipoib_neigh_get(struct net_device *dev, u8 *daddr);
468struct ipoib_neigh *ipoib_neigh_alloc(u8 *daddr,
469 struct net_device *dev);
470void ipoib_neigh_free(struct ipoib_neigh *neigh);
471void ipoib_del_neighs_by_gid(struct net_device *dev, u8 *gid);
472
473extern struct workqueue_struct *ipoib_workqueue;
474
475/* functions */
476
477int ipoib_rx_poll(struct napi_struct *napi, int budget);
478int ipoib_tx_poll(struct napi_struct *napi, int budget);
479void ipoib_ib_rx_completion(struct ib_cq *cq, void *ctx_ptr);
480void ipoib_ib_tx_completion(struct ib_cq *cq, void *ctx_ptr);
481
482struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
483 struct ib_pd *pd, struct rdma_ah_attr *attr);
484void ipoib_free_ah(struct kref *kref);
485static inline void ipoib_put_ah(struct ipoib_ah *ah)
486{
487 kref_put(&ah->ref, ipoib_free_ah);
488}
489int ipoib_open(struct net_device *dev);
490void ipoib_intf_free(struct net_device *dev);
491int ipoib_add_pkey_attr(struct net_device *dev);
492int ipoib_add_umcast_attr(struct net_device *dev);
493
494int ipoib_send(struct net_device *dev, struct sk_buff *skb,
495 struct ib_ah *address, u32 dqpn);
496void ipoib_reap_ah(struct work_struct *work);
497
498struct ipoib_path *__path_find(struct net_device *dev, void *gid);
499void ipoib_mark_paths_invalid(struct net_device *dev);
500void ipoib_flush_paths(struct net_device *dev);
501struct net_device *ipoib_intf_alloc(struct ib_device *hca, u8 port,
502 const char *format);
503int ipoib_intf_init(struct ib_device *hca, u8 port, const char *format,
504 struct net_device *dev);
505void ipoib_ib_tx_timer_func(struct timer_list *t);
506void ipoib_ib_dev_flush_light(struct work_struct *work);
507void ipoib_ib_dev_flush_normal(struct work_struct *work);
508void ipoib_ib_dev_flush_heavy(struct work_struct *work);
509void ipoib_pkey_event(struct work_struct *work);
510void ipoib_ib_dev_cleanup(struct net_device *dev);
511
512int ipoib_ib_dev_open_default(struct net_device *dev);
513int ipoib_ib_dev_open(struct net_device *dev);
514int ipoib_ib_dev_stop(struct net_device *dev);
515void ipoib_ib_dev_up(struct net_device *dev);
516void ipoib_ib_dev_down(struct net_device *dev);
517int ipoib_ib_dev_stop_default(struct net_device *dev);
518void ipoib_pkey_dev_check_presence(struct net_device *dev);
519
520void ipoib_mcast_join_task(struct work_struct *work);
521void ipoib_mcast_carrier_on_task(struct work_struct *work);
522void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb);
523
524void ipoib_mcast_restart_task(struct work_struct *work);
525void ipoib_mcast_start_thread(struct net_device *dev);
526int ipoib_mcast_stop_thread(struct net_device *dev);
527
528void ipoib_mcast_dev_down(struct net_device *dev);
529void ipoib_mcast_dev_flush(struct net_device *dev);
530
531int ipoib_dma_map_tx(struct ib_device *ca, struct ipoib_tx_buf *tx_req);
532void ipoib_dma_unmap_tx(struct ipoib_dev_priv *priv,
533 struct ipoib_tx_buf *tx_req);
534
535struct rtnl_link_ops *ipoib_get_link_ops(void);
536
537static inline void ipoib_build_sge(struct ipoib_dev_priv *priv,
538 struct ipoib_tx_buf *tx_req)
539{
540 int i, off;
541 struct sk_buff *skb = tx_req->skb;
542 skb_frag_t *frags = skb_shinfo(skb)->frags;
543 int nr_frags = skb_shinfo(skb)->nr_frags;
544 u64 *mapping = tx_req->mapping;
545
546 if (skb_headlen(skb)) {
547 priv->tx_sge[0].addr = mapping[0];
548 priv->tx_sge[0].length = skb_headlen(skb);
549 off = 1;
550 } else
551 off = 0;
552
553 for (i = 0; i < nr_frags; ++i) {
554 priv->tx_sge[i + off].addr = mapping[i + off];
555 priv->tx_sge[i + off].length = skb_frag_size(&frags[i]);
556 }
557 priv->tx_wr.wr.num_sge = nr_frags + off;
558}
559
560#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
561struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev);
562int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter);
563void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
564 union ib_gid *gid,
565 unsigned long *created,
566 unsigned int *queuelen,
567 unsigned int *complete,
568 unsigned int *send_only);
569
570struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev);
571int ipoib_path_iter_next(struct ipoib_path_iter *iter);
572void ipoib_path_iter_read(struct ipoib_path_iter *iter,
573 struct ipoib_path *path);
574#endif
575
576int ipoib_mcast_attach(struct net_device *dev, struct ib_device *hca,
577 union ib_gid *mgid, u16 mlid, int set_qkey, u32 qkey);
578int ipoib_mcast_detach(struct net_device *dev, struct ib_device *hca,
579 union ib_gid *mgid, u16 mlid);
580void ipoib_mcast_remove_list(struct list_head *remove_list);
581void ipoib_check_and_add_mcast_sendonly(struct ipoib_dev_priv *priv, u8 *mgid,
582 struct list_head *remove_list);
583
584int ipoib_init_qp(struct net_device *dev);
585int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca);
586void ipoib_transport_dev_cleanup(struct net_device *dev);
587
588void ipoib_event(struct ib_event_handler *handler,
589 struct ib_event *record);
590
591int ipoib_vlan_add(struct net_device *pdev, unsigned short pkey);
592int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey);
593
594int __ipoib_vlan_add(struct ipoib_dev_priv *ppriv, struct ipoib_dev_priv *priv,
595 u16 pkey, int child_type);
596
597int __init ipoib_netlink_init(void);
598void __exit ipoib_netlink_fini(void);
599
600void ipoib_set_umcast(struct net_device *ndev, int umcast_val);
601int ipoib_set_mode(struct net_device *dev, const char *buf);
602
603void ipoib_setup_common(struct net_device *dev);
604
605void ipoib_pkey_open(struct ipoib_dev_priv *priv);
606void ipoib_drain_cq(struct net_device *dev);
607
608void ipoib_set_ethtool_ops(struct net_device *dev);
609
610#define IPOIB_FLAGS_RC 0x80
611#define IPOIB_FLAGS_UC 0x40
612
613/* We don't support UC connections at the moment */
614#define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
615
616#ifdef CONFIG_INFINIBAND_IPOIB_CM
617
618extern int ipoib_max_conn_qp;
619
620static inline int ipoib_cm_admin_enabled(struct net_device *dev)
621{
622 struct ipoib_dev_priv *priv = ipoib_priv(dev);
623 return IPOIB_CM_SUPPORTED(dev->dev_addr) &&
624 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
625}
626
627static inline int ipoib_cm_enabled(struct net_device *dev, u8 *hwaddr)
628{
629 struct ipoib_dev_priv *priv = ipoib_priv(dev);
630 return IPOIB_CM_SUPPORTED(hwaddr) &&
631 test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
632}
633
634static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
635
636{
637 return test_bit(IPOIB_FLAG_OPER_UP, &neigh->cm->flags);
638}
639
640static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
641{
642 return neigh->cm;
643}
644
645static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
646{
647 neigh->cm = tx;
648}
649
650static inline int ipoib_cm_has_srq(struct net_device *dev)
651{
652 struct ipoib_dev_priv *priv = ipoib_priv(dev);
653 return !!priv->cm.srq;
654}
655
656static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
657{
658 struct ipoib_dev_priv *priv = ipoib_priv(dev);
659 return priv->cm.max_cm_mtu;
660}
661
662void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx);
663int ipoib_cm_dev_open(struct net_device *dev);
664void ipoib_cm_dev_stop(struct net_device *dev);
665int ipoib_cm_dev_init(struct net_device *dev);
666int ipoib_cm_add_mode_attr(struct net_device *dev);
667void ipoib_cm_dev_cleanup(struct net_device *dev);
668struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
669 struct ipoib_neigh *neigh);
670void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx);
671void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
672 unsigned int mtu);
673void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc);
674void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc);
675#else
676
677struct ipoib_cm_tx;
678
679#define ipoib_max_conn_qp 0
680
681static inline int ipoib_cm_admin_enabled(struct net_device *dev)
682{
683 return 0;
684}
685static inline int ipoib_cm_enabled(struct net_device *dev, u8 *hwaddr)
686
687{
688 return 0;
689}
690
691static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
692
693{
694 return 0;
695}
696
697static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
698{
699 return NULL;
700}
701
702static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
703{
704}
705
706static inline int ipoib_cm_has_srq(struct net_device *dev)
707{
708 return 0;
709}
710
711static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
712{
713 return 0;
714}
715
716static inline
717void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx)
718{
719 return;
720}
721
722static inline
723int ipoib_cm_dev_open(struct net_device *dev)
724{
725 return 0;
726}
727
728static inline
729void ipoib_cm_dev_stop(struct net_device *dev)
730{
731 return;
732}
733
734static inline
735int ipoib_cm_dev_init(struct net_device *dev)
736{
737 return -EOPNOTSUPP;
738}
739
740static inline
741void ipoib_cm_dev_cleanup(struct net_device *dev)
742{
743 return;
744}
745
746static inline
747struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
748 struct ipoib_neigh *neigh)
749{
750 return NULL;
751}
752
753static inline
754void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx)
755{
756 return;
757}
758
759static inline
760int ipoib_cm_add_mode_attr(struct net_device *dev)
761{
762 return 0;
763}
764
765static inline void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
766 unsigned int mtu)
767{
768 dev_kfree_skb_any(skb);
769}
770
771static inline void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
772{
773}
774
775static inline void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
776{
777}
778#endif
779
780#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
781void ipoib_create_debug_files(struct net_device *dev);
782void ipoib_delete_debug_files(struct net_device *dev);
783void ipoib_register_debugfs(void);
784void ipoib_unregister_debugfs(void);
785#else
786static inline void ipoib_create_debug_files(struct net_device *dev) { }
787static inline void ipoib_delete_debug_files(struct net_device *dev) { }
788static inline void ipoib_register_debugfs(void) { }
789static inline void ipoib_unregister_debugfs(void) { }
790#endif
791
792#define ipoib_printk(level, priv, format, arg...) \
793 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
794#define ipoib_warn(priv, format, arg...) \
795do { \
796 static DEFINE_RATELIMIT_STATE(_rs, \
797 10 * HZ /*10 seconds */, \
798 100); \
799 if (__ratelimit(&_rs)) \
800 ipoib_printk(KERN_WARNING, priv, format , ## arg);\
801} while (0)
802
803extern int ipoib_sendq_size;
804extern int ipoib_recvq_size;
805
806extern struct ib_sa_client ipoib_sa_client;
807
808#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
809extern int ipoib_debug_level;
810
811#define ipoib_dbg(priv, format, arg...) \
812 do { \
813 if (ipoib_debug_level > 0) \
814 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
815 } while (0)
816#define ipoib_dbg_mcast(priv, format, arg...) \
817 do { \
818 if (mcast_debug_level > 0) \
819 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
820 } while (0)
821#else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
822#define ipoib_dbg(priv, format, arg...) \
823 do { (void) (priv); } while (0)
824#define ipoib_dbg_mcast(priv, format, arg...) \
825 do { (void) (priv); } while (0)
826#endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
827
828#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
829#define ipoib_dbg_data(priv, format, arg...) \
830 do { \
831 if (data_debug_level > 0) \
832 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
833 } while (0)
834#else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
835#define ipoib_dbg_data(priv, format, arg...) \
836 do { (void) (priv); } while (0)
837#endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
838
839#define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
840
841extern const char ipoib_driver_version[];
842
843#endif /* _IPOIB_H */