xref: /freebsd/sys/dev/ipmi/ipmi.c (revision 7899f917b1c0ea178f1d2be0cfb452086d079d23)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2006 IronPort Systems Inc. <ambrisko@ironport.com>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/bus.h>
32 #include <sys/condvar.h>
33 #include <sys/conf.h>
34 #include <sys/eventhandler.h>
35 #include <sys/kernel.h>
36 #include <sys/lock.h>
37 #include <sys/malloc.h>
38 #include <sys/module.h>
39 #include <sys/mutex.h>
40 #include <sys/poll.h>
41 #include <sys/reboot.h>
42 #include <sys/rman.h>
43 #include <sys/selinfo.h>
44 #include <sys/sysctl.h>
45 #include <sys/watchdog.h>
46 
47 #ifdef LOCAL_MODULE
48 #include <ipmi.h>
49 #include <ipmivars.h>
50 #else
51 #include <sys/ipmi.h>
52 #include <dev/ipmi/ipmivars.h>
53 #endif
54 
55 #ifdef IPMICTL_SEND_COMMAND_32
56 #include <sys/abi_compat.h>
57 #endif
58 
59 /*
60  * Driver request structures are allocated on the stack via alloca() to
61  * avoid calling malloc(), especially for the watchdog handler.
62  * To avoid too much stack growth, a previously allocated structure can
63  * be reused via IPMI_INIT_DRIVER_REQUEST(), but the caller should ensure
64  * that there is adequate reply/request space in the original allocation.
65  */
66 #define	IPMI_INIT_DRIVER_REQUEST(req, addr, cmd, reqlen, replylen)	\
67 	bzero((req), sizeof(struct ipmi_request));			\
68 	ipmi_init_request((req), NULL, 0, (addr), (cmd), (reqlen), (replylen))
69 
70 #define	IPMI_ALLOC_DRIVER_REQUEST(req, addr, cmd, reqlen, replylen)	\
71 	(req) = __builtin_alloca(sizeof(struct ipmi_request) +		\
72 	    (reqlen) + (replylen));					\
73 	IPMI_INIT_DRIVER_REQUEST((req), (addr), (cmd), (reqlen),	\
74 	    (replylen))
75 
76 static d_ioctl_t ipmi_ioctl;
77 static d_poll_t ipmi_poll;
78 static d_open_t ipmi_open;
79 static void ipmi_dtor(void *arg);
80 
81 int ipmi_attached = 0;
82 
83 static int on = 1;
84 static bool wd_in_shutdown = false;
85 static int wd_timer_actions = IPMI_SET_WD_ACTION_POWER_CYCLE;
86 static int wd_shutdown_countdown = 0; /* sec */
87 static int wd_startup_countdown = 0; /* sec */
88 static int wd_pretimeout_countdown = 120; /* sec */
89 static int cycle_wait = 10; /* sec */
90 static int wd_init_enable = 1;
91 
92 static SYSCTL_NODE(_hw, OID_AUTO, ipmi, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
93     "IPMI driver parameters");
94 SYSCTL_INT(_hw_ipmi, OID_AUTO, on, CTLFLAG_RWTUN,
95 	&on, 0, "");
96 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_init_enable, CTLFLAG_RWTUN,
97 	&wd_init_enable, 1, "Enable watchdog initialization");
98 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_timer_actions, CTLFLAG_RWTUN,
99 	&wd_timer_actions, 0,
100 	"IPMI watchdog timer actions (including pre-timeout interrupt)");
101 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_shutdown_countdown, CTLFLAG_RWTUN,
102 	&wd_shutdown_countdown, 0,
103 	"IPMI watchdog countdown for shutdown (seconds)");
104 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_startup_countdown, CTLFLAG_RDTUN,
105 	&wd_startup_countdown, 0,
106 	"IPMI watchdog countdown initialized during startup (seconds)");
107 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_pretimeout_countdown, CTLFLAG_RWTUN,
108 	&wd_pretimeout_countdown, 0,
109 	"IPMI watchdog pre-timeout countdown (seconds)");
110 SYSCTL_INT(_hw_ipmi, OID_AUTO, cycle_wait, CTLFLAG_RWTUN,
111 	&cycle_wait, 0,
112 	"IPMI power cycle on reboot delay time (seconds)");
113 
114 static struct cdevsw ipmi_cdevsw = {
115 	.d_version =    D_VERSION,
116 	.d_open =	ipmi_open,
117 	.d_ioctl =	ipmi_ioctl,
118 	.d_poll =	ipmi_poll,
119 	.d_name =	"ipmi",
120 };
121 
122 static MALLOC_DEFINE(M_IPMI, "ipmi", "ipmi");
123 
124 static int
125 ipmi_open(struct cdev *cdev, int flags, int fmt, struct thread *td)
126 {
127 	struct ipmi_device *dev;
128 	struct ipmi_softc *sc;
129 	int error;
130 
131 	if (!on)
132 		return (ENOENT);
133 
134 	/* Initialize the per file descriptor data. */
135 	dev = malloc(sizeof(struct ipmi_device), M_IPMI, M_WAITOK | M_ZERO);
136 	error = devfs_set_cdevpriv(dev, ipmi_dtor);
137 	if (error) {
138 		free(dev, M_IPMI);
139 		return (error);
140 	}
141 
142 	sc = cdev->si_drv1;
143 	TAILQ_INIT(&dev->ipmi_completed_requests);
144 	dev->ipmi_address = IPMI_BMC_SLAVE_ADDR;
145 	dev->ipmi_lun = IPMI_BMC_SMS_LUN;
146 	dev->ipmi_softc = sc;
147 	IPMI_LOCK(sc);
148 	sc->ipmi_opened++;
149 	IPMI_UNLOCK(sc);
150 
151 	return (0);
152 }
153 
154 static int
155 ipmi_poll(struct cdev *cdev, int poll_events, struct thread *td)
156 {
157 	struct ipmi_device *dev;
158 	struct ipmi_softc *sc;
159 	int revents = 0;
160 
161 	if (devfs_get_cdevpriv((void **)&dev))
162 		return (0);
163 
164 	sc = cdev->si_drv1;
165 	IPMI_LOCK(sc);
166 	if (poll_events & (POLLIN | POLLRDNORM)) {
167 		if (!TAILQ_EMPTY(&dev->ipmi_completed_requests))
168 		    revents |= poll_events & (POLLIN | POLLRDNORM);
169 		if (dev->ipmi_requests == 0)
170 		    revents |= POLLERR;
171 	}
172 
173 	if (revents == 0) {
174 		if (poll_events & (POLLIN | POLLRDNORM))
175 			selrecord(td, &dev->ipmi_select);
176 	}
177 	IPMI_UNLOCK(sc);
178 
179 	return (revents);
180 }
181 
182 static void
183 ipmi_purge_completed_requests(struct ipmi_device *dev)
184 {
185 	struct ipmi_request *req;
186 
187 	while (!TAILQ_EMPTY(&dev->ipmi_completed_requests)) {
188 		req = TAILQ_FIRST(&dev->ipmi_completed_requests);
189 		TAILQ_REMOVE(&dev->ipmi_completed_requests, req, ir_link);
190 		dev->ipmi_requests--;
191 		ipmi_free_request(req);
192 	}
193 }
194 
195 static void
196 ipmi_dtor(void *arg)
197 {
198 	struct ipmi_request *req, *nreq;
199 	struct ipmi_device *dev;
200 	struct ipmi_softc *sc;
201 
202 	dev = arg;
203 	sc = dev->ipmi_softc;
204 
205 	IPMI_LOCK(sc);
206 	if (dev->ipmi_requests) {
207 		/* Throw away any pending requests for this device. */
208 		TAILQ_FOREACH_SAFE(req, &sc->ipmi_pending_requests_highpri, ir_link,
209 		    nreq) {
210 			if (req->ir_owner == dev) {
211 				TAILQ_REMOVE(&sc->ipmi_pending_requests_highpri, req,
212 				    ir_link);
213 				dev->ipmi_requests--;
214 				ipmi_free_request(req);
215 			}
216 		}
217 		TAILQ_FOREACH_SAFE(req, &sc->ipmi_pending_requests, ir_link,
218 		    nreq) {
219 			if (req->ir_owner == dev) {
220 				TAILQ_REMOVE(&sc->ipmi_pending_requests, req,
221 				    ir_link);
222 				dev->ipmi_requests--;
223 				ipmi_free_request(req);
224 			}
225 		}
226 
227 		/* Throw away any pending completed requests for this device. */
228 		ipmi_purge_completed_requests(dev);
229 
230 		/*
231 		 * If we still have outstanding requests, they must be stuck
232 		 * in an interface driver, so wait for those to drain.
233 		 */
234 		dev->ipmi_closing = 1;
235 		while (dev->ipmi_requests > 0) {
236 			msleep(&dev->ipmi_requests, &sc->ipmi_requests_lock,
237 			    PWAIT, "ipmidrain", 0);
238 			ipmi_purge_completed_requests(dev);
239 		}
240 	}
241 	sc->ipmi_opened--;
242 	IPMI_UNLOCK(sc);
243 
244 	/* Cleanup. */
245 	free(dev, M_IPMI);
246 }
247 
248 static u_char
249 ipmi_ipmb_checksum(u_char *data, int len)
250 {
251 	u_char sum = 0;
252 
253 	for (; len; len--)
254 		sum += *data++;
255 	return (-sum);
256 }
257 
258 static int
259 ipmi_ioctl(struct cdev *cdev, u_long cmd, caddr_t data,
260     int flags, struct thread *td)
261 {
262 	struct ipmi_softc *sc;
263 	struct ipmi_device *dev;
264 	struct ipmi_request *kreq;
265 	struct ipmi_req *req = (struct ipmi_req *)data;
266 	struct ipmi_recv *recv = (struct ipmi_recv *)data;
267 	struct ipmi_addr addr;
268 #ifdef IPMICTL_SEND_COMMAND_32
269 	struct ipmi_req32 *req32 = (struct ipmi_req32 *)data;
270 	struct ipmi_recv32 *recv32 = (struct ipmi_recv32 *)data;
271 	union {
272 		struct ipmi_req req;
273 		struct ipmi_recv recv;
274 	} thunk32;
275 #endif
276 	int error, len;
277 
278 	error = devfs_get_cdevpriv((void **)&dev);
279 	if (error)
280 		return (error);
281 
282 	sc = cdev->si_drv1;
283 
284 #ifdef IPMICTL_SEND_COMMAND_32
285 	/* Convert 32-bit structures to native. */
286 	switch (cmd) {
287 	case IPMICTL_SEND_COMMAND_32:
288 		req = &thunk32.req;
289 		req->addr = PTRIN(req32->addr);
290 		req->addr_len = req32->addr_len;
291 		req->msgid = req32->msgid;
292 		req->msg.netfn = req32->msg.netfn;
293 		req->msg.cmd = req32->msg.cmd;
294 		req->msg.data_len = req32->msg.data_len;
295 		req->msg.data = PTRIN(req32->msg.data);
296 		break;
297 	case IPMICTL_RECEIVE_MSG_TRUNC_32:
298 	case IPMICTL_RECEIVE_MSG_32:
299 		recv = &thunk32.recv;
300 		recv->addr = PTRIN(recv32->addr);
301 		recv->addr_len = recv32->addr_len;
302 		recv->msg.data_len = recv32->msg.data_len;
303 		recv->msg.data = PTRIN(recv32->msg.data);
304 		break;
305 	}
306 #endif
307 
308 	switch (cmd) {
309 #ifdef IPMICTL_SEND_COMMAND_32
310 	case IPMICTL_SEND_COMMAND_32:
311 #endif
312 	case IPMICTL_SEND_COMMAND:
313 		error = copyin(req->addr, &addr, sizeof(addr));
314 		if (error)
315 			return (error);
316 
317 		if (addr.addr_type == IPMI_SYSTEM_INTERFACE_ADDR_TYPE) {
318 			struct ipmi_system_interface_addr *saddr =
319 			    (struct ipmi_system_interface_addr *)&addr;
320 
321 			kreq = ipmi_alloc_request(dev, req->msgid,
322 			    IPMI_ADDR(req->msg.netfn, saddr->lun & 0x3),
323 			    req->msg.cmd, req->msg.data_len, IPMI_MAX_RX);
324 			error = copyin(req->msg.data, kreq->ir_request,
325 			    req->msg.data_len);
326 			if (error) {
327 				ipmi_free_request(kreq);
328 				return (error);
329 			}
330 			IPMI_LOCK(sc);
331 			dev->ipmi_requests++;
332 			error = sc->ipmi_enqueue_request(sc, kreq);
333 			IPMI_UNLOCK(sc);
334 			if (error)
335 				return (error);
336 			break;
337 		}
338 
339 		/* Special processing for IPMB commands */
340 		struct ipmi_ipmb_addr *iaddr = (struct ipmi_ipmb_addr *)&addr;
341 
342 		IPMI_ALLOC_DRIVER_REQUEST(kreq, IPMI_ADDR(IPMI_APP_REQUEST, 0),
343 		    IPMI_SEND_MSG, req->msg.data_len + 8, IPMI_MAX_RX);
344 		/* Construct the SEND MSG header */
345 		kreq->ir_request[0] = iaddr->channel;
346 		kreq->ir_request[1] = iaddr->slave_addr;
347 		kreq->ir_request[2] = IPMI_ADDR(req->msg.netfn, iaddr->lun);
348 		kreq->ir_request[3] =
349 		    ipmi_ipmb_checksum(&kreq->ir_request[1], 2);
350 		kreq->ir_request[4] = dev->ipmi_address;
351 		kreq->ir_request[5] = IPMI_ADDR(0, dev->ipmi_lun);
352 		kreq->ir_request[6] = req->msg.cmd;
353 		/* Copy the message data */
354 		if (req->msg.data_len > 0) {
355 			error = copyin(req->msg.data, &kreq->ir_request[7],
356 			    req->msg.data_len);
357 			if (error != 0)
358 				return (error);
359 		}
360 		kreq->ir_request[req->msg.data_len + 7] =
361 		    ipmi_ipmb_checksum(&kreq->ir_request[4],
362 		    req->msg.data_len + 3);
363 		error = ipmi_submit_driver_request(sc, kreq, MAX_TIMEOUT);
364 		if (error != 0)
365 			return (error);
366 
367 		kreq = ipmi_alloc_request(dev, req->msgid,
368 		    IPMI_ADDR(IPMI_APP_REQUEST, 0), IPMI_GET_MSG,
369 		    0, IPMI_MAX_RX);
370 		kreq->ir_ipmb = true;
371 		kreq->ir_ipmb_addr = IPMI_ADDR(req->msg.netfn, 0);
372 		kreq->ir_ipmb_command = req->msg.cmd;
373 		IPMI_LOCK(sc);
374 		dev->ipmi_requests++;
375 		error = sc->ipmi_enqueue_request(sc, kreq);
376 		IPMI_UNLOCK(sc);
377 		if (error != 0)
378 			return (error);
379 		break;
380 #ifdef IPMICTL_SEND_COMMAND_32
381 	case IPMICTL_RECEIVE_MSG_TRUNC_32:
382 	case IPMICTL_RECEIVE_MSG_32:
383 #endif
384 	case IPMICTL_RECEIVE_MSG_TRUNC:
385 	case IPMICTL_RECEIVE_MSG:
386 		error = copyin(recv->addr, &addr, sizeof(addr));
387 		if (error)
388 			return (error);
389 
390 		IPMI_LOCK(sc);
391 		kreq = TAILQ_FIRST(&dev->ipmi_completed_requests);
392 		if (kreq == NULL) {
393 			IPMI_UNLOCK(sc);
394 			return (EAGAIN);
395 		}
396 		if (kreq->ir_error != 0) {
397 			error = kreq->ir_error;
398 			TAILQ_REMOVE(&dev->ipmi_completed_requests, kreq,
399 			    ir_link);
400 			dev->ipmi_requests--;
401 			IPMI_UNLOCK(sc);
402 			ipmi_free_request(kreq);
403 			return (error);
404 		}
405 
406 		recv->recv_type = IPMI_RESPONSE_RECV_TYPE;
407 		recv->msgid = kreq->ir_msgid;
408 		if (kreq->ir_ipmb) {
409 			addr.channel = IPMI_IPMB_CHANNEL;
410 			recv->msg.netfn =
411 			    IPMI_REPLY_ADDR(kreq->ir_ipmb_addr) >> 2;
412 			recv->msg.cmd = kreq->ir_ipmb_command;
413 			/* Get the compcode of response */
414 			kreq->ir_compcode = kreq->ir_reply[6];
415 			/* Move the reply head past response header */
416 			kreq->ir_reply += 7;
417 			len = kreq->ir_replylen - 7;
418 		} else {
419 			addr.channel = IPMI_BMC_CHANNEL;
420 			recv->msg.netfn = IPMI_REPLY_ADDR(kreq->ir_addr) >> 2;
421 			recv->msg.cmd = kreq->ir_command;
422 			len = kreq->ir_replylen + 1;
423 		}
424 
425 		if (recv->msg.data_len < len &&
426 		    (cmd == IPMICTL_RECEIVE_MSG
427 #ifdef IPMICTL_RECEIVE_MSG_32
428 		    || cmd == IPMICTL_RECEIVE_MSG_32
429 #endif
430 		    )) {
431 			IPMI_UNLOCK(sc);
432 			return (EMSGSIZE);
433 		}
434 		TAILQ_REMOVE(&dev->ipmi_completed_requests, kreq, ir_link);
435 		dev->ipmi_requests--;
436 		IPMI_UNLOCK(sc);
437 		len = min(recv->msg.data_len, len);
438 		recv->msg.data_len = len;
439 		error = copyout(&addr, recv->addr,sizeof(addr));
440 		if (error == 0)
441 			error = copyout(&kreq->ir_compcode, recv->msg.data, 1);
442 		if (error == 0)
443 			error = copyout(kreq->ir_reply, recv->msg.data + 1,
444 			    len - 1);
445 		ipmi_free_request(kreq);
446 		if (error)
447 			return (error);
448 		break;
449 	case IPMICTL_SET_MY_ADDRESS_CMD:
450 		IPMI_LOCK(sc);
451 		dev->ipmi_address = *(int*)data;
452 		IPMI_UNLOCK(sc);
453 		break;
454 	case IPMICTL_GET_MY_ADDRESS_CMD:
455 		IPMI_LOCK(sc);
456 		*(int*)data = dev->ipmi_address;
457 		IPMI_UNLOCK(sc);
458 		break;
459 	case IPMICTL_SET_MY_LUN_CMD:
460 		IPMI_LOCK(sc);
461 		dev->ipmi_lun = *(int*)data & 0x3;
462 		IPMI_UNLOCK(sc);
463 		break;
464 	case IPMICTL_GET_MY_LUN_CMD:
465 		IPMI_LOCK(sc);
466 		*(int*)data = dev->ipmi_lun;
467 		IPMI_UNLOCK(sc);
468 		break;
469 	case IPMICTL_SET_GETS_EVENTS_CMD:
470 		/*
471 		device_printf(sc->ipmi_dev,
472 		    "IPMICTL_SET_GETS_EVENTS_CMD NA\n");
473 		*/
474 		break;
475 	case IPMICTL_REGISTER_FOR_CMD:
476 	case IPMICTL_UNREGISTER_FOR_CMD:
477 		return (EOPNOTSUPP);
478 	default:
479 		device_printf(sc->ipmi_dev, "Unknown IOCTL %lX\n", cmd);
480 		return (ENOIOCTL);
481 	}
482 
483 #ifdef IPMICTL_SEND_COMMAND_32
484 	/* Update changed fields in 32-bit structures. */
485 	switch (cmd) {
486 	case IPMICTL_RECEIVE_MSG_TRUNC_32:
487 	case IPMICTL_RECEIVE_MSG_32:
488 		recv32->recv_type = recv->recv_type;
489 		recv32->msgid = recv->msgid;
490 		recv32->msg.netfn = recv->msg.netfn;
491 		recv32->msg.cmd = recv->msg.cmd;
492 		recv32->msg.data_len = recv->msg.data_len;
493 		break;
494 	}
495 #endif
496 	return (0);
497 }
498 
499 /*
500  * Request management.
501  */
502 
503 __inline void
504 ipmi_init_request(struct ipmi_request *req, struct ipmi_device *dev, long msgid,
505     uint8_t addr, uint8_t command, size_t requestlen, size_t replylen)
506 {
507 
508 	req->ir_owner = dev;
509 	req->ir_msgid = msgid;
510 	req->ir_addr = addr;
511 	req->ir_command = command;
512 	if (requestlen) {
513 		req->ir_request = (char *)&req[1];
514 		req->ir_requestlen = requestlen;
515 	}
516 	if (replylen) {
517 		req->ir_reply = (char *)&req[1] + requestlen;
518 		req->ir_replybuflen = replylen;
519 	}
520 }
521 
522 /* Allocate a new request with request and reply buffers. */
523 struct ipmi_request *
524 ipmi_alloc_request(struct ipmi_device *dev, long msgid, uint8_t addr,
525     uint8_t command, size_t requestlen, size_t replylen)
526 {
527 	struct ipmi_request *req;
528 
529 	req = malloc(sizeof(struct ipmi_request) + requestlen + replylen,
530 	    M_IPMI, M_WAITOK | M_ZERO);
531 	ipmi_init_request(req, dev, msgid, addr, command, requestlen, replylen);
532 	return (req);
533 }
534 
535 /* Free a request no longer in use. */
536 void
537 ipmi_free_request(struct ipmi_request *req)
538 {
539 
540 	free(req, M_IPMI);
541 }
542 
543 /* Store a processed request on the appropriate completion queue. */
544 void
545 ipmi_complete_request(struct ipmi_softc *sc, struct ipmi_request *req)
546 {
547 	struct ipmi_device *dev;
548 
549 	IPMI_LOCK_ASSERT(sc);
550 
551 	/*
552 	 * Anonymous requests (from inside the driver) always have a
553 	 * waiter that we awaken.
554 	 */
555 	if (req->ir_owner == NULL)
556 		wakeup(req);
557 	else {
558 		dev = req->ir_owner;
559 		TAILQ_INSERT_TAIL(&dev->ipmi_completed_requests, req, ir_link);
560 		selwakeup(&dev->ipmi_select);
561 		if (dev->ipmi_closing)
562 			wakeup(&dev->ipmi_requests);
563 	}
564 }
565 
566 /* Perform an internal driver request. */
567 int
568 ipmi_submit_driver_request(struct ipmi_softc *sc, struct ipmi_request *req,
569     int timo)
570 {
571 
572 	return (sc->ipmi_driver_request(sc, req, timo));
573 }
574 
575 /*
576  * Helper routine for polled system interfaces that use
577  * ipmi_polled_enqueue_request() to queue requests.  This request
578  * waits until there is a pending request and then returns the first
579  * request.  If the driver is shutting down, it returns NULL.
580  */
581 struct ipmi_request *
582 ipmi_dequeue_request(struct ipmi_softc *sc)
583 {
584 	struct ipmi_request *req;
585 
586 	IPMI_LOCK_ASSERT(sc);
587 
588 	while (!sc->ipmi_detaching && TAILQ_EMPTY(&sc->ipmi_pending_requests) &&
589 	    TAILQ_EMPTY(&sc->ipmi_pending_requests_highpri))
590 		cv_wait(&sc->ipmi_request_added, &sc->ipmi_requests_lock);
591 	if (sc->ipmi_detaching)
592 		return (NULL);
593 
594 	req = TAILQ_FIRST(&sc->ipmi_pending_requests_highpri);
595 	if (req != NULL)
596 		TAILQ_REMOVE(&sc->ipmi_pending_requests_highpri, req, ir_link);
597 	else {
598 		req = TAILQ_FIRST(&sc->ipmi_pending_requests);
599 		TAILQ_REMOVE(&sc->ipmi_pending_requests, req, ir_link);
600 	}
601 	return (req);
602 }
603 
604 /* Default implementation of ipmi_enqueue_request() for polled interfaces. */
605 int
606 ipmi_polled_enqueue_request(struct ipmi_softc *sc, struct ipmi_request *req)
607 {
608 
609 	IPMI_LOCK_ASSERT(sc);
610 
611 	TAILQ_INSERT_TAIL(&sc->ipmi_pending_requests, req, ir_link);
612 	cv_signal(&sc->ipmi_request_added);
613 	return (0);
614 }
615 
616 int
617 ipmi_polled_enqueue_request_highpri(struct ipmi_softc *sc, struct ipmi_request *req)
618 {
619 
620 	IPMI_LOCK_ASSERT(sc);
621 
622 	TAILQ_INSERT_TAIL(&sc->ipmi_pending_requests_highpri, req, ir_link);
623 	cv_signal(&sc->ipmi_request_added);
624 	return (0);
625 }
626 
627 /*
628  * Watchdog event handler.
629  */
630 
631 static int
632 ipmi_reset_watchdog(struct ipmi_softc *sc)
633 {
634 	struct ipmi_request *req;
635 	int error;
636 
637 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
638 	    IPMI_RESET_WDOG, 0, 0);
639 	error = ipmi_submit_driver_request(sc, req, 0);
640 	if (error) {
641 		device_printf(sc->ipmi_dev, "Failed to reset watchdog\n");
642 	} else if (req->ir_compcode == 0x80) {
643 		error = ENOENT;
644 	} else if (req->ir_compcode != 0) {
645 		device_printf(sc->ipmi_dev, "Watchdog reset returned 0x%x\n",
646 		    req->ir_compcode);
647 		error = EINVAL;
648 	}
649 	return (error);
650 }
651 
652 static int
653 ipmi_set_watchdog(struct ipmi_softc *sc, unsigned int sec)
654 {
655 	struct ipmi_request *req;
656 	int error;
657 
658 	if (sec > 0xffff / 10)
659 		return (EINVAL);
660 
661 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
662 	    IPMI_SET_WDOG, 6, 0);
663 	if (sec) {
664 		req->ir_request[0] = IPMI_SET_WD_TIMER_DONT_STOP
665 		    | IPMI_SET_WD_TIMER_SMS_OS;
666 		req->ir_request[1] = (wd_timer_actions & 0xff);
667 		req->ir_request[2] = min(0xff,
668 		    min(wd_pretimeout_countdown, (sec + 2) / 4));
669 		req->ir_request[3] = 0;	/* Timer use */
670 		req->ir_request[4] = (sec * 10) & 0xff;
671 		req->ir_request[5] = (sec * 10) >> 8;
672 	} else {
673 		req->ir_request[0] = IPMI_SET_WD_TIMER_SMS_OS;
674 		req->ir_request[1] = 0;
675 		req->ir_request[2] = 0;
676 		req->ir_request[3] = 0;	/* Timer use */
677 		req->ir_request[4] = 0;
678 		req->ir_request[5] = 0;
679 	}
680 	error = ipmi_submit_driver_request(sc, req, 0);
681 	if (error) {
682 		device_printf(sc->ipmi_dev, "Failed to set watchdog\n");
683 	} else if (req->ir_compcode != 0) {
684 		device_printf(sc->ipmi_dev, "Watchdog set returned 0x%x\n",
685 		    req->ir_compcode);
686 		error = EINVAL;
687 	}
688 	return (error);
689 }
690 
691 static void
692 ipmi_wd_event(void *arg, unsigned int cmd, int *error)
693 {
694 	struct ipmi_softc *sc = arg;
695 	unsigned int timeout;
696 	int e;
697 
698 	/* Ignore requests while disabled. */
699 	if (!on)
700 		return;
701 
702 	/*
703 	 * To prevent infinite hangs, we don't let anyone pat or change
704 	 * the watchdog when we're shutting down. (See ipmi_shutdown_event().)
705 	 * However, we do want to keep patting the watchdog while we are doing
706 	 * a coredump.
707 	 */
708 	if (wd_in_shutdown) {
709 		if (dumping && sc->ipmi_watchdog_active)
710 			ipmi_reset_watchdog(sc);
711 		return;
712 	}
713 
714 	cmd &= WD_INTERVAL;
715 	if (cmd > 0 && cmd <= 63) {
716 		timeout = ((uint64_t)1 << cmd) / 1000000000;
717 		if (timeout == 0)
718 			timeout = 1;
719 		if (timeout != sc->ipmi_watchdog_active ||
720 		    wd_timer_actions != sc->ipmi_watchdog_actions ||
721 		    wd_pretimeout_countdown != sc->ipmi_watchdog_pretimeout) {
722 			e = ipmi_set_watchdog(sc, timeout);
723 			if (e == 0) {
724 				sc->ipmi_watchdog_active = timeout;
725 				sc->ipmi_watchdog_actions = wd_timer_actions;
726 				sc->ipmi_watchdog_pretimeout = wd_pretimeout_countdown;
727 			} else {
728 				(void)ipmi_set_watchdog(sc, 0);
729 				sc->ipmi_watchdog_active = 0;
730 				sc->ipmi_watchdog_actions = 0;
731 				sc->ipmi_watchdog_pretimeout = 0;
732 			}
733 		}
734 		if (sc->ipmi_watchdog_active != 0) {
735 			e = ipmi_reset_watchdog(sc);
736 			if (e == 0) {
737 				*error = 0;
738 			} else {
739 				(void)ipmi_set_watchdog(sc, 0);
740 				sc->ipmi_watchdog_active = 0;
741 				sc->ipmi_watchdog_actions = 0;
742 				sc->ipmi_watchdog_pretimeout = 0;
743 			}
744 		}
745 	} else if (atomic_readandclear_int(&sc->ipmi_watchdog_active) != 0) {
746 		sc->ipmi_watchdog_actions = 0;
747 		sc->ipmi_watchdog_pretimeout = 0;
748 
749 		e = ipmi_set_watchdog(sc, 0);
750 		if (e != 0 && cmd == 0)
751 			*error = EOPNOTSUPP;
752 	}
753 }
754 
755 static void
756 ipmi_shutdown_event(void *arg, int howto)
757 {
758 	struct ipmi_softc *sc = arg;
759 
760 	/* Ignore event if disabled. */
761 	if (!on)
762 		return;
763 
764 	/*
765 	 * Positive wd_shutdown_countdown value will re-arm watchdog;
766 	 * Zero value in wd_shutdown_countdown will disable watchdog;
767 	 * Negative value in wd_shutdown_countdown will keep existing state;
768 	 *
769 	 * System halt is a special case of shutdown where wd_shutdown_countdown
770 	 * is ignored and watchdog is disabled to ensure that the system remains
771 	 * halted as requested.
772 	 *
773 	 * Revert to using a power cycle to ensure that the watchdog will
774 	 * do something useful here.  Having the watchdog send an NMI
775 	 * instead is useless during shutdown, and might be ignored if an
776 	 * NMI already triggered.
777 	 */
778 
779 	wd_in_shutdown = true;
780 	if (wd_shutdown_countdown == 0 || (howto & RB_HALT) != 0) {
781 		/* disable watchdog */
782 		ipmi_set_watchdog(sc, 0);
783 		sc->ipmi_watchdog_active = 0;
784 	} else if (wd_shutdown_countdown > 0) {
785 		/* set desired action and time, and, reset watchdog */
786 		wd_timer_actions = IPMI_SET_WD_ACTION_POWER_CYCLE;
787 		ipmi_set_watchdog(sc, wd_shutdown_countdown);
788 		sc->ipmi_watchdog_active = wd_shutdown_countdown;
789 		ipmi_reset_watchdog(sc);
790 	}
791 }
792 
793 static void
794 ipmi_power_cycle(void *arg, int howto)
795 {
796 	struct ipmi_softc *sc = arg;
797 	struct ipmi_request *req;
798 
799 	/*
800 	 * Ignore everything except power cycling requests
801 	 */
802 	if ((howto & RB_POWERCYCLE) == 0)
803 		return;
804 
805 	device_printf(sc->ipmi_dev, "Power cycling using IPMI\n");
806 
807 	/*
808 	 * Send a CHASSIS_CONTROL command to the CHASSIS device, subcommand 2
809 	 * as described in IPMI v2.0 spec section 28.3.
810 	 */
811 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_CHASSIS_REQUEST, 0),
812 	    IPMI_CHASSIS_CONTROL, 1, 0);
813 	req->ir_request[0] = IPMI_CC_POWER_CYCLE;
814 
815 	ipmi_submit_driver_request(sc, req, MAX_TIMEOUT);
816 
817 	if (req->ir_error != 0 || req->ir_compcode != 0) {
818 		device_printf(sc->ipmi_dev, "Power cycling via IPMI failed code %#x %#x\n",
819 		    req->ir_error, req->ir_compcode);
820 		return;
821 	}
822 
823 	/*
824 	 * BMCs are notoriously slow, give it cycle_wait seconds for the power
825 	 * down leg of the power cycle. If that fails, fallback to the next
826 	 * hanlder in the shutdown_final chain and/or the platform failsafe.
827 	 */
828 	DELAY(cycle_wait * 1000 * 1000);
829 	device_printf(sc->ipmi_dev, "Power cycling via IPMI timed out\n");
830 }
831 
832 static void
833 ipmi_startup(void *arg)
834 {
835 	struct ipmi_softc *sc = arg;
836 	struct ipmi_request *req;
837 	device_t dev;
838 	int error, i;
839 
840 	config_intrhook_disestablish(&sc->ipmi_ich);
841 	dev = sc->ipmi_dev;
842 
843 	/* Initialize interface-independent state. */
844 	mtx_init(&sc->ipmi_requests_lock, "ipmi requests", NULL, MTX_DEF);
845 	mtx_init(&sc->ipmi_io_lock, "ipmi io", NULL, MTX_DEF);
846 	cv_init(&sc->ipmi_request_added, "ipmireq");
847 	TAILQ_INIT(&sc->ipmi_pending_requests_highpri);
848 	TAILQ_INIT(&sc->ipmi_pending_requests);
849 
850 	/* Initialize interface-dependent state. */
851 	error = sc->ipmi_startup(sc);
852 	if (error) {
853 		device_printf(dev, "Failed to initialize interface: %d\n",
854 		    error);
855 		return;
856 	}
857 
858 	/* Send a GET_DEVICE_ID request. */
859 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
860 	    IPMI_GET_DEVICE_ID, 0, 15);
861 
862 	error = ipmi_submit_driver_request(sc, req, MAX_TIMEOUT);
863 	if (error == EWOULDBLOCK) {
864 		device_printf(dev, "Timed out waiting for GET_DEVICE_ID\n");
865 		return;
866 	} else if (error) {
867 		device_printf(dev, "Failed GET_DEVICE_ID: %d\n", error);
868 		return;
869 	} else if (req->ir_compcode != 0) {
870 		device_printf(dev,
871 		    "Bad completion code for GET_DEVICE_ID: %d\n",
872 		    req->ir_compcode);
873 		return;
874 	} else if (req->ir_replylen < 5) {
875 		device_printf(dev, "Short reply for GET_DEVICE_ID: %d\n",
876 		    req->ir_replylen);
877 		return;
878 	}
879 
880 	device_printf(dev, "IPMI device rev. %d, firmware rev. %d.%d%d, "
881 	    "version %d.%d, device support mask %#x\n",
882 	    req->ir_reply[1] & 0x0f,
883 	    req->ir_reply[2] & 0x7f, req->ir_reply[3] >> 4, req->ir_reply[3] & 0x0f,
884 	    req->ir_reply[4] & 0x0f, req->ir_reply[4] >> 4, req->ir_reply[5]);
885 
886 	sc->ipmi_dev_support = req->ir_reply[5];
887 
888 	IPMI_INIT_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
889 	    IPMI_CLEAR_FLAGS, 1, 0);
890 
891 	ipmi_submit_driver_request(sc, req, 0);
892 
893 	/* XXX: Magic numbers */
894 	if (req->ir_compcode == 0xc0) {
895 		device_printf(dev, "Clear flags is busy\n");
896 	}
897 	if (req->ir_compcode == 0xc1) {
898 		device_printf(dev, "Clear flags illegal\n");
899 	}
900 
901 	for (i = 0; i < 8; i++) {
902 		IPMI_INIT_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
903 		    IPMI_GET_CHANNEL_INFO, 1, 0);
904 		req->ir_request[0] = i;
905 
906 		error = ipmi_submit_driver_request(sc, req, 0);
907 
908 		if (error != 0 || req->ir_compcode != 0)
909 			break;
910 	}
911 	device_printf(dev, "Number of channels %d\n", i);
912 
913 	/*
914 	 * Probe for watchdog, but only for backends which support
915 	 * polled driver requests.
916 	 */
917 	if (wd_init_enable && sc->ipmi_driver_requests_polled) {
918 		IPMI_INIT_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
919 		    IPMI_GET_WDOG, 0, 0);
920 
921 		error = ipmi_submit_driver_request(sc, req, 0);
922 
923 		if (error == 0 && req->ir_compcode == 0x00) {
924 			device_printf(dev, "Attached watchdog\n");
925 			/* register the watchdog event handler */
926 			sc->ipmi_watchdog_tag = EVENTHANDLER_REGISTER(
927 				watchdog_list, ipmi_wd_event, sc, 0);
928 			sc->ipmi_shutdown_tag = EVENTHANDLER_REGISTER(
929 				shutdown_pre_sync, ipmi_shutdown_event,
930 				sc, 0);
931 		}
932 	}
933 
934 	sc->ipmi_cdev = make_dev(&ipmi_cdevsw, device_get_unit(dev),
935 	    UID_ROOT, GID_OPERATOR, 0660, "ipmi%d", device_get_unit(dev));
936 	if (sc->ipmi_cdev == NULL) {
937 		device_printf(dev, "Failed to create cdev\n");
938 		return;
939 	}
940 	sc->ipmi_cdev->si_drv1 = sc;
941 
942 	/*
943 	 * Set initial watchdog state. If desired, set an initial
944 	 * watchdog on startup. Or, if the watchdog device is
945 	 * disabled, clear any existing watchdog.
946 	 */
947 	if (on && wd_startup_countdown > 0) {
948 		if (ipmi_set_watchdog(sc, wd_startup_countdown) == 0 &&
949 		    ipmi_reset_watchdog(sc) == 0) {
950 			sc->ipmi_watchdog_active = wd_startup_countdown;
951 			sc->ipmi_watchdog_actions = wd_timer_actions;
952 			sc->ipmi_watchdog_pretimeout = wd_pretimeout_countdown;
953 		} else
954 			(void)ipmi_set_watchdog(sc, 0);
955 		ipmi_reset_watchdog(sc);
956 	} else if (!on)
957 		(void)ipmi_set_watchdog(sc, 0);
958 	/*
959 	 * Power cycle the system off using IPMI. We use last - 2 since we don't
960 	 * handle all the other kinds of reboots. We'll let others handle them.
961 	 * We only try to do this if the BMC supports the Chassis device.
962 	 */
963 	if (sc->ipmi_dev_support & IPMI_ADS_CHASSIS) {
964 		device_printf(dev, "Establishing power cycle handler\n");
965 		sc->ipmi_power_cycle_tag = EVENTHANDLER_REGISTER(shutdown_final,
966 		    ipmi_power_cycle, sc, SHUTDOWN_PRI_LAST - 2);
967 	}
968 }
969 
970 int
971 ipmi_attach(device_t dev)
972 {
973 	struct ipmi_softc *sc = device_get_softc(dev);
974 	int error;
975 
976 	if (sc->ipmi_irq_res != NULL && sc->ipmi_intr != NULL) {
977 		error = bus_setup_intr(dev, sc->ipmi_irq_res, INTR_TYPE_MISC,
978 		    NULL, sc->ipmi_intr, sc, &sc->ipmi_irq);
979 		if (error) {
980 			device_printf(dev, "can't set up interrupt\n");
981 			return (error);
982 		}
983 	}
984 
985 	bzero(&sc->ipmi_ich, sizeof(struct intr_config_hook));
986 	sc->ipmi_ich.ich_func = ipmi_startup;
987 	sc->ipmi_ich.ich_arg = sc;
988 	if (config_intrhook_establish(&sc->ipmi_ich) != 0) {
989 		device_printf(dev, "can't establish configuration hook\n");
990 		return (ENOMEM);
991 	}
992 
993 	ipmi_attached = 1;
994 	return (0);
995 }
996 
997 int
998 ipmi_detach(device_t dev)
999 {
1000 	struct ipmi_softc *sc;
1001 
1002 	sc = device_get_softc(dev);
1003 
1004 	/* Fail if there are any open handles. */
1005 	IPMI_LOCK(sc);
1006 	if (sc->ipmi_opened) {
1007 		IPMI_UNLOCK(sc);
1008 		return (EBUSY);
1009 	}
1010 	IPMI_UNLOCK(sc);
1011 	if (sc->ipmi_cdev)
1012 		destroy_dev(sc->ipmi_cdev);
1013 
1014 	/* Detach from watchdog handling and turn off watchdog. */
1015 	if (sc->ipmi_shutdown_tag)
1016 		EVENTHANDLER_DEREGISTER(shutdown_pre_sync,
1017 		sc->ipmi_shutdown_tag);
1018 	if (sc->ipmi_watchdog_tag) {
1019 		EVENTHANDLER_DEREGISTER(watchdog_list, sc->ipmi_watchdog_tag);
1020 		ipmi_set_watchdog(sc, 0);
1021 	}
1022 
1023 	/* Detach from shutdown handling for power cycle reboot */
1024 	if (sc->ipmi_power_cycle_tag)
1025 		EVENTHANDLER_DEREGISTER(shutdown_final, sc->ipmi_power_cycle_tag);
1026 
1027 	/* XXX: should use shutdown callout I think. */
1028 	/* If the backend uses a kthread, shut it down. */
1029 	IPMI_LOCK(sc);
1030 	sc->ipmi_detaching = 1;
1031 	if (sc->ipmi_kthread) {
1032 		cv_broadcast(&sc->ipmi_request_added);
1033 		msleep(sc->ipmi_kthread, &sc->ipmi_requests_lock, 0,
1034 		    "ipmi_wait", 0);
1035 	}
1036 	IPMI_UNLOCK(sc);
1037 	if (sc->ipmi_irq)
1038 		bus_teardown_intr(dev, sc->ipmi_irq_res, sc->ipmi_irq);
1039 
1040 	ipmi_release_resources(dev);
1041 	mtx_destroy(&sc->ipmi_io_lock);
1042 	mtx_destroy(&sc->ipmi_requests_lock);
1043 	return (0);
1044 }
1045 
1046 void
1047 ipmi_release_resources(device_t dev)
1048 {
1049 	struct ipmi_softc *sc;
1050 	int i;
1051 
1052 	sc = device_get_softc(dev);
1053 	if (sc->ipmi_irq)
1054 		bus_teardown_intr(dev, sc->ipmi_irq_res, sc->ipmi_irq);
1055 	if (sc->ipmi_irq_res)
1056 		bus_release_resource(dev, SYS_RES_IRQ, sc->ipmi_irq_rid,
1057 		    sc->ipmi_irq_res);
1058 	for (i = 0; i < MAX_RES; i++)
1059 		if (sc->ipmi_io_res[i])
1060 			bus_release_resource(dev, sc->ipmi_io_type,
1061 			    sc->ipmi_io_rid + i, sc->ipmi_io_res[i]);
1062 }
1063 
1064 /* XXX: Why? */
1065 static void
1066 ipmi_unload(void *arg)
1067 {
1068 	device_t *	devs;
1069 	int		count;
1070 	int		i;
1071 
1072 	if (devclass_get_devices(devclass_find("ipmi"), &devs, &count) != 0)
1073 		return;
1074 	for (i = 0; i < count; i++)
1075 		device_delete_child(device_get_parent(devs[i]), devs[i]);
1076 	free(devs, M_TEMP);
1077 }
1078 SYSUNINIT(ipmi_unload, SI_SUB_DRIVERS, SI_ORDER_FIRST, ipmi_unload, NULL);
1079 
1080 #ifdef IMPI_DEBUG
1081 static void
1082 dump_buf(u_char *data, int len)
1083 {
1084 	char buf[20];
1085 	char line[1024];
1086 	char temp[30];
1087 	int count = 0;
1088 	int i=0;
1089 
1090 	printf("Address %p len %d\n", data, len);
1091 	if (len > 256)
1092 		len = 256;
1093 	line[0] = '\000';
1094 	for (; len > 0; len--, data++) {
1095 		sprintf(temp, "%02x ", *data);
1096 		strcat(line, temp);
1097 		if (*data >= ' ' && *data <= '~')
1098 			buf[count] = *data;
1099 		else if (*data >= 'A' && *data <= 'Z')
1100 			buf[count] = *data;
1101 		else
1102 			buf[count] = '.';
1103 		if (++count == 16) {
1104 			buf[count] = '\000';
1105 			count = 0;
1106 			printf("  %3x  %s %s\n", i, line, buf);
1107 			i+=16;
1108 			line[0] = '\000';
1109 		}
1110 	}
1111 	buf[count] = '\000';
1112 
1113 	for (; count != 16; count++) {
1114 		strcat(line, "   ");
1115 	}
1116 	printf("  %3x  %s %s\n", i, line, buf);
1117 }
1118 #endif
1119