4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
21 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
22 * CA 95054 USA or visit www.sun.com if you need additional information or
28 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
29 * Use is subject to license terms.
31 * Copyright (c) 2012, Intel Corporation.
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__
46 #include "../../include/linux/libcfs/libcfs.h"
47 #include "../../include/linux/lnet/lnet.h"
48 #include "../../include/linux/lnet/lib-lnet.h"
49 #include "../../include/linux/lnet/lib-types.h"
50 #include "../../include/linux/lnet/lnetst.h"
55 #ifndef MADE_WITHOUT_COMPROMISE
56 #define MADE_WITHOUT_COMPROMISE
59 #define SWI_STATE_NEWBORN 0
60 #define SWI_STATE_REPLY_SUBMITTED 1
61 #define SWI_STATE_REPLY_SENT 2
62 #define SWI_STATE_REQUEST_SUBMITTED 3
63 #define SWI_STATE_REQUEST_SENT 4
64 #define SWI_STATE_REPLY_RECEIVED 5
65 #define SWI_STATE_BULK_STARTED 6
66 #define SWI_STATE_DONE 10
70 struct srpc_service_cd;
72 struct sfw_test_instance;
74 /* services below SRPC_FRAMEWORK_SERVICE_MAX_ID are framework
75 * services, e.g. create/modify session.
77 #define SRPC_SERVICE_DEBUG 0
78 #define SRPC_SERVICE_MAKE_SESSION 1
79 #define SRPC_SERVICE_REMOVE_SESSION 2
80 #define SRPC_SERVICE_BATCH 3
81 #define SRPC_SERVICE_TEST 4
82 #define SRPC_SERVICE_QUERY_STAT 5
83 #define SRPC_SERVICE_JOIN 6
84 #define SRPC_FRAMEWORK_SERVICE_MAX_ID 10
85 /* other services start from SRPC_FRAMEWORK_SERVICE_MAX_ID+1 */
86 #define SRPC_SERVICE_BRW 11
87 #define SRPC_SERVICE_PING 12
88 #define SRPC_SERVICE_MAX_ID 12
90 #define SRPC_REQUEST_PORTAL 50
91 /* a lazy portal for framework RPC requests */
92 #define SRPC_FRAMEWORK_REQUEST_PORTAL 51
93 /* all reply/bulk RDMAs go to this portal */
94 #define SRPC_RDMA_PORTAL 52
96 static inline enum srpc_msg_type
97 srpc_service2request(int service)
102 case SRPC_SERVICE_DEBUG:
103 return SRPC_MSG_DEBUG_REQST;
105 case SRPC_SERVICE_MAKE_SESSION:
106 return SRPC_MSG_MKSN_REQST;
108 case SRPC_SERVICE_REMOVE_SESSION:
109 return SRPC_MSG_RMSN_REQST;
111 case SRPC_SERVICE_BATCH:
112 return SRPC_MSG_BATCH_REQST;
114 case SRPC_SERVICE_TEST:
115 return SRPC_MSG_TEST_REQST;
117 case SRPC_SERVICE_QUERY_STAT:
118 return SRPC_MSG_STAT_REQST;
120 case SRPC_SERVICE_BRW:
121 return SRPC_MSG_BRW_REQST;
123 case SRPC_SERVICE_PING:
124 return SRPC_MSG_PING_REQST;
126 case SRPC_SERVICE_JOIN:
127 return SRPC_MSG_JOIN_REQST;
131 static inline enum srpc_msg_type
132 srpc_service2reply(int service)
134 return srpc_service2request(service) + 1;
137 enum srpc_event_type {
138 SRPC_BULK_REQ_RCVD = 1, /* passive bulk request(PUT sink/GET source)
140 SRPC_BULK_PUT_SENT = 2, /* active bulk PUT sent (source) */
141 SRPC_BULK_GET_RPLD = 3, /* active bulk GET replied (sink) */
142 SRPC_REPLY_RCVD = 4, /* incoming reply received */
143 SRPC_REPLY_SENT = 5, /* outgoing reply sent */
144 SRPC_REQUEST_RCVD = 6, /* incoming request received */
145 SRPC_REQUEST_SENT = 7, /* outgoing request sent */
150 enum srpc_event_type ev_type; /* what's up */
151 lnet_event_kind_t ev_lnet; /* LNet event type */
152 int ev_fired; /* LNet event fired? */
153 int ev_status; /* LNet event status */
154 void *ev_data; /* owning server/client RPC */
157 /* bulk descriptor */
159 int bk_len; /* len of bulk data */
160 lnet_handle_md_t bk_mdh;
161 int bk_sink; /* sink/source */
162 int bk_niov; /* # iov in bk_iovs */
163 lnet_kiov_t bk_iovs[0];
166 /* message buffer descriptor */
168 struct list_head buf_list; /* chain on srpc_service::*_msgq */
169 struct srpc_msg buf_msg;
170 lnet_handle_md_t buf_mdh;
172 lnet_process_id_t buf_peer;
176 typedef int (*swi_action_t) (struct swi_workitem *);
178 struct swi_workitem {
179 struct cfs_wi_sched *swi_sched;
180 struct cfs_workitem swi_workitem;
181 swi_action_t swi_action;
185 /* server-side state of a RPC */
186 struct srpc_server_rpc {
187 /* chain on srpc_service::*_rpcq */
188 struct list_head srpc_list;
189 struct srpc_service_cd *srpc_scd;
190 struct swi_workitem srpc_wi;
191 struct srpc_event srpc_ev; /* bulk/reply event */
192 lnet_nid_t srpc_self;
193 lnet_process_id_t srpc_peer;
194 struct srpc_msg srpc_replymsg;
195 lnet_handle_md_t srpc_replymdh;
196 struct srpc_buffer *srpc_reqstbuf;
197 struct srpc_bulk *srpc_bulk;
199 unsigned int srpc_aborted; /* being given up */
201 void (*srpc_done)(struct srpc_server_rpc *);
204 /* client-side state of a RPC */
205 struct srpc_client_rpc {
206 struct list_head crpc_list; /* chain on user's lists */
207 spinlock_t crpc_lock; /* serialize */
209 atomic_t crpc_refcount;
210 int crpc_timeout; /* # seconds to wait for reply */
211 struct stt_timer crpc_timer;
212 struct swi_workitem crpc_wi;
213 lnet_process_id_t crpc_dest;
215 void (*crpc_done)(struct srpc_client_rpc *);
216 void (*crpc_fini)(struct srpc_client_rpc *);
217 int crpc_status; /* completion status */
218 void *crpc_priv; /* caller data */
221 unsigned int crpc_aborted:1; /* being given up */
222 unsigned int crpc_closed:1; /* completed */
225 struct srpc_event crpc_bulkev; /* bulk event */
226 struct srpc_event crpc_reqstev; /* request event */
227 struct srpc_event crpc_replyev; /* reply event */
229 /* bulk, request(reqst), and reply exchanged on wire */
230 struct srpc_msg crpc_reqstmsg;
231 struct srpc_msg crpc_replymsg;
232 lnet_handle_md_t crpc_reqstmdh;
233 lnet_handle_md_t crpc_replymdh;
234 struct srpc_bulk crpc_bulk;
237 #define srpc_client_rpc_size(rpc) \
238 offsetof(struct srpc_client_rpc, crpc_bulk.bk_iovs[(rpc)->crpc_bulk.bk_niov])
240 #define srpc_client_rpc_addref(rpc) \
242 CDEBUG(D_NET, "RPC[%p] -> %s (%d)++\n", \
243 (rpc), libcfs_id2str((rpc)->crpc_dest), \
244 atomic_read(&(rpc)->crpc_refcount)); \
245 LASSERT(atomic_read(&(rpc)->crpc_refcount) > 0); \
246 atomic_inc(&(rpc)->crpc_refcount); \
249 #define srpc_client_rpc_decref(rpc) \
251 CDEBUG(D_NET, "RPC[%p] -> %s (%d)--\n", \
252 (rpc), libcfs_id2str((rpc)->crpc_dest), \
253 atomic_read(&(rpc)->crpc_refcount)); \
254 LASSERT(atomic_read(&(rpc)->crpc_refcount) > 0); \
255 if (atomic_dec_and_test(&(rpc)->crpc_refcount)) \
256 srpc_destroy_client_rpc(rpc); \
259 #define srpc_event_pending(rpc) (!(rpc)->crpc_bulkev.ev_fired || \
260 !(rpc)->crpc_reqstev.ev_fired || \
261 !(rpc)->crpc_replyev.ev_fired)
263 /* CPU partition data of srpc service */
264 struct srpc_service_cd {
267 /** backref to service */
268 struct srpc_service *scd_svc;
270 struct srpc_event scd_ev;
271 /** free RPC descriptors */
272 struct list_head scd_rpc_free;
273 /** in-flight RPCs */
274 struct list_head scd_rpc_active;
275 /** workitem for posting buffer */
276 struct swi_workitem scd_buf_wi;
279 /** error code for scd_buf_wi */
281 /** timestamp for scd_buf_err */
282 time64_t scd_buf_err_stamp;
283 /** total # request buffers */
285 /** # posted request buffers */
287 /** in progress of buffer posting */
289 /** allocate more buffers if scd_buf_nposted < scd_buf_low */
291 /** increase/decrease some buffers */
293 /** posted message buffers */
294 struct list_head scd_buf_posted;
295 /** blocked for RPC descriptor */
296 struct list_head scd_buf_blocked;
299 /* number of server workitems (mini-thread) for testing service */
300 #define SFW_TEST_WI_MIN 256
301 #define SFW_TEST_WI_MAX 2048
302 /* extra buffers for tolerating buggy peers, or unbalanced number
303 * of peers between partitions */
304 #define SFW_TEST_WI_EXTRA 64
306 /* number of server workitems (mini-thread) for framework service */
307 #define SFW_FRWK_WI_MIN 16
308 #define SFW_FRWK_WI_MAX 256
310 struct srpc_service {
311 int sv_id; /* service id */
312 const char *sv_name; /* human readable name */
313 int sv_wi_total; /* total server workitems */
316 /* percpt data for srpc_service */
317 struct srpc_service_cd **sv_cpt_data;
318 /* Service callbacks:
319 * - sv_handler: process incoming RPC request
320 * - sv_bulk_ready: notify bulk data
322 int (*sv_handler)(struct srpc_server_rpc *);
323 int (*sv_bulk_ready)(struct srpc_server_rpc *, int);
327 struct list_head sn_list; /* chain on fw_zombie_sessions */
328 lst_sid_t sn_id; /* unique identifier */
329 unsigned int sn_timeout; /* # seconds' inactivity to expire */
331 unsigned int sn_features;
332 struct stt_timer sn_timer;
333 struct list_head sn_batches; /* list of batches */
334 char sn_name[LST_NAME_SIZE];
335 atomic_t sn_refcount;
336 atomic_t sn_brw_errors;
337 atomic_t sn_ping_errors;
338 unsigned long sn_started;
341 #define sfw_sid_equal(sid0, sid1) ((sid0).ses_nid == (sid1).ses_nid && \
342 (sid0).ses_stamp == (sid1).ses_stamp)
345 struct list_head bat_list; /* chain on sn_batches */
346 lst_bid_t bat_id; /* batch id */
347 int bat_error; /* error code of batch */
348 struct sfw_session *bat_session; /* batch's session */
349 atomic_t bat_nactive; /* # of active tests */
350 struct list_head bat_tests; /* test instances */
353 struct sfw_test_client_ops {
354 int (*tso_init)(struct sfw_test_instance *tsi); /* initialize test
356 void (*tso_fini)(struct sfw_test_instance *tsi); /* finalize test
358 int (*tso_prep_rpc)(struct sfw_test_unit *tsu,
359 lnet_process_id_t dest,
360 struct srpc_client_rpc **rpc); /* prep a tests rpc */
361 void (*tso_done_rpc)(struct sfw_test_unit *tsu,
362 struct srpc_client_rpc *rpc); /* done a test rpc */
365 struct sfw_test_instance {
366 struct list_head tsi_list; /* chain on batch */
367 int tsi_service; /* test type */
368 struct sfw_batch *tsi_batch; /* batch */
369 struct sfw_test_client_ops *tsi_ops; /* test client operation
372 /* public parameter for all test units */
373 unsigned int tsi_is_client:1; /* is test client */
374 unsigned int tsi_stoptsu_onerr:1; /* stop tsu on error */
375 int tsi_concur; /* concurrency */
376 int tsi_loop; /* loop count */
378 /* status of test instance */
379 spinlock_t tsi_lock; /* serialize */
380 unsigned int tsi_stopping:1; /* test is stopping */
381 atomic_t tsi_nactive; /* # of active test
383 struct list_head tsi_units; /* test units */
384 struct list_head tsi_free_rpcs; /* free rpcs */
385 struct list_head tsi_active_rpcs; /* active rpcs */
388 struct test_ping_req ping; /* ping parameter */
389 struct test_bulk_req bulk_v0; /* bulk parameter */
390 struct test_bulk_req_v1 bulk_v1; /* bulk v1 parameter */
394 /* XXX: trailing (PAGE_SIZE % sizeof(lnet_process_id_t)) bytes at the end of
395 * pages are not used */
396 #define SFW_MAX_CONCUR LST_MAX_CONCUR
397 #define SFW_ID_PER_PAGE (PAGE_SIZE / sizeof(lnet_process_id_packed_t))
398 #define SFW_MAX_NDESTS (LNET_MAX_IOV * SFW_ID_PER_PAGE)
399 #define sfw_id_pages(n) (((n) + SFW_ID_PER_PAGE - 1) / SFW_ID_PER_PAGE)
401 struct sfw_test_unit {
402 struct list_head tsu_list; /* chain on lst_test_instance */
403 lnet_process_id_t tsu_dest; /* id of dest node */
404 int tsu_loop; /* loop count of the test */
405 struct sfw_test_instance *tsu_instance; /* pointer to test instance */
406 void *tsu_private; /* private data */
407 struct swi_workitem tsu_worker; /* workitem of the test unit */
410 struct sfw_test_case {
411 struct list_head tsc_list; /* chain on fw_tests */
412 struct srpc_service *tsc_srv_service; /* test service */
413 struct sfw_test_client_ops *tsc_cli_ops; /* ops of test client */
416 struct srpc_client_rpc *
417 sfw_create_rpc(lnet_process_id_t peer, int service,
418 unsigned features, int nbulkiov, int bulklen,
419 void (*done)(struct srpc_client_rpc *), void *priv);
420 int sfw_create_test_rpc(struct sfw_test_unit *tsu,
421 lnet_process_id_t peer, unsigned features,
422 int nblk, int blklen, struct srpc_client_rpc **rpc);
423 void sfw_abort_rpc(struct srpc_client_rpc *rpc);
424 void sfw_post_rpc(struct srpc_client_rpc *rpc);
425 void sfw_client_rpc_done(struct srpc_client_rpc *rpc);
426 void sfw_unpack_message(struct srpc_msg *msg);
427 void sfw_free_pages(struct srpc_server_rpc *rpc);
428 void sfw_add_bulk_page(struct srpc_bulk *bk, struct page *pg, int i);
429 int sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
431 int sfw_make_session(struct srpc_mksn_reqst *request,
432 struct srpc_mksn_reply *reply);
434 struct srpc_client_rpc *
435 srpc_create_client_rpc(lnet_process_id_t peer, int service,
436 int nbulkiov, int bulklen,
437 void (*rpc_done)(struct srpc_client_rpc *),
438 void (*rpc_fini)(struct srpc_client_rpc *), void *priv);
439 void srpc_post_rpc(struct srpc_client_rpc *rpc);
440 void srpc_abort_rpc(struct srpc_client_rpc *rpc, int why);
441 void srpc_free_bulk(struct srpc_bulk *bk);
442 struct srpc_bulk *srpc_alloc_bulk(int cpt, unsigned bulk_npg,
443 unsigned bulk_len, int sink);
444 int srpc_send_rpc(struct swi_workitem *wi);
445 int srpc_send_reply(struct srpc_server_rpc *rpc);
446 int srpc_add_service(struct srpc_service *sv);
447 int srpc_remove_service(struct srpc_service *sv);
448 void srpc_shutdown_service(struct srpc_service *sv);
449 void srpc_abort_service(struct srpc_service *sv);
450 int srpc_finish_service(struct srpc_service *sv);
451 int srpc_service_add_buffers(struct srpc_service *sv, int nbuffer);
452 void srpc_service_remove_buffers(struct srpc_service *sv, int nbuffer);
453 void srpc_get_counters(srpc_counters_t *cnt);
454 void srpc_set_counters(const srpc_counters_t *cnt);
456 extern struct cfs_wi_sched *lst_sched_serial;
457 extern struct cfs_wi_sched **lst_sched_test;
460 srpc_serv_is_framework(struct srpc_service *svc)
462 return svc->sv_id < SRPC_FRAMEWORK_SERVICE_MAX_ID;
466 swi_wi_action(struct cfs_workitem *wi)
468 struct swi_workitem *swi;
470 swi = container_of(wi, struct swi_workitem, swi_workitem);
472 return swi->swi_action(swi);
476 swi_init_workitem(struct swi_workitem *swi, void *data,
477 swi_action_t action, struct cfs_wi_sched *sched)
479 swi->swi_sched = sched;
480 swi->swi_action = action;
481 swi->swi_state = SWI_STATE_NEWBORN;
482 cfs_wi_init(&swi->swi_workitem, data, swi_wi_action);
486 swi_schedule_workitem(struct swi_workitem *wi)
488 cfs_wi_schedule(wi->swi_sched, &wi->swi_workitem);
492 swi_exit_workitem(struct swi_workitem *swi)
494 cfs_wi_exit(swi->swi_sched, &swi->swi_workitem);
498 swi_deschedule_workitem(struct swi_workitem *swi)
500 return cfs_wi_deschedule(swi->swi_sched, &swi->swi_workitem);
503 int sfw_startup(void);
504 int srpc_startup(void);
505 void sfw_shutdown(void);
506 void srpc_shutdown(void);
509 srpc_destroy_client_rpc(struct srpc_client_rpc *rpc)
512 LASSERT(!srpc_event_pending(rpc));
513 LASSERT(!atomic_read(&rpc->crpc_refcount));
516 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
518 (*rpc->crpc_fini)(rpc);
522 srpc_init_client_rpc(struct srpc_client_rpc *rpc, lnet_process_id_t peer,
523 int service, int nbulkiov, int bulklen,
524 void (*rpc_done)(struct srpc_client_rpc *),
525 void (*rpc_fini)(struct srpc_client_rpc *), void *priv)
527 LASSERT(nbulkiov <= LNET_MAX_IOV);
529 memset(rpc, 0, offsetof(struct srpc_client_rpc,
530 crpc_bulk.bk_iovs[nbulkiov]));
532 INIT_LIST_HEAD(&rpc->crpc_list);
533 swi_init_workitem(&rpc->crpc_wi, rpc, srpc_send_rpc,
534 lst_sched_test[lnet_cpt_of_nid(peer.nid)]);
535 spin_lock_init(&rpc->crpc_lock);
536 atomic_set(&rpc->crpc_refcount, 1); /* 1 ref for caller */
538 rpc->crpc_dest = peer;
539 rpc->crpc_priv = priv;
540 rpc->crpc_service = service;
541 rpc->crpc_bulk.bk_len = bulklen;
542 rpc->crpc_bulk.bk_niov = nbulkiov;
543 rpc->crpc_done = rpc_done;
544 rpc->crpc_fini = rpc_fini;
545 LNetInvalidateHandle(&rpc->crpc_reqstmdh);
546 LNetInvalidateHandle(&rpc->crpc_replymdh);
547 LNetInvalidateHandle(&rpc->crpc_bulk.bk_mdh);
549 /* no event is expected at this point */
550 rpc->crpc_bulkev.ev_fired = 1;
551 rpc->crpc_reqstev.ev_fired = 1;
552 rpc->crpc_replyev.ev_fired = 1;
554 rpc->crpc_reqstmsg.msg_magic = SRPC_MSG_MAGIC;
555 rpc->crpc_reqstmsg.msg_version = SRPC_MSG_VERSION;
556 rpc->crpc_reqstmsg.msg_type = srpc_service2request(service);
559 static inline const char *
560 swi_state2str(int state)
562 #define STATE2STR(x) case x: return #x
566 STATE2STR(SWI_STATE_NEWBORN);
567 STATE2STR(SWI_STATE_REPLY_SUBMITTED);
568 STATE2STR(SWI_STATE_REPLY_SENT);
569 STATE2STR(SWI_STATE_REQUEST_SUBMITTED);
570 STATE2STR(SWI_STATE_REQUEST_SENT);
571 STATE2STR(SWI_STATE_REPLY_RECEIVED);
572 STATE2STR(SWI_STATE_BULK_STARTED);
573 STATE2STR(SWI_STATE_DONE);
578 #define selftest_wait_events() \
580 set_current_state(TASK_UNINTERRUPTIBLE); \
581 schedule_timeout(cfs_time_seconds(1) / 10); \
584 #define lst_wait_until(cond, lock, fmt, ...) \
588 CDEBUG(is_power_of_2(++__I) ? D_WARNING : D_NET, \
589 fmt, ## __VA_ARGS__); \
590 spin_unlock(&(lock)); \
592 selftest_wait_events(); \
594 spin_lock(&(lock)); \
599 srpc_wait_service_shutdown(struct srpc_service *sv)
603 LASSERT(sv->sv_shuttingdown);
605 while (!srpc_finish_service(sv)) {
607 CDEBUG(((i & -i) == i) ? D_WARNING : D_NET,
608 "Waiting for %s service to shutdown...\n",
610 selftest_wait_events();
614 extern struct sfw_test_client_ops brw_test_client;
615 void brw_init_test_client(void);
617 extern struct srpc_service brw_test_service;
618 void brw_init_test_service(void);
620 extern struct sfw_test_client_ops ping_test_client;
621 void ping_init_test_client(void);
623 extern struct srpc_service ping_test_service;
624 void ping_init_test_service(void);
626 #endif /* __SELFTEST_SELFTEST_H__ */