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 < 30203
82 #error Invalid GCC version. Must use 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_NTX 64 /* # tx descs */
130 #define IBNAL_NTX_NBLK 128 /* # reserved tx descs */
131 /* reduced from 256 to ensure we register < 255 pages per region.
132 * this can change if we register all memory. */
134 #define IBNAL_PEER_HASH_SIZE 101 /* # peer lists */
136 #define IBNAL_RESCHED 100 /* # scheduler loops before reschedule */
138 #define IBNAL_CONCURRENT_PEERS 1000 /* # nodes all talking at once to me */
140 #define IBNAL_RDMA_BASE 0x0eeb0000
141 #define IBNAL_CKSUM 0
142 #define IBNAL_WHOLE_MEM 1
144 # error "incompatible with voltaire adaptor-tavor (REGISTER_RAM_IN_ONE_PHY_MR)"
147 /* default vals for runtime tunables */
148 #define IBNAL_IO_TIMEOUT 50 /* default comms timeout (seconds) */
150 /************************/
151 /* derived constants... */
153 /* TX messages (shared by all connections) */
154 #define IBNAL_TX_MSGS (IBNAL_NTX + IBNAL_NTX_NBLK)
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)
159 # define IBNAL_MAX_RDMA_FRAGS PTL_MD_MAX_IOV
161 # define IBNAL_MAX_RDMA_FRAGS 1
164 /* RX messages (per connection) */
165 #define IBNAL_RX_MSGS IBNAL_MSG_QUEUE_SIZE
166 #define IBNAL_RX_MSG_BYTES (IBNAL_RX_MSGS * IBNAL_MSG_SIZE)
167 #define IBNAL_RX_MSG_PAGES ((IBNAL_RX_MSG_BYTES + PAGE_SIZE - 1)/PAGE_SIZE)
169 #define IBNAL_CQ_ENTRIES (IBNAL_TX_MSGS * (1 + IBNAL_MAX_RDMA_FRAGS) + \
170 IBNAL_RX_MSGS * IBNAL_CONCURRENT_PEERS)
174 int kib_io_timeout; /* comms timeout (seconds) */
175 struct ctl_table_header *kib_sysctl; /* sysctl interface */
180 int ibp_npages; /* # pages */
181 int ibp_mapped; /* mapped? */
182 __u64 ibp_vaddr; /* mapped region vaddr */
183 __u32 ibp_lkey; /* mapped region lkey */
184 __u32 ibp_rkey; /* mapped region rkey */
185 vv_mem_reg_h_t ibp_handle; /* mapped region handle */
186 struct page *ibp_pages[0];
191 vv_mem_reg_h_t md_handle;
199 int kib_init; /* initialisation state */
200 __u64 kib_incarnation; /* which one am I */
201 int kib_shutdown; /* shut down? */
202 atomic_t kib_nthreads; /* # live threads */
204 __u64 kib_svc_id; /* service number I listen on */
205 vv_gid_t kib_port_gid; /* device/port GID */
206 vv_p_key_t kib_port_pkey; /* device/port pkey */
208 struct semaphore kib_nid_mutex; /* serialise NID ops */
209 cm_cep_handle_t kib_listen_handle; /* IB listen handle */
211 rwlock_t kib_global_lock; /* stabilize peer/conn ops */
212 spinlock_t kib_vverbs_lock; /* serialize vverbs calls */
213 int kib_ready; /* CQ callback fired */
214 int kib_checking_cq; /* a scheduler is checking the CQ */
216 struct list_head *kib_peers; /* hash table of all my known peers */
217 int kib_peer_hash_size; /* size of kib_peers */
218 atomic_t kib_npeers; /* # peers extant */
219 atomic_t kib_nconns; /* # connections extant */
221 void *kib_connd; /* the connd task (serialisation assertions) */
222 struct list_head kib_connd_peers; /* peers wanting to get connected */
223 struct list_head kib_connd_pcreqs; /* passive connection requests */
224 struct list_head kib_connd_conns; /* connections to setup/teardown */
225 struct list_head kib_connd_zombies; /* connections with zero refcount */
226 wait_queue_head_t kib_connd_waitq; /* connection daemon sleeps here */
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 struct list_head kib_idle_nblk_txs; /* idle reserved tx descriptors */
239 wait_queue_head_t kib_idle_tx_waitq; /* block here for tx descriptor */
240 __u64 kib_next_tx_cookie; /* RDMA completion cookie */
241 spinlock_t kib_tx_lock; /* serialise */
243 vv_hca_h_t kib_hca; /* The HCA */
244 vv_hca_attrib_t kib_hca_attrs; /* its properties */
245 int kib_port; /* port on the device */
246 vv_port_attrib_t kib_port_attr; /* its properties */
248 vv_pd_h_t kib_pd; /* protection domain */
249 vv_cq_h_t kib_cq; /* completion queue */
253 #define IBNAL_INIT_NOTHING 0
254 #define IBNAL_INIT_DATA 1
255 #define IBNAL_INIT_LIB 2
256 #define IBNAL_INIT_HCA 3
257 #define IBNAL_INIT_ASYNC 4
258 #define IBNAL_INIT_PD 5
259 #define IBNAL_INIT_TXD 6
260 #define IBNAL_INIT_CQ 7
261 #define IBNAL_INIT_ALL 8
263 /************************************************************************
264 * IB Wire message format.
265 * These are sent in sender's byte order (i.e. receiver flips).
268 typedef struct kib_connparams
270 __u32 ibcp_queue_depth;
271 __u32 ibcp_max_msg_size;
272 __u32 ibcp_max_frags;
273 } kib_connparams_t __attribute__((packed));
277 ptl_hdr_t ibim_hdr; /* portals header */
278 char ibim_payload[0]; /* piggy-backed payload */
279 } kib_immediate_msg_t __attribute__((packed));
281 /* YEUCH! the __u64 address is split into 2 __u32 fields to ensure proper
282 * packing. Otherwise we can't fit enough frags into an IBNAL message (<=
283 * smallest page size on any arch). */
286 __u32 rf_nob; /* # of bytes */
287 __u32 rf_addr_lo; /* lo 4 bytes of vaddr */
288 __u32 rf_addr_hi; /* hi 4 bytes of vaddr */
289 } kib_rdma_frag_t __attribute__((packed));
293 __u32 rd_key; /* local/remote key */
294 __u32 rd_nfrag; /* # fragments */
295 kib_rdma_frag_t rd_frags[0]; /* buffer frags */
296 } kib_rdma_desc_t __attribute__((packed));
298 /* CAVEAT EMPTOR! We don't actually put ibprm_rd on the wire; it's just there
299 * to remember the source buffers while we wait for the PUT_ACK */
303 ptl_hdr_t ibprm_hdr; /* portals header */
304 __u64 ibprm_cookie; /* opaque completion cookie */
305 kib_rdma_frag_t ibprm_rd; /* source buffer */
306 } kib_putreq_msg_t __attribute__((packed));
310 __u64 ibpam_src_cookie; /* reflected completion cookie */
311 __u64 ibpam_dst_cookie; /* opaque completion cookie */
312 kib_rdma_desc_t ibpam_rd; /* sender's sink buffer */
313 } kib_putack_msg_t __attribute__((packed));
317 ptl_hdr_t ibgm_hdr; /* portals header */
318 __u64 ibgm_cookie; /* opaque completion cookie */
319 kib_rdma_desc_t ibgm_rd; /* rdma descriptor */
320 } kib_get_msg_t __attribute__((packed));
324 __u64 ibcm_cookie; /* opaque completion cookie */
325 __s32 ibcm_status; /* < 0 failure: >= 0 length */
326 } kib_completion_msg_t __attribute__((packed));
330 /* First 2 fields fixed FOR ALL TIME */
331 __u32 ibm_magic; /* I'm an openibnal message */
332 __u16 ibm_version; /* this is my version number */
334 __u8 ibm_type; /* msg type */
335 __u8 ibm_credits; /* returned credits */
336 __u32 ibm_nob; /* # bytes in whole message */
337 __u32 ibm_cksum; /* checksum (0 == no checksum) */
338 __u64 ibm_srcnid; /* sender's NID */
339 __u64 ibm_srcstamp; /* sender's incarnation */
340 __u64 ibm_dstnid; /* destination's NID */
341 __u64 ibm_dststamp; /* destination's incarnation */
344 kib_connparams_t connparams;
345 kib_immediate_msg_t immediate;
346 kib_putreq_msg_t putreq;
347 kib_putack_msg_t putack;
349 kib_completion_msg_t completion;
350 } ibm_u __attribute__((packed));
351 } kib_msg_t __attribute__((packed));
353 #define IBNAL_MSG_MAGIC 0x0be91b91 /* unique magic */
354 #define IBNAL_MSG_VERSION 4 /* current protocol version */
356 #define IBNAL_MSG_CONNREQ 0xc0 /* connection request */
357 #define IBNAL_MSG_CONNACK 0xc1 /* connection acknowledge */
358 #define IBNAL_MSG_NOOP 0xd0 /* nothing (just credits) */
359 #define IBNAL_MSG_IMMEDIATE 0xd1 /* immediate */
360 #define IBNAL_MSG_PUT_REQ 0xd2 /* putreq (src->sink) */
361 #define IBNAL_MSG_PUT_NAK 0xd3 /* completion (sink->src) */
362 #define IBNAL_MSG_PUT_ACK 0xd4 /* putack (sink->src) */
363 #define IBNAL_MSG_PUT_DONE 0xd5 /* completion (src->sink) */
364 #define IBNAL_MSG_GET_REQ 0xd6 /* getreq (sink->src) */
365 #define IBNAL_MSG_GET_DONE 0xd7 /* completion (src->sink: all OK) */
367 /***********************************************************************/
369 typedef struct kib_rx /* receive message */
371 struct list_head rx_list; /* queue for attention */
372 struct kib_conn *rx_conn; /* owning conn */
373 int rx_responded; /* responded to peer? */
374 int rx_posted; /* posted? */
376 vv_l_key_t rx_lkey; /* local key */
378 __u64 rx_vaddr; /* pre-mapped buffer (hca vaddr) */
380 kib_msg_t *rx_msg; /* pre-mapped buffer (host vaddr) */
381 vv_wr_t rx_wrq; /* receive work item */
382 vv_scatgat_t rx_gl; /* and its memory */
386 # define KIBNAL_RX_VADDR(rx) ((__u64)((unsigned long)((rx)->rx_msg)))
387 # define KIBNAL_RX_LKEY(rx) ((rx)->rx_lkey)
389 # define KIBNAL_RX_VADDR(rx) ((rx)->rx_vaddr)
390 # define KIBNAL_RX_LKEY(rx) ((rx)->rx_conn->ibc_rx_pages->ibp_lkey)
393 typedef struct kib_tx /* transmit message */
395 struct list_head tx_list; /* queue on idle_txs ibc_tx_queue etc. */
396 int tx_isnblk; /* I'm reserved for non-blocking sends */
397 struct kib_conn *tx_conn; /* owning conn */
398 int tx_mapped; /* mapped for RDMA? */
399 int tx_sending; /* # tx callbacks outstanding */
400 int tx_waiting; /* waiting for peer */
401 int tx_status; /* completion status */
402 unsigned long tx_deadline; /* completion deadline */
403 __u64 tx_cookie; /* completion cookie */
404 lib_msg_t *tx_libmsg[2]; /* lib msgs to finalize on completion */
406 vv_l_key_t tx_lkey; /* local key for message buffer */
408 kib_md_t tx_md; /* RDMA mapping (active/passive) */
409 __u64 tx_vaddr; /* pre-mapped buffer (hca vaddr) */
411 kib_msg_t *tx_msg; /* message buffer (host vaddr) */
412 int tx_nwrq; /* # send work items */
413 vv_wr_t *tx_wrq; /* send work items... */
414 vv_scatgat_t *tx_gl; /* ...and their memory */
415 kib_rdma_desc_t *tx_rd; /* rdma descriptor (src buffers) */
419 # define KIBNAL_TX_VADDR(tx) ((__u64)((unsigned long)((tx)->tx_msg)))
420 # define KIBNAL_TX_LKEY(tx) ((tx)->tx_lkey)
422 # define KIBNAL_TX_VADDR(tx) ((tx)->tx_vaddr)
423 # define KIBNAL_TX_LKEY(tx) (kibnal_data.kib_tx_pages->ibp_lkey)
426 #define KIB_TX_UNMAPPED 0
427 #define KIB_TX_MAPPED 1
429 /* Passive connection request (listener callback) queued for handling by connd */
430 typedef struct kib_pcreq
432 struct list_head pcr_list; /* queue for handling by connd */
433 cm_cep_handle_t pcr_cep; /* listening handle */
434 cm_request_data_t pcr_cmreq; /* request data */
437 typedef struct kib_connvars
439 /* connection-in-progress variables */
448 ib_path_record_v2_t cv_path;
449 ibat_arp_data_t cv_arp;
450 ibat_stat_t cv_arprc;
451 cm_conn_data_t cv_conndata;
454 typedef struct kib_conn
456 struct kib_peer *ibc_peer; /* owning peer */
457 struct list_head ibc_list; /* stash on peer's conn list */
458 __u64 ibc_incarnation; /* which instance of the peer */
459 atomic_t ibc_refcount; /* # users */
460 int ibc_state; /* what's happening */
461 atomic_t ibc_nob; /* # bytes buffered */
462 int ibc_nsends_posted; /* # uncompleted sends */
463 int ibc_credits; /* # credits I have */
464 int ibc_outstanding_credits; /* # credits to return */
465 int ibc_disconnect; /* some disconnect callback fired */
466 int ibc_comms_error; /* set on comms error */
467 struct list_head ibc_early_rxs; /* rxs completed before ESTABLISHED */
468 struct list_head ibc_tx_queue; /* send queue */
469 struct list_head ibc_active_txs; /* active tx awaiting completion */
470 spinlock_t ibc_lock; /* serialise */
471 kib_rx_t *ibc_rxs; /* the rx descs */
472 kib_pages_t *ibc_rx_pages; /* premapped rx msg pages */
473 vv_qp_h_t ibc_qp; /* queue pair */
474 cm_cep_handle_t ibc_cep; /* connection endpoint */
475 kib_connvars_t *ibc_connvars; /* in-progress connection state */
478 #define IBNAL_CONN_INIT_NOTHING 0 /* incomplete init */
479 #define IBNAL_CONN_INIT 1 /* completed init */
480 #define IBNAL_CONN_ACTIVE_ARP 2 /* active arping */
481 #define IBNAL_CONN_ACTIVE_CONNECT 3 /* active sending req */
482 #define IBNAL_CONN_ACTIVE_CHECK_REPLY 4 /* active checking reply */
483 #define IBNAL_CONN_ACTIVE_RTU 5 /* active sending rtu */
484 #define IBNAL_CONN_PASSIVE_WAIT 6 /* passive waiting for rtu */
485 #define IBNAL_CONN_ESTABLISHED 7 /* connection established */
486 #define IBNAL_CONN_DISCONNECT1 8 /* disconnect phase 1 */
487 #define IBNAL_CONN_DISCONNECT2 9 /* disconnect phase 2 */
488 #define IBNAL_CONN_DISCONNECTED 10 /* disconnect complete */
490 typedef struct kib_peer
492 struct list_head ibp_list; /* stash on global peer list */
493 struct list_head ibp_connd_list; /* schedule on kib_connd_peers */
494 ptl_nid_t ibp_nid; /* who's on the other end(s) */
495 __u32 ibp_ip; /* IP to query for peer conn params */
496 int ibp_port; /* port to qery for peer conn params */
497 __u64 ibp_incarnation; /* peer's incarnation */
498 atomic_t ibp_refcount; /* # users */
499 int ibp_persistence; /* "known" peer refs */
500 struct list_head ibp_conns; /* all active connections */
501 struct list_head ibp_tx_queue; /* msgs waiting for a conn */
502 int ibp_connecting; /* connecting+accepting */
503 unsigned long ibp_reconnect_time; /* when reconnect may be attempted */
504 unsigned long ibp_reconnect_interval; /* exponential backoff */
508 extern lib_nal_t kibnal_lib;
509 extern kib_data_t kibnal_data;
510 extern kib_tunables_t kibnal_tunables;
512 extern void kibnal_init_msg(kib_msg_t *msg, int type, int body_nob);
513 extern void kibnal_pack_msg(kib_msg_t *msg, int credits,
514 ptl_nid_t dstnid, __u64 dststamp);
515 extern int kibnal_unpack_msg(kib_msg_t *msg, int nob);
516 extern kib_peer_t *kibnal_create_peer(ptl_nid_t nid);
517 extern void kibnal_destroy_peer(kib_peer_t *peer);
518 extern int kibnal_del_peer(ptl_nid_t nid, int single_share);
519 extern kib_peer_t *kibnal_find_peer_locked(ptl_nid_t nid);
520 extern void kibnal_unlink_peer_locked(kib_peer_t *peer);
521 extern int kibnal_close_stale_conns_locked(kib_peer_t *peer,
523 extern kib_conn_t *kibnal_create_conn(cm_cep_handle_t cep);
524 extern void kibnal_listen_callback(cm_cep_handle_t cep, cm_conn_data_t *info, void *arg);
526 extern int kibnal_alloc_pages(kib_pages_t **pp, int npages, int access);
527 extern void kibnal_free_pages(kib_pages_t *p);
529 extern void kibnal_check_sends(kib_conn_t *conn);
530 extern void kibnal_close_conn_locked(kib_conn_t *conn, int error);
531 extern void kibnal_destroy_conn(kib_conn_t *conn);
532 extern int kibnal_thread_start(int (*fn)(void *arg), void *arg);
533 extern int kibnal_scheduler(void *arg);
534 extern int kibnal_connd(void *arg);
535 extern void kibnal_init_tx_msg(kib_tx_t *tx, int type, int body_nob);
536 extern void kibnal_close_conn(kib_conn_t *conn, int why);
537 extern int kibnal_set_qp_state(kib_conn_t *conn, vv_qp_state_t new_state);
538 extern void kibnal_async_callback(vv_event_record_t ev);
539 extern void kibnal_cq_callback(unsigned long context);
540 extern void kibnal_passive_connreq(kib_pcreq_t *pcr, int reject);
541 extern void kibnal_pause(int ticks);
542 extern void kibnal_queue_tx(kib_tx_t *tx, kib_conn_t *conn);
543 extern int kibnal_init_rdma(kib_tx_t *tx, int type, int nob,
544 kib_rdma_desc_t *dstrd, __u64 dstcookie);
547 wrq_signals_completion (vv_wr_t *wrq)
549 return wrq->completion_notification != 0;
553 kibnal_conn_addref (kib_conn_t *conn)
555 CDEBUG(D_NET, "++conn[%p] (%d)\n",
556 conn, atomic_read(&conn->ibc_refcount));
557 LASSERT(atomic_read(&conn->ibc_refcount) > 0);
558 atomic_inc(&conn->ibc_refcount);
562 kibnal_conn_decref (kib_conn_t *conn)
566 CDEBUG(D_NET, "--conn[%p] (%d)\n",
567 conn, atomic_read(&conn->ibc_refcount));
568 LASSERT(atomic_read(&conn->ibc_refcount) > 0);
569 if (atomic_dec_and_test(&conn->ibc_refcount)) {
570 spin_lock_irqsave(&kibnal_data.kib_connd_lock, flags);
571 list_add_tail(&conn->ibc_list, &kibnal_data.kib_connd_zombies);
572 wake_up(&kibnal_data.kib_connd_waitq);
573 spin_unlock_irqrestore(&kibnal_data.kib_connd_lock, flags);
578 kibnal_peer_addref (kib_peer_t *peer)
580 CDEBUG(D_NET, "++peer[%p] -> "LPX64" (%d)\n",
581 peer, peer->ibp_nid, atomic_read (&peer->ibp_refcount));
582 LASSERT(atomic_read(&peer->ibp_refcount) > 0);
583 atomic_inc(&peer->ibp_refcount);
587 kibnal_peer_decref (kib_peer_t *peer)
589 CDEBUG(D_NET, "--peer[%p] -> "LPX64" (%d)\n",
590 peer, peer->ibp_nid, atomic_read (&peer->ibp_refcount));
592 LASSERT(atomic_read(&peer->ibp_refcount) > 0);
593 if (atomic_dec_and_test (&peer->ibp_refcount))
594 kibnal_destroy_peer (peer);
597 static inline struct list_head *
598 kibnal_nid2peerlist (ptl_nid_t nid)
600 unsigned int hash = ((unsigned int)nid) % kibnal_data.kib_peer_hash_size;
602 return (&kibnal_data.kib_peers [hash]);
606 kibnal_peer_active (kib_peer_t *peer)
608 /* Am I in the peer hash table? */
609 return (!list_empty(&peer->ibp_list));
613 kibnal_queue_tx_locked (kib_tx_t *tx, kib_conn_t *conn)
615 /* CAVEAT EMPTOR: tx takes caller's ref on conn */
617 LASSERT (tx->tx_nwrq > 0); /* work items set up */
618 if (tx->tx_conn == NULL) {
619 kibnal_conn_addref(conn);
622 LASSERT (tx->tx_conn == conn);
623 LASSERT (tx->tx_msg->ibm_type == IBNAL_MSG_PUT_DONE);
625 tx->tx_deadline = jiffies + kibnal_tunables.kib_io_timeout * HZ;
626 list_add_tail(&tx->tx_list, &conn->ibc_tx_queue);
630 kibnal_page2phys (struct page *p)
632 return page_to_phys(p);
635 /* CAVEAT EMPTOR: We rely on tx/rx descriptor alignment to allow us to use the
636 * lowest 2 bits of the work request id to stash the work item type (the op
637 * field is not valid when the wc completes in error). */
639 #define IBNAL_WID_TX 0
640 #define IBNAL_WID_RX 1
641 #define IBNAL_WID_RDMA 2
642 #define IBNAL_WID_MASK 3UL
644 static inline vv_wr_id_t
645 kibnal_ptr2wreqid (void *ptr, int type)
647 unsigned long lptr = (unsigned long)ptr;
649 LASSERT ((lptr & IBNAL_WID_MASK) == 0);
650 LASSERT ((type & ~IBNAL_WID_MASK) == 0);
651 return (vv_wr_id_t)(lptr | type);
655 kibnal_wreqid2ptr (vv_wr_id_t wreqid)
657 return (void *)(((unsigned long)wreqid) & ~IBNAL_WID_MASK);
661 kibnal_wreqid2type (vv_wr_id_t wreqid)
663 return (wreqid & IBNAL_WID_MASK);
667 kibnal_set_conn_state (kib_conn_t *conn, int state)
669 conn->ibc_state = state;
674 kibnal_rf_addr (kib_rdma_frag_t *rf)
676 return (((__u64)rf->rf_addr_hi)<<32) | ((__u64)rf->rf_addr_lo);
680 kibnal_rf_set (kib_rdma_frag_t *rf, __u64 addr, int nob)
682 rf->rf_addr_lo = addr & 0xffffffff;
683 rf->rf_addr_hi = (addr >> 32) & 0xffffffff;
688 kibnal_rd_size (kib_rdma_desc_t *rd)
693 for (i = size = 0; i < rd->rd_nfrag; i++)
694 size += rd->rd_frags[i].rf_nob;