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 (c) 2007, 2010, Oracle and/or its affiliates. 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_DONE 10
99 struct sfw_test_instance;
102 * A workitems is deferred work with these semantics:
103 * - a workitem always runs in thread context.
104 * - a workitem can be concurrent with other workitems but is strictly
105 * serialized with respect to itself.
106 * - no CPU affinity, a workitem does not necessarily run on the same CPU
107 * that schedules it. However, this might change in the future.
108 * - if a workitem is scheduled again before it has a chance to run, it
110 * - if a workitem is scheduled while it runs, it runs again after it
111 * completes; this ensures that events occurring while other events are
112 * being processed receive due attention. This behavior also allows a
113 * workitem to reschedule itself.
116 * - a workitem can sleep but it should be aware of how that sleep might
118 * - a workitem runs inside a kernel thread so there's no user space to access.
119 * - do not use a workitem if the scheduling latency can't be tolerated.
121 * When wi_action returns non-zero, it means the workitem has either been
122 * freed or reused and workitem scheduler won't touch it any more.
124 typedef int (*swi_action_t) (struct swi_workitem *);
125 typedef struct swi_workitem {
126 struct list_head wi_list; /* chain on runq */
128 swi_action_t wi_action;
130 unsigned int wi_running:1;
131 unsigned int wi_scheduled:1;
135 swi_init_workitem (swi_workitem_t *wi, void *data, swi_action_t action)
137 CFS_INIT_LIST_HEAD(&wi->wi_list);
140 wi->wi_scheduled = 0;
142 wi->wi_action = action;
143 wi->wi_state = SWI_STATE_NEWBORN;
146 #define SWI_RESCHED 128 /* # workitem scheduler loops before reschedule */
148 /* services below SRPC_FRAMEWORK_SERVICE_MAX_ID are framework
149 * services, e.g. create/modify session.
151 #define SRPC_SERVICE_DEBUG 0
152 #define SRPC_SERVICE_MAKE_SESSION 1
153 #define SRPC_SERVICE_REMOVE_SESSION 2
154 #define SRPC_SERVICE_BATCH 3
155 #define SRPC_SERVICE_TEST 4
156 #define SRPC_SERVICE_QUERY_STAT 5
157 #define SRPC_SERVICE_JOIN 6
158 #define SRPC_FRAMEWORK_SERVICE_MAX_ID 10
159 /* other services start from SRPC_FRAMEWORK_SERVICE_MAX_ID+1 */
160 #define SRPC_SERVICE_BRW 11
161 #define SRPC_SERVICE_PING 12
162 #define SRPC_SERVICE_MAX_ID 12
164 #define SRPC_REQUEST_PORTAL 50
165 /* a lazy portal for framework RPC requests */
166 #define SRPC_FRAMEWORK_REQUEST_PORTAL 51
167 /* all reply/bulk RDMAs go to this portal */
168 #define SRPC_RDMA_PORTAL 52
170 static inline srpc_msg_type_t
171 srpc_service2request (int service)
176 case SRPC_SERVICE_DEBUG:
177 return SRPC_MSG_DEBUG_REQST;
179 case SRPC_SERVICE_MAKE_SESSION:
180 return SRPC_MSG_MKSN_REQST;
182 case SRPC_SERVICE_REMOVE_SESSION:
183 return SRPC_MSG_RMSN_REQST;
185 case SRPC_SERVICE_BATCH:
186 return SRPC_MSG_BATCH_REQST;
188 case SRPC_SERVICE_TEST:
189 return SRPC_MSG_TEST_REQST;
191 case SRPC_SERVICE_QUERY_STAT:
192 return SRPC_MSG_STAT_REQST;
194 case SRPC_SERVICE_BRW:
195 return SRPC_MSG_BRW_REQST;
197 case SRPC_SERVICE_PING:
198 return SRPC_MSG_PING_REQST;
200 case SRPC_SERVICE_JOIN:
201 return SRPC_MSG_JOIN_REQST;
205 static inline srpc_msg_type_t
206 srpc_service2reply (int service)
208 return srpc_service2request(service) + 1;
212 SRPC_BULK_REQ_RCVD = 1, /* passive bulk request(PUT sink/GET source) received */
213 SRPC_BULK_PUT_SENT = 2, /* active bulk PUT sent (source) */
214 SRPC_BULK_GET_RPLD = 3, /* active bulk GET replied (sink) */
215 SRPC_REPLY_RCVD = 4, /* incoming reply received */
216 SRPC_REPLY_SENT = 5, /* outgoing reply sent */
217 SRPC_REQUEST_RCVD = 6, /* incoming request received */
218 SRPC_REQUEST_SENT = 7, /* outgoing request sent */
223 srpc_event_type_t ev_type; /* what's up */
224 lnet_event_kind_t ev_lnet; /* LNet event type */
225 int ev_fired; /* LNet event fired? */
226 int ev_status; /* LNet event status */
227 void *ev_data; /* owning server/client RPC */
231 int bk_len; /* len of bulk data */
232 lnet_handle_md_t bk_mdh;
233 int bk_sink; /* sink/source */
234 int bk_niov; /* # iov in bk_iovs */
236 lnet_kiov_t bk_iovs[0];
238 cfs_page_t **bk_pages;
239 lnet_md_iovec_t bk_iovs[0];
241 } srpc_bulk_t; /* bulk descriptor */
243 /* message buffer descriptor */
245 struct list_head buf_list; /* chain on srpc_service::*_msgq */
247 lnet_handle_md_t buf_mdh;
249 lnet_process_id_t buf_peer;
252 /* server-side state of a RPC */
253 typedef struct srpc_server_rpc {
254 struct list_head srpc_list; /* chain on srpc_service::*_rpcq */
255 struct srpc_service *srpc_service;
256 swi_workitem_t srpc_wi;
257 srpc_event_t srpc_ev; /* bulk/reply event */
258 lnet_nid_t srpc_self;
259 lnet_process_id_t srpc_peer;
260 srpc_msg_t srpc_replymsg;
261 lnet_handle_md_t srpc_replymdh;
262 srpc_buffer_t *srpc_reqstbuf;
263 srpc_bulk_t *srpc_bulk;
265 unsigned int srpc_aborted; /* being given up */
267 void (*srpc_done)(struct srpc_server_rpc *);
270 /* client-side state of a RPC */
271 typedef struct srpc_client_rpc {
272 struct list_head crpc_list; /* chain on user's lists */
273 spinlock_t crpc_lock; /* serialize */
275 atomic_t crpc_refcount;
276 int crpc_timeout; /* # seconds to wait for reply */
277 stt_timer_t crpc_timer;
278 swi_workitem_t crpc_wi;
279 lnet_process_id_t crpc_dest;
281 void (*crpc_done)(struct srpc_client_rpc *);
282 void (*crpc_fini)(struct srpc_client_rpc *);
283 int crpc_status; /* completion status */
284 void *crpc_priv; /* caller data */
287 unsigned int crpc_aborted:1; /* being given up */
288 unsigned int crpc_closed:1; /* completed */
291 srpc_event_t crpc_bulkev; /* bulk event */
292 srpc_event_t crpc_reqstev; /* request event */
293 srpc_event_t crpc_replyev; /* reply event */
295 /* bulk, request(reqst), and reply exchanged on wire */
296 srpc_msg_t crpc_reqstmsg;
297 srpc_msg_t crpc_replymsg;
298 lnet_handle_md_t crpc_reqstmdh;
299 lnet_handle_md_t crpc_replymdh;
300 srpc_bulk_t crpc_bulk;
303 #define srpc_client_rpc_size(rpc) \
304 offsetof(srpc_client_rpc_t, crpc_bulk.bk_iovs[(rpc)->crpc_bulk.bk_niov])
306 #define srpc_client_rpc_addref(rpc) \
308 CDEBUG(D_NET, "RPC[%p] -> %s (%d)++\n", \
309 (rpc), libcfs_id2str((rpc)->crpc_dest), \
310 atomic_read(&(rpc)->crpc_refcount)); \
311 LASSERT(atomic_read(&(rpc)->crpc_refcount) > 0); \
312 atomic_inc(&(rpc)->crpc_refcount); \
315 #define srpc_client_rpc_decref(rpc) \
317 CDEBUG(D_NET, "RPC[%p] -> %s (%d)--\n", \
318 (rpc), libcfs_id2str((rpc)->crpc_dest), \
319 atomic_read(&(rpc)->crpc_refcount)); \
320 LASSERT(atomic_read(&(rpc)->crpc_refcount) > 0); \
321 if (atomic_dec_and_test(&(rpc)->crpc_refcount)) \
322 srpc_destroy_client_rpc(rpc); \
325 #define srpc_event_pending(rpc) ((rpc)->crpc_bulkev.ev_fired == 0 || \
326 (rpc)->crpc_reqstev.ev_fired == 0 || \
327 (rpc)->crpc_replyev.ev_fired == 0)
329 typedef struct srpc_service {
330 int sv_id; /* service id */
331 const char *sv_name; /* human readable name */
332 int sv_nprune; /* # posted RPC to be pruned */
333 int sv_concur; /* max # concurrent RPCs */
337 srpc_event_t sv_ev; /* LNet event */
338 int sv_nposted_msg; /* # posted message buffers */
339 struct list_head sv_free_rpcq; /* free RPC descriptors */
340 struct list_head sv_active_rpcq; /* in-flight RPCs */
341 struct list_head sv_posted_msgq; /* posted message buffers */
342 struct list_head sv_blocked_msgq; /* blocked for RPC descriptor */
344 /* Service callbacks:
345 * - sv_handler: process incoming RPC request
346 * - sv_bulk_ready: notify bulk data
348 int (*sv_handler) (srpc_server_rpc_t *);
349 int (*sv_bulk_ready) (srpc_server_rpc_t *, int);
352 #define SFW_POST_BUFFERS 256
353 #define SFW_SERVICE_CONCURRENCY (SFW_POST_BUFFERS/2)
356 struct list_head sn_list; /* chain on fw_zombie_sessions */
357 lst_sid_t sn_id; /* unique identifier */
358 unsigned int sn_timeout; /* # seconds' inactivity to expire */
360 stt_timer_t sn_timer;
361 struct list_head sn_batches; /* list of batches */
362 char sn_name[LST_NAME_SIZE];
363 atomic_t sn_refcount;
364 atomic_t sn_brw_errors;
365 atomic_t sn_ping_errors;
368 #define sfw_sid_equal(sid0, sid1) ((sid0).ses_nid == (sid1).ses_nid && \
369 (sid0).ses_stamp == (sid1).ses_stamp)
372 struct list_head bat_list; /* chain on sn_batches */
373 lst_bid_t bat_id; /* batch id */
374 int bat_error; /* error code of batch */
375 sfw_session_t *bat_session; /* batch's session */
376 atomic_t bat_nactive; /* # of active tests */
377 struct list_head bat_tests; /* test instances */
381 int (*tso_init)(struct sfw_test_instance *tsi); /* intialize test client */
382 void (*tso_fini)(struct sfw_test_instance *tsi); /* finalize test client */
383 int (*tso_prep_rpc)(struct sfw_test_unit *tsu,
384 lnet_process_id_t dest,
385 srpc_client_rpc_t **rpc); /* prep a tests rpc */
386 void (*tso_done_rpc)(struct sfw_test_unit *tsu,
387 srpc_client_rpc_t *rpc); /* done a test rpc */
388 } sfw_test_client_ops_t;
390 typedef struct sfw_test_instance {
391 struct list_head tsi_list; /* chain on batch */
392 int tsi_service; /* test type */
393 sfw_batch_t *tsi_batch; /* batch */
394 sfw_test_client_ops_t *tsi_ops; /* test client operations */
396 /* public parameter for all test units */
397 int tsi_is_client:1; /* is test client */
398 int tsi_stoptsu_onerr:1; /* stop tsu on error */
399 int tsi_concur; /* concurrency */
400 int tsi_loop; /* loop count */
402 /* status of test instance */
403 spinlock_t tsi_lock; /* serialize */
404 int tsi_stopping:1; /* test is stopping */
405 atomic_t tsi_nactive; /* # of active test unit */
406 struct list_head tsi_units; /* test units */
407 struct list_head tsi_free_rpcs; /* free rpcs */
408 struct list_head tsi_active_rpcs; /* active rpcs */
411 test_bulk_req_t bulk; /* bulk parameter */
412 test_ping_req_t ping; /* ping parameter */
414 } sfw_test_instance_t;
416 /* XXX: trailing (CFS_PAGE_SIZE % sizeof(lnet_process_id_t)) bytes at
417 * the end of pages are not used */
418 #define SFW_MAX_CONCUR LST_MAX_CONCUR
419 #define SFW_ID_PER_PAGE (CFS_PAGE_SIZE / sizeof(lnet_process_id_packed_t))
420 #define SFW_MAX_NDESTS (LNET_MAX_IOV * SFW_ID_PER_PAGE)
421 #define sfw_id_pages(n) (((n) + SFW_ID_PER_PAGE - 1) / SFW_ID_PER_PAGE)
423 typedef struct sfw_test_unit {
424 struct list_head tsu_list; /* chain on lst_test_instance */
425 lnet_process_id_t tsu_dest; /* id of dest node */
426 int tsu_loop; /* loop count of the test */
427 sfw_test_instance_t *tsu_instance; /* pointer to test instance */
428 void *tsu_private; /* private data */
429 swi_workitem_t tsu_worker; /* workitem of the test unit */
433 struct list_head tsc_list; /* chain on fw_tests */
434 srpc_service_t *tsc_srv_service; /* test service */
435 sfw_test_client_ops_t *tsc_cli_ops; /* ops of test client */
440 sfw_create_rpc(lnet_process_id_t peer, int service, int nbulkiov, int bulklen,
441 void (*done) (srpc_client_rpc_t *), void *priv);
442 int sfw_create_test_rpc(sfw_test_unit_t *tsu, lnet_process_id_t peer,
443 int nblk, int blklen, srpc_client_rpc_t **rpc);
444 void sfw_abort_rpc(srpc_client_rpc_t *rpc);
445 void sfw_post_rpc(srpc_client_rpc_t *rpc);
446 void sfw_client_rpc_done(srpc_client_rpc_t *rpc);
447 void sfw_unpack_message(srpc_msg_t *msg);
448 void sfw_free_pages(srpc_server_rpc_t *rpc);
449 void sfw_add_bulk_page(srpc_bulk_t *bk, cfs_page_t *pg, int i);
450 int sfw_alloc_pages(srpc_server_rpc_t *rpc, int npages, int sink);
451 int sfw_make_session (srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply);
454 srpc_create_client_rpc(lnet_process_id_t peer, int service,
455 int nbulkiov, int bulklen,
456 void (*rpc_done)(srpc_client_rpc_t *),
457 void (*rpc_fini)(srpc_client_rpc_t *), void *priv);
458 void srpc_post_rpc(srpc_client_rpc_t *rpc);
459 void srpc_abort_rpc(srpc_client_rpc_t *rpc, int why);
460 void srpc_free_bulk(srpc_bulk_t *bk);
461 srpc_bulk_t *srpc_alloc_bulk(int npages, int sink);
462 int srpc_send_rpc(swi_workitem_t *wi);
463 int srpc_send_reply(srpc_server_rpc_t *rpc);
464 int srpc_add_service(srpc_service_t *sv);
465 int srpc_remove_service(srpc_service_t *sv);
466 void srpc_shutdown_service(srpc_service_t *sv);
467 void srpc_abort_service(srpc_service_t *sv);
468 int srpc_finish_service(srpc_service_t *sv);
469 int srpc_service_add_buffers(srpc_service_t *sv, int nbuffer);
470 void srpc_service_remove_buffers(srpc_service_t *sv, int nbuffer);
471 void srpc_get_counters(srpc_counters_t *cnt);
472 void srpc_set_counters(const srpc_counters_t *cnt);
474 void swi_kill_workitem(swi_workitem_t *wi);
475 void swi_schedule_workitem(swi_workitem_t *wi);
476 void swi_schedule_serial_workitem(swi_workitem_t *wi);
477 int swi_startup(void);
478 int sfw_startup(void);
479 int srpc_startup(void);
480 void swi_shutdown(void);
481 void sfw_shutdown(void);
482 void srpc_shutdown(void);
485 srpc_destroy_client_rpc (srpc_client_rpc_t *rpc)
487 LASSERT (rpc != NULL);
488 LASSERT (!srpc_event_pending(rpc));
489 LASSERT (atomic_read(&rpc->crpc_refcount) == 0);
491 LASSERT (rpc->crpc_bulk.bk_pages == NULL);
494 if (rpc->crpc_fini == NULL) {
495 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
497 (*rpc->crpc_fini) (rpc);
504 srpc_init_client_rpc (srpc_client_rpc_t *rpc, lnet_process_id_t peer,
505 int service, int nbulkiov, int bulklen,
506 void (*rpc_done)(srpc_client_rpc_t *),
507 void (*rpc_fini)(srpc_client_rpc_t *), void *priv)
509 LASSERT (nbulkiov <= LNET_MAX_IOV);
511 memset(rpc, 0, offsetof(srpc_client_rpc_t,
512 crpc_bulk.bk_iovs[nbulkiov]));
514 CFS_INIT_LIST_HEAD(&rpc->crpc_list);
515 swi_init_workitem(&rpc->crpc_wi, rpc, srpc_send_rpc);
516 spin_lock_init(&rpc->crpc_lock);
517 atomic_set(&rpc->crpc_refcount, 1); /* 1 ref for caller */
519 rpc->crpc_dest = peer;
520 rpc->crpc_priv = priv;
521 rpc->crpc_service = service;
522 rpc->crpc_bulk.bk_len = bulklen;
523 rpc->crpc_bulk.bk_niov = nbulkiov;
524 rpc->crpc_done = rpc_done;
525 rpc->crpc_fini = rpc_fini;
528 rpc->crpc_bulk.bk_mdh = LNET_INVALID_HANDLE;
530 /* no event is expected at this point */
531 rpc->crpc_bulkev.ev_fired =
532 rpc->crpc_reqstev.ev_fired =
533 rpc->crpc_replyev.ev_fired = 1;
535 rpc->crpc_reqstmsg.msg_magic = SRPC_MSG_MAGIC;
536 rpc->crpc_reqstmsg.msg_version = SRPC_MSG_VERSION;
537 rpc->crpc_reqstmsg.msg_type = srpc_service2request(service);
541 static inline const char *
542 swi_state2str (int state)
544 #define STATE2STR(x) case x: return #x
548 STATE2STR(SWI_STATE_NEWBORN);
549 STATE2STR(SWI_STATE_REPLY_SUBMITTED);
550 STATE2STR(SWI_STATE_REPLY_SENT);
551 STATE2STR(SWI_STATE_REQUEST_SUBMITTED);
552 STATE2STR(SWI_STATE_REQUEST_SENT);
553 STATE2STR(SWI_STATE_REPLY_RECEIVED);
554 STATE2STR(SWI_STATE_BULK_STARTED);
555 STATE2STR(SWI_STATE_DONE);
560 #define UNUSED(x) ( (void)(x) )
564 int stt_poll_interval(void);
565 int sfw_session_removed(void);
567 int stt_check_events(void);
568 int swi_check_events(void);
569 int srpc_check_event(int timeout);
571 int lnet_selftest_init(void);
572 void lnet_selftest_fini(void);
573 int selftest_wait_events(void);
577 #define selftest_wait_events() cfs_pause(cfs_time_seconds(1))
581 #define lst_wait_until(cond, lock, fmt, a...) \
586 CDEBUG(((__I & (-__I)) == __I) ? D_WARNING : \
589 spin_unlock(&(lock)); \
591 selftest_wait_events(); \
593 spin_lock(&(lock)); \
598 srpc_wait_service_shutdown (srpc_service_t *sv)
602 spin_lock(&sv->sv_lock);
603 LASSERT (sv->sv_shuttingdown);
604 spin_unlock(&sv->sv_lock);
606 while (srpc_finish_service(sv) == 0) {
608 CDEBUG (((i & -i) == i) ? D_WARNING : D_NET,
609 "Waiting for %s service to shutdown...\n",
611 selftest_wait_events();
615 #endif /* __SELFTEST_SELFTEST_H__ */