1 /* 2 * GPL HEADER START 3 * 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 6 * This program is free software; you can redistribute it and/or modify 7 * it under the terms of the GNU General Public License version 2 only, 8 * as published by the Free Software Foundation. 9 * 10 * This program is distributed in the hope that it will be useful, but 11 * WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 13 * General Public License version 2 for more details (a copy is included 14 * in the LICENSE file that accompanied this code). 15 * 16 * You should have received a copy of the GNU General Public License 17 * version 2 along with this program; If not, see 18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf 19 * 20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, 21 * CA 95054 USA or visit www.sun.com if you need additional information or 22 * have any questions. 23 * 24 * GPL HEADER END 25 */ 26 /* 27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. 28 * Use is subject to license terms. 29 * 30 * Copyright (c) 2012, Intel Corporation. 31 */ 32 /* 33 * This file is part of Lustre, http://www.lustre.org/ 34 * Lustre is a trademark of Sun Microsystems, Inc. 35 * 36 * lnet/include/lnet/lib-types.h 37 * 38 * Types used by the library side routines that do not need to be 39 * exposed to the user application 40 */ 41 42 #ifndef __LNET_LIB_TYPES_H__ 43 #define __LNET_LIB_TYPES_H__ 44 45 #include "linux/lib-types.h" 46 47 #include "../libcfs/libcfs.h" 48 #include <linux/list.h> 49 #include "types.h" 50 51 #define WIRE_ATTR __attribute__((packed)) 52 53 /* Packed version of lnet_process_id_t to transfer via network */ 54 typedef struct { 55 lnet_nid_t nid; 56 lnet_pid_t pid; /* node id / process id */ 57 } WIRE_ATTR lnet_process_id_packed_t; 58 59 /* The wire handle's interface cookie only matches one network interface in 60 * one epoch (i.e. new cookie when the interface restarts or the node 61 * reboots). The object cookie only matches one object on that interface 62 * during that object's lifetime (i.e. no cookie re-use). */ 63 typedef struct { 64 __u64 wh_interface_cookie; 65 __u64 wh_object_cookie; 66 } WIRE_ATTR lnet_handle_wire_t; 67 68 typedef enum { 69 LNET_MSG_ACK = 0, 70 LNET_MSG_PUT, 71 LNET_MSG_GET, 72 LNET_MSG_REPLY, 73 LNET_MSG_HELLO, 74 } lnet_msg_type_t; 75 76 /* The variant fields of the portals message header are aligned on an 8 77 * byte boundary in the message header. Note that all types used in these 78 * wire structs MUST be fixed size and the smaller types are placed at the 79 * end. */ 80 typedef struct lnet_ack { 81 lnet_handle_wire_t dst_wmd; 82 __u64 match_bits; 83 __u32 mlength; 84 } WIRE_ATTR lnet_ack_t; 85 86 typedef struct lnet_put { 87 lnet_handle_wire_t ack_wmd; 88 __u64 match_bits; 89 __u64 hdr_data; 90 __u32 ptl_index; 91 __u32 offset; 92 } WIRE_ATTR lnet_put_t; 93 94 typedef struct lnet_get { 95 lnet_handle_wire_t return_wmd; 96 __u64 match_bits; 97 __u32 ptl_index; 98 __u32 src_offset; 99 __u32 sink_length; 100 } WIRE_ATTR lnet_get_t; 101 102 typedef struct lnet_reply { 103 lnet_handle_wire_t dst_wmd; 104 } WIRE_ATTR lnet_reply_t; 105 106 typedef struct lnet_hello { 107 __u64 incarnation; 108 __u32 type; 109 } WIRE_ATTR lnet_hello_t; 110 111 typedef struct { 112 lnet_nid_t dest_nid; 113 lnet_nid_t src_nid; 114 lnet_pid_t dest_pid; 115 lnet_pid_t src_pid; 116 __u32 type; /* lnet_msg_type_t */ 117 __u32 payload_length; /* payload data to follow */ 118 /*<------__u64 aligned------->*/ 119 union { 120 lnet_ack_t ack; 121 lnet_put_t put; 122 lnet_get_t get; 123 lnet_reply_t reply; 124 lnet_hello_t hello; 125 } msg; 126 } WIRE_ATTR lnet_hdr_t; 127 128 /* A HELLO message contains a magic number and protocol version 129 * code in the header's dest_nid, the peer's NID in the src_nid, and 130 * LNET_MSG_HELLO in the type field. All other common fields are zero 131 * (including payload_size; i.e. no payload). 132 * This is for use by byte-stream LNDs (e.g. TCP/IP) to check the peer is 133 * running the same protocol and to find out its NID. These LNDs should 134 * exchange HELLO messages when a connection is first established. Individual 135 * LNDs can put whatever else they fancy in lnet_hdr_t::msg. 136 */ 137 typedef struct { 138 __u32 magic; /* LNET_PROTO_TCP_MAGIC */ 139 __u16 version_major; /* increment on incompatible change */ 140 __u16 version_minor; /* increment on compatible change */ 141 } WIRE_ATTR lnet_magicversion_t; 142 143 /* PROTO MAGIC for LNDs */ 144 #define LNET_PROTO_IB_MAGIC 0x0be91b91 145 #define LNET_PROTO_RA_MAGIC 0x0be91b92 146 #define LNET_PROTO_QSW_MAGIC 0x0be91b93 147 #define LNET_PROTO_GNI_MAGIC 0xb00fbabe /* ask Kim */ 148 #define LNET_PROTO_TCP_MAGIC 0xeebc0ded 149 #define LNET_PROTO_PTL_MAGIC 0x50746C4E /* 'PtlN' unique magic */ 150 #define LNET_PROTO_MX_MAGIC 0x4d583130 /* 'MX10'! */ 151 #define LNET_PROTO_ACCEPTOR_MAGIC 0xacce7100 152 #define LNET_PROTO_PING_MAGIC 0x70696E67 /* 'ping' */ 153 154 /* Placeholder for a future "unified" protocol across all LNDs */ 155 /* Current LNDs that receive a request with this magic will respond with a 156 * "stub" reply using their current protocol */ 157 #define LNET_PROTO_MAGIC 0x45726963 /* ! */ 158 159 #define LNET_PROTO_TCP_VERSION_MAJOR 1 160 #define LNET_PROTO_TCP_VERSION_MINOR 0 161 162 /* Acceptor connection request */ 163 typedef struct { 164 __u32 acr_magic; /* PTL_ACCEPTOR_PROTO_MAGIC */ 165 __u32 acr_version; /* protocol version */ 166 __u64 acr_nid; /* target NID */ 167 } WIRE_ATTR lnet_acceptor_connreq_t; 168 169 #define LNET_PROTO_ACCEPTOR_VERSION 1 170 171 /* forward refs */ 172 struct lnet_libmd; 173 174 typedef struct lnet_msg { 175 struct list_head msg_activelist; 176 struct list_head msg_list; /* Q for credits/MD */ 177 178 lnet_process_id_t msg_target; 179 /* where is it from, it's only for building event */ 180 lnet_nid_t msg_from; 181 __u32 msg_type; 182 183 /* committed for sending */ 184 unsigned int msg_tx_committed:1; 185 /* CPT # this message committed for sending */ 186 unsigned int msg_tx_cpt:15; 187 /* committed for receiving */ 188 unsigned int msg_rx_committed:1; 189 /* CPT # this message committed for receiving */ 190 unsigned int msg_rx_cpt:15; 191 /* queued for tx credit */ 192 unsigned int msg_tx_delayed:1; 193 /* queued for RX buffer */ 194 unsigned int msg_rx_delayed:1; 195 /* ready for pending on RX delay list */ 196 unsigned int msg_rx_ready_delay:1; 197 198 unsigned int msg_vmflush:1; /* VM trying to free memory */ 199 unsigned int msg_target_is_router:1; /* sending to a router */ 200 unsigned int msg_routing:1; /* being forwarded */ 201 unsigned int msg_ack:1; /* ack on finalize (PUT) */ 202 unsigned int msg_sending:1; /* outgoing message */ 203 unsigned int msg_receiving:1; /* being received */ 204 unsigned int msg_txcredit:1; /* taken an NI send credit */ 205 unsigned int msg_peertxcredit:1; /* taken a peer send credit */ 206 unsigned int msg_rtrcredit:1; /* taken a global router credit */ 207 unsigned int msg_peerrtrcredit:1; /* taken a peer router credit */ 208 unsigned int msg_onactivelist:1; /* on the activelist */ 209 210 struct lnet_peer *msg_txpeer; /* peer I'm sending to */ 211 struct lnet_peer *msg_rxpeer; /* peer I received from */ 212 213 void *msg_private; 214 struct lnet_libmd *msg_md; 215 216 unsigned int msg_len; 217 unsigned int msg_wanted; 218 unsigned int msg_offset; 219 unsigned int msg_niov; 220 struct iovec *msg_iov; 221 lnet_kiov_t *msg_kiov; 222 223 lnet_event_t msg_ev; 224 lnet_hdr_t msg_hdr; 225 } lnet_msg_t; 226 227 typedef struct lnet_libhandle { 228 struct list_head lh_hash_chain; 229 __u64 lh_cookie; 230 } lnet_libhandle_t; 231 232 #define lh_entry(ptr, type, member) \ 233 ((type *)((char *)(ptr)-(char *)(&((type *)0)->member))) 234 235 typedef struct lnet_eq { 236 struct list_head eq_list; 237 lnet_libhandle_t eq_lh; 238 lnet_seq_t eq_enq_seq; 239 lnet_seq_t eq_deq_seq; 240 unsigned int eq_size; 241 lnet_eq_handler_t eq_callback; 242 lnet_event_t *eq_events; 243 int **eq_refs; /* percpt refcount for EQ */ 244 } lnet_eq_t; 245 246 typedef struct lnet_me { 247 struct list_head me_list; 248 lnet_libhandle_t me_lh; 249 lnet_process_id_t me_match_id; 250 unsigned int me_portal; 251 unsigned int me_pos; /* hash offset in mt_hash */ 252 __u64 me_match_bits; 253 __u64 me_ignore_bits; 254 lnet_unlink_t me_unlink; 255 struct lnet_libmd *me_md; 256 } lnet_me_t; 257 258 typedef struct lnet_libmd { 259 struct list_head md_list; 260 lnet_libhandle_t md_lh; 261 lnet_me_t *md_me; 262 char *md_start; 263 unsigned int md_offset; 264 unsigned int md_length; 265 unsigned int md_max_size; 266 int md_threshold; 267 int md_refcount; 268 unsigned int md_options; 269 unsigned int md_flags; 270 void *md_user_ptr; 271 lnet_eq_t *md_eq; 272 unsigned int md_niov; /* # frags */ 273 union { 274 struct iovec iov[LNET_MAX_IOV]; 275 lnet_kiov_t kiov[LNET_MAX_IOV]; 276 } md_iov; 277 } lnet_libmd_t; 278 279 #define LNET_MD_FLAG_ZOMBIE (1 << 0) 280 #define LNET_MD_FLAG_AUTO_UNLINK (1 << 1) 281 #define LNET_MD_FLAG_ABORTED (1 << 2) 282 283 #ifdef LNET_USE_LIB_FREELIST 284 typedef struct { 285 void *fl_objs; /* single contiguous array of objects */ 286 int fl_nobjs; /* the number of them */ 287 int fl_objsize; /* the size (including overhead) of each of them */ 288 struct list_head fl_list; /* where they are enqueued */ 289 } lnet_freelist_t; 290 291 typedef struct { 292 struct list_head fo_list; /* enqueue on fl_list */ 293 void *fo_contents; /* aligned contents */ 294 } lnet_freeobj_t; 295 #endif 296 297 typedef struct { 298 /* info about peers we are trying to fail */ 299 struct list_head tp_list; /* ln_test_peers */ 300 lnet_nid_t tp_nid; /* matching nid */ 301 unsigned int tp_threshold; /* # failures to simulate */ 302 } lnet_test_peer_t; 303 304 #define LNET_COOKIE_TYPE_MD 1 305 #define LNET_COOKIE_TYPE_ME 2 306 #define LNET_COOKIE_TYPE_EQ 3 307 #define LNET_COOKIE_TYPE_BITS 2 308 #define LNET_COOKIE_MASK ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL) 309 310 struct lnet_ni; /* forward ref */ 311 312 typedef struct lnet_lnd { 313 /* fields managed by portals */ 314 struct list_head lnd_list; /* stash in the LND table */ 315 int lnd_refcount; /* # active instances */ 316 317 /* fields initialised by the LND */ 318 unsigned int lnd_type; 319 320 int (*lnd_startup)(struct lnet_ni *ni); 321 void (*lnd_shutdown)(struct lnet_ni *ni); 322 int (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg); 323 324 /* In data movement APIs below, payload buffers are described as a set 325 * of 'niov' fragments which are... 326 * EITHER 327 * in virtual memory (struct iovec *iov != NULL) 328 * OR 329 * in pages (kernel only: plt_kiov_t *kiov != NULL). 330 * The LND may NOT overwrite these fragment descriptors. 331 * An 'offset' and may specify a byte offset within the set of 332 * fragments to start from 333 */ 334 335 /* Start sending a preformatted message. 'private' is NULL for PUT and 336 * GET messages; otherwise this is a response to an incoming message 337 * and 'private' is the 'private' passed to lnet_parse(). Return 338 * non-zero for immediate failure, otherwise complete later with 339 * lnet_finalize() */ 340 int (*lnd_send)(struct lnet_ni *ni, void *private, lnet_msg_t *msg); 341 342 /* Start receiving 'mlen' bytes of payload data, skipping the following 343 * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to 344 * lnet_parse(). Return non-zero for immediate failure, otherwise 345 * complete later with lnet_finalize(). This also gives back a receive 346 * credit if the LND does flow control. */ 347 int (*lnd_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg, 348 int delayed, unsigned int niov, 349 struct iovec *iov, lnet_kiov_t *kiov, 350 unsigned int offset, unsigned int mlen, unsigned int rlen); 351 352 /* lnet_parse() has had to delay processing of this message 353 * (e.g. waiting for a forwarding buffer or send credits). Give the 354 * LND a chance to free urgently needed resources. If called, return 0 355 * for success and do NOT give back a receive credit; that has to wait 356 * until lnd_recv() gets called. On failure return < 0 and 357 * release resources; lnd_recv() will not be called. */ 358 int (*lnd_eager_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg, 359 void **new_privatep); 360 361 /* notification of peer health */ 362 void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive); 363 364 /* query of peer aliveness */ 365 void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, unsigned long *when); 366 367 /* accept a new connection */ 368 int (*lnd_accept)(struct lnet_ni *ni, struct socket *sock); 369 370 } lnd_t; 371 372 #define LNET_NI_STATUS_UP 0x15aac0de 373 #define LNET_NI_STATUS_DOWN 0xdeadface 374 #define LNET_NI_STATUS_INVALID 0x00000000 375 typedef struct { 376 lnet_nid_t ns_nid; 377 __u32 ns_status; 378 __u32 ns_unused; 379 } WIRE_ATTR lnet_ni_status_t; 380 381 struct lnet_tx_queue { 382 int tq_credits; /* # tx credits free */ 383 int tq_credits_min; /* lowest it's been */ 384 int tq_credits_max; /* total # tx credits */ 385 struct list_head tq_delayed; /* delayed TXs */ 386 }; 387 388 #define LNET_MAX_INTERFACES 16 389 390 typedef struct lnet_ni { 391 spinlock_t ni_lock; 392 struct list_head ni_list; /* chain on ln_nis */ 393 struct list_head ni_cptlist; /* chain on ln_nis_cpt */ 394 int ni_maxtxcredits; /* # tx credits */ 395 /* # per-peer send credits */ 396 int ni_peertxcredits; 397 /* # per-peer router buffer credits */ 398 int ni_peerrtrcredits; 399 /* seconds to consider peer dead */ 400 int ni_peertimeout; 401 int ni_ncpts; /* number of CPTs */ 402 __u32 *ni_cpts; /* bond NI on some CPTs */ 403 lnet_nid_t ni_nid; /* interface's NID */ 404 void *ni_data; /* instance-specific data */ 405 lnd_t *ni_lnd; /* procedural interface */ 406 struct lnet_tx_queue **ni_tx_queues; /* percpt TX queues */ 407 int **ni_refs; /* percpt reference count */ 408 long ni_last_alive; /* when I was last alive */ 409 lnet_ni_status_t *ni_status; /* my health status */ 410 /* equivalent interfaces to use */ 411 char *ni_interfaces[LNET_MAX_INTERFACES]; 412 } lnet_ni_t; 413 414 #define LNET_PROTO_PING_MATCHBITS 0x8000000000000000LL 415 416 /* NB: value of these features equal to LNET_PROTO_PING_VERSION_x 417 * of old LNet, so there shouldn't be any compatibility issue */ 418 #define LNET_PING_FEAT_INVAL (0) /* no feature */ 419 #define LNET_PING_FEAT_BASE (1 << 0) /* just a ping */ 420 #define LNET_PING_FEAT_NI_STATUS (1 << 1) /* return NI status */ 421 422 #define LNET_PING_FEAT_MASK (LNET_PING_FEAT_BASE | \ 423 LNET_PING_FEAT_NI_STATUS) 424 425 typedef struct { 426 __u32 pi_magic; 427 __u32 pi_features; 428 lnet_pid_t pi_pid; 429 __u32 pi_nnis; 430 lnet_ni_status_t pi_ni[0]; 431 } WIRE_ATTR lnet_ping_info_t; 432 433 /* router checker data, per router */ 434 #define LNET_MAX_RTR_NIS 16 435 #define LNET_PINGINFO_SIZE offsetof(lnet_ping_info_t, pi_ni[LNET_MAX_RTR_NIS]) 436 typedef struct { 437 /* chain on the_lnet.ln_zombie_rcd or ln_deathrow_rcd */ 438 struct list_head rcd_list; 439 lnet_handle_md_t rcd_mdh; /* ping buffer MD */ 440 struct lnet_peer *rcd_gateway; /* reference to gateway */ 441 lnet_ping_info_t *rcd_pinginfo; /* ping buffer */ 442 } lnet_rc_data_t; 443 444 typedef struct lnet_peer { 445 struct list_head lp_hashlist; /* chain on peer hash */ 446 struct list_head lp_txq; /* messages blocking for tx credits */ 447 struct list_head lp_rtrq; /* messages blocking for router credits */ 448 struct list_head lp_rtr_list; /* chain on router list */ 449 int lp_txcredits; /* # tx credits available */ 450 int lp_mintxcredits; /* low water mark */ 451 int lp_rtrcredits; /* # router credits */ 452 int lp_minrtrcredits; /* low water mark */ 453 unsigned int lp_alive:1; /* alive/dead? */ 454 unsigned int lp_notify:1; /* notification outstanding? */ 455 unsigned int lp_notifylnd:1; /* outstanding notification for LND? */ 456 unsigned int lp_notifying:1; /* some thread is handling notification */ 457 unsigned int lp_ping_notsent; /* SEND event outstanding from ping */ 458 int lp_alive_count; /* # times router went dead<->alive */ 459 long lp_txqnob; /* bytes queued for sending */ 460 unsigned long lp_timestamp; /* time of last aliveness news */ 461 unsigned long lp_ping_timestamp; /* time of last ping attempt */ 462 unsigned long lp_ping_deadline; /* != 0 if ping reply expected */ 463 unsigned long lp_last_alive; /* when I was last alive */ 464 unsigned long lp_last_query; /* when lp_ni was queried last time */ 465 lnet_ni_t *lp_ni; /* interface peer is on */ 466 lnet_nid_t lp_nid; /* peer's NID */ 467 int lp_refcount; /* # refs */ 468 int lp_cpt; /* CPT this peer attached on */ 469 /* # refs from lnet_route_t::lr_gateway */ 470 int lp_rtr_refcount; 471 /* returned RC ping features */ 472 unsigned int lp_ping_feats; 473 struct list_head lp_routes; /* routers on this peer */ 474 lnet_rc_data_t *lp_rcd; /* router checker state */ 475 } lnet_peer_t; 476 477 /* peer hash size */ 478 #define LNET_PEER_HASH_BITS 9 479 #define LNET_PEER_HASH_SIZE (1 << LNET_PEER_HASH_BITS) 480 481 /* peer hash table */ 482 struct lnet_peer_table { 483 int pt_version; /* /proc validity stamp */ 484 int pt_number; /* # peers extant */ 485 struct list_head pt_deathrow; /* zombie peers */ 486 struct list_head *pt_hash; /* NID->peer hash */ 487 }; 488 489 /* peer aliveness is enabled only on routers for peers in a network where the 490 * lnet_ni_t::ni_peertimeout has been set to a positive value */ 491 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \ 492 (lp)->lp_ni->ni_peertimeout > 0) 493 494 typedef struct { 495 struct list_head lr_list; /* chain on net */ 496 struct list_head lr_gwlist; /* chain on gateway */ 497 lnet_peer_t *lr_gateway; /* router node */ 498 __u32 lr_net; /* remote network number */ 499 int lr_seq; /* sequence for round-robin */ 500 unsigned int lr_downis; /* number of down NIs */ 501 unsigned int lr_hops; /* how far I am */ 502 unsigned int lr_priority; /* route priority */ 503 } lnet_route_t; 504 505 #define LNET_REMOTE_NETS_HASH_DEFAULT (1U << 7) 506 #define LNET_REMOTE_NETS_HASH_MAX (1U << 16) 507 #define LNET_REMOTE_NETS_HASH_SIZE (1 << the_lnet.ln_remote_nets_hbits) 508 509 typedef struct { 510 struct list_head lrn_list; /* chain on ln_remote_nets_hash */ 511 struct list_head lrn_routes; /* routes to me */ 512 __u32 lrn_net; /* my net number */ 513 } lnet_remotenet_t; 514 515 typedef struct { 516 struct list_head rbp_bufs; /* my free buffer pool */ 517 struct list_head rbp_msgs; /* messages blocking for a buffer */ 518 int rbp_npages; /* # pages in each buffer */ 519 int rbp_nbuffers; /* # buffers */ 520 int rbp_credits; /* # free buffers / blocked messages */ 521 int rbp_mincredits; /* low water mark */ 522 } lnet_rtrbufpool_t; 523 524 typedef struct { 525 struct list_head rb_list; /* chain on rbp_bufs */ 526 lnet_rtrbufpool_t *rb_pool; /* owning pool */ 527 lnet_kiov_t rb_kiov[0]; /* the buffer space */ 528 } lnet_rtrbuf_t; 529 530 typedef struct { 531 __u32 msgs_alloc; 532 __u32 msgs_max; 533 __u32 errors; 534 __u32 send_count; 535 __u32 recv_count; 536 __u32 route_count; 537 __u32 drop_count; 538 __u64 send_length; 539 __u64 recv_length; 540 __u64 route_length; 541 __u64 drop_length; 542 } WIRE_ATTR lnet_counters_t; 543 544 #define LNET_PEER_HASHSIZE 503 /* prime! */ 545 546 #define LNET_NRBPOOLS 3 /* # different router buffer pools */ 547 548 enum { 549 /* Didn't match anything */ 550 LNET_MATCHMD_NONE = (1 << 0), 551 /* Matched OK */ 552 LNET_MATCHMD_OK = (1 << 1), 553 /* Must be discarded */ 554 LNET_MATCHMD_DROP = (1 << 2), 555 /* match and buffer is exhausted */ 556 LNET_MATCHMD_EXHAUSTED = (1 << 3), 557 /* match or drop */ 558 LNET_MATCHMD_FINISH = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP), 559 }; 560 561 /* Options for lnet_portal_t::ptl_options */ 562 #define LNET_PTL_LAZY (1 << 0) 563 #define LNET_PTL_MATCH_UNIQUE (1 << 1) /* unique match, for RDMA */ 564 #define LNET_PTL_MATCH_WILDCARD (1 << 2) /* wildcard match, request portal */ 565 566 /* parameter for matching operations (GET, PUT) */ 567 struct lnet_match_info { 568 __u64 mi_mbits; 569 lnet_process_id_t mi_id; 570 unsigned int mi_opc; 571 unsigned int mi_portal; 572 unsigned int mi_rlength; 573 unsigned int mi_roffset; 574 }; 575 576 /* ME hash of RDMA portal */ 577 #define LNET_MT_HASH_BITS 8 578 #define LNET_MT_HASH_SIZE (1 << LNET_MT_HASH_BITS) 579 #define LNET_MT_HASH_MASK (LNET_MT_HASH_SIZE - 1) 580 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash, 581 * the last entry is reserved for MEs with ignore-bits */ 582 #define LNET_MT_HASH_IGNORE LNET_MT_HASH_SIZE 583 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which 584 * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the 585 * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */ 586 #define LNET_MT_BITS_U64 6 /* 2^6 bits */ 587 #define LNET_MT_EXHAUSTED_BITS (LNET_MT_HASH_BITS - LNET_MT_BITS_U64) 588 #define LNET_MT_EXHAUSTED_BMAP ((1 << LNET_MT_EXHAUSTED_BITS) + 1) 589 590 /* portal match table */ 591 struct lnet_match_table { 592 /* reserved for upcoming patches, CPU partition ID */ 593 unsigned int mt_cpt; 594 unsigned int mt_portal; /* portal index */ 595 /* match table is set as "enabled" if there's non-exhausted MD 596 * attached on mt_mhash, it's only valid for wildcard portal */ 597 unsigned int mt_enabled; 598 /* bitmap to flag whether MEs on mt_hash are exhausted or not */ 599 __u64 mt_exhausted[LNET_MT_EXHAUSTED_BMAP]; 600 struct list_head *mt_mhash; /* matching hash */ 601 }; 602 603 /* these are only useful for wildcard portal */ 604 /* Turn off message rotor for wildcard portals */ 605 #define LNET_PTL_ROTOR_OFF 0 606 /* round-robin dispatch all PUT messages for wildcard portals */ 607 #define LNET_PTL_ROTOR_ON 1 608 /* round-robin dispatch routed PUT message for wildcard portals */ 609 #define LNET_PTL_ROTOR_RR_RT 2 610 /* dispatch routed PUT message by hashing source NID for wildcard portals */ 611 #define LNET_PTL_ROTOR_HASH_RT 3 612 613 typedef struct lnet_portal { 614 spinlock_t ptl_lock; 615 unsigned int ptl_index; /* portal ID, reserved */ 616 /* flags on this portal: lazy, unique... */ 617 unsigned int ptl_options; 618 /* list of messages which are stealing buffer */ 619 struct list_head ptl_msg_stealing; 620 /* messages blocking for MD */ 621 struct list_head ptl_msg_delayed; 622 /* Match table for each CPT */ 623 struct lnet_match_table **ptl_mtables; 624 /* spread rotor of incoming "PUT" */ 625 int ptl_rotor; 626 /* # active entries for this portal */ 627 int ptl_mt_nmaps; 628 /* array of active entries' cpu-partition-id */ 629 int ptl_mt_maps[0]; 630 } lnet_portal_t; 631 632 #define LNET_LH_HASH_BITS 12 633 #define LNET_LH_HASH_SIZE (1ULL << LNET_LH_HASH_BITS) 634 #define LNET_LH_HASH_MASK (LNET_LH_HASH_SIZE - 1) 635 636 /* resource container (ME, MD, EQ) */ 637 struct lnet_res_container { 638 unsigned int rec_type; /* container type */ 639 __u64 rec_lh_cookie; /* cookie generator */ 640 struct list_head rec_active; /* active resource list */ 641 struct list_head *rec_lh_hash; /* handle hash */ 642 #ifdef LNET_USE_LIB_FREELIST 643 lnet_freelist_t rec_freelist; /* freelist for resources */ 644 #endif 645 }; 646 647 /* message container */ 648 struct lnet_msg_container { 649 int msc_init; /* initialized or not */ 650 /* max # threads finalizing */ 651 int msc_nfinalizers; 652 /* msgs waiting to complete finalizing */ 653 struct list_head msc_finalizing; 654 struct list_head msc_active; /* active message list */ 655 /* threads doing finalization */ 656 void **msc_finalizers; 657 #ifdef LNET_USE_LIB_FREELIST 658 lnet_freelist_t msc_freelist; /* freelist for messages */ 659 #endif 660 }; 661 662 /* Router Checker states */ 663 #define LNET_RC_STATE_SHUTDOWN 0 /* not started */ 664 #define LNET_RC_STATE_RUNNING 1 /* started up OK */ 665 #define LNET_RC_STATE_STOPPING 2 /* telling thread to stop */ 666 667 typedef struct { 668 /* CPU partition table of LNet */ 669 struct cfs_cpt_table *ln_cpt_table; 670 /* number of CPTs in ln_cpt_table */ 671 unsigned int ln_cpt_number; 672 unsigned int ln_cpt_bits; 673 674 /* protect LNet resources (ME/MD/EQ) */ 675 struct cfs_percpt_lock *ln_res_lock; 676 /* # portals */ 677 int ln_nportals; 678 /* the vector of portals */ 679 lnet_portal_t **ln_portals; 680 /* percpt ME containers */ 681 struct lnet_res_container **ln_me_containers; 682 /* percpt MD container */ 683 struct lnet_res_container **ln_md_containers; 684 685 /* Event Queue container */ 686 struct lnet_res_container ln_eq_container; 687 wait_queue_head_t ln_eq_waitq; 688 spinlock_t ln_eq_wait_lock; 689 unsigned int ln_remote_nets_hbits; 690 691 /* protect NI, peer table, credits, routers, rtrbuf... */ 692 struct cfs_percpt_lock *ln_net_lock; 693 /* percpt message containers for active/finalizing/freed message */ 694 struct lnet_msg_container **ln_msg_containers; 695 lnet_counters_t **ln_counters; 696 struct lnet_peer_table **ln_peer_tables; 697 /* failure simulation */ 698 struct list_head ln_test_peers; 699 700 struct list_head ln_nis; /* LND instances */ 701 /* NIs bond on specific CPT(s) */ 702 struct list_head ln_nis_cpt; 703 /* dying LND instances */ 704 struct list_head ln_nis_zombie; 705 lnet_ni_t *ln_loni; /* the loopback NI */ 706 /* NI to wait for events in */ 707 lnet_ni_t *ln_eq_waitni; 708 709 /* remote networks with routes to them */ 710 struct list_head *ln_remote_nets_hash; 711 /* validity stamp */ 712 __u64 ln_remote_nets_version; 713 /* list of all known routers */ 714 struct list_head ln_routers; 715 /* validity stamp */ 716 __u64 ln_routers_version; 717 /* percpt router buffer pools */ 718 lnet_rtrbufpool_t **ln_rtrpools; 719 720 lnet_handle_md_t ln_ping_target_md; 721 lnet_handle_eq_t ln_ping_target_eq; 722 lnet_ping_info_t *ln_ping_info; 723 724 /* router checker startup/shutdown state */ 725 int ln_rc_state; 726 /* router checker's event queue */ 727 lnet_handle_eq_t ln_rc_eqh; 728 /* rcd still pending on net */ 729 struct list_head ln_rcd_deathrow; 730 /* rcd ready for free */ 731 struct list_head ln_rcd_zombie; 732 /* serialise startup/shutdown */ 733 struct semaphore ln_rc_signal; 734 735 struct mutex ln_api_mutex; 736 struct mutex ln_lnd_mutex; 737 int ln_init; /* LNetInit() called? */ 738 /* Have I called LNetNIInit myself? */ 739 int ln_niinit_self; 740 /* LNetNIInit/LNetNIFini counter */ 741 int ln_refcount; 742 /* shutdown in progress */ 743 int ln_shutdown; 744 745 int ln_routing; /* am I a router? */ 746 lnet_pid_t ln_pid; /* requested pid */ 747 /* uniquely identifies this ni in this epoch */ 748 __u64 ln_interface_cookie; 749 /* registered LNDs */ 750 struct list_head ln_lnds; 751 752 /* space for network names */ 753 char *ln_network_tokens; 754 int ln_network_tokens_nob; 755 /* test protocol compatibility flags */ 756 int ln_testprotocompat; 757 758 } lnet_t; 759 760 #endif 761