Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / char / ipmi / ipmi_si_intf.c
diff --git a/kernel/drivers/char/ipmi/ipmi_si_intf.c b/kernel/drivers/char/ipmi/ipmi_si_intf.c
new file mode 100644 (file)
index 0000000..8a45e92
--- /dev/null
@@ -0,0 +1,3874 @@
+/*
+ * ipmi_si.c
+ *
+ * The interface to the IPMI driver for the system interfaces (KCS, SMIC,
+ * BT).
+ *
+ * Author: MontaVista Software, Inc.
+ *         Corey Minyard <minyard@mvista.com>
+ *         source@mvista.com
+ *
+ * Copyright 2002 MontaVista Software Inc.
+ * Copyright 2006 IBM Corp., Christian Krafft <krafft@de.ibm.com>
+ *
+ *  This program is free software; you can redistribute it and/or modify it
+ *  under the terms of the GNU General Public License as published by the
+ *  Free Software Foundation; either version 2 of the License, or (at your
+ *  option) any later version.
+ *
+ *
+ *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ *  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
+ *  INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ *  BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ *  OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ *  ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR
+ *  TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+ *  USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ *  You should have received a copy of the GNU General Public License along
+ *  with this program; if not, write to the Free Software Foundation, Inc.,
+ *  675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+/*
+ * This file holds the "policy" for the interface to the SMI state
+ * machine.  It does the configuration, handles timers and interrupts,
+ * and drives the real SMI state machine.
+ */
+
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/sched.h>
+#include <linux/seq_file.h>
+#include <linux/timer.h>
+#include <linux/errno.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+#include <linux/delay.h>
+#include <linux/list.h>
+#include <linux/pci.h>
+#include <linux/ioport.h>
+#include <linux/notifier.h>
+#include <linux/mutex.h>
+#include <linux/kthread.h>
+#include <asm/irq.h>
+#include <linux/interrupt.h>
+#include <linux/rcupdate.h>
+#include <linux/ipmi.h>
+#include <linux/ipmi_smi.h>
+#include <asm/io.h>
+#include "ipmi_si_sm.h"
+#include <linux/dmi.h>
+#include <linux/string.h>
+#include <linux/ctype.h>
+#include <linux/pnp.h>
+#include <linux/of_device.h>
+#include <linux/of_platform.h>
+#include <linux/of_address.h>
+#include <linux/of_irq.h>
+
+#ifdef CONFIG_PARISC
+#include <asm/hardware.h>      /* for register_parisc_driver() stuff */
+#include <asm/parisc-device.h>
+#endif
+
+#define PFX "ipmi_si: "
+
+/* Measure times between events in the driver. */
+#undef DEBUG_TIMING
+
+/* Call every 10 ms. */
+#define SI_TIMEOUT_TIME_USEC   10000
+#define SI_USEC_PER_JIFFY      (1000000/HZ)
+#define SI_TIMEOUT_JIFFIES     (SI_TIMEOUT_TIME_USEC/SI_USEC_PER_JIFFY)
+#define SI_SHORT_TIMEOUT_USEC  250 /* .25ms when the SM request a
+                                     short timeout */
+
+enum si_intf_state {
+       SI_NORMAL,
+       SI_GETTING_FLAGS,
+       SI_GETTING_EVENTS,
+       SI_CLEARING_FLAGS,
+       SI_GETTING_MESSAGES,
+       SI_CHECKING_ENABLES,
+       SI_SETTING_ENABLES
+       /* FIXME - add watchdog stuff. */
+};
+
+/* Some BT-specific defines we need here. */
+#define IPMI_BT_INTMASK_REG            2
+#define IPMI_BT_INTMASK_CLEAR_IRQ_BIT  2
+#define IPMI_BT_INTMASK_ENABLE_IRQ_BIT 1
+
+enum si_type {
+    SI_KCS, SI_SMIC, SI_BT
+};
+static char *si_to_str[] = { "kcs", "smic", "bt" };
+
+#define DEVICE_NAME "ipmi_si"
+
+static struct platform_driver ipmi_driver;
+
+/*
+ * Indexes into stats[] in smi_info below.
+ */
+enum si_stat_indexes {
+       /*
+        * Number of times the driver requested a timer while an operation
+        * was in progress.
+        */
+       SI_STAT_short_timeouts = 0,
+
+       /*
+        * Number of times the driver requested a timer while nothing was in
+        * progress.
+        */
+       SI_STAT_long_timeouts,
+
+       /* Number of times the interface was idle while being polled. */
+       SI_STAT_idles,
+
+       /* Number of interrupts the driver handled. */
+       SI_STAT_interrupts,
+
+       /* Number of time the driver got an ATTN from the hardware. */
+       SI_STAT_attentions,
+
+       /* Number of times the driver requested flags from the hardware. */
+       SI_STAT_flag_fetches,
+
+       /* Number of times the hardware didn't follow the state machine. */
+       SI_STAT_hosed_count,
+
+       /* Number of completed messages. */
+       SI_STAT_complete_transactions,
+
+       /* Number of IPMI events received from the hardware. */
+       SI_STAT_events,
+
+       /* Number of watchdog pretimeouts. */
+       SI_STAT_watchdog_pretimeouts,
+
+       /* Number of asynchronous messages received. */
+       SI_STAT_incoming_messages,
+
+
+       /* This *must* remain last, add new values above this. */
+       SI_NUM_STATS
+};
+
+struct smi_info {
+       int                    intf_num;
+       ipmi_smi_t             intf;
+       struct si_sm_data      *si_sm;
+       struct si_sm_handlers  *handlers;
+       enum si_type           si_type;
+       spinlock_t             si_lock;
+       struct ipmi_smi_msg    *waiting_msg;
+       struct ipmi_smi_msg    *curr_msg;
+       enum si_intf_state     si_state;
+
+       /*
+        * Used to handle the various types of I/O that can occur with
+        * IPMI
+        */
+       struct si_sm_io io;
+       int (*io_setup)(struct smi_info *info);
+       void (*io_cleanup)(struct smi_info *info);
+       int (*irq_setup)(struct smi_info *info);
+       void (*irq_cleanup)(struct smi_info *info);
+       unsigned int io_size;
+       enum ipmi_addr_src addr_source; /* ACPI, PCI, SMBIOS, hardcode, etc. */
+       void (*addr_source_cleanup)(struct smi_info *info);
+       void *addr_source_data;
+
+       /*
+        * Per-OEM handler, called from handle_flags().  Returns 1
+        * when handle_flags() needs to be re-run or 0 indicating it
+        * set si_state itself.
+        */
+       int (*oem_data_avail_handler)(struct smi_info *smi_info);
+
+       /*
+        * Flags from the last GET_MSG_FLAGS command, used when an ATTN
+        * is set to hold the flags until we are done handling everything
+        * from the flags.
+        */
+#define RECEIVE_MSG_AVAIL      0x01
+#define EVENT_MSG_BUFFER_FULL  0x02
+#define WDT_PRE_TIMEOUT_INT    0x08
+#define OEM0_DATA_AVAIL     0x20
+#define OEM1_DATA_AVAIL     0x40
+#define OEM2_DATA_AVAIL     0x80
+#define OEM_DATA_AVAIL      (OEM0_DATA_AVAIL | \
+                            OEM1_DATA_AVAIL | \
+                            OEM2_DATA_AVAIL)
+       unsigned char       msg_flags;
+
+       /* Does the BMC have an event buffer? */
+       bool                has_event_buffer;
+
+       /*
+        * If set to true, this will request events the next time the
+        * state machine is idle.
+        */
+       atomic_t            req_events;
+
+       /*
+        * If true, run the state machine to completion on every send
+        * call.  Generally used after a panic to make sure stuff goes
+        * out.
+        */
+       bool                run_to_completion;
+
+       /* The I/O port of an SI interface. */
+       int                 port;
+
+       /*
+        * The space between start addresses of the two ports.  For
+        * instance, if the first port is 0xca2 and the spacing is 4, then
+        * the second port is 0xca6.
+        */
+       unsigned int        spacing;
+
+       /* zero if no irq; */
+       int                 irq;
+
+       /* The timer for this si. */
+       struct timer_list   si_timer;
+
+       /* This flag is set, if the timer is running (timer_pending() isn't enough) */
+       bool                timer_running;
+
+       /* The time (in jiffies) the last timeout occurred at. */
+       unsigned long       last_timeout_jiffies;
+
+       /* Are we waiting for the events, pretimeouts, received msgs? */
+       atomic_t            need_watch;
+
+       /*
+        * The driver will disable interrupts when it gets into a
+        * situation where it cannot handle messages due to lack of
+        * memory.  Once that situation clears up, it will re-enable
+        * interrupts.
+        */
+       bool interrupt_disabled;
+
+       /*
+        * Does the BMC support events?
+        */
+       bool supports_event_msg_buff;
+
+       /*
+        * Can we clear the global enables receive irq bit?
+        */
+       bool cannot_clear_recv_irq_bit;
+
+       /*
+        * Did we get an attention that we did not handle?
+        */
+       bool got_attn;
+
+       /* From the get device id response... */
+       struct ipmi_device_id device_id;
+
+       /* Driver model stuff. */
+       struct device *dev;
+       struct platform_device *pdev;
+
+       /*
+        * True if we allocated the device, false if it came from
+        * someplace else (like PCI).
+        */
+       bool dev_registered;
+
+       /* Slave address, could be reported from DMI. */
+       unsigned char slave_addr;
+
+       /* Counters and things for the proc filesystem. */
+       atomic_t stats[SI_NUM_STATS];
+
+       struct task_struct *thread;
+
+       struct list_head link;
+       union ipmi_smi_info_union addr_info;
+};
+
+#define smi_inc_stat(smi, stat) \
+       atomic_inc(&(smi)->stats[SI_STAT_ ## stat])
+#define smi_get_stat(smi, stat) \
+       ((unsigned int) atomic_read(&(smi)->stats[SI_STAT_ ## stat]))
+
+#define SI_MAX_PARMS 4
+
+static int force_kipmid[SI_MAX_PARMS];
+static int num_force_kipmid;
+#ifdef CONFIG_PCI
+static bool pci_registered;
+#endif
+#ifdef CONFIG_ACPI
+static bool pnp_registered;
+#endif
+#ifdef CONFIG_PARISC
+static bool parisc_registered;
+#endif
+
+static unsigned int kipmid_max_busy_us[SI_MAX_PARMS];
+static int num_max_busy_us;
+
+static bool unload_when_empty = true;
+
+static int add_smi(struct smi_info *smi);
+static int try_smi_init(struct smi_info *smi);
+static void cleanup_one_si(struct smi_info *to_clean);
+static void cleanup_ipmi_si(void);
+
+#ifdef DEBUG_TIMING
+void debug_timestamp(char *msg)
+{
+       struct timespec64 t;
+
+       getnstimeofday64(&t);
+       pr_debug("**%s: %lld.%9.9ld\n", msg, (long long) t.tv_sec, t.tv_nsec);
+}
+#else
+#define debug_timestamp(x)
+#endif
+
+static ATOMIC_NOTIFIER_HEAD(xaction_notifier_list);
+static int register_xaction_notifier(struct notifier_block *nb)
+{
+       return atomic_notifier_chain_register(&xaction_notifier_list, nb);
+}
+
+static void deliver_recv_msg(struct smi_info *smi_info,
+                            struct ipmi_smi_msg *msg)
+{
+       /* Deliver the message to the upper layer. */
+       if (smi_info->intf)
+               ipmi_smi_msg_received(smi_info->intf, msg);
+       else
+               ipmi_free_smi_msg(msg);
+}
+
+static void return_hosed_msg(struct smi_info *smi_info, int cCode)
+{
+       struct ipmi_smi_msg *msg = smi_info->curr_msg;
+
+       if (cCode < 0 || cCode > IPMI_ERR_UNSPECIFIED)
+               cCode = IPMI_ERR_UNSPECIFIED;
+       /* else use it as is */
+
+       /* Make it a response */
+       msg->rsp[0] = msg->data[0] | 4;
+       msg->rsp[1] = msg->data[1];
+       msg->rsp[2] = cCode;
+       msg->rsp_size = 3;
+
+       smi_info->curr_msg = NULL;
+       deliver_recv_msg(smi_info, msg);
+}
+
+static enum si_sm_result start_next_msg(struct smi_info *smi_info)
+{
+       int              rv;
+
+       if (!smi_info->waiting_msg) {
+               smi_info->curr_msg = NULL;
+               rv = SI_SM_IDLE;
+       } else {
+               int err;
+
+               smi_info->curr_msg = smi_info->waiting_msg;
+               smi_info->waiting_msg = NULL;
+               debug_timestamp("Start2");
+               err = atomic_notifier_call_chain(&xaction_notifier_list,
+                               0, smi_info);
+               if (err & NOTIFY_STOP_MASK) {
+                       rv = SI_SM_CALL_WITHOUT_DELAY;
+                       goto out;
+               }
+               err = smi_info->handlers->start_transaction(
+                       smi_info->si_sm,
+                       smi_info->curr_msg->data,
+                       smi_info->curr_msg->data_size);
+               if (err)
+                       return_hosed_msg(smi_info, err);
+
+               rv = SI_SM_CALL_WITHOUT_DELAY;
+       }
+ out:
+       return rv;
+}
+
+static void start_check_enables(struct smi_info *smi_info)
+{
+       unsigned char msg[2];
+
+       msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
+       msg[1] = IPMI_GET_BMC_GLOBAL_ENABLES_CMD;
+
+       smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
+       smi_info->si_state = SI_CHECKING_ENABLES;
+}
+
+static void start_clear_flags(struct smi_info *smi_info)
+{
+       unsigned char msg[3];
+
+       /* Make sure the watchdog pre-timeout flag is not set at startup. */
+       msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
+       msg[1] = IPMI_CLEAR_MSG_FLAGS_CMD;
+       msg[2] = WDT_PRE_TIMEOUT_INT;
+
+       smi_info->handlers->start_transaction(smi_info->si_sm, msg, 3);
+       smi_info->si_state = SI_CLEARING_FLAGS;
+}
+
+static void start_getting_msg_queue(struct smi_info *smi_info)
+{
+       smi_info->curr_msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
+       smi_info->curr_msg->data[1] = IPMI_GET_MSG_CMD;
+       smi_info->curr_msg->data_size = 2;
+
+       smi_info->handlers->start_transaction(
+               smi_info->si_sm,
+               smi_info->curr_msg->data,
+               smi_info->curr_msg->data_size);
+       smi_info->si_state = SI_GETTING_MESSAGES;
+}
+
+static void start_getting_events(struct smi_info *smi_info)
+{
+       smi_info->curr_msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
+       smi_info->curr_msg->data[1] = IPMI_READ_EVENT_MSG_BUFFER_CMD;
+       smi_info->curr_msg->data_size = 2;
+
+       smi_info->handlers->start_transaction(
+               smi_info->si_sm,
+               smi_info->curr_msg->data,
+               smi_info->curr_msg->data_size);
+       smi_info->si_state = SI_GETTING_EVENTS;
+}
+
+static void smi_mod_timer(struct smi_info *smi_info, unsigned long new_val)
+{
+       smi_info->last_timeout_jiffies = jiffies;
+       mod_timer(&smi_info->si_timer, new_val);
+       smi_info->timer_running = true;
+}
+
+/*
+ * When we have a situtaion where we run out of memory and cannot
+ * allocate messages, we just leave them in the BMC and run the system
+ * polled until we can allocate some memory.  Once we have some
+ * memory, we will re-enable the interrupt.
+ *
+ * Note that we cannot just use disable_irq(), since the interrupt may
+ * be shared.
+ */
+static inline bool disable_si_irq(struct smi_info *smi_info)
+{
+       if ((smi_info->irq) && (!smi_info->interrupt_disabled)) {
+               smi_info->interrupt_disabled = true;
+               start_check_enables(smi_info);
+               return true;
+       }
+       return false;
+}
+
+static inline bool enable_si_irq(struct smi_info *smi_info)
+{
+       if ((smi_info->irq) && (smi_info->interrupt_disabled)) {
+               smi_info->interrupt_disabled = false;
+               start_check_enables(smi_info);
+               return true;
+       }
+       return false;
+}
+
+/*
+ * Allocate a message.  If unable to allocate, start the interrupt
+ * disable process and return NULL.  If able to allocate but
+ * interrupts are disabled, free the message and return NULL after
+ * starting the interrupt enable process.
+ */
+static struct ipmi_smi_msg *alloc_msg_handle_irq(struct smi_info *smi_info)
+{
+       struct ipmi_smi_msg *msg;
+
+       msg = ipmi_alloc_smi_msg();
+       if (!msg) {
+               if (!disable_si_irq(smi_info))
+                       smi_info->si_state = SI_NORMAL;
+       } else if (enable_si_irq(smi_info)) {
+               ipmi_free_smi_msg(msg);
+               msg = NULL;
+       }
+       return msg;
+}
+
+static void handle_flags(struct smi_info *smi_info)
+{
+ retry:
+       if (smi_info->msg_flags & WDT_PRE_TIMEOUT_INT) {
+               /* Watchdog pre-timeout */
+               smi_inc_stat(smi_info, watchdog_pretimeouts);
+
+               start_clear_flags(smi_info);
+               smi_info->msg_flags &= ~WDT_PRE_TIMEOUT_INT;
+               if (smi_info->intf)
+                       ipmi_smi_watchdog_pretimeout(smi_info->intf);
+       } else if (smi_info->msg_flags & RECEIVE_MSG_AVAIL) {
+               /* Messages available. */
+               smi_info->curr_msg = alloc_msg_handle_irq(smi_info);
+               if (!smi_info->curr_msg)
+                       return;
+
+               start_getting_msg_queue(smi_info);
+       } else if (smi_info->msg_flags & EVENT_MSG_BUFFER_FULL) {
+               /* Events available. */
+               smi_info->curr_msg = alloc_msg_handle_irq(smi_info);
+               if (!smi_info->curr_msg)
+                       return;
+
+               start_getting_events(smi_info);
+       } else if (smi_info->msg_flags & OEM_DATA_AVAIL &&
+                  smi_info->oem_data_avail_handler) {
+               if (smi_info->oem_data_avail_handler(smi_info))
+                       goto retry;
+       } else
+               smi_info->si_state = SI_NORMAL;
+}
+
+/*
+ * Global enables we care about.
+ */
+#define GLOBAL_ENABLES_MASK (IPMI_BMC_EVT_MSG_BUFF | IPMI_BMC_RCV_MSG_INTR | \
+                            IPMI_BMC_EVT_MSG_INTR)
+
+static u8 current_global_enables(struct smi_info *smi_info, u8 base,
+                                bool *irq_on)
+{
+       u8 enables = 0;
+
+       if (smi_info->supports_event_msg_buff)
+               enables |= IPMI_BMC_EVT_MSG_BUFF;
+
+       if ((smi_info->irq && !smi_info->interrupt_disabled) ||
+           smi_info->cannot_clear_recv_irq_bit)
+               enables |= IPMI_BMC_RCV_MSG_INTR;
+
+       if (smi_info->supports_event_msg_buff &&
+           smi_info->irq && !smi_info->interrupt_disabled)
+
+               enables |= IPMI_BMC_EVT_MSG_INTR;
+
+       *irq_on = enables & (IPMI_BMC_EVT_MSG_INTR | IPMI_BMC_RCV_MSG_INTR);
+
+       return enables;
+}
+
+static void check_bt_irq(struct smi_info *smi_info, bool irq_on)
+{
+       u8 irqstate = smi_info->io.inputb(&smi_info->io, IPMI_BT_INTMASK_REG);
+
+       irqstate &= IPMI_BT_INTMASK_ENABLE_IRQ_BIT;
+
+       if ((bool)irqstate == irq_on)
+               return;
+
+       if (irq_on)
+               smi_info->io.outputb(&smi_info->io, IPMI_BT_INTMASK_REG,
+                                    IPMI_BT_INTMASK_ENABLE_IRQ_BIT);
+       else
+               smi_info->io.outputb(&smi_info->io, IPMI_BT_INTMASK_REG, 0);
+}
+
+static void handle_transaction_done(struct smi_info *smi_info)
+{
+       struct ipmi_smi_msg *msg;
+
+       debug_timestamp("Done");
+       switch (smi_info->si_state) {
+       case SI_NORMAL:
+               if (!smi_info->curr_msg)
+                       break;
+
+               smi_info->curr_msg->rsp_size
+                       = smi_info->handlers->get_result(
+                               smi_info->si_sm,
+                               smi_info->curr_msg->rsp,
+                               IPMI_MAX_MSG_LENGTH);
+
+               /*
+                * Do this here becase deliver_recv_msg() releases the
+                * lock, and a new message can be put in during the
+                * time the lock is released.
+                */
+               msg = smi_info->curr_msg;
+               smi_info->curr_msg = NULL;
+               deliver_recv_msg(smi_info, msg);
+               break;
+
+       case SI_GETTING_FLAGS:
+       {
+               unsigned char msg[4];
+               unsigned int  len;
+
+               /* We got the flags from the SMI, now handle them. */
+               len = smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
+               if (msg[2] != 0) {
+                       /* Error fetching flags, just give up for now. */
+                       smi_info->si_state = SI_NORMAL;
+               } else if (len < 4) {
+                       /*
+                        * Hmm, no flags.  That's technically illegal, but
+                        * don't use uninitialized data.
+                        */
+                       smi_info->si_state = SI_NORMAL;
+               } else {
+                       smi_info->msg_flags = msg[3];
+                       handle_flags(smi_info);
+               }
+               break;
+       }
+
+       case SI_CLEARING_FLAGS:
+       {
+               unsigned char msg[3];
+
+               /* We cleared the flags. */
+               smi_info->handlers->get_result(smi_info->si_sm, msg, 3);
+               if (msg[2] != 0) {
+                       /* Error clearing flags */
+                       dev_warn(smi_info->dev,
+                                "Error clearing flags: %2.2x\n", msg[2]);
+               }
+               smi_info->si_state = SI_NORMAL;
+               break;
+       }
+
+       case SI_GETTING_EVENTS:
+       {
+               smi_info->curr_msg->rsp_size
+                       = smi_info->handlers->get_result(
+                               smi_info->si_sm,
+                               smi_info->curr_msg->rsp,
+                               IPMI_MAX_MSG_LENGTH);
+
+               /*
+                * Do this here becase deliver_recv_msg() releases the
+                * lock, and a new message can be put in during the
+                * time the lock is released.
+                */
+               msg = smi_info->curr_msg;
+               smi_info->curr_msg = NULL;
+               if (msg->rsp[2] != 0) {
+                       /* Error getting event, probably done. */
+                       msg->done(msg);
+
+                       /* Take off the event flag. */
+                       smi_info->msg_flags &= ~EVENT_MSG_BUFFER_FULL;
+                       handle_flags(smi_info);
+               } else {
+                       smi_inc_stat(smi_info, events);
+
+                       /*
+                        * Do this before we deliver the message
+                        * because delivering the message releases the
+                        * lock and something else can mess with the
+                        * state.
+                        */
+                       handle_flags(smi_info);
+
+                       deliver_recv_msg(smi_info, msg);
+               }
+               break;
+       }
+
+       case SI_GETTING_MESSAGES:
+       {
+               smi_info->curr_msg->rsp_size
+                       = smi_info->handlers->get_result(
+                               smi_info->si_sm,
+                               smi_info->curr_msg->rsp,
+                               IPMI_MAX_MSG_LENGTH);
+
+               /*
+                * Do this here becase deliver_recv_msg() releases the
+                * lock, and a new message can be put in during the
+                * time the lock is released.
+                */
+               msg = smi_info->curr_msg;
+               smi_info->curr_msg = NULL;
+               if (msg->rsp[2] != 0) {
+                       /* Error getting event, probably done. */
+                       msg->done(msg);
+
+                       /* Take off the msg flag. */
+                       smi_info->msg_flags &= ~RECEIVE_MSG_AVAIL;
+                       handle_flags(smi_info);
+               } else {
+                       smi_inc_stat(smi_info, incoming_messages);
+
+                       /*
+                        * Do this before we deliver the message
+                        * because delivering the message releases the
+                        * lock and something else can mess with the
+                        * state.
+                        */
+                       handle_flags(smi_info);
+
+                       deliver_recv_msg(smi_info, msg);
+               }
+               break;
+       }
+
+       case SI_CHECKING_ENABLES:
+       {
+               unsigned char msg[4];
+               u8 enables;
+               bool irq_on;
+
+               /* We got the flags from the SMI, now handle them. */
+               smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
+               if (msg[2] != 0) {
+                       dev_warn(smi_info->dev,
+                                "Couldn't get irq info: %x.\n", msg[2]);
+                       dev_warn(smi_info->dev,
+                                "Maybe ok, but ipmi might run very slowly.\n");
+                       smi_info->si_state = SI_NORMAL;
+                       break;
+               }
+               enables = current_global_enables(smi_info, 0, &irq_on);
+               if (smi_info->si_type == SI_BT)
+                       /* BT has its own interrupt enable bit. */
+                       check_bt_irq(smi_info, irq_on);
+               if (enables != (msg[3] & GLOBAL_ENABLES_MASK)) {
+                       /* Enables are not correct, fix them. */
+                       msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
+                       msg[1] = IPMI_SET_BMC_GLOBAL_ENABLES_CMD;
+                       msg[2] = enables | (msg[3] & ~GLOBAL_ENABLES_MASK);
+                       smi_info->handlers->start_transaction(
+                               smi_info->si_sm, msg, 3);
+                       smi_info->si_state = SI_SETTING_ENABLES;
+               } else if (smi_info->supports_event_msg_buff) {
+                       smi_info->curr_msg = ipmi_alloc_smi_msg();
+                       if (!smi_info->curr_msg) {
+                               smi_info->si_state = SI_NORMAL;
+                               break;
+                       }
+                       start_getting_msg_queue(smi_info);
+               } else {
+                       smi_info->si_state = SI_NORMAL;
+               }
+               break;
+       }
+
+       case SI_SETTING_ENABLES:
+       {
+               unsigned char msg[4];
+
+               smi_info->handlers->get_result(smi_info->si_sm, msg, 4);
+               if (msg[2] != 0)
+                       dev_warn(smi_info->dev,
+                                "Could not set the global enables: 0x%x.\n",
+                                msg[2]);
+
+               if (smi_info->supports_event_msg_buff) {
+                       smi_info->curr_msg = ipmi_alloc_smi_msg();
+                       if (!smi_info->curr_msg) {
+                               smi_info->si_state = SI_NORMAL;
+                               break;
+                       }
+                       start_getting_msg_queue(smi_info);
+               } else {
+                       smi_info->si_state = SI_NORMAL;
+               }
+               break;
+       }
+       }
+}
+
+/*
+ * Called on timeouts and events.  Timeouts should pass the elapsed
+ * time, interrupts should pass in zero.  Must be called with
+ * si_lock held and interrupts disabled.
+ */
+static enum si_sm_result smi_event_handler(struct smi_info *smi_info,
+                                          int time)
+{
+       enum si_sm_result si_sm_result;
+
+ restart:
+       /*
+        * There used to be a loop here that waited a little while
+        * (around 25us) before giving up.  That turned out to be
+        * pointless, the minimum delays I was seeing were in the 300us
+        * range, which is far too long to wait in an interrupt.  So
+        * we just run until the state machine tells us something
+        * happened or it needs a delay.
+        */
+       si_sm_result = smi_info->handlers->event(smi_info->si_sm, time);
+       time = 0;
+       while (si_sm_result == SI_SM_CALL_WITHOUT_DELAY)
+               si_sm_result = smi_info->handlers->event(smi_info->si_sm, 0);
+
+       if (si_sm_result == SI_SM_TRANSACTION_COMPLETE) {
+               smi_inc_stat(smi_info, complete_transactions);
+
+               handle_transaction_done(smi_info);
+               si_sm_result = smi_info->handlers->event(smi_info->si_sm, 0);
+       } else if (si_sm_result == SI_SM_HOSED) {
+               smi_inc_stat(smi_info, hosed_count);
+
+               /*
+                * Do the before return_hosed_msg, because that
+                * releases the lock.
+                */
+               smi_info->si_state = SI_NORMAL;
+               if (smi_info->curr_msg != NULL) {
+                       /*
+                        * If we were handling a user message, format
+                        * a response to send to the upper layer to
+                        * tell it about the error.
+                        */
+                       return_hosed_msg(smi_info, IPMI_ERR_UNSPECIFIED);
+               }
+               si_sm_result = smi_info->handlers->event(smi_info->si_sm, 0);
+       }
+
+       /*
+        * We prefer handling attn over new messages.  But don't do
+        * this if there is not yet an upper layer to handle anything.
+        */
+       if (likely(smi_info->intf) &&
+           (si_sm_result == SI_SM_ATTN || smi_info->got_attn)) {
+               unsigned char msg[2];
+
+               if (smi_info->si_state != SI_NORMAL) {
+                       /*
+                        * We got an ATTN, but we are doing something else.
+                        * Handle the ATTN later.
+                        */
+                       smi_info->got_attn = true;
+               } else {
+                       smi_info->got_attn = false;
+                       smi_inc_stat(smi_info, attentions);
+
+                       /*
+                        * Got a attn, send down a get message flags to see
+                        * what's causing it.  It would be better to handle
+                        * this in the upper layer, but due to the way
+                        * interrupts work with the SMI, that's not really
+                        * possible.
+                        */
+                       msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
+                       msg[1] = IPMI_GET_MSG_FLAGS_CMD;
+
+                       smi_info->handlers->start_transaction(
+                               smi_info->si_sm, msg, 2);
+                       smi_info->si_state = SI_GETTING_FLAGS;
+                       goto restart;
+               }
+       }
+
+       /* If we are currently idle, try to start the next message. */
+       if (si_sm_result == SI_SM_IDLE) {
+               smi_inc_stat(smi_info, idles);
+
+               si_sm_result = start_next_msg(smi_info);
+               if (si_sm_result != SI_SM_IDLE)
+                       goto restart;
+       }
+
+       if ((si_sm_result == SI_SM_IDLE)
+           && (atomic_read(&smi_info->req_events))) {
+               /*
+                * We are idle and the upper layer requested that I fetch
+                * events, so do so.
+                */
+               atomic_set(&smi_info->req_events, 0);
+
+               /*
+                * Take this opportunity to check the interrupt and
+                * message enable state for the BMC.  The BMC can be
+                * asynchronously reset, and may thus get interrupts
+                * disable and messages disabled.
+                */
+               if (smi_info->supports_event_msg_buff || smi_info->irq) {
+                       start_check_enables(smi_info);
+               } else {
+                       smi_info->curr_msg = alloc_msg_handle_irq(smi_info);
+                       if (!smi_info->curr_msg)
+                               goto out;
+
+                       start_getting_events(smi_info);
+               }
+               goto restart;
+       }
+ out:
+       return si_sm_result;
+}
+
+static void check_start_timer_thread(struct smi_info *smi_info)
+{
+       if (smi_info->si_state == SI_NORMAL && smi_info->curr_msg == NULL) {
+               smi_mod_timer(smi_info, jiffies + SI_TIMEOUT_JIFFIES);
+
+               if (smi_info->thread)
+                       wake_up_process(smi_info->thread);
+
+               start_next_msg(smi_info);
+               smi_event_handler(smi_info, 0);
+       }
+}
+
+static void sender(void                *send_info,
+                  struct ipmi_smi_msg *msg)
+{
+       struct smi_info   *smi_info = send_info;
+       enum si_sm_result result;
+       unsigned long     flags;
+
+       debug_timestamp("Enqueue");
+
+       if (smi_info->run_to_completion) {
+               /*
+                * If we are running to completion, start it and run
+                * transactions until everything is clear.
+                */
+               smi_info->waiting_msg = msg;
+
+               /*
+                * Run to completion means we are single-threaded, no
+                * need for locks.
+                */
+
+               result = smi_event_handler(smi_info, 0);
+               while (result != SI_SM_IDLE) {
+                       udelay(SI_SHORT_TIMEOUT_USEC);
+                       result = smi_event_handler(smi_info,
+                                                  SI_SHORT_TIMEOUT_USEC);
+               }
+               return;
+       }
+
+       spin_lock_irqsave(&smi_info->si_lock, flags);
+       /*
+        * The following two lines don't need to be under the lock for
+        * the lock's sake, but they do need SMP memory barriers to
+        * avoid getting things out of order.  We are already claiming
+        * the lock, anyway, so just do it under the lock to avoid the
+        * ordering problem.
+        */
+       BUG_ON(smi_info->waiting_msg);
+       smi_info->waiting_msg = msg;
+       check_start_timer_thread(smi_info);
+       spin_unlock_irqrestore(&smi_info->si_lock, flags);
+}
+
+static void set_run_to_completion(void *send_info, bool i_run_to_completion)
+{
+       struct smi_info   *smi_info = send_info;
+       enum si_sm_result result;
+
+       smi_info->run_to_completion = i_run_to_completion;
+       if (i_run_to_completion) {
+               result = smi_event_handler(smi_info, 0);
+               while (result != SI_SM_IDLE) {
+                       udelay(SI_SHORT_TIMEOUT_USEC);
+                       result = smi_event_handler(smi_info,
+                                                  SI_SHORT_TIMEOUT_USEC);
+               }
+       }
+}
+
+/*
+ * Use -1 in the nsec value of the busy waiting timespec to tell that
+ * we are spinning in kipmid looking for something and not delaying
+ * between checks
+ */
+static inline void ipmi_si_set_not_busy(struct timespec64 *ts)
+{
+       ts->tv_nsec = -1;
+}
+static inline int ipmi_si_is_busy(struct timespec64 *ts)
+{
+       return ts->tv_nsec != -1;
+}
+
+static inline int ipmi_thread_busy_wait(enum si_sm_result smi_result,
+                                       const struct smi_info *smi_info,
+                                       struct timespec64 *busy_until)
+{
+       unsigned int max_busy_us = 0;
+
+       if (smi_info->intf_num < num_max_busy_us)
+               max_busy_us = kipmid_max_busy_us[smi_info->intf_num];
+       if (max_busy_us == 0 || smi_result != SI_SM_CALL_WITH_DELAY)
+               ipmi_si_set_not_busy(busy_until);
+       else if (!ipmi_si_is_busy(busy_until)) {
+               getnstimeofday64(busy_until);
+               timespec64_add_ns(busy_until, max_busy_us*NSEC_PER_USEC);
+       } else {
+               struct timespec64 now;
+
+               getnstimeofday64(&now);
+               if (unlikely(timespec64_compare(&now, busy_until) > 0)) {
+                       ipmi_si_set_not_busy(busy_until);
+                       return 0;
+               }
+       }
+       return 1;
+}
+
+
+/*
+ * A busy-waiting loop for speeding up IPMI operation.
+ *
+ * Lousy hardware makes this hard.  This is only enabled for systems
+ * that are not BT and do not have interrupts.  It starts spinning
+ * when an operation is complete or until max_busy tells it to stop
+ * (if that is enabled).  See the paragraph on kimid_max_busy_us in
+ * Documentation/IPMI.txt for details.
+ */
+static int ipmi_thread(void *data)
+{
+       struct smi_info *smi_info = data;
+       unsigned long flags;
+       enum si_sm_result smi_result;
+       struct timespec64 busy_until;
+
+       ipmi_si_set_not_busy(&busy_until);
+       set_user_nice(current, MAX_NICE);
+       while (!kthread_should_stop()) {
+               int busy_wait;
+
+               spin_lock_irqsave(&(smi_info->si_lock), flags);
+               smi_result = smi_event_handler(smi_info, 0);
+
+               /*
+                * If the driver is doing something, there is a possible
+                * race with the timer.  If the timer handler see idle,
+                * and the thread here sees something else, the timer
+                * handler won't restart the timer even though it is
+                * required.  So start it here if necessary.
+                */
+               if (smi_result != SI_SM_IDLE && !smi_info->timer_running)
+                       smi_mod_timer(smi_info, jiffies + SI_TIMEOUT_JIFFIES);
+
+               spin_unlock_irqrestore(&(smi_info->si_lock), flags);
+               busy_wait = ipmi_thread_busy_wait(smi_result, smi_info,
+                                                 &busy_until);
+               if (smi_result == SI_SM_CALL_WITHOUT_DELAY)
+                       ; /* do nothing */
+               else if (smi_result == SI_SM_CALL_WITH_DELAY && busy_wait)
+                       schedule();
+               else if (smi_result == SI_SM_IDLE) {
+                       if (atomic_read(&smi_info->need_watch)) {
+                               schedule_timeout_interruptible(100);
+                       } else {
+                               /* Wait to be woken up when we are needed. */
+                               __set_current_state(TASK_INTERRUPTIBLE);
+                               schedule();
+                       }
+               } else
+                       schedule_timeout_interruptible(1);
+       }
+       return 0;
+}
+
+
+static void poll(void *send_info)
+{
+       struct smi_info *smi_info = send_info;
+       unsigned long flags = 0;
+       bool run_to_completion = smi_info->run_to_completion;
+
+       /*
+        * Make sure there is some delay in the poll loop so we can
+        * drive time forward and timeout things.
+        */
+       udelay(10);
+       if (!run_to_completion)
+               spin_lock_irqsave(&smi_info->si_lock, flags);
+       smi_event_handler(smi_info, 10);
+       if (!run_to_completion)
+               spin_unlock_irqrestore(&smi_info->si_lock, flags);
+}
+
+static void request_events(void *send_info)
+{
+       struct smi_info *smi_info = send_info;
+
+       if (!smi_info->has_event_buffer)
+               return;
+
+       atomic_set(&smi_info->req_events, 1);
+}
+
+static void set_need_watch(void *send_info, bool enable)
+{
+       struct smi_info *smi_info = send_info;
+       unsigned long flags;
+
+       atomic_set(&smi_info->need_watch, enable);
+       spin_lock_irqsave(&smi_info->si_lock, flags);
+       check_start_timer_thread(smi_info);
+       spin_unlock_irqrestore(&smi_info->si_lock, flags);
+}
+
+static int initialized;
+
+static void smi_timeout(unsigned long data)
+{
+       struct smi_info   *smi_info = (struct smi_info *) data;
+       enum si_sm_result smi_result;
+       unsigned long     flags;
+       unsigned long     jiffies_now;
+       long              time_diff;
+       long              timeout;
+
+       spin_lock_irqsave(&(smi_info->si_lock), flags);
+       debug_timestamp("Timer");
+
+       jiffies_now = jiffies;
+       time_diff = (((long)jiffies_now - (long)smi_info->last_timeout_jiffies)
+                    * SI_USEC_PER_JIFFY);
+       smi_result = smi_event_handler(smi_info, time_diff);
+
+       if ((smi_info->irq) && (!smi_info->interrupt_disabled)) {
+               /* Running with interrupts, only do long timeouts. */
+               timeout = jiffies + SI_TIMEOUT_JIFFIES;
+               smi_inc_stat(smi_info, long_timeouts);
+               goto do_mod_timer;
+       }
+
+       /*
+        * If the state machine asks for a short delay, then shorten
+        * the timer timeout.
+        */
+       if (smi_result == SI_SM_CALL_WITH_DELAY) {
+               smi_inc_stat(smi_info, short_timeouts);
+               timeout = jiffies + 1;
+       } else {
+               smi_inc_stat(smi_info, long_timeouts);
+               timeout = jiffies + SI_TIMEOUT_JIFFIES;
+       }
+
+ do_mod_timer:
+       if (smi_result != SI_SM_IDLE)
+               smi_mod_timer(smi_info, timeout);
+       else
+               smi_info->timer_running = false;
+       spin_unlock_irqrestore(&(smi_info->si_lock), flags);
+}
+
+static irqreturn_t si_irq_handler(int irq, void *data)
+{
+       struct smi_info *smi_info = data;
+       unsigned long   flags;
+
+       spin_lock_irqsave(&(smi_info->si_lock), flags);
+
+       smi_inc_stat(smi_info, interrupts);
+
+       debug_timestamp("Interrupt");
+
+       smi_event_handler(smi_info, 0);
+       spin_unlock_irqrestore(&(smi_info->si_lock), flags);
+       return IRQ_HANDLED;
+}
+
+static irqreturn_t si_bt_irq_handler(int irq, void *data)
+{
+       struct smi_info *smi_info = data;
+       /* We need to clear the IRQ flag for the BT interface. */
+       smi_info->io.outputb(&smi_info->io, IPMI_BT_INTMASK_REG,
+                            IPMI_BT_INTMASK_CLEAR_IRQ_BIT
+                            | IPMI_BT_INTMASK_ENABLE_IRQ_BIT);
+       return si_irq_handler(irq, data);
+}
+
+static int smi_start_processing(void       *send_info,
+                               ipmi_smi_t intf)
+{
+       struct smi_info *new_smi = send_info;
+       int             enable = 0;
+
+       new_smi->intf = intf;
+
+       /* Try to claim any interrupts. */
+       if (new_smi->irq_setup)
+               new_smi->irq_setup(new_smi);
+
+       /* Set up the timer that drives the interface. */
+       setup_timer(&new_smi->si_timer, smi_timeout, (long)new_smi);
+       smi_mod_timer(new_smi, jiffies + SI_TIMEOUT_JIFFIES);
+
+       /*
+        * Check if the user forcefully enabled the daemon.
+        */
+       if (new_smi->intf_num < num_force_kipmid)
+               enable = force_kipmid[new_smi->intf_num];
+       /*
+        * The BT interface is efficient enough to not need a thread,
+        * and there is no need for a thread if we have interrupts.
+        */
+       else if ((new_smi->si_type != SI_BT) && (!new_smi->irq))
+               enable = 1;
+
+       if (enable) {
+               new_smi->thread = kthread_run(ipmi_thread, new_smi,
+                                             "kipmi%d", new_smi->intf_num);
+               if (IS_ERR(new_smi->thread)) {
+                       dev_notice(new_smi->dev, "Could not start"
+                                  " kernel thread due to error %ld, only using"
+                                  " timers to drive the interface\n",
+                                  PTR_ERR(new_smi->thread));
+                       new_smi->thread = NULL;
+               }
+       }
+
+       return 0;
+}
+
+static int get_smi_info(void *send_info, struct ipmi_smi_info *data)
+{
+       struct smi_info *smi = send_info;
+
+       data->addr_src = smi->addr_source;
+       data->dev = smi->dev;
+       data->addr_info = smi->addr_info;
+       get_device(smi->dev);
+
+       return 0;
+}
+
+static void set_maintenance_mode(void *send_info, bool enable)
+{
+       struct smi_info   *smi_info = send_info;
+
+       if (!enable)
+               atomic_set(&smi_info->req_events, 0);
+}
+
+static struct ipmi_smi_handlers handlers = {
+       .owner                  = THIS_MODULE,
+       .start_processing       = smi_start_processing,
+       .get_smi_info           = get_smi_info,
+       .sender                 = sender,
+       .request_events         = request_events,
+       .set_need_watch         = set_need_watch,
+       .set_maintenance_mode   = set_maintenance_mode,
+       .set_run_to_completion  = set_run_to_completion,
+       .poll                   = poll,
+};
+
+/*
+ * There can be 4 IO ports passed in (with or without IRQs), 4 addresses,
+ * a default IO port, and 1 ACPI/SPMI address.  That sets SI_MAX_DRIVERS.
+ */
+
+static LIST_HEAD(smi_infos);
+static DEFINE_MUTEX(smi_infos_lock);
+static int smi_num; /* Used to sequence the SMIs */
+
+#define DEFAULT_REGSPACING     1
+#define DEFAULT_REGSIZE                1
+
+#ifdef CONFIG_ACPI
+static bool          si_tryacpi = 1;
+#endif
+#ifdef CONFIG_DMI
+static bool          si_trydmi = 1;
+#endif
+static bool          si_tryplatform = 1;
+#ifdef CONFIG_PCI
+static bool          si_trypci = 1;
+#endif
+static bool          si_trydefaults = IS_ENABLED(CONFIG_IPMI_SI_PROBE_DEFAULTS);
+static char          *si_type[SI_MAX_PARMS];
+#define MAX_SI_TYPE_STR 30
+static char          si_type_str[MAX_SI_TYPE_STR];
+static unsigned long addrs[SI_MAX_PARMS];
+static unsigned int num_addrs;
+static unsigned int  ports[SI_MAX_PARMS];
+static unsigned int num_ports;
+static int           irqs[SI_MAX_PARMS];
+static unsigned int num_irqs;
+static int           regspacings[SI_MAX_PARMS];
+static unsigned int num_regspacings;
+static int           regsizes[SI_MAX_PARMS];
+static unsigned int num_regsizes;
+static int           regshifts[SI_MAX_PARMS];
+static unsigned int num_regshifts;
+static int slave_addrs[SI_MAX_PARMS]; /* Leaving 0 chooses the default value */
+static unsigned int num_slave_addrs;
+
+#define IPMI_IO_ADDR_SPACE  0
+#define IPMI_MEM_ADDR_SPACE 1
+static char *addr_space_to_str[] = { "i/o", "mem" };
+
+static int hotmod_handler(const char *val, struct kernel_param *kp);
+
+module_param_call(hotmod, hotmod_handler, NULL, NULL, 0200);
+MODULE_PARM_DESC(hotmod, "Add and remove interfaces.  See"
+                " Documentation/IPMI.txt in the kernel sources for the"
+                " gory details.");
+
+#ifdef CONFIG_ACPI
+module_param_named(tryacpi, si_tryacpi, bool, 0);
+MODULE_PARM_DESC(tryacpi, "Setting this to zero will disable the"
+                " default scan of the interfaces identified via ACPI");
+#endif
+#ifdef CONFIG_DMI
+module_param_named(trydmi, si_trydmi, bool, 0);
+MODULE_PARM_DESC(trydmi, "Setting this to zero will disable the"
+                " default scan of the interfaces identified via DMI");
+#endif
+module_param_named(tryplatform, si_tryplatform, bool, 0);
+MODULE_PARM_DESC(tryacpi, "Setting this to zero will disable the"
+                " default scan of the interfaces identified via platform"
+                " interfaces like openfirmware");
+#ifdef CONFIG_PCI
+module_param_named(trypci, si_trypci, bool, 0);
+MODULE_PARM_DESC(tryacpi, "Setting this to zero will disable the"
+                " default scan of the interfaces identified via pci");
+#endif
+module_param_named(trydefaults, si_trydefaults, bool, 0);
+MODULE_PARM_DESC(trydefaults, "Setting this to 'false' will disable the"
+                " default scan of the KCS and SMIC interface at the standard"
+                " address");
+module_param_string(type, si_type_str, MAX_SI_TYPE_STR, 0);
+MODULE_PARM_DESC(type, "Defines the type of each interface, each"
+                " interface separated by commas.  The types are 'kcs',"
+                " 'smic', and 'bt'.  For example si_type=kcs,bt will set"
+                " the first interface to kcs and the second to bt");
+module_param_array(addrs, ulong, &num_addrs, 0);
+MODULE_PARM_DESC(addrs, "Sets the memory address of each interface, the"
+                " addresses separated by commas.  Only use if an interface"
+                " is in memory.  Otherwise, set it to zero or leave"
+                " it blank.");
+module_param_array(ports, uint, &num_ports, 0);
+MODULE_PARM_DESC(ports, "Sets the port address of each interface, the"
+                " addresses separated by commas.  Only use if an interface"
+                " is a port.  Otherwise, set it to zero or leave"
+                " it blank.");
+module_param_array(irqs, int, &num_irqs, 0);
+MODULE_PARM_DESC(irqs, "Sets the interrupt of each interface, the"
+                " addresses separated by commas.  Only use if an interface"
+                " has an interrupt.  Otherwise, set it to zero or leave"
+                " it blank.");
+module_param_array(regspacings, int, &num_regspacings, 0);
+MODULE_PARM_DESC(regspacings, "The number of bytes between the start address"
+                " and each successive register used by the interface.  For"
+                " instance, if the start address is 0xca2 and the spacing"
+                " is 2, then the second address is at 0xca4.  Defaults"
+                " to 1.");
+module_param_array(regsizes, int, &num_regsizes, 0);
+MODULE_PARM_DESC(regsizes, "The size of the specific IPMI register in bytes."
+                " This should generally be 1, 2, 4, or 8 for an 8-bit,"
+                " 16-bit, 32-bit, or 64-bit register.  Use this if you"
+                " the 8-bit IPMI register has to be read from a larger"
+                " register.");
+module_param_array(regshifts, int, &num_regshifts, 0);
+MODULE_PARM_DESC(regshifts, "The amount to shift the data read from the."
+                " IPMI register, in bits.  For instance, if the data"
+                " is read from a 32-bit word and the IPMI data is in"
+                " bit 8-15, then the shift would be 8");
+module_param_array(slave_addrs, int, &num_slave_addrs, 0);
+MODULE_PARM_DESC(slave_addrs, "Set the default IPMB slave address for"
+                " the controller.  Normally this is 0x20, but can be"
+                " overridden by this parm.  This is an array indexed"
+                " by interface number.");
+module_param_array(force_kipmid, int, &num_force_kipmid, 0);
+MODULE_PARM_DESC(force_kipmid, "Force the kipmi daemon to be enabled (1) or"
+                " disabled(0).  Normally the IPMI driver auto-detects"
+                " this, but the value may be overridden by this parm.");
+module_param(unload_when_empty, bool, 0);
+MODULE_PARM_DESC(unload_when_empty, "Unload the module if no interfaces are"
+                " specified or found, default is 1.  Setting to 0"
+                " is useful for hot add of devices using hotmod.");
+module_param_array(kipmid_max_busy_us, uint, &num_max_busy_us, 0644);
+MODULE_PARM_DESC(kipmid_max_busy_us,
+                "Max time (in microseconds) to busy-wait for IPMI data before"
+                " sleeping. 0 (default) means to wait forever. Set to 100-500"
+                " if kipmid is using up a lot of CPU time.");
+
+
+static void std_irq_cleanup(struct smi_info *info)
+{
+       if (info->si_type == SI_BT)
+               /* Disable the interrupt in the BT interface. */
+               info->io.outputb(&info->io, IPMI_BT_INTMASK_REG, 0);
+       free_irq(info->irq, info);
+}
+
+static int std_irq_setup(struct smi_info *info)
+{
+       int rv;
+
+       if (!info->irq)
+               return 0;
+
+       if (info->si_type == SI_BT) {
+               rv = request_irq(info->irq,
+                                si_bt_irq_handler,
+                                IRQF_SHARED,
+                                DEVICE_NAME,
+                                info);
+               if (!rv)
+                       /* Enable the interrupt in the BT interface. */
+                       info->io.outputb(&info->io, IPMI_BT_INTMASK_REG,
+                                        IPMI_BT_INTMASK_ENABLE_IRQ_BIT);
+       } else
+               rv = request_irq(info->irq,
+                                si_irq_handler,
+                                IRQF_SHARED,
+                                DEVICE_NAME,
+                                info);
+       if (rv) {
+               dev_warn(info->dev, "%s unable to claim interrupt %d,"
+                        " running polled\n",
+                        DEVICE_NAME, info->irq);
+               info->irq = 0;
+       } else {
+               info->irq_cleanup = std_irq_cleanup;
+               dev_info(info->dev, "Using irq %d\n", info->irq);
+       }
+
+       return rv;
+}
+
+static unsigned char port_inb(struct si_sm_io *io, unsigned int offset)
+{
+       unsigned int addr = io->addr_data;
+
+       return inb(addr + (offset * io->regspacing));
+}
+
+static void port_outb(struct si_sm_io *io, unsigned int offset,
+                     unsigned char b)
+{
+       unsigned int addr = io->addr_data;
+
+       outb(b, addr + (offset * io->regspacing));
+}
+
+static unsigned char port_inw(struct si_sm_io *io, unsigned int offset)
+{
+       unsigned int addr = io->addr_data;
+
+       return (inw(addr + (offset * io->regspacing)) >> io->regshift) & 0xff;
+}
+
+static void port_outw(struct si_sm_io *io, unsigned int offset,
+                     unsigned char b)
+{
+       unsigned int addr = io->addr_data;
+
+       outw(b << io->regshift, addr + (offset * io->regspacing));
+}
+
+static unsigned char port_inl(struct si_sm_io *io, unsigned int offset)
+{
+       unsigned int addr = io->addr_data;
+
+       return (inl(addr + (offset * io->regspacing)) >> io->regshift) & 0xff;
+}
+
+static void port_outl(struct si_sm_io *io, unsigned int offset,
+                     unsigned char b)
+{
+       unsigned int addr = io->addr_data;
+
+       outl(b << io->regshift, addr+(offset * io->regspacing));
+}
+
+static void port_cleanup(struct smi_info *info)
+{
+       unsigned int addr = info->io.addr_data;
+       int          idx;
+
+       if (addr) {
+               for (idx = 0; idx < info->io_size; idx++)
+                       release_region(addr + idx * info->io.regspacing,
+                                      info->io.regsize);
+       }
+}
+
+static int port_setup(struct smi_info *info)
+{
+       unsigned int addr = info->io.addr_data;
+       int          idx;
+
+       if (!addr)
+               return -ENODEV;
+
+       info->io_cleanup = port_cleanup;
+
+       /*
+        * Figure out the actual inb/inw/inl/etc routine to use based
+        * upon the register size.
+        */
+       switch (info->io.regsize) {
+       case 1:
+               info->io.inputb = port_inb;
+               info->io.outputb = port_outb;
+               break;
+       case 2:
+               info->io.inputb = port_inw;
+               info->io.outputb = port_outw;
+               break;
+       case 4:
+               info->io.inputb = port_inl;
+               info->io.outputb = port_outl;
+               break;
+       default:
+               dev_warn(info->dev, "Invalid register size: %d\n",
+                        info->io.regsize);
+               return -EINVAL;
+       }
+
+       /*
+        * Some BIOSes reserve disjoint I/O regions in their ACPI
+        * tables.  This causes problems when trying to register the
+        * entire I/O region.  Therefore we must register each I/O
+        * port separately.
+        */
+       for (idx = 0; idx < info->io_size; idx++) {
+               if (request_region(addr + idx * info->io.regspacing,
+                                  info->io.regsize, DEVICE_NAME) == NULL) {
+                       /* Undo allocations */
+                       while (idx--) {
+                               release_region(addr + idx * info->io.regspacing,
+                                              info->io.regsize);
+                       }
+                       return -EIO;
+               }
+       }
+       return 0;
+}
+
+static unsigned char intf_mem_inb(struct si_sm_io *io, unsigned int offset)
+{
+       return readb((io->addr)+(offset * io->regspacing));
+}
+
+static void intf_mem_outb(struct si_sm_io *io, unsigned int offset,
+                    unsigned char b)
+{
+       writeb(b, (io->addr)+(offset * io->regspacing));
+}
+
+static unsigned char intf_mem_inw(struct si_sm_io *io, unsigned int offset)
+{
+       return (readw((io->addr)+(offset * io->regspacing)) >> io->regshift)
+               & 0xff;
+}
+
+static void intf_mem_outw(struct si_sm_io *io, unsigned int offset,
+                    unsigned char b)
+{
+       writeb(b << io->regshift, (io->addr)+(offset * io->regspacing));
+}
+
+static unsigned char intf_mem_inl(struct si_sm_io *io, unsigned int offset)
+{
+       return (readl((io->addr)+(offset * io->regspacing)) >> io->regshift)
+               & 0xff;
+}
+
+static void intf_mem_outl(struct si_sm_io *io, unsigned int offset,
+                    unsigned char b)
+{
+       writel(b << io->regshift, (io->addr)+(offset * io->regspacing));
+}
+
+#ifdef readq
+static unsigned char mem_inq(struct si_sm_io *io, unsigned int offset)
+{
+       return (readq((io->addr)+(offset * io->regspacing)) >> io->regshift)
+               & 0xff;
+}
+
+static void mem_outq(struct si_sm_io *io, unsigned int offset,
+                    unsigned char b)
+{
+       writeq(b << io->regshift, (io->addr)+(offset * io->regspacing));
+}
+#endif
+
+static void mem_cleanup(struct smi_info *info)
+{
+       unsigned long addr = info->io.addr_data;
+       int           mapsize;
+
+       if (info->io.addr) {
+               iounmap(info->io.addr);
+
+               mapsize = ((info->io_size * info->io.regspacing)
+                          - (info->io.regspacing - info->io.regsize));
+
+               release_mem_region(addr, mapsize);
+       }
+}
+
+static int mem_setup(struct smi_info *info)
+{
+       unsigned long addr = info->io.addr_data;
+       int           mapsize;
+
+       if (!addr)
+               return -ENODEV;
+
+       info->io_cleanup = mem_cleanup;
+
+       /*
+        * Figure out the actual readb/readw/readl/etc routine to use based
+        * upon the register size.
+        */
+       switch (info->io.regsize) {
+       case 1:
+               info->io.inputb = intf_mem_inb;
+               info->io.outputb = intf_mem_outb;
+               break;
+       case 2:
+               info->io.inputb = intf_mem_inw;
+               info->io.outputb = intf_mem_outw;
+               break;
+       case 4:
+               info->io.inputb = intf_mem_inl;
+               info->io.outputb = intf_mem_outl;
+               break;
+#ifdef readq
+       case 8:
+               info->io.inputb = mem_inq;
+               info->io.outputb = mem_outq;
+               break;
+#endif
+       default:
+               dev_warn(info->dev, "Invalid register size: %d\n",
+                        info->io.regsize);
+               return -EINVAL;
+       }
+
+       /*
+        * Calculate the total amount of memory to claim.  This is an
+        * unusual looking calculation, but it avoids claiming any
+        * more memory than it has to.  It will claim everything
+        * between the first address to the end of the last full
+        * register.
+        */
+       mapsize = ((info->io_size * info->io.regspacing)
+                  - (info->io.regspacing - info->io.regsize));
+
+       if (request_mem_region(addr, mapsize, DEVICE_NAME) == NULL)
+               return -EIO;
+
+       info->io.addr = ioremap(addr, mapsize);
+       if (info->io.addr == NULL) {
+               release_mem_region(addr, mapsize);
+               return -EIO;
+       }
+       return 0;
+}
+
+/*
+ * Parms come in as <op1>[:op2[:op3...]].  ops are:
+ *   add|remove,kcs|bt|smic,mem|i/o,<address>[,<opt1>[,<opt2>[,...]]]
+ * Options are:
+ *   rsp=<regspacing>
+ *   rsi=<regsize>
+ *   rsh=<regshift>
+ *   irq=<irq>
+ *   ipmb=<ipmb addr>
+ */
+enum hotmod_op { HM_ADD, HM_REMOVE };
+struct hotmod_vals {
+       char *name;
+       int  val;
+};
+static struct hotmod_vals hotmod_ops[] = {
+       { "add",        HM_ADD },
+       { "remove",     HM_REMOVE },
+       { NULL }
+};
+static struct hotmod_vals hotmod_si[] = {
+       { "kcs",        SI_KCS },
+       { "smic",       SI_SMIC },
+       { "bt",         SI_BT },
+       { NULL }
+};
+static struct hotmod_vals hotmod_as[] = {
+       { "mem",        IPMI_MEM_ADDR_SPACE },
+       { "i/o",        IPMI_IO_ADDR_SPACE },
+       { NULL }
+};
+
+static int parse_str(struct hotmod_vals *v, int *val, char *name, char **curr)
+{
+       char *s;
+       int  i;
+
+       s = strchr(*curr, ',');
+       if (!s) {
+               printk(KERN_WARNING PFX "No hotmod %s given.\n", name);
+               return -EINVAL;
+       }
+       *s = '\0';
+       s++;
+       for (i = 0; v[i].name; i++) {
+               if (strcmp(*curr, v[i].name) == 0) {
+                       *val = v[i].val;
+                       *curr = s;
+                       return 0;
+               }
+       }
+
+       printk(KERN_WARNING PFX "Invalid hotmod %s '%s'\n", name, *curr);
+       return -EINVAL;
+}
+
+static int check_hotmod_int_op(const char *curr, const char *option,
+                              const char *name, int *val)
+{
+       char *n;
+
+       if (strcmp(curr, name) == 0) {
+               if (!option) {
+                       printk(KERN_WARNING PFX
+                              "No option given for '%s'\n",
+                              curr);
+                       return -EINVAL;
+               }
+               *val = simple_strtoul(option, &n, 0);
+               if ((*n != '\0') || (*option == '\0')) {
+                       printk(KERN_WARNING PFX
+                              "Bad option given for '%s'\n",
+                              curr);
+                       return -EINVAL;
+               }
+               return 1;
+       }
+       return 0;
+}
+
+static struct smi_info *smi_info_alloc(void)
+{
+       struct smi_info *info = kzalloc(sizeof(*info), GFP_KERNEL);
+
+       if (info)
+               spin_lock_init(&info->si_lock);
+       return info;
+}
+
+static int hotmod_handler(const char *val, struct kernel_param *kp)
+{
+       char *str = kstrdup(val, GFP_KERNEL);
+       int  rv;
+       char *next, *curr, *s, *n, *o;
+       enum hotmod_op op;
+       enum si_type si_type;
+       int  addr_space;
+       unsigned long addr;
+       int regspacing;
+       int regsize;
+       int regshift;
+       int irq;
+       int ipmb;
+       int ival;
+       int len;
+       struct smi_info *info;
+
+       if (!str)
+               return -ENOMEM;
+
+       /* Kill any trailing spaces, as we can get a "\n" from echo. */
+       len = strlen(str);
+       ival = len - 1;
+       while ((ival >= 0) && isspace(str[ival])) {
+               str[ival] = '\0';
+               ival--;
+       }
+
+       for (curr = str; curr; curr = next) {
+               regspacing = 1;
+               regsize = 1;
+               regshift = 0;
+               irq = 0;
+               ipmb = 0; /* Choose the default if not specified */
+
+               next = strchr(curr, ':');
+               if (next) {
+                       *next = '\0';
+                       next++;
+               }
+
+               rv = parse_str(hotmod_ops, &ival, "operation", &curr);
+               if (rv)
+                       break;
+               op = ival;
+
+               rv = parse_str(hotmod_si, &ival, "interface type", &curr);
+               if (rv)
+                       break;
+               si_type = ival;
+
+               rv = parse_str(hotmod_as, &addr_space, "address space", &curr);
+               if (rv)
+                       break;
+
+               s = strchr(curr, ',');
+               if (s) {
+                       *s = '\0';
+                       s++;
+               }
+               addr = simple_strtoul(curr, &n, 0);
+               if ((*n != '\0') || (*curr == '\0')) {
+                       printk(KERN_WARNING PFX "Invalid hotmod address"
+                              " '%s'\n", curr);
+                       break;
+               }
+
+               while (s) {
+                       curr = s;
+                       s = strchr(curr, ',');
+                       if (s) {
+                               *s = '\0';
+                               s++;
+                       }
+                       o = strchr(curr, '=');
+                       if (o) {
+                               *o = '\0';
+                               o++;
+                       }
+                       rv = check_hotmod_int_op(curr, o, "rsp", &regspacing);
+                       if (rv < 0)
+                               goto out;
+                       else if (rv)
+                               continue;
+                       rv = check_hotmod_int_op(curr, o, "rsi", &regsize);
+                       if (rv < 0)
+                               goto out;
+                       else if (rv)
+                               continue;
+                       rv = check_hotmod_int_op(curr, o, "rsh", &regshift);
+                       if (rv < 0)
+                               goto out;
+                       else if (rv)
+                               continue;
+                       rv = check_hotmod_int_op(curr, o, "irq", &irq);
+                       if (rv < 0)
+                               goto out;
+                       else if (rv)
+                               continue;
+                       rv = check_hotmod_int_op(curr, o, "ipmb", &ipmb);
+                       if (rv < 0)
+                               goto out;
+                       else if (rv)
+                               continue;
+
+                       rv = -EINVAL;
+                       printk(KERN_WARNING PFX
+                              "Invalid hotmod option '%s'\n",
+                              curr);
+                       goto out;
+               }
+
+               if (op == HM_ADD) {
+                       info = smi_info_alloc();
+                       if (!info) {
+                               rv = -ENOMEM;
+                               goto out;
+                       }
+
+                       info->addr_source = SI_HOTMOD;
+                       info->si_type = si_type;
+                       info->io.addr_data = addr;
+                       info->io.addr_type = addr_space;
+                       if (addr_space == IPMI_MEM_ADDR_SPACE)
+                               info->io_setup = mem_setup;
+                       else
+                               info->io_setup = port_setup;
+
+                       info->io.addr = NULL;
+                       info->io.regspacing = regspacing;
+                       if (!info->io.regspacing)
+                               info->io.regspacing = DEFAULT_REGSPACING;
+                       info->io.regsize = regsize;
+                       if (!info->io.regsize)
+                               info->io.regsize = DEFAULT_REGSPACING;
+                       info->io.regshift = regshift;
+                       info->irq = irq;
+                       if (info->irq)
+                               info->irq_setup = std_irq_setup;
+                       info->slave_addr = ipmb;
+
+                       rv = add_smi(info);
+                       if (rv) {
+                               kfree(info);
+                               goto out;
+                       }
+                       rv = try_smi_init(info);
+                       if (rv) {
+                               cleanup_one_si(info);
+                               goto out;
+                       }
+               } else {
+                       /* remove */
+                       struct smi_info *e, *tmp_e;
+
+                       mutex_lock(&smi_infos_lock);
+                       list_for_each_entry_safe(e, tmp_e, &smi_infos, link) {
+                               if (e->io.addr_type != addr_space)
+                                       continue;
+                               if (e->si_type != si_type)
+                                       continue;
+                               if (e->io.addr_data == addr)
+                                       cleanup_one_si(e);
+                       }
+                       mutex_unlock(&smi_infos_lock);
+               }
+       }
+       rv = len;
+ out:
+       kfree(str);
+       return rv;
+}
+
+static int hardcode_find_bmc(void)
+{
+       int ret = -ENODEV;
+       int             i;
+       struct smi_info *info;
+
+       for (i = 0; i < SI_MAX_PARMS; i++) {
+               if (!ports[i] && !addrs[i])
+                       continue;
+
+               info = smi_info_alloc();
+               if (!info)
+                       return -ENOMEM;
+
+               info->addr_source = SI_HARDCODED;
+               printk(KERN_INFO PFX "probing via hardcoded address\n");
+
+               if (!si_type[i] || strcmp(si_type[i], "kcs") == 0) {
+                       info->si_type = SI_KCS;
+               } else if (strcmp(si_type[i], "smic") == 0) {
+                       info->si_type = SI_SMIC;
+               } else if (strcmp(si_type[i], "bt") == 0) {
+                       info->si_type = SI_BT;
+               } else {
+                       printk(KERN_WARNING PFX "Interface type specified "
+                              "for interface %d, was invalid: %s\n",
+                              i, si_type[i]);
+                       kfree(info);
+                       continue;
+               }
+
+               if (ports[i]) {
+                       /* An I/O port */
+                       info->io_setup = port_setup;
+                       info->io.addr_data = ports[i];
+                       info->io.addr_type = IPMI_IO_ADDR_SPACE;
+               } else if (addrs[i]) {
+                       /* A memory port */
+                       info->io_setup = mem_setup;
+                       info->io.addr_data = addrs[i];
+                       info->io.addr_type = IPMI_MEM_ADDR_SPACE;
+               } else {
+                       printk(KERN_WARNING PFX "Interface type specified "
+                              "for interface %d, but port and address were "
+                              "not set or set to zero.\n", i);
+                       kfree(info);
+                       continue;
+               }
+
+               info->io.addr = NULL;
+               info->io.regspacing = regspacings[i];
+               if (!info->io.regspacing)
+                       info->io.regspacing = DEFAULT_REGSPACING;
+               info->io.regsize = regsizes[i];
+               if (!info->io.regsize)
+                       info->io.regsize = DEFAULT_REGSPACING;
+               info->io.regshift = regshifts[i];
+               info->irq = irqs[i];
+               if (info->irq)
+                       info->irq_setup = std_irq_setup;
+               info->slave_addr = slave_addrs[i];
+
+               if (!add_smi(info)) {
+                       if (try_smi_init(info))
+                               cleanup_one_si(info);
+                       ret = 0;
+               } else {
+                       kfree(info);
+               }
+       }
+       return ret;
+}
+
+#ifdef CONFIG_ACPI
+
+#include <linux/acpi.h>
+
+/*
+ * Once we get an ACPI failure, we don't try any more, because we go
+ * through the tables sequentially.  Once we don't find a table, there
+ * are no more.
+ */
+static int acpi_failure;
+
+/* For GPE-type interrupts. */
+static u32 ipmi_acpi_gpe(acpi_handle gpe_device,
+       u32 gpe_number, void *context)
+{
+       struct smi_info *smi_info = context;
+       unsigned long   flags;
+
+       spin_lock_irqsave(&(smi_info->si_lock), flags);
+
+       smi_inc_stat(smi_info, interrupts);
+
+       debug_timestamp("ACPI_GPE");
+
+       smi_event_handler(smi_info, 0);
+       spin_unlock_irqrestore(&(smi_info->si_lock), flags);
+
+       return ACPI_INTERRUPT_HANDLED;
+}
+
+static void acpi_gpe_irq_cleanup(struct smi_info *info)
+{
+       if (!info->irq)
+               return;
+
+       acpi_remove_gpe_handler(NULL, info->irq, &ipmi_acpi_gpe);
+}
+
+static int acpi_gpe_irq_setup(struct smi_info *info)
+{
+       acpi_status status;
+
+       if (!info->irq)
+               return 0;
+
+       status = acpi_install_gpe_handler(NULL,
+                                         info->irq,
+                                         ACPI_GPE_LEVEL_TRIGGERED,
+                                         &ipmi_acpi_gpe,
+                                         info);
+       if (status != AE_OK) {
+               dev_warn(info->dev, "%s unable to claim ACPI GPE %d,"
+                        " running polled\n", DEVICE_NAME, info->irq);
+               info->irq = 0;
+               return -EINVAL;
+       } else {
+               info->irq_cleanup = acpi_gpe_irq_cleanup;
+               dev_info(info->dev, "Using ACPI GPE %d\n", info->irq);
+               return 0;
+       }
+}
+
+/*
+ * Defined at
+ * http://h21007.www2.hp.com/portal/download/files/unprot/hpspmi.pdf
+ */
+struct SPMITable {
+       s8      Signature[4];
+       u32     Length;
+       u8      Revision;
+       u8      Checksum;
+       s8      OEMID[6];
+       s8      OEMTableID[8];
+       s8      OEMRevision[4];
+       s8      CreatorID[4];
+       s8      CreatorRevision[4];
+       u8      InterfaceType;
+       u8      IPMIlegacy;
+       s16     SpecificationRevision;
+
+       /*
+        * Bit 0 - SCI interrupt supported
+        * Bit 1 - I/O APIC/SAPIC
+        */
+       u8      InterruptType;
+
+       /*
+        * If bit 0 of InterruptType is set, then this is the SCI
+        * interrupt in the GPEx_STS register.
+        */
+       u8      GPE;
+
+       s16     Reserved;
+
+       /*
+        * If bit 1 of InterruptType is set, then this is the I/O
+        * APIC/SAPIC interrupt.
+        */
+       u32     GlobalSystemInterrupt;
+
+       /* The actual register address. */
+       struct acpi_generic_address addr;
+
+       u8      UID[4];
+
+       s8      spmi_id[1]; /* A '\0' terminated array starts here. */
+};
+
+static int try_init_spmi(struct SPMITable *spmi)
+{
+       struct smi_info  *info;
+       int rv;
+
+       if (spmi->IPMIlegacy != 1) {
+               printk(KERN_INFO PFX "Bad SPMI legacy %d\n", spmi->IPMIlegacy);
+               return -ENODEV;
+       }
+
+       info = smi_info_alloc();
+       if (!info) {
+               printk(KERN_ERR PFX "Could not allocate SI data (3)\n");
+               return -ENOMEM;
+       }
+
+       info->addr_source = SI_SPMI;
+       printk(KERN_INFO PFX "probing via SPMI\n");
+
+       /* Figure out the interface type. */
+       switch (spmi->InterfaceType) {
+       case 1: /* KCS */
+               info->si_type = SI_KCS;
+               break;
+       case 2: /* SMIC */
+               info->si_type = SI_SMIC;
+               break;
+       case 3: /* BT */
+               info->si_type = SI_BT;
+               break;
+       case 4: /* SSIF, just ignore */
+               kfree(info);
+               return -EIO;
+       default:
+               printk(KERN_INFO PFX "Unknown ACPI/SPMI SI type %d\n",
+                      spmi->InterfaceType);
+               kfree(info);
+               return -EIO;
+       }
+
+       if (spmi->InterruptType & 1) {
+               /* We've got a GPE interrupt. */
+               info->irq = spmi->GPE;
+               info->irq_setup = acpi_gpe_irq_setup;
+       } else if (spmi->InterruptType & 2) {
+               /* We've got an APIC/SAPIC interrupt. */
+               info->irq = spmi->GlobalSystemInterrupt;
+               info->irq_setup = std_irq_setup;
+       } else {
+               /* Use the default interrupt setting. */
+               info->irq = 0;
+               info->irq_setup = NULL;
+       }
+
+       if (spmi->addr.bit_width) {
+               /* A (hopefully) properly formed register bit width. */
+               info->io.regspacing = spmi->addr.bit_width / 8;
+       } else {
+               info->io.regspacing = DEFAULT_REGSPACING;
+       }
+       info->io.regsize = info->io.regspacing;
+       info->io.regshift = spmi->addr.bit_offset;
+
+       if (spmi->addr.space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
+               info->io_setup = mem_setup;
+               info->io.addr_type = IPMI_MEM_ADDR_SPACE;
+       } else if (spmi->addr.space_id == ACPI_ADR_SPACE_SYSTEM_IO) {
+               info->io_setup = port_setup;
+               info->io.addr_type = IPMI_IO_ADDR_SPACE;
+       } else {
+               kfree(info);
+               printk(KERN_WARNING PFX "Unknown ACPI I/O Address type\n");
+               return -EIO;
+       }
+       info->io.addr_data = spmi->addr.address;
+
+       pr_info("ipmi_si: SPMI: %s %#lx regsize %d spacing %d irq %d\n",
+                (info->io.addr_type == IPMI_IO_ADDR_SPACE) ? "io" : "mem",
+                info->io.addr_data, info->io.regsize, info->io.regspacing,
+                info->irq);
+
+       rv = add_smi(info);
+       if (rv)
+               kfree(info);
+
+       return rv;
+}
+
+static void spmi_find_bmc(void)
+{
+       acpi_status      status;
+       struct SPMITable *spmi;
+       int              i;
+
+       if (acpi_disabled)
+               return;
+
+       if (acpi_failure)
+               return;
+
+       for (i = 0; ; i++) {
+               status = acpi_get_table(ACPI_SIG_SPMI, i+1,
+                                       (struct acpi_table_header **)&spmi);
+               if (status != AE_OK)
+                       return;
+
+               try_init_spmi(spmi);
+       }
+}
+
+static int ipmi_pnp_probe(struct pnp_dev *dev,
+                                   const struct pnp_device_id *dev_id)
+{
+       struct acpi_device *acpi_dev;
+       struct smi_info *info;
+       struct resource *res, *res_second;
+       acpi_handle handle;
+       acpi_status status;
+       unsigned long long tmp;
+       int rv = -EINVAL;
+
+       acpi_dev = pnp_acpi_device(dev);
+       if (!acpi_dev)
+               return -ENODEV;
+
+       info = smi_info_alloc();
+       if (!info)
+               return -ENOMEM;
+
+       info->addr_source = SI_ACPI;
+       printk(KERN_INFO PFX "probing via ACPI\n");
+
+       handle = acpi_dev->handle;
+       info->addr_info.acpi_info.acpi_handle = handle;
+
+       /* _IFT tells us the interface type: KCS, BT, etc */
+       status = acpi_evaluate_integer(handle, "_IFT", NULL, &tmp);
+       if (ACPI_FAILURE(status)) {
+               dev_err(&dev->dev, "Could not find ACPI IPMI interface type\n");
+               goto err_free;
+       }
+
+       switch (tmp) {
+       case 1:
+               info->si_type = SI_KCS;
+               break;
+       case 2:
+               info->si_type = SI_SMIC;
+               break;
+       case 3:
+               info->si_type = SI_BT;
+               break;
+       case 4: /* SSIF, just ignore */
+               rv = -ENODEV;
+               goto err_free;
+       default:
+               dev_info(&dev->dev, "unknown IPMI type %lld\n", tmp);
+               goto err_free;
+       }
+
+       res = pnp_get_resource(dev, IORESOURCE_IO, 0);
+       if (res) {
+               info->io_setup = port_setup;
+               info->io.addr_type = IPMI_IO_ADDR_SPACE;
+       } else {
+               res = pnp_get_resource(dev, IORESOURCE_MEM, 0);
+               if (res) {
+                       info->io_setup = mem_setup;
+                       info->io.addr_type = IPMI_MEM_ADDR_SPACE;
+               }
+       }
+       if (!res) {
+               dev_err(&dev->dev, "no I/O or memory address\n");
+               goto err_free;
+       }
+       info->io.addr_data = res->start;
+
+       info->io.regspacing = DEFAULT_REGSPACING;
+       res_second = pnp_get_resource(dev,
+                              (info->io.addr_type == IPMI_IO_ADDR_SPACE) ?
+                                       IORESOURCE_IO : IORESOURCE_MEM,
+                              1);
+       if (res_second) {
+               if (res_second->start > info->io.addr_data)
+                       info->io.regspacing = res_second->start - info->io.addr_data;
+       }
+       info->io.regsize = DEFAULT_REGSPACING;
+       info->io.regshift = 0;
+
+       /* If _GPE exists, use it; otherwise use standard interrupts */
+       status = acpi_evaluate_integer(handle, "_GPE", NULL, &tmp);
+       if (ACPI_SUCCESS(status)) {
+               info->irq = tmp;
+               info->irq_setup = acpi_gpe_irq_setup;
+       } else if (pnp_irq_valid(dev, 0)) {
+               info->irq = pnp_irq(dev, 0);
+               info->irq_setup = std_irq_setup;
+       }
+
+       info->dev = &dev->dev;
+       pnp_set_drvdata(dev, info);
+
+       dev_info(info->dev, "%pR regsize %d spacing %d irq %d\n",
+                res, info->io.regsize, info->io.regspacing,
+                info->irq);
+
+       rv = add_smi(info);
+       if (rv)
+               kfree(info);
+
+       return rv;
+
+err_free:
+       kfree(info);
+       return rv;
+}
+
+static void ipmi_pnp_remove(struct pnp_dev *dev)
+{
+       struct smi_info *info = pnp_get_drvdata(dev);
+
+       cleanup_one_si(info);
+}
+
+static const struct pnp_device_id pnp_dev_table[] = {
+       {"IPI0001", 0},
+       {"", 0},
+};
+
+static struct pnp_driver ipmi_pnp_driver = {
+       .name           = DEVICE_NAME,
+       .probe          = ipmi_pnp_probe,
+       .remove         = ipmi_pnp_remove,
+       .id_table       = pnp_dev_table,
+};
+
+MODULE_DEVICE_TABLE(pnp, pnp_dev_table);
+#endif
+
+#ifdef CONFIG_DMI
+struct dmi_ipmi_data {
+       u8              type;
+       u8              addr_space;
+       unsigned long   base_addr;
+       u8              irq;
+       u8              offset;
+       u8              slave_addr;
+};
+
+static int decode_dmi(const struct dmi_header *dm,
+                               struct dmi_ipmi_data *dmi)
+{
+       const u8        *data = (const u8 *)dm;
+       unsigned long   base_addr;
+       u8              reg_spacing;
+       u8              len = dm->length;
+
+       dmi->type = data[4];
+
+       memcpy(&base_addr, data+8, sizeof(unsigned long));
+       if (len >= 0x11) {
+               if (base_addr & 1) {
+                       /* I/O */
+                       base_addr &= 0xFFFE;
+                       dmi->addr_space = IPMI_IO_ADDR_SPACE;
+               } else
+                       /* Memory */
+                       dmi->addr_space = IPMI_MEM_ADDR_SPACE;
+
+               /* If bit 4 of byte 0x10 is set, then the lsb for the address
+                  is odd. */
+               dmi->base_addr = base_addr | ((data[0x10] & 0x10) >> 4);
+
+               dmi->irq = data[0x11];
+
+               /* The top two bits of byte 0x10 hold the register spacing. */
+               reg_spacing = (data[0x10] & 0xC0) >> 6;
+               switch (reg_spacing) {
+               case 0x00: /* Byte boundaries */
+                   dmi->offset = 1;
+                   break;
+               case 0x01: /* 32-bit boundaries */
+                   dmi->offset = 4;
+                   break;
+               case 0x02: /* 16-byte boundaries */
+                   dmi->offset = 16;
+                   break;
+               default:
+                   /* Some other interface, just ignore it. */
+                   return -EIO;
+               }
+       } else {
+               /* Old DMI spec. */
+               /*
+                * Note that technically, the lower bit of the base
+                * address should be 1 if the address is I/O and 0 if
+                * the address is in memory.  So many systems get that
+                * wrong (and all that I have seen are I/O) so we just
+                * ignore that bit and assume I/O.  Systems that use
+                * memory should use the newer spec, anyway.
+                */
+               dmi->base_addr = base_addr & 0xfffe;
+               dmi->addr_space = IPMI_IO_ADDR_SPACE;
+               dmi->offset = 1;
+       }
+
+       dmi->slave_addr = data[6];
+
+       return 0;
+}
+
+static void try_init_dmi(struct dmi_ipmi_data *ipmi_data)
+{
+       struct smi_info *info;
+
+       info = smi_info_alloc();
+       if (!info) {
+               printk(KERN_ERR PFX "Could not allocate SI data\n");
+               return;
+       }
+
+       info->addr_source = SI_SMBIOS;
+       printk(KERN_INFO PFX "probing via SMBIOS\n");
+
+       switch (ipmi_data->type) {
+       case 0x01: /* KCS */
+               info->si_type = SI_KCS;
+               break;
+       case 0x02: /* SMIC */
+               info->si_type = SI_SMIC;
+               break;
+       case 0x03: /* BT */
+               info->si_type = SI_BT;
+               break;
+       default:
+               kfree(info);
+               return;
+       }
+
+       switch (ipmi_data->addr_space) {
+       case IPMI_MEM_ADDR_SPACE:
+               info->io_setup = mem_setup;
+               info->io.addr_type = IPMI_MEM_ADDR_SPACE;
+               break;
+
+       case IPMI_IO_ADDR_SPACE:
+               info->io_setup = port_setup;
+               info->io.addr_type = IPMI_IO_ADDR_SPACE;
+               break;
+
+       default:
+               kfree(info);
+               printk(KERN_WARNING PFX "Unknown SMBIOS I/O Address type: %d\n",
+                      ipmi_data->addr_space);
+               return;
+       }
+       info->io.addr_data = ipmi_data->base_addr;
+
+       info->io.regspacing = ipmi_data->offset;
+       if (!info->io.regspacing)
+               info->io.regspacing = DEFAULT_REGSPACING;
+       info->io.regsize = DEFAULT_REGSPACING;
+       info->io.regshift = 0;
+
+       info->slave_addr = ipmi_data->slave_addr;
+
+       info->irq = ipmi_data->irq;
+       if (info->irq)
+               info->irq_setup = std_irq_setup;
+
+       pr_info("ipmi_si: SMBIOS: %s %#lx regsize %d spacing %d irq %d\n",
+                (info->io.addr_type == IPMI_IO_ADDR_SPACE) ? "io" : "mem",
+                info->io.addr_data, info->io.regsize, info->io.regspacing,
+                info->irq);
+
+       if (add_smi(info))
+               kfree(info);
+}
+
+static void dmi_find_bmc(void)
+{
+       const struct dmi_device *dev = NULL;
+       struct dmi_ipmi_data data;
+       int                  rv;
+
+       while ((dev = dmi_find_device(DMI_DEV_TYPE_IPMI, NULL, dev))) {
+               memset(&data, 0, sizeof(data));
+               rv = decode_dmi((const struct dmi_header *) dev->device_data,
+                               &data);
+               if (!rv)
+                       try_init_dmi(&data);
+       }
+}
+#endif /* CONFIG_DMI */
+
+#ifdef CONFIG_PCI
+
+#define PCI_ERMC_CLASSCODE             0x0C0700
+#define PCI_ERMC_CLASSCODE_MASK                0xffffff00
+#define PCI_ERMC_CLASSCODE_TYPE_MASK   0xff
+#define PCI_ERMC_CLASSCODE_TYPE_SMIC   0x00
+#define PCI_ERMC_CLASSCODE_TYPE_KCS    0x01
+#define PCI_ERMC_CLASSCODE_TYPE_BT     0x02
+
+#define PCI_HP_VENDOR_ID    0x103C
+#define PCI_MMC_DEVICE_ID   0x121A
+#define PCI_MMC_ADDR_CW     0x10
+
+static void ipmi_pci_cleanup(struct smi_info *info)
+{
+       struct pci_dev *pdev = info->addr_source_data;
+
+       pci_disable_device(pdev);
+}
+
+static int ipmi_pci_probe_regspacing(struct smi_info *info)
+{
+       if (info->si_type == SI_KCS) {
+               unsigned char   status;
+               int             regspacing;
+
+               info->io.regsize = DEFAULT_REGSIZE;
+               info->io.regshift = 0;
+               info->io_size = 2;
+               info->handlers = &kcs_smi_handlers;
+
+               /* detect 1, 4, 16byte spacing */
+               for (regspacing = DEFAULT_REGSPACING; regspacing <= 16;) {
+                       info->io.regspacing = regspacing;
+                       if (info->io_setup(info)) {
+                               dev_err(info->dev,
+                                       "Could not setup I/O space\n");
+                               return DEFAULT_REGSPACING;
+                       }
+                       /* write invalid cmd */
+                       info->io.outputb(&info->io, 1, 0x10);
+                       /* read status back */
+                       status = info->io.inputb(&info->io, 1);
+                       info->io_cleanup(info);
+                       if (status)
+                               return regspacing;
+                       regspacing *= 4;
+               }
+       }
+       return DEFAULT_REGSPACING;
+}
+
+static int ipmi_pci_probe(struct pci_dev *pdev,
+                                   const struct pci_device_id *ent)
+{
+       int rv;
+       int class_type = pdev->class & PCI_ERMC_CLASSCODE_TYPE_MASK;
+       struct smi_info *info;
+
+       info = smi_info_alloc();
+       if (!info)
+               return -ENOMEM;
+
+       info->addr_source = SI_PCI;
+       dev_info(&pdev->dev, "probing via PCI");
+
+       switch (class_type) {
+       case PCI_ERMC_CLASSCODE_TYPE_SMIC:
+               info->si_type = SI_SMIC;
+               break;
+
+       case PCI_ERMC_CLASSCODE_TYPE_KCS:
+               info->si_type = SI_KCS;
+               break;
+
+       case PCI_ERMC_CLASSCODE_TYPE_BT:
+               info->si_type = SI_BT;
+               break;
+
+       default:
+               kfree(info);
+               dev_info(&pdev->dev, "Unknown IPMI type: %d\n", class_type);
+               return -ENOMEM;
+       }
+
+       rv = pci_enable_device(pdev);
+       if (rv) {
+               dev_err(&pdev->dev, "couldn't enable PCI device\n");
+               kfree(info);
+               return rv;
+       }
+
+       info->addr_source_cleanup = ipmi_pci_cleanup;
+       info->addr_source_data = pdev;
+
+       if (pci_resource_flags(pdev, 0) & IORESOURCE_IO) {
+               info->io_setup = port_setup;
+               info->io.addr_type = IPMI_IO_ADDR_SPACE;
+       } else {
+               info->io_setup = mem_setup;
+               info->io.addr_type = IPMI_MEM_ADDR_SPACE;
+       }
+       info->io.addr_data = pci_resource_start(pdev, 0);
+
+       info->io.regspacing = ipmi_pci_probe_regspacing(info);
+       info->io.regsize = DEFAULT_REGSIZE;
+       info->io.regshift = 0;
+
+       info->irq = pdev->irq;
+       if (info->irq)
+               info->irq_setup = std_irq_setup;
+
+       info->dev = &pdev->dev;
+       pci_set_drvdata(pdev, info);
+
+       dev_info(&pdev->dev, "%pR regsize %d spacing %d irq %d\n",
+               &pdev->resource[0], info->io.regsize, info->io.regspacing,
+               info->irq);
+
+       rv = add_smi(info);
+       if (rv) {
+               kfree(info);
+               pci_disable_device(pdev);
+       }
+
+       return rv;
+}
+
+static void ipmi_pci_remove(struct pci_dev *pdev)
+{
+       struct smi_info *info = pci_get_drvdata(pdev);
+       cleanup_one_si(info);
+       pci_disable_device(pdev);
+}
+
+static struct pci_device_id ipmi_pci_devices[] = {
+       { PCI_DEVICE(PCI_HP_VENDOR_ID, PCI_MMC_DEVICE_ID) },
+       { PCI_DEVICE_CLASS(PCI_ERMC_CLASSCODE, PCI_ERMC_CLASSCODE_MASK) },
+       { 0, }
+};
+MODULE_DEVICE_TABLE(pci, ipmi_pci_devices);
+
+static struct pci_driver ipmi_pci_driver = {
+       .name =         DEVICE_NAME,
+       .id_table =     ipmi_pci_devices,
+       .probe =        ipmi_pci_probe,
+       .remove =       ipmi_pci_remove,
+};
+#endif /* CONFIG_PCI */
+
+static const struct of_device_id ipmi_match[];
+static int ipmi_probe(struct platform_device *dev)
+{
+#ifdef CONFIG_OF
+       const struct of_device_id *match;
+       struct smi_info *info;
+       struct resource resource;
+       const __be32 *regsize, *regspacing, *regshift;
+       struct device_node *np = dev->dev.of_node;
+       int ret;
+       int proplen;
+
+       dev_info(&dev->dev, "probing via device tree\n");
+
+       match = of_match_device(ipmi_match, &dev->dev);
+       if (!match)
+               return -EINVAL;
+
+       if (!of_device_is_available(np))
+               return -EINVAL;
+
+       ret = of_address_to_resource(np, 0, &resource);
+       if (ret) {
+               dev_warn(&dev->dev, PFX "invalid address from OF\n");
+               return ret;
+       }
+
+       regsize = of_get_property(np, "reg-size", &proplen);
+       if (regsize && proplen != 4) {
+               dev_warn(&dev->dev, PFX "invalid regsize from OF\n");
+               return -EINVAL;
+       }
+
+       regspacing = of_get_property(np, "reg-spacing", &proplen);
+       if (regspacing && proplen != 4) {
+               dev_warn(&dev->dev, PFX "invalid regspacing from OF\n");
+               return -EINVAL;
+       }
+
+       regshift = of_get_property(np, "reg-shift", &proplen);
+       if (regshift && proplen != 4) {
+               dev_warn(&dev->dev, PFX "invalid regshift from OF\n");
+               return -EINVAL;
+       }
+
+       info = smi_info_alloc();
+
+       if (!info) {
+               dev_err(&dev->dev,
+                       "could not allocate memory for OF probe\n");
+               return -ENOMEM;
+       }
+
+       info->si_type           = (enum si_type) match->data;
+       info->addr_source       = SI_DEVICETREE;
+       info->irq_setup         = std_irq_setup;
+
+       if (resource.flags & IORESOURCE_IO) {
+               info->io_setup          = port_setup;
+               info->io.addr_type      = IPMI_IO_ADDR_SPACE;
+       } else {
+               info->io_setup          = mem_setup;
+               info->io.addr_type      = IPMI_MEM_ADDR_SPACE;
+       }
+
+       info->io.addr_data      = resource.start;
+
+       info->io.regsize        = regsize ? be32_to_cpup(regsize) : DEFAULT_REGSIZE;
+       info->io.regspacing     = regspacing ? be32_to_cpup(regspacing) : DEFAULT_REGSPACING;
+       info->io.regshift       = regshift ? be32_to_cpup(regshift) : 0;
+
+       info->irq               = irq_of_parse_and_map(dev->dev.of_node, 0);
+       info->dev               = &dev->dev;
+
+       dev_dbg(&dev->dev, "addr 0x%lx regsize %d spacing %d irq %d\n",
+               info->io.addr_data, info->io.regsize, info->io.regspacing,
+               info->irq);
+
+       dev_set_drvdata(&dev->dev, info);
+
+       ret = add_smi(info);
+       if (ret) {
+               kfree(info);
+               return ret;
+       }
+#endif
+       return 0;
+}
+
+static int ipmi_remove(struct platform_device *dev)
+{
+#ifdef CONFIG_OF
+       cleanup_one_si(dev_get_drvdata(&dev->dev));
+#endif
+       return 0;
+}
+
+static const struct of_device_id ipmi_match[] =
+{
+       { .type = "ipmi", .compatible = "ipmi-kcs",
+         .data = (void *)(unsigned long) SI_KCS },
+       { .type = "ipmi", .compatible = "ipmi-smic",
+         .data = (void *)(unsigned long) SI_SMIC },
+       { .type = "ipmi", .compatible = "ipmi-bt",
+         .data = (void *)(unsigned long) SI_BT },
+       {},
+};
+
+static struct platform_driver ipmi_driver = {
+       .driver = {
+               .name = DEVICE_NAME,
+               .of_match_table = ipmi_match,
+       },
+       .probe          = ipmi_probe,
+       .remove         = ipmi_remove,
+};
+
+#ifdef CONFIG_PARISC
+static int ipmi_parisc_probe(struct parisc_device *dev)
+{
+       struct smi_info *info;
+       int rv;
+
+       info = smi_info_alloc();
+
+       if (!info) {
+               dev_err(&dev->dev,
+                       "could not allocate memory for PARISC probe\n");
+               return -ENOMEM;
+       }
+
+       info->si_type           = SI_KCS;
+       info->addr_source       = SI_DEVICETREE;
+       info->io_setup          = mem_setup;
+       info->io.addr_type      = IPMI_MEM_ADDR_SPACE;
+       info->io.addr_data      = dev->hpa.start;
+       info->io.regsize        = 1;
+       info->io.regspacing     = 1;
+       info->io.regshift       = 0;
+       info->irq               = 0; /* no interrupt */
+       info->irq_setup         = NULL;
+       info->dev               = &dev->dev;
+
+       dev_dbg(&dev->dev, "addr 0x%lx\n", info->io.addr_data);
+
+       dev_set_drvdata(&dev->dev, info);
+
+       rv = add_smi(info);
+       if (rv) {
+               kfree(info);
+               return rv;
+       }
+
+       return 0;
+}
+
+static int ipmi_parisc_remove(struct parisc_device *dev)
+{
+       cleanup_one_si(dev_get_drvdata(&dev->dev));
+       return 0;
+}
+
+static struct parisc_device_id ipmi_parisc_tbl[] = {
+       { HPHW_MC, HVERSION_REV_ANY_ID, 0x004, 0xC0 },
+       { 0, }
+};
+
+static struct parisc_driver ipmi_parisc_driver = {
+       .name =         "ipmi",
+       .id_table =     ipmi_parisc_tbl,
+       .probe =        ipmi_parisc_probe,
+       .remove =       ipmi_parisc_remove,
+};
+#endif /* CONFIG_PARISC */
+
+static int wait_for_msg_done(struct smi_info *smi_info)
+{
+       enum si_sm_result     smi_result;
+
+       smi_result = smi_info->handlers->event(smi_info->si_sm, 0);
+       for (;;) {
+               if (smi_result == SI_SM_CALL_WITH_DELAY ||
+                   smi_result == SI_SM_CALL_WITH_TICK_DELAY) {
+                       schedule_timeout_uninterruptible(1);
+                       smi_result = smi_info->handlers->event(
+                               smi_info->si_sm, jiffies_to_usecs(1));
+               } else if (smi_result == SI_SM_CALL_WITHOUT_DELAY) {
+                       smi_result = smi_info->handlers->event(
+                               smi_info->si_sm, 0);
+               } else
+                       break;
+       }
+       if (smi_result == SI_SM_HOSED)
+               /*
+                * We couldn't get the state machine to run, so whatever's at
+                * the port is probably not an IPMI SMI interface.
+                */
+               return -ENODEV;
+
+       return 0;
+}
+
+static int try_get_dev_id(struct smi_info *smi_info)
+{
+       unsigned char         msg[2];
+       unsigned char         *resp;
+       unsigned long         resp_len;
+       int                   rv = 0;
+
+       resp = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
+       if (!resp)
+               return -ENOMEM;
+
+       /*
+        * Do a Get Device ID command, since it comes back with some
+        * useful info.
+        */
+       msg[0] = IPMI_NETFN_APP_REQUEST << 2;
+       msg[1] = IPMI_GET_DEVICE_ID_CMD;
+       smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
+
+       rv = wait_for_msg_done(smi_info);
+       if (rv)
+               goto out;
+
+       resp_len = smi_info->handlers->get_result(smi_info->si_sm,
+                                                 resp, IPMI_MAX_MSG_LENGTH);
+
+       /* Check and record info from the get device id, in case we need it. */
+       rv = ipmi_demangle_device_id(resp, resp_len, &smi_info->device_id);
+
+ out:
+       kfree(resp);
+       return rv;
+}
+
+/*
+ * Some BMCs do not support clearing the receive irq bit in the global
+ * enables (even if they don't support interrupts on the BMC).  Check
+ * for this and handle it properly.
+ */
+static void check_clr_rcv_irq(struct smi_info *smi_info)
+{
+       unsigned char         msg[3];
+       unsigned char         *resp;
+       unsigned long         resp_len;
+       int                   rv;
+
+       resp = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
+       if (!resp) {
+               printk(KERN_WARNING PFX "Out of memory allocating response for"
+                      " global enables command, cannot check recv irq bit"
+                      " handling.\n");
+               return;
+       }
+
+       msg[0] = IPMI_NETFN_APP_REQUEST << 2;
+       msg[1] = IPMI_GET_BMC_GLOBAL_ENABLES_CMD;
+       smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
+
+       rv = wait_for_msg_done(smi_info);
+       if (rv) {
+               printk(KERN_WARNING PFX "Error getting response from get"
+                      " global enables command, cannot check recv irq bit"
+                      " handling.\n");
+               goto out;
+       }
+
+       resp_len = smi_info->handlers->get_result(smi_info->si_sm,
+                                                 resp, IPMI_MAX_MSG_LENGTH);
+
+       if (resp_len < 4 ||
+                       resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
+                       resp[1] != IPMI_GET_BMC_GLOBAL_ENABLES_CMD   ||
+                       resp[2] != 0) {
+               printk(KERN_WARNING PFX "Invalid return from get global"
+                      " enables command, cannot check recv irq bit"
+                      " handling.\n");
+               rv = -EINVAL;
+               goto out;
+       }
+
+       if ((resp[3] & IPMI_BMC_RCV_MSG_INTR) == 0)
+               /* Already clear, should work ok. */
+               goto out;
+
+       msg[0] = IPMI_NETFN_APP_REQUEST << 2;
+       msg[1] = IPMI_SET_BMC_GLOBAL_ENABLES_CMD;
+       msg[2] = resp[3] & ~IPMI_BMC_RCV_MSG_INTR;
+       smi_info->handlers->start_transaction(smi_info->si_sm, msg, 3);
+
+       rv = wait_for_msg_done(smi_info);
+       if (rv) {
+               printk(KERN_WARNING PFX "Error getting response from set"
+                      " global enables command, cannot check recv irq bit"
+                      " handling.\n");
+               goto out;
+       }
+
+       resp_len = smi_info->handlers->get_result(smi_info->si_sm,
+                                                 resp, IPMI_MAX_MSG_LENGTH);
+
+       if (resp_len < 3 ||
+                       resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
+                       resp[1] != IPMI_SET_BMC_GLOBAL_ENABLES_CMD) {
+               printk(KERN_WARNING PFX "Invalid return from get global"
+                      " enables command, cannot check recv irq bit"
+                      " handling.\n");
+               rv = -EINVAL;
+               goto out;
+       }
+
+       if (resp[2] != 0) {
+               /*
+                * An error when setting the event buffer bit means
+                * clearing the bit is not supported.
+                */
+               printk(KERN_WARNING PFX "The BMC does not support clearing"
+                      " the recv irq bit, compensating, but the BMC needs to"
+                      " be fixed.\n");
+               smi_info->cannot_clear_recv_irq_bit = true;
+       }
+ out:
+       kfree(resp);
+}
+
+static int try_enable_event_buffer(struct smi_info *smi_info)
+{
+       unsigned char         msg[3];
+       unsigned char         *resp;
+       unsigned long         resp_len;
+       int                   rv = 0;
+
+       resp = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
+       if (!resp)
+               return -ENOMEM;
+
+       msg[0] = IPMI_NETFN_APP_REQUEST << 2;
+       msg[1] = IPMI_GET_BMC_GLOBAL_ENABLES_CMD;
+       smi_info->handlers->start_transaction(smi_info->si_sm, msg, 2);
+
+       rv = wait_for_msg_done(smi_info);
+       if (rv) {
+               printk(KERN_WARNING PFX "Error getting response from get"
+                      " global enables command, the event buffer is not"
+                      " enabled.\n");
+               goto out;
+       }
+
+       resp_len = smi_info->handlers->get_result(smi_info->si_sm,
+                                                 resp, IPMI_MAX_MSG_LENGTH);
+
+       if (resp_len < 4 ||
+                       resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
+                       resp[1] != IPMI_GET_BMC_GLOBAL_ENABLES_CMD   ||
+                       resp[2] != 0) {
+               printk(KERN_WARNING PFX "Invalid return from get global"
+                      " enables command, cannot enable the event buffer.\n");
+               rv = -EINVAL;
+               goto out;
+       }
+
+       if (resp[3] & IPMI_BMC_EVT_MSG_BUFF) {
+               /* buffer is already enabled, nothing to do. */
+               smi_info->supports_event_msg_buff = true;
+               goto out;
+       }
+
+       msg[0] = IPMI_NETFN_APP_REQUEST << 2;
+       msg[1] = IPMI_SET_BMC_GLOBAL_ENABLES_CMD;
+       msg[2] = resp[3] | IPMI_BMC_EVT_MSG_BUFF;
+       smi_info->handlers->start_transaction(smi_info->si_sm, msg, 3);
+
+       rv = wait_for_msg_done(smi_info);
+       if (rv) {
+               printk(KERN_WARNING PFX "Error getting response from set"
+                      " global, enables command, the event buffer is not"
+                      " enabled.\n");
+               goto out;
+       }
+
+       resp_len = smi_info->handlers->get_result(smi_info->si_sm,
+                                                 resp, IPMI_MAX_MSG_LENGTH);
+
+       if (resp_len < 3 ||
+                       resp[0] != (IPMI_NETFN_APP_REQUEST | 1) << 2 ||
+                       resp[1] != IPMI_SET_BMC_GLOBAL_ENABLES_CMD) {
+               printk(KERN_WARNING PFX "Invalid return from get global,"
+                      "enables command, not enable the event buffer.\n");
+               rv = -EINVAL;
+               goto out;
+       }
+
+       if (resp[2] != 0)
+               /*
+                * An error when setting the event buffer bit means
+                * that the event buffer is not supported.
+                */
+               rv = -ENOENT;
+       else
+               smi_info->supports_event_msg_buff = true;
+
+ out:
+       kfree(resp);
+       return rv;
+}
+
+static int smi_type_proc_show(struct seq_file *m, void *v)
+{
+       struct smi_info *smi = m->private;
+
+       seq_printf(m, "%s\n", si_to_str[smi->si_type]);
+
+       return 0;
+}
+
+static int smi_type_proc_open(struct inode *inode, struct file *file)
+{
+       return single_open(file, smi_type_proc_show, PDE_DATA(inode));
+}
+
+static const struct file_operations smi_type_proc_ops = {
+       .open           = smi_type_proc_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = single_release,
+};
+
+static int smi_si_stats_proc_show(struct seq_file *m, void *v)
+{
+       struct smi_info *smi = m->private;
+
+       seq_printf(m, "interrupts_enabled:    %d\n",
+                      smi->irq && !smi->interrupt_disabled);
+       seq_printf(m, "short_timeouts:        %u\n",
+                      smi_get_stat(smi, short_timeouts));
+       seq_printf(m, "long_timeouts:         %u\n",
+                      smi_get_stat(smi, long_timeouts));
+       seq_printf(m, "idles:                 %u\n",
+                      smi_get_stat(smi, idles));
+       seq_printf(m, "interrupts:            %u\n",
+                      smi_get_stat(smi, interrupts));
+       seq_printf(m, "attentions:            %u\n",
+                      smi_get_stat(smi, attentions));
+       seq_printf(m, "flag_fetches:          %u\n",
+                      smi_get_stat(smi, flag_fetches));
+       seq_printf(m, "hosed_count:           %u\n",
+                      smi_get_stat(smi, hosed_count));
+       seq_printf(m, "complete_transactions: %u\n",
+                      smi_get_stat(smi, complete_transactions));
+       seq_printf(m, "events:                %u\n",
+                      smi_get_stat(smi, events));
+       seq_printf(m, "watchdog_pretimeouts:  %u\n",
+                      smi_get_stat(smi, watchdog_pretimeouts));
+       seq_printf(m, "incoming_messages:     %u\n",
+                      smi_get_stat(smi, incoming_messages));
+       return 0;
+}
+
+static int smi_si_stats_proc_open(struct inode *inode, struct file *file)
+{
+       return single_open(file, smi_si_stats_proc_show, PDE_DATA(inode));
+}
+
+static const struct file_operations smi_si_stats_proc_ops = {
+       .open           = smi_si_stats_proc_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = single_release,
+};
+
+static int smi_params_proc_show(struct seq_file *m, void *v)
+{
+       struct smi_info *smi = m->private;
+
+       seq_printf(m,
+                  "%s,%s,0x%lx,rsp=%d,rsi=%d,rsh=%d,irq=%d,ipmb=%d\n",
+                  si_to_str[smi->si_type],
+                  addr_space_to_str[smi->io.addr_type],
+                  smi->io.addr_data,
+                  smi->io.regspacing,
+                  smi->io.regsize,
+                  smi->io.regshift,
+                  smi->irq,
+                  smi->slave_addr);
+
+       return 0;
+}
+
+static int smi_params_proc_open(struct inode *inode, struct file *file)
+{
+       return single_open(file, smi_params_proc_show, PDE_DATA(inode));
+}
+
+static const struct file_operations smi_params_proc_ops = {
+       .open           = smi_params_proc_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = single_release,
+};
+
+/*
+ * oem_data_avail_to_receive_msg_avail
+ * @info - smi_info structure with msg_flags set
+ *
+ * Converts flags from OEM_DATA_AVAIL to RECEIVE_MSG_AVAIL
+ * Returns 1 indicating need to re-run handle_flags().
+ */
+static int oem_data_avail_to_receive_msg_avail(struct smi_info *smi_info)
+{
+       smi_info->msg_flags = ((smi_info->msg_flags & ~OEM_DATA_AVAIL) |
+                              RECEIVE_MSG_AVAIL);
+       return 1;
+}
+
+/*
+ * setup_dell_poweredge_oem_data_handler
+ * @info - smi_info.device_id must be populated
+ *
+ * Systems that match, but have firmware version < 1.40 may assert
+ * OEM0_DATA_AVAIL on their own, without being told via Set Flags that
+ * it's safe to do so.  Such systems will de-assert OEM1_DATA_AVAIL
+ * upon receipt of IPMI_GET_MSG_CMD, so we should treat these flags
+ * as RECEIVE_MSG_AVAIL instead.
+ *
+ * As Dell has no plans to release IPMI 1.5 firmware that *ever*
+ * assert the OEM[012] bits, and if it did, the driver would have to
+ * change to handle that properly, we don't actually check for the
+ * firmware version.
+ * Device ID = 0x20                BMC on PowerEdge 8G servers
+ * Device Revision = 0x80
+ * Firmware Revision1 = 0x01       BMC version 1.40
+ * Firmware Revision2 = 0x40       BCD encoded
+ * IPMI Version = 0x51             IPMI 1.5
+ * Manufacturer ID = A2 02 00      Dell IANA
+ *
+ * Additionally, PowerEdge systems with IPMI < 1.5 may also assert
+ * OEM0_DATA_AVAIL and needs to be treated as RECEIVE_MSG_AVAIL.
+ *
+ */
+#define DELL_POWEREDGE_8G_BMC_DEVICE_ID  0x20
+#define DELL_POWEREDGE_8G_BMC_DEVICE_REV 0x80
+#define DELL_POWEREDGE_8G_BMC_IPMI_VERSION 0x51
+#define DELL_IANA_MFR_ID 0x0002a2
+static void setup_dell_poweredge_oem_data_handler(struct smi_info *smi_info)
+{
+       struct ipmi_device_id *id = &smi_info->device_id;
+       if (id->manufacturer_id == DELL_IANA_MFR_ID) {
+               if (id->device_id       == DELL_POWEREDGE_8G_BMC_DEVICE_ID  &&
+                   id->device_revision == DELL_POWEREDGE_8G_BMC_DEVICE_REV &&
+                   id->ipmi_version   == DELL_POWEREDGE_8G_BMC_IPMI_VERSION) {
+                       smi_info->oem_data_avail_handler =
+                               oem_data_avail_to_receive_msg_avail;
+               } else if (ipmi_version_major(id) < 1 ||
+                          (ipmi_version_major(id) == 1 &&
+                           ipmi_version_minor(id) < 5)) {
+                       smi_info->oem_data_avail_handler =
+                               oem_data_avail_to_receive_msg_avail;
+               }
+       }
+}
+
+#define CANNOT_RETURN_REQUESTED_LENGTH 0xCA
+static void return_hosed_msg_badsize(struct smi_info *smi_info)
+{
+       struct ipmi_smi_msg *msg = smi_info->curr_msg;
+
+       /* Make it a response */
+       msg->rsp[0] = msg->data[0] | 4;
+       msg->rsp[1] = msg->data[1];
+       msg->rsp[2] = CANNOT_RETURN_REQUESTED_LENGTH;
+       msg->rsp_size = 3;
+       smi_info->curr_msg = NULL;
+       deliver_recv_msg(smi_info, msg);
+}
+
+/*
+ * dell_poweredge_bt_xaction_handler
+ * @info - smi_info.device_id must be populated
+ *
+ * Dell PowerEdge servers with the BT interface (x6xx and 1750) will
+ * not respond to a Get SDR command if the length of the data
+ * requested is exactly 0x3A, which leads to command timeouts and no
+ * data returned.  This intercepts such commands, and causes userspace
+ * callers to try again with a different-sized buffer, which succeeds.
+ */
+
+#define STORAGE_NETFN 0x0A
+#define STORAGE_CMD_GET_SDR 0x23
+static int dell_poweredge_bt_xaction_handler(struct notifier_block *self,
+                                            unsigned long unused,
+                                            void *in)
+{
+       struct smi_info *smi_info = in;
+       unsigned char *data = smi_info->curr_msg->data;
+       unsigned int size   = smi_info->curr_msg->data_size;
+       if (size >= 8 &&
+           (data[0]>>2) == STORAGE_NETFN &&
+           data[1] == STORAGE_CMD_GET_SDR &&
+           data[7] == 0x3A) {
+               return_hosed_msg_badsize(smi_info);
+               return NOTIFY_STOP;
+       }
+       return NOTIFY_DONE;
+}
+
+static struct notifier_block dell_poweredge_bt_xaction_notifier = {
+       .notifier_call  = dell_poweredge_bt_xaction_handler,
+};
+
+/*
+ * setup_dell_poweredge_bt_xaction_handler
+ * @info - smi_info.device_id must be filled in already
+ *
+ * Fills in smi_info.device_id.start_transaction_pre_hook
+ * when we know what function to use there.
+ */
+static void
+setup_dell_poweredge_bt_xaction_handler(struct smi_info *smi_info)
+{
+       struct ipmi_device_id *id = &smi_info->device_id;
+       if (id->manufacturer_id == DELL_IANA_MFR_ID &&
+           smi_info->si_type == SI_BT)
+               register_xaction_notifier(&dell_poweredge_bt_xaction_notifier);
+}
+
+/*
+ * setup_oem_data_handler
+ * @info - smi_info.device_id must be filled in already
+ *
+ * Fills in smi_info.device_id.oem_data_available_handler
+ * when we know what function to use there.
+ */
+
+static void setup_oem_data_handler(struct smi_info *smi_info)
+{
+       setup_dell_poweredge_oem_data_handler(smi_info);
+}
+
+static void setup_xaction_handlers(struct smi_info *smi_info)
+{
+       setup_dell_poweredge_bt_xaction_handler(smi_info);
+}
+
+static inline void wait_for_timer_and_thread(struct smi_info *smi_info)
+{
+       if (smi_info->thread != NULL)
+               kthread_stop(smi_info->thread);
+       if (smi_info->timer_running)
+               del_timer_sync(&smi_info->si_timer);
+}
+
+static struct ipmi_default_vals
+{
+       int type;
+       int port;
+} ipmi_defaults[] =
+{
+       { .type = SI_KCS, .port = 0xca2 },
+       { .type = SI_SMIC, .port = 0xca9 },
+       { .type = SI_BT, .port = 0xe4 },
+       { .port = 0 }
+};
+
+static void default_find_bmc(void)
+{
+       struct smi_info *info;
+       int             i;
+
+       for (i = 0; ; i++) {
+               if (!ipmi_defaults[i].port)
+                       break;
+#ifdef CONFIG_PPC
+               if (check_legacy_ioport(ipmi_defaults[i].port))
+                       continue;
+#endif
+               info = smi_info_alloc();
+               if (!info)
+                       return;
+
+               info->addr_source = SI_DEFAULT;
+
+               info->si_type = ipmi_defaults[i].type;
+               info->io_setup = port_setup;
+               info->io.addr_data = ipmi_defaults[i].port;
+               info->io.addr_type = IPMI_IO_ADDR_SPACE;
+
+               info->io.addr = NULL;
+               info->io.regspacing = DEFAULT_REGSPACING;
+               info->io.regsize = DEFAULT_REGSPACING;
+               info->io.regshift = 0;
+
+               if (add_smi(info) == 0) {
+                       if ((try_smi_init(info)) == 0) {
+                               /* Found one... */
+                               printk(KERN_INFO PFX "Found default %s"
+                               " state machine at %s address 0x%lx\n",
+                               si_to_str[info->si_type],
+                               addr_space_to_str[info->io.addr_type],
+                               info->io.addr_data);
+                       } else
+                               cleanup_one_si(info);
+               } else {
+                       kfree(info);
+               }
+       }
+}
+
+static int is_new_interface(struct smi_info *info)
+{
+       struct smi_info *e;
+
+       list_for_each_entry(e, &smi_infos, link) {
+               if (e->io.addr_type != info->io.addr_type)
+                       continue;
+               if (e->io.addr_data == info->io.addr_data)
+                       return 0;
+       }
+
+       return 1;
+}
+
+static int add_smi(struct smi_info *new_smi)
+{
+       int rv = 0;
+
+       printk(KERN_INFO PFX "Adding %s-specified %s state machine",
+              ipmi_addr_src_to_str(new_smi->addr_source),
+              si_to_str[new_smi->si_type]);
+       mutex_lock(&smi_infos_lock);
+       if (!is_new_interface(new_smi)) {
+               printk(KERN_CONT " duplicate interface\n");
+               rv = -EBUSY;
+               goto out_err;
+       }
+
+       printk(KERN_CONT "\n");
+
+       /* So we know not to free it unless we have allocated one. */
+       new_smi->intf = NULL;
+       new_smi->si_sm = NULL;
+       new_smi->handlers = NULL;
+
+       list_add_tail(&new_smi->link, &smi_infos);
+
+out_err:
+       mutex_unlock(&smi_infos_lock);
+       return rv;
+}
+
+static int try_smi_init(struct smi_info *new_smi)
+{
+       int rv = 0;
+       int i;
+
+       printk(KERN_INFO PFX "Trying %s-specified %s state"
+              " machine at %s address 0x%lx, slave address 0x%x,"
+              " irq %d\n",
+              ipmi_addr_src_to_str(new_smi->addr_source),
+              si_to_str[new_smi->si_type],
+              addr_space_to_str[new_smi->io.addr_type],
+              new_smi->io.addr_data,
+              new_smi->slave_addr, new_smi->irq);
+
+       switch (new_smi->si_type) {
+       case SI_KCS:
+               new_smi->handlers = &kcs_smi_handlers;
+               break;
+
+       case SI_SMIC:
+               new_smi->handlers = &smic_smi_handlers;
+               break;
+
+       case SI_BT:
+               new_smi->handlers = &bt_smi_handlers;
+               break;
+
+       default:
+               /* No support for anything else yet. */
+               rv = -EIO;
+               goto out_err;
+       }
+
+       /* Allocate the state machine's data and initialize it. */
+       new_smi->si_sm = kmalloc(new_smi->handlers->size(), GFP_KERNEL);
+       if (!new_smi->si_sm) {
+               printk(KERN_ERR PFX
+                      "Could not allocate state machine memory\n");
+               rv = -ENOMEM;
+               goto out_err;
+       }
+       new_smi->io_size = new_smi->handlers->init_data(new_smi->si_sm,
+                                                       &new_smi->io);
+
+       /* Now that we know the I/O size, we can set up the I/O. */
+       rv = new_smi->io_setup(new_smi);
+       if (rv) {
+               printk(KERN_ERR PFX "Could not set up I/O space\n");
+               goto out_err;
+       }
+
+       /* Do low-level detection first. */
+       if (new_smi->handlers->detect(new_smi->si_sm)) {
+               if (new_smi->addr_source)
+                       printk(KERN_INFO PFX "Interface detection failed\n");
+               rv = -ENODEV;
+               goto out_err;
+       }
+
+       /*
+        * Attempt a get device id command.  If it fails, we probably
+        * don't have a BMC here.
+        */
+       rv = try_get_dev_id(new_smi);
+       if (rv) {
+               if (new_smi->addr_source)
+                       printk(KERN_INFO PFX "There appears to be no BMC"
+                              " at this location\n");
+               goto out_err;
+       }
+
+       check_clr_rcv_irq(new_smi);
+
+       setup_oem_data_handler(new_smi);
+       setup_xaction_handlers(new_smi);
+
+       new_smi->waiting_msg = NULL;
+       new_smi->curr_msg = NULL;
+       atomic_set(&new_smi->req_events, 0);
+       new_smi->run_to_completion = false;
+       for (i = 0; i < SI_NUM_STATS; i++)
+               atomic_set(&new_smi->stats[i], 0);
+
+       new_smi->interrupt_disabled = true;
+       atomic_set(&new_smi->need_watch, 0);
+       new_smi->intf_num = smi_num;
+       smi_num++;
+
+       rv = try_enable_event_buffer(new_smi);
+       if (rv == 0)
+               new_smi->has_event_buffer = true;
+
+       /*
+        * Start clearing the flags before we enable interrupts or the
+        * timer to avoid racing with the timer.
+        */
+       start_clear_flags(new_smi);
+
+       /*
+        * IRQ is defined to be set when non-zero.  req_events will
+        * cause a global flags check that will enable interrupts.
+        */
+       if (new_smi->irq) {
+               new_smi->interrupt_disabled = false;
+               atomic_set(&new_smi->req_events, 1);
+       }
+
+       if (!new_smi->dev) {
+               /*
+                * If we don't already have a device from something
+                * else (like PCI), then register a new one.
+                */
+               new_smi->pdev = platform_device_alloc("ipmi_si",
+                                                     new_smi->intf_num);
+               if (!new_smi->pdev) {
+                       printk(KERN_ERR PFX
+                              "Unable to allocate platform device\n");
+                       goto out_err;
+               }
+               new_smi->dev = &new_smi->pdev->dev;
+               new_smi->dev->driver = &ipmi_driver.driver;
+
+               rv = platform_device_add(new_smi->pdev);
+               if (rv) {
+                       printk(KERN_ERR PFX
+                              "Unable to register system interface device:"
+                              " %d\n",
+                              rv);
+                       goto out_err;
+               }
+               new_smi->dev_registered = true;
+       }
+
+       rv = ipmi_register_smi(&handlers,
+                              new_smi,
+                              &new_smi->device_id,
+                              new_smi->dev,
+                              new_smi->slave_addr);
+       if (rv) {
+               dev_err(new_smi->dev, "Unable to register device: error %d\n",
+                       rv);
+               goto out_err_stop_timer;
+       }
+
+       rv = ipmi_smi_add_proc_entry(new_smi->intf, "type",
+                                    &smi_type_proc_ops,
+                                    new_smi);
+       if (rv) {
+               dev_err(new_smi->dev, "Unable to create proc entry: %d\n", rv);
+               goto out_err_stop_timer;
+       }
+
+       rv = ipmi_smi_add_proc_entry(new_smi->intf, "si_stats",
+                                    &smi_si_stats_proc_ops,
+                                    new_smi);
+       if (rv) {
+               dev_err(new_smi->dev, "Unable to create proc entry: %d\n", rv);
+               goto out_err_stop_timer;
+       }
+
+       rv = ipmi_smi_add_proc_entry(new_smi->intf, "params",
+                                    &smi_params_proc_ops,
+                                    new_smi);
+       if (rv) {
+               dev_err(new_smi->dev, "Unable to create proc entry: %d\n", rv);
+               goto out_err_stop_timer;
+       }
+
+       dev_info(new_smi->dev, "IPMI %s interface initialized\n",
+                si_to_str[new_smi->si_type]);
+
+       return 0;
+
+ out_err_stop_timer:
+       wait_for_timer_and_thread(new_smi);
+
+ out_err:
+       new_smi->interrupt_disabled = true;
+
+       if (new_smi->intf) {
+               ipmi_smi_t intf = new_smi->intf;
+               new_smi->intf = NULL;
+               ipmi_unregister_smi(intf);
+       }
+
+       if (new_smi->irq_cleanup) {
+               new_smi->irq_cleanup(new_smi);
+               new_smi->irq_cleanup = NULL;
+       }
+
+       /*
+        * Wait until we know that we are out of any interrupt
+        * handlers might have been running before we freed the
+        * interrupt.
+        */
+       synchronize_sched();
+
+       if (new_smi->si_sm) {
+               if (new_smi->handlers)
+                       new_smi->handlers->cleanup(new_smi->si_sm);
+               kfree(new_smi->si_sm);
+               new_smi->si_sm = NULL;
+       }
+       if (new_smi->addr_source_cleanup) {
+               new_smi->addr_source_cleanup(new_smi);
+               new_smi->addr_source_cleanup = NULL;
+       }
+       if (new_smi->io_cleanup) {
+               new_smi->io_cleanup(new_smi);
+               new_smi->io_cleanup = NULL;
+       }
+
+       if (new_smi->dev_registered) {
+               platform_device_unregister(new_smi->pdev);
+               new_smi->dev_registered = false;
+       }
+
+       return rv;
+}
+
+static int init_ipmi_si(void)
+{
+       int  i;
+       char *str;
+       int  rv;
+       struct smi_info *e;
+       enum ipmi_addr_src type = SI_INVALID;
+
+       if (initialized)
+               return 0;
+       initialized = 1;
+
+       if (si_tryplatform) {
+               rv = platform_driver_register(&ipmi_driver);
+               if (rv) {
+                       printk(KERN_ERR PFX "Unable to register "
+                              "driver: %d\n", rv);
+                       return rv;
+               }
+       }
+
+       /* Parse out the si_type string into its components. */
+       str = si_type_str;
+       if (*str != '\0') {
+               for (i = 0; (i < SI_MAX_PARMS) && (*str != '\0'); i++) {
+                       si_type[i] = str;
+                       str = strchr(str, ',');
+                       if (str) {
+                               *str = '\0';
+                               str++;
+                       } else {
+                               break;
+                       }
+               }
+       }
+
+       printk(KERN_INFO "IPMI System Interface driver.\n");
+
+       /* If the user gave us a device, they presumably want us to use it */
+       if (!hardcode_find_bmc())
+               return 0;
+
+#ifdef CONFIG_PCI
+       if (si_trypci) {
+               rv = pci_register_driver(&ipmi_pci_driver);
+               if (rv)
+                       printk(KERN_ERR PFX "Unable to register "
+                              "PCI driver: %d\n", rv);
+               else
+                       pci_registered = true;
+       }
+#endif
+
+#ifdef CONFIG_ACPI
+       if (si_tryacpi) {
+               pnp_register_driver(&ipmi_pnp_driver);
+               pnp_registered = true;
+       }
+#endif
+
+#ifdef CONFIG_DMI
+       if (si_trydmi)
+               dmi_find_bmc();
+#endif
+
+#ifdef CONFIG_ACPI
+       if (si_tryacpi)
+               spmi_find_bmc();
+#endif
+
+#ifdef CONFIG_PARISC
+       register_parisc_driver(&ipmi_parisc_driver);
+       parisc_registered = true;
+       /* poking PC IO addresses will crash machine, don't do it */
+       si_trydefaults = 0;
+#endif
+
+       /* We prefer devices with interrupts, but in the case of a machine
+          with multiple BMCs we assume that there will be several instances
+          of a given type so if we succeed in registering a type then also
+          try to register everything else of the same type */
+
+       mutex_lock(&smi_infos_lock);
+       list_for_each_entry(e, &smi_infos, link) {
+               /* Try to register a device if it has an IRQ and we either
+                  haven't successfully registered a device yet or this
+                  device has the same type as one we successfully registered */
+               if (e->irq && (!type || e->addr_source == type)) {
+                       if (!try_smi_init(e)) {
+                               type = e->addr_source;
+                       }
+               }
+       }
+
+       /* type will only have been set if we successfully registered an si */
+       if (type) {
+               mutex_unlock(&smi_infos_lock);
+               return 0;
+       }
+
+       /* Fall back to the preferred device */
+
+       list_for_each_entry(e, &smi_infos, link) {
+               if (!e->irq && (!type || e->addr_source == type)) {
+                       if (!try_smi_init(e)) {
+                               type = e->addr_source;
+                       }
+               }
+       }
+       mutex_unlock(&smi_infos_lock);
+
+       if (type)
+               return 0;
+
+       if (si_trydefaults) {
+               mutex_lock(&smi_infos_lock);
+               if (list_empty(&smi_infos)) {
+                       /* No BMC was found, try defaults. */
+                       mutex_unlock(&smi_infos_lock);
+                       default_find_bmc();
+               } else
+                       mutex_unlock(&smi_infos_lock);
+       }
+
+       mutex_lock(&smi_infos_lock);
+       if (unload_when_empty && list_empty(&smi_infos)) {
+               mutex_unlock(&smi_infos_lock);
+               cleanup_ipmi_si();
+               printk(KERN_WARNING PFX
+                      "Unable to find any System Interface(s)\n");
+               return -ENODEV;
+       } else {
+               mutex_unlock(&smi_infos_lock);
+               return 0;
+       }
+}
+module_init(init_ipmi_si);
+
+static void cleanup_one_si(struct smi_info *to_clean)
+{
+       int           rv = 0;
+
+       if (!to_clean)
+               return;
+
+       if (to_clean->intf) {
+               ipmi_smi_t intf = to_clean->intf;
+
+               to_clean->intf = NULL;
+               rv = ipmi_unregister_smi(intf);
+               if (rv) {
+                       pr_err(PFX "Unable to unregister device: errno=%d\n",
+                              rv);
+               }
+       }
+
+       if (to_clean->dev)
+               dev_set_drvdata(to_clean->dev, NULL);
+
+       list_del(&to_clean->link);
+
+       /*
+        * Make sure that interrupts, the timer and the thread are
+        * stopped and will not run again.
+        */
+       if (to_clean->irq_cleanup)
+               to_clean->irq_cleanup(to_clean);
+       wait_for_timer_and_thread(to_clean);
+
+       /*
+        * Timeouts are stopped, now make sure the interrupts are off
+        * in the BMC.  Note that timers and CPU interrupts are off,
+        * so no need for locks.
+        */
+       while (to_clean->curr_msg || (to_clean->si_state != SI_NORMAL)) {
+               poll(to_clean);
+               schedule_timeout_uninterruptible(1);
+       }
+       disable_si_irq(to_clean);
+       while (to_clean->curr_msg || (to_clean->si_state != SI_NORMAL)) {
+               poll(to_clean);
+               schedule_timeout_uninterruptible(1);
+       }
+
+       if (to_clean->handlers)
+               to_clean->handlers->cleanup(to_clean->si_sm);
+
+       kfree(to_clean->si_sm);
+
+       if (to_clean->addr_source_cleanup)
+               to_clean->addr_source_cleanup(to_clean);
+       if (to_clean->io_cleanup)
+               to_clean->io_cleanup(to_clean);
+
+       if (to_clean->dev_registered)
+               platform_device_unregister(to_clean->pdev);
+
+       kfree(to_clean);
+}
+
+static void cleanup_ipmi_si(void)
+{
+       struct smi_info *e, *tmp_e;
+
+       if (!initialized)
+               return;
+
+#ifdef CONFIG_PCI
+       if (pci_registered)
+               pci_unregister_driver(&ipmi_pci_driver);
+#endif
+#ifdef CONFIG_ACPI
+       if (pnp_registered)
+               pnp_unregister_driver(&ipmi_pnp_driver);
+#endif
+#ifdef CONFIG_PARISC
+       if (parisc_registered)
+               unregister_parisc_driver(&ipmi_parisc_driver);
+#endif
+
+       platform_driver_unregister(&ipmi_driver);
+
+       mutex_lock(&smi_infos_lock);
+       list_for_each_entry_safe(e, tmp_e, &smi_infos, link)
+               cleanup_one_si(e);
+       mutex_unlock(&smi_infos_lock);
+}
+module_exit(cleanup_ipmi_si);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>");
+MODULE_DESCRIPTION("Interface to the IPMI driver for the KCS, SMIC, and BT"
+                  " system interfaces.");