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 [sun.com URL with a
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/viblnd/viblnd.h
38 * Author: Eric Barton <eric@bartonsoftware.com>
39 * Author: Frank Zago <fzago@systemfabricworks.com>
43 # define EXPORT_SYMTAB
45 #ifndef AUTOCONF_INCLUDED
46 #include <linux/config.h>
48 #include <linux/module.h>
49 #include <linux/kernel.h>
51 #include <linux/string.h>
52 #include <linux/stat.h>
53 #include <linux/errno.h>
54 #include <linux/smp_lock.h>
55 #include <linux/unistd.h>
56 #include <linux/uio.h>
58 #include <asm/system.h>
59 #include <asm/uaccess.h>
62 #include <linux/init.h>
64 #include <linux/file.h>
65 #include <linux/stat.h>
66 #include <linux/list.h>
67 #include <linux/kmod.h>
68 #include <linux/sysctl.h>
69 #include <linux/random.h>
74 #define DEBUG_SUBSYSTEM S_LND
76 #include <libcfs/libcfs.h>
77 #include <lnet/lnet.h>
78 #include <lnet/lib-lnet.h>
80 /* CPU_{L,B}E #defines needed by Voltaire headers */
81 #include <asm/byteorder.h>
86 #ifdef __LITTLE_ENDIAN__
95 /* GCC 3.2.2, miscompiles this driver.
96 * See http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9853. */
97 #define GCC_VERSION ((__GNUC__*100 + __GNUC_MINOR__)*100 + __GNUC_PATCHLEVEL__)
98 #if (GCC_VERSION >= 30000) && (GCC_VERSION < 30203)
99 # error Invalid GCC version. Must use GCC < 3.0.0 || GCC >= 3.2.3
103 # define IBNAL_N_SCHED num_online_cpus() /* # schedulers */
105 # define IBNAL_N_SCHED 1 /* # schedulers */
108 #define IBNAL_USE_FMR 1
110 /* tunables fixed at compile time */
111 #define IBNAL_PEER_HASH_SIZE 101 /* # peer lists */
112 #define IBNAL_RESCHED 100 /* # scheduler loops before reschedule */
113 #define IBNAL_MSG_QUEUE_SIZE 8 /* # messages/RDMAs in-flight */
114 #define IBNAL_CREDIT_HIGHWATER 7 /* when eagerly to return credits */
115 #define IBNAL_MSG_SIZE (4<<10) /* max size of queued messages (inc hdr) */
117 /* constants derived from sdp-connection.c */
119 #define IBNAL_PKEY 0xffff
120 #define IBNAL_PKEY_IDX 0
121 #define IBNAL_SGID_IDX 0
122 #define IBNAL_SERVICE_LEVEL 0
123 #define IBNAL_STATIC_RATE 0
124 #define IBNAL_EE_FLOW_CNT 1
125 #define IBNAL_LOCAL_SUB 1
126 #define IBNAL_TRAFFIC_CLASS 0
127 #define IBNAL_SOURCE_PATH_BIT 0
128 #define IBNAL_OUS_DST_RD 1
129 #define IBNAL_IB_MTU vv_mtu_1024
131 /* constants derived from sdp-hca-params.h */
132 #define PATH_RATE_2_5GB 2
135 #define IBNAL_R_2_STATIC_RATE(r) ((r) == PATH_RATE_2_5GB ? MLX_IPD_1x : MLX_IPD_4x)
137 /* other low-level IB constants */
138 #define IBNAL_PKT_LIFETIME 5
139 #define IBNAL_ARB_INITIATOR_DEPTH 0
140 #define IBNAL_ARB_RESP_RES 0
141 #define IBNAL_FAILOVER_ACCEPTED 0
143 /************************/
144 /* derived constants... */
146 /* TX messages (shared by all connections) */
147 #define IBNAL_TX_MSGS() (*kibnal_tunables.kib_ntx)
148 #define IBNAL_TX_MSG_BYTES() (IBNAL_TX_MSGS() * IBNAL_MSG_SIZE)
149 #define IBNAL_TX_MSG_PAGES() ((IBNAL_TX_MSG_BYTES() + PAGE_SIZE - 1)/PAGE_SIZE)
152 # define IBNAL_MAX_RDMA_FRAGS 1
153 # define IBNAL_CONCURRENT_SENDS IBNAL_RX_MSGS
155 # define IBNAL_MAX_RDMA_FRAGS LNET_MAX_IOV
156 # define IBNAL_CONCURRENT_SENDS IBNAL_MSG_QUEUE_SIZE
159 /* RX messages (per connection) */
160 #define IBNAL_RX_MSGS (IBNAL_MSG_QUEUE_SIZE*2)
161 #define IBNAL_RX_MSG_BYTES (IBNAL_RX_MSGS * IBNAL_MSG_SIZE)
162 #define IBNAL_RX_MSG_PAGES ((IBNAL_RX_MSG_BYTES + PAGE_SIZE - 1)/PAGE_SIZE)
164 #define IBNAL_CQ_ENTRIES() (IBNAL_TX_MSGS() * (1 + IBNAL_MAX_RDMA_FRAGS) + \
165 IBNAL_RX_MSGS * *kibnal_tunables.kib_concurrent_peers)
169 unsigned int *kib_service_number; /* IB service number */
170 int *kib_min_reconnect_interval; /* first failed connection retry... */
171 int *kib_max_reconnect_interval; /* ...exponentially increasing to this */
172 int *kib_concurrent_peers; /* max # nodes all talking to me */
173 int *kib_cksum; /* checksum kib_msg_t? */
174 int *kib_timeout; /* comms timeout (seconds) */
175 int *kib_ntx; /* # tx descs */
176 int *kib_credits; /* # concurrent sends */
177 int *kib_peercredits; /* # concurrent sends to 1 peer */
178 int *kib_arp_retries; /* # times to retry ARP */
179 char **kib_hca_basename; /* HCA base name */
180 char **kib_ipif_basename; /* IPoIB interface base name */
181 int *kib_local_ack_timeout; /* IB RC QP ack timeout... */
182 int *kib_retry_cnt; /* ...and retry */
183 int *kib_rnr_cnt; /* RNR retries... */
184 int *kib_rnr_nak_timer; /* ...and interval */
185 int *kib_keepalive; /* keepalive interval */
186 int *kib_concurrent_sends; /* send work queue sizing */
188 int *kib_fmr_remaps; /* # FMR maps before unmap required */
190 #if defined(CONFIG_SYSCTL) && !CFS_SYSFS_MODULE_PARM
191 cfs_sysctl_table_header_t *kib_sysctl; /* sysctl interface */
197 int ibp_npages; /* # pages */
198 struct page *ibp_pages[0];
204 vv_fmr_h_t md_fmrhandle; /* FMR handle */
205 int md_fmrcount; /* # mappings left */
206 int md_active; /* mapping in use? */
207 __u32 md_lkey; /* local key */
208 __u32 md_rkey; /* remote key */
209 __u64 md_addr; /* IO VM address */
215 int kib_init; /* initialisation state */
216 __u64 kib_incarnation; /* which one am I */
217 int kib_shutdown; /* shut down? */
218 atomic_t kib_nthreads; /* # live threads */
219 lnet_ni_t *kib_ni; /* _the_ nal instance */
221 vv_gid_t kib_port_gid; /* device/port GID */
222 vv_p_key_t kib_port_pkey; /* device/port pkey */
224 cm_cep_handle_t kib_listen_handle; /* IB listen handle */
226 rwlock_t kib_global_lock; /* stabilize peer/conn ops */
227 int kib_ready; /* CQ callback fired */
228 int kib_checking_cq; /* a scheduler is checking the CQ */
230 struct list_head *kib_peers; /* hash table of all my known peers */
231 int kib_peer_hash_size; /* size of kib_peers */
232 atomic_t kib_npeers; /* # peers extant */
233 atomic_t kib_nconns; /* # connections extant */
235 void *kib_connd; /* the connd task (serialisation assertions) */
236 struct list_head kib_connd_peers; /* peers wanting to get connected */
237 struct list_head kib_connd_pcreqs; /* passive connection requests */
238 struct list_head kib_connd_conns; /* connections to setup/teardown */
239 struct list_head kib_connd_zombies; /* connections with zero refcount */
240 wait_queue_head_t kib_connd_waitq; /* connection daemon sleeps here */
241 spinlock_t kib_connd_lock; /* serialise */
243 wait_queue_head_t kib_sched_waitq; /* schedulers sleep here */
244 spinlock_t kib_sched_lock; /* serialise */
246 struct kib_tx *kib_tx_descs; /* all the tx descriptors */
247 kib_pages_t *kib_tx_pages; /* premapped tx msg pages */
249 struct list_head kib_idle_txs; /* idle tx descriptors */
250 __u64 kib_next_tx_cookie; /* RDMA completion cookie */
251 spinlock_t kib_tx_lock; /* serialise */
253 vv_hca_h_t kib_hca; /* The HCA */
254 vv_hca_attrib_t kib_hca_attrs; /* its properties */
255 int kib_port; /* port on the device */
256 vv_port_attrib_t kib_port_attr; /* its properties */
258 vv_pd_h_t kib_pd; /* protection domain */
259 vv_cq_h_t kib_cq; /* completion queue */
263 #define IBNAL_INIT_NOTHING 0
264 #define IBNAL_INIT_DATA 1
265 #define IBNAL_INIT_LIB 2
266 #define IBNAL_INIT_HCA 3
267 #define IBNAL_INIT_ASYNC 4
268 #define IBNAL_INIT_PD 5
269 #define IBNAL_INIT_TXD 6
270 #define IBNAL_INIT_CQ 7
271 #define IBNAL_INIT_ALL 8
273 #include "viblnd_wire.h"
275 /***********************************************************************/
277 typedef struct kib_rx /* receive message */
279 struct list_head rx_list; /* queue for attention */
280 struct kib_conn *rx_conn; /* owning conn */
281 int rx_nob; /* # bytes received (-1 while posted) */
282 vv_l_key_t rx_lkey; /* local key */
283 kib_msg_t *rx_msg; /* pre-mapped buffer (host vaddr) */
284 vv_wr_t rx_wrq; /* receive work item */
285 vv_scatgat_t rx_gl; /* and its memory */
288 typedef struct kib_tx /* transmit message */
290 struct list_head tx_list; /* queue on idle_txs ibc_tx_queue etc. */
291 struct kib_conn *tx_conn; /* owning conn */
292 int tx_sending; /* # tx callbacks outstanding */
293 int tx_queued; /* queued for sending */
294 int tx_waiting; /* waiting for peer */
295 int tx_status; /* completion status */
296 unsigned long tx_deadline; /* completion deadline */
297 __u64 tx_cookie; /* completion cookie */
298 lnet_msg_t *tx_lntmsg[2]; /* lnet msgs to finalize on completion */
299 vv_l_key_t tx_lkey; /* local key for message buffer */
300 kib_msg_t *tx_msg; /* message buffer (host vaddr) */
301 int tx_nwrq; /* # send work items */
303 vv_wr_t tx_wrq[2]; /* send work items... */
304 vv_scatgat_t tx_gl[2]; /* ...and their memory */
305 kib_rdma_desc_t tx_rd[1]; /* rdma descriptor */
306 kib_md_t tx_md; /* FMR mapping descriptor */
307 __u64 *tx_pages; /* page phys addrs */
309 vv_wr_t *tx_wrq; /* send work items... */
310 vv_scatgat_t *tx_gl; /* ...and their memory */
311 kib_rdma_desc_t *tx_rd; /* rdma descriptor (src buffers) */
315 /* Passive connection request (listener callback) queued for handling by connd */
316 typedef struct kib_pcreq
318 struct list_head pcr_list; /* queue for handling by connd */
319 cm_cep_handle_t pcr_cep; /* listening handle */
320 cm_request_data_t pcr_cmreq; /* request data */
323 typedef struct kib_connvars
325 /* connection-in-progress variables */
334 ib_path_record_v2_t cv_path;
335 ibat_arp_data_t cv_arp;
336 ibat_stat_t cv_arprc;
337 cm_conn_data_t cv_conndata;
340 typedef struct kib_conn
342 struct kib_peer *ibc_peer; /* owning peer */
343 struct list_head ibc_list; /* stash on peer's conn list */
344 __u64 ibc_incarnation; /* which instance of the peer */
345 __u64 ibc_txseq; /* tx sequence number */
346 __u64 ibc_rxseq; /* rx sequence number */
347 __u32 ibc_version; /* peer protocol version */
348 atomic_t ibc_refcount; /* # users */
349 int ibc_state; /* what's happening */
350 int ibc_nsends_posted; /* # uncompleted sends */
351 int ibc_credits; /* # credits I have */
352 int ibc_outstanding_credits; /* # credits to return */
353 int ibc_reserved_credits; /* # credits for ACK/DONE msgs */
354 int ibc_disconnect; /* some disconnect callback fired */
355 int ibc_comms_error; /* set on comms error */
356 unsigned long ibc_last_send; /* time of last send */
357 struct list_head ibc_early_rxs; /* rxs completed before ESTABLISHED */
358 struct list_head ibc_tx_queue_nocred; /* sends that don't need a cred */
359 struct list_head ibc_tx_queue_rsrvd; /* sends that need a reserved cred */
360 struct list_head ibc_tx_queue; /* send queue */
361 struct list_head ibc_active_txs; /* active tx awaiting completion */
362 spinlock_t ibc_lock; /* serialise */
363 kib_rx_t *ibc_rxs; /* the rx descs */
364 kib_pages_t *ibc_rx_pages; /* premapped rx msg pages */
365 vv_qp_h_t ibc_qp; /* queue pair */
366 cm_cep_handle_t ibc_cep; /* connection endpoint */
367 kib_connvars_t *ibc_connvars; /* in-progress connection state */
370 #define IBNAL_CONN_INIT_NOTHING 0 /* incomplete init */
371 #define IBNAL_CONN_INIT_QP 1 /* QP allocated */
372 #define IBNAL_CONN_INIT 2 /* completed init */
373 #define IBNAL_CONN_ACTIVE_ARP 3 /* active arping */
374 #define IBNAL_CONN_ACTIVE_CONNECT 4 /* active sending req */
375 #define IBNAL_CONN_ACTIVE_CHECK_REPLY 5 /* active checking reply */
376 #define IBNAL_CONN_ACTIVE_RTU 6 /* active sending rtu */
377 #define IBNAL_CONN_PASSIVE_WAIT 7 /* passive waiting for rtu */
378 #define IBNAL_CONN_ESTABLISHED 8 /* connection established */
379 #define IBNAL_CONN_DISCONNECT1 9 /* disconnect phase 1 */
380 #define IBNAL_CONN_DISCONNECT2 10 /* disconnect phase 2 */
381 #define IBNAL_CONN_DISCONNECTED 11 /* disconnect complete */
383 typedef struct kib_peer
385 struct list_head ibp_list; /* stash on global peer list */
386 struct list_head ibp_connd_list; /* schedule on kib_connd_peers */
387 lnet_nid_t ibp_nid; /* who's on the other end(s) */
388 __u32 ibp_ip; /* IP to query for peer conn params */
389 int ibp_port; /* port to qery for peer conn params */
390 __u64 ibp_incarnation; /* peer's incarnation */
391 atomic_t ibp_refcount; /* # users */
392 int ibp_persistence; /* "known" peer refs */
393 struct list_head ibp_conns; /* all active connections */
394 struct list_head ibp_tx_queue; /* msgs waiting for a conn */
395 int ibp_connecting; /* current active connection attempts */
396 int ibp_accepting; /* current passive connection attempts */
397 int ibp_arp_count; /* # arp attempts */
398 unsigned long ibp_reconnect_time; /* when reconnect may be attempted */
399 unsigned long ibp_reconnect_interval; /* exponential backoff */
400 int ibp_error; /* errno on closing this peer */
401 cfs_time_t ibp_last_alive; /* when (in jiffies) I was last alive */
405 extern kib_data_t kibnal_data;
406 extern kib_tunables_t kibnal_tunables;
408 int kibnal_startup (lnet_ni_t *ni);
409 void kibnal_shutdown (lnet_ni_t *ni);
410 int kibnal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
411 int kibnal_send (lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
412 extern int kibnal_eager_recv (lnet_ni_t *ni, void *private,
413 lnet_msg_t *lntmsg, void **new_private);
414 int kibnal_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
415 int delayed, unsigned int niov,
416 struct iovec *iov, lnet_kiov_t *kiov,
417 unsigned int offset, unsigned int mlen, unsigned int rlen);
418 extern void kibnal_init_msg(kib_msg_t *msg, int type, int body_nob);
419 extern void kibnal_pack_msg(kib_msg_t *msg, __u32 version, int credits,
420 lnet_nid_t dstnid, __u64 dststamp, __u64 seq);
421 extern int kibnal_unpack_msg(kib_msg_t *msg, __u32 expected_version, int nob);
422 extern int kibnal_create_peer(kib_peer_t **peerp, lnet_nid_t nid);
423 extern void kibnal_destroy_peer(kib_peer_t *peer);
424 extern int kibnal_add_persistent_peer (lnet_nid_t nid, __u32 ip);
425 extern int kibnal_del_peer(lnet_nid_t nid);
426 extern kib_peer_t *kibnal_find_peer_locked(lnet_nid_t nid);
427 extern void kibnal_unlink_peer_locked(kib_peer_t *peer);
428 extern void kibnal_peer_alive(kib_peer_t *peer);
429 extern int kibnal_close_stale_conns_locked(kib_peer_t *peer,
431 extern kib_conn_t *kibnal_create_conn(cm_cep_handle_t cep);
432 extern void kibnal_listen_callback(cm_cep_handle_t cep, cm_conn_data_t *info, void *arg);
434 extern int kibnal_alloc_pages(kib_pages_t **pp, int npages, int access);
435 extern void kibnal_free_pages(kib_pages_t *p);
437 extern void kibnal_check_sends(kib_conn_t *conn);
438 extern void kibnal_close_conn_locked(kib_conn_t *conn, int error);
439 extern void kibnal_destroy_conn(kib_conn_t *conn);
440 extern int kibnal_thread_start(int (*fn)(void *arg), void *arg);
441 extern int kibnal_scheduler(void *arg);
442 extern int kibnal_connd(void *arg);
443 extern void kibnal_init_tx_msg(kib_tx_t *tx, int type, int body_nob);
444 extern void kibnal_close_conn(kib_conn_t *conn, int why);
445 extern int kibnal_set_qp_state(kib_conn_t *conn, vv_qp_state_t new_state);
446 extern void kibnal_async_callback(vv_event_record_t ev);
447 extern void kibnal_cq_callback(unsigned long context);
448 extern void kibnal_passive_connreq(kib_pcreq_t *pcr, int reject);
449 extern void kibnal_txlist_done (struct list_head *txlist, int status);
450 extern void kibnal_queue_tx(kib_tx_t *tx, kib_conn_t *conn);
451 extern int kibnal_init_rdma(kib_tx_t *tx, int type, int nob,
452 kib_rdma_desc_t *dstrd, __u64 dstcookie);
453 extern int kibnal_tunables_init(void);
454 extern void kibnal_tunables_fini(void);
456 #define kibnal_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 kibnal_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(&kibnal_data.kib_connd_lock, flags); \
473 list_add_tail(&(conn)->ibc_list, \
474 &kibnal_data.kib_connd_zombies); \
475 wake_up(&kibnal_data.kib_connd_waitq); \
476 spin_unlock_irqrestore(&kibnal_data.kib_connd_lock, flags); \
480 #define kibnal_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 kibnal_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 kibnal_destroy_peer(peer); \
499 static inline struct list_head *
500 kibnal_nid2peerlist (lnet_nid_t nid)
502 unsigned int hash = ((unsigned int)nid) % kibnal_data.kib_peer_hash_size;
504 return (&kibnal_data.kib_peers [hash]);
508 kibnal_peer_active (kib_peer_t *peer)
510 /* Am I in the peer hash table? */
511 return (!list_empty(&peer->ibp_list));
515 kibnal_queue_tx_locked (kib_tx_t *tx, kib_conn_t *conn)
519 LASSERT (tx->tx_nwrq > 0); /* work items set up */
520 LASSERT (!tx->tx_queued); /* not queued for sending already */
523 tx->tx_deadline = jiffies + (*kibnal_tunables.kib_timeout * HZ);
525 if (tx->tx_conn == NULL) {
526 kibnal_conn_addref(conn);
528 LASSERT (tx->tx_msg->ibm_type != IBNAL_MSG_PUT_DONE);
530 LASSERT (tx->tx_conn == conn);
531 LASSERT (tx->tx_msg->ibm_type == IBNAL_MSG_PUT_DONE);
534 if (conn->ibc_version == IBNAL_MSG_VERSION_RDMAREPLYNOTRSRVD) {
535 /* All messages have simple credit control */
536 q = &conn->ibc_tx_queue;
538 LASSERT (conn->ibc_version == IBNAL_MSG_VERSION);
540 switch (tx->tx_msg->ibm_type) {
541 case IBNAL_MSG_PUT_REQ:
542 case IBNAL_MSG_GET_REQ:
543 /* RDMA request: reserve a buffer for the RDMA reply
545 q = &conn->ibc_tx_queue_rsrvd;
548 case IBNAL_MSG_PUT_NAK:
549 case IBNAL_MSG_PUT_ACK:
550 case IBNAL_MSG_PUT_DONE:
551 case IBNAL_MSG_GET_DONE:
552 /* RDMA reply/completion: no credits; peer has reserved
554 q = &conn->ibc_tx_queue_nocred;
558 case IBNAL_MSG_IMMEDIATE:
559 /* Otherwise: consume a credit before sending */
560 q = &conn->ibc_tx_queue;
569 list_add_tail(&tx->tx_list, q);
573 kibnal_send_keepalive(kib_conn_t *conn)
575 return (*kibnal_tunables.kib_keepalive > 0) &&
576 time_after(jiffies, conn->ibc_last_send +
577 *kibnal_tunables.kib_keepalive*HZ);
580 #ifndef IBNAL_VOIDSTAR_SGADDR
581 # define IBNAL_VOIDSTAR_SGADDR 0
584 #if IBNAL_VOIDSTAR_SGADDR
585 # if defined(CONFIG_HIGHMEM)
586 # if defined(CONFIG_X86) && defined(CONFIG_HIGHMEM4G)
587 /* truncation to void* doesn't matter if 0 <= physmem < 4G
588 * so allow x86 with 32 bit phys addrs */
589 # elif defined(CONFIG_IA64)
590 /* OK anyway on 64-bit arch */
592 # error "Can't support HIGHMEM when vv_scatgat_t::v_address is void *"
595 # define KIBNAL_ADDR2SG(a) ((void *)((unsigned long)(a)))
596 # define KIBNAL_SG2ADDR(a) ((__u64)((unsigned long)(a)))
597 static inline __u64 kibnal_addr2net (__u64 addr)
600 vv_return_t vvrc = vv_va2advertise_addr(kibnal_data.kib_hca,
601 KIBNAL_ADDR2SG(addr),
603 LASSERT (vvrc == vv_return_ok);
604 return KIBNAL_SG2ADDR(netaddr);
607 # define KIBNAL_ADDR2SG(a) a
608 # define KIBNAL_SG2ADDR(a) a
609 static inline __u64 kibnal_addr2net (__u64 addr)
612 vv_return_t vvrc = vv_va2advertise_addr(kibnal_data.kib_hca,
615 LASSERT (vvrc == vv_return_ok);
620 /* CAVEAT EMPTOR: We rely on tx/rx descriptor alignment to allow us to use the
621 * lowest 2 bits of the work request id to stash the work item type (the op
622 * field is not valid when the wc completes in error). */
624 #define IBNAL_WID_TX 0
625 #define IBNAL_WID_RX 1
626 #define IBNAL_WID_RDMA 2
627 #define IBNAL_WID_MASK 3UL
629 static inline vv_wr_id_t
630 kibnal_ptr2wreqid (void *ptr, int type)
632 unsigned long lptr = (unsigned long)ptr;
634 LASSERT ((lptr & IBNAL_WID_MASK) == 0);
635 LASSERT ((type & ~IBNAL_WID_MASK) == 0);
636 return (vv_wr_id_t)(lptr | type);
640 kibnal_wreqid2ptr (vv_wr_id_t wreqid)
642 return (void *)(((unsigned long)wreqid) & ~IBNAL_WID_MASK);
646 kibnal_wreqid2type (vv_wr_id_t wreqid)
648 return (wreqid & IBNAL_WID_MASK);
652 kibnal_set_conn_state (kib_conn_t *conn, int state)
654 conn->ibc_state = state;
661 kibnal_rd_size (kib_rdma_desc_t *rd)
668 kibnal_rf_addr (kib_rdma_frag_t *rf)
670 return (((__u64)rf->rf_addr_hi)<<32) | ((__u64)rf->rf_addr_lo);
674 kibnal_rf_set (kib_rdma_frag_t *rf, __u64 addr, int nob)
676 rf->rf_addr_lo = addr & 0xffffffff;
677 rf->rf_addr_hi = (addr >> 32) & 0xffffffff;
682 kibnal_rd_size (kib_rdma_desc_t *rd)
687 for (i = size = 0; i < rd->rd_nfrag; i++)
688 size += rd->rd_frags[i].rf_nob;