xref: /titanic_52/usr/src/lib/fm/libldom/sparc/ldmsvcs_utils.c (revision 9525b14bcdeb5b5f6f95ab27c2f48f18bd2ec829)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <stdlib.h>
27 #include <stdio.h>
28 
29 #include <strings.h>
30 #include <sys/types.h>
31 #include <sys/stat.h>
32 #include <time.h>
33 #include <fcntl.h>
34 #include <unistd.h>
35 #include <errno.h>
36 #include <assert.h>
37 #include <umem.h>
38 #include <alloca.h>
39 #include <sys/processor.h>
40 #include <poll.h>
41 #include <pthread.h>
42 #include <signal.h>
43 #include <values.h>
44 #include <libscf.h>
45 
46 #include <ctype.h>
47 
48 #include "ldmsvcs_utils.h"
49 #include "ldom_alloc.h"
50 
51 #define	ASSERT(cnd) \
52 	((void) ((cnd) || ((void) fprintf(stderr, \
53 		"assertion failure in %s:%d: %s\n", \
54 		__FILE__, __LINE__, #cnd), 0)))
55 
56 #define	FDS_VLDC \
57 	"/devices/virtual-devices@100/channel-devices@200/" \
58 	"/virtual-channel-client@1:ldmfma"
59 
60 /* allow timeouts in sec that are nearly forever but small enough for an int */
61 #define	LDM_TIMEOUT_CEILING	(MAXINT / 2)
62 
63 #define	MIN(x, y)	((x) < (y) ? (x) : (y))
64 
65 /*
66  * functions in this file are for version 1.0 of FMA domain services
67  */
68 static ds_ver_t ds_vers[] = {
69 	{ 1, 0 }
70 };
71 
72 #define	DS_NUM_VER	(sizeof (ds_vers) / sizeof (ds_ver_t))
73 
74 /*
75  * information for each channel
76  */
77 struct ldmsvcs_info {
78 	pthread_mutex_t mt;
79 	pthread_cond_t cv;
80 	fds_channel_t fds_chan;
81 	fds_reg_svcs_t fmas_svcs;
82 	int cv_twait;
83 };
84 
85 /*
86  * struct listdata_s and struct poller_s are used to maintain the state of
87  * the poller thread.  this thread is used to manage incoming messages and
88  * pass those messages onto the correct requesting thread.  see the "poller
89  * functions" section for more details.
90  */
91 struct listdata_s {
92 	enum {
93 		UNUSED,
94 		PENDING,
95 		ARRIVED
96 	} status;
97 	uint64_t req_num;
98 	int fd;
99 	size_t datalen;
100 };
101 
102 static struct poller_s {
103 	pthread_mutex_t mt;
104 	pthread_cond_t cv;
105 	pthread_t polling_tid;
106 	int doreset;
107 	int doexit;
108 	int nclients;
109 	struct listdata_s **list;
110 	int list_len;
111 	int pending_count;
112 } pollbase = {
113 	PTHREAD_MUTEX_INITIALIZER,
114 	PTHREAD_COND_INITIALIZER,
115 	0,
116 	1,
117 	0,
118 	0,
119 	NULL,
120 	0,
121 	0
122 };
123 
124 
125 static struct ldmsvcs_info *channel_init(struct ldom_hdl *lhp);
126 static int channel_openreset(struct ldmsvcs_info *lsp);
127 static int read_msg(struct ldmsvcs_info *lsp);
128 
129 static int
130 get_smf_int_val(char *prop_nm, int min, int max, int default_val)
131 {
132 	scf_simple_prop_t	*prop;		/* SMF property */
133 	int64_t			*valp;		/* prop value ptr */
134 	int64_t			val;		/* prop value to return */
135 
136 	val = default_val;
137 	if ((prop = scf_simple_prop_get(NULL, LDM_SVC_NM, LDM_PROP_GROUP_NM,
138 	    prop_nm)) != NULL) {
139 		if ((valp = scf_simple_prop_next_integer(prop)) != NULL) {
140 			val = *valp;
141 			if (val < min)
142 				val = min;
143 			else if (val > max)
144 				val = max;
145 		}
146 		scf_simple_prop_free(prop);
147 	}
148 	return ((int)val);
149 }
150 
151 static void
152 channel_close(struct ldmsvcs_info *lsp)
153 {
154 	(void) pthread_mutex_lock(&lsp->mt);
155 
156 	if (lsp->fds_chan.state == CHANNEL_OPEN ||
157 	    lsp->fds_chan.state == CHANNEL_READY) {
158 		(void) close(lsp->fds_chan.fd);
159 		lsp->cv_twait = get_smf_int_val(LDM_INIT_TO_PROP_NM,
160 		    0, LDM_TIMEOUT_CEILING, LDM_INIT_WAIT_TIME);
161 		lsp->fds_chan.state = CHANNEL_CLOSED;
162 	}
163 
164 	(void) pthread_mutex_unlock(&lsp->mt);
165 }
166 
167 /*
168  * read size bytes of data from a streaming fd into buf
169  */
170 static int
171 read_stream(int fd, void *buf, size_t size)
172 {
173 	pollfd_t pollfd;
174 	ssize_t rv;
175 	size_t data_left;
176 	ptrdiff_t currentp;
177 
178 	pollfd.events = POLLIN;
179 	pollfd.revents = 0;
180 	pollfd.fd = fd;
181 
182 	currentp = (ptrdiff_t)buf;
183 	data_left = size;
184 
185 	/*
186 	 * data may come in bits and pieces
187 	 */
188 	do {
189 		if ((rv = read(fd, (void *)currentp, data_left)) < 0) {
190 			if (errno == EAGAIN && poll(&pollfd, 1, -1) > 0)
191 				continue;	/* retry */
192 			else
193 				return (1);
194 		}
195 
196 		data_left -= rv;
197 		currentp += rv;
198 	} while (data_left > 0);
199 
200 	return (0);
201 }
202 
203 
204 /*
205  * poller functions
206  *
207  * at init time, a thread is created for the purpose of monitoring incoming
208  * messages and doing one of the following:
209  *
210  * 1. doing the initial handshake and version negotiation
211  *
212  * 2. handing incoming data off to the requesting thread (which is an fmd
213  * module or scheme thread)
214  */
215 static int
216 poller_handle_data(int fd, size_t payloadsize)
217 {
218 	uint64_t *req_num;
219 	void *pr;
220 	size_t prlen;
221 	int i;
222 
223 	prlen = sizeof (ds_data_handle_t) + sizeof (uint64_t);
224 
225 	if (payloadsize < prlen)
226 		return (1);
227 
228 	pr = alloca(prlen);
229 
230 	if (read_stream(fd, pr, prlen) != 0)
231 		return (1);
232 
233 	req_num = (uint64_t *)((ptrdiff_t)pr + sizeof (ds_data_handle_t));
234 
235 	(void) pthread_mutex_lock(&pollbase.mt);
236 
237 	for (i = 0; i < pollbase.list_len; i++) {
238 		if (pollbase.list[i]->req_num == *req_num) {
239 			ASSERT(pollbase.list[i]->status == PENDING);
240 
241 			pollbase.list[i]->status = ARRIVED;
242 			pollbase.list[i]->fd = fd;
243 			pollbase.list[i]->datalen = payloadsize - prlen;
244 
245 			pollbase.pending_count--;
246 			(void) pthread_cond_broadcast(&pollbase.cv);
247 			break;
248 		}
249 	}
250 
251 	/*
252 	 * now wait for receiving thread to read in the data
253 	 */
254 	if (i < pollbase.list_len) {
255 		while (pollbase.list[i]->status == ARRIVED)
256 			(void) pthread_cond_wait(&pollbase.cv, &pollbase.mt);
257 	}
258 
259 	(void) pthread_mutex_unlock(&pollbase.mt);
260 
261 	return (0);
262 }
263 
264 
265 /*
266  * note that this function is meant to handle only DS_DATA messages
267  */
268 static int
269 poller_recv_data(struct ldom_hdl *lhp, uint64_t req_num, int index,
270 		void **resp, size_t *resplen)
271 {
272 	struct timespec twait;
273 	int ier;
274 
275 	ier = 0;
276 	twait.tv_sec = time(NULL) + lhp->lsinfo->cv_twait;
277 	twait.tv_nsec = 0;
278 
279 	(void) pthread_mutex_lock(&pollbase.mt);
280 
281 	ASSERT(pollbase.list[index]->req_num == req_num);
282 
283 	while (pollbase.list[index]->status == PENDING &&
284 	    pollbase.doreset == 0 && ier == 0)
285 		ier = pthread_cond_timedwait(&pollbase.cv, &pollbase.mt,
286 		    &twait);
287 
288 	if (ier == 0) {
289 		if (pollbase.doreset == 0) {
290 			ASSERT(pollbase.list[index]->status == ARRIVED);
291 
292 			/*
293 			 * need to add req_num to beginning of resp
294 			 */
295 			*resplen = pollbase.list[index]->datalen +
296 			    sizeof (uint64_t);
297 			*resp = lhp->allocp(*resplen);
298 			*((uint64_t *)*resp) = req_num;
299 
300 			if (read_stream(pollbase.list[index]->fd,
301 			    (void *)((ptrdiff_t)*resp + sizeof (uint64_t)),
302 			    *resplen - sizeof (uint64_t)) != 0)
303 				ier = ETIMEDOUT;
304 
305 			pollbase.list[index]->status = UNUSED;
306 			pollbase.list[index]->req_num = 0;
307 			(void) pthread_cond_broadcast(&pollbase.cv);
308 		} else {
309 			if (--(pollbase.pending_count) == 0)
310 				(void) pthread_cond_broadcast(&pollbase.cv);
311 		}
312 	}
313 
314 	(void) pthread_mutex_unlock(&pollbase.mt);
315 
316 	ASSERT(ier == 0 || ier == ETIMEDOUT);
317 
318 	return (ier);
319 }
320 
321 
322 static void
323 poller_add_client(void)
324 {
325 	(void) pthread_mutex_lock(&pollbase.mt);
326 	pollbase.nclients++;
327 	(void) pthread_mutex_unlock(&pollbase.mt);
328 }
329 
330 
331 static void
332 poller_remove_client(void)
333 {
334 	(void) pthread_mutex_lock(&pollbase.mt);
335 	pollbase.nclients--;
336 	ASSERT(pollbase.nclients >= 0);
337 	(void) pthread_mutex_unlock(&pollbase.mt);
338 }
339 
340 
341 static int
342 poller_add_pending(uint64_t req_num)
343 {
344 	int newlen, index, i, j;
345 
346 	(void) pthread_mutex_lock(&pollbase.mt);
347 	pollbase.pending_count++;
348 
349 	for (j = 0, index = -1; j < 2 && index == -1; j++) {
350 		for (i = 0; i < pollbase.list_len; i++) {
351 			if (pollbase.list[i]->status == UNUSED) {
352 				pollbase.list[i]->status = PENDING;
353 				pollbase.list[i]->req_num = req_num;
354 				pollbase.list[i]->datalen = 0;
355 				index = i;
356 				break;
357 			}
358 		}
359 
360 		if (index == -1) {
361 			struct listdata_s **newlist, **oldlist;
362 
363 			/*
364 			 * get to this point if list is not long enough.
365 			 * check for a runaway list.  since requests are
366 			 * synchronous (clients send a request and need to
367 			 * wait for the result before returning) the size
368 			 * of the list cannot be much more than the number
369 			 * of clients.
370 			 */
371 			ASSERT(pollbase.list_len < pollbase.nclients + 1);
372 
373 			newlen = pollbase.list_len + 5;
374 			newlist = ldom_alloc(newlen *
375 			    sizeof (struct listdata_s *));
376 
377 			for (i = 0; i < pollbase.list_len; i++)
378 				newlist[i] = pollbase.list[i];
379 
380 			oldlist = pollbase.list;
381 			pollbase.list = newlist;
382 			ldom_free(oldlist, pollbase.list_len *
383 			    sizeof (struct listdata_s *));
384 
385 			for (i = pollbase.list_len; i < newlen; i++) {
386 				pollbase.list[i] =
387 				    ldom_alloc(sizeof (struct listdata_s));
388 				pollbase.list[i]->status = UNUSED;
389 			}
390 
391 			pollbase.list_len = newlen;
392 		}
393 	}
394 
395 	(void) pthread_mutex_unlock(&pollbase.mt);
396 	ASSERT(index != -1);
397 
398 	return (index);
399 }
400 
401 
402 static void
403 poller_delete_pending(uint64_t req_num, int index)
404 {
405 	(void) pthread_mutex_lock(&pollbase.mt);
406 
407 	ASSERT(pollbase.list[index]->req_num == req_num);
408 	pollbase.list[index]->status = UNUSED;
409 
410 	if (--(pollbase.pending_count) == 0 && pollbase.doreset == 1)
411 		(void) pthread_cond_broadcast(&pollbase.cv);
412 
413 	(void) pthread_mutex_unlock(&pollbase.mt);
414 }
415 
416 
417 static void
418 poller_shutdown(boolean_t wait)
419 {
420 	(void) pthread_mutex_lock(&pollbase.mt);
421 
422 	pollbase.doexit = 1;
423 
424 	(void) pthread_mutex_unlock(&pollbase.mt);
425 
426 	if (wait == B_TRUE) {
427 		/* stop the poller thread  and wait for it to end */
428 		(void) pthread_kill(pollbase.polling_tid, SIGTERM);
429 		(void) pthread_join(pollbase.polling_tid, NULL);
430 	}
431 }
432 
433 
434 /*
435  * perform the polling of incoming messages.  manage any resets (usually
436  * due to one end of the connection being closed) as well as exit
437  * conditions.
438  */
439 static void *
440 poller_loop(void *arg)
441 {
442 	struct ldmsvcs_info *lsp;
443 	pollfd_t pollfd;
444 	int ier;
445 
446 	lsp = (struct ldmsvcs_info *)arg;
447 
448 	for (;;) {
449 		(void) pthread_mutex_lock(&pollbase.mt);
450 
451 		if (pollbase.doexit) {
452 			(void) pthread_mutex_unlock(&pollbase.mt);
453 			break;
454 		}
455 
456 		if (pollbase.doreset) {
457 			int i;
458 
459 			while (pollbase.pending_count > 0)
460 				(void) pthread_cond_wait(&pollbase.cv,
461 				    &pollbase.mt);
462 
463 			ASSERT(pollbase.pending_count == 0);
464 			for (i = 0; i < pollbase.list_len; i++)
465 				pollbase.list[i]->status = UNUSED;
466 
467 			pollbase.doreset = 0;
468 		}
469 		(void) pthread_mutex_unlock(&pollbase.mt);
470 
471 		if ((ier = channel_openreset(lsp)) == 1) {
472 			continue;
473 		} else if (ier == 2) {
474 			/*
475 			 * start exit preparations
476 			 */
477 			poller_shutdown(B_FALSE);
478 			continue;
479 		}
480 
481 		pollfd.events = POLLIN;
482 		pollfd.revents = 0;
483 		pollfd.fd = lsp->fds_chan.fd;
484 
485 		if (poll(&pollfd, 1, -1) <= 0 || read_msg(lsp) != 0) {
486 			/*
487 			 * read error and/or fd got closed
488 			 */
489 			(void) pthread_mutex_lock(&pollbase.mt);
490 			pollbase.doreset = 1;
491 			(void) pthread_mutex_unlock(&pollbase.mt);
492 
493 			channel_close(lsp);
494 		}
495 	}
496 
497 	return (NULL);
498 }
499 
500 
501 /*
502  * create the polling thread
503  */
504 static int
505 poller_init(struct ldmsvcs_info *lsp)
506 {
507 	int rc = 0;
508 
509 	(void) pthread_mutex_lock(&pollbase.mt);
510 
511 	if (pollbase.polling_tid == 0) {
512 		pthread_attr_t *attr = NULL;
513 
514 		/*
515 		 * create a joinable polling thread for receiving messages
516 		 */
517 		if (pthread_create(&pollbase.polling_tid, attr,
518 		    poller_loop, lsp) != 0)
519 			rc = 1;
520 	}
521 
522 	(void) pthread_mutex_unlock(&pollbase.mt);
523 
524 	return (rc);
525 }
526 
527 /*
528  * Cleanup the polling thread
529  */
530 static void
531 poller_fini(void)
532 {
533 	int i;
534 
535 	/* stop the poller thread */
536 	poller_shutdown(B_TRUE);
537 
538 	(void) pthread_mutex_lock(&pollbase.mt);
539 
540 	/* Free up the list of outstanding requests */
541 	if (pollbase.list != NULL) {
542 		for (i = 0; i < pollbase.list_len; i++) {
543 			if (pollbase.list[i]) {
544 				ldom_free(pollbase.list[i],
545 				    sizeof (struct listdata_s));
546 			}
547 		}
548 		ldom_free(pollbase.list, pollbase.list_len *
549 		    sizeof (struct listdata_s *));
550 		pollbase.list = NULL;
551 		pollbase.list_len = 0;
552 	}
553 
554 	(void) pthread_mutex_unlock(&pollbase.mt);
555 }
556 
557 /*
558  * utilities for message handlers
559  */
560 static int
561 fds_send(struct ldmsvcs_info *lsp, void *msg, size_t msglen)
562 {
563 	static pthread_mutex_t mt = PTHREAD_MUTEX_INITIALIZER;
564 
565 	(void) pthread_mutex_lock(&mt);
566 
567 	if (write(lsp->fds_chan.fd, msg, msglen) != msglen) {
568 		channel_close(lsp);
569 		(void) pthread_mutex_unlock(&mt);
570 		return (ETIMEDOUT);
571 	}
572 
573 	(void) pthread_mutex_unlock(&mt);
574 	return (0);
575 }
576 
577 
578 /*
579  * Find the max and min version supported
580  */
581 static void
582 fds_min_max_versions(uint16_t *min_major, uint16_t *max_major)
583 {
584 	int i;
585 
586 	*min_major = ds_vers[0].major;
587 	*max_major = *min_major;
588 
589 	for (i = 1; i < DS_NUM_VER; i++) {
590 		if (ds_vers[i].major < *min_major)
591 			*min_major = ds_vers[i].major;
592 
593 		if (ds_vers[i].major > *max_major)
594 			*max_major = ds_vers[i].major;
595 	}
596 }
597 
598 /*
599  * check whether the major and minor numbers requested by remote ds client
600  * can be satisfied.  if the requested major is supported, true is
601  * returned, and the agreed minor is returned in new_minor.  if the
602  * requested major is not supported, the routine returns false, and the
603  * closest major is returned in *new_major, upon which the ds client should
604  * renegotiate.  the closest major is the just lower that the requested
605  * major number.
606  */
607 static boolean_t
608 fds_negotiate_version(uint16_t req_major, uint16_t *new_majorp,
609     uint16_t *new_minorp)
610 {
611 	int i = 0;
612 	uint16_t major, lower_major;
613 	uint16_t min_major, max_major;
614 	boolean_t found_match = B_FALSE;
615 
616 	fds_min_max_versions(&min_major, &max_major);
617 
618 	/*
619 	 * if the minimum version supported is greater than the version
620 	 * requested, return the lowest version supported
621 	 */
622 	if (min_major > req_major) {
623 		*new_majorp = min_major;
624 		return (B_FALSE);
625 	}
626 
627 	/*
628 	 * if the largest version supported is lower than the version
629 	 * requested, return the largest version supported
630 	 */
631 	if (max_major < req_major) {
632 		*new_majorp = max_major;
633 		return (B_FALSE);
634 	}
635 
636 	/*
637 	 * now we know that the requested version lies between the min and
638 	 * max versions supported.  check if the requested major can be
639 	 * found in supported versions.
640 	 */
641 	lower_major = min_major;
642 	for (i = 0; i < DS_NUM_VER; i++) {
643 		major = ds_vers[i].major;
644 		if (major == req_major) {
645 			found_match = B_TRUE;
646 			*new_minorp = ds_vers[i].minor;
647 			*new_majorp = major;
648 			break;
649 		} else if ((major < req_major) && (major > lower_major))
650 			lower_major = major;
651 	}
652 
653 	/*
654 	 * If  no match is found, return the closest available number
655 	 */
656 	if (!found_match)
657 		*new_majorp = lower_major;
658 
659 	return (found_match);
660 }
661 
662 
663 /*
664  * return 0 if service is added; 1 if service is a duplicate
665  */
666 static int
667 fds_svc_add(struct ldmsvcs_info *lsp, ds_reg_req_t *req, int minor)
668 {
669 	fds_svc_t *svc;
670 	int i, rc;
671 
672 	svc = NULL;
673 	for (i = 0; i < lsp->fmas_svcs.nsvcs; i++) {
674 		if (strcmp(lsp->fmas_svcs.tbl[i]->name, req->svc_id) == 0) {
675 			svc = lsp->fmas_svcs.tbl[i];
676 			break;
677 		}
678 	}
679 
680 	if (svc == NULL)
681 		return (0);	/* we don't need this service */
682 
683 	(void) pthread_mutex_lock(&lsp->fmas_svcs.mt);
684 
685 	/*
686 	 * duplicate registration is OK --- we retain the previous entry
687 	 * (which has not been unregistered anyway)
688 	 */
689 	if (svc->state == DS_SVC_ACTIVE) {
690 		rc = 1;
691 	} else {
692 		svc->state = DS_SVC_ACTIVE;
693 		svc->hdl = req->svc_handle;
694 		svc->ver.major = req->major_vers;
695 		svc->ver.minor = minor;
696 
697 		rc = 0;
698 		(void) pthread_cond_broadcast(&lsp->fmas_svcs.cv);
699 	}
700 
701 	(void) pthread_mutex_unlock(&lsp->fmas_svcs.mt);
702 
703 	return (rc);
704 }
705 
706 
707 static void
708 fds_svc_reset(struct ldmsvcs_info *lsp, int index)
709 {
710 	int i, start, end;
711 
712 	if (index >= 0) {
713 		start = index;
714 		end = index + 1;
715 	} else {
716 		start = 0;
717 		end = lsp->fmas_svcs.nsvcs;
718 	}
719 
720 	(void) pthread_mutex_lock(&lsp->fmas_svcs.mt);
721 
722 	for (i = start; i < end; i++) {
723 		lsp->fmas_svcs.tbl[i]->hdl = 0;
724 		lsp->fmas_svcs.tbl[i]->state = DS_SVC_INVAL;
725 		lsp->fmas_svcs.tbl[i]->ver.major =
726 		    ds_vers[DS_NUM_VER - 1].major;
727 		lsp->fmas_svcs.tbl[i]->ver.minor =
728 		    ds_vers[DS_NUM_VER - 1].minor;
729 	}
730 
731 	(void) pthread_mutex_unlock(&lsp->fmas_svcs.mt);
732 }
733 
734 
735 static int
736 fds_svc_remove(struct ldmsvcs_info *lsp, ds_svc_hdl_t svc_handle)
737 {
738 	int i;
739 
740 	for (i = 0; i < lsp->fmas_svcs.nsvcs; i++) {
741 		if (lsp->fmas_svcs.tbl[i]->hdl == svc_handle) {
742 			fds_svc_reset(lsp, i);
743 			return (0);
744 		}
745 	}
746 
747 	return (1);
748 }
749 
750 
751 /*
752  * message handlers
753  */
754 /*ARGSUSED*/
755 static void
756 ds_handle_msg_noop(struct ldmsvcs_info *lsp, void *buf, size_t len)
757 {
758 }
759 
760 static void
761 ds_handle_init_req(struct ldmsvcs_info *lsp, void *buf, size_t len)
762 {
763 	ds_init_req_t *req;
764 	uint16_t new_major, new_minor;
765 	size_t msglen;
766 
767 	req = (ds_init_req_t *)buf;
768 
769 	/* sanity check the incoming message */
770 	if (len != sizeof (ds_init_req_t)) {
771 		channel_close(lsp);
772 		return;
773 	}
774 
775 	/*
776 	 * Check version info. ACK only if the major numbers exactly
777 	 * match. The service entity can retry with a new minor
778 	 * based on the response sent as part of the NACK.
779 	 */
780 	if (fds_negotiate_version(req->major_vers, &new_major, &new_minor)) {
781 		ds_hdr_t *H;
782 		ds_init_ack_t *R;
783 
784 		msglen = sizeof (ds_hdr_t) + sizeof (ds_init_ack_t);
785 		H = alloca(msglen);
786 		R = (void *)((ptrdiff_t)H + sizeof (ds_hdr_t));
787 
788 		H->msg_type = DS_INIT_ACK;
789 		H->payload_len = sizeof (ds_init_ack_t);
790 		R->minor_vers = MIN(new_minor, req->minor_vers);
791 
792 		if (fds_send(lsp, H, msglen) != 0)
793 			return;
794 
795 		(void) pthread_mutex_lock(&lsp->mt);
796 		ASSERT(lsp->fds_chan.state == CHANNEL_OPEN);
797 		lsp->fds_chan.state = CHANNEL_READY;
798 
799 		/*
800 		 * Now the channel is ready after the handshake completes.
801 		 * Reset the timeout to a smaller value for receiving messages
802 		 * from the domain services.
803 		 */
804 		lsp->cv_twait = get_smf_int_val(LDM_RUNNING_TO_PROP_NM,
805 		    0, LDM_TIMEOUT_CEILING, LDM_RUNNING_WAIT_TIME);
806 
807 		(void) pthread_mutex_unlock(&lsp->mt);
808 	} else {
809 		ds_hdr_t *H;
810 		ds_init_nack_t *R;
811 
812 		msglen = sizeof (ds_hdr_t) + sizeof (ds_init_nack_t);
813 		H = alloca(msglen);
814 		R = (void *)((ptrdiff_t)H + sizeof (ds_hdr_t));
815 
816 		H->msg_type = DS_INIT_NACK;
817 		H->payload_len = sizeof (ds_init_nack_t);
818 		R->major_vers = new_major;
819 
820 		(void) fds_send(lsp, H, msglen);
821 		/*
822 		 * do not update state; remote end may attempt to initiate
823 		 * connection with a different version
824 		 */
825 	}
826 }
827 
828 
829 /*ARGSUSED*/
830 static void
831 ds_handle_reg_req(struct ldmsvcs_info *lsp, void *buf, size_t len)
832 {
833 	ds_reg_req_t *req;
834 	char *msg;
835 	uint16_t new_major, new_minor;
836 	size_t msglen;
837 	int dup_svcreg = 0;
838 
839 	req = (ds_reg_req_t *)buf;
840 	msg = (char *)req->svc_id;
841 
842 	/*
843 	 * Service must be NULL terminated
844 	 */
845 	if (req->svc_id == NULL || strlen(req->svc_id) == 0 ||
846 	    msg[strlen(req->svc_id)] != '\0') {
847 		channel_close(lsp);
848 		return;
849 	}
850 
851 	if (fds_negotiate_version(req->major_vers, &new_major, &new_minor) &&
852 	    (dup_svcreg = fds_svc_add(lsp, req,
853 	    MIN(new_minor, req->minor_vers))) == 0) {
854 
855 		/*
856 		 * Check version info. ACK only if the major numbers
857 		 * exactly match. The service entity can retry with a new
858 		 * minor based on the response sent as part of the NACK.
859 		 */
860 		ds_hdr_t *H;
861 		ds_reg_ack_t *R;
862 
863 		msglen = sizeof (ds_hdr_t) + sizeof (ds_reg_ack_t);
864 		H = alloca(msglen);
865 		bzero(H, msglen);
866 		R = (void *)((ptrdiff_t)H + sizeof (ds_hdr_t));
867 
868 		H->msg_type = DS_REG_ACK;
869 		H->payload_len = sizeof (ds_reg_ack_t);
870 		R->svc_handle = req->svc_handle;
871 		R->minor_vers = MIN(new_minor, req->minor_vers);
872 
873 		(void) fds_send(lsp, H, msglen);
874 	} else {
875 		ds_hdr_t *H;
876 		ds_reg_nack_t *R;
877 
878 		msglen = sizeof (ds_hdr_t) + sizeof (ds_reg_nack_t);
879 		H = alloca(msglen);
880 		bzero(H, msglen);
881 		R = (void *)((ptrdiff_t)H + sizeof (ds_hdr_t));
882 
883 		H->msg_type = DS_REG_NACK;
884 		H->payload_len = sizeof (ds_reg_nack_t);
885 		R->svc_handle = req->svc_handle;
886 		R->major_vers = new_major;
887 
888 		if (dup_svcreg)
889 			R->result = DS_REG_DUP;
890 		else
891 			R->result = DS_REG_VER_NACK;
892 
893 		(void) fds_send(lsp, H, msglen);
894 	}
895 }
896 
897 
898 /*ARGSUSED*/
899 static void
900 ds_handle_unreg(struct ldmsvcs_info *lsp, void *buf, size_t len)
901 {
902 	ds_unreg_req_t *req;
903 	size_t msglen;
904 
905 	req = (ds_unreg_req_t *)buf;
906 
907 	if (fds_svc_remove(lsp, req->svc_handle) == 0) {
908 		ds_hdr_t *H;
909 		ds_unreg_ack_t *R;
910 
911 		msglen = sizeof (ds_hdr_t) + sizeof (ds_unreg_ack_t);
912 		H = alloca(msglen);
913 		R = (void *)((ptrdiff_t)H + sizeof (ds_hdr_t));
914 
915 		H->msg_type = DS_REG_ACK;
916 		H->payload_len = sizeof (ds_unreg_ack_t);
917 		R->svc_handle = req->svc_handle;
918 
919 		(void) fds_send(lsp, H, msglen);
920 	} else {
921 		ds_hdr_t *H;
922 		ds_unreg_nack_t *R;
923 
924 		msglen = sizeof (ds_hdr_t) + sizeof (ds_unreg_nack_t);
925 		H = alloca(msglen);
926 		R = (void *)((ptrdiff_t)H + sizeof (ds_hdr_t));
927 
928 		H->msg_type = DS_REG_NACK;
929 		H->payload_len = sizeof (ds_unreg_nack_t);
930 		R->svc_handle = req->svc_handle;
931 
932 		(void) fds_send(lsp, H, msglen);
933 	}
934 }
935 
936 
937 /*
938  * Message handler lookup table (v1.0 only for now) Future
939  * versions can add their own lookup table.
940  */
941 typedef void (*ds_msg_handler_t)(struct ldmsvcs_info *lsp,
942 				void *buf, size_t len);
943 
944 static const ds_msg_handler_t ds_msg_handlers[] = {
945 	ds_handle_init_req,		/* DS_INIT_REQ */
946 	ds_handle_msg_noop,		/* DS_INIT_ACK */
947 	ds_handle_msg_noop,		/* DS_INIT_NACK */
948 	ds_handle_reg_req,		/* DS_REG_REQ */
949 	ds_handle_msg_noop,		/* DS_REG_ACK */
950 	ds_handle_msg_noop,		/* DS_REG_NACK */
951 	ds_handle_unreg,		/* DS_UNREG */
952 	ds_handle_msg_noop,		/* DS_UNREG_ACK */
953 	ds_handle_msg_noop,		/* DS_UNREG_NACK */
954 	ds_handle_msg_noop,		/* DS_DATA */
955 	ds_handle_msg_noop		/* DS_NACK */
956 };
957 
958 
959 /*
960  * message and service internal functions
961  */
962 static void
963 fds_svc_alloc(struct ldmsvcs_info *lsp)
964 {
965 	int i;
966 	static char *name[] = { LDM_DS_NAME_CPU, LDM_DS_NAME_MEM,
967 			LDM_DS_NAME_PRI, LDM_DS_NAME_IOD, NULL };
968 
969 	(void) pthread_mutex_init(&lsp->fmas_svcs.mt, NULL);
970 	(void) pthread_cond_init(&lsp->fmas_svcs.cv, NULL);
971 
972 	for (lsp->fmas_svcs.nsvcs = 0; name[lsp->fmas_svcs.nsvcs] != NULL;
973 	    lsp->fmas_svcs.nsvcs++)
974 		;
975 
976 	lsp->fmas_svcs.tbl = (fds_svc_t **)ldom_alloc(sizeof (fds_svc_t *) *
977 	    lsp->fmas_svcs.nsvcs);
978 
979 	for (i = 0; i < lsp->fmas_svcs.nsvcs; i++) {
980 		lsp->fmas_svcs.tbl[i] =
981 		    (fds_svc_t *)ldom_alloc(sizeof (fds_svc_t));
982 		bzero(lsp->fmas_svcs.tbl[i], sizeof (fds_svc_t));
983 		lsp->fmas_svcs.tbl[i]->name = name[i];
984 	}
985 }
986 
987 
988 static fds_svc_t *
989 fds_svc_lookup(struct ldmsvcs_info *lsp, char *name)
990 {
991 	struct timespec twait;
992 	fds_svc_t *svc;
993 	int i, ier;
994 
995 	if (pthread_mutex_lock(&lsp->fmas_svcs.mt) == EINVAL)
996 		return (NULL);	/* uninitialized or destroyed mutex */
997 
998 	svc = NULL;
999 	for (i = 0; i < lsp->fmas_svcs.nsvcs; i++) {
1000 		if (strcmp(lsp->fmas_svcs.tbl[i]->name, name) == 0) {
1001 			svc = lsp->fmas_svcs.tbl[i];
1002 			break;
1003 		}
1004 	}
1005 
1006 	ASSERT(svc != NULL);
1007 
1008 	if (svc->state == DS_SVC_INACTIVE) {
1009 		/* service is not registered */
1010 		ier = ETIMEDOUT;
1011 	} else {
1012 		ier = 0;
1013 		twait.tv_sec = time(NULL) + lsp->cv_twait;
1014 		twait.tv_nsec = 0;
1015 
1016 		while (svc->state != DS_SVC_ACTIVE && ier == 0 &&
1017 		    lsp->fds_chan.state != CHANNEL_UNUSABLE)
1018 			ier = pthread_cond_timedwait(&lsp->fmas_svcs.cv,
1019 			    &lsp->fmas_svcs.mt, &twait);
1020 
1021 		/*
1022 		 * By now, the ds service should have registered already.
1023 		 * If it does not, ldmd probably does not support this service.
1024 		 * Then mark the service state as inactive.
1025 		 */
1026 		if (ier == ETIMEDOUT) {
1027 			svc->state = DS_SVC_INACTIVE;
1028 		}
1029 	}
1030 
1031 	(void) pthread_mutex_unlock(&lsp->fmas_svcs.mt);
1032 
1033 	if (ier == 0)
1034 		return (svc);
1035 	else
1036 		return (NULL);
1037 }
1038 
1039 
1040 static uint64_t
1041 fds_svc_req_num(void)
1042 {
1043 	static uint64_t req_num = 1;
1044 
1045 	return (req_num++);
1046 }
1047 
1048 
1049 /*
1050  * return 0 if successful, 1 if otherwise
1051  */
1052 static int
1053 read_msg(struct ldmsvcs_info *lsp)
1054 {
1055 	ds_hdr_t header;
1056 	void *msg_buf;
1057 
1058 	/*
1059 	 * read the header
1060 	 */
1061 	if (read_stream(lsp->fds_chan.fd, &header, sizeof (ds_hdr_t)) != 0)
1062 		return (1);
1063 
1064 	if (header.msg_type >=
1065 	    sizeof (ds_msg_handlers) / sizeof (ds_msg_handler_t))
1066 		return (1);
1067 
1068 	/*
1069 	 * handle data as a special case
1070 	 */
1071 	if (header.msg_type == 9)
1072 		return (poller_handle_data(lsp->fds_chan.fd,
1073 		    header.payload_len));
1074 
1075 	/*
1076 	 * all other types of messages should be small
1077 	 */
1078 	ASSERT(header.payload_len < 1024);
1079 	msg_buf = alloca(header.payload_len);
1080 
1081 	/*
1082 	 * read the payload
1083 	 */
1084 	if (read_stream(lsp->fds_chan.fd, msg_buf, header.payload_len) != 0)
1085 		return (1);
1086 
1087 	(*ds_msg_handlers[header.msg_type])(lsp, msg_buf, header.payload_len);
1088 
1089 	return (0);
1090 }
1091 
1092 
1093 /*
1094  * return values:
1095  *  0 - success
1096  *  1 - problem with opening the channel
1097  *  2 - channed not opened; request to exit has been detected
1098  */
1099 static int
1100 channel_openreset(struct ldmsvcs_info *lsp)
1101 {
1102 	int ier;
1103 
1104 	ier = pthread_mutex_lock(&lsp->mt);
1105 
1106 	if (ier == EINVAL || lsp->fds_chan.state == CHANNEL_EXIT ||
1107 	    lsp->fds_chan.state == CHANNEL_UNUSABLE) {
1108 		(void) pthread_mutex_unlock(&lsp->mt);
1109 		return (2);
1110 	}
1111 
1112 	if (lsp->fds_chan.state == CHANNEL_UNINITIALIZED ||
1113 	    lsp->fds_chan.state == CHANNEL_CLOSED) {
1114 		(void) pthread_cond_broadcast(&lsp->cv);
1115 
1116 		if ((lsp->fds_chan.fd = open(FDS_VLDC, O_RDWR)) < 0) {
1117 			lsp->fds_chan.state = CHANNEL_UNUSABLE;
1118 			lsp->cv_twait = get_smf_int_val(LDM_RUNNING_TO_PROP_NM,
1119 			    0, LDM_TIMEOUT_CEILING, LDM_RUNNING_WAIT_TIME);
1120 			(void) pthread_mutex_unlock(&lsp->mt);
1121 			(void) pthread_cond_broadcast(&lsp->fmas_svcs.cv);
1122 
1123 			return (2);
1124 		} else {
1125 			vldc_opt_op_t op;
1126 
1127 			op.op_sel = VLDC_OP_SET;
1128 			op.opt_sel = VLDC_OPT_MODE;
1129 			op.opt_val = LDC_MODE_RELIABLE;
1130 
1131 			if (ioctl(lsp->fds_chan.fd, VLDC_IOCTL_OPT_OP,
1132 			    &op) != 0) {
1133 				(void) close(lsp->fds_chan.fd);
1134 				(void) pthread_mutex_unlock(&lsp->mt);
1135 				return (1);
1136 			}
1137 		}
1138 		lsp->fds_chan.state = CHANNEL_OPEN;
1139 	}
1140 
1141 	if (lsp->fds_chan.state == CHANNEL_OPEN) {
1142 		/*
1143 		 * reset various channel parameters
1144 		 */
1145 		lsp->fds_chan.ver.major = 0;
1146 		lsp->fds_chan.ver.minor = 0;
1147 		fds_svc_reset(lsp, -1);
1148 	}
1149 	(void) pthread_mutex_unlock(&lsp->mt);
1150 
1151 	return (0);
1152 }
1153 
1154 
1155 static void
1156 channel_fini(void)
1157 {
1158 	int i;
1159 	struct ldmsvcs_info *lsp;
1160 
1161 	/*
1162 	 * End the poller thread
1163 	 */
1164 	poller_fini();
1165 
1166 	if ((lsp = channel_init(NULL)) == NULL)
1167 		return;
1168 
1169 	(void) pthread_mutex_lock(&lsp->mt);
1170 
1171 	lsp->fds_chan.state = CHANNEL_EXIT;
1172 	(void) close(lsp->fds_chan.fd);
1173 
1174 	(void) pthread_mutex_unlock(&lsp->mt);
1175 
1176 	/* Free the ldom service structure */
1177 	for (i = 0; i < lsp->fmas_svcs.nsvcs; i++) {
1178 		ldom_free(lsp->fmas_svcs.tbl[i], sizeof (fds_svc_t));
1179 	}
1180 	ldom_free(lsp->fmas_svcs.tbl,
1181 	    lsp->fmas_svcs.nsvcs * sizeof (fds_svc_t *));
1182 	ldom_free(lsp, sizeof (struct ldmsvcs_info));
1183 }
1184 
1185 
1186 static struct ldmsvcs_info *
1187 channel_init(struct ldom_hdl *lhp)
1188 {
1189 	static pthread_mutex_t mt = PTHREAD_MUTEX_INITIALIZER;
1190 	static pthread_cond_t cv = PTHREAD_COND_INITIALIZER;
1191 	static struct ldmsvcs_info *root = NULL;
1192 	static int busy_init = 0;
1193 
1194 	struct timespec twait;
1195 	int expired;
1196 
1197 	(void) pthread_mutex_lock(&mt);
1198 
1199 	while (busy_init == 1)
1200 		(void) pthread_cond_wait(&cv, &mt);
1201 
1202 	if (root != NULL || (lhp == NULL && root == NULL)) {
1203 		(void) pthread_mutex_unlock(&mt);
1204 		return (root);
1205 	}
1206 
1207 	/*
1208 	 * get to this point if we need to open the channel
1209 	 */
1210 	busy_init = 1;
1211 	(void) pthread_mutex_unlock(&mt);
1212 
1213 	root = (struct ldmsvcs_info *)
1214 	    ldom_alloc(sizeof (struct ldmsvcs_info));
1215 	bzero(root, sizeof (struct ldmsvcs_info));
1216 
1217 	root->fds_chan.state = CHANNEL_UNINITIALIZED;
1218 	root->cv_twait = get_smf_int_val(LDM_INIT_TO_PROP_NM,
1219 	    0, LDM_TIMEOUT_CEILING, LDM_INIT_WAIT_TIME);
1220 
1221 	if (pthread_mutex_init(&root->mt, NULL) != 0 ||
1222 	    pthread_cond_init(&root->cv, NULL) != 0) {
1223 		ldom_free(root, sizeof (struct ldmsvcs_info));
1224 		return (NULL);
1225 	}
1226 
1227 	fds_svc_alloc(root);
1228 	fds_svc_reset(root, -1);
1229 
1230 	(void) poller_init(root);
1231 
1232 	expired = 0;
1233 	twait.tv_sec = time(NULL) + 10;
1234 	twait.tv_nsec = 0;
1235 
1236 	(void) pthread_mutex_lock(&root->mt);
1237 
1238 	/*
1239 	 * wait for channel to become uninitialized.  this should be quick.
1240 	 */
1241 	while (root->fds_chan.state == CHANNEL_UNINITIALIZED && expired == 0)
1242 		expired = pthread_cond_timedwait(&root->cv, &root->mt, &twait);
1243 
1244 	if (root->fds_chan.state == CHANNEL_UNUSABLE)
1245 		expired = 1;
1246 
1247 	(void) pthread_mutex_unlock(&root->mt);
1248 
1249 	(void) pthread_mutex_lock(&mt);
1250 	busy_init = 0;
1251 	(void) pthread_mutex_unlock(&mt);
1252 	(void) pthread_cond_broadcast(&cv);
1253 
1254 	(void) atexit(channel_fini);
1255 
1256 	if (expired == 0)
1257 		return (root);
1258 	else
1259 		return (NULL);
1260 }
1261 
1262 
1263 static int
1264 sendrecv(struct ldom_hdl *lhp, uint64_t req_num,
1265 	void *msg, size_t msglen, ds_svc_hdl_t *svc_hdl, char *svcname,
1266 	void **resp, size_t *resplen)
1267 {
1268 	struct ldmsvcs_info *lsp;
1269 	fds_svc_t *svc;
1270 	int maxretries, index, i, ier;
1271 
1272 	lsp = lhp->lsinfo;
1273 	i = 0;
1274 	maxretries = 1;
1275 
1276 	do {
1277 		/*
1278 		 * if any of the calls in this loop fail, retry some number
1279 		 * of times before giving up.
1280 		 */
1281 		if ((svc = fds_svc_lookup(lsp, svcname)) == NULL) {
1282 			(void) pthread_mutex_lock(&lsp->mt);
1283 
1284 			if (lsp->fds_chan.state != CHANNEL_READY)
1285 				ier = ETIMEDOUT;	/* channel not ready */
1286 			else
1287 				ier = ENOTSUP;		/* service not ready */
1288 
1289 			(void) pthread_mutex_unlock(&lsp->mt);
1290 
1291 			continue;
1292 		} else {
1293 			ier = 0;
1294 			*svc_hdl = svc->hdl;
1295 		}
1296 
1297 		index = poller_add_pending(req_num);
1298 
1299 		if ((ier = fds_send(lsp, msg, msglen)) != 0 ||
1300 		    (ier = poller_recv_data(lhp, req_num, index, resp,
1301 		    resplen)) != 0)
1302 			poller_delete_pending(req_num, index);
1303 
1304 	} while (i++ < maxretries && ier != 0);
1305 
1306 	ASSERT(ier == 0 || ier == ETIMEDOUT || ier == ENOTSUP);
1307 
1308 	return (ier);
1309 }
1310 
1311 
1312 /*
1313  * input:
1314  *   msg_type - requested operation: FMA_CPU_REQ_STATUS or FMA_CPU_REQ_OFFLINE
1315  *   cpuid - physical cpu id
1316  *
1317  * normal return values:
1318  *   P_OFFLINE - cpu is offline
1319  *   P_ONLINE - cpu is online
1320  *
1321  * abnormal return values:
1322  *   ETIMEDOUT - LDOM manager is not responding
1323  *   ENOTSUP - LDOM service for cpu offlining/status is not available
1324  *   ENOMSG - got an unexpected response from the LDOM cpu service
1325  */
1326 static int
1327 cpu_request(struct ldom_hdl *lhp, uint32_t msg_type, uint32_t cpuid)
1328 {
1329 	ds_hdr_t *H;
1330 	ds_data_handle_t *D;
1331 	fma_cpu_service_req_t *R;
1332 
1333 	char *svcname = LDM_DS_NAME_CPU;
1334 	fma_cpu_resp_t *respmsg;
1335 	void *resp;
1336 	size_t resplen, reqmsglen;
1337 	int rc;
1338 
1339 	if (lhp->lsinfo == NULL)
1340 		return (ENOMSG);
1341 
1342 	reqmsglen = sizeof (ds_hdr_t) + sizeof (ds_data_handle_t) +
1343 	    sizeof (fma_cpu_service_req_t);
1344 
1345 	H = lhp->allocp(reqmsglen);
1346 	D = (void *)((ptrdiff_t)H + sizeof (ds_hdr_t));
1347 	R = (void *)((ptrdiff_t)D + sizeof (ds_data_handle_t));
1348 
1349 	H->msg_type = DS_DATA;
1350 	H->payload_len = sizeof (ds_data_handle_t) +
1351 	    sizeof (fma_cpu_service_req_t);
1352 
1353 	R->req_num = fds_svc_req_num();
1354 	R->msg_type = msg_type;
1355 	R->cpu_id = cpuid;
1356 
1357 	if ((rc = sendrecv(lhp, R->req_num, H, reqmsglen,
1358 	    &D->svc_handle, svcname, &resp, &resplen)) != 0) {
1359 		lhp->freep(H, reqmsglen);
1360 		return (rc);
1361 	}
1362 
1363 	lhp->freep(H, reqmsglen);
1364 
1365 	ASSERT(resplen == sizeof (fma_cpu_resp_t));
1366 	respmsg = (fma_cpu_resp_t *)resp;
1367 
1368 	rc = ENOMSG;
1369 	if (respmsg->result == FMA_CPU_RESP_OK) {
1370 		if (respmsg->status == FMA_CPU_STAT_ONLINE)
1371 			rc = P_ONLINE;
1372 		else if (respmsg->status == FMA_CPU_STAT_OFFLINE)
1373 			rc = P_OFFLINE;
1374 	} else {
1375 		if (msg_type == FMA_CPU_REQ_OFFLINE &&
1376 		    respmsg->status == FMA_CPU_STAT_OFFLINE)
1377 			rc = P_OFFLINE;
1378 	}
1379 
1380 	lhp->freep(resp, resplen);
1381 
1382 	return (rc);
1383 }
1384 
1385 
1386 /*
1387  * input:
1388  *   msg_type - requested operation: FMA_MEM_REQ_STATUS or FMA_MEM_REQ_RETIRE
1389  *   pa - starting address of memory page
1390  *   pgsize - memory page size in bytes
1391  *
1392  * normal return values for msg_type == FMA_MEM_REQ_STATUS:
1393  *   0 - page is retired
1394  *   EAGAIN - page is scheduled for retirement
1395  *   EIO - page not scheduled for retirement
1396  *   EINVAL - error
1397  *
1398  * normal return values for msg_type == FMA_MEM_REQ_RETIRE:
1399  *   0 - success in retiring page
1400  *   EIO - page is already retired
1401  *   EAGAIN - page is scheduled for retirement
1402  *   EINVAL - error
1403  *
1404  * abnormal return values (regardless of msg_type)
1405  *   ETIMEDOUT - LDOM manager is not responding
1406  *   ENOTSUP - LDOM service for cpu offlining/status is not available
1407  *   ENOMSG - got an unexpected response from the LDOM cpu service
1408  */
1409 static int
1410 mem_request(struct ldom_hdl *lhp, uint32_t msg_type, uint64_t pa,
1411 	    uint64_t pgsize)
1412 {
1413 	ds_hdr_t *H;
1414 	ds_data_handle_t *D;
1415 	fma_mem_service_req_t *R;
1416 
1417 	char *svcname = LDM_DS_NAME_MEM;
1418 	fma_mem_resp_t *respmsg;
1419 	void *resp;
1420 	size_t resplen, reqmsglen;
1421 	int rc;
1422 
1423 	if (lhp->lsinfo == NULL)
1424 		return (ENOMSG);
1425 
1426 	reqmsglen = sizeof (ds_hdr_t) + sizeof (ds_data_handle_t) +
1427 	    sizeof (fma_mem_service_req_t);
1428 
1429 	H = lhp->allocp(reqmsglen);
1430 	bzero(H, reqmsglen);
1431 	D = (void *)((ptrdiff_t)H + sizeof (ds_hdr_t));
1432 	R = (void *)((ptrdiff_t)D + sizeof (ds_data_handle_t));
1433 
1434 	H->msg_type = DS_DATA;
1435 	H->payload_len = sizeof (ds_data_handle_t) +
1436 	    sizeof (fma_mem_service_req_t);
1437 
1438 	R->req_num = fds_svc_req_num();
1439 	R->msg_type = msg_type;
1440 	R->real_addr = pa;
1441 	R->length = pgsize;
1442 
1443 	if ((rc = sendrecv(lhp, R->req_num, H, reqmsglen,
1444 	    &D->svc_handle, svcname, &resp, &resplen)) != 0) {
1445 		lhp->freep(H, reqmsglen);
1446 		return (rc);
1447 	}
1448 
1449 	lhp->freep(H, reqmsglen);
1450 
1451 	ASSERT(resplen == sizeof (fma_mem_resp_t));
1452 	respmsg = (fma_mem_resp_t *)resp;
1453 
1454 	rc = ENOMSG;
1455 	if (msg_type == FMA_MEM_REQ_STATUS) {
1456 		if (respmsg->result == FMA_MEM_RESP_OK) {
1457 			if (respmsg->status == FMA_MEM_STAT_RETIRED)
1458 				rc = 0;		/* page is retired */
1459 			else if (respmsg->status == FMA_MEM_STAT_NOTRETIRED)
1460 				rc = EIO;	/* page is not scheduled */
1461 		} else if (respmsg->result == FMA_MEM_RESP_FAILURE) {
1462 			if (respmsg->status == FMA_MEM_STAT_NOTRETIRED)
1463 				rc = EAGAIN;	/* page is scheduled */
1464 			else if (respmsg->status == FMA_MEM_STAT_ILLEGAL)
1465 				rc = EINVAL;
1466 		}
1467 	} else if (msg_type == FMA_MEM_REQ_RETIRE) {
1468 		if (respmsg->result == FMA_MEM_RESP_OK) {
1469 			if (respmsg->status == FMA_MEM_STAT_RETIRED)
1470 				rc = 0;		/* is successfully retired */
1471 		} else if (respmsg->result == FMA_MEM_RESP_FAILURE) {
1472 			if (respmsg->status == FMA_MEM_STAT_RETIRED)
1473 				rc = EIO;	/* is already retired */
1474 			else if (respmsg->status == FMA_MEM_STAT_NOTRETIRED)
1475 				rc = EAGAIN;	/* is scheduled to retire */
1476 			else if (respmsg->status == FMA_MEM_STAT_ILLEGAL)
1477 				rc = EINVAL;
1478 		}
1479 	} else if (msg_type == FMA_MEM_REQ_RESURRECT) {
1480 		if (respmsg->result == FMA_MEM_RESP_OK) {
1481 			if (respmsg->status == FMA_MEM_STAT_NOTRETIRED)
1482 				rc = 0;		/* is successfully unretired */
1483 		} if (respmsg->result == FMA_MEM_RESP_FAILURE) {
1484 			if (respmsg->status == FMA_MEM_STAT_RETIRED)
1485 				rc = EAGAIN; 	/* page couldn't be locked */
1486 			else if (respmsg->status == FMA_MEM_STAT_NOTRETIRED)
1487 				rc = EIO;	/* page isn't retired already */
1488 			else if (respmsg->status == FMA_MEM_STAT_ILLEGAL)
1489 				rc = EINVAL;
1490 		}
1491 	}
1492 
1493 	lhp->freep(resp, resplen);
1494 
1495 	return (rc);
1496 }
1497 
1498 
1499 /*
1500  * APIs
1501  */
1502 int
1503 ldmsvcs_check_channel(void)
1504 {
1505 	struct stat buf;
1506 
1507 	if (stat(FDS_VLDC, &buf) == 0)
1508 		return (0);	/* vldc exists */
1509 	else if (errno == ENOENT || errno == ENOTDIR)
1510 		return (1);	/* vldc does not exist */
1511 	else
1512 		return (-1);	/* miscellaneous error */
1513 }
1514 
1515 
1516 /*ARGSUSED*/
1517 void
1518 ldmsvcs_init(struct ldom_hdl *lhp)
1519 {
1520 	if (ldmsvcs_check_channel() != 0)
1521 		return;
1522 
1523 	lhp->lsinfo = channel_init(lhp);
1524 	poller_add_client();
1525 }
1526 
1527 
1528 /*ARGSUSED*/
1529 void
1530 ldmsvcs_fini(struct ldom_hdl *lhp)
1531 {
1532 	if (ldmsvcs_check_channel() != 0)
1533 		return;
1534 
1535 	poller_remove_client();
1536 }
1537 
1538 
1539 /*ARGSUSED*/
1540 ssize_t
1541 ldmsvcs_get_core_md(struct ldom_hdl *lhp, uint64_t **buf)
1542 {
1543 	ds_hdr_t *H;
1544 	ds_data_handle_t *D;
1545 	fma_req_pri_t *R;
1546 
1547 	char *svcname = LDM_DS_NAME_PRI;
1548 	void *resp;
1549 	size_t resplen, reqmsglen;
1550 	ssize_t buflen;
1551 	int rc;
1552 
1553 	if (lhp->lsinfo == NULL)
1554 		return (-1);
1555 
1556 	reqmsglen = sizeof (ds_hdr_t) + sizeof (ds_data_handle_t) +
1557 	    sizeof (fma_req_pri_t);
1558 
1559 	H = lhp->allocp(reqmsglen);
1560 	D = (void *)((ptrdiff_t)H + sizeof (ds_hdr_t));
1561 	R = (void *)((ptrdiff_t)D + sizeof (ds_data_handle_t));
1562 
1563 	H->msg_type = DS_DATA;
1564 	H->payload_len = sizeof (ds_data_handle_t) +
1565 	    sizeof (fma_req_pri_t);
1566 
1567 	R->req_num = fds_svc_req_num();
1568 
1569 	if ((rc = sendrecv(lhp, R->req_num, H, reqmsglen,
1570 	    &D->svc_handle, svcname, &resp, &resplen)) != 0) {
1571 		lhp->freep(H, reqmsglen);
1572 		errno = rc;
1573 		return (-1);
1574 	}
1575 
1576 	lhp->freep(H, reqmsglen);
1577 
1578 	/*
1579 	 * resp should contain the req_num immediately followed by the PRI
1580 	 * (the latter may or may not be present).  unfortunately, the
1581 	 * current compiler flags cause a warning for the following
1582 	 * definition
1583 	 *
1584 	 * typedef struct {
1585 	 *    uint64_t req_num;
1586 	 *    uint8_t pri[];
1587 	 *  } fma_pri_resp_t;
1588 	 *
1589 	 * so we do not use the struct here.
1590 	 */
1591 	if (resplen <= sizeof (uint64_t)) {
1592 		lhp->freep(resp, resplen);
1593 		if (resplen == sizeof (uint64_t))
1594 			return (0);
1595 		else
1596 			return (-1);
1597 	}
1598 
1599 	buflen = resplen - sizeof (uint64_t);
1600 	*buf = lhp->allocp(buflen);
1601 
1602 	bcopy((void *)((ptrdiff_t)resp + sizeof (uint64_t)), *buf, buflen);
1603 	lhp->freep(resp, resplen);
1604 
1605 	return (buflen);
1606 }
1607 
1608 
1609 /*
1610  * see cpu_request() for a description of return values
1611  */
1612 int
1613 ldmsvcs_cpu_req_status(struct ldom_hdl *lhp, uint32_t cpuid)
1614 {
1615 	return (cpu_request(lhp, FMA_CPU_REQ_STATUS, cpuid));
1616 }
1617 
1618 
1619 int
1620 ldmsvcs_cpu_req_offline(struct ldom_hdl *lhp, uint32_t cpuid)
1621 {
1622 	return (cpu_request(lhp, FMA_CPU_REQ_OFFLINE, cpuid));
1623 }
1624 
1625 int
1626 ldmsvcs_cpu_req_online(struct ldom_hdl *lhp, uint32_t cpuid)
1627 {
1628 	return (cpu_request(lhp, FMA_CPU_REQ_ONLINE, cpuid));
1629 }
1630 
1631 /*
1632  * see mem_request() for a description of return values
1633  */
1634 int
1635 ldmsvcs_mem_req_status(struct ldom_hdl *lhp, uint64_t pa)
1636 {
1637 	return (mem_request(lhp, FMA_MEM_REQ_STATUS, pa, getpagesize()));
1638 }
1639 
1640 int
1641 ldmsvcs_mem_req_retire(struct ldom_hdl *lhp, uint64_t pa)
1642 {
1643 	return (mem_request(lhp, FMA_MEM_REQ_RETIRE, pa, getpagesize()));
1644 }
1645 
1646 int
1647 ldmsvcs_mem_req_unretire(struct ldom_hdl *lhp, uint64_t pa)
1648 {
1649 	return (mem_request(lhp, FMA_MEM_REQ_RESURRECT, pa, getpagesize()));
1650 }
1651 
1652 int
1653 ldmsvcs_io_req_id(struct ldom_hdl *lhp, uint64_t addr, uint_t type,
1654     uint64_t *virt_addr, char *name, int name_len, uint64_t *did)
1655 {
1656 
1657 	ds_hdr_t *H;
1658 	ds_data_handle_t *D;
1659 	fma_io_req_t *R;
1660 
1661 	char *svcname = LDM_DS_NAME_IOD;
1662 	void *resp;
1663 	fma_io_resp_t *iop;
1664 	size_t resplen, reqmsglen;
1665 	int offset;
1666 	int rc;
1667 
1668 	if (lhp->lsinfo == NULL)
1669 		return (-1);
1670 
1671 	reqmsglen = sizeof (ds_hdr_t) + sizeof (ds_data_handle_t) +
1672 	    sizeof (fma_io_req_t);
1673 
1674 	H = lhp->allocp(reqmsglen);
1675 	D = (void *)((ptrdiff_t)H + sizeof (ds_hdr_t));
1676 	R = (void *)((ptrdiff_t)D + sizeof (ds_data_handle_t));
1677 
1678 	H->msg_type = DS_DATA;
1679 	H->payload_len = sizeof (ds_data_handle_t) + sizeof (fma_io_req_t);
1680 
1681 	R->req_num = fds_svc_req_num();
1682 	R->msg_type = type;
1683 	R->rsrc_address = addr;
1684 
1685 	rc = ENOMSG;
1686 	if ((rc = sendrecv(lhp, R->req_num, H, reqmsglen,
1687 	    &D->svc_handle, svcname, &resp, &resplen)) != 0) {
1688 		lhp->freep(H, reqmsglen);
1689 		return (rc);
1690 	}
1691 	lhp->freep(H, reqmsglen);
1692 
1693 	/*
1694 	 * resp should contain the req_num, status, virtual addr, domain id
1695 	 * and the domain name. The domain name may or may not be present.
1696 	 */
1697 	offset = sizeof (fma_io_resp_t);
1698 	if (resplen < offset) {
1699 		lhp->freep(resp, resplen);
1700 		return (-1);
1701 	}
1702 
1703 	iop = (fma_io_resp_t *)resp;
1704 	switch (iop->result) {
1705 	case FMA_IO_RESP_OK:
1706 		/* success */
1707 		rc = 0;
1708 		*virt_addr = iop->virt_rsrc_address;
1709 		*did = iop->domain_id;
1710 		if (name == NULL || name_len <= 0)
1711 			break;
1712 		*name = '\0';
1713 		if (resplen > offset) {
1714 			(void) strncpy(name, (char *)((ptrdiff_t)resp + offset),
1715 			    name_len);
1716 		}
1717 		break;
1718 	default:
1719 		rc = -1;
1720 		break;
1721 	}
1722 
1723 	lhp->freep(resp, resplen);
1724 	return (rc);
1725 }
1726 
1727 /* end file */
1728