xref: /freebsd/sys/dev/ipmi/ipmi.c (revision 7ef62cebc2f965b0f640263e179276928885e33d)
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/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/bus.h>
35 #include <sys/condvar.h>
36 #include <sys/conf.h>
37 #include <sys/eventhandler.h>
38 #include <sys/kernel.h>
39 #include <sys/lock.h>
40 #include <sys/malloc.h>
41 #include <sys/module.h>
42 #include <sys/mutex.h>
43 #include <sys/poll.h>
44 #include <sys/reboot.h>
45 #include <sys/rman.h>
46 #include <sys/selinfo.h>
47 #include <sys/sysctl.h>
48 #include <sys/watchdog.h>
49 
50 #ifdef LOCAL_MODULE
51 #include <ipmi.h>
52 #include <ipmivars.h>
53 #else
54 #include <sys/ipmi.h>
55 #include <dev/ipmi/ipmivars.h>
56 #endif
57 
58 #ifdef IPMICTL_SEND_COMMAND_32
59 #include <sys/abi_compat.h>
60 #endif
61 
62 /*
63  * Driver request structures are allocated on the stack via alloca() to
64  * avoid calling malloc(), especially for the watchdog handler.
65  * To avoid too much stack growth, a previously allocated structure can
66  * be reused via IPMI_INIT_DRIVER_REQUEST(), but the caller should ensure
67  * that there is adequate reply/request space in the original allocation.
68  */
69 #define	IPMI_INIT_DRIVER_REQUEST(req, addr, cmd, reqlen, replylen)	\
70 	bzero((req), sizeof(struct ipmi_request));			\
71 	ipmi_init_request((req), NULL, 0, (addr), (cmd), (reqlen), (replylen))
72 
73 #define	IPMI_ALLOC_DRIVER_REQUEST(req, addr, cmd, reqlen, replylen)	\
74 	(req) = __builtin_alloca(sizeof(struct ipmi_request) +		\
75 	    (reqlen) + (replylen));					\
76 	IPMI_INIT_DRIVER_REQUEST((req), (addr), (cmd), (reqlen),	\
77 	    (replylen))
78 
79 static d_ioctl_t ipmi_ioctl;
80 static d_poll_t ipmi_poll;
81 static d_open_t ipmi_open;
82 static void ipmi_dtor(void *arg);
83 
84 int ipmi_attached = 0;
85 
86 static int on = 1;
87 static bool wd_in_shutdown = false;
88 static int wd_timer_actions = IPMI_SET_WD_ACTION_POWER_CYCLE;
89 static int wd_shutdown_countdown = 0; /* sec */
90 static int wd_startup_countdown = 0; /* sec */
91 static int wd_pretimeout_countdown = 120; /* sec */
92 static int cycle_wait = 10; /* sec */
93 static int wd_init_enable = 1;
94 
95 static SYSCTL_NODE(_hw, OID_AUTO, ipmi, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
96     "IPMI driver parameters");
97 SYSCTL_INT(_hw_ipmi, OID_AUTO, on, CTLFLAG_RWTUN,
98 	&on, 0, "");
99 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_init_enable, CTLFLAG_RWTUN,
100 	&wd_init_enable, 1, "Enable watchdog initialization");
101 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_timer_actions, CTLFLAG_RWTUN,
102 	&wd_timer_actions, 0,
103 	"IPMI watchdog timer actions (including pre-timeout interrupt)");
104 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_shutdown_countdown, CTLFLAG_RWTUN,
105 	&wd_shutdown_countdown, 0,
106 	"IPMI watchdog countdown for shutdown (seconds)");
107 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_startup_countdown, CTLFLAG_RDTUN,
108 	&wd_startup_countdown, 0,
109 	"IPMI watchdog countdown initialized during startup (seconds)");
110 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_pretimeout_countdown, CTLFLAG_RWTUN,
111 	&wd_pretimeout_countdown, 0,
112 	"IPMI watchdog pre-timeout countdown (seconds)");
113 SYSCTL_INT(_hw_ipmi, OID_AUTO, cycle_wait, CTLFLAG_RWTUN,
114 	&cycle_wait, 0,
115 	"IPMI power cycle on reboot delay time (seconds)");
116 
117 static struct cdevsw ipmi_cdevsw = {
118 	.d_version =    D_VERSION,
119 	.d_open =	ipmi_open,
120 	.d_ioctl =	ipmi_ioctl,
121 	.d_poll =	ipmi_poll,
122 	.d_name =	"ipmi",
123 };
124 
125 static MALLOC_DEFINE(M_IPMI, "ipmi", "ipmi");
126 
127 static int
128 ipmi_open(struct cdev *cdev, int flags, int fmt, struct thread *td)
129 {
130 	struct ipmi_device *dev;
131 	struct ipmi_softc *sc;
132 	int error;
133 
134 	if (!on)
135 		return (ENOENT);
136 
137 	/* Initialize the per file descriptor data. */
138 	dev = malloc(sizeof(struct ipmi_device), M_IPMI, M_WAITOK | M_ZERO);
139 	error = devfs_set_cdevpriv(dev, ipmi_dtor);
140 	if (error) {
141 		free(dev, M_IPMI);
142 		return (error);
143 	}
144 
145 	sc = cdev->si_drv1;
146 	TAILQ_INIT(&dev->ipmi_completed_requests);
147 	dev->ipmi_address = IPMI_BMC_SLAVE_ADDR;
148 	dev->ipmi_lun = IPMI_BMC_SMS_LUN;
149 	dev->ipmi_softc = sc;
150 	IPMI_LOCK(sc);
151 	sc->ipmi_opened++;
152 	IPMI_UNLOCK(sc);
153 
154 	return (0);
155 }
156 
157 static int
158 ipmi_poll(struct cdev *cdev, int poll_events, struct thread *td)
159 {
160 	struct ipmi_device *dev;
161 	struct ipmi_softc *sc;
162 	int revents = 0;
163 
164 	if (devfs_get_cdevpriv((void **)&dev))
165 		return (0);
166 
167 	sc = cdev->si_drv1;
168 	IPMI_LOCK(sc);
169 	if (poll_events & (POLLIN | POLLRDNORM)) {
170 		if (!TAILQ_EMPTY(&dev->ipmi_completed_requests))
171 		    revents |= poll_events & (POLLIN | POLLRDNORM);
172 		if (dev->ipmi_requests == 0)
173 		    revents |= POLLERR;
174 	}
175 
176 	if (revents == 0) {
177 		if (poll_events & (POLLIN | POLLRDNORM))
178 			selrecord(td, &dev->ipmi_select);
179 	}
180 	IPMI_UNLOCK(sc);
181 
182 	return (revents);
183 }
184 
185 static void
186 ipmi_purge_completed_requests(struct ipmi_device *dev)
187 {
188 	struct ipmi_request *req;
189 
190 	while (!TAILQ_EMPTY(&dev->ipmi_completed_requests)) {
191 		req = TAILQ_FIRST(&dev->ipmi_completed_requests);
192 		TAILQ_REMOVE(&dev->ipmi_completed_requests, req, ir_link);
193 		dev->ipmi_requests--;
194 		ipmi_free_request(req);
195 	}
196 }
197 
198 static void
199 ipmi_dtor(void *arg)
200 {
201 	struct ipmi_request *req, *nreq;
202 	struct ipmi_device *dev;
203 	struct ipmi_softc *sc;
204 
205 	dev = arg;
206 	sc = dev->ipmi_softc;
207 
208 	IPMI_LOCK(sc);
209 	if (dev->ipmi_requests) {
210 		/* Throw away any pending requests for this device. */
211 		TAILQ_FOREACH_SAFE(req, &sc->ipmi_pending_requests_highpri, ir_link,
212 		    nreq) {
213 			if (req->ir_owner == dev) {
214 				TAILQ_REMOVE(&sc->ipmi_pending_requests_highpri, req,
215 				    ir_link);
216 				dev->ipmi_requests--;
217 				ipmi_free_request(req);
218 			}
219 		}
220 		TAILQ_FOREACH_SAFE(req, &sc->ipmi_pending_requests, ir_link,
221 		    nreq) {
222 			if (req->ir_owner == dev) {
223 				TAILQ_REMOVE(&sc->ipmi_pending_requests, req,
224 				    ir_link);
225 				dev->ipmi_requests--;
226 				ipmi_free_request(req);
227 			}
228 		}
229 
230 		/* Throw away any pending completed requests for this device. */
231 		ipmi_purge_completed_requests(dev);
232 
233 		/*
234 		 * If we still have outstanding requests, they must be stuck
235 		 * in an interface driver, so wait for those to drain.
236 		 */
237 		dev->ipmi_closing = 1;
238 		while (dev->ipmi_requests > 0) {
239 			msleep(&dev->ipmi_requests, &sc->ipmi_requests_lock,
240 			    PWAIT, "ipmidrain", 0);
241 			ipmi_purge_completed_requests(dev);
242 		}
243 	}
244 	sc->ipmi_opened--;
245 	IPMI_UNLOCK(sc);
246 
247 	/* Cleanup. */
248 	free(dev, M_IPMI);
249 }
250 
251 static u_char
252 ipmi_ipmb_checksum(u_char *data, int len)
253 {
254 	u_char sum = 0;
255 
256 	for (; len; len--)
257 		sum += *data++;
258 	return (-sum);
259 }
260 
261 static int
262 ipmi_ioctl(struct cdev *cdev, u_long cmd, caddr_t data,
263     int flags, struct thread *td)
264 {
265 	struct ipmi_softc *sc;
266 	struct ipmi_device *dev;
267 	struct ipmi_request *kreq;
268 	struct ipmi_req *req = (struct ipmi_req *)data;
269 	struct ipmi_recv *recv = (struct ipmi_recv *)data;
270 	struct ipmi_addr addr;
271 #ifdef IPMICTL_SEND_COMMAND_32
272 	struct ipmi_req32 *req32 = (struct ipmi_req32 *)data;
273 	struct ipmi_recv32 *recv32 = (struct ipmi_recv32 *)data;
274 	union {
275 		struct ipmi_req req;
276 		struct ipmi_recv recv;
277 	} thunk32;
278 #endif
279 	int error, len;
280 
281 	error = devfs_get_cdevpriv((void **)&dev);
282 	if (error)
283 		return (error);
284 
285 	sc = cdev->si_drv1;
286 
287 #ifdef IPMICTL_SEND_COMMAND_32
288 	/* Convert 32-bit structures to native. */
289 	switch (cmd) {
290 	case IPMICTL_SEND_COMMAND_32:
291 		req = &thunk32.req;
292 		req->addr = PTRIN(req32->addr);
293 		req->addr_len = req32->addr_len;
294 		req->msgid = req32->msgid;
295 		req->msg.netfn = req32->msg.netfn;
296 		req->msg.cmd = req32->msg.cmd;
297 		req->msg.data_len = req32->msg.data_len;
298 		req->msg.data = PTRIN(req32->msg.data);
299 		break;
300 	case IPMICTL_RECEIVE_MSG_TRUNC_32:
301 	case IPMICTL_RECEIVE_MSG_32:
302 		recv = &thunk32.recv;
303 		recv->addr = PTRIN(recv32->addr);
304 		recv->addr_len = recv32->addr_len;
305 		recv->msg.data_len = recv32->msg.data_len;
306 		recv->msg.data = PTRIN(recv32->msg.data);
307 		break;
308 	}
309 #endif
310 
311 	switch (cmd) {
312 #ifdef IPMICTL_SEND_COMMAND_32
313 	case IPMICTL_SEND_COMMAND_32:
314 #endif
315 	case IPMICTL_SEND_COMMAND:
316 		error = copyin(req->addr, &addr, sizeof(addr));
317 		if (error)
318 			return (error);
319 
320 		if (addr.addr_type == IPMI_SYSTEM_INTERFACE_ADDR_TYPE) {
321 			struct ipmi_system_interface_addr *saddr =
322 			    (struct ipmi_system_interface_addr *)&addr;
323 
324 			kreq = ipmi_alloc_request(dev, req->msgid,
325 			    IPMI_ADDR(req->msg.netfn, saddr->lun & 0x3),
326 			    req->msg.cmd, req->msg.data_len, IPMI_MAX_RX);
327 			error = copyin(req->msg.data, kreq->ir_request,
328 			    req->msg.data_len);
329 			if (error) {
330 				ipmi_free_request(kreq);
331 				return (error);
332 			}
333 			IPMI_LOCK(sc);
334 			dev->ipmi_requests++;
335 			error = sc->ipmi_enqueue_request(sc, kreq);
336 			IPMI_UNLOCK(sc);
337 			if (error)
338 				return (error);
339 			break;
340 		}
341 
342 		/* Special processing for IPMB commands */
343 		struct ipmi_ipmb_addr *iaddr = (struct ipmi_ipmb_addr *)&addr;
344 
345 		IPMI_ALLOC_DRIVER_REQUEST(kreq, IPMI_ADDR(IPMI_APP_REQUEST, 0),
346 		    IPMI_SEND_MSG, req->msg.data_len + 8, IPMI_MAX_RX);
347 		/* Construct the SEND MSG header */
348 		kreq->ir_request[0] = iaddr->channel;
349 		kreq->ir_request[1] = iaddr->slave_addr;
350 		kreq->ir_request[2] = IPMI_ADDR(req->msg.netfn, iaddr->lun);
351 		kreq->ir_request[3] =
352 		    ipmi_ipmb_checksum(&kreq->ir_request[1], 2);
353 		kreq->ir_request[4] = dev->ipmi_address;
354 		kreq->ir_request[5] = IPMI_ADDR(0, dev->ipmi_lun);
355 		kreq->ir_request[6] = req->msg.cmd;
356 		/* Copy the message data */
357 		if (req->msg.data_len > 0) {
358 			error = copyin(req->msg.data, &kreq->ir_request[7],
359 			    req->msg.data_len);
360 			if (error != 0)
361 				return (error);
362 		}
363 		kreq->ir_request[req->msg.data_len + 7] =
364 		    ipmi_ipmb_checksum(&kreq->ir_request[4],
365 		    req->msg.data_len + 3);
366 		error = ipmi_submit_driver_request(sc, kreq, MAX_TIMEOUT);
367 		if (error != 0)
368 			return (error);
369 
370 		kreq = ipmi_alloc_request(dev, req->msgid,
371 		    IPMI_ADDR(IPMI_APP_REQUEST, 0), IPMI_GET_MSG,
372 		    0, IPMI_MAX_RX);
373 		kreq->ir_ipmb = true;
374 		kreq->ir_ipmb_addr = IPMI_ADDR(req->msg.netfn, 0);
375 		kreq->ir_ipmb_command = req->msg.cmd;
376 		IPMI_LOCK(sc);
377 		dev->ipmi_requests++;
378 		error = sc->ipmi_enqueue_request(sc, kreq);
379 		IPMI_UNLOCK(sc);
380 		if (error != 0)
381 			return (error);
382 		break;
383 #ifdef IPMICTL_SEND_COMMAND_32
384 	case IPMICTL_RECEIVE_MSG_TRUNC_32:
385 	case IPMICTL_RECEIVE_MSG_32:
386 #endif
387 	case IPMICTL_RECEIVE_MSG_TRUNC:
388 	case IPMICTL_RECEIVE_MSG:
389 		error = copyin(recv->addr, &addr, sizeof(addr));
390 		if (error)
391 			return (error);
392 
393 		IPMI_LOCK(sc);
394 		kreq = TAILQ_FIRST(&dev->ipmi_completed_requests);
395 		if (kreq == NULL) {
396 			IPMI_UNLOCK(sc);
397 			return (EAGAIN);
398 		}
399 		if (kreq->ir_error != 0) {
400 			error = kreq->ir_error;
401 			TAILQ_REMOVE(&dev->ipmi_completed_requests, kreq,
402 			    ir_link);
403 			dev->ipmi_requests--;
404 			IPMI_UNLOCK(sc);
405 			ipmi_free_request(kreq);
406 			return (error);
407 		}
408 
409 		recv->recv_type = IPMI_RESPONSE_RECV_TYPE;
410 		recv->msgid = kreq->ir_msgid;
411 		if (kreq->ir_ipmb) {
412 			addr.channel = IPMI_IPMB_CHANNEL;
413 			recv->msg.netfn =
414 			    IPMI_REPLY_ADDR(kreq->ir_ipmb_addr) >> 2;
415 			recv->msg.cmd = kreq->ir_ipmb_command;
416 			/* Get the compcode of response */
417 			kreq->ir_compcode = kreq->ir_reply[6];
418 			/* Move the reply head past response header */
419 			kreq->ir_reply += 7;
420 			len = kreq->ir_replylen - 7;
421 		} else {
422 			addr.channel = IPMI_BMC_CHANNEL;
423 			recv->msg.netfn = IPMI_REPLY_ADDR(kreq->ir_addr) >> 2;
424 			recv->msg.cmd = kreq->ir_command;
425 			len = kreq->ir_replylen + 1;
426 		}
427 
428 		if (recv->msg.data_len < len &&
429 		    (cmd == IPMICTL_RECEIVE_MSG
430 #ifdef IPMICTL_RECEIVE_MSG_32
431 		    || cmd == IPMICTL_RECEIVE_MSG_32
432 #endif
433 		    )) {
434 			IPMI_UNLOCK(sc);
435 			return (EMSGSIZE);
436 		}
437 		TAILQ_REMOVE(&dev->ipmi_completed_requests, kreq, ir_link);
438 		dev->ipmi_requests--;
439 		IPMI_UNLOCK(sc);
440 		len = min(recv->msg.data_len, len);
441 		recv->msg.data_len = len;
442 		error = copyout(&addr, recv->addr,sizeof(addr));
443 		if (error == 0)
444 			error = copyout(&kreq->ir_compcode, recv->msg.data, 1);
445 		if (error == 0)
446 			error = copyout(kreq->ir_reply, recv->msg.data + 1,
447 			    len - 1);
448 		ipmi_free_request(kreq);
449 		if (error)
450 			return (error);
451 		break;
452 	case IPMICTL_SET_MY_ADDRESS_CMD:
453 		IPMI_LOCK(sc);
454 		dev->ipmi_address = *(int*)data;
455 		IPMI_UNLOCK(sc);
456 		break;
457 	case IPMICTL_GET_MY_ADDRESS_CMD:
458 		IPMI_LOCK(sc);
459 		*(int*)data = dev->ipmi_address;
460 		IPMI_UNLOCK(sc);
461 		break;
462 	case IPMICTL_SET_MY_LUN_CMD:
463 		IPMI_LOCK(sc);
464 		dev->ipmi_lun = *(int*)data & 0x3;
465 		IPMI_UNLOCK(sc);
466 		break;
467 	case IPMICTL_GET_MY_LUN_CMD:
468 		IPMI_LOCK(sc);
469 		*(int*)data = dev->ipmi_lun;
470 		IPMI_UNLOCK(sc);
471 		break;
472 	case IPMICTL_SET_GETS_EVENTS_CMD:
473 		/*
474 		device_printf(sc->ipmi_dev,
475 		    "IPMICTL_SET_GETS_EVENTS_CMD NA\n");
476 		*/
477 		break;
478 	case IPMICTL_REGISTER_FOR_CMD:
479 	case IPMICTL_UNREGISTER_FOR_CMD:
480 		return (EOPNOTSUPP);
481 	default:
482 		device_printf(sc->ipmi_dev, "Unknown IOCTL %lX\n", cmd);
483 		return (ENOIOCTL);
484 	}
485 
486 #ifdef IPMICTL_SEND_COMMAND_32
487 	/* Update changed fields in 32-bit structures. */
488 	switch (cmd) {
489 	case IPMICTL_RECEIVE_MSG_TRUNC_32:
490 	case IPMICTL_RECEIVE_MSG_32:
491 		recv32->recv_type = recv->recv_type;
492 		recv32->msgid = recv->msgid;
493 		recv32->msg.netfn = recv->msg.netfn;
494 		recv32->msg.cmd = recv->msg.cmd;
495 		recv32->msg.data_len = recv->msg.data_len;
496 		break;
497 	}
498 #endif
499 	return (0);
500 }
501 
502 /*
503  * Request management.
504  */
505 
506 __inline void
507 ipmi_init_request(struct ipmi_request *req, struct ipmi_device *dev, long msgid,
508     uint8_t addr, uint8_t command, size_t requestlen, size_t replylen)
509 {
510 
511 	req->ir_owner = dev;
512 	req->ir_msgid = msgid;
513 	req->ir_addr = addr;
514 	req->ir_command = command;
515 	if (requestlen) {
516 		req->ir_request = (char *)&req[1];
517 		req->ir_requestlen = requestlen;
518 	}
519 	if (replylen) {
520 		req->ir_reply = (char *)&req[1] + requestlen;
521 		req->ir_replybuflen = replylen;
522 	}
523 }
524 
525 /* Allocate a new request with request and reply buffers. */
526 struct ipmi_request *
527 ipmi_alloc_request(struct ipmi_device *dev, long msgid, uint8_t addr,
528     uint8_t command, size_t requestlen, size_t replylen)
529 {
530 	struct ipmi_request *req;
531 
532 	req = malloc(sizeof(struct ipmi_request) + requestlen + replylen,
533 	    M_IPMI, M_WAITOK | M_ZERO);
534 	ipmi_init_request(req, dev, msgid, addr, command, requestlen, replylen);
535 	return (req);
536 }
537 
538 /* Free a request no longer in use. */
539 void
540 ipmi_free_request(struct ipmi_request *req)
541 {
542 
543 	free(req, M_IPMI);
544 }
545 
546 /* Store a processed request on the appropriate completion queue. */
547 void
548 ipmi_complete_request(struct ipmi_softc *sc, struct ipmi_request *req)
549 {
550 	struct ipmi_device *dev;
551 
552 	IPMI_LOCK_ASSERT(sc);
553 
554 	/*
555 	 * Anonymous requests (from inside the driver) always have a
556 	 * waiter that we awaken.
557 	 */
558 	if (req->ir_owner == NULL)
559 		wakeup(req);
560 	else {
561 		dev = req->ir_owner;
562 		TAILQ_INSERT_TAIL(&dev->ipmi_completed_requests, req, ir_link);
563 		selwakeup(&dev->ipmi_select);
564 		if (dev->ipmi_closing)
565 			wakeup(&dev->ipmi_requests);
566 	}
567 }
568 
569 /* Perform an internal driver request. */
570 int
571 ipmi_submit_driver_request(struct ipmi_softc *sc, struct ipmi_request *req,
572     int timo)
573 {
574 
575 	return (sc->ipmi_driver_request(sc, req, timo));
576 }
577 
578 /*
579  * Helper routine for polled system interfaces that use
580  * ipmi_polled_enqueue_request() to queue requests.  This request
581  * waits until there is a pending request and then returns the first
582  * request.  If the driver is shutting down, it returns NULL.
583  */
584 struct ipmi_request *
585 ipmi_dequeue_request(struct ipmi_softc *sc)
586 {
587 	struct ipmi_request *req;
588 
589 	IPMI_LOCK_ASSERT(sc);
590 
591 	while (!sc->ipmi_detaching && TAILQ_EMPTY(&sc->ipmi_pending_requests) &&
592 	    TAILQ_EMPTY(&sc->ipmi_pending_requests_highpri))
593 		cv_wait(&sc->ipmi_request_added, &sc->ipmi_requests_lock);
594 	if (sc->ipmi_detaching)
595 		return (NULL);
596 
597 	req = TAILQ_FIRST(&sc->ipmi_pending_requests_highpri);
598 	if (req != NULL)
599 		TAILQ_REMOVE(&sc->ipmi_pending_requests_highpri, req, ir_link);
600 	else {
601 		req = TAILQ_FIRST(&sc->ipmi_pending_requests);
602 		TAILQ_REMOVE(&sc->ipmi_pending_requests, req, ir_link);
603 	}
604 	return (req);
605 }
606 
607 /* Default implementation of ipmi_enqueue_request() for polled interfaces. */
608 int
609 ipmi_polled_enqueue_request(struct ipmi_softc *sc, struct ipmi_request *req)
610 {
611 
612 	IPMI_LOCK_ASSERT(sc);
613 
614 	TAILQ_INSERT_TAIL(&sc->ipmi_pending_requests, req, ir_link);
615 	cv_signal(&sc->ipmi_request_added);
616 	return (0);
617 }
618 
619 int
620 ipmi_polled_enqueue_request_highpri(struct ipmi_softc *sc, struct ipmi_request *req)
621 {
622 
623 	IPMI_LOCK_ASSERT(sc);
624 
625 	TAILQ_INSERT_TAIL(&sc->ipmi_pending_requests_highpri, req, ir_link);
626 	cv_signal(&sc->ipmi_request_added);
627 	return (0);
628 }
629 
630 /*
631  * Watchdog event handler.
632  */
633 
634 static int
635 ipmi_reset_watchdog(struct ipmi_softc *sc)
636 {
637 	struct ipmi_request *req;
638 	int error;
639 
640 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
641 	    IPMI_RESET_WDOG, 0, 0);
642 	error = ipmi_submit_driver_request(sc, req, 0);
643 	if (error) {
644 		device_printf(sc->ipmi_dev, "Failed to reset watchdog\n");
645 	} else if (req->ir_compcode == 0x80) {
646 		error = ENOENT;
647 	} else if (req->ir_compcode != 0) {
648 		device_printf(sc->ipmi_dev, "Watchdog reset returned 0x%x\n",
649 		    req->ir_compcode);
650 		error = EINVAL;
651 	}
652 	return (error);
653 }
654 
655 static int
656 ipmi_set_watchdog(struct ipmi_softc *sc, unsigned int sec)
657 {
658 	struct ipmi_request *req;
659 	int error;
660 
661 	if (sec > 0xffff / 10)
662 		return (EINVAL);
663 
664 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
665 	    IPMI_SET_WDOG, 6, 0);
666 	if (sec) {
667 		req->ir_request[0] = IPMI_SET_WD_TIMER_DONT_STOP
668 		    | IPMI_SET_WD_TIMER_SMS_OS;
669 		req->ir_request[1] = (wd_timer_actions & 0xff);
670 		req->ir_request[2] = min(0xff,
671 		    min(wd_pretimeout_countdown, (sec + 2) / 4));
672 		req->ir_request[3] = 0;	/* Timer use */
673 		req->ir_request[4] = (sec * 10) & 0xff;
674 		req->ir_request[5] = (sec * 10) >> 8;
675 	} else {
676 		req->ir_request[0] = IPMI_SET_WD_TIMER_SMS_OS;
677 		req->ir_request[1] = 0;
678 		req->ir_request[2] = 0;
679 		req->ir_request[3] = 0;	/* Timer use */
680 		req->ir_request[4] = 0;
681 		req->ir_request[5] = 0;
682 	}
683 	error = ipmi_submit_driver_request(sc, req, 0);
684 	if (error) {
685 		device_printf(sc->ipmi_dev, "Failed to set watchdog\n");
686 	} else if (req->ir_compcode != 0) {
687 		device_printf(sc->ipmi_dev, "Watchdog set returned 0x%x\n",
688 		    req->ir_compcode);
689 		error = EINVAL;
690 	}
691 	return (error);
692 }
693 
694 static void
695 ipmi_wd_event(void *arg, unsigned int cmd, int *error)
696 {
697 	struct ipmi_softc *sc = arg;
698 	unsigned int timeout;
699 	int e;
700 
701 	/* Ignore requests while disabled. */
702 	if (!on)
703 		return;
704 
705 	/*
706 	 * To prevent infinite hangs, we don't let anyone pat or change
707 	 * the watchdog when we're shutting down. (See ipmi_shutdown_event().)
708 	 * However, we do want to keep patting the watchdog while we are doing
709 	 * a coredump.
710 	 */
711 	if (wd_in_shutdown) {
712 		if (dumping && sc->ipmi_watchdog_active)
713 			ipmi_reset_watchdog(sc);
714 		return;
715 	}
716 
717 	cmd &= WD_INTERVAL;
718 	if (cmd > 0 && cmd <= 63) {
719 		timeout = ((uint64_t)1 << cmd) / 1000000000;
720 		if (timeout == 0)
721 			timeout = 1;
722 		if (timeout != sc->ipmi_watchdog_active ||
723 		    wd_timer_actions != sc->ipmi_watchdog_actions ||
724 		    wd_pretimeout_countdown != sc->ipmi_watchdog_pretimeout) {
725 			e = ipmi_set_watchdog(sc, timeout);
726 			if (e == 0) {
727 				sc->ipmi_watchdog_active = timeout;
728 				sc->ipmi_watchdog_actions = wd_timer_actions;
729 				sc->ipmi_watchdog_pretimeout = wd_pretimeout_countdown;
730 			} else {
731 				(void)ipmi_set_watchdog(sc, 0);
732 				sc->ipmi_watchdog_active = 0;
733 				sc->ipmi_watchdog_actions = 0;
734 				sc->ipmi_watchdog_pretimeout = 0;
735 			}
736 		}
737 		if (sc->ipmi_watchdog_active != 0) {
738 			e = ipmi_reset_watchdog(sc);
739 			if (e == 0) {
740 				*error = 0;
741 			} else {
742 				(void)ipmi_set_watchdog(sc, 0);
743 				sc->ipmi_watchdog_active = 0;
744 				sc->ipmi_watchdog_actions = 0;
745 				sc->ipmi_watchdog_pretimeout = 0;
746 			}
747 		}
748 	} else if (atomic_readandclear_int(&sc->ipmi_watchdog_active) != 0) {
749 		sc->ipmi_watchdog_actions = 0;
750 		sc->ipmi_watchdog_pretimeout = 0;
751 
752 		e = ipmi_set_watchdog(sc, 0);
753 		if (e != 0 && cmd == 0)
754 			*error = EOPNOTSUPP;
755 	}
756 }
757 
758 static void
759 ipmi_shutdown_event(void *arg, unsigned int cmd, int *error)
760 {
761 	struct ipmi_softc *sc = arg;
762 
763 	/* Ignore event if disabled. */
764 	if (!on)
765 		return;
766 
767 	/*
768 	 * Positive wd_shutdown_countdown value will re-arm watchdog;
769 	 * Zero value in wd_shutdown_countdown will disable watchdog;
770 	 * Negative value in wd_shutdown_countdown will keep existing state;
771 	 *
772 	 * Revert to using a power cycle to ensure that the watchdog will
773 	 * do something useful here.  Having the watchdog send an NMI
774 	 * instead is useless during shutdown, and might be ignored if an
775 	 * NMI already triggered.
776 	 */
777 
778 	wd_in_shutdown = true;
779 	if (wd_shutdown_countdown == 0) {
780 		/* disable watchdog */
781 		ipmi_set_watchdog(sc, 0);
782 		sc->ipmi_watchdog_active = 0;
783 	} else if (wd_shutdown_countdown > 0) {
784 		/* set desired action and time, and, reset watchdog */
785 		wd_timer_actions = IPMI_SET_WD_ACTION_POWER_CYCLE;
786 		ipmi_set_watchdog(sc, wd_shutdown_countdown);
787 		sc->ipmi_watchdog_active = wd_shutdown_countdown;
788 		ipmi_reset_watchdog(sc);
789 	}
790 }
791 
792 static void
793 ipmi_power_cycle(void *arg, int howto)
794 {
795 	struct ipmi_softc *sc = arg;
796 	struct ipmi_request *req;
797 
798 	/*
799 	 * Ignore everything except power cycling requests
800 	 */
801 	if ((howto & RB_POWERCYCLE) == 0)
802 		return;
803 
804 	device_printf(sc->ipmi_dev, "Power cycling using IPMI\n");
805 
806 	/*
807 	 * Send a CHASSIS_CONTROL command to the CHASSIS device, subcommand 2
808 	 * as described in IPMI v2.0 spec section 28.3.
809 	 */
810 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_CHASSIS_REQUEST, 0),
811 	    IPMI_CHASSIS_CONTROL, 1, 0);
812 	req->ir_request[0] = IPMI_CC_POWER_CYCLE;
813 
814 	ipmi_submit_driver_request(sc, req, MAX_TIMEOUT);
815 
816 	if (req->ir_error != 0 || req->ir_compcode != 0) {
817 		device_printf(sc->ipmi_dev, "Power cycling via IPMI failed code %#x %#x\n",
818 		    req->ir_error, req->ir_compcode);
819 		return;
820 	}
821 
822 	/*
823 	 * BMCs are notoriously slow, give it cycle_wait seconds for the power
824 	 * down leg of the power cycle. If that fails, fallback to the next
825 	 * hanlder in the shutdown_final chain and/or the platform failsafe.
826 	 */
827 	DELAY(cycle_wait * 1000 * 1000);
828 	device_printf(sc->ipmi_dev, "Power cycling via IPMI timed out\n");
829 }
830 
831 static void
832 ipmi_startup(void *arg)
833 {
834 	struct ipmi_softc *sc = arg;
835 	struct ipmi_request *req;
836 	device_t dev;
837 	int error, i;
838 
839 	config_intrhook_disestablish(&sc->ipmi_ich);
840 	dev = sc->ipmi_dev;
841 
842 	/* Initialize interface-independent state. */
843 	mtx_init(&sc->ipmi_requests_lock, "ipmi requests", NULL, MTX_DEF);
844 	mtx_init(&sc->ipmi_io_lock, "ipmi io", NULL, MTX_DEF);
845 	cv_init(&sc->ipmi_request_added, "ipmireq");
846 	TAILQ_INIT(&sc->ipmi_pending_requests_highpri);
847 	TAILQ_INIT(&sc->ipmi_pending_requests);
848 
849 	/* Initialize interface-dependent state. */
850 	error = sc->ipmi_startup(sc);
851 	if (error) {
852 		device_printf(dev, "Failed to initialize interface: %d\n",
853 		    error);
854 		return;
855 	}
856 
857 	/* Send a GET_DEVICE_ID request. */
858 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
859 	    IPMI_GET_DEVICE_ID, 0, 15);
860 
861 	error = ipmi_submit_driver_request(sc, req, MAX_TIMEOUT);
862 	if (error == EWOULDBLOCK) {
863 		device_printf(dev, "Timed out waiting for GET_DEVICE_ID\n");
864 		return;
865 	} else if (error) {
866 		device_printf(dev, "Failed GET_DEVICE_ID: %d\n", error);
867 		return;
868 	} else if (req->ir_compcode != 0) {
869 		device_printf(dev,
870 		    "Bad completion code for GET_DEVICE_ID: %d\n",
871 		    req->ir_compcode);
872 		return;
873 	} else if (req->ir_replylen < 5) {
874 		device_printf(dev, "Short reply for GET_DEVICE_ID: %d\n",
875 		    req->ir_replylen);
876 		return;
877 	}
878 
879 	device_printf(dev, "IPMI device rev. %d, firmware rev. %d.%d%d, "
880 	    "version %d.%d, device support mask %#x\n",
881 	    req->ir_reply[1] & 0x0f,
882 	    req->ir_reply[2] & 0x7f, req->ir_reply[3] >> 4, req->ir_reply[3] & 0x0f,
883 	    req->ir_reply[4] & 0x0f, req->ir_reply[4] >> 4, req->ir_reply[5]);
884 
885 	sc->ipmi_dev_support = req->ir_reply[5];
886 
887 	IPMI_INIT_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
888 	    IPMI_CLEAR_FLAGS, 1, 0);
889 
890 	ipmi_submit_driver_request(sc, req, 0);
891 
892 	/* XXX: Magic numbers */
893 	if (req->ir_compcode == 0xc0) {
894 		device_printf(dev, "Clear flags is busy\n");
895 	}
896 	if (req->ir_compcode == 0xc1) {
897 		device_printf(dev, "Clear flags illegal\n");
898 	}
899 
900 	for (i = 0; i < 8; i++) {
901 		IPMI_INIT_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
902 		    IPMI_GET_CHANNEL_INFO, 1, 0);
903 		req->ir_request[0] = i;
904 
905 		error = ipmi_submit_driver_request(sc, req, 0);
906 
907 		if (error != 0 || req->ir_compcode != 0)
908 			break;
909 	}
910 	device_printf(dev, "Number of channels %d\n", i);
911 
912 	/*
913 	 * Probe for watchdog, but only for backends which support
914 	 * polled driver requests.
915 	 */
916 	if (wd_init_enable && sc->ipmi_driver_requests_polled) {
917 		IPMI_INIT_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
918 		    IPMI_GET_WDOG, 0, 0);
919 
920 		error = ipmi_submit_driver_request(sc, req, 0);
921 
922 		if (error == 0 && req->ir_compcode == 0x00) {
923 			device_printf(dev, "Attached watchdog\n");
924 			/* register the watchdog event handler */
925 			sc->ipmi_watchdog_tag = EVENTHANDLER_REGISTER(
926 				watchdog_list, ipmi_wd_event, sc, 0);
927 			sc->ipmi_shutdown_tag = EVENTHANDLER_REGISTER(
928 				shutdown_pre_sync, ipmi_shutdown_event,
929 				sc, 0);
930 		}
931 	}
932 
933 	sc->ipmi_cdev = make_dev(&ipmi_cdevsw, device_get_unit(dev),
934 	    UID_ROOT, GID_OPERATOR, 0660, "ipmi%d", device_get_unit(dev));
935 	if (sc->ipmi_cdev == NULL) {
936 		device_printf(dev, "Failed to create cdev\n");
937 		return;
938 	}
939 	sc->ipmi_cdev->si_drv1 = sc;
940 
941 	/*
942 	 * Set initial watchdog state. If desired, set an initial
943 	 * watchdog on startup. Or, if the watchdog device is
944 	 * disabled, clear any existing watchdog.
945 	 */
946 	if (on && wd_startup_countdown > 0) {
947 		if (ipmi_set_watchdog(sc, wd_startup_countdown) == 0 &&
948 		    ipmi_reset_watchdog(sc) == 0) {
949 			sc->ipmi_watchdog_active = wd_startup_countdown;
950 			sc->ipmi_watchdog_actions = wd_timer_actions;
951 			sc->ipmi_watchdog_pretimeout = wd_pretimeout_countdown;
952 		} else
953 			(void)ipmi_set_watchdog(sc, 0);
954 		ipmi_reset_watchdog(sc);
955 	} else if (!on)
956 		(void)ipmi_set_watchdog(sc, 0);
957 	/*
958 	 * Power cycle the system off using IPMI. We use last - 2 since we don't
959 	 * handle all the other kinds of reboots. We'll let others handle them.
960 	 * We only try to do this if the BMC supports the Chassis device.
961 	 */
962 	if (sc->ipmi_dev_support & IPMI_ADS_CHASSIS) {
963 		device_printf(dev, "Establishing power cycle handler\n");
964 		sc->ipmi_power_cycle_tag = EVENTHANDLER_REGISTER(shutdown_final,
965 		    ipmi_power_cycle, sc, SHUTDOWN_PRI_LAST - 2);
966 	}
967 }
968 
969 int
970 ipmi_attach(device_t dev)
971 {
972 	struct ipmi_softc *sc = device_get_softc(dev);
973 	int error;
974 
975 	if (sc->ipmi_irq_res != NULL && sc->ipmi_intr != NULL) {
976 		error = bus_setup_intr(dev, sc->ipmi_irq_res, INTR_TYPE_MISC,
977 		    NULL, sc->ipmi_intr, sc, &sc->ipmi_irq);
978 		if (error) {
979 			device_printf(dev, "can't set up interrupt\n");
980 			return (error);
981 		}
982 	}
983 
984 	bzero(&sc->ipmi_ich, sizeof(struct intr_config_hook));
985 	sc->ipmi_ich.ich_func = ipmi_startup;
986 	sc->ipmi_ich.ich_arg = sc;
987 	if (config_intrhook_establish(&sc->ipmi_ich) != 0) {
988 		device_printf(dev, "can't establish configuration hook\n");
989 		return (ENOMEM);
990 	}
991 
992 	ipmi_attached = 1;
993 	return (0);
994 }
995 
996 int
997 ipmi_detach(device_t dev)
998 {
999 	struct ipmi_softc *sc;
1000 
1001 	sc = device_get_softc(dev);
1002 
1003 	/* Fail if there are any open handles. */
1004 	IPMI_LOCK(sc);
1005 	if (sc->ipmi_opened) {
1006 		IPMI_UNLOCK(sc);
1007 		return (EBUSY);
1008 	}
1009 	IPMI_UNLOCK(sc);
1010 	if (sc->ipmi_cdev)
1011 		destroy_dev(sc->ipmi_cdev);
1012 
1013 	/* Detach from watchdog handling and turn off watchdog. */
1014 	if (sc->ipmi_shutdown_tag)
1015 		EVENTHANDLER_DEREGISTER(shutdown_pre_sync,
1016 		sc->ipmi_shutdown_tag);
1017 	if (sc->ipmi_watchdog_tag) {
1018 		EVENTHANDLER_DEREGISTER(watchdog_list, sc->ipmi_watchdog_tag);
1019 		ipmi_set_watchdog(sc, 0);
1020 	}
1021 
1022 	/* Detach from shutdown handling for power cycle reboot */
1023 	if (sc->ipmi_power_cycle_tag)
1024 		EVENTHANDLER_DEREGISTER(shutdown_final, sc->ipmi_power_cycle_tag);
1025 
1026 	/* XXX: should use shutdown callout I think. */
1027 	/* If the backend uses a kthread, shut it down. */
1028 	IPMI_LOCK(sc);
1029 	sc->ipmi_detaching = 1;
1030 	if (sc->ipmi_kthread) {
1031 		cv_broadcast(&sc->ipmi_request_added);
1032 		msleep(sc->ipmi_kthread, &sc->ipmi_requests_lock, 0,
1033 		    "ipmi_wait", 0);
1034 	}
1035 	IPMI_UNLOCK(sc);
1036 	if (sc->ipmi_irq)
1037 		bus_teardown_intr(dev, sc->ipmi_irq_res, sc->ipmi_irq);
1038 
1039 	ipmi_release_resources(dev);
1040 	mtx_destroy(&sc->ipmi_io_lock);
1041 	mtx_destroy(&sc->ipmi_requests_lock);
1042 	return (0);
1043 }
1044 
1045 void
1046 ipmi_release_resources(device_t dev)
1047 {
1048 	struct ipmi_softc *sc;
1049 	int i;
1050 
1051 	sc = device_get_softc(dev);
1052 	if (sc->ipmi_irq)
1053 		bus_teardown_intr(dev, sc->ipmi_irq_res, sc->ipmi_irq);
1054 	if (sc->ipmi_irq_res)
1055 		bus_release_resource(dev, SYS_RES_IRQ, sc->ipmi_irq_rid,
1056 		    sc->ipmi_irq_res);
1057 	for (i = 0; i < MAX_RES; i++)
1058 		if (sc->ipmi_io_res[i])
1059 			bus_release_resource(dev, sc->ipmi_io_type,
1060 			    sc->ipmi_io_rid + i, sc->ipmi_io_res[i]);
1061 }
1062 
1063 /* XXX: Why? */
1064 static void
1065 ipmi_unload(void *arg)
1066 {
1067 	device_t *	devs;
1068 	int		count;
1069 	int		i;
1070 
1071 	if (devclass_get_devices(devclass_find("ipmi"), &devs, &count) != 0)
1072 		return;
1073 	for (i = 0; i < count; i++)
1074 		device_delete_child(device_get_parent(devs[i]), devs[i]);
1075 	free(devs, M_TEMP);
1076 }
1077 SYSUNINIT(ipmi_unload, SI_SUB_DRIVERS, SI_ORDER_FIRST, ipmi_unload, NULL);
1078 
1079 #ifdef IMPI_DEBUG
1080 static void
1081 dump_buf(u_char *data, int len)
1082 {
1083 	char buf[20];
1084 	char line[1024];
1085 	char temp[30];
1086 	int count = 0;
1087 	int i=0;
1088 
1089 	printf("Address %p len %d\n", data, len);
1090 	if (len > 256)
1091 		len = 256;
1092 	line[0] = '\000';
1093 	for (; len > 0; len--, data++) {
1094 		sprintf(temp, "%02x ", *data);
1095 		strcat(line, temp);
1096 		if (*data >= ' ' && *data <= '~')
1097 			buf[count] = *data;
1098 		else if (*data >= 'A' && *data <= 'Z')
1099 			buf[count] = *data;
1100 		else
1101 			buf[count] = '.';
1102 		if (++count == 16) {
1103 			buf[count] = '\000';
1104 			count = 0;
1105 			printf("  %3x  %s %s\n", i, line, buf);
1106 			i+=16;
1107 			line[0] = '\000';
1108 		}
1109 	}
1110 	buf[count] = '\000';
1111 
1112 	for (; count != 16; count++) {
1113 		strcat(line, "   ");
1114 	}
1115 	printf("  %3x  %s %s\n", i, line, buf);
1116 }
1117 #endif
1118