xref: /freebsd-13-stable/sys/dev/ipmi/ipmi.c (revision 9ca7451d86dcf8fb072491028e3b0fb50b2299c5)
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 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/bus.h>
33 #include <sys/condvar.h>
34 #include <sys/conf.h>
35 #include <sys/eventhandler.h>
36 #include <sys/kernel.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/module.h>
40 #include <sys/mutex.h>
41 #include <sys/poll.h>
42 #include <sys/reboot.h>
43 #include <sys/rman.h>
44 #include <sys/selinfo.h>
45 #include <sys/sysctl.h>
46 #include <sys/watchdog.h>
47 
48 #ifdef LOCAL_MODULE
49 #include <ipmi.h>
50 #include <ipmivars.h>
51 #else
52 #include <sys/ipmi.h>
53 #include <dev/ipmi/ipmivars.h>
54 #endif
55 
56 #ifdef IPMICTL_SEND_COMMAND_32
57 #include <sys/abi_compat.h>
58 #endif
59 
60 /*
61  * Driver request structures are allocated on the stack via alloca() to
62  * avoid calling malloc(), especially for the watchdog handler.
63  * To avoid too much stack growth, a previously allocated structure can
64  * be reused via IPMI_INIT_DRIVER_REQUEST(), but the caller should ensure
65  * that there is adequate reply/request space in the original allocation.
66  */
67 #define	IPMI_INIT_DRIVER_REQUEST(req, addr, cmd, reqlen, replylen)	\
68 	bzero((req), sizeof(struct ipmi_request));			\
69 	ipmi_init_request((req), NULL, 0, (addr), (cmd), (reqlen), (replylen))
70 
71 #define	IPMI_ALLOC_DRIVER_REQUEST(req, addr, cmd, reqlen, replylen)	\
72 	(req) = __builtin_alloca(sizeof(struct ipmi_request) +		\
73 	    (reqlen) + (replylen));					\
74 	IPMI_INIT_DRIVER_REQUEST((req), (addr), (cmd), (reqlen),	\
75 	    (replylen))
76 
77 static d_ioctl_t ipmi_ioctl;
78 static d_poll_t ipmi_poll;
79 static d_open_t ipmi_open;
80 static void ipmi_dtor(void *arg);
81 
82 int ipmi_attached = 0;
83 
84 static int on = 1;
85 static bool wd_in_shutdown = false;
86 static int wd_timer_actions = IPMI_SET_WD_ACTION_POWER_CYCLE;
87 static int wd_shutdown_countdown = 0; /* sec */
88 static int wd_startup_countdown = 0; /* sec */
89 static int wd_pretimeout_countdown = 120; /* sec */
90 static int cycle_wait = 10; /* sec */
91 static int wd_init_enable = 1;
92 
93 static SYSCTL_NODE(_hw, OID_AUTO, ipmi, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
94     "IPMI driver parameters");
95 SYSCTL_INT(_hw_ipmi, OID_AUTO, on, CTLFLAG_RWTUN,
96 	&on, 0, "");
97 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_init_enable, CTLFLAG_RWTUN,
98 	&wd_init_enable, 1, "Enable watchdog initialization");
99 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_timer_actions, CTLFLAG_RWTUN,
100 	&wd_timer_actions, 0,
101 	"IPMI watchdog timer actions (including pre-timeout interrupt)");
102 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_shutdown_countdown, CTLFLAG_RWTUN,
103 	&wd_shutdown_countdown, 0,
104 	"IPMI watchdog countdown for shutdown (seconds)");
105 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_startup_countdown, CTLFLAG_RDTUN,
106 	&wd_startup_countdown, 0,
107 	"IPMI watchdog countdown initialized during startup (seconds)");
108 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_pretimeout_countdown, CTLFLAG_RWTUN,
109 	&wd_pretimeout_countdown, 0,
110 	"IPMI watchdog pre-timeout countdown (seconds)");
111 SYSCTL_INT(_hw_ipmi, OID_AUTO, cyle_wait, CTLFLAG_RWTUN,
112 	&cycle_wait, 0,
113 	"IPMI power cycle on reboot delay time (seconds)");
114 
115 static struct cdevsw ipmi_cdevsw = {
116 	.d_version =    D_VERSION,
117 	.d_open =	ipmi_open,
118 	.d_ioctl =	ipmi_ioctl,
119 	.d_poll =	ipmi_poll,
120 	.d_name =	"ipmi",
121 };
122 
123 static MALLOC_DEFINE(M_IPMI, "ipmi", "ipmi");
124 
125 static int
ipmi_open(struct cdev * cdev,int flags,int fmt,struct thread * td)126 ipmi_open(struct cdev *cdev, int flags, int fmt, struct thread *td)
127 {
128 	struct ipmi_device *dev;
129 	struct ipmi_softc *sc;
130 	int error;
131 
132 	if (!on)
133 		return (ENOENT);
134 
135 	/* Initialize the per file descriptor data. */
136 	dev = malloc(sizeof(struct ipmi_device), M_IPMI, M_WAITOK | M_ZERO);
137 	error = devfs_set_cdevpriv(dev, ipmi_dtor);
138 	if (error) {
139 		free(dev, M_IPMI);
140 		return (error);
141 	}
142 
143 	sc = cdev->si_drv1;
144 	TAILQ_INIT(&dev->ipmi_completed_requests);
145 	dev->ipmi_address = IPMI_BMC_SLAVE_ADDR;
146 	dev->ipmi_lun = IPMI_BMC_SMS_LUN;
147 	dev->ipmi_softc = sc;
148 	IPMI_LOCK(sc);
149 	sc->ipmi_opened++;
150 	IPMI_UNLOCK(sc);
151 
152 	return (0);
153 }
154 
155 static int
ipmi_poll(struct cdev * cdev,int poll_events,struct thread * td)156 ipmi_poll(struct cdev *cdev, int poll_events, struct thread *td)
157 {
158 	struct ipmi_device *dev;
159 	struct ipmi_softc *sc;
160 	int revents = 0;
161 
162 	if (devfs_get_cdevpriv((void **)&dev))
163 		return (0);
164 
165 	sc = cdev->si_drv1;
166 	IPMI_LOCK(sc);
167 	if (poll_events & (POLLIN | POLLRDNORM)) {
168 		if (!TAILQ_EMPTY(&dev->ipmi_completed_requests))
169 		    revents |= poll_events & (POLLIN | POLLRDNORM);
170 		if (dev->ipmi_requests == 0)
171 		    revents |= POLLERR;
172 	}
173 
174 	if (revents == 0) {
175 		if (poll_events & (POLLIN | POLLRDNORM))
176 			selrecord(td, &dev->ipmi_select);
177 	}
178 	IPMI_UNLOCK(sc);
179 
180 	return (revents);
181 }
182 
183 static void
ipmi_purge_completed_requests(struct ipmi_device * dev)184 ipmi_purge_completed_requests(struct ipmi_device *dev)
185 {
186 	struct ipmi_request *req;
187 
188 	while (!TAILQ_EMPTY(&dev->ipmi_completed_requests)) {
189 		req = TAILQ_FIRST(&dev->ipmi_completed_requests);
190 		TAILQ_REMOVE(&dev->ipmi_completed_requests, req, ir_link);
191 		dev->ipmi_requests--;
192 		ipmi_free_request(req);
193 	}
194 }
195 
196 static void
ipmi_dtor(void * arg)197 ipmi_dtor(void *arg)
198 {
199 	struct ipmi_request *req, *nreq;
200 	struct ipmi_device *dev;
201 	struct ipmi_softc *sc;
202 
203 	dev = arg;
204 	sc = dev->ipmi_softc;
205 
206 	IPMI_LOCK(sc);
207 	if (dev->ipmi_requests) {
208 		/* Throw away any pending requests for this device. */
209 		TAILQ_FOREACH_SAFE(req, &sc->ipmi_pending_requests, ir_link,
210 		    nreq) {
211 			if (req->ir_owner == dev) {
212 				TAILQ_REMOVE(&sc->ipmi_pending_requests, req,
213 				    ir_link);
214 				dev->ipmi_requests--;
215 				ipmi_free_request(req);
216 			}
217 		}
218 
219 		/* Throw away any pending completed requests for this device. */
220 		ipmi_purge_completed_requests(dev);
221 
222 		/*
223 		 * If we still have outstanding requests, they must be stuck
224 		 * in an interface driver, so wait for those to drain.
225 		 */
226 		dev->ipmi_closing = 1;
227 		while (dev->ipmi_requests > 0) {
228 			msleep(&dev->ipmi_requests, &sc->ipmi_requests_lock,
229 			    PWAIT, "ipmidrain", 0);
230 			ipmi_purge_completed_requests(dev);
231 		}
232 	}
233 	sc->ipmi_opened--;
234 	IPMI_UNLOCK(sc);
235 
236 	/* Cleanup. */
237 	free(dev, M_IPMI);
238 }
239 
240 static u_char
ipmi_ipmb_checksum(u_char * data,int len)241 ipmi_ipmb_checksum(u_char *data, int len)
242 {
243 	u_char sum = 0;
244 
245 	for (; len; len--)
246 		sum += *data++;
247 	return (-sum);
248 }
249 
250 static int
ipmi_ioctl(struct cdev * cdev,u_long cmd,caddr_t data,int flags,struct thread * td)251 ipmi_ioctl(struct cdev *cdev, u_long cmd, caddr_t data,
252     int flags, struct thread *td)
253 {
254 	struct ipmi_softc *sc;
255 	struct ipmi_device *dev;
256 	struct ipmi_request *kreq;
257 	struct ipmi_req *req = (struct ipmi_req *)data;
258 	struct ipmi_recv *recv = (struct ipmi_recv *)data;
259 	struct ipmi_addr addr;
260 #ifdef IPMICTL_SEND_COMMAND_32
261 	struct ipmi_req32 *req32 = (struct ipmi_req32 *)data;
262 	struct ipmi_recv32 *recv32 = (struct ipmi_recv32 *)data;
263 	union {
264 		struct ipmi_req req;
265 		struct ipmi_recv recv;
266 	} thunk32;
267 #endif
268 	int error, len;
269 
270 	error = devfs_get_cdevpriv((void **)&dev);
271 	if (error)
272 		return (error);
273 
274 	sc = cdev->si_drv1;
275 
276 #ifdef IPMICTL_SEND_COMMAND_32
277 	/* Convert 32-bit structures to native. */
278 	switch (cmd) {
279 	case IPMICTL_SEND_COMMAND_32:
280 		req = &thunk32.req;
281 		req->addr = PTRIN(req32->addr);
282 		req->addr_len = req32->addr_len;
283 		req->msgid = req32->msgid;
284 		req->msg.netfn = req32->msg.netfn;
285 		req->msg.cmd = req32->msg.cmd;
286 		req->msg.data_len = req32->msg.data_len;
287 		req->msg.data = PTRIN(req32->msg.data);
288 		break;
289 	case IPMICTL_RECEIVE_MSG_TRUNC_32:
290 	case IPMICTL_RECEIVE_MSG_32:
291 		recv = &thunk32.recv;
292 		recv->addr = PTRIN(recv32->addr);
293 		recv->addr_len = recv32->addr_len;
294 		recv->msg.data_len = recv32->msg.data_len;
295 		recv->msg.data = PTRIN(recv32->msg.data);
296 		break;
297 	}
298 #endif
299 
300 	switch (cmd) {
301 #ifdef IPMICTL_SEND_COMMAND_32
302 	case IPMICTL_SEND_COMMAND_32:
303 #endif
304 	case IPMICTL_SEND_COMMAND:
305 		error = copyin(req->addr, &addr, sizeof(addr));
306 		if (error)
307 			return (error);
308 
309 		if (addr.addr_type == IPMI_SYSTEM_INTERFACE_ADDR_TYPE) {
310 			struct ipmi_system_interface_addr *saddr =
311 			    (struct ipmi_system_interface_addr *)&addr;
312 
313 			kreq = ipmi_alloc_request(dev, req->msgid,
314 			    IPMI_ADDR(req->msg.netfn, saddr->lun & 0x3),
315 			    req->msg.cmd, req->msg.data_len, IPMI_MAX_RX);
316 			error = copyin(req->msg.data, kreq->ir_request,
317 			    req->msg.data_len);
318 			if (error) {
319 				ipmi_free_request(kreq);
320 				return (error);
321 			}
322 			IPMI_LOCK(sc);
323 			dev->ipmi_requests++;
324 			error = sc->ipmi_enqueue_request(sc, kreq);
325 			IPMI_UNLOCK(sc);
326 			if (error)
327 				return (error);
328 			break;
329 		}
330 
331 		/* Special processing for IPMB commands */
332 		struct ipmi_ipmb_addr *iaddr = (struct ipmi_ipmb_addr *)&addr;
333 
334 		IPMI_ALLOC_DRIVER_REQUEST(kreq, IPMI_ADDR(IPMI_APP_REQUEST, 0),
335 		    IPMI_SEND_MSG, req->msg.data_len + 8, IPMI_MAX_RX);
336 		/* Construct the SEND MSG header */
337 		kreq->ir_request[0] = iaddr->channel;
338 		kreq->ir_request[1] = iaddr->slave_addr;
339 		kreq->ir_request[2] = IPMI_ADDR(req->msg.netfn, iaddr->lun);
340 		kreq->ir_request[3] =
341 		    ipmi_ipmb_checksum(&kreq->ir_request[1], 2);
342 		kreq->ir_request[4] = dev->ipmi_address;
343 		kreq->ir_request[5] = IPMI_ADDR(0, dev->ipmi_lun);
344 		kreq->ir_request[6] = req->msg.cmd;
345 		/* Copy the message data */
346 		if (req->msg.data_len > 0) {
347 			error = copyin(req->msg.data, &kreq->ir_request[7],
348 			    req->msg.data_len);
349 			if (error != 0)
350 				return (error);
351 		}
352 		kreq->ir_request[req->msg.data_len + 7] =
353 		    ipmi_ipmb_checksum(&kreq->ir_request[4],
354 		    req->msg.data_len + 3);
355 		error = ipmi_submit_driver_request(sc, kreq, MAX_TIMEOUT);
356 		if (error != 0)
357 			return (error);
358 
359 		kreq = ipmi_alloc_request(dev, req->msgid,
360 		    IPMI_ADDR(IPMI_APP_REQUEST, 0), IPMI_GET_MSG,
361 		    0, IPMI_MAX_RX);
362 		kreq->ir_ipmb = true;
363 		kreq->ir_ipmb_addr = IPMI_ADDR(req->msg.netfn, 0);
364 		kreq->ir_ipmb_command = req->msg.cmd;
365 		IPMI_LOCK(sc);
366 		dev->ipmi_requests++;
367 		error = sc->ipmi_enqueue_request(sc, kreq);
368 		IPMI_UNLOCK(sc);
369 		if (error != 0)
370 			return (error);
371 		break;
372 #ifdef IPMICTL_SEND_COMMAND_32
373 	case IPMICTL_RECEIVE_MSG_TRUNC_32:
374 	case IPMICTL_RECEIVE_MSG_32:
375 #endif
376 	case IPMICTL_RECEIVE_MSG_TRUNC:
377 	case IPMICTL_RECEIVE_MSG:
378 		error = copyin(recv->addr, &addr, sizeof(addr));
379 		if (error)
380 			return (error);
381 
382 		IPMI_LOCK(sc);
383 		kreq = TAILQ_FIRST(&dev->ipmi_completed_requests);
384 		if (kreq == NULL) {
385 			IPMI_UNLOCK(sc);
386 			return (EAGAIN);
387 		}
388 		if (kreq->ir_error != 0) {
389 			error = kreq->ir_error;
390 			TAILQ_REMOVE(&dev->ipmi_completed_requests, kreq,
391 			    ir_link);
392 			dev->ipmi_requests--;
393 			IPMI_UNLOCK(sc);
394 			ipmi_free_request(kreq);
395 			return (error);
396 		}
397 
398 		recv->recv_type = IPMI_RESPONSE_RECV_TYPE;
399 		recv->msgid = kreq->ir_msgid;
400 		if (kreq->ir_ipmb) {
401 			addr.channel = IPMI_IPMB_CHANNEL;
402 			recv->msg.netfn =
403 			    IPMI_REPLY_ADDR(kreq->ir_ipmb_addr) >> 2;
404 			recv->msg.cmd = kreq->ir_ipmb_command;
405 			/* Get the compcode of response */
406 			kreq->ir_compcode = kreq->ir_reply[6];
407 			/* Move the reply head past response header */
408 			kreq->ir_reply += 7;
409 			len = kreq->ir_replylen - 7;
410 		} else {
411 			addr.channel = IPMI_BMC_CHANNEL;
412 			recv->msg.netfn = IPMI_REPLY_ADDR(kreq->ir_addr) >> 2;
413 			recv->msg.cmd = kreq->ir_command;
414 			len = kreq->ir_replylen + 1;
415 		}
416 
417 		if (recv->msg.data_len < len &&
418 		    (cmd == IPMICTL_RECEIVE_MSG
419 #ifdef IPMICTL_RECEIVE_MSG_32
420 		    || cmd == IPMICTL_RECEIVE_MSG_32
421 #endif
422 		    )) {
423 			IPMI_UNLOCK(sc);
424 			return (EMSGSIZE);
425 		}
426 		TAILQ_REMOVE(&dev->ipmi_completed_requests, kreq, ir_link);
427 		dev->ipmi_requests--;
428 		IPMI_UNLOCK(sc);
429 		len = min(recv->msg.data_len, len);
430 		recv->msg.data_len = len;
431 		error = copyout(&addr, recv->addr,sizeof(addr));
432 		if (error == 0)
433 			error = copyout(&kreq->ir_compcode, recv->msg.data, 1);
434 		if (error == 0)
435 			error = copyout(kreq->ir_reply, recv->msg.data + 1,
436 			    len - 1);
437 		ipmi_free_request(kreq);
438 		if (error)
439 			return (error);
440 		break;
441 	case IPMICTL_SET_MY_ADDRESS_CMD:
442 		IPMI_LOCK(sc);
443 		dev->ipmi_address = *(int*)data;
444 		IPMI_UNLOCK(sc);
445 		break;
446 	case IPMICTL_GET_MY_ADDRESS_CMD:
447 		IPMI_LOCK(sc);
448 		*(int*)data = dev->ipmi_address;
449 		IPMI_UNLOCK(sc);
450 		break;
451 	case IPMICTL_SET_MY_LUN_CMD:
452 		IPMI_LOCK(sc);
453 		dev->ipmi_lun = *(int*)data & 0x3;
454 		IPMI_UNLOCK(sc);
455 		break;
456 	case IPMICTL_GET_MY_LUN_CMD:
457 		IPMI_LOCK(sc);
458 		*(int*)data = dev->ipmi_lun;
459 		IPMI_UNLOCK(sc);
460 		break;
461 	case IPMICTL_SET_GETS_EVENTS_CMD:
462 		/*
463 		device_printf(sc->ipmi_dev,
464 		    "IPMICTL_SET_GETS_EVENTS_CMD NA\n");
465 		*/
466 		break;
467 	case IPMICTL_REGISTER_FOR_CMD:
468 	case IPMICTL_UNREGISTER_FOR_CMD:
469 		return (EOPNOTSUPP);
470 	default:
471 		device_printf(sc->ipmi_dev, "Unknown IOCTL %lX\n", cmd);
472 		return (ENOIOCTL);
473 	}
474 
475 #ifdef IPMICTL_SEND_COMMAND_32
476 	/* Update changed fields in 32-bit structures. */
477 	switch (cmd) {
478 	case IPMICTL_RECEIVE_MSG_TRUNC_32:
479 	case IPMICTL_RECEIVE_MSG_32:
480 		recv32->recv_type = recv->recv_type;
481 		recv32->msgid = recv->msgid;
482 		recv32->msg.netfn = recv->msg.netfn;
483 		recv32->msg.cmd = recv->msg.cmd;
484 		recv32->msg.data_len = recv->msg.data_len;
485 		break;
486 	}
487 #endif
488 	return (0);
489 }
490 
491 /*
492  * Request management.
493  */
494 
495 __inline void
ipmi_init_request(struct ipmi_request * req,struct ipmi_device * dev,long msgid,uint8_t addr,uint8_t command,size_t requestlen,size_t replylen)496 ipmi_init_request(struct ipmi_request *req, struct ipmi_device *dev, long msgid,
497     uint8_t addr, uint8_t command, size_t requestlen, size_t replylen)
498 {
499 
500 	req->ir_owner = dev;
501 	req->ir_msgid = msgid;
502 	req->ir_addr = addr;
503 	req->ir_command = command;
504 	if (requestlen) {
505 		req->ir_request = (char *)&req[1];
506 		req->ir_requestlen = requestlen;
507 	}
508 	if (replylen) {
509 		req->ir_reply = (char *)&req[1] + requestlen;
510 		req->ir_replybuflen = replylen;
511 	}
512 }
513 
514 /* Allocate a new request with request and reply buffers. */
515 struct ipmi_request *
ipmi_alloc_request(struct ipmi_device * dev,long msgid,uint8_t addr,uint8_t command,size_t requestlen,size_t replylen)516 ipmi_alloc_request(struct ipmi_device *dev, long msgid, uint8_t addr,
517     uint8_t command, size_t requestlen, size_t replylen)
518 {
519 	struct ipmi_request *req;
520 
521 	req = malloc(sizeof(struct ipmi_request) + requestlen + replylen,
522 	    M_IPMI, M_WAITOK | M_ZERO);
523 	ipmi_init_request(req, dev, msgid, addr, command, requestlen, replylen);
524 	return (req);
525 }
526 
527 /* Free a request no longer in use. */
528 void
ipmi_free_request(struct ipmi_request * req)529 ipmi_free_request(struct ipmi_request *req)
530 {
531 
532 	free(req, M_IPMI);
533 }
534 
535 /* Store a processed request on the appropriate completion queue. */
536 void
ipmi_complete_request(struct ipmi_softc * sc,struct ipmi_request * req)537 ipmi_complete_request(struct ipmi_softc *sc, struct ipmi_request *req)
538 {
539 	struct ipmi_device *dev;
540 
541 	IPMI_LOCK_ASSERT(sc);
542 
543 	/*
544 	 * Anonymous requests (from inside the driver) always have a
545 	 * waiter that we awaken.
546 	 */
547 	if (req->ir_owner == NULL)
548 		wakeup(req);
549 	else {
550 		dev = req->ir_owner;
551 		TAILQ_INSERT_TAIL(&dev->ipmi_completed_requests, req, ir_link);
552 		selwakeup(&dev->ipmi_select);
553 		if (dev->ipmi_closing)
554 			wakeup(&dev->ipmi_requests);
555 	}
556 }
557 
558 /* Perform an internal driver request. */
559 int
ipmi_submit_driver_request(struct ipmi_softc * sc,struct ipmi_request * req,int timo)560 ipmi_submit_driver_request(struct ipmi_softc *sc, struct ipmi_request *req,
561     int timo)
562 {
563 
564 	return (sc->ipmi_driver_request(sc, req, timo));
565 }
566 
567 /*
568  * Helper routine for polled system interfaces that use
569  * ipmi_polled_enqueue_request() to queue requests.  This request
570  * waits until there is a pending request and then returns the first
571  * request.  If the driver is shutting down, it returns NULL.
572  */
573 struct ipmi_request *
ipmi_dequeue_request(struct ipmi_softc * sc)574 ipmi_dequeue_request(struct ipmi_softc *sc)
575 {
576 	struct ipmi_request *req;
577 
578 	IPMI_LOCK_ASSERT(sc);
579 
580 	while (!sc->ipmi_detaching && TAILQ_EMPTY(&sc->ipmi_pending_requests))
581 		cv_wait(&sc->ipmi_request_added, &sc->ipmi_requests_lock);
582 	if (sc->ipmi_detaching)
583 		return (NULL);
584 
585 	req = TAILQ_FIRST(&sc->ipmi_pending_requests);
586 	TAILQ_REMOVE(&sc->ipmi_pending_requests, req, ir_link);
587 	return (req);
588 }
589 
590 /* Default implementation of ipmi_enqueue_request() for polled interfaces. */
591 int
ipmi_polled_enqueue_request(struct ipmi_softc * sc,struct ipmi_request * req)592 ipmi_polled_enqueue_request(struct ipmi_softc *sc, struct ipmi_request *req)
593 {
594 
595 	IPMI_LOCK_ASSERT(sc);
596 
597 	TAILQ_INSERT_TAIL(&sc->ipmi_pending_requests, req, ir_link);
598 	cv_signal(&sc->ipmi_request_added);
599 	return (0);
600 }
601 
602 /*
603  * Watchdog event handler.
604  */
605 
606 static int
ipmi_reset_watchdog(struct ipmi_softc * sc)607 ipmi_reset_watchdog(struct ipmi_softc *sc)
608 {
609 	struct ipmi_request *req;
610 	int error;
611 
612 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
613 	    IPMI_RESET_WDOG, 0, 0);
614 	error = ipmi_submit_driver_request(sc, req, 0);
615 	if (error) {
616 		device_printf(sc->ipmi_dev, "Failed to reset watchdog\n");
617 	} else if (req->ir_compcode == 0x80) {
618 		error = ENOENT;
619 	} else if (req->ir_compcode != 0) {
620 		device_printf(sc->ipmi_dev, "Watchdog reset returned 0x%x\n",
621 		    req->ir_compcode);
622 		error = EINVAL;
623 	}
624 	return (error);
625 }
626 
627 static int
ipmi_set_watchdog(struct ipmi_softc * sc,unsigned int sec)628 ipmi_set_watchdog(struct ipmi_softc *sc, unsigned int sec)
629 {
630 	struct ipmi_request *req;
631 	int error;
632 
633 	if (sec > 0xffff / 10)
634 		return (EINVAL);
635 
636 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
637 	    IPMI_SET_WDOG, 6, 0);
638 	if (sec) {
639 		req->ir_request[0] = IPMI_SET_WD_TIMER_DONT_STOP
640 		    | IPMI_SET_WD_TIMER_SMS_OS;
641 		req->ir_request[1] = (wd_timer_actions & 0xff);
642 		req->ir_request[2] = min(0xff,
643 		    min(wd_pretimeout_countdown, (sec + 2) / 4));
644 		req->ir_request[3] = 0;	/* Timer use */
645 		req->ir_request[4] = (sec * 10) & 0xff;
646 		req->ir_request[5] = (sec * 10) >> 8;
647 	} else {
648 		req->ir_request[0] = IPMI_SET_WD_TIMER_SMS_OS;
649 		req->ir_request[1] = 0;
650 		req->ir_request[2] = 0;
651 		req->ir_request[3] = 0;	/* Timer use */
652 		req->ir_request[4] = 0;
653 		req->ir_request[5] = 0;
654 	}
655 	error = ipmi_submit_driver_request(sc, req, 0);
656 	if (error) {
657 		device_printf(sc->ipmi_dev, "Failed to set watchdog\n");
658 	} else if (req->ir_compcode != 0) {
659 		device_printf(sc->ipmi_dev, "Watchdog set returned 0x%x\n",
660 		    req->ir_compcode);
661 		error = EINVAL;
662 	}
663 	return (error);
664 }
665 
666 static void
ipmi_wd_event(void * arg,unsigned int cmd,int * error)667 ipmi_wd_event(void *arg, unsigned int cmd, int *error)
668 {
669 	struct ipmi_softc *sc = arg;
670 	unsigned int timeout;
671 	int e;
672 
673 	/* Ignore requests while disabled. */
674 	if (!on)
675 		return;
676 
677 	/*
678 	 * To prevent infinite hangs, we don't let anyone pat or change
679 	 * the watchdog when we're shutting down. (See ipmi_shutdown_event().)
680 	 * However, we do want to keep patting the watchdog while we are doing
681 	 * a coredump.
682 	 */
683 	if (wd_in_shutdown) {
684 		if (dumping && sc->ipmi_watchdog_active)
685 			ipmi_reset_watchdog(sc);
686 		return;
687 	}
688 
689 	cmd &= WD_INTERVAL;
690 	if (cmd > 0 && cmd <= 63) {
691 		timeout = ((uint64_t)1 << cmd) / 1000000000;
692 		if (timeout == 0)
693 			timeout = 1;
694 		if (timeout != sc->ipmi_watchdog_active ||
695 		    wd_timer_actions != sc->ipmi_watchdog_actions ||
696 		    wd_pretimeout_countdown != sc->ipmi_watchdog_pretimeout) {
697 			e = ipmi_set_watchdog(sc, timeout);
698 			if (e == 0) {
699 				sc->ipmi_watchdog_active = timeout;
700 				sc->ipmi_watchdog_actions = wd_timer_actions;
701 				sc->ipmi_watchdog_pretimeout = wd_pretimeout_countdown;
702 			} else {
703 				(void)ipmi_set_watchdog(sc, 0);
704 				sc->ipmi_watchdog_active = 0;
705 				sc->ipmi_watchdog_actions = 0;
706 				sc->ipmi_watchdog_pretimeout = 0;
707 			}
708 		}
709 		if (sc->ipmi_watchdog_active != 0) {
710 			e = ipmi_reset_watchdog(sc);
711 			if (e == 0) {
712 				*error = 0;
713 			} else {
714 				(void)ipmi_set_watchdog(sc, 0);
715 				sc->ipmi_watchdog_active = 0;
716 				sc->ipmi_watchdog_actions = 0;
717 				sc->ipmi_watchdog_pretimeout = 0;
718 			}
719 		}
720 	} else if (atomic_readandclear_int(&sc->ipmi_watchdog_active) != 0) {
721 		sc->ipmi_watchdog_actions = 0;
722 		sc->ipmi_watchdog_pretimeout = 0;
723 
724 		e = ipmi_set_watchdog(sc, 0);
725 		if (e != 0 && cmd == 0)
726 			*error = EOPNOTSUPP;
727 	}
728 }
729 
730 static void
ipmi_shutdown_event(void * arg,int howto)731 ipmi_shutdown_event(void *arg, int howto)
732 {
733 	struct ipmi_softc *sc = arg;
734 
735 	/* Ignore event if disabled. */
736 	if (!on)
737 		return;
738 
739 	/*
740 	 * Positive wd_shutdown_countdown value will re-arm watchdog;
741 	 * Zero value in wd_shutdown_countdown will disable watchdog;
742 	 * Negative value in wd_shutdown_countdown will keep existing state;
743 	 *
744 	 * System halt is a special case of shutdown where wd_shutdown_countdown
745 	 * is ignored and watchdog is disabled to ensure that the system remains
746 	 * halted as requested.
747 	 *
748 	 * Revert to using a power cycle to ensure that the watchdog will
749 	 * do something useful here.  Having the watchdog send an NMI
750 	 * instead is useless during shutdown, and might be ignored if an
751 	 * NMI already triggered.
752 	 */
753 
754 	wd_in_shutdown = true;
755 	if (wd_shutdown_countdown == 0 || (howto & RB_HALT) != 0) {
756 		/* disable watchdog */
757 		ipmi_set_watchdog(sc, 0);
758 		sc->ipmi_watchdog_active = 0;
759 	} else if (wd_shutdown_countdown > 0) {
760 		/* set desired action and time, and, reset watchdog */
761 		wd_timer_actions = IPMI_SET_WD_ACTION_POWER_CYCLE;
762 		ipmi_set_watchdog(sc, wd_shutdown_countdown);
763 		sc->ipmi_watchdog_active = wd_shutdown_countdown;
764 		ipmi_reset_watchdog(sc);
765 	}
766 }
767 
768 static void
ipmi_power_cycle(void * arg,int howto)769 ipmi_power_cycle(void *arg, int howto)
770 {
771 	struct ipmi_softc *sc = arg;
772 	struct ipmi_request *req;
773 
774 	/*
775 	 * Ignore everything except power cycling requests
776 	 */
777 	if ((howto & RB_POWERCYCLE) == 0)
778 		return;
779 
780 	device_printf(sc->ipmi_dev, "Power cycling using IPMI\n");
781 
782 	/*
783 	 * Send a CHASSIS_CONTROL command to the CHASSIS device, subcommand 2
784 	 * as described in IPMI v2.0 spec section 28.3.
785 	 */
786 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_CHASSIS_REQUEST, 0),
787 	    IPMI_CHASSIS_CONTROL, 1, 0);
788 	req->ir_request[0] = IPMI_CC_POWER_CYCLE;
789 
790 	ipmi_submit_driver_request(sc, req, MAX_TIMEOUT);
791 
792 	if (req->ir_error != 0 || req->ir_compcode != 0) {
793 		device_printf(sc->ipmi_dev, "Power cycling via IPMI failed code %#x %#x\n",
794 		    req->ir_error, req->ir_compcode);
795 		return;
796 	}
797 
798 	/*
799 	 * BMCs are notoriously slow, give it cyle_wait seconds for the power
800 	 * down leg of the power cycle. If that fails, fallback to the next
801 	 * hanlder in the shutdown_final chain and/or the platform failsafe.
802 	 */
803 	DELAY(cycle_wait * 1000 * 1000);
804 	device_printf(sc->ipmi_dev, "Power cycling via IPMI timed out\n");
805 }
806 
807 static void
ipmi_startup(void * arg)808 ipmi_startup(void *arg)
809 {
810 	struct ipmi_softc *sc = arg;
811 	struct ipmi_request *req;
812 	device_t dev;
813 	int error, i;
814 
815 	config_intrhook_disestablish(&sc->ipmi_ich);
816 	dev = sc->ipmi_dev;
817 
818 	/* Initialize interface-independent state. */
819 	mtx_init(&sc->ipmi_requests_lock, "ipmi requests", NULL, MTX_DEF);
820 	mtx_init(&sc->ipmi_io_lock, "ipmi io", NULL, MTX_DEF);
821 	cv_init(&sc->ipmi_request_added, "ipmireq");
822 	TAILQ_INIT(&sc->ipmi_pending_requests);
823 
824 	/* Initialize interface-dependent state. */
825 	error = sc->ipmi_startup(sc);
826 	if (error) {
827 		device_printf(dev, "Failed to initialize interface: %d\n",
828 		    error);
829 		return;
830 	}
831 
832 	/* Send a GET_DEVICE_ID request. */
833 	IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
834 	    IPMI_GET_DEVICE_ID, 0, 15);
835 
836 	error = ipmi_submit_driver_request(sc, req, MAX_TIMEOUT);
837 	if (error == EWOULDBLOCK) {
838 		device_printf(dev, "Timed out waiting for GET_DEVICE_ID\n");
839 		return;
840 	} else if (error) {
841 		device_printf(dev, "Failed GET_DEVICE_ID: %d\n", error);
842 		return;
843 	} else if (req->ir_compcode != 0) {
844 		device_printf(dev,
845 		    "Bad completion code for GET_DEVICE_ID: %d\n",
846 		    req->ir_compcode);
847 		return;
848 	} else if (req->ir_replylen < 5) {
849 		device_printf(dev, "Short reply for GET_DEVICE_ID: %d\n",
850 		    req->ir_replylen);
851 		return;
852 	}
853 
854 	device_printf(dev, "IPMI device rev. %d, firmware rev. %d.%d%d, "
855 	    "version %d.%d, device support mask %#x\n",
856 	    req->ir_reply[1] & 0x0f,
857 	    req->ir_reply[2] & 0x7f, req->ir_reply[3] >> 4, req->ir_reply[3] & 0x0f,
858 	    req->ir_reply[4] & 0x0f, req->ir_reply[4] >> 4, req->ir_reply[5]);
859 
860 	sc->ipmi_dev_support = req->ir_reply[5];
861 
862 	IPMI_INIT_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
863 	    IPMI_CLEAR_FLAGS, 1, 0);
864 
865 	ipmi_submit_driver_request(sc, req, 0);
866 
867 	/* XXX: Magic numbers */
868 	if (req->ir_compcode == 0xc0) {
869 		device_printf(dev, "Clear flags is busy\n");
870 	}
871 	if (req->ir_compcode == 0xc1) {
872 		device_printf(dev, "Clear flags illegal\n");
873 	}
874 
875 	for (i = 0; i < 8; i++) {
876 		IPMI_INIT_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
877 		    IPMI_GET_CHANNEL_INFO, 1, 0);
878 		req->ir_request[0] = i;
879 
880 		error = ipmi_submit_driver_request(sc, req, 0);
881 
882 		if (error != 0 || req->ir_compcode != 0)
883 			break;
884 	}
885 	device_printf(dev, "Number of channels %d\n", i);
886 
887 	/*
888 	 * Probe for watchdog, but only for backends which support
889 	 * polled driver requests.
890 	 */
891 	if (wd_init_enable && sc->ipmi_driver_requests_polled) {
892 		IPMI_INIT_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
893 		    IPMI_GET_WDOG, 0, 0);
894 
895 		error = ipmi_submit_driver_request(sc, req, 0);
896 
897 		if (error == 0 && req->ir_compcode == 0x00) {
898 			device_printf(dev, "Attached watchdog\n");
899 			/* register the watchdog event handler */
900 			sc->ipmi_watchdog_tag = EVENTHANDLER_REGISTER(
901 				watchdog_list, ipmi_wd_event, sc, 0);
902 			sc->ipmi_shutdown_tag = EVENTHANDLER_REGISTER(
903 				shutdown_pre_sync, ipmi_shutdown_event,
904 				sc, 0);
905 		}
906 	}
907 
908 	sc->ipmi_cdev = make_dev(&ipmi_cdevsw, device_get_unit(dev),
909 	    UID_ROOT, GID_OPERATOR, 0660, "ipmi%d", device_get_unit(dev));
910 	if (sc->ipmi_cdev == NULL) {
911 		device_printf(dev, "Failed to create cdev\n");
912 		return;
913 	}
914 	sc->ipmi_cdev->si_drv1 = sc;
915 
916 	/*
917 	 * Set initial watchdog state. If desired, set an initial
918 	 * watchdog on startup. Or, if the watchdog device is
919 	 * disabled, clear any existing watchdog.
920 	 */
921 	if (on && wd_startup_countdown > 0) {
922 		if (ipmi_set_watchdog(sc, wd_startup_countdown) == 0 &&
923 		    ipmi_reset_watchdog(sc) == 0) {
924 			sc->ipmi_watchdog_active = wd_startup_countdown;
925 			sc->ipmi_watchdog_actions = wd_timer_actions;
926 			sc->ipmi_watchdog_pretimeout = wd_pretimeout_countdown;
927 		} else
928 			(void)ipmi_set_watchdog(sc, 0);
929 		ipmi_reset_watchdog(sc);
930 	} else if (!on)
931 		(void)ipmi_set_watchdog(sc, 0);
932 	/*
933 	 * Power cycle the system off using IPMI. We use last - 2 since we don't
934 	 * handle all the other kinds of reboots. We'll let others handle them.
935 	 * We only try to do this if the BMC supports the Chassis device.
936 	 */
937 	if (sc->ipmi_dev_support & IPMI_ADS_CHASSIS) {
938 		device_printf(dev, "Establishing power cycle handler\n");
939 		sc->ipmi_power_cycle_tag = EVENTHANDLER_REGISTER(shutdown_final,
940 		    ipmi_power_cycle, sc, SHUTDOWN_PRI_LAST - 2);
941 	}
942 }
943 
944 int
ipmi_attach(device_t dev)945 ipmi_attach(device_t dev)
946 {
947 	struct ipmi_softc *sc = device_get_softc(dev);
948 	int error;
949 
950 	if (sc->ipmi_irq_res != NULL && sc->ipmi_intr != NULL) {
951 		error = bus_setup_intr(dev, sc->ipmi_irq_res, INTR_TYPE_MISC,
952 		    NULL, sc->ipmi_intr, sc, &sc->ipmi_irq);
953 		if (error) {
954 			device_printf(dev, "can't set up interrupt\n");
955 			return (error);
956 		}
957 	}
958 
959 	bzero(&sc->ipmi_ich, sizeof(struct intr_config_hook));
960 	sc->ipmi_ich.ich_func = ipmi_startup;
961 	sc->ipmi_ich.ich_arg = sc;
962 	if (config_intrhook_establish(&sc->ipmi_ich) != 0) {
963 		device_printf(dev, "can't establish configuration hook\n");
964 		return (ENOMEM);
965 	}
966 
967 	ipmi_attached = 1;
968 	return (0);
969 }
970 
971 int
ipmi_detach(device_t dev)972 ipmi_detach(device_t dev)
973 {
974 	struct ipmi_softc *sc;
975 
976 	sc = device_get_softc(dev);
977 
978 	/* Fail if there are any open handles. */
979 	IPMI_LOCK(sc);
980 	if (sc->ipmi_opened) {
981 		IPMI_UNLOCK(sc);
982 		return (EBUSY);
983 	}
984 	IPMI_UNLOCK(sc);
985 	if (sc->ipmi_cdev)
986 		destroy_dev(sc->ipmi_cdev);
987 
988 	/* Detach from watchdog handling and turn off watchdog. */
989 	if (sc->ipmi_shutdown_tag)
990 		EVENTHANDLER_DEREGISTER(shutdown_pre_sync,
991 		sc->ipmi_shutdown_tag);
992 	if (sc->ipmi_watchdog_tag) {
993 		EVENTHANDLER_DEREGISTER(watchdog_list, sc->ipmi_watchdog_tag);
994 		ipmi_set_watchdog(sc, 0);
995 	}
996 
997 	/* Detach from shutdown handling for power cycle reboot */
998 	if (sc->ipmi_power_cycle_tag)
999 		EVENTHANDLER_DEREGISTER(shutdown_final, sc->ipmi_power_cycle_tag);
1000 
1001 	/* XXX: should use shutdown callout I think. */
1002 	/* If the backend uses a kthread, shut it down. */
1003 	IPMI_LOCK(sc);
1004 	sc->ipmi_detaching = 1;
1005 	if (sc->ipmi_kthread) {
1006 		cv_broadcast(&sc->ipmi_request_added);
1007 		msleep(sc->ipmi_kthread, &sc->ipmi_requests_lock, 0,
1008 		    "ipmi_wait", 0);
1009 	}
1010 	IPMI_UNLOCK(sc);
1011 	if (sc->ipmi_irq)
1012 		bus_teardown_intr(dev, sc->ipmi_irq_res, sc->ipmi_irq);
1013 
1014 	ipmi_release_resources(dev);
1015 	mtx_destroy(&sc->ipmi_io_lock);
1016 	mtx_destroy(&sc->ipmi_requests_lock);
1017 	return (0);
1018 }
1019 
1020 void
ipmi_release_resources(device_t dev)1021 ipmi_release_resources(device_t dev)
1022 {
1023 	struct ipmi_softc *sc;
1024 	int i;
1025 
1026 	sc = device_get_softc(dev);
1027 	if (sc->ipmi_irq)
1028 		bus_teardown_intr(dev, sc->ipmi_irq_res, sc->ipmi_irq);
1029 	if (sc->ipmi_irq_res)
1030 		bus_release_resource(dev, SYS_RES_IRQ, sc->ipmi_irq_rid,
1031 		    sc->ipmi_irq_res);
1032 	for (i = 0; i < MAX_RES; i++)
1033 		if (sc->ipmi_io_res[i])
1034 			bus_release_resource(dev, sc->ipmi_io_type,
1035 			    sc->ipmi_io_rid + i, sc->ipmi_io_res[i]);
1036 }
1037 
1038 devclass_t ipmi_devclass;
1039 
1040 /* XXX: Why? */
1041 static void
ipmi_unload(void * arg)1042 ipmi_unload(void *arg)
1043 {
1044 	device_t *	devs;
1045 	int		count;
1046 	int		i;
1047 
1048 	if (ipmi_devclass == NULL)
1049 		return;
1050 	if (devclass_get_devices(ipmi_devclass, &devs, &count) != 0)
1051 		return;
1052 	for (i = 0; i < count; i++)
1053 		device_delete_child(device_get_parent(devs[i]), devs[i]);
1054 	free(devs, M_TEMP);
1055 }
1056 SYSUNINIT(ipmi_unload, SI_SUB_DRIVERS, SI_ORDER_FIRST, ipmi_unload, NULL);
1057 
1058 #ifdef IMPI_DEBUG
1059 static void
dump_buf(u_char * data,int len)1060 dump_buf(u_char *data, int len)
1061 {
1062 	char buf[20];
1063 	char line[1024];
1064 	char temp[30];
1065 	int count = 0;
1066 	int i=0;
1067 
1068 	printf("Address %p len %d\n", data, len);
1069 	if (len > 256)
1070 		len = 256;
1071 	line[0] = '\000';
1072 	for (; len > 0; len--, data++) {
1073 		sprintf(temp, "%02x ", *data);
1074 		strcat(line, temp);
1075 		if (*data >= ' ' && *data <= '~')
1076 			buf[count] = *data;
1077 		else if (*data >= 'A' && *data <= 'Z')
1078 			buf[count] = *data;
1079 		else
1080 			buf[count] = '.';
1081 		if (++count == 16) {
1082 			buf[count] = '\000';
1083 			count = 0;
1084 			printf("  %3x  %s %s\n", i, line, buf);
1085 			i+=16;
1086 			line[0] = '\000';
1087 		}
1088 	}
1089 	buf[count] = '\000';
1090 
1091 	for (; count != 16; count++) {
1092 		strcat(line, "   ");
1093 	}
1094 	printf("  %3x  %s %s\n", i, line, buf);
1095 }
1096 #endif
1097