xref: /freebsd/sys/dev/ipmi/ipmi.c (revision b1c5f60ce87cc2f179dfb81de507d9b7bf59564c)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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 #ifdef IPMB
80 static int ipmi_ipmb_checksum(u_char, int);
81 static int ipmi_ipmb_send_message(device_t, u_char, u_char, u_char,
82      u_char, u_char, int)
83 #endif
84 
85 static d_ioctl_t ipmi_ioctl;
86 static d_poll_t ipmi_poll;
87 static d_open_t ipmi_open;
88 static void ipmi_dtor(void *arg);
89 
90 int ipmi_attached = 0;
91 
92 static int on = 1;
93 static bool wd_in_shutdown = false;
94 static int wd_timer_actions = IPMI_SET_WD_ACTION_POWER_CYCLE;
95 static int wd_shutdown_countdown = 0; /* sec */
96 static int wd_startup_countdown = 0; /* sec */
97 static int wd_pretimeout_countdown = 120; /* sec */
98 static int cycle_wait = 10; /* sec */
99 static int wd_init_enable = 1;
100 
101 static SYSCTL_NODE(_hw, OID_AUTO, ipmi, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
102     "IPMI driver parameters");
103 SYSCTL_INT(_hw_ipmi, OID_AUTO, on, CTLFLAG_RWTUN,
104 	&on, 0, "");
105 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_init_enable, CTLFLAG_RWTUN,
106 	&wd_init_enable, 1, "Enable watchdog initialization");
107 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_timer_actions, CTLFLAG_RWTUN,
108 	&wd_timer_actions, 0,
109 	"IPMI watchdog timer actions (including pre-timeout interrupt)");
110 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_shutdown_countdown, CTLFLAG_RWTUN,
111 	&wd_shutdown_countdown, 0,
112 	"IPMI watchdog countdown for shutdown (seconds)");
113 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_startup_countdown, CTLFLAG_RDTUN,
114 	&wd_startup_countdown, 0,
115 	"IPMI watchdog countdown initialized during startup (seconds)");
116 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_pretimeout_countdown, CTLFLAG_RWTUN,
117 	&wd_pretimeout_countdown, 0,
118 	"IPMI watchdog pre-timeout countdown (seconds)");
119 SYSCTL_INT(_hw_ipmi, OID_AUTO, cyle_wait, CTLFLAG_RWTUN,
120 	&cycle_wait, 0,
121 	"IPMI power cycle on reboot delay time (seconds)");
122 
123 static struct cdevsw ipmi_cdevsw = {
124 	.d_version =    D_VERSION,
125 	.d_open =	ipmi_open,
126 	.d_ioctl =	ipmi_ioctl,
127 	.d_poll =	ipmi_poll,
128 	.d_name =	"ipmi",
129 };
130 
131 static MALLOC_DEFINE(M_IPMI, "ipmi", "ipmi");
132 
133 static int
134 ipmi_open(struct cdev *cdev, int flags, int fmt, struct thread *td)
135 {
136 	struct ipmi_device *dev;
137 	struct ipmi_softc *sc;
138 	int error;
139 
140 	if (!on)
141 		return (ENOENT);
142 
143 	/* Initialize the per file descriptor data. */
144 	dev = malloc(sizeof(struct ipmi_device), M_IPMI, M_WAITOK | M_ZERO);
145 	error = devfs_set_cdevpriv(dev, ipmi_dtor);
146 	if (error) {
147 		free(dev, M_IPMI);
148 		return (error);
149 	}
150 
151 	sc = cdev->si_drv1;
152 	TAILQ_INIT(&dev->ipmi_completed_requests);
153 	dev->ipmi_address = IPMI_BMC_SLAVE_ADDR;
154 	dev->ipmi_lun = IPMI_BMC_SMS_LUN;
155 	dev->ipmi_softc = sc;
156 	IPMI_LOCK(sc);
157 	sc->ipmi_opened++;
158 	IPMI_UNLOCK(sc);
159 
160 	return (0);
161 }
162 
163 static int
164 ipmi_poll(struct cdev *cdev, int poll_events, struct thread *td)
165 {
166 	struct ipmi_device *dev;
167 	struct ipmi_softc *sc;
168 	int revents = 0;
169 
170 	if (devfs_get_cdevpriv((void **)&dev))
171 		return (0);
172 
173 	sc = cdev->si_drv1;
174 	IPMI_LOCK(sc);
175 	if (poll_events & (POLLIN | POLLRDNORM)) {
176 		if (!TAILQ_EMPTY(&dev->ipmi_completed_requests))
177 		    revents |= poll_events & (POLLIN | POLLRDNORM);
178 		if (dev->ipmi_requests == 0)
179 		    revents |= POLLERR;
180 	}
181 
182 	if (revents == 0) {
183 		if (poll_events & (POLLIN | POLLRDNORM))
184 			selrecord(td, &dev->ipmi_select);
185 	}
186 	IPMI_UNLOCK(sc);
187 
188 	return (revents);
189 }
190 
191 static void
192 ipmi_purge_completed_requests(struct ipmi_device *dev)
193 {
194 	struct ipmi_request *req;
195 
196 	while (!TAILQ_EMPTY(&dev->ipmi_completed_requests)) {
197 		req = TAILQ_FIRST(&dev->ipmi_completed_requests);
198 		TAILQ_REMOVE(&dev->ipmi_completed_requests, req, ir_link);
199 		dev->ipmi_requests--;
200 		ipmi_free_request(req);
201 	}
202 }
203 
204 static void
205 ipmi_dtor(void *arg)
206 {
207 	struct ipmi_request *req, *nreq;
208 	struct ipmi_device *dev;
209 	struct ipmi_softc *sc;
210 
211 	dev = arg;
212 	sc = dev->ipmi_softc;
213 
214 	IPMI_LOCK(sc);
215 	if (dev->ipmi_requests) {
216 		/* Throw away any pending requests for this device. */
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 #ifdef IPMB
249 static int
250 ipmi_ipmb_checksum(u_char *data, int len)
251 {
252 	u_char sum = 0;
253 
254 	for (; len; len--) {
255 		sum += *data++;
256 	}
257 	return (-sum);
258 }
259 
260 /* XXX: Needs work */
261 static int
262 ipmi_ipmb_send_message(device_t dev, u_char channel, u_char netfn,
263     u_char command, u_char seq, u_char *data, int data_len)
264 {
265 	struct ipmi_softc *sc = device_get_softc(dev);
266 	struct ipmi_request *req;
267 	u_char slave_addr = 0x52;
268 	int error;
269 
270 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
271 	    IPMI_SEND_MSG, data_len + 8, 0);
272 	req->ir_request[0] = channel;
273 	req->ir_request[1] = slave_addr;
274 	req->ir_request[2] = IPMI_ADDR(netfn, 0);
275 	req->ir_request[3] = ipmi_ipmb_checksum(&req->ir_request[1], 2);
276 	req->ir_request[4] = sc->ipmi_address;
277 	req->ir_request[5] = IPMI_ADDR(seq, sc->ipmi_lun);
278 	req->ir_request[6] = command;
279 
280 	bcopy(data, &req->ir_request[7], data_len);
281 	temp[data_len + 7] = ipmi_ipmb_checksum(&req->ir_request[4],
282 	    data_len + 3);
283 
284 	ipmi_submit_driver_request(sc, req);
285 	error = req->ir_error;
286 
287 	return (error);
288 }
289 
290 static int
291 ipmi_handle_attn(struct ipmi_softc *sc)
292 {
293 	struct ipmi_request *req;
294 	int error;
295 
296 	device_printf(sc->ipmi_dev, "BMC has a message\n");
297 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
298 	    IPMI_GET_MSG_FLAGS, 0, 1);
299 
300 	ipmi_submit_driver_request(sc, req);
301 
302 	if (req->ir_error == 0 && req->ir_compcode == 0) {
303 		if (req->ir_reply[0] & IPMI_MSG_BUFFER_FULL) {
304 			device_printf(sc->ipmi_dev, "message buffer full");
305 		}
306 		if (req->ir_reply[0] & IPMI_WDT_PRE_TIMEOUT) {
307 			device_printf(sc->ipmi_dev,
308 			    "watchdog about to go off");
309 		}
310 		if (req->ir_reply[0] & IPMI_MSG_AVAILABLE) {
311 			IPMI_ALLOC_DRIVER_REQUEST(req,
312 			    IPMI_ADDR(IPMI_APP_REQUEST, 0), IPMI_GET_MSG, 0,
313 			    16);
314 
315 			device_printf(sc->ipmi_dev, "throw out message ");
316 			dump_buf(temp, 16);
317 		}
318 	}
319 	error = req->ir_error;
320 
321 	return (error);
322 }
323 #endif
324 
325 static int
326 ipmi_ioctl(struct cdev *cdev, u_long cmd, caddr_t data,
327     int flags, struct thread *td)
328 {
329 	struct ipmi_softc *sc;
330 	struct ipmi_device *dev;
331 	struct ipmi_request *kreq;
332 	struct ipmi_req *req = (struct ipmi_req *)data;
333 	struct ipmi_recv *recv = (struct ipmi_recv *)data;
334 	struct ipmi_addr addr;
335 #ifdef IPMICTL_SEND_COMMAND_32
336 	struct ipmi_req32 *req32 = (struct ipmi_req32 *)data;
337 	struct ipmi_recv32 *recv32 = (struct ipmi_recv32 *)data;
338 	union {
339 		struct ipmi_req req;
340 		struct ipmi_recv recv;
341 	} thunk32;
342 #endif
343 	int error, len;
344 
345 	error = devfs_get_cdevpriv((void **)&dev);
346 	if (error)
347 		return (error);
348 
349 	sc = cdev->si_drv1;
350 
351 #ifdef IPMICTL_SEND_COMMAND_32
352 	/* Convert 32-bit structures to native. */
353 	switch (cmd) {
354 	case IPMICTL_SEND_COMMAND_32:
355 		req = &thunk32.req;
356 		req->addr = PTRIN(req32->addr);
357 		req->addr_len = req32->addr_len;
358 		req->msgid = req32->msgid;
359 		req->msg.netfn = req32->msg.netfn;
360 		req->msg.cmd = req32->msg.cmd;
361 		req->msg.data_len = req32->msg.data_len;
362 		req->msg.data = PTRIN(req32->msg.data);
363 		break;
364 	case IPMICTL_RECEIVE_MSG_TRUNC_32:
365 	case IPMICTL_RECEIVE_MSG_32:
366 		recv = &thunk32.recv;
367 		recv->addr = PTRIN(recv32->addr);
368 		recv->addr_len = recv32->addr_len;
369 		recv->msg.data_len = recv32->msg.data_len;
370 		recv->msg.data = PTRIN(recv32->msg.data);
371 		break;
372 	}
373 #endif
374 
375 	switch (cmd) {
376 #ifdef IPMICTL_SEND_COMMAND_32
377 	case IPMICTL_SEND_COMMAND_32:
378 #endif
379 	case IPMICTL_SEND_COMMAND:
380 		/*
381 		 * XXX: Need to add proper handling of this.
382 		 */
383 		error = copyin(req->addr, &addr, sizeof(addr));
384 		if (error)
385 			return (error);
386 
387 		IPMI_LOCK(sc);
388 		/* clear out old stuff in queue of stuff done */
389 		/* XXX: This seems odd. */
390 		while ((kreq = TAILQ_FIRST(&dev->ipmi_completed_requests))) {
391 			TAILQ_REMOVE(&dev->ipmi_completed_requests, kreq,
392 			    ir_link);
393 			dev->ipmi_requests--;
394 			ipmi_free_request(kreq);
395 		}
396 		IPMI_UNLOCK(sc);
397 
398 		kreq = ipmi_alloc_request(dev, req->msgid,
399 		    IPMI_ADDR(req->msg.netfn, 0), req->msg.cmd,
400 		    req->msg.data_len, IPMI_MAX_RX);
401 		error = copyin(req->msg.data, kreq->ir_request,
402 		    req->msg.data_len);
403 		if (error) {
404 			ipmi_free_request(kreq);
405 			return (error);
406 		}
407 		IPMI_LOCK(sc);
408 		dev->ipmi_requests++;
409 		error = sc->ipmi_enqueue_request(sc, kreq);
410 		IPMI_UNLOCK(sc);
411 		if (error)
412 			return (error);
413 		break;
414 #ifdef IPMICTL_SEND_COMMAND_32
415 	case IPMICTL_RECEIVE_MSG_TRUNC_32:
416 	case IPMICTL_RECEIVE_MSG_32:
417 #endif
418 	case IPMICTL_RECEIVE_MSG_TRUNC:
419 	case IPMICTL_RECEIVE_MSG:
420 		error = copyin(recv->addr, &addr, sizeof(addr));
421 		if (error)
422 			return (error);
423 
424 		IPMI_LOCK(sc);
425 		kreq = TAILQ_FIRST(&dev->ipmi_completed_requests);
426 		if (kreq == NULL) {
427 			IPMI_UNLOCK(sc);
428 			return (EAGAIN);
429 		}
430 		addr.channel = IPMI_BMC_CHANNEL;
431 		/* XXX */
432 		recv->recv_type = IPMI_RESPONSE_RECV_TYPE;
433 		recv->msgid = kreq->ir_msgid;
434 		recv->msg.netfn = IPMI_REPLY_ADDR(kreq->ir_addr) >> 2;
435 		recv->msg.cmd = kreq->ir_command;
436 		error = kreq->ir_error;
437 		if (error) {
438 			TAILQ_REMOVE(&dev->ipmi_completed_requests, kreq,
439 			    ir_link);
440 			dev->ipmi_requests--;
441 			IPMI_UNLOCK(sc);
442 			ipmi_free_request(kreq);
443 			return (error);
444 		}
445 		len = kreq->ir_replylen + 1;
446 		if (recv->msg.data_len < len &&
447 		    (cmd == IPMICTL_RECEIVE_MSG
448 #ifdef IPMICTL_RECEIVE_MSG_32
449 		     || cmd == IPMICTL_RECEIVE_MSG_32
450 #endif
451 		    )) {
452 			IPMI_UNLOCK(sc);
453 			return (EMSGSIZE);
454 		}
455 		TAILQ_REMOVE(&dev->ipmi_completed_requests, kreq, ir_link);
456 		dev->ipmi_requests--;
457 		IPMI_UNLOCK(sc);
458 		len = min(recv->msg.data_len, len);
459 		recv->msg.data_len = len;
460 		error = copyout(&addr, recv->addr,sizeof(addr));
461 		if (error == 0)
462 			error = copyout(&kreq->ir_compcode, recv->msg.data, 1);
463 		if (error == 0)
464 			error = copyout(kreq->ir_reply, recv->msg.data + 1,
465 			    len - 1);
466 		ipmi_free_request(kreq);
467 		if (error)
468 			return (error);
469 		break;
470 	case IPMICTL_SET_MY_ADDRESS_CMD:
471 		IPMI_LOCK(sc);
472 		dev->ipmi_address = *(int*)data;
473 		IPMI_UNLOCK(sc);
474 		break;
475 	case IPMICTL_GET_MY_ADDRESS_CMD:
476 		IPMI_LOCK(sc);
477 		*(int*)data = dev->ipmi_address;
478 		IPMI_UNLOCK(sc);
479 		break;
480 	case IPMICTL_SET_MY_LUN_CMD:
481 		IPMI_LOCK(sc);
482 		dev->ipmi_lun = *(int*)data & 0x3;
483 		IPMI_UNLOCK(sc);
484 		break;
485 	case IPMICTL_GET_MY_LUN_CMD:
486 		IPMI_LOCK(sc);
487 		*(int*)data = dev->ipmi_lun;
488 		IPMI_UNLOCK(sc);
489 		break;
490 	case IPMICTL_SET_GETS_EVENTS_CMD:
491 		/*
492 		device_printf(sc->ipmi_dev,
493 		    "IPMICTL_SET_GETS_EVENTS_CMD NA\n");
494 		*/
495 		break;
496 	case IPMICTL_REGISTER_FOR_CMD:
497 	case IPMICTL_UNREGISTER_FOR_CMD:
498 		return (EOPNOTSUPP);
499 	default:
500 		device_printf(sc->ipmi_dev, "Unknown IOCTL %lX\n", cmd);
501 		return (ENOIOCTL);
502 	}
503 
504 #ifdef IPMICTL_SEND_COMMAND_32
505 	/* Update changed fields in 32-bit structures. */
506 	switch (cmd) {
507 	case IPMICTL_RECEIVE_MSG_TRUNC_32:
508 	case IPMICTL_RECEIVE_MSG_32:
509 		recv32->recv_type = recv->recv_type;
510 		recv32->msgid = recv->msgid;
511 		recv32->msg.netfn = recv->msg.netfn;
512 		recv32->msg.cmd = recv->msg.cmd;
513 		recv32->msg.data_len = recv->msg.data_len;
514 		break;
515 	}
516 #endif
517 	return (0);
518 }
519 
520 /*
521  * Request management.
522  */
523 
524 static __inline void
525 ipmi_init_request(struct ipmi_request *req, struct ipmi_device *dev, long msgid,
526     uint8_t addr, uint8_t command, size_t requestlen, size_t replylen)
527 {
528 
529 	req->ir_owner = dev;
530 	req->ir_msgid = msgid;
531 	req->ir_addr = addr;
532 	req->ir_command = command;
533 	if (requestlen) {
534 		req->ir_request = (char *)&req[1];
535 		req->ir_requestlen = requestlen;
536 	}
537 	if (replylen) {
538 		req->ir_reply = (char *)&req[1] + requestlen;
539 		req->ir_replybuflen = replylen;
540 	}
541 }
542 
543 /* Allocate a new request with request and reply buffers. */
544 struct ipmi_request *
545 ipmi_alloc_request(struct ipmi_device *dev, long msgid, uint8_t addr,
546     uint8_t command, size_t requestlen, size_t replylen)
547 {
548 	struct ipmi_request *req;
549 
550 	req = malloc(sizeof(struct ipmi_request) + requestlen + replylen,
551 	    M_IPMI, M_WAITOK | M_ZERO);
552 	ipmi_init_request(req, dev, msgid, addr, command, requestlen, replylen);
553 	return (req);
554 }
555 
556 /* Free a request no longer in use. */
557 void
558 ipmi_free_request(struct ipmi_request *req)
559 {
560 
561 	free(req, M_IPMI);
562 }
563 
564 /* Store a processed request on the appropriate completion queue. */
565 void
566 ipmi_complete_request(struct ipmi_softc *sc, struct ipmi_request *req)
567 {
568 	struct ipmi_device *dev;
569 
570 	IPMI_LOCK_ASSERT(sc);
571 
572 	/*
573 	 * Anonymous requests (from inside the driver) always have a
574 	 * waiter that we awaken.
575 	 */
576 	if (req->ir_owner == NULL)
577 		wakeup(req);
578 	else {
579 		dev = req->ir_owner;
580 		TAILQ_INSERT_TAIL(&dev->ipmi_completed_requests, req, ir_link);
581 		selwakeup(&dev->ipmi_select);
582 		if (dev->ipmi_closing)
583 			wakeup(&dev->ipmi_requests);
584 	}
585 }
586 
587 /* Perform an internal driver request. */
588 int
589 ipmi_submit_driver_request(struct ipmi_softc *sc, struct ipmi_request *req,
590     int timo)
591 {
592 
593 	return (sc->ipmi_driver_request(sc, req, timo));
594 }
595 
596 /*
597  * Helper routine for polled system interfaces that use
598  * ipmi_polled_enqueue_request() to queue requests.  This request
599  * waits until there is a pending request and then returns the first
600  * request.  If the driver is shutting down, it returns NULL.
601  */
602 struct ipmi_request *
603 ipmi_dequeue_request(struct ipmi_softc *sc)
604 {
605 	struct ipmi_request *req;
606 
607 	IPMI_LOCK_ASSERT(sc);
608 
609 	while (!sc->ipmi_detaching && TAILQ_EMPTY(&sc->ipmi_pending_requests))
610 		cv_wait(&sc->ipmi_request_added, &sc->ipmi_requests_lock);
611 	if (sc->ipmi_detaching)
612 		return (NULL);
613 
614 	req = TAILQ_FIRST(&sc->ipmi_pending_requests);
615 	TAILQ_REMOVE(&sc->ipmi_pending_requests, req, ir_link);
616 	return (req);
617 }
618 
619 /* Default implementation of ipmi_enqueue_request() for polled interfaces. */
620 int
621 ipmi_polled_enqueue_request(struct ipmi_softc *sc, struct ipmi_request *req)
622 {
623 
624 	IPMI_LOCK_ASSERT(sc);
625 
626 	TAILQ_INSERT_TAIL(&sc->ipmi_pending_requests, req, ir_link);
627 	cv_signal(&sc->ipmi_request_added);
628 	return (0);
629 }
630 
631 /*
632  * Watchdog event handler.
633  */
634 
635 static int
636 ipmi_reset_watchdog(struct ipmi_softc *sc)
637 {
638 	struct ipmi_request *req;
639 	int error;
640 
641 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
642 	    IPMI_RESET_WDOG, 0, 0);
643 	error = ipmi_submit_driver_request(sc, req, 0);
644 	if (error) {
645 		device_printf(sc->ipmi_dev, "Failed to reset watchdog\n");
646 	} else if (req->ir_compcode == 0x80) {
647 		error = ENOENT;
648 	} else if (req->ir_compcode != 0) {
649 		device_printf(sc->ipmi_dev, "Watchdog reset returned 0x%x\n",
650 		    req->ir_compcode);
651 		error = EINVAL;
652 	}
653 	return (error);
654 }
655 
656 static int
657 ipmi_set_watchdog(struct ipmi_softc *sc, unsigned int sec)
658 {
659 	struct ipmi_request *req;
660 	int error;
661 
662 	if (sec > 0xffff / 10)
663 		return (EINVAL);
664 
665 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
666 	    IPMI_SET_WDOG, 6, 0);
667 	if (sec) {
668 		req->ir_request[0] = IPMI_SET_WD_TIMER_DONT_STOP
669 		    | IPMI_SET_WD_TIMER_SMS_OS;
670 		req->ir_request[1] = (wd_timer_actions & 0xff);
671 		req->ir_request[2] = min(0xff,
672 		    min(wd_pretimeout_countdown, (sec + 2) / 4));
673 		req->ir_request[3] = 0;	/* Timer use */
674 		req->ir_request[4] = (sec * 10) & 0xff;
675 		req->ir_request[5] = (sec * 10) >> 8;
676 	} else {
677 		req->ir_request[0] = IPMI_SET_WD_TIMER_SMS_OS;
678 		req->ir_request[1] = 0;
679 		req->ir_request[2] = 0;
680 		req->ir_request[3] = 0;	/* Timer use */
681 		req->ir_request[4] = 0;
682 		req->ir_request[5] = 0;
683 	}
684 	error = ipmi_submit_driver_request(sc, req, 0);
685 	if (error) {
686 		device_printf(sc->ipmi_dev, "Failed to set watchdog\n");
687 	} else if (req->ir_compcode != 0) {
688 		device_printf(sc->ipmi_dev, "Watchdog set returned 0x%x\n",
689 		    req->ir_compcode);
690 		error = EINVAL;
691 	}
692 	return (error);
693 }
694 
695 static void
696 ipmi_wd_event(void *arg, unsigned int cmd, int *error)
697 {
698 	struct ipmi_softc *sc = arg;
699 	unsigned int timeout;
700 	int e;
701 
702 	/* Ignore requests while disabled. */
703 	if (!on)
704 		return;
705 
706 	/*
707 	 * To prevent infinite hangs, we don't let anyone pat or change
708 	 * the watchdog when we're shutting down. (See ipmi_shutdown_event().)
709 	 * However, we do want to keep patting the watchdog while we are doing
710 	 * a coredump.
711 	 */
712 	if (wd_in_shutdown) {
713 		if (dumping && sc->ipmi_watchdog_active)
714 			ipmi_reset_watchdog(sc);
715 		return;
716 	}
717 
718 	cmd &= WD_INTERVAL;
719 	if (cmd > 0 && cmd <= 63) {
720 		timeout = ((uint64_t)1 << cmd) / 1000000000;
721 		if (timeout == 0)
722 			timeout = 1;
723 		if (timeout != sc->ipmi_watchdog_active ||
724 		    wd_timer_actions != sc->ipmi_watchdog_actions ||
725 		    wd_pretimeout_countdown != sc->ipmi_watchdog_pretimeout) {
726 			e = ipmi_set_watchdog(sc, timeout);
727 			if (e == 0) {
728 				sc->ipmi_watchdog_active = timeout;
729 				sc->ipmi_watchdog_actions = wd_timer_actions;
730 				sc->ipmi_watchdog_pretimeout = wd_pretimeout_countdown;
731 			} else {
732 				(void)ipmi_set_watchdog(sc, 0);
733 				sc->ipmi_watchdog_active = 0;
734 				sc->ipmi_watchdog_actions = 0;
735 				sc->ipmi_watchdog_pretimeout = 0;
736 			}
737 		}
738 		if (sc->ipmi_watchdog_active != 0) {
739 			e = ipmi_reset_watchdog(sc);
740 			if (e == 0) {
741 				*error = 0;
742 			} else {
743 				(void)ipmi_set_watchdog(sc, 0);
744 				sc->ipmi_watchdog_active = 0;
745 				sc->ipmi_watchdog_actions = 0;
746 				sc->ipmi_watchdog_pretimeout = 0;
747 			}
748 		}
749 	} else if (atomic_readandclear_int(&sc->ipmi_watchdog_active) != 0) {
750 		sc->ipmi_watchdog_actions = 0;
751 		sc->ipmi_watchdog_pretimeout = 0;
752 
753 		e = ipmi_set_watchdog(sc, 0);
754 		if (e != 0 && cmd == 0)
755 			*error = EOPNOTSUPP;
756 	}
757 }
758 
759 static void
760 ipmi_shutdown_event(void *arg, unsigned int cmd, int *error)
761 {
762 	struct ipmi_softc *sc = arg;
763 
764 	/* Ignore event if disabled. */
765 	if (!on)
766 		return;
767 
768 	/*
769 	 * Positive wd_shutdown_countdown value will re-arm watchdog;
770 	 * Zero value in wd_shutdown_countdown will disable watchdog;
771 	 * Negative value in wd_shutdown_countdown will keep existing state;
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) {
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 cyle_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);
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 devclass_t ipmi_devclass;
1064 
1065 /* XXX: Why? */
1066 static void
1067 ipmi_unload(void *arg)
1068 {
1069 	device_t *	devs;
1070 	int		count;
1071 	int		i;
1072 
1073 	if (ipmi_devclass == NULL)
1074 		return;
1075 	if (devclass_get_devices(ipmi_devclass, &devs, &count) != 0)
1076 		return;
1077 	for (i = 0; i < count; i++)
1078 		device_delete_child(device_get_parent(devs[i]), devs[i]);
1079 	free(devs, M_TEMP);
1080 }
1081 SYSUNINIT(ipmi_unload, SI_SUB_DRIVERS, SI_ORDER_FIRST, ipmi_unload, NULL);
1082 
1083 #ifdef IMPI_DEBUG
1084 static void
1085 dump_buf(u_char *data, int len)
1086 {
1087 	char buf[20];
1088 	char line[1024];
1089 	char temp[30];
1090 	int count = 0;
1091 	int i=0;
1092 
1093 	printf("Address %p len %d\n", data, len);
1094 	if (len > 256)
1095 		len = 256;
1096 	line[0] = '\000';
1097 	for (; len > 0; len--, data++) {
1098 		sprintf(temp, "%02x ", *data);
1099 		strcat(line, temp);
1100 		if (*data >= ' ' && *data <= '~')
1101 			buf[count] = *data;
1102 		else if (*data >= 'A' && *data <= 'Z')
1103 			buf[count] = *data;
1104 		else
1105 			buf[count] = '.';
1106 		if (++count == 16) {
1107 			buf[count] = '\000';
1108 			count = 0;
1109 			printf("  %3x  %s %s\n", i, line, buf);
1110 			i+=16;
1111 			line[0] = '\000';
1112 		}
1113 	}
1114 	buf[count] = '\000';
1115 
1116 	for (; count != 16; count++) {
1117 		strcat(line, "   ");
1118 	}
1119 	printf("  %3x  %s %s\n", i, line, buf);
1120 }
1121 #endif
1122