1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lnet/klnds/ptllnd/ptllnd.h
38 * Author: PJ Kirner <pjkirner@clusterfs.com>
42 # define EXPORT_SYMTAB
44 #ifndef AUTOCONF_INCLUDED
45 #include <linux/config.h>
47 #include <linux/module.h>
48 #include <linux/kernel.h>
50 #include <linux/string.h>
51 #include <linux/stat.h>
52 #include <linux/errno.h>
53 #include <linux/smp_lock.h>
54 #include <linux/unistd.h>
55 #include <linux/uio.h>
57 #include <asm/system.h>
58 #include <asm/uaccess.h>
61 #include <linux/init.h>
63 #include <linux/file.h>
64 #include <linux/stat.h>
65 #include <linux/list.h>
66 #include <linux/kmod.h>
67 #include <linux/sysctl.h>
68 #include <linux/random.h>
74 #define DEBUG_SUBSYSTEM S_LND
76 #include <libcfs/libcfs.h>
77 #include <lnet/lnet.h>
78 #include <lnet/lib-lnet.h>
79 #include <lnet/lnet-sysctl.h>
80 #include <portals/p30.h>
82 #include <portals/ptltrace.h>
84 #include <lnet/ptllnd.h> /* Depends on portals/p30.h */
87 * Define this to enable console debug logging
90 //#define PJK_DEBUGGING
93 # define PTLLND_N_SCHED num_online_cpus() /* # schedulers */
95 # define PTLLND_N_SCHED 1 /* # schedulers */
98 #define PTLLND_CREDIT_HIGHWATER ((*kptllnd_tunables.kptl_peercredits)-1)
99 /* when eagerly to return credits */
103 int *kptl_ntx; /* # tx descs to pre-allocate */
104 int *kptl_max_nodes; /* max # nodes all talking to me */
105 int *kptl_max_procs_per_node; /* max # processes per node */
106 int *kptl_checksum; /* checksum kptl_msg_t? */
107 int *kptl_timeout; /* comms timeout (seconds) */
108 int *kptl_portal; /* portal number */
109 int *kptl_pid; /* portals PID (self + kernel peers) */
110 int *kptl_rxb_npages; /* number of pages for rx buffer */
111 int *kptl_rxb_nspare; /* number of spare rx buffers */
112 int *kptl_credits; /* number of credits */
113 int *kptl_peercredits; /* number of credits */
114 int *kptl_max_msg_size; /* max immd message size*/
115 int *kptl_peer_hash_table_size; /* # slots in peer hash table */
116 int *kptl_reschedule_loops; /* scheduler yield loops */
117 int *kptl_ack_puts; /* make portals ack PUTs */
119 int *kptl_ptltrace_on_timeout; /* dump pltrace on timeout? */
120 char **kptl_ptltrace_basename; /* ptltrace dump file basename */
123 int *kptl_simulation_bitmap;/* simulation bitmap */
126 #if defined(CONFIG_SYSCTL) && !CFS_SYSFS_MODULE_PARM
127 cfs_sysctl_table_header_t *kptl_sysctl; /* sysctl interface */
131 #include "lnet/ptllnd_wire.h"
133 /***********************************************************************/
135 typedef struct kptl_data kptl_data_t;
136 typedef struct kptl_rx_buffer kptl_rx_buffer_t;
137 typedef struct kptl_peer kptl_peer_t;
143 #define PTLLND_EVENTARG_TYPE_MSG 0x1
144 #define PTLLND_EVENTARG_TYPE_RDMA 0x2
145 #define PTLLND_EVENTARG_TYPE_BUF 0x3
147 typedef struct kptl_rx /* receive message */
149 struct list_head rx_list; /* queue for attention */
150 kptl_rx_buffer_t *rx_rxb; /* the rx buffer pointer */
151 kptl_msg_t *rx_msg; /* received message */
152 int rx_nob; /* received message size */
153 unsigned long rx_treceived; /* time received */
154 ptl_process_id_t rx_initiator; /* sender's address */
156 ptl_uid_t rx_uid; /* sender's uid */
158 kptl_peer_t *rx_peer; /* pointer to peer */
159 char rx_space[0]; /* copy of incoming request */
162 #define PTLLND_POSTRX_DONT_POST 0 /* don't post */
163 #define PTLLND_POSTRX_NO_CREDIT 1 /* post: no credits */
164 #define PTLLND_POSTRX_PEER_CREDIT 2 /* post: give peer back 1 credit */
166 typedef struct kptl_rx_buffer_pool
168 spinlock_t rxbp_lock;
169 struct list_head rxbp_list; /* all allocated buffers */
170 int rxbp_count; /* # allocated buffers */
171 int rxbp_reserved; /* # requests to buffer */
172 int rxbp_shutdown; /* shutdown flag */
173 } kptl_rx_buffer_pool_t;
175 struct kptl_rx_buffer
177 kptl_rx_buffer_pool_t *rxb_pool;
178 struct list_head rxb_list; /* for the rxb_pool list */
179 struct list_head rxb_repost_list;/* for the kptl_sched_rxbq list */
180 int rxb_posted:1; /* on the net */
181 int rxb_idle:1; /* all done */
182 kptl_eventarg_t rxb_eventarg; /* event->md.user_ptr */
183 int rxb_refcount; /* reference count */
184 ptl_handle_md_t rxb_mdh; /* the portals memory descriptor (MD) handle */
185 char *rxb_buffer; /* the buffer */
191 TX_TYPE_RESERVED = 0,
192 TX_TYPE_SMALL_MESSAGE = 1,
193 TX_TYPE_PUT_REQUEST = 2,
194 TX_TYPE_GET_REQUEST = 3,
195 TX_TYPE_PUT_RESPONSE = 4,
196 TX_TYPE_GET_RESPONSE = 5,
200 #ifdef _USING_LUSTRE_PORTALS_
201 struct iovec iov[PTL_MD_MAX_IOV];
202 lnet_kiov_t kiov[PTL_MD_MAX_IOV];
204 ptl_md_iovec_t iov[PTL_MD_MAX_IOV];
208 typedef struct kptl_tx /* transmit message */
210 struct list_head tx_list; /* queue on idle_txs etc */
211 atomic_t tx_refcount; /* reference count*/
212 enum kptl_tx_type tx_type; /* small msg/{put,get}{req,resp} */
213 int tx_active:1; /* queued on the peer */
214 int tx_idle:1; /* on the free list */
215 int tx_acked:1; /* portals ACK wanted (for debug only) */
216 kptl_eventarg_t tx_msg_eventarg; /* event->md.user_ptr */
217 kptl_eventarg_t tx_rdma_eventarg; /* event->md.user_ptr */
218 int tx_status; /* the status of this tx descriptor */
219 ptl_handle_md_t tx_rdma_mdh; /* RDMA buffer */
220 ptl_handle_md_t tx_msg_mdh; /* the portals MD handle for the initial message */
221 lnet_msg_t *tx_lnet_msg; /* LNET message to finalize */
222 lnet_msg_t *tx_lnet_replymsg; /* LNET reply message to finalize */
223 kptl_msg_t *tx_msg; /* the message data */
224 kptl_peer_t *tx_peer; /* the peer this is waiting on */
225 unsigned long tx_deadline; /* deadline */
226 unsigned long tx_tposted; /* time posted */
227 ptl_md_t tx_rdma_md; /* rdma descriptor */
228 kptl_fragvec_t *tx_frags; /* buffer fragments */
231 enum kptllnd_peer_state
233 PEER_STATE_UNINITIALIZED = 0,
234 PEER_STATE_ALLOCATED = 1,
235 PEER_STATE_WAITING_HELLO = 2,
236 PEER_STATE_ACTIVE = 3,
237 PEER_STATE_CLOSING = 4,
238 PEER_STATE_ZOMBIE = 5,
243 struct list_head peer_list;
244 atomic_t peer_refcount; /* The current refrences */
245 enum kptllnd_peer_state peer_state;
246 spinlock_t peer_lock; /* serialize */
247 struct list_head peer_noops; /* PTLLND_MSG_TYPE_NOOP txs */
248 struct list_head peer_sendq; /* txs waiting for mh handles */
249 struct list_head peer_activeq; /* txs awaiting completion */
250 lnet_process_id_t peer_id; /* Peer's LNET id */
251 ptl_process_id_t peer_ptlid; /* Peer's portals id */
252 __u64 peer_incarnation; /* peer's incarnation */
253 __u64 peer_myincarnation; /* my incarnation at HELLO */
254 int peer_sent_hello; /* have I sent HELLO? */
255 int peer_credits; /* number of send credits */
256 int peer_outstanding_credits;/* number of peer credits to return */
257 int peer_sent_credits; /* #msg buffers posted for peer */
258 int peer_max_msg_size; /* peer's rx buffer size */
259 int peer_error; /* errno on closing this peer */
260 int peer_retry_noop; /* need to retry returning credits */
261 int peer_check_stamp; /* watchdog check stamp */
262 cfs_time_t peer_last_alive; /* when (in jiffies) I was last alive */
263 __u64 peer_next_matchbits; /* Next value to register RDMA from peer */
264 __u64 peer_last_matchbits_seen; /* last matchbits used to RDMA to peer */
269 int kptl_init; /* initialisation state */
270 volatile int kptl_shutdown; /* shut down? */
271 atomic_t kptl_nthreads; /* # live threads */
272 lnet_ni_t *kptl_ni; /* _the_ LND instance */
273 ptl_handle_ni_t kptl_nih; /* network inteface handle */
274 ptl_process_id_t kptl_portals_id; /* Portals ID of interface */
275 __u64 kptl_incarnation; /* which one am I */
276 ptl_handle_eq_t kptl_eqh; /* Event Queue (EQ) */
278 spinlock_t kptl_sched_lock; /* serialise... */
279 wait_queue_head_t kptl_sched_waitq; /* schedulers sleep here */
280 struct list_head kptl_sched_txq; /* tx requiring attention */
281 struct list_head kptl_sched_rxq; /* rx requiring attention */
282 struct list_head kptl_sched_rxbq; /* rxb requiring reposting */
284 wait_queue_head_t kptl_watchdog_waitq; /* watchdog sleeps here */
286 kptl_rx_buffer_pool_t kptl_rx_buffer_pool; /* rx buffer pool */
287 cfs_mem_cache_t* kptl_rx_cache; /* rx descripter cache */
289 atomic_t kptl_ntx; /* # tx descs allocated */
290 spinlock_t kptl_tx_lock; /* serialise idle tx list*/
291 struct list_head kptl_idle_txs; /* idle tx descriptors */
293 rwlock_t kptl_peer_rw_lock; /* lock for peer table */
294 struct list_head *kptl_peers; /* hash table of all my known peers */
295 struct list_head kptl_closing_peers; /* peers being closed */
296 struct list_head kptl_zombie_peers; /* peers waiting for refs to drain */
297 int kptl_peer_hash_size; /* size of kptl_peers */
298 int kptl_npeers; /* # peers extant */
299 int kptl_n_active_peers; /* # active peers */
300 int kptl_expected_peers; /* # peers I can buffer HELLOs from */
302 kptl_msg_t *kptl_nak_msg; /* common NAK message */
303 spinlock_t kptl_ptlid2str_lock; /* serialise str ops */
308 PTLLND_INIT_NOTHING = 0,
313 extern kptl_tunables_t kptllnd_tunables;
314 extern kptl_data_t kptllnd_data;
316 static inline lnet_nid_t
317 kptllnd_ptl2lnetnid(ptl_nid_t ptl_nid)
319 #ifdef _USING_LUSTRE_PORTALS_
320 return LNET_MKNID(LNET_NIDNET(kptllnd_data.kptl_ni->ni_nid),
321 LNET_NIDADDR(ptl_nid));
323 return LNET_MKNID(LNET_NIDNET(kptllnd_data.kptl_ni->ni_nid),
328 static inline ptl_nid_t
329 kptllnd_lnet2ptlnid(lnet_nid_t lnet_nid)
331 #ifdef _USING_LUSTRE_PORTALS_
332 return LNET_MKNID(LNET_NIDNET(kptllnd_data.kptl_portals_id.nid),
333 LNET_NIDADDR(lnet_nid));
335 return LNET_NIDADDR(lnet_nid);
339 int kptllnd_startup(lnet_ni_t *ni);
340 void kptllnd_shutdown(lnet_ni_t *ni);
341 int kptllnd_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
342 int kptllnd_send(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
343 int kptllnd_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
344 int delayed, unsigned int niov,
345 struct iovec *iov, lnet_kiov_t *kiov,
346 unsigned int offset, unsigned int mlen, unsigned int rlen);
347 int kptllnd_eager_recv(struct lnet_ni *ni, void *private,
348 lnet_msg_t *msg, void **new_privatep);
349 void kptllnd_eq_callback(ptl_event_t *evp);
350 int kptllnd_scheduler(void *arg);
351 int kptllnd_watchdog(void *arg);
352 int kptllnd_thread_start(int (*fn)(void *arg), void *arg);
353 int kptllnd_tunables_init(void);
354 void kptllnd_tunables_fini(void);
356 const char *kptllnd_evtype2str(int evtype);
357 const char *kptllnd_msgtype2str(int msgtype);
358 const char *kptllnd_errtype2str(int errtype);
361 kptllnd_eventarg2obj (kptl_eventarg_t *eva)
363 switch (eva->eva_type) {
366 case PTLLND_EVENTARG_TYPE_BUF:
367 return list_entry(eva, kptl_rx_buffer_t, rxb_eventarg);
368 case PTLLND_EVENTARG_TYPE_RDMA:
369 return list_entry(eva, kptl_tx_t, tx_rdma_eventarg);
370 case PTLLND_EVENTARG_TYPE_MSG:
371 return list_entry(eva, kptl_tx_t, tx_msg_eventarg);
376 * RX BUFFER SUPPORT FUNCTIONS
378 void kptllnd_rx_buffer_pool_init(kptl_rx_buffer_pool_t *rxbp);
379 void kptllnd_rx_buffer_pool_fini(kptl_rx_buffer_pool_t *rxbp);
380 int kptllnd_rx_buffer_pool_reserve(kptl_rx_buffer_pool_t *rxbp, int count);
381 void kptllnd_rx_buffer_pool_unreserve(kptl_rx_buffer_pool_t *rxbp, int count);
382 void kptllnd_rx_buffer_callback(ptl_event_t *ev);
383 void kptllnd_rx_buffer_post(kptl_rx_buffer_t *rxb);
386 kptllnd_rx_buffer_size(void)
388 return PAGE_SIZE * (*kptllnd_tunables.kptl_rxb_npages);
392 kptllnd_rx_buffer_addref(kptl_rx_buffer_t *rxb)
396 spin_lock_irqsave(&rxb->rxb_pool->rxbp_lock, flags);
398 spin_unlock_irqrestore(&rxb->rxb_pool->rxbp_lock, flags);
402 kptllnd_rx_buffer_decref_locked(kptl_rx_buffer_t *rxb)
404 if (--(rxb->rxb_refcount) == 0) {
405 spin_lock(&kptllnd_data.kptl_sched_lock);
407 list_add_tail(&rxb->rxb_repost_list,
408 &kptllnd_data.kptl_sched_rxbq);
409 wake_up(&kptllnd_data.kptl_sched_waitq);
411 spin_unlock(&kptllnd_data.kptl_sched_lock);
416 kptllnd_rx_buffer_decref(kptl_rx_buffer_t *rxb)
421 spin_lock_irqsave(&rxb->rxb_pool->rxbp_lock, flags);
422 count = --(rxb->rxb_refcount);
423 spin_unlock_irqrestore(&rxb->rxb_pool->rxbp_lock, flags);
426 kptllnd_rx_buffer_post(rxb);
430 * RX SUPPORT FUNCTIONS
432 void kptllnd_rx_parse(kptl_rx_t *rx);
433 void kptllnd_rx_done(kptl_rx_t *rx, int post_credit);
436 * PEER SUPPORT FUNCTIONS
438 int kptllnd_get_peer_info(int index,
439 lnet_process_id_t *id,
440 int *state, int *sent_hello,
441 int *refcount, __u64 *incarnation,
442 __u64 *next_matchbits, __u64 *last_matchbits_seen,
443 int *nsendq, int *nactiveq,
444 int *credits, int *outstanding_credits);
445 void kptllnd_peer_destroy(kptl_peer_t *peer);
446 int kptllnd_peer_del(lnet_process_id_t id);
447 void kptllnd_peer_close_locked(kptl_peer_t *peer, int why);
448 void kptllnd_peer_close(kptl_peer_t *peer, int why);
449 void kptllnd_handle_closing_peers(void);
450 int kptllnd_peer_connect(kptl_tx_t *tx, lnet_nid_t nid);
451 void kptllnd_peer_check_sends(kptl_peer_t *peer);
452 void kptllnd_peer_check_bucket(int idx, int stamp);
453 void kptllnd_tx_launch(kptl_peer_t *peer, kptl_tx_t *tx, int nfrag);
454 int kptllnd_find_target(kptl_peer_t **peerp, lnet_process_id_t target);
455 kptl_peer_t *kptllnd_peer_handle_hello(ptl_process_id_t initiator,
457 kptl_peer_t *kptllnd_id2peer_locked(lnet_process_id_t id);
458 void kptllnd_peer_alive(kptl_peer_t *peer);
461 kptllnd_peer_addref (kptl_peer_t *peer)
463 atomic_inc(&peer->peer_refcount);
467 kptllnd_peer_decref (kptl_peer_t *peer)
469 if (atomic_dec_and_test(&peer->peer_refcount))
470 kptllnd_peer_destroy(peer);
474 kptllnd_set_tx_peer(kptl_tx_t *tx, kptl_peer_t *peer)
476 LASSERT (tx->tx_peer == NULL);
478 kptllnd_peer_addref(peer);
482 static inline struct list_head *
483 kptllnd_nid2peerlist(lnet_nid_t nid)
485 unsigned int hash = ((unsigned int)nid) %
486 kptllnd_data.kptl_peer_hash_size;
488 return &kptllnd_data.kptl_peers[hash];
491 static inline kptl_peer_t *
492 kptllnd_id2peer(lnet_process_id_t id)
497 read_lock_irqsave(&kptllnd_data.kptl_peer_rw_lock, flags);
498 peer = kptllnd_id2peer_locked(id);
499 read_unlock_irqrestore(&kptllnd_data.kptl_peer_rw_lock, flags);
505 kptllnd_reserve_buffers(int n)
507 return kptllnd_rx_buffer_pool_reserve(&kptllnd_data.kptl_rx_buffer_pool,
512 kptllnd_peer_reserve_buffers(void)
514 return kptllnd_reserve_buffers(*kptllnd_tunables.kptl_peercredits);
518 kptllnd_peer_unreserve_buffers(void)
520 kptllnd_rx_buffer_pool_unreserve(&kptllnd_data.kptl_rx_buffer_pool,
521 *kptllnd_tunables.kptl_peercredits);
525 * TX SUPPORT FUNCTIONS
527 int kptllnd_setup_tx_descs(void);
528 void kptllnd_cleanup_tx_descs(void);
529 void kptllnd_tx_fini(kptl_tx_t *tx);
530 kptl_tx_t *kptllnd_get_idle_tx(enum kptl_tx_type purpose);
531 void kptllnd_tx_callback(ptl_event_t *ev);
532 const char *kptllnd_tx_typestr(int type);
535 kptllnd_tx_addref(kptl_tx_t *tx)
537 atomic_inc(&tx->tx_refcount);
541 kptllnd_tx_decref(kptl_tx_t *tx)
543 LASSERT (!in_interrupt()); /* Thread context only */
545 if (atomic_dec_and_test(&tx->tx_refcount))
550 * MESSAGE SUPPORT FUNCTIONS
552 void kptllnd_init_msg(kptl_msg_t *msg, int type, int body_nob);
553 void kptllnd_msg_pack(kptl_msg_t *msg, kptl_peer_t *peer);
554 int kptllnd_msg_unpack(kptl_msg_t *msg, int nob);
557 * MISC SUPPORT FUNCTIONS
559 void kptllnd_init_rdma_md(kptl_tx_t *tx, unsigned int niov,
560 struct iovec *iov, lnet_kiov_t *kiov,
561 unsigned int offset, unsigned int nob);
562 char *kptllnd_ptlid2str(ptl_process_id_t id);
564 void kptllnd_init_ptltrace(void);
565 void kptllnd_dump_ptltrace(void);
568 #define SIMULATION_FAIL_TX_PUT_ALLOC 0 /* 0x00000001 */
569 #define SIMULATION_FAIL_TX_GET_ALLOC 1 /* 0x00000002 */
570 #define SIMULATION_FAIL_TX 2 /* 0x00000004 */
571 #define SIMULATION_FAIL_RX_ALLOC 3 /* 0x00000008 */
573 #define IS_SIMULATION_ENABLED(x) \
574 (((*kptllnd_tunables.kptl_simulation_bitmap) & 1<< SIMULATION_##x) != 0)
576 #define IS_SIMULATION_ENABLED(x) 0