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/selftest/selftest.h
38 * Author: Isaac Huang <isaac@clusterfs.com>
40 #ifndef __SELFTEST_SELFTEST_H__
41 #define __SELFTEST_SELFTEST_H__
47 /* XXX workaround XXX */
48 #ifdef HAVE_SYS_TYPES_H
49 #include <sys/types.h>
52 /* TODO: remove these when libcfs provides proper primitives for userspace
54 * Dummy implementations of spinlock_t and atomic_t work since userspace
55 * selftest is completely single-threaded, even using multi-threaded usocklnd.
57 typedef struct { } spinlock_t;
58 static inline void spin_lock(spinlock_t *l) {return;}
59 static inline void spin_unlock(spinlock_t *l) {return;}
60 static inline void spin_lock_init(spinlock_t *l) {return;}
62 typedef struct { volatile int counter; } atomic_t;
63 #define atomic_read(a) ((a)->counter)
64 #define atomic_set(a,b) do {(a)->counter = b; } while (0)
65 #define atomic_dec_and_test(a) ((--((a)->counter)) == 0)
66 #define atomic_inc(a) (((a)->counter)++)
67 #define atomic_dec(a) do { (a)->counter--; } while (0)
71 #include <libcfs/kp30.h>
72 #include <libcfs/libcfs.h>
73 #include <lnet/lnet.h>
74 #include <lnet/lib-lnet.h>
75 #include <lnet/lib-types.h>
76 #include <lnet/lnetst.h>
81 #ifndef MADE_WITHOUT_COMPROMISE
82 #define MADE_WITHOUT_COMPROMISE
86 #define SWI_STATE_NEWBORN 0
87 #define SWI_STATE_REPLY_SUBMITTED 1
88 #define SWI_STATE_REPLY_SENT 2
89 #define SWI_STATE_REQUEST_SUBMITTED 3
90 #define SWI_STATE_REQUEST_SENT 4
91 #define SWI_STATE_REPLY_RECEIVED 5
92 #define SWI_STATE_BULK_STARTED 6
93 #define SWI_STATE_BULK_ERRORED 7
94 #define SWI_STATE_DONE 10
100 struct sfw_test_instance;
103 * A workitems is deferred work with these semantics:
104 * - a workitem always runs in thread context.
105 * - a workitem can be concurrent with other workitems but is strictly
106 * serialized with respect to itself.
107 * - no CPU affinity, a workitem does not necessarily run on the same CPU
108 * that schedules it. However, this might change in the future.
109 * - if a workitem is scheduled again before it has a chance to run, it
111 * - if a workitem is scheduled while it runs, it runs again after it
112 * completes; this ensures that events occurring while other events are
113 * being processed receive due attention. This behavior also allows a
114 * workitem to reschedule itself.
117 * - a workitem can sleep but it should be aware of how that sleep might
119 * - a workitem runs inside a kernel thread so there's no user space to access.
120 * - do not use a workitem if the scheduling latency can't be tolerated.
122 * When wi_action returns non-zero, it means the workitem has either been
123 * freed or reused and workitem scheduler won't touch it any more.
125 typedef int (*swi_action_t) (struct swi_workitem *);
126 typedef struct swi_workitem {
127 struct list_head wi_list; /* chain on runq */
129 swi_action_t wi_action;
131 unsigned int wi_running:1;
132 unsigned int wi_scheduled:1;
136 swi_init_workitem (swi_workitem_t *wi, void *data, swi_action_t action)
138 CFS_INIT_LIST_HEAD(&wi->wi_list);
141 wi->wi_scheduled = 0;
143 wi->wi_action = action;
144 wi->wi_state = SWI_STATE_NEWBORN;
147 #define SWI_RESCHED 128 /* # workitem scheduler loops before reschedule */
149 /* services below SRPC_FRAMEWORK_SERVICE_MAX_ID are framework
150 * services, e.g. create/modify session.
152 #define SRPC_SERVICE_DEBUG 0
153 #define SRPC_SERVICE_MAKE_SESSION 1
154 #define SRPC_SERVICE_REMOVE_SESSION 2
155 #define SRPC_SERVICE_BATCH 3
156 #define SRPC_SERVICE_TEST 4
157 #define SRPC_SERVICE_QUERY_STAT 5
158 #define SRPC_SERVICE_JOIN 6
159 #define SRPC_FRAMEWORK_SERVICE_MAX_ID 10
160 /* other services start from SRPC_FRAMEWORK_SERVICE_MAX_ID+1 */
161 #define SRPC_SERVICE_BRW 11
162 #define SRPC_SERVICE_PING 12
163 #define SRPC_SERVICE_MAX_ID 12
165 #define SRPC_REQUEST_PORTAL 50
166 /* a lazy portal for framework RPC requests */
167 #define SRPC_FRAMEWORK_REQUEST_PORTAL 51
168 /* all reply/bulk RDMAs go to this portal */
169 #define SRPC_RDMA_PORTAL 52
171 static inline srpc_msg_type_t
172 srpc_service2request (int service)
177 case SRPC_SERVICE_DEBUG:
178 return SRPC_MSG_DEBUG_REQST;
180 case SRPC_SERVICE_MAKE_SESSION:
181 return SRPC_MSG_MKSN_REQST;
183 case SRPC_SERVICE_REMOVE_SESSION:
184 return SRPC_MSG_RMSN_REQST;
186 case SRPC_SERVICE_BATCH:
187 return SRPC_MSG_BATCH_REQST;
189 case SRPC_SERVICE_TEST:
190 return SRPC_MSG_TEST_REQST;
192 case SRPC_SERVICE_QUERY_STAT:
193 return SRPC_MSG_STAT_REQST;
195 case SRPC_SERVICE_BRW:
196 return SRPC_MSG_BRW_REQST;
198 case SRPC_SERVICE_PING:
199 return SRPC_MSG_PING_REQST;
201 case SRPC_SERVICE_JOIN:
202 return SRPC_MSG_JOIN_REQST;
206 static inline srpc_msg_type_t
207 srpc_service2reply (int service)
209 return srpc_service2request(service) + 1;
213 SRPC_BULK_REQ_RCVD = 0, /* passive bulk request(PUT sink/GET source) received */
214 SRPC_BULK_PUT_SENT = 1, /* active bulk PUT sent (source) */
215 SRPC_BULK_GET_RPLD = 2, /* active bulk GET replied (sink) */
216 SRPC_REPLY_RCVD = 3, /* incoming reply received */
217 SRPC_REPLY_SENT = 4, /* outgoing reply sent */
218 SRPC_REQUEST_RCVD = 5, /* incoming request received */
219 SRPC_REQUEST_SENT = 6, /* outgoing request sent */
224 srpc_event_type_t ev_type; /* what's up */
225 lnet_event_kind_t ev_lnet; /* LNet event type */
226 int ev_fired; /* LNet event fired? */
227 int ev_status; /* LNet event status */
228 void *ev_data; /* owning server/client RPC */
232 int bk_len; /* len of bulk data */
233 lnet_handle_md_t bk_mdh;
234 int bk_sink; /* sink/source */
235 int bk_niov; /* # iov in bk_iovs */
237 lnet_kiov_t bk_iovs[0];
239 cfs_page_t **bk_pages;
240 lnet_md_iovec_t bk_iovs[0];
242 } srpc_bulk_t; /* bulk descriptor */
244 typedef struct srpc_peer {
245 struct list_head stp_list; /* chain on peer hash */
246 struct list_head stp_rpcq; /* q of non-control RPCs */
247 struct list_head stp_ctl_rpcq; /* q of control RPCs */
248 spinlock_t stp_lock; /* serialize */
250 int stp_credits; /* available credits */
253 /* message buffer descriptor */
255 struct list_head buf_list; /* chain on srpc_service::*_msgq */
257 lnet_handle_md_t buf_mdh;
259 lnet_process_id_t buf_peer;
262 /* server-side state of a RPC */
263 typedef struct srpc_server_rpc {
264 struct list_head srpc_list; /* chain on srpc_service::*_rpcq */
265 struct srpc_service *srpc_service;
266 swi_workitem_t srpc_wi;
267 srpc_event_t srpc_ev; /* bulk/reply event */
268 lnet_nid_t srpc_self;
269 lnet_process_id_t srpc_peer;
270 srpc_msg_t srpc_replymsg;
271 lnet_handle_md_t srpc_replymdh;
272 srpc_buffer_t *srpc_reqstbuf;
273 srpc_bulk_t *srpc_bulk;
276 void (*srpc_done)(struct srpc_server_rpc *);
279 /* client-side state of a RPC */
280 typedef struct srpc_client_rpc {
281 struct list_head crpc_list; /* chain on user's lists */
282 struct list_head crpc_privl; /* chain on srpc_peer_t::*rpcq */
283 spinlock_t crpc_lock; /* serialize */
285 atomic_t crpc_refcount;
286 int crpc_timeout; /* # seconds to wait for reply */
287 stt_timer_t crpc_timer;
288 swi_workitem_t crpc_wi;
289 lnet_process_id_t crpc_dest;
290 srpc_peer_t *crpc_peer;
292 void (*crpc_done)(struct srpc_client_rpc *);
293 void (*crpc_fini)(struct srpc_client_rpc *);
294 int crpc_status; /* completion status */
295 void *crpc_priv; /* caller data */
298 unsigned int crpc_aborted:1; /* being given up */
299 unsigned int crpc_closed:1; /* completed */
302 srpc_event_t crpc_bulkev; /* bulk event */
303 srpc_event_t crpc_reqstev; /* request event */
304 srpc_event_t crpc_replyev; /* reply event */
306 /* bulk, request(reqst), and reply exchanged on wire */
307 srpc_msg_t crpc_reqstmsg;
308 srpc_msg_t crpc_replymsg;
309 lnet_handle_md_t crpc_reqstmdh;
310 lnet_handle_md_t crpc_replymdh;
311 srpc_bulk_t crpc_bulk;
314 #define srpc_client_rpc_size(rpc) \
315 offsetof(srpc_client_rpc_t, crpc_bulk.bk_iovs[(rpc)->crpc_bulk.bk_niov])
317 #define srpc_client_rpc_addref(rpc) \
319 CDEBUG(D_NET, "RPC[%p] -> %s (%d)++\n", \
320 (rpc), libcfs_id2str((rpc)->crpc_dest), \
321 atomic_read(&(rpc)->crpc_refcount)); \
322 LASSERT(atomic_read(&(rpc)->crpc_refcount) > 0); \
323 atomic_inc(&(rpc)->crpc_refcount); \
326 #define srpc_client_rpc_decref(rpc) \
328 CDEBUG(D_NET, "RPC[%p] -> %s (%d)--\n", \
329 (rpc), libcfs_id2str((rpc)->crpc_dest), \
330 atomic_read(&(rpc)->crpc_refcount)); \
331 LASSERT(atomic_read(&(rpc)->crpc_refcount) > 0); \
332 if (atomic_dec_and_test(&(rpc)->crpc_refcount)) \
333 srpc_destroy_client_rpc(rpc); \
336 #define srpc_event_pending(rpc) ((rpc)->crpc_bulkev.ev_fired == 0 || \
337 (rpc)->crpc_reqstev.ev_fired == 0 || \
338 (rpc)->crpc_replyev.ev_fired == 0)
340 typedef struct srpc_service {
341 int sv_id; /* service id */
342 const char *sv_name; /* human readable name */
343 int sv_nprune; /* # posted RPC to be pruned */
344 int sv_concur; /* max # concurrent RPCs */
348 srpc_event_t sv_ev; /* LNet event */
349 int sv_nposted_msg; /* # posted message buffers */
350 struct list_head sv_free_rpcq; /* free RPC descriptors */
351 struct list_head sv_active_rpcq; /* in-flight RPCs */
352 struct list_head sv_posted_msgq; /* posted message buffers */
353 struct list_head sv_blocked_msgq; /* blocked for RPC descriptor */
355 /* Service callbacks:
356 * - sv_handler: process incoming RPC request
357 * - sv_bulk_ready: notify bulk data
359 int (*sv_handler) (srpc_server_rpc_t *);
360 int (*sv_bulk_ready) (srpc_server_rpc_t *, int);
363 #define SFW_POST_BUFFERS 8
364 #define SFW_SERVICE_CONCURRENCY (SFW_POST_BUFFERS/2)
367 struct list_head sn_list; /* chain on fw_zombie_sessions */
368 lst_sid_t sn_id; /* unique identifier */
369 unsigned int sn_timeout; /* # seconds' inactivity to expire */
371 stt_timer_t sn_timer;
372 struct list_head sn_batches; /* list of batches */
373 char sn_name[LST_NAME_SIZE];
374 atomic_t sn_refcount;
375 atomic_t sn_brw_errors;
376 atomic_t sn_ping_errors;
379 #define sfw_sid_equal(sid0, sid1) ((sid0).ses_nid == (sid1).ses_nid && \
380 (sid0).ses_stamp == (sid1).ses_stamp)
383 struct list_head bat_list; /* chain on sn_batches */
384 lst_bid_t bat_id; /* batch id */
385 int bat_error; /* error code of batch */
386 sfw_session_t *bat_session; /* batch's session */
387 atomic_t bat_nactive; /* # of active tests */
388 struct list_head bat_tests; /* test instances */
392 int (*tso_init)(struct sfw_test_instance *tsi); /* intialize test client */
393 void (*tso_fini)(struct sfw_test_instance *tsi); /* finalize test client */
394 int (*tso_prep_rpc)(struct sfw_test_unit *tsu,
395 lnet_process_id_t dest,
396 srpc_client_rpc_t **rpc); /* prep a tests rpc */
397 void (*tso_done_rpc)(struct sfw_test_unit *tsu,
398 srpc_client_rpc_t *rpc); /* done a test rpc */
399 } sfw_test_client_ops_t;
401 typedef struct sfw_test_instance {
402 struct list_head tsi_list; /* chain on batch */
403 int tsi_service; /* test type */
404 sfw_batch_t *tsi_batch; /* batch */
405 sfw_test_client_ops_t *tsi_ops; /* test client operations */
407 /* public parameter for all test units */
408 int tsi_is_client:1; /* is test client */
409 int tsi_stoptsu_onerr:1; /* stop tsu on error */
410 int tsi_concur; /* concurrency */
411 int tsi_loop; /* loop count */
413 /* status of test instance */
414 spinlock_t tsi_lock; /* serialize */
415 int tsi_stopping:1; /* test is stopping */
416 atomic_t tsi_nactive; /* # of active test unit */
417 struct list_head tsi_units; /* test units */
418 struct list_head tsi_free_rpcs; /* free rpcs */
419 struct list_head tsi_active_rpcs; /* active rpcs */
422 test_bulk_req_t bulk; /* bulk parameter */
423 test_ping_req_t ping; /* ping parameter */
425 } sfw_test_instance_t;
427 /* XXX: trailing (CFS_PAGE_SIZE % sizeof(lnet_process_id_t)) bytes at
428 * the end of pages are not used */
429 #define SFW_MAX_CONCUR LST_MAX_CONCUR
430 #define SFW_ID_PER_PAGE (CFS_PAGE_SIZE / sizeof(lnet_process_id_t))
431 #define SFW_MAX_NDESTS (LNET_MAX_IOV * SFW_ID_PER_PAGE)
432 #define sfw_id_pages(n) (((n) + SFW_ID_PER_PAGE - 1) / SFW_ID_PER_PAGE)
434 typedef struct sfw_test_unit {
435 struct list_head tsu_list; /* chain on lst_test_instance */
436 lnet_process_id_t tsu_dest; /* id of dest node */
437 int tsu_loop; /* loop count of the test */
438 sfw_test_instance_t *tsu_instance; /* pointer to test instance */
439 void *tsu_private; /* private data */
440 swi_workitem_t tsu_worker; /* workitem of the test unit */
444 struct list_head tsc_list; /* chain on fw_tests */
445 srpc_service_t *tsc_srv_service; /* test service */
446 sfw_test_client_ops_t *tsc_cli_ops; /* ops of test client */
451 sfw_create_rpc(lnet_process_id_t peer, int service, int nbulkiov, int bulklen,
452 void (*done) (srpc_client_rpc_t *), void *priv);
453 int sfw_create_test_rpc(sfw_test_unit_t *tsu, lnet_process_id_t peer,
454 int nblk, int blklen, srpc_client_rpc_t **rpc);
455 void sfw_abort_rpc(srpc_client_rpc_t *rpc);
456 void sfw_post_rpc(srpc_client_rpc_t *rpc);
457 void sfw_client_rpc_done(srpc_client_rpc_t *rpc);
458 void sfw_unpack_message(srpc_msg_t *msg);
459 void sfw_free_pages(srpc_server_rpc_t *rpc);
460 void sfw_add_bulk_page(srpc_bulk_t *bk, cfs_page_t *pg, int i);
461 int sfw_alloc_pages(srpc_server_rpc_t *rpc, int npages, int sink);
464 srpc_create_client_rpc(lnet_process_id_t peer, int service,
465 int nbulkiov, int bulklen,
466 void (*rpc_done)(srpc_client_rpc_t *),
467 void (*rpc_fini)(srpc_client_rpc_t *), void *priv);
468 void srpc_post_rpc(srpc_client_rpc_t *rpc);
469 void srpc_abort_rpc(srpc_client_rpc_t *rpc, int why);
470 void srpc_free_bulk(srpc_bulk_t *bk);
471 srpc_bulk_t *srpc_alloc_bulk(int npages, int sink);
472 int srpc_send_rpc(swi_workitem_t *wi);
473 int srpc_send_reply(srpc_server_rpc_t *rpc);
474 int srpc_add_service(srpc_service_t *sv);
475 int srpc_remove_service(srpc_service_t *sv);
476 void srpc_shutdown_service(srpc_service_t *sv);
477 int srpc_finish_service(srpc_service_t *sv);
478 int srpc_service_add_buffers(srpc_service_t *sv, int nbuffer);
479 void srpc_service_remove_buffers(srpc_service_t *sv, int nbuffer);
480 void srpc_get_counters(srpc_counters_t *cnt);
481 void srpc_set_counters(const srpc_counters_t *cnt);
483 void swi_kill_workitem(swi_workitem_t *wi);
484 void swi_schedule_workitem(swi_workitem_t *wi);
485 void swi_schedule_serial_workitem(swi_workitem_t *wi);
486 int swi_startup(void);
487 int sfw_startup(void);
488 int srpc_startup(void);
489 void swi_shutdown(void);
490 void sfw_shutdown(void);
491 void srpc_shutdown(void);
494 srpc_destroy_client_rpc (srpc_client_rpc_t *rpc)
496 LASSERT (rpc != NULL);
497 LASSERT (!srpc_event_pending(rpc));
498 LASSERT (list_empty(&rpc->crpc_privl));
499 LASSERT (atomic_read(&rpc->crpc_refcount) == 0);
501 LASSERT (rpc->crpc_bulk.bk_pages == NULL);
504 if (rpc->crpc_fini == NULL) {
505 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
507 (*rpc->crpc_fini) (rpc);
514 srpc_init_client_rpc (srpc_client_rpc_t *rpc, lnet_process_id_t peer,
515 int service, int nbulkiov, int bulklen,
516 void (*rpc_done)(srpc_client_rpc_t *),
517 void (*rpc_fini)(srpc_client_rpc_t *), void *priv)
519 LASSERT (nbulkiov <= LNET_MAX_IOV);
521 memset(rpc, 0, offsetof(srpc_client_rpc_t,
522 crpc_bulk.bk_iovs[nbulkiov]));
524 CFS_INIT_LIST_HEAD(&rpc->crpc_list);
525 CFS_INIT_LIST_HEAD(&rpc->crpc_privl);
526 swi_init_workitem(&rpc->crpc_wi, rpc, srpc_send_rpc);
527 spin_lock_init(&rpc->crpc_lock);
528 atomic_set(&rpc->crpc_refcount, 1); /* 1 ref for caller */
530 rpc->crpc_dest = peer;
531 rpc->crpc_priv = priv;
532 rpc->crpc_service = service;
533 rpc->crpc_bulk.bk_len = bulklen;
534 rpc->crpc_bulk.bk_niov = nbulkiov;
535 rpc->crpc_done = rpc_done;
536 rpc->crpc_fini = rpc_fini;
539 rpc->crpc_bulk.bk_mdh = LNET_INVALID_HANDLE;
541 /* no event is expected at this point */
542 rpc->crpc_bulkev.ev_fired =
543 rpc->crpc_reqstev.ev_fired =
544 rpc->crpc_replyev.ev_fired = 1;
546 rpc->crpc_reqstmsg.msg_magic = SRPC_MSG_MAGIC;
547 rpc->crpc_reqstmsg.msg_version = SRPC_MSG_VERSION;
548 rpc->crpc_reqstmsg.msg_type = srpc_service2request(service);
552 static inline const char *
553 swi_state2str (int state)
555 #define STATE2STR(x) case x: return #x
559 STATE2STR(SWI_STATE_NEWBORN);
560 STATE2STR(SWI_STATE_REPLY_SUBMITTED);
561 STATE2STR(SWI_STATE_REPLY_SENT);
562 STATE2STR(SWI_STATE_REQUEST_SUBMITTED);
563 STATE2STR(SWI_STATE_REQUEST_SENT);
564 STATE2STR(SWI_STATE_REPLY_RECEIVED);
565 STATE2STR(SWI_STATE_BULK_STARTED);
566 STATE2STR(SWI_STATE_BULK_ERRORED);
567 STATE2STR(SWI_STATE_DONE);
572 #define UNUSED(x) ( (void)(x) )
576 int stt_poll_interval(void);
577 int sfw_session_removed(void);
579 int stt_check_events(void);
580 int swi_check_events(void);
581 int srpc_check_event(int timeout);
583 int lnet_selftest_init(void);
584 void lnet_selftest_fini(void);
585 int selftest_wait_events(void);
589 #define selftest_wait_events() cfs_pause(cfs_time_seconds(1))
593 #define lst_wait_until(cond, lock, fmt, a...) \
598 CDEBUG(((__I & (-__I)) == __I) ? D_WARNING : \
601 spin_unlock(&(lock)); \
603 selftest_wait_events(); \
605 spin_lock(&(lock)); \
610 srpc_wait_service_shutdown (srpc_service_t *sv)
614 spin_lock(&sv->sv_lock);
615 LASSERT (sv->sv_shuttingdown);
616 spin_unlock(&sv->sv_lock);
618 while (srpc_finish_service(sv) == 0) {
620 CDEBUG (((i & -i) == i) ? D_WARNING : D_NET,
621 "Waiting for %s service to shutdown...\n",
623 selftest_wait_events();
627 #endif /* __SELFTEST_SELFTEST_H__ */