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 (c) 2004, 2010, Oracle and/or its affiliates. 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/openiblnd/openiblnd.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>
72 #define DEBUG_SUBSYSTEM S_LND
74 #include <libcfs/libcfs.h>
75 #include <lnet/lnet.h>
76 #include <lnet/lib-lnet.h>
77 #include <lnet/lnet-sysctl.h>
79 #include <ts_ib_core.h>
81 #include <ts_ib_sa_client.h>
86 typedef struct ib_qp ib_qp_t;
87 typedef struct ib_mr ib_mr_t;
88 typedef struct ib_fmr ib_fmr_t;
89 typedef struct ib_pd ib_pd_t;
90 typedef struct ib_cq ib_cq_t;
91 typedef struct ib_fmr_pool ib_fmr_pool_t;
98 typedef void ib_fmr_t;
100 typedef void ib_cq_t;
101 typedef void ib_fmr_pool_t;
103 #define IB_ACCESS_LOCAL_WRITE TS_IB_ACCESS_LOCAL_WRITE
104 #define IB_WQ_SIGNAL_SELECTABLE TS_IB_ACCESS_LOCAL_WRITE
105 #define IB_TRANSPORT_RC TS_IB_TRANSPORT_RC
106 #define IB_QP_STATE_INIT TS_IB_QP_STATE_INIT
107 #define IB_QP_ATTRIBUTE_STATE TS_IB_QP_ATTRIBUTE_STATE
108 #define IB_QP_ATTRIBUTE_PORT TS_IB_QP_ATTRIBUTE_PORT
109 #define IB_QP_ATTRIBUTE_PKEY_INDEX TS_IB_QP_ATTRIBUTE_PKEY_INDEX
110 #define IB_QP_ATTRIBUTE_RDMA_ATOMIC_ENABLE TS_IB_QP_ATTRIBUTE_RDMA_ATOMIC_ENABLE
111 #define IB_ACCESS_LOCAL_WRITE TS_IB_ACCESS_LOCAL_WRITE
112 #define IB_ACCESS_REMOTE_WRITE TS_IB_ACCESS_REMOTE_WRITE
113 #define IB_ACCESS_REMOTE_READ TS_IB_ACCESS_REMOTE_READ
114 #define IB_CQ_CALLBACK_INTERRU TS_IB_CQ_CALLBACK_INTERRUPTPT
115 #define IB_CQ_PROVIDER_REARM TS_IB_CQ_PROVIDER_REARM
116 #define IB_CQ_CALLBACK_INTERRUPT TS_IB_CQ_CALLBACK_INTERRUPT
117 #define IB_COMPLETION_STATUS_SUCCESS TS_IB_COMPLETION_STATUS_SUCCESS
118 #define IB_OP_SEND TS_IB_OP_SEND
119 #define IB_OP_RDMA_WRITE TS_IB_OP_RDMA_WRITE
120 #define IB_OP_RDMA_READ TS_IB_OP_RDMA_READ
125 # define IBNAL_N_SCHED num_online_cpus() /* # schedulers */
127 # define IBNAL_N_SCHED 1 /* # schedulers */
131 //#define IBNAL_CALLBACK_CTXT IB_CQ_CALLBACK_PROCESS
132 #define IBNAL_CALLBACK_CTXT IB_CQ_CALLBACK_INTERRUPT
135 /* tunables fixed at compile time */
136 #define IBNAL_PEER_HASH_SIZE 101 /* # peer lists */
137 #define IBNAL_RESCHED 100 /* # scheduler loops before reschedule */
138 #define IBNAL_MSG_QUEUE_SIZE 8 /* # messages/RDMAs in-flight */
139 #define IBNAL_CREDIT_HIGHWATER 6 /* when to eagerly return credits */
140 #define IBNAL_MSG_SIZE (4<<10) /* max size of queued messages (inc hdr) */
141 #define IBNAL_RDMA_BASE 0x0eeb0000
144 #define IBNAL_RETRY 7 /* # times to retry */
145 #define IBNAL_RNR_RETRY 7 /* */
146 #define IBNAL_CM_RETRY 7 /* # times to retry connection */
147 #define IBNAL_FLOW_CONTROL 1
148 #define IBNAL_RESPONDER_RESOURCES 8
150 /************************/
151 /* derived constants... */
153 /* TX messages (shared by all connections) */
154 #define IBNAL_TX_MSGS() (*kibnal_tunables.kib_ntx)
155 #define IBNAL_TX_MSG_BYTES() (IBNAL_TX_MSGS() * IBNAL_MSG_SIZE)
156 #define IBNAL_TX_MSG_PAGES() ((IBNAL_TX_MSG_BYTES() + PAGE_SIZE - 1)/PAGE_SIZE)
158 /* RX messages (per connection) */
159 #define IBNAL_RX_MSGS (IBNAL_MSG_QUEUE_SIZE * 2)
160 #define IBNAL_RX_MSG_BYTES (IBNAL_RX_MSGS * IBNAL_MSG_SIZE)
161 #define IBNAL_RX_MSG_PAGES ((IBNAL_RX_MSG_BYTES + PAGE_SIZE - 1)/PAGE_SIZE)
163 /* we may have up to 2 completions per transmit +
164 1 completion per receive, per connection */
165 #define IBNAL_CQ_ENTRIES() ((2*IBNAL_TX_MSGS()) + \
166 (IBNAL_RX_MSGS * *kibnal_tunables.kib_concurrent_peers))
170 char **kib_ipif_basename; /* IPoIB interface base name */
171 int *kib_n_connd; /* # connection daemons */
172 int *kib_min_reconnect_interval; /* min connect retry seconds... */
173 int *kib_max_reconnect_interval; /* max connect retry seconds */
174 int *kib_concurrent_peers; /* max # peers */
175 int *kib_cksum; /* checksum kib_msg_t? */
176 int *kib_timeout; /* comms timeout (seconds) */
177 int *kib_keepalive; /* keepalive (seconds) */
178 int *kib_ntx; /* # tx descs */
179 int *kib_credits; /* # concurrent sends */
180 int *kib_peercredits; /* # concurrent sends to 1 peer */
182 cfs_sysctl_table_header_t *kib_sysctl; /* sysctl interface */
187 int ibp_npages; /* # pages */
188 int ibp_mapped; /* mapped? */
189 __u64 ibp_vaddr; /* mapped region vaddr */
190 __u32 ibp_lkey; /* mapped region lkey */
191 __u32 ibp_rkey; /* mapped region rkey */
192 ib_mr_t *ibp_handle; /* mapped region handle */
193 struct page *ibp_pages[0];
198 int kib_init; /* initialisation state */
199 __u64 kib_incarnation; /* which one am I */
200 int kib_shutdown; /* shut down? */
201 atomic_t kib_nthreads; /* # live threads */
202 lnet_ni_t *kib_ni; /* _the_ openib interface */
204 __u64 kib_svc_id; /* service number I listen on */
205 tTS_IB_GID kib_svc_gid; /* device/port GID */
206 __u16 kib_svc_pkey; /* device/port pkey */
208 void *kib_listen_handle; /* IB listen handle */
210 rwlock_t kib_global_lock; /* stabilize peer/conn ops */
212 struct list_head *kib_peers; /* hash table of all my known peers */
213 int kib_peer_hash_size; /* size of kib_peers */
214 int kib_nonewpeers; /* prevent new peers? */
215 atomic_t kib_npeers; /* # peers extant */
216 atomic_t kib_nconns; /* # connections extant */
218 struct list_head kib_reaper_conns; /* connections to reap */
219 wait_queue_head_t kib_reaper_waitq; /* reaper sleeps here */
220 unsigned long kib_reaper_waketime; /* when reaper will wake */
221 spinlock_t kib_reaper_lock; /* serialise */
223 struct list_head kib_connd_peers; /* peers waiting for a connection */
224 struct list_head kib_connd_acceptq; /* accepted sockets to handle */
225 wait_queue_head_t kib_connd_waitq; /* connection daemons sleep here */
226 int kib_connd_connecting; /* # connds connecting */
227 spinlock_t kib_connd_lock; /* serialise */
229 wait_queue_head_t kib_sched_waitq; /* schedulers sleep here */
230 struct list_head kib_sched_txq; /* tx requiring attention */
231 struct list_head kib_sched_rxq; /* rx requiring attention */
232 spinlock_t kib_sched_lock; /* serialise */
234 struct kib_tx *kib_tx_descs; /* all the tx descriptors */
235 kib_pages_t *kib_tx_pages; /* premapped tx msg pages */
237 struct list_head kib_idle_txs; /* idle tx descriptors */
238 __u64 kib_next_tx_cookie; /* RDMA completion cookie */
239 spinlock_t kib_tx_lock; /* serialise */
241 int kib_hca_idx; /* my HCA number */
242 struct ib_device *kib_device; /* "the" device */
243 struct ib_device_properties kib_device_props; /* its properties */
244 int kib_port; /* port on the device */
245 struct ib_port_properties kib_port_props; /* its properties */
246 ib_pd_t *kib_pd; /* protection domain */
248 ib_fmr_pool_t *kib_fmr_pool; /* fast memory region pool */
250 ib_cq_t *kib_cq; /* completion queue */
254 #define IBNAL_INIT_NOTHING 0
255 #define IBNAL_INIT_DATA 1
256 #define IBNAL_INIT_LIB 2
257 #define IBNAL_INIT_PD 3
258 #define IBNAL_INIT_FMR 4
259 #define IBNAL_INIT_TXD 5
260 #define IBNAL_INIT_CQ 6
261 #define IBNAL_INIT_ALL 7
263 typedef struct kib_acceptsock /* accepted socket queued for connd */
265 struct list_head ibas_list; /* queue for attention */
266 struct socket *ibas_sock; /* the accepted socket */
269 /************************************************************************
270 * IB Wire message format.
271 * These are sent in sender's byte order (i.e. receiver flips).
272 * They may be sent via TCP/IP (service ID,GID,PKEY query/response),
273 * as private data in the connection request/response, or "normally".
276 typedef struct kib_svcrsp /* service response */
278 __u64 ibsr_svc_id; /* service's id */
279 __u8 ibsr_svc_gid[16]; /* service's gid */
280 __u16 ibsr_svc_pkey; /* service's pkey */
281 } WIRE_ATTR kib_svcrsp_t;
283 typedef struct kib_connparams
285 __u32 ibcp_queue_depth;
286 } WIRE_ATTR kib_connparams_t;
301 __u32 rd_key; /* remote key */
302 __u32 rd_nob; /* # of bytes */
303 __u64 rd_addr; /* remote io vaddr */
304 } WIRE_ATTR kib_rdma_desc_t;
308 lnet_hdr_t ibim_hdr; /* portals header */
309 char ibim_payload[0]; /* piggy-backed payload */
310 } WIRE_ATTR kib_immediate_msg_t;
314 lnet_hdr_t ibrm_hdr; /* portals header */
315 __u64 ibrm_cookie; /* opaque completion cookie */
316 kib_rdma_desc_t ibrm_desc; /* where to suck/blow */
317 } WIRE_ATTR kib_rdma_msg_t;
321 __u64 ibcm_cookie; /* opaque completion cookie */
322 __u32 ibcm_status; /* completion status */
323 } WIRE_ATTR kib_completion_msg_t;
327 /* First 2 fields fixed FOR ALL TIME */
328 __u32 ibm_magic; /* I'm an openibnal message */
329 __u16 ibm_version; /* this is my version number */
331 __u8 ibm_type; /* msg type */
332 __u8 ibm_credits; /* returned credits */
333 __u32 ibm_nob; /* # bytes in whole message */
334 __u32 ibm_cksum; /* checksum (0 == no checksum) */
335 __u64 ibm_srcnid; /* sender's NID */
336 __u64 ibm_srcstamp; /* sender's incarnation */
337 __u64 ibm_dstnid; /* destination's NID */
338 __u64 ibm_dststamp; /* destination's incarnation */
341 kib_connparams_t connparams;
342 kib_immediate_msg_t immediate;
344 kib_completion_msg_t completion;
346 } WIRE_ATTR kib_msg_t;
348 #define IBNAL_MSG_MAGIC LNET_PROTO_OPENIB_MAGIC /* unique magic */
349 #define IBNAL_MSG_VERSION_RDMAREPLYNOTRSRVD 2 /* previous protocol version */
350 #define IBNAL_MSG_VERSION 3 /* current protocol version */
352 #define IBNAL_MSG_SVCQRY 0xb0 /* service query */
353 #define IBNAL_MSG_SVCRSP 0xb1 /* service response */
354 #define IBNAL_MSG_CONNREQ 0xc0 /* connection request */
355 #define IBNAL_MSG_CONNACK 0xc1 /* connection acknowledge */
356 #define IBNAL_MSG_NOOP 0xd0 /* nothing (just credits) */
357 #define IBNAL_MSG_IMMEDIATE 0xd1 /* portals hdr + payload */
358 #define IBNAL_MSG_PUT_RDMA 0xd2 /* portals PUT hdr + source rdma desc */
359 #define IBNAL_MSG_PUT_DONE 0xd3 /* signal PUT rdma completion */
360 #define IBNAL_MSG_GET_RDMA 0xd4 /* portals GET hdr + sink rdma desc */
361 #define IBNAL_MSG_GET_DONE 0xd5 /* signal GET rdma completion */
363 /***********************************************************************/
365 typedef struct kib_rx /* receive message */
367 struct list_head rx_list; /* queue for attention */
368 struct kib_conn *rx_conn; /* owning conn */
369 int rx_nob; /* # bytes received (-1 while posted) */
370 __u64 rx_vaddr; /* pre-mapped buffer (hca vaddr) */
371 kib_msg_t *rx_msg; /* pre-mapped buffer (host vaddr) */
372 struct ib_receive_param rx_sp; /* receive work item */
373 struct ib_gather_scatter rx_gl; /* and it's memory */
376 typedef struct kib_tx /* transmit message */
378 struct list_head tx_list; /* queue on idle_txs ibc_tx_queue etc. */
379 struct kib_conn *tx_conn; /* owning conn */
380 int tx_mapped; /* mapped for RDMA? */
381 int tx_sending; /* # tx callbacks outstanding */
382 int tx_status; /* completion status */
383 unsigned long tx_deadline; /* completion deadline */
384 int tx_passive_rdma; /* peer sucks/blows */
385 int tx_passive_rdma_wait; /* waiting for peer to complete */
386 __u64 tx_passive_rdma_cookie; /* completion cookie */
387 lnet_msg_t *tx_lntmsg[2]; /* ptl msgs to finalize on completion */
388 kib_md_t tx_md; /* RDMA mapping (active/passive) */
389 __u64 tx_vaddr; /* pre-mapped buffer (hca vaddr) */
390 kib_msg_t *tx_msg; /* pre-mapped buffer (host vaddr) */
391 int tx_nsp; /* # send work items */
392 struct ib_send_param tx_sp[2]; /* send work items... */
393 struct ib_gather_scatter tx_gl[2]; /* ...and their memory */
396 #define KIB_TX_UNMAPPED 0
397 #define KIB_TX_MAPPED 1
398 #define KIB_TX_MAPPED_FMR 2
400 typedef struct kib_connreq
402 /* active connection-in-progress state */
403 struct kib_conn *cr_conn;
407 kib_svcrsp_t cr_svcrsp;
408 struct ib_path_record cr_path;
409 struct ib_cm_active_param cr_connparam;
412 typedef struct kib_conn
414 struct kib_peer *ibc_peer; /* owning peer */
415 struct list_head ibc_list; /* stash on peer's conn list */
416 __u64 ibc_incarnation; /* which instance of the peer */
417 int ibc_version; /* peer protocol version */
418 atomic_t ibc_refcount; /* # users */
419 int ibc_state; /* what's happening */
420 int ibc_nsends_posted; /* # uncompleted sends */
421 int ibc_credits; /* # credits I have */
422 int ibc_outstanding_credits; /* # credits to return */
423 int ibc_reserved_credits; /* # credits for ACK/DONE msgs */
424 unsigned long ibc_last_send; /* time of last send */
425 struct list_head ibc_tx_queue_nocred; /* sends that don't need a credit */
426 struct list_head ibc_tx_queue_rsrvd; /* sends that need a reserved cred */
427 struct list_head ibc_tx_queue; /* send queue */
428 struct list_head ibc_active_txs; /* active tx awaiting completion */
429 spinlock_t ibc_lock; /* serialise */
430 kib_rx_t *ibc_rxs; /* the rx descs */
431 kib_pages_t *ibc_rx_pages; /* premapped rx msg pages */
432 ib_qp_t *ibc_qp; /* queue pair */
433 __u32 ibc_qpn; /* queue pair number */
434 tTS_IB_CM_COMM_ID ibc_comm_id; /* connection ID? */
435 kib_connreq_t *ibc_connreq; /* connection request state */
438 #define IBNAL_CONN_INIT_NOTHING 0 /* initial state */
439 #define IBNAL_CONN_INIT_QP 1 /* ibc_qp set up */
440 #define IBNAL_CONN_CONNECTING 2 /* started to connect */
441 #define IBNAL_CONN_ESTABLISHED 3 /* connection established */
442 #define IBNAL_CONN_DEATHROW 4 /* waiting to be closed */
443 #define IBNAL_CONN_ZOMBIE 5 /* waiting to be freed */
445 typedef struct kib_peer
447 struct list_head ibp_list; /* stash on global peer list */
448 struct list_head ibp_connd_list; /* schedule on kib_connd_peers */
449 lnet_nid_t ibp_nid; /* who's on the other end(s) */
450 __u32 ibp_ip; /* IP to query for peer conn params */
451 int ibp_port; /* port to qery for peer conn params */
452 __u64 ibp_incarnation; /* peer's incarnation */
453 atomic_t ibp_refcount; /* # users */
454 int ibp_persistence; /* "known" peer refs */
455 struct list_head ibp_conns; /* all active connections */
456 struct list_head ibp_tx_queue; /* msgs waiting for a conn */
457 int ibp_connecting; /* current active connection attempts */
458 int ibp_accepting; /* current passive connection attempts */
459 unsigned long ibp_reconnect_time; /* when reconnect may be attempted */
460 unsigned long ibp_reconnect_interval; /* exponential backoff */
461 int ibp_error; /* errno on closing this peer */
462 cfs_time_t ibp_last_alive; /* when (in jiffies) I was last alive */
465 extern kib_data_t kibnal_data;
466 extern kib_tunables_t kibnal_tunables;
468 /******************************************************************************/
470 /* these are purposely avoiding using local vars so they don't increase
471 * stack consumption. */
473 #define kibnal_conn_addref(conn) \
475 CDEBUG(D_NET, "conn[%p] (%d)++\n", \
476 (conn), atomic_read(&(conn)->ibc_refcount)); \
477 LASSERT(atomic_read(&(conn)->ibc_refcount) > 0); \
478 atomic_inc(&(conn)->ibc_refcount); \
481 #define kibnal_conn_decref(conn) \
483 unsigned long flags; \
485 CDEBUG(D_NET, "conn[%p] (%d)--\n", \
486 (conn), atomic_read(&(conn)->ibc_refcount)); \
487 LASSERT(atomic_read(&(conn)->ibc_refcount) > 0); \
488 if (atomic_dec_and_test(&(conn)->ibc_refcount)) { \
489 spin_lock_irqsave(&kibnal_data.kib_reaper_lock, flags); \
490 list_add_tail(&(conn)->ibc_list, \
491 &kibnal_data.kib_reaper_conns); \
492 wake_up(&kibnal_data.kib_reaper_waitq); \
493 spin_unlock_irqrestore(&kibnal_data.kib_reaper_lock, flags); \
497 #define kibnal_peer_addref(peer) \
499 CDEBUG(D_NET, "peer[%p] -> %s (%d)++\n", \
500 (peer), libcfs_nid2str((peer)->ibp_nid), \
501 atomic_read (&(peer)->ibp_refcount)); \
502 LASSERT(atomic_read(&(peer)->ibp_refcount) > 0); \
503 atomic_inc(&(peer)->ibp_refcount); \
506 #define kibnal_peer_decref(peer) \
508 CDEBUG(D_NET, "peer[%p] -> %s (%d)--\n", \
509 (peer), libcfs_nid2str((peer)->ibp_nid), \
510 atomic_read (&(peer)->ibp_refcount)); \
511 LASSERT(atomic_read(&(peer)->ibp_refcount) > 0); \
512 if (atomic_dec_and_test(&(peer)->ibp_refcount)) \
513 kibnal_destroy_peer(peer); \
516 /******************************************************************************/
518 static inline struct list_head *
519 kibnal_nid2peerlist (lnet_nid_t nid)
521 unsigned int hash = ((unsigned int)nid) % kibnal_data.kib_peer_hash_size;
523 return (&kibnal_data.kib_peers [hash]);
527 kibnal_peer_active(kib_peer_t *peer)
529 /* Am I in the peer hash table? */
530 return (!list_empty(&peer->ibp_list));
534 kibnal_queue_tx_locked (kib_tx_t *tx, kib_conn_t *conn)
538 LASSERT (tx->tx_nsp > 0); /* work items set up */
539 LASSERT (tx->tx_conn == NULL); /* only set here */
541 kibnal_conn_addref(conn);
543 tx->tx_deadline = jiffies + *kibnal_tunables.kib_timeout * HZ;
545 if (conn->ibc_version == IBNAL_MSG_VERSION_RDMAREPLYNOTRSRVD) {
546 /* All messages have simple credit control */
547 q = &conn->ibc_tx_queue;
549 LASSERT (conn->ibc_version == IBNAL_MSG_VERSION);
551 switch (tx->tx_msg->ibm_type) {
552 case IBNAL_MSG_PUT_RDMA:
553 case IBNAL_MSG_GET_RDMA:
554 /* RDMA request: reserve a buffer for the RDMA reply
556 q = &conn->ibc_tx_queue_rsrvd;
559 case IBNAL_MSG_PUT_DONE:
560 case IBNAL_MSG_GET_DONE:
561 /* RDMA completion: no credits; peer has reserved a
563 q = &conn->ibc_tx_queue_nocred;
567 case IBNAL_MSG_IMMEDIATE:
568 /* Otherwise: consume a credit before sending */
569 q = &conn->ibc_tx_queue;
578 list_add_tail(&tx->tx_list, q);
582 kibnal_send_keepalive(kib_conn_t *conn)
584 return (*kibnal_tunables.kib_keepalive > 0) &&
585 time_after(jiffies, conn->ibc_last_send +
586 *kibnal_tunables.kib_keepalive*HZ);
590 * We rely on tx/rx descriptor alignment to allow us to use the lowest bit
591 * of the work request id as a flag to determine if the completion is for a
592 * transmit or a receive. It seems that that the CQ entry's 'op' field
593 * isn't always set correctly on completions that occur after QP teardown. */
596 kibnal_ptr2wreqid (void *ptr, int isrx)
598 unsigned long lptr = (unsigned long)ptr;
600 LASSERT ((lptr & 1) == 0);
601 return (__u64)(lptr | (isrx ? 1 : 0));
605 kibnal_wreqid2ptr (__u64 wreqid)
607 return (void *)(((unsigned long)wreqid) & ~1UL);
611 kibnal_wreqid_is_rx (__u64 wreqid)
613 return (wreqid & 1) != 0;
616 #if (IB_NTXRXPARAMS == 3)
618 kibnal_ib_send(ib_qp_t *qp, struct ib_send_param *p)
620 return ib_send(qp, p, 1);
624 kibnal_ib_receive(ib_qp_t *qp, struct ib_receive_param *p)
626 return ib_receive(qp, p, 1);
628 #elif (IB_NTXRXPARAMS == 4)
630 kibnal_ib_send(ib_qp_t *qp, struct ib_send_param *p)
632 return ib_send(qp, p, 1, NULL);
636 kibnal_ib_receive(ib_qp_t *qp, struct ib_receive_param *p)
638 return ib_receive(qp, p, 1, NULL);
641 #error "IB_NTXRXPARAMS not set correctly"
644 int kibnal_startup (lnet_ni_t *ni);
645 void kibnal_shutdown (lnet_ni_t *ni);
646 int kibnal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
647 int kibnal_send (lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
648 int kibnal_eager_recv (lnet_ni_t *ni, void *private,
649 lnet_msg_t *lntmsg, void **new_private);
650 int kibnal_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
651 int delayed, unsigned int niov,
652 struct iovec *iov, lnet_kiov_t *kiov,
653 unsigned int offset, unsigned int mlen, unsigned int rlen);
654 int kibnal_accept(lnet_ni_t *ni, struct socket *sock);
656 extern void kibnal_init_msg(kib_msg_t *msg, int type, int body_nob);
657 extern void kibnal_pack_msg(kib_msg_t *msg, int version, int credits,
658 lnet_nid_t dstnid, __u64 dststamp);
659 extern int kibnal_unpack_msg(kib_msg_t *msg, int expected_version, int nob);
660 extern void kibnal_handle_svcqry (struct socket *sock);
661 extern int kibnal_make_svcqry (kib_conn_t *conn);
662 extern void kibnal_free_acceptsock (kib_acceptsock_t *as);
663 extern int kibnal_create_peer (kib_peer_t **peerp, lnet_nid_t nid);
664 extern void kibnal_destroy_peer (kib_peer_t *peer);
665 extern int kibnal_add_persistent_peer(lnet_nid_t nid, __u32 ip, int port);
666 extern int kibnal_del_peer (lnet_nid_t nid);
667 extern kib_peer_t *kibnal_find_peer_locked (lnet_nid_t nid);
668 extern void kibnal_unlink_peer_locked (kib_peer_t *peer);
669 extern void kibnal_peer_alive(kib_peer_t *peer);
670 extern int kibnal_close_stale_conns_locked (kib_peer_t *peer,
672 extern kib_conn_t *kibnal_create_conn (void);
673 extern void kibnal_destroy_conn (kib_conn_t *conn);
674 extern int kibnal_alloc_pages (kib_pages_t **pp, int npages, int access);
675 extern void kibnal_free_pages (kib_pages_t *p);
677 extern void kibnal_check_sends (kib_conn_t *conn);
679 extern tTS_IB_CM_CALLBACK_RETURN
680 kibnal_bad_conn_callback (tTS_IB_CM_EVENT event, tTS_IB_CM_COMM_ID cid,
681 void *param, void *arg);
682 extern tTS_IB_CM_CALLBACK_RETURN
683 kibnal_conn_callback (tTS_IB_CM_EVENT event, tTS_IB_CM_COMM_ID cid,
684 void *param, void *arg);
685 extern tTS_IB_CM_CALLBACK_RETURN
686 kibnal_passive_conn_callback (tTS_IB_CM_EVENT event, tTS_IB_CM_COMM_ID cid,
687 void *param, void *arg);
689 extern void kibnal_close_conn_locked (kib_conn_t *conn, int error);
690 extern void kibnal_destroy_conn (kib_conn_t *conn);
691 extern int kibnal_thread_start (int (*fn)(void *arg), void *arg);
692 extern int kibnal_scheduler(void *arg);
693 extern int kibnal_connd (void *arg);
694 extern int kibnal_reaper (void *arg);
695 extern void kibnal_callback (ib_cq_t *cq, struct ib_cq_entry *e, void *arg);
696 extern void kibnal_txlist_done (struct list_head *txlist, int status);
697 extern void kibnal_init_tx_msg (kib_tx_t *tx, int type, int body_nob);
698 extern int kibnal_close_conn (kib_conn_t *conn, int why);
699 extern void kibnal_start_active_rdma (int type, int status,
700 kib_rx_t *rx, lnet_msg_t *lntmsg,
702 struct iovec *iov, lnet_kiov_t *kiov,
703 int offset, int nob);
705 extern int kibnal_tunables_init(void);
706 extern void kibnal_tunables_fini(void);