]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/dev/ipmi/ipmi.c
efirt: When present, attempt to use EFI runtime services to shutdown
[FreeBSD/FreeBSD.git] / sys / dev / ipmi / ipmi.c
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/kernel.h>
38 #include <sys/malloc.h>
39 #include <sys/module.h>
40 #include <sys/poll.h>
41 #include <sys/reboot.h>
42 #include <sys/rman.h>
43 #include <sys/selinfo.h>
44 #include <sys/sysctl.h>
45 #include <sys/watchdog.h>
46
47 #ifdef LOCAL_MODULE
48 #include <ipmi.h>
49 #include <ipmivars.h>
50 #else
51 #include <sys/ipmi.h>
52 #include <dev/ipmi/ipmivars.h>
53 #endif
54
55 /*
56  * Driver request structures are allocated on the stack via alloca() to
57  * avoid calling malloc(), especially for the watchdog handler.
58  * To avoid too much stack growth, a previously allocated structure can
59  * be reused via IPMI_INIT_DRIVER_REQUEST(), but the caller should ensure
60  * that there is adequate reply/request space in the original allocation.
61  */
62 #define IPMI_INIT_DRIVER_REQUEST(req, addr, cmd, reqlen, replylen)      \
63         bzero((req), sizeof(struct ipmi_request));                      \
64         ipmi_init_request((req), NULL, 0, (addr), (cmd), (reqlen), (replylen))
65
66 #define IPMI_ALLOC_DRIVER_REQUEST(req, addr, cmd, reqlen, replylen)     \
67         (req) = __builtin_alloca(sizeof(struct ipmi_request) +          \
68             (reqlen) + (replylen));                                     \
69         IPMI_INIT_DRIVER_REQUEST((req), (addr), (cmd), (reqlen),        \
70             (replylen))
71
72 #ifdef IPMB
73 static int ipmi_ipmb_checksum(u_char, int);
74 static int ipmi_ipmb_send_message(device_t, u_char, u_char, u_char,
75      u_char, u_char, int)
76 #endif
77
78 static d_ioctl_t ipmi_ioctl;
79 static d_poll_t ipmi_poll;
80 static d_open_t ipmi_open;
81 static void ipmi_dtor(void *arg);
82
83 int ipmi_attached = 0;
84
85 static int on = 1;
86 static bool wd_in_shutdown = false;
87 static int wd_timer_actions = IPMI_SET_WD_ACTION_POWER_CYCLE;
88 static int wd_shutdown_countdown = 0; /* sec */
89 static int wd_startup_countdown = 0; /* sec */
90 static int wd_pretimeout_countdown = 120; /* sec */
91 static int cycle_wait = 10; /* sec */
92
93 static SYSCTL_NODE(_hw, OID_AUTO, ipmi, CTLFLAG_RD, 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_timer_actions, CTLFLAG_RW,
98         &wd_timer_actions, 0,
99         "IPMI watchdog timer actions (including pre-timeout interrupt)");
100 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_shutdown_countdown, CTLFLAG_RW,
101         &wd_shutdown_countdown, 0,
102         "IPMI watchdog countdown for shutdown (seconds)");
103 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_startup_countdown, CTLFLAG_RDTUN,
104         &wd_startup_countdown, 0,
105         "IPMI watchdog countdown initialized during startup (seconds)");
106 SYSCTL_INT(_hw_ipmi, OID_AUTO, wd_pretimeout_countdown, CTLFLAG_RW,
107         &wd_pretimeout_countdown, 0,
108         "IPMI watchdog pre-timeout countdown (seconds)");
109 SYSCTL_INT(_hw_ipmi, OID_AUTO, cyle_wait, CTLFLAG_RWTUN,
110         &cycle_wait, 0,
111         "IPMI power cycle on reboot delay time (seconds)");
112
113 static struct cdevsw ipmi_cdevsw = {
114         .d_version =    D_VERSION,
115         .d_open =       ipmi_open,
116         .d_ioctl =      ipmi_ioctl,
117         .d_poll =       ipmi_poll,
118         .d_name =       "ipmi",
119 };
120
121 static MALLOC_DEFINE(M_IPMI, "ipmi", "ipmi");
122
123 static int
124 ipmi_open(struct cdev *cdev, int flags, int fmt, struct thread *td)
125 {
126         struct ipmi_device *dev;
127         struct ipmi_softc *sc;
128         int error;
129
130         if (!on)
131                 return (ENOENT);
132
133         /* Initialize the per file descriptor data. */
134         dev = malloc(sizeof(struct ipmi_device), M_IPMI, M_WAITOK | M_ZERO);
135         error = devfs_set_cdevpriv(dev, ipmi_dtor);
136         if (error) {
137                 free(dev, M_IPMI);
138                 return (error);
139         }
140
141         sc = cdev->si_drv1;
142         TAILQ_INIT(&dev->ipmi_completed_requests);
143         dev->ipmi_address = IPMI_BMC_SLAVE_ADDR;
144         dev->ipmi_lun = IPMI_BMC_SMS_LUN;
145         dev->ipmi_softc = sc;
146         IPMI_LOCK(sc);
147         sc->ipmi_opened++;
148         IPMI_UNLOCK(sc);
149
150         return (0);
151 }
152
153 static int
154 ipmi_poll(struct cdev *cdev, int poll_events, struct thread *td)
155 {
156         struct ipmi_device *dev;
157         struct ipmi_softc *sc;
158         int revents = 0;
159
160         if (devfs_get_cdevpriv((void **)&dev))
161                 return (0);
162
163         sc = cdev->si_drv1;
164         IPMI_LOCK(sc);
165         if (poll_events & (POLLIN | POLLRDNORM)) {
166                 if (!TAILQ_EMPTY(&dev->ipmi_completed_requests))
167                     revents |= poll_events & (POLLIN | POLLRDNORM);
168                 if (dev->ipmi_requests == 0)
169                     revents |= POLLERR;
170         }
171
172         if (revents == 0) {
173                 if (poll_events & (POLLIN | POLLRDNORM))
174                         selrecord(td, &dev->ipmi_select);
175         }
176         IPMI_UNLOCK(sc);
177
178         return (revents);
179 }
180
181 static void
182 ipmi_purge_completed_requests(struct ipmi_device *dev)
183 {
184         struct ipmi_request *req;
185
186         while (!TAILQ_EMPTY(&dev->ipmi_completed_requests)) {
187                 req = TAILQ_FIRST(&dev->ipmi_completed_requests);
188                 TAILQ_REMOVE(&dev->ipmi_completed_requests, req, ir_link);
189                 dev->ipmi_requests--;
190                 ipmi_free_request(req);
191         }
192 }
193
194 static void
195 ipmi_dtor(void *arg)
196 {
197         struct ipmi_request *req, *nreq;
198         struct ipmi_device *dev;
199         struct ipmi_softc *sc;
200
201         dev = arg;
202         sc = dev->ipmi_softc;
203
204         IPMI_LOCK(sc);
205         if (dev->ipmi_requests) {
206                 /* Throw away any pending requests for this device. */
207                 TAILQ_FOREACH_SAFE(req, &sc->ipmi_pending_requests, ir_link,
208                     nreq) {
209                         if (req->ir_owner == dev) {
210                                 TAILQ_REMOVE(&sc->ipmi_pending_requests, req,
211                                     ir_link);
212                                 dev->ipmi_requests--;
213                                 ipmi_free_request(req);
214                         }
215                 }
216
217                 /* Throw away any pending completed requests for this device. */
218                 ipmi_purge_completed_requests(dev);
219
220                 /*
221                  * If we still have outstanding requests, they must be stuck
222                  * in an interface driver, so wait for those to drain.
223                  */
224                 dev->ipmi_closing = 1;
225                 while (dev->ipmi_requests > 0) {
226                         msleep(&dev->ipmi_requests, &sc->ipmi_requests_lock,
227                             PWAIT, "ipmidrain", 0);
228                         ipmi_purge_completed_requests(dev);
229                 }
230         }
231         sc->ipmi_opened--;
232         IPMI_UNLOCK(sc);
233
234         /* Cleanup. */
235         free(dev, M_IPMI);
236 }
237
238 #ifdef IPMB
239 static int
240 ipmi_ipmb_checksum(u_char *data, int len)
241 {
242         u_char sum = 0;
243
244         for (; len; len--) {
245                 sum += *data++;
246         }
247         return (-sum);
248 }
249
250 /* XXX: Needs work */
251 static int
252 ipmi_ipmb_send_message(device_t dev, u_char channel, u_char netfn,
253     u_char command, u_char seq, u_char *data, int data_len)
254 {
255         struct ipmi_softc *sc = device_get_softc(dev);
256         struct ipmi_request *req;
257         u_char slave_addr = 0x52;
258         int error;
259
260         IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
261             IPMI_SEND_MSG, data_len + 8, 0);
262         req->ir_request[0] = channel;
263         req->ir_request[1] = slave_addr;
264         req->ir_request[2] = IPMI_ADDR(netfn, 0);
265         req->ir_request[3] = ipmi_ipmb_checksum(&req->ir_request[1], 2);
266         req->ir_request[4] = sc->ipmi_address;
267         req->ir_request[5] = IPMI_ADDR(seq, sc->ipmi_lun);
268         req->ir_request[6] = command;
269
270         bcopy(data, &req->ir_request[7], data_len);
271         temp[data_len + 7] = ipmi_ipmb_checksum(&req->ir_request[4],
272             data_len + 3);
273
274         ipmi_submit_driver_request(sc, req);
275         error = req->ir_error;
276
277         return (error);
278 }
279
280 static int
281 ipmi_handle_attn(struct ipmi_softc *sc)
282 {
283         struct ipmi_request *req;
284         int error;
285
286         device_printf(sc->ipmi_dev, "BMC has a message\n");
287         IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
288             IPMI_GET_MSG_FLAGS, 0, 1);
289
290         ipmi_submit_driver_request(sc, req);
291
292         if (req->ir_error == 0 && req->ir_compcode == 0) {
293                 if (req->ir_reply[0] & IPMI_MSG_BUFFER_FULL) {
294                         device_printf(sc->ipmi_dev, "message buffer full");
295                 }
296                 if (req->ir_reply[0] & IPMI_WDT_PRE_TIMEOUT) {
297                         device_printf(sc->ipmi_dev,
298                             "watchdog about to go off");
299                 }
300                 if (req->ir_reply[0] & IPMI_MSG_AVAILABLE) {
301                         IPMI_ALLOC_DRIVER_REQUEST(req,
302                             IPMI_ADDR(IPMI_APP_REQUEST, 0), IPMI_GET_MSG, 0,
303                             16);
304
305                         device_printf(sc->ipmi_dev, "throw out message ");
306                         dump_buf(temp, 16);
307                 }
308         }
309         error = req->ir_error;
310
311         return (error);
312 }
313 #endif
314
315 #ifdef IPMICTL_SEND_COMMAND_32
316 #define PTRIN(p)        ((void *)(uintptr_t)(p))
317 #define PTROUT(p)       ((uintptr_t)(p))
318 #endif
319
320 static int
321 ipmi_ioctl(struct cdev *cdev, u_long cmd, caddr_t data,
322     int flags, struct thread *td)
323 {
324         struct ipmi_softc *sc;
325         struct ipmi_device *dev;
326         struct ipmi_request *kreq;
327         struct ipmi_req *req = (struct ipmi_req *)data;
328         struct ipmi_recv *recv = (struct ipmi_recv *)data;
329         struct ipmi_addr addr;
330 #ifdef IPMICTL_SEND_COMMAND_32
331         struct ipmi_req32 *req32 = (struct ipmi_req32 *)data;
332         struct ipmi_recv32 *recv32 = (struct ipmi_recv32 *)data;
333         union {
334                 struct ipmi_req req;
335                 struct ipmi_recv recv;
336         } thunk32;
337 #endif
338         int error, len;
339
340         error = devfs_get_cdevpriv((void **)&dev);
341         if (error)
342                 return (error);
343
344         sc = cdev->si_drv1;
345
346 #ifdef IPMICTL_SEND_COMMAND_32
347         /* Convert 32-bit structures to native. */
348         switch (cmd) {
349         case IPMICTL_SEND_COMMAND_32:
350                 req = &thunk32.req;
351                 req->addr = PTRIN(req32->addr);
352                 req->addr_len = req32->addr_len;
353                 req->msgid = req32->msgid;
354                 req->msg.netfn = req32->msg.netfn;
355                 req->msg.cmd = req32->msg.cmd;
356                 req->msg.data_len = req32->msg.data_len;
357                 req->msg.data = PTRIN(req32->msg.data);
358                 break;
359         case IPMICTL_RECEIVE_MSG_TRUNC_32:
360         case IPMICTL_RECEIVE_MSG_32:
361                 recv = &thunk32.recv;
362                 recv->addr = PTRIN(recv32->addr);
363                 recv->addr_len = recv32->addr_len;
364                 recv->msg.data_len = recv32->msg.data_len;
365                 recv->msg.data = PTRIN(recv32->msg.data);
366                 break;
367         }
368 #endif
369
370         switch (cmd) {
371 #ifdef IPMICTL_SEND_COMMAND_32
372         case IPMICTL_SEND_COMMAND_32:
373 #endif
374         case IPMICTL_SEND_COMMAND:
375                 /*
376                  * XXX: Need to add proper handling of this.
377                  */
378                 error = copyin(req->addr, &addr, sizeof(addr));
379                 if (error)
380                         return (error);
381
382                 IPMI_LOCK(sc);
383                 /* clear out old stuff in queue of stuff done */
384                 /* XXX: This seems odd. */
385                 while ((kreq = TAILQ_FIRST(&dev->ipmi_completed_requests))) {
386                         TAILQ_REMOVE(&dev->ipmi_completed_requests, kreq,
387                             ir_link);
388                         dev->ipmi_requests--;
389                         ipmi_free_request(kreq);
390                 }
391                 IPMI_UNLOCK(sc);
392
393                 kreq = ipmi_alloc_request(dev, req->msgid,
394                     IPMI_ADDR(req->msg.netfn, 0), req->msg.cmd,
395                     req->msg.data_len, IPMI_MAX_RX);
396                 error = copyin(req->msg.data, kreq->ir_request,
397                     req->msg.data_len);
398                 if (error) {
399                         ipmi_free_request(kreq);
400                         return (error);
401                 }
402                 IPMI_LOCK(sc);
403                 dev->ipmi_requests++;
404                 error = sc->ipmi_enqueue_request(sc, kreq);
405                 IPMI_UNLOCK(sc);
406                 if (error)
407                         return (error);
408                 break;
409 #ifdef IPMICTL_SEND_COMMAND_32
410         case IPMICTL_RECEIVE_MSG_TRUNC_32:
411         case IPMICTL_RECEIVE_MSG_32:
412 #endif
413         case IPMICTL_RECEIVE_MSG_TRUNC:
414         case IPMICTL_RECEIVE_MSG:
415                 error = copyin(recv->addr, &addr, sizeof(addr));
416                 if (error)
417                         return (error);
418
419                 IPMI_LOCK(sc);
420                 kreq = TAILQ_FIRST(&dev->ipmi_completed_requests);
421                 if (kreq == NULL) {
422                         IPMI_UNLOCK(sc);
423                         return (EAGAIN);
424                 }
425                 addr.channel = IPMI_BMC_CHANNEL;
426                 /* XXX */
427                 recv->recv_type = IPMI_RESPONSE_RECV_TYPE;
428                 recv->msgid = kreq->ir_msgid;
429                 recv->msg.netfn = IPMI_REPLY_ADDR(kreq->ir_addr) >> 2;
430                 recv->msg.cmd = kreq->ir_command;
431                 error = kreq->ir_error;
432                 if (error) {
433                         TAILQ_REMOVE(&dev->ipmi_completed_requests, kreq,
434                             ir_link);
435                         dev->ipmi_requests--;
436                         IPMI_UNLOCK(sc);
437                         ipmi_free_request(kreq);
438                         return (error);
439                 }
440                 len = kreq->ir_replylen + 1;
441                 if (recv->msg.data_len < len &&
442                     (cmd == IPMICTL_RECEIVE_MSG
443 #ifdef IPMICTL_RECEIVE_MSG_32
444                      || cmd == IPMICTL_RECEIVE_MSG_32
445 #endif
446                     )) {
447                         IPMI_UNLOCK(sc);
448                         return (EMSGSIZE);
449                 }
450                 TAILQ_REMOVE(&dev->ipmi_completed_requests, kreq, ir_link);
451                 dev->ipmi_requests--;
452                 IPMI_UNLOCK(sc);
453                 len = min(recv->msg.data_len, len);
454                 recv->msg.data_len = len;
455                 error = copyout(&addr, recv->addr,sizeof(addr));
456                 if (error == 0)
457                         error = copyout(&kreq->ir_compcode, recv->msg.data, 1);
458                 if (error == 0)
459                         error = copyout(kreq->ir_reply, recv->msg.data + 1,
460                             len - 1);
461                 ipmi_free_request(kreq);
462                 if (error)
463                         return (error);
464                 break;
465         case IPMICTL_SET_MY_ADDRESS_CMD:
466                 IPMI_LOCK(sc);
467                 dev->ipmi_address = *(int*)data;
468                 IPMI_UNLOCK(sc);
469                 break;
470         case IPMICTL_GET_MY_ADDRESS_CMD:
471                 IPMI_LOCK(sc);
472                 *(int*)data = dev->ipmi_address;
473                 IPMI_UNLOCK(sc);
474                 break;
475         case IPMICTL_SET_MY_LUN_CMD:
476                 IPMI_LOCK(sc);
477                 dev->ipmi_lun = *(int*)data & 0x3;
478                 IPMI_UNLOCK(sc);
479                 break;
480         case IPMICTL_GET_MY_LUN_CMD:
481                 IPMI_LOCK(sc);
482                 *(int*)data = dev->ipmi_lun;
483                 IPMI_UNLOCK(sc);
484                 break;
485         case IPMICTL_SET_GETS_EVENTS_CMD:
486                 /*
487                 device_printf(sc->ipmi_dev,
488                     "IPMICTL_SET_GETS_EVENTS_CMD NA\n");
489                 */
490                 break;
491         case IPMICTL_REGISTER_FOR_CMD:
492         case IPMICTL_UNREGISTER_FOR_CMD:
493                 return (EOPNOTSUPP);
494         default:
495                 device_printf(sc->ipmi_dev, "Unknown IOCTL %lX\n", cmd);
496                 return (ENOIOCTL);
497         }
498
499 #ifdef IPMICTL_SEND_COMMAND_32
500         /* Update changed fields in 32-bit structures. */
501         switch (cmd) {
502         case IPMICTL_RECEIVE_MSG_TRUNC_32:
503         case IPMICTL_RECEIVE_MSG_32:
504                 recv32->recv_type = recv->recv_type;
505                 recv32->msgid = recv->msgid;
506                 recv32->msg.netfn = recv->msg.netfn;
507                 recv32->msg.cmd = recv->msg.cmd;
508                 recv32->msg.data_len = recv->msg.data_len;
509                 break;
510         }
511 #endif
512         return (0);
513 }
514
515 /*
516  * Request management.
517  */
518
519 static __inline void
520 ipmi_init_request(struct ipmi_request *req, struct ipmi_device *dev, long msgid,
521     uint8_t addr, uint8_t command, size_t requestlen, size_t replylen)
522 {
523
524         req->ir_owner = dev;
525         req->ir_msgid = msgid;
526         req->ir_addr = addr;
527         req->ir_command = command;
528         if (requestlen) {
529                 req->ir_request = (char *)&req[1];
530                 req->ir_requestlen = requestlen;
531         }
532         if (replylen) {
533                 req->ir_reply = (char *)&req[1] + requestlen;
534                 req->ir_replybuflen = replylen;
535         }
536 }
537
538 /* Allocate a new request with request and reply buffers. */
539 struct ipmi_request *
540 ipmi_alloc_request(struct ipmi_device *dev, long msgid, uint8_t addr,
541     uint8_t command, size_t requestlen, size_t replylen)
542 {
543         struct ipmi_request *req;
544
545         req = malloc(sizeof(struct ipmi_request) + requestlen + replylen,
546             M_IPMI, M_WAITOK | M_ZERO);
547         ipmi_init_request(req, dev, msgid, addr, command, requestlen, replylen);
548         return (req);
549 }
550
551 /* Free a request no longer in use. */
552 void
553 ipmi_free_request(struct ipmi_request *req)
554 {
555
556         free(req, M_IPMI);
557 }
558
559 /* Store a processed request on the appropriate completion queue. */
560 void
561 ipmi_complete_request(struct ipmi_softc *sc, struct ipmi_request *req)
562 {
563         struct ipmi_device *dev;
564
565         IPMI_LOCK_ASSERT(sc);
566
567         /*
568          * Anonymous requests (from inside the driver) always have a
569          * waiter that we awaken.
570          */
571         if (req->ir_owner == NULL)
572                 wakeup(req);
573         else {
574                 dev = req->ir_owner;
575                 TAILQ_INSERT_TAIL(&dev->ipmi_completed_requests, req, ir_link);
576                 selwakeup(&dev->ipmi_select);
577                 if (dev->ipmi_closing)
578                         wakeup(&dev->ipmi_requests);
579         }
580 }
581
582 /* Perform an internal driver request. */
583 int
584 ipmi_submit_driver_request(struct ipmi_softc *sc, struct ipmi_request *req,
585     int timo)
586 {
587
588         return (sc->ipmi_driver_request(sc, req, timo));
589 }
590
591 /*
592  * Helper routine for polled system interfaces that use
593  * ipmi_polled_enqueue_request() to queue requests.  This request
594  * waits until there is a pending request and then returns the first
595  * request.  If the driver is shutting down, it returns NULL.
596  */
597 struct ipmi_request *
598 ipmi_dequeue_request(struct ipmi_softc *sc)
599 {
600         struct ipmi_request *req;
601
602         IPMI_LOCK_ASSERT(sc);
603
604         while (!sc->ipmi_detaching && TAILQ_EMPTY(&sc->ipmi_pending_requests))
605                 cv_wait(&sc->ipmi_request_added, &sc->ipmi_requests_lock);
606         if (sc->ipmi_detaching)
607                 return (NULL);
608
609         req = TAILQ_FIRST(&sc->ipmi_pending_requests);
610         TAILQ_REMOVE(&sc->ipmi_pending_requests, req, ir_link);
611         return (req);
612 }
613
614 /* Default implementation of ipmi_enqueue_request() for polled interfaces. */
615 int
616 ipmi_polled_enqueue_request(struct ipmi_softc *sc, struct ipmi_request *req)
617 {
618
619         IPMI_LOCK_ASSERT(sc);
620
621         TAILQ_INSERT_TAIL(&sc->ipmi_pending_requests, req, ir_link);
622         cv_signal(&sc->ipmi_request_added);
623         return (0);
624 }
625
626 /*
627  * Watchdog event handler.
628  */
629
630 static int
631 ipmi_reset_watchdog(struct ipmi_softc *sc)
632 {
633         struct ipmi_request *req;
634         int error;
635
636         IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
637             IPMI_RESET_WDOG, 0, 0);
638         error = ipmi_submit_driver_request(sc, req, 0);
639         if (error)
640                 device_printf(sc->ipmi_dev, "Failed to reset watchdog\n");
641         return (error);
642 }
643
644 static int
645 ipmi_set_watchdog(struct ipmi_softc *sc, unsigned int sec)
646 {
647         struct ipmi_request *req;
648         int error;
649
650         if (sec > 0xffff / 10)
651                 return (EINVAL);
652
653         IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
654             IPMI_SET_WDOG, 6, 0);
655         if (sec) {
656                 req->ir_request[0] = IPMI_SET_WD_TIMER_DONT_STOP
657                     | IPMI_SET_WD_TIMER_SMS_OS;
658                 req->ir_request[1] = (wd_timer_actions & 0xff);
659                 req->ir_request[2] = (wd_pretimeout_countdown & 0xff);
660                 req->ir_request[3] = 0; /* Timer use */
661                 req->ir_request[4] = (sec * 10) & 0xff;
662                 req->ir_request[5] = (sec * 10) >> 8;
663         } else {
664                 req->ir_request[0] = IPMI_SET_WD_TIMER_SMS_OS;
665                 req->ir_request[1] = 0;
666                 req->ir_request[2] = 0;
667                 req->ir_request[3] = 0; /* Timer use */
668                 req->ir_request[4] = 0;
669                 req->ir_request[5] = 0;
670         }
671         error = ipmi_submit_driver_request(sc, req, 0);
672         if (error)
673                 device_printf(sc->ipmi_dev, "Failed to set watchdog\n");
674         return (error);
675 }
676
677 static void
678 ipmi_wd_event(void *arg, unsigned int cmd, int *error)
679 {
680         struct ipmi_softc *sc = arg;
681         unsigned int timeout;
682         int e;
683
684         /* Ignore requests while disabled. */
685         if (!on)
686                 return;
687
688         /*
689          * To prevent infinite hangs, we don't let anyone pat or change
690          * the watchdog when we're shutting down. (See ipmi_shutdown_event().)
691          * However, we do want to keep patting the watchdog while we are doing
692          * a coredump.
693          */
694         if (wd_in_shutdown) {
695                 if (dumping && sc->ipmi_watchdog_active)
696                         ipmi_reset_watchdog(sc);
697                 return;
698         }
699
700         cmd &= WD_INTERVAL;
701         if (cmd > 0 && cmd <= 63) {
702                 timeout = ((uint64_t)1 << cmd) / 1000000000;
703                 if (timeout == 0)
704                         timeout = 1;
705                 if (timeout != sc->ipmi_watchdog_active ||
706                     wd_timer_actions != sc->ipmi_watchdog_actions ||
707                     wd_pretimeout_countdown != sc->ipmi_watchdog_pretimeout) {
708                         e = ipmi_set_watchdog(sc, timeout);
709                         if (e == 0) {
710                                 sc->ipmi_watchdog_active = timeout;
711                                 sc->ipmi_watchdog_actions = wd_timer_actions;
712                                 sc->ipmi_watchdog_pretimeout = wd_pretimeout_countdown;
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                 if (sc->ipmi_watchdog_active != 0) {
721                         e = ipmi_reset_watchdog(sc);
722                         if (e == 0) {
723                                 *error = 0;
724                         } else {
725                                 (void)ipmi_set_watchdog(sc, 0);
726                                 sc->ipmi_watchdog_active = 0;
727                                 sc->ipmi_watchdog_actions = 0;
728                                 sc->ipmi_watchdog_pretimeout = 0;
729                         }
730                 }
731         } else if (atomic_readandclear_int(&sc->ipmi_watchdog_active) != 0) {
732                 sc->ipmi_watchdog_actions = 0;
733                 sc->ipmi_watchdog_pretimeout = 0;
734
735                 e = ipmi_set_watchdog(sc, 0);
736                 if (e != 0 && cmd == 0)
737                         *error = EOPNOTSUPP;
738         }
739 }
740
741 static void
742 ipmi_shutdown_event(void *arg, unsigned int cmd, int *error)
743 {
744         struct ipmi_softc *sc = arg;
745
746         /* Ignore event if disabled. */
747         if (!on)
748                 return;
749
750         /*
751          * Positive wd_shutdown_countdown value will re-arm watchdog;
752          * Zero value in wd_shutdown_countdown will disable watchdog;
753          * Negative value in wd_shutdown_countdown will keep existing state;
754          *
755          * Revert to using a power cycle to ensure that the watchdog will
756          * do something useful here.  Having the watchdog send an NMI
757          * instead is useless during shutdown, and might be ignored if an
758          * NMI already triggered.
759          */
760
761         wd_in_shutdown = true;
762         if (wd_shutdown_countdown == 0) {
763                 /* disable watchdog */
764                 ipmi_set_watchdog(sc, 0);
765                 sc->ipmi_watchdog_active = 0;
766         } else if (wd_shutdown_countdown > 0) {
767                 /* set desired action and time, and, reset watchdog */
768                 wd_timer_actions = IPMI_SET_WD_ACTION_POWER_CYCLE;
769                 ipmi_set_watchdog(sc, wd_shutdown_countdown);
770                 sc->ipmi_watchdog_active = wd_shutdown_countdown;
771                 ipmi_reset_watchdog(sc);
772         }
773 }
774
775 static void
776 ipmi_power_cycle(void *arg, int howto)
777 {
778         struct ipmi_softc *sc = arg;
779         struct ipmi_request *req;
780
781         /*
782          * Ignore everything except power cycling requests
783          */
784         if ((howto & RB_POWERCYCLE) == 0)
785                 return;
786
787         device_printf(sc->ipmi_dev, "Power cycling using IPMI\n");
788
789         /*
790          * Send a CHASSIS_CONTROL command to the CHASSIS device, subcommand 2
791          * as described in IPMI v2.0 spec section 28.3.
792          */
793         IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_CHASSIS_REQUEST, 0),
794             IPMI_CHASSIS_CONTROL, 1, 0);
795         req->ir_request[0] = IPMI_CC_POWER_CYCLE;
796
797         ipmi_submit_driver_request(sc, req, MAX_TIMEOUT);
798
799         if (req->ir_error != 0 || req->ir_compcode != 0) {
800                 device_printf(sc->ipmi_dev, "Power cycling via IPMI failed code %#x %#x\n",
801                     req->ir_error, req->ir_compcode);
802                 return;
803         }
804
805         /*
806          * BMCs are notoriously slow, give it cyle_wait seconds for the power
807          * down leg of the power cycle. If that fails, fallback to the next
808          * hanlder in the shutdown_final chain and/or the platform failsafe.
809          */
810         DELAY(cycle_wait * 1000 * 1000);
811         device_printf(sc->ipmi_dev, "Power cycling via IPMI timed out\n");
812 }
813
814 static void
815 ipmi_startup(void *arg)
816 {
817         struct ipmi_softc *sc = arg;
818         struct ipmi_request *req;
819         device_t dev;
820         int error, i;
821
822         config_intrhook_disestablish(&sc->ipmi_ich);
823         dev = sc->ipmi_dev;
824
825         /* Initialize interface-independent state. */
826         mtx_init(&sc->ipmi_requests_lock, "ipmi requests", NULL, MTX_DEF);
827         mtx_init(&sc->ipmi_io_lock, "ipmi io", NULL, MTX_DEF);
828         cv_init(&sc->ipmi_request_added, "ipmireq");
829         TAILQ_INIT(&sc->ipmi_pending_requests);
830
831         /* Initialize interface-dependent state. */
832         error = sc->ipmi_startup(sc);
833         if (error) {
834                 device_printf(dev, "Failed to initialize interface: %d\n",
835                     error);
836                 return;
837         }
838
839         /* Send a GET_DEVICE_ID request. */
840         IPMI_ALLOC_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
841             IPMI_GET_DEVICE_ID, 0, 15);
842
843         error = ipmi_submit_driver_request(sc, req, MAX_TIMEOUT);
844         if (error == EWOULDBLOCK) {
845                 device_printf(dev, "Timed out waiting for GET_DEVICE_ID\n");
846                 return;
847         } else if (error) {
848                 device_printf(dev, "Failed GET_DEVICE_ID: %d\n", error);
849                 return;
850         } else if (req->ir_compcode != 0) {
851                 device_printf(dev,
852                     "Bad completion code for GET_DEVICE_ID: %d\n",
853                     req->ir_compcode);
854                 return;
855         } else if (req->ir_replylen < 5) {
856                 device_printf(dev, "Short reply for GET_DEVICE_ID: %d\n",
857                     req->ir_replylen);
858                 return;
859         }
860
861         device_printf(dev, "IPMI device rev. %d, firmware rev. %d.%d%d, "
862             "version %d.%d, device support mask %#x\n",
863             req->ir_reply[1] & 0x0f,
864             req->ir_reply[2] & 0x7f, req->ir_reply[3] >> 4, req->ir_reply[3] & 0x0f,
865             req->ir_reply[4] & 0x0f, req->ir_reply[4] >> 4, req->ir_reply[5]);
866
867         sc->ipmi_dev_support = req->ir_reply[5];
868
869         IPMI_INIT_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
870             IPMI_CLEAR_FLAGS, 1, 0);
871
872         ipmi_submit_driver_request(sc, req, 0);
873
874         /* XXX: Magic numbers */
875         if (req->ir_compcode == 0xc0) {
876                 device_printf(dev, "Clear flags is busy\n");
877         }
878         if (req->ir_compcode == 0xc1) {
879                 device_printf(dev, "Clear flags illegal\n");
880         }
881
882         for (i = 0; i < 8; i++) {
883                 IPMI_INIT_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
884                     IPMI_GET_CHANNEL_INFO, 1, 0);
885                 req->ir_request[0] = i;
886
887                 ipmi_submit_driver_request(sc, req, 0);
888
889                 if (req->ir_compcode != 0)
890                         break;
891         }
892         device_printf(dev, "Number of channels %d\n", i);
893
894         /*
895          * Probe for watchdog, but only for backends which support
896          * polled driver requests.
897          */
898         if (sc->ipmi_driver_requests_polled) {
899                 IPMI_INIT_DRIVER_REQUEST(req, IPMI_ADDR(IPMI_APP_REQUEST, 0),
900                     IPMI_GET_WDOG, 0, 0);
901
902                 ipmi_submit_driver_request(sc, req, 0);
903
904                 if (req->ir_compcode == 0x00) {
905                         device_printf(dev, "Attached watchdog\n");
906                         /* register the watchdog event handler */
907                         sc->ipmi_watchdog_tag = EVENTHANDLER_REGISTER(
908                                 watchdog_list, ipmi_wd_event, sc, 0);
909                         sc->ipmi_shutdown_tag = EVENTHANDLER_REGISTER(
910                                 shutdown_pre_sync, ipmi_shutdown_event,
911                                 sc, 0);
912                 }
913         }
914
915         sc->ipmi_cdev = make_dev(&ipmi_cdevsw, device_get_unit(dev),
916             UID_ROOT, GID_OPERATOR, 0660, "ipmi%d", device_get_unit(dev));
917         if (sc->ipmi_cdev == NULL) {
918                 device_printf(dev, "Failed to create cdev\n");
919                 return;
920         }
921         sc->ipmi_cdev->si_drv1 = sc;
922
923         /*
924          * Set initial watchdog state. If desired, set an initial
925          * watchdog on startup. Or, if the watchdog device is
926          * disabled, clear any existing watchdog.
927          */
928         if (on && wd_startup_countdown > 0) {
929                 wd_timer_actions = IPMI_SET_WD_ACTION_POWER_CYCLE;
930                 if (ipmi_set_watchdog(sc, wd_startup_countdown) == 0 &&
931                     ipmi_reset_watchdog(sc) == 0) {
932                         sc->ipmi_watchdog_active = wd_startup_countdown;
933                         sc->ipmi_watchdog_actions = wd_timer_actions;
934                         sc->ipmi_watchdog_pretimeout = wd_pretimeout_countdown;
935                 } else
936                         (void)ipmi_set_watchdog(sc, 0);
937                 ipmi_reset_watchdog(sc);
938         } else if (!on)
939                 (void)ipmi_set_watchdog(sc, 0);
940         /*
941          * Power cycle the system off using IPMI. We use last - 2 since we don't
942          * handle all the other kinds of reboots. We'll let others handle them.
943          * We only try to do this if the BMC supports the Chassis device.
944          */
945         if (sc->ipmi_dev_support & IPMI_ADS_CHASSIS) {
946                 device_printf(dev, "Establishing power cycle handler\n");
947                 sc->ipmi_power_cycle_tag = EVENTHANDLER_REGISTER(shutdown_final,
948                     ipmi_power_cycle, sc, SHUTDOWN_PRI_LAST - 2);
949         }
950 }
951
952 int
953 ipmi_attach(device_t dev)
954 {
955         struct ipmi_softc *sc = device_get_softc(dev);
956         int error;
957
958         if (sc->ipmi_irq_res != NULL && sc->ipmi_intr != NULL) {
959                 error = bus_setup_intr(dev, sc->ipmi_irq_res, INTR_TYPE_MISC,
960                     NULL, sc->ipmi_intr, sc, &sc->ipmi_irq);
961                 if (error) {
962                         device_printf(dev, "can't set up interrupt\n");
963                         return (error);
964                 }
965         }
966
967         bzero(&sc->ipmi_ich, sizeof(struct intr_config_hook));
968         sc->ipmi_ich.ich_func = ipmi_startup;
969         sc->ipmi_ich.ich_arg = sc;
970         if (config_intrhook_establish(&sc->ipmi_ich) != 0) {
971                 device_printf(dev, "can't establish configuration hook\n");
972                 return (ENOMEM);
973         }
974
975         ipmi_attached = 1;
976         return (0);
977 }
978
979 int
980 ipmi_detach(device_t dev)
981 {
982         struct ipmi_softc *sc;
983
984         sc = device_get_softc(dev);
985
986         /* Fail if there are any open handles. */
987         IPMI_LOCK(sc);
988         if (sc->ipmi_opened) {
989                 IPMI_UNLOCK(sc);
990                 return (EBUSY);
991         }
992         IPMI_UNLOCK(sc);
993         if (sc->ipmi_cdev)
994                 destroy_dev(sc->ipmi_cdev);
995
996         /* Detach from watchdog handling and turn off watchdog. */
997         if (sc->ipmi_shutdown_tag)
998                 EVENTHANDLER_DEREGISTER(shutdown_pre_sync,
999                 sc->ipmi_shutdown_tag);
1000         if (sc->ipmi_watchdog_tag) {
1001                 EVENTHANDLER_DEREGISTER(watchdog_list, sc->ipmi_watchdog_tag);
1002                 ipmi_set_watchdog(sc, 0);
1003         }
1004
1005         /* Detach from shutdown handling for power cycle reboot */
1006         if (sc->ipmi_power_cycle_tag)
1007                 EVENTHANDLER_DEREGISTER(shutdown_final, sc->ipmi_power_cycle_tag);
1008
1009         /* XXX: should use shutdown callout I think. */
1010         /* If the backend uses a kthread, shut it down. */
1011         IPMI_LOCK(sc);
1012         sc->ipmi_detaching = 1;
1013         if (sc->ipmi_kthread) {
1014                 cv_broadcast(&sc->ipmi_request_added);
1015                 msleep(sc->ipmi_kthread, &sc->ipmi_requests_lock, 0,
1016                     "ipmi_wait", 0);
1017         }
1018         IPMI_UNLOCK(sc);
1019         if (sc->ipmi_irq)
1020                 bus_teardown_intr(dev, sc->ipmi_irq_res, sc->ipmi_irq);
1021
1022         ipmi_release_resources(dev);
1023         mtx_destroy(&sc->ipmi_io_lock);
1024         mtx_destroy(&sc->ipmi_requests_lock);
1025         return (0);
1026 }
1027
1028 void
1029 ipmi_release_resources(device_t dev)
1030 {
1031         struct ipmi_softc *sc;
1032         int i;
1033
1034         sc = device_get_softc(dev);
1035         if (sc->ipmi_irq)
1036                 bus_teardown_intr(dev, sc->ipmi_irq_res, sc->ipmi_irq);
1037         if (sc->ipmi_irq_res)
1038                 bus_release_resource(dev, SYS_RES_IRQ, sc->ipmi_irq_rid,
1039                     sc->ipmi_irq_res);
1040         for (i = 0; i < MAX_RES; i++)
1041                 if (sc->ipmi_io_res[i])
1042                         bus_release_resource(dev, sc->ipmi_io_type,
1043                             sc->ipmi_io_rid + i, sc->ipmi_io_res[i]);
1044 }
1045
1046 devclass_t ipmi_devclass;
1047
1048 /* XXX: Why? */
1049 static void
1050 ipmi_unload(void *arg)
1051 {
1052         device_t *      devs;
1053         int             count;
1054         int             i;
1055
1056         if (ipmi_devclass == NULL)
1057                 return;
1058         if (devclass_get_devices(ipmi_devclass, &devs, &count) != 0)
1059                 return;
1060         for (i = 0; i < count; i++)
1061                 device_delete_child(device_get_parent(devs[i]), devs[i]);
1062         free(devs, M_TEMP);
1063 }
1064 SYSUNINIT(ipmi_unload, SI_SUB_DRIVERS, SI_ORDER_FIRST, ipmi_unload, NULL);
1065
1066 #ifdef IMPI_DEBUG
1067 static void
1068 dump_buf(u_char *data, int len)
1069 {
1070         char buf[20];
1071         char line[1024];
1072         char temp[30];
1073         int count = 0;
1074         int i=0;
1075
1076         printf("Address %p len %d\n", data, len);
1077         if (len > 256)
1078                 len = 256;
1079         line[0] = '\000';
1080         for (; len > 0; len--, data++) {
1081                 sprintf(temp, "%02x ", *data);
1082                 strcat(line, temp);
1083                 if (*data >= ' ' && *data <= '~')
1084                         buf[count] = *data;
1085                 else if (*data >= 'A' && *data <= 'Z')
1086                         buf[count] = *data;
1087                 else
1088                         buf[count] = '.';
1089                 if (++count == 16) {
1090                         buf[count] = '\000';
1091                         count = 0;
1092                         printf("  %3x  %s %s\n", i, line, buf);
1093                         i+=16;
1094                         line[0] = '\000';
1095                 }
1096         }
1097         buf[count] = '\000';
1098
1099         for (; count != 16; count++) {
1100                 strcat(line, "   ");
1101         }
1102         printf("  %3x  %s %s\n", i, line, buf);
1103 }
1104 #endif