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 srpc_msg_type_t
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 srpc_msg_type_t
132 srpc_service2reply(int service
)
134 return srpc_service2request(service
) + 1;
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 srpc_event_type_t 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 */
158 int bk_len
; /* len of bulk data */
159 lnet_handle_md_t bk_mdh
;
160 int bk_sink
; /* sink/source */
161 int bk_niov
; /* # iov in bk_iovs */
162 lnet_kiov_t bk_iovs
[0];
163 } srpc_bulk_t
; /* bulk descriptor */
165 /* message buffer descriptor */
166 typedef struct srpc_buffer
{
167 struct list_head buf_list
; /* chain on srpc_service::*_msgq */
169 lnet_handle_md_t buf_mdh
;
171 lnet_process_id_t buf_peer
;
175 typedef int (*swi_action_t
) (struct swi_workitem
*);
177 typedef struct swi_workitem
{
178 struct cfs_wi_sched
*swi_sched
;
179 cfs_workitem_t swi_workitem
;
180 swi_action_t swi_action
;
184 /* server-side state of a RPC */
185 struct srpc_server_rpc
{
186 /* chain on srpc_service::*_rpcq */
187 struct list_head srpc_list
;
188 struct srpc_service_cd
*srpc_scd
;
189 swi_workitem_t srpc_wi
;
190 srpc_event_t srpc_ev
; /* bulk/reply event */
191 lnet_nid_t srpc_self
;
192 lnet_process_id_t srpc_peer
;
193 srpc_msg_t srpc_replymsg
;
194 lnet_handle_md_t srpc_replymdh
;
195 srpc_buffer_t
*srpc_reqstbuf
;
196 srpc_bulk_t
*srpc_bulk
;
198 unsigned int srpc_aborted
; /* being given up */
200 void (*srpc_done
)(struct srpc_server_rpc
*);
203 /* client-side state of a RPC */
204 typedef struct srpc_client_rpc
{
205 struct list_head crpc_list
; /* chain on user's lists */
206 spinlock_t crpc_lock
; /* serialize */
208 atomic_t crpc_refcount
;
209 int crpc_timeout
; /* # seconds to wait for reply */
210 struct stt_timer crpc_timer
;
211 swi_workitem_t crpc_wi
;
212 lnet_process_id_t crpc_dest
;
214 void (*crpc_done
)(struct srpc_client_rpc
*);
215 void (*crpc_fini
)(struct srpc_client_rpc
*);
216 int crpc_status
; /* completion status */
217 void *crpc_priv
; /* caller data */
220 unsigned int crpc_aborted
:1; /* being given up */
221 unsigned int crpc_closed
:1; /* completed */
224 srpc_event_t crpc_bulkev
; /* bulk event */
225 srpc_event_t crpc_reqstev
; /* request event */
226 srpc_event_t crpc_replyev
; /* reply event */
228 /* bulk, request(reqst), and reply exchanged on wire */
229 srpc_msg_t crpc_reqstmsg
;
230 srpc_msg_t crpc_replymsg
;
231 lnet_handle_md_t crpc_reqstmdh
;
232 lnet_handle_md_t crpc_replymdh
;
233 srpc_bulk_t crpc_bulk
;
236 #define srpc_client_rpc_size(rpc) \
237 offsetof(srpc_client_rpc_t, crpc_bulk.bk_iovs[(rpc)->crpc_bulk.bk_niov])
239 #define srpc_client_rpc_addref(rpc) \
241 CDEBUG(D_NET, "RPC[%p] -> %s (%d)++\n", \
242 (rpc), libcfs_id2str((rpc)->crpc_dest), \
243 atomic_read(&(rpc)->crpc_refcount)); \
244 LASSERT(atomic_read(&(rpc)->crpc_refcount) > 0); \
245 atomic_inc(&(rpc)->crpc_refcount); \
248 #define srpc_client_rpc_decref(rpc) \
250 CDEBUG(D_NET, "RPC[%p] -> %s (%d)--\n", \
251 (rpc), libcfs_id2str((rpc)->crpc_dest), \
252 atomic_read(&(rpc)->crpc_refcount)); \
253 LASSERT(atomic_read(&(rpc)->crpc_refcount) > 0); \
254 if (atomic_dec_and_test(&(rpc)->crpc_refcount)) \
255 srpc_destroy_client_rpc(rpc); \
258 #define srpc_event_pending(rpc) (!(rpc)->crpc_bulkev.ev_fired || \
259 !(rpc)->crpc_reqstev.ev_fired || \
260 !(rpc)->crpc_replyev.ev_fired)
262 /* CPU partition data of srpc service */
263 struct srpc_service_cd
{
266 /** backref to service */
267 struct srpc_service
*scd_svc
;
270 /** free RPC descriptors */
271 struct list_head scd_rpc_free
;
272 /** in-flight RPCs */
273 struct list_head scd_rpc_active
;
274 /** workitem for posting buffer */
275 swi_workitem_t scd_buf_wi
;
278 /** error code for scd_buf_wi */
280 /** timestamp for scd_buf_err */
281 time64_t scd_buf_err_stamp
;
282 /** total # request buffers */
284 /** # posted request buffers */
286 /** in progress of buffer posting */
288 /** allocate more buffers if scd_buf_nposted < scd_buf_low */
290 /** increase/decrease some buffers */
292 /** posted message buffers */
293 struct list_head scd_buf_posted
;
294 /** blocked for RPC descriptor */
295 struct list_head scd_buf_blocked
;
298 /* number of server workitems (mini-thread) for testing service */
299 #define SFW_TEST_WI_MIN 256
300 #define SFW_TEST_WI_MAX 2048
301 /* extra buffers for tolerating buggy peers, or unbalanced number
302 * of peers between partitions */
303 #define SFW_TEST_WI_EXTRA 64
305 /* number of server workitems (mini-thread) for framework service */
306 #define SFW_FRWK_WI_MIN 16
307 #define SFW_FRWK_WI_MAX 256
309 typedef struct srpc_service
{
310 int sv_id
; /* service id */
311 const char *sv_name
; /* human readable name */
312 int sv_wi_total
; /* total server workitems */
315 /* percpt data for srpc_service */
316 struct srpc_service_cd
**sv_cpt_data
;
317 /* Service callbacks:
318 * - sv_handler: process incoming RPC request
319 * - sv_bulk_ready: notify bulk data
321 int (*sv_handler
)(struct srpc_server_rpc
*);
322 int (*sv_bulk_ready
)(struct srpc_server_rpc
*, int);
326 struct list_head sn_list
; /* chain on fw_zombie_sessions */
327 lst_sid_t sn_id
; /* unique identifier */
328 unsigned int sn_timeout
; /* # seconds' inactivity to expire */
330 unsigned int sn_features
;
331 struct stt_timer sn_timer
;
332 struct list_head sn_batches
; /* list of batches */
333 char sn_name
[LST_NAME_SIZE
];
334 atomic_t sn_refcount
;
335 atomic_t sn_brw_errors
;
336 atomic_t sn_ping_errors
;
337 unsigned long sn_started
;
340 #define sfw_sid_equal(sid0, sid1) ((sid0).ses_nid == (sid1).ses_nid && \
341 (sid0).ses_stamp == (sid1).ses_stamp)
344 struct list_head bat_list
; /* chain on sn_batches */
345 lst_bid_t bat_id
; /* batch id */
346 int bat_error
; /* error code of batch */
347 sfw_session_t
*bat_session
; /* batch's session */
348 atomic_t bat_nactive
; /* # of active tests */
349 struct list_head bat_tests
; /* test instances */
353 int (*tso_init
)(struct sfw_test_instance
*tsi
); /* initialize test
355 void (*tso_fini
)(struct sfw_test_instance
*tsi
); /* finalize test
357 int (*tso_prep_rpc
)(struct sfw_test_unit
*tsu
,
358 lnet_process_id_t dest
,
359 srpc_client_rpc_t
**rpc
); /* prep a tests rpc */
360 void (*tso_done_rpc
)(struct sfw_test_unit
*tsu
,
361 srpc_client_rpc_t
*rpc
); /* done a test rpc */
362 } sfw_test_client_ops_t
;
364 typedef struct sfw_test_instance
{
365 struct list_head tsi_list
; /* chain on batch */
366 int tsi_service
; /* test type */
367 sfw_batch_t
*tsi_batch
; /* batch */
368 sfw_test_client_ops_t
*tsi_ops
; /* test client operation
371 /* public parameter for all test units */
372 unsigned int tsi_is_client
:1; /* is test client */
373 unsigned int tsi_stoptsu_onerr
:1; /* stop tsu on error */
374 int tsi_concur
; /* concurrency */
375 int tsi_loop
; /* loop count */
377 /* status of test instance */
378 spinlock_t tsi_lock
; /* serialize */
379 unsigned int tsi_stopping
:1; /* test is stopping */
380 atomic_t tsi_nactive
; /* # of active test
382 struct list_head tsi_units
; /* test units */
383 struct list_head tsi_free_rpcs
; /* free rpcs */
384 struct list_head tsi_active_rpcs
; /* active rpcs */
387 test_ping_req_t ping
; /* ping parameter */
388 test_bulk_req_t bulk_v0
; /* bulk parameter */
389 test_bulk_req_v1_t bulk_v1
; /* bulk v1 parameter */
391 } sfw_test_instance_t
;
393 /* XXX: trailing (PAGE_SIZE % sizeof(lnet_process_id_t)) bytes at the end of
394 * pages are not used */
395 #define SFW_MAX_CONCUR LST_MAX_CONCUR
396 #define SFW_ID_PER_PAGE (PAGE_SIZE / sizeof(lnet_process_id_packed_t))
397 #define SFW_MAX_NDESTS (LNET_MAX_IOV * SFW_ID_PER_PAGE)
398 #define sfw_id_pages(n) (((n) + SFW_ID_PER_PAGE - 1) / SFW_ID_PER_PAGE)
400 typedef struct sfw_test_unit
{
401 struct list_head tsu_list
; /* chain on lst_test_instance */
402 lnet_process_id_t tsu_dest
; /* id of dest node */
403 int tsu_loop
; /* loop count of the test */
404 sfw_test_instance_t
*tsu_instance
; /* pointer to test instance */
405 void *tsu_private
; /* private data */
406 swi_workitem_t tsu_worker
; /* workitem of the test unit */
409 typedef struct sfw_test_case
{
410 struct list_head tsc_list
; /* chain on fw_tests */
411 srpc_service_t
*tsc_srv_service
; /* test service */
412 sfw_test_client_ops_t
*tsc_cli_ops
; /* ops of test client */
416 sfw_create_rpc(lnet_process_id_t peer
, int service
,
417 unsigned features
, int nbulkiov
, int bulklen
,
418 void (*done
)(srpc_client_rpc_t
*), void *priv
);
419 int sfw_create_test_rpc(sfw_test_unit_t
*tsu
,
420 lnet_process_id_t peer
, unsigned features
,
421 int nblk
, int blklen
, srpc_client_rpc_t
**rpc
);
422 void sfw_abort_rpc(srpc_client_rpc_t
*rpc
);
423 void sfw_post_rpc(srpc_client_rpc_t
*rpc
);
424 void sfw_client_rpc_done(srpc_client_rpc_t
*rpc
);
425 void sfw_unpack_message(srpc_msg_t
*msg
);
426 void sfw_free_pages(struct srpc_server_rpc
*rpc
);
427 void sfw_add_bulk_page(srpc_bulk_t
*bk
, struct page
*pg
, int i
);
428 int sfw_alloc_pages(struct srpc_server_rpc
*rpc
, int cpt
, int npages
, int len
,
430 int sfw_make_session(srpc_mksn_reqst_t
*request
, srpc_mksn_reply_t
*reply
);
433 srpc_create_client_rpc(lnet_process_id_t peer
, int service
,
434 int nbulkiov
, int bulklen
,
435 void (*rpc_done
)(srpc_client_rpc_t
*),
436 void (*rpc_fini
)(srpc_client_rpc_t
*), void *priv
);
437 void srpc_post_rpc(srpc_client_rpc_t
*rpc
);
438 void srpc_abort_rpc(srpc_client_rpc_t
*rpc
, int why
);
439 void srpc_free_bulk(srpc_bulk_t
*bk
);
440 srpc_bulk_t
*srpc_alloc_bulk(int cpt
, unsigned bulk_npg
, unsigned bulk_len
,
442 int srpc_send_rpc(swi_workitem_t
*wi
);
443 int srpc_send_reply(struct srpc_server_rpc
*rpc
);
444 int srpc_add_service(srpc_service_t
*sv
);
445 int srpc_remove_service(srpc_service_t
*sv
);
446 void srpc_shutdown_service(srpc_service_t
*sv
);
447 void srpc_abort_service(srpc_service_t
*sv
);
448 int srpc_finish_service(srpc_service_t
*sv
);
449 int srpc_service_add_buffers(srpc_service_t
*sv
, int nbuffer
);
450 void srpc_service_remove_buffers(srpc_service_t
*sv
, int nbuffer
);
451 void srpc_get_counters(srpc_counters_t
*cnt
);
452 void srpc_set_counters(const srpc_counters_t
*cnt
);
454 extern struct cfs_wi_sched
*lst_sched_serial
;
455 extern struct cfs_wi_sched
**lst_sched_test
;
458 srpc_serv_is_framework(struct srpc_service
*svc
)
460 return svc
->sv_id
< SRPC_FRAMEWORK_SERVICE_MAX_ID
;
464 swi_wi_action(cfs_workitem_t
*wi
)
466 swi_workitem_t
*swi
= container_of(wi
, swi_workitem_t
, swi_workitem
);
468 return swi
->swi_action(swi
);
472 swi_init_workitem(swi_workitem_t
*swi
, void *data
,
473 swi_action_t action
, struct cfs_wi_sched
*sched
)
475 swi
->swi_sched
= sched
;
476 swi
->swi_action
= action
;
477 swi
->swi_state
= SWI_STATE_NEWBORN
;
478 cfs_wi_init(&swi
->swi_workitem
, data
, swi_wi_action
);
482 swi_schedule_workitem(swi_workitem_t
*wi
)
484 cfs_wi_schedule(wi
->swi_sched
, &wi
->swi_workitem
);
488 swi_exit_workitem(swi_workitem_t
*swi
)
490 cfs_wi_exit(swi
->swi_sched
, &swi
->swi_workitem
);
494 swi_deschedule_workitem(swi_workitem_t
*swi
)
496 return cfs_wi_deschedule(swi
->swi_sched
, &swi
->swi_workitem
);
499 int sfw_startup(void);
500 int srpc_startup(void);
501 void sfw_shutdown(void);
502 void srpc_shutdown(void);
505 srpc_destroy_client_rpc(srpc_client_rpc_t
*rpc
)
508 LASSERT(!srpc_event_pending(rpc
));
509 LASSERT(!atomic_read(&rpc
->crpc_refcount
));
512 LIBCFS_FREE(rpc
, srpc_client_rpc_size(rpc
));
514 (*rpc
->crpc_fini
)(rpc
);
518 srpc_init_client_rpc(srpc_client_rpc_t
*rpc
, lnet_process_id_t peer
,
519 int service
, int nbulkiov
, int bulklen
,
520 void (*rpc_done
)(srpc_client_rpc_t
*),
521 void (*rpc_fini
)(srpc_client_rpc_t
*), void *priv
)
523 LASSERT(nbulkiov
<= LNET_MAX_IOV
);
525 memset(rpc
, 0, offsetof(srpc_client_rpc_t
,
526 crpc_bulk
.bk_iovs
[nbulkiov
]));
528 INIT_LIST_HEAD(&rpc
->crpc_list
);
529 swi_init_workitem(&rpc
->crpc_wi
, rpc
, srpc_send_rpc
,
530 lst_sched_test
[lnet_cpt_of_nid(peer
.nid
)]);
531 spin_lock_init(&rpc
->crpc_lock
);
532 atomic_set(&rpc
->crpc_refcount
, 1); /* 1 ref for caller */
534 rpc
->crpc_dest
= peer
;
535 rpc
->crpc_priv
= priv
;
536 rpc
->crpc_service
= service
;
537 rpc
->crpc_bulk
.bk_len
= bulklen
;
538 rpc
->crpc_bulk
.bk_niov
= nbulkiov
;
539 rpc
->crpc_done
= rpc_done
;
540 rpc
->crpc_fini
= rpc_fini
;
541 LNetInvalidateHandle(&rpc
->crpc_reqstmdh
);
542 LNetInvalidateHandle(&rpc
->crpc_replymdh
);
543 LNetInvalidateHandle(&rpc
->crpc_bulk
.bk_mdh
);
545 /* no event is expected at this point */
546 rpc
->crpc_bulkev
.ev_fired
= 1;
547 rpc
->crpc_reqstev
.ev_fired
= 1;
548 rpc
->crpc_replyev
.ev_fired
= 1;
550 rpc
->crpc_reqstmsg
.msg_magic
= SRPC_MSG_MAGIC
;
551 rpc
->crpc_reqstmsg
.msg_version
= SRPC_MSG_VERSION
;
552 rpc
->crpc_reqstmsg
.msg_type
= srpc_service2request(service
);
555 static inline const char *
556 swi_state2str(int state
)
558 #define STATE2STR(x) case x: return #x
562 STATE2STR(SWI_STATE_NEWBORN
);
563 STATE2STR(SWI_STATE_REPLY_SUBMITTED
);
564 STATE2STR(SWI_STATE_REPLY_SENT
);
565 STATE2STR(SWI_STATE_REQUEST_SUBMITTED
);
566 STATE2STR(SWI_STATE_REQUEST_SENT
);
567 STATE2STR(SWI_STATE_REPLY_RECEIVED
);
568 STATE2STR(SWI_STATE_BULK_STARTED
);
569 STATE2STR(SWI_STATE_DONE
);
574 #define selftest_wait_events() \
576 set_current_state(TASK_UNINTERRUPTIBLE); \
577 schedule_timeout(cfs_time_seconds(1) / 10); \
580 #define lst_wait_until(cond, lock, fmt, ...) \
584 CDEBUG(is_power_of_2(++__I) ? D_WARNING : D_NET, \
585 fmt, ## __VA_ARGS__); \
586 spin_unlock(&(lock)); \
588 selftest_wait_events(); \
590 spin_lock(&(lock)); \
595 srpc_wait_service_shutdown(srpc_service_t
*sv
)
599 LASSERT(sv
->sv_shuttingdown
);
601 while (!srpc_finish_service(sv
)) {
603 CDEBUG(((i
& -i
) == i
) ? D_WARNING
: D_NET
,
604 "Waiting for %s service to shutdown...\n",
606 selftest_wait_events();
610 extern sfw_test_client_ops_t brw_test_client
;
611 void brw_init_test_client(void);
613 extern srpc_service_t brw_test_service
;
614 void brw_init_test_service(void);
616 extern sfw_test_client_ops_t ping_test_client
;
617 void ping_init_test_client(void);
619 extern srpc_service_t ping_test_service
;
620 void ping_init_test_service(void);
622 #endif /* __SELFTEST_SELFTEST_H__ */