drivers/char/ipmi/ipmi_si_intf.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * ipmi_si.c
   4  *
   5  * The interface to the IPMI driver for the system interfaces (KCS, SMIC,
   6  * BT).
   7  *
   8  * Author: MontaVista Software, Inc.
   9  *         Corey Minyard <minyard@mvista.com>
  10  *         source@mvista.com
  11  *
  12  * Copyright 2002 MontaVista Software Inc.
  13  * Copyright 2006 IBM Corp., Christian Krafft <krafft@de.ibm.com>
  14  */
  15
  16 /*
  17  * This file holds the "policy" for the interface to the SMI state
  18  * machine.  It does the configuration, handles timers and interrupts,
  19  * and drives the real SMI state machine.
  20  */
  21
  22 #include <linux/module.h>
  23 #include <linux/moduleparam.h>
  24 #include <linux/sched.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/timer.h>
  27 #include <linux/errno.h>
  28 #include <linux/spinlock.h>
  29 #include <linux/slab.h>
  30 #include <linux/delay.h>
  31 #include <linux/list.h>
  32 #include <linux/notifier.h>
  33 #include <linux/mutex.h>
  34 #include <linux/kthread.h>
  35 #include <asm/irq.h>
  36 #include <linux/interrupt.h>
  37 #include <linux/rcupdate.h>
  38 #include <linux/ipmi.h>
  39 #include <linux/ipmi_smi.h>
  40 #include "ipmi_si.h"
  41 #include <linux/string.h>
  42 #include <linux/ctype.h>
  43
  44 #define PFX "ipmi_si: "
  45
  46 /* Measure times between events in the driver. */
  47 #undef DEBUG_TIMING
  48
  49 /* Call every 10 ms. */
  50 #define SI_TIMEOUT_TIME_USEC    10000
  51 #define SI_USEC_PER_JIFFY       (1000000/HZ)
  52 #define SI_TIMEOUT_JIFFIES      (SI_TIMEOUT_TIME_USEC/SI_USEC_PER_JIFFY)
  53 #define SI_SHORT_TIMEOUT_USEC  250 /* .25ms when the SM request a
  54                                       short timeout */
  55
  56 enum si_intf_state {
  57         SI_NORMAL,
  58         SI_GETTING_FLAGS,
  59         SI_GETTING_EVENTS,
  60         SI_CLEARING_FLAGS,
  61         SI_GETTING_MESSAGES,
  62         SI_CHECKING_ENABLES,
  63         SI_SETTING_ENABLES
  64         /* FIXME - add watchdog stuff. */
  65 };
  66
  67 /* Some BT-specific defines we need here. */
  68 #define IPMI_BT_INTMASK_REG             2
  69 #define IPMI_BT_INTMASK_CLEAR_IRQ_BIT   2
  70 #define IPMI_BT_INTMASK_ENABLE_IRQ_BIT  1
  71
  72 static const char * const si_to_str[] = { "invalid", "kcs", "smic", "bt" };
  73
  74 static int initialized;
  75
  76 /*
  77  * Indexes into stats[] in smi_info below.
  78  */
  79 enum si_stat_indexes {
  80         /*
  81          * Number of times the driver requested a timer while an operation
  82          * was in progress.
  83          */
  84         SI_STAT_short_timeouts = 0,
  85
  86         /*
  87          * Number of times the driver requested a timer while nothing was in
  88          * progress.
  89          */
  90         SI_STAT_long_timeouts,
  91
  92         /* Number of times the interface was idle while being polled. */
  93         SI_STAT_idles,
  94
  95         /* Number of interrupts the driver handled. */
  96         SI_STAT_interrupts,
  97
  98         /* Number of time the driver got an ATTN from the hardware. */
  99         SI_STAT_attentions,
 100
 101         /* Number of times the driver requested flags from the hardware. */
 102         SI_STAT_flag_fetches,
 103
 104         /* Number of times the hardware didn't follow the state machine. */
 105         SI_STAT_hosed_count,
 106
 107         /* Number of completed messages. */
 108         SI_STAT_complete_transactions,
 109
 110         /* Number of IPMI events received from the hardware. */
 111         SI_STAT_events,
 112
 113         /* Number of watchdog pretimeouts. */
 114         SI_STAT_watchdog_pretimeouts,
 115
 116         /* Number of asynchronous messages received. */
 117         SI_STAT_incoming_messages,
 118
 119
 120         /* This *must* remain last, add new values above this. */
 121         SI_NUM_STATS
 122 };
 123
 124 struct smi_info {
 125         int                    si_num;
 126         struct ipmi_smi        *intf;
 127         struct si_sm_data      *si_sm;
 128         const struct si_sm_handlers *handlers;
 129         spinlock_t             si_lock;
 130         struct ipmi_smi_msg    *waiting_msg;
 131         struct ipmi_smi_msg    *curr_msg;
 132         enum si_intf_state     si_state;
 133
 134         /*
 135          * Used to handle the various types of I/O that can occur with
 136          * IPMI
 137          */
 138         struct si_sm_io io;
 139
 140         /*
 141          * Per-OEM handler, called from handle_flags().  Returns 1
 142          * when handle_flags() needs to be re-run or 0 indicating it
 143          * set si_state itself.
 144          */
 145         int (*oem_data_avail_handler)(struct smi_info *smi_info);
 146
 147         /*
 148          * Flags from the last GET_MSG_FLAGS command, used when an ATTN
 149          * is set to hold the flags until we are done handling everything
 150          * from the flags.
 151          */
 152 #define RECEIVE_MSG_AVAIL       0x01
 153 #define EVENT_MSG_BUFFER_FULL   0x02
 154 #define WDT_PRE_TIMEOUT_INT     0x08
 155 #define OEM0_DATA_AVAIL     0x20
 156 #define OEM1_DATA_AVAIL     0x40
 157 #define OEM2_DATA_AVAIL     0x80
 158 #define OEM_DATA_AVAIL      (OEM0_DATA_AVAIL | \
 159                              OEM1_DATA_AVAIL | \
 160                              OEM2_DATA_AVAIL)
 161         unsigned char       msg_flags;
 162
 163         /* Does the BMC have an event buffer? */
 164         bool                has_event_buffer;
 165
 166         /*
 167          * If set to true, this will request events the next time the
 168          * state machine is idle.
 169          */
 170         atomic_t            req_events;
 171
 172         /*
 173          * If true, run the state machine to completion on every send
 174          * call.  Generally used after a panic to make sure stuff goes
 175          * out.
 176          */
 177         bool                run_to_completion;
 178
 179         /* The timer for this si. */
 180         struct timer_list   si_timer;
 181
 182         /* This flag is set, if the timer can be set */
 183         bool                timer_can_start;
 184
 185         /* This flag is set, if the timer is running (timer_pending() isn't enough) */
 186         bool                timer_running;
 187
 188         /* The time (in jiffies) the last timeout occurred at. */
 189         unsigned long       last_timeout_jiffies;
 190
 191         /* Are we waiting for the events, pretimeouts, received msgs? */
 192         atomic_t            need_watch;
 193
 194         /*
 195          * The driver will disable interrupts when it gets into a
 196          * situation where it cannot handle messages due to lack of
 197          * memory.  Once that situation clears up, it will re-enable
 198          * interrupts.
 199          */
 200         bool interrupt_disabled;
 201
 202         /*
 203          * Does the BMC support events?
 204          */
 205         bool supports_event_msg_buff;
 206
 207         /*
 208          * Can we disable interrupts the global enables receive irq
 209          * bit?  There are currently two forms of brokenness, some
 210          * systems cannot disable the bit (which is technically within
 211          * the spec but a bad idea) and some systems have the bit
 212          * forced to zero even though interrupts work (which is
 213          * clearly outside the spec).  The next bool tells which form
 214          * of brokenness is present.
 215          */
 216         bool cannot_disable_irq;
 217
 218         /*
 219          * Some systems are broken and cannot set the irq enable
 220          * bit, even if they support interrupts.
 221          */
 222         bool irq_enable_broken;
 223
 224         /*
 225          * Did we get an attention that we did not handle?
 226          */
 227         bool got_attn;
 228
 229         /* From the get device id response... */
 230         struct ipmi_device_id device_id;
 231
 232         /* Default driver model device. */
 233         struct platform_device *pdev;
 234
 235         /* Have we added the device group to the device? */
 236         bool dev_group_added;
 237
 238         /* Have we added the platform device? */
 239         bool pdev_registered;
 240
 241         /* Counters and things for the proc filesystem. */
 242         atomic_t stats[SI_NUM_STATS];
 243
 244         struct task_struct *thread;
 245
 246         struct list_head link;
 247 };
 248
 249 #define smi_inc_stat(smi, stat) \
 250         atomic_inc(&(smi)->stats[SI_STAT_ ## stat])
 251 #define smi_get_stat(smi, stat) \
 252         ((unsigned int) atomic_read(&(smi)->stats[SI_STAT_ ## stat]))
 253
 254 #define IPMI_MAX_INTFS 4
 255 static int force_kipmid[IPMI_MAX_INTFS];
 256 static int num_force_kipmid;
 257
 258 static unsigned int kipmid_max_busy_us[IPMI_MAX_INTFS];
 259 static int num_max_busy_us;
 260
 261 static bool unload_when_empty = true;
 262
 263 static int try_smi_init(struct smi_info *smi);
 264 static void cleanup_one_si(struct smi_info *smi_info);
 265 static void cleanup_ipmi_si(void);
 266
 267 #ifdef DEBUG_TIMING
 268 void debug_timestamp(char *msg)
 269 {
 270         struct timespec64 t;
 271
 272         getnstimeofday64(&t);
 273         pr_debug("**%s: %lld.%9.9ld\n", msg, (long long) t.tv_sec, t.tv_nsec);
 274 }
 275 #else
 276 #define debug_timestamp(x)
 277 #endif
 278
 279 static ATOMIC_NOTIFIER_HEAD(xaction_notifier_list);
 280 static int register_xaction_notifier(struct notifier_block *nb)
 281 {
 282         return atomic_notifier_chain_register(&xaction_notifier_list, nb);
 283 }
 284
 285 static void deliver_recv_msg(struct smi_info *smi_info,
 286                              struct ipmi_smi_msg *msg)
 287 {
 288         /* Deliver the message to the upper layer. */
 289         ipmi_smi_msg_received(smi_info->intf, msg);
 290 }
 291
 292 static void return_hosed_msg(struct smi_info *smi_info, int cCode)
 293 {
 294         struct ipmi_smi_msg *msg = smi_info->curr_msg;
 295
 296         if (cCode < 0 || cCode > IPMI_ERR_UNSPECIFIED)
 297                 cCode = IPMI_ERR_UNSPECIFIED;
 298         /* else use it as is */
 299
 300         /* Make it a response */
 301         msg->rsp[0] = msg->data[0] | 4;
 302         msg->rsp[1] = msg->data[1];
 303         msg->rsp[2] = cCode;
 304         msg->rsp_size = 3;
 305
 306         smi_info->curr_msg = NULL;
 307         deliver_recv_msg(smi_info, msg);
 308 }
 309
 310 static enum si_sm_result start_next_msg(struct smi_info *smi_info)
 311 {
 312         int              rv;
 313
 314         if (!smi_info->waiting_msg) {
 315                 smi_info->curr_msg = NULL;
 316                 rv = SI_SM_IDLE;
 317         } else {
 318                 int err;
 319
 320                 smi_info->curr_msg = smi_info->waiting_msg;
 321                 smi_info->waiting_msg = NULL;
 322                 debug_timestamp("Start2");
 323                 err = atomic_notifier_call_chain(&xaction_notifier_list,
 324                                 0, smi_info);
 325                 if (err & NOTIFY_STOP_MASK) {
 326                         rv = SI_SM_CALL_WITHOUT_DELAY;
 327                         goto out;
 328                 }
 329                 err = smi_info->handlers->start_transaction(
 330                         smi_info->si_sm,
 331                         smi_info->curr_msg->data,
 332                         smi_info->curr_msg->data_size);
 333                 if (err)
 334                         return_hosed_msg(smi_info, err);
 335
 336                 rv = SI_SM_CALL_WITHOUT_DELAY;
 337         }
 338 out:
 339         return rv;
 340 }
 341
 342 static void smi_mod_timer(struct smi_info *smi_info, unsigned long new_val)
 343 {
 344         if (!smi_info->timer_can_start)
 345                 return;
 346         smi_info->last_timeout_jiffies = jiffies;
 347         mod_timer(&smi_info->si_timer, new_val);
 348         smi_info->timer_running = true;
 349 }
 350
 351 /*
 352  * Start a new message and (re)start the timer and thread.
 353  */
 354 static void start_new_msg(struct smi_info *smi_info, unsigned char *msg,
 355                           unsigned int size)
 356 {
 357         smi_mod_timer(smi_info, jiffies + SI_TIMEOUT_JIFFIES);
 358
 359         if (smi_info->thread)
 360                 wake_up_process(smi_info->thread);
 361
 362         smi_info->handlers->start_transaction(smi_info->si_sm, msg, size);
 363 }
 364
 365 static void start_check_enables(struct smi_info *smi_info)
 366 {
 367         unsigned char msg[2];
 368
 369         msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
 370         msg[1] = IPMI_GET_BMC_GLOBAL_ENABLES_CMD;
 371
 372         start_new_msg(smi_info, msg, 2);
 373         smi_info->si_state = SI_CHECKING_ENABLES;
 374 }
 375
 376 static void start_clear_flags(struct smi_info *smi_info)
 377 {
 378         unsigned char msg[3];
 379
 380         /* Make sure the watchdog pre-timeout flag is not set at startup. */
 381         msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
 382         msg[1] = IPMI_CLEAR_MSG_FLAGS_CMD;
 383         msg[2] = WDT_PRE_TIMEOUT_INT;
 384
 385         start_new_msg(smi_info, msg, 3);
 386         smi_info->si_state = SI_CLEARING_FLAGS;
 387 }
 388
 389 static void start_getting_msg_queue(struct smi_info *smi_info)
 390 {
 391         smi_info->curr_msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
 392         smi_info->curr_msg->data[1] = IPMI_GET_MSG_CMD;
 393         smi_info->curr_msg->data_size = 2;
 394
 395         start_new_msg(smi_info, smi_info->curr_msg->data,
 396                       smi_info->curr_msg->data_size);
 397         smi_info->si_state = SI_GETTING_MESSAGES;
 398 }
 399
 400 static void start_getting_events(struct smi_info *smi_info)
 401 {
 402         smi_info->curr_msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
 403         smi_info->curr_msg->data[1] = IPMI_READ_EVENT_MSG_BUFFER_CMD;
 404         smi_info->curr_msg->data_size = 2;
 405
 406         start_new_msg(smi_info, smi_info->curr_msg->data,
 407                       smi_info->curr_msg->data_size);
 408         smi_info->si_state = SI_GETTING_EVENTS;
 409 }
 410
 411 /*
 412  * When we have a situtaion where we run out of memory and cannot
 413  * allocate messages, we just leave them in the BMC and run the system
 414  * polled until we can allocate some memory.  Once we have some
 415  * memory, we will re-enable the interrupt.
 416  *
 417  * Note that we cannot just use disable_irq(), since the interrupt may
 418  * be shared.
 419  */
 420 static inline bool disable_si_irq(struct smi_info *smi_info)
 421 {
 422         if ((smi_info->io.irq) && (!smi_info->interrupt_disabled)) {
 423                 smi_info->interrupt_disabled = true;
 424                 start_check_enables(smi_info);
 425                 return true;
 426         }
 427         return false;
 428 }
 429
 430 static inline bool enable_si_irq(struct smi_info *smi_info)
 431 {
 432         if ((smi_info->io.irq) && (smi_info->interrupt_disabled)) {
 433                 smi_info->interrupt_disabled = false;
 434                 start_check_enables(smi_info);
 435                 return true;
 436         }
 437         return false;
 438 }
 439
 440 /*
 441  * Allocate a message.  If unable to allocate, start the interrupt
 442  * disable process and return NULL.  If able to allocate but
 443  * interrupts are disabled, free the message and return NULL after
 444  * starting the interrupt enable process.
 445  */
 446 static struct ipmi_smi_msg *alloc_msg_handle_irq(struct smi_info *smi_info)
 447 {
 448         struct ipmi_smi_msg *msg;
 449
 450         msg = ipmi_alloc_smi_msg();
 451         if (!msg) {
 452                 if (!disable_si_irq(smi_info))
 453                         smi_info->si_state = SI_NORMAL;
 454         } else if (enable_si_irq(smi_info)) {
 455                 ipmi_free_smi_msg(msg);
 456                 msg = NULL;
 457         }
 458         return msg;
 459 }
 460
 461 static void handle_flags(struct smi_info *smi_info)
 462 {
 463 retry:
 464         if (smi_info->msg_flags & WDT_PRE_TIMEOUT_INT) {
 465                 /* Watchdog pre-timeout */
 466                 smi_inc_stat(smi_info, watchdog_pretimeouts);
 467
 468                 start_clear_flags(smi_info);
 469                 smi_info->msg_flags &= ~WDT_PRE_TIMEOUT_INT;
 470                 ipmi_smi_watchdog_pretimeout(smi_info->intf);
 471         } else if (smi_info->msg_flags & RECEIVE_MSG_AVAIL) {
 472                 /* Messages available. */
 473                 smi_info->curr_msg = alloc_msg_handle_irq(smi_info);
 474                 if (!smi_info->curr_msg)
 475                         return;
 476
 477                 start_getting_msg_queue(smi_info);
 478         } else if (smi_info->msg_flags & EVENT_MSG_BUFFER_FULL) {
 479                 /* Events available. */
 480                 smi_info->curr_msg = alloc_msg_handle_irq(smi_info);
 481                 if (!smi_info->curr_msg)
 482                         return;
 483
 484                 start_getting_events(smi_info);
 485         } else if (smi_info->msg_flags & OEM_DATA_AVAIL &&
 486                    smi_info->oem_data_avail_handler) {
 487                 if (smi_info->oem_data_avail_handler(smi_info))
 488                         goto retry;
 489         } else
 490                 smi_info->si_state = SI_NORMAL;
 491 }
 492
 493 /*
 494  * Global enables we care about.
 495  */
 496 #define GLOBAL_ENABLES_MASK (IPMI_BMC_EVT_MSG_BUFF | IPMI_BMC_RCV_MSG_INTR | \
 497                              IPMI_BMC_EVT_MSG_INTR)
 498
 499 static u8 current_global_enables(struct smi_info *smi_info, u8 base,
 500                                  bool *irq_on)
 501 {
 502         u8 enables = 0;
 503
 504         if (smi_info->supports_event_msg_buff)
 505                 enables |= IPMI_BMC_EVT_MSG_BUFF;
 506
 507         if (((smi_info->io.irq && !smi_info->interrupt_disabled) ||
 508              smi_info->cannot_disable_irq) &&
 509             !smi_info->irq_enable_broken)
 510                 enables |= IPMI_BMC_RCV_MSG_INTR;
 511
 512         if (smi_info->supports_event_msg_buff &&
 513             smi_info->io.irq && !smi_info->interrupt_disabled &&
 514             !smi_info->irq_enable_broken)
 515                 enables |= IPMI_BMC_EVT_MSG_INTR;
 516
 517         *irq_on = enables & (IPMI_BMC_EVT_MSG_INTR | IPMI_BMC_RCV_MSG_INTR);
 518
 519         return enables;
 520 }
 521
 522 static void check_bt_irq(struct smi_info *smi_info, bool irq_on)
 523 {
 524         u8 irqstate = smi_info->io.inputb(&smi_info->io, IPMI_BT_INTMASK_REG);
 525
 526         irqstate &= IPMI_BT_INTMASK_ENABLE_IRQ_BIT;
 527
 528         if ((bool)irqstate == irq_on)
 529                 return;
 530
 531         if (irq_on)
 532                 smi_info->io.outputb(&smi_info->io, IPMI_BT_INTMASK_REG,
 533                                      IPMI_BT_INTMASK_ENABLE_IRQ_BIT);
 534         else
 535                 smi_info->io.outputb(&smi_info->io, IPMI_BT_INTMASK_REG, 0);
 536 }
 537
 538 static void handle_transaction_done(struct smi_info *smi_info)
 539 {
 540         struct ipmi_smi_msg *msg;
 541
 542         debug_timestamp("Done");
 543         switch (smi_info->si_state) {
 544         case SI_NORMAL:
 545                 if (!smi_info->curr_msg)
 546                         break;
 547
 548                 smi_info->curr_msg->rsp_size
 549                         = smi_info->handlers->get_result(
 550                                 smi_info->si_sm,
 551                                 smi_info->curr_msg->rsp,
 552                                 IPMI_MAX_MSG_LENGTH);
 553
 554                 /*
 555                  * Do this here becase deliver_recv_msg() releases the
 556                  * lock, and a new message can be put in during the
 557                  * time the lock is released.
 558                  */
 559                 msg = smi_info->curr_msg;
 560                 smi_info->curr_msg = NULL;
 561                 deliver_recv_msg(smi_info, msg);
 562                 break;
 563
 564         case SI_GETTING_FLAGS:
 565         {
 566                 unsigned char msg[4];
 567                 unsigned int  len;
 568
 569                 /* We got the flags from the SMI, now handle them. */
 570                 len = smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
 571                 if (msg[2] != 0) {
 572                         /* Error fetching flags, just give up for now. */
 573                         smi_info->si_state = SI_NORMAL;
 574                 } else if (len < 4) {
 575                         /*
 576                          * Hmm, no flags.  That's technically illegal, but
 577                          * don't use uninitialized data.
 578                          */
 579                         smi_info->si_state = SI_NORMAL;
 580                 } else {
 581                         smi_info->msg_flags = msg[3];
 582                         handle_flags(smi_info);
 583                 }
 584                 break;
 585         }
 586
 587         case SI_CLEARING_FLAGS:
 588         {
 589                 unsigned char msg[3];
 590
 591                 /* We cleared the flags. */
 592                 smi_info->handlers->get_result(smi_info->si_sm, msg, 3);
 593                 if (msg[2] != 0) {
 594                         /* Error clearing flags */
 595                         dev_warn(smi_info->io.dev,
 596                                  "Error clearing flags: %2.2x\n", msg[2]);
 597                 }
 598                 smi_info->si_state = SI_NORMAL;
 599                 break;
 600         }
 601
 602         case SI_GETTING_EVENTS:
 603         {
 604                 smi_info->curr_msg->rsp_size
 605                         = smi_info->handlers->get_result(
 606                                 smi_info->si_sm,
 607                                 smi_info->curr_msg->rsp,
 608                                 IPMI_MAX_MSG_LENGTH);
 609
 610                 /*
 611                  * Do this here becase deliver_recv_msg() releases the
 612                  * lock, and a new message can be put in during the
 613                  * time the lock is released.
 614                  */
 615                 msg = smi_info->curr_msg;
 616                 smi_info->curr_msg = NULL;
 617                 if (msg->rsp[2] != 0) {
 618                         /* Error getting event, probably done. */
 619                         msg->done(msg);
 620
 621                         /* Take off the event flag. */
 622                         smi_info->msg_flags &= ~EVENT_MSG_BUFFER_FULL;
 623                         handle_flags(smi_info);
 624                 } else {
 625                         smi_inc_stat(smi_info, events);
 626
 627                         /*
 628                          * Do this before we deliver the message
 629                          * because delivering the message releases the
 630                          * lock and something else can mess with the
 631                          * state.
 632                          */
 633                         handle_flags(smi_info);
 634
 635                         deliver_recv_msg(smi_info, msg);
 636                 }
 637                 break;
 638         }
 639
 640         case SI_GETTING_MESSAGES:
 641         {
 642                 smi_info->curr_msg->rsp_size
 643                         = smi_info->handlers->get_result(
 644                                 smi_info->si_sm,
 645                                 smi_info->curr_msg->rsp,
 646                                 IPMI_MAX_MSG_LENGTH);
 647
 648                 /*
 649                  * Do this here becase deliver_recv_msg() releases the
 650                  * lock, and a new message can be put in during the
 651                  * time the lock is released.
 652                  */
 653                 msg = smi_info->curr_msg;
 654                 smi_info->curr_msg = NULL;
 655                 if (msg->rsp[2] != 0) {
 656                         /* Error getting event, probably done. */
 657                         msg->done(msg);
 658
 659                         /* Take off the msg flag. */
 660                         smi_info->msg_flags &= ~RECEIVE_MSG_AVAIL;
 661                         handle_flags(smi_info);
 662                 } else {
 663                         smi_inc_stat(smi_info, incoming_messages);
 664
 665                         /*
 666                          * Do this before we deliver the message
 667                          * because delivering the message releases the
 668                          * lock and something else can mess with the
 669                          * state.
 670                          */
 671                         handle_flags(smi_info);
 672
 673                         deliver_recv_msg(smi_info, msg);
 674                 }
 675                 break;
 676         }
 677
 678         case SI_CHECKING_ENABLES:
 679         {
 680                 unsigned char msg[4];
 681                 u8 enables;
 682                 bool irq_on;
 683
 684                 /* We got the flags from the SMI, now handle them. */
 685                 smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
 686                 if (msg[2] != 0) {
 687                         dev_warn(smi_info->io.dev,
 688                                  "Couldn't get irq info: %x.\n", msg[2]);
 689                         dev_warn(smi_info->io.dev,
 690                                  "Maybe ok, but ipmi might run very slowly.\n");
 691                         smi_info->si_state = SI_NORMAL;
 692                         break;
 693                 }
 694                 enables = current_global_enables(smi_info, 0, &irq_on);
 695                 if (smi_info->io.si_type == SI_BT)
 696                         /* BT has its own interrupt enable bit. */
 697                         check_bt_irq(smi_info, irq_on);
 698                 if (enables != (msg[3] & GLOBAL_ENABLES_MASK)) {
 699                         /* Enables are not correct, fix them. */
 700                         msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
 701                         msg[1] = IPMI_SET_BMC_GLOBAL_ENABLES_CMD;
 702                         msg[2] = enables | (msg[3] & ~GLOBAL_ENABLES_MASK);
 703                         smi_info->handlers->start_transaction(
 704                                 smi_info->si_sm, msg, 3);
 705                         smi_info->si_state = SI_SETTING_ENABLES;
 706                 } else if (smi_info->supports_event_msg_buff) {
 707                         smi_info->curr_msg = ipmi_alloc_smi_msg();
 708                         if (!smi_info->curr_msg) {
 709                                 smi_info->si_state = SI_NORMAL;
 710                                 break;
 711                         }
 712                         start_getting_events(smi_info);
 713                 } else {
 714                         smi_info->si_state = SI_NORMAL;
 715                 }
 716                 break;
 717         }
 718
 719         case SI_SETTING_ENABLES:
 720         {
 721                 unsigned char msg[4];
 722
 723                 smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
 724                 if (msg[2] != 0)
 725                         dev_warn(smi_info->io.dev,
 726                                  "Could not set the global enables: 0x%x.\n",
 727                                  msg[2]);
 728
 729                 if (smi_info->supports_event_msg_buff) {
 730                         smi_info->curr_msg = ipmi_alloc_smi_msg();
 731                         if (!smi_info->curr_msg) {
 732                                 smi_info->si_state = SI_NORMAL;
 733                                 break;
 734                         }
 735                         start_getting_events(smi_info);
 736                 } else {
 737                         smi_info->si_state = SI_NORMAL;
 738                 }
 739                 break;
 740         }
 741         }
 742 }
 743
 744 /*
 745  * Called on timeouts and events.  Timeouts should pass the elapsed
 746  * time, interrupts should pass in zero.  Must be called with
 747  * si_lock held and interrupts disabled.
 748  */
 749 static enum si_sm_result smi_event_handler(struct smi_info *smi_info,
 750                                            int time)
 751 {
 752         enum si_sm_result si_sm_result;
 753
 754 restart:
 755         /*
 756          * There used to be a loop here that waited a little while
 757          * (around 25us) before giving up.  That turned out to be
 758          * pointless, the minimum delays I was seeing were in the 300us
 759          * range, which is far too long to wait in an interrupt.  So
 760          * we just run until the state machine tells us something
 761          * happened or it needs a delay.
 762          */
 763         si_sm_result = smi_info->handlers->event(smi_info->si_sm, time);
 764         time = 0;
 765         while (si_sm_result == SI_SM_CALL_WITHOUT_DELAY)
 766                 si_sm_result = smi_info->handlers->event(smi_info->si_sm, 0);
 767
 768         if (si_sm_result == SI_SM_TRANSACTION_COMPLETE) {
 769                 smi_inc_stat(smi_info, complete_transactions);
 770
 771                 handle_transaction_done(smi_info);
 772                 goto restart;
 773         } else if (si_sm_result == SI_SM_HOSED) {
 774                 smi_inc_stat(smi_info, hosed_count);
 775
 776                 /*
 777                  * Do the before return_hosed_msg, because that
 778                  * releases the lock.
 779                  */
 780                 smi_info->si_state = SI_NORMAL;
 781                 if (smi_info->curr_msg != NULL) {
 782                         /*
 783                          * If we were handling a user message, format
 784                          * a response to send to the upper layer to
 785                          * tell it about the error.
 786                          */
 787                         return_hosed_msg(smi_info, IPMI_ERR_UNSPECIFIED);
 788                 }
 789                 goto restart;
 790         }
 791
 792         /*
 793          * We prefer handling attn over new messages.  But don't do
 794          * this if there is not yet an upper layer to handle anything.
 795          */
 796         if (si_sm_result == SI_SM_ATTN || smi_info->got_attn) {
 797                 unsigned char msg[2];
 798
 799                 if (smi_info->si_state != SI_NORMAL) {
 800                         /*
 801                          * We got an ATTN, but we are doing something else.
 802                          * Handle the ATTN later.
 803                          */
 804                         smi_info->got_attn = true;
 805                 } else {
 806                         smi_info->got_attn = false;
 807                         smi_inc_stat(smi_info, attentions);
 808
 809                         /*
 810                          * Got a attn, send down a get message flags to see
 811                          * what's causing it.  It would be better to handle
 812                          * this in the upper layer, but due to the way
 813                          * interrupts work with the SMI, that's not really
 814                          * possible.
 815                          */
 816                         msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
 817                         msg[1] = IPMI_GET_MSG_FLAGS_CMD;
 818
 819                         start_new_msg(smi_info, msg, 2);
 820                         smi_info->si_state = SI_GETTING_FLAGS;
 821                         goto restart;
 822                 }
 823         }
 824
 825         /* If we are currently idle, try to start the next message. */
 826         if (si_sm_result == SI_SM_IDLE) {
 827                 smi_inc_stat(smi_info, idles);
 828
 829                 si_sm_result = start_next_msg(smi_info);
 830                 if (si_sm_result != SI_SM_IDLE)
 831                         goto restart;
 832         }
 833
 834         if ((si_sm_result == SI_SM_IDLE)
 835             && (atomic_read(&smi_info->req_events))) {
 836                 /*
 837                  * We are idle and the upper layer requested that I fetch
 838                  * events, so do so.
 839                  */
 840                 atomic_set(&smi_info->req_events, 0);
 841
 842                 /*
 843                  * Take this opportunity to check the interrupt and
 844                  * message enable state for the BMC.  The BMC can be
 845                  * asynchronously reset, and may thus get interrupts
 846                  * disable and messages disabled.
 847                  */
 848                 if (smi_info->supports_event_msg_buff || smi_info->io.irq) {
 849                         start_check_enables(smi_info);
 850                 } else {
 851                         smi_info->curr_msg = alloc_msg_handle_irq(smi_info);
 852                         if (!smi_info->curr_msg)
 853                                 goto out;
 854
 855                         start_getting_events(smi_info);
 856                 }
 857                 goto restart;
 858         }
 859
 860         if (si_sm_result == SI_SM_IDLE && smi_info->timer_running) {
 861                 /* Ok it if fails, the timer will just go off. */
 862                 if (del_timer(&smi_info->si_timer))
 863                         smi_info->timer_running = false;
 864         }
 865
 866 out:
 867         return si_sm_result;
 868 }
 869
 870 static void check_start_timer_thread(struct smi_info *smi_info)
 871 {
 872         if (smi_info->si_state == SI_NORMAL && smi_info->curr_msg == NULL) {
 873                 smi_mod_timer(smi_info, jiffies + SI_TIMEOUT_JIFFIES);
 874
 875                 if (smi_info->thread)
 876                         wake_up_process(smi_info->thread);
 877
 878                 start_next_msg(smi_info);
 879                 smi_event_handler(smi_info, 0);
 880         }
 881 }
 882
 883 static void flush_messages(void *send_info)
 884 {
 885         struct smi_info *smi_info = send_info;
 886         enum si_sm_result result;
 887
 888         /*
 889          * Currently, this function is called only in run-to-completion
 890          * mode.  This means we are single-threaded, no need for locks.
 891          */
 892         result = smi_event_handler(smi_info, 0);
 893         while (result != SI_SM_IDLE) {
 894                 udelay(SI_SHORT_TIMEOUT_USEC);
 895                 result = smi_event_handler(smi_info, SI_SHORT_TIMEOUT_USEC);
 896         }
 897 }
 898
 899 static void sender(void                *send_info,
 900                    struct ipmi_smi_msg *msg)
 901 {
 902         struct smi_info   *smi_info = send_info;
 903         unsigned long     flags;
 904
 905         debug_timestamp("Enqueue");
 906
 907         if (smi_info->run_to_completion) {
 908                 /*
 909                  * If we are running to completion, start it.  Upper
 910                  * layer will call flush_messages to clear it out.
 911                  */
 912                 smi_info->waiting_msg = msg;
 913                 return;
 914         }
 915
 916         spin_lock_irqsave(&smi_info->si_lock, flags);
 917         /*
 918          * The following two lines don't need to be under the lock for
 919          * the lock's sake, but they do need SMP memory barriers to
 920          * avoid getting things out of order.  We are already claiming
 921          * the lock, anyway, so just do it under the lock to avoid the
 922          * ordering problem.
 923          */
 924         BUG_ON(smi_info->waiting_msg);
 925         smi_info->waiting_msg = msg;
 926         check_start_timer_thread(smi_info);
 927         spin_unlock_irqrestore(&smi_info->si_lock, flags);
 928 }
 929
 930 static void set_run_to_completion(void *send_info, bool i_run_to_completion)
 931 {
 932         struct smi_info   *smi_info = send_info;
 933
 934         smi_info->run_to_completion = i_run_to_completion;
 935         if (i_run_to_completion)
 936                 flush_messages(smi_info);
 937 }
 938
 939 /*
 940  * Use -1 in the nsec value of the busy waiting timespec to tell that
 941  * we are spinning in kipmid looking for something and not delaying
 942  * between checks
 943  */
 944 static inline void ipmi_si_set_not_busy(struct timespec64 *ts)
 945 {
 946         ts->tv_nsec = -1;
 947 }
 948 static inline int ipmi_si_is_busy(struct timespec64 *ts)
 949 {
 950         return ts->tv_nsec != -1;
 951 }
 952
 953 static inline int ipmi_thread_busy_wait(enum si_sm_result smi_result,
 954                                         const struct smi_info *smi_info,
 955                                         struct timespec64 *busy_until)
 956 {
 957         unsigned int max_busy_us = 0;
 958
 959         if (smi_info->si_num < num_max_busy_us)
 960                 max_busy_us = kipmid_max_busy_us[smi_info->si_num];
 961         if (max_busy_us == 0 || smi_result != SI_SM_CALL_WITH_DELAY)
 962                 ipmi_si_set_not_busy(busy_until);
 963         else if (!ipmi_si_is_busy(busy_until)) {
 964                 getnstimeofday64(busy_until);
 965                 timespec64_add_ns(busy_until, max_busy_us*NSEC_PER_USEC);
 966         } else {
 967                 struct timespec64 now;
 968
 969                 getnstimeofday64(&now);
 970                 if (unlikely(timespec64_compare(&now, busy_until) > 0)) {
 971                         ipmi_si_set_not_busy(busy_until);
 972                         return 0;
 973                 }
 974         }
 975         return 1;
 976 }
 977
 978
 979 /*
 980  * A busy-waiting loop for speeding up IPMI operation.
 981  *
 982  * Lousy hardware makes this hard.  This is only enabled for systems
 983  * that are not BT and do not have interrupts.  It starts spinning
 984  * when an operation is complete or until max_busy tells it to stop
 985  * (if that is enabled).  See the paragraph on kimid_max_busy_us in
 986  * Documentation/IPMI.txt for details.
 987  */
 988 static int ipmi_thread(void *data)
 989 {
 990         struct smi_info *smi_info = data;
 991         unsigned long flags;
 992         enum si_sm_result smi_result;
 993         struct timespec64 busy_until;
 994
 995         ipmi_si_set_not_busy(&busy_until);
 996         set_user_nice(current, MAX_NICE);
 997         while (!kthread_should_stop()) {
 998                 int busy_wait;
 999
1000                 spin_lock_irqsave(&(smi_info->si_lock), flags);
1001                 smi_result = smi_event_handler(smi_info, 0);
1002
1003                 /*
1004                  * If the driver is doing something, there is a possible
1005                  * race with the timer.  If the timer handler see idle,
1006                  * and the thread here sees something else, the timer
1007                  * handler won't restart the timer even though it is
1008                  * required.  So start it here if necessary.
1009                  */
1010                 if (smi_result != SI_SM_IDLE && !smi_info->timer_running)
1011                         smi_mod_timer(smi_info, jiffies + SI_TIMEOUT_JIFFIES);
1012
1013                 spin_unlock_irqrestore(&(smi_info->si_lock), flags);
1014                 busy_wait = ipmi_thread_busy_wait(smi_result, smi_info,
1015                                                   &busy_until);
1016                 if (smi_result == SI_SM_CALL_WITHOUT_DELAY)
1017                         ; /* do nothing */
1018                 else if (smi_result == SI_SM_CALL_WITH_DELAY && busy_wait)
1019                         schedule();
1020                 else if (smi_result == SI_SM_IDLE) {
1021                         if (atomic_read(&smi_info->need_watch)) {
1022                                 schedule_timeout_interruptible(100);
1023                         } else {
1024                                 /* Wait to be woken up when we are needed. */
1025                                 __set_current_state(TASK_INTERRUPTIBLE);
1026                                 schedule();
1027                         }
1028                 } else
1029                         schedule_timeout_interruptible(1);
1030         }
1031         return 0;
1032 }
1033
1034
1035 static void poll(void *send_info)
1036 {
1037         struct smi_info *smi_info = send_info;
1038         unsigned long flags = 0;
1039         bool run_to_completion = smi_info->run_to_completion;
1040
1041         /*
1042          * Make sure there is some delay in the poll loop so we can
1043          * drive time forward and timeout things.
1044          */
1045         udelay(10);
1046         if (!run_to_completion)
1047                 spin_lock_irqsave(&smi_info->si_lock, flags);
1048         smi_event_handler(smi_info, 10);
1049         if (!run_to_completion)
1050                 spin_unlock_irqrestore(&smi_info->si_lock, flags);
1051 }
1052
1053 static void request_events(void *send_info)
1054 {
1055         struct smi_info *smi_info = send_info;
1056
1057         if (!smi_info->has_event_buffer)
1058                 return;
1059
1060         atomic_set(&smi_info->req_events, 1);
1061 }
1062
1063 static void set_need_watch(void *send_info, bool enable)
1064 {
1065         struct smi_info *smi_info = send_info;
1066         unsigned long flags;
1067
1068         atomic_set(&smi_info->need_watch, enable);
1069         spin_lock_irqsave(&smi_info->si_lock, flags);
1070         check_start_timer_thread(smi_info);
1071         spin_unlock_irqrestore(&smi_info->si_lock, flags);
1072 }
1073
1074 static void smi_timeout(struct timer_list *t)
1075 {
1076         struct smi_info   *smi_info = from_timer(smi_info, t, si_timer);
1077         enum si_sm_result smi_result;
1078         unsigned long     flags;
1079         unsigned long     jiffies_now;
1080         long              time_diff;
1081         long              timeout;
1082
1083         spin_lock_irqsave(&(smi_info->si_lock), flags);
1084         debug_timestamp("Timer");
1085
1086         jiffies_now = jiffies;
1087         time_diff = (((long)jiffies_now - (long)smi_info->last_timeout_jiffies)
1088                      * SI_USEC_PER_JIFFY);
1089         smi_result = smi_event_handler(smi_info, time_diff);
1090
1091         if ((smi_info->io.irq) && (!smi_info->interrupt_disabled)) {
1092                 /* Running with interrupts, only do long timeouts. */
1093                 timeout = jiffies + SI_TIMEOUT_JIFFIES;
1094                 smi_inc_stat(smi_info, long_timeouts);
1095                 goto do_mod_timer;
1096         }
1097
1098         /*
1099          * If the state machine asks for a short delay, then shorten
1100          * the timer timeout.
1101          */
1102         if (smi_result == SI_SM_CALL_WITH_DELAY) {
1103                 smi_inc_stat(smi_info, short_timeouts);
1104                 timeout = jiffies + 1;
1105         } else {
1106                 smi_inc_stat(smi_info, long_timeouts);
1107                 timeout = jiffies + SI_TIMEOUT_JIFFIES;
1108         }
1109
1110 do_mod_timer:
1111         if (smi_result != SI_SM_IDLE)
1112                 smi_mod_timer(smi_info, timeout);
1113         else
1114                 smi_info->timer_running = false;
1115         spin_unlock_irqrestore(&(smi_info->si_lock), flags);
1116 }
1117
1118 irqreturn_t ipmi_si_irq_handler(int irq, void *data)
1119 {
1120         struct smi_info *smi_info = data;
1121         unsigned long   flags;
1122
1123         if (smi_info->io.si_type == SI_BT)
1124                 /* We need to clear the IRQ flag for the BT interface. */
1125                 smi_info->io.outputb(&smi_info->io, IPMI_BT_INTMASK_REG,
1126                                      IPMI_BT_INTMASK_CLEAR_IRQ_BIT
1127                                      | IPMI_BT_INTMASK_ENABLE_IRQ_BIT);
1128
1129         spin_lock_irqsave(&(smi_info->si_lock), flags);
1130
1131         smi_inc_stat(smi_info, interrupts);
1132
1133         debug_timestamp("Interrupt");
1134
1135         smi_event_handler(smi_info, 0);
1136         spin_unlock_irqrestore(&(smi_info->si_lock), flags);
1137         return IRQ_HANDLED;
1138 }
1139
1140 static int smi_start_processing(void            *send_info,
1141                                 struct ipmi_smi *intf)
1142 {
1143         struct smi_info *new_smi = send_info;
1144         int             enable = 0;
1145
1146         new_smi->intf = intf;
1147
1148         /* Set up the timer that drives the interface. */
1149         timer_setup(&new_smi->si_timer, smi_timeout, 0);
1150         new_smi->timer_can_start = true;
1151         smi_mod_timer(new_smi, jiffies + SI_TIMEOUT_JIFFIES);
1152
1153         /* Try to claim any interrupts. */
1154         if (new_smi->io.irq_setup) {
1155                 new_smi->io.irq_handler_data = new_smi;
1156                 new_smi->io.irq_setup(&new_smi->io);
1157         }
1158
1159         /*
1160          * Check if the user forcefully enabled the daemon.
1161          */
1162         if (new_smi->si_num < num_force_kipmid)
1163                 enable = force_kipmid[new_smi->si_num];
1164         /*
1165          * The BT interface is efficient enough to not need a thread,
1166          * and there is no need for a thread if we have interrupts.
1167          */
1168         else if ((new_smi->io.si_type != SI_BT) && (!new_smi->io.irq))
1169                 enable = 1;
1170
1171         if (enable) {
1172                 new_smi->thread = kthread_run(ipmi_thread, new_smi,
1173                                               "kipmi%d", new_smi->si_num);
1174                 if (IS_ERR(new_smi->thread)) {
1175                         dev_notice(new_smi->io.dev, "Could not start"
1176                                    " kernel thread due to error %ld, only using"
1177                                    " timers to drive the interface\n",
1178                                    PTR_ERR(new_smi->thread));
1179                         new_smi->thread = NULL;
1180                 }
1181         }
1182
1183         return 0;
1184 }
1185
1186 static int get_smi_info(void *send_info, struct ipmi_smi_info *data)
1187 {
1188         struct smi_info *smi = send_info;
1189
1190         data->addr_src = smi->io.addr_source;
1191         data->dev = smi->io.dev;
1192         data->addr_info = smi->io.addr_info;
1193         get_device(smi->io.dev);
1194
1195         return 0;
1196 }
1197
1198 static void set_maintenance_mode(void *send_info, bool enable)
1199 {
1200         struct smi_info   *smi_info = send_info;
1201
1202         if (!enable)
1203                 atomic_set(&smi_info->req_events, 0);
1204 }
1205
1206 static void shutdown_smi(void *send_info);
1207 static const struct ipmi_smi_handlers handlers = {
1208         .owner                  = THIS_MODULE,
1209         .start_processing       = smi_start_processing,
1210         .shutdown               = shutdown_smi,
1211         .get_smi_info           = get_smi_info,
1212         .sender                 = sender,
1213         .request_events         = request_events,
1214         .set_need_watch         = set_need_watch,
1215         .set_maintenance_mode   = set_maintenance_mode,
1216         .set_run_to_completion  = set_run_to_completion,
1217         .flush_messages         = flush_messages,
1218         .poll                   = poll,
1219 };
1220
1221 static LIST_HEAD(smi_infos);
1222 static DEFINE_MUTEX(smi_infos_lock);
1223 static int smi_num; /* Used to sequence the SMIs */
1224
1225 static const char * const addr_space_to_str[] = { "i/o", "mem" };
1226
1227 module_param_array(force_kipmid, int, &num_force_kipmid, 0);
1228 MODULE_PARM_DESC(force_kipmid, "Force the kipmi daemon to be enabled (1) or"
1229                  " disabled(0).  Normally the IPMI driver auto-detects"
1230                  " this, but the value may be overridden by this parm.");
1231 module_param(unload_when_empty, bool, 0);
1232 MODULE_PARM_DESC(unload_when_empty, "Unload the module if no interfaces are"
1233                  " specified or found, default is 1.  Setting to 0"
1234                  " is useful for hot add of devices using hotmod.");
1235 module_param_array(kipmid_max_busy_us, uint, &num_max_busy_us, 0644);
1236 MODULE_PARM_DESC(kipmid_max_busy_us,
1237                  "Max time (in microseconds) to busy-wait for IPMI data before"
1238                  " sleeping. 0 (default) means to wait forever. Set to 100-500"
1239                  " if kipmid is using up a lot of CPU time.");
1240
1241 void ipmi_irq_finish_setup(struct si_sm_io *io)
1242 {
1243         if (io->si_type == SI_BT)
1244                 /* Enable the interrupt in the BT interface. */
1245                 io->outputb(io, IPMI_BT_INTMASK_REG,
1246                             IPMI_BT_INTMASK_ENABLE_IRQ_BIT);
1247 }
1248
1249 void ipmi_irq_start_cleanup(struct si_sm_io *io)
1250 {
1251         if (io->si_type == SI_BT)
1252                 /* Disable the interrupt in the BT interface. */
1253                 io->outputb(io, IPMI_BT_INTMASK_REG, 0);
1254 }
1255
1256 static void std_irq_cleanup(struct si_sm_io *io)
1257 {
1258         ipmi_irq_start_cleanup(io);
1259         free_irq(io->irq, io->irq_handler_data);
1260 }
1261
1262 int ipmi_std_irq_setup(struct si_sm_io *io)
1263 {
1264         int rv;
1265
1266         if (!io->irq)
1267                 return 0;
1268
1269         rv = request_irq(io->irq,
1270                          ipmi_si_irq_handler,
1271                          IRQF_SHARED,
1272                          DEVICE_NAME,
1273                          io->irq_handler_data);
1274         if (rv) {
1275                 dev_warn(io->dev, "%s unable to claim interrupt %d,"
1276                          " running polled\n",
1277                          DEVICE_NAME, io->irq);
1278                 io->irq = 0;
1279         } else {
1280                 io->irq_cleanup = std_irq_cleanup;
1281                 ipmi_irq_finish_setup(io);
1282                 dev_info(io->dev, "Using irq %d\n", io->irq);
1283         }
1284
1285         return rv;
1286 }
1287
1288 static int wait_for_msg_done(struct smi_info *smi_info)
1289 {
1290         enum si_sm_result     smi_result;
1291
1292         smi_result = smi_info->handlers->event(smi_info->si_sm, 0);
1293         for (;;) {
1294                 if (smi_result == SI_SM_CALL_WITH_DELAY ||
1295                     smi_result == SI_SM_CALL_WITH_TICK_DELAY) {
1296                         schedule_timeout_uninterruptible(1);
1297                         smi_result = smi_info->handlers->event(
1298                                 smi_info->si_sm, jiffies_to_usecs(1));
1299                 } else if (smi_result == SI_SM_CALL_WITHOUT_DELAY) {
1300                         smi_result = smi_info->handlers->event(
1301                                 smi_info->si_sm, 0);
1302                 } else
1303                         break;
1304         }
1305         if (smi_result == SI_SM_HOSED)
1306                 /*
1307                  * We couldn't get the state machine to run, so whatever's at
1308                  * the port is probably not an IPMI SMI interface.
1309                  */
1310                 return -ENODEV;
1311
1312         return 0;
1313 }
1314
1315 static int try_get_dev_id(struct smi_info *smi_info)
1316 {
1317         unsigned char         msg[2];
1318         unsigned char         *resp;
1319         unsigned long         resp_len;
1320         int                   rv = 0;
1321
1322         resp = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
1323         if (!resp)
1324                 return -ENOMEM;
1325
1326         /*
1327          * Do a Get Device ID command, since it comes back with some
1328          * useful info.
1329          */
1330         msg[0] = IPMI_NETFN_APP_REQUEST << 2;
1331         msg[1] = IPMI_GET_DEVICE_ID_CMD;
1332         smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
1333
1334         rv = wait_for_msg_done(smi_info);
1335         if (rv)
1336                 goto out;
1337
1338         resp_len = smi_info->handlers->get_result(smi_info->si_sm,
1339                                                   resp, IPMI_MAX_MSG_LENGTH);
1340
1341         /* Check and record info from the get device id, in case we need it. */
1342         rv = ipmi_demangle_device_id(resp[0] >> 2, resp[1],
1343                         resp + 2, resp_len - 2, &smi_info->device_id);
1344
1345 out:
1346         kfree(resp);
1347         return rv;
1348 }
1349
1350 static int get_global_enables(struct smi_info *smi_info, u8 *enables)
1351 {
1352         unsigned char         msg[3];
1353         unsigned char         *resp;
1354         unsigned long         resp_len;
1355         int                   rv;
1356
1357         resp = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
1358         if (!resp)
1359                 return -ENOMEM;
1360
1361         msg[0] = IPMI_NETFN_APP_REQUEST << 2;
1362         msg[1] = IPMI_GET_BMC_GLOBAL_ENABLES_CMD;
1363         smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
1364
1365         rv = wait_for_msg_done(smi_info);
1366         if (rv) {
1367                 dev_warn(smi_info->io.dev,
1368                          "Error getting response from get global enables command: %d\n",
1369                          rv);
1370                 goto out;
1371         }
1372
1373         resp_len = smi_info->handlers->get_result(smi_info->si_sm,
1374                                                   resp, IPMI_MAX_MSG_LENGTH);
1375
1376         if (resp_len < 4 ||
1377                         resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
1378                         resp[1] != IPMI_GET_BMC_GLOBAL_ENABLES_CMD   ||
1379                         resp[2] != 0) {
1380                 dev_warn(smi_info->io.dev,
1381                          "Invalid return from get global enables command: %ld %x %x %x\n",
1382                          resp_len, resp[0], resp[1], resp[2]);
1383                 rv = -EINVAL;
1384                 goto out;
1385         } else {
1386                 *enables = resp[3];
1387         }
1388
1389 out:
1390         kfree(resp);
1391         return rv;
1392 }
1393
1394 /*
1395  * Returns 1 if it gets an error from the command.
1396  */
1397 static int set_global_enables(struct smi_info *smi_info, u8 enables)
1398 {
1399         unsigned char         msg[3];
1400         unsigned char         *resp;
1401         unsigned long         resp_len;
1402         int                   rv;
1403
1404         resp = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
1405         if (!resp)
1406                 return -ENOMEM;
1407
1408         msg[0] = IPMI_NETFN_APP_REQUEST << 2;
1409         msg[1] = IPMI_SET_BMC_GLOBAL_ENABLES_CMD;
1410         msg[2] = enables;
1411         smi_info->handlers->start_transaction(smi_info->si_sm, msg, 3);
1412
1413         rv = wait_for_msg_done(smi_info);
1414         if (rv) {
1415                 dev_warn(smi_info->io.dev,
1416                          "Error getting response from set global enables command: %d\n",
1417                          rv);
1418                 goto out;
1419         }
1420
1421         resp_len = smi_info->handlers->get_result(smi_info->si_sm,
1422                                                   resp, IPMI_MAX_MSG_LENGTH);
1423
1424         if (resp_len < 3 ||
1425                         resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
1426                         resp[1] != IPMI_SET_BMC_GLOBAL_ENABLES_CMD) {
1427                 dev_warn(smi_info->io.dev,
1428                          "Invalid return from set global enables command: %ld %x %x\n",
1429                          resp_len, resp[0], resp[1]);
1430                 rv = -EINVAL;
1431                 goto out;
1432         }
1433
1434         if (resp[2] != 0)
1435                 rv = 1;
1436
1437 out:
1438         kfree(resp);
1439         return rv;
1440 }
1441
1442 /*
1443  * Some BMCs do not support clearing the receive irq bit in the global
1444  * enables (even if they don't support interrupts on the BMC).  Check
1445  * for this and handle it properly.
1446  */
1447 static void check_clr_rcv_irq(struct smi_info *smi_info)
1448 {
1449         u8 enables = 0;
1450         int rv;
1451
1452         rv = get_global_enables(smi_info, &enables);
1453         if (!rv) {
1454                 if ((enables & IPMI_BMC_RCV_MSG_INTR) == 0)
1455                         /* Already clear, should work ok. */
1456                         return;
1457
1458                 enables &= ~IPMI_BMC_RCV_MSG_INTR;
1459                 rv = set_global_enables(smi_info, enables);
1460         }
1461
1462         if (rv < 0) {
1463                 dev_err(smi_info->io.dev,
1464                         "Cannot check clearing the rcv irq: %d\n", rv);
1465                 return;
1466         }
1467
1468         if (rv) {
1469                 /*
1470                  * An error when setting the event buffer bit means
1471                  * clearing the bit is not supported.
1472                  */
1473                 dev_warn(smi_info->io.dev,
1474                          "The BMC does not support clearing the recv irq bit, compensating, but the BMC needs to be fixed.\n");
1475                 smi_info->cannot_disable_irq = true;
1476         }
1477 }
1478
1479 /*
1480  * Some BMCs do not support setting the interrupt bits in the global
1481  * enables even if they support interrupts.  Clearly bad, but we can
1482  * compensate.
1483  */
1484 static void check_set_rcv_irq(struct smi_info *smi_info)
1485 {
1486         u8 enables = 0;
1487         int rv;
1488
1489         if (!smi_info->io.irq)
1490                 return;
1491
1492         rv = get_global_enables(smi_info, &enables);
1493         if (!rv) {
1494                 enables |= IPMI_BMC_RCV_MSG_INTR;
1495                 rv = set_global_enables(smi_info, enables);
1496         }
1497
1498         if (rv < 0) {
1499                 dev_err(smi_info->io.dev,
1500                         "Cannot check setting the rcv irq: %d\n", rv);
1501                 return;
1502         }
1503
1504         if (rv) {
1505                 /*
1506                  * An error when setting the event buffer bit means
1507                  * setting the bit is not supported.
1508                  */
1509                 dev_warn(smi_info->io.dev,
1510                          "The BMC does not support setting the recv irq bit, compensating, but the BMC needs to be fixed.\n");
1511                 smi_info->cannot_disable_irq = true;
1512                 smi_info->irq_enable_broken = true;
1513         }
1514 }
1515
1516 static int try_enable_event_buffer(struct smi_info *smi_info)
1517 {
1518         unsigned char         msg[3];
1519         unsigned char         *resp;
1520         unsigned long         resp_len;
1521         int                   rv = 0;
1522
1523         resp = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
1524         if (!resp)
1525                 return -ENOMEM;
1526
1527         msg[0] = IPMI_NETFN_APP_REQUEST << 2;
1528         msg[1] = IPMI_GET_BMC_GLOBAL_ENABLES_CMD;
1529         smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
1530
1531         rv = wait_for_msg_done(smi_info);
1532         if (rv) {
1533                 pr_warn(PFX "Error getting response from get global enables command, the event buffer is not enabled.\n");
1534                 goto out;
1535         }
1536
1537         resp_len = smi_info->handlers->get_result(smi_info->si_sm,
1538                                                   resp, IPMI_MAX_MSG_LENGTH);
1539
1540         if (resp_len < 4 ||
1541                         resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
1542                         resp[1] != IPMI_GET_BMC_GLOBAL_ENABLES_CMD   ||
1543                         resp[2] != 0) {
1544                 pr_warn(PFX "Invalid return from get global enables command, cannot enable the event buffer.\n");
1545                 rv = -EINVAL;
1546                 goto out;
1547         }
1548
1549         if (resp[3] & IPMI_BMC_EVT_MSG_BUFF) {
1550                 /* buffer is already enabled, nothing to do. */
1551                 smi_info->supports_event_msg_buff = true;
1552                 goto out;
1553         }
1554
1555         msg[0] = IPMI_NETFN_APP_REQUEST << 2;
1556         msg[1] = IPMI_SET_BMC_GLOBAL_ENABLES_CMD;
1557         msg[2] = resp[3] | IPMI_BMC_EVT_MSG_BUFF;
1558         smi_info->handlers->start_transaction(smi_info->si_sm, msg, 3);
1559
1560         rv = wait_for_msg_done(smi_info);
1561         if (rv) {
1562                 pr_warn(PFX "Error getting response from set global, enables command, the event buffer is not enabled.\n");
1563                 goto out;
1564         }
1565
1566         resp_len = smi_info->handlers->get_result(smi_info->si_sm,
1567                                                   resp, IPMI_MAX_MSG_LENGTH);
1568
1569         if (resp_len < 3 ||
1570                         resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
1571                         resp[1] != IPMI_SET_BMC_GLOBAL_ENABLES_CMD) {
1572                 pr_warn(PFX "Invalid return from get global, enables command, not enable the event buffer.\n");
1573                 rv = -EINVAL;
1574                 goto out;
1575         }
1576
1577         if (resp[2] != 0)
1578                 /*
1579                  * An error when setting the event buffer bit means
1580                  * that the event buffer is not supported.
1581                  */
1582                 rv = -ENOENT;
1583         else
1584                 smi_info->supports_event_msg_buff = true;
1585
1586 out:
1587         kfree(resp);
1588         return rv;
1589 }
1590
1591 #define IPMI_SI_ATTR(name) \
1592 static ssize_t ipmi_##name##_show(struct device *dev,                   \
1593                                   struct device_attribute *attr,        \
1594                                   char *buf)                            \
1595 {                                                                       \
1596         struct smi_info *smi_info = dev_get_drvdata(dev);               \
1597                                                                         \
1598         return snprintf(buf, 10, "%u\n", smi_get_stat(smi_info, name)); \
1599 }                                                                       \
1600 static DEVICE_ATTR(name, S_IRUGO, ipmi_##name##_show, NULL)
1601
1602 static ssize_t ipmi_type_show(struct device *dev,
1603                               struct device_attribute *attr,
1604                               char *buf)
1605 {
1606         struct smi_info *smi_info = dev_get_drvdata(dev);
1607
1608         return snprintf(buf, 10, "%s\n", si_to_str[smi_info->io.si_type]);
1609 }
1610 static DEVICE_ATTR(type, S_IRUGO, ipmi_type_show, NULL);
1611
1612 static ssize_t ipmi_interrupts_enabled_show(struct device *dev,
1613                                             struct device_attribute *attr,
1614                                             char *buf)
1615 {
1616         struct smi_info *smi_info = dev_get_drvdata(dev);
1617         int enabled = smi_info->io.irq && !smi_info->interrupt_disabled;
1618
1619         return snprintf(buf, 10, "%d\n", enabled);
1620 }
1621 static DEVICE_ATTR(interrupts_enabled, S_IRUGO,
1622                    ipmi_interrupts_enabled_show, NULL);
1623
1624 IPMI_SI_ATTR(short_timeouts);
1625 IPMI_SI_ATTR(long_timeouts);
1626 IPMI_SI_ATTR(idles);
1627 IPMI_SI_ATTR(interrupts);
1628 IPMI_SI_ATTR(attentions);
1629 IPMI_SI_ATTR(flag_fetches);
1630 IPMI_SI_ATTR(hosed_count);
1631 IPMI_SI_ATTR(complete_transactions);
1632 IPMI_SI_ATTR(events);
1633 IPMI_SI_ATTR(watchdog_pretimeouts);
1634 IPMI_SI_ATTR(incoming_messages);
1635
1636 static ssize_t ipmi_params_show(struct device *dev,
1637                                 struct device_attribute *attr,
1638                                 char *buf)
1639 {
1640         struct smi_info *smi_info = dev_get_drvdata(dev);
1641
1642         return snprintf(buf, 200,
1643                         "%s,%s,0x%lx,rsp=%d,rsi=%d,rsh=%d,irq=%d,ipmb=%d\n",
1644                         si_to_str[smi_info->io.si_type],
1645                         addr_space_to_str[smi_info->io.addr_type],
1646                         smi_info->io.addr_data,
1647                         smi_info->io.regspacing,
1648                         smi_info->io.regsize,
1649                         smi_info->io.regshift,
1650                         smi_info->io.irq,
1651                         smi_info->io.slave_addr);
1652 }
1653 static DEVICE_ATTR(params, S_IRUGO, ipmi_params_show, NULL);
1654
1655 static struct attribute *ipmi_si_dev_attrs[] = {
1656         &dev_attr_type.attr,
1657         &dev_attr_interrupts_enabled.attr,
1658         &dev_attr_short_timeouts.attr,
1659         &dev_attr_long_timeouts.attr,
1660         &dev_attr_idles.attr,
1661         &dev_attr_interrupts.attr,
1662         &dev_attr_attentions.attr,
1663         &dev_attr_flag_fetches.attr,
1664         &dev_attr_hosed_count.attr,
1665         &dev_attr_complete_transactions.attr,
1666         &dev_attr_events.attr,
1667         &dev_attr_watchdog_pretimeouts.attr,
1668         &dev_attr_incoming_messages.attr,
1669         &dev_attr_params.attr,
1670         NULL
1671 };
1672
1673 static const struct attribute_group ipmi_si_dev_attr_group = {
1674         .attrs          = ipmi_si_dev_attrs,
1675 };
1676
1677 /*
1678  * oem_data_avail_to_receive_msg_avail
1679  * @info - smi_info structure with msg_flags set
1680  *
1681  * Converts flags from OEM_DATA_AVAIL to RECEIVE_MSG_AVAIL
1682  * Returns 1 indicating need to re-run handle_flags().
1683  */
1684 static int oem_data_avail_to_receive_msg_avail(struct smi_info *smi_info)
1685 {
1686         smi_info->msg_flags = ((smi_info->msg_flags & ~OEM_DATA_AVAIL) |
1687                                RECEIVE_MSG_AVAIL);
1688         return 1;
1689 }
1690
1691 /*
1692  * setup_dell_poweredge_oem_data_handler
1693  * @info - smi_info.device_id must be populated
1694  *
1695  * Systems that match, but have firmware version < 1.40 may assert
1696  * OEM0_DATA_AVAIL on their own, without being told via Set Flags that
1697  * it's safe to do so.  Such systems will de-assert OEM1_DATA_AVAIL
1698  * upon receipt of IPMI_GET_MSG_CMD, so we should treat these flags
1699  * as RECEIVE_MSG_AVAIL instead.
1700  *
1701  * As Dell has no plans to release IPMI 1.5 firmware that *ever*
1702  * assert the OEM[012] bits, and if it did, the driver would have to
1703  * change to handle that properly, we don't actually check for the
1704  * firmware version.
1705  * Device ID = 0x20                BMC on PowerEdge 8G servers
1706  * Device Revision = 0x80
1707  * Firmware Revision1 = 0x01       BMC version 1.40
1708  * Firmware Revision2 = 0x40       BCD encoded
1709  * IPMI Version = 0x51             IPMI 1.5
1710  * Manufacturer ID = A2 02 00      Dell IANA
1711  *
1712  * Additionally, PowerEdge systems with IPMI < 1.5 may also assert
1713  * OEM0_DATA_AVAIL and needs to be treated as RECEIVE_MSG_AVAIL.
1714  *
1715  */
1716 #define DELL_POWEREDGE_8G_BMC_DEVICE_ID  0x20
1717 #define DELL_POWEREDGE_8G_BMC_DEVICE_REV 0x80
1718 #define DELL_POWEREDGE_8G_BMC_IPMI_VERSION 0x51
1719 #define DELL_IANA_MFR_ID 0x0002a2
1720 static void setup_dell_poweredge_oem_data_handler(struct smi_info *smi_info)
1721 {
1722         struct ipmi_device_id *id = &smi_info->device_id;
1723         if (id->manufacturer_id == DELL_IANA_MFR_ID) {
1724                 if (id->device_id       == DELL_POWEREDGE_8G_BMC_DEVICE_ID  &&
1725                     id->device_revision == DELL_POWEREDGE_8G_BMC_DEVICE_REV &&
1726                     id->ipmi_version   == DELL_POWEREDGE_8G_BMC_IPMI_VERSION) {
1727                         smi_info->oem_data_avail_handler =
1728                                 oem_data_avail_to_receive_msg_avail;
1729                 } else if (ipmi_version_major(id) < 1 ||
1730                            (ipmi_version_major(id) == 1 &&
1731                             ipmi_version_minor(id) < 5)) {
1732                         smi_info->oem_data_avail_handler =
1733                                 oem_data_avail_to_receive_msg_avail;
1734                 }
1735         }
1736 }
1737
1738 #define CANNOT_RETURN_REQUESTED_LENGTH 0xCA
1739 static void return_hosed_msg_badsize(struct smi_info *smi_info)
1740 {
1741         struct ipmi_smi_msg *msg = smi_info->curr_msg;
1742
1743         /* Make it a response */
1744         msg->rsp[0] = msg->data[0] | 4;
1745         msg->rsp[1] = msg->data[1];
1746         msg->rsp[2] = CANNOT_RETURN_REQUESTED_LENGTH;
1747         msg->rsp_size = 3;
1748         smi_info->curr_msg = NULL;
1749         deliver_recv_msg(smi_info, msg);
1750 }
1751
1752 /*
1753  * dell_poweredge_bt_xaction_handler
1754  * @info - smi_info.device_id must be populated
1755  *
1756  * Dell PowerEdge servers with the BT interface (x6xx and 1750) will
1757  * not respond to a Get SDR command if the length of the data
1758  * requested is exactly 0x3A, which leads to command timeouts and no
1759  * data returned.  This intercepts such commands, and causes userspace
1760  * callers to try again with a different-sized buffer, which succeeds.
1761  */
1762
1763 #define STORAGE_NETFN 0x0A
1764 #define STORAGE_CMD_GET_SDR 0x23
1765 static int dell_poweredge_bt_xaction_handler(struct notifier_block *self,
1766                                              unsigned long unused,
1767                                              void *in)
1768 {
1769         struct smi_info *smi_info = in;
1770         unsigned char *data = smi_info->curr_msg->data;
1771         unsigned int size   = smi_info->curr_msg->data_size;
1772         if (size >= 8 &&
1773             (data[0]>>2) == STORAGE_NETFN &&
1774             data[1] == STORAGE_CMD_GET_SDR &&
1775             data[7] == 0x3A) {
1776                 return_hosed_msg_badsize(smi_info);
1777                 return NOTIFY_STOP;
1778         }
1779         return NOTIFY_DONE;
1780 }
1781
1782 static struct notifier_block dell_poweredge_bt_xaction_notifier = {
1783         .notifier_call  = dell_poweredge_bt_xaction_handler,
1784 };
1785
1786 /*
1787  * setup_dell_poweredge_bt_xaction_handler
1788  * @info - smi_info.device_id must be filled in already
1789  *
1790  * Fills in smi_info.device_id.start_transaction_pre_hook
1791  * when we know what function to use there.
1792  */
1793 static void
1794 setup_dell_poweredge_bt_xaction_handler(struct smi_info *smi_info)
1795 {
1796         struct ipmi_device_id *id = &smi_info->device_id;
1797         if (id->manufacturer_id == DELL_IANA_MFR_ID &&
1798             smi_info->io.si_type == SI_BT)
1799                 register_xaction_notifier(&dell_poweredge_bt_xaction_notifier);
1800 }
1801
1802 /*
1803  * setup_oem_data_handler
1804  * @info - smi_info.device_id must be filled in already
1805  *
1806  * Fills in smi_info.device_id.oem_data_available_handler
1807  * when we know what function to use there.
1808  */
1809
1810 static void setup_oem_data_handler(struct smi_info *smi_info)
1811 {
1812         setup_dell_poweredge_oem_data_handler(smi_info);
1813 }
1814
1815 static void setup_xaction_handlers(struct smi_info *smi_info)
1816 {
1817         setup_dell_poweredge_bt_xaction_handler(smi_info);
1818 }
1819
1820 static void check_for_broken_irqs(struct smi_info *smi_info)
1821 {
1822         check_clr_rcv_irq(smi_info);
1823         check_set_rcv_irq(smi_info);
1824 }
1825
1826 static inline void stop_timer_and_thread(struct smi_info *smi_info)
1827 {
1828         if (smi_info->thread != NULL) {
1829                 kthread_stop(smi_info->thread);
1830                 smi_info->thread = NULL;
1831         }
1832
1833         smi_info->timer_can_start = false;
1834         if (smi_info->timer_running)
1835                 del_timer_sync(&smi_info->si_timer);
1836 }
1837
1838 static struct smi_info *find_dup_si(struct smi_info *info)
1839 {
1840         struct smi_info *e;
1841
1842         list_for_each_entry(e, &smi_infos, link) {
1843                 if (e->io.addr_type != info->io.addr_type)
1844                         continue;
1845                 if (e->io.addr_data == info->io.addr_data) {
1846                         /*
1847                          * This is a cheap hack, ACPI doesn't have a defined
1848                          * slave address but SMBIOS does.  Pick it up from
1849                          * any source that has it available.
1850                          */
1851                         if (info->io.slave_addr && !e->io.slave_addr)
1852                                 e->io.slave_addr = info->io.slave_addr;
1853                         return e;
1854                 }
1855         }
1856
1857         return NULL;
1858 }
1859
1860 int ipmi_si_add_smi(struct si_sm_io *io)
1861 {
1862         int rv = 0;
1863         struct smi_info *new_smi, *dup;
1864
1865         if (!io->io_setup) {
1866                 if (io->addr_type == IPMI_IO_ADDR_SPACE) {
1867                         io->io_setup = ipmi_si_port_setup;
1868                 } else if (io->addr_type == IPMI_MEM_ADDR_SPACE) {
1869                         io->io_setup = ipmi_si_mem_setup;
1870                 } else {
1871                         return -EINVAL;
1872                 }
1873         }
1874
1875         new_smi = kzalloc(sizeof(*new_smi), GFP_KERNEL);
1876         if (!new_smi)
1877                 return -ENOMEM;
1878         spin_lock_init(&new_smi->si_lock);
1879
1880         new_smi->io = *io;
1881
1882         mutex_lock(&smi_infos_lock);
1883         dup = find_dup_si(new_smi);
1884         if (dup) {
1885                 if (new_smi->io.addr_source == SI_ACPI &&
1886                     dup->io.addr_source == SI_SMBIOS) {
1887                         /* We prefer ACPI over SMBIOS. */
1888                         dev_info(dup->io.dev,
1889                                  "Removing SMBIOS-specified %s state machine in favor of ACPI\n",
1890                                  si_to_str[new_smi->io.si_type]);
1891                         cleanup_one_si(dup);
1892                 } else {
1893                         dev_info(new_smi->io.dev,
1894                                  "%s-specified %s state machine: duplicate\n",
1895                                  ipmi_addr_src_to_str(new_smi->io.addr_source),
1896                                  si_to_str[new_smi->io.si_type]);
1897                         rv = -EBUSY;
1898                         kfree(new_smi);
1899                         goto out_err;
1900                 }
1901         }
1902
1903         pr_info(PFX "Adding %s-specified %s state machine\n",
1904                 ipmi_addr_src_to_str(new_smi->io.addr_source),
1905                 si_to_str[new_smi->io.si_type]);
1906
1907         list_add_tail(&new_smi->link, &smi_infos);
1908
1909         if (initialized)
1910                 rv = try_smi_init(new_smi);
1911 out_err:
1912         mutex_unlock(&smi_infos_lock);
1913         return rv;
1914 }
1915
1916 /*
1917  * Try to start up an interface.  Must be called with smi_infos_lock
1918  * held, primarily to keep smi_num consistent, we only one to do these
1919  * one at a time.
1920  */
1921 static int try_smi_init(struct smi_info *new_smi)
1922 {
1923         int rv = 0;
1924         int i;
1925         char *init_name = NULL;
1926
1927         pr_info(PFX "Trying %s-specified %s state machine at %s address 0x%lx, slave address 0x%x, irq %d\n",
1928                 ipmi_addr_src_to_str(new_smi->io.addr_source),
1929                 si_to_str[new_smi->io.si_type],
1930                 addr_space_to_str[new_smi->io.addr_type],
1931                 new_smi->io.addr_data,
1932                 new_smi->io.slave_addr, new_smi->io.irq);
1933
1934         switch (new_smi->io.si_type) {
1935         case SI_KCS:
1936                 new_smi->handlers = &kcs_smi_handlers;
1937                 break;
1938
1939         case SI_SMIC:
1940                 new_smi->handlers = &smic_smi_handlers;
1941                 break;
1942
1943         case SI_BT:
1944                 new_smi->handlers = &bt_smi_handlers;
1945                 break;
1946
1947         default:
1948                 /* No support for anything else yet. */
1949                 rv = -EIO;
1950                 goto out_err;
1951         }
1952
1953         new_smi->si_num = smi_num;
1954
1955         /* Do this early so it's available for logs. */
1956         if (!new_smi->io.dev) {
1957                 init_name = kasprintf(GFP_KERNEL, "ipmi_si.%d",
1958                                       new_smi->si_num);
1959
1960                 /*
1961                  * If we don't already have a device from something
1962                  * else (like PCI), then register a new one.
1963                  */
1964                 new_smi->pdev = platform_device_alloc("ipmi_si",
1965                                                       new_smi->si_num);
1966                 if (!new_smi->pdev) {
1967                         pr_err(PFX "Unable to allocate platform device\n");
1968                         rv = -ENOMEM;
1969                         goto out_err;
1970                 }
1971                 new_smi->io.dev = &new_smi->pdev->dev;
1972                 new_smi->io.dev->driver = &ipmi_platform_driver.driver;
1973                 /* Nulled by device_add() */
1974                 new_smi->io.dev->init_name = init_name;
1975         }
1976
1977         /* Allocate the state machine's data and initialize it. */
1978         new_smi->si_sm = kmalloc(new_smi->handlers->size(), GFP_KERNEL);
1979         if (!new_smi->si_sm) {
1980                 rv = -ENOMEM;
1981                 goto out_err;
1982         }
1983         new_smi->io.io_size = new_smi->handlers->init_data(new_smi->si_sm,
1984                                                            &new_smi->io);
1985
1986         /* Now that we know the I/O size, we can set up the I/O. */
1987         rv = new_smi->io.io_setup(&new_smi->io);
1988         if (rv) {
1989                 dev_err(new_smi->io.dev, "Could not set up I/O space\n");
1990                 goto out_err;
1991         }
1992
1993         /* Do low-level detection first. */
1994         if (new_smi->handlers->detect(new_smi->si_sm)) {
1995                 if (new_smi->io.addr_source)
1996                         dev_err(new_smi->io.dev,
1997                                 "Interface detection failed\n");
1998                 rv = -ENODEV;
1999                 goto out_err;
2000         }
2001
2002         /*
2003          * Attempt a get device id command.  If it fails, we probably
2004          * don't have a BMC here.
2005          */
2006         rv = try_get_dev_id(new_smi);
2007         if (rv) {
2008                 if (new_smi->io.addr_source)
2009                         dev_err(new_smi->io.dev,
2010                                "There appears to be no BMC at this location\n");
2011                 goto out_err;
2012         }
2013
2014         setup_oem_data_handler(new_smi);
2015         setup_xaction_handlers(new_smi);
2016         check_for_broken_irqs(new_smi);
2017
2018         new_smi->waiting_msg = NULL;
2019         new_smi->curr_msg = NULL;
2020         atomic_set(&new_smi->req_events, 0);
2021         new_smi->run_to_completion = false;
2022         for (i = 0; i < SI_NUM_STATS; i++)
2023                 atomic_set(&new_smi->stats[i], 0);
2024
2025         new_smi->interrupt_disabled = true;
2026         atomic_set(&new_smi->need_watch, 0);
2027
2028         rv = try_enable_event_buffer(new_smi);
2029         if (rv == 0)
2030                 new_smi->has_event_buffer = true;
2031
2032         /*
2033          * Start clearing the flags before we enable interrupts or the
2034          * timer to avoid racing with the timer.
2035          */
2036         start_clear_flags(new_smi);
2037
2038         /*
2039          * IRQ is defined to be set when non-zero.  req_events will
2040          * cause a global flags check that will enable interrupts.
2041          */
2042         if (new_smi->io.irq) {
2043                 new_smi->interrupt_disabled = false;
2044                 atomic_set(&new_smi->req_events, 1);
2045         }
2046
2047         if (new_smi->pdev && !new_smi->pdev_registered) {
2048                 rv = platform_device_add(new_smi->pdev);
2049                 if (rv) {
2050                         dev_err(new_smi->io.dev,
2051                                 "Unable to register system interface device: %d\n",
2052                                 rv);
2053                         goto out_err;
2054                 }
2055                 new_smi->pdev_registered = true;
2056         }
2057
2058         dev_set_drvdata(new_smi->io.dev, new_smi);
2059         rv = device_add_group(new_smi->io.dev, &ipmi_si_dev_attr_group);
2060         if (rv) {
2061                 dev_err(new_smi->io.dev,
2062                         "Unable to add device attributes: error %d\n",
2063                         rv);
2064                 goto out_err;
2065         }
2066         new_smi->dev_group_added = true;
2067
2068         rv = ipmi_register_smi(&handlers,
2069                                new_smi,
2070                                new_smi->io.dev,
2071                                new_smi->io.slave_addr);
2072         if (rv) {
2073                 dev_err(new_smi->io.dev,
2074                         "Unable to register device: error %d\n",
2075                         rv);
2076                 goto out_err;
2077         }
2078
2079         /* Don't increment till we know we have succeeded. */
2080         smi_num++;
2081
2082         dev_info(new_smi->io.dev, "IPMI %s interface initialized\n",
2083                  si_to_str[new_smi->io.si_type]);
2084
2085         WARN_ON(new_smi->io.dev->init_name != NULL);
2086         kfree(init_name);
2087
2088         return 0;
2089
2090 out_err:
2091         if (new_smi->intf) {
2092                 ipmi_unregister_smi(new_smi->intf);
2093                 new_smi->intf = NULL;
2094         }
2095
2096         kfree(init_name);
2097
2098         return rv;
2099 }
2100
2101 static int init_ipmi_si(void)
2102 {
2103         struct smi_info *e;
2104         enum ipmi_addr_src type = SI_INVALID;
2105
2106         if (initialized)
2107                 return 0;
2108
2109         pr_info("IPMI System Interface driver.\n");
2110
2111         /* If the user gave us a device, they presumably want us to use it */
2112         if (!ipmi_si_hardcode_find_bmc())
2113                 goto do_scan;
2114
2115         ipmi_si_platform_init();
2116
2117         ipmi_si_pci_init();
2118
2119         ipmi_si_parisc_init();
2120
2121         /* We prefer devices with interrupts, but in the case of a machine
2122            with multiple BMCs we assume that there will be several instances
2123            of a given type so if we succeed in registering a type then also
2124            try to register everything else of the same type */
2125 do_scan:
2126         mutex_lock(&smi_infos_lock);
2127         list_for_each_entry(e, &smi_infos, link) {
2128                 /* Try to register a device if it has an IRQ and we either
2129                    haven't successfully registered a device yet or this
2130                    device has the same type as one we successfully registered */
2131                 if (e->io.irq && (!type || e->io.addr_source == type)) {
2132                         if (!try_smi_init(e)) {
2133                                 type = e->io.addr_source;
2134                         }
2135                 }
2136         }
2137
2138         /* type will only have been set if we successfully registered an si */
2139         if (type)
2140                 goto skip_fallback_noirq;
2141
2142         /* Fall back to the preferred device */
2143
2144         list_for_each_entry(e, &smi_infos, link) {
2145                 if (!e->io.irq && (!type || e->io.addr_source == type)) {
2146                         if (!try_smi_init(e)) {
2147                                 type = e->io.addr_source;
2148                         }
2149                 }
2150         }
2151
2152 skip_fallback_noirq:
2153         initialized = 1;
2154         mutex_unlock(&smi_infos_lock);
2155
2156         if (type)
2157                 return 0;
2158
2159         mutex_lock(&smi_infos_lock);
2160         if (unload_when_empty && list_empty(&smi_infos)) {
2161                 mutex_unlock(&smi_infos_lock);
2162                 cleanup_ipmi_si();
2163                 pr_warn(PFX "Unable to find any System Interface(s)\n");
2164                 return -ENODEV;
2165         } else {
2166                 mutex_unlock(&smi_infos_lock);
2167                 return 0;
2168         }
2169 }
2170 module_init(init_ipmi_si);
2171
2172 static void shutdown_smi(void *send_info)
2173 {
2174         struct smi_info *smi_info = send_info;
2175
2176         if (smi_info->dev_group_added) {
2177                 device_remove_group(smi_info->io.dev, &ipmi_si_dev_attr_group);
2178                 smi_info->dev_group_added = false;
2179         }
2180         if (smi_info->io.dev)
2181                 dev_set_drvdata(smi_info->io.dev, NULL);
2182
2183         /*
2184          * Make sure that interrupts, the timer and the thread are
2185          * stopped and will not run again.
2186          */
2187         smi_info->interrupt_disabled = true;
2188         if (smi_info->io.irq_cleanup) {
2189                 smi_info->io.irq_cleanup(&smi_info->io);
2190                 smi_info->io.irq_cleanup = NULL;
2191         }
2192         stop_timer_and_thread(smi_info);
2193
2194         /*
2195          * Wait until we know that we are out of any interrupt
2196          * handlers might have been running before we freed the
2197          * interrupt.
2198          */
2199         synchronize_sched();
2200
2201         /*
2202          * Timeouts are stopped, now make sure the interrupts are off
2203          * in the BMC.  Note that timers and CPU interrupts are off,
2204          * so no need for locks.
2205          */
2206         while (smi_info->curr_msg || (smi_info->si_state != SI_NORMAL)) {
2207                 poll(smi_info);
2208                 schedule_timeout_uninterruptible(1);
2209         }
2210         if (smi_info->handlers)
2211                 disable_si_irq(smi_info);
2212         while (smi_info->curr_msg || (smi_info->si_state != SI_NORMAL)) {
2213                 poll(smi_info);
2214                 schedule_timeout_uninterruptible(1);
2215         }
2216         if (smi_info->handlers)
2217                 smi_info->handlers->cleanup(smi_info->si_sm);
2218
2219         if (smi_info->io.addr_source_cleanup) {
2220                 smi_info->io.addr_source_cleanup(&smi_info->io);
2221                 smi_info->io.addr_source_cleanup = NULL;
2222         }
2223         if (smi_info->io.io_cleanup) {
2224                 smi_info->io.io_cleanup(&smi_info->io);
2225                 smi_info->io.io_cleanup = NULL;
2226         }
2227
2228         kfree(smi_info->si_sm);
2229         smi_info->si_sm = NULL;
2230 }
2231
2232 /*
2233  * Must be called with smi_infos_lock held, to serialize the
2234  * smi_info->intf check.
2235  */
2236 static void cleanup_one_si(struct smi_info *smi_info)
2237 {
2238         if (!smi_info)
2239                 return;
2240
2241         list_del(&smi_info->link);
2242
2243         if (smi_info->intf) {
2244                 ipmi_unregister_smi(smi_info->intf);
2245                 smi_info->intf = NULL;
2246         }
2247
2248         if (smi_info->pdev) {
2249                 if (smi_info->pdev_registered)
2250                         platform_device_unregister(smi_info->pdev);
2251                 else
2252                         platform_device_put(smi_info->pdev);
2253         }
2254
2255         kfree(smi_info);
2256 }
2257
2258 int ipmi_si_remove_by_dev(struct device *dev)
2259 {
2260         struct smi_info *e;
2261         int rv = -ENOENT;
2262
2263         mutex_lock(&smi_infos_lock);
2264         list_for_each_entry(e, &smi_infos, link) {
2265                 if (e->io.dev == dev) {
2266                         cleanup_one_si(e);
2267                         rv = 0;
2268                         break;
2269                 }
2270         }
2271         mutex_unlock(&smi_infos_lock);
2272
2273         return rv;
2274 }
2275
2276 void ipmi_si_remove_by_data(int addr_space, enum si_type si_type,
2277                             unsigned long addr)
2278 {
2279         /* remove */
2280         struct smi_info *e, *tmp_e;
2281
2282         mutex_lock(&smi_infos_lock);
2283         list_for_each_entry_safe(e, tmp_e, &smi_infos, link) {
2284                 if (e->io.addr_type != addr_space)
2285                         continue;
2286                 if (e->io.si_type != si_type)
2287                         continue;
2288                 if (e->io.addr_data == addr)
2289                         cleanup_one_si(e);
2290         }
2291         mutex_unlock(&smi_infos_lock);
2292 }
2293
2294 static void cleanup_ipmi_si(void)
2295 {
2296         struct smi_info *e, *tmp_e;
2297
2298         if (!initialized)
2299                 return;
2300
2301         ipmi_si_pci_shutdown();
2302
2303         ipmi_si_parisc_shutdown();
2304
2305         ipmi_si_platform_shutdown();
2306
2307         mutex_lock(&smi_infos_lock);
2308         list_for_each_entry_safe(e, tmp_e, &smi_infos, link)
2309                 cleanup_one_si(e);
2310         mutex_unlock(&smi_infos_lock);
2311 }
2312 module_exit(cleanup_ipmi_si);
2313
2314 MODULE_ALIAS("platform:dmi-ipmi-si");
2315 MODULE_LICENSE("GPL");
2316 MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>");
2317 MODULE_DESCRIPTION("Interface to the IPMI driver for the KCS, SMIC, and BT"
2318                    " system interfaces.");