Merge branch 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
[sfrench/cifs-2.6.git] / drivers / char / ipmi / ipmi_si_intf.c
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 <linux/sched.h>
45 #include <linux/seq_file.h>
46 #include <linux/timer.h>
47 #include <linux/errno.h>
48 #include <linux/spinlock.h>
49 #include <linux/slab.h>
50 #include <linux/delay.h>
51 #include <linux/list.h>
52 #include <linux/notifier.h>
53 #include <linux/mutex.h>
54 #include <linux/kthread.h>
55 #include <asm/irq.h>
56 #include <linux/interrupt.h>
57 #include <linux/rcupdate.h>
58 #include <linux/ipmi.h>
59 #include <linux/ipmi_smi.h>
60 #include "ipmi_si.h"
61 #include <linux/string.h>
62 #include <linux/ctype.h>
63
64 #define PFX "ipmi_si: "
65
66 /* Measure times between events in the driver. */
67 #undef DEBUG_TIMING
68
69 /* Call every 10 ms. */
70 #define SI_TIMEOUT_TIME_USEC    10000
71 #define SI_USEC_PER_JIFFY       (1000000/HZ)
72 #define SI_TIMEOUT_JIFFIES      (SI_TIMEOUT_TIME_USEC/SI_USEC_PER_JIFFY)
73 #define SI_SHORT_TIMEOUT_USEC  250 /* .25ms when the SM request a
74                                       short timeout */
75
76 enum si_intf_state {
77         SI_NORMAL,
78         SI_GETTING_FLAGS,
79         SI_GETTING_EVENTS,
80         SI_CLEARING_FLAGS,
81         SI_GETTING_MESSAGES,
82         SI_CHECKING_ENABLES,
83         SI_SETTING_ENABLES
84         /* FIXME - add watchdog stuff. */
85 };
86
87 /* Some BT-specific defines we need here. */
88 #define IPMI_BT_INTMASK_REG             2
89 #define IPMI_BT_INTMASK_CLEAR_IRQ_BIT   2
90 #define IPMI_BT_INTMASK_ENABLE_IRQ_BIT  1
91
92 static const char * const si_to_str[] = { "invalid", "kcs", "smic", "bt" };
93
94 static int initialized;
95
96 /*
97  * Indexes into stats[] in smi_info below.
98  */
99 enum si_stat_indexes {
100         /*
101          * Number of times the driver requested a timer while an operation
102          * was in progress.
103          */
104         SI_STAT_short_timeouts = 0,
105
106         /*
107          * Number of times the driver requested a timer while nothing was in
108          * progress.
109          */
110         SI_STAT_long_timeouts,
111
112         /* Number of times the interface was idle while being polled. */
113         SI_STAT_idles,
114
115         /* Number of interrupts the driver handled. */
116         SI_STAT_interrupts,
117
118         /* Number of time the driver got an ATTN from the hardware. */
119         SI_STAT_attentions,
120
121         /* Number of times the driver requested flags from the hardware. */
122         SI_STAT_flag_fetches,
123
124         /* Number of times the hardware didn't follow the state machine. */
125         SI_STAT_hosed_count,
126
127         /* Number of completed messages. */
128         SI_STAT_complete_transactions,
129
130         /* Number of IPMI events received from the hardware. */
131         SI_STAT_events,
132
133         /* Number of watchdog pretimeouts. */
134         SI_STAT_watchdog_pretimeouts,
135
136         /* Number of asynchronous messages received. */
137         SI_STAT_incoming_messages,
138
139
140         /* This *must* remain last, add new values above this. */
141         SI_NUM_STATS
142 };
143
144 struct smi_info {
145         int                    intf_num;
146         ipmi_smi_t             intf;
147         struct si_sm_data      *si_sm;
148         const struct si_sm_handlers *handlers;
149         spinlock_t             si_lock;
150         struct ipmi_smi_msg    *waiting_msg;
151         struct ipmi_smi_msg    *curr_msg;
152         enum si_intf_state     si_state;
153
154         /*
155          * Used to handle the various types of I/O that can occur with
156          * IPMI
157          */
158         struct si_sm_io io;
159
160         /*
161          * Per-OEM handler, called from handle_flags().  Returns 1
162          * when handle_flags() needs to be re-run or 0 indicating it
163          * set si_state itself.
164          */
165         int (*oem_data_avail_handler)(struct smi_info *smi_info);
166
167         /*
168          * Flags from the last GET_MSG_FLAGS command, used when an ATTN
169          * is set to hold the flags until we are done handling everything
170          * from the flags.
171          */
172 #define RECEIVE_MSG_AVAIL       0x01
173 #define EVENT_MSG_BUFFER_FULL   0x02
174 #define WDT_PRE_TIMEOUT_INT     0x08
175 #define OEM0_DATA_AVAIL     0x20
176 #define OEM1_DATA_AVAIL     0x40
177 #define OEM2_DATA_AVAIL     0x80
178 #define OEM_DATA_AVAIL      (OEM0_DATA_AVAIL | \
179                              OEM1_DATA_AVAIL | \
180                              OEM2_DATA_AVAIL)
181         unsigned char       msg_flags;
182
183         /* Does the BMC have an event buffer? */
184         bool                has_event_buffer;
185
186         /*
187          * If set to true, this will request events the next time the
188          * state machine is idle.
189          */
190         atomic_t            req_events;
191
192         /*
193          * If true, run the state machine to completion on every send
194          * call.  Generally used after a panic to make sure stuff goes
195          * out.
196          */
197         bool                run_to_completion;
198
199         /* The timer for this si. */
200         struct timer_list   si_timer;
201
202         /* This flag is set, if the timer can be set */
203         bool                timer_can_start;
204
205         /* This flag is set, if the timer is running (timer_pending() isn't enough) */
206         bool                timer_running;
207
208         /* The time (in jiffies) the last timeout occurred at. */
209         unsigned long       last_timeout_jiffies;
210
211         /* Are we waiting for the events, pretimeouts, received msgs? */
212         atomic_t            need_watch;
213
214         /*
215          * The driver will disable interrupts when it gets into a
216          * situation where it cannot handle messages due to lack of
217          * memory.  Once that situation clears up, it will re-enable
218          * interrupts.
219          */
220         bool interrupt_disabled;
221
222         /*
223          * Does the BMC support events?
224          */
225         bool supports_event_msg_buff;
226
227         /*
228          * Can we disable interrupts the global enables receive irq
229          * bit?  There are currently two forms of brokenness, some
230          * systems cannot disable the bit (which is technically within
231          * the spec but a bad idea) and some systems have the bit
232          * forced to zero even though interrupts work (which is
233          * clearly outside the spec).  The next bool tells which form
234          * of brokenness is present.
235          */
236         bool cannot_disable_irq;
237
238         /*
239          * Some systems are broken and cannot set the irq enable
240          * bit, even if they support interrupts.
241          */
242         bool irq_enable_broken;
243
244         /*
245          * Did we get an attention that we did not handle?
246          */
247         bool got_attn;
248
249         /* From the get device id response... */
250         struct ipmi_device_id device_id;
251
252         /* Default driver model device. */
253         struct platform_device *pdev;
254
255         /* Counters and things for the proc filesystem. */
256         atomic_t stats[SI_NUM_STATS];
257
258         struct task_struct *thread;
259
260         struct list_head link;
261 };
262
263 #define smi_inc_stat(smi, stat) \
264         atomic_inc(&(smi)->stats[SI_STAT_ ## stat])
265 #define smi_get_stat(smi, stat) \
266         ((unsigned int) atomic_read(&(smi)->stats[SI_STAT_ ## stat]))
267
268 #define IPMI_MAX_INTFS 4
269 static int force_kipmid[IPMI_MAX_INTFS];
270 static int num_force_kipmid;
271
272 static unsigned int kipmid_max_busy_us[IPMI_MAX_INTFS];
273 static int num_max_busy_us;
274
275 static bool unload_when_empty = true;
276
277 static int try_smi_init(struct smi_info *smi);
278 static void cleanup_one_si(struct smi_info *to_clean);
279 static void cleanup_ipmi_si(void);
280
281 #ifdef DEBUG_TIMING
282 void debug_timestamp(char *msg)
283 {
284         struct timespec64 t;
285
286         getnstimeofday64(&t);
287         pr_debug("**%s: %lld.%9.9ld\n", msg, (long long) t.tv_sec, t.tv_nsec);
288 }
289 #else
290 #define debug_timestamp(x)
291 #endif
292
293 static ATOMIC_NOTIFIER_HEAD(xaction_notifier_list);
294 static int register_xaction_notifier(struct notifier_block *nb)
295 {
296         return atomic_notifier_chain_register(&xaction_notifier_list, nb);
297 }
298
299 static void deliver_recv_msg(struct smi_info *smi_info,
300                              struct ipmi_smi_msg *msg)
301 {
302         /* Deliver the message to the upper layer. */
303         if (smi_info->intf)
304                 ipmi_smi_msg_received(smi_info->intf, msg);
305         else
306                 ipmi_free_smi_msg(msg);
307 }
308
309 static void return_hosed_msg(struct smi_info *smi_info, int cCode)
310 {
311         struct ipmi_smi_msg *msg = smi_info->curr_msg;
312
313         if (cCode < 0 || cCode > IPMI_ERR_UNSPECIFIED)
314                 cCode = IPMI_ERR_UNSPECIFIED;
315         /* else use it as is */
316
317         /* Make it a response */
318         msg->rsp[0] = msg->data[0] | 4;
319         msg->rsp[1] = msg->data[1];
320         msg->rsp[2] = cCode;
321         msg->rsp_size = 3;
322
323         smi_info->curr_msg = NULL;
324         deliver_recv_msg(smi_info, msg);
325 }
326
327 static enum si_sm_result start_next_msg(struct smi_info *smi_info)
328 {
329         int              rv;
330
331         if (!smi_info->waiting_msg) {
332                 smi_info->curr_msg = NULL;
333                 rv = SI_SM_IDLE;
334         } else {
335                 int err;
336
337                 smi_info->curr_msg = smi_info->waiting_msg;
338                 smi_info->waiting_msg = NULL;
339                 debug_timestamp("Start2");
340                 err = atomic_notifier_call_chain(&xaction_notifier_list,
341                                 0, smi_info);
342                 if (err & NOTIFY_STOP_MASK) {
343                         rv = SI_SM_CALL_WITHOUT_DELAY;
344                         goto out;
345                 }
346                 err = smi_info->handlers->start_transaction(
347                         smi_info->si_sm,
348                         smi_info->curr_msg->data,
349                         smi_info->curr_msg->data_size);
350                 if (err)
351                         return_hosed_msg(smi_info, err);
352
353                 rv = SI_SM_CALL_WITHOUT_DELAY;
354         }
355 out:
356         return rv;
357 }
358
359 static void smi_mod_timer(struct smi_info *smi_info, unsigned long new_val)
360 {
361         if (!smi_info->timer_can_start)
362                 return;
363         smi_info->last_timeout_jiffies = jiffies;
364         mod_timer(&smi_info->si_timer, new_val);
365         smi_info->timer_running = true;
366 }
367
368 /*
369  * Start a new message and (re)start the timer and thread.
370  */
371 static void start_new_msg(struct smi_info *smi_info, unsigned char *msg,
372                           unsigned int size)
373 {
374         smi_mod_timer(smi_info, jiffies + SI_TIMEOUT_JIFFIES);
375
376         if (smi_info->thread)
377                 wake_up_process(smi_info->thread);
378
379         smi_info->handlers->start_transaction(smi_info->si_sm, msg, size);
380 }
381
382 static void start_check_enables(struct smi_info *smi_info)
383 {
384         unsigned char msg[2];
385
386         msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
387         msg[1] = IPMI_GET_BMC_GLOBAL_ENABLES_CMD;
388
389         start_new_msg(smi_info, msg, 2);
390         smi_info->si_state = SI_CHECKING_ENABLES;
391 }
392
393 static void start_clear_flags(struct smi_info *smi_info)
394 {
395         unsigned char msg[3];
396
397         /* Make sure the watchdog pre-timeout flag is not set at startup. */
398         msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
399         msg[1] = IPMI_CLEAR_MSG_FLAGS_CMD;
400         msg[2] = WDT_PRE_TIMEOUT_INT;
401
402         start_new_msg(smi_info, msg, 3);
403         smi_info->si_state = SI_CLEARING_FLAGS;
404 }
405
406 static void start_getting_msg_queue(struct smi_info *smi_info)
407 {
408         smi_info->curr_msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
409         smi_info->curr_msg->data[1] = IPMI_GET_MSG_CMD;
410         smi_info->curr_msg->data_size = 2;
411
412         start_new_msg(smi_info, smi_info->curr_msg->data,
413                       smi_info->curr_msg->data_size);
414         smi_info->si_state = SI_GETTING_MESSAGES;
415 }
416
417 static void start_getting_events(struct smi_info *smi_info)
418 {
419         smi_info->curr_msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
420         smi_info->curr_msg->data[1] = IPMI_READ_EVENT_MSG_BUFFER_CMD;
421         smi_info->curr_msg->data_size = 2;
422
423         start_new_msg(smi_info, smi_info->curr_msg->data,
424                       smi_info->curr_msg->data_size);
425         smi_info->si_state = SI_GETTING_EVENTS;
426 }
427
428 /*
429  * When we have a situtaion where we run out of memory and cannot
430  * allocate messages, we just leave them in the BMC and run the system
431  * polled until we can allocate some memory.  Once we have some
432  * memory, we will re-enable the interrupt.
433  *
434  * Note that we cannot just use disable_irq(), since the interrupt may
435  * be shared.
436  */
437 static inline bool disable_si_irq(struct smi_info *smi_info)
438 {
439         if ((smi_info->io.irq) && (!smi_info->interrupt_disabled)) {
440                 smi_info->interrupt_disabled = true;
441                 start_check_enables(smi_info);
442                 return true;
443         }
444         return false;
445 }
446
447 static inline bool enable_si_irq(struct smi_info *smi_info)
448 {
449         if ((smi_info->io.irq) && (smi_info->interrupt_disabled)) {
450                 smi_info->interrupt_disabled = false;
451                 start_check_enables(smi_info);
452                 return true;
453         }
454         return false;
455 }
456
457 /*
458  * Allocate a message.  If unable to allocate, start the interrupt
459  * disable process and return NULL.  If able to allocate but
460  * interrupts are disabled, free the message and return NULL after
461  * starting the interrupt enable process.
462  */
463 static struct ipmi_smi_msg *alloc_msg_handle_irq(struct smi_info *smi_info)
464 {
465         struct ipmi_smi_msg *msg;
466
467         msg = ipmi_alloc_smi_msg();
468         if (!msg) {
469                 if (!disable_si_irq(smi_info))
470                         smi_info->si_state = SI_NORMAL;
471         } else if (enable_si_irq(smi_info)) {
472                 ipmi_free_smi_msg(msg);
473                 msg = NULL;
474         }
475         return msg;
476 }
477
478 static void handle_flags(struct smi_info *smi_info)
479 {
480 retry:
481         if (smi_info->msg_flags & WDT_PRE_TIMEOUT_INT) {
482                 /* Watchdog pre-timeout */
483                 smi_inc_stat(smi_info, watchdog_pretimeouts);
484
485                 start_clear_flags(smi_info);
486                 smi_info->msg_flags &= ~WDT_PRE_TIMEOUT_INT;
487                 if (smi_info->intf)
488                         ipmi_smi_watchdog_pretimeout(smi_info->intf);
489         } else if (smi_info->msg_flags & RECEIVE_MSG_AVAIL) {
490                 /* Messages available. */
491                 smi_info->curr_msg = alloc_msg_handle_irq(smi_info);
492                 if (!smi_info->curr_msg)
493                         return;
494
495                 start_getting_msg_queue(smi_info);
496         } else if (smi_info->msg_flags & EVENT_MSG_BUFFER_FULL) {
497                 /* Events available. */
498                 smi_info->curr_msg = alloc_msg_handle_irq(smi_info);
499                 if (!smi_info->curr_msg)
500                         return;
501
502                 start_getting_events(smi_info);
503         } else if (smi_info->msg_flags & OEM_DATA_AVAIL &&
504                    smi_info->oem_data_avail_handler) {
505                 if (smi_info->oem_data_avail_handler(smi_info))
506                         goto retry;
507         } else
508                 smi_info->si_state = SI_NORMAL;
509 }
510
511 /*
512  * Global enables we care about.
513  */
514 #define GLOBAL_ENABLES_MASK (IPMI_BMC_EVT_MSG_BUFF | IPMI_BMC_RCV_MSG_INTR | \
515                              IPMI_BMC_EVT_MSG_INTR)
516
517 static u8 current_global_enables(struct smi_info *smi_info, u8 base,
518                                  bool *irq_on)
519 {
520         u8 enables = 0;
521
522         if (smi_info->supports_event_msg_buff)
523                 enables |= IPMI_BMC_EVT_MSG_BUFF;
524
525         if (((smi_info->io.irq && !smi_info->interrupt_disabled) ||
526              smi_info->cannot_disable_irq) &&
527             !smi_info->irq_enable_broken)
528                 enables |= IPMI_BMC_RCV_MSG_INTR;
529
530         if (smi_info->supports_event_msg_buff &&
531             smi_info->io.irq && !smi_info->interrupt_disabled &&
532             !smi_info->irq_enable_broken)
533                 enables |= IPMI_BMC_EVT_MSG_INTR;
534
535         *irq_on = enables & (IPMI_BMC_EVT_MSG_INTR | IPMI_BMC_RCV_MSG_INTR);
536
537         return enables;
538 }
539
540 static void check_bt_irq(struct smi_info *smi_info, bool irq_on)
541 {
542         u8 irqstate = smi_info->io.inputb(&smi_info->io, IPMI_BT_INTMASK_REG);
543
544         irqstate &= IPMI_BT_INTMASK_ENABLE_IRQ_BIT;
545
546         if ((bool)irqstate == irq_on)
547                 return;
548
549         if (irq_on)
550                 smi_info->io.outputb(&smi_info->io, IPMI_BT_INTMASK_REG,
551                                      IPMI_BT_INTMASK_ENABLE_IRQ_BIT);
552         else
553                 smi_info->io.outputb(&smi_info->io, IPMI_BT_INTMASK_REG, 0);
554 }
555
556 static void handle_transaction_done(struct smi_info *smi_info)
557 {
558         struct ipmi_smi_msg *msg;
559
560         debug_timestamp("Done");
561         switch (smi_info->si_state) {
562         case SI_NORMAL:
563                 if (!smi_info->curr_msg)
564                         break;
565
566                 smi_info->curr_msg->rsp_size
567                         = smi_info->handlers->get_result(
568                                 smi_info->si_sm,
569                                 smi_info->curr_msg->rsp,
570                                 IPMI_MAX_MSG_LENGTH);
571
572                 /*
573                  * Do this here becase deliver_recv_msg() releases the
574                  * lock, and a new message can be put in during the
575                  * time the lock is released.
576                  */
577                 msg = smi_info->curr_msg;
578                 smi_info->curr_msg = NULL;
579                 deliver_recv_msg(smi_info, msg);
580                 break;
581
582         case SI_GETTING_FLAGS:
583         {
584                 unsigned char msg[4];
585                 unsigned int  len;
586
587                 /* We got the flags from the SMI, now handle them. */
588                 len = smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
589                 if (msg[2] != 0) {
590                         /* Error fetching flags, just give up for now. */
591                         smi_info->si_state = SI_NORMAL;
592                 } else if (len < 4) {
593                         /*
594                          * Hmm, no flags.  That's technically illegal, but
595                          * don't use uninitialized data.
596                          */
597                         smi_info->si_state = SI_NORMAL;
598                 } else {
599                         smi_info->msg_flags = msg[3];
600                         handle_flags(smi_info);
601                 }
602                 break;
603         }
604
605         case SI_CLEARING_FLAGS:
606         {
607                 unsigned char msg[3];
608
609                 /* We cleared the flags. */
610                 smi_info->handlers->get_result(smi_info->si_sm, msg, 3);
611                 if (msg[2] != 0) {
612                         /* Error clearing flags */
613                         dev_warn(smi_info->io.dev,
614                                  "Error clearing flags: %2.2x\n", msg[2]);
615                 }
616                 smi_info->si_state = SI_NORMAL;
617                 break;
618         }
619
620         case SI_GETTING_EVENTS:
621         {
622                 smi_info->curr_msg->rsp_size
623                         = smi_info->handlers->get_result(
624                                 smi_info->si_sm,
625                                 smi_info->curr_msg->rsp,
626                                 IPMI_MAX_MSG_LENGTH);
627
628                 /*
629                  * Do this here becase deliver_recv_msg() releases the
630                  * lock, and a new message can be put in during the
631                  * time the lock is released.
632                  */
633                 msg = smi_info->curr_msg;
634                 smi_info->curr_msg = NULL;
635                 if (msg->rsp[2] != 0) {
636                         /* Error getting event, probably done. */
637                         msg->done(msg);
638
639                         /* Take off the event flag. */
640                         smi_info->msg_flags &= ~EVENT_MSG_BUFFER_FULL;
641                         handle_flags(smi_info);
642                 } else {
643                         smi_inc_stat(smi_info, events);
644
645                         /*
646                          * Do this before we deliver the message
647                          * because delivering the message releases the
648                          * lock and something else can mess with the
649                          * state.
650                          */
651                         handle_flags(smi_info);
652
653                         deliver_recv_msg(smi_info, msg);
654                 }
655                 break;
656         }
657
658         case SI_GETTING_MESSAGES:
659         {
660                 smi_info->curr_msg->rsp_size
661                         = smi_info->handlers->get_result(
662                                 smi_info->si_sm,
663                                 smi_info->curr_msg->rsp,
664                                 IPMI_MAX_MSG_LENGTH);
665
666                 /*
667                  * Do this here becase deliver_recv_msg() releases the
668                  * lock, and a new message can be put in during the
669                  * time the lock is released.
670                  */
671                 msg = smi_info->curr_msg;
672                 smi_info->curr_msg = NULL;
673                 if (msg->rsp[2] != 0) {
674                         /* Error getting event, probably done. */
675                         msg->done(msg);
676
677                         /* Take off the msg flag. */
678                         smi_info->msg_flags &= ~RECEIVE_MSG_AVAIL;
679                         handle_flags(smi_info);
680                 } else {
681                         smi_inc_stat(smi_info, incoming_messages);
682
683                         /*
684                          * Do this before we deliver the message
685                          * because delivering the message releases the
686                          * lock and something else can mess with the
687                          * state.
688                          */
689                         handle_flags(smi_info);
690
691                         deliver_recv_msg(smi_info, msg);
692                 }
693                 break;
694         }
695
696         case SI_CHECKING_ENABLES:
697         {
698                 unsigned char msg[4];
699                 u8 enables;
700                 bool irq_on;
701
702                 /* We got the flags from the SMI, now handle them. */
703                 smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
704                 if (msg[2] != 0) {
705                         dev_warn(smi_info->io.dev,
706                                  "Couldn't get irq info: %x.\n", msg[2]);
707                         dev_warn(smi_info->io.dev,
708                                  "Maybe ok, but ipmi might run very slowly.\n");
709                         smi_info->si_state = SI_NORMAL;
710                         break;
711                 }
712                 enables = current_global_enables(smi_info, 0, &irq_on);
713                 if (smi_info->io.si_type == SI_BT)
714                         /* BT has its own interrupt enable bit. */
715                         check_bt_irq(smi_info, irq_on);
716                 if (enables != (msg[3] & GLOBAL_ENABLES_MASK)) {
717                         /* Enables are not correct, fix them. */
718                         msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
719                         msg[1] = IPMI_SET_BMC_GLOBAL_ENABLES_CMD;
720                         msg[2] = enables | (msg[3] & ~GLOBAL_ENABLES_MASK);
721                         smi_info->handlers->start_transaction(
722                                 smi_info->si_sm, msg, 3);
723                         smi_info->si_state = SI_SETTING_ENABLES;
724                 } else if (smi_info->supports_event_msg_buff) {
725                         smi_info->curr_msg = ipmi_alloc_smi_msg();
726                         if (!smi_info->curr_msg) {
727                                 smi_info->si_state = SI_NORMAL;
728                                 break;
729                         }
730                         start_getting_events(smi_info);
731                 } else {
732                         smi_info->si_state = SI_NORMAL;
733                 }
734                 break;
735         }
736
737         case SI_SETTING_ENABLES:
738         {
739                 unsigned char msg[4];
740
741                 smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
742                 if (msg[2] != 0)
743                         dev_warn(smi_info->io.dev,
744                                  "Could not set the global enables: 0x%x.\n",
745                                  msg[2]);
746
747                 if (smi_info->supports_event_msg_buff) {
748                         smi_info->curr_msg = ipmi_alloc_smi_msg();
749                         if (!smi_info->curr_msg) {
750                                 smi_info->si_state = SI_NORMAL;
751                                 break;
752                         }
753                         start_getting_events(smi_info);
754                 } else {
755                         smi_info->si_state = SI_NORMAL;
756                 }
757                 break;
758         }
759         }
760 }
761
762 /*
763  * Called on timeouts and events.  Timeouts should pass the elapsed
764  * time, interrupts should pass in zero.  Must be called with
765  * si_lock held and interrupts disabled.
766  */
767 static enum si_sm_result smi_event_handler(struct smi_info *smi_info,
768                                            int time)
769 {
770         enum si_sm_result si_sm_result;
771
772 restart:
773         /*
774          * There used to be a loop here that waited a little while
775          * (around 25us) before giving up.  That turned out to be
776          * pointless, the minimum delays I was seeing were in the 300us
777          * range, which is far too long to wait in an interrupt.  So
778          * we just run until the state machine tells us something
779          * happened or it needs a delay.
780          */
781         si_sm_result = smi_info->handlers->event(smi_info->si_sm, time);
782         time = 0;
783         while (si_sm_result == SI_SM_CALL_WITHOUT_DELAY)
784                 si_sm_result = smi_info->handlers->event(smi_info->si_sm, 0);
785
786         if (si_sm_result == SI_SM_TRANSACTION_COMPLETE) {
787                 smi_inc_stat(smi_info, complete_transactions);
788
789                 handle_transaction_done(smi_info);
790                 goto restart;
791         } else if (si_sm_result == SI_SM_HOSED) {
792                 smi_inc_stat(smi_info, hosed_count);
793
794                 /*
795                  * Do the before return_hosed_msg, because that
796                  * releases the lock.
797                  */
798                 smi_info->si_state = SI_NORMAL;
799                 if (smi_info->curr_msg != NULL) {
800                         /*
801                          * If we were handling a user message, format
802                          * a response to send to the upper layer to
803                          * tell it about the error.
804                          */
805                         return_hosed_msg(smi_info, IPMI_ERR_UNSPECIFIED);
806                 }
807                 goto restart;
808         }
809
810         /*
811          * We prefer handling attn over new messages.  But don't do
812          * this if there is not yet an upper layer to handle anything.
813          */
814         if (likely(smi_info->intf) &&
815             (si_sm_result == SI_SM_ATTN || smi_info->got_attn)) {
816                 unsigned char msg[2];
817
818                 if (smi_info->si_state != SI_NORMAL) {
819                         /*
820                          * We got an ATTN, but we are doing something else.
821                          * Handle the ATTN later.
822                          */
823                         smi_info->got_attn = true;
824                 } else {
825                         smi_info->got_attn = false;
826                         smi_inc_stat(smi_info, attentions);
827
828                         /*
829                          * Got a attn, send down a get message flags to see
830                          * what's causing it.  It would be better to handle
831                          * this in the upper layer, but due to the way
832                          * interrupts work with the SMI, that's not really
833                          * possible.
834                          */
835                         msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
836                         msg[1] = IPMI_GET_MSG_FLAGS_CMD;
837
838                         start_new_msg(smi_info, msg, 2);
839                         smi_info->si_state = SI_GETTING_FLAGS;
840                         goto restart;
841                 }
842         }
843
844         /* If we are currently idle, try to start the next message. */
845         if (si_sm_result == SI_SM_IDLE) {
846                 smi_inc_stat(smi_info, idles);
847
848                 si_sm_result = start_next_msg(smi_info);
849                 if (si_sm_result != SI_SM_IDLE)
850                         goto restart;
851         }
852
853         if ((si_sm_result == SI_SM_IDLE)
854             && (atomic_read(&smi_info->req_events))) {
855                 /*
856                  * We are idle and the upper layer requested that I fetch
857                  * events, so do so.
858                  */
859                 atomic_set(&smi_info->req_events, 0);
860
861                 /*
862                  * Take this opportunity to check the interrupt and
863                  * message enable state for the BMC.  The BMC can be
864                  * asynchronously reset, and may thus get interrupts
865                  * disable and messages disabled.
866                  */
867                 if (smi_info->supports_event_msg_buff || smi_info->io.irq) {
868                         start_check_enables(smi_info);
869                 } else {
870                         smi_info->curr_msg = alloc_msg_handle_irq(smi_info);
871                         if (!smi_info->curr_msg)
872                                 goto out;
873
874                         start_getting_events(smi_info);
875                 }
876                 goto restart;
877         }
878
879         if (si_sm_result == SI_SM_IDLE && smi_info->timer_running) {
880                 /* Ok it if fails, the timer will just go off. */
881                 if (del_timer(&smi_info->si_timer))
882                         smi_info->timer_running = false;
883         }
884
885 out:
886         return si_sm_result;
887 }
888
889 static void check_start_timer_thread(struct smi_info *smi_info)
890 {
891         if (smi_info->si_state == SI_NORMAL && smi_info->curr_msg == NULL) {
892                 smi_mod_timer(smi_info, jiffies + SI_TIMEOUT_JIFFIES);
893
894                 if (smi_info->thread)
895                         wake_up_process(smi_info->thread);
896
897                 start_next_msg(smi_info);
898                 smi_event_handler(smi_info, 0);
899         }
900 }
901
902 static void flush_messages(void *send_info)
903 {
904         struct smi_info *smi_info = send_info;
905         enum si_sm_result result;
906
907         /*
908          * Currently, this function is called only in run-to-completion
909          * mode.  This means we are single-threaded, no need for locks.
910          */
911         result = smi_event_handler(smi_info, 0);
912         while (result != SI_SM_IDLE) {
913                 udelay(SI_SHORT_TIMEOUT_USEC);
914                 result = smi_event_handler(smi_info, SI_SHORT_TIMEOUT_USEC);
915         }
916 }
917
918 static void sender(void                *send_info,
919                    struct ipmi_smi_msg *msg)
920 {
921         struct smi_info   *smi_info = send_info;
922         unsigned long     flags;
923
924         debug_timestamp("Enqueue");
925
926         if (smi_info->run_to_completion) {
927                 /*
928                  * If we are running to completion, start it.  Upper
929                  * layer will call flush_messages to clear it out.
930                  */
931                 smi_info->waiting_msg = msg;
932                 return;
933         }
934
935         spin_lock_irqsave(&smi_info->si_lock, flags);
936         /*
937          * The following two lines don't need to be under the lock for
938          * the lock's sake, but they do need SMP memory barriers to
939          * avoid getting things out of order.  We are already claiming
940          * the lock, anyway, so just do it under the lock to avoid the
941          * ordering problem.
942          */
943         BUG_ON(smi_info->waiting_msg);
944         smi_info->waiting_msg = msg;
945         check_start_timer_thread(smi_info);
946         spin_unlock_irqrestore(&smi_info->si_lock, flags);
947 }
948
949 static void set_run_to_completion(void *send_info, bool i_run_to_completion)
950 {
951         struct smi_info   *smi_info = send_info;
952
953         smi_info->run_to_completion = i_run_to_completion;
954         if (i_run_to_completion)
955                 flush_messages(smi_info);
956 }
957
958 /*
959  * Use -1 in the nsec value of the busy waiting timespec to tell that
960  * we are spinning in kipmid looking for something and not delaying
961  * between checks
962  */
963 static inline void ipmi_si_set_not_busy(struct timespec64 *ts)
964 {
965         ts->tv_nsec = -1;
966 }
967 static inline int ipmi_si_is_busy(struct timespec64 *ts)
968 {
969         return ts->tv_nsec != -1;
970 }
971
972 static inline int ipmi_thread_busy_wait(enum si_sm_result smi_result,
973                                         const struct smi_info *smi_info,
974                                         struct timespec64 *busy_until)
975 {
976         unsigned int max_busy_us = 0;
977
978         if (smi_info->intf_num < num_max_busy_us)
979                 max_busy_us = kipmid_max_busy_us[smi_info->intf_num];
980         if (max_busy_us == 0 || smi_result != SI_SM_CALL_WITH_DELAY)
981                 ipmi_si_set_not_busy(busy_until);
982         else if (!ipmi_si_is_busy(busy_until)) {
983                 getnstimeofday64(busy_until);
984                 timespec64_add_ns(busy_until, max_busy_us*NSEC_PER_USEC);
985         } else {
986                 struct timespec64 now;
987
988                 getnstimeofday64(&now);
989                 if (unlikely(timespec64_compare(&now, busy_until) > 0)) {
990                         ipmi_si_set_not_busy(busy_until);
991                         return 0;
992                 }
993         }
994         return 1;
995 }
996
997
998 /*
999  * A busy-waiting loop for speeding up IPMI operation.
1000  *
1001  * Lousy hardware makes this hard.  This is only enabled for systems
1002  * that are not BT and do not have interrupts.  It starts spinning
1003  * when an operation is complete or until max_busy tells it to stop
1004  * (if that is enabled).  See the paragraph on kimid_max_busy_us in
1005  * Documentation/IPMI.txt for details.
1006  */
1007 static int ipmi_thread(void *data)
1008 {
1009         struct smi_info *smi_info = data;
1010         unsigned long flags;
1011         enum si_sm_result smi_result;
1012         struct timespec64 busy_until;
1013
1014         ipmi_si_set_not_busy(&busy_until);
1015         set_user_nice(current, MAX_NICE);
1016         while (!kthread_should_stop()) {
1017                 int busy_wait;
1018
1019                 spin_lock_irqsave(&(smi_info->si_lock), flags);
1020                 smi_result = smi_event_handler(smi_info, 0);
1021
1022                 /*
1023                  * If the driver is doing something, there is a possible
1024                  * race with the timer.  If the timer handler see idle,
1025                  * and the thread here sees something else, the timer
1026                  * handler won't restart the timer even though it is
1027                  * required.  So start it here if necessary.
1028                  */
1029                 if (smi_result != SI_SM_IDLE && !smi_info->timer_running)
1030                         smi_mod_timer(smi_info, jiffies + SI_TIMEOUT_JIFFIES);
1031
1032                 spin_unlock_irqrestore(&(smi_info->si_lock), flags);
1033                 busy_wait = ipmi_thread_busy_wait(smi_result, smi_info,
1034                                                   &busy_until);
1035                 if (smi_result == SI_SM_CALL_WITHOUT_DELAY)
1036                         ; /* do nothing */
1037                 else if (smi_result == SI_SM_CALL_WITH_DELAY && busy_wait)
1038                         schedule();
1039                 else if (smi_result == SI_SM_IDLE) {
1040                         if (atomic_read(&smi_info->need_watch)) {
1041                                 schedule_timeout_interruptible(100);
1042                         } else {
1043                                 /* Wait to be woken up when we are needed. */
1044                                 __set_current_state(TASK_INTERRUPTIBLE);
1045                                 schedule();
1046                         }
1047                 } else
1048                         schedule_timeout_interruptible(1);
1049         }
1050         return 0;
1051 }
1052
1053
1054 static void poll(void *send_info)
1055 {
1056         struct smi_info *smi_info = send_info;
1057         unsigned long flags = 0;
1058         bool run_to_completion = smi_info->run_to_completion;
1059
1060         /*
1061          * Make sure there is some delay in the poll loop so we can
1062          * drive time forward and timeout things.
1063          */
1064         udelay(10);
1065         if (!run_to_completion)
1066                 spin_lock_irqsave(&smi_info->si_lock, flags);
1067         smi_event_handler(smi_info, 10);
1068         if (!run_to_completion)
1069                 spin_unlock_irqrestore(&smi_info->si_lock, flags);
1070 }
1071
1072 static void request_events(void *send_info)
1073 {
1074         struct smi_info *smi_info = send_info;
1075
1076         if (!smi_info->has_event_buffer)
1077                 return;
1078
1079         atomic_set(&smi_info->req_events, 1);
1080 }
1081
1082 static void set_need_watch(void *send_info, bool enable)
1083 {
1084         struct smi_info *smi_info = send_info;
1085         unsigned long flags;
1086
1087         atomic_set(&smi_info->need_watch, enable);
1088         spin_lock_irqsave(&smi_info->si_lock, flags);
1089         check_start_timer_thread(smi_info);
1090         spin_unlock_irqrestore(&smi_info->si_lock, flags);
1091 }
1092
1093 static void smi_timeout(struct timer_list *t)
1094 {
1095         struct smi_info   *smi_info = from_timer(smi_info, t, si_timer);
1096         enum si_sm_result smi_result;
1097         unsigned long     flags;
1098         unsigned long     jiffies_now;
1099         long              time_diff;
1100         long              timeout;
1101
1102         spin_lock_irqsave(&(smi_info->si_lock), flags);
1103         debug_timestamp("Timer");
1104
1105         jiffies_now = jiffies;
1106         time_diff = (((long)jiffies_now - (long)smi_info->last_timeout_jiffies)
1107                      * SI_USEC_PER_JIFFY);
1108         smi_result = smi_event_handler(smi_info, time_diff);
1109
1110         if ((smi_info->io.irq) && (!smi_info->interrupt_disabled)) {
1111                 /* Running with interrupts, only do long timeouts. */
1112                 timeout = jiffies + SI_TIMEOUT_JIFFIES;
1113                 smi_inc_stat(smi_info, long_timeouts);
1114                 goto do_mod_timer;
1115         }
1116
1117         /*
1118          * If the state machine asks for a short delay, then shorten
1119          * the timer timeout.
1120          */
1121         if (smi_result == SI_SM_CALL_WITH_DELAY) {
1122                 smi_inc_stat(smi_info, short_timeouts);
1123                 timeout = jiffies + 1;
1124         } else {
1125                 smi_inc_stat(smi_info, long_timeouts);
1126                 timeout = jiffies + SI_TIMEOUT_JIFFIES;
1127         }
1128
1129 do_mod_timer:
1130         if (smi_result != SI_SM_IDLE)
1131                 smi_mod_timer(smi_info, timeout);
1132         else
1133                 smi_info->timer_running = false;
1134         spin_unlock_irqrestore(&(smi_info->si_lock), flags);
1135 }
1136
1137 irqreturn_t ipmi_si_irq_handler(int irq, void *data)
1138 {
1139         struct smi_info *smi_info = data;
1140         unsigned long   flags;
1141
1142         if (smi_info->io.si_type == SI_BT)
1143                 /* We need to clear the IRQ flag for the BT interface. */
1144                 smi_info->io.outputb(&smi_info->io, IPMI_BT_INTMASK_REG,
1145                                      IPMI_BT_INTMASK_CLEAR_IRQ_BIT
1146                                      | IPMI_BT_INTMASK_ENABLE_IRQ_BIT);
1147
1148         spin_lock_irqsave(&(smi_info->si_lock), flags);
1149
1150         smi_inc_stat(smi_info, interrupts);
1151
1152         debug_timestamp("Interrupt");
1153
1154         smi_event_handler(smi_info, 0);
1155         spin_unlock_irqrestore(&(smi_info->si_lock), flags);
1156         return IRQ_HANDLED;
1157 }
1158
1159 static int smi_start_processing(void       *send_info,
1160                                 ipmi_smi_t intf)
1161 {
1162         struct smi_info *new_smi = send_info;
1163         int             enable = 0;
1164
1165         new_smi->intf = intf;
1166
1167         /* Set up the timer that drives the interface. */
1168         timer_setup(&new_smi->si_timer, smi_timeout, 0);
1169         new_smi->timer_can_start = true;
1170         smi_mod_timer(new_smi, jiffies + SI_TIMEOUT_JIFFIES);
1171
1172         /* Try to claim any interrupts. */
1173         if (new_smi->io.irq_setup) {
1174                 new_smi->io.irq_handler_data = new_smi;
1175                 new_smi->io.irq_setup(&new_smi->io);
1176         }
1177
1178         /*
1179          * Check if the user forcefully enabled the daemon.
1180          */
1181         if (new_smi->intf_num < num_force_kipmid)
1182                 enable = force_kipmid[new_smi->intf_num];
1183         /*
1184          * The BT interface is efficient enough to not need a thread,
1185          * and there is no need for a thread if we have interrupts.
1186          */
1187         else if ((new_smi->io.si_type != SI_BT) && (!new_smi->io.irq))
1188                 enable = 1;
1189
1190         if (enable) {
1191                 new_smi->thread = kthread_run(ipmi_thread, new_smi,
1192                                               "kipmi%d", new_smi->intf_num);
1193                 if (IS_ERR(new_smi->thread)) {
1194                         dev_notice(new_smi->io.dev, "Could not start"
1195                                    " kernel thread due to error %ld, only using"
1196                                    " timers to drive the interface\n",
1197                                    PTR_ERR(new_smi->thread));
1198                         new_smi->thread = NULL;
1199                 }
1200         }
1201
1202         return 0;
1203 }
1204
1205 static int get_smi_info(void *send_info, struct ipmi_smi_info *data)
1206 {
1207         struct smi_info *smi = send_info;
1208
1209         data->addr_src = smi->io.addr_source;
1210         data->dev = smi->io.dev;
1211         data->addr_info = smi->io.addr_info;
1212         get_device(smi->io.dev);
1213
1214         return 0;
1215 }
1216
1217 static void set_maintenance_mode(void *send_info, bool enable)
1218 {
1219         struct smi_info   *smi_info = send_info;
1220
1221         if (!enable)
1222                 atomic_set(&smi_info->req_events, 0);
1223 }
1224
1225 static const struct ipmi_smi_handlers handlers = {
1226         .owner                  = THIS_MODULE,
1227         .start_processing       = smi_start_processing,
1228         .get_smi_info           = get_smi_info,
1229         .sender                 = sender,
1230         .request_events         = request_events,
1231         .set_need_watch         = set_need_watch,
1232         .set_maintenance_mode   = set_maintenance_mode,
1233         .set_run_to_completion  = set_run_to_completion,
1234         .flush_messages         = flush_messages,
1235         .poll                   = poll,
1236 };
1237
1238 static LIST_HEAD(smi_infos);
1239 static DEFINE_MUTEX(smi_infos_lock);
1240 static int smi_num; /* Used to sequence the SMIs */
1241
1242 static const char * const addr_space_to_str[] = { "i/o", "mem" };
1243
1244 module_param_array(force_kipmid, int, &num_force_kipmid, 0);
1245 MODULE_PARM_DESC(force_kipmid, "Force the kipmi daemon to be enabled (1) or"
1246                  " disabled(0).  Normally the IPMI driver auto-detects"
1247                  " this, but the value may be overridden by this parm.");
1248 module_param(unload_when_empty, bool, 0);
1249 MODULE_PARM_DESC(unload_when_empty, "Unload the module if no interfaces are"
1250                  " specified or found, default is 1.  Setting to 0"
1251                  " is useful for hot add of devices using hotmod.");
1252 module_param_array(kipmid_max_busy_us, uint, &num_max_busy_us, 0644);
1253 MODULE_PARM_DESC(kipmid_max_busy_us,
1254                  "Max time (in microseconds) to busy-wait for IPMI data before"
1255                  " sleeping. 0 (default) means to wait forever. Set to 100-500"
1256                  " if kipmid is using up a lot of CPU time.");
1257
1258 void ipmi_irq_finish_setup(struct si_sm_io *io)
1259 {
1260         if (io->si_type == SI_BT)
1261                 /* Enable the interrupt in the BT interface. */
1262                 io->outputb(io, IPMI_BT_INTMASK_REG,
1263                             IPMI_BT_INTMASK_ENABLE_IRQ_BIT);
1264 }
1265
1266 void ipmi_irq_start_cleanup(struct si_sm_io *io)
1267 {
1268         if (io->si_type == SI_BT)
1269                 /* Disable the interrupt in the BT interface. */
1270                 io->outputb(io, IPMI_BT_INTMASK_REG, 0);
1271 }
1272
1273 static void std_irq_cleanup(struct si_sm_io *io)
1274 {
1275         ipmi_irq_start_cleanup(io);
1276         free_irq(io->irq, io->irq_handler_data);
1277 }
1278
1279 int ipmi_std_irq_setup(struct si_sm_io *io)
1280 {
1281         int rv;
1282
1283         if (!io->irq)
1284                 return 0;
1285
1286         rv = request_irq(io->irq,
1287                          ipmi_si_irq_handler,
1288                          IRQF_SHARED,
1289                          DEVICE_NAME,
1290                          io->irq_handler_data);
1291         if (rv) {
1292                 dev_warn(io->dev, "%s unable to claim interrupt %d,"
1293                          " running polled\n",
1294                          DEVICE_NAME, io->irq);
1295                 io->irq = 0;
1296         } else {
1297                 io->irq_cleanup = std_irq_cleanup;
1298                 ipmi_irq_finish_setup(io);
1299                 dev_info(io->dev, "Using irq %d\n", io->irq);
1300         }
1301
1302         return rv;
1303 }
1304
1305 static int wait_for_msg_done(struct smi_info *smi_info)
1306 {
1307         enum si_sm_result     smi_result;
1308
1309         smi_result = smi_info->handlers->event(smi_info->si_sm, 0);
1310         for (;;) {
1311                 if (smi_result == SI_SM_CALL_WITH_DELAY ||
1312                     smi_result == SI_SM_CALL_WITH_TICK_DELAY) {
1313                         schedule_timeout_uninterruptible(1);
1314                         smi_result = smi_info->handlers->event(
1315                                 smi_info->si_sm, jiffies_to_usecs(1));
1316                 } else if (smi_result == SI_SM_CALL_WITHOUT_DELAY) {
1317                         smi_result = smi_info->handlers->event(
1318                                 smi_info->si_sm, 0);
1319                 } else
1320                         break;
1321         }
1322         if (smi_result == SI_SM_HOSED)
1323                 /*
1324                  * We couldn't get the state machine to run, so whatever's at
1325                  * the port is probably not an IPMI SMI interface.
1326                  */
1327                 return -ENODEV;
1328
1329         return 0;
1330 }
1331
1332 static int try_get_dev_id(struct smi_info *smi_info)
1333 {
1334         unsigned char         msg[2];
1335         unsigned char         *resp;
1336         unsigned long         resp_len;
1337         int                   rv = 0;
1338
1339         resp = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
1340         if (!resp)
1341                 return -ENOMEM;
1342
1343         /*
1344          * Do a Get Device ID command, since it comes back with some
1345          * useful info.
1346          */
1347         msg[0] = IPMI_NETFN_APP_REQUEST << 2;
1348         msg[1] = IPMI_GET_DEVICE_ID_CMD;
1349         smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
1350
1351         rv = wait_for_msg_done(smi_info);
1352         if (rv)
1353                 goto out;
1354
1355         resp_len = smi_info->handlers->get_result(smi_info->si_sm,
1356                                                   resp, IPMI_MAX_MSG_LENGTH);
1357
1358         /* Check and record info from the get device id, in case we need it. */
1359         rv = ipmi_demangle_device_id(resp[0] >> 2, resp[1],
1360                         resp + 2, resp_len - 2, &smi_info->device_id);
1361
1362 out:
1363         kfree(resp);
1364         return rv;
1365 }
1366
1367 static int get_global_enables(struct smi_info *smi_info, u8 *enables)
1368 {
1369         unsigned char         msg[3];
1370         unsigned char         *resp;
1371         unsigned long         resp_len;
1372         int                   rv;
1373
1374         resp = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
1375         if (!resp)
1376                 return -ENOMEM;
1377
1378         msg[0] = IPMI_NETFN_APP_REQUEST << 2;
1379         msg[1] = IPMI_GET_BMC_GLOBAL_ENABLES_CMD;
1380         smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
1381
1382         rv = wait_for_msg_done(smi_info);
1383         if (rv) {
1384                 dev_warn(smi_info->io.dev,
1385                          "Error getting response from get global enables command: %d\n",
1386                          rv);
1387                 goto out;
1388         }
1389
1390         resp_len = smi_info->handlers->get_result(smi_info->si_sm,
1391                                                   resp, IPMI_MAX_MSG_LENGTH);
1392
1393         if (resp_len < 4 ||
1394                         resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
1395                         resp[1] != IPMI_GET_BMC_GLOBAL_ENABLES_CMD   ||
1396                         resp[2] != 0) {
1397                 dev_warn(smi_info->io.dev,
1398                          "Invalid return from get global enables command: %ld %x %x %x\n",
1399                          resp_len, resp[0], resp[1], resp[2]);
1400                 rv = -EINVAL;
1401                 goto out;
1402         } else {
1403                 *enables = resp[3];
1404         }
1405
1406 out:
1407         kfree(resp);
1408         return rv;
1409 }
1410
1411 /*
1412  * Returns 1 if it gets an error from the command.
1413  */
1414 static int set_global_enables(struct smi_info *smi_info, u8 enables)
1415 {
1416         unsigned char         msg[3];
1417         unsigned char         *resp;
1418         unsigned long         resp_len;
1419         int                   rv;
1420
1421         resp = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
1422         if (!resp)
1423                 return -ENOMEM;
1424
1425         msg[0] = IPMI_NETFN_APP_REQUEST << 2;
1426         msg[1] = IPMI_SET_BMC_GLOBAL_ENABLES_CMD;
1427         msg[2] = enables;
1428         smi_info->handlers->start_transaction(smi_info->si_sm, msg, 3);
1429
1430         rv = wait_for_msg_done(smi_info);
1431         if (rv) {
1432                 dev_warn(smi_info->io.dev,
1433                          "Error getting response from set global enables command: %d\n",
1434                          rv);
1435                 goto out;
1436         }
1437
1438         resp_len = smi_info->handlers->get_result(smi_info->si_sm,
1439                                                   resp, IPMI_MAX_MSG_LENGTH);
1440
1441         if (resp_len < 3 ||
1442                         resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
1443                         resp[1] != IPMI_SET_BMC_GLOBAL_ENABLES_CMD) {
1444                 dev_warn(smi_info->io.dev,
1445                          "Invalid return from set global enables command: %ld %x %x\n",
1446                          resp_len, resp[0], resp[1]);
1447                 rv = -EINVAL;
1448                 goto out;
1449         }
1450
1451         if (resp[2] != 0)
1452                 rv = 1;
1453
1454 out:
1455         kfree(resp);
1456         return rv;
1457 }
1458
1459 /*
1460  * Some BMCs do not support clearing the receive irq bit in the global
1461  * enables (even if they don't support interrupts on the BMC).  Check
1462  * for this and handle it properly.
1463  */
1464 static void check_clr_rcv_irq(struct smi_info *smi_info)
1465 {
1466         u8 enables = 0;
1467         int rv;
1468
1469         rv = get_global_enables(smi_info, &enables);
1470         if (!rv) {
1471                 if ((enables & IPMI_BMC_RCV_MSG_INTR) == 0)
1472                         /* Already clear, should work ok. */
1473                         return;
1474
1475                 enables &= ~IPMI_BMC_RCV_MSG_INTR;
1476                 rv = set_global_enables(smi_info, enables);
1477         }
1478
1479         if (rv < 0) {
1480                 dev_err(smi_info->io.dev,
1481                         "Cannot check clearing the rcv irq: %d\n", rv);
1482                 return;
1483         }
1484
1485         if (rv) {
1486                 /*
1487                  * An error when setting the event buffer bit means
1488                  * clearing the bit is not supported.
1489                  */
1490                 dev_warn(smi_info->io.dev,
1491                          "The BMC does not support clearing the recv irq bit, compensating, but the BMC needs to be fixed.\n");
1492                 smi_info->cannot_disable_irq = true;
1493         }
1494 }
1495
1496 /*
1497  * Some BMCs do not support setting the interrupt bits in the global
1498  * enables even if they support interrupts.  Clearly bad, but we can
1499  * compensate.
1500  */
1501 static void check_set_rcv_irq(struct smi_info *smi_info)
1502 {
1503         u8 enables = 0;
1504         int rv;
1505
1506         if (!smi_info->io.irq)
1507                 return;
1508
1509         rv = get_global_enables(smi_info, &enables);
1510         if (!rv) {
1511                 enables |= IPMI_BMC_RCV_MSG_INTR;
1512                 rv = set_global_enables(smi_info, enables);
1513         }
1514
1515         if (rv < 0) {
1516                 dev_err(smi_info->io.dev,
1517                         "Cannot check setting the rcv irq: %d\n", rv);
1518                 return;
1519         }
1520
1521         if (rv) {
1522                 /*
1523                  * An error when setting the event buffer bit means
1524                  * setting the bit is not supported.
1525                  */
1526                 dev_warn(smi_info->io.dev,
1527                          "The BMC does not support setting the recv irq bit, compensating, but the BMC needs to be fixed.\n");
1528                 smi_info->cannot_disable_irq = true;
1529                 smi_info->irq_enable_broken = true;
1530         }
1531 }
1532
1533 static int try_enable_event_buffer(struct smi_info *smi_info)
1534 {
1535         unsigned char         msg[3];
1536         unsigned char         *resp;
1537         unsigned long         resp_len;
1538         int                   rv = 0;
1539
1540         resp = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
1541         if (!resp)
1542                 return -ENOMEM;
1543
1544         msg[0] = IPMI_NETFN_APP_REQUEST << 2;
1545         msg[1] = IPMI_GET_BMC_GLOBAL_ENABLES_CMD;
1546         smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
1547
1548         rv = wait_for_msg_done(smi_info);
1549         if (rv) {
1550                 pr_warn(PFX "Error getting response from get global enables command, the event buffer is not enabled.\n");
1551                 goto out;
1552         }
1553
1554         resp_len = smi_info->handlers->get_result(smi_info->si_sm,
1555                                                   resp, IPMI_MAX_MSG_LENGTH);
1556
1557         if (resp_len < 4 ||
1558                         resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
1559                         resp[1] != IPMI_GET_BMC_GLOBAL_ENABLES_CMD   ||
1560                         resp[2] != 0) {
1561                 pr_warn(PFX "Invalid return from get global enables command, cannot enable the event buffer.\n");
1562                 rv = -EINVAL;
1563                 goto out;
1564         }
1565
1566         if (resp[3] & IPMI_BMC_EVT_MSG_BUFF) {
1567                 /* buffer is already enabled, nothing to do. */
1568                 smi_info->supports_event_msg_buff = true;
1569                 goto out;
1570         }
1571
1572         msg[0] = IPMI_NETFN_APP_REQUEST << 2;
1573         msg[1] = IPMI_SET_BMC_GLOBAL_ENABLES_CMD;
1574         msg[2] = resp[3] | IPMI_BMC_EVT_MSG_BUFF;
1575         smi_info->handlers->start_transaction(smi_info->si_sm, msg, 3);
1576
1577         rv = wait_for_msg_done(smi_info);
1578         if (rv) {
1579                 pr_warn(PFX "Error getting response from set global, enables command, the event buffer is not enabled.\n");
1580                 goto out;
1581         }
1582
1583         resp_len = smi_info->handlers->get_result(smi_info->si_sm,
1584                                                   resp, IPMI_MAX_MSG_LENGTH);
1585
1586         if (resp_len < 3 ||
1587                         resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
1588                         resp[1] != IPMI_SET_BMC_GLOBAL_ENABLES_CMD) {
1589                 pr_warn(PFX "Invalid return from get global, enables command, not enable the event buffer.\n");
1590                 rv = -EINVAL;
1591                 goto out;
1592         }
1593
1594         if (resp[2] != 0)
1595                 /*
1596                  * An error when setting the event buffer bit means
1597                  * that the event buffer is not supported.
1598                  */
1599                 rv = -ENOENT;
1600         else
1601                 smi_info->supports_event_msg_buff = true;
1602
1603 out:
1604         kfree(resp);
1605         return rv;
1606 }
1607
1608 #ifdef CONFIG_IPMI_PROC_INTERFACE
1609 static int smi_type_proc_show(struct seq_file *m, void *v)
1610 {
1611         struct smi_info *smi = m->private;
1612
1613         seq_printf(m, "%s\n", si_to_str[smi->io.si_type]);
1614
1615         return 0;
1616 }
1617
1618 static int smi_type_proc_open(struct inode *inode, struct file *file)
1619 {
1620         return single_open(file, smi_type_proc_show, PDE_DATA(inode));
1621 }
1622
1623 static const struct file_operations smi_type_proc_ops = {
1624         .open           = smi_type_proc_open,
1625         .read           = seq_read,
1626         .llseek         = seq_lseek,
1627         .release        = single_release,
1628 };
1629
1630 static int smi_si_stats_proc_show(struct seq_file *m, void *v)
1631 {
1632         struct smi_info *smi = m->private;
1633
1634         seq_printf(m, "interrupts_enabled:    %d\n",
1635                        smi->io.irq && !smi->interrupt_disabled);
1636         seq_printf(m, "short_timeouts:        %u\n",
1637                        smi_get_stat(smi, short_timeouts));
1638         seq_printf(m, "long_timeouts:         %u\n",
1639                        smi_get_stat(smi, long_timeouts));
1640         seq_printf(m, "idles:                 %u\n",
1641                        smi_get_stat(smi, idles));
1642         seq_printf(m, "interrupts:            %u\n",
1643                        smi_get_stat(smi, interrupts));
1644         seq_printf(m, "attentions:            %u\n",
1645                        smi_get_stat(smi, attentions));
1646         seq_printf(m, "flag_fetches:          %u\n",
1647                        smi_get_stat(smi, flag_fetches));
1648         seq_printf(m, "hosed_count:           %u\n",
1649                        smi_get_stat(smi, hosed_count));
1650         seq_printf(m, "complete_transactions: %u\n",
1651                        smi_get_stat(smi, complete_transactions));
1652         seq_printf(m, "events:                %u\n",
1653                        smi_get_stat(smi, events));
1654         seq_printf(m, "watchdog_pretimeouts:  %u\n",
1655                        smi_get_stat(smi, watchdog_pretimeouts));
1656         seq_printf(m, "incoming_messages:     %u\n",
1657                        smi_get_stat(smi, incoming_messages));
1658         return 0;
1659 }
1660
1661 static int smi_si_stats_proc_open(struct inode *inode, struct file *file)
1662 {
1663         return single_open(file, smi_si_stats_proc_show, PDE_DATA(inode));
1664 }
1665
1666 static const struct file_operations smi_si_stats_proc_ops = {
1667         .open           = smi_si_stats_proc_open,
1668         .read           = seq_read,
1669         .llseek         = seq_lseek,
1670         .release        = single_release,
1671 };
1672
1673 static int smi_params_proc_show(struct seq_file *m, void *v)
1674 {
1675         struct smi_info *smi = m->private;
1676
1677         seq_printf(m,
1678                    "%s,%s,0x%lx,rsp=%d,rsi=%d,rsh=%d,irq=%d,ipmb=%d\n",
1679                    si_to_str[smi->io.si_type],
1680                    addr_space_to_str[smi->io.addr_type],
1681                    smi->io.addr_data,
1682                    smi->io.regspacing,
1683                    smi->io.regsize,
1684                    smi->io.regshift,
1685                    smi->io.irq,
1686                    smi->io.slave_addr);
1687
1688         return 0;
1689 }
1690
1691 static int smi_params_proc_open(struct inode *inode, struct file *file)
1692 {
1693         return single_open(file, smi_params_proc_show, PDE_DATA(inode));
1694 }
1695
1696 static const struct file_operations smi_params_proc_ops = {
1697         .open           = smi_params_proc_open,
1698         .read           = seq_read,
1699         .llseek         = seq_lseek,
1700         .release        = single_release,
1701 };
1702 #endif
1703
1704 #define IPMI_SI_ATTR(name) \
1705 static ssize_t ipmi_##name##_show(struct device *dev,                   \
1706                                   struct device_attribute *attr,        \
1707                                   char *buf)                            \
1708 {                                                                       \
1709         struct smi_info *smi_info = dev_get_drvdata(dev);               \
1710                                                                         \
1711         return snprintf(buf, 10, "%u\n", smi_get_stat(smi_info, name)); \
1712 }                                                                       \
1713 static DEVICE_ATTR(name, S_IRUGO, ipmi_##name##_show, NULL)
1714
1715 static ssize_t ipmi_type_show(struct device *dev,
1716                               struct device_attribute *attr,
1717                               char *buf)
1718 {
1719         struct smi_info *smi_info = dev_get_drvdata(dev);
1720
1721         return snprintf(buf, 10, "%s\n", si_to_str[smi_info->io.si_type]);
1722 }
1723 static DEVICE_ATTR(type, S_IRUGO, ipmi_type_show, NULL);
1724
1725 static ssize_t ipmi_interrupts_enabled_show(struct device *dev,
1726                                             struct device_attribute *attr,
1727                                             char *buf)
1728 {
1729         struct smi_info *smi_info = dev_get_drvdata(dev);
1730         int enabled = smi_info->io.irq && !smi_info->interrupt_disabled;
1731
1732         return snprintf(buf, 10, "%d\n", enabled);
1733 }
1734 static DEVICE_ATTR(interrupts_enabled, S_IRUGO,
1735                    ipmi_interrupts_enabled_show, NULL);
1736
1737 IPMI_SI_ATTR(short_timeouts);
1738 IPMI_SI_ATTR(long_timeouts);
1739 IPMI_SI_ATTR(idles);
1740 IPMI_SI_ATTR(interrupts);
1741 IPMI_SI_ATTR(attentions);
1742 IPMI_SI_ATTR(flag_fetches);
1743 IPMI_SI_ATTR(hosed_count);
1744 IPMI_SI_ATTR(complete_transactions);
1745 IPMI_SI_ATTR(events);
1746 IPMI_SI_ATTR(watchdog_pretimeouts);
1747 IPMI_SI_ATTR(incoming_messages);
1748
1749 static ssize_t ipmi_params_show(struct device *dev,
1750                                 struct device_attribute *attr,
1751                                 char *buf)
1752 {
1753         struct smi_info *smi_info = dev_get_drvdata(dev);
1754
1755         return snprintf(buf, 200,
1756                         "%s,%s,0x%lx,rsp=%d,rsi=%d,rsh=%d,irq=%d,ipmb=%d\n",
1757                         si_to_str[smi_info->io.si_type],
1758                         addr_space_to_str[smi_info->io.addr_type],
1759                         smi_info->io.addr_data,
1760                         smi_info->io.regspacing,
1761                         smi_info->io.regsize,
1762                         smi_info->io.regshift,
1763                         smi_info->io.irq,
1764                         smi_info->io.slave_addr);
1765 }
1766 static DEVICE_ATTR(params, S_IRUGO, ipmi_params_show, NULL);
1767
1768 static struct attribute *ipmi_si_dev_attrs[] = {
1769         &dev_attr_type.attr,
1770         &dev_attr_interrupts_enabled.attr,
1771         &dev_attr_short_timeouts.attr,
1772         &dev_attr_long_timeouts.attr,
1773         &dev_attr_idles.attr,
1774         &dev_attr_interrupts.attr,
1775         &dev_attr_attentions.attr,
1776         &dev_attr_flag_fetches.attr,
1777         &dev_attr_hosed_count.attr,
1778         &dev_attr_complete_transactions.attr,
1779         &dev_attr_events.attr,
1780         &dev_attr_watchdog_pretimeouts.attr,
1781         &dev_attr_incoming_messages.attr,
1782         &dev_attr_params.attr,
1783         NULL
1784 };
1785
1786 static const struct attribute_group ipmi_si_dev_attr_group = {
1787         .attrs          = ipmi_si_dev_attrs,
1788 };
1789
1790 /*
1791  * oem_data_avail_to_receive_msg_avail
1792  * @info - smi_info structure with msg_flags set
1793  *
1794  * Converts flags from OEM_DATA_AVAIL to RECEIVE_MSG_AVAIL
1795  * Returns 1 indicating need to re-run handle_flags().
1796  */
1797 static int oem_data_avail_to_receive_msg_avail(struct smi_info *smi_info)
1798 {
1799         smi_info->msg_flags = ((smi_info->msg_flags & ~OEM_DATA_AVAIL) |
1800                                RECEIVE_MSG_AVAIL);
1801         return 1;
1802 }
1803
1804 /*
1805  * setup_dell_poweredge_oem_data_handler
1806  * @info - smi_info.device_id must be populated
1807  *
1808  * Systems that match, but have firmware version < 1.40 may assert
1809  * OEM0_DATA_AVAIL on their own, without being told via Set Flags that
1810  * it's safe to do so.  Such systems will de-assert OEM1_DATA_AVAIL
1811  * upon receipt of IPMI_GET_MSG_CMD, so we should treat these flags
1812  * as RECEIVE_MSG_AVAIL instead.
1813  *
1814  * As Dell has no plans to release IPMI 1.5 firmware that *ever*
1815  * assert the OEM[012] bits, and if it did, the driver would have to
1816  * change to handle that properly, we don't actually check for the
1817  * firmware version.
1818  * Device ID = 0x20                BMC on PowerEdge 8G servers
1819  * Device Revision = 0x80
1820  * Firmware Revision1 = 0x01       BMC version 1.40
1821  * Firmware Revision2 = 0x40       BCD encoded
1822  * IPMI Version = 0x51             IPMI 1.5
1823  * Manufacturer ID = A2 02 00      Dell IANA
1824  *
1825  * Additionally, PowerEdge systems with IPMI < 1.5 may also assert
1826  * OEM0_DATA_AVAIL and needs to be treated as RECEIVE_MSG_AVAIL.
1827  *
1828  */
1829 #define DELL_POWEREDGE_8G_BMC_DEVICE_ID  0x20
1830 #define DELL_POWEREDGE_8G_BMC_DEVICE_REV 0x80
1831 #define DELL_POWEREDGE_8G_BMC_IPMI_VERSION 0x51
1832 #define DELL_IANA_MFR_ID 0x0002a2
1833 static void setup_dell_poweredge_oem_data_handler(struct smi_info *smi_info)
1834 {
1835         struct ipmi_device_id *id = &smi_info->device_id;
1836         if (id->manufacturer_id == DELL_IANA_MFR_ID) {
1837                 if (id->device_id       == DELL_POWEREDGE_8G_BMC_DEVICE_ID  &&
1838                     id->device_revision == DELL_POWEREDGE_8G_BMC_DEVICE_REV &&
1839                     id->ipmi_version   == DELL_POWEREDGE_8G_BMC_IPMI_VERSION) {
1840                         smi_info->oem_data_avail_handler =
1841                                 oem_data_avail_to_receive_msg_avail;
1842                 } else if (ipmi_version_major(id) < 1 ||
1843                            (ipmi_version_major(id) == 1 &&
1844                             ipmi_version_minor(id) < 5)) {
1845                         smi_info->oem_data_avail_handler =
1846                                 oem_data_avail_to_receive_msg_avail;
1847                 }
1848         }
1849 }
1850
1851 #define CANNOT_RETURN_REQUESTED_LENGTH 0xCA
1852 static void return_hosed_msg_badsize(struct smi_info *smi_info)
1853 {
1854         struct ipmi_smi_msg *msg = smi_info->curr_msg;
1855
1856         /* Make it a response */
1857         msg->rsp[0] = msg->data[0] | 4;
1858         msg->rsp[1] = msg->data[1];
1859         msg->rsp[2] = CANNOT_RETURN_REQUESTED_LENGTH;
1860         msg->rsp_size = 3;
1861         smi_info->curr_msg = NULL;
1862         deliver_recv_msg(smi_info, msg);
1863 }
1864
1865 /*
1866  * dell_poweredge_bt_xaction_handler
1867  * @info - smi_info.device_id must be populated
1868  *
1869  * Dell PowerEdge servers with the BT interface (x6xx and 1750) will
1870  * not respond to a Get SDR command if the length of the data
1871  * requested is exactly 0x3A, which leads to command timeouts and no
1872  * data returned.  This intercepts such commands, and causes userspace
1873  * callers to try again with a different-sized buffer, which succeeds.
1874  */
1875
1876 #define STORAGE_NETFN 0x0A
1877 #define STORAGE_CMD_GET_SDR 0x23
1878 static int dell_poweredge_bt_xaction_handler(struct notifier_block *self,
1879                                              unsigned long unused,
1880                                              void *in)
1881 {
1882         struct smi_info *smi_info = in;
1883         unsigned char *data = smi_info->curr_msg->data;
1884         unsigned int size   = smi_info->curr_msg->data_size;
1885         if (size >= 8 &&
1886             (data[0]>>2) == STORAGE_NETFN &&
1887             data[1] == STORAGE_CMD_GET_SDR &&
1888             data[7] == 0x3A) {
1889                 return_hosed_msg_badsize(smi_info);
1890                 return NOTIFY_STOP;
1891         }
1892         return NOTIFY_DONE;
1893 }
1894
1895 static struct notifier_block dell_poweredge_bt_xaction_notifier = {
1896         .notifier_call  = dell_poweredge_bt_xaction_handler,
1897 };
1898
1899 /*
1900  * setup_dell_poweredge_bt_xaction_handler
1901  * @info - smi_info.device_id must be filled in already
1902  *
1903  * Fills in smi_info.device_id.start_transaction_pre_hook
1904  * when we know what function to use there.
1905  */
1906 static void
1907 setup_dell_poweredge_bt_xaction_handler(struct smi_info *smi_info)
1908 {
1909         struct ipmi_device_id *id = &smi_info->device_id;
1910         if (id->manufacturer_id == DELL_IANA_MFR_ID &&
1911             smi_info->io.si_type == SI_BT)
1912                 register_xaction_notifier(&dell_poweredge_bt_xaction_notifier);
1913 }
1914
1915 /*
1916  * setup_oem_data_handler
1917  * @info - smi_info.device_id must be filled in already
1918  *
1919  * Fills in smi_info.device_id.oem_data_available_handler
1920  * when we know what function to use there.
1921  */
1922
1923 static void setup_oem_data_handler(struct smi_info *smi_info)
1924 {
1925         setup_dell_poweredge_oem_data_handler(smi_info);
1926 }
1927
1928 static void setup_xaction_handlers(struct smi_info *smi_info)
1929 {
1930         setup_dell_poweredge_bt_xaction_handler(smi_info);
1931 }
1932
1933 static void check_for_broken_irqs(struct smi_info *smi_info)
1934 {
1935         check_clr_rcv_irq(smi_info);
1936         check_set_rcv_irq(smi_info);
1937 }
1938
1939 static inline void stop_timer_and_thread(struct smi_info *smi_info)
1940 {
1941         if (smi_info->thread != NULL)
1942                 kthread_stop(smi_info->thread);
1943
1944         smi_info->timer_can_start = false;
1945         if (smi_info->timer_running)
1946                 del_timer_sync(&smi_info->si_timer);
1947 }
1948
1949 static struct smi_info *find_dup_si(struct smi_info *info)
1950 {
1951         struct smi_info *e;
1952
1953         list_for_each_entry(e, &smi_infos, link) {
1954                 if (e->io.addr_type != info->io.addr_type)
1955                         continue;
1956                 if (e->io.addr_data == info->io.addr_data) {
1957                         /*
1958                          * This is a cheap hack, ACPI doesn't have a defined
1959                          * slave address but SMBIOS does.  Pick it up from
1960                          * any source that has it available.
1961                          */
1962                         if (info->io.slave_addr && !e->io.slave_addr)
1963                                 e->io.slave_addr = info->io.slave_addr;
1964                         return e;
1965                 }
1966         }
1967
1968         return NULL;
1969 }
1970
1971 int ipmi_si_add_smi(struct si_sm_io *io)
1972 {
1973         int rv = 0;
1974         struct smi_info *new_smi, *dup;
1975
1976         if (!io->io_setup) {
1977                 if (io->addr_type == IPMI_IO_ADDR_SPACE) {
1978                         io->io_setup = ipmi_si_port_setup;
1979                 } else if (io->addr_type == IPMI_MEM_ADDR_SPACE) {
1980                         io->io_setup = ipmi_si_mem_setup;
1981                 } else {
1982                         return -EINVAL;
1983                 }
1984         }
1985
1986         new_smi = kzalloc(sizeof(*new_smi), GFP_KERNEL);
1987         if (!new_smi)
1988                 return -ENOMEM;
1989         spin_lock_init(&new_smi->si_lock);
1990
1991         new_smi->io = *io;
1992
1993         mutex_lock(&smi_infos_lock);
1994         dup = find_dup_si(new_smi);
1995         if (dup) {
1996                 if (new_smi->io.addr_source == SI_ACPI &&
1997                     dup->io.addr_source == SI_SMBIOS) {
1998                         /* We prefer ACPI over SMBIOS. */
1999                         dev_info(dup->io.dev,
2000                                  "Removing SMBIOS-specified %s state machine in favor of ACPI\n",
2001                                  si_to_str[new_smi->io.si_type]);
2002                         cleanup_one_si(dup);
2003                 } else {
2004                         dev_info(new_smi->io.dev,
2005                                  "%s-specified %s state machine: duplicate\n",
2006                                  ipmi_addr_src_to_str(new_smi->io.addr_source),
2007                                  si_to_str[new_smi->io.si_type]);
2008                         rv = -EBUSY;
2009                         kfree(new_smi);
2010                         goto out_err;
2011                 }
2012         }
2013
2014         pr_info(PFX "Adding %s-specified %s state machine\n",
2015                 ipmi_addr_src_to_str(new_smi->io.addr_source),
2016                 si_to_str[new_smi->io.si_type]);
2017
2018         /* So we know not to free it unless we have allocated one. */
2019         new_smi->intf = NULL;
2020         new_smi->si_sm = NULL;
2021         new_smi->handlers = NULL;
2022
2023         list_add_tail(&new_smi->link, &smi_infos);
2024
2025         if (initialized) {
2026                 rv = try_smi_init(new_smi);
2027                 if (rv) {
2028                         mutex_unlock(&smi_infos_lock);
2029                         cleanup_one_si(new_smi);
2030                         return rv;
2031                 }
2032         }
2033 out_err:
2034         mutex_unlock(&smi_infos_lock);
2035         return rv;
2036 }
2037
2038 /*
2039  * Try to start up an interface.  Must be called with smi_infos_lock
2040  * held, primarily to keep smi_num consistent, we only one to do these
2041  * one at a time.
2042  */
2043 static int try_smi_init(struct smi_info *new_smi)
2044 {
2045         int rv = 0;
2046         int i;
2047         char *init_name = NULL;
2048
2049         pr_info(PFX "Trying %s-specified %s state machine at %s address 0x%lx, slave address 0x%x, irq %d\n",
2050                 ipmi_addr_src_to_str(new_smi->io.addr_source),
2051                 si_to_str[new_smi->io.si_type],
2052                 addr_space_to_str[new_smi->io.addr_type],
2053                 new_smi->io.addr_data,
2054                 new_smi->io.slave_addr, new_smi->io.irq);
2055
2056         switch (new_smi->io.si_type) {
2057         case SI_KCS:
2058                 new_smi->handlers = &kcs_smi_handlers;
2059                 break;
2060
2061         case SI_SMIC:
2062                 new_smi->handlers = &smic_smi_handlers;
2063                 break;
2064
2065         case SI_BT:
2066                 new_smi->handlers = &bt_smi_handlers;
2067                 break;
2068
2069         default:
2070                 /* No support for anything else yet. */
2071                 rv = -EIO;
2072                 goto out_err;
2073         }
2074
2075         new_smi->intf_num = smi_num;
2076
2077         /* Do this early so it's available for logs. */
2078         if (!new_smi->io.dev) {
2079                 init_name = kasprintf(GFP_KERNEL, "ipmi_si.%d",
2080                                       new_smi->intf_num);
2081
2082                 /*
2083                  * If we don't already have a device from something
2084                  * else (like PCI), then register a new one.
2085                  */
2086                 new_smi->pdev = platform_device_alloc("ipmi_si",
2087                                                       new_smi->intf_num);
2088                 if (!new_smi->pdev) {
2089                         pr_err(PFX "Unable to allocate platform device\n");
2090                         goto out_err;
2091                 }
2092                 new_smi->io.dev = &new_smi->pdev->dev;
2093                 new_smi->io.dev->driver = &ipmi_platform_driver.driver;
2094                 /* Nulled by device_add() */
2095                 new_smi->io.dev->init_name = init_name;
2096         }
2097
2098         /* Allocate the state machine's data and initialize it. */
2099         new_smi->si_sm = kmalloc(new_smi->handlers->size(), GFP_KERNEL);
2100         if (!new_smi->si_sm) {
2101                 rv = -ENOMEM;
2102                 goto out_err;
2103         }
2104         new_smi->io.io_size = new_smi->handlers->init_data(new_smi->si_sm,
2105                                                            &new_smi->io);
2106
2107         /* Now that we know the I/O size, we can set up the I/O. */
2108         rv = new_smi->io.io_setup(&new_smi->io);
2109         if (rv) {
2110                 dev_err(new_smi->io.dev, "Could not set up I/O space\n");
2111                 goto out_err;
2112         }
2113
2114         /* Do low-level detection first. */
2115         if (new_smi->handlers->detect(new_smi->si_sm)) {
2116                 if (new_smi->io.addr_source)
2117                         dev_err(new_smi->io.dev,
2118                                 "Interface detection failed\n");
2119                 rv = -ENODEV;
2120                 goto out_err;
2121         }
2122
2123         /*
2124          * Attempt a get device id command.  If it fails, we probably
2125          * don't have a BMC here.
2126          */
2127         rv = try_get_dev_id(new_smi);
2128         if (rv) {
2129                 if (new_smi->io.addr_source)
2130                         dev_err(new_smi->io.dev,
2131                                "There appears to be no BMC at this location\n");
2132                 goto out_err;
2133         }
2134
2135         setup_oem_data_handler(new_smi);
2136         setup_xaction_handlers(new_smi);
2137         check_for_broken_irqs(new_smi);
2138
2139         new_smi->waiting_msg = NULL;
2140         new_smi->curr_msg = NULL;
2141         atomic_set(&new_smi->req_events, 0);
2142         new_smi->run_to_completion = false;
2143         for (i = 0; i < SI_NUM_STATS; i++)
2144                 atomic_set(&new_smi->stats[i], 0);
2145
2146         new_smi->interrupt_disabled = true;
2147         atomic_set(&new_smi->need_watch, 0);
2148
2149         rv = try_enable_event_buffer(new_smi);
2150         if (rv == 0)
2151                 new_smi->has_event_buffer = true;
2152
2153         /*
2154          * Start clearing the flags before we enable interrupts or the
2155          * timer to avoid racing with the timer.
2156          */
2157         start_clear_flags(new_smi);
2158
2159         /*
2160          * IRQ is defined to be set when non-zero.  req_events will
2161          * cause a global flags check that will enable interrupts.
2162          */
2163         if (new_smi->io.irq) {
2164                 new_smi->interrupt_disabled = false;
2165                 atomic_set(&new_smi->req_events, 1);
2166         }
2167
2168         if (new_smi->pdev) {
2169                 rv = platform_device_add(new_smi->pdev);
2170                 if (rv) {
2171                         dev_err(new_smi->io.dev,
2172                                 "Unable to register system interface device: %d\n",
2173                                 rv);
2174                         goto out_err;
2175                 }
2176         }
2177
2178         dev_set_drvdata(new_smi->io.dev, new_smi);
2179         rv = device_add_group(new_smi->io.dev, &ipmi_si_dev_attr_group);
2180         if (rv) {
2181                 dev_err(new_smi->io.dev,
2182                         "Unable to add device attributes: error %d\n",
2183                         rv);
2184                 goto out_err_stop_timer;
2185         }
2186
2187         rv = ipmi_register_smi(&handlers,
2188                                new_smi,
2189                                new_smi->io.dev,
2190                                new_smi->io.slave_addr);
2191         if (rv) {
2192                 dev_err(new_smi->io.dev,
2193                         "Unable to register device: error %d\n",
2194                         rv);
2195                 goto out_err_remove_attrs;
2196         }
2197
2198 #ifdef CONFIG_IPMI_PROC_INTERFACE
2199         rv = ipmi_smi_add_proc_entry(new_smi->intf, "type",
2200                                      &smi_type_proc_ops,
2201                                      new_smi);
2202         if (rv) {
2203                 dev_err(new_smi->io.dev,
2204                         "Unable to create proc entry: %d\n", rv);
2205                 goto out_err_stop_timer;
2206         }
2207
2208         rv = ipmi_smi_add_proc_entry(new_smi->intf, "si_stats",
2209                                      &smi_si_stats_proc_ops,
2210                                      new_smi);
2211         if (rv) {
2212                 dev_err(new_smi->io.dev,
2213                         "Unable to create proc entry: %d\n", rv);
2214                 goto out_err_stop_timer;
2215         }
2216
2217         rv = ipmi_smi_add_proc_entry(new_smi->intf, "params",
2218                                      &smi_params_proc_ops,
2219                                      new_smi);
2220         if (rv) {
2221                 dev_err(new_smi->io.dev,
2222                         "Unable to create proc entry: %d\n", rv);
2223                 goto out_err_stop_timer;
2224         }
2225 #endif
2226
2227         /* Don't increment till we know we have succeeded. */
2228         smi_num++;
2229
2230         dev_info(new_smi->io.dev, "IPMI %s interface initialized\n",
2231                  si_to_str[new_smi->io.si_type]);
2232
2233         WARN_ON(new_smi->io.dev->init_name != NULL);
2234         kfree(init_name);
2235
2236         return 0;
2237
2238 out_err_remove_attrs:
2239         device_remove_group(new_smi->io.dev, &ipmi_si_dev_attr_group);
2240         dev_set_drvdata(new_smi->io.dev, NULL);
2241
2242 out_err_stop_timer:
2243         stop_timer_and_thread(new_smi);
2244
2245 out_err:
2246         new_smi->interrupt_disabled = true;
2247
2248         if (new_smi->intf) {
2249                 ipmi_smi_t intf = new_smi->intf;
2250                 new_smi->intf = NULL;
2251                 ipmi_unregister_smi(intf);
2252         }
2253
2254         if (new_smi->io.irq_cleanup) {
2255                 new_smi->io.irq_cleanup(&new_smi->io);
2256                 new_smi->io.irq_cleanup = NULL;
2257         }
2258
2259         /*
2260          * Wait until we know that we are out of any interrupt
2261          * handlers might have been running before we freed the
2262          * interrupt.
2263          */
2264         synchronize_sched();
2265
2266         if (new_smi->si_sm) {
2267                 if (new_smi->handlers)
2268                         new_smi->handlers->cleanup(new_smi->si_sm);
2269                 kfree(new_smi->si_sm);
2270                 new_smi->si_sm = NULL;
2271         }
2272         if (new_smi->io.addr_source_cleanup) {
2273                 new_smi->io.addr_source_cleanup(&new_smi->io);
2274                 new_smi->io.addr_source_cleanup = NULL;
2275         }
2276         if (new_smi->io.io_cleanup) {
2277                 new_smi->io.io_cleanup(&new_smi->io);
2278                 new_smi->io.io_cleanup = NULL;
2279         }
2280
2281         if (new_smi->pdev) {
2282                 platform_device_unregister(new_smi->pdev);
2283                 new_smi->pdev = NULL;
2284         } else if (new_smi->pdev) {
2285                 platform_device_put(new_smi->pdev);
2286         }
2287
2288         kfree(init_name);
2289
2290         return rv;
2291 }
2292
2293 static int init_ipmi_si(void)
2294 {
2295         struct smi_info *e;
2296         enum ipmi_addr_src type = SI_INVALID;
2297
2298         if (initialized)
2299                 return 0;
2300
2301         pr_info("IPMI System Interface driver.\n");
2302
2303         /* If the user gave us a device, they presumably want us to use it */
2304         if (!ipmi_si_hardcode_find_bmc())
2305                 goto do_scan;
2306
2307         ipmi_si_platform_init();
2308
2309         ipmi_si_pci_init();
2310
2311         ipmi_si_parisc_init();
2312
2313         /* We prefer devices with interrupts, but in the case of a machine
2314            with multiple BMCs we assume that there will be several instances
2315            of a given type so if we succeed in registering a type then also
2316            try to register everything else of the same type */
2317 do_scan:
2318         mutex_lock(&smi_infos_lock);
2319         list_for_each_entry(e, &smi_infos, link) {
2320                 /* Try to register a device if it has an IRQ and we either
2321                    haven't successfully registered a device yet or this
2322                    device has the same type as one we successfully registered */
2323                 if (e->io.irq && (!type || e->io.addr_source == type)) {
2324                         if (!try_smi_init(e)) {
2325                                 type = e->io.addr_source;
2326                         }
2327                 }
2328         }
2329
2330         /* type will only have been set if we successfully registered an si */
2331         if (type)
2332                 goto skip_fallback_noirq;
2333
2334         /* Fall back to the preferred device */
2335
2336         list_for_each_entry(e, &smi_infos, link) {
2337                 if (!e->io.irq && (!type || e->io.addr_source == type)) {
2338                         if (!try_smi_init(e)) {
2339                                 type = e->io.addr_source;
2340                         }
2341                 }
2342         }
2343
2344 skip_fallback_noirq:
2345         initialized = 1;
2346         mutex_unlock(&smi_infos_lock);
2347
2348         if (type)
2349                 return 0;
2350
2351         mutex_lock(&smi_infos_lock);
2352         if (unload_when_empty && list_empty(&smi_infos)) {
2353                 mutex_unlock(&smi_infos_lock);
2354                 cleanup_ipmi_si();
2355                 pr_warn(PFX "Unable to find any System Interface(s)\n");
2356                 return -ENODEV;
2357         } else {
2358                 mutex_unlock(&smi_infos_lock);
2359                 return 0;
2360         }
2361 }
2362 module_init(init_ipmi_si);
2363
2364 static void cleanup_one_si(struct smi_info *to_clean)
2365 {
2366         int           rv = 0;
2367
2368         if (!to_clean)
2369                 return;
2370
2371         if (to_clean->intf) {
2372                 ipmi_smi_t intf = to_clean->intf;
2373
2374                 to_clean->intf = NULL;
2375                 rv = ipmi_unregister_smi(intf);
2376                 if (rv) {
2377                         pr_err(PFX "Unable to unregister device: errno=%d\n",
2378                                rv);
2379                 }
2380         }
2381
2382         device_remove_group(to_clean->io.dev, &ipmi_si_dev_attr_group);
2383         dev_set_drvdata(to_clean->io.dev, NULL);
2384
2385         list_del(&to_clean->link);
2386
2387         /*
2388          * Make sure that interrupts, the timer and the thread are
2389          * stopped and will not run again.
2390          */
2391         if (to_clean->io.irq_cleanup)
2392                 to_clean->io.irq_cleanup(&to_clean->io);
2393         stop_timer_and_thread(to_clean);
2394
2395         /*
2396          * Timeouts are stopped, now make sure the interrupts are off
2397          * in the BMC.  Note that timers and CPU interrupts are off,
2398          * so no need for locks.
2399          */
2400         while (to_clean->curr_msg || (to_clean->si_state != SI_NORMAL)) {
2401                 poll(to_clean);
2402                 schedule_timeout_uninterruptible(1);
2403         }
2404         if (to_clean->handlers)
2405                 disable_si_irq(to_clean);
2406         while (to_clean->curr_msg || (to_clean->si_state != SI_NORMAL)) {
2407                 poll(to_clean);
2408                 schedule_timeout_uninterruptible(1);
2409         }
2410
2411         if (to_clean->handlers)
2412                 to_clean->handlers->cleanup(to_clean->si_sm);
2413
2414         kfree(to_clean->si_sm);
2415
2416         if (to_clean->io.addr_source_cleanup)
2417                 to_clean->io.addr_source_cleanup(&to_clean->io);
2418         if (to_clean->io.io_cleanup)
2419                 to_clean->io.io_cleanup(&to_clean->io);
2420
2421         if (to_clean->pdev)
2422                 platform_device_unregister(to_clean->pdev);
2423
2424         kfree(to_clean);
2425 }
2426
2427 int ipmi_si_remove_by_dev(struct device *dev)
2428 {
2429         struct smi_info *e;
2430         int rv = -ENOENT;
2431
2432         mutex_lock(&smi_infos_lock);
2433         list_for_each_entry(e, &smi_infos, link) {
2434                 if (e->io.dev == dev) {
2435                         cleanup_one_si(e);
2436                         rv = 0;
2437                         break;
2438                 }
2439         }
2440         mutex_unlock(&smi_infos_lock);
2441
2442         return rv;
2443 }
2444
2445 void ipmi_si_remove_by_data(int addr_space, enum si_type si_type,
2446                             unsigned long addr)
2447 {
2448         /* remove */
2449         struct smi_info *e, *tmp_e;
2450
2451         mutex_lock(&smi_infos_lock);
2452         list_for_each_entry_safe(e, tmp_e, &smi_infos, link) {
2453                 if (e->io.addr_type != addr_space)
2454                         continue;
2455                 if (e->io.si_type != si_type)
2456                         continue;
2457                 if (e->io.addr_data == addr)
2458                         cleanup_one_si(e);
2459         }
2460         mutex_unlock(&smi_infos_lock);
2461 }
2462
2463 static void cleanup_ipmi_si(void)
2464 {
2465         struct smi_info *e, *tmp_e;
2466
2467         if (!initialized)
2468                 return;
2469
2470         ipmi_si_pci_shutdown();
2471
2472         ipmi_si_parisc_shutdown();
2473
2474         ipmi_si_platform_shutdown();
2475
2476         mutex_lock(&smi_infos_lock);
2477         list_for_each_entry_safe(e, tmp_e, &smi_infos, link)
2478                 cleanup_one_si(e);
2479         mutex_unlock(&smi_infos_lock);
2480 }
2481 module_exit(cleanup_ipmi_si);
2482
2483 MODULE_ALIAS("platform:dmi-ipmi-si");
2484 MODULE_LICENSE("GPL");
2485 MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>");
2486 MODULE_DESCRIPTION("Interface to the IPMI driver for the KCS, SMIC, and BT"
2487                    " system interfaces.");