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
29 #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>
51 #include <linux/random.h>
56 #define DEBUG_SUBSYSTEM S_NAL
58 #include <libcfs/kp30.h>
59 #include <portals/p30.h>
60 #include <portals/lib-p30.h>
61 #include <portals/nal.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 /* sdp-connection.c */
93 #define IBNAL_PKEY 0xffff
94 #define IBNAL_PKEY_IDX 0
95 #define IBNAL_SGID_IDX 0
96 #define IBNAL_SERVICE_LEVEL 0
97 #define IBNAL_STATIC_RATE 0
98 #define IBNAL_RETRY_CNT 7
99 #define IBNAL_RNR_CNT 7
100 #define IBNAL_EE_FLOW_CNT 1
101 #define IBNAL_LOCAL_SUB 1
102 #define IBNAL_TRAFFIC_CLASS 0
103 #define IBNAL_SOURCE_PATH_BIT 0
104 #define IBNAL_OUS_DST_RD 32
105 #define IBNAL_IB_MTU vv_mtu_1024
107 /* sdp-hca-params.h */
108 #define PATH_RATE_2_5GB 2
111 #define IBNAL_R_2_STATIC_RATE(r) ((r) == PATH_RATE_2_5GB ? MLX_IPD_1x : MLX_IPD_4x)
113 /* other low-level IB constants */
114 #define IBNAL_LOCAL_ACK_TIMEOUT 0x12
115 #define IBNAL_PKT_LIFETIME 5
116 #define IBNAL_ARB_INITIATOR_DEPTH 0
117 #define IBNAL_ARB_RESP_RES 0
118 #define IBNAL_FAILOVER_ACCEPTED 0
119 #define IBNAL_SERVICE_NUMBER 0x11b9a2 /* Fixed service number */
121 #define IBNAL_MIN_RECONNECT_INTERVAL HZ /* first failed connection retry... */
122 #define IBNAL_MAX_RECONNECT_INTERVAL (60*HZ) /* ...exponentially increasing to this */
124 #define IBNAL_MSG_SIZE (4<<10) /* max size of queued messages (inc hdr) */
126 #define IBNAL_MSG_QUEUE_SIZE 8 /* # messages/RDMAs in-flight */
127 #define IBNAL_CREDIT_HIGHWATER 7 /* when to eagerly return credits */
129 #define IBNAL_ARP_RETRIES 3 /* How many times to retry ARP */
131 #define IBNAL_NTX 64 /* # tx descs */
132 #define IBNAL_NTX_NBLK 128 /* # reserved tx descs */
133 /* reduced from 256 to ensure we register < 255 pages per region.
134 * this can change if we register all memory. */
136 #define IBNAL_PEER_HASH_SIZE 101 /* # peer lists */
138 #define IBNAL_RESCHED 100 /* # scheduler loops before reschedule */
140 #define IBNAL_CONCURRENT_PEERS 1000 /* # nodes all talking at once to me */
142 #define IBNAL_RDMA_BASE 0x0eeb0000
143 #define IBNAL_CKSUM 0
144 #define IBNAL_WHOLE_MEM 1
146 # error "incompatible with voltaire adaptor-tavor (REGISTER_RAM_IN_ONE_PHY_MR)"
149 /* default vals for runtime tunables */
150 #define IBNAL_IO_TIMEOUT 50 /* default comms timeout (seconds) */
152 /************************/
153 /* derived constants... */
155 /* TX messages (shared by all connections) */
156 #define IBNAL_TX_MSGS (IBNAL_NTX + IBNAL_NTX_NBLK)
157 #define IBNAL_TX_MSG_BYTES (IBNAL_TX_MSGS * IBNAL_MSG_SIZE)
158 #define IBNAL_TX_MSG_PAGES ((IBNAL_TX_MSG_BYTES + PAGE_SIZE - 1)/PAGE_SIZE)
161 # define IBNAL_MAX_RDMA_FRAGS PTL_MD_MAX_IOV
163 # define IBNAL_MAX_RDMA_FRAGS 1
166 /* RX messages (per connection) */
167 #define IBNAL_RX_MSGS IBNAL_MSG_QUEUE_SIZE
168 #define IBNAL_RX_MSG_BYTES (IBNAL_RX_MSGS * IBNAL_MSG_SIZE)
169 #define IBNAL_RX_MSG_PAGES ((IBNAL_RX_MSG_BYTES + PAGE_SIZE - 1)/PAGE_SIZE)
171 #define IBNAL_CQ_ENTRIES (IBNAL_TX_MSGS * (1 + IBNAL_MAX_RDMA_FRAGS) + \
172 IBNAL_RX_MSGS * IBNAL_CONCURRENT_PEERS)
176 int kib_io_timeout; /* comms timeout (seconds) */
177 struct ctl_table_header *kib_sysctl; /* sysctl interface */
182 int ibp_npages; /* # pages */
183 int ibp_mapped; /* mapped? */
184 __u64 ibp_vaddr; /* mapped region vaddr */
185 __u32 ibp_lkey; /* mapped region lkey */
186 __u32 ibp_rkey; /* mapped region rkey */
187 vv_mem_reg_h_t ibp_handle; /* mapped region handle */
188 struct page *ibp_pages[0];
193 vv_mem_reg_h_t md_handle;
201 int kib_init; /* initialisation state */
202 __u64 kib_incarnation; /* which one am I */
203 int kib_shutdown; /* shut down? */
204 atomic_t kib_nthreads; /* # live threads */
206 __u64 kib_svc_id; /* service number I listen on */
207 vv_gid_t kib_port_gid; /* device/port GID */
208 vv_p_key_t kib_port_pkey; /* device/port pkey */
210 struct semaphore kib_nid_mutex; /* serialise NID ops */
211 cm_cep_handle_t kib_listen_handle; /* IB listen handle */
213 rwlock_t kib_global_lock; /* stabilize peer/conn ops */
214 spinlock_t kib_vverbs_lock; /* serialize vverbs calls */
215 int kib_ready; /* CQ callback fired */
216 int kib_checking_cq; /* a scheduler is checking the CQ */
218 struct list_head *kib_peers; /* hash table of all my known peers */
219 int kib_peer_hash_size; /* size of kib_peers */
220 atomic_t kib_npeers; /* # peers extant */
221 atomic_t kib_nconns; /* # connections extant */
223 void *kib_connd; /* the connd task (serialisation assertions) */
224 struct list_head kib_connd_peers; /* peers wanting to get connected */
225 struct list_head kib_connd_pcreqs; /* passive connection requests */
226 struct list_head kib_connd_conns; /* connections to setup/teardown */
227 struct list_head kib_connd_zombies; /* connections with zero refcount */
228 wait_queue_head_t kib_connd_waitq; /* connection daemon sleeps here */
229 spinlock_t kib_connd_lock; /* serialise */
231 wait_queue_head_t kib_sched_waitq; /* schedulers sleep here */
232 struct list_head kib_sched_txq; /* tx requiring attention */
233 struct list_head kib_sched_rxq; /* rx requiring attention */
234 spinlock_t kib_sched_lock; /* serialise */
236 struct kib_tx *kib_tx_descs; /* all the tx descriptors */
237 kib_pages_t *kib_tx_pages; /* premapped tx msg pages */
239 struct list_head kib_idle_txs; /* idle tx descriptors */
240 struct list_head kib_idle_nblk_txs; /* idle reserved tx descriptors */
241 wait_queue_head_t kib_idle_tx_waitq; /* block here for tx descriptor */
242 __u64 kib_next_tx_cookie; /* RDMA completion cookie */
243 spinlock_t kib_tx_lock; /* serialise */
245 vv_hca_h_t kib_hca; /* The HCA */
246 vv_hca_attrib_t kib_hca_attrs; /* its properties */
247 int kib_port; /* port on the device */
248 vv_port_attrib_t kib_port_attr; /* its properties */
250 vv_pd_h_t kib_pd; /* protection domain */
251 vv_cq_h_t kib_cq; /* completion queue */
255 #define IBNAL_INIT_NOTHING 0
256 #define IBNAL_INIT_DATA 1
257 #define IBNAL_INIT_LIB 2
258 #define IBNAL_INIT_HCA 3
259 #define IBNAL_INIT_ASYNC 4
260 #define IBNAL_INIT_PD 5
261 #define IBNAL_INIT_TXD 6
262 #define IBNAL_INIT_CQ 7
263 #define IBNAL_INIT_ALL 8
265 #include "vibnal_wire.h"
267 /***********************************************************************/
269 typedef struct kib_rx /* receive message */
271 struct list_head rx_list; /* queue for attention */
272 struct kib_conn *rx_conn; /* owning conn */
273 int rx_responded; /* responded to peer? */
274 int rx_posted; /* posted? */
276 vv_l_key_t rx_lkey; /* local key */
278 __u64 rx_vaddr; /* pre-mapped buffer (hca vaddr) */
280 kib_msg_t *rx_msg; /* pre-mapped buffer (host vaddr) */
281 vv_wr_t rx_wrq; /* receive work item */
282 vv_scatgat_t rx_gl; /* and its memory */
286 # define KIBNAL_RX_VADDR(rx) ((__u64)((unsigned long)((rx)->rx_msg)))
287 # define KIBNAL_RX_LKEY(rx) ((rx)->rx_lkey)
289 # define KIBNAL_RX_VADDR(rx) ((rx)->rx_vaddr)
290 # define KIBNAL_RX_LKEY(rx) ((rx)->rx_conn->ibc_rx_pages->ibp_lkey)
293 typedef struct kib_tx /* transmit message */
295 struct list_head tx_list; /* queue on idle_txs ibc_tx_queue etc. */
296 int tx_isnblk; /* I'm reserved for non-blocking sends */
297 struct kib_conn *tx_conn; /* owning conn */
298 int tx_mapped; /* mapped for RDMA? */
299 int tx_sending; /* # tx callbacks outstanding */
300 int tx_queued; /* queued for sending */
301 int tx_waiting; /* waiting for peer */
302 int tx_status; /* completion status */
303 unsigned long tx_deadline; /* completion deadline */
304 __u64 tx_cookie; /* completion cookie */
305 lib_msg_t *tx_libmsg[2]; /* lib msgs to finalize on completion */
307 vv_l_key_t tx_lkey; /* local key for message buffer */
309 kib_md_t tx_md; /* RDMA mapping (active/passive) */
310 __u64 tx_vaddr; /* pre-mapped buffer (hca vaddr) */
312 kib_msg_t *tx_msg; /* message buffer (host vaddr) */
313 int tx_nwrq; /* # send work items */
314 vv_wr_t *tx_wrq; /* send work items... */
315 vv_scatgat_t *tx_gl; /* ...and their memory */
316 kib_rdma_desc_t *tx_rd; /* rdma descriptor (src buffers) */
320 # define KIBNAL_TX_VADDR(tx) ((__u64)((unsigned long)((tx)->tx_msg)))
321 # define KIBNAL_TX_LKEY(tx) ((tx)->tx_lkey)
323 # define KIBNAL_TX_VADDR(tx) ((tx)->tx_vaddr)
324 # define KIBNAL_TX_LKEY(tx) (kibnal_data.kib_tx_pages->ibp_lkey)
327 #define KIB_TX_UNMAPPED 0
328 #define KIB_TX_MAPPED 1
330 /* Passive connection request (listener callback) queued for handling by connd */
331 typedef struct kib_pcreq
333 struct list_head pcr_list; /* queue for handling by connd */
334 cm_cep_handle_t pcr_cep; /* listening handle */
335 cm_request_data_t pcr_cmreq; /* request data */
338 typedef struct kib_connvars
340 /* connection-in-progress variables */
349 ib_path_record_v2_t cv_path;
350 ibat_arp_data_t cv_arp;
351 ibat_stat_t cv_arprc;
352 cm_conn_data_t cv_conndata;
355 typedef struct kib_conn
357 struct kib_peer *ibc_peer; /* owning peer */
358 struct list_head ibc_list; /* stash on peer's conn list */
359 __u64 ibc_incarnation; /* which instance of the peer */
360 __u64 ibc_txseq; /* tx sequence number */
361 __u64 ibc_rxseq; /* rx sequence number */
362 atomic_t ibc_refcount; /* # users */
363 int ibc_state; /* what's happening */
364 atomic_t ibc_nob; /* # bytes buffered */
365 int ibc_nsends_posted; /* # uncompleted sends */
366 int ibc_credits; /* # credits I have */
367 int ibc_outstanding_credits; /* # credits to return */
368 int ibc_disconnect; /* some disconnect callback fired */
369 int ibc_comms_error; /* set on comms error */
370 struct list_head ibc_early_rxs; /* rxs completed before ESTABLISHED */
371 struct list_head ibc_tx_queue; /* send queue */
372 struct list_head ibc_active_txs; /* active tx awaiting completion */
373 spinlock_t ibc_lock; /* serialise */
374 kib_rx_t *ibc_rxs; /* the rx descs */
375 kib_pages_t *ibc_rx_pages; /* premapped rx msg pages */
376 vv_qp_h_t ibc_qp; /* queue pair */
377 cm_cep_handle_t ibc_cep; /* connection endpoint */
378 kib_connvars_t *ibc_connvars; /* in-progress connection state */
381 #define IBNAL_CONN_INIT_NOTHING 0 /* incomplete init */
382 #define IBNAL_CONN_INIT 1 /* completed init */
383 #define IBNAL_CONN_ACTIVE_ARP 2 /* active arping */
384 #define IBNAL_CONN_ACTIVE_CONNECT 3 /* active sending req */
385 #define IBNAL_CONN_ACTIVE_CHECK_REPLY 4 /* active checking reply */
386 #define IBNAL_CONN_ACTIVE_RTU 5 /* active sending rtu */
387 #define IBNAL_CONN_PASSIVE_WAIT 6 /* passive waiting for rtu */
388 #define IBNAL_CONN_ESTABLISHED 7 /* connection established */
389 #define IBNAL_CONN_DISCONNECT1 8 /* disconnect phase 1 */
390 #define IBNAL_CONN_DISCONNECT2 9 /* disconnect phase 2 */
391 #define IBNAL_CONN_DISCONNECTED 10 /* disconnect complete */
393 typedef struct kib_peer
395 struct list_head ibp_list; /* stash on global peer list */
396 struct list_head ibp_connd_list; /* schedule on kib_connd_peers */
397 ptl_nid_t ibp_nid; /* who's on the other end(s) */
398 __u32 ibp_ip; /* IP to query for peer conn params */
399 int ibp_port; /* port to qery for peer conn params */
400 __u64 ibp_incarnation; /* peer's incarnation */
401 atomic_t ibp_refcount; /* # users */
402 int ibp_persistence; /* "known" peer refs */
403 struct list_head ibp_conns; /* all active connections */
404 struct list_head ibp_tx_queue; /* msgs waiting for a conn */
405 int ibp_connecting; /* connecting+accepting */
406 int ibp_arp_count; /* # arp attempts */
407 unsigned long ibp_reconnect_time; /* when reconnect may be attempted */
408 unsigned long ibp_reconnect_interval; /* exponential backoff */
412 extern lib_nal_t kibnal_lib;
413 extern kib_data_t kibnal_data;
414 extern kib_tunables_t kibnal_tunables;
416 extern void kibnal_init_msg(kib_msg_t *msg, int type, int body_nob);
417 extern void kibnal_pack_msg(kib_msg_t *msg, int credits, ptl_nid_t dstnid,
418 __u64 dststamp, __u64 seq);
419 extern int kibnal_unpack_msg(kib_msg_t *msg, int nob);
420 extern kib_peer_t *kibnal_create_peer(ptl_nid_t nid);
421 extern void kibnal_destroy_peer(kib_peer_t *peer);
422 extern int kibnal_del_peer(ptl_nid_t nid, int single_share);
423 extern kib_peer_t *kibnal_find_peer_locked(ptl_nid_t nid);
424 extern void kibnal_unlink_peer_locked(kib_peer_t *peer);
425 extern int kibnal_close_stale_conns_locked(kib_peer_t *peer,
427 extern kib_conn_t *kibnal_create_conn(cm_cep_handle_t cep);
428 extern void kibnal_listen_callback(cm_cep_handle_t cep, cm_conn_data_t *info, void *arg);
430 extern int kibnal_alloc_pages(kib_pages_t **pp, int npages, int access);
431 extern void kibnal_free_pages(kib_pages_t *p);
433 extern void kibnal_check_sends(kib_conn_t *conn);
434 extern void kibnal_close_conn_locked(kib_conn_t *conn, int error);
435 extern void kibnal_destroy_conn(kib_conn_t *conn);
436 extern int kibnal_thread_start(int (*fn)(void *arg), void *arg);
437 extern int kibnal_scheduler(void *arg);
438 extern int kibnal_connd(void *arg);
439 extern void kibnal_init_tx_msg(kib_tx_t *tx, int type, int body_nob);
440 extern void kibnal_close_conn(kib_conn_t *conn, int why);
441 extern int kibnal_set_qp_state(kib_conn_t *conn, vv_qp_state_t new_state);
442 extern void kibnal_async_callback(vv_event_record_t ev);
443 extern void kibnal_cq_callback(unsigned long context);
444 extern void kibnal_passive_connreq(kib_pcreq_t *pcr, int reject);
445 extern void kibnal_pause(int ticks);
446 extern void kibnal_queue_tx(kib_tx_t *tx, kib_conn_t *conn);
447 extern int kibnal_init_rdma(kib_tx_t *tx, int type, int nob,
448 kib_rdma_desc_t *dstrd, __u64 dstcookie);
451 wrq_signals_completion (vv_wr_t *wrq)
453 return wrq->completion_notification != 0;
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] -> "LPX64" (%d)++\n", \
483 (peer), (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] -> "LPX64" (%d)--\n", \
492 (peer), (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 (ptl_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)
517 /* CAVEAT EMPTOR: tx takes caller's ref on conn */
519 LASSERT (tx->tx_nwrq > 0); /* work items set up */
520 LASSERT (!tx->tx_queued); /* not queued for sending already */
522 if (tx->tx_conn == NULL) {
523 kibnal_conn_addref(conn);
526 LASSERT (tx->tx_conn == conn);
527 LASSERT (tx->tx_msg->ibm_type == IBNAL_MSG_PUT_DONE);
530 tx->tx_deadline = jiffies + kibnal_tunables.kib_io_timeout * HZ;
531 list_add_tail(&tx->tx_list, &conn->ibc_tx_queue);
535 kibnal_page2phys (struct page *p)
537 #if IBNAL_32BIT_PAGE2PHYS
538 CLASSERT (sizeof(typeof(page_to_phys(p))) == 4);
539 CLASSERT (sizeof(unsigned long) == 4);
540 /* page_to_phys returns a 32 bit physical address. This must be a 32
541 * bit machine with <= 4G memory and we must ensure we don't sign
542 * extend when converting to 64 bits. */
543 return (unsigned long)page_to_phys(p);
545 CLASSERT (sizeof(typeof(page_to_phys(p))) == 8);
546 /* page_to_phys returns a 64 bit physical address :) */
547 return page_to_phys(p);
551 #if IBNAL_VOIDSTAR_SGADDR
553 # if CONFIG_X86 && CONFIG_HIGHMEM4G
554 /* truncation to void* doesn't matter if 0 <= physmem < 4G
555 * so allow x86 with 32 bit phys addrs */
557 /* OK anyway on 64-bit arch */
559 # error "Can't support HIGHMEM when vv_scatgat_t::v_address is void *"
562 # define KIBNAL_ADDR2SG(a) ((void *)((unsigned long)(a)))
563 # define KIBNAL_SG2ADDR(a) ((__u64)((unsigned long)(a)))
564 static inline __u64 kibnal_addr2net (__u64 addr)
567 vv_return_t vvrc = vv_va2advertise_addr(kibnal_data.kib_hca,
568 KIBNAL_ADDR2SG(addr),
570 LASSERT (vvrc == vv_return_ok);
571 return KIBNAL_SG2ADDR(netaddr);
574 # define KIBNAL_ADDR2SG(a) a
575 # define KIBNAL_SG2ADDR(a) a
576 static inline __u64 kibnal_addr2net (__u64 addr)
579 vv_return_t vvrc = vv_va2advertise_addr(kibnal_data.kib_hca,
582 LASSERT (vvrc == vv_return_ok);
587 /* CAVEAT EMPTOR: We rely on tx/rx descriptor alignment to allow us to use the
588 * lowest 2 bits of the work request id to stash the work item type (the op
589 * field is not valid when the wc completes in error). */
591 #define IBNAL_WID_TX 0
592 #define IBNAL_WID_RX 1
593 #define IBNAL_WID_RDMA 2
594 #define IBNAL_WID_MASK 3UL
596 static inline vv_wr_id_t
597 kibnal_ptr2wreqid (void *ptr, int type)
599 unsigned long lptr = (unsigned long)ptr;
601 LASSERT ((lptr & IBNAL_WID_MASK) == 0);
602 LASSERT ((type & ~IBNAL_WID_MASK) == 0);
603 return (vv_wr_id_t)(lptr | type);
607 kibnal_wreqid2ptr (vv_wr_id_t wreqid)
609 return (void *)(((unsigned long)wreqid) & ~IBNAL_WID_MASK);
613 kibnal_wreqid2type (vv_wr_id_t wreqid)
615 return (wreqid & IBNAL_WID_MASK);
619 kibnal_set_conn_state (kib_conn_t *conn, int state)
621 conn->ibc_state = state;
626 kibnal_rf_addr (kib_rdma_frag_t *rf)
628 return (((__u64)rf->rf_addr_hi)<<32) | ((__u64)rf->rf_addr_lo);
632 kibnal_rf_set (kib_rdma_frag_t *rf, __u64 addr, int nob)
634 rf->rf_addr_lo = addr & 0xffffffff;
635 rf->rf_addr_hi = (addr >> 32) & 0xffffffff;
640 kibnal_rd_size (kib_rdma_desc_t *rd)
645 for (i = size = 0; i < rd->rd_nfrag; i++)
646 size += rd->rd_frags[i].rf_nob;