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>
6 * Author: Frank Zago <fzago@systemfabricworks.com>
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
26 # define EXPORT_SYMTAB
28 #ifdef HAVE_KERNEL_CONFIG_H
29 #include <linux/config.h>
31 #include <linux/module.h>
32 #include <linux/kernel.h>
34 #include <linux/string.h>
35 #include <linux/stat.h>
36 #include <linux/errno.h>
37 #include <linux/smp_lock.h>
38 #include <linux/unistd.h>
39 #include <linux/uio.h>
41 #include <asm/system.h>
42 #include <asm/uaccess.h>
45 #include <linux/init.h>
47 #include <linux/file.h>
48 #include <linux/stat.h>
49 #include <linux/list.h>
50 #include <linux/kmod.h>
51 #include <linux/sysctl.h>
52 #include <linux/random.h>
57 #define DEBUG_SUBSYSTEM S_LND
59 #include <libcfs/kp30.h>
60 #include <lnet/lnet.h>
61 #include <lnet/lib-lnet.h>
63 /* CPU_{L,B}E #defines needed by Voltaire headers */
64 #include <asm/byteorder.h>
69 #ifdef __LITTLE_ENDIAN__
78 /* GCC 3.2.2, miscompiles this driver.
79 * See http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9853. */
80 #define GCC_VERSION ((__GNUC__*100 + __GNUC_MINOR__)*100 + __GNUC_PATCHLEVEL__)
81 #if (GCC_VERSION >= 30000) && (GCC_VERSION < 30203)
82 # error Invalid GCC version. Must use GCC < 3.0.0 || GCC >= 3.2.3
86 # define IBNAL_N_SCHED num_online_cpus() /* # schedulers */
88 # define IBNAL_N_SCHED 1 /* # schedulers */
91 #define IBNAL_USE_FMR 1
93 /* tunables fixed at compile time */
94 #define IBNAL_PEER_HASH_SIZE 101 /* # peer lists */
95 #define IBNAL_RESCHED 100 /* # scheduler loops before reschedule */
96 #define IBNAL_MSG_QUEUE_SIZE 8 /* # messages/RDMAs in-flight */
97 #define IBNAL_CREDIT_HIGHWATER 7 /* when eagerly to return credits */
98 #define IBNAL_MSG_SIZE (4<<10) /* max size of queued messages (inc hdr) */
100 /* constants derived from sdp-connection.c */
102 #define IBNAL_PKEY 0xffff
103 #define IBNAL_PKEY_IDX 0
104 #define IBNAL_SGID_IDX 0
105 #define IBNAL_SERVICE_LEVEL 0
106 #define IBNAL_STATIC_RATE 0
107 #define IBNAL_EE_FLOW_CNT 1
108 #define IBNAL_LOCAL_SUB 1
109 #define IBNAL_TRAFFIC_CLASS 0
110 #define IBNAL_SOURCE_PATH_BIT 0
111 #define IBNAL_OUS_DST_RD 1
112 #define IBNAL_IB_MTU vv_mtu_1024
114 /* constants derived from sdp-hca-params.h */
115 #define PATH_RATE_2_5GB 2
118 #define IBNAL_R_2_STATIC_RATE(r) ((r) == PATH_RATE_2_5GB ? MLX_IPD_1x : MLX_IPD_4x)
120 /* other low-level IB constants */
121 #define IBNAL_PKT_LIFETIME 5
122 #define IBNAL_ARB_INITIATOR_DEPTH 0
123 #define IBNAL_ARB_RESP_RES 0
124 #define IBNAL_FAILOVER_ACCEPTED 0
126 /************************/
127 /* derived constants... */
129 /* TX messages (shared by all connections) */
130 #define IBNAL_TX_MSGS() (*kibnal_tunables.kib_ntx)
131 #define IBNAL_TX_MSG_BYTES() (IBNAL_TX_MSGS() * IBNAL_MSG_SIZE)
132 #define IBNAL_TX_MSG_PAGES() ((IBNAL_TX_MSG_BYTES() + PAGE_SIZE - 1)/PAGE_SIZE)
135 # define IBNAL_MAX_RDMA_FRAGS 1
136 # define IBNAL_CONCURRENT_SENDS IBNAL_RX_MSGS
138 # define IBNAL_MAX_RDMA_FRAGS LNET_MAX_IOV
139 # define IBNAL_CONCURRENT_SENDS IBNAL_MSG_QUEUE_SIZE
142 /* RX messages (per connection) */
143 #define IBNAL_RX_MSGS (IBNAL_MSG_QUEUE_SIZE*2)
144 #define IBNAL_RX_MSG_BYTES (IBNAL_RX_MSGS * IBNAL_MSG_SIZE)
145 #define IBNAL_RX_MSG_PAGES ((IBNAL_RX_MSG_BYTES + PAGE_SIZE - 1)/PAGE_SIZE)
147 #define IBNAL_CQ_ENTRIES() (IBNAL_TX_MSGS() * (1 + IBNAL_MAX_RDMA_FRAGS) + \
148 IBNAL_RX_MSGS * *kibnal_tunables.kib_concurrent_peers)
152 unsigned int *kib_service_number; /* IB service number */
153 int *kib_min_reconnect_interval; /* first failed connection retry... */
154 int *kib_max_reconnect_interval; /* ...exponentially increasing to this */
155 int *kib_concurrent_peers; /* max # nodes all talking to me */
156 int *kib_cksum; /* checksum kib_msg_t? */
157 int *kib_timeout; /* comms timeout (seconds) */
158 int *kib_ntx; /* # tx descs */
159 int *kib_credits; /* # concurrent sends */
160 int *kib_peercredits; /* # concurrent sends to 1 peer */
161 int *kib_arp_retries; /* # times to retry ARP */
162 char **kib_hca_basename; /* HCA base name */
163 char **kib_ipif_basename; /* IPoIB interface base name */
164 int *kib_local_ack_timeout; /* IB RC QP ack timeout... */
165 int *kib_retry_cnt; /* ...and retry */
166 int *kib_rnr_cnt; /* RNR retries... */
167 int *kib_rnr_nak_timer; /* ...and interval */
168 int *kib_keepalive; /* keepalive interval */
169 int *kib_concurrent_sends; /* send work queue sizing */
171 int *kib_fmr_remaps; /* # FMR maps before unmap required */
173 #if CONFIG_SYSCTL && !CFS_SYSFS_MODULE_PARM
174 struct ctl_table_header *kib_sysctl; /* sysctl interface */
180 int ibp_npages; /* # pages */
181 struct page *ibp_pages[0];
187 vv_fmr_h_t md_fmrhandle; /* FMR handle */
188 int md_fmrcount; /* # mappings left */
189 int md_active; /* mapping in use? */
190 __u32 md_lkey; /* local key */
191 __u32 md_rkey; /* remote key */
192 __u64 md_addr; /* IO VM address */
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_ nal instance */
204 vv_gid_t kib_port_gid; /* device/port GID */
205 vv_p_key_t kib_port_pkey; /* device/port pkey */
207 cm_cep_handle_t kib_listen_handle; /* IB listen handle */
209 rwlock_t kib_global_lock; /* stabilize peer/conn ops */
210 int kib_ready; /* CQ callback fired */
211 int kib_checking_cq; /* a scheduler is checking the CQ */
213 struct list_head *kib_peers; /* hash table of all my known peers */
214 int kib_peer_hash_size; /* size of kib_peers */
215 atomic_t kib_npeers; /* # peers extant */
216 atomic_t kib_nconns; /* # connections extant */
218 void *kib_connd; /* the connd task (serialisation assertions) */
219 struct list_head kib_connd_peers; /* peers wanting to get connected */
220 struct list_head kib_connd_pcreqs; /* passive connection requests */
221 struct list_head kib_connd_conns; /* connections to setup/teardown */
222 struct list_head kib_connd_zombies; /* connections with zero refcount */
223 wait_queue_head_t kib_connd_waitq; /* connection daemon sleeps here */
224 spinlock_t kib_connd_lock; /* serialise */
226 wait_queue_head_t kib_sched_waitq; /* schedulers sleep here */
227 spinlock_t kib_sched_lock; /* serialise */
229 struct kib_tx *kib_tx_descs; /* all the tx descriptors */
230 kib_pages_t *kib_tx_pages; /* premapped tx msg pages */
232 struct list_head kib_idle_txs; /* idle tx descriptors */
233 __u64 kib_next_tx_cookie; /* RDMA completion cookie */
234 spinlock_t kib_tx_lock; /* serialise */
236 vv_hca_h_t kib_hca; /* The HCA */
237 vv_hca_attrib_t kib_hca_attrs; /* its properties */
238 int kib_port; /* port on the device */
239 vv_port_attrib_t kib_port_attr; /* its properties */
241 vv_pd_h_t kib_pd; /* protection domain */
242 vv_cq_h_t kib_cq; /* completion queue */
246 #define IBNAL_INIT_NOTHING 0
247 #define IBNAL_INIT_DATA 1
248 #define IBNAL_INIT_LIB 2
249 #define IBNAL_INIT_HCA 3
250 #define IBNAL_INIT_ASYNC 4
251 #define IBNAL_INIT_PD 5
252 #define IBNAL_INIT_TXD 6
253 #define IBNAL_INIT_CQ 7
254 #define IBNAL_INIT_ALL 8
256 #include "viblnd_wire.h"
258 /***********************************************************************/
260 typedef struct kib_rx /* receive message */
262 struct list_head rx_list; /* queue for attention */
263 struct kib_conn *rx_conn; /* owning conn */
264 int rx_nob; /* # bytes received (-1 while posted) */
265 vv_l_key_t rx_lkey; /* local key */
266 kib_msg_t *rx_msg; /* pre-mapped buffer (host vaddr) */
267 vv_wr_t rx_wrq; /* receive work item */
268 vv_scatgat_t rx_gl; /* and its memory */
271 typedef struct kib_tx /* transmit message */
273 struct list_head tx_list; /* queue on idle_txs ibc_tx_queue etc. */
274 struct kib_conn *tx_conn; /* owning conn */
275 int tx_sending; /* # tx callbacks outstanding */
276 int tx_queued; /* queued for sending */
277 int tx_waiting; /* waiting for peer */
278 int tx_status; /* completion status */
279 unsigned long tx_deadline; /* completion deadline */
280 __u64 tx_cookie; /* completion cookie */
281 lnet_msg_t *tx_lntmsg[2]; /* lnet msgs to finalize on completion */
282 vv_l_key_t tx_lkey; /* local key for message buffer */
283 kib_msg_t *tx_msg; /* message buffer (host vaddr) */
284 int tx_nwrq; /* # send work items */
286 vv_wr_t tx_wrq[2]; /* send work items... */
287 vv_scatgat_t tx_gl[2]; /* ...and their memory */
288 kib_rdma_desc_t tx_rd[1]; /* rdma descriptor */
289 kib_md_t tx_md; /* FMR mapping descriptor */
290 __u64 *tx_pages; /* page phys addrs */
292 vv_wr_t *tx_wrq; /* send work items... */
293 vv_scatgat_t *tx_gl; /* ...and their memory */
294 kib_rdma_desc_t *tx_rd; /* rdma descriptor (src buffers) */
298 /* Passive connection request (listener callback) queued for handling by connd */
299 typedef struct kib_pcreq
301 struct list_head pcr_list; /* queue for handling by connd */
302 cm_cep_handle_t pcr_cep; /* listening handle */
303 cm_request_data_t pcr_cmreq; /* request data */
306 typedef struct kib_connvars
308 /* connection-in-progress variables */
317 ib_path_record_v2_t cv_path;
318 ibat_arp_data_t cv_arp;
319 ibat_stat_t cv_arprc;
320 cm_conn_data_t cv_conndata;
323 typedef struct kib_conn
325 struct kib_peer *ibc_peer; /* owning peer */
326 struct list_head ibc_list; /* stash on peer's conn list */
327 __u64 ibc_incarnation; /* which instance of the peer */
328 __u64 ibc_txseq; /* tx sequence number */
329 __u64 ibc_rxseq; /* rx sequence number */
330 __u32 ibc_version; /* peer protocol version */
331 atomic_t ibc_refcount; /* # users */
332 int ibc_state; /* what's happening */
333 int ibc_nsends_posted; /* # uncompleted sends */
334 int ibc_credits; /* # credits I have */
335 int ibc_outstanding_credits; /* # credits to return */
336 int ibc_reserved_credits; /* # credits for ACK/DONE msgs */
337 int ibc_disconnect; /* some disconnect callback fired */
338 int ibc_comms_error; /* set on comms error */
339 unsigned long ibc_last_send; /* time of last send */
340 struct list_head ibc_early_rxs; /* rxs completed before ESTABLISHED */
341 struct list_head ibc_tx_queue_nocred; /* sends that don't need a cred */
342 struct list_head ibc_tx_queue_rsrvd; /* sends that need a reserved cred */
343 struct list_head ibc_tx_queue; /* send queue */
344 struct list_head ibc_active_txs; /* active tx awaiting completion */
345 spinlock_t ibc_lock; /* serialise */
346 kib_rx_t *ibc_rxs; /* the rx descs */
347 kib_pages_t *ibc_rx_pages; /* premapped rx msg pages */
348 vv_qp_h_t ibc_qp; /* queue pair */
349 cm_cep_handle_t ibc_cep; /* connection endpoint */
350 kib_connvars_t *ibc_connvars; /* in-progress connection state */
353 #define IBNAL_CONN_INIT_NOTHING 0 /* incomplete init */
354 #define IBNAL_CONN_INIT_QP 1 /* QP allocated */
355 #define IBNAL_CONN_INIT 2 /* completed init */
356 #define IBNAL_CONN_ACTIVE_ARP 3 /* active arping */
357 #define IBNAL_CONN_ACTIVE_CONNECT 4 /* active sending req */
358 #define IBNAL_CONN_ACTIVE_CHECK_REPLY 5 /* active checking reply */
359 #define IBNAL_CONN_ACTIVE_RTU 6 /* active sending rtu */
360 #define IBNAL_CONN_PASSIVE_WAIT 7 /* passive waiting for rtu */
361 #define IBNAL_CONN_ESTABLISHED 8 /* connection established */
362 #define IBNAL_CONN_DISCONNECT1 9 /* disconnect phase 1 */
363 #define IBNAL_CONN_DISCONNECT2 10 /* disconnect phase 2 */
364 #define IBNAL_CONN_DISCONNECTED 11 /* disconnect complete */
366 typedef struct kib_peer
368 struct list_head ibp_list; /* stash on global peer list */
369 struct list_head ibp_connd_list; /* schedule on kib_connd_peers */
370 lnet_nid_t ibp_nid; /* who's on the other end(s) */
371 __u32 ibp_ip; /* IP to query for peer conn params */
372 int ibp_port; /* port to qery for peer conn params */
373 __u64 ibp_incarnation; /* peer's incarnation */
374 atomic_t ibp_refcount; /* # users */
375 int ibp_persistence; /* "known" peer refs */
376 struct list_head ibp_conns; /* all active connections */
377 struct list_head ibp_tx_queue; /* msgs waiting for a conn */
378 int ibp_connecting; /* current active connection attempts */
379 int ibp_accepting; /* current passive connection attempts */
380 int ibp_arp_count; /* # arp attempts */
381 unsigned long ibp_reconnect_time; /* when reconnect may be attempted */
382 unsigned long ibp_reconnect_interval; /* exponential backoff */
383 int ibp_error; /* errno on closing this peer */
384 cfs_time_t ibp_last_alive; /* when (in jiffies) I was last alive */
388 extern kib_data_t kibnal_data;
389 extern kib_tunables_t kibnal_tunables;
391 int kibnal_startup (lnet_ni_t *ni);
392 void kibnal_shutdown (lnet_ni_t *ni);
393 int kibnal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
394 int kibnal_send (lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
395 extern int kibnal_eager_recv (lnet_ni_t *ni, void *private,
396 lnet_msg_t *lntmsg, void **new_private);
397 int kibnal_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
398 int delayed, unsigned int niov,
399 struct iovec *iov, lnet_kiov_t *kiov,
400 unsigned int offset, unsigned int mlen, unsigned int rlen);
401 extern void kibnal_init_msg(kib_msg_t *msg, int type, int body_nob);
402 extern void kibnal_pack_msg(kib_msg_t *msg, __u32 version, int credits,
403 lnet_nid_t dstnid, __u64 dststamp, __u64 seq);
404 extern int kibnal_unpack_msg(kib_msg_t *msg, __u32 expected_version, int nob);
405 extern int kibnal_create_peer(kib_peer_t **peerp, lnet_nid_t nid);
406 extern void kibnal_destroy_peer(kib_peer_t *peer);
407 extern int kibnal_add_persistent_peer (lnet_nid_t nid, __u32 ip);
408 extern int kibnal_del_peer(lnet_nid_t nid);
409 extern kib_peer_t *kibnal_find_peer_locked(lnet_nid_t nid);
410 extern void kibnal_unlink_peer_locked(kib_peer_t *peer);
411 extern void kibnal_peer_alive(kib_peer_t *peer);
412 extern int kibnal_close_stale_conns_locked(kib_peer_t *peer,
414 extern kib_conn_t *kibnal_create_conn(cm_cep_handle_t cep);
415 extern void kibnal_listen_callback(cm_cep_handle_t cep, cm_conn_data_t *info, void *arg);
417 extern int kibnal_alloc_pages(kib_pages_t **pp, int npages, int access);
418 extern void kibnal_free_pages(kib_pages_t *p);
420 extern void kibnal_check_sends(kib_conn_t *conn);
421 extern void kibnal_close_conn_locked(kib_conn_t *conn, int error);
422 extern void kibnal_destroy_conn(kib_conn_t *conn);
423 extern int kibnal_thread_start(int (*fn)(void *arg), void *arg);
424 extern int kibnal_scheduler(void *arg);
425 extern int kibnal_connd(void *arg);
426 extern void kibnal_init_tx_msg(kib_tx_t *tx, int type, int body_nob);
427 extern void kibnal_close_conn(kib_conn_t *conn, int why);
428 extern int kibnal_set_qp_state(kib_conn_t *conn, vv_qp_state_t new_state);
429 extern void kibnal_async_callback(vv_event_record_t ev);
430 extern void kibnal_cq_callback(unsigned long context);
431 extern void kibnal_passive_connreq(kib_pcreq_t *pcr, int reject);
432 extern void kibnal_txlist_done (struct list_head *txlist, int status);
433 extern void kibnal_queue_tx(kib_tx_t *tx, kib_conn_t *conn);
434 extern int kibnal_init_rdma(kib_tx_t *tx, int type, int nob,
435 kib_rdma_desc_t *dstrd, __u64 dstcookie);
436 extern int kibnal_tunables_init(void);
437 extern void kibnal_tunables_fini(void);
439 #define kibnal_conn_addref(conn) \
441 CDEBUG(D_NET, "conn[%p] (%d)++\n", \
442 (conn), atomic_read(&(conn)->ibc_refcount)); \
443 LASSERT(atomic_read(&(conn)->ibc_refcount) > 0); \
444 atomic_inc(&(conn)->ibc_refcount); \
447 #define kibnal_conn_decref(conn) \
449 unsigned long flags; \
451 CDEBUG(D_NET, "conn[%p] (%d)--\n", \
452 (conn), atomic_read(&(conn)->ibc_refcount)); \
453 LASSERT(atomic_read(&(conn)->ibc_refcount) > 0); \
454 if (atomic_dec_and_test(&(conn)->ibc_refcount)) { \
455 spin_lock_irqsave(&kibnal_data.kib_connd_lock, flags); \
456 list_add_tail(&(conn)->ibc_list, \
457 &kibnal_data.kib_connd_zombies); \
458 wake_up(&kibnal_data.kib_connd_waitq); \
459 spin_unlock_irqrestore(&kibnal_data.kib_connd_lock, flags); \
463 #define kibnal_peer_addref(peer) \
465 CDEBUG(D_NET, "peer[%p] -> %s (%d)++\n", \
466 (peer), libcfs_nid2str((peer)->ibp_nid), \
467 atomic_read (&(peer)->ibp_refcount)); \
468 LASSERT(atomic_read(&(peer)->ibp_refcount) > 0); \
469 atomic_inc(&(peer)->ibp_refcount); \
472 #define kibnal_peer_decref(peer) \
474 CDEBUG(D_NET, "peer[%p] -> %s (%d)--\n", \
475 (peer), libcfs_nid2str((peer)->ibp_nid), \
476 atomic_read (&(peer)->ibp_refcount)); \
477 LASSERT(atomic_read(&(peer)->ibp_refcount) > 0); \
478 if (atomic_dec_and_test(&(peer)->ibp_refcount)) \
479 kibnal_destroy_peer(peer); \
482 static inline struct list_head *
483 kibnal_nid2peerlist (lnet_nid_t nid)
485 unsigned int hash = ((unsigned int)nid) % kibnal_data.kib_peer_hash_size;
487 return (&kibnal_data.kib_peers [hash]);
491 kibnal_peer_active (kib_peer_t *peer)
493 /* Am I in the peer hash table? */
494 return (!list_empty(&peer->ibp_list));
498 kibnal_queue_tx_locked (kib_tx_t *tx, kib_conn_t *conn)
502 LASSERT (tx->tx_nwrq > 0); /* work items set up */
503 LASSERT (!tx->tx_queued); /* not queued for sending already */
506 tx->tx_deadline = jiffies + (*kibnal_tunables.kib_timeout * HZ);
508 if (tx->tx_conn == NULL) {
509 kibnal_conn_addref(conn);
511 LASSERT (tx->tx_msg->ibm_type != IBNAL_MSG_PUT_DONE);
513 LASSERT (tx->tx_conn == conn);
514 LASSERT (tx->tx_msg->ibm_type == IBNAL_MSG_PUT_DONE);
517 if (conn->ibc_version == IBNAL_MSG_VERSION_RDMAREPLYNOTRSRVD) {
518 /* All messages have simple credit control */
519 q = &conn->ibc_tx_queue;
521 LASSERT (conn->ibc_version == IBNAL_MSG_VERSION);
523 switch (tx->tx_msg->ibm_type) {
524 case IBNAL_MSG_PUT_REQ:
525 case IBNAL_MSG_GET_REQ:
526 /* RDMA request: reserve a buffer for the RDMA reply
528 q = &conn->ibc_tx_queue_rsrvd;
531 case IBNAL_MSG_PUT_NAK:
532 case IBNAL_MSG_PUT_ACK:
533 case IBNAL_MSG_PUT_DONE:
534 case IBNAL_MSG_GET_DONE:
535 /* RDMA reply/completion: no credits; peer has reserved
537 q = &conn->ibc_tx_queue_nocred;
541 case IBNAL_MSG_IMMEDIATE:
542 /* Otherwise: consume a credit before sending */
543 q = &conn->ibc_tx_queue;
552 list_add_tail(&tx->tx_list, q);
556 kibnal_send_keepalive(kib_conn_t *conn)
558 return (*kibnal_tunables.kib_keepalive > 0) &&
559 time_after(jiffies, conn->ibc_last_send +
560 *kibnal_tunables.kib_keepalive*HZ);
563 #ifndef IBNAL_VOIDSTAR_SGADDR
564 # define IBNAL_VOIDSTAR_SGADDR 0
567 #if IBNAL_VOIDSTAR_SGADDR
569 # if CONFIG_X86 && CONFIG_HIGHMEM4G
570 /* truncation to void* doesn't matter if 0 <= physmem < 4G
571 * so allow x86 with 32 bit phys addrs */
573 /* OK anyway on 64-bit arch */
575 # error "Can't support HIGHMEM when vv_scatgat_t::v_address is void *"
578 # define KIBNAL_ADDR2SG(a) ((void *)((unsigned long)(a)))
579 # define KIBNAL_SG2ADDR(a) ((__u64)((unsigned long)(a)))
580 static inline __u64 kibnal_addr2net (__u64 addr)
583 vv_return_t vvrc = vv_va2advertise_addr(kibnal_data.kib_hca,
584 KIBNAL_ADDR2SG(addr),
586 LASSERT (vvrc == vv_return_ok);
587 return KIBNAL_SG2ADDR(netaddr);
590 # define KIBNAL_ADDR2SG(a) a
591 # define KIBNAL_SG2ADDR(a) a
592 static inline __u64 kibnal_addr2net (__u64 addr)
595 vv_return_t vvrc = vv_va2advertise_addr(kibnal_data.kib_hca,
598 LASSERT (vvrc == vv_return_ok);
603 /* CAVEAT EMPTOR: We rely on tx/rx descriptor alignment to allow us to use the
604 * lowest 2 bits of the work request id to stash the work item type (the op
605 * field is not valid when the wc completes in error). */
607 #define IBNAL_WID_TX 0
608 #define IBNAL_WID_RX 1
609 #define IBNAL_WID_RDMA 2
610 #define IBNAL_WID_MASK 3UL
612 static inline vv_wr_id_t
613 kibnal_ptr2wreqid (void *ptr, int type)
615 unsigned long lptr = (unsigned long)ptr;
617 LASSERT ((lptr & IBNAL_WID_MASK) == 0);
618 LASSERT ((type & ~IBNAL_WID_MASK) == 0);
619 return (vv_wr_id_t)(lptr | type);
623 kibnal_wreqid2ptr (vv_wr_id_t wreqid)
625 return (void *)(((unsigned long)wreqid) & ~IBNAL_WID_MASK);
629 kibnal_wreqid2type (vv_wr_id_t wreqid)
631 return (wreqid & IBNAL_WID_MASK);
635 kibnal_set_conn_state (kib_conn_t *conn, int state)
637 conn->ibc_state = state;
644 kibnal_rd_size (kib_rdma_desc_t *rd)
651 kibnal_rf_addr (kib_rdma_frag_t *rf)
653 return (((__u64)rf->rf_addr_hi)<<32) | ((__u64)rf->rf_addr_lo);
657 kibnal_rf_set (kib_rdma_frag_t *rf, __u64 addr, int nob)
659 rf->rf_addr_lo = addr & 0xffffffff;
660 rf->rf_addr_hi = (addr >> 32) & 0xffffffff;
665 kibnal_rd_size (kib_rdma_desc_t *rd)
670 for (i = size = 0; i < rd->rd_nfrag; i++)
671 size += rd->rd_frags[i].rf_nob;