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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lnet/selftest/selftest.h
34 * Author: Isaac Huang <isaac@clusterfs.com>
36 #ifndef __SELFTEST_SELFTEST_H__
37 #define __SELFTEST_SELFTEST_H__
41 #include <libcfs/libcfs.h>
43 #include <lnet/lib-lnet.h>
44 #include <lnet/lib-types.h>
45 #include <uapi/linux/lnet/lnetst.h>
50 #ifndef MADE_WITHOUT_COMPROMISE
51 #define MADE_WITHOUT_COMPROMISE
55 #define SWI_STATE_NEWBORN 0
56 #define SWI_STATE_REPLY_SUBMITTED 1
57 #define SWI_STATE_REPLY_SENT 2
58 #define SWI_STATE_REQUEST_SUBMITTED 3
59 #define SWI_STATE_REQUEST_SENT 4
60 #define SWI_STATE_REPLY_RECEIVED 5
61 #define SWI_STATE_BULK_STARTED 6
62 #define SWI_STATE_DONE 10
66 struct srpc_service_cd;
68 struct sfw_test_instance;
70 /* services below SRPC_FRAMEWORK_SERVICE_MAX_ID are framework
71 * services, e.g. create/modify session.
73 #define SRPC_SERVICE_DEBUG 0
74 #define SRPC_SERVICE_MAKE_SESSION 1
75 #define SRPC_SERVICE_REMOVE_SESSION 2
76 #define SRPC_SERVICE_BATCH 3
77 #define SRPC_SERVICE_TEST 4
78 #define SRPC_SERVICE_QUERY_STAT 5
79 #define SRPC_SERVICE_JOIN 6
80 #define SRPC_FRAMEWORK_SERVICE_MAX_ID 10
81 /* other services start from SRPC_FRAMEWORK_SERVICE_MAX_ID+1 */
82 #define SRPC_SERVICE_BRW 11
83 #define SRPC_SERVICE_PING 12
84 #define SRPC_SERVICE_MAX_ID 12
86 #define SRPC_REQUEST_PORTAL 50
87 /* a lazy portal for framework RPC requests */
88 #define SRPC_FRAMEWORK_REQUEST_PORTAL 51
89 /* all reply/bulk RDMAs go to this portal */
90 #define SRPC_RDMA_PORTAL 52
92 static inline enum srpc_msg_type
93 srpc_service2request (int service)
98 case SRPC_SERVICE_DEBUG:
99 return SRPC_MSG_DEBUG_REQST;
101 case SRPC_SERVICE_MAKE_SESSION:
102 return SRPC_MSG_MKSN_REQST;
104 case SRPC_SERVICE_REMOVE_SESSION:
105 return SRPC_MSG_RMSN_REQST;
107 case SRPC_SERVICE_BATCH:
108 return SRPC_MSG_BATCH_REQST;
110 case SRPC_SERVICE_TEST:
111 return SRPC_MSG_TEST_REQST;
113 case SRPC_SERVICE_QUERY_STAT:
114 return SRPC_MSG_STAT_REQST;
116 case SRPC_SERVICE_BRW:
117 return SRPC_MSG_BRW_REQST;
119 case SRPC_SERVICE_PING:
120 return SRPC_MSG_PING_REQST;
122 case SRPC_SERVICE_JOIN:
123 return SRPC_MSG_JOIN_REQST;
127 static inline enum srpc_msg_type
128 srpc_service2reply (int service)
130 return srpc_service2request(service) + 1;
133 enum srpc_event_type {
134 SRPC_BULK_REQ_RCVD = 1, /* passive bulk request(PUT sink/GET source) received */
135 SRPC_BULK_PUT_SENT = 2, /* active bulk PUT sent (source) */
136 SRPC_BULK_GET_RPLD = 3, /* active bulk GET replied (sink) */
137 SRPC_REPLY_RCVD = 4, /* incoming reply received */
138 SRPC_REPLY_SENT = 5, /* outgoing reply sent */
139 SRPC_REQUEST_RCVD = 6, /* incoming request received */
140 SRPC_REQUEST_SENT = 7, /* outgoing request sent */
145 enum srpc_event_type ev_type; /* what's up */
146 enum lnet_event_kind ev_lnet; /* LNet event type */
147 int ev_fired; /* LNet event fired? */
148 int ev_status; /* LNet event status */
149 void *ev_data; /* owning server/client RPC */
152 /* bulk descriptor */
154 int bk_len; /* len of bulk data */
155 struct lnet_handle_md bk_mdh;
156 int bk_sink; /* sink/source */
157 int bk_niov; /* # iov in bk_iovs */
158 struct bio_vec bk_iovs[0];
161 /* message buffer descriptor */
163 struct list_head buf_list; /* chain on srpc_service::*_msgq */
164 struct srpc_msg buf_msg;
165 struct lnet_handle_md buf_mdh;
167 struct lnet_process_id buf_peer;
171 typedef int (*swi_action_t)(struct swi_workitem *);
173 struct swi_workitem {
174 struct cfs_wi_sched *swi_sched;
175 struct cfs_workitem swi_workitem;
176 swi_action_t swi_action;
180 /* server-side state of a RPC */
181 struct srpc_server_rpc {
182 /* chain on srpc_service::*_rpcq */
183 struct list_head srpc_list;
184 struct srpc_service_cd *srpc_scd;
185 struct swi_workitem srpc_wi;
186 struct srpc_event srpc_ev; /* bulk/reply event */
187 lnet_nid_t srpc_self;
188 struct lnet_process_id srpc_peer;
189 struct srpc_msg srpc_replymsg;
190 struct lnet_handle_md srpc_replymdh;
191 struct srpc_buffer *srpc_reqstbuf;
192 struct srpc_bulk *srpc_bulk;
194 unsigned int srpc_aborted; /* being given up */
196 void (*srpc_done)(struct srpc_server_rpc *);
199 /* client-side state of a RPC */
200 struct srpc_client_rpc {
201 struct list_head crpc_list; /* chain on user's lists */
202 spinlock_t crpc_lock; /* serialize */
204 atomic_t crpc_refcount;
205 /* # seconds to wait for reply */
207 struct stt_timer crpc_timer;
208 struct swi_workitem crpc_wi;
209 struct lnet_process_id crpc_dest;
211 void (*crpc_done)(struct srpc_client_rpc *);
212 void (*crpc_fini)(struct srpc_client_rpc *);
213 int crpc_status; /* completion status */
214 void *crpc_priv; /* caller data */
217 unsigned int crpc_aborted:1; /* being given up */
218 unsigned int crpc_closed:1; /* completed */
221 struct srpc_event crpc_bulkev; /* bulk event */
222 struct srpc_event crpc_reqstev; /* request event */
223 struct srpc_event crpc_replyev; /* reply event */
225 /* bulk, request(reqst), and reply exchanged on wire */
226 struct srpc_msg crpc_reqstmsg;
227 struct srpc_msg crpc_replymsg;
228 struct lnet_handle_md crpc_reqstmdh;
229 struct lnet_handle_md crpc_replymdh;
230 struct srpc_bulk crpc_bulk;
233 #define srpc_client_rpc_size(rpc) \
234 offsetof(struct srpc_client_rpc, crpc_bulk.bk_iovs[(rpc)->crpc_bulk.bk_niov])
236 #define srpc_client_rpc_addref(rpc) \
238 CDEBUG(D_NET, "RPC[%p] -> %s (%d)++\n", \
239 (rpc), libcfs_id2str((rpc)->crpc_dest), \
240 atomic_read(&(rpc)->crpc_refcount)); \
241 LASSERT(atomic_read(&(rpc)->crpc_refcount) > 0); \
242 atomic_inc(&(rpc)->crpc_refcount); \
245 #define srpc_client_rpc_decref(rpc) \
247 CDEBUG(D_NET, "RPC[%p] -> %s (%d)--\n", \
248 (rpc), libcfs_id2str((rpc)->crpc_dest), \
249 atomic_read(&(rpc)->crpc_refcount)); \
250 LASSERT(atomic_read(&(rpc)->crpc_refcount) > 0); \
251 if (atomic_dec_and_test(&(rpc)->crpc_refcount)) \
252 srpc_destroy_client_rpc(rpc); \
255 #define srpc_event_pending(rpc) ((rpc)->crpc_bulkev.ev_fired == 0 || \
256 (rpc)->crpc_reqstev.ev_fired == 0 || \
257 (rpc)->crpc_replyev.ev_fired == 0)
259 /* CPU partition data of srpc service */
260 struct srpc_service_cd {
263 /** backref to service */
264 struct srpc_service *scd_svc;
266 struct srpc_event scd_ev;
267 /** free RPC descriptors */
268 struct list_head scd_rpc_free;
269 /** in-flight RPCs */
270 struct list_head scd_rpc_active;
271 /** workitem for posting buffer */
272 struct swi_workitem scd_buf_wi;
275 /** error code for scd_buf_wi */
277 /** timestamp for scd_buf_err */
278 time64_t scd_buf_err_stamp;
279 /** total # request buffers */
281 /** # posted request buffers */
283 /** in progress of buffer posting */
285 /** allocate more buffers if scd_buf_nposted < scd_buf_low */
287 /** increase/decrease some buffers */
289 /** posted message buffers */
290 struct list_head scd_buf_posted;
291 /** blocked for RPC descriptor */
292 struct list_head scd_buf_blocked;
295 /* number of server workitems (mini-thread) for testing service */
296 #define SFW_TEST_WI_MIN 256
297 #define SFW_TEST_WI_MAX 2048
298 /* extra buffers for tolerating buggy peers, or unbalanced number
299 * of peers between partitions */
300 #define SFW_TEST_WI_EXTRA 64
302 /* number of server workitems (mini-thread) for framework service */
303 #define SFW_FRWK_WI_MIN 16
304 #define SFW_FRWK_WI_MAX 256
306 struct srpc_service {
307 int sv_id; /* service id */
308 const char *sv_name; /* human readable name */
309 int sv_wi_total; /* total server workitems */
312 /* percpt data for srpc_service */
313 struct srpc_service_cd **sv_cpt_data;
314 /* Service callbacks:
315 * - sv_handler: process incoming RPC request
316 * - sv_bulk_ready: notify bulk data
318 int (*sv_handler)(struct srpc_server_rpc *);
319 int (*sv_bulk_ready)(struct srpc_server_rpc *, int);
323 /* chain on fw_zombie_sessions */
324 struct list_head sn_list;
325 struct lst_sid sn_id; /* unique identifier */
326 /* # seconds' inactivity to expire */
327 unsigned int sn_timeout;
329 unsigned int sn_features;
330 struct stt_timer sn_timer;
331 struct list_head sn_batches; /* list of batches */
332 char sn_name[LST_NAME_SIZE];
333 atomic_t sn_refcount;
334 atomic_t sn_brw_errors;
335 atomic_t sn_ping_errors;
339 #define sfw_sid_equal(sid0, sid1) ((sid0).ses_nid == (sid1).ses_nid && \
340 (sid0).ses_stamp == (sid1).ses_stamp)
343 struct list_head bat_list; /* chain on sn_batches */
344 struct lst_bid bat_id; /* batch id */
345 int bat_error; /* error code of batch */
346 struct sfw_session *bat_session; /* batch's session */
347 atomic_t bat_nactive; /* # of active tests */
348 struct list_head bat_tests; /* test instances */
351 struct sfw_test_client_ops {
352 int (*tso_init)(struct sfw_test_instance *tsi); /* intailize test client */
353 void (*tso_fini)(struct sfw_test_instance *tsi); /* finalize test client */
354 int (*tso_prep_rpc)(struct sfw_test_unit *tsu,
355 struct lnet_process_id dest,
356 struct srpc_client_rpc **rpc); /* prep a tests rpc */
357 void (*tso_done_rpc)(struct sfw_test_unit *tsu,
358 struct srpc_client_rpc *rpc); /* done a test rpc */
361 struct sfw_test_instance {
362 struct list_head tsi_list; /* chain on batch */
363 int tsi_service; /* test type */
364 struct sfw_batch *tsi_batch; /* batch */
365 struct sfw_test_client_ops *tsi_ops; /* test client operations */
367 /* public parameter for all test units */
368 unsigned int tsi_is_client:1; /* is test client */
369 unsigned int tsi_stoptsu_onerr:1; /* stop tsu on error */
370 int tsi_concur; /* concurrency */
371 int tsi_loop; /* loop count */
373 /* status of test instance */
374 spinlock_t tsi_lock; /* serialize */
375 unsigned int tsi_stopping:1; /* test is stopping */
376 atomic_t tsi_nactive; /* # of active test unit */
377 struct list_head tsi_units; /* test units */
378 struct list_head tsi_free_rpcs; /* free rpcs */
379 struct list_head tsi_active_rpcs;/* active rpcs */
382 struct test_ping_req ping; /* ping parameter */
383 struct test_bulk_req bulk_v0; /* bulk parameter */
384 struct test_bulk_req_v1 bulk_v1; /* bulk v1 parameter */
388 /* XXX: trailing (PAGE_SIZE % sizeof(struct lnet_process_id)) bytes at
389 * the end of pages are not used */
390 #define SFW_MAX_CONCUR LST_MAX_CONCUR
391 #define SFW_ID_PER_PAGE (PAGE_SIZE / sizeof(struct lnet_process_id_packed))
392 #define SFW_MAX_NDESTS (LNET_MAX_IOV * SFW_ID_PER_PAGE)
393 #define sfw_id_pages(n) (((n) + SFW_ID_PER_PAGE - 1) / SFW_ID_PER_PAGE)
395 struct sfw_test_unit {
396 struct list_head tsu_list; /* chain on lst_test_instance */
397 struct lnet_process_id tsu_dest; /* id of dest node */
398 int tsu_loop; /* loop count of the test */
399 struct sfw_test_instance *tsu_instance; /* pointer to test instance */
400 void *tsu_private; /* private data */
401 struct swi_workitem tsu_worker; /* workitem of the test unit */
404 struct sfw_test_case {
405 struct list_head tsc_list; /* chain on fw_tests */
406 struct srpc_service *tsc_srv_service; /* test service */
407 struct sfw_test_client_ops *tsc_cli_ops; /* ops of test client */
410 struct srpc_client_rpc *
411 sfw_create_rpc(struct lnet_process_id peer, int service,
412 unsigned features, int nbulkiov, int bulklen,
413 void (*done)(struct srpc_client_rpc *), void *priv);
414 int sfw_create_test_rpc(struct sfw_test_unit *tsu,
415 struct lnet_process_id peer, unsigned int features,
416 int nblk, int blklen, struct srpc_client_rpc **rpc);
417 void sfw_abort_rpc(struct srpc_client_rpc *rpc);
418 void sfw_post_rpc(struct srpc_client_rpc *rpc);
419 void sfw_client_rpc_done(struct srpc_client_rpc *rpc);
420 void sfw_unpack_message(struct srpc_msg *msg);
421 void sfw_free_pages(struct srpc_server_rpc *rpc);
422 void sfw_add_bulk_page(struct srpc_bulk *bk, struct page *pg, int i);
423 int sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
425 int sfw_make_session(struct srpc_mksn_reqst *request,
426 struct srpc_mksn_reply *reply);
428 struct srpc_client_rpc *
429 srpc_create_client_rpc(struct lnet_process_id peer, int service,
430 int nbulkiov, int bulklen,
431 void (*rpc_done)(struct srpc_client_rpc *),
432 void (*rpc_fini)(struct srpc_client_rpc *), void *priv);
433 void srpc_post_rpc(struct srpc_client_rpc *rpc);
434 void srpc_abort_rpc(struct srpc_client_rpc *rpc, int why);
435 void srpc_free_bulk(struct srpc_bulk *bk);
436 struct srpc_bulk *srpc_alloc_bulk(int cpt, unsigned int off,
437 unsigned int bulk_npg, unsigned int bulk_len,
439 int srpc_send_rpc(struct swi_workitem *wi);
440 int srpc_send_reply(struct srpc_server_rpc *rpc);
441 int srpc_add_service(struct srpc_service *sv);
442 int srpc_remove_service(struct srpc_service *sv);
443 void srpc_shutdown_service(struct srpc_service *sv);
444 void srpc_abort_service(struct srpc_service *sv);
445 int srpc_finish_service(struct srpc_service *sv);
446 int srpc_service_add_buffers(struct srpc_service *sv, int nbuffer);
447 void srpc_service_remove_buffers(struct srpc_service *sv, int nbuffer);
448 void srpc_get_counters(struct srpc_counters *cnt);
449 void srpc_set_counters(const struct srpc_counters *cnt);
451 extern struct cfs_wi_sched *lst_sched_serial;
452 extern struct cfs_wi_sched **lst_sched_test;
455 srpc_serv_is_framework(struct srpc_service *svc)
457 return svc->sv_id < SRPC_FRAMEWORK_SERVICE_MAX_ID;
461 swi_wi_action(struct cfs_workitem *wi)
463 struct swi_workitem *swi;
465 swi = container_of(wi, struct swi_workitem, swi_workitem);
466 return swi->swi_action(swi);
470 swi_init_workitem(struct swi_workitem *swi,
471 swi_action_t action, struct cfs_wi_sched *sched)
473 swi->swi_sched = sched;
474 swi->swi_action = action;
475 swi->swi_state = SWI_STATE_NEWBORN;
476 cfs_wi_init(&swi->swi_workitem, swi_wi_action);
480 swi_schedule_workitem(struct swi_workitem *wi)
482 cfs_wi_schedule(wi->swi_sched, &wi->swi_workitem);
486 swi_exit_workitem(struct swi_workitem *swi)
488 cfs_wi_exit(swi->swi_sched, &swi->swi_workitem);
492 swi_deschedule_workitem(struct swi_workitem *swi)
494 return cfs_wi_deschedule(swi->swi_sched, &swi->swi_workitem);
497 int sfw_startup(void);
498 int srpc_startup(void);
499 void sfw_shutdown(void);
500 void srpc_shutdown(void);
503 srpc_destroy_client_rpc(struct srpc_client_rpc *rpc)
505 LASSERT (rpc != NULL);
506 LASSERT (!srpc_event_pending(rpc));
507 LASSERT (atomic_read(&rpc->crpc_refcount) == 0);
509 if (rpc->crpc_fini == NULL) {
510 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
512 (*rpc->crpc_fini) (rpc);
517 srpc_init_client_rpc(struct srpc_client_rpc *rpc, struct lnet_process_id peer,
518 int service, int nbulkiov, int bulklen,
519 void (*rpc_done)(struct srpc_client_rpc *),
520 void (*rpc_fini)(struct srpc_client_rpc *), void *priv)
522 LASSERT(nbulkiov <= LNET_MAX_IOV);
524 memset(rpc, 0, offsetof(struct srpc_client_rpc,
525 crpc_bulk.bk_iovs[nbulkiov]));
527 INIT_LIST_HEAD(&rpc->crpc_list);
528 swi_init_workitem(&rpc->crpc_wi, srpc_send_rpc,
529 lst_sched_test[lnet_cpt_of_nid(peer.nid, NULL)]);
530 spin_lock_init(&rpc->crpc_lock);
531 atomic_set(&rpc->crpc_refcount, 1); /* 1 ref for caller */
533 rpc->crpc_dest = peer;
534 rpc->crpc_priv = priv;
535 rpc->crpc_service = service;
536 rpc->crpc_bulk.bk_len = bulklen;
537 rpc->crpc_bulk.bk_niov = nbulkiov;
538 rpc->crpc_done = rpc_done;
539 rpc->crpc_fini = rpc_fini;
540 LNetInvalidateMDHandle(&rpc->crpc_reqstmdh);
541 LNetInvalidateMDHandle(&rpc->crpc_replymdh);
542 LNetInvalidateMDHandle(&rpc->crpc_bulk.bk_mdh);
544 /* no event is expected at this point */
545 rpc->crpc_bulkev.ev_fired =
546 rpc->crpc_reqstev.ev_fired =
547 rpc->crpc_replyev.ev_fired = 1;
549 rpc->crpc_reqstmsg.msg_magic = SRPC_MSG_MAGIC;
550 rpc->crpc_reqstmsg.msg_version = SRPC_MSG_VERSION;
551 rpc->crpc_reqstmsg.msg_type = srpc_service2request(service);
554 static inline const char *
555 swi_state2str (int state)
557 #define STATE2STR(x) case x: return #x
561 STATE2STR(SWI_STATE_NEWBORN);
562 STATE2STR(SWI_STATE_REPLY_SUBMITTED);
563 STATE2STR(SWI_STATE_REPLY_SENT);
564 STATE2STR(SWI_STATE_REQUEST_SUBMITTED);
565 STATE2STR(SWI_STATE_REQUEST_SENT);
566 STATE2STR(SWI_STATE_REPLY_RECEIVED);
567 STATE2STR(SWI_STATE_BULK_STARTED);
568 STATE2STR(SWI_STATE_DONE);
573 #define lst_wait_until(cond, lock, fmt, ...) \
577 CDEBUG(is_power_of_2(++__I) ? D_WARNING : D_NET, \
578 fmt, ## __VA_ARGS__); \
579 spin_unlock(&(lock)); \
581 schedule_timeout_uninterruptible( \
582 cfs_time_seconds(1) / 10); \
584 spin_lock(&(lock)); \
589 srpc_wait_service_shutdown(struct srpc_service *sv)
593 LASSERT(sv->sv_shuttingdown);
595 while (srpc_finish_service(sv) == 0) {
597 CDEBUG(((i & -i) == i) ? D_WARNING : D_NET,
598 "Waiting for %s service to shutdown...\n",
600 schedule_timeout_uninterruptible(cfs_time_seconds(1) / 10);
604 extern struct sfw_test_client_ops ping_test_client;
605 extern struct srpc_service ping_test_service;
606 void ping_init_test_client(void);
607 void ping_init_test_service(void);
609 extern struct sfw_test_client_ops brw_test_client;
610 extern struct srpc_service brw_test_service;
611 void brw_init_test_client(void);
612 void brw_init_test_service(void);
614 #endif /* __SELFTEST_SELFTEST_H__ */