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
23 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
24 * CA 95054 USA or visit www.sun.com if you need additional information or
30 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
31 * Use is subject to license terms.
34 * This file is part of Lustre, http://www.lustre.org/
35 * Lustre is a trademark of Sun Microsystems, Inc.
37 * lnet/selftest/selftest.h
39 * Author: Isaac Huang <isaac@clusterfs.com>
41 #ifndef __SELFTEST_SELFTEST_H__
42 #define __SELFTEST_SELFTEST_H__
48 /* XXX workaround XXX */
49 #ifdef HAVE_SYS_TYPES_H
50 #include <sys/types.h>
54 #include <libcfs/libcfs.h>
55 #include <lnet/lnet.h>
56 #include <lnet/lib-lnet.h>
57 #include <lnet/lib-types.h>
58 #include <lnet/lnetst.h>
63 #ifndef MADE_WITHOUT_COMPROMISE
64 #define MADE_WITHOUT_COMPROMISE
68 #define SWI_STATE_NEWBORN 0
69 #define SWI_STATE_REPLY_SUBMITTED 1
70 #define SWI_STATE_REPLY_SENT 2
71 #define SWI_STATE_REQUEST_SUBMITTED 3
72 #define SWI_STATE_REQUEST_SENT 4
73 #define SWI_STATE_REPLY_RECEIVED 5
74 #define SWI_STATE_BULK_STARTED 6
75 #define SWI_STATE_DONE 10
80 struct sfw_test_instance;
82 /* services below SRPC_FRAMEWORK_SERVICE_MAX_ID are framework
83 * services, e.g. create/modify session.
85 #define SRPC_SERVICE_DEBUG 0
86 #define SRPC_SERVICE_MAKE_SESSION 1
87 #define SRPC_SERVICE_REMOVE_SESSION 2
88 #define SRPC_SERVICE_BATCH 3
89 #define SRPC_SERVICE_TEST 4
90 #define SRPC_SERVICE_QUERY_STAT 5
91 #define SRPC_SERVICE_JOIN 6
92 #define SRPC_FRAMEWORK_SERVICE_MAX_ID 10
93 /* other services start from SRPC_FRAMEWORK_SERVICE_MAX_ID+1 */
94 #define SRPC_SERVICE_BRW 11
95 #define SRPC_SERVICE_PING 12
96 #define SRPC_SERVICE_MAX_ID 12
98 #define SRPC_REQUEST_PORTAL 50
99 /* a lazy portal for framework RPC requests */
100 #define SRPC_FRAMEWORK_REQUEST_PORTAL 51
101 /* all reply/bulk RDMAs go to this portal */
102 #define SRPC_RDMA_PORTAL 52
104 static inline srpc_msg_type_t
105 srpc_service2request (int service)
110 case SRPC_SERVICE_DEBUG:
111 return SRPC_MSG_DEBUG_REQST;
113 case SRPC_SERVICE_MAKE_SESSION:
114 return SRPC_MSG_MKSN_REQST;
116 case SRPC_SERVICE_REMOVE_SESSION:
117 return SRPC_MSG_RMSN_REQST;
119 case SRPC_SERVICE_BATCH:
120 return SRPC_MSG_BATCH_REQST;
122 case SRPC_SERVICE_TEST:
123 return SRPC_MSG_TEST_REQST;
125 case SRPC_SERVICE_QUERY_STAT:
126 return SRPC_MSG_STAT_REQST;
128 case SRPC_SERVICE_BRW:
129 return SRPC_MSG_BRW_REQST;
131 case SRPC_SERVICE_PING:
132 return SRPC_MSG_PING_REQST;
134 case SRPC_SERVICE_JOIN:
135 return SRPC_MSG_JOIN_REQST;
139 static inline srpc_msg_type_t
140 srpc_service2reply (int service)
142 return srpc_service2request(service) + 1;
146 SRPC_BULK_REQ_RCVD = 1, /* passive bulk request(PUT sink/GET source) received */
147 SRPC_BULK_PUT_SENT = 2, /* active bulk PUT sent (source) */
148 SRPC_BULK_GET_RPLD = 3, /* active bulk GET replied (sink) */
149 SRPC_REPLY_RCVD = 4, /* incoming reply received */
150 SRPC_REPLY_SENT = 5, /* outgoing reply sent */
151 SRPC_REQUEST_RCVD = 6, /* incoming request received */
152 SRPC_REQUEST_SENT = 7, /* outgoing request sent */
157 srpc_event_type_t ev_type; /* what's up */
158 lnet_event_kind_t ev_lnet; /* LNet event type */
159 int ev_fired; /* LNet event fired? */
160 int ev_status; /* LNet event status */
161 void *ev_data; /* owning server/client RPC */
165 int bk_len; /* len of bulk data */
166 lnet_handle_md_t bk_mdh;
167 int bk_sink; /* sink/source */
168 int bk_niov; /* # iov in bk_iovs */
170 lnet_kiov_t bk_iovs[0];
172 cfs_page_t **bk_pages;
173 lnet_md_iovec_t bk_iovs[0];
175 } srpc_bulk_t; /* bulk descriptor */
177 /* message buffer descriptor */
179 cfs_list_t buf_list; /* chain on srpc_service::*_msgq */
181 lnet_handle_md_t buf_mdh;
183 lnet_process_id_t buf_peer;
187 typedef int (*swi_action_t) (struct swi_workitem *);
189 typedef struct swi_workitem {
190 cfs_workitem_t swi_workitem;
191 swi_action_t swi_action;
195 /* server-side state of a RPC */
196 typedef struct srpc_server_rpc {
197 cfs_list_t srpc_list; /* chain on srpc_service::*_rpcq */
198 struct srpc_service *srpc_service;
199 swi_workitem_t srpc_wi;
200 srpc_event_t srpc_ev; /* bulk/reply event */
201 lnet_nid_t srpc_self;
202 lnet_process_id_t srpc_peer;
203 srpc_msg_t srpc_replymsg;
204 lnet_handle_md_t srpc_replymdh;
205 srpc_buffer_t *srpc_reqstbuf;
206 srpc_bulk_t *srpc_bulk;
208 unsigned int srpc_aborted; /* being given up */
210 void (*srpc_done)(struct srpc_server_rpc *);
213 /* client-side state of a RPC */
214 typedef struct srpc_client_rpc {
215 cfs_list_t crpc_list; /* chain on user's lists */
216 cfs_spinlock_t crpc_lock; /* serialize */
218 cfs_atomic_t crpc_refcount;
219 int crpc_timeout; /* # seconds to wait for reply */
220 stt_timer_t crpc_timer;
221 swi_workitem_t crpc_wi;
222 lnet_process_id_t crpc_dest;
224 void (*crpc_done)(struct srpc_client_rpc *);
225 void (*crpc_fini)(struct srpc_client_rpc *);
226 int crpc_status; /* completion status */
227 void *crpc_priv; /* caller data */
230 unsigned int crpc_aborted:1; /* being given up */
231 unsigned int crpc_closed:1; /* completed */
234 srpc_event_t crpc_bulkev; /* bulk event */
235 srpc_event_t crpc_reqstev; /* request event */
236 srpc_event_t crpc_replyev; /* reply event */
238 /* bulk, request(reqst), and reply exchanged on wire */
239 srpc_msg_t crpc_reqstmsg;
240 srpc_msg_t crpc_replymsg;
241 lnet_handle_md_t crpc_reqstmdh;
242 lnet_handle_md_t crpc_replymdh;
243 srpc_bulk_t crpc_bulk;
246 #define srpc_client_rpc_size(rpc) \
247 offsetof(srpc_client_rpc_t, crpc_bulk.bk_iovs[(rpc)->crpc_bulk.bk_niov])
249 #define srpc_client_rpc_addref(rpc) \
251 CDEBUG(D_NET, "RPC[%p] -> %s (%d)++\n", \
252 (rpc), libcfs_id2str((rpc)->crpc_dest), \
253 cfs_atomic_read(&(rpc)->crpc_refcount)); \
254 LASSERT(cfs_atomic_read(&(rpc)->crpc_refcount) > 0); \
255 cfs_atomic_inc(&(rpc)->crpc_refcount); \
258 #define srpc_client_rpc_decref(rpc) \
260 CDEBUG(D_NET, "RPC[%p] -> %s (%d)--\n", \
261 (rpc), libcfs_id2str((rpc)->crpc_dest), \
262 cfs_atomic_read(&(rpc)->crpc_refcount)); \
263 LASSERT(cfs_atomic_read(&(rpc)->crpc_refcount) > 0); \
264 if (cfs_atomic_dec_and_test(&(rpc)->crpc_refcount)) \
265 srpc_destroy_client_rpc(rpc); \
268 #define srpc_event_pending(rpc) ((rpc)->crpc_bulkev.ev_fired == 0 || \
269 (rpc)->crpc_reqstev.ev_fired == 0 || \
270 (rpc)->crpc_replyev.ev_fired == 0)
272 typedef struct srpc_service {
273 int sv_id; /* service id */
274 const char *sv_name; /* human readable name */
275 int sv_nprune; /* # posted RPC to be pruned */
276 int sv_concur; /* max # concurrent RPCs */
278 cfs_spinlock_t sv_lock;
280 srpc_event_t sv_ev; /* LNet event */
281 int sv_nposted_msg; /* # posted message buffers */
282 cfs_list_t sv_free_rpcq; /* free RPC descriptors */
283 cfs_list_t sv_active_rpcq; /* in-flight RPCs */
284 cfs_list_t sv_posted_msgq; /* posted message buffers */
285 cfs_list_t sv_blocked_msgq; /* blocked for RPC descriptor */
287 /* Service callbacks:
288 * - sv_handler: process incoming RPC request
289 * - sv_bulk_ready: notify bulk data
291 int (*sv_handler) (srpc_server_rpc_t *);
292 int (*sv_bulk_ready) (srpc_server_rpc_t *, int);
295 #define SFW_POST_BUFFERS 256
296 #define SFW_SERVICE_CONCURRENCY (SFW_POST_BUFFERS/2)
299 cfs_list_t sn_list; /* chain on fw_zombie_sessions */
300 lst_sid_t sn_id; /* unique identifier */
301 unsigned int sn_timeout; /* # seconds' inactivity to expire */
303 stt_timer_t sn_timer;
304 cfs_list_t sn_batches; /* list of batches */
305 char sn_name[LST_NAME_SIZE];
306 cfs_atomic_t sn_refcount;
307 cfs_atomic_t sn_brw_errors;
308 cfs_atomic_t sn_ping_errors;
311 #define sfw_sid_equal(sid0, sid1) ((sid0).ses_nid == (sid1).ses_nid && \
312 (sid0).ses_stamp == (sid1).ses_stamp)
315 cfs_list_t bat_list; /* chain on sn_batches */
316 lst_bid_t bat_id; /* batch id */
317 int bat_error; /* error code of batch */
318 sfw_session_t *bat_session; /* batch's session */
319 cfs_atomic_t bat_nactive; /* # of active tests */
320 cfs_list_t bat_tests; /* test instances */
324 int (*tso_init)(struct sfw_test_instance *tsi); /* intialize test client */
325 void (*tso_fini)(struct sfw_test_instance *tsi); /* finalize test client */
326 int (*tso_prep_rpc)(struct sfw_test_unit *tsu,
327 lnet_process_id_t dest,
328 srpc_client_rpc_t **rpc); /* prep a tests rpc */
329 void (*tso_done_rpc)(struct sfw_test_unit *tsu,
330 srpc_client_rpc_t *rpc); /* done a test rpc */
331 } sfw_test_client_ops_t;
333 typedef struct sfw_test_instance {
334 cfs_list_t tsi_list; /* chain on batch */
335 int tsi_service; /* test type */
336 sfw_batch_t *tsi_batch; /* batch */
337 sfw_test_client_ops_t *tsi_ops; /* test client operations */
339 /* public parameter for all test units */
340 int tsi_is_client:1; /* is test client */
341 int tsi_stoptsu_onerr:1; /* stop tsu on error */
342 int tsi_concur; /* concurrency */
343 int tsi_loop; /* loop count */
345 /* status of test instance */
346 cfs_spinlock_t tsi_lock; /* serialize */
347 int tsi_stopping:1; /* test is stopping */
348 cfs_atomic_t tsi_nactive; /* # of active test unit */
349 cfs_list_t tsi_units; /* test units */
350 cfs_list_t tsi_free_rpcs; /* free rpcs */
351 cfs_list_t tsi_active_rpcs; /* active rpcs */
354 test_bulk_req_t bulk; /* bulk parameter */
355 test_ping_req_t ping; /* ping parameter */
357 } sfw_test_instance_t;
359 /* XXX: trailing (CFS_PAGE_SIZE % sizeof(lnet_process_id_t)) bytes at
360 * the end of pages are not used */
361 #define SFW_MAX_CONCUR LST_MAX_CONCUR
362 #define SFW_ID_PER_PAGE (CFS_PAGE_SIZE / sizeof(lnet_process_id_packed_t))
363 #define SFW_MAX_NDESTS (LNET_MAX_IOV * SFW_ID_PER_PAGE)
364 #define sfw_id_pages(n) (((n) + SFW_ID_PER_PAGE - 1) / SFW_ID_PER_PAGE)
366 typedef struct sfw_test_unit {
367 cfs_list_t tsu_list; /* chain on lst_test_instance */
368 lnet_process_id_t tsu_dest; /* id of dest node */
369 int tsu_loop; /* loop count of the test */
370 sfw_test_instance_t *tsu_instance; /* pointer to test instance */
371 void *tsu_private; /* private data */
372 swi_workitem_t tsu_worker; /* workitem of the test unit */
376 cfs_list_t tsc_list; /* chain on fw_tests */
377 srpc_service_t *tsc_srv_service; /* test service */
378 sfw_test_client_ops_t *tsc_cli_ops; /* ops of test client */
382 sfw_create_rpc(lnet_process_id_t peer, int service, int nbulkiov, int bulklen,
383 void (*done) (srpc_client_rpc_t *), void *priv);
384 int sfw_create_test_rpc(sfw_test_unit_t *tsu, lnet_process_id_t peer,
385 int nblk, int blklen, srpc_client_rpc_t **rpc);
386 void sfw_abort_rpc(srpc_client_rpc_t *rpc);
387 void sfw_post_rpc(srpc_client_rpc_t *rpc);
388 void sfw_client_rpc_done(srpc_client_rpc_t *rpc);
389 void sfw_unpack_message(srpc_msg_t *msg);
390 void sfw_free_pages(srpc_server_rpc_t *rpc);
391 void sfw_add_bulk_page(srpc_bulk_t *bk, cfs_page_t *pg, int i);
392 int sfw_alloc_pages(srpc_server_rpc_t *rpc, int npages, int sink);
393 int sfw_make_session (srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply);
396 srpc_create_client_rpc(lnet_process_id_t peer, int service,
397 int nbulkiov, int bulklen,
398 void (*rpc_done)(srpc_client_rpc_t *),
399 void (*rpc_fini)(srpc_client_rpc_t *), void *priv);
400 void srpc_post_rpc(srpc_client_rpc_t *rpc);
401 void srpc_abort_rpc(srpc_client_rpc_t *rpc, int why);
402 void srpc_free_bulk(srpc_bulk_t *bk);
403 srpc_bulk_t *srpc_alloc_bulk(int npages, int sink);
404 int srpc_send_rpc(swi_workitem_t *wi);
405 int srpc_send_reply(srpc_server_rpc_t *rpc);
406 int srpc_add_service(srpc_service_t *sv);
407 int srpc_remove_service(srpc_service_t *sv);
408 void srpc_shutdown_service(srpc_service_t *sv);
409 void srpc_abort_service(srpc_service_t *sv);
410 int srpc_finish_service(srpc_service_t *sv);
411 int srpc_service_add_buffers(srpc_service_t *sv, int nbuffer);
412 void srpc_service_remove_buffers(srpc_service_t *sv, int nbuffer);
413 void srpc_get_counters(srpc_counters_t *cnt);
414 void srpc_set_counters(const srpc_counters_t *cnt);
417 swi_wi_action(cfs_workitem_t *wi)
419 swi_workitem_t *swi = container_of(wi, swi_workitem_t, swi_workitem);
421 return swi->swi_action(swi);
425 swi_init_workitem (swi_workitem_t *swi, void *data,
426 swi_action_t action, short sched_id)
428 swi->swi_action = action;
429 swi->swi_state = SWI_STATE_NEWBORN;
430 cfs_wi_init(&swi->swi_workitem, data, swi_wi_action, sched_id);
434 swi_schedule_workitem(swi_workitem_t *wi)
436 cfs_wi_schedule(&wi->swi_workitem);
440 swi_kill_workitem(swi_workitem_t *swi)
442 cfs_wi_exit(&swi->swi_workitem);
447 swi_check_events(void)
449 return cfs_wi_check_events();
453 int sfw_startup(void);
454 int srpc_startup(void);
455 void sfw_shutdown(void);
456 void srpc_shutdown(void);
459 srpc_destroy_client_rpc (srpc_client_rpc_t *rpc)
461 LASSERT (rpc != NULL);
462 LASSERT (!srpc_event_pending(rpc));
463 LASSERT (cfs_atomic_read(&rpc->crpc_refcount) == 0);
465 LASSERT (rpc->crpc_bulk.bk_pages == NULL);
468 if (rpc->crpc_fini == NULL) {
469 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
471 (*rpc->crpc_fini) (rpc);
478 srpc_init_client_rpc (srpc_client_rpc_t *rpc, lnet_process_id_t peer,
479 int service, int nbulkiov, int bulklen,
480 void (*rpc_done)(srpc_client_rpc_t *),
481 void (*rpc_fini)(srpc_client_rpc_t *), void *priv)
483 LASSERT (nbulkiov <= LNET_MAX_IOV);
485 memset(rpc, 0, offsetof(srpc_client_rpc_t,
486 crpc_bulk.bk_iovs[nbulkiov]));
488 CFS_INIT_LIST_HEAD(&rpc->crpc_list);
489 swi_init_workitem(&rpc->crpc_wi, rpc, srpc_send_rpc,
491 cfs_spin_lock_init(&rpc->crpc_lock);
492 cfs_atomic_set(&rpc->crpc_refcount, 1); /* 1 ref for caller */
494 rpc->crpc_dest = peer;
495 rpc->crpc_priv = priv;
496 rpc->crpc_service = service;
497 rpc->crpc_bulk.bk_len = bulklen;
498 rpc->crpc_bulk.bk_niov = nbulkiov;
499 rpc->crpc_done = rpc_done;
500 rpc->crpc_fini = rpc_fini;
501 LNetInvalidateHandle(&rpc->crpc_reqstmdh);
502 LNetInvalidateHandle(&rpc->crpc_replymdh);
503 LNetInvalidateHandle(&rpc->crpc_bulk.bk_mdh);
505 /* no event is expected at this point */
506 rpc->crpc_bulkev.ev_fired =
507 rpc->crpc_reqstev.ev_fired =
508 rpc->crpc_replyev.ev_fired = 1;
510 rpc->crpc_reqstmsg.msg_magic = SRPC_MSG_MAGIC;
511 rpc->crpc_reqstmsg.msg_version = SRPC_MSG_VERSION;
512 rpc->crpc_reqstmsg.msg_type = srpc_service2request(service);
516 static inline const char *
517 swi_state2str (int state)
519 #define STATE2STR(x) case x: return #x
523 STATE2STR(SWI_STATE_NEWBORN);
524 STATE2STR(SWI_STATE_REPLY_SUBMITTED);
525 STATE2STR(SWI_STATE_REPLY_SENT);
526 STATE2STR(SWI_STATE_REQUEST_SUBMITTED);
527 STATE2STR(SWI_STATE_REQUEST_SENT);
528 STATE2STR(SWI_STATE_REPLY_RECEIVED);
529 STATE2STR(SWI_STATE_BULK_STARTED);
530 STATE2STR(SWI_STATE_DONE);
535 #define UNUSED(x) ( (void)(x) )
539 int stt_poll_interval(void);
540 int sfw_session_removed(void);
542 int stt_check_events(void);
543 int srpc_check_event(int timeout);
545 int lnet_selftest_init(void);
546 void lnet_selftest_fini(void);
547 int selftest_wait_events(void);
551 #define selftest_wait_events() cfs_pause(cfs_time_seconds(1))
555 #define lst_wait_until(cond, lock, fmt, ...) \
559 CDEBUG(IS_PO2(++__I) ? D_WARNING : D_NET, \
560 fmt, ## __VA_ARGS__); \
561 cfs_spin_unlock(&(lock)); \
563 selftest_wait_events(); \
565 cfs_spin_lock(&(lock)); \
570 srpc_wait_service_shutdown (srpc_service_t *sv)
574 cfs_spin_lock(&sv->sv_lock);
575 LASSERT (sv->sv_shuttingdown);
576 cfs_spin_unlock(&sv->sv_lock);
578 while (srpc_finish_service(sv) == 0) {
580 CDEBUG (((i & -i) == i) ? D_WARNING : D_NET,
581 "Waiting for %s service to shutdown...\n",
583 selftest_wait_events();
587 #endif /* __SELFTEST_SELFTEST_H__ */