1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2004 Cluster File Systems, Inc.
5 * Author: Eric Barton <eric@bartonsoftware.com>
7 * This file is part of Lustre, http://www.lustre.org.
9 * Lustre is free software; you can redistribute it and/or
10 * modify it under the terms of version 2 of the GNU General Public
11 * License as published by the Free Software Foundation.
13 * Lustre is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with Lustre; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
25 # define EXPORT_SYMTAB
27 #ifndef AUTOCONF_INCLUDED
28 #include <linux/config.h>
30 #include <linux/module.h>
31 #include <linux/kernel.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/errno.h>
36 #include <linux/smp_lock.h>
37 #include <linux/unistd.h>
38 #include <linux/uio.h>
40 #include <asm/system.h>
41 #include <asm/uaccess.h>
44 #include <linux/init.h>
46 #include <linux/file.h>
47 #include <linux/stat.h>
48 #include <linux/list.h>
49 #include <linux/kmod.h>
50 #include <linux/sysctl.h>
52 #define DEBUG_SUBSYSTEM S_LND
54 #include <libcfs/kp30.h>
55 #include <lnet/lnet.h>
56 #include <lnet/lib-lnet.h>
58 #include <linux/iba/ibt.h>
60 #define GCC_VERSION (__GNUC__ * 10000 \
61 + __GNUC_MINOR__ * 100 \
62 + __GNUC_PATCHLEVEL__)
64 /* Test for GCC > 3.2.2 */
65 #if GCC_VERSION <= 30202
66 /* GCC 3.2.2, and presumably several versions before it, will
67 * miscompile this driver. See
68 * http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9853. */
69 #error Invalid GCC version. Must use GCC >= 3.2.3
73 # define IBNAL_N_SCHED num_online_cpus() /* # schedulers */
75 # define IBNAL_N_SCHED 1 /* # schedulers */
78 #define IBNAL_USE_FMR 0 /* map on demand v. use whole mem mapping */
79 #define KIBLND_DETAILED_DEBUG 0
81 /* tunables fixed at compile time */
82 #define IBNAL_PEER_HASH_SIZE 101 /* # peer lists */
83 #define IBNAL_RESCHED 100 /* # scheduler loops before reschedule */
84 #define IBNAL_MSG_QUEUE_SIZE 8 /* # messages/RDMAs in-flight */
85 #define IBNAL_CREDIT_HIGHWATER 7 /* when to eagerly return credits */
86 #define IBNAL_MSG_SIZE (4<<10) /* max size of queued messages (inc hdr) */
87 #define IBNAL_RDMA_BASE 0x0eeb0000
88 #define IBNAL_STARTING_PSN 1
91 /* 7 indicates infinite retry attempts, Infinicon recommended 5 */
92 #define IBNAL_RETRY 5 /* # times to retry */
93 #define IBNAL_RNR_RETRY 5 /* */
94 #define IBNAL_CM_RETRY 5 /* # times to retry connection */
95 #define IBNAL_FLOW_CONTROL 1
96 #define IBNAL_ACK_TIMEOUT 20 /* supposedly 4 secs */
97 #define IBNAL_EE_FLOW 1
98 #define IBNAL_LOCAL_SUB 1
99 #define IBNAL_FAILOVER_ACCEPTED 0
101 /************************/
102 /* derived constants... */
104 /* TX messages (shared by all connections) */
105 #define IBNAL_TX_MSGS() (*kibnal_tunables.kib_ntx)
106 #define IBNAL_TX_MSG_BYTES() (IBNAL_TX_MSGS() * IBNAL_MSG_SIZE)
107 #define IBNAL_TX_MSG_PAGES() ((IBNAL_TX_MSG_BYTES() + PAGE_SIZE - 1)/PAGE_SIZE)
110 # define IBNAL_MAX_RDMA_FRAGS 1
111 # define IBNAL_CONCURRENT_SENDS IBNAL_RX_MSGS
113 # define IBNAL_MAX_RDMA_FRAGS LNET_MAX_IOV
114 # define IBNAL_CONCURRENT_SENDS IBNAL_MSG_QUEUE_SIZE
117 /* RX messages (per connection) */
118 #define IBNAL_RX_MSGS (IBNAL_MSG_QUEUE_SIZE * 2)
119 #define IBNAL_RX_MSG_BYTES (IBNAL_RX_MSGS * IBNAL_MSG_SIZE)
120 #define IBNAL_RX_MSG_PAGES ((IBNAL_RX_MSG_BYTES + PAGE_SIZE - 1)/PAGE_SIZE)
122 #define IBNAL_CQ_ENTRIES() (IBNAL_TX_MSGS() * (1 + IBNAL_MAX_RDMA_FRAGS) + \
123 (IBNAL_RX_MSGS * *kibnal_tunables.kib_concurrent_peers))
127 char **kib_hca_basename; /* HCA base name */
128 char **kib_ipif_basename; /* IPoIB interface base name */
129 char **kib_service_name; /* global service name */
130 unsigned int *kib_service_number; /* global service number */
131 int *kib_min_reconnect_interval; /* min connect retry seconds... */
132 int *kib_max_reconnect_interval; /* max connect retry seconds */
133 int *kib_concurrent_peers; /* max # peers */
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 int *kib_sd_retries; /* # concurrent sends to 1 peer */
141 int *kib_concurrent_sends; /* send work queue sizing */
142 #if CONFIG_SYSCTL && !CFS_SYSFS_MODULE_PARM
143 cfs_sysctl_table_header_t *kib_sysctl; /* sysctl interface */
147 /* NB The Infinicon stack has specific typedefs for some things
148 * (e.g. IB_{L,R}_KEY), that just map back to __u32 etc */
151 int ibp_npages; /* # pages */
152 struct page *ibp_pages[0];
165 int kib_init; /* initialisation state */
166 __u64 kib_incarnation; /* which one am I */
167 int kib_shutdown; /* shut down? */
168 atomic_t kib_nthreads; /* # live threads */
169 lnet_ni_t *kib_ni; /* _the_ iib instance */
171 __u64 kib_port_guid; /* my GUID (lo 64 of GID)*/
172 __u16 kib_port_pkey; /* my pkey, whatever that is */
173 struct semaphore kib_listener_signal; /* signal completion */
174 IB_HANDLE kib_listener_cep; /* connection end point */
176 rwlock_t kib_global_lock; /* stabilize peer/conn ops */
177 int kib_ready; /* CQ callback fired */
178 int kib_checking_cq; /* a scheduler is checking the CQ */
180 struct list_head *kib_peers; /* hash table of all my known peers */
181 int kib_peer_hash_size; /* size of kib_peers */
182 atomic_t kib_npeers; /* # peers extant */
183 atomic_t kib_nconns; /* # connections extant */
185 struct list_head kib_connd_zombies; /* connections to free */
186 struct list_head kib_connd_conns; /* connections to progress */
187 struct list_head kib_connd_peers; /* peers waiting for a connection */
188 wait_queue_head_t kib_connd_waitq; /* connection daemon sleep here */
189 spinlock_t kib_connd_lock; /* serialise */
191 wait_queue_head_t kib_sched_waitq; /* schedulers sleep here */
192 spinlock_t kib_sched_lock; /* serialise */
194 struct kib_tx *kib_tx_descs; /* all the tx descriptors */
195 kib_pages_t *kib_tx_pages; /* premapped tx msg pages */
197 struct list_head kib_idle_txs; /* idle tx descriptors */
198 __u64 kib_next_tx_cookie; /* RDMA completion cookie */
199 spinlock_t kib_tx_lock; /* serialise */
201 IB_HANDLE kib_hca; /* The HCA */
202 int kib_port; /* port on the device */
203 IB_HANDLE kib_pd; /* protection domain */
204 IB_HANDLE kib_sd; /* SD handle */
205 IB_HANDLE kib_cq; /* completion queue */
206 kib_md_t kib_whole_mem; /* whole-mem registration */
208 int kib_hca_idx; /* my HCA number */
209 uint64 kib_hca_guids[8]; /* all the HCA guids */
210 IB_CA_ATTRIBUTES kib_hca_attrs; /* where to get HCA attrs */
212 COMMAND_CONTROL_PARAMETERS kib_sdretry; /* control SD query retries */
215 #define IBNAL_INIT_NOTHING 0
216 #define IBNAL_INIT_DATA 1
217 #define IBNAL_INIT_LIB 2
218 #define IBNAL_INIT_HCA 3
219 #define IBNAL_INIT_PORTATTRS 4
220 #define IBNAL_INIT_SD 5
221 #define IBNAL_INIT_PD 6
222 #define IBNAL_INIT_MD 7
223 #define IBNAL_INIT_TXD 8
224 #define IBNAL_INIT_CQ 9
225 #define IBNAL_INIT_ALL 10
227 /************************************************************************
228 * Wire message structs.
229 * These are sent in sender's byte order (i.e. receiver flips).
230 * CAVEAT EMPTOR: other structs communicated between nodes (e.g. MAD
231 * private data and SM service info), is LE on the wire.
234 typedef struct kib_connparams
236 __u32 ibcp_queue_depth;
237 __u32 ibcp_max_msg_size;
238 __u32 ibcp_max_frags;
239 } WIRE_ATTR kib_connparams_t;
243 lnet_hdr_t ibim_hdr; /* portals header */
244 char ibim_payload[0]; /* piggy-backed payload */
245 } WIRE_ATTR kib_immediate_msg_t;
250 __u64 rd_addr; /* IO VMA address */
251 __u32 rd_nob; /* # of bytes */
252 __u32 rd_key; /* remote key */
253 } WIRE_ATTR kib_rdma_desc_t;
257 __u32 rf_nob; /* # of bytes */
258 __u64 rf_addr; /* remote io vaddr */
259 } WIRE_ATTR kib_rdma_frag_t;
263 __u32 rd_key; /* local/remote key */
264 __u32 rd_nfrag; /* # fragments */
265 kib_rdma_frag_t rd_frags[0]; /* buffer frags */
266 } WIRE_ATTR kib_rdma_desc_t;
271 lnet_hdr_t ibprm_hdr; /* LNET header */
272 __u64 ibprm_cookie; /* opaque completion cookie */
273 } WIRE_ATTR kib_putreq_msg_t;
277 __u64 ibpam_src_cookie; /* reflected completion cookie */
278 __u64 ibpam_dst_cookie; /* opaque completion cookie */
279 kib_rdma_desc_t ibpam_rd; /* sender's sink buffer */
280 } WIRE_ATTR kib_putack_msg_t;
284 lnet_hdr_t ibgm_hdr; /* LNET header */
285 __u64 ibgm_cookie; /* opaque completion cookie */
286 kib_rdma_desc_t ibgm_rd; /* sender's sink buffer */
287 } WIRE_ATTR kib_get_msg_t;
291 __u64 ibcm_cookie; /* opaque completion cookie */
292 __u32 ibcm_status; /* completion status */
293 } WIRE_ATTR kib_completion_msg_t;
297 /* First 2 fields fixed FOR ALL TIME */
298 __u32 ibm_magic; /* I'm an openibnal message */
299 __u16 ibm_version; /* this is my version number */
301 __u8 ibm_type; /* msg type */
302 __u8 ibm_credits; /* returned credits */
303 __u32 ibm_nob; /* # bytes in whole message */
304 __u32 ibm_cksum; /* checksum (0 == no checksum) */
305 __u64 ibm_srcnid; /* sender's NID */
306 __u64 ibm_srcstamp; /* sender's incarnation */
307 __u64 ibm_dstnid; /* destination's NID */
308 __u64 ibm_dststamp; /* destination's incarnation */
309 __u64 ibm_seq; /* sequence number */
312 kib_connparams_t connparams;
313 kib_immediate_msg_t immediate;
314 kib_putreq_msg_t putreq;
315 kib_putack_msg_t putack;
317 kib_completion_msg_t completion;
319 } WIRE_ATTR kib_msg_t;
321 #define IBNAL_MSG_MAGIC LNET_PROTO_IIB_MAGIC /* unique magic */
322 #define IBNAL_MSG_VERSION 2 /* current protocol version */
323 #define IBNAL_MSG_VERSION_RDMAREPLYNOTRSRVD 1 /* previous version */
325 #define IBNAL_MSG_CONNREQ 0xc0 /* connection request */
326 #define IBNAL_MSG_CONNACK 0xc1 /* connection acknowledge */
327 #define IBNAL_MSG_NOOP 0xd0 /* nothing (just credits) */
328 #define IBNAL_MSG_IMMEDIATE 0xd1 /* immediate */
329 #define IBNAL_MSG_PUT_REQ 0xd2 /* putreq (src->sink) */
330 #define IBNAL_MSG_PUT_NAK 0xd3 /* completion (sink->src) */
331 #define IBNAL_MSG_PUT_ACK 0xd4 /* putack (sink->src) */
332 #define IBNAL_MSG_PUT_DONE 0xd5 /* completion (src->sink) */
333 #define IBNAL_MSG_GET_REQ 0xd6 /* getreq (sink->src) */
334 #define IBNAL_MSG_GET_DONE 0xd7 /* completion (src->sink: all OK) */
336 /* connection rejection reasons */
337 #define IBNAL_REJECT_CONN_RACE 0 /* You lost connection race */
338 #define IBNAL_REJECT_NO_RESOURCES 1 /* Out of memory/conns etc */
339 #define IBNAL_REJECT_FATAL 2 /* 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 __u64 rx_hca_msg; /* pre-mapped buffer (hca vaddr) */
349 kib_msg_t *rx_msg; /* pre-mapped buffer (host vaddr) */
351 IB_LOCAL_DATASEGMENT rx_gl; /* and its memory */
354 typedef struct kib_tx /* transmit message */
356 struct list_head tx_list; /* queue on idle_txs ibc_tx_queue etc. */
357 struct kib_conn *tx_conn; /* owning conn */
358 int tx_mapped; /* mapped for RDMA? */
359 int tx_sending; /* # tx callbacks outstanding */
360 int tx_queued; /* queued for sending */
361 int tx_waiting; /* waiting for peer */
362 int tx_status; /* completion status */
363 unsigned long tx_deadline; /* completion deadline */
364 __u64 tx_cookie; /* completion cookie */
365 lnet_msg_t *tx_lntmsg[2]; /* lnet msgs to finalize on completion */
366 kib_msg_t *tx_msg; /* pre-mapped buffer (host vaddr) */
367 __u64 tx_hca_msg; /* pre-mapped buffer (HCA vaddr) */
368 int tx_nwrq; /* # send work items */
370 IB_WORK_REQ2 tx_wrq[2]; /* send work items... */
371 IB_LOCAL_DATASEGMENT tx_gl[2]; /* ...and their memory */
372 kib_rdma_desc_t tx_rd[1]; /* rdma descriptor */
373 kib_md_t tx_md; /* mapping */
374 __u64 *tx_pages; /* page phys addrs */
376 IB_WORK_REQ2 *tx_wrq; /* send work items... */
377 IB_LOCAL_DATASEGMENT *tx_gl; /* ...and their memory */
378 kib_rdma_desc_t *tx_rd; /* rdma descriptor (src buffers) */
384 /* scratchpad during connection establishment */
385 IB_QP_ATTRIBUTES_QUERY cv_qpattrs;
387 IB_SERVICE_RECORD cv_svcrec;
388 IB_PATH_RECORD cv_path;
389 CM_CONN_INFO cv_cmci;
392 typedef struct kib_conn
394 struct kib_peer *ibc_peer; /* owning peer */
395 struct list_head ibc_list; /* stash on peer's conn list */
396 __u64 ibc_incarnation; /* which instance of the peer */
397 __u64 ibc_txseq; /* tx sequence number */
398 __u64 ibc_rxseq; /* rx sequence number */
399 __u32 ibc_version; /* peer protocol version */
400 atomic_t ibc_refcount; /* # users */
401 int ibc_state; /* what's happening */
402 int ibc_nsends_posted; /* # uncompleted sends */
403 int ibc_credits; /* # credits I have */
404 int ibc_outstanding_credits; /* # credits to return */
405 int ibc_reserved_credits; /* # credits for ACK/DONE msgs */
406 unsigned long ibc_last_send; /* time of last send */
407 struct list_head ibc_early_rxs; /* rxs completed before ESTABLISHED */
408 struct list_head ibc_tx_queue_nocred; /* sends that don't need a cred */
409 struct list_head ibc_tx_queue_rsrvd; /* sends that need a reserved cred */
410 struct list_head ibc_tx_queue; /* send queue */
411 struct list_head ibc_active_txs; /* active tx awaiting completion */
412 spinlock_t ibc_lock; /* serialise */
413 kib_rx_t *ibc_rxs; /* the rx descs */
414 kib_pages_t *ibc_rx_pages; /* premapped rx msg pages */
415 IB_HANDLE ibc_qp; /* queue pair */
416 IB_HANDLE ibc_cep; /* CM endpoint */
417 kib_connvars_t *ibc_cvars; /* connection scratchpad */
420 #define IBNAL_CONN_INIT_NOTHING 0 /* initial state */
421 #define IBNAL_CONN_INIT_QP 1 /* ibc_qp set up */
422 #define IBNAL_CONN_CONNECTING 2 /* started to connect */
423 #define IBNAL_CONN_ESTABLISHED 3 /* connection established */
424 #define IBNAL_CONN_DISCONNECTING 4 /* to send disconnect req */
425 #define IBNAL_CONN_DISCONNECTED 5 /* no more QP or CM traffic */
427 /* types of connection */
428 #define IBNAL_CONN_ACTIVE 0 /* active connect */
429 #define IBNAL_CONN_PASSIVE 1 /* passive connect */
430 #define IBNAL_CONN_WAITING 2 /* waiting for connect */
432 typedef struct kib_peer
434 struct list_head ibp_list; /* stash on global peer list */
435 struct list_head ibp_connd_list; /* schedule on kib_connd_peers */
436 lnet_nid_t ibp_nid; /* who's on the other end(s) */
437 atomic_t ibp_refcount; /* # users */
438 int ibp_persistence; /* "known" peer refs */
439 int ibp_version; /* protocol version */
440 struct list_head ibp_conns; /* all active connections */
441 struct list_head ibp_tx_queue; /* msgs waiting for a conn */
442 int ibp_connecting; /* active connects in progress */
443 int ibp_accepting; /* passive connects in progress */
444 int ibp_passivewait; /* waiting for peer to connect */
445 unsigned long ibp_passivewait_deadline; /* when passive wait must complete */
446 unsigned long ibp_reconnect_time; /* when reconnect may be attempted */
447 unsigned long ibp_reconnect_interval; /* exponential backoff */
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 kibnal_data;
454 extern kib_tunables_t kibnal_tunables;
456 /******************************************************************************/
458 /* these are purposely avoiding using local vars so they don't increase
459 * stack consumption. */
461 #define kibnal_conn_addref(conn) \
463 CDEBUG(D_NET, "conn[%p] (%d)++\n", \
464 (conn), atomic_read(&(conn)->ibc_refcount)); \
465 LASSERT(atomic_read(&(conn)->ibc_refcount) > 0); \
466 atomic_inc(&(conn)->ibc_refcount); \
469 #define kibnal_conn_decref(conn) \
471 unsigned long flags; \
473 CDEBUG(D_NET, "conn[%p] (%d)--\n", \
474 (conn), atomic_read(&(conn)->ibc_refcount)); \
475 LASSERT(atomic_read(&(conn)->ibc_refcount) > 0); \
476 if (atomic_dec_and_test(&(conn)->ibc_refcount)) { \
477 spin_lock_irqsave(&kibnal_data.kib_connd_lock, flags); \
478 list_add_tail(&(conn)->ibc_list, \
479 &kibnal_data.kib_connd_zombies); \
480 wake_up(&kibnal_data.kib_connd_waitq); \
481 spin_unlock_irqrestore(&kibnal_data.kib_connd_lock, flags); \
485 #define kibnal_peer_addref(peer) \
487 CDEBUG(D_NET, "peer[%p] -> %s (%d)++\n", \
488 (peer), libcfs_nid2str((peer)->ibp_nid), \
489 atomic_read (&(peer)->ibp_refcount)); \
490 LASSERT(atomic_read(&(peer)->ibp_refcount) > 0); \
491 atomic_inc(&(peer)->ibp_refcount); \
494 #define kibnal_peer_decref(peer) \
496 CDEBUG(D_NET, "peer[%p] -> %s (%d)--\n", \
497 (peer), libcfs_nid2str((peer)->ibp_nid), \
498 atomic_read (&(peer)->ibp_refcount)); \
499 LASSERT(atomic_read(&(peer)->ibp_refcount) > 0); \
500 if (atomic_dec_and_test(&(peer)->ibp_refcount)) \
501 kibnal_destroy_peer(peer); \
504 /******************************************************************************/
506 static inline struct list_head *
507 kibnal_nid2peerlist (lnet_nid_t nid)
509 unsigned int hash = ((unsigned int)nid) % kibnal_data.kib_peer_hash_size;
511 return (&kibnal_data.kib_peers [hash]);
515 kibnal_peer_active(kib_peer_t *peer)
517 /* Am I in the peer hash table? */
518 return (!list_empty(&peer->ibp_list));
522 kibnal_peer_connecting(kib_peer_t *peer)
524 /* Am I expecting a connection to materialise? */
525 return (peer->ibp_connecting != 0 ||
526 peer->ibp_accepting != 0 ||
527 peer->ibp_passivewait);
531 kibnal_queue_tx_locked (kib_tx_t *tx, kib_conn_t *conn)
535 LASSERT (tx->tx_nwrq > 0); /* work items set up */
536 LASSERT (!tx->tx_queued); /* not queued for sending already */
539 tx->tx_deadline = jiffies + (*kibnal_tunables.kib_timeout * HZ);
541 if (tx->tx_conn == NULL) {
542 kibnal_conn_addref(conn);
544 LASSERT (tx->tx_msg->ibm_type != IBNAL_MSG_PUT_DONE);
546 LASSERT (tx->tx_conn == conn);
547 LASSERT (tx->tx_msg->ibm_type == IBNAL_MSG_PUT_DONE);
550 if (conn->ibc_version == IBNAL_MSG_VERSION_RDMAREPLYNOTRSRVD) {
551 /* All messages have simple credit control */
552 q = &conn->ibc_tx_queue;
554 LASSERT (conn->ibc_version == IBNAL_MSG_VERSION);
556 switch (tx->tx_msg->ibm_type) {
557 case IBNAL_MSG_PUT_REQ:
558 case IBNAL_MSG_GET_REQ:
559 /* RDMA request: reserve a buffer for the RDMA reply
561 q = &conn->ibc_tx_queue_rsrvd;
564 case IBNAL_MSG_PUT_NAK:
565 case IBNAL_MSG_PUT_ACK:
566 case IBNAL_MSG_PUT_DONE:
567 case IBNAL_MSG_GET_DONE:
568 /* RDMA reply/completion: no credits; peer has reserved
570 q = &conn->ibc_tx_queue_nocred;
574 case IBNAL_MSG_IMMEDIATE:
575 /* Otherwise: consume a credit before sending */
576 q = &conn->ibc_tx_queue;
585 list_add_tail(&tx->tx_list, q);
589 kibnal_send_keepalive(kib_conn_t *conn)
591 return (*kibnal_tunables.kib_keepalive > 0) &&
592 time_after(jiffies, conn->ibc_last_send +
593 *kibnal_tunables.kib_keepalive*HZ);
596 #define KIBNAL_SERVICE_KEY_MASK (IB_SERVICE_RECORD_COMP_SERVICENAME | \
597 IB_SERVICE_RECORD_COMP_SERVICEDATA8_1 | \
598 IB_SERVICE_RECORD_COMP_SERVICEDATA8_2 | \
599 IB_SERVICE_RECORD_COMP_SERVICEDATA8_3 | \
600 IB_SERVICE_RECORD_COMP_SERVICEDATA8_4 | \
601 IB_SERVICE_RECORD_COMP_SERVICEDATA8_5 | \
602 IB_SERVICE_RECORD_COMP_SERVICEDATA8_6 | \
603 IB_SERVICE_RECORD_COMP_SERVICEDATA8_7 | \
604 IB_SERVICE_RECORD_COMP_SERVICEDATA8_8)
607 kibnal_service_nid_field(IB_SERVICE_RECORD *srv)
609 /* must be consistent with KIBNAL_SERVICE_KEY_MASK */
610 return (__u64 *)srv->ServiceData8;
614 kibnal_set_service_keys(IB_SERVICE_RECORD *srv, lnet_nid_t nid)
616 char *svc_name = *kibnal_tunables.kib_service_name;
618 LASSERT (strlen(svc_name) < sizeof(srv->ServiceName));
619 memset (srv->ServiceName, 0, sizeof(srv->ServiceName));
620 strcpy (srv->ServiceName, svc_name);
622 *kibnal_service_nid_field(srv) = cpu_to_le64(nid);
625 /* CAVEAT EMPTOR: We rely on tx/rx descriptor alignment to allow us to use the
626 * lowest 2 bits of the work request id to stash the work item type (the op
627 * field is not valid when the wc completes in error). */
629 #define IBNAL_WID_TX 0
630 #define IBNAL_WID_RX 1
631 #define IBNAL_WID_RDMA 2
632 #define IBNAL_WID_MASK 3UL
635 kibnal_ptr2wreqid (void *ptr, int type)
637 unsigned long lptr = (unsigned long)ptr;
639 LASSERT ((lptr & IBNAL_WID_MASK) == 0);
640 LASSERT ((type & ~IBNAL_WID_MASK) == 0);
641 return (__u64)(lptr | type);
645 kibnal_wreqid2ptr (__u64 wreqid)
647 return (void *)(((unsigned long)wreqid) & ~IBNAL_WID_MASK);
651 kibnal_wreqid2type (__u64 wreqid)
653 return (wreqid & IBNAL_WID_MASK);
657 kibnal_set_conn_state (kib_conn_t *conn, int state)
659 CDEBUG(D_NET,"%p state %d\n", conn, state);
660 conn->ibc_state = state;
667 kibnal_rd_size (kib_rdma_desc_t *rd)
674 kibnal_rd_size (kib_rdma_desc_t *rd)
679 for (i = size = 0; i < rd->rd_nfrag; i++)
680 size += rd->rd_frags[i].rf_nob;
686 int kibnal_startup (lnet_ni_t *ni);
687 void kibnal_shutdown (lnet_ni_t *ni);
688 int kibnal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
689 int kibnal_send (lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
690 int kibnal_eager_recv (lnet_ni_t *ni, void *private,
691 lnet_msg_t *lntmsg, void **new_private);
692 int kibnal_recv (lnet_ni_t *ni, void *private, lnet_msg_t *msg,
693 int delayed, unsigned int niov,
694 struct iovec *iov, lnet_kiov_t *kiov,
695 unsigned int offset, unsigned int mlen, unsigned int rlen);
696 void kibnal_init_msg(kib_msg_t *msg, int type, int body_nob);
697 void kibnal_pack_msg(kib_msg_t *msg, __u32 version, int credits,
698 lnet_nid_t dstnid, __u64 dststamp, __u64 seq);
699 void kibnal_pack_connmsg(kib_msg_t *msg, __u32 version, int nob, int type,
700 lnet_nid_t dstnid, __u64 dststamp);
701 int kibnal_unpack_msg(kib_msg_t *msg, __u32 expected_version, int nob);
702 IB_HANDLE kibnal_create_cep(lnet_nid_t nid);
703 int kibnal_create_peer (kib_peer_t **peerp, lnet_nid_t nid);
704 void kibnal_destroy_peer (kib_peer_t *peer);
705 kib_peer_t *kibnal_find_peer_locked (lnet_nid_t nid);
706 int kibnal_del_peer (lnet_nid_t nid);
707 void kibnal_peer_alive (kib_peer_t *peer);
708 void kibnal_unlink_peer_locked (kib_peer_t *peer);
709 int kibnal_add_persistent_peer (lnet_nid_t nid);
710 int kibnal_close_stale_conns_locked (kib_peer_t *peer,
712 int kibnal_conn_rts(kib_conn_t *conn,
713 __u32 qpn, __u8 resp_res, __u8 init_depth, __u32 psn);
714 kib_conn_t *kibnal_create_conn (lnet_nid_t nid, int proto_version);
715 void kibnal_destroy_conn (kib_conn_t *conn);
716 void kibnal_listen_callback(IB_HANDLE cep, CM_CONN_INFO *info, void *arg);
717 int kibnal_alloc_pages (kib_pages_t **pp, int npages);
718 void kibnal_free_pages (kib_pages_t *p);
719 void kibnal_queue_tx (kib_tx_t *tx, kib_conn_t *conn);
720 void kibnal_txlist_done (struct list_head *txlist, int status);
721 int kibnal_post_receives (kib_conn_t *conn);
722 int kibnal_init_rdma (kib_tx_t *tx, int type, int nob,
723 kib_rdma_desc_t *dstrd, __u64 dstcookie);
724 void kibnal_check_sends (kib_conn_t *conn);
725 void kibnal_close_conn_locked (kib_conn_t *conn, int error);
726 int kibnal_thread_start (int (*fn)(void *arg), void *arg);
727 int kibnal_scheduler(void *arg);
728 int kibnal_connd (void *arg);
729 void kibnal_init_tx_msg (kib_tx_t *tx, int type, int body_nob);
730 void kibnal_close_conn (kib_conn_t *conn, int why);
731 void kibnal_start_active_rdma (int type, int status,
732 kib_rx_t *rx, lnet_msg_t *lntmsg,
734 struct iovec *iov, lnet_kiov_t *kiov,
735 unsigned int offset, unsigned int nob);
736 void kibnal_hca_async_callback (void *hca_arg, IB_EVENT_RECORD *ev);
737 void kibnal_hca_callback (void *hca_arg, void *cq_arg);
738 int kibnal_tunables_init (void);
739 void kibnal_tunables_fini (void);