1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lnet/klnds/o2iblnd/o2iblnd.h
38 * Author: Eric Barton <eric@bartonsoftware.com>
42 # define EXPORT_SYMTAB
44 #ifndef AUTOCONF_INCLUDED
45 #include <linux/config.h>
47 #include <linux/module.h>
48 #include <linux/kernel.h>
50 #include <linux/string.h>
51 #include <linux/stat.h>
52 #include <linux/errno.h>
53 #include <linux/smp_lock.h>
54 #include <linux/unistd.h>
55 #include <linux/uio.h>
57 #include <asm/system.h>
58 #include <asm/uaccess.h>
61 #include <linux/init.h>
63 #include <linux/file.h>
64 #include <linux/stat.h>
65 #include <linux/list.h>
66 #include <linux/kmod.h>
67 #include <linux/sysctl.h>
68 #include <linux/random.h>
69 #include <linux/pci.h>
74 #define DEBUG_SUBSYSTEM S_LND
76 #include <libcfs/libcfs.h>
77 #include <lnet/lnet.h>
78 #include <lnet/lib-lnet.h>
79 #include <lnet/lnet-sysctl.h>
85 #include <rdma/rdma_cm.h>
86 #include <rdma/ib_cm.h>
87 #include <rdma/ib_verbs.h>
88 #include <rdma/ib_fmr_pool.h>
90 /* tunables fixed at compile time */
92 # define IBLND_N_SCHED num_online_cpus() /* # schedulers */
94 # define IBLND_N_SCHED 1 /* # schedulers */
97 #define IBLND_PEER_HASH_SIZE 101 /* # peer lists */
98 #define IBLND_RESCHED 100 /* # scheduler loops before reschedule */
99 #define IBLND_MSG_QUEUE_SIZE 8 /* # messages/RDMAs in-flight */
100 #define IBLND_CREDIT_HIGHWATER 7 /* when eagerly to return credits */
101 #define IBLND_MSG_SIZE (4<<10) /* max size of queued messages (inc hdr) */
103 #define IBLND_MAP_ON_DEMAND 0
104 #if IBLND_MAP_ON_DEMAND
105 # define IBLND_MAX_RDMA_FRAGS 1
107 # define IBLND_MAX_RDMA_FRAGS LNET_MAX_IOV
110 /************************/
111 /* derived constants... */
113 /* TX messages (shared by all connections) */
114 #define IBLND_TX_MSGS() (*kiblnd_tunables.kib_ntx)
115 #define IBLND_TX_MSG_BYTES() (IBLND_TX_MSGS() * IBLND_MSG_SIZE)
116 #define IBLND_TX_MSG_PAGES() ((IBLND_TX_MSG_BYTES() + PAGE_SIZE - 1)/PAGE_SIZE)
118 /* RX messages (per connection) */
119 #define IBLND_RX_MSGS (IBLND_MSG_QUEUE_SIZE * 2)
120 #define IBLND_RX_MSG_BYTES (IBLND_RX_MSGS * IBLND_MSG_SIZE)
121 #define IBLND_RX_MSG_PAGES ((IBLND_RX_MSG_BYTES + PAGE_SIZE - 1)/PAGE_SIZE)
123 /* WRs and CQEs (per connection) */
124 #define IBLND_RECV_WRS IBLND_RX_MSGS
125 #define IBLND_SEND_WRS ((*kiblnd_tunables.kib_concurrent_sends) * \
126 (1 + IBLND_MAX_RDMA_FRAGS))
127 #define IBLND_CQ_ENTRIES() (IBLND_RECV_WRS + IBLND_SEND_WRS)
131 unsigned int *kib_service; /* IB service number */
132 int *kib_min_reconnect_interval; /* first failed connection retry... */
133 int *kib_max_reconnect_interval; /* ...exponentially increasing to this */
134 int *kib_cksum; /* checksum kib_msg_t? */
135 int *kib_timeout; /* comms timeout (seconds) */
136 int *kib_keepalive; /* keepalive timeout (seconds) */
137 int *kib_ntx; /* # tx descs */
138 int *kib_credits; /* # concurrent sends */
139 int *kib_peercredits; /* # concurrent sends to 1 peer */
140 char **kib_default_ipif; /* default IPoIB interface */
141 int *kib_retry_count;
142 int *kib_rnr_retry_count;
143 int *kib_concurrent_sends; /* send work queue sizing */
144 int *kib_ib_mtu; /* IB MTU */
145 #if IBLND_MAP_ON_DEMAND
146 int *kib_fmr_pool_size; /* # FMRs in pool */
147 int *kib_fmr_flush_trigger; /* When to trigger FMR flush */
148 int *kib_fmr_cache; /* enable FMR pool cache? */
150 #if defined(CONFIG_SYSCTL) && !CFS_SYSFS_MODULE_PARM
151 cfs_sysctl_table_header_t *kib_sysctl; /* sysctl interface */
157 int ibp_npages; /* # pages */
158 struct page *ibp_pages[0];
163 struct list_head ibd_list; /* chain on kib_devs */
164 __u32 ibd_ifip; /* IPoIB interface IP */
165 char ibd_ifname[32]; /* IPoIB interface name */
166 int ibd_nnets; /* # nets extant */
168 struct rdma_cm_id *ibd_cmid; /* IB listener (bound to 1 device) */
169 struct ib_pd *ibd_pd; /* PD for the device */
170 struct ib_mr *ibd_mr; /* MR for non RDMA I/O */
175 __u64 ibn_incarnation; /* my epoch */
176 int ibn_init; /* initialisation state */
177 int ibn_shutdown; /* shutting down? */
179 atomic_t ibn_npeers; /* # peers extant */
180 atomic_t ibn_nconns; /* # connections extant */
182 struct kib_tx *ibn_tx_descs; /* all the tx descriptors */
183 kib_pages_t *ibn_tx_pages; /* premapped tx msg pages */
184 struct list_head ibn_idle_txs; /* idle tx descriptors */
185 spinlock_t ibn_tx_lock; /* serialise */
187 #if IBLND_MAP_ON_DEMAND
188 struct ib_fmr_pool *ibn_fmrpool; /* FMR pool for RDMA I/O */
191 kib_dev_t *ibn_dev; /* underlying IB device */
196 int kib_init; /* initialisation state */
197 int kib_shutdown; /* shut down? */
198 struct list_head kib_devs; /* IB devices extant */
199 atomic_t kib_nthreads; /* # live threads */
200 rwlock_t kib_global_lock; /* stabilize net/dev/peer/conn ops */
202 struct list_head *kib_peers; /* hash table of all my known peers */
203 int kib_peer_hash_size; /* size of kib_peers */
205 void *kib_connd; /* the connd task (serialisation assertions) */
206 struct list_head kib_connd_conns; /* connections to setup/teardown */
207 struct list_head kib_connd_zombies; /* connections with zero refcount */
208 wait_queue_head_t kib_connd_waitq; /* connection daemon sleeps here */
209 spinlock_t kib_connd_lock; /* serialise */
211 wait_queue_head_t kib_sched_waitq; /* schedulers sleep here */
212 struct list_head kib_sched_conns; /* conns to check for rx completions */
213 spinlock_t kib_sched_lock; /* serialise */
215 __u64 kib_next_tx_cookie; /* RDMA completion cookie */
216 struct ib_qp_attr kib_error_qpa; /* QP->ERROR */
219 #define IBLND_INIT_NOTHING 0
220 #define IBLND_INIT_DATA 1
221 #define IBLND_INIT_ALL 2
223 /************************************************************************
224 * IB Wire message format.
225 * These are sent in sender's byte order (i.e. receiver flips).
228 typedef struct kib_connparams
230 __u16 ibcp_queue_depth;
231 __u16 ibcp_max_frags;
232 __u32 ibcp_max_msg_size;
233 } WIRE_ATTR kib_connparams_t;
237 lnet_hdr_t ibim_hdr; /* portals header */
238 char ibim_payload[0]; /* piggy-backed payload */
239 } WIRE_ATTR kib_immediate_msg_t;
241 #if IBLND_MAP_ON_DEMAND
244 __u64 rd_addr; /* IO VMA address */
245 __u32 rd_nob; /* # of bytes */
246 __u32 rd_key; /* remote key */
247 } WIRE_ATTR kib_rdma_desc_t;
251 __u32 rf_nob; /* # bytes this frag */
252 __u64 rf_addr; /* CAVEAT EMPTOR: misaligned!! */
253 } WIRE_ATTR kib_rdma_frag_t;
257 __u32 rd_key; /* local/remote key */
258 __u32 rd_nfrags; /* # fragments */
259 kib_rdma_frag_t rd_frags[0]; /* buffer frags */
260 } WIRE_ATTR kib_rdma_desc_t;
265 lnet_hdr_t ibprm_hdr; /* portals header */
266 __u64 ibprm_cookie; /* opaque completion cookie */
267 } WIRE_ATTR kib_putreq_msg_t;
271 __u64 ibpam_src_cookie; /* reflected completion cookie */
272 __u64 ibpam_dst_cookie; /* opaque completion cookie */
273 kib_rdma_desc_t ibpam_rd; /* sender's sink buffer */
274 } WIRE_ATTR kib_putack_msg_t;
278 lnet_hdr_t ibgm_hdr; /* portals header */
279 __u64 ibgm_cookie; /* opaque completion cookie */
280 kib_rdma_desc_t ibgm_rd; /* rdma descriptor */
281 } WIRE_ATTR kib_get_msg_t;
285 __u64 ibcm_cookie; /* opaque completion cookie */
286 __s32 ibcm_status; /* < 0 failure: >= 0 length */
287 } WIRE_ATTR kib_completion_msg_t;
291 /* First 2 fields fixed FOR ALL TIME */
292 __u32 ibm_magic; /* I'm an openibnal message */
293 __u16 ibm_version; /* this is my version number */
295 __u8 ibm_type; /* msg type */
296 __u8 ibm_credits; /* returned credits */
297 __u32 ibm_nob; /* # bytes in whole message */
298 __u32 ibm_cksum; /* checksum (0 == no checksum) */
299 __u64 ibm_srcnid; /* sender's NID */
300 __u64 ibm_srcstamp; /* sender's incarnation */
301 __u64 ibm_dstnid; /* destination's NID */
302 __u64 ibm_dststamp; /* destination's incarnation */
305 kib_connparams_t connparams;
306 kib_immediate_msg_t immediate;
307 kib_putreq_msg_t putreq;
308 kib_putack_msg_t putack;
310 kib_completion_msg_t completion;
312 } WIRE_ATTR kib_msg_t;
314 #define IBLND_MSG_MAGIC LNET_PROTO_IB_MAGIC /* unique magic */
316 #define IBLND_MSG_VERSION 0x11
318 #define IBLND_MSG_CONNREQ 0xc0 /* connection request */
319 #define IBLND_MSG_CONNACK 0xc1 /* connection acknowledge */
320 #define IBLND_MSG_NOOP 0xd0 /* nothing (just credits) */
321 #define IBLND_MSG_IMMEDIATE 0xd1 /* immediate */
322 #define IBLND_MSG_PUT_REQ 0xd2 /* putreq (src->sink) */
323 #define IBLND_MSG_PUT_NAK 0xd3 /* completion (sink->src) */
324 #define IBLND_MSG_PUT_ACK 0xd4 /* putack (sink->src) */
325 #define IBLND_MSG_PUT_DONE 0xd5 /* completion (src->sink) */
326 #define IBLND_MSG_GET_REQ 0xd6 /* getreq (sink->src) */
327 #define IBLND_MSG_GET_DONE 0xd7 /* completion (src->sink: all OK) */
330 __u32 ibr_magic; /* sender's magic */
331 __u16 ibr_version; /* sender's version */
332 __u8 ibr_why; /* reject reason */
333 } WIRE_ATTR kib_rej_t;
336 /* connection rejection reasons */
337 #define IBLND_REJECT_CONN_RACE 1 /* You lost connection race */
338 #define IBLND_REJECT_NO_RESOURCES 2 /* Out of memory/conns etc */
339 #define IBLND_REJECT_FATAL 3 /* Anything else */
341 /***********************************************************************/
343 typedef struct kib_rx /* receive message */
345 struct list_head rx_list; /* queue for attention */
346 struct kib_conn *rx_conn; /* owning conn */
347 int rx_nob; /* # bytes received (-1 while posted) */
348 enum ib_wc_status rx_status; /* completion status */
349 kib_msg_t *rx_msg; /* message buffer (host vaddr) */
350 __u64 rx_msgaddr; /* message buffer (I/O addr) */
351 DECLARE_PCI_UNMAP_ADDR (rx_msgunmap); /* for dma_unmap_single() */
352 struct ib_recv_wr rx_wrq; /* receive work item... */
353 struct ib_sge rx_sge; /* ...and its memory */
356 #define IBLND_POSTRX_DONT_POST 0 /* don't post */
357 #define IBLND_POSTRX_NO_CREDIT 1 /* post: no credits */
358 #define IBLND_POSTRX_PEER_CREDIT 2 /* post: give peer back 1 credit */
359 #define IBLND_POSTRX_RSRVD_CREDIT 3 /* post: give myself back 1 reserved credit */
361 typedef struct kib_tx /* transmit message */
363 struct list_head tx_list; /* queue on idle_txs ibc_tx_queue etc. */
364 struct kib_conn *tx_conn; /* owning conn */
365 int tx_sending; /* # tx callbacks outstanding */
366 int tx_queued; /* queued for sending */
367 int tx_waiting; /* waiting for peer */
368 int tx_status; /* LNET completion status */
369 unsigned long tx_deadline; /* completion deadline */
370 __u64 tx_cookie; /* completion cookie */
371 lnet_msg_t *tx_lntmsg[2]; /* lnet msgs to finalize on completion */
372 kib_msg_t *tx_msg; /* message buffer (host vaddr) */
373 __u64 tx_msgaddr; /* message buffer (I/O addr) */
374 DECLARE_PCI_UNMAP_ADDR (tx_msgunmap); /* for dma_unmap_single() */
375 int tx_nwrq; /* # send work items */
376 #if IBLND_MAP_ON_DEMAND
377 struct ib_send_wr tx_wrq[2]; /* send work items... */
378 struct ib_sge tx_sge[2]; /* ...and their memory */
379 kib_rdma_desc_t tx_rd[1]; /* rdma descriptor */
380 __u64 *tx_pages; /* rdma phys page addrs */
381 struct ib_pool_fmr *tx_fmr; /* rdma mapping (mapped if != NULL) */
383 struct ib_send_wr *tx_wrq; /* send work items... */
384 struct ib_sge *tx_sge; /* ...and their memory */
385 kib_rdma_desc_t *tx_rd; /* rdma descriptor */
386 int tx_nfrags; /* # entries in... */
387 struct scatterlist *tx_frags; /* dma_map_sg descriptor */
388 int tx_dmadir; /* dma direction */
392 typedef struct kib_connvars
394 /* connection-in-progress variables */
398 typedef struct kib_conn
400 struct kib_peer *ibc_peer; /* owning peer */
401 struct list_head ibc_list; /* stash on peer's conn list */
402 struct list_head ibc_sched_list; /* schedule for attention */
403 __u64 ibc_incarnation; /* which instance of the peer */
404 atomic_t ibc_refcount; /* # users */
405 int ibc_state; /* what's happening */
406 int ibc_nsends_posted; /* # uncompleted sends */
407 int ibc_credits; /* # credits I have */
408 int ibc_outstanding_credits; /* # credits to return */
409 int ibc_reserved_credits;/* # ACK/DONE msg credits */
410 int ibc_comms_error; /* set on comms error */
411 int ibc_nrx:8; /* receive buffers owned */
412 int ibc_scheduled:1; /* scheduled for attention */
413 int ibc_ready:1; /* CQ callback fired */
414 unsigned long ibc_last_send; /* time of last send */
415 struct list_head ibc_early_rxs; /* rxs completed before ESTABLISHED */
416 struct list_head ibc_tx_noops; /* IBLND_MSG_NOOPs */
417 struct list_head ibc_tx_queue; /* sends that need a credit */
418 struct list_head ibc_tx_queue_nocred;/* sends that don't need a credit */
419 struct list_head ibc_tx_queue_rsrvd; /* sends that need to reserve an ACK/DONE msg */
420 struct list_head ibc_active_txs; /* active tx awaiting completion */
421 spinlock_t ibc_lock; /* serialise */
422 kib_rx_t *ibc_rxs; /* the rx descs */
423 kib_pages_t *ibc_rx_pages; /* premapped rx msg pages */
425 struct rdma_cm_id *ibc_cmid; /* CM id */
426 struct ib_cq *ibc_cq; /* completion queue */
428 kib_connvars_t *ibc_connvars; /* in-progress connection state */
431 #define IBLND_CONN_INIT 0 /* being intialised */
432 #define IBLND_CONN_ACTIVE_CONNECT 1 /* active sending req */
433 #define IBLND_CONN_PASSIVE_WAIT 2 /* passive waiting for rtu */
434 #define IBLND_CONN_ESTABLISHED 3 /* connection established */
435 #define IBLND_CONN_CLOSING 4 /* being closed */
436 #define IBLND_CONN_DISCONNECTED 5 /* disconnected */
438 typedef struct kib_peer
440 struct list_head ibp_list; /* stash on global peer list */
441 lnet_nid_t ibp_nid; /* who's on the other end(s) */
442 lnet_ni_t *ibp_ni; /* LNet interface */
443 atomic_t ibp_refcount; /* # users */
444 struct list_head ibp_conns; /* all active connections */
445 struct list_head ibp_tx_queue; /* msgs waiting for a conn */
446 int ibp_connecting; /* current active connection attempts */
447 int ibp_accepting; /* current passive connection attempts */
448 int ibp_error; /* errno on closing this peer */
449 cfs_time_t ibp_last_alive; /* when (in jiffies) I was last alive */
453 extern kib_data_t kiblnd_data;
454 extern kib_tunables_t kiblnd_tunables;
456 #define kiblnd_conn_addref(conn) \
458 CDEBUG(D_NET, "conn[%p] (%d)++\n", \
459 (conn), atomic_read(&(conn)->ibc_refcount)); \
460 LASSERT(atomic_read(&(conn)->ibc_refcount) > 0); \
461 atomic_inc(&(conn)->ibc_refcount); \
464 #define kiblnd_conn_decref(conn) \
466 unsigned long flags; \
468 CDEBUG(D_NET, "conn[%p] (%d)--\n", \
469 (conn), atomic_read(&(conn)->ibc_refcount)); \
470 LASSERT(atomic_read(&(conn)->ibc_refcount) > 0); \
471 if (atomic_dec_and_test(&(conn)->ibc_refcount)) { \
472 spin_lock_irqsave(&kiblnd_data.kib_connd_lock, flags); \
473 list_add_tail(&(conn)->ibc_list, \
474 &kiblnd_data.kib_connd_zombies); \
475 wake_up(&kiblnd_data.kib_connd_waitq); \
476 spin_unlock_irqrestore(&kiblnd_data.kib_connd_lock, flags); \
480 #define kiblnd_peer_addref(peer) \
482 CDEBUG(D_NET, "peer[%p] -> %s (%d)++\n", \
483 (peer), libcfs_nid2str((peer)->ibp_nid), \
484 atomic_read (&(peer)->ibp_refcount)); \
485 LASSERT(atomic_read(&(peer)->ibp_refcount) > 0); \
486 atomic_inc(&(peer)->ibp_refcount); \
489 #define kiblnd_peer_decref(peer) \
491 CDEBUG(D_NET, "peer[%p] -> %s (%d)--\n", \
492 (peer), libcfs_nid2str((peer)->ibp_nid), \
493 atomic_read (&(peer)->ibp_refcount)); \
494 LASSERT(atomic_read(&(peer)->ibp_refcount) > 0); \
495 if (atomic_dec_and_test(&(peer)->ibp_refcount)) \
496 kiblnd_destroy_peer(peer); \
499 static inline struct list_head *
500 kiblnd_nid2peerlist (lnet_nid_t nid)
502 unsigned int hash = ((unsigned int)nid) % kiblnd_data.kib_peer_hash_size;
504 return (&kiblnd_data.kib_peers [hash]);
508 kiblnd_peer_active (kib_peer_t *peer)
510 /* Am I in the peer hash table? */
511 return (!list_empty(&peer->ibp_list));
514 static inline kib_conn_t *
515 kiblnd_get_conn_locked (kib_peer_t *peer)
517 LASSERT (!list_empty(&peer->ibp_conns));
519 /* just return the first connection */
520 return list_entry(peer->ibp_conns.next, kib_conn_t, ibc_list);
524 kiblnd_send_keepalive(kib_conn_t *conn)
526 return (*kiblnd_tunables.kib_keepalive > 0) &&
527 time_after(jiffies, conn->ibc_last_send +
528 *kiblnd_tunables.kib_keepalive*HZ);
532 kiblnd_send_noop(kib_conn_t *conn)
534 LASSERT (conn->ibc_state >= IBLND_CONN_ESTABLISHED);
536 if (conn->ibc_outstanding_credits < IBLND_CREDIT_HIGHWATER &&
537 !kiblnd_send_keepalive(conn))
538 return 0; /* No need to send NOOP */
540 if (!list_empty(&conn->ibc_tx_noops) || /* NOOP already queued */
541 !list_empty(&conn->ibc_tx_queue_nocred) || /* can be piggybacked */
542 conn->ibc_credits == 0) /* no credit */
545 if (conn->ibc_credits == 1 && /* last credit reserved for */
546 conn->ibc_outstanding_credits == 0) /* giving back credits */
549 /* No tx to piggyback NOOP onto or no credit to send a tx */
550 return (list_empty(&conn->ibc_tx_queue) || conn->ibc_credits == 1);
554 kiblnd_abort_receives(kib_conn_t *conn)
556 ib_modify_qp(conn->ibc_cmid->qp,
557 &kiblnd_data.kib_error_qpa, IB_QP_STATE);
560 /* CAVEAT EMPTOR: We rely on descriptor alignment to allow us to use the
561 * lowest bits of the work request id to stash the work item type. */
563 #define IBLND_WID_TX 0
564 #define IBLND_WID_RDMA 1
565 #define IBLND_WID_RX 2
566 #define IBLND_WID_MASK 3UL
569 kiblnd_ptr2wreqid (void *ptr, int type)
571 unsigned long lptr = (unsigned long)ptr;
573 LASSERT ((lptr & IBLND_WID_MASK) == 0);
574 LASSERT ((type & ~IBLND_WID_MASK) == 0);
575 return (__u64)(lptr | type);
579 kiblnd_wreqid2ptr (__u64 wreqid)
581 return (void *)(((unsigned long)wreqid) & ~IBLND_WID_MASK);
585 kiblnd_wreqid2type (__u64 wreqid)
587 return (wreqid & IBLND_WID_MASK);
591 kiblnd_set_conn_state (kib_conn_t *conn, int state)
593 conn->ibc_state = state;
597 #if IBLND_MAP_ON_DEMAND
599 kiblnd_rd_size (kib_rdma_desc_t *rd)
605 kiblnd_rd_size (kib_rdma_desc_t *rd)
610 for (i = size = 0; i < rd->rd_nfrags; i++)
611 size += rd->rd_frags[i].rf_nob;
617 #ifdef HAVE_OFED_IB_DMA_MAP
619 static inline __u64 kiblnd_dma_map_single(struct ib_device *dev,
620 void *msg, size_t size,
621 enum dma_data_direction direction)
623 return ib_dma_map_single(dev, msg, size, direction);
626 static inline void kiblnd_dma_unmap_single(struct ib_device *dev,
627 __u64 addr, size_t size,
628 enum dma_data_direction direction)
630 ib_dma_unmap_single(dev, addr, size, direction);
633 #define KIBLND_UNMAP_ADDR_SET(p, m, a) do {} while (0)
634 #define KIBLND_UNMAP_ADDR(p, m, a) (a)
636 static inline int kiblnd_dma_map_sg(struct ib_device *dev,
637 struct scatterlist *sg, int nents,
638 enum dma_data_direction direction)
640 return ib_dma_map_sg(dev, sg, nents, direction);
643 static inline void kiblnd_dma_unmap_sg(struct ib_device *dev,
644 struct scatterlist *sg, int nents,
645 enum dma_data_direction direction)
647 ib_dma_unmap_sg(dev, sg, nents, direction);
650 static inline __u64 kiblnd_sg_dma_address(struct ib_device *dev,
651 struct scatterlist *sg)
653 return ib_sg_dma_address(dev, sg);
656 static inline unsigned int kiblnd_sg_dma_len(struct ib_device *dev,
657 struct scatterlist *sg)
659 return ib_sg_dma_len(dev, sg);
662 /* XXX We use KIBLND_CONN_PARAM(e) as writable buffer, it's not strictly
663 * right because OFED1.2 defines it as const, to use it we have to add
664 * (void *) cast to overcome "const" */
666 #define KIBLND_CONN_PARAM(e) ((e)->param.conn.private_data)
667 #define KIBLND_CONN_PARAM_LEN(e) ((e)->param.conn.private_data_len)
671 static inline dma_addr_t kiblnd_dma_map_single(struct ib_device *dev,
672 void *msg, size_t size,
673 enum dma_data_direction direction)
675 return dma_map_single(dev->dma_device, msg, size, direction);
678 static inline void kiblnd_dma_unmap_single(struct ib_device *dev,
679 dma_addr_t addr, size_t size,
680 enum dma_data_direction direction)
682 dma_unmap_single(dev->dma_device, addr, size, direction);
685 #define KIBLND_UNMAP_ADDR_SET(p, m, a) pci_unmap_addr_set(p, m, a)
686 #define KIBLND_UNMAP_ADDR(p, m, a) pci_unmap_addr(p, m)
688 static inline int kiblnd_dma_map_sg(struct ib_device *dev,
689 struct scatterlist *sg, int nents,
690 enum dma_data_direction direction)
692 return dma_map_sg(dev->dma_device, sg, nents, direction);
695 static inline void kiblnd_dma_unmap_sg(struct ib_device *dev,
696 struct scatterlist *sg, int nents,
697 enum dma_data_direction direction)
699 return dma_unmap_sg(dev->dma_device, sg, nents, direction);
703 static inline dma_addr_t kiblnd_sg_dma_address(struct ib_device *dev,
704 struct scatterlist *sg)
706 return sg_dma_address(sg);
710 static inline unsigned int kiblnd_sg_dma_len(struct ib_device *dev,
711 struct scatterlist *sg)
713 return sg_dma_len(sg);
716 #define KIBLND_CONN_PARAM(e) ((e)->private_data)
717 #define KIBLND_CONN_PARAM_LEN(e) ((e)->private_data_len)
721 int kiblnd_startup (lnet_ni_t *ni);
722 void kiblnd_shutdown (lnet_ni_t *ni);
723 int kiblnd_ctl (lnet_ni_t *ni, unsigned int cmd, void *arg);
725 int kiblnd_tunables_init(void);
726 void kiblnd_tunables_fini(void);
728 int kiblnd_connd (void *arg);
729 int kiblnd_scheduler(void *arg);
730 int kiblnd_thread_start (int (*fn)(void *arg), void *arg);
732 int kiblnd_alloc_pages (kib_pages_t **pp, int npages);
733 void kiblnd_free_pages (kib_pages_t *p);
735 int kiblnd_cm_callback(struct rdma_cm_id *cmid,
736 struct rdma_cm_event *event);
738 int kiblnd_create_peer (lnet_ni_t *ni, kib_peer_t **peerp, lnet_nid_t nid);
739 void kiblnd_destroy_peer (kib_peer_t *peer);
740 void kiblnd_destroy_dev (kib_dev_t *dev);
741 void kiblnd_unlink_peer_locked (kib_peer_t *peer);
742 void kiblnd_peer_alive (kib_peer_t *peer);
743 kib_peer_t *kiblnd_find_peer_locked (lnet_nid_t nid);
744 void kiblnd_peer_connect_failed (kib_peer_t *peer, int active, int error);
745 int kiblnd_close_stale_conns_locked (kib_peer_t *peer, __u64 incarnation);
747 void kiblnd_connreq_done(kib_conn_t *conn, int status);
748 kib_conn_t *kiblnd_create_conn (kib_peer_t *peer, struct rdma_cm_id *cmid,
750 void kiblnd_destroy_conn (kib_conn_t *conn);
751 void kiblnd_close_conn (kib_conn_t *conn, int error);
752 void kiblnd_close_conn_locked (kib_conn_t *conn, int error);
754 int kiblnd_init_rdma (lnet_ni_t *ni, kib_tx_t *tx, int type,
755 int nob, kib_rdma_desc_t *dstrd, __u64 dstcookie);
757 void kiblnd_queue_tx_locked (kib_tx_t *tx, kib_conn_t *conn);
758 void kiblnd_queue_tx (kib_tx_t *tx, kib_conn_t *conn);
759 void kiblnd_init_tx_msg (lnet_ni_t *ni, kib_tx_t *tx, int type, int body_nob);
760 void kiblnd_txlist_done (lnet_ni_t *ni, struct list_head *txlist, int status);
761 void kiblnd_check_sends (kib_conn_t *conn);
763 void kiblnd_qp_event(struct ib_event *event, void *arg);
764 void kiblnd_cq_event(struct ib_event *event, void *arg);
765 void kiblnd_cq_completion(struct ib_cq *cq, void *arg);
767 void kiblnd_init_msg (kib_msg_t *msg, int type, int body_nob);
768 void kiblnd_pack_msg (lnet_ni_t *ni, kib_msg_t *msg,
769 int credits, lnet_nid_t dstnid, __u64 dststamp);
770 int kiblnd_unpack_msg(kib_msg_t *msg, int nob);
771 int kiblnd_post_rx (kib_rx_t *rx, int credit);
773 int kiblnd_send(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
774 int kiblnd_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg, int delayed,
775 unsigned int niov, struct iovec *iov, lnet_kiov_t *kiov,
776 unsigned int offset, unsigned int mlen, unsigned int rlen);