xref: /openbmc/linux/drivers/char/ipmi/ipmi_si_intf.c (revision 9cdb81c7)
1 /*
2  * ipmi_si.c
3  *
4  * The interface to the IPMI driver for the system interfaces (KCS, SMIC,
5  * BT).
6  *
7  * Author: MontaVista Software, Inc.
8  *         Corey Minyard <minyard@mvista.com>
9  *         source@mvista.com
10  *
11  * Copyright 2002 MontaVista Software Inc.
12  * Copyright 2006 IBM Corp., Christian Krafft <krafft@de.ibm.com>
13  *
14  *  This program is free software; you can redistribute it and/or modify it
15  *  under the terms of the GNU General Public License as published by the
16  *  Free Software Foundation; either version 2 of the License, or (at your
17  *  option) any later version.
18  *
19  *
20  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
21  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
22  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23  *  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
24  *  INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
25  *  BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
26  *  OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
27  *  ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR
28  *  TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
29  *  USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  *
31  *  You should have received a copy of the GNU General Public License along
32  *  with this program; if not, write to the Free Software Foundation, Inc.,
33  *  675 Mass Ave, Cambridge, MA 02139, USA.
34  */
35 
36 /*
37  * This file holds the "policy" for the interface to the SMI state
38  * machine.  It does the configuration, handles timers and interrupts,
39  * and drives the real SMI state machine.
40  */
41 
42 #include <linux/module.h>
43 #include <linux/moduleparam.h>
44 #include <asm/system.h>
45 #include <linux/sched.h>
46 #include <linux/seq_file.h>
47 #include <linux/timer.h>
48 #include <linux/errno.h>
49 #include <linux/spinlock.h>
50 #include <linux/slab.h>
51 #include <linux/delay.h>
52 #include <linux/list.h>
53 #include <linux/pci.h>
54 #include <linux/ioport.h>
55 #include <linux/notifier.h>
56 #include <linux/mutex.h>
57 #include <linux/kthread.h>
58 #include <asm/irq.h>
59 #include <linux/interrupt.h>
60 #include <linux/rcupdate.h>
61 #include <linux/ipmi.h>
62 #include <linux/ipmi_smi.h>
63 #include <asm/io.h>
64 #include "ipmi_si_sm.h"
65 #include <linux/init.h>
66 #include <linux/dmi.h>
67 #include <linux/string.h>
68 #include <linux/ctype.h>
69 #include <linux/pnp.h>
70 #include <linux/of_device.h>
71 #include <linux/of_platform.h>
72 #include <linux/of_address.h>
73 #include <linux/of_irq.h>
74 
75 #define PFX "ipmi_si: "
76 
77 /* Measure times between events in the driver. */
78 #undef DEBUG_TIMING
79 
80 /* Call every 10 ms. */
81 #define SI_TIMEOUT_TIME_USEC	10000
82 #define SI_USEC_PER_JIFFY	(1000000/HZ)
83 #define SI_TIMEOUT_JIFFIES	(SI_TIMEOUT_TIME_USEC/SI_USEC_PER_JIFFY)
84 #define SI_SHORT_TIMEOUT_USEC  250 /* .25ms when the SM request a
85 				      short timeout */
86 
87 enum si_intf_state {
88 	SI_NORMAL,
89 	SI_GETTING_FLAGS,
90 	SI_GETTING_EVENTS,
91 	SI_CLEARING_FLAGS,
92 	SI_CLEARING_FLAGS_THEN_SET_IRQ,
93 	SI_GETTING_MESSAGES,
94 	SI_ENABLE_INTERRUPTS1,
95 	SI_ENABLE_INTERRUPTS2,
96 	SI_DISABLE_INTERRUPTS1,
97 	SI_DISABLE_INTERRUPTS2
98 	/* FIXME - add watchdog stuff. */
99 };
100 
101 /* Some BT-specific defines we need here. */
102 #define IPMI_BT_INTMASK_REG		2
103 #define IPMI_BT_INTMASK_CLEAR_IRQ_BIT	2
104 #define IPMI_BT_INTMASK_ENABLE_IRQ_BIT	1
105 
106 enum si_type {
107     SI_KCS, SI_SMIC, SI_BT
108 };
109 static char *si_to_str[] = { "kcs", "smic", "bt" };
110 
111 static char *ipmi_addr_src_to_str[] = { NULL, "hotmod", "hardcoded", "SPMI",
112 					"ACPI", "SMBIOS", "PCI",
113 					"device-tree", "default" };
114 
115 #define DEVICE_NAME "ipmi_si"
116 
117 static struct platform_driver ipmi_driver;
118 
119 /*
120  * Indexes into stats[] in smi_info below.
121  */
122 enum si_stat_indexes {
123 	/*
124 	 * Number of times the driver requested a timer while an operation
125 	 * was in progress.
126 	 */
127 	SI_STAT_short_timeouts = 0,
128 
129 	/*
130 	 * Number of times the driver requested a timer while nothing was in
131 	 * progress.
132 	 */
133 	SI_STAT_long_timeouts,
134 
135 	/* Number of times the interface was idle while being polled. */
136 	SI_STAT_idles,
137 
138 	/* Number of interrupts the driver handled. */
139 	SI_STAT_interrupts,
140 
141 	/* Number of time the driver got an ATTN from the hardware. */
142 	SI_STAT_attentions,
143 
144 	/* Number of times the driver requested flags from the hardware. */
145 	SI_STAT_flag_fetches,
146 
147 	/* Number of times the hardware didn't follow the state machine. */
148 	SI_STAT_hosed_count,
149 
150 	/* Number of completed messages. */
151 	SI_STAT_complete_transactions,
152 
153 	/* Number of IPMI events received from the hardware. */
154 	SI_STAT_events,
155 
156 	/* Number of watchdog pretimeouts. */
157 	SI_STAT_watchdog_pretimeouts,
158 
159 	/* Number of asyncronous messages received. */
160 	SI_STAT_incoming_messages,
161 
162 
163 	/* This *must* remain last, add new values above this. */
164 	SI_NUM_STATS
165 };
166 
167 struct smi_info {
168 	int                    intf_num;
169 	ipmi_smi_t             intf;
170 	struct si_sm_data      *si_sm;
171 	struct si_sm_handlers  *handlers;
172 	enum si_type           si_type;
173 	spinlock_t             si_lock;
174 	spinlock_t             msg_lock;
175 	struct list_head       xmit_msgs;
176 	struct list_head       hp_xmit_msgs;
177 	struct ipmi_smi_msg    *curr_msg;
178 	enum si_intf_state     si_state;
179 
180 	/*
181 	 * Used to handle the various types of I/O that can occur with
182 	 * IPMI
183 	 */
184 	struct si_sm_io io;
185 	int (*io_setup)(struct smi_info *info);
186 	void (*io_cleanup)(struct smi_info *info);
187 	int (*irq_setup)(struct smi_info *info);
188 	void (*irq_cleanup)(struct smi_info *info);
189 	unsigned int io_size;
190 	enum ipmi_addr_src addr_source; /* ACPI, PCI, SMBIOS, hardcode, etc. */
191 	void (*addr_source_cleanup)(struct smi_info *info);
192 	void *addr_source_data;
193 
194 	/*
195 	 * Per-OEM handler, called from handle_flags().  Returns 1
196 	 * when handle_flags() needs to be re-run or 0 indicating it
197 	 * set si_state itself.
198 	 */
199 	int (*oem_data_avail_handler)(struct smi_info *smi_info);
200 
201 	/*
202 	 * Flags from the last GET_MSG_FLAGS command, used when an ATTN
203 	 * is set to hold the flags until we are done handling everything
204 	 * from the flags.
205 	 */
206 #define RECEIVE_MSG_AVAIL	0x01
207 #define EVENT_MSG_BUFFER_FULL	0x02
208 #define WDT_PRE_TIMEOUT_INT	0x08
209 #define OEM0_DATA_AVAIL     0x20
210 #define OEM1_DATA_AVAIL     0x40
211 #define OEM2_DATA_AVAIL     0x80
212 #define OEM_DATA_AVAIL      (OEM0_DATA_AVAIL | \
213 			     OEM1_DATA_AVAIL | \
214 			     OEM2_DATA_AVAIL)
215 	unsigned char       msg_flags;
216 
217 	/* Does the BMC have an event buffer? */
218 	char		    has_event_buffer;
219 
220 	/*
221 	 * If set to true, this will request events the next time the
222 	 * state machine is idle.
223 	 */
224 	atomic_t            req_events;
225 
226 	/*
227 	 * If true, run the state machine to completion on every send
228 	 * call.  Generally used after a panic to make sure stuff goes
229 	 * out.
230 	 */
231 	int                 run_to_completion;
232 
233 	/* The I/O port of an SI interface. */
234 	int                 port;
235 
236 	/*
237 	 * The space between start addresses of the two ports.  For
238 	 * instance, if the first port is 0xca2 and the spacing is 4, then
239 	 * the second port is 0xca6.
240 	 */
241 	unsigned int        spacing;
242 
243 	/* zero if no irq; */
244 	int                 irq;
245 
246 	/* The timer for this si. */
247 	struct timer_list   si_timer;
248 
249 	/* The time (in jiffies) the last timeout occurred at. */
250 	unsigned long       last_timeout_jiffies;
251 
252 	/* Used to gracefully stop the timer without race conditions. */
253 	atomic_t            stop_operation;
254 
255 	/*
256 	 * The driver will disable interrupts when it gets into a
257 	 * situation where it cannot handle messages due to lack of
258 	 * memory.  Once that situation clears up, it will re-enable
259 	 * interrupts.
260 	 */
261 	int interrupt_disabled;
262 
263 	/* From the get device id response... */
264 	struct ipmi_device_id device_id;
265 
266 	/* Driver model stuff. */
267 	struct device *dev;
268 	struct platform_device *pdev;
269 
270 	/*
271 	 * True if we allocated the device, false if it came from
272 	 * someplace else (like PCI).
273 	 */
274 	int dev_registered;
275 
276 	/* Slave address, could be reported from DMI. */
277 	unsigned char slave_addr;
278 
279 	/* Counters and things for the proc filesystem. */
280 	atomic_t stats[SI_NUM_STATS];
281 
282 	struct task_struct *thread;
283 
284 	struct list_head link;
285 	union ipmi_smi_info_union addr_info;
286 };
287 
288 #define smi_inc_stat(smi, stat) \
289 	atomic_inc(&(smi)->stats[SI_STAT_ ## stat])
290 #define smi_get_stat(smi, stat) \
291 	((unsigned int) atomic_read(&(smi)->stats[SI_STAT_ ## stat]))
292 
293 #define SI_MAX_PARMS 4
294 
295 static int force_kipmid[SI_MAX_PARMS];
296 static int num_force_kipmid;
297 #ifdef CONFIG_PCI
298 static int pci_registered;
299 #endif
300 #ifdef CONFIG_ACPI
301 static int pnp_registered;
302 #endif
303 
304 static unsigned int kipmid_max_busy_us[SI_MAX_PARMS];
305 static int num_max_busy_us;
306 
307 static int unload_when_empty = 1;
308 
309 static int add_smi(struct smi_info *smi);
310 static int try_smi_init(struct smi_info *smi);
311 static void cleanup_one_si(struct smi_info *to_clean);
312 static void cleanup_ipmi_si(void);
313 
314 static ATOMIC_NOTIFIER_HEAD(xaction_notifier_list);
315 static int register_xaction_notifier(struct notifier_block *nb)
316 {
317 	return atomic_notifier_chain_register(&xaction_notifier_list, nb);
318 }
319 
320 static void deliver_recv_msg(struct smi_info *smi_info,
321 			     struct ipmi_smi_msg *msg)
322 {
323 	/* Deliver the message to the upper layer with the lock
324 	   released. */
325 
326 	if (smi_info->run_to_completion) {
327 		ipmi_smi_msg_received(smi_info->intf, msg);
328 	} else {
329 		spin_unlock(&(smi_info->si_lock));
330 		ipmi_smi_msg_received(smi_info->intf, msg);
331 		spin_lock(&(smi_info->si_lock));
332 	}
333 }
334 
335 static void return_hosed_msg(struct smi_info *smi_info, int cCode)
336 {
337 	struct ipmi_smi_msg *msg = smi_info->curr_msg;
338 
339 	if (cCode < 0 || cCode > IPMI_ERR_UNSPECIFIED)
340 		cCode = IPMI_ERR_UNSPECIFIED;
341 	/* else use it as is */
342 
343 	/* Make it a response */
344 	msg->rsp[0] = msg->data[0] | 4;
345 	msg->rsp[1] = msg->data[1];
346 	msg->rsp[2] = cCode;
347 	msg->rsp_size = 3;
348 
349 	smi_info->curr_msg = NULL;
350 	deliver_recv_msg(smi_info, msg);
351 }
352 
353 static enum si_sm_result start_next_msg(struct smi_info *smi_info)
354 {
355 	int              rv;
356 	struct list_head *entry = NULL;
357 #ifdef DEBUG_TIMING
358 	struct timeval t;
359 #endif
360 
361 	/*
362 	 * No need to save flags, we aleady have interrupts off and we
363 	 * already hold the SMI lock.
364 	 */
365 	if (!smi_info->run_to_completion)
366 		spin_lock(&(smi_info->msg_lock));
367 
368 	/* Pick the high priority queue first. */
369 	if (!list_empty(&(smi_info->hp_xmit_msgs))) {
370 		entry = smi_info->hp_xmit_msgs.next;
371 	} else if (!list_empty(&(smi_info->xmit_msgs))) {
372 		entry = smi_info->xmit_msgs.next;
373 	}
374 
375 	if (!entry) {
376 		smi_info->curr_msg = NULL;
377 		rv = SI_SM_IDLE;
378 	} else {
379 		int err;
380 
381 		list_del(entry);
382 		smi_info->curr_msg = list_entry(entry,
383 						struct ipmi_smi_msg,
384 						link);
385 #ifdef DEBUG_TIMING
386 		do_gettimeofday(&t);
387 		printk(KERN_DEBUG "**Start2: %d.%9.9d\n", t.tv_sec, t.tv_usec);
388 #endif
389 		err = atomic_notifier_call_chain(&xaction_notifier_list,
390 				0, smi_info);
391 		if (err & NOTIFY_STOP_MASK) {
392 			rv = SI_SM_CALL_WITHOUT_DELAY;
393 			goto out;
394 		}
395 		err = smi_info->handlers->start_transaction(
396 			smi_info->si_sm,
397 			smi_info->curr_msg->data,
398 			smi_info->curr_msg->data_size);
399 		if (err)
400 			return_hosed_msg(smi_info, err);
401 
402 		rv = SI_SM_CALL_WITHOUT_DELAY;
403 	}
404  out:
405 	if (!smi_info->run_to_completion)
406 		spin_unlock(&(smi_info->msg_lock));
407 
408 	return rv;
409 }
410 
411 static void start_enable_irq(struct smi_info *smi_info)
412 {
413 	unsigned char msg[2];
414 
415 	/*
416 	 * If we are enabling interrupts, we have to tell the
417 	 * BMC to use them.
418 	 */
419 	msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
420 	msg[1] = IPMI_GET_BMC_GLOBAL_ENABLES_CMD;
421 
422 	smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
423 	smi_info->si_state = SI_ENABLE_INTERRUPTS1;
424 }
425 
426 static void start_disable_irq(struct smi_info *smi_info)
427 {
428 	unsigned char msg[2];
429 
430 	msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
431 	msg[1] = IPMI_GET_BMC_GLOBAL_ENABLES_CMD;
432 
433 	smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
434 	smi_info->si_state = SI_DISABLE_INTERRUPTS1;
435 }
436 
437 static void start_clear_flags(struct smi_info *smi_info)
438 {
439 	unsigned char msg[3];
440 
441 	/* Make sure the watchdog pre-timeout flag is not set at startup. */
442 	msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
443 	msg[1] = IPMI_CLEAR_MSG_FLAGS_CMD;
444 	msg[2] = WDT_PRE_TIMEOUT_INT;
445 
446 	smi_info->handlers->start_transaction(smi_info->si_sm, msg, 3);
447 	smi_info->si_state = SI_CLEARING_FLAGS;
448 }
449 
450 /*
451  * When we have a situtaion where we run out of memory and cannot
452  * allocate messages, we just leave them in the BMC and run the system
453  * polled until we can allocate some memory.  Once we have some
454  * memory, we will re-enable the interrupt.
455  */
456 static inline void disable_si_irq(struct smi_info *smi_info)
457 {
458 	if ((smi_info->irq) && (!smi_info->interrupt_disabled)) {
459 		start_disable_irq(smi_info);
460 		smi_info->interrupt_disabled = 1;
461 		if (!atomic_read(&smi_info->stop_operation))
462 			mod_timer(&smi_info->si_timer,
463 				  jiffies + SI_TIMEOUT_JIFFIES);
464 	}
465 }
466 
467 static inline void enable_si_irq(struct smi_info *smi_info)
468 {
469 	if ((smi_info->irq) && (smi_info->interrupt_disabled)) {
470 		start_enable_irq(smi_info);
471 		smi_info->interrupt_disabled = 0;
472 	}
473 }
474 
475 static void handle_flags(struct smi_info *smi_info)
476 {
477  retry:
478 	if (smi_info->msg_flags & WDT_PRE_TIMEOUT_INT) {
479 		/* Watchdog pre-timeout */
480 		smi_inc_stat(smi_info, watchdog_pretimeouts);
481 
482 		start_clear_flags(smi_info);
483 		smi_info->msg_flags &= ~WDT_PRE_TIMEOUT_INT;
484 		spin_unlock(&(smi_info->si_lock));
485 		ipmi_smi_watchdog_pretimeout(smi_info->intf);
486 		spin_lock(&(smi_info->si_lock));
487 	} else if (smi_info->msg_flags & RECEIVE_MSG_AVAIL) {
488 		/* Messages available. */
489 		smi_info->curr_msg = ipmi_alloc_smi_msg();
490 		if (!smi_info->curr_msg) {
491 			disable_si_irq(smi_info);
492 			smi_info->si_state = SI_NORMAL;
493 			return;
494 		}
495 		enable_si_irq(smi_info);
496 
497 		smi_info->curr_msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
498 		smi_info->curr_msg->data[1] = IPMI_GET_MSG_CMD;
499 		smi_info->curr_msg->data_size = 2;
500 
501 		smi_info->handlers->start_transaction(
502 			smi_info->si_sm,
503 			smi_info->curr_msg->data,
504 			smi_info->curr_msg->data_size);
505 		smi_info->si_state = SI_GETTING_MESSAGES;
506 	} else if (smi_info->msg_flags & EVENT_MSG_BUFFER_FULL) {
507 		/* Events available. */
508 		smi_info->curr_msg = ipmi_alloc_smi_msg();
509 		if (!smi_info->curr_msg) {
510 			disable_si_irq(smi_info);
511 			smi_info->si_state = SI_NORMAL;
512 			return;
513 		}
514 		enable_si_irq(smi_info);
515 
516 		smi_info->curr_msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
517 		smi_info->curr_msg->data[1] = IPMI_READ_EVENT_MSG_BUFFER_CMD;
518 		smi_info->curr_msg->data_size = 2;
519 
520 		smi_info->handlers->start_transaction(
521 			smi_info->si_sm,
522 			smi_info->curr_msg->data,
523 			smi_info->curr_msg->data_size);
524 		smi_info->si_state = SI_GETTING_EVENTS;
525 	} else if (smi_info->msg_flags & OEM_DATA_AVAIL &&
526 		   smi_info->oem_data_avail_handler) {
527 		if (smi_info->oem_data_avail_handler(smi_info))
528 			goto retry;
529 	} else
530 		smi_info->si_state = SI_NORMAL;
531 }
532 
533 static void handle_transaction_done(struct smi_info *smi_info)
534 {
535 	struct ipmi_smi_msg *msg;
536 #ifdef DEBUG_TIMING
537 	struct timeval t;
538 
539 	do_gettimeofday(&t);
540 	printk(KERN_DEBUG "**Done: %d.%9.9d\n", t.tv_sec, t.tv_usec);
541 #endif
542 	switch (smi_info->si_state) {
543 	case SI_NORMAL:
544 		if (!smi_info->curr_msg)
545 			break;
546 
547 		smi_info->curr_msg->rsp_size
548 			= smi_info->handlers->get_result(
549 				smi_info->si_sm,
550 				smi_info->curr_msg->rsp,
551 				IPMI_MAX_MSG_LENGTH);
552 
553 		/*
554 		 * Do this here becase deliver_recv_msg() releases the
555 		 * lock, and a new message can be put in during the
556 		 * time the lock is released.
557 		 */
558 		msg = smi_info->curr_msg;
559 		smi_info->curr_msg = NULL;
560 		deliver_recv_msg(smi_info, msg);
561 		break;
562 
563 	case SI_GETTING_FLAGS:
564 	{
565 		unsigned char msg[4];
566 		unsigned int  len;
567 
568 		/* We got the flags from the SMI, now handle them. */
569 		len = smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
570 		if (msg[2] != 0) {
571 			/* Error fetching flags, just give up for now. */
572 			smi_info->si_state = SI_NORMAL;
573 		} else if (len < 4) {
574 			/*
575 			 * Hmm, no flags.  That's technically illegal, but
576 			 * don't use uninitialized data.
577 			 */
578 			smi_info->si_state = SI_NORMAL;
579 		} else {
580 			smi_info->msg_flags = msg[3];
581 			handle_flags(smi_info);
582 		}
583 		break;
584 	}
585 
586 	case SI_CLEARING_FLAGS:
587 	case SI_CLEARING_FLAGS_THEN_SET_IRQ:
588 	{
589 		unsigned char msg[3];
590 
591 		/* We cleared the flags. */
592 		smi_info->handlers->get_result(smi_info->si_sm, msg, 3);
593 		if (msg[2] != 0) {
594 			/* Error clearing flags */
595 			dev_warn(smi_info->dev,
596 				 "Error clearing flags: %2.2x\n", msg[2]);
597 		}
598 		if (smi_info->si_state == SI_CLEARING_FLAGS_THEN_SET_IRQ)
599 			start_enable_irq(smi_info);
600 		else
601 			smi_info->si_state = SI_NORMAL;
602 		break;
603 	}
604 
605 	case SI_GETTING_EVENTS:
606 	{
607 		smi_info->curr_msg->rsp_size
608 			= smi_info->handlers->get_result(
609 				smi_info->si_sm,
610 				smi_info->curr_msg->rsp,
611 				IPMI_MAX_MSG_LENGTH);
612 
613 		/*
614 		 * Do this here becase deliver_recv_msg() releases the
615 		 * lock, and a new message can be put in during the
616 		 * time the lock is released.
617 		 */
618 		msg = smi_info->curr_msg;
619 		smi_info->curr_msg = NULL;
620 		if (msg->rsp[2] != 0) {
621 			/* Error getting event, probably done. */
622 			msg->done(msg);
623 
624 			/* Take off the event flag. */
625 			smi_info->msg_flags &= ~EVENT_MSG_BUFFER_FULL;
626 			handle_flags(smi_info);
627 		} else {
628 			smi_inc_stat(smi_info, events);
629 
630 			/*
631 			 * Do this before we deliver the message
632 			 * because delivering the message releases the
633 			 * lock and something else can mess with the
634 			 * state.
635 			 */
636 			handle_flags(smi_info);
637 
638 			deliver_recv_msg(smi_info, msg);
639 		}
640 		break;
641 	}
642 
643 	case SI_GETTING_MESSAGES:
644 	{
645 		smi_info->curr_msg->rsp_size
646 			= smi_info->handlers->get_result(
647 				smi_info->si_sm,
648 				smi_info->curr_msg->rsp,
649 				IPMI_MAX_MSG_LENGTH);
650 
651 		/*
652 		 * Do this here becase deliver_recv_msg() releases the
653 		 * lock, and a new message can be put in during the
654 		 * time the lock is released.
655 		 */
656 		msg = smi_info->curr_msg;
657 		smi_info->curr_msg = NULL;
658 		if (msg->rsp[2] != 0) {
659 			/* Error getting event, probably done. */
660 			msg->done(msg);
661 
662 			/* Take off the msg flag. */
663 			smi_info->msg_flags &= ~RECEIVE_MSG_AVAIL;
664 			handle_flags(smi_info);
665 		} else {
666 			smi_inc_stat(smi_info, incoming_messages);
667 
668 			/*
669 			 * Do this before we deliver the message
670 			 * because delivering the message releases the
671 			 * lock and something else can mess with the
672 			 * state.
673 			 */
674 			handle_flags(smi_info);
675 
676 			deliver_recv_msg(smi_info, msg);
677 		}
678 		break;
679 	}
680 
681 	case SI_ENABLE_INTERRUPTS1:
682 	{
683 		unsigned char msg[4];
684 
685 		/* We got the flags from the SMI, now handle them. */
686 		smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
687 		if (msg[2] != 0) {
688 			dev_warn(smi_info->dev, "Could not enable interrupts"
689 				 ", failed get, using polled mode.\n");
690 			smi_info->si_state = SI_NORMAL;
691 		} else {
692 			msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
693 			msg[1] = IPMI_SET_BMC_GLOBAL_ENABLES_CMD;
694 			msg[2] = (msg[3] |
695 				  IPMI_BMC_RCV_MSG_INTR |
696 				  IPMI_BMC_EVT_MSG_INTR);
697 			smi_info->handlers->start_transaction(
698 				smi_info->si_sm, msg, 3);
699 			smi_info->si_state = SI_ENABLE_INTERRUPTS2;
700 		}
701 		break;
702 	}
703 
704 	case SI_ENABLE_INTERRUPTS2:
705 	{
706 		unsigned char msg[4];
707 
708 		/* We got the flags from the SMI, now handle them. */
709 		smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
710 		if (msg[2] != 0)
711 			dev_warn(smi_info->dev, "Could not enable interrupts"
712 				 ", failed set, using polled mode.\n");
713 		else
714 			smi_info->interrupt_disabled = 0;
715 		smi_info->si_state = SI_NORMAL;
716 		break;
717 	}
718 
719 	case SI_DISABLE_INTERRUPTS1:
720 	{
721 		unsigned char msg[4];
722 
723 		/* We got the flags from the SMI, now handle them. */
724 		smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
725 		if (msg[2] != 0) {
726 			dev_warn(smi_info->dev, "Could not disable interrupts"
727 				 ", failed get.\n");
728 			smi_info->si_state = SI_NORMAL;
729 		} else {
730 			msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
731 			msg[1] = IPMI_SET_BMC_GLOBAL_ENABLES_CMD;
732 			msg[2] = (msg[3] &
733 				  ~(IPMI_BMC_RCV_MSG_INTR |
734 				    IPMI_BMC_EVT_MSG_INTR));
735 			smi_info->handlers->start_transaction(
736 				smi_info->si_sm, msg, 3);
737 			smi_info->si_state = SI_DISABLE_INTERRUPTS2;
738 		}
739 		break;
740 	}
741 
742 	case SI_DISABLE_INTERRUPTS2:
743 	{
744 		unsigned char msg[4];
745 
746 		/* We got the flags from the SMI, now handle them. */
747 		smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
748 		if (msg[2] != 0) {
749 			dev_warn(smi_info->dev, "Could not disable interrupts"
750 				 ", failed set.\n");
751 		}
752 		smi_info->si_state = SI_NORMAL;
753 		break;
754 	}
755 	}
756 }
757 
758 /*
759  * Called on timeouts and events.  Timeouts should pass the elapsed
760  * time, interrupts should pass in zero.  Must be called with
761  * si_lock held and interrupts disabled.
762  */
763 static enum si_sm_result smi_event_handler(struct smi_info *smi_info,
764 					   int time)
765 {
766 	enum si_sm_result si_sm_result;
767 
768  restart:
769 	/*
770 	 * There used to be a loop here that waited a little while
771 	 * (around 25us) before giving up.  That turned out to be
772 	 * pointless, the minimum delays I was seeing were in the 300us
773 	 * range, which is far too long to wait in an interrupt.  So
774 	 * we just run until the state machine tells us something
775 	 * happened or it needs a delay.
776 	 */
777 	si_sm_result = smi_info->handlers->event(smi_info->si_sm, time);
778 	time = 0;
779 	while (si_sm_result == SI_SM_CALL_WITHOUT_DELAY)
780 		si_sm_result = smi_info->handlers->event(smi_info->si_sm, 0);
781 
782 	if (si_sm_result == SI_SM_TRANSACTION_COMPLETE) {
783 		smi_inc_stat(smi_info, complete_transactions);
784 
785 		handle_transaction_done(smi_info);
786 		si_sm_result = smi_info->handlers->event(smi_info->si_sm, 0);
787 	} else if (si_sm_result == SI_SM_HOSED) {
788 		smi_inc_stat(smi_info, hosed_count);
789 
790 		/*
791 		 * Do the before return_hosed_msg, because that
792 		 * releases the lock.
793 		 */
794 		smi_info->si_state = SI_NORMAL;
795 		if (smi_info->curr_msg != NULL) {
796 			/*
797 			 * If we were handling a user message, format
798 			 * a response to send to the upper layer to
799 			 * tell it about the error.
800 			 */
801 			return_hosed_msg(smi_info, IPMI_ERR_UNSPECIFIED);
802 		}
803 		si_sm_result = smi_info->handlers->event(smi_info->si_sm, 0);
804 	}
805 
806 	/*
807 	 * We prefer handling attn over new messages.  But don't do
808 	 * this if there is not yet an upper layer to handle anything.
809 	 */
810 	if (likely(smi_info->intf) && si_sm_result == SI_SM_ATTN) {
811 		unsigned char msg[2];
812 
813 		smi_inc_stat(smi_info, attentions);
814 
815 		/*
816 		 * Got a attn, send down a get message flags to see
817 		 * what's causing it.  It would be better to handle
818 		 * this in the upper layer, but due to the way
819 		 * interrupts work with the SMI, that's not really
820 		 * possible.
821 		 */
822 		msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
823 		msg[1] = IPMI_GET_MSG_FLAGS_CMD;
824 
825 		smi_info->handlers->start_transaction(
826 			smi_info->si_sm, msg, 2);
827 		smi_info->si_state = SI_GETTING_FLAGS;
828 		goto restart;
829 	}
830 
831 	/* If we are currently idle, try to start the next message. */
832 	if (si_sm_result == SI_SM_IDLE) {
833 		smi_inc_stat(smi_info, idles);
834 
835 		si_sm_result = start_next_msg(smi_info);
836 		if (si_sm_result != SI_SM_IDLE)
837 			goto restart;
838 	}
839 
840 	if ((si_sm_result == SI_SM_IDLE)
841 	    && (atomic_read(&smi_info->req_events))) {
842 		/*
843 		 * We are idle and the upper layer requested that I fetch
844 		 * events, so do so.
845 		 */
846 		atomic_set(&smi_info->req_events, 0);
847 
848 		smi_info->curr_msg = ipmi_alloc_smi_msg();
849 		if (!smi_info->curr_msg)
850 			goto out;
851 
852 		smi_info->curr_msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
853 		smi_info->curr_msg->data[1] = IPMI_READ_EVENT_MSG_BUFFER_CMD;
854 		smi_info->curr_msg->data_size = 2;
855 
856 		smi_info->handlers->start_transaction(
857 			smi_info->si_sm,
858 			smi_info->curr_msg->data,
859 			smi_info->curr_msg->data_size);
860 		smi_info->si_state = SI_GETTING_EVENTS;
861 		goto restart;
862 	}
863  out:
864 	return si_sm_result;
865 }
866 
867 static void sender(void                *send_info,
868 		   struct ipmi_smi_msg *msg,
869 		   int                 priority)
870 {
871 	struct smi_info   *smi_info = send_info;
872 	enum si_sm_result result;
873 	unsigned long     flags;
874 #ifdef DEBUG_TIMING
875 	struct timeval    t;
876 #endif
877 
878 	if (atomic_read(&smi_info->stop_operation)) {
879 		msg->rsp[0] = msg->data[0] | 4;
880 		msg->rsp[1] = msg->data[1];
881 		msg->rsp[2] = IPMI_ERR_UNSPECIFIED;
882 		msg->rsp_size = 3;
883 		deliver_recv_msg(smi_info, msg);
884 		return;
885 	}
886 
887 #ifdef DEBUG_TIMING
888 	do_gettimeofday(&t);
889 	printk("**Enqueue: %d.%9.9d\n", t.tv_sec, t.tv_usec);
890 #endif
891 
892 	/*
893 	 * last_timeout_jiffies is updated here to avoid
894 	 * smi_timeout() handler passing very large time_diff
895 	 * value to smi_event_handler() that causes
896 	 * the send command to abort.
897 	 */
898 	smi_info->last_timeout_jiffies = jiffies;
899 
900 	mod_timer(&smi_info->si_timer, jiffies + SI_TIMEOUT_JIFFIES);
901 
902 	if (smi_info->thread)
903 		wake_up_process(smi_info->thread);
904 
905 	if (smi_info->run_to_completion) {
906 		/*
907 		 * If we are running to completion, then throw it in
908 		 * the list and run transactions until everything is
909 		 * clear.  Priority doesn't matter here.
910 		 */
911 
912 		/*
913 		 * Run to completion means we are single-threaded, no
914 		 * need for locks.
915 		 */
916 		list_add_tail(&(msg->link), &(smi_info->xmit_msgs));
917 
918 		result = smi_event_handler(smi_info, 0);
919 		while (result != SI_SM_IDLE) {
920 			udelay(SI_SHORT_TIMEOUT_USEC);
921 			result = smi_event_handler(smi_info,
922 						   SI_SHORT_TIMEOUT_USEC);
923 		}
924 		return;
925 	}
926 
927 	spin_lock_irqsave(&smi_info->msg_lock, flags);
928 	if (priority > 0)
929 		list_add_tail(&msg->link, &smi_info->hp_xmit_msgs);
930 	else
931 		list_add_tail(&msg->link, &smi_info->xmit_msgs);
932 	spin_unlock_irqrestore(&smi_info->msg_lock, flags);
933 
934 	spin_lock_irqsave(&smi_info->si_lock, flags);
935 	if (smi_info->si_state == SI_NORMAL && smi_info->curr_msg == NULL)
936 		start_next_msg(smi_info);
937 	spin_unlock_irqrestore(&smi_info->si_lock, flags);
938 }
939 
940 static void set_run_to_completion(void *send_info, int i_run_to_completion)
941 {
942 	struct smi_info   *smi_info = send_info;
943 	enum si_sm_result result;
944 
945 	smi_info->run_to_completion = i_run_to_completion;
946 	if (i_run_to_completion) {
947 		result = smi_event_handler(smi_info, 0);
948 		while (result != SI_SM_IDLE) {
949 			udelay(SI_SHORT_TIMEOUT_USEC);
950 			result = smi_event_handler(smi_info,
951 						   SI_SHORT_TIMEOUT_USEC);
952 		}
953 	}
954 }
955 
956 /*
957  * Use -1 in the nsec value of the busy waiting timespec to tell that
958  * we are spinning in kipmid looking for something and not delaying
959  * between checks
960  */
961 static inline void ipmi_si_set_not_busy(struct timespec *ts)
962 {
963 	ts->tv_nsec = -1;
964 }
965 static inline int ipmi_si_is_busy(struct timespec *ts)
966 {
967 	return ts->tv_nsec != -1;
968 }
969 
970 static int ipmi_thread_busy_wait(enum si_sm_result smi_result,
971 				 const struct smi_info *smi_info,
972 				 struct timespec *busy_until)
973 {
974 	unsigned int max_busy_us = 0;
975 
976 	if (smi_info->intf_num < num_max_busy_us)
977 		max_busy_us = kipmid_max_busy_us[smi_info->intf_num];
978 	if (max_busy_us == 0 || smi_result != SI_SM_CALL_WITH_DELAY)
979 		ipmi_si_set_not_busy(busy_until);
980 	else if (!ipmi_si_is_busy(busy_until)) {
981 		getnstimeofday(busy_until);
982 		timespec_add_ns(busy_until, max_busy_us*NSEC_PER_USEC);
983 	} else {
984 		struct timespec now;
985 		getnstimeofday(&now);
986 		if (unlikely(timespec_compare(&now, busy_until) > 0)) {
987 			ipmi_si_set_not_busy(busy_until);
988 			return 0;
989 		}
990 	}
991 	return 1;
992 }
993 
994 
995 /*
996  * A busy-waiting loop for speeding up IPMI operation.
997  *
998  * Lousy hardware makes this hard.  This is only enabled for systems
999  * that are not BT and do not have interrupts.  It starts spinning
1000  * when an operation is complete or until max_busy tells it to stop
1001  * (if that is enabled).  See the paragraph on kimid_max_busy_us in
1002  * Documentation/IPMI.txt for details.
1003  */
1004 static int ipmi_thread(void *data)
1005 {
1006 	struct smi_info *smi_info = data;
1007 	unsigned long flags;
1008 	enum si_sm_result smi_result;
1009 	struct timespec busy_until;
1010 
1011 	ipmi_si_set_not_busy(&busy_until);
1012 	set_user_nice(current, 19);
1013 	while (!kthread_should_stop()) {
1014 		int busy_wait;
1015 
1016 		spin_lock_irqsave(&(smi_info->si_lock), flags);
1017 		smi_result = smi_event_handler(smi_info, 0);
1018 		spin_unlock_irqrestore(&(smi_info->si_lock), flags);
1019 		busy_wait = ipmi_thread_busy_wait(smi_result, smi_info,
1020 						  &busy_until);
1021 		if (smi_result == SI_SM_CALL_WITHOUT_DELAY)
1022 			; /* do nothing */
1023 		else if (smi_result == SI_SM_CALL_WITH_DELAY && busy_wait)
1024 			schedule();
1025 		else if (smi_result == SI_SM_IDLE)
1026 			schedule_timeout_interruptible(100);
1027 		else
1028 			schedule_timeout_interruptible(1);
1029 	}
1030 	return 0;
1031 }
1032 
1033 
1034 static void poll(void *send_info)
1035 {
1036 	struct smi_info *smi_info = send_info;
1037 	unsigned long flags;
1038 
1039 	/*
1040 	 * Make sure there is some delay in the poll loop so we can
1041 	 * drive time forward and timeout things.
1042 	 */
1043 	udelay(10);
1044 	spin_lock_irqsave(&smi_info->si_lock, flags);
1045 	smi_event_handler(smi_info, 10);
1046 	spin_unlock_irqrestore(&smi_info->si_lock, flags);
1047 }
1048 
1049 static void request_events(void *send_info)
1050 {
1051 	struct smi_info *smi_info = send_info;
1052 
1053 	if (atomic_read(&smi_info->stop_operation) ||
1054 				!smi_info->has_event_buffer)
1055 		return;
1056 
1057 	atomic_set(&smi_info->req_events, 1);
1058 }
1059 
1060 static int initialized;
1061 
1062 static void smi_timeout(unsigned long data)
1063 {
1064 	struct smi_info   *smi_info = (struct smi_info *) data;
1065 	enum si_sm_result smi_result;
1066 	unsigned long     flags;
1067 	unsigned long     jiffies_now;
1068 	long              time_diff;
1069 	long		  timeout;
1070 #ifdef DEBUG_TIMING
1071 	struct timeval    t;
1072 #endif
1073 
1074 	spin_lock_irqsave(&(smi_info->si_lock), flags);
1075 #ifdef DEBUG_TIMING
1076 	do_gettimeofday(&t);
1077 	printk(KERN_DEBUG "**Timer: %d.%9.9d\n", t.tv_sec, t.tv_usec);
1078 #endif
1079 	jiffies_now = jiffies;
1080 	time_diff = (((long)jiffies_now - (long)smi_info->last_timeout_jiffies)
1081 		     * SI_USEC_PER_JIFFY);
1082 	smi_result = smi_event_handler(smi_info, time_diff);
1083 
1084 	spin_unlock_irqrestore(&(smi_info->si_lock), flags);
1085 
1086 	smi_info->last_timeout_jiffies = jiffies_now;
1087 
1088 	if ((smi_info->irq) && (!smi_info->interrupt_disabled)) {
1089 		/* Running with interrupts, only do long timeouts. */
1090 		timeout = jiffies + SI_TIMEOUT_JIFFIES;
1091 		smi_inc_stat(smi_info, long_timeouts);
1092 		goto do_mod_timer;
1093 	}
1094 
1095 	/*
1096 	 * If the state machine asks for a short delay, then shorten
1097 	 * the timer timeout.
1098 	 */
1099 	if (smi_result == SI_SM_CALL_WITH_DELAY) {
1100 		smi_inc_stat(smi_info, short_timeouts);
1101 		timeout = jiffies + 1;
1102 	} else {
1103 		smi_inc_stat(smi_info, long_timeouts);
1104 		timeout = jiffies + SI_TIMEOUT_JIFFIES;
1105 	}
1106 
1107  do_mod_timer:
1108 	if (smi_result != SI_SM_IDLE)
1109 		mod_timer(&(smi_info->si_timer), timeout);
1110 }
1111 
1112 static irqreturn_t si_irq_handler(int irq, void *data)
1113 {
1114 	struct smi_info *smi_info = data;
1115 	unsigned long   flags;
1116 #ifdef DEBUG_TIMING
1117 	struct timeval  t;
1118 #endif
1119 
1120 	spin_lock_irqsave(&(smi_info->si_lock), flags);
1121 
1122 	smi_inc_stat(smi_info, interrupts);
1123 
1124 #ifdef DEBUG_TIMING
1125 	do_gettimeofday(&t);
1126 	printk(KERN_DEBUG "**Interrupt: %d.%9.9d\n", t.tv_sec, t.tv_usec);
1127 #endif
1128 	smi_event_handler(smi_info, 0);
1129 	spin_unlock_irqrestore(&(smi_info->si_lock), flags);
1130 	return IRQ_HANDLED;
1131 }
1132 
1133 static irqreturn_t si_bt_irq_handler(int irq, void *data)
1134 {
1135 	struct smi_info *smi_info = data;
1136 	/* We need to clear the IRQ flag for the BT interface. */
1137 	smi_info->io.outputb(&smi_info->io, IPMI_BT_INTMASK_REG,
1138 			     IPMI_BT_INTMASK_CLEAR_IRQ_BIT
1139 			     | IPMI_BT_INTMASK_ENABLE_IRQ_BIT);
1140 	return si_irq_handler(irq, data);
1141 }
1142 
1143 static int smi_start_processing(void       *send_info,
1144 				ipmi_smi_t intf)
1145 {
1146 	struct smi_info *new_smi = send_info;
1147 	int             enable = 0;
1148 
1149 	new_smi->intf = intf;
1150 
1151 	/* Try to claim any interrupts. */
1152 	if (new_smi->irq_setup)
1153 		new_smi->irq_setup(new_smi);
1154 
1155 	/* Set up the timer that drives the interface. */
1156 	setup_timer(&new_smi->si_timer, smi_timeout, (long)new_smi);
1157 	new_smi->last_timeout_jiffies = jiffies;
1158 	mod_timer(&new_smi->si_timer, jiffies + SI_TIMEOUT_JIFFIES);
1159 
1160 	/*
1161 	 * Check if the user forcefully enabled the daemon.
1162 	 */
1163 	if (new_smi->intf_num < num_force_kipmid)
1164 		enable = force_kipmid[new_smi->intf_num];
1165 	/*
1166 	 * The BT interface is efficient enough to not need a thread,
1167 	 * and there is no need for a thread if we have interrupts.
1168 	 */
1169 	else if ((new_smi->si_type != SI_BT) && (!new_smi->irq))
1170 		enable = 1;
1171 
1172 	if (enable) {
1173 		new_smi->thread = kthread_run(ipmi_thread, new_smi,
1174 					      "kipmi%d", new_smi->intf_num);
1175 		if (IS_ERR(new_smi->thread)) {
1176 			dev_notice(new_smi->dev, "Could not start"
1177 				   " kernel thread due to error %ld, only using"
1178 				   " timers to drive the interface\n",
1179 				   PTR_ERR(new_smi->thread));
1180 			new_smi->thread = NULL;
1181 		}
1182 	}
1183 
1184 	return 0;
1185 }
1186 
1187 static int get_smi_info(void *send_info, struct ipmi_smi_info *data)
1188 {
1189 	struct smi_info *smi = send_info;
1190 
1191 	data->addr_src = smi->addr_source;
1192 	data->dev = smi->dev;
1193 	data->addr_info = smi->addr_info;
1194 	get_device(smi->dev);
1195 
1196 	return 0;
1197 }
1198 
1199 static void set_maintenance_mode(void *send_info, int enable)
1200 {
1201 	struct smi_info   *smi_info = send_info;
1202 
1203 	if (!enable)
1204 		atomic_set(&smi_info->req_events, 0);
1205 }
1206 
1207 static struct ipmi_smi_handlers handlers = {
1208 	.owner                  = THIS_MODULE,
1209 	.start_processing       = smi_start_processing,
1210 	.get_smi_info		= get_smi_info,
1211 	.sender			= sender,
1212 	.request_events		= request_events,
1213 	.set_maintenance_mode   = set_maintenance_mode,
1214 	.set_run_to_completion  = set_run_to_completion,
1215 	.poll			= poll,
1216 };
1217 
1218 /*
1219  * There can be 4 IO ports passed in (with or without IRQs), 4 addresses,
1220  * a default IO port, and 1 ACPI/SPMI address.  That sets SI_MAX_DRIVERS.
1221  */
1222 
1223 static LIST_HEAD(smi_infos);
1224 static DEFINE_MUTEX(smi_infos_lock);
1225 static int smi_num; /* Used to sequence the SMIs */
1226 
1227 #define DEFAULT_REGSPACING	1
1228 #define DEFAULT_REGSIZE		1
1229 
1230 static bool          si_trydefaults = 1;
1231 static char          *si_type[SI_MAX_PARMS];
1232 #define MAX_SI_TYPE_STR 30
1233 static char          si_type_str[MAX_SI_TYPE_STR];
1234 static unsigned long addrs[SI_MAX_PARMS];
1235 static unsigned int num_addrs;
1236 static unsigned int  ports[SI_MAX_PARMS];
1237 static unsigned int num_ports;
1238 static int           irqs[SI_MAX_PARMS];
1239 static unsigned int num_irqs;
1240 static int           regspacings[SI_MAX_PARMS];
1241 static unsigned int num_regspacings;
1242 static int           regsizes[SI_MAX_PARMS];
1243 static unsigned int num_regsizes;
1244 static int           regshifts[SI_MAX_PARMS];
1245 static unsigned int num_regshifts;
1246 static int slave_addrs[SI_MAX_PARMS]; /* Leaving 0 chooses the default value */
1247 static unsigned int num_slave_addrs;
1248 
1249 #define IPMI_IO_ADDR_SPACE  0
1250 #define IPMI_MEM_ADDR_SPACE 1
1251 static char *addr_space_to_str[] = { "i/o", "mem" };
1252 
1253 static int hotmod_handler(const char *val, struct kernel_param *kp);
1254 
1255 module_param_call(hotmod, hotmod_handler, NULL, NULL, 0200);
1256 MODULE_PARM_DESC(hotmod, "Add and remove interfaces.  See"
1257 		 " Documentation/IPMI.txt in the kernel sources for the"
1258 		 " gory details.");
1259 
1260 module_param_named(trydefaults, si_trydefaults, bool, 0);
1261 MODULE_PARM_DESC(trydefaults, "Setting this to 'false' will disable the"
1262 		 " default scan of the KCS and SMIC interface at the standard"
1263 		 " address");
1264 module_param_string(type, si_type_str, MAX_SI_TYPE_STR, 0);
1265 MODULE_PARM_DESC(type, "Defines the type of each interface, each"
1266 		 " interface separated by commas.  The types are 'kcs',"
1267 		 " 'smic', and 'bt'.  For example si_type=kcs,bt will set"
1268 		 " the first interface to kcs and the second to bt");
1269 module_param_array(addrs, ulong, &num_addrs, 0);
1270 MODULE_PARM_DESC(addrs, "Sets the memory address of each interface, the"
1271 		 " addresses separated by commas.  Only use if an interface"
1272 		 " is in memory.  Otherwise, set it to zero or leave"
1273 		 " it blank.");
1274 module_param_array(ports, uint, &num_ports, 0);
1275 MODULE_PARM_DESC(ports, "Sets the port address of each interface, the"
1276 		 " addresses separated by commas.  Only use if an interface"
1277 		 " is a port.  Otherwise, set it to zero or leave"
1278 		 " it blank.");
1279 module_param_array(irqs, int, &num_irqs, 0);
1280 MODULE_PARM_DESC(irqs, "Sets the interrupt of each interface, the"
1281 		 " addresses separated by commas.  Only use if an interface"
1282 		 " has an interrupt.  Otherwise, set it to zero or leave"
1283 		 " it blank.");
1284 module_param_array(regspacings, int, &num_regspacings, 0);
1285 MODULE_PARM_DESC(regspacings, "The number of bytes between the start address"
1286 		 " and each successive register used by the interface.  For"
1287 		 " instance, if the start address is 0xca2 and the spacing"
1288 		 " is 2, then the second address is at 0xca4.  Defaults"
1289 		 " to 1.");
1290 module_param_array(regsizes, int, &num_regsizes, 0);
1291 MODULE_PARM_DESC(regsizes, "The size of the specific IPMI register in bytes."
1292 		 " This should generally be 1, 2, 4, or 8 for an 8-bit,"
1293 		 " 16-bit, 32-bit, or 64-bit register.  Use this if you"
1294 		 " the 8-bit IPMI register has to be read from a larger"
1295 		 " register.");
1296 module_param_array(regshifts, int, &num_regshifts, 0);
1297 MODULE_PARM_DESC(regshifts, "The amount to shift the data read from the."
1298 		 " IPMI register, in bits.  For instance, if the data"
1299 		 " is read from a 32-bit word and the IPMI data is in"
1300 		 " bit 8-15, then the shift would be 8");
1301 module_param_array(slave_addrs, int, &num_slave_addrs, 0);
1302 MODULE_PARM_DESC(slave_addrs, "Set the default IPMB slave address for"
1303 		 " the controller.  Normally this is 0x20, but can be"
1304 		 " overridden by this parm.  This is an array indexed"
1305 		 " by interface number.");
1306 module_param_array(force_kipmid, int, &num_force_kipmid, 0);
1307 MODULE_PARM_DESC(force_kipmid, "Force the kipmi daemon to be enabled (1) or"
1308 		 " disabled(0).  Normally the IPMI driver auto-detects"
1309 		 " this, but the value may be overridden by this parm.");
1310 module_param(unload_when_empty, int, 0);
1311 MODULE_PARM_DESC(unload_when_empty, "Unload the module if no interfaces are"
1312 		 " specified or found, default is 1.  Setting to 0"
1313 		 " is useful for hot add of devices using hotmod.");
1314 module_param_array(kipmid_max_busy_us, uint, &num_max_busy_us, 0644);
1315 MODULE_PARM_DESC(kipmid_max_busy_us,
1316 		 "Max time (in microseconds) to busy-wait for IPMI data before"
1317 		 " sleeping. 0 (default) means to wait forever. Set to 100-500"
1318 		 " if kipmid is using up a lot of CPU time.");
1319 
1320 
1321 static void std_irq_cleanup(struct smi_info *info)
1322 {
1323 	if (info->si_type == SI_BT)
1324 		/* Disable the interrupt in the BT interface. */
1325 		info->io.outputb(&info->io, IPMI_BT_INTMASK_REG, 0);
1326 	free_irq(info->irq, info);
1327 }
1328 
1329 static int std_irq_setup(struct smi_info *info)
1330 {
1331 	int rv;
1332 
1333 	if (!info->irq)
1334 		return 0;
1335 
1336 	if (info->si_type == SI_BT) {
1337 		rv = request_irq(info->irq,
1338 				 si_bt_irq_handler,
1339 				 IRQF_SHARED | IRQF_DISABLED,
1340 				 DEVICE_NAME,
1341 				 info);
1342 		if (!rv)
1343 			/* Enable the interrupt in the BT interface. */
1344 			info->io.outputb(&info->io, IPMI_BT_INTMASK_REG,
1345 					 IPMI_BT_INTMASK_ENABLE_IRQ_BIT);
1346 	} else
1347 		rv = request_irq(info->irq,
1348 				 si_irq_handler,
1349 				 IRQF_SHARED | IRQF_DISABLED,
1350 				 DEVICE_NAME,
1351 				 info);
1352 	if (rv) {
1353 		dev_warn(info->dev, "%s unable to claim interrupt %d,"
1354 			 " running polled\n",
1355 			 DEVICE_NAME, info->irq);
1356 		info->irq = 0;
1357 	} else {
1358 		info->irq_cleanup = std_irq_cleanup;
1359 		dev_info(info->dev, "Using irq %d\n", info->irq);
1360 	}
1361 
1362 	return rv;
1363 }
1364 
1365 static unsigned char port_inb(struct si_sm_io *io, unsigned int offset)
1366 {
1367 	unsigned int addr = io->addr_data;
1368 
1369 	return inb(addr + (offset * io->regspacing));
1370 }
1371 
1372 static void port_outb(struct si_sm_io *io, unsigned int offset,
1373 		      unsigned char b)
1374 {
1375 	unsigned int addr = io->addr_data;
1376 
1377 	outb(b, addr + (offset * io->regspacing));
1378 }
1379 
1380 static unsigned char port_inw(struct si_sm_io *io, unsigned int offset)
1381 {
1382 	unsigned int addr = io->addr_data;
1383 
1384 	return (inw(addr + (offset * io->regspacing)) >> io->regshift) & 0xff;
1385 }
1386 
1387 static void port_outw(struct si_sm_io *io, unsigned int offset,
1388 		      unsigned char b)
1389 {
1390 	unsigned int addr = io->addr_data;
1391 
1392 	outw(b << io->regshift, addr + (offset * io->regspacing));
1393 }
1394 
1395 static unsigned char port_inl(struct si_sm_io *io, unsigned int offset)
1396 {
1397 	unsigned int addr = io->addr_data;
1398 
1399 	return (inl(addr + (offset * io->regspacing)) >> io->regshift) & 0xff;
1400 }
1401 
1402 static void port_outl(struct si_sm_io *io, unsigned int offset,
1403 		      unsigned char b)
1404 {
1405 	unsigned int addr = io->addr_data;
1406 
1407 	outl(b << io->regshift, addr+(offset * io->regspacing));
1408 }
1409 
1410 static void port_cleanup(struct smi_info *info)
1411 {
1412 	unsigned int addr = info->io.addr_data;
1413 	int          idx;
1414 
1415 	if (addr) {
1416 		for (idx = 0; idx < info->io_size; idx++)
1417 			release_region(addr + idx * info->io.regspacing,
1418 				       info->io.regsize);
1419 	}
1420 }
1421 
1422 static int port_setup(struct smi_info *info)
1423 {
1424 	unsigned int addr = info->io.addr_data;
1425 	int          idx;
1426 
1427 	if (!addr)
1428 		return -ENODEV;
1429 
1430 	info->io_cleanup = port_cleanup;
1431 
1432 	/*
1433 	 * Figure out the actual inb/inw/inl/etc routine to use based
1434 	 * upon the register size.
1435 	 */
1436 	switch (info->io.regsize) {
1437 	case 1:
1438 		info->io.inputb = port_inb;
1439 		info->io.outputb = port_outb;
1440 		break;
1441 	case 2:
1442 		info->io.inputb = port_inw;
1443 		info->io.outputb = port_outw;
1444 		break;
1445 	case 4:
1446 		info->io.inputb = port_inl;
1447 		info->io.outputb = port_outl;
1448 		break;
1449 	default:
1450 		dev_warn(info->dev, "Invalid register size: %d\n",
1451 			 info->io.regsize);
1452 		return -EINVAL;
1453 	}
1454 
1455 	/*
1456 	 * Some BIOSes reserve disjoint I/O regions in their ACPI
1457 	 * tables.  This causes problems when trying to register the
1458 	 * entire I/O region.  Therefore we must register each I/O
1459 	 * port separately.
1460 	 */
1461 	for (idx = 0; idx < info->io_size; idx++) {
1462 		if (request_region(addr + idx * info->io.regspacing,
1463 				   info->io.regsize, DEVICE_NAME) == NULL) {
1464 			/* Undo allocations */
1465 			while (idx--) {
1466 				release_region(addr + idx * info->io.regspacing,
1467 					       info->io.regsize);
1468 			}
1469 			return -EIO;
1470 		}
1471 	}
1472 	return 0;
1473 }
1474 
1475 static unsigned char intf_mem_inb(struct si_sm_io *io, unsigned int offset)
1476 {
1477 	return readb((io->addr)+(offset * io->regspacing));
1478 }
1479 
1480 static void intf_mem_outb(struct si_sm_io *io, unsigned int offset,
1481 		     unsigned char b)
1482 {
1483 	writeb(b, (io->addr)+(offset * io->regspacing));
1484 }
1485 
1486 static unsigned char intf_mem_inw(struct si_sm_io *io, unsigned int offset)
1487 {
1488 	return (readw((io->addr)+(offset * io->regspacing)) >> io->regshift)
1489 		& 0xff;
1490 }
1491 
1492 static void intf_mem_outw(struct si_sm_io *io, unsigned int offset,
1493 		     unsigned char b)
1494 {
1495 	writeb(b << io->regshift, (io->addr)+(offset * io->regspacing));
1496 }
1497 
1498 static unsigned char intf_mem_inl(struct si_sm_io *io, unsigned int offset)
1499 {
1500 	return (readl((io->addr)+(offset * io->regspacing)) >> io->regshift)
1501 		& 0xff;
1502 }
1503 
1504 static void intf_mem_outl(struct si_sm_io *io, unsigned int offset,
1505 		     unsigned char b)
1506 {
1507 	writel(b << io->regshift, (io->addr)+(offset * io->regspacing));
1508 }
1509 
1510 #ifdef readq
1511 static unsigned char mem_inq(struct si_sm_io *io, unsigned int offset)
1512 {
1513 	return (readq((io->addr)+(offset * io->regspacing)) >> io->regshift)
1514 		& 0xff;
1515 }
1516 
1517 static void mem_outq(struct si_sm_io *io, unsigned int offset,
1518 		     unsigned char b)
1519 {
1520 	writeq(b << io->regshift, (io->addr)+(offset * io->regspacing));
1521 }
1522 #endif
1523 
1524 static void mem_cleanup(struct smi_info *info)
1525 {
1526 	unsigned long addr = info->io.addr_data;
1527 	int           mapsize;
1528 
1529 	if (info->io.addr) {
1530 		iounmap(info->io.addr);
1531 
1532 		mapsize = ((info->io_size * info->io.regspacing)
1533 			   - (info->io.regspacing - info->io.regsize));
1534 
1535 		release_mem_region(addr, mapsize);
1536 	}
1537 }
1538 
1539 static int mem_setup(struct smi_info *info)
1540 {
1541 	unsigned long addr = info->io.addr_data;
1542 	int           mapsize;
1543 
1544 	if (!addr)
1545 		return -ENODEV;
1546 
1547 	info->io_cleanup = mem_cleanup;
1548 
1549 	/*
1550 	 * Figure out the actual readb/readw/readl/etc routine to use based
1551 	 * upon the register size.
1552 	 */
1553 	switch (info->io.regsize) {
1554 	case 1:
1555 		info->io.inputb = intf_mem_inb;
1556 		info->io.outputb = intf_mem_outb;
1557 		break;
1558 	case 2:
1559 		info->io.inputb = intf_mem_inw;
1560 		info->io.outputb = intf_mem_outw;
1561 		break;
1562 	case 4:
1563 		info->io.inputb = intf_mem_inl;
1564 		info->io.outputb = intf_mem_outl;
1565 		break;
1566 #ifdef readq
1567 	case 8:
1568 		info->io.inputb = mem_inq;
1569 		info->io.outputb = mem_outq;
1570 		break;
1571 #endif
1572 	default:
1573 		dev_warn(info->dev, "Invalid register size: %d\n",
1574 			 info->io.regsize);
1575 		return -EINVAL;
1576 	}
1577 
1578 	/*
1579 	 * Calculate the total amount of memory to claim.  This is an
1580 	 * unusual looking calculation, but it avoids claiming any
1581 	 * more memory than it has to.  It will claim everything
1582 	 * between the first address to the end of the last full
1583 	 * register.
1584 	 */
1585 	mapsize = ((info->io_size * info->io.regspacing)
1586 		   - (info->io.regspacing - info->io.regsize));
1587 
1588 	if (request_mem_region(addr, mapsize, DEVICE_NAME) == NULL)
1589 		return -EIO;
1590 
1591 	info->io.addr = ioremap(addr, mapsize);
1592 	if (info->io.addr == NULL) {
1593 		release_mem_region(addr, mapsize);
1594 		return -EIO;
1595 	}
1596 	return 0;
1597 }
1598 
1599 /*
1600  * Parms come in as <op1>[:op2[:op3...]].  ops are:
1601  *   add|remove,kcs|bt|smic,mem|i/o,<address>[,<opt1>[,<opt2>[,...]]]
1602  * Options are:
1603  *   rsp=<regspacing>
1604  *   rsi=<regsize>
1605  *   rsh=<regshift>
1606  *   irq=<irq>
1607  *   ipmb=<ipmb addr>
1608  */
1609 enum hotmod_op { HM_ADD, HM_REMOVE };
1610 struct hotmod_vals {
1611 	char *name;
1612 	int  val;
1613 };
1614 static struct hotmod_vals hotmod_ops[] = {
1615 	{ "add",	HM_ADD },
1616 	{ "remove",	HM_REMOVE },
1617 	{ NULL }
1618 };
1619 static struct hotmod_vals hotmod_si[] = {
1620 	{ "kcs",	SI_KCS },
1621 	{ "smic",	SI_SMIC },
1622 	{ "bt",		SI_BT },
1623 	{ NULL }
1624 };
1625 static struct hotmod_vals hotmod_as[] = {
1626 	{ "mem",	IPMI_MEM_ADDR_SPACE },
1627 	{ "i/o",	IPMI_IO_ADDR_SPACE },
1628 	{ NULL }
1629 };
1630 
1631 static int parse_str(struct hotmod_vals *v, int *val, char *name, char **curr)
1632 {
1633 	char *s;
1634 	int  i;
1635 
1636 	s = strchr(*curr, ',');
1637 	if (!s) {
1638 		printk(KERN_WARNING PFX "No hotmod %s given.\n", name);
1639 		return -EINVAL;
1640 	}
1641 	*s = '\0';
1642 	s++;
1643 	for (i = 0; hotmod_ops[i].name; i++) {
1644 		if (strcmp(*curr, v[i].name) == 0) {
1645 			*val = v[i].val;
1646 			*curr = s;
1647 			return 0;
1648 		}
1649 	}
1650 
1651 	printk(KERN_WARNING PFX "Invalid hotmod %s '%s'\n", name, *curr);
1652 	return -EINVAL;
1653 }
1654 
1655 static int check_hotmod_int_op(const char *curr, const char *option,
1656 			       const char *name, int *val)
1657 {
1658 	char *n;
1659 
1660 	if (strcmp(curr, name) == 0) {
1661 		if (!option) {
1662 			printk(KERN_WARNING PFX
1663 			       "No option given for '%s'\n",
1664 			       curr);
1665 			return -EINVAL;
1666 		}
1667 		*val = simple_strtoul(option, &n, 0);
1668 		if ((*n != '\0') || (*option == '\0')) {
1669 			printk(KERN_WARNING PFX
1670 			       "Bad option given for '%s'\n",
1671 			       curr);
1672 			return -EINVAL;
1673 		}
1674 		return 1;
1675 	}
1676 	return 0;
1677 }
1678 
1679 static struct smi_info *smi_info_alloc(void)
1680 {
1681 	struct smi_info *info = kzalloc(sizeof(*info), GFP_KERNEL);
1682 
1683 	if (info) {
1684 		spin_lock_init(&info->si_lock);
1685 		spin_lock_init(&info->msg_lock);
1686 	}
1687 	return info;
1688 }
1689 
1690 static int hotmod_handler(const char *val, struct kernel_param *kp)
1691 {
1692 	char *str = kstrdup(val, GFP_KERNEL);
1693 	int  rv;
1694 	char *next, *curr, *s, *n, *o;
1695 	enum hotmod_op op;
1696 	enum si_type si_type;
1697 	int  addr_space;
1698 	unsigned long addr;
1699 	int regspacing;
1700 	int regsize;
1701 	int regshift;
1702 	int irq;
1703 	int ipmb;
1704 	int ival;
1705 	int len;
1706 	struct smi_info *info;
1707 
1708 	if (!str)
1709 		return -ENOMEM;
1710 
1711 	/* Kill any trailing spaces, as we can get a "\n" from echo. */
1712 	len = strlen(str);
1713 	ival = len - 1;
1714 	while ((ival >= 0) && isspace(str[ival])) {
1715 		str[ival] = '\0';
1716 		ival--;
1717 	}
1718 
1719 	for (curr = str; curr; curr = next) {
1720 		regspacing = 1;
1721 		regsize = 1;
1722 		regshift = 0;
1723 		irq = 0;
1724 		ipmb = 0; /* Choose the default if not specified */
1725 
1726 		next = strchr(curr, ':');
1727 		if (next) {
1728 			*next = '\0';
1729 			next++;
1730 		}
1731 
1732 		rv = parse_str(hotmod_ops, &ival, "operation", &curr);
1733 		if (rv)
1734 			break;
1735 		op = ival;
1736 
1737 		rv = parse_str(hotmod_si, &ival, "interface type", &curr);
1738 		if (rv)
1739 			break;
1740 		si_type = ival;
1741 
1742 		rv = parse_str(hotmod_as, &addr_space, "address space", &curr);
1743 		if (rv)
1744 			break;
1745 
1746 		s = strchr(curr, ',');
1747 		if (s) {
1748 			*s = '\0';
1749 			s++;
1750 		}
1751 		addr = simple_strtoul(curr, &n, 0);
1752 		if ((*n != '\0') || (*curr == '\0')) {
1753 			printk(KERN_WARNING PFX "Invalid hotmod address"
1754 			       " '%s'\n", curr);
1755 			break;
1756 		}
1757 
1758 		while (s) {
1759 			curr = s;
1760 			s = strchr(curr, ',');
1761 			if (s) {
1762 				*s = '\0';
1763 				s++;
1764 			}
1765 			o = strchr(curr, '=');
1766 			if (o) {
1767 				*o = '\0';
1768 				o++;
1769 			}
1770 			rv = check_hotmod_int_op(curr, o, "rsp", &regspacing);
1771 			if (rv < 0)
1772 				goto out;
1773 			else if (rv)
1774 				continue;
1775 			rv = check_hotmod_int_op(curr, o, "rsi", &regsize);
1776 			if (rv < 0)
1777 				goto out;
1778 			else if (rv)
1779 				continue;
1780 			rv = check_hotmod_int_op(curr, o, "rsh", &regshift);
1781 			if (rv < 0)
1782 				goto out;
1783 			else if (rv)
1784 				continue;
1785 			rv = check_hotmod_int_op(curr, o, "irq", &irq);
1786 			if (rv < 0)
1787 				goto out;
1788 			else if (rv)
1789 				continue;
1790 			rv = check_hotmod_int_op(curr, o, "ipmb", &ipmb);
1791 			if (rv < 0)
1792 				goto out;
1793 			else if (rv)
1794 				continue;
1795 
1796 			rv = -EINVAL;
1797 			printk(KERN_WARNING PFX
1798 			       "Invalid hotmod option '%s'\n",
1799 			       curr);
1800 			goto out;
1801 		}
1802 
1803 		if (op == HM_ADD) {
1804 			info = smi_info_alloc();
1805 			if (!info) {
1806 				rv = -ENOMEM;
1807 				goto out;
1808 			}
1809 
1810 			info->addr_source = SI_HOTMOD;
1811 			info->si_type = si_type;
1812 			info->io.addr_data = addr;
1813 			info->io.addr_type = addr_space;
1814 			if (addr_space == IPMI_MEM_ADDR_SPACE)
1815 				info->io_setup = mem_setup;
1816 			else
1817 				info->io_setup = port_setup;
1818 
1819 			info->io.addr = NULL;
1820 			info->io.regspacing = regspacing;
1821 			if (!info->io.regspacing)
1822 				info->io.regspacing = DEFAULT_REGSPACING;
1823 			info->io.regsize = regsize;
1824 			if (!info->io.regsize)
1825 				info->io.regsize = DEFAULT_REGSPACING;
1826 			info->io.regshift = regshift;
1827 			info->irq = irq;
1828 			if (info->irq)
1829 				info->irq_setup = std_irq_setup;
1830 			info->slave_addr = ipmb;
1831 
1832 			if (!add_smi(info)) {
1833 				if (try_smi_init(info))
1834 					cleanup_one_si(info);
1835 			} else {
1836 				kfree(info);
1837 			}
1838 		} else {
1839 			/* remove */
1840 			struct smi_info *e, *tmp_e;
1841 
1842 			mutex_lock(&smi_infos_lock);
1843 			list_for_each_entry_safe(e, tmp_e, &smi_infos, link) {
1844 				if (e->io.addr_type != addr_space)
1845 					continue;
1846 				if (e->si_type != si_type)
1847 					continue;
1848 				if (e->io.addr_data == addr)
1849 					cleanup_one_si(e);
1850 			}
1851 			mutex_unlock(&smi_infos_lock);
1852 		}
1853 	}
1854 	rv = len;
1855  out:
1856 	kfree(str);
1857 	return rv;
1858 }
1859 
1860 static int __devinit hardcode_find_bmc(void)
1861 {
1862 	int ret = -ENODEV;
1863 	int             i;
1864 	struct smi_info *info;
1865 
1866 	for (i = 0; i < SI_MAX_PARMS; i++) {
1867 		if (!ports[i] && !addrs[i])
1868 			continue;
1869 
1870 		info = smi_info_alloc();
1871 		if (!info)
1872 			return -ENOMEM;
1873 
1874 		info->addr_source = SI_HARDCODED;
1875 		printk(KERN_INFO PFX "probing via hardcoded address\n");
1876 
1877 		if (!si_type[i] || strcmp(si_type[i], "kcs") == 0) {
1878 			info->si_type = SI_KCS;
1879 		} else if (strcmp(si_type[i], "smic") == 0) {
1880 			info->si_type = SI_SMIC;
1881 		} else if (strcmp(si_type[i], "bt") == 0) {
1882 			info->si_type = SI_BT;
1883 		} else {
1884 			printk(KERN_WARNING PFX "Interface type specified "
1885 			       "for interface %d, was invalid: %s\n",
1886 			       i, si_type[i]);
1887 			kfree(info);
1888 			continue;
1889 		}
1890 
1891 		if (ports[i]) {
1892 			/* An I/O port */
1893 			info->io_setup = port_setup;
1894 			info->io.addr_data = ports[i];
1895 			info->io.addr_type = IPMI_IO_ADDR_SPACE;
1896 		} else if (addrs[i]) {
1897 			/* A memory port */
1898 			info->io_setup = mem_setup;
1899 			info->io.addr_data = addrs[i];
1900 			info->io.addr_type = IPMI_MEM_ADDR_SPACE;
1901 		} else {
1902 			printk(KERN_WARNING PFX "Interface type specified "
1903 			       "for interface %d, but port and address were "
1904 			       "not set or set to zero.\n", i);
1905 			kfree(info);
1906 			continue;
1907 		}
1908 
1909 		info->io.addr = NULL;
1910 		info->io.regspacing = regspacings[i];
1911 		if (!info->io.regspacing)
1912 			info->io.regspacing = DEFAULT_REGSPACING;
1913 		info->io.regsize = regsizes[i];
1914 		if (!info->io.regsize)
1915 			info->io.regsize = DEFAULT_REGSPACING;
1916 		info->io.regshift = regshifts[i];
1917 		info->irq = irqs[i];
1918 		if (info->irq)
1919 			info->irq_setup = std_irq_setup;
1920 		info->slave_addr = slave_addrs[i];
1921 
1922 		if (!add_smi(info)) {
1923 			if (try_smi_init(info))
1924 				cleanup_one_si(info);
1925 			ret = 0;
1926 		} else {
1927 			kfree(info);
1928 		}
1929 	}
1930 	return ret;
1931 }
1932 
1933 #ifdef CONFIG_ACPI
1934 
1935 #include <linux/acpi.h>
1936 
1937 /*
1938  * Once we get an ACPI failure, we don't try any more, because we go
1939  * through the tables sequentially.  Once we don't find a table, there
1940  * are no more.
1941  */
1942 static int acpi_failure;
1943 
1944 /* For GPE-type interrupts. */
1945 static u32 ipmi_acpi_gpe(acpi_handle gpe_device,
1946 	u32 gpe_number, void *context)
1947 {
1948 	struct smi_info *smi_info = context;
1949 	unsigned long   flags;
1950 #ifdef DEBUG_TIMING
1951 	struct timeval t;
1952 #endif
1953 
1954 	spin_lock_irqsave(&(smi_info->si_lock), flags);
1955 
1956 	smi_inc_stat(smi_info, interrupts);
1957 
1958 #ifdef DEBUG_TIMING
1959 	do_gettimeofday(&t);
1960 	printk("**ACPI_GPE: %d.%9.9d\n", t.tv_sec, t.tv_usec);
1961 #endif
1962 	smi_event_handler(smi_info, 0);
1963 	spin_unlock_irqrestore(&(smi_info->si_lock), flags);
1964 
1965 	return ACPI_INTERRUPT_HANDLED;
1966 }
1967 
1968 static void acpi_gpe_irq_cleanup(struct smi_info *info)
1969 {
1970 	if (!info->irq)
1971 		return;
1972 
1973 	acpi_remove_gpe_handler(NULL, info->irq, &ipmi_acpi_gpe);
1974 }
1975 
1976 static int acpi_gpe_irq_setup(struct smi_info *info)
1977 {
1978 	acpi_status status;
1979 
1980 	if (!info->irq)
1981 		return 0;
1982 
1983 	/* FIXME - is level triggered right? */
1984 	status = acpi_install_gpe_handler(NULL,
1985 					  info->irq,
1986 					  ACPI_GPE_LEVEL_TRIGGERED,
1987 					  &ipmi_acpi_gpe,
1988 					  info);
1989 	if (status != AE_OK) {
1990 		dev_warn(info->dev, "%s unable to claim ACPI GPE %d,"
1991 			 " running polled\n", DEVICE_NAME, info->irq);
1992 		info->irq = 0;
1993 		return -EINVAL;
1994 	} else {
1995 		info->irq_cleanup = acpi_gpe_irq_cleanup;
1996 		dev_info(info->dev, "Using ACPI GPE %d\n", info->irq);
1997 		return 0;
1998 	}
1999 }
2000 
2001 /*
2002  * Defined at
2003  * http://h21007.www2.hp.com/portal/download/files/unprot/hpspmi.pdf
2004  */
2005 struct SPMITable {
2006 	s8	Signature[4];
2007 	u32	Length;
2008 	u8	Revision;
2009 	u8	Checksum;
2010 	s8	OEMID[6];
2011 	s8	OEMTableID[8];
2012 	s8	OEMRevision[4];
2013 	s8	CreatorID[4];
2014 	s8	CreatorRevision[4];
2015 	u8	InterfaceType;
2016 	u8	IPMIlegacy;
2017 	s16	SpecificationRevision;
2018 
2019 	/*
2020 	 * Bit 0 - SCI interrupt supported
2021 	 * Bit 1 - I/O APIC/SAPIC
2022 	 */
2023 	u8	InterruptType;
2024 
2025 	/*
2026 	 * If bit 0 of InterruptType is set, then this is the SCI
2027 	 * interrupt in the GPEx_STS register.
2028 	 */
2029 	u8	GPE;
2030 
2031 	s16	Reserved;
2032 
2033 	/*
2034 	 * If bit 1 of InterruptType is set, then this is the I/O
2035 	 * APIC/SAPIC interrupt.
2036 	 */
2037 	u32	GlobalSystemInterrupt;
2038 
2039 	/* The actual register address. */
2040 	struct acpi_generic_address addr;
2041 
2042 	u8	UID[4];
2043 
2044 	s8      spmi_id[1]; /* A '\0' terminated array starts here. */
2045 };
2046 
2047 static int __devinit try_init_spmi(struct SPMITable *spmi)
2048 {
2049 	struct smi_info  *info;
2050 
2051 	if (spmi->IPMIlegacy != 1) {
2052 		printk(KERN_INFO PFX "Bad SPMI legacy %d\n", spmi->IPMIlegacy);
2053 		return -ENODEV;
2054 	}
2055 
2056 	info = smi_info_alloc();
2057 	if (!info) {
2058 		printk(KERN_ERR PFX "Could not allocate SI data (3)\n");
2059 		return -ENOMEM;
2060 	}
2061 
2062 	info->addr_source = SI_SPMI;
2063 	printk(KERN_INFO PFX "probing via SPMI\n");
2064 
2065 	/* Figure out the interface type. */
2066 	switch (spmi->InterfaceType) {
2067 	case 1:	/* KCS */
2068 		info->si_type = SI_KCS;
2069 		break;
2070 	case 2:	/* SMIC */
2071 		info->si_type = SI_SMIC;
2072 		break;
2073 	case 3:	/* BT */
2074 		info->si_type = SI_BT;
2075 		break;
2076 	default:
2077 		printk(KERN_INFO PFX "Unknown ACPI/SPMI SI type %d\n",
2078 		       spmi->InterfaceType);
2079 		kfree(info);
2080 		return -EIO;
2081 	}
2082 
2083 	if (spmi->InterruptType & 1) {
2084 		/* We've got a GPE interrupt. */
2085 		info->irq = spmi->GPE;
2086 		info->irq_setup = acpi_gpe_irq_setup;
2087 	} else if (spmi->InterruptType & 2) {
2088 		/* We've got an APIC/SAPIC interrupt. */
2089 		info->irq = spmi->GlobalSystemInterrupt;
2090 		info->irq_setup = std_irq_setup;
2091 	} else {
2092 		/* Use the default interrupt setting. */
2093 		info->irq = 0;
2094 		info->irq_setup = NULL;
2095 	}
2096 
2097 	if (spmi->addr.bit_width) {
2098 		/* A (hopefully) properly formed register bit width. */
2099 		info->io.regspacing = spmi->addr.bit_width / 8;
2100 	} else {
2101 		info->io.regspacing = DEFAULT_REGSPACING;
2102 	}
2103 	info->io.regsize = info->io.regspacing;
2104 	info->io.regshift = spmi->addr.bit_offset;
2105 
2106 	if (spmi->addr.space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
2107 		info->io_setup = mem_setup;
2108 		info->io.addr_type = IPMI_MEM_ADDR_SPACE;
2109 	} else if (spmi->addr.space_id == ACPI_ADR_SPACE_SYSTEM_IO) {
2110 		info->io_setup = port_setup;
2111 		info->io.addr_type = IPMI_IO_ADDR_SPACE;
2112 	} else {
2113 		kfree(info);
2114 		printk(KERN_WARNING PFX "Unknown ACPI I/O Address type\n");
2115 		return -EIO;
2116 	}
2117 	info->io.addr_data = spmi->addr.address;
2118 
2119 	pr_info("ipmi_si: SPMI: %s %#lx regsize %d spacing %d irq %d\n",
2120 		 (info->io.addr_type == IPMI_IO_ADDR_SPACE) ? "io" : "mem",
2121 		 info->io.addr_data, info->io.regsize, info->io.regspacing,
2122 		 info->irq);
2123 
2124 	if (add_smi(info))
2125 		kfree(info);
2126 
2127 	return 0;
2128 }
2129 
2130 static void __devinit spmi_find_bmc(void)
2131 {
2132 	acpi_status      status;
2133 	struct SPMITable *spmi;
2134 	int              i;
2135 
2136 	if (acpi_disabled)
2137 		return;
2138 
2139 	if (acpi_failure)
2140 		return;
2141 
2142 	for (i = 0; ; i++) {
2143 		status = acpi_get_table(ACPI_SIG_SPMI, i+1,
2144 					(struct acpi_table_header **)&spmi);
2145 		if (status != AE_OK)
2146 			return;
2147 
2148 		try_init_spmi(spmi);
2149 	}
2150 }
2151 
2152 static int __devinit ipmi_pnp_probe(struct pnp_dev *dev,
2153 				    const struct pnp_device_id *dev_id)
2154 {
2155 	struct acpi_device *acpi_dev;
2156 	struct smi_info *info;
2157 	struct resource *res, *res_second;
2158 	acpi_handle handle;
2159 	acpi_status status;
2160 	unsigned long long tmp;
2161 
2162 	acpi_dev = pnp_acpi_device(dev);
2163 	if (!acpi_dev)
2164 		return -ENODEV;
2165 
2166 	info = smi_info_alloc();
2167 	if (!info)
2168 		return -ENOMEM;
2169 
2170 	info->addr_source = SI_ACPI;
2171 	printk(KERN_INFO PFX "probing via ACPI\n");
2172 
2173 	handle = acpi_dev->handle;
2174 	info->addr_info.acpi_info.acpi_handle = handle;
2175 
2176 	/* _IFT tells us the interface type: KCS, BT, etc */
2177 	status = acpi_evaluate_integer(handle, "_IFT", NULL, &tmp);
2178 	if (ACPI_FAILURE(status))
2179 		goto err_free;
2180 
2181 	switch (tmp) {
2182 	case 1:
2183 		info->si_type = SI_KCS;
2184 		break;
2185 	case 2:
2186 		info->si_type = SI_SMIC;
2187 		break;
2188 	case 3:
2189 		info->si_type = SI_BT;
2190 		break;
2191 	default:
2192 		dev_info(&dev->dev, "unknown IPMI type %lld\n", tmp);
2193 		goto err_free;
2194 	}
2195 
2196 	res = pnp_get_resource(dev, IORESOURCE_IO, 0);
2197 	if (res) {
2198 		info->io_setup = port_setup;
2199 		info->io.addr_type = IPMI_IO_ADDR_SPACE;
2200 	} else {
2201 		res = pnp_get_resource(dev, IORESOURCE_MEM, 0);
2202 		if (res) {
2203 			info->io_setup = mem_setup;
2204 			info->io.addr_type = IPMI_MEM_ADDR_SPACE;
2205 		}
2206 	}
2207 	if (!res) {
2208 		dev_err(&dev->dev, "no I/O or memory address\n");
2209 		goto err_free;
2210 	}
2211 	info->io.addr_data = res->start;
2212 
2213 	info->io.regspacing = DEFAULT_REGSPACING;
2214 	res_second = pnp_get_resource(dev,
2215 			       (info->io.addr_type == IPMI_IO_ADDR_SPACE) ?
2216 					IORESOURCE_IO : IORESOURCE_MEM,
2217 			       1);
2218 	if (res_second) {
2219 		if (res_second->start > info->io.addr_data)
2220 			info->io.regspacing = res_second->start - info->io.addr_data;
2221 	}
2222 	info->io.regsize = DEFAULT_REGSPACING;
2223 	info->io.regshift = 0;
2224 
2225 	/* If _GPE exists, use it; otherwise use standard interrupts */
2226 	status = acpi_evaluate_integer(handle, "_GPE", NULL, &tmp);
2227 	if (ACPI_SUCCESS(status)) {
2228 		info->irq = tmp;
2229 		info->irq_setup = acpi_gpe_irq_setup;
2230 	} else if (pnp_irq_valid(dev, 0)) {
2231 		info->irq = pnp_irq(dev, 0);
2232 		info->irq_setup = std_irq_setup;
2233 	}
2234 
2235 	info->dev = &dev->dev;
2236 	pnp_set_drvdata(dev, info);
2237 
2238 	dev_info(info->dev, "%pR regsize %d spacing %d irq %d\n",
2239 		 res, info->io.regsize, info->io.regspacing,
2240 		 info->irq);
2241 
2242 	if (add_smi(info))
2243 		goto err_free;
2244 
2245 	return 0;
2246 
2247 err_free:
2248 	kfree(info);
2249 	return -EINVAL;
2250 }
2251 
2252 static void __devexit ipmi_pnp_remove(struct pnp_dev *dev)
2253 {
2254 	struct smi_info *info = pnp_get_drvdata(dev);
2255 
2256 	cleanup_one_si(info);
2257 }
2258 
2259 static const struct pnp_device_id pnp_dev_table[] = {
2260 	{"IPI0001", 0},
2261 	{"", 0},
2262 };
2263 
2264 static struct pnp_driver ipmi_pnp_driver = {
2265 	.name		= DEVICE_NAME,
2266 	.probe		= ipmi_pnp_probe,
2267 	.remove		= __devexit_p(ipmi_pnp_remove),
2268 	.id_table	= pnp_dev_table,
2269 };
2270 #endif
2271 
2272 #ifdef CONFIG_DMI
2273 struct dmi_ipmi_data {
2274 	u8   		type;
2275 	u8   		addr_space;
2276 	unsigned long	base_addr;
2277 	u8   		irq;
2278 	u8              offset;
2279 	u8              slave_addr;
2280 };
2281 
2282 static int __devinit decode_dmi(const struct dmi_header *dm,
2283 				struct dmi_ipmi_data *dmi)
2284 {
2285 	const u8	*data = (const u8 *)dm;
2286 	unsigned long  	base_addr;
2287 	u8		reg_spacing;
2288 	u8              len = dm->length;
2289 
2290 	dmi->type = data[4];
2291 
2292 	memcpy(&base_addr, data+8, sizeof(unsigned long));
2293 	if (len >= 0x11) {
2294 		if (base_addr & 1) {
2295 			/* I/O */
2296 			base_addr &= 0xFFFE;
2297 			dmi->addr_space = IPMI_IO_ADDR_SPACE;
2298 		} else
2299 			/* Memory */
2300 			dmi->addr_space = IPMI_MEM_ADDR_SPACE;
2301 
2302 		/* If bit 4 of byte 0x10 is set, then the lsb for the address
2303 		   is odd. */
2304 		dmi->base_addr = base_addr | ((data[0x10] & 0x10) >> 4);
2305 
2306 		dmi->irq = data[0x11];
2307 
2308 		/* The top two bits of byte 0x10 hold the register spacing. */
2309 		reg_spacing = (data[0x10] & 0xC0) >> 6;
2310 		switch (reg_spacing) {
2311 		case 0x00: /* Byte boundaries */
2312 		    dmi->offset = 1;
2313 		    break;
2314 		case 0x01: /* 32-bit boundaries */
2315 		    dmi->offset = 4;
2316 		    break;
2317 		case 0x02: /* 16-byte boundaries */
2318 		    dmi->offset = 16;
2319 		    break;
2320 		default:
2321 		    /* Some other interface, just ignore it. */
2322 		    return -EIO;
2323 		}
2324 	} else {
2325 		/* Old DMI spec. */
2326 		/*
2327 		 * Note that technically, the lower bit of the base
2328 		 * address should be 1 if the address is I/O and 0 if
2329 		 * the address is in memory.  So many systems get that
2330 		 * wrong (and all that I have seen are I/O) so we just
2331 		 * ignore that bit and assume I/O.  Systems that use
2332 		 * memory should use the newer spec, anyway.
2333 		 */
2334 		dmi->base_addr = base_addr & 0xfffe;
2335 		dmi->addr_space = IPMI_IO_ADDR_SPACE;
2336 		dmi->offset = 1;
2337 	}
2338 
2339 	dmi->slave_addr = data[6];
2340 
2341 	return 0;
2342 }
2343 
2344 static void __devinit try_init_dmi(struct dmi_ipmi_data *ipmi_data)
2345 {
2346 	struct smi_info *info;
2347 
2348 	info = smi_info_alloc();
2349 	if (!info) {
2350 		printk(KERN_ERR PFX "Could not allocate SI data\n");
2351 		return;
2352 	}
2353 
2354 	info->addr_source = SI_SMBIOS;
2355 	printk(KERN_INFO PFX "probing via SMBIOS\n");
2356 
2357 	switch (ipmi_data->type) {
2358 	case 0x01: /* KCS */
2359 		info->si_type = SI_KCS;
2360 		break;
2361 	case 0x02: /* SMIC */
2362 		info->si_type = SI_SMIC;
2363 		break;
2364 	case 0x03: /* BT */
2365 		info->si_type = SI_BT;
2366 		break;
2367 	default:
2368 		kfree(info);
2369 		return;
2370 	}
2371 
2372 	switch (ipmi_data->addr_space) {
2373 	case IPMI_MEM_ADDR_SPACE:
2374 		info->io_setup = mem_setup;
2375 		info->io.addr_type = IPMI_MEM_ADDR_SPACE;
2376 		break;
2377 
2378 	case IPMI_IO_ADDR_SPACE:
2379 		info->io_setup = port_setup;
2380 		info->io.addr_type = IPMI_IO_ADDR_SPACE;
2381 		break;
2382 
2383 	default:
2384 		kfree(info);
2385 		printk(KERN_WARNING PFX "Unknown SMBIOS I/O Address type: %d\n",
2386 		       ipmi_data->addr_space);
2387 		return;
2388 	}
2389 	info->io.addr_data = ipmi_data->base_addr;
2390 
2391 	info->io.regspacing = ipmi_data->offset;
2392 	if (!info->io.regspacing)
2393 		info->io.regspacing = DEFAULT_REGSPACING;
2394 	info->io.regsize = DEFAULT_REGSPACING;
2395 	info->io.regshift = 0;
2396 
2397 	info->slave_addr = ipmi_data->slave_addr;
2398 
2399 	info->irq = ipmi_data->irq;
2400 	if (info->irq)
2401 		info->irq_setup = std_irq_setup;
2402 
2403 	pr_info("ipmi_si: SMBIOS: %s %#lx regsize %d spacing %d irq %d\n",
2404 		 (info->io.addr_type == IPMI_IO_ADDR_SPACE) ? "io" : "mem",
2405 		 info->io.addr_data, info->io.regsize, info->io.regspacing,
2406 		 info->irq);
2407 
2408 	if (add_smi(info))
2409 		kfree(info);
2410 }
2411 
2412 static void __devinit dmi_find_bmc(void)
2413 {
2414 	const struct dmi_device *dev = NULL;
2415 	struct dmi_ipmi_data data;
2416 	int                  rv;
2417 
2418 	while ((dev = dmi_find_device(DMI_DEV_TYPE_IPMI, NULL, dev))) {
2419 		memset(&data, 0, sizeof(data));
2420 		rv = decode_dmi((const struct dmi_header *) dev->device_data,
2421 				&data);
2422 		if (!rv)
2423 			try_init_dmi(&data);
2424 	}
2425 }
2426 #endif /* CONFIG_DMI */
2427 
2428 #ifdef CONFIG_PCI
2429 
2430 #define PCI_ERMC_CLASSCODE		0x0C0700
2431 #define PCI_ERMC_CLASSCODE_MASK		0xffffff00
2432 #define PCI_ERMC_CLASSCODE_TYPE_MASK	0xff
2433 #define PCI_ERMC_CLASSCODE_TYPE_SMIC	0x00
2434 #define PCI_ERMC_CLASSCODE_TYPE_KCS	0x01
2435 #define PCI_ERMC_CLASSCODE_TYPE_BT	0x02
2436 
2437 #define PCI_HP_VENDOR_ID    0x103C
2438 #define PCI_MMC_DEVICE_ID   0x121A
2439 #define PCI_MMC_ADDR_CW     0x10
2440 
2441 static void ipmi_pci_cleanup(struct smi_info *info)
2442 {
2443 	struct pci_dev *pdev = info->addr_source_data;
2444 
2445 	pci_disable_device(pdev);
2446 }
2447 
2448 static int __devinit ipmi_pci_probe(struct pci_dev *pdev,
2449 				    const struct pci_device_id *ent)
2450 {
2451 	int rv;
2452 	int class_type = pdev->class & PCI_ERMC_CLASSCODE_TYPE_MASK;
2453 	struct smi_info *info;
2454 
2455 	info = smi_info_alloc();
2456 	if (!info)
2457 		return -ENOMEM;
2458 
2459 	info->addr_source = SI_PCI;
2460 	dev_info(&pdev->dev, "probing via PCI");
2461 
2462 	switch (class_type) {
2463 	case PCI_ERMC_CLASSCODE_TYPE_SMIC:
2464 		info->si_type = SI_SMIC;
2465 		break;
2466 
2467 	case PCI_ERMC_CLASSCODE_TYPE_KCS:
2468 		info->si_type = SI_KCS;
2469 		break;
2470 
2471 	case PCI_ERMC_CLASSCODE_TYPE_BT:
2472 		info->si_type = SI_BT;
2473 		break;
2474 
2475 	default:
2476 		kfree(info);
2477 		dev_info(&pdev->dev, "Unknown IPMI type: %d\n", class_type);
2478 		return -ENOMEM;
2479 	}
2480 
2481 	rv = pci_enable_device(pdev);
2482 	if (rv) {
2483 		dev_err(&pdev->dev, "couldn't enable PCI device\n");
2484 		kfree(info);
2485 		return rv;
2486 	}
2487 
2488 	info->addr_source_cleanup = ipmi_pci_cleanup;
2489 	info->addr_source_data = pdev;
2490 
2491 	if (pci_resource_flags(pdev, 0) & IORESOURCE_IO) {
2492 		info->io_setup = port_setup;
2493 		info->io.addr_type = IPMI_IO_ADDR_SPACE;
2494 	} else {
2495 		info->io_setup = mem_setup;
2496 		info->io.addr_type = IPMI_MEM_ADDR_SPACE;
2497 	}
2498 	info->io.addr_data = pci_resource_start(pdev, 0);
2499 
2500 	info->io.regspacing = DEFAULT_REGSPACING;
2501 	info->io.regsize = DEFAULT_REGSPACING;
2502 	info->io.regshift = 0;
2503 
2504 	info->irq = pdev->irq;
2505 	if (info->irq)
2506 		info->irq_setup = std_irq_setup;
2507 
2508 	info->dev = &pdev->dev;
2509 	pci_set_drvdata(pdev, info);
2510 
2511 	dev_info(&pdev->dev, "%pR regsize %d spacing %d irq %d\n",
2512 		&pdev->resource[0], info->io.regsize, info->io.regspacing,
2513 		info->irq);
2514 
2515 	if (add_smi(info))
2516 		kfree(info);
2517 
2518 	return 0;
2519 }
2520 
2521 static void __devexit ipmi_pci_remove(struct pci_dev *pdev)
2522 {
2523 	struct smi_info *info = pci_get_drvdata(pdev);
2524 	cleanup_one_si(info);
2525 }
2526 
2527 #ifdef CONFIG_PM
2528 static int ipmi_pci_suspend(struct pci_dev *pdev, pm_message_t state)
2529 {
2530 	return 0;
2531 }
2532 
2533 static int ipmi_pci_resume(struct pci_dev *pdev)
2534 {
2535 	return 0;
2536 }
2537 #endif
2538 
2539 static struct pci_device_id ipmi_pci_devices[] = {
2540 	{ PCI_DEVICE(PCI_HP_VENDOR_ID, PCI_MMC_DEVICE_ID) },
2541 	{ PCI_DEVICE_CLASS(PCI_ERMC_CLASSCODE, PCI_ERMC_CLASSCODE_MASK) },
2542 	{ 0, }
2543 };
2544 MODULE_DEVICE_TABLE(pci, ipmi_pci_devices);
2545 
2546 static struct pci_driver ipmi_pci_driver = {
2547 	.name =         DEVICE_NAME,
2548 	.id_table =     ipmi_pci_devices,
2549 	.probe =        ipmi_pci_probe,
2550 	.remove =       __devexit_p(ipmi_pci_remove),
2551 #ifdef CONFIG_PM
2552 	.suspend =      ipmi_pci_suspend,
2553 	.resume =       ipmi_pci_resume,
2554 #endif
2555 };
2556 #endif /* CONFIG_PCI */
2557 
2558 static struct of_device_id ipmi_match[];
2559 static int __devinit ipmi_probe(struct platform_device *dev)
2560 {
2561 #ifdef CONFIG_OF
2562 	const struct of_device_id *match;
2563 	struct smi_info *info;
2564 	struct resource resource;
2565 	const __be32 *regsize, *regspacing, *regshift;
2566 	struct device_node *np = dev->dev.of_node;
2567 	int ret;
2568 	int proplen;
2569 
2570 	dev_info(&dev->dev, "probing via device tree\n");
2571 
2572 	match = of_match_device(ipmi_match, &dev->dev);
2573 	if (!match)
2574 		return -EINVAL;
2575 
2576 	ret = of_address_to_resource(np, 0, &resource);
2577 	if (ret) {
2578 		dev_warn(&dev->dev, PFX "invalid address from OF\n");
2579 		return ret;
2580 	}
2581 
2582 	regsize = of_get_property(np, "reg-size", &proplen);
2583 	if (regsize && proplen != 4) {
2584 		dev_warn(&dev->dev, PFX "invalid regsize from OF\n");
2585 		return -EINVAL;
2586 	}
2587 
2588 	regspacing = of_get_property(np, "reg-spacing", &proplen);
2589 	if (regspacing && proplen != 4) {
2590 		dev_warn(&dev->dev, PFX "invalid regspacing from OF\n");
2591 		return -EINVAL;
2592 	}
2593 
2594 	regshift = of_get_property(np, "reg-shift", &proplen);
2595 	if (regshift && proplen != 4) {
2596 		dev_warn(&dev->dev, PFX "invalid regshift from OF\n");
2597 		return -EINVAL;
2598 	}
2599 
2600 	info = smi_info_alloc();
2601 
2602 	if (!info) {
2603 		dev_err(&dev->dev,
2604 			"could not allocate memory for OF probe\n");
2605 		return -ENOMEM;
2606 	}
2607 
2608 	info->si_type		= (enum si_type) match->data;
2609 	info->addr_source	= SI_DEVICETREE;
2610 	info->irq_setup		= std_irq_setup;
2611 
2612 	if (resource.flags & IORESOURCE_IO) {
2613 		info->io_setup		= port_setup;
2614 		info->io.addr_type	= IPMI_IO_ADDR_SPACE;
2615 	} else {
2616 		info->io_setup		= mem_setup;
2617 		info->io.addr_type	= IPMI_MEM_ADDR_SPACE;
2618 	}
2619 
2620 	info->io.addr_data	= resource.start;
2621 
2622 	info->io.regsize	= regsize ? be32_to_cpup(regsize) : DEFAULT_REGSIZE;
2623 	info->io.regspacing	= regspacing ? be32_to_cpup(regspacing) : DEFAULT_REGSPACING;
2624 	info->io.regshift	= regshift ? be32_to_cpup(regshift) : 0;
2625 
2626 	info->irq		= irq_of_parse_and_map(dev->dev.of_node, 0);
2627 	info->dev		= &dev->dev;
2628 
2629 	dev_dbg(&dev->dev, "addr 0x%lx regsize %d spacing %d irq %d\n",
2630 		info->io.addr_data, info->io.regsize, info->io.regspacing,
2631 		info->irq);
2632 
2633 	dev_set_drvdata(&dev->dev, info);
2634 
2635 	if (add_smi(info)) {
2636 		kfree(info);
2637 		return -EBUSY;
2638 	}
2639 #endif
2640 	return 0;
2641 }
2642 
2643 static int __devexit ipmi_remove(struct platform_device *dev)
2644 {
2645 #ifdef CONFIG_OF
2646 	cleanup_one_si(dev_get_drvdata(&dev->dev));
2647 #endif
2648 	return 0;
2649 }
2650 
2651 static struct of_device_id ipmi_match[] =
2652 {
2653 	{ .type = "ipmi", .compatible = "ipmi-kcs",
2654 	  .data = (void *)(unsigned long) SI_KCS },
2655 	{ .type = "ipmi", .compatible = "ipmi-smic",
2656 	  .data = (void *)(unsigned long) SI_SMIC },
2657 	{ .type = "ipmi", .compatible = "ipmi-bt",
2658 	  .data = (void *)(unsigned long) SI_BT },
2659 	{},
2660 };
2661 
2662 static struct platform_driver ipmi_driver = {
2663 	.driver = {
2664 		.name = DEVICE_NAME,
2665 		.owner = THIS_MODULE,
2666 		.of_match_table = ipmi_match,
2667 	},
2668 	.probe		= ipmi_probe,
2669 	.remove		= __devexit_p(ipmi_remove),
2670 };
2671 
2672 static int wait_for_msg_done(struct smi_info *smi_info)
2673 {
2674 	enum si_sm_result     smi_result;
2675 
2676 	smi_result = smi_info->handlers->event(smi_info->si_sm, 0);
2677 	for (;;) {
2678 		if (smi_result == SI_SM_CALL_WITH_DELAY ||
2679 		    smi_result == SI_SM_CALL_WITH_TICK_DELAY) {
2680 			schedule_timeout_uninterruptible(1);
2681 			smi_result = smi_info->handlers->event(
2682 				smi_info->si_sm, 100);
2683 		} else if (smi_result == SI_SM_CALL_WITHOUT_DELAY) {
2684 			smi_result = smi_info->handlers->event(
2685 				smi_info->si_sm, 0);
2686 		} else
2687 			break;
2688 	}
2689 	if (smi_result == SI_SM_HOSED)
2690 		/*
2691 		 * We couldn't get the state machine to run, so whatever's at
2692 		 * the port is probably not an IPMI SMI interface.
2693 		 */
2694 		return -ENODEV;
2695 
2696 	return 0;
2697 }
2698 
2699 static int try_get_dev_id(struct smi_info *smi_info)
2700 {
2701 	unsigned char         msg[2];
2702 	unsigned char         *resp;
2703 	unsigned long         resp_len;
2704 	int                   rv = 0;
2705 
2706 	resp = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
2707 	if (!resp)
2708 		return -ENOMEM;
2709 
2710 	/*
2711 	 * Do a Get Device ID command, since it comes back with some
2712 	 * useful info.
2713 	 */
2714 	msg[0] = IPMI_NETFN_APP_REQUEST << 2;
2715 	msg[1] = IPMI_GET_DEVICE_ID_CMD;
2716 	smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
2717 
2718 	rv = wait_for_msg_done(smi_info);
2719 	if (rv)
2720 		goto out;
2721 
2722 	resp_len = smi_info->handlers->get_result(smi_info->si_sm,
2723 						  resp, IPMI_MAX_MSG_LENGTH);
2724 
2725 	/* Check and record info from the get device id, in case we need it. */
2726 	rv = ipmi_demangle_device_id(resp, resp_len, &smi_info->device_id);
2727 
2728  out:
2729 	kfree(resp);
2730 	return rv;
2731 }
2732 
2733 static int try_enable_event_buffer(struct smi_info *smi_info)
2734 {
2735 	unsigned char         msg[3];
2736 	unsigned char         *resp;
2737 	unsigned long         resp_len;
2738 	int                   rv = 0;
2739 
2740 	resp = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
2741 	if (!resp)
2742 		return -ENOMEM;
2743 
2744 	msg[0] = IPMI_NETFN_APP_REQUEST << 2;
2745 	msg[1] = IPMI_GET_BMC_GLOBAL_ENABLES_CMD;
2746 	smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
2747 
2748 	rv = wait_for_msg_done(smi_info);
2749 	if (rv) {
2750 		printk(KERN_WARNING PFX "Error getting response from get"
2751 		       " global enables command, the event buffer is not"
2752 		       " enabled.\n");
2753 		goto out;
2754 	}
2755 
2756 	resp_len = smi_info->handlers->get_result(smi_info->si_sm,
2757 						  resp, IPMI_MAX_MSG_LENGTH);
2758 
2759 	if (resp_len < 4 ||
2760 			resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
2761 			resp[1] != IPMI_GET_BMC_GLOBAL_ENABLES_CMD   ||
2762 			resp[2] != 0) {
2763 		printk(KERN_WARNING PFX "Invalid return from get global"
2764 		       " enables command, cannot enable the event buffer.\n");
2765 		rv = -EINVAL;
2766 		goto out;
2767 	}
2768 
2769 	if (resp[3] & IPMI_BMC_EVT_MSG_BUFF)
2770 		/* buffer is already enabled, nothing to do. */
2771 		goto out;
2772 
2773 	msg[0] = IPMI_NETFN_APP_REQUEST << 2;
2774 	msg[1] = IPMI_SET_BMC_GLOBAL_ENABLES_CMD;
2775 	msg[2] = resp[3] | IPMI_BMC_EVT_MSG_BUFF;
2776 	smi_info->handlers->start_transaction(smi_info->si_sm, msg, 3);
2777 
2778 	rv = wait_for_msg_done(smi_info);
2779 	if (rv) {
2780 		printk(KERN_WARNING PFX "Error getting response from set"
2781 		       " global, enables command, the event buffer is not"
2782 		       " enabled.\n");
2783 		goto out;
2784 	}
2785 
2786 	resp_len = smi_info->handlers->get_result(smi_info->si_sm,
2787 						  resp, IPMI_MAX_MSG_LENGTH);
2788 
2789 	if (resp_len < 3 ||
2790 			resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
2791 			resp[1] != IPMI_SET_BMC_GLOBAL_ENABLES_CMD) {
2792 		printk(KERN_WARNING PFX "Invalid return from get global,"
2793 		       "enables command, not enable the event buffer.\n");
2794 		rv = -EINVAL;
2795 		goto out;
2796 	}
2797 
2798 	if (resp[2] != 0)
2799 		/*
2800 		 * An error when setting the event buffer bit means
2801 		 * that the event buffer is not supported.
2802 		 */
2803 		rv = -ENOENT;
2804  out:
2805 	kfree(resp);
2806 	return rv;
2807 }
2808 
2809 static int smi_type_proc_show(struct seq_file *m, void *v)
2810 {
2811 	struct smi_info *smi = m->private;
2812 
2813 	return seq_printf(m, "%s\n", si_to_str[smi->si_type]);
2814 }
2815 
2816 static int smi_type_proc_open(struct inode *inode, struct file *file)
2817 {
2818 	return single_open(file, smi_type_proc_show, PDE(inode)->data);
2819 }
2820 
2821 static const struct file_operations smi_type_proc_ops = {
2822 	.open		= smi_type_proc_open,
2823 	.read		= seq_read,
2824 	.llseek		= seq_lseek,
2825 	.release	= single_release,
2826 };
2827 
2828 static int smi_si_stats_proc_show(struct seq_file *m, void *v)
2829 {
2830 	struct smi_info *smi = m->private;
2831 
2832 	seq_printf(m, "interrupts_enabled:    %d\n",
2833 		       smi->irq && !smi->interrupt_disabled);
2834 	seq_printf(m, "short_timeouts:        %u\n",
2835 		       smi_get_stat(smi, short_timeouts));
2836 	seq_printf(m, "long_timeouts:         %u\n",
2837 		       smi_get_stat(smi, long_timeouts));
2838 	seq_printf(m, "idles:                 %u\n",
2839 		       smi_get_stat(smi, idles));
2840 	seq_printf(m, "interrupts:            %u\n",
2841 		       smi_get_stat(smi, interrupts));
2842 	seq_printf(m, "attentions:            %u\n",
2843 		       smi_get_stat(smi, attentions));
2844 	seq_printf(m, "flag_fetches:          %u\n",
2845 		       smi_get_stat(smi, flag_fetches));
2846 	seq_printf(m, "hosed_count:           %u\n",
2847 		       smi_get_stat(smi, hosed_count));
2848 	seq_printf(m, "complete_transactions: %u\n",
2849 		       smi_get_stat(smi, complete_transactions));
2850 	seq_printf(m, "events:                %u\n",
2851 		       smi_get_stat(smi, events));
2852 	seq_printf(m, "watchdog_pretimeouts:  %u\n",
2853 		       smi_get_stat(smi, watchdog_pretimeouts));
2854 	seq_printf(m, "incoming_messages:     %u\n",
2855 		       smi_get_stat(smi, incoming_messages));
2856 	return 0;
2857 }
2858 
2859 static int smi_si_stats_proc_open(struct inode *inode, struct file *file)
2860 {
2861 	return single_open(file, smi_si_stats_proc_show, PDE(inode)->data);
2862 }
2863 
2864 static const struct file_operations smi_si_stats_proc_ops = {
2865 	.open		= smi_si_stats_proc_open,
2866 	.read		= seq_read,
2867 	.llseek		= seq_lseek,
2868 	.release	= single_release,
2869 };
2870 
2871 static int smi_params_proc_show(struct seq_file *m, void *v)
2872 {
2873 	struct smi_info *smi = m->private;
2874 
2875 	return seq_printf(m,
2876 		       "%s,%s,0x%lx,rsp=%d,rsi=%d,rsh=%d,irq=%d,ipmb=%d\n",
2877 		       si_to_str[smi->si_type],
2878 		       addr_space_to_str[smi->io.addr_type],
2879 		       smi->io.addr_data,
2880 		       smi->io.regspacing,
2881 		       smi->io.regsize,
2882 		       smi->io.regshift,
2883 		       smi->irq,
2884 		       smi->slave_addr);
2885 }
2886 
2887 static int smi_params_proc_open(struct inode *inode, struct file *file)
2888 {
2889 	return single_open(file, smi_params_proc_show, PDE(inode)->data);
2890 }
2891 
2892 static const struct file_operations smi_params_proc_ops = {
2893 	.open		= smi_params_proc_open,
2894 	.read		= seq_read,
2895 	.llseek		= seq_lseek,
2896 	.release	= single_release,
2897 };
2898 
2899 /*
2900  * oem_data_avail_to_receive_msg_avail
2901  * @info - smi_info structure with msg_flags set
2902  *
2903  * Converts flags from OEM_DATA_AVAIL to RECEIVE_MSG_AVAIL
2904  * Returns 1 indicating need to re-run handle_flags().
2905  */
2906 static int oem_data_avail_to_receive_msg_avail(struct smi_info *smi_info)
2907 {
2908 	smi_info->msg_flags = ((smi_info->msg_flags & ~OEM_DATA_AVAIL) |
2909 			       RECEIVE_MSG_AVAIL);
2910 	return 1;
2911 }
2912 
2913 /*
2914  * setup_dell_poweredge_oem_data_handler
2915  * @info - smi_info.device_id must be populated
2916  *
2917  * Systems that match, but have firmware version < 1.40 may assert
2918  * OEM0_DATA_AVAIL on their own, without being told via Set Flags that
2919  * it's safe to do so.  Such systems will de-assert OEM1_DATA_AVAIL
2920  * upon receipt of IPMI_GET_MSG_CMD, so we should treat these flags
2921  * as RECEIVE_MSG_AVAIL instead.
2922  *
2923  * As Dell has no plans to release IPMI 1.5 firmware that *ever*
2924  * assert the OEM[012] bits, and if it did, the driver would have to
2925  * change to handle that properly, we don't actually check for the
2926  * firmware version.
2927  * Device ID = 0x20                BMC on PowerEdge 8G servers
2928  * Device Revision = 0x80
2929  * Firmware Revision1 = 0x01       BMC version 1.40
2930  * Firmware Revision2 = 0x40       BCD encoded
2931  * IPMI Version = 0x51             IPMI 1.5
2932  * Manufacturer ID = A2 02 00      Dell IANA
2933  *
2934  * Additionally, PowerEdge systems with IPMI < 1.5 may also assert
2935  * OEM0_DATA_AVAIL and needs to be treated as RECEIVE_MSG_AVAIL.
2936  *
2937  */
2938 #define DELL_POWEREDGE_8G_BMC_DEVICE_ID  0x20
2939 #define DELL_POWEREDGE_8G_BMC_DEVICE_REV 0x80
2940 #define DELL_POWEREDGE_8G_BMC_IPMI_VERSION 0x51
2941 #define DELL_IANA_MFR_ID 0x0002a2
2942 static void setup_dell_poweredge_oem_data_handler(struct smi_info *smi_info)
2943 {
2944 	struct ipmi_device_id *id = &smi_info->device_id;
2945 	if (id->manufacturer_id == DELL_IANA_MFR_ID) {
2946 		if (id->device_id       == DELL_POWEREDGE_8G_BMC_DEVICE_ID  &&
2947 		    id->device_revision == DELL_POWEREDGE_8G_BMC_DEVICE_REV &&
2948 		    id->ipmi_version   == DELL_POWEREDGE_8G_BMC_IPMI_VERSION) {
2949 			smi_info->oem_data_avail_handler =
2950 				oem_data_avail_to_receive_msg_avail;
2951 		} else if (ipmi_version_major(id) < 1 ||
2952 			   (ipmi_version_major(id) == 1 &&
2953 			    ipmi_version_minor(id) < 5)) {
2954 			smi_info->oem_data_avail_handler =
2955 				oem_data_avail_to_receive_msg_avail;
2956 		}
2957 	}
2958 }
2959 
2960 #define CANNOT_RETURN_REQUESTED_LENGTH 0xCA
2961 static void return_hosed_msg_badsize(struct smi_info *smi_info)
2962 {
2963 	struct ipmi_smi_msg *msg = smi_info->curr_msg;
2964 
2965 	/* Make it a response */
2966 	msg->rsp[0] = msg->data[0] | 4;
2967 	msg->rsp[1] = msg->data[1];
2968 	msg->rsp[2] = CANNOT_RETURN_REQUESTED_LENGTH;
2969 	msg->rsp_size = 3;
2970 	smi_info->curr_msg = NULL;
2971 	deliver_recv_msg(smi_info, msg);
2972 }
2973 
2974 /*
2975  * dell_poweredge_bt_xaction_handler
2976  * @info - smi_info.device_id must be populated
2977  *
2978  * Dell PowerEdge servers with the BT interface (x6xx and 1750) will
2979  * not respond to a Get SDR command if the length of the data
2980  * requested is exactly 0x3A, which leads to command timeouts and no
2981  * data returned.  This intercepts such commands, and causes userspace
2982  * callers to try again with a different-sized buffer, which succeeds.
2983  */
2984 
2985 #define STORAGE_NETFN 0x0A
2986 #define STORAGE_CMD_GET_SDR 0x23
2987 static int dell_poweredge_bt_xaction_handler(struct notifier_block *self,
2988 					     unsigned long unused,
2989 					     void *in)
2990 {
2991 	struct smi_info *smi_info = in;
2992 	unsigned char *data = smi_info->curr_msg->data;
2993 	unsigned int size   = smi_info->curr_msg->data_size;
2994 	if (size >= 8 &&
2995 	    (data[0]>>2) == STORAGE_NETFN &&
2996 	    data[1] == STORAGE_CMD_GET_SDR &&
2997 	    data[7] == 0x3A) {
2998 		return_hosed_msg_badsize(smi_info);
2999 		return NOTIFY_STOP;
3000 	}
3001 	return NOTIFY_DONE;
3002 }
3003 
3004 static struct notifier_block dell_poweredge_bt_xaction_notifier = {
3005 	.notifier_call	= dell_poweredge_bt_xaction_handler,
3006 };
3007 
3008 /*
3009  * setup_dell_poweredge_bt_xaction_handler
3010  * @info - smi_info.device_id must be filled in already
3011  *
3012  * Fills in smi_info.device_id.start_transaction_pre_hook
3013  * when we know what function to use there.
3014  */
3015 static void
3016 setup_dell_poweredge_bt_xaction_handler(struct smi_info *smi_info)
3017 {
3018 	struct ipmi_device_id *id = &smi_info->device_id;
3019 	if (id->manufacturer_id == DELL_IANA_MFR_ID &&
3020 	    smi_info->si_type == SI_BT)
3021 		register_xaction_notifier(&dell_poweredge_bt_xaction_notifier);
3022 }
3023 
3024 /*
3025  * setup_oem_data_handler
3026  * @info - smi_info.device_id must be filled in already
3027  *
3028  * Fills in smi_info.device_id.oem_data_available_handler
3029  * when we know what function to use there.
3030  */
3031 
3032 static void setup_oem_data_handler(struct smi_info *smi_info)
3033 {
3034 	setup_dell_poweredge_oem_data_handler(smi_info);
3035 }
3036 
3037 static void setup_xaction_handlers(struct smi_info *smi_info)
3038 {
3039 	setup_dell_poweredge_bt_xaction_handler(smi_info);
3040 }
3041 
3042 static inline void wait_for_timer_and_thread(struct smi_info *smi_info)
3043 {
3044 	if (smi_info->intf) {
3045 		/*
3046 		 * The timer and thread are only running if the
3047 		 * interface has been started up and registered.
3048 		 */
3049 		if (smi_info->thread != NULL)
3050 			kthread_stop(smi_info->thread);
3051 		del_timer_sync(&smi_info->si_timer);
3052 	}
3053 }
3054 
3055 static __devinitdata struct ipmi_default_vals
3056 {
3057 	int type;
3058 	int port;
3059 } ipmi_defaults[] =
3060 {
3061 	{ .type = SI_KCS, .port = 0xca2 },
3062 	{ .type = SI_SMIC, .port = 0xca9 },
3063 	{ .type = SI_BT, .port = 0xe4 },
3064 	{ .port = 0 }
3065 };
3066 
3067 static void __devinit default_find_bmc(void)
3068 {
3069 	struct smi_info *info;
3070 	int             i;
3071 
3072 	for (i = 0; ; i++) {
3073 		if (!ipmi_defaults[i].port)
3074 			break;
3075 #ifdef CONFIG_PPC
3076 		if (check_legacy_ioport(ipmi_defaults[i].port))
3077 			continue;
3078 #endif
3079 		info = smi_info_alloc();
3080 		if (!info)
3081 			return;
3082 
3083 		info->addr_source = SI_DEFAULT;
3084 
3085 		info->si_type = ipmi_defaults[i].type;
3086 		info->io_setup = port_setup;
3087 		info->io.addr_data = ipmi_defaults[i].port;
3088 		info->io.addr_type = IPMI_IO_ADDR_SPACE;
3089 
3090 		info->io.addr = NULL;
3091 		info->io.regspacing = DEFAULT_REGSPACING;
3092 		info->io.regsize = DEFAULT_REGSPACING;
3093 		info->io.regshift = 0;
3094 
3095 		if (add_smi(info) == 0) {
3096 			if ((try_smi_init(info)) == 0) {
3097 				/* Found one... */
3098 				printk(KERN_INFO PFX "Found default %s"
3099 				" state machine at %s address 0x%lx\n",
3100 				si_to_str[info->si_type],
3101 				addr_space_to_str[info->io.addr_type],
3102 				info->io.addr_data);
3103 			} else
3104 				cleanup_one_si(info);
3105 		} else {
3106 			kfree(info);
3107 		}
3108 	}
3109 }
3110 
3111 static int is_new_interface(struct smi_info *info)
3112 {
3113 	struct smi_info *e;
3114 
3115 	list_for_each_entry(e, &smi_infos, link) {
3116 		if (e->io.addr_type != info->io.addr_type)
3117 			continue;
3118 		if (e->io.addr_data == info->io.addr_data)
3119 			return 0;
3120 	}
3121 
3122 	return 1;
3123 }
3124 
3125 static int add_smi(struct smi_info *new_smi)
3126 {
3127 	int rv = 0;
3128 
3129 	printk(KERN_INFO PFX "Adding %s-specified %s state machine",
3130 			ipmi_addr_src_to_str[new_smi->addr_source],
3131 			si_to_str[new_smi->si_type]);
3132 	mutex_lock(&smi_infos_lock);
3133 	if (!is_new_interface(new_smi)) {
3134 		printk(KERN_CONT " duplicate interface\n");
3135 		rv = -EBUSY;
3136 		goto out_err;
3137 	}
3138 
3139 	printk(KERN_CONT "\n");
3140 
3141 	/* So we know not to free it unless we have allocated one. */
3142 	new_smi->intf = NULL;
3143 	new_smi->si_sm = NULL;
3144 	new_smi->handlers = NULL;
3145 
3146 	list_add_tail(&new_smi->link, &smi_infos);
3147 
3148 out_err:
3149 	mutex_unlock(&smi_infos_lock);
3150 	return rv;
3151 }
3152 
3153 static int try_smi_init(struct smi_info *new_smi)
3154 {
3155 	int rv = 0;
3156 	int i;
3157 
3158 	printk(KERN_INFO PFX "Trying %s-specified %s state"
3159 	       " machine at %s address 0x%lx, slave address 0x%x,"
3160 	       " irq %d\n",
3161 	       ipmi_addr_src_to_str[new_smi->addr_source],
3162 	       si_to_str[new_smi->si_type],
3163 	       addr_space_to_str[new_smi->io.addr_type],
3164 	       new_smi->io.addr_data,
3165 	       new_smi->slave_addr, new_smi->irq);
3166 
3167 	switch (new_smi->si_type) {
3168 	case SI_KCS:
3169 		new_smi->handlers = &kcs_smi_handlers;
3170 		break;
3171 
3172 	case SI_SMIC:
3173 		new_smi->handlers = &smic_smi_handlers;
3174 		break;
3175 
3176 	case SI_BT:
3177 		new_smi->handlers = &bt_smi_handlers;
3178 		break;
3179 
3180 	default:
3181 		/* No support for anything else yet. */
3182 		rv = -EIO;
3183 		goto out_err;
3184 	}
3185 
3186 	/* Allocate the state machine's data and initialize it. */
3187 	new_smi->si_sm = kmalloc(new_smi->handlers->size(), GFP_KERNEL);
3188 	if (!new_smi->si_sm) {
3189 		printk(KERN_ERR PFX
3190 		       "Could not allocate state machine memory\n");
3191 		rv = -ENOMEM;
3192 		goto out_err;
3193 	}
3194 	new_smi->io_size = new_smi->handlers->init_data(new_smi->si_sm,
3195 							&new_smi->io);
3196 
3197 	/* Now that we know the I/O size, we can set up the I/O. */
3198 	rv = new_smi->io_setup(new_smi);
3199 	if (rv) {
3200 		printk(KERN_ERR PFX "Could not set up I/O space\n");
3201 		goto out_err;
3202 	}
3203 
3204 	/* Do low-level detection first. */
3205 	if (new_smi->handlers->detect(new_smi->si_sm)) {
3206 		if (new_smi->addr_source)
3207 			printk(KERN_INFO PFX "Interface detection failed\n");
3208 		rv = -ENODEV;
3209 		goto out_err;
3210 	}
3211 
3212 	/*
3213 	 * Attempt a get device id command.  If it fails, we probably
3214 	 * don't have a BMC here.
3215 	 */
3216 	rv = try_get_dev_id(new_smi);
3217 	if (rv) {
3218 		if (new_smi->addr_source)
3219 			printk(KERN_INFO PFX "There appears to be no BMC"
3220 			       " at this location\n");
3221 		goto out_err;
3222 	}
3223 
3224 	setup_oem_data_handler(new_smi);
3225 	setup_xaction_handlers(new_smi);
3226 
3227 	INIT_LIST_HEAD(&(new_smi->xmit_msgs));
3228 	INIT_LIST_HEAD(&(new_smi->hp_xmit_msgs));
3229 	new_smi->curr_msg = NULL;
3230 	atomic_set(&new_smi->req_events, 0);
3231 	new_smi->run_to_completion = 0;
3232 	for (i = 0; i < SI_NUM_STATS; i++)
3233 		atomic_set(&new_smi->stats[i], 0);
3234 
3235 	new_smi->interrupt_disabled = 1;
3236 	atomic_set(&new_smi->stop_operation, 0);
3237 	new_smi->intf_num = smi_num;
3238 	smi_num++;
3239 
3240 	rv = try_enable_event_buffer(new_smi);
3241 	if (rv == 0)
3242 		new_smi->has_event_buffer = 1;
3243 
3244 	/*
3245 	 * Start clearing the flags before we enable interrupts or the
3246 	 * timer to avoid racing with the timer.
3247 	 */
3248 	start_clear_flags(new_smi);
3249 	/* IRQ is defined to be set when non-zero. */
3250 	if (new_smi->irq)
3251 		new_smi->si_state = SI_CLEARING_FLAGS_THEN_SET_IRQ;
3252 
3253 	if (!new_smi->dev) {
3254 		/*
3255 		 * If we don't already have a device from something
3256 		 * else (like PCI), then register a new one.
3257 		 */
3258 		new_smi->pdev = platform_device_alloc("ipmi_si",
3259 						      new_smi->intf_num);
3260 		if (!new_smi->pdev) {
3261 			printk(KERN_ERR PFX
3262 			       "Unable to allocate platform device\n");
3263 			goto out_err;
3264 		}
3265 		new_smi->dev = &new_smi->pdev->dev;
3266 		new_smi->dev->driver = &ipmi_driver.driver;
3267 
3268 		rv = platform_device_add(new_smi->pdev);
3269 		if (rv) {
3270 			printk(KERN_ERR PFX
3271 			       "Unable to register system interface device:"
3272 			       " %d\n",
3273 			       rv);
3274 			goto out_err;
3275 		}
3276 		new_smi->dev_registered = 1;
3277 	}
3278 
3279 	rv = ipmi_register_smi(&handlers,
3280 			       new_smi,
3281 			       &new_smi->device_id,
3282 			       new_smi->dev,
3283 			       "bmc",
3284 			       new_smi->slave_addr);
3285 	if (rv) {
3286 		dev_err(new_smi->dev, "Unable to register device: error %d\n",
3287 			rv);
3288 		goto out_err_stop_timer;
3289 	}
3290 
3291 	rv = ipmi_smi_add_proc_entry(new_smi->intf, "type",
3292 				     &smi_type_proc_ops,
3293 				     new_smi);
3294 	if (rv) {
3295 		dev_err(new_smi->dev, "Unable to create proc entry: %d\n", rv);
3296 		goto out_err_stop_timer;
3297 	}
3298 
3299 	rv = ipmi_smi_add_proc_entry(new_smi->intf, "si_stats",
3300 				     &smi_si_stats_proc_ops,
3301 				     new_smi);
3302 	if (rv) {
3303 		dev_err(new_smi->dev, "Unable to create proc entry: %d\n", rv);
3304 		goto out_err_stop_timer;
3305 	}
3306 
3307 	rv = ipmi_smi_add_proc_entry(new_smi->intf, "params",
3308 				     &smi_params_proc_ops,
3309 				     new_smi);
3310 	if (rv) {
3311 		dev_err(new_smi->dev, "Unable to create proc entry: %d\n", rv);
3312 		goto out_err_stop_timer;
3313 	}
3314 
3315 	dev_info(new_smi->dev, "IPMI %s interface initialized\n",
3316 		 si_to_str[new_smi->si_type]);
3317 
3318 	return 0;
3319 
3320  out_err_stop_timer:
3321 	atomic_inc(&new_smi->stop_operation);
3322 	wait_for_timer_and_thread(new_smi);
3323 
3324  out_err:
3325 	new_smi->interrupt_disabled = 1;
3326 
3327 	if (new_smi->intf) {
3328 		ipmi_unregister_smi(new_smi->intf);
3329 		new_smi->intf = NULL;
3330 	}
3331 
3332 	if (new_smi->irq_cleanup) {
3333 		new_smi->irq_cleanup(new_smi);
3334 		new_smi->irq_cleanup = NULL;
3335 	}
3336 
3337 	/*
3338 	 * Wait until we know that we are out of any interrupt
3339 	 * handlers might have been running before we freed the
3340 	 * interrupt.
3341 	 */
3342 	synchronize_sched();
3343 
3344 	if (new_smi->si_sm) {
3345 		if (new_smi->handlers)
3346 			new_smi->handlers->cleanup(new_smi->si_sm);
3347 		kfree(new_smi->si_sm);
3348 		new_smi->si_sm = NULL;
3349 	}
3350 	if (new_smi->addr_source_cleanup) {
3351 		new_smi->addr_source_cleanup(new_smi);
3352 		new_smi->addr_source_cleanup = NULL;
3353 	}
3354 	if (new_smi->io_cleanup) {
3355 		new_smi->io_cleanup(new_smi);
3356 		new_smi->io_cleanup = NULL;
3357 	}
3358 
3359 	if (new_smi->dev_registered) {
3360 		platform_device_unregister(new_smi->pdev);
3361 		new_smi->dev_registered = 0;
3362 	}
3363 
3364 	return rv;
3365 }
3366 
3367 static int __devinit init_ipmi_si(void)
3368 {
3369 	int  i;
3370 	char *str;
3371 	int  rv;
3372 	struct smi_info *e;
3373 	enum ipmi_addr_src type = SI_INVALID;
3374 
3375 	if (initialized)
3376 		return 0;
3377 	initialized = 1;
3378 
3379 	rv = platform_driver_register(&ipmi_driver);
3380 	if (rv) {
3381 		printk(KERN_ERR PFX "Unable to register driver: %d\n", rv);
3382 		return rv;
3383 	}
3384 
3385 
3386 	/* Parse out the si_type string into its components. */
3387 	str = si_type_str;
3388 	if (*str != '\0') {
3389 		for (i = 0; (i < SI_MAX_PARMS) && (*str != '\0'); i++) {
3390 			si_type[i] = str;
3391 			str = strchr(str, ',');
3392 			if (str) {
3393 				*str = '\0';
3394 				str++;
3395 			} else {
3396 				break;
3397 			}
3398 		}
3399 	}
3400 
3401 	printk(KERN_INFO "IPMI System Interface driver.\n");
3402 
3403 	/* If the user gave us a device, they presumably want us to use it */
3404 	if (!hardcode_find_bmc())
3405 		return 0;
3406 
3407 #ifdef CONFIG_PCI
3408 	rv = pci_register_driver(&ipmi_pci_driver);
3409 	if (rv)
3410 		printk(KERN_ERR PFX "Unable to register PCI driver: %d\n", rv);
3411 	else
3412 		pci_registered = 1;
3413 #endif
3414 
3415 #ifdef CONFIG_ACPI
3416 	pnp_register_driver(&ipmi_pnp_driver);
3417 	pnp_registered = 1;
3418 #endif
3419 
3420 #ifdef CONFIG_DMI
3421 	dmi_find_bmc();
3422 #endif
3423 
3424 #ifdef CONFIG_ACPI
3425 	spmi_find_bmc();
3426 #endif
3427 
3428 	/* We prefer devices with interrupts, but in the case of a machine
3429 	   with multiple BMCs we assume that there will be several instances
3430 	   of a given type so if we succeed in registering a type then also
3431 	   try to register everything else of the same type */
3432 
3433 	mutex_lock(&smi_infos_lock);
3434 	list_for_each_entry(e, &smi_infos, link) {
3435 		/* Try to register a device if it has an IRQ and we either
3436 		   haven't successfully registered a device yet or this
3437 		   device has the same type as one we successfully registered */
3438 		if (e->irq && (!type || e->addr_source == type)) {
3439 			if (!try_smi_init(e)) {
3440 				type = e->addr_source;
3441 			}
3442 		}
3443 	}
3444 
3445 	/* type will only have been set if we successfully registered an si */
3446 	if (type) {
3447 		mutex_unlock(&smi_infos_lock);
3448 		return 0;
3449 	}
3450 
3451 	/* Fall back to the preferred device */
3452 
3453 	list_for_each_entry(e, &smi_infos, link) {
3454 		if (!e->irq && (!type || e->addr_source == type)) {
3455 			if (!try_smi_init(e)) {
3456 				type = e->addr_source;
3457 			}
3458 		}
3459 	}
3460 	mutex_unlock(&smi_infos_lock);
3461 
3462 	if (type)
3463 		return 0;
3464 
3465 	if (si_trydefaults) {
3466 		mutex_lock(&smi_infos_lock);
3467 		if (list_empty(&smi_infos)) {
3468 			/* No BMC was found, try defaults. */
3469 			mutex_unlock(&smi_infos_lock);
3470 			default_find_bmc();
3471 		} else
3472 			mutex_unlock(&smi_infos_lock);
3473 	}
3474 
3475 	mutex_lock(&smi_infos_lock);
3476 	if (unload_when_empty && list_empty(&smi_infos)) {
3477 		mutex_unlock(&smi_infos_lock);
3478 		cleanup_ipmi_si();
3479 		printk(KERN_WARNING PFX
3480 		       "Unable to find any System Interface(s)\n");
3481 		return -ENODEV;
3482 	} else {
3483 		mutex_unlock(&smi_infos_lock);
3484 		return 0;
3485 	}
3486 }
3487 module_init(init_ipmi_si);
3488 
3489 static void cleanup_one_si(struct smi_info *to_clean)
3490 {
3491 	int           rv = 0;
3492 	unsigned long flags;
3493 
3494 	if (!to_clean)
3495 		return;
3496 
3497 	list_del(&to_clean->link);
3498 
3499 	/* Tell the driver that we are shutting down. */
3500 	atomic_inc(&to_clean->stop_operation);
3501 
3502 	/*
3503 	 * Make sure the timer and thread are stopped and will not run
3504 	 * again.
3505 	 */
3506 	wait_for_timer_and_thread(to_clean);
3507 
3508 	/*
3509 	 * Timeouts are stopped, now make sure the interrupts are off
3510 	 * for the device.  A little tricky with locks to make sure
3511 	 * there are no races.
3512 	 */
3513 	spin_lock_irqsave(&to_clean->si_lock, flags);
3514 	while (to_clean->curr_msg || (to_clean->si_state != SI_NORMAL)) {
3515 		spin_unlock_irqrestore(&to_clean->si_lock, flags);
3516 		poll(to_clean);
3517 		schedule_timeout_uninterruptible(1);
3518 		spin_lock_irqsave(&to_clean->si_lock, flags);
3519 	}
3520 	disable_si_irq(to_clean);
3521 	spin_unlock_irqrestore(&to_clean->si_lock, flags);
3522 	while (to_clean->curr_msg || (to_clean->si_state != SI_NORMAL)) {
3523 		poll(to_clean);
3524 		schedule_timeout_uninterruptible(1);
3525 	}
3526 
3527 	/* Clean up interrupts and make sure that everything is done. */
3528 	if (to_clean->irq_cleanup)
3529 		to_clean->irq_cleanup(to_clean);
3530 	while (to_clean->curr_msg || (to_clean->si_state != SI_NORMAL)) {
3531 		poll(to_clean);
3532 		schedule_timeout_uninterruptible(1);
3533 	}
3534 
3535 	if (to_clean->intf)
3536 		rv = ipmi_unregister_smi(to_clean->intf);
3537 
3538 	if (rv) {
3539 		printk(KERN_ERR PFX "Unable to unregister device: errno=%d\n",
3540 		       rv);
3541 	}
3542 
3543 	if (to_clean->handlers)
3544 		to_clean->handlers->cleanup(to_clean->si_sm);
3545 
3546 	kfree(to_clean->si_sm);
3547 
3548 	if (to_clean->addr_source_cleanup)
3549 		to_clean->addr_source_cleanup(to_clean);
3550 	if (to_clean->io_cleanup)
3551 		to_clean->io_cleanup(to_clean);
3552 
3553 	if (to_clean->dev_registered)
3554 		platform_device_unregister(to_clean->pdev);
3555 
3556 	kfree(to_clean);
3557 }
3558 
3559 static void cleanup_ipmi_si(void)
3560 {
3561 	struct smi_info *e, *tmp_e;
3562 
3563 	if (!initialized)
3564 		return;
3565 
3566 #ifdef CONFIG_PCI
3567 	if (pci_registered)
3568 		pci_unregister_driver(&ipmi_pci_driver);
3569 #endif
3570 #ifdef CONFIG_ACPI
3571 	if (pnp_registered)
3572 		pnp_unregister_driver(&ipmi_pnp_driver);
3573 #endif
3574 
3575 	platform_driver_unregister(&ipmi_driver);
3576 
3577 	mutex_lock(&smi_infos_lock);
3578 	list_for_each_entry_safe(e, tmp_e, &smi_infos, link)
3579 		cleanup_one_si(e);
3580 	mutex_unlock(&smi_infos_lock);
3581 }
3582 module_exit(cleanup_ipmi_si);
3583 
3584 MODULE_LICENSE("GPL");
3585 MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>");
3586 MODULE_DESCRIPTION("Interface to the IPMI driver for the KCS, SMIC, and BT"
3587 		   " system interfaces.");
3588