drivers/staging/lustre/include/linux/lnet/lib-types.h

   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
 160 #define LNET_PROTO_TCP_VERSION_MAJOR    1
 161 #define LNET_PROTO_TCP_VERSION_MINOR    0
 162
 163 /* Acceptor connection request */
 164 typedef struct {
 165         __u32       acr_magic;            /* PTL_ACCEPTOR_PROTO_MAGIC */
 166         __u32       acr_version;                /* protocol version */
 167         __u64       acr_nid;                /* target NID */
 168 } WIRE_ATTR lnet_acceptor_connreq_t;
 169
 170 #define LNET_PROTO_ACCEPTOR_VERSION       1
 171
 172 /* forward refs */
 173 struct lnet_libmd;
 174
 175 typedef struct lnet_msg {
 176         struct list_head            msg_activelist;
 177         struct list_head            msg_list;      /* Q for credits/MD */
 178
 179         lnet_process_id_t     msg_target;
 180         /* where is it from, it's only for building event */
 181         lnet_nid_t              msg_from;
 182         __u32                   msg_type;
 183
 184         /* committed for sending */
 185         unsigned int            msg_tx_committed:1;
 186         /* CPT # this message committed for sending */
 187         unsigned int            msg_tx_cpt:15;
 188         /* committed for receiving */
 189         unsigned int            msg_rx_committed:1;
 190         /* CPT # this message committed for receiving */
 191         unsigned int            msg_rx_cpt:15;
 192         /* queued for tx credit */
 193         unsigned int            msg_tx_delayed:1;
 194         /* queued for RX buffer */
 195         unsigned int            msg_rx_delayed:1;
 196         /* ready for pending on RX delay list */
 197         unsigned int            msg_rx_ready_delay:1;
 198
 199         unsigned int      msg_vmflush:1;      /* VM trying to free memory */
 200         unsigned int      msg_target_is_router:1; /* sending to a router */
 201         unsigned int      msg_routing:1;      /* being forwarded */
 202         unsigned int      msg_ack:1;      /* ack on finalize (PUT) */
 203         unsigned int      msg_sending:1;      /* outgoing message */
 204         unsigned int      msg_receiving:1;    /* being received */
 205         unsigned int      msg_txcredit:1;     /* taken an NI send credit */
 206         unsigned int      msg_peertxcredit:1; /* taken a peer send credit */
 207         unsigned int      msg_rtrcredit:1;    /* taken a global router credit */
 208         unsigned int      msg_peerrtrcredit:1; /* taken a peer router credit */
 209         unsigned int      msg_onactivelist:1; /* on the activelist */
 210
 211         struct lnet_peer     *msg_txpeer;        /* peer I'm sending to */
 212         struct lnet_peer     *msg_rxpeer;        /* peer I received from */
 213
 214         void             *msg_private;
 215         struct lnet_libmd    *msg_md;
 216
 217         unsigned int      msg_len;
 218         unsigned int      msg_wanted;
 219         unsigned int      msg_offset;
 220         unsigned int      msg_niov;
 221         struct iovec     *msg_iov;
 222         lnet_kiov_t       *msg_kiov;
 223
 224         lnet_event_t      msg_ev;
 225         lnet_hdr_t          msg_hdr;
 226 } lnet_msg_t;
 227
 228
 229 typedef struct lnet_libhandle {
 230         struct list_head            lh_hash_chain;
 231         __u64            lh_cookie;
 232 } lnet_libhandle_t;
 233
 234 #define lh_entry(ptr, type, member) \
 235         ((type *)((char *)(ptr)-(char *)(&((type *)0)->member)))
 236
 237 typedef struct lnet_eq {
 238         struct list_head                eq_list;
 239         lnet_libhandle_t        eq_lh;
 240         lnet_seq_t              eq_enq_seq;
 241         lnet_seq_t              eq_deq_seq;
 242         unsigned int            eq_size;
 243         lnet_eq_handler_t       eq_callback;
 244         lnet_event_t            *eq_events;
 245         int                     **eq_refs;      /* percpt refcount for EQ */
 246 } lnet_eq_t;
 247
 248 typedef struct lnet_me {
 249         struct list_head             me_list;
 250         lnet_libhandle_t       me_lh;
 251         lnet_process_id_t      me_match_id;
 252         unsigned int       me_portal;
 253         unsigned int       me_pos;              /* hash offset in mt_hash */
 254         __u64             me_match_bits;
 255         __u64             me_ignore_bits;
 256         lnet_unlink_t     me_unlink;
 257         struct lnet_libmd     *me_md;
 258 } lnet_me_t;
 259
 260 typedef struct lnet_libmd {
 261         struct list_head            md_list;
 262         lnet_libhandle_t      md_lh;
 263         lnet_me_t           *md_me;
 264         char             *md_start;
 265         unsigned int      md_offset;
 266         unsigned int      md_length;
 267         unsigned int      md_max_size;
 268         int                md_threshold;
 269         int                md_refcount;
 270         unsigned int      md_options;
 271         unsigned int      md_flags;
 272         void             *md_user_ptr;
 273         lnet_eq_t           *md_eq;
 274         unsigned int      md_niov;              /* # frags */
 275         union {
 276                 struct iovec  iov[LNET_MAX_IOV];
 277                 lnet_kiov_t   kiov[LNET_MAX_IOV];
 278         } md_iov;
 279 } lnet_libmd_t;
 280
 281 #define LNET_MD_FLAG_ZOMBIE        (1 << 0)
 282 #define LNET_MD_FLAG_AUTO_UNLINK      (1 << 1)
 283 #define LNET_MD_FLAG_ABORTED     (1 << 2)
 284
 285 #ifdef LNET_USE_LIB_FREELIST
 286 typedef struct {
 287         void              *fl_objs;       /* single contiguous array of objects */
 288         int                 fl_nobjs;    /* the number of them */
 289         int                 fl_objsize;       /* the size (including overhead) of each of them */
 290         struct list_head             fl_list;     /* where they are enqueued */
 291 } lnet_freelist_t;
 292
 293 typedef struct {
 294         struct list_head             fo_list;        /* enqueue on fl_list */
 295         void              *fo_contents;  /* aligned contents */
 296 } lnet_freeobj_t;
 297 #endif
 298
 299 typedef struct {
 300         /* info about peers we are trying to fail */
 301         struct list_head             tp_list;        /* ln_test_peers */
 302         lnet_nid_t           tp_nid;          /* matching nid */
 303         unsigned int       tp_threshold;        /* # failures to simulate */
 304 } lnet_test_peer_t;
 305
 306 #define LNET_COOKIE_TYPE_MD    1
 307 #define LNET_COOKIE_TYPE_ME    2
 308 #define LNET_COOKIE_TYPE_EQ    3
 309 #define LNET_COOKIE_TYPE_BITS  2
 310 #define LNET_COOKIE_MASK        ((1ULL << LNET_COOKIE_TYPE_BITS) - 1ULL)
 311
 312 struct lnet_ni;                           /* forward ref */
 313
 314 typedef struct lnet_lnd {
 315         /* fields managed by portals */
 316         struct list_head            lnd_list;        /* stash in the LND table */
 317         int                lnd_refcount;         /* # active instances */
 318
 319         /* fields initialised by the LND */
 320         unsigned int      lnd_type;
 321
 322         int  (*lnd_startup)(struct lnet_ni *ni);
 323         void (*lnd_shutdown)(struct lnet_ni *ni);
 324         int  (*lnd_ctl)(struct lnet_ni *ni, unsigned int cmd, void *arg);
 325
 326         /* In data movement APIs below, payload buffers are described as a set
 327          * of 'niov' fragments which are...
 328          * EITHER
 329          *    in virtual memory (struct iovec *iov != NULL)
 330          * OR
 331          *    in pages (kernel only: plt_kiov_t *kiov != NULL).
 332          * The LND may NOT overwrite these fragment descriptors.
 333          * An 'offset' and may specify a byte offset within the set of
 334          * fragments to start from
 335          */
 336
 337         /* Start sending a preformatted message.  'private' is NULL for PUT and
 338          * GET messages; otherwise this is a response to an incoming message
 339          * and 'private' is the 'private' passed to lnet_parse().  Return
 340          * non-zero for immediate failure, otherwise complete later with
 341          * lnet_finalize() */
 342         int (*lnd_send)(struct lnet_ni *ni, void *private, lnet_msg_t *msg);
 343
 344         /* Start receiving 'mlen' bytes of payload data, skipping the following
 345          * 'rlen' - 'mlen' bytes. 'private' is the 'private' passed to
 346          * lnet_parse().  Return non-zero for immediate failure, otherwise
 347          * complete later with lnet_finalize().  This also gives back a receive
 348          * credit if the LND does flow control. */
 349         int (*lnd_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg,
 350                         int delayed, unsigned int niov,
 351                         struct iovec *iov, lnet_kiov_t *kiov,
 352                         unsigned int offset, unsigned int mlen, unsigned int rlen);
 353
 354         /* lnet_parse() has had to delay processing of this message
 355          * (e.g. waiting for a forwarding buffer or send credits).  Give the
 356          * LND a chance to free urgently needed resources.  If called, return 0
 357          * for success and do NOT give back a receive credit; that has to wait
 358          * until lnd_recv() gets called.  On failure return < 0 and
 359          * release resources; lnd_recv() will not be called. */
 360         int (*lnd_eager_recv)(struct lnet_ni *ni, void *private, lnet_msg_t *msg,
 361                               void **new_privatep);
 362
 363         /* notification of peer health */
 364         void (*lnd_notify)(struct lnet_ni *ni, lnet_nid_t peer, int alive);
 365
 366         /* query of peer aliveness */
 367         void (*lnd_query)(struct lnet_ni *ni, lnet_nid_t peer, cfs_time_t *when);
 368
 369         /* accept a new connection */
 370         int (*lnd_accept)(struct lnet_ni *ni, socket_t *sock);
 371
 372 } lnd_t;
 373
 374 #define LNET_NI_STATUS_UP      0x15aac0de
 375 #define LNET_NI_STATUS_DOWN    0xdeadface
 376 #define LNET_NI_STATUS_INVALID 0x00000000
 377 typedef struct {
 378         lnet_nid_t ns_nid;
 379         __u32      ns_status;
 380         __u32      ns_unused;
 381 } WIRE_ATTR lnet_ni_status_t;
 382
 383 struct lnet_tx_queue {
 384         int                     tq_credits;     /* # tx credits free */
 385         int                     tq_credits_min; /* lowest it's been */
 386         int                     tq_credits_max; /* total # tx credits */
 387         struct list_head                tq_delayed;     /* delayed TXs */
 388 };
 389
 390 #define LNET_MAX_INTERFACES   16
 391
 392 typedef struct lnet_ni {
 393         spinlock_t              ni_lock;
 394         struct list_head                ni_list;        /* chain on ln_nis */
 395         struct list_head                ni_cptlist;     /* chain on ln_nis_cpt */
 396         int                     ni_maxtxcredits; /* # tx credits  */
 397         /* # per-peer send credits */
 398         int                     ni_peertxcredits;
 399         /* # per-peer router buffer credits */
 400         int                     ni_peerrtrcredits;
 401         /* seconds to consider peer dead */
 402         int                     ni_peertimeout;
 403         int                     ni_ncpts;       /* number of CPTs */
 404         __u32                   *ni_cpts;       /* bond NI on some CPTs */
 405         lnet_nid_t              ni_nid;         /* interface's NID */
 406         void                    *ni_data;       /* instance-specific data */
 407         lnd_t                   *ni_lnd;        /* procedural interface */
 408         struct lnet_tx_queue    **ni_tx_queues; /* percpt TX queues */
 409         int                     **ni_refs;      /* percpt reference count */
 410         long                    ni_last_alive;  /* when I was last alive */
 411         lnet_ni_status_t        *ni_status;     /* my health status */
 412         /* equivalent interfaces to use */
 413         char                    *ni_interfaces[LNET_MAX_INTERFACES];
 414 } lnet_ni_t;
 415
 416 #define LNET_PROTO_PING_MATCHBITS       0x8000000000000000LL
 417
 418 /* NB: value of these features equal to LNET_PROTO_PING_VERSION_x
 419  * of old LNet, so there shouldn't be any compatibility issue */
 420 #define LNET_PING_FEAT_INVAL            (0)             /* no feature */
 421 #define LNET_PING_FEAT_BASE             (1 << 0)        /* just a ping */
 422 #define LNET_PING_FEAT_NI_STATUS        (1 << 1)        /* return NI status */
 423
 424 #define LNET_PING_FEAT_MASK             (LNET_PING_FEAT_BASE | \
 425                                          LNET_PING_FEAT_NI_STATUS)
 426
 427 typedef struct {
 428         __u32                   pi_magic;
 429         __u32                   pi_features;
 430         lnet_pid_t              pi_pid;
 431         __u32                   pi_nnis;
 432         lnet_ni_status_t        pi_ni[0];
 433 } WIRE_ATTR lnet_ping_info_t;
 434
 435 /* router checker data, per router */
 436 #define LNET_MAX_RTR_NIS   16
 437 #define LNET_PINGINFO_SIZE offsetof(lnet_ping_info_t, pi_ni[LNET_MAX_RTR_NIS])
 438 typedef struct {
 439         /* chain on the_lnet.ln_zombie_rcd or ln_deathrow_rcd */
 440         struct list_head                rcd_list;
 441         lnet_handle_md_t        rcd_mdh;        /* ping buffer MD */
 442         struct lnet_peer        *rcd_gateway;   /* reference to gateway */
 443         lnet_ping_info_t        *rcd_pinginfo;  /* ping buffer */
 444 } lnet_rc_data_t;
 445
 446 typedef struct lnet_peer {
 447         struct list_head        lp_hashlist;      /* chain on peer hash */
 448         struct list_head        lp_txq;        /* messages blocking for tx credits */
 449         struct list_head        lp_rtrq;              /* messages blocking for router credits */
 450         struct list_head        lp_rtr_list;      /* chain on router list */
 451         int            lp_txcredits;     /* # tx credits available */
 452         int            lp_mintxcredits;      /* low water mark */
 453         int            lp_rtrcredits;   /* # router credits */
 454         int            lp_minrtrcredits;     /* low water mark */
 455         unsigned int      lp_alive:1;      /* alive/dead? */
 456         unsigned int      lp_notify:1;    /* notification outstanding? */
 457         unsigned int      lp_notifylnd:1;       /* outstanding notification for LND? */
 458         unsigned int      lp_notifying:1;       /* some thread is handling notification */
 459         unsigned int      lp_ping_notsent;      /* SEND event outstanding from ping */
 460         int            lp_alive_count;       /* # times router went dead<->alive */
 461         long          lp_txqnob;            /* bytes queued for sending */
 462         cfs_time_t      lp_timestamp;    /* time of last aliveness news */
 463         cfs_time_t      lp_ping_timestamp;    /* time of last ping attempt */
 464         cfs_time_t      lp_ping_deadline;     /* != 0 if ping reply expected */
 465         cfs_time_t      lp_last_alive;  /* when I was last alive */
 466         cfs_time_t      lp_last_query;  /* when lp_ni was queried last time */
 467         lnet_ni_t       *lp_ni;         /* interface peer is on */
 468         lnet_nid_t      lp_nid;        /* peer's NID */
 469         int            lp_refcount;       /* # refs */
 470         int                     lp_cpt;         /* CPT this peer attached on */
 471         /* # refs from lnet_route_t::lr_gateway */
 472         int                     lp_rtr_refcount;
 473         /* returned RC ping features */
 474         unsigned int            lp_ping_feats;
 475         struct list_head                lp_routes;      /* routers on this peer */
 476         lnet_rc_data_t          *lp_rcd;        /* router checker state */
 477 } lnet_peer_t;
 478
 479 /* peer hash size */
 480 #define LNET_PEER_HASH_BITS     9
 481 #define LNET_PEER_HASH_SIZE     (1 << LNET_PEER_HASH_BITS)
 482
 483 /* peer hash table */
 484 struct lnet_peer_table {
 485         int                     pt_version;     /* /proc validity stamp */
 486         int                     pt_number;      /* # peers extant */
 487         struct list_head                pt_deathrow;    /* zombie peers */
 488         struct list_head                *pt_hash;       /* NID->peer hash */
 489 };
 490
 491 /* peer aliveness is enabled only on routers for peers in a network where the
 492  * lnet_ni_t::ni_peertimeout has been set to a positive value */
 493 #define lnet_peer_aliveness_enabled(lp) (the_lnet.ln_routing != 0 && \
 494                                          (lp)->lp_ni->ni_peertimeout > 0)
 495
 496 typedef struct {
 497         struct list_head                lr_list;        /* chain on net */
 498         struct list_head                lr_gwlist;      /* chain on gateway */
 499         lnet_peer_t             *lr_gateway;    /* router node */
 500         __u32                   lr_net;         /* remote network number */
 501         int                     lr_seq;         /* sequence for round-robin */
 502         unsigned int            lr_downis;      /* number of down NIs */
 503         unsigned int            lr_hops;        /* how far I am */
 504         unsigned int            lr_priority;    /* route priority */
 505 } lnet_route_t;
 506
 507 #define LNET_REMOTE_NETS_HASH_DEFAULT   (1U << 7)
 508 #define LNET_REMOTE_NETS_HASH_MAX       (1U << 16)
 509 #define LNET_REMOTE_NETS_HASH_SIZE      (1 << the_lnet.ln_remote_nets_hbits)
 510
 511 typedef struct {
 512         struct list_head              lrn_list;       /* chain on ln_remote_nets_hash */
 513         struct list_head              lrn_routes;     /* routes to me */
 514         __u32              lrn_net;     /* my net number */
 515 } lnet_remotenet_t;
 516
 517 typedef struct {
 518         struct list_head rbp_bufs;           /* my free buffer pool */
 519         struct list_head rbp_msgs;           /* messages blocking for a buffer */
 520         int     rbp_npages;        /* # pages in each buffer */
 521         int     rbp_nbuffers;    /* # buffers */
 522         int     rbp_credits;      /* # free buffers / blocked messages */
 523         int     rbp_mincredits;       /* low water mark */
 524 } lnet_rtrbufpool_t;
 525
 526 typedef struct {
 527         struct list_head             rb_list;        /* chain on rbp_bufs */
 528         lnet_rtrbufpool_t     *rb_pool;      /* owning pool */
 529         lnet_kiov_t         rb_kiov[0];   /* the buffer space */
 530 } lnet_rtrbuf_t;
 531
 532 typedef struct {
 533         __u32   msgs_alloc;
 534         __u32   msgs_max;
 535         __u32   errors;
 536         __u32   send_count;
 537         __u32   recv_count;
 538         __u32   route_count;
 539         __u32   drop_count;
 540         __u64   send_length;
 541         __u64   recv_length;
 542         __u64   route_length;
 543         __u64   drop_length;
 544 } WIRE_ATTR lnet_counters_t;
 545
 546 #define LNET_PEER_HASHSIZE   503                /* prime! */
 547
 548 #define LNET_NRBPOOLS    3               /* # different router buffer pools */
 549
 550 enum {
 551         /* Didn't match anything */
 552         LNET_MATCHMD_NONE       = (1 << 0),
 553         /* Matched OK */
 554         LNET_MATCHMD_OK         = (1 << 1),
 555         /* Must be discarded */
 556         LNET_MATCHMD_DROP       = (1 << 2),
 557         /* match and buffer is exhausted */
 558         LNET_MATCHMD_EXHAUSTED  = (1 << 3),
 559         /* match or drop */
 560         LNET_MATCHMD_FINISH     = (LNET_MATCHMD_OK | LNET_MATCHMD_DROP),
 561 };
 562
 563 /* Options for lnet_portal_t::ptl_options */
 564 #define LNET_PTL_LAZY          (1 << 0)
 565 #define LNET_PTL_MATCH_UNIQUE       (1 << 1)    /* unique match, for RDMA */
 566 #define LNET_PTL_MATCH_WILDCARD     (1 << 2)    /* wildcard match, request portal */
 567
 568 /* parameter for matching operations (GET, PUT) */
 569 struct lnet_match_info {
 570         __u64                   mi_mbits;
 571         lnet_process_id_t       mi_id;
 572         unsigned int            mi_opc;
 573         unsigned int            mi_portal;
 574         unsigned int            mi_rlength;
 575         unsigned int            mi_roffset;
 576 };
 577
 578 /* ME hash of RDMA portal */
 579 #define LNET_MT_HASH_BITS               8
 580 #define LNET_MT_HASH_SIZE               (1 << LNET_MT_HASH_BITS)
 581 #define LNET_MT_HASH_MASK               (LNET_MT_HASH_SIZE - 1)
 582 /* we allocate (LNET_MT_HASH_SIZE + 1) entries for lnet_match_table::mt_hash,
 583  * the last entry is reserved for MEs with ignore-bits */
 584 #define LNET_MT_HASH_IGNORE             LNET_MT_HASH_SIZE
 585 /* __u64 has 2^6 bits, so need 2^(LNET_MT_HASH_BITS - LNET_MT_BITS_U64) which
 586  * is 4 __u64s as bit-map, and add an extra __u64 (only use one bit) for the
 587  * ME-list with ignore-bits, which is mtable::mt_hash[LNET_MT_HASH_IGNORE] */
 588 #define LNET_MT_BITS_U64                6       /* 2^6 bits */
 589 #define LNET_MT_EXHAUSTED_BITS          (LNET_MT_HASH_BITS - LNET_MT_BITS_U64)
 590 #define LNET_MT_EXHAUSTED_BMAP          ((1 << LNET_MT_EXHAUSTED_BITS) + 1)
 591
 592 /* portal match table */
 593 struct lnet_match_table {
 594         /* reserved for upcoming patches, CPU partition ID */
 595         unsigned int            mt_cpt;
 596         unsigned int            mt_portal;      /* portal index */
 597         /* match table is set as "enabled" if there's non-exhausted MD
 598          * attached on mt_mhash, it's only valid for wildcard portal */
 599         unsigned int            mt_enabled;
 600         /* bitmap to flag whether MEs on mt_hash are exhausted or not */
 601         __u64                   mt_exhausted[LNET_MT_EXHAUSTED_BMAP];
 602         struct list_head                *mt_mhash;      /* matching hash */
 603 };
 604
 605 /* these are only useful for wildcard portal */
 606 /* Turn off message rotor for wildcard portals */
 607 #define LNET_PTL_ROTOR_OFF      0
 608 /* round-robin dispatch all PUT messages for wildcard portals */
 609 #define LNET_PTL_ROTOR_ON       1
 610 /* round-robin dispatch routed PUT message for wildcard portals */
 611 #define LNET_PTL_ROTOR_RR_RT    2
 612 /* dispatch routed PUT message by hashing source NID for wildcard portals */
 613 #define LNET_PTL_ROTOR_HASH_RT  3
 614
 615 typedef struct lnet_portal {
 616         spinlock_t              ptl_lock;
 617         unsigned int            ptl_index;      /* portal ID, reserved */
 618         /* flags on this portal: lazy, unique... */
 619         unsigned int            ptl_options;
 620         /* list of messages which are stealing buffer */
 621         struct list_head                ptl_msg_stealing;
 622         /* messages blocking for MD */
 623         struct list_head                ptl_msg_delayed;
 624         /* Match table for each CPT */
 625         struct lnet_match_table **ptl_mtables;
 626         /* spread rotor of incoming "PUT" */
 627         int                     ptl_rotor;
 628         /* # active entries for this portal */
 629         int                  ptl_mt_nmaps;
 630         /* array of active entries' cpu-partition-id */
 631         int                  ptl_mt_maps[0];
 632 } lnet_portal_t;
 633
 634 #define LNET_LH_HASH_BITS       12
 635 #define LNET_LH_HASH_SIZE       (1ULL << LNET_LH_HASH_BITS)
 636 #define LNET_LH_HASH_MASK       (LNET_LH_HASH_SIZE - 1)
 637
 638 /* resource container (ME, MD, EQ) */
 639 struct lnet_res_container {
 640         unsigned int            rec_type;       /* container type */
 641         __u64                   rec_lh_cookie;  /* cookie generator */
 642         struct list_head                rec_active;     /* active resource list */
 643         struct list_head                *rec_lh_hash;   /* handle hash */
 644 #ifdef LNET_USE_LIB_FREELIST
 645         lnet_freelist_t         rec_freelist;   /* freelist for resources */
 646 #endif
 647 };
 648
 649 /* message container */
 650 struct lnet_msg_container {
 651         int                     msc_init;       /* initialized or not */
 652         /* max # threads finalizing */
 653         int                     msc_nfinalizers;
 654         /* msgs waiting to complete finalizing */
 655         struct list_head                msc_finalizing;
 656         struct list_head                msc_active;     /* active message list */
 657         /* threads doing finalization */
 658         void                    **msc_finalizers;
 659 #ifdef LNET_USE_LIB_FREELIST
 660         lnet_freelist_t         msc_freelist;   /* freelist for messages */
 661 #endif
 662 };
 663
 664 /* Router Checker states */
 665 #define LNET_RC_STATE_SHUTDOWN          0       /* not started */
 666 #define LNET_RC_STATE_RUNNING           1       /* started up OK */
 667 #define LNET_RC_STATE_STOPPING          2       /* telling thread to stop */
 668
 669 typedef struct {
 670         /* CPU partition table of LNet */
 671         struct cfs_cpt_table            *ln_cpt_table;
 672         /* number of CPTs in ln_cpt_table */
 673         unsigned int                    ln_cpt_number;
 674         unsigned int                    ln_cpt_bits;
 675
 676         /* protect LNet resources (ME/MD/EQ) */
 677         struct cfs_percpt_lock          *ln_res_lock;
 678         /* # portals */
 679         int                             ln_nportals;
 680         /* the vector of portals */
 681         lnet_portal_t                   **ln_portals;
 682         /* percpt ME containers */
 683         struct lnet_res_container       **ln_me_containers;
 684         /* percpt MD container */
 685         struct lnet_res_container       **ln_md_containers;
 686
 687         /* Event Queue container */
 688         struct lnet_res_container       ln_eq_container;
 689         wait_queue_head_t                       ln_eq_waitq;
 690         spinlock_t                      ln_eq_wait_lock;
 691         unsigned int                    ln_remote_nets_hbits;
 692
 693         /* protect NI, peer table, credits, routers, rtrbuf... */
 694         struct cfs_percpt_lock          *ln_net_lock;
 695         /* percpt message containers for active/finalizing/freed message */
 696         struct lnet_msg_container       **ln_msg_containers;
 697         lnet_counters_t                 **ln_counters;
 698         struct lnet_peer_table          **ln_peer_tables;
 699         /* failure simulation */
 700         struct list_head                        ln_test_peers;
 701
 702         struct list_head                        ln_nis;         /* LND instances */
 703         /* NIs bond on specific CPT(s) */
 704         struct list_head                        ln_nis_cpt;
 705         /* dying LND instances */
 706         struct list_head                        ln_nis_zombie;
 707         lnet_ni_t                       *ln_loni;       /* the loopback NI */
 708         /* NI to wait for events in */
 709         lnet_ni_t                       *ln_eq_waitni;
 710
 711         /* remote networks with routes to them */
 712         struct list_head                        *ln_remote_nets_hash;
 713         /* validity stamp */
 714         __u64                           ln_remote_nets_version;
 715         /* list of all known routers */
 716         struct list_head                        ln_routers;
 717         /* validity stamp */
 718         __u64                           ln_routers_version;
 719         /* percpt router buffer pools */
 720         lnet_rtrbufpool_t               **ln_rtrpools;
 721
 722         lnet_handle_md_t                ln_ping_target_md;
 723         lnet_handle_eq_t                ln_ping_target_eq;
 724         lnet_ping_info_t                *ln_ping_info;
 725
 726         /* router checker startup/shutdown state */
 727         int                             ln_rc_state;
 728         /* router checker's event queue */
 729         lnet_handle_eq_t                ln_rc_eqh;
 730         /* rcd still pending on net */
 731         struct list_head                        ln_rcd_deathrow;
 732         /* rcd ready for free */
 733         struct list_head                        ln_rcd_zombie;
 734         /* serialise startup/shutdown */
 735         struct semaphore                ln_rc_signal;
 736
 737         struct mutex                    ln_api_mutex;
 738         struct mutex                    ln_lnd_mutex;
 739         int                             ln_init;        /* LNetInit() called? */
 740         /* Have I called LNetNIInit myself? */
 741         int                             ln_niinit_self;
 742         /* LNetNIInit/LNetNIFini counter */
 743         int                             ln_refcount;
 744         /* shutdown in progress */
 745         int                             ln_shutdown;
 746
 747         int                             ln_routing;     /* am I a router? */
 748         lnet_pid_t                      ln_pid;         /* requested pid */
 749         /* uniquely identifies this ni in this epoch */
 750         __u64                           ln_interface_cookie;
 751         /* registered LNDs */
 752         struct list_head                        ln_lnds;
 753
 754         /* space for network names */
 755         char                            *ln_network_tokens;
 756         int                             ln_network_tokens_nob;
 757         /* test protocol compatibility flags */
 758         int                             ln_testprotocompat;
 759
 760 } lnet_t;
 761
 762 #endif