Linux-2.6.12-rc2

Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.

Let it rip!

11 files changed, 9546 insertions, 0 deletions

diff --git a/drivers/char/ipmi/Kconfig b/drivers/char/ipmi/Kconfig
new file mode 100644
index 000000000000..a6dcb2918157
--- /dev/null
+++ b/drivers/char/ipmi/Kconfig
@@ -0,0 +1,67 @@
+#
+# IPMI device configuration
+#
+
+menu "IPMI"
+config IPMI_HANDLER
+       tristate 'IPMI top-level message handler'
+       help
+         This enables the central IPMI message handler, required for IPMI
+	 to work.
+
+         IPMI is a standard for managing sensors (temperature,
+         voltage, etc.) in a system.
+
+         See <file:Documentation/IPMI.txt> for more details on the driver.
+
+	 If unsure, say N.
+
+config IPMI_PANIC_EVENT
+       bool 'Generate a panic event to all BMCs on a panic'
+       depends on IPMI_HANDLER
+       help
+         When a panic occurs, this will cause the IPMI message handler to
+	 generate an IPMI event describing the panic to each interface
+	 registered with the message handler.
+
+config IPMI_PANIC_STRING
+	bool 'Generate OEM events containing the panic string'
+	depends on IPMI_PANIC_EVENT
+	help
+	  When a panic occurs, this will cause the IPMI message handler to
+	  generate IPMI OEM type f0 events holding the IPMB address of the
+	  panic generator (byte 4 of the event), a sequence number for the
+	  string (byte 5 of the event) and part of the string (the rest of the
+	  event).  Bytes 1, 2, and 3 are the normal usage for an OEM event.
+	  You can fetch these events and use the sequence numbers to piece the
+	  string together.
+
+config IPMI_DEVICE_INTERFACE
+       tristate 'Device interface for IPMI'
+       depends on IPMI_HANDLER
+       help
+         This provides an IOCTL interface to the IPMI message handler so
+	 userland processes may use IPMI.  It supports poll() and select().
+
+config IPMI_SI
+       tristate 'IPMI System Interface handler'
+       depends on IPMI_HANDLER
+       help
+         Provides a driver for System Interfaces (KCS, SMIC, BT).
+	 Currently, only KCS and SMIC are supported.  If
+	 you are using IPMI, you should probably say "y" here.
+
+config IPMI_WATCHDOG
+       tristate 'IPMI Watchdog Timer'
+       depends on IPMI_HANDLER
+       help
+         This enables the IPMI watchdog timer.
+
+config IPMI_POWEROFF
+       tristate 'IPMI Poweroff'
+       depends on IPMI_HANDLER
+       help
+         This enables a function to power off the system with IPMI if
+	 the IPMI management controller is capable of this.
+
+endmenu
diff --git a/drivers/char/ipmi/Makefile b/drivers/char/ipmi/Makefile
new file mode 100644
index 000000000000..553f0a408eda
--- /dev/null
+++ b/drivers/char/ipmi/Makefile
@@ -0,0 +1,15 @@
+#
+# Makefile for the ipmi drivers.
+#
+
+ipmi_si-objs := ipmi_si_intf.o ipmi_kcs_sm.o ipmi_smic_sm.o ipmi_bt_sm.o
+
+obj-$(CONFIG_IPMI_HANDLER) += ipmi_msghandler.o
+obj-$(CONFIG_IPMI_DEVICE_INTERFACE) += ipmi_devintf.o
+obj-$(CONFIG_IPMI_SI) += ipmi_si.o
+obj-$(CONFIG_IPMI_WATCHDOG) += ipmi_watchdog.o
+obj-$(CONFIG_IPMI_POWEROFF) += ipmi_poweroff.o
+
+ipmi_si.o:	$(ipmi_si-objs)
+	$(LD) -r -o $@ $(ipmi_si-objs)
+
diff --git a/drivers/char/ipmi/ipmi_bt_sm.c b/drivers/char/ipmi/ipmi_bt_sm.c
new file mode 100644
index 000000000000..225b330115bb
--- /dev/null
+++ b/drivers/char/ipmi/ipmi_bt_sm.c
@@ -0,0 +1,513 @@
+/*
+ *  ipmi_bt_sm.c
+ *
+ *  The state machine for an Open IPMI BT sub-driver under ipmi_si.c, part
+ *  of the driver architecture at http://sourceforge.net/project/openipmi
+ *
+ *  Author:	Rocky Craig <first.last@hp.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.  */
+
+#include <linux/kernel.h> /* For printk. */
+#include <linux/string.h>
+#include <linux/ipmi_msgdefs.h>		/* for completion codes */
+#include "ipmi_si_sm.h"
+
+#define IPMI_BT_VERSION "v33"
+
+static int bt_debug = 0x00;	/* Production value 0, see following flags */
+
+#define	BT_DEBUG_ENABLE	1
+#define BT_DEBUG_MSG	2
+#define BT_DEBUG_STATES	4
+
+/* Typical "Get BT Capabilities" values are 2-3 retries, 5-10 seconds,
+   and 64 byte buffers.  However, one HP implementation wants 255 bytes of
+   buffer (with a documented message of 160 bytes) so go for the max.
+   Since the Open IPMI architecture is single-message oriented at this
+   stage, the queue depth of BT is of no concern. */
+
+#define BT_NORMAL_TIMEOUT	2000000	/* seconds in microseconds */
+#define BT_RETRY_LIMIT		2
+#define BT_RESET_DELAY		6000000	/* 6 seconds after warm reset */
+
+enum bt_states {
+	BT_STATE_IDLE,
+	BT_STATE_XACTION_START,
+	BT_STATE_WRITE_BYTES,
+	BT_STATE_WRITE_END,
+	BT_STATE_WRITE_CONSUME,
+	BT_STATE_B2H_WAIT,
+	BT_STATE_READ_END,
+	BT_STATE_RESET1,		/* These must come last */
+	BT_STATE_RESET2,
+	BT_STATE_RESET3,
+	BT_STATE_RESTART,
+	BT_STATE_HOSED
+};
+
+struct si_sm_data {
+	enum bt_states	state;
+	enum bt_states	last_state;	/* assist printing and resets */
+	unsigned char	seq;		/* BT sequence number */
+	struct si_sm_io	*io;
+        unsigned char	write_data[IPMI_MAX_MSG_LENGTH];
+        int		write_count;
+        unsigned char	read_data[IPMI_MAX_MSG_LENGTH];
+        int		read_count;
+        int		truncated;
+        long		timeout;
+        unsigned int	error_retries;	/* end of "common" fields */
+	int		nonzero_status;	/* hung BMCs stay all 0 */
+};
+
+#define BT_CLR_WR_PTR	0x01	/* See IPMI 1.5 table 11.6.4 */
+#define BT_CLR_RD_PTR	0x02
+#define BT_H2B_ATN	0x04
+#define BT_B2H_ATN	0x08
+#define BT_SMS_ATN	0x10
+#define BT_OEM0		0x20
+#define BT_H_BUSY	0x40
+#define BT_B_BUSY	0x80
+
+/* Some bits are toggled on each write: write once to set it, once
+   more to clear it; writing a zero does nothing.  To absolutely
+   clear it, check its state and write if set.  This avoids the "get
+   current then use as mask" scheme to modify one bit.  Note that the
+   variable "bt" is hardcoded into these macros. */
+
+#define BT_STATUS	bt->io->inputb(bt->io, 0)
+#define BT_CONTROL(x)	bt->io->outputb(bt->io, 0, x)
+
+#define BMC2HOST	bt->io->inputb(bt->io, 1)
+#define HOST2BMC(x)	bt->io->outputb(bt->io, 1, x)
+
+#define BT_INTMASK_R	bt->io->inputb(bt->io, 2)
+#define BT_INTMASK_W(x)	bt->io->outputb(bt->io, 2, x)
+
+/* Convenience routines for debugging.  These are not multi-open safe!
+   Note the macros have hardcoded variables in them. */
+
+static char *state2txt(unsigned char state)
+{
+	switch (state) {
+		case BT_STATE_IDLE:		return("IDLE");
+		case BT_STATE_XACTION_START:	return("XACTION");
+		case BT_STATE_WRITE_BYTES:	return("WR_BYTES");
+		case BT_STATE_WRITE_END:	return("WR_END");
+		case BT_STATE_WRITE_CONSUME:	return("WR_CONSUME");
+		case BT_STATE_B2H_WAIT:		return("B2H_WAIT");
+		case BT_STATE_READ_END:		return("RD_END");
+		case BT_STATE_RESET1:		return("RESET1");
+		case BT_STATE_RESET2:		return("RESET2");
+		case BT_STATE_RESET3:		return("RESET3");
+		case BT_STATE_RESTART:		return("RESTART");
+		case BT_STATE_HOSED:		return("HOSED");
+	}
+	return("BAD STATE");
+}
+#define STATE2TXT state2txt(bt->state)
+
+static char *status2txt(unsigned char status, char *buf)
+{
+	strcpy(buf, "[ ");
+	if (status & BT_B_BUSY) strcat(buf, "B_BUSY ");
+	if (status & BT_H_BUSY) strcat(buf, "H_BUSY ");
+	if (status & BT_OEM0) strcat(buf, "OEM0 ");
+	if (status & BT_SMS_ATN) strcat(buf, "SMS ");
+	if (status & BT_B2H_ATN) strcat(buf, "B2H ");
+	if (status & BT_H2B_ATN) strcat(buf, "H2B ");
+	strcat(buf, "]");
+	return buf;
+}
+#define STATUS2TXT(buf) status2txt(status, buf)
+
+/* This will be called from within this module on a hosed condition */
+#define FIRST_SEQ	0
+static unsigned int bt_init_data(struct si_sm_data *bt, struct si_sm_io *io)
+{
+	bt->state = BT_STATE_IDLE;
+	bt->last_state = BT_STATE_IDLE;
+	bt->seq = FIRST_SEQ;
+	bt->io = io;
+	bt->write_count = 0;
+	bt->read_count = 0;
+	bt->error_retries = 0;
+	bt->nonzero_status = 0;
+	bt->truncated = 0;
+	bt->timeout = BT_NORMAL_TIMEOUT;
+	return 3; /* We claim 3 bytes of space; ought to check SPMI table */
+}
+
+static int bt_start_transaction(struct si_sm_data *bt,
+				unsigned char *data,
+				unsigned int size)
+{
+	unsigned int i;
+
+	if ((size < 2) || (size > IPMI_MAX_MSG_LENGTH)) return -1;
+
+	if ((bt->state != BT_STATE_IDLE) && (bt->state != BT_STATE_HOSED))
+		return -2;
+
+	if (bt_debug & BT_DEBUG_MSG) {
+    		printk(KERN_WARNING "+++++++++++++++++++++++++++++++++++++\n");
+		printk(KERN_WARNING "BT: write seq=0x%02X:", bt->seq);
+		for (i = 0; i < size; i ++) printk (" %02x", data[i]);
+		printk("\n");
+	}
+	bt->write_data[0] = size + 1;	/* all data plus seq byte */
+	bt->write_data[1] = *data;	/* NetFn/LUN */
+	bt->write_data[2] = bt->seq;
+	memcpy(bt->write_data + 3, data + 1, size - 1);
+	bt->write_count = size + 2;
+
+	bt->error_retries = 0;
+	bt->nonzero_status = 0;
+	bt->read_count = 0;
+	bt->truncated = 0;
+	bt->state = BT_STATE_XACTION_START;
+	bt->last_state = BT_STATE_IDLE;
+	bt->timeout = BT_NORMAL_TIMEOUT;
+	return 0;
+}
+
+/* After the upper state machine has been told SI_SM_TRANSACTION_COMPLETE
+   it calls this.  Strip out the length and seq bytes. */
+
+static int bt_get_result(struct si_sm_data *bt,
+			   unsigned char *data,
+			   unsigned int length)
+{
+	int i, msg_len;
+
+	msg_len = bt->read_count - 2;		/* account for length & seq */
+	/* Always NetFn, Cmd, cCode */
+	if (msg_len < 3 || msg_len > IPMI_MAX_MSG_LENGTH) {
+		printk(KERN_WARNING "BT results: bad msg_len = %d\n", msg_len);
+		data[0] = bt->write_data[1] | 0x4;	/* Kludge a response */
+		data[1] = bt->write_data[3];
+		data[2] = IPMI_ERR_UNSPECIFIED;
+		msg_len = 3;
+	} else {
+		data[0] = bt->read_data[1];
+		data[1] = bt->read_data[3];
+		if (length < msg_len) bt->truncated = 1;
+		if (bt->truncated) {	/* can be set in read_all_bytes() */
+			data[2] = IPMI_ERR_MSG_TRUNCATED;
+			msg_len = 3;
+		} else memcpy(data + 2, bt->read_data + 4, msg_len - 2);
+
+		if (bt_debug & BT_DEBUG_MSG) {
+			printk (KERN_WARNING "BT: res (raw)");
+			for (i = 0; i < msg_len; i++) printk(" %02x", data[i]);
+			printk ("\n");
+		}
+	}
+	bt->read_count = 0;	/* paranoia */
+	return msg_len;
+}
+
+/* This bit's functionality is optional */
+#define BT_BMC_HWRST	0x80
+
+static void reset_flags(struct si_sm_data *bt)
+{
+	if (BT_STATUS & BT_H_BUSY) BT_CONTROL(BT_H_BUSY);
+	if (BT_STATUS & BT_B_BUSY) BT_CONTROL(BT_B_BUSY);
+	BT_CONTROL(BT_CLR_WR_PTR);
+	BT_CONTROL(BT_SMS_ATN);
+	BT_INTMASK_W(BT_BMC_HWRST);
+#ifdef DEVELOPMENT_ONLY_NOT_FOR_PRODUCTION
+	if (BT_STATUS & BT_B2H_ATN) {
+		int i;
+		BT_CONTROL(BT_H_BUSY);
+		BT_CONTROL(BT_B2H_ATN);
+		BT_CONTROL(BT_CLR_RD_PTR);
+		for (i = 0; i < IPMI_MAX_MSG_LENGTH + 2; i++) BMC2HOST;
+		BT_CONTROL(BT_H_BUSY);
+	}
+#endif
+}
+
+static inline void write_all_bytes(struct si_sm_data *bt)
+{
+	int i;
+
+	if (bt_debug & BT_DEBUG_MSG) {
+    		printk(KERN_WARNING "BT: write %d bytes seq=0x%02X",
+			bt->write_count, bt->seq);
+		for (i = 0; i < bt->write_count; i++)
+			printk (" %02x", bt->write_data[i]);
+		printk ("\n");
+	}
+	for (i = 0; i < bt->write_count; i++) HOST2BMC(bt->write_data[i]);
+}
+
+static inline int read_all_bytes(struct si_sm_data *bt)
+{
+	unsigned char i;
+
+	bt->read_data[0] = BMC2HOST;
+	bt->read_count = bt->read_data[0];
+	if (bt_debug & BT_DEBUG_MSG)
+    		printk(KERN_WARNING "BT: read %d bytes:", bt->read_count);
+
+	/* minimum: length, NetFn, Seq, Cmd, cCode == 5 total, or 4 more
+	   following the length byte. */
+	if (bt->read_count < 4 || bt->read_count >= IPMI_MAX_MSG_LENGTH) {
+		if (bt_debug & BT_DEBUG_MSG)
+			printk("bad length %d\n", bt->read_count);
+		bt->truncated = 1;
+		return 1;	/* let next XACTION START clean it up */
+	}
+	for (i = 1; i <= bt->read_count; i++) bt->read_data[i] = BMC2HOST;
+	bt->read_count++;	/* account for the length byte */
+
+	if (bt_debug & BT_DEBUG_MSG) {
+	    	for (i = 0; i < bt->read_count; i++)
+			printk (" %02x", bt->read_data[i]);
+	    	printk ("\n");
+	}
+	if (bt->seq != bt->write_data[2])	/* idiot check */
+		printk(KERN_WARNING "BT: internal error: sequence mismatch\n");
+
+	/* per the spec, the (NetFn, Seq, Cmd) tuples should match */
+	if ((bt->read_data[3] == bt->write_data[3]) &&		/* Cmd */
+        	(bt->read_data[2] == bt->write_data[2]) &&	/* Sequence */
+        	((bt->read_data[1] & 0xF8) == (bt->write_data[1] & 0xF8)))
+			return 1;
+
+	if (bt_debug & BT_DEBUG_MSG) printk(KERN_WARNING "BT: bad packet: "
+		"want 0x(%02X, %02X, %02X) got (%02X, %02X, %02X)\n",
+		bt->write_data[1], bt->write_data[2], bt->write_data[3],
+		bt->read_data[1],  bt->read_data[2],  bt->read_data[3]);
+	return 0;
+}
+
+/* Modifies bt->state appropriately, need to get into the bt_event() switch */
+
+static void error_recovery(struct si_sm_data *bt, char *reason)
+{
+	unsigned char status;
+	char buf[40]; /* For getting status */
+
+	bt->timeout = BT_NORMAL_TIMEOUT; /* various places want to retry */
+
+	status = BT_STATUS;
+	printk(KERN_WARNING "BT: %s in %s %s ", reason, STATE2TXT,
+	       STATUS2TXT(buf));
+
+	(bt->error_retries)++;
+	if (bt->error_retries > BT_RETRY_LIMIT) {
+		printk("retry limit (%d) exceeded\n", BT_RETRY_LIMIT);
+		bt->state = BT_STATE_HOSED;
+		if (!bt->nonzero_status)
+			printk(KERN_ERR "IPMI: BT stuck, try power cycle\n");
+		else if (bt->seq == FIRST_SEQ + BT_RETRY_LIMIT) {
+			/* most likely during insmod */
+			printk(KERN_WARNING "IPMI: BT reset (takes 5 secs)\n");
+        		bt->state = BT_STATE_RESET1;
+		}
+	return;
+	}
+
+	/* Sometimes the BMC queues get in an "off-by-one" state...*/
+	if ((bt->state == BT_STATE_B2H_WAIT) && (status & BT_B2H_ATN)) {
+    		printk("retry B2H_WAIT\n");
+		return;
+	}
+
+	printk("restart command\n");
+	bt->state = BT_STATE_RESTART;
+}
+
+/* Check the status and (possibly) advance the BT state machine.  The
+   default return is SI_SM_CALL_WITH_DELAY. */
+
+static enum si_sm_result bt_event(struct si_sm_data *bt, long time)
+{
+	unsigned char status;
+	char buf[40]; /* For getting status */
+	int i;
+
+	status = BT_STATUS;
+	bt->nonzero_status |= status;
+
+	if ((bt_debug & BT_DEBUG_STATES) && (bt->state != bt->last_state))
+		printk(KERN_WARNING "BT: %s %s TO=%ld - %ld \n",
+			STATE2TXT,
+			STATUS2TXT(buf),
+			bt->timeout,
+			time);
+	bt->last_state = bt->state;
+
+	if (bt->state == BT_STATE_HOSED) return SI_SM_HOSED;
+
+	if (bt->state != BT_STATE_IDLE) {	/* do timeout test */
+
+		/* Certain states, on error conditions, can lock up a CPU
+		   because they are effectively in an infinite loop with
+		   CALL_WITHOUT_DELAY (right back here with time == 0).
+		   Prevent infinite lockup by ALWAYS decrementing timeout. */
+
+    	/* FIXME: bt_event is sometimes called with time > BT_NORMAL_TIMEOUT
+              (noticed in ipmi_smic_sm.c January 2004) */
+
+		if ((time <= 0) || (time >= BT_NORMAL_TIMEOUT)) time = 100;
+		bt->timeout -= time;
+		if ((bt->timeout < 0) && (bt->state < BT_STATE_RESET1)) {
+			error_recovery(bt, "timed out");
+			return SI_SM_CALL_WITHOUT_DELAY;
+		}
+	}
+
+	switch (bt->state) {
+
+    	case BT_STATE_IDLE:	/* check for asynchronous messages */
+		if (status & BT_SMS_ATN) {
+			BT_CONTROL(BT_SMS_ATN);	/* clear it */
+			return SI_SM_ATTN;
+		}
+		return SI_SM_IDLE;
+
+	case BT_STATE_XACTION_START:
+		if (status & BT_H_BUSY) {
+			BT_CONTROL(BT_H_BUSY);
+			break;
+		}
+    		if (status & BT_B2H_ATN) break;
+		bt->state = BT_STATE_WRITE_BYTES;
+		return SI_SM_CALL_WITHOUT_DELAY;	/* for logging */
+
+	case BT_STATE_WRITE_BYTES:
+		if (status & (BT_B_BUSY | BT_H2B_ATN)) break;
+		BT_CONTROL(BT_CLR_WR_PTR);
+		write_all_bytes(bt);
+		BT_CONTROL(BT_H2B_ATN);	/* clears too fast to catch? */
+		bt->state = BT_STATE_WRITE_CONSUME;
+		return SI_SM_CALL_WITHOUT_DELAY; /* it MIGHT sail through */
+
+	case BT_STATE_WRITE_CONSUME: /* BMCs usually blow right thru here */
+        	if (status & (BT_H2B_ATN | BT_B_BUSY)) break;
+		bt->state = BT_STATE_B2H_WAIT;
+		/* fall through with status */
+
+	/* Stay in BT_STATE_B2H_WAIT until a packet matches.  However, spinning
+	   hard here, constantly reading status, seems to hold off the
+	   generation of B2H_ATN so ALWAYS return CALL_WITH_DELAY. */
+
+	case BT_STATE_B2H_WAIT:
+    		if (!(status & BT_B2H_ATN)) break;
+
+		/* Assume ordered, uncached writes: no need to wait */
+		if (!(status & BT_H_BUSY)) BT_CONTROL(BT_H_BUSY); /* set */
+		BT_CONTROL(BT_B2H_ATN);		/* clear it, ACK to the BMC */
+		BT_CONTROL(BT_CLR_RD_PTR);	/* reset the queue */
+		i = read_all_bytes(bt);
+		BT_CONTROL(BT_H_BUSY);		/* clear */
+		if (!i) break;			/* Try this state again */
+		bt->state = BT_STATE_READ_END;
+		return SI_SM_CALL_WITHOUT_DELAY;	/* for logging */
+
+    	case BT_STATE_READ_END:
+
+		/* I could wait on BT_H_BUSY to go clear for a truly clean
+		   exit.  However, this is already done in XACTION_START
+		   and the (possible) extra loop/status/possible wait affects
+		   performance.  So, as long as it works, just ignore H_BUSY */
+
+#ifdef MAKE_THIS_TRUE_IF_NECESSARY
+
+		if (status & BT_H_BUSY) break;
+#endif
+		bt->seq++;
+		bt->state = BT_STATE_IDLE;
+		return SI_SM_TRANSACTION_COMPLETE;
+
+	case BT_STATE_RESET1:
+    		reset_flags(bt);
+    		bt->timeout = BT_RESET_DELAY;
+		bt->state = BT_STATE_RESET2;
+		break;
+
+	case BT_STATE_RESET2:		/* Send a soft reset */
+		BT_CONTROL(BT_CLR_WR_PTR);
+		HOST2BMC(3);		/* number of bytes following */
+		HOST2BMC(0x18);		/* NetFn/LUN == Application, LUN 0 */
+		HOST2BMC(42);		/* Sequence number */
+		HOST2BMC(3);		/* Cmd == Soft reset */
+		BT_CONTROL(BT_H2B_ATN);
+		bt->state = BT_STATE_RESET3;
+		break;
+
+	case BT_STATE_RESET3:
+		if (bt->timeout > 0) return SI_SM_CALL_WITH_DELAY;
+		bt->state = BT_STATE_RESTART;	/* printk in debug modes */
+		break;
+
+	case BT_STATE_RESTART:		/* don't reset retries! */
+		bt->write_data[2] = ++bt->seq;
+		bt->read_count = 0;
+		bt->nonzero_status = 0;
+		bt->timeout = BT_NORMAL_TIMEOUT;
+		bt->state = BT_STATE_XACTION_START;
+		break;
+
+	default:	/* HOSED is supposed to be caught much earlier */
+		error_recovery(bt, "internal logic error");
+		break;
+  	}
+  	return SI_SM_CALL_WITH_DELAY;
+}
+
+static int bt_detect(struct si_sm_data *bt)
+{
+	/* It's impossible for the BT status and interrupt registers to be
+	   all 1's, (assuming a properly functioning, self-initialized BMC)
+	   but that's what you get from reading a bogus address, so we
+	   test that first.  The calling routine uses negative logic. */
+
+	if ((BT_STATUS == 0xFF) && (BT_INTMASK_R == 0xFF)) return 1;
+	reset_flags(bt);
+	return 0;
+}
+
+static void bt_cleanup(struct si_sm_data *bt)
+{
+}
+
+static int bt_size(void)
+{
+	return sizeof(struct si_sm_data);
+}
+
+struct si_sm_handlers bt_smi_handlers =
+{
+	.version           = IPMI_BT_VERSION,
+	.init_data         = bt_init_data,
+	.start_transaction = bt_start_transaction,
+	.get_result        = bt_get_result,
+	.event             = bt_event,
+	.detect            = bt_detect,
+	.cleanup           = bt_cleanup,
+	.size              = bt_size,
+};
diff --git a/drivers/char/ipmi/ipmi_devintf.c b/drivers/char/ipmi/ipmi_devintf.c
new file mode 100644
index 000000000000..49d67f5384a2
--- /dev/null
+++ b/drivers/char/ipmi/ipmi_devintf.c
@@ -0,0 +1,582 @@
+/*
+ * ipmi_devintf.c
+ *
+ * Linux device interface for the IPMI message handler.
+ *
+ * Author: MontaVista Software, Inc.
+ *         Corey Minyard <minyard@mvista.com>
+ *         source@mvista.com
+ *
+ * Copyright 2002 MontaVista Software Inc.
+ *
+ *  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.
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/errno.h>
+#include <asm/system.h>
+#include <linux/sched.h>
+#include <linux/poll.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+#include <linux/devfs_fs_kernel.h>
+#include <linux/ipmi.h>
+#include <asm/semaphore.h>
+#include <linux/init.h>
+
+#define IPMI_DEVINTF_VERSION "v33"
+
+struct ipmi_file_private
+{
+	ipmi_user_t          user;
+	spinlock_t           recv_msg_lock;
+	struct list_head     recv_msgs;
+	struct file          *file;
+	struct fasync_struct *fasync_queue;
+	wait_queue_head_t    wait;
+	struct semaphore     recv_sem;
+	int                  default_retries;
+	unsigned int         default_retry_time_ms;
+};
+
+static void file_receive_handler(struct ipmi_recv_msg *msg,
+				 void                 *handler_data)
+{
+	struct ipmi_file_private *priv = handler_data;
+	int                      was_empty;
+	unsigned long            flags;
+
+	spin_lock_irqsave(&(priv->recv_msg_lock), flags);
+
+	was_empty = list_empty(&(priv->recv_msgs));
+	list_add_tail(&(msg->link), &(priv->recv_msgs));
+
+	if (was_empty) {
+		wake_up_interruptible(&priv->wait);
+		kill_fasync(&priv->fasync_queue, SIGIO, POLL_IN);
+	}
+
+	spin_unlock_irqrestore(&(priv->recv_msg_lock), flags);
+}
+
+static unsigned int ipmi_poll(struct file *file, poll_table *wait)
+{
+	struct ipmi_file_private *priv = file->private_data;
+	unsigned int             mask = 0;
+	unsigned long            flags;
+
+	poll_wait(file, &priv->wait, wait);
+
+	spin_lock_irqsave(&priv->recv_msg_lock, flags);
+
+	if (! list_empty(&(priv->recv_msgs)))
+		mask |= (POLLIN | POLLRDNORM);
+
+	spin_unlock_irqrestore(&priv->recv_msg_lock, flags);
+
+	return mask;
+}
+
+static int ipmi_fasync(int fd, struct file *file, int on)
+{
+	struct ipmi_file_private *priv = file->private_data;
+	int                      result;
+
+	result = fasync_helper(fd, file, on, &priv->fasync_queue);
+
+	return (result);
+}
+
+static struct ipmi_user_hndl ipmi_hndlrs =
+{
+	.ipmi_recv_hndl	= file_receive_handler,
+};
+
+static int ipmi_open(struct inode *inode, struct file *file)
+{
+	int                      if_num = iminor(inode);
+	int                      rv;
+	struct ipmi_file_private *priv;
+
+
+	priv = kmalloc(sizeof(*priv), GFP_KERNEL);
+	if (!priv)
+		return -ENOMEM;
+
+	priv->file = file;
+
+	rv = ipmi_create_user(if_num,
+			      &ipmi_hndlrs,
+			      priv,
+			      &(priv->user));
+	if (rv) {
+		kfree(priv);
+		return rv;
+	}
+
+	file->private_data = priv;
+
+	spin_lock_init(&(priv->recv_msg_lock));
+	INIT_LIST_HEAD(&(priv->recv_msgs));
+	init_waitqueue_head(&priv->wait);
+	priv->fasync_queue = NULL;
+	sema_init(&(priv->recv_sem), 1);
+
+	/* Use the low-level defaults. */
+	priv->default_retries = -1;
+	priv->default_retry_time_ms = 0;
+
+	return 0;
+}
+
+static int ipmi_release(struct inode *inode, struct file *file)
+{
+	struct ipmi_file_private *priv = file->private_data;
+	int                      rv;
+
+	rv = ipmi_destroy_user(priv->user);
+	if (rv)
+		return rv;
+
+	ipmi_fasync (-1, file, 0);
+
+	/* FIXME - free the messages in the list. */
+	kfree(priv);
+
+	return 0;
+}
+
+static int handle_send_req(ipmi_user_t     user,
+			   struct ipmi_req *req,
+			   int             retries,
+			   unsigned int    retry_time_ms)
+{
+	int              rv;
+	struct ipmi_addr addr;
+	struct kernel_ipmi_msg msg;
+
+	if (req->addr_len > sizeof(struct ipmi_addr))
+		return -EINVAL;
+
+	if (copy_from_user(&addr, req->addr, req->addr_len))
+		return -EFAULT;
+
+	msg.netfn = req->msg.netfn;
+	msg.cmd = req->msg.cmd;
+	msg.data_len = req->msg.data_len;
+	msg.data = kmalloc(IPMI_MAX_MSG_LENGTH, GFP_KERNEL);
+	if (!msg.data)
+		return -ENOMEM;
+
+	/* From here out we cannot return, we must jump to "out" for
+	   error exits to free msgdata. */
+
+	rv = ipmi_validate_addr(&addr, req->addr_len);
+	if (rv)
+		goto out;
+
+	if (req->msg.data != NULL) {
+		if (req->msg.data_len > IPMI_MAX_MSG_LENGTH) {
+			rv = -EMSGSIZE;
+			goto out;
+		}
+
+		if (copy_from_user(msg.data,
+				   req->msg.data,
+				   req->msg.data_len))
+		{
+			rv = -EFAULT;
+			goto out;
+		}
+	} else {
+		msg.data_len = 0;
+	}
+
+	rv = ipmi_request_settime(user,
+				  &addr,
+				  req->msgid,
+				  &msg,
+				  NULL,
+				  0,
+				  retries,
+				  retry_time_ms);
+ out:
+	kfree(msg.data);
+	return rv;
+}
+
+static int ipmi_ioctl(struct inode  *inode,
+		      struct file   *file,
+		      unsigned int  cmd,
+		      unsigned long data)
+{
+	int                      rv = -EINVAL;
+	struct ipmi_file_private *priv = file->private_data;
+	void __user *arg = (void __user *)data;
+
+	switch (cmd) 
+	{
+	case IPMICTL_SEND_COMMAND:
+	{
+		struct ipmi_req req;
+
+		if (copy_from_user(&req, arg, sizeof(req))) {
+			rv = -EFAULT;
+			break;
+		}
+
+		rv = handle_send_req(priv->user,
+				     &req,
+				     priv->default_retries,
+				     priv->default_retry_time_ms);
+		break;
+	}
+
+	case IPMICTL_SEND_COMMAND_SETTIME:
+	{
+		struct ipmi_req_settime req;
+
+		if (copy_from_user(&req, arg, sizeof(req))) {
+			rv = -EFAULT;
+			break;
+		}
+
+		rv = handle_send_req(priv->user,
+				     &req.req,
+				     req.retries,
+				     req.retry_time_ms);
+		break;
+	}
+
+	case IPMICTL_RECEIVE_MSG:
+	case IPMICTL_RECEIVE_MSG_TRUNC:
+	{
+		struct ipmi_recv      rsp;
+		int              addr_len;
+		struct list_head *entry;
+		struct ipmi_recv_msg  *msg;
+		unsigned long    flags;
+		
+
+		rv = 0;
+		if (copy_from_user(&rsp, arg, sizeof(rsp))) {
+			rv = -EFAULT;
+			break;
+		}
+
+		/* We claim a semaphore because we don't want two
+                   users getting something from the queue at a time.
+                   Since we have to release the spinlock before we can
+                   copy the data to the user, it's possible another
+                   user will grab something from the queue, too.  Then
+                   the messages might get out of order if something
+                   fails and the message gets put back onto the
+                   queue.  This semaphore prevents that problem. */
+		down(&(priv->recv_sem));
+
+		/* Grab the message off the list. */
+		spin_lock_irqsave(&(priv->recv_msg_lock), flags);
+		if (list_empty(&(priv->recv_msgs))) {
+			spin_unlock_irqrestore(&(priv->recv_msg_lock), flags);
+			rv = -EAGAIN;
+			goto recv_err;
+		}
+		entry = priv->recv_msgs.next;
+		msg = list_entry(entry, struct ipmi_recv_msg, link);
+		list_del(entry);
+		spin_unlock_irqrestore(&(priv->recv_msg_lock), flags);
+
+		addr_len = ipmi_addr_length(msg->addr.addr_type);
+		if (rsp.addr_len < addr_len)
+		{
+			rv = -EINVAL;
+			goto recv_putback_on_err;
+		}
+
+		if (copy_to_user(rsp.addr, &(msg->addr), addr_len)) {
+			rv = -EFAULT;
+			goto recv_putback_on_err;
+		}
+		rsp.addr_len = addr_len;
+
+		rsp.recv_type = msg->recv_type;
+		rsp.msgid = msg->msgid;
+		rsp.msg.netfn = msg->msg.netfn;
+		rsp.msg.cmd = msg->msg.cmd;
+
+		if (msg->msg.data_len > 0) {
+			if (rsp.msg.data_len < msg->msg.data_len) {
+				rv = -EMSGSIZE;
+				if (cmd == IPMICTL_RECEIVE_MSG_TRUNC) {
+					msg->msg.data_len = rsp.msg.data_len;
+				} else {
+					goto recv_putback_on_err;
+				}
+			}
+
+			if (copy_to_user(rsp.msg.data,
+					 msg->msg.data,
+					 msg->msg.data_len))
+			{
+				rv = -EFAULT;
+				goto recv_putback_on_err;
+			}
+			rsp.msg.data_len = msg->msg.data_len;
+		} else {
+			rsp.msg.data_len = 0;
+		}
+
+		if (copy_to_user(arg, &rsp, sizeof(rsp))) {
+			rv = -EFAULT;
+			goto recv_putback_on_err;
+		}
+
+		up(&(priv->recv_sem));
+		ipmi_free_recv_msg(msg);
+		break;
+
+	recv_putback_on_err:
+		/* If we got an error, put the message back onto
+		   the head of the queue. */
+		spin_lock_irqsave(&(priv->recv_msg_lock), flags);
+		list_add(entry, &(priv->recv_msgs));
+		spin_unlock_irqrestore(&(priv->recv_msg_lock), flags);
+		up(&(priv->recv_sem));
+		break;
+
+	recv_err:
+		up(&(priv->recv_sem));
+		break;
+	}
+
+	case IPMICTL_REGISTER_FOR_CMD:
+	{
+		struct ipmi_cmdspec val;
+
+		if (copy_from_user(&val, arg, sizeof(val))) {
+			rv = -EFAULT;
+			break;
+		}
+
+		rv = ipmi_register_for_cmd(priv->user, val.netfn, val.cmd);
+		break;
+	}
+
+	case IPMICTL_UNREGISTER_FOR_CMD:
+	{
+		struct ipmi_cmdspec   val;
+
+		if (copy_from_user(&val, arg, sizeof(val))) {
+			rv = -EFAULT;
+			break;
+		}
+
+		rv = ipmi_unregister_for_cmd(priv->user, val.netfn, val.cmd);
+		break;
+	}
+
+	case IPMICTL_SET_GETS_EVENTS_CMD:
+	{
+		int val;
+
+		if (copy_from_user(&val, arg, sizeof(val))) {
+			rv = -EFAULT;
+			break;
+		}
+
+		rv = ipmi_set_gets_events(priv->user, val);
+		break;
+	}
+
+	case IPMICTL_SET_MY_ADDRESS_CMD:
+	{
+		unsigned int val;
+
+		if (copy_from_user(&val, arg, sizeof(val))) {
+			rv = -EFAULT;
+			break;
+		}
+
+		ipmi_set_my_address(priv->user, val);
+		rv = 0;
+		break;
+	}
+
+	case IPMICTL_GET_MY_ADDRESS_CMD:
+	{
+		unsigned int val;
+
+		val = ipmi_get_my_address(priv->user);
+
+		if (copy_to_user(arg, &val, sizeof(val))) {
+			rv = -EFAULT;
+			break;
+		}
+		rv = 0;
+		break;
+	}
+
+	case IPMICTL_SET_MY_LUN_CMD:
+	{
+		unsigned int val;
+
+		if (copy_from_user(&val, arg, sizeof(val))) {
+			rv = -EFAULT;
+			break;
+		}
+
+		ipmi_set_my_LUN(priv->user, val);
+		rv = 0;
+		break;
+	}
+
+	case IPMICTL_GET_MY_LUN_CMD:
+	{
+		unsigned int val;
+
+		val = ipmi_get_my_LUN(priv->user);
+
+		if (copy_to_user(arg, &val, sizeof(val))) {
+			rv = -EFAULT;
+			break;
+		}
+		rv = 0;
+		break;
+	}
+	case IPMICTL_SET_TIMING_PARMS_CMD:
+	{
+		struct ipmi_timing_parms parms;
+
+		if (copy_from_user(&parms, arg, sizeof(parms))) {
+			rv = -EFAULT;
+			break;
+		}
+
+		priv->default_retries = parms.retries;
+		priv->default_retry_time_ms = parms.retry_time_ms;
+		rv = 0;
+		break;
+	}
+
+	case IPMICTL_GET_TIMING_PARMS_CMD:
+	{
+		struct ipmi_timing_parms parms;
+
+		parms.retries = priv->default_retries;
+		parms.retry_time_ms = priv->default_retry_time_ms;
+
+		if (copy_to_user(arg, &parms, sizeof(parms))) {
+			rv = -EFAULT;
+			break;
+		}
+
+		rv = 0;
+		break;
+	}
+	}
+  
+	return rv;
+}
+
+
+static struct file_operations ipmi_fops = {
+	.owner		= THIS_MODULE,
+	.ioctl		= ipmi_ioctl,
+	.open		= ipmi_open,
+	.release	= ipmi_release,
+	.fasync		= ipmi_fasync,
+	.poll		= ipmi_poll,
+};
+
+#define DEVICE_NAME     "ipmidev"
+
+static int ipmi_major = 0;
+module_param(ipmi_major, int, 0);
+MODULE_PARM_DESC(ipmi_major, "Sets the major number of the IPMI device.  By"
+		 " default, or if you set it to zero, it will choose the next"
+		 " available device.  Setting it to -1 will disable the"
+		 " interface.  Other values will set the major device number"
+		 " to that value.");
+
+static void ipmi_new_smi(int if_num)
+{
+	devfs_mk_cdev(MKDEV(ipmi_major, if_num),
+		      S_IFCHR | S_IRUSR | S_IWUSR,
+		      "ipmidev/%d", if_num);
+}
+
+static void ipmi_smi_gone(int if_num)
+{
+	devfs_remove("ipmidev/%d", if_num);
+}
+
+static struct ipmi_smi_watcher smi_watcher =
+{
+	.owner    = THIS_MODULE,
+	.new_smi  = ipmi_new_smi,
+	.smi_gone = ipmi_smi_gone,
+};
+
+static __init int init_ipmi_devintf(void)
+{
+	int rv;
+
+	if (ipmi_major < 0)
+		return -EINVAL;
+
+	printk(KERN_INFO "ipmi device interface version "
+	       IPMI_DEVINTF_VERSION "\n");
+
+	rv = register_chrdev(ipmi_major, DEVICE_NAME, &ipmi_fops);
+	if (rv < 0) {
+		printk(KERN_ERR "ipmi: can't get major %d\n", ipmi_major);
+		return rv;
+	}
+
+	if (ipmi_major == 0) {
+		ipmi_major = rv;
+	}
+
+	devfs_mk_dir(DEVICE_NAME);
+
+	rv = ipmi_smi_watcher_register(&smi_watcher);
+	if (rv) {
+		unregister_chrdev(ipmi_major, DEVICE_NAME);
+		printk(KERN_WARNING "ipmi: can't register smi watcher\n");
+		return rv;
+	}
+
+	return 0;
+}
+module_init(init_ipmi_devintf);
+
+static __exit void cleanup_ipmi(void)
+{
+	ipmi_smi_watcher_unregister(&smi_watcher);
+	devfs_remove(DEVICE_NAME);
+	unregister_chrdev(ipmi_major, DEVICE_NAME);
+}
+module_exit(cleanup_ipmi);
+
+MODULE_LICENSE("GPL");
diff --git a/drivers/char/ipmi/ipmi_kcs_sm.c b/drivers/char/ipmi/ipmi_kcs_sm.c
new file mode 100644
index 000000000000..48cce24329be
--- /dev/null
+++ b/drivers/char/ipmi/ipmi_kcs_sm.c
@@ -0,0 +1,500 @@
+/*
+ * ipmi_kcs_sm.c
+ *
+ * State machine for handling IPMI KCS interfaces.
+ *
+ * Author: MontaVista Software, Inc.
+ *         Corey Minyard <minyard@mvista.com>
+ *         source@mvista.com
+ *
+ * Copyright 2002 MontaVista Software Inc.
+ *
+ *  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 state machine is taken from the state machine in the IPMI spec,
+ * pretty much verbatim.  If you have questions about the states, see
+ * that document.
+ */
+
+#include <linux/kernel.h> /* For printk. */
+#include <linux/string.h>
+#include <linux/ipmi_msgdefs.h>		/* for completion codes */
+#include "ipmi_si_sm.h"
+
+#define IPMI_KCS_VERSION "v33"
+
+/* Set this if you want a printout of why the state machine was hosed
+   when it gets hosed. */
+#define DEBUG_HOSED_REASON
+
+/* Print the state machine state on entry every time. */
+#undef DEBUG_STATE
+
+/* The states the KCS driver may be in. */
+enum kcs_states {
+	KCS_IDLE,		/* The KCS interface is currently
+                                   doing nothing. */
+	KCS_START_OP,		/* We are starting an operation.  The
+				   data is in the output buffer, but
+				   nothing has been done to the
+				   interface yet.  This was added to
+				   the state machine in the spec to
+				   wait for the initial IBF. */
+	KCS_WAIT_WRITE_START,	/* We have written a write cmd to the
+				   interface. */
+	KCS_WAIT_WRITE,		/* We are writing bytes to the
+                                   interface. */
+	KCS_WAIT_WRITE_END,	/* We have written the write end cmd
+                                   to the interface, and still need to
+                                   write the last byte. */
+	KCS_WAIT_READ,		/* We are waiting to read data from
+				   the interface. */
+	KCS_ERROR0,		/* State to transition to the error
+				   handler, this was added to the
+				   state machine in the spec to be
+				   sure IBF was there. */
+	KCS_ERROR1,		/* First stage error handler, wait for
+                                   the interface to respond. */
+	KCS_ERROR2,		/* The abort cmd has been written,
+				   wait for the interface to
+				   respond. */
+	KCS_ERROR3,		/* We wrote some data to the
+				   interface, wait for it to switch to
+				   read mode. */
+	KCS_HOSED		/* The hardware failed to follow the
+				   state machine. */
+};
+
+#define MAX_KCS_READ_SIZE 80
+#define MAX_KCS_WRITE_SIZE 80
+
+/* Timeouts in microseconds. */
+#define IBF_RETRY_TIMEOUT 1000000
+#define OBF_RETRY_TIMEOUT 1000000
+#define MAX_ERROR_RETRIES 10
+
+struct si_sm_data
+{
+	enum kcs_states  state;
+	struct si_sm_io *io;
+	unsigned char    write_data[MAX_KCS_WRITE_SIZE];
+	int              write_pos;
+	int              write_count;
+	int              orig_write_count;
+	unsigned char    read_data[MAX_KCS_READ_SIZE];
+	int              read_pos;
+	int	         truncated;
+
+	unsigned int  error_retries;
+	long          ibf_timeout;
+	long          obf_timeout;
+};
+
+static unsigned int init_kcs_data(struct si_sm_data *kcs,
+				  struct si_sm_io *io)
+{
+	kcs->state = KCS_IDLE;
+	kcs->io = io;
+	kcs->write_pos = 0;
+	kcs->write_count = 0;
+	kcs->orig_write_count = 0;
+	kcs->read_pos = 0;
+	kcs->error_retries = 0;
+	kcs->truncated = 0;
+	kcs->ibf_timeout = IBF_RETRY_TIMEOUT;
+	kcs->obf_timeout = OBF_RETRY_TIMEOUT;
+
+	/* Reserve 2 I/O bytes. */
+	return 2;
+}
+
+static inline unsigned char read_status(struct si_sm_data *kcs)
+{
+	return kcs->io->inputb(kcs->io, 1);
+}
+
+static inline unsigned char read_data(struct si_sm_data *kcs)
+{
+	return kcs->io->inputb(kcs->io, 0);
+}
+
+static inline void write_cmd(struct si_sm_data *kcs, unsigned char data)
+{
+	kcs->io->outputb(kcs->io, 1, data);
+}
+
+static inline void write_data(struct si_sm_data *kcs, unsigned char data)
+{
+	kcs->io->outputb(kcs->io, 0, data);
+}
+
+/* Control codes. */
+#define KCS_GET_STATUS_ABORT	0x60
+#define KCS_WRITE_START		0x61
+#define KCS_WRITE_END		0x62
+#define KCS_READ_BYTE		0x68
+
+/* Status bits. */
+#define GET_STATUS_STATE(status) (((status) >> 6) & 0x03)
+#define KCS_IDLE_STATE	0
+#define KCS_READ_STATE	1
+#define KCS_WRITE_STATE	2
+#define KCS_ERROR_STATE	3
+#define GET_STATUS_ATN(status) ((status) & 0x04)
+#define GET_STATUS_IBF(status) ((status) & 0x02)
+#define GET_STATUS_OBF(status) ((status) & 0x01)
+
+
+static inline void write_next_byte(struct si_sm_data *kcs)
+{
+	write_data(kcs, kcs->write_data[kcs->write_pos]);
+	(kcs->write_pos)++;
+	(kcs->write_count)--;
+}
+
+static inline void start_error_recovery(struct si_sm_data *kcs, char *reason)
+{
+	(kcs->error_retries)++;
+	if (kcs->error_retries > MAX_ERROR_RETRIES) {
+#ifdef DEBUG_HOSED_REASON
+		printk("ipmi_kcs_sm: kcs hosed: %s\n", reason);
+#endif
+		kcs->state = KCS_HOSED;
+	} else {
+		kcs->state = KCS_ERROR0;
+	}
+}
+
+static inline void read_next_byte(struct si_sm_data *kcs)
+{
+	if (kcs->read_pos >= MAX_KCS_READ_SIZE) {
+		/* Throw the data away and mark it truncated. */
+		read_data(kcs);
+		kcs->truncated = 1;
+	} else {
+		kcs->read_data[kcs->read_pos] = read_data(kcs);
+		(kcs->read_pos)++;
+	}
+	write_data(kcs, KCS_READ_BYTE);
+}
+
+static inline int check_ibf(struct si_sm_data *kcs, unsigned char status,
+			    long time)
+{
+	if (GET_STATUS_IBF(status)) {
+		kcs->ibf_timeout -= time;
+		if (kcs->ibf_timeout < 0) {
+			start_error_recovery(kcs, "IBF not ready in time");
+			kcs->ibf_timeout = IBF_RETRY_TIMEOUT;
+			return 1;
+		}
+		return 0;
+	}
+	kcs->ibf_timeout = IBF_RETRY_TIMEOUT;
+	return 1;
+}
+
+static inline int check_obf(struct si_sm_data *kcs, unsigned char status,
+			    long time)
+{
+	if (! GET_STATUS_OBF(status)) {
+		kcs->obf_timeout -= time;
+		if (kcs->obf_timeout < 0) {
+		    start_error_recovery(kcs, "OBF not ready in time");
+		    return 1;
+		}
+		return 0;
+	}
+	kcs->obf_timeout = OBF_RETRY_TIMEOUT;
+	return 1;
+}
+
+static void clear_obf(struct si_sm_data *kcs, unsigned char status)
+{
+	if (GET_STATUS_OBF(status))
+		read_data(kcs);
+}
+
+static void restart_kcs_transaction(struct si_sm_data *kcs)
+{
+	kcs->write_count = kcs->orig_write_count;
+	kcs->write_pos = 0;
+	kcs->read_pos = 0;
+	kcs->state = KCS_WAIT_WRITE_START;
+	kcs->ibf_timeout = IBF_RETRY_TIMEOUT;
+	kcs->obf_timeout = OBF_RETRY_TIMEOUT;
+	write_cmd(kcs, KCS_WRITE_START);
+}
+
+static int start_kcs_transaction(struct si_sm_data *kcs, unsigned char *data,
+				 unsigned int size)
+{
+	if ((size < 2) || (size > MAX_KCS_WRITE_SIZE)) {
+		return -1;
+	}
+
+	if ((kcs->state != KCS_IDLE) && (kcs->state != KCS_HOSED)) {
+		return -2;
+	}
+
+	kcs->error_retries = 0;
+	memcpy(kcs->write_data, data, size);
+	kcs->write_count = size;
+	kcs->orig_write_count = size;
+	kcs->write_pos = 0;
+	kcs->read_pos = 0;
+	kcs->state = KCS_START_OP;
+	kcs->ibf_timeout = IBF_RETRY_TIMEOUT;
+	kcs->obf_timeout = OBF_RETRY_TIMEOUT;
+	return 0;
+}
+
+static int get_kcs_result(struct si_sm_data *kcs, unsigned char *data,
+			  unsigned int length)
+{
+	if (length < kcs->read_pos) {
+		kcs->read_pos = length;
+		kcs->truncated = 1;
+	}
+
+	memcpy(data, kcs->read_data, kcs->read_pos);
+
+	if ((length >= 3) && (kcs->read_pos < 3)) {
+		/* Guarantee that we return at least 3 bytes, with an
+		   error in the third byte if it is too short. */
+		data[2] = IPMI_ERR_UNSPECIFIED;
+		kcs->read_pos = 3;
+	}
+	if (kcs->truncated) {
+		/* Report a truncated error.  We might overwrite
+		   another error, but that's too bad, the user needs
+		   to know it was truncated. */
+		data[2] = IPMI_ERR_MSG_TRUNCATED;
+		kcs->truncated = 0;
+	}
+
+	return kcs->read_pos;
+}
+
+/* This implements the state machine defined in the IPMI manual, see
+   that for details on how this works.  Divide that flowchart into
+   sections delimited by "Wait for IBF" and this will become clear. */
+static enum si_sm_result kcs_event(struct si_sm_data *kcs, long time)
+{
+	unsigned char status;
+	unsigned char state;
+
+	status = read_status(kcs);
+
+#ifdef DEBUG_STATE
+	printk("  State = %d, %x\n", kcs->state, status);
+#endif
+	/* All states wait for ibf, so just do it here. */
+	if (!check_ibf(kcs, status, time))
+		return SI_SM_CALL_WITH_DELAY;
+
+	/* Just about everything looks at the KCS state, so grab that, too. */
+	state = GET_STATUS_STATE(status);
+
+	switch (kcs->state) {
+	case KCS_IDLE:
+		/* If there's and interrupt source, turn it off. */
+		clear_obf(kcs, status);
+
+		if (GET_STATUS_ATN(status))
+			return SI_SM_ATTN;
+		else
+			return SI_SM_IDLE;
+
+	case KCS_START_OP:
+		if (state != KCS_IDLE) {
+			start_error_recovery(kcs,
+					     "State machine not idle at start");
+			break;
+		}
+
+		clear_obf(kcs, status);
+		write_cmd(kcs, KCS_WRITE_START);
+		kcs->state = KCS_WAIT_WRITE_START;
+		break;
+
+	case KCS_WAIT_WRITE_START:
+		if (state != KCS_WRITE_STATE) {
+			start_error_recovery(
+				kcs,
+				"Not in write state at write start");
+			break;
+		}
+		read_data(kcs);
+		if (kcs->write_count == 1) {
+			write_cmd(kcs, KCS_WRITE_END);
+			kcs->state = KCS_WAIT_WRITE_END;
+		} else {
+			write_next_byte(kcs);
+			kcs->state = KCS_WAIT_WRITE;
+		}
+		break;
+
+	case KCS_WAIT_WRITE:
+		if (state != KCS_WRITE_STATE) {
+			start_error_recovery(kcs,
+					     "Not in write state for write");
+			break;
+		}
+		clear_obf(kcs, status);
+		if (kcs->write_count == 1) {
+			write_cmd(kcs, KCS_WRITE_END);
+			kcs->state = KCS_WAIT_WRITE_END;
+		} else {
+			write_next_byte(kcs);
+		}
+		break;
+		
+	case KCS_WAIT_WRITE_END:
+		if (state != KCS_WRITE_STATE) {
+			start_error_recovery(kcs,
+					     "Not in write state for write end");
+			break;
+		}
+		clear_obf(kcs, status);
+		write_next_byte(kcs);
+		kcs->state = KCS_WAIT_READ;
+		break;
+
+	case KCS_WAIT_READ:
+		if ((state != KCS_READ_STATE) && (state != KCS_IDLE_STATE)) {
+			start_error_recovery(
+				kcs,
+				"Not in read or idle in read state");
+			break;
+		}
+
+		if (state == KCS_READ_STATE) {
+			if (! check_obf(kcs, status, time))
+				return SI_SM_CALL_WITH_DELAY;
+			read_next_byte(kcs);
+		} else {
+			/* We don't implement this exactly like the state
+			   machine in the spec.  Some broken hardware
+			   does not write the final dummy byte to the
+			   read register.  Thus obf will never go high
+			   here.  We just go straight to idle, and we
+			   handle clearing out obf in idle state if it
+			   happens to come in. */
+			clear_obf(kcs, status);
+			kcs->orig_write_count = 0;
+			kcs->state = KCS_IDLE;
+			return SI_SM_TRANSACTION_COMPLETE;
+		}
+		break;
+
+	case KCS_ERROR0:
+		clear_obf(kcs, status);
+		write_cmd(kcs, KCS_GET_STATUS_ABORT);
+		kcs->state = KCS_ERROR1;
+		break;
+
+	case KCS_ERROR1:
+		clear_obf(kcs, status);
+		write_data(kcs, 0);
+		kcs->state = KCS_ERROR2;
+		break;
+		
+	case KCS_ERROR2:
+		if (state != KCS_READ_STATE) {
+			start_error_recovery(kcs,
+					     "Not in read state for error2");
+			break;
+		}
+		if (! check_obf(kcs, status, time))
+			return SI_SM_CALL_WITH_DELAY;
+
+		clear_obf(kcs, status);
+		write_data(kcs, KCS_READ_BYTE);
+		kcs->state = KCS_ERROR3;
+		break;
+		
+	case KCS_ERROR3:
+		if (state != KCS_IDLE_STATE) {
+			start_error_recovery(kcs,
+					     "Not in idle state for error3");
+			break;
+		}
+
+		if (! check_obf(kcs, status, time))
+			return SI_SM_CALL_WITH_DELAY;
+
+		clear_obf(kcs, status);
+		if (kcs->orig_write_count) {
+			restart_kcs_transaction(kcs);
+		} else {
+			kcs->state = KCS_IDLE;
+			return SI_SM_TRANSACTION_COMPLETE;
+		}
+		break;
+			
+	case KCS_HOSED:
+		break;
+	}
+
+	if (kcs->state == KCS_HOSED) {
+		init_kcs_data(kcs, kcs->io);
+		return SI_SM_HOSED;
+	}
+
+	return SI_SM_CALL_WITHOUT_DELAY;
+}
+
+static int kcs_size(void)
+{
+	return sizeof(struct si_sm_data);
+}
+
+static int kcs_detect(struct si_sm_data *kcs)
+{
+	/* It's impossible for the KCS status register to be all 1's,
+	   (assuming a properly functioning, self-initialized BMC)
+	   but that's what you get from reading a bogus address, so we
+	   test that first. */
+	if (read_status(kcs) == 0xff)
+		return 1;
+
+	return 0;
+}
+
+static void kcs_cleanup(struct si_sm_data *kcs)
+{
+}
+
+struct si_sm_handlers kcs_smi_handlers =
+{
+	.version           = IPMI_KCS_VERSION,
+	.init_data         = init_kcs_data,
+	.start_transaction = start_kcs_transaction,
+	.get_result        = get_kcs_result,
+	.event             = kcs_event,
+	.detect            = kcs_detect,
+	.cleanup           = kcs_cleanup,
+	.size              = kcs_size,
+};
diff --git a/drivers/char/ipmi/ipmi_msghandler.c b/drivers/char/ipmi/ipmi_msghandler.c
new file mode 100644
index 000000000000..a6606a1aced7
--- /dev/null
+++ b/drivers/char/ipmi/ipmi_msghandler.c
@@ -0,0 +1,3174 @@
+/*
+ * ipmi_msghandler.c
+ *
+ * Incoming and outgoing message routing for an IPMI interface.
+ *
+ * Author: MontaVista Software, Inc.
+ *         Corey Minyard <minyard@mvista.com>
+ *         source@mvista.com
+ *
+ * Copyright 2002 MontaVista Software Inc.
+ *
+ *  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.
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/errno.h>
+#include <asm/system.h>
+#include <linux/sched.h>
+#include <linux/poll.h>
+#include <linux/spinlock.h>
+#include <linux/rwsem.h>
+#include <linux/slab.h>
+#include <linux/ipmi.h>
+#include <linux/ipmi_smi.h>
+#include <linux/notifier.h>
+#include <linux/init.h>
+#include <linux/proc_fs.h>
+
+#define PFX "IPMI message handler: "
+#define IPMI_MSGHANDLER_VERSION "v33"
+
+static struct ipmi_recv_msg *ipmi_alloc_recv_msg(void);
+static int ipmi_init_msghandler(void);
+
+static int initialized = 0;
+
+static struct proc_dir_entry *proc_ipmi_root = NULL;
+
+#define MAX_EVENTS_IN_QUEUE	25
+
+/* Don't let a message sit in a queue forever, always time it with at lest
+   the max message timer.  This is in milliseconds. */
+#define MAX_MSG_TIMEOUT		60000
+
+struct ipmi_user
+{
+	struct list_head link;
+
+	/* The upper layer that handles receive messages. */
+	struct ipmi_user_hndl *handler;
+	void             *handler_data;
+
+	/* The interface this user is bound to. */
+	ipmi_smi_t intf;
+
+	/* Does this interface receive IPMI events? */
+	int gets_events;
+};
+
+struct cmd_rcvr
+{
+	struct list_head link;
+
+	ipmi_user_t   user;
+	unsigned char netfn;
+	unsigned char cmd;
+};
+
+struct seq_table
+{
+	unsigned int         inuse : 1;
+	unsigned int         broadcast : 1;
+
+	unsigned long        timeout;
+	unsigned long        orig_timeout;
+	unsigned int         retries_left;
+
+	/* To verify on an incoming send message response that this is
+           the message that the response is for, we keep a sequence id
+           and increment it every time we send a message. */
+	long                 seqid;
+
+	/* This is held so we can properly respond to the message on a
+           timeout, and it is used to hold the temporary data for
+           retransmission, too. */
+	struct ipmi_recv_msg *recv_msg;
+};
+
+/* Store the information in a msgid (long) to allow us to find a
+   sequence table entry from the msgid. */
+#define STORE_SEQ_IN_MSGID(seq, seqid) (((seq&0xff)<<26) | (seqid&0x3ffffff))
+
+#define GET_SEQ_FROM_MSGID(msgid, seq, seqid) \
+	do {								\
+		seq = ((msgid >> 26) & 0x3f);				\
+		seqid = (msgid & 0x3fffff);				\
+        } while(0)
+
+#define NEXT_SEQID(seqid) (((seqid) + 1) & 0x3fffff)
+
+struct ipmi_channel
+{
+	unsigned char medium;
+	unsigned char protocol;
+};
+
+struct ipmi_proc_entry
+{
+	char                   *name;
+	struct ipmi_proc_entry *next;
+};
+
+#define IPMI_IPMB_NUM_SEQ	64
+#define IPMI_MAX_CHANNELS       8
+struct ipmi_smi
+{
+	/* What interface number are we? */
+	int intf_num;
+
+	/* The list of upper layers that are using me.  We read-lock
+           this when delivering messages to the upper layer to keep
+           the user from going away while we are processing the
+           message.  This means that you cannot add or delete a user
+           from the receive callback. */
+	rwlock_t                users_lock;
+	struct list_head        users;
+
+	/* Used for wake ups at startup. */
+	wait_queue_head_t waitq;
+
+	/* The IPMI version of the BMC on the other end. */
+	unsigned char       version_major;
+	unsigned char       version_minor;
+
+	/* This is the lower-layer's sender routine. */
+	struct ipmi_smi_handlers *handlers;
+	void                     *send_info;
+
+	/* A list of proc entries for this interface.  This does not
+	   need a lock, only one thread creates it and only one thread
+	   destroys it. */
+	struct ipmi_proc_entry *proc_entries;
+
+	/* A table of sequence numbers for this interface.  We use the
+           sequence numbers for IPMB messages that go out of the
+           interface to match them up with their responses.  A routine
+           is called periodically to time the items in this list. */
+	spinlock_t       seq_lock;
+	struct seq_table seq_table[IPMI_IPMB_NUM_SEQ];
+	int curr_seq;
+
+	/* Messages that were delayed for some reason (out of memory,
+           for instance), will go in here to be processed later in a
+           periodic timer interrupt. */
+	spinlock_t       waiting_msgs_lock;
+	struct list_head waiting_msgs;
+
+	/* The list of command receivers that are registered for commands
+	   on this interface. */
+	rwlock_t	 cmd_rcvr_lock;
+	struct list_head cmd_rcvrs;
+
+	/* Events that were queues because no one was there to receive
+           them. */
+	spinlock_t       events_lock; /* For dealing with event stuff. */
+	struct list_head waiting_events;
+	unsigned int     waiting_events_count; /* How many events in queue? */
+
+	/* This will be non-null if someone registers to receive all
+	   IPMI commands (this is for interface emulation).  There
+	   may not be any things in the cmd_rcvrs list above when
+	   this is registered. */
+	ipmi_user_t all_cmd_rcvr;
+
+	/* My slave address.  This is initialized to IPMI_BMC_SLAVE_ADDR,
+	   but may be changed by the user. */
+	unsigned char my_address;
+
+	/* My LUN.  This should generally stay the SMS LUN, but just in
+	   case... */
+	unsigned char my_lun;
+
+	/* The event receiver for my BMC, only really used at panic
+	   shutdown as a place to store this. */
+	unsigned char event_receiver;
+	unsigned char event_receiver_lun;
+	unsigned char local_sel_device;
+	unsigned char local_event_generator;
+
+	/* A cheap hack, if this is non-null and a message to an
+	   interface comes in with a NULL user, call this routine with
+	   it.  Note that the message will still be freed by the
+	   caller.  This only works on the system interface. */
+	void (*null_user_handler)(ipmi_smi_t intf, struct ipmi_smi_msg *msg);
+
+	/* When we are scanning the channels for an SMI, this will
+	   tell which channel we are scanning. */
+	int curr_channel;
+
+	/* Channel information */
+	struct ipmi_channel channels[IPMI_MAX_CHANNELS];
+
+	/* Proc FS stuff. */
+	struct proc_dir_entry *proc_dir;
+	char                  proc_dir_name[10];
+
+	spinlock_t   counter_lock; /* For making counters atomic. */
+
+	/* Commands we got that were invalid. */
+	unsigned int sent_invalid_commands;
+
+	/* Commands we sent to the MC. */
+	unsigned int sent_local_commands;
+	/* Responses from the MC that were delivered to a user. */
+	unsigned int handled_local_responses;
+	/* Responses from the MC that were not delivered to a user. */
+	unsigned int unhandled_local_responses;
+
+	/* Commands we sent out to the IPMB bus. */
+	unsigned int sent_ipmb_commands;
+	/* Commands sent on the IPMB that had errors on the SEND CMD */
+	unsigned int sent_ipmb_command_errs;
+	/* Each retransmit increments this count. */
+	unsigned int retransmitted_ipmb_commands;
+	/* When a message times out (runs out of retransmits) this is
+           incremented. */
+	unsigned int timed_out_ipmb_commands;
+
+	/* This is like above, but for broadcasts.  Broadcasts are
+           *not* included in the above count (they are expected to
+           time out). */
+	unsigned int timed_out_ipmb_broadcasts;
+
+	/* Responses I have sent to the IPMB bus. */
+	unsigned int sent_ipmb_responses;
+
+	/* The response was delivered to the user. */
+	unsigned int handled_ipmb_responses;
+	/* The response had invalid data in it. */
+	unsigned int invalid_ipmb_responses;
+	/* The response didn't have anyone waiting for it. */
+	unsigned int unhandled_ipmb_responses;
+
+	/* Commands we sent out to the IPMB bus. */
+	unsigned int sent_lan_commands;
+	/* Commands sent on the IPMB that had errors on the SEND CMD */
+	unsigned int sent_lan_command_errs;
+	/* Each retransmit increments this count. */
+	unsigned int retransmitted_lan_commands;
+	/* When a message times out (runs out of retransmits) this is
+           incremented. */
+	unsigned int timed_out_lan_commands;
+
+	/* Responses I have sent to the IPMB bus. */
+	unsigned int sent_lan_responses;
+
+	/* The response was delivered to the user. */
+	unsigned int handled_lan_responses;
+	/* The response had invalid data in it. */
+	unsigned int invalid_lan_responses;
+	/* The response didn't have anyone waiting for it. */
+	unsigned int unhandled_lan_responses;
+
+	/* The command was delivered to the user. */
+	unsigned int handled_commands;
+	/* The command had invalid data in it. */
+	unsigned int invalid_commands;
+	/* The command didn't have anyone waiting for it. */
+	unsigned int unhandled_commands;
+
+	/* Invalid data in an event. */
+	unsigned int invalid_events;
+	/* Events that were received with the proper format. */
+	unsigned int events;
+};
+
+#define MAX_IPMI_INTERFACES 4
+static ipmi_smi_t ipmi_interfaces[MAX_IPMI_INTERFACES];
+
+/* Used to keep interfaces from going away while operations are
+   operating on interfaces.  Grab read if you are not modifying the
+   interfaces, write if you are. */
+static DECLARE_RWSEM(interfaces_sem);
+
+/* Directly protects the ipmi_interfaces data structure.  This is
+   claimed in the timer interrupt. */
+static DEFINE_SPINLOCK(interfaces_lock);
+
+/* List of watchers that want to know when smi's are added and
+   deleted. */
+static struct list_head smi_watchers = LIST_HEAD_INIT(smi_watchers);
+static DECLARE_RWSEM(smi_watchers_sem);
+
+int ipmi_smi_watcher_register(struct ipmi_smi_watcher *watcher)
+{
+	int i;
+
+	down_read(&interfaces_sem);
+	down_write(&smi_watchers_sem);
+	list_add(&(watcher->link), &smi_watchers);
+	for (i=0; i<MAX_IPMI_INTERFACES; i++) {
+		if (ipmi_interfaces[i] != NULL) {
+			watcher->new_smi(i);
+		}
+	}
+	up_write(&smi_watchers_sem);
+	up_read(&interfaces_sem);
+	return 0;
+}
+
+int ipmi_smi_watcher_unregister(struct ipmi_smi_watcher *watcher)
+{
+	down_write(&smi_watchers_sem);
+	list_del(&(watcher->link));
+	up_write(&smi_watchers_sem);
+	return 0;
+}
+
+static void
+call_smi_watchers(int i)
+{
+	struct ipmi_smi_watcher *w;
+
+	down_read(&smi_watchers_sem);
+	list_for_each_entry(w, &smi_watchers, link) {
+		if (try_module_get(w->owner)) {
+			w->new_smi(i);
+			module_put(w->owner);
+		}
+	}
+	up_read(&smi_watchers_sem);
+}
+
+static int
+ipmi_addr_equal(struct ipmi_addr *addr1, struct ipmi_addr *addr2)
+{
+	if (addr1->addr_type != addr2->addr_type)
+		return 0;
+
+	if (addr1->channel != addr2->channel)
+		return 0;
+
+	if (addr1->addr_type == IPMI_SYSTEM_INTERFACE_ADDR_TYPE) {
+		struct ipmi_system_interface_addr *smi_addr1
+		    = (struct ipmi_system_interface_addr *) addr1;
+		struct ipmi_system_interface_addr *smi_addr2
+		    = (struct ipmi_system_interface_addr *) addr2;
+		return (smi_addr1->lun == smi_addr2->lun);
+	}
+
+	if ((addr1->addr_type == IPMI_IPMB_ADDR_TYPE)
+	    || (addr1->addr_type == IPMI_IPMB_BROADCAST_ADDR_TYPE))
+	{
+		struct ipmi_ipmb_addr *ipmb_addr1
+		    = (struct ipmi_ipmb_addr *) addr1;
+		struct ipmi_ipmb_addr *ipmb_addr2
+		    = (struct ipmi_ipmb_addr *) addr2;
+
+		return ((ipmb_addr1->slave_addr == ipmb_addr2->slave_addr)
+			&& (ipmb_addr1->lun == ipmb_addr2->lun));
+	}
+
+	if (addr1->addr_type == IPMI_LAN_ADDR_TYPE) {
+		struct ipmi_lan_addr *lan_addr1
+			= (struct ipmi_lan_addr *) addr1;
+		struct ipmi_lan_addr *lan_addr2
+		    = (struct ipmi_lan_addr *) addr2;
+
+		return ((lan_addr1->remote_SWID == lan_addr2->remote_SWID)
+			&& (lan_addr1->local_SWID == lan_addr2->local_SWID)
+			&& (lan_addr1->session_handle
+			    == lan_addr2->session_handle)
+			&& (lan_addr1->lun == lan_addr2->lun));
+	}
+
+	return 1;
+}
+
+int ipmi_validate_addr(struct ipmi_addr *addr, int len)
+{
+	if (len < sizeof(struct ipmi_system_interface_addr)) {
+		return -EINVAL;
+	}
+
+	if (addr->addr_type == IPMI_SYSTEM_INTERFACE_ADDR_TYPE) {
+		if (addr->channel != IPMI_BMC_CHANNEL)
+			return -EINVAL;
+		return 0;
+	}
+
+	if ((addr->channel == IPMI_BMC_CHANNEL)
+	    || (addr->channel >= IPMI_NUM_CHANNELS)
+	    || (addr->channel < 0))
+		return -EINVAL;
+
+	if ((addr->addr_type == IPMI_IPMB_ADDR_TYPE)
+	    || (addr->addr_type == IPMI_IPMB_BROADCAST_ADDR_TYPE))
+	{
+		if (len < sizeof(struct ipmi_ipmb_addr)) {
+			return -EINVAL;
+		}
+		return 0;
+	}
+
+	if (addr->addr_type == IPMI_LAN_ADDR_TYPE) {
+		if (len < sizeof(struct ipmi_lan_addr)) {
+			return -EINVAL;
+		}
+		return 0;
+	}
+
+	return -EINVAL;
+}
+
+unsigned int ipmi_addr_length(int addr_type)
+{
+	if (addr_type == IPMI_SYSTEM_INTERFACE_ADDR_TYPE)
+		return sizeof(struct ipmi_system_interface_addr);
+
+	if ((addr_type == IPMI_IPMB_ADDR_TYPE)
+	    || (addr_type == IPMI_IPMB_BROADCAST_ADDR_TYPE))
+	{
+		return sizeof(struct ipmi_ipmb_addr);
+	}
+
+	if (addr_type == IPMI_LAN_ADDR_TYPE)
+		return sizeof(struct ipmi_lan_addr);
+
+	return 0;
+}
+
+static void deliver_response(struct ipmi_recv_msg *msg)
+{
+	msg->user->handler->ipmi_recv_hndl(msg, msg->user->handler_data);
+}
+
+/* Find the next sequence number not being used and add the given
+   message with the given timeout to the sequence table.  This must be
+   called with the interface's seq_lock held. */
+static int intf_next_seq(ipmi_smi_t           intf,
+			 struct ipmi_recv_msg *recv_msg,
+			 unsigned long        timeout,
+			 int                  retries,
+			 int                  broadcast,
+			 unsigned char        *seq,
+			 long                 *seqid)
+{
+	int          rv = 0;
+	unsigned int i;
+
+	for (i=intf->curr_seq;
+	     (i+1)%IPMI_IPMB_NUM_SEQ != intf->curr_seq;
+	     i=(i+1)%IPMI_IPMB_NUM_SEQ)
+	{
+		if (! intf->seq_table[i].inuse)
+			break;
+	}
+
+	if (! intf->seq_table[i].inuse) {
+		intf->seq_table[i].recv_msg = recv_msg;
+
+		/* Start with the maximum timeout, when the send response
+		   comes in we will start the real timer. */
+		intf->seq_table[i].timeout = MAX_MSG_TIMEOUT;
+		intf->seq_table[i].orig_timeout = timeout;
+		intf->seq_table[i].retries_left = retries;
+		intf->seq_table[i].broadcast = broadcast;
+		intf->seq_table[i].inuse = 1;
+		intf->seq_table[i].seqid = NEXT_SEQID(intf->seq_table[i].seqid);
+		*seq = i;
+		*seqid = intf->seq_table[i].seqid;
+		intf->curr_seq = (i+1)%IPMI_IPMB_NUM_SEQ;
+	} else {
+		rv = -EAGAIN;
+	}
+	
+	return rv;
+}
+
+/* Return the receive message for the given sequence number and
+   release the sequence number so it can be reused.  Some other data
+   is passed in to be sure the message matches up correctly (to help
+   guard against message coming in after their timeout and the
+   sequence number being reused). */
+static int intf_find_seq(ipmi_smi_t           intf,
+			 unsigned char        seq,
+			 short                channel,
+			 unsigned char        cmd,
+			 unsigned char        netfn,
+			 struct ipmi_addr     *addr,
+			 struct ipmi_recv_msg **recv_msg)
+{
+	int           rv = -ENODEV;
+	unsigned long flags;
+
+	if (seq >= IPMI_IPMB_NUM_SEQ)
+		return -EINVAL;
+
+	spin_lock_irqsave(&(intf->seq_lock), flags);
+	if (intf->seq_table[seq].inuse) {
+		struct ipmi_recv_msg *msg = intf->seq_table[seq].recv_msg;
+
+		if ((msg->addr.channel == channel)
+		    && (msg->msg.cmd == cmd)
+		    && (msg->msg.netfn == netfn)
+		    && (ipmi_addr_equal(addr, &(msg->addr))))
+		{
+			*recv_msg = msg;
+			intf->seq_table[seq].inuse = 0;
+			rv = 0;
+		}
+	}
+	spin_unlock_irqrestore(&(intf->seq_lock), flags);
+
+	return rv;
+}
+
+
+/* Start the timer for a specific sequence table entry. */
+static int intf_start_seq_timer(ipmi_smi_t intf,
+				long       msgid)
+{
+	int           rv = -ENODEV;
+	unsigned long flags;
+	unsigned char seq;
+	unsigned long seqid;
+
+
+	GET_SEQ_FROM_MSGID(msgid, seq, seqid);
+
+	spin_lock_irqsave(&(intf->seq_lock), flags);
+	/* We do this verification because the user can be deleted
+           while a message is outstanding. */
+	if ((intf->seq_table[seq].inuse)
+	    && (intf->seq_table[seq].seqid == seqid))
+	{
+		struct seq_table *ent = &(intf->seq_table[seq]);
+		ent->timeout = ent->orig_timeout;
+		rv = 0;
+	}
+	spin_unlock_irqrestore(&(intf->seq_lock), flags);
+
+	return rv;
+}
+
+/* Got an error for the send message for a specific sequence number. */
+static int intf_err_seq(ipmi_smi_t   intf,
+			long         msgid,
+			unsigned int err)
+{
+	int                  rv = -ENODEV;
+	unsigned long        flags;
+	unsigned char        seq;
+	unsigned long        seqid;
+	struct ipmi_recv_msg *msg = NULL;
+
+
+	GET_SEQ_FROM_MSGID(msgid, seq, seqid);
+
+	spin_lock_irqsave(&(intf->seq_lock), flags);
+	/* We do this verification because the user can be deleted
+           while a message is outstanding. */
+	if ((intf->seq_table[seq].inuse)
+	    && (intf->seq_table[seq].seqid == seqid))
+	{
+		struct seq_table *ent = &(intf->seq_table[seq]);
+
+		ent->inuse = 0;
+		msg = ent->recv_msg;
+		rv = 0;
+	}
+	spin_unlock_irqrestore(&(intf->seq_lock), flags);
+
+	if (msg) {
+		msg->recv_type = IPMI_RESPONSE_RECV_TYPE;
+		msg->msg_data[0] = err;
+		msg->msg.netfn |= 1; /* Convert to a response. */
+		msg->msg.data_len = 1;
+		msg->msg.data = msg->msg_data;
+		deliver_response(msg);
+	}
+
+	return rv;
+}
+
+
+int ipmi_create_user(unsigned int          if_num,
+		     struct ipmi_user_hndl *handler,
+		     void                  *handler_data,
+		     ipmi_user_t           *user)
+{
+	unsigned long flags;
+	ipmi_user_t   new_user;
+	int           rv = 0;
+	ipmi_smi_t    intf;
+
+	/* There is no module usecount here, because it's not
+           required.  Since this can only be used by and called from
+           other modules, they will implicitly use this module, and
+           thus this can't be removed unless the other modules are
+           removed. */
+
+	if (handler == NULL)
+		return -EINVAL;
+
+	/* Make sure the driver is actually initialized, this handles
+	   problems with initialization order. */
+	if (!initialized) {
+		rv = ipmi_init_msghandler();
+		if (rv)
+			return rv;
+
+		/* The init code doesn't return an error if it was turned
+		   off, but it won't initialize.  Check that. */
+		if (!initialized)
+			return -ENODEV;
+	}
+
+	new_user = kmalloc(sizeof(*new_user), GFP_KERNEL);
+	if (! new_user)
+		return -ENOMEM;
+
+	down_read(&interfaces_sem);
+	if ((if_num > MAX_IPMI_INTERFACES) || ipmi_interfaces[if_num] == NULL)
+	{
+		rv = -EINVAL;
+		goto out_unlock;
+	}
+
+	intf = ipmi_interfaces[if_num];
+
+	new_user->handler = handler;
+	new_user->handler_data = handler_data;
+	new_user->intf = intf;
+	new_user->gets_events = 0;
+
+	if (!try_module_get(intf->handlers->owner)) {
+		rv = -ENODEV;
+		goto out_unlock;
+	}
+
+	if (intf->handlers->inc_usecount) {
+		rv = intf->handlers->inc_usecount(intf->send_info);
+		if (rv) {
+			module_put(intf->handlers->owner);
+			goto out_unlock;
+		}
+	}
+
+	write_lock_irqsave(&intf->users_lock, flags);
+	list_add_tail(&new_user->link, &intf->users);
+	write_unlock_irqrestore(&intf->users_lock, flags);
+
+ out_unlock:	
+	if (rv) {
+		kfree(new_user);
+	} else {
+		*user = new_user;
+	}
+
+	up_read(&interfaces_sem);
+	return rv;
+}
+
+static int ipmi_destroy_user_nolock(ipmi_user_t user)
+{
+	int              rv = -ENODEV;
+	ipmi_user_t      t_user;
+	struct cmd_rcvr  *rcvr, *rcvr2;
+	int              i;
+	unsigned long    flags;
+
+	/* Find the user and delete them from the list. */
+	list_for_each_entry(t_user, &(user->intf->users), link) {
+		if (t_user == user) {
+			list_del(&t_user->link);
+			rv = 0;
+			break;
+		}
+	}
+
+	if (rv) {
+		goto out_unlock;
+	}
+
+	/* Remove the user from the interfaces sequence table. */
+	spin_lock_irqsave(&(user->intf->seq_lock), flags);
+	for (i=0; i<IPMI_IPMB_NUM_SEQ; i++) {
+		if (user->intf->seq_table[i].inuse
+		    && (user->intf->seq_table[i].recv_msg->user == user))
+		{
+			user->intf->seq_table[i].inuse = 0;
+		}
+	}
+	spin_unlock_irqrestore(&(user->intf->seq_lock), flags);
+
+	/* Remove the user from the command receiver's table. */
+	write_lock_irqsave(&(user->intf->cmd_rcvr_lock), flags);
+	list_for_each_entry_safe(rcvr, rcvr2, &(user->intf->cmd_rcvrs), link) {
+		if (rcvr->user == user) {
+			list_del(&rcvr->link);
+			kfree(rcvr);
+		}
+	}
+	write_unlock_irqrestore(&(user->intf->cmd_rcvr_lock), flags);
+
+	kfree(user);
+
+ out_unlock:
+
+	return rv;
+}
+
+int ipmi_destroy_user(ipmi_user_t user)
+{
+	int           rv;
+	ipmi_smi_t    intf = user->intf;
+	unsigned long flags;
+
+	down_read(&interfaces_sem);
+	write_lock_irqsave(&intf->users_lock, flags);
+	rv = ipmi_destroy_user_nolock(user);
+	if (!rv) {
+		module_put(intf->handlers->owner);
+		if (intf->handlers->dec_usecount)
+			intf->handlers->dec_usecount(intf->send_info);
+	}
+		
+	write_unlock_irqrestore(&intf->users_lock, flags);
+	up_read(&interfaces_sem);
+	return rv;
+}
+
+void ipmi_get_version(ipmi_user_t   user,
+		      unsigned char *major,
+		      unsigned char *minor)
+{
+	*major = user->intf->version_major;
+	*minor = user->intf->version_minor;
+}
+
+void ipmi_set_my_address(ipmi_user_t   user,
+			 unsigned char address)
+{
+	user->intf->my_address = address;
+}
+
+unsigned char ipmi_get_my_address(ipmi_user_t user)
+{
+	return user->intf->my_address;
+}
+
+void ipmi_set_my_LUN(ipmi_user_t   user,
+		     unsigned char LUN)
+{
+	user->intf->my_lun = LUN & 0x3;
+}
+
+unsigned char ipmi_get_my_LUN(ipmi_user_t user)
+{
+	return user->intf->my_lun;
+}
+
+int ipmi_set_gets_events(ipmi_user_t user, int val)
+{
+	unsigned long         flags;
+	struct ipmi_recv_msg  *msg, *msg2;
+
+	read_lock(&(user->intf->users_lock));
+	spin_lock_irqsave(&(user->intf->events_lock), flags);
+	user->gets_events = val;
+
+	if (val) {
+		/* Deliver any queued events. */
+		list_for_each_entry_safe(msg, msg2, &(user->intf->waiting_events), link) {
+			list_del(&msg->link);
+			msg->user = user;
+			deliver_response(msg);
+		}
+	}
+	
+	spin_unlock_irqrestore(&(user->intf->events_lock), flags);
+	read_unlock(&(user->intf->users_lock));
+
+	return 0;
+}
+
+int ipmi_register_for_cmd(ipmi_user_t   user,
+			  unsigned char netfn,
+			  unsigned char cmd)
+{
+	struct cmd_rcvr  *cmp;
+	unsigned long    flags;
+	struct cmd_rcvr  *rcvr;
+	int              rv = 0;
+
+
+	rcvr = kmalloc(sizeof(*rcvr), GFP_KERNEL);
+	if (! rcvr)
+		return -ENOMEM;
+
+	read_lock(&(user->intf->users_lock));
+	write_lock_irqsave(&(user->intf->cmd_rcvr_lock), flags);
+	if (user->intf->all_cmd_rcvr != NULL) {
+		rv = -EBUSY;
+		goto out_unlock;
+	}
+
+	/* Make sure the command/netfn is not already registered. */
+	list_for_each_entry(cmp, &(user->intf->cmd_rcvrs), link) {
+		if ((cmp->netfn == netfn) && (cmp->cmd == cmd)) {
+			rv = -EBUSY;
+			break;
+		}
+	}
+
+	if (! rv) {
+		rcvr->cmd = cmd;
+		rcvr->netfn = netfn;
+		rcvr->user = user;
+		list_add_tail(&(rcvr->link), &(user->intf->cmd_rcvrs));
+	}
+ out_unlock:
+	write_unlock_irqrestore(&(user->intf->cmd_rcvr_lock), flags);
+	read_unlock(&(user->intf->users_lock));
+
+	if (rv)
+		kfree(rcvr);
+
+	return rv;
+}
+
+int ipmi_unregister_for_cmd(ipmi_user_t   user,
+			    unsigned char netfn,
+			    unsigned char cmd)
+{
+	unsigned long    flags;
+	struct cmd_rcvr  *rcvr;
+	int              rv = -ENOENT;
+
+	read_lock(&(user->intf->users_lock));
+	write_lock_irqsave(&(user->intf->cmd_rcvr_lock), flags);
+	/* Make sure the command/netfn is not already registered. */
+	list_for_each_entry(rcvr, &(user->intf->cmd_rcvrs), link) {
+		if ((rcvr->netfn == netfn) && (rcvr->cmd == cmd)) {
+			rv = 0;
+			list_del(&rcvr->link);
+			kfree(rcvr);
+			break;
+		}
+	}
+	write_unlock_irqrestore(&(user->intf->cmd_rcvr_lock), flags);
+	read_unlock(&(user->intf->users_lock));
+
+	return rv;
+}
+
+void ipmi_user_set_run_to_completion(ipmi_user_t user, int val)
+{
+	user->intf->handlers->set_run_to_completion(user->intf->send_info,
+						    val);
+}
+
+static unsigned char
+ipmb_checksum(unsigned char *data, int size)
+{
+	unsigned char csum = 0;
+	
+	for (; size > 0; size--, data++)
+		csum += *data;
+
+	return -csum;
+}
+
+static inline void format_ipmb_msg(struct ipmi_smi_msg   *smi_msg,
+				   struct kernel_ipmi_msg *msg,
+				   struct ipmi_ipmb_addr *ipmb_addr,
+				   long                  msgid,
+				   unsigned char         ipmb_seq,
+				   int                   broadcast,
+				   unsigned char         source_address,
+				   unsigned char         source_lun)
+{
+	int i = broadcast;
+
+	/* Format the IPMB header data. */
+	smi_msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
+	smi_msg->data[1] = IPMI_SEND_MSG_CMD;
+	smi_msg->data[2] = ipmb_addr->channel;
+	if (broadcast)
+		smi_msg->data[3] = 0;
+	smi_msg->data[i+3] = ipmb_addr->slave_addr;
+	smi_msg->data[i+4] = (msg->netfn << 2) | (ipmb_addr->lun & 0x3);
+	smi_msg->data[i+5] = ipmb_checksum(&(smi_msg->data[i+3]), 2);
+	smi_msg->data[i+6] = source_address;
+	smi_msg->data[i+7] = (ipmb_seq << 2) | source_lun;
+	smi_msg->data[i+8] = msg->cmd;
+
+	/* Now tack on the data to the message. */
+	if (msg->data_len > 0)
+		memcpy(&(smi_msg->data[i+9]), msg->data,
+		       msg->data_len);
+	smi_msg->data_size = msg->data_len + 9;
+
+	/* Now calculate the checksum and tack it on. */
+	smi_msg->data[i+smi_msg->data_size]
+		= ipmb_checksum(&(smi_msg->data[i+6]),
+				smi_msg->data_size-6);
+
+	/* Add on the checksum size and the offset from the
+	   broadcast. */
+	smi_msg->data_size += 1 + i;
+
+	smi_msg->msgid = msgid;
+}
+
+static inline void format_lan_msg(struct ipmi_smi_msg   *smi_msg,
+				  struct kernel_ipmi_msg *msg,
+				  struct ipmi_lan_addr  *lan_addr,
+				  long                  msgid,
+				  unsigned char         ipmb_seq,
+				  unsigned char         source_lun)
+{
+	/* Format the IPMB header data. */
+	smi_msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
+	smi_msg->data[1] = IPMI_SEND_MSG_CMD;
+	smi_msg->data[2] = lan_addr->channel;
+	smi_msg->data[3] = lan_addr->session_handle;
+	smi_msg->data[4] = lan_addr->remote_SWID;
+	smi_msg->data[5] = (msg->netfn << 2) | (lan_addr->lun & 0x3);
+	smi_msg->data[6] = ipmb_checksum(&(smi_msg->data[4]), 2);
+	smi_msg->data[7] = lan_addr->local_SWID;
+	smi_msg->data[8] = (ipmb_seq << 2) | source_lun;
+	smi_msg->data[9] = msg->cmd;
+
+	/* Now tack on the data to the message. */
+	if (msg->data_len > 0)
+		memcpy(&(smi_msg->data[10]), msg->data,
+		       msg->data_len);
+	smi_msg->data_size = msg->data_len + 10;
+
+	/* Now calculate the checksum and tack it on. */
+	smi_msg->data[smi_msg->data_size]
+		= ipmb_checksum(&(smi_msg->data[7]),
+				smi_msg->data_size-7);
+
+	/* Add on the checksum size and the offset from the
+	   broadcast. */
+	smi_msg->data_size += 1;
+
+	smi_msg->msgid = msgid;
+}
+
+/* Separate from ipmi_request so that the user does not have to be
+   supplied in certain circumstances (mainly at panic time).  If
+   messages are supplied, they will be freed, even if an error
+   occurs. */
+static inline int i_ipmi_request(ipmi_user_t          user,
+				 ipmi_smi_t           intf,
+				 struct ipmi_addr     *addr,
+				 long                 msgid,
+				 struct kernel_ipmi_msg *msg,
+				 void                 *user_msg_data,
+				 void                 *supplied_smi,
+				 struct ipmi_recv_msg *supplied_recv,
+				 int                  priority,
+				 unsigned char        source_address,
+				 unsigned char        source_lun,
+				 int                  retries,
+				 unsigned int         retry_time_ms)
+{
+	int                  rv = 0;
+	struct ipmi_smi_msg  *smi_msg;
+	struct ipmi_recv_msg *recv_msg;
+	unsigned long        flags;
+
+
+	if (supplied_recv) {
+		recv_msg = supplied_recv;
+	} else {
+		recv_msg = ipmi_alloc_recv_msg();
+		if (recv_msg == NULL) {
+			return -ENOMEM;
+		}
+	}
+	recv_msg->user_msg_data = user_msg_data;
+
+	if (supplied_smi) {
+		smi_msg = (struct ipmi_smi_msg *) supplied_smi;
+	} else {
+		smi_msg = ipmi_alloc_smi_msg();
+		if (smi_msg == NULL) {
+			ipmi_free_recv_msg(recv_msg);
+			return -ENOMEM;
+		}
+	}
+
+	recv_msg->user = user;
+	recv_msg->msgid = msgid;
+	/* Store the message to send in the receive message so timeout
+	   responses can get the proper response data. */
+	recv_msg->msg = *msg;
+
+	if (addr->addr_type == IPMI_SYSTEM_INTERFACE_ADDR_TYPE) {
+		struct ipmi_system_interface_addr *smi_addr;
+
+		if (msg->netfn & 1) {
+			/* Responses are not allowed to the SMI. */
+			rv = -EINVAL;
+			goto out_err;
+		}
+
+		smi_addr = (struct ipmi_system_interface_addr *) addr;
+		if (smi_addr->lun > 3) {
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->sent_invalid_commands++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			rv = -EINVAL;
+			goto out_err;
+		}
+
+		memcpy(&recv_msg->addr, smi_addr, sizeof(*smi_addr));
+
+		if ((msg->netfn == IPMI_NETFN_APP_REQUEST)
+		    && ((msg->cmd == IPMI_SEND_MSG_CMD)
+			|| (msg->cmd == IPMI_GET_MSG_CMD)
+			|| (msg->cmd == IPMI_READ_EVENT_MSG_BUFFER_CMD)))
+		{
+			/* We don't let the user do these, since we manage
+			   the sequence numbers. */
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->sent_invalid_commands++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			rv = -EINVAL;
+			goto out_err;
+		}
+
+		if ((msg->data_len + 2) > IPMI_MAX_MSG_LENGTH) {
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->sent_invalid_commands++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			rv = -EMSGSIZE;
+			goto out_err;
+		}
+
+		smi_msg->data[0] = (msg->netfn << 2) | (smi_addr->lun & 0x3);
+		smi_msg->data[1] = msg->cmd;
+		smi_msg->msgid = msgid;
+		smi_msg->user_data = recv_msg;
+		if (msg->data_len > 0)
+			memcpy(&(smi_msg->data[2]), msg->data, msg->data_len);
+		smi_msg->data_size = msg->data_len + 2;
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->sent_local_commands++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+	} else if ((addr->addr_type == IPMI_IPMB_ADDR_TYPE)
+		   || (addr->addr_type == IPMI_IPMB_BROADCAST_ADDR_TYPE))
+	{
+		struct ipmi_ipmb_addr *ipmb_addr;
+		unsigned char         ipmb_seq;
+		long                  seqid;
+		int                   broadcast = 0;
+
+		if (addr->channel > IPMI_NUM_CHANNELS) {
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->sent_invalid_commands++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			rv = -EINVAL;
+			goto out_err;
+		}
+
+		if (intf->channels[addr->channel].medium
+		    != IPMI_CHANNEL_MEDIUM_IPMB)
+		{
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->sent_invalid_commands++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			rv = -EINVAL;
+			goto out_err;
+		}
+
+		if (retries < 0) {
+		    if (addr->addr_type == IPMI_IPMB_BROADCAST_ADDR_TYPE)
+			retries = 0; /* Don't retry broadcasts. */
+		    else
+			retries = 4;
+		}
+		if (addr->addr_type == IPMI_IPMB_BROADCAST_ADDR_TYPE) {
+		    /* Broadcasts add a zero at the beginning of the
+		       message, but otherwise is the same as an IPMB
+		       address. */
+		    addr->addr_type = IPMI_IPMB_ADDR_TYPE;
+		    broadcast = 1;
+		}
+
+
+		/* Default to 1 second retries. */
+		if (retry_time_ms == 0)
+		    retry_time_ms = 1000;
+
+		/* 9 for the header and 1 for the checksum, plus
+                   possibly one for the broadcast. */
+		if ((msg->data_len + 10 + broadcast) > IPMI_MAX_MSG_LENGTH) {
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->sent_invalid_commands++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			rv = -EMSGSIZE;
+			goto out_err;
+		}
+
+		ipmb_addr = (struct ipmi_ipmb_addr *) addr;
+		if (ipmb_addr->lun > 3) {
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->sent_invalid_commands++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			rv = -EINVAL;
+			goto out_err;
+		}
+
+		memcpy(&recv_msg->addr, ipmb_addr, sizeof(*ipmb_addr));
+
+		if (recv_msg->msg.netfn & 0x1) {
+			/* It's a response, so use the user's sequence
+                           from msgid. */
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->sent_ipmb_responses++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			format_ipmb_msg(smi_msg, msg, ipmb_addr, msgid,
+					msgid, broadcast,
+					source_address, source_lun);
+
+			/* Save the receive message so we can use it
+			   to deliver the response. */
+			smi_msg->user_data = recv_msg;
+		} else {
+			/* It's a command, so get a sequence for it. */
+
+			spin_lock_irqsave(&(intf->seq_lock), flags);
+
+			spin_lock(&intf->counter_lock);
+			intf->sent_ipmb_commands++;
+			spin_unlock(&intf->counter_lock);
+
+			/* Create a sequence number with a 1 second
+                           timeout and 4 retries. */
+			rv = intf_next_seq(intf,
+					   recv_msg,
+					   retry_time_ms,
+					   retries,
+					   broadcast,
+					   &ipmb_seq,
+					   &seqid);
+			if (rv) {
+				/* We have used up all the sequence numbers,
+				   probably, so abort. */
+				spin_unlock_irqrestore(&(intf->seq_lock),
+						       flags);
+				goto out_err;
+			}
+
+			/* Store the sequence number in the message,
+                           so that when the send message response
+                           comes back we can start the timer. */
+			format_ipmb_msg(smi_msg, msg, ipmb_addr,
+					STORE_SEQ_IN_MSGID(ipmb_seq, seqid),
+					ipmb_seq, broadcast,
+					source_address, source_lun);
+
+			/* Copy the message into the recv message data, so we
+			   can retransmit it later if necessary. */
+			memcpy(recv_msg->msg_data, smi_msg->data,
+			       smi_msg->data_size);
+			recv_msg->msg.data = recv_msg->msg_data;
+			recv_msg->msg.data_len = smi_msg->data_size;
+
+			/* We don't unlock until here, because we need
+                           to copy the completed message into the
+                           recv_msg before we release the lock.
+                           Otherwise, race conditions may bite us.  I
+                           know that's pretty paranoid, but I prefer
+                           to be correct. */
+			spin_unlock_irqrestore(&(intf->seq_lock), flags);
+		}
+	} else if (addr->addr_type == IPMI_LAN_ADDR_TYPE) {
+		struct ipmi_lan_addr  *lan_addr;
+		unsigned char         ipmb_seq;
+		long                  seqid;
+
+		if (addr->channel > IPMI_NUM_CHANNELS) {
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->sent_invalid_commands++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			rv = -EINVAL;
+			goto out_err;
+		}
+
+		if ((intf->channels[addr->channel].medium
+		    != IPMI_CHANNEL_MEDIUM_8023LAN)
+		    && (intf->channels[addr->channel].medium
+			!= IPMI_CHANNEL_MEDIUM_ASYNC))
+		{
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->sent_invalid_commands++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			rv = -EINVAL;
+			goto out_err;
+		}
+
+		retries = 4;
+
+		/* Default to 1 second retries. */
+		if (retry_time_ms == 0)
+		    retry_time_ms = 1000;
+
+		/* 11 for the header and 1 for the checksum. */
+		if ((msg->data_len + 12) > IPMI_MAX_MSG_LENGTH) {
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->sent_invalid_commands++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			rv = -EMSGSIZE;
+			goto out_err;
+		}
+
+		lan_addr = (struct ipmi_lan_addr *) addr;
+		if (lan_addr->lun > 3) {
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->sent_invalid_commands++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			rv = -EINVAL;
+			goto out_err;
+		}
+
+		memcpy(&recv_msg->addr, lan_addr, sizeof(*lan_addr));
+
+		if (recv_msg->msg.netfn & 0x1) {
+			/* It's a response, so use the user's sequence
+                           from msgid. */
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->sent_lan_responses++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			format_lan_msg(smi_msg, msg, lan_addr, msgid,
+				       msgid, source_lun);
+
+			/* Save the receive message so we can use it
+			   to deliver the response. */
+			smi_msg->user_data = recv_msg;
+		} else {
+			/* It's a command, so get a sequence for it. */
+
+			spin_lock_irqsave(&(intf->seq_lock), flags);
+
+			spin_lock(&intf->counter_lock);
+			intf->sent_lan_commands++;
+			spin_unlock(&intf->counter_lock);
+
+			/* Create a sequence number with a 1 second
+                           timeout and 4 retries. */
+			rv = intf_next_seq(intf,
+					   recv_msg,
+					   retry_time_ms,
+					   retries,
+					   0,
+					   &ipmb_seq,
+					   &seqid);
+			if (rv) {
+				/* We have used up all the sequence numbers,
+				   probably, so abort. */
+				spin_unlock_irqrestore(&(intf->seq_lock),
+						       flags);
+				goto out_err;
+			}
+
+			/* Store the sequence number in the message,
+                           so that when the send message response
+                           comes back we can start the timer. */
+			format_lan_msg(smi_msg, msg, lan_addr,
+				       STORE_SEQ_IN_MSGID(ipmb_seq, seqid),
+				       ipmb_seq, source_lun);
+
+			/* Copy the message into the recv message data, so we
+			   can retransmit it later if necessary. */
+			memcpy(recv_msg->msg_data, smi_msg->data,
+			       smi_msg->data_size);
+			recv_msg->msg.data = recv_msg->msg_data;
+			recv_msg->msg.data_len = smi_msg->data_size;
+
+			/* We don't unlock until here, because we need
+                           to copy the completed message into the
+                           recv_msg before we release the lock.
+                           Otherwise, race conditions may bite us.  I
+                           know that's pretty paranoid, but I prefer
+                           to be correct. */
+			spin_unlock_irqrestore(&(intf->seq_lock), flags);
+		}
+	} else {
+	    /* Unknown address type. */
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->sent_invalid_commands++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+		rv = -EINVAL;
+		goto out_err;
+	}
+
+#ifdef DEBUG_MSGING
+	{
+		int m;
+		for (m=0; m<smi_msg->data_size; m++)
+			printk(" %2.2x", smi_msg->data[m]);
+		printk("\n");
+	}
+#endif
+	intf->handlers->sender(intf->send_info, smi_msg, priority);
+
+	return 0;
+
+ out_err:
+	ipmi_free_smi_msg(smi_msg);
+	ipmi_free_recv_msg(recv_msg);
+	return rv;
+}
+
+int ipmi_request_settime(ipmi_user_t      user,
+			 struct ipmi_addr *addr,
+			 long             msgid,
+			 struct kernel_ipmi_msg  *msg,
+			 void             *user_msg_data,
+			 int              priority,
+			 int              retries,
+			 unsigned int     retry_time_ms)
+{
+	return i_ipmi_request(user,
+			      user->intf,
+			      addr,
+			      msgid,
+			      msg,
+			      user_msg_data,
+			      NULL, NULL,
+			      priority,
+			      user->intf->my_address,
+			      user->intf->my_lun,
+			      retries,
+			      retry_time_ms);
+}
+
+int ipmi_request_supply_msgs(ipmi_user_t          user,
+			     struct ipmi_addr     *addr,
+			     long                 msgid,
+			     struct kernel_ipmi_msg *msg,
+			     void                 *user_msg_data,
+			     void                 *supplied_smi,
+			     struct ipmi_recv_msg *supplied_recv,
+			     int                  priority)
+{
+	return i_ipmi_request(user,
+			      user->intf,
+			      addr,
+			      msgid,
+			      msg,
+			      user_msg_data,
+			      supplied_smi,
+			      supplied_recv,
+			      priority,
+			      user->intf->my_address,
+			      user->intf->my_lun,
+			      -1, 0);
+}
+
+static int ipmb_file_read_proc(char *page, char **start, off_t off,
+			       int count, int *eof, void *data)
+{
+	char       *out = (char *) page;
+	ipmi_smi_t intf = data;
+
+	return sprintf(out, "%x\n", intf->my_address);
+}
+
+static int version_file_read_proc(char *page, char **start, off_t off,
+				  int count, int *eof, void *data)
+{
+	char       *out = (char *) page;
+	ipmi_smi_t intf = data;
+
+	return sprintf(out, "%d.%d\n",
+		       intf->version_major, intf->version_minor);
+}
+
+static int stat_file_read_proc(char *page, char **start, off_t off,
+			       int count, int *eof, void *data)
+{
+	char       *out = (char *) page;
+	ipmi_smi_t intf = data;
+
+	out += sprintf(out, "sent_invalid_commands:       %d\n",
+		       intf->sent_invalid_commands);
+	out += sprintf(out, "sent_local_commands:         %d\n",
+		       intf->sent_local_commands);
+	out += sprintf(out, "handled_local_responses:     %d\n",
+		       intf->handled_local_responses);
+	out += sprintf(out, "unhandled_local_responses:   %d\n",
+		       intf->unhandled_local_responses);
+	out += sprintf(out, "sent_ipmb_commands:          %d\n",
+		       intf->sent_ipmb_commands);
+	out += sprintf(out, "sent_ipmb_command_errs:      %d\n",
+		       intf->sent_ipmb_command_errs);
+	out += sprintf(out, "retransmitted_ipmb_commands: %d\n",
+		       intf->retransmitted_ipmb_commands);
+	out += sprintf(out, "timed_out_ipmb_commands:     %d\n",
+		       intf->timed_out_ipmb_commands);
+	out += sprintf(out, "timed_out_ipmb_broadcasts:   %d\n",
+		       intf->timed_out_ipmb_broadcasts);
+	out += sprintf(out, "sent_ipmb_responses:         %d\n",
+		       intf->sent_ipmb_responses);
+	out += sprintf(out, "handled_ipmb_responses:      %d\n",
+		       intf->handled_ipmb_responses);
+	out += sprintf(out, "invalid_ipmb_responses:      %d\n",
+		       intf->invalid_ipmb_responses);
+	out += sprintf(out, "unhandled_ipmb_responses:    %d\n",
+		       intf->unhandled_ipmb_responses);
+	out += sprintf(out, "sent_lan_commands:           %d\n",
+		       intf->sent_lan_commands);
+	out += sprintf(out, "sent_lan_command_errs:       %d\n",
+		       intf->sent_lan_command_errs);
+	out += sprintf(out, "retransmitted_lan_commands:  %d\n",
+		       intf->retransmitted_lan_commands);
+	out += sprintf(out, "timed_out_lan_commands:      %d\n",
+		       intf->timed_out_lan_commands);
+	out += sprintf(out, "sent_lan_responses:          %d\n",
+		       intf->sent_lan_responses);
+	out += sprintf(out, "handled_lan_responses:       %d\n",
+		       intf->handled_lan_responses);
+	out += sprintf(out, "invalid_lan_responses:       %d\n",
+		       intf->invalid_lan_responses);
+	out += sprintf(out, "unhandled_lan_responses:     %d\n",
+		       intf->unhandled_lan_responses);
+	out += sprintf(out, "handled_commands:            %d\n",
+		       intf->handled_commands);
+	out += sprintf(out, "invalid_commands:            %d\n",
+		       intf->invalid_commands);
+	out += sprintf(out, "unhandled_commands:          %d\n",
+		       intf->unhandled_commands);
+	out += sprintf(out, "invalid_events:              %d\n",
+		       intf->invalid_events);
+	out += sprintf(out, "events:                      %d\n",
+		       intf->events);
+
+	return (out - ((char *) page));
+}
+
+int ipmi_smi_add_proc_entry(ipmi_smi_t smi, char *name,
+			    read_proc_t *read_proc, write_proc_t *write_proc,
+			    void *data, struct module *owner)
+{
+	struct proc_dir_entry  *file;
+	int                    rv = 0;
+	struct ipmi_proc_entry *entry;
+
+	/* Create a list element. */
+	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
+	if (!entry)
+		return -ENOMEM;
+	entry->name = kmalloc(strlen(name)+1, GFP_KERNEL);
+	if (!entry->name) {
+		kfree(entry);
+		return -ENOMEM;
+	}
+	strcpy(entry->name, name);
+
+	file = create_proc_entry(name, 0, smi->proc_dir);
+	if (!file) {
+		kfree(entry->name);
+		kfree(entry);
+		rv = -ENOMEM;
+	} else {
+		file->nlink = 1;
+		file->data = data;
+		file->read_proc = read_proc;
+		file->write_proc = write_proc;
+		file->owner = owner;
+
+		/* Stick it on the list. */
+		entry->next = smi->proc_entries;
+		smi->proc_entries = entry;
+	}
+
+	return rv;
+}
+
+static int add_proc_entries(ipmi_smi_t smi, int num)
+{
+	int rv = 0;
+
+	sprintf(smi->proc_dir_name, "%d", num);
+	smi->proc_dir = proc_mkdir(smi->proc_dir_name, proc_ipmi_root);
+	if (!smi->proc_dir)
+		rv = -ENOMEM;
+	else {
+		smi->proc_dir->owner = THIS_MODULE;
+	}
+
+	if (rv == 0)
+		rv = ipmi_smi_add_proc_entry(smi, "stats",
+					     stat_file_read_proc, NULL,
+					     smi, THIS_MODULE);
+
+	if (rv == 0)
+		rv = ipmi_smi_add_proc_entry(smi, "ipmb",
+					     ipmb_file_read_proc, NULL,
+					     smi, THIS_MODULE);
+
+	if (rv == 0)
+		rv = ipmi_smi_add_proc_entry(smi, "version",
+					     version_file_read_proc, NULL,
+					     smi, THIS_MODULE);
+
+	return rv;
+}
+
+static void remove_proc_entries(ipmi_smi_t smi)
+{
+	struct ipmi_proc_entry *entry;
+
+	while (smi->proc_entries) {
+		entry = smi->proc_entries;
+		smi->proc_entries = entry->next;
+
+		remove_proc_entry(entry->name, smi->proc_dir);
+		kfree(entry->name);
+		kfree(entry);
+	}
+	remove_proc_entry(smi->proc_dir_name, proc_ipmi_root);
+}
+
+static int
+send_channel_info_cmd(ipmi_smi_t intf, int chan)
+{
+	struct kernel_ipmi_msg            msg;
+	unsigned char                     data[1];
+	struct ipmi_system_interface_addr si;
+
+	si.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
+	si.channel = IPMI_BMC_CHANNEL;
+	si.lun = 0;
+
+	msg.netfn = IPMI_NETFN_APP_REQUEST;
+	msg.cmd = IPMI_GET_CHANNEL_INFO_CMD;
+	msg.data = data;
+	msg.data_len = 1;
+	data[0] = chan;
+	return i_ipmi_request(NULL,
+			      intf,
+			      (struct ipmi_addr *) &si,
+			      0,
+			      &msg,
+			      NULL,
+			      NULL,
+			      NULL,
+			      0,
+			      intf->my_address,
+			      intf->my_lun,
+			      -1, 0);
+}
+
+static void
+channel_handler(ipmi_smi_t intf, struct ipmi_smi_msg *msg)
+{
+	int rv = 0;
+	int chan;
+
+	if ((msg->rsp[0] == (IPMI_NETFN_APP_RESPONSE << 2))
+	    && (msg->rsp[1] == IPMI_GET_CHANNEL_INFO_CMD))
+	{
+		/* It's the one we want */
+		if (msg->rsp[2] != 0) {
+			/* Got an error from the channel, just go on. */
+
+			if (msg->rsp[2] == IPMI_INVALID_COMMAND_ERR) {
+				/* If the MC does not support this
+				   command, that is legal.  We just
+				   assume it has one IPMB at channel
+				   zero. */
+				intf->channels[0].medium
+					= IPMI_CHANNEL_MEDIUM_IPMB;
+				intf->channels[0].protocol
+					= IPMI_CHANNEL_PROTOCOL_IPMB;
+				rv = -ENOSYS;
+
+				intf->curr_channel = IPMI_MAX_CHANNELS;
+				wake_up(&intf->waitq);
+				goto out;
+			}
+			goto next_channel;
+		}
+		if (msg->rsp_size < 6) {
+			/* Message not big enough, just go on. */
+			goto next_channel;
+		}
+		chan = intf->curr_channel;
+		intf->channels[chan].medium = msg->rsp[4] & 0x7f;
+		intf->channels[chan].protocol = msg->rsp[5] & 0x1f;
+
+	next_channel:
+		intf->curr_channel++;
+		if (intf->curr_channel >= IPMI_MAX_CHANNELS)
+			wake_up(&intf->waitq);
+		else
+			rv = send_channel_info_cmd(intf, intf->curr_channel);
+
+		if (rv) {
+			/* Got an error somehow, just give up. */
+			intf->curr_channel = IPMI_MAX_CHANNELS;
+			wake_up(&intf->waitq);
+
+			printk(KERN_WARNING PFX
+			       "Error sending channel information: %d\n",
+			       rv);
+		}
+	}
+ out:
+	return;
+}
+
+int ipmi_register_smi(struct ipmi_smi_handlers *handlers,
+		      void		       *send_info,
+		      unsigned char            version_major,
+		      unsigned char            version_minor,
+		      unsigned char            slave_addr,
+		      ipmi_smi_t               *intf)
+{
+	int              i, j;
+	int              rv;
+	ipmi_smi_t       new_intf;
+	unsigned long    flags;
+
+
+	/* Make sure the driver is actually initialized, this handles
+	   problems with initialization order. */
+	if (!initialized) {
+		rv = ipmi_init_msghandler();
+		if (rv)
+			return rv;
+		/* The init code doesn't return an error if it was turned
+		   off, but it won't initialize.  Check that. */
+		if (!initialized)
+			return -ENODEV;
+	}
+
+	new_intf = kmalloc(sizeof(*new_intf), GFP_KERNEL);
+	if (!new_intf)
+		return -ENOMEM;
+	memset(new_intf, 0, sizeof(*new_intf));
+
+	new_intf->proc_dir = NULL;
+
+	rv = -ENOMEM;
+
+	down_write(&interfaces_sem);
+	for (i=0; i<MAX_IPMI_INTERFACES; i++) {
+		if (ipmi_interfaces[i] == NULL) {
+			new_intf->intf_num = i;
+			new_intf->version_major = version_major;
+			new_intf->version_minor = version_minor;
+			if (slave_addr == 0)
+				new_intf->my_address = IPMI_BMC_SLAVE_ADDR;
+			else
+				new_intf->my_address = slave_addr;
+			new_intf->my_lun = 2;  /* the SMS LUN. */
+			rwlock_init(&(new_intf->users_lock));
+			INIT_LIST_HEAD(&(new_intf->users));
+			new_intf->handlers = handlers;
+			new_intf->send_info = send_info;
+			spin_lock_init(&(new_intf->seq_lock));
+			for (j=0; j<IPMI_IPMB_NUM_SEQ; j++) {
+				new_intf->seq_table[j].inuse = 0;
+				new_intf->seq_table[j].seqid = 0;
+			}
+			new_intf->curr_seq = 0;
+			spin_lock_init(&(new_intf->waiting_msgs_lock));
+			INIT_LIST_HEAD(&(new_intf->waiting_msgs));
+			spin_lock_init(&(new_intf->events_lock));
+			INIT_LIST_HEAD(&(new_intf->waiting_events));
+			new_intf->waiting_events_count = 0;
+			rwlock_init(&(new_intf->cmd_rcvr_lock));
+			init_waitqueue_head(&new_intf->waitq);
+			INIT_LIST_HEAD(&(new_intf->cmd_rcvrs));
+			new_intf->all_cmd_rcvr = NULL;
+
+			spin_lock_init(&(new_intf->counter_lock));
+
+			spin_lock_irqsave(&interfaces_lock, flags);
+			ipmi_interfaces[i] = new_intf;
+			spin_unlock_irqrestore(&interfaces_lock, flags);
+
+			rv = 0;
+			*intf = new_intf;
+			break;
+		}
+	}
+
+	downgrade_write(&interfaces_sem);
+
+	if (rv == 0)
+		rv = add_proc_entries(*intf, i);
+
+	if (rv == 0) {
+		if ((version_major > 1)
+		    || ((version_major == 1) && (version_minor >= 5)))
+		{
+			/* Start scanning the channels to see what is
+			   available. */
+			(*intf)->null_user_handler = channel_handler;
+			(*intf)->curr_channel = 0;
+			rv = send_channel_info_cmd(*intf, 0);
+			if (rv)
+				goto out;
+
+			/* Wait for the channel info to be read. */
+			up_read(&interfaces_sem);
+			wait_event((*intf)->waitq,
+				   ((*intf)->curr_channel>=IPMI_MAX_CHANNELS));
+			down_read(&interfaces_sem);
+
+			if (ipmi_interfaces[i] != new_intf)
+				/* Well, it went away.  Just return. */
+				goto out;
+		} else {
+			/* Assume a single IPMB channel at zero. */
+			(*intf)->channels[0].medium = IPMI_CHANNEL_MEDIUM_IPMB;
+			(*intf)->channels[0].protocol
+				= IPMI_CHANNEL_PROTOCOL_IPMB;
+  		}
+
+		/* Call all the watcher interfaces to tell
+		   them that a new interface is available. */
+		call_smi_watchers(i);
+	}
+
+ out:
+	up_read(&interfaces_sem);
+
+	if (rv) {
+		if (new_intf->proc_dir)
+			remove_proc_entries(new_intf);
+		kfree(new_intf);
+	}
+
+	return rv;
+}
+
+static void free_recv_msg_list(struct list_head *q)
+{
+	struct ipmi_recv_msg *msg, *msg2;
+
+	list_for_each_entry_safe(msg, msg2, q, link) {
+		list_del(&msg->link);
+		ipmi_free_recv_msg(msg);
+	}
+}
+
+static void free_cmd_rcvr_list(struct list_head *q)
+{
+	struct cmd_rcvr  *rcvr, *rcvr2;
+
+	list_for_each_entry_safe(rcvr, rcvr2, q, link) {
+		list_del(&rcvr->link);
+		kfree(rcvr);
+	}
+}
+
+static void clean_up_interface_data(ipmi_smi_t intf)
+{
+	int i;
+
+	free_recv_msg_list(&(intf->waiting_msgs));
+	free_recv_msg_list(&(intf->waiting_events));
+	free_cmd_rcvr_list(&(intf->cmd_rcvrs));
+
+	for (i=0; i<IPMI_IPMB_NUM_SEQ; i++) {
+		if ((intf->seq_table[i].inuse)
+		    && (intf->seq_table[i].recv_msg))
+		{
+			ipmi_free_recv_msg(intf->seq_table[i].recv_msg);
+		}	
+	}
+}
+
+int ipmi_unregister_smi(ipmi_smi_t intf)
+{
+	int                     rv = -ENODEV;
+	int                     i;
+	struct ipmi_smi_watcher *w;
+	unsigned long           flags;
+
+	down_write(&interfaces_sem);
+	if (list_empty(&(intf->users)))
+	{
+		for (i=0; i<MAX_IPMI_INTERFACES; i++) {
+			if (ipmi_interfaces[i] == intf) {
+				remove_proc_entries(intf);
+				spin_lock_irqsave(&interfaces_lock, flags);
+				ipmi_interfaces[i] = NULL;
+				clean_up_interface_data(intf);
+				spin_unlock_irqrestore(&interfaces_lock,flags);
+				kfree(intf);
+				rv = 0;
+				goto out_call_watcher;
+			}
+		}
+	} else {
+		rv = -EBUSY;
+	}
+	up_write(&interfaces_sem);
+
+	return rv;
+
+ out_call_watcher:
+	downgrade_write(&interfaces_sem);
+
+	/* Call all the watcher interfaces to tell them that
+	   an interface is gone. */
+	down_read(&smi_watchers_sem);
+	list_for_each_entry(w, &smi_watchers, link) {
+		w->smi_gone(i);
+	}
+	up_read(&smi_watchers_sem);
+	up_read(&interfaces_sem);
+	return 0;
+}
+
+static int handle_ipmb_get_msg_rsp(ipmi_smi_t          intf,
+				   struct ipmi_smi_msg *msg)
+{
+	struct ipmi_ipmb_addr ipmb_addr;
+	struct ipmi_recv_msg  *recv_msg;
+	unsigned long         flags;
+
+	
+	/* This is 11, not 10, because the response must contain a
+	 * completion code. */
+	if (msg->rsp_size < 11) {
+		/* Message not big enough, just ignore it. */
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->invalid_ipmb_responses++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+		return 0;
+	}
+
+	if (msg->rsp[2] != 0) {
+		/* An error getting the response, just ignore it. */
+		return 0;
+	}
+
+	ipmb_addr.addr_type = IPMI_IPMB_ADDR_TYPE;
+	ipmb_addr.slave_addr = msg->rsp[6];
+	ipmb_addr.channel = msg->rsp[3] & 0x0f;
+	ipmb_addr.lun = msg->rsp[7] & 3;
+
+	/* It's a response from a remote entity.  Look up the sequence
+	   number and handle the response. */
+	if (intf_find_seq(intf,
+			  msg->rsp[7] >> 2,
+			  msg->rsp[3] & 0x0f,
+			  msg->rsp[8],
+			  (msg->rsp[4] >> 2) & (~1),
+			  (struct ipmi_addr *) &(ipmb_addr),
+			  &recv_msg))
+	{
+		/* We were unable to find the sequence number,
+		   so just nuke the message. */
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->unhandled_ipmb_responses++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+		return 0;
+	}
+
+	memcpy(recv_msg->msg_data,
+	       &(msg->rsp[9]),
+	       msg->rsp_size - 9);
+	/* THe other fields matched, so no need to set them, except
+           for netfn, which needs to be the response that was
+           returned, not the request value. */
+	recv_msg->msg.netfn = msg->rsp[4] >> 2;
+	recv_msg->msg.data = recv_msg->msg_data;
+	recv_msg->msg.data_len = msg->rsp_size - 10;
+	recv_msg->recv_type = IPMI_RESPONSE_RECV_TYPE;
+	spin_lock_irqsave(&intf->counter_lock, flags);
+	intf->handled_ipmb_responses++;
+	spin_unlock_irqrestore(&intf->counter_lock, flags);
+	deliver_response(recv_msg);
+
+	return 0;
+}
+
+static int handle_ipmb_get_msg_cmd(ipmi_smi_t          intf,
+				   struct ipmi_smi_msg *msg)
+{
+	struct cmd_rcvr       *rcvr;
+	int                   rv = 0;
+	unsigned char         netfn;
+	unsigned char         cmd;
+	ipmi_user_t           user = NULL;
+	struct ipmi_ipmb_addr *ipmb_addr;
+	struct ipmi_recv_msg  *recv_msg;
+	unsigned long         flags;
+
+	if (msg->rsp_size < 10) {
+		/* Message not big enough, just ignore it. */
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->invalid_commands++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+		return 0;
+	}
+
+	if (msg->rsp[2] != 0) {
+		/* An error getting the response, just ignore it. */
+		return 0;
+	}
+
+	netfn = msg->rsp[4] >> 2;
+	cmd = msg->rsp[8];
+
+	read_lock(&(intf->cmd_rcvr_lock));
+	
+	if (intf->all_cmd_rcvr) {
+		user = intf->all_cmd_rcvr;
+	} else {
+		/* Find the command/netfn. */
+		list_for_each_entry(rcvr, &(intf->cmd_rcvrs), link) {
+			if ((rcvr->netfn == netfn) && (rcvr->cmd == cmd)) {
+				user = rcvr->user;
+				break;
+			}
+		}
+	}
+	read_unlock(&(intf->cmd_rcvr_lock));
+
+	if (user == NULL) {
+		/* We didn't find a user, deliver an error response. */
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->unhandled_commands++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+
+		msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
+		msg->data[1] = IPMI_SEND_MSG_CMD;
+		msg->data[2] = msg->rsp[3];
+		msg->data[3] = msg->rsp[6];
+                msg->data[4] = ((netfn + 1) << 2) | (msg->rsp[7] & 0x3);
+		msg->data[5] = ipmb_checksum(&(msg->data[3]), 2);
+		msg->data[6] = intf->my_address;
+                /* rqseq/lun */
+                msg->data[7] = (msg->rsp[7] & 0xfc) | (msg->rsp[4] & 0x3);
+		msg->data[8] = msg->rsp[8]; /* cmd */
+		msg->data[9] = IPMI_INVALID_CMD_COMPLETION_CODE;
+		msg->data[10] = ipmb_checksum(&(msg->data[6]), 4);
+		msg->data_size = 11;
+
+#ifdef DEBUG_MSGING
+	{
+		int m;
+		printk("Invalid command:");
+		for (m=0; m<msg->data_size; m++)
+			printk(" %2.2x", msg->data[m]);
+		printk("\n");
+	}
+#endif
+		intf->handlers->sender(intf->send_info, msg, 0);
+
+		rv = -1; /* We used the message, so return the value that
+			    causes it to not be freed or queued. */
+	} else {
+		/* Deliver the message to the user. */
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->handled_commands++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+
+		recv_msg = ipmi_alloc_recv_msg();
+		if (! recv_msg) {
+			/* We couldn't allocate memory for the
+                           message, so requeue it for handling
+                           later. */
+			rv = 1;
+		} else {
+			/* Extract the source address from the data. */
+			ipmb_addr = (struct ipmi_ipmb_addr *) &recv_msg->addr;
+			ipmb_addr->addr_type = IPMI_IPMB_ADDR_TYPE;
+			ipmb_addr->slave_addr = msg->rsp[6];
+			ipmb_addr->lun = msg->rsp[7] & 3;
+			ipmb_addr->channel = msg->rsp[3] & 0xf;
+
+			/* Extract the rest of the message information
+			   from the IPMB header.*/
+			recv_msg->user = user;
+			recv_msg->recv_type = IPMI_CMD_RECV_TYPE;
+			recv_msg->msgid = msg->rsp[7] >> 2;
+			recv_msg->msg.netfn = msg->rsp[4] >> 2;
+			recv_msg->msg.cmd = msg->rsp[8];
+			recv_msg->msg.data = recv_msg->msg_data;
+
+			/* We chop off 10, not 9 bytes because the checksum
+			   at the end also needs to be removed. */
+			recv_msg->msg.data_len = msg->rsp_size - 10;
+			memcpy(recv_msg->msg_data,
+			       &(msg->rsp[9]),
+			       msg->rsp_size - 10);
+			deliver_response(recv_msg);
+		}
+	}
+
+	return rv;
+}
+
+static int handle_lan_get_msg_rsp(ipmi_smi_t          intf,
+				  struct ipmi_smi_msg *msg)
+{
+	struct ipmi_lan_addr  lan_addr;
+	struct ipmi_recv_msg  *recv_msg;
+	unsigned long         flags;
+
+
+	/* This is 13, not 12, because the response must contain a
+	 * completion code. */
+	if (msg->rsp_size < 13) {
+		/* Message not big enough, just ignore it. */
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->invalid_lan_responses++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+		return 0;
+	}
+
+	if (msg->rsp[2] != 0) {
+		/* An error getting the response, just ignore it. */
+		return 0;
+	}
+
+	lan_addr.addr_type = IPMI_LAN_ADDR_TYPE;
+	lan_addr.session_handle = msg->rsp[4];
+	lan_addr.remote_SWID = msg->rsp[8];
+	lan_addr.local_SWID = msg->rsp[5];
+	lan_addr.channel = msg->rsp[3] & 0x0f;
+	lan_addr.privilege = msg->rsp[3] >> 4;
+	lan_addr.lun = msg->rsp[9] & 3;
+
+	/* It's a response from a remote entity.  Look up the sequence
+	   number and handle the response. */
+	if (intf_find_seq(intf,
+			  msg->rsp[9] >> 2,
+			  msg->rsp[3] & 0x0f,
+			  msg->rsp[10],
+			  (msg->rsp[6] >> 2) & (~1),
+			  (struct ipmi_addr *) &(lan_addr),
+			  &recv_msg))
+	{
+		/* We were unable to find the sequence number,
+		   so just nuke the message. */
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->unhandled_lan_responses++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+		return 0;
+	}
+
+	memcpy(recv_msg->msg_data,
+	       &(msg->rsp[11]),
+	       msg->rsp_size - 11);
+	/* The other fields matched, so no need to set them, except
+           for netfn, which needs to be the response that was
+           returned, not the request value. */
+	recv_msg->msg.netfn = msg->rsp[6] >> 2;
+	recv_msg->msg.data = recv_msg->msg_data;
+	recv_msg->msg.data_len = msg->rsp_size - 12;
+	recv_msg->recv_type = IPMI_RESPONSE_RECV_TYPE;
+	spin_lock_irqsave(&intf->counter_lock, flags);
+	intf->handled_lan_responses++;
+	spin_unlock_irqrestore(&intf->counter_lock, flags);
+	deliver_response(recv_msg);
+
+	return 0;
+}
+
+static int handle_lan_get_msg_cmd(ipmi_smi_t          intf,
+				  struct ipmi_smi_msg *msg)
+{
+	struct cmd_rcvr       *rcvr;
+	int                   rv = 0;
+	unsigned char         netfn;
+	unsigned char         cmd;
+	ipmi_user_t           user = NULL;
+	struct ipmi_lan_addr  *lan_addr;
+	struct ipmi_recv_msg  *recv_msg;
+	unsigned long         flags;
+
+	if (msg->rsp_size < 12) {
+		/* Message not big enough, just ignore it. */
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->invalid_commands++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+		return 0;
+	}
+
+	if (msg->rsp[2] != 0) {
+		/* An error getting the response, just ignore it. */
+		return 0;
+	}
+
+	netfn = msg->rsp[6] >> 2;
+	cmd = msg->rsp[10];
+
+	read_lock(&(intf->cmd_rcvr_lock));
+
+	if (intf->all_cmd_rcvr) {
+		user = intf->all_cmd_rcvr;
+	} else {
+		/* Find the command/netfn. */
+		list_for_each_entry(rcvr, &(intf->cmd_rcvrs), link) {
+			if ((rcvr->netfn == netfn) && (rcvr->cmd == cmd)) {
+				user = rcvr->user;
+				break;
+			}
+		}
+	}
+	read_unlock(&(intf->cmd_rcvr_lock));
+
+	if (user == NULL) {
+		/* We didn't find a user, deliver an error response. */
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->unhandled_commands++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+
+		rv = 0; /* Don't do anything with these messages, just
+			   allow them to be freed. */
+	} else {
+		/* Deliver the message to the user. */
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->handled_commands++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+
+		recv_msg = ipmi_alloc_recv_msg();
+		if (! recv_msg) {
+			/* We couldn't allocate memory for the
+                           message, so requeue it for handling
+                           later. */
+			rv = 1;
+		} else {
+			/* Extract the source address from the data. */
+			lan_addr = (struct ipmi_lan_addr *) &recv_msg->addr;
+			lan_addr->addr_type = IPMI_LAN_ADDR_TYPE;
+			lan_addr->session_handle = msg->rsp[4];
+			lan_addr->remote_SWID = msg->rsp[8];
+			lan_addr->local_SWID = msg->rsp[5];
+			lan_addr->lun = msg->rsp[9] & 3;
+			lan_addr->channel = msg->rsp[3] & 0xf;
+			lan_addr->privilege = msg->rsp[3] >> 4;
+
+			/* Extract the rest of the message information
+			   from the IPMB header.*/
+			recv_msg->user = user;
+			recv_msg->recv_type = IPMI_CMD_RECV_TYPE;
+			recv_msg->msgid = msg->rsp[9] >> 2;
+			recv_msg->msg.netfn = msg->rsp[6] >> 2;
+			recv_msg->msg.cmd = msg->rsp[10];
+			recv_msg->msg.data = recv_msg->msg_data;
+
+			/* We chop off 12, not 11 bytes because the checksum
+			   at the end also needs to be removed. */
+			recv_msg->msg.data_len = msg->rsp_size - 12;
+			memcpy(recv_msg->msg_data,
+			       &(msg->rsp[11]),
+			       msg->rsp_size - 12);
+			deliver_response(recv_msg);
+		}
+	}
+
+	return rv;
+}
+
+static void copy_event_into_recv_msg(struct ipmi_recv_msg *recv_msg,
+				     struct ipmi_smi_msg  *msg)
+{
+	struct ipmi_system_interface_addr *smi_addr;
+	
+	recv_msg->msgid = 0;
+	smi_addr = (struct ipmi_system_interface_addr *) &(recv_msg->addr);
+	smi_addr->addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
+	smi_addr->channel = IPMI_BMC_CHANNEL;
+	smi_addr->lun = msg->rsp[0] & 3;
+	recv_msg->recv_type = IPMI_ASYNC_EVENT_RECV_TYPE;
+	recv_msg->msg.netfn = msg->rsp[0] >> 2;
+	recv_msg->msg.cmd = msg->rsp[1];
+	memcpy(recv_msg->msg_data, &(msg->rsp[3]), msg->rsp_size - 3);
+	recv_msg->msg.data = recv_msg->msg_data;
+	recv_msg->msg.data_len = msg->rsp_size - 3;
+}
+
+/* This will be called with the intf->users_lock read-locked, so no need
+   to do that here. */
+static int handle_read_event_rsp(ipmi_smi_t          intf,
+				 struct ipmi_smi_msg *msg)
+{
+	struct ipmi_recv_msg *recv_msg, *recv_msg2;
+	struct list_head     msgs;
+	ipmi_user_t          user;
+	int                  rv = 0;
+	int                  deliver_count = 0;
+	unsigned long        flags;
+
+	if (msg->rsp_size < 19) {
+		/* Message is too small to be an IPMB event. */
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->invalid_events++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+		return 0;
+	}
+
+	if (msg->rsp[2] != 0) {
+		/* An error getting the event, just ignore it. */
+		return 0;
+	}
+
+	INIT_LIST_HEAD(&msgs);
+
+	spin_lock_irqsave(&(intf->events_lock), flags);
+
+	spin_lock(&intf->counter_lock);
+	intf->events++;
+	spin_unlock(&intf->counter_lock);
+
+	/* Allocate and fill in one message for every user that is getting
+	   events. */
+	list_for_each_entry(user, &(intf->users), link) {
+		if (! user->gets_events)
+			continue;
+
+		recv_msg = ipmi_alloc_recv_msg();
+		if (! recv_msg) {
+			list_for_each_entry_safe(recv_msg, recv_msg2, &msgs, link) {
+				list_del(&recv_msg->link);
+				ipmi_free_recv_msg(recv_msg);
+			}
+			/* We couldn't allocate memory for the
+                           message, so requeue it for handling
+                           later. */
+			rv = 1;
+			goto out;
+		}
+
+		deliver_count++;
+
+		copy_event_into_recv_msg(recv_msg, msg);
+		recv_msg->user = user;
+		list_add_tail(&(recv_msg->link), &msgs);
+	}
+
+	if (deliver_count) {
+		/* Now deliver all the messages. */
+		list_for_each_entry_safe(recv_msg, recv_msg2, &msgs, link) {
+			list_del(&recv_msg->link);
+			deliver_response(recv_msg);
+		}
+	} else if (intf->waiting_events_count < MAX_EVENTS_IN_QUEUE) {
+		/* No one to receive the message, put it in queue if there's
+		   not already too many things in the queue. */
+		recv_msg = ipmi_alloc_recv_msg();
+		if (! recv_msg) {
+			/* We couldn't allocate memory for the
+                           message, so requeue it for handling
+                           later. */
+			rv = 1;
+			goto out;
+		}
+
+		copy_event_into_recv_msg(recv_msg, msg);
+		list_add_tail(&(recv_msg->link), &(intf->waiting_events));
+	} else {
+		/* There's too many things in the queue, discard this
+		   message. */
+		printk(KERN_WARNING PFX "Event queue full, discarding an"
+		       " incoming event\n");
+	}
+
+ out:
+	spin_unlock_irqrestore(&(intf->events_lock), flags);
+
+	return rv;
+}
+
+static int handle_bmc_rsp(ipmi_smi_t          intf,
+			  struct ipmi_smi_msg *msg)
+{
+	struct ipmi_recv_msg *recv_msg;
+	int                  found = 0;
+	struct ipmi_user     *user;
+	unsigned long        flags;
+
+	recv_msg = (struct ipmi_recv_msg *) msg->user_data;
+
+	/* Make sure the user still exists. */
+	list_for_each_entry(user, &(intf->users), link) {
+		if (user == recv_msg->user) {
+			/* Found it, so we can deliver it */
+			found = 1;
+			break;
+		}
+	}
+
+	if (!found) {
+		/* Special handling for NULL users. */
+		if (!recv_msg->user && intf->null_user_handler){
+			intf->null_user_handler(intf, msg);
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->handled_local_responses++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+		}else{
+			/* The user for the message went away, so give up. */
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			intf->unhandled_local_responses++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+		}
+		ipmi_free_recv_msg(recv_msg);
+	} else {
+		struct ipmi_system_interface_addr *smi_addr;
+
+		spin_lock_irqsave(&intf->counter_lock, flags);
+		intf->handled_local_responses++;
+		spin_unlock_irqrestore(&intf->counter_lock, flags);
+		recv_msg->recv_type = IPMI_RESPONSE_RECV_TYPE;
+		recv_msg->msgid = msg->msgid;
+		smi_addr = ((struct ipmi_system_interface_addr *)
+			    &(recv_msg->addr));
+		smi_addr->addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
+		smi_addr->channel = IPMI_BMC_CHANNEL;
+		smi_addr->lun = msg->rsp[0] & 3;
+		recv_msg->msg.netfn = msg->rsp[0] >> 2;
+		recv_msg->msg.cmd = msg->rsp[1];
+		memcpy(recv_msg->msg_data,
+		       &(msg->rsp[2]),
+		       msg->rsp_size - 2);
+		recv_msg->msg.data = recv_msg->msg_data;
+		recv_msg->msg.data_len = msg->rsp_size - 2;
+		deliver_response(recv_msg);
+	}
+
+	return 0;
+}
+
+/* Handle a new message.  Return 1 if the message should be requeued,
+   0 if the message should be freed, or -1 if the message should not
+   be freed or requeued. */
+static int handle_new_recv_msg(ipmi_smi_t          intf,
+			       struct ipmi_smi_msg *msg)
+{
+	int requeue;
+	int chan;
+
+#ifdef DEBUG_MSGING
+	int m;
+	printk("Recv:");
+	for (m=0; m<msg->rsp_size; m++)
+		printk(" %2.2x", msg->rsp[m]);
+	printk("\n");
+#endif
+	if (msg->rsp_size < 2) {
+		/* Message is too small to be correct. */
+		printk(KERN_WARNING PFX "BMC returned to small a message"
+		       " for netfn %x cmd %x, got %d bytes\n",
+		       (msg->data[0] >> 2) | 1, msg->data[1], msg->rsp_size);
+
+		/* Generate an error response for the message. */
+		msg->rsp[0] = msg->data[0] | (1 << 2);
+		msg->rsp[1] = msg->data[1];
+		msg->rsp[2] = IPMI_ERR_UNSPECIFIED;
+		msg->rsp_size = 3;
+	} else if (((msg->rsp[0] >> 2) != ((msg->data[0] >> 2) | 1))/* Netfn */
+		   || (msg->rsp[1] != msg->data[1]))		  /* Command */
+	{
+		/* The response is not even marginally correct. */
+		printk(KERN_WARNING PFX "BMC returned incorrect response,"
+		       " expected netfn %x cmd %x, got netfn %x cmd %x\n",
+		       (msg->data[0] >> 2) | 1, msg->data[1],
+		       msg->rsp[0] >> 2, msg->rsp[1]);
+
+		/* Generate an error response for the message. */
+		msg->rsp[0] = msg->data[0] | (1 << 2);
+		msg->rsp[1] = msg->data[1];
+		msg->rsp[2] = IPMI_ERR_UNSPECIFIED;
+		msg->rsp_size = 3;
+	}
+
+	if ((msg->rsp[0] == ((IPMI_NETFN_APP_REQUEST|1) << 2))
+	    && (msg->rsp[1] == IPMI_SEND_MSG_CMD)
+	    && (msg->user_data != NULL))
+	{
+		/* It's a response to a response we sent.  For this we
+		   deliver a send message response to the user. */
+		struct ipmi_recv_msg *recv_msg = msg->user_data;
+
+		requeue = 0;
+		if (msg->rsp_size < 2)
+			/* Message is too small to be correct. */
+			goto out;
+
+		chan = msg->data[2] & 0x0f;
+		if (chan >= IPMI_MAX_CHANNELS)
+			/* Invalid channel number */
+			goto out;
+
+		if (recv_msg) {
+			recv_msg->recv_type = IPMI_RESPONSE_RESPONSE_TYPE;
+			recv_msg->msg.data = recv_msg->msg_data;
+			recv_msg->msg.data_len = 1;
+			recv_msg->msg_data[0] = msg->rsp[2];
+			deliver_response(recv_msg);
+		}
+	} else if ((msg->rsp[0] == ((IPMI_NETFN_APP_REQUEST|1) << 2))
+		   && (msg->rsp[1] == IPMI_GET_MSG_CMD))
+	{
+		/* It's from the receive queue. */
+		chan = msg->rsp[3] & 0xf;
+		if (chan >= IPMI_MAX_CHANNELS) {
+			/* Invalid channel number */
+			requeue = 0;
+			goto out;
+		}
+
+		switch (intf->channels[chan].medium) {
+		case IPMI_CHANNEL_MEDIUM_IPMB:
+			if (msg->rsp[4] & 0x04) {
+				/* It's a response, so find the
+				   requesting message and send it up. */
+				requeue = handle_ipmb_get_msg_rsp(intf, msg);
+			} else {
+				/* It's a command to the SMS from some other
+				   entity.  Handle that. */
+				requeue = handle_ipmb_get_msg_cmd(intf, msg);
+			}
+			break;
+
+		case IPMI_CHANNEL_MEDIUM_8023LAN:
+		case IPMI_CHANNEL_MEDIUM_ASYNC:
+			if (msg->rsp[6] & 0x04) {
+				/* It's a response, so find the
+				   requesting message and send it up. */
+				requeue = handle_lan_get_msg_rsp(intf, msg);
+			} else {
+				/* It's a command to the SMS from some other
+				   entity.  Handle that. */
+				requeue = handle_lan_get_msg_cmd(intf, msg);
+			}
+			break;
+
+		default:
+			/* We don't handle the channel type, so just
+			 * free the message. */
+			requeue = 0;
+		}
+
+	} else if ((msg->rsp[0] == ((IPMI_NETFN_APP_REQUEST|1) << 2))
+		   && (msg->rsp[1] == IPMI_READ_EVENT_MSG_BUFFER_CMD))
+	{
+		/* It's an asyncronous event. */
+		requeue = handle_read_event_rsp(intf, msg);
+	} else {
+		/* It's a response from the local BMC. */
+		requeue = handle_bmc_rsp(intf, msg);
+	}
+
+ out:
+	return requeue;
+}
+
+/* Handle a new message from the lower layer. */
+void ipmi_smi_msg_received(ipmi_smi_t          intf,
+			   struct ipmi_smi_msg *msg)
+{
+	unsigned long flags;
+	int           rv;
+
+
+	/* Lock the user lock so the user can't go away while we are
+	   working on it. */
+	read_lock(&(intf->users_lock));
+
+	if ((msg->data_size >= 2)
+	    && (msg->data[0] == (IPMI_NETFN_APP_REQUEST << 2))
+	    && (msg->data[1] == IPMI_SEND_MSG_CMD)
+	    && (msg->user_data == NULL)) {
+		/* This is the local response to a command send, start
+                   the timer for these.  The user_data will not be
+                   NULL if this is a response send, and we will let
+                   response sends just go through. */
+
+		/* Check for errors, if we get certain errors (ones
+                   that mean basically we can try again later), we
+                   ignore them and start the timer.  Otherwise we
+                   report the error immediately. */
+		if ((msg->rsp_size >= 3) && (msg->rsp[2] != 0)
+		    && (msg->rsp[2] != IPMI_NODE_BUSY_ERR)
+		    && (msg->rsp[2] != IPMI_LOST_ARBITRATION_ERR))
+		{
+			int chan = msg->rsp[3] & 0xf;
+
+			/* Got an error sending the message, handle it. */
+			spin_lock_irqsave(&intf->counter_lock, flags);
+			if (chan >= IPMI_MAX_CHANNELS)
+				; /* This shouldn't happen */
+			else if ((intf->channels[chan].medium
+				  == IPMI_CHANNEL_MEDIUM_8023LAN)
+				 || (intf->channels[chan].medium
+				     == IPMI_CHANNEL_MEDIUM_ASYNC))
+				intf->sent_lan_command_errs++;
+			else
+				intf->sent_ipmb_command_errs++;
+			spin_unlock_irqrestore(&intf->counter_lock, flags);
+			intf_err_seq(intf, msg->msgid, msg->rsp[2]);
+		} else {
+			/* The message was sent, start the timer. */
+			intf_start_seq_timer(intf, msg->msgid);
+		}
+
+		ipmi_free_smi_msg(msg);
+		goto out_unlock;
+	}
+
+	/* To preserve message order, if the list is not empty, we
+           tack this message onto the end of the list. */
+	spin_lock_irqsave(&(intf->waiting_msgs_lock), flags);
+	if (!list_empty(&(intf->waiting_msgs))) {
+		list_add_tail(&(msg->link), &(intf->waiting_msgs));
+		spin_unlock(&(intf->waiting_msgs_lock));
+		goto out_unlock;
+	}
+	spin_unlock_irqrestore(&(intf->waiting_msgs_lock), flags);
+		
+	rv = handle_new_recv_msg(intf, msg);
+	if (rv > 0) {
+		/* Could not handle the message now, just add it to a
+                   list to handle later. */
+		spin_lock(&(intf->waiting_msgs_lock));
+		list_add_tail(&(msg->link), &(intf->waiting_msgs));
+		spin_unlock(&(intf->waiting_msgs_lock));
+	} else if (rv == 0) {
+		ipmi_free_smi_msg(msg);
+	}
+
+ out_unlock:
+	read_unlock(&(intf->users_lock));
+}
+
+void ipmi_smi_watchdog_pretimeout(ipmi_smi_t intf)
+{
+	ipmi_user_t user;
+
+	read_lock(&(intf->users_lock));
+	list_for_each_entry(user, &(intf->users), link) {
+		if (! user->handler->ipmi_watchdog_pretimeout)
+			continue;
+
+		user->handler->ipmi_watchdog_pretimeout(user->handler_data);
+	}
+	read_unlock(&(intf->users_lock));
+}
+
+static void
+handle_msg_timeout(struct ipmi_recv_msg *msg)
+{
+	msg->recv_type = IPMI_RESPONSE_RECV_TYPE;
+	msg->msg_data[0] = IPMI_TIMEOUT_COMPLETION_CODE;
+	msg->msg.netfn |= 1; /* Convert to a response. */
+	msg->msg.data_len = 1;
+	msg->msg.data = msg->msg_data;
+	deliver_response(msg);
+}
+
+static void
+send_from_recv_msg(ipmi_smi_t intf, struct ipmi_recv_msg *recv_msg,
+		   struct ipmi_smi_msg *smi_msg,
+		   unsigned char seq, long seqid)
+{
+	if (!smi_msg)
+		smi_msg = ipmi_alloc_smi_msg();
+	if (!smi_msg)
+		/* If we can't allocate the message, then just return, we
+		   get 4 retries, so this should be ok. */
+		return;
+
+	memcpy(smi_msg->data, recv_msg->msg.data, recv_msg->msg.data_len);
+	smi_msg->data_size = recv_msg->msg.data_len;
+	smi_msg->msgid = STORE_SEQ_IN_MSGID(seq, seqid);
+		
+	/* Send the new message.  We send with a zero priority.  It
+	   timed out, I doubt time is that critical now, and high
+	   priority messages are really only for messages to the local
+	   MC, which don't get resent. */
+	intf->handlers->sender(intf->send_info, smi_msg, 0);
+
+#ifdef DEBUG_MSGING
+	{
+		int m;
+		printk("Resend: ");
+		for (m=0; m<smi_msg->data_size; m++)
+			printk(" %2.2x", smi_msg->data[m]);
+		printk("\n");
+	}
+#endif
+}
+
+static void
+ipmi_timeout_handler(long timeout_period)
+{
+	ipmi_smi_t           intf;
+	struct list_head     timeouts;
+	struct ipmi_recv_msg *msg, *msg2;
+	struct ipmi_smi_msg  *smi_msg, *smi_msg2;
+	unsigned long        flags;
+	int                  i, j;
+
+	INIT_LIST_HEAD(&timeouts);
+
+	spin_lock(&interfaces_lock);
+	for (i=0; i<MAX_IPMI_INTERFACES; i++) {
+		intf = ipmi_interfaces[i];
+		if (intf == NULL)
+			continue;
+
+		read_lock(&(intf->users_lock));
+
+		/* See if any waiting messages need to be processed. */
+		spin_lock_irqsave(&(intf->waiting_msgs_lock), flags);
+		list_for_each_entry_safe(smi_msg, smi_msg2, &(intf->waiting_msgs), link) {
+			if (! handle_new_recv_msg(intf, smi_msg)) {
+				list_del(&smi_msg->link);
+				ipmi_free_smi_msg(smi_msg);
+			} else {
+				/* To preserve message order, quit if we
+				   can't handle a message. */
+				break;
+			}
+		}
+		spin_unlock_irqrestore(&(intf->waiting_msgs_lock), flags);
+
+		/* Go through the seq table and find any messages that
+		   have timed out, putting them in the timeouts
+		   list. */
+		spin_lock_irqsave(&(intf->seq_lock), flags);
+		for (j=0; j<IPMI_IPMB_NUM_SEQ; j++) {
+			struct seq_table *ent = &(intf->seq_table[j]);
+			if (!ent->inuse)
+				continue;
+
+			ent->timeout -= timeout_period;
+			if (ent->timeout > 0)
+				continue;
+
+			if (ent->retries_left == 0) {
+				/* The message has used all its retries. */
+				ent->inuse = 0;
+				msg = ent->recv_msg;
+				list_add_tail(&(msg->link), &timeouts);
+				spin_lock(&intf->counter_lock);
+				if (ent->broadcast)
+					intf->timed_out_ipmb_broadcasts++;
+				else if (ent->recv_msg->addr.addr_type
+					 == IPMI_LAN_ADDR_TYPE)
+					intf->timed_out_lan_commands++;
+				else
+					intf->timed_out_ipmb_commands++;
+				spin_unlock(&intf->counter_lock);
+			} else {
+				/* More retries, send again. */
+
+				/* Start with the max timer, set to normal
+				   timer after the message is sent. */
+				ent->timeout = MAX_MSG_TIMEOUT;
+				ent->retries_left--;
+				send_from_recv_msg(intf, ent->recv_msg, NULL,
+						   j, ent->seqid);
+				spin_lock(&intf->counter_lock);
+				if (ent->recv_msg->addr.addr_type
+				    == IPMI_LAN_ADDR_TYPE)
+					intf->retransmitted_lan_commands++;
+				else
+					intf->retransmitted_ipmb_commands++;
+				spin_unlock(&intf->counter_lock);
+			}
+		}
+		spin_unlock_irqrestore(&(intf->seq_lock), flags);
+
+		list_for_each_entry_safe(msg, msg2, &timeouts, link) {
+			handle_msg_timeout(msg);
+		}
+
+		read_unlock(&(intf->users_lock));
+	}
+	spin_unlock(&interfaces_lock);
+}
+
+static void ipmi_request_event(void)
+{
+	ipmi_smi_t intf;
+	int        i;
+
+	spin_lock(&interfaces_lock);
+	for (i=0; i<MAX_IPMI_INTERFACES; i++) {
+		intf = ipmi_interfaces[i];
+		if (intf == NULL)
+			continue;
+
+		intf->handlers->request_events(intf->send_info);
+	}
+	spin_unlock(&interfaces_lock);
+}
+
+static struct timer_list ipmi_timer;
+
+/* Call every ~100 ms. */
+#define IPMI_TIMEOUT_TIME	100
+
+/* How many jiffies does it take to get to the timeout time. */
+#define IPMI_TIMEOUT_JIFFIES	((IPMI_TIMEOUT_TIME * HZ) / 1000)
+
+/* Request events from the queue every second (this is the number of
+   IPMI_TIMEOUT_TIMES between event requests).  Hopefully, in the
+   future, IPMI will add a way to know immediately if an event is in
+   the queue and this silliness can go away. */
+#define IPMI_REQUEST_EV_TIME	(1000 / (IPMI_TIMEOUT_TIME))
+
+static volatile int stop_operation = 0;
+static volatile int timer_stopped = 0;
+static unsigned int ticks_to_req_ev = IPMI_REQUEST_EV_TIME;
+
+static void ipmi_timeout(unsigned long data)
+{
+	if (stop_operation) {
+		timer_stopped = 1;
+		return;
+	}
+
+	ticks_to_req_ev--;
+	if (ticks_to_req_ev == 0) {
+		ipmi_request_event();
+		ticks_to_req_ev = IPMI_REQUEST_EV_TIME;
+	}
+
+	ipmi_timeout_handler(IPMI_TIMEOUT_TIME);
+
+	ipmi_timer.expires += IPMI_TIMEOUT_JIFFIES;
+	add_timer(&ipmi_timer);
+}
+
+
+static atomic_t smi_msg_inuse_count = ATOMIC_INIT(0);
+static atomic_t recv_msg_inuse_count = ATOMIC_INIT(0);
+
+/* FIXME - convert these to slabs. */
+static void free_smi_msg(struct ipmi_smi_msg *msg)
+{
+	atomic_dec(&smi_msg_inuse_count);
+	kfree(msg);
+}
+
+struct ipmi_smi_msg *ipmi_alloc_smi_msg(void)
+{
+	struct ipmi_smi_msg *rv;
+	rv = kmalloc(sizeof(struct ipmi_smi_msg), GFP_ATOMIC);
+	if (rv) {
+		rv->done = free_smi_msg;
+		rv->user_data = NULL;
+		atomic_inc(&smi_msg_inuse_count);
+	}
+	return rv;
+}
+
+static void free_recv_msg(struct ipmi_recv_msg *msg)
+{
+	atomic_dec(&recv_msg_inuse_count);
+	kfree(msg);
+}
+
+struct ipmi_recv_msg *ipmi_alloc_recv_msg(void)
+{
+	struct ipmi_recv_msg *rv;
+
+	rv = kmalloc(sizeof(struct ipmi_recv_msg), GFP_ATOMIC);
+	if (rv) {
+		rv->done = free_recv_msg;
+		atomic_inc(&recv_msg_inuse_count);
+	}
+	return rv;
+}
+
+#ifdef CONFIG_IPMI_PANIC_EVENT
+
+static void dummy_smi_done_handler(struct ipmi_smi_msg *msg)
+{
+}
+
+static void dummy_recv_done_handler(struct ipmi_recv_msg *msg)
+{
+}
+
+#ifdef CONFIG_IPMI_PANIC_STRING
+static void event_receiver_fetcher(ipmi_smi_t intf, struct ipmi_smi_msg *msg)
+{
+	if ((msg->rsp[0] == (IPMI_NETFN_SENSOR_EVENT_RESPONSE << 2))
+	    && (msg->rsp[1] == IPMI_GET_EVENT_RECEIVER_CMD)
+	    && (msg->rsp[2] == IPMI_CC_NO_ERROR))
+	{
+		/* A get event receiver command, save it. */
+		intf->event_receiver = msg->rsp[3];
+		intf->event_receiver_lun = msg->rsp[4] & 0x3;
+	}
+}
+
+static void device_id_fetcher(ipmi_smi_t intf, struct ipmi_smi_msg *msg)
+{
+	if ((msg->rsp[0] == (IPMI_NETFN_APP_RESPONSE << 2))
+	    && (msg->rsp[1] == IPMI_GET_DEVICE_ID_CMD)
+	    && (msg->rsp[2] == IPMI_CC_NO_ERROR))
+	{
+		/* A get device id command, save if we are an event
+		   receiver or generator. */
+		intf->local_sel_device = (msg->rsp[8] >> 2) & 1;
+		intf->local_event_generator = (msg->rsp[8] >> 5) & 1;
+	}
+}
+#endif
+
+static void send_panic_events(char *str)
+{
+	struct kernel_ipmi_msg            msg;
+	ipmi_smi_t                        intf;
+	unsigned char                     data[16];
+	int                               i;
+	struct ipmi_system_interface_addr *si;
+	struct ipmi_addr                  addr;
+	struct ipmi_smi_msg               smi_msg;
+	struct ipmi_recv_msg              recv_msg;
+
+	si = (struct ipmi_system_interface_addr *) &addr;
+	si->addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
+	si->channel = IPMI_BMC_CHANNEL;
+	si->lun = 0;
+
+	/* Fill in an event telling that we have failed. */
+	msg.netfn = 0x04; /* Sensor or Event. */
+	msg.cmd = 2; /* Platform event command. */
+	msg.data = data;
+	msg.data_len = 8;
+	data[0] = 0x21; /* Kernel generator ID, IPMI table 5-4 */
+	data[1] = 0x03; /* This is for IPMI 1.0. */
+	data[2] = 0x20; /* OS Critical Stop, IPMI table 36-3 */
+	data[4] = 0x6f; /* Sensor specific, IPMI table 36-1 */
+	data[5] = 0xa1; /* Runtime stop OEM bytes 2 & 3. */
+
+	/* Put a few breadcrumbs in.  Hopefully later we can add more things
+	   to make the panic events more useful. */
+	if (str) {
+		data[3] = str[0];
+		data[6] = str[1];
+		data[7] = str[2];
+	}
+
+	smi_msg.done = dummy_smi_done_handler;
+	recv_msg.done = dummy_recv_done_handler;
+
+	/* For every registered interface, send the event. */
+	for (i=0; i<MAX_IPMI_INTERFACES; i++) {
+		intf = ipmi_interfaces[i];
+		if (intf == NULL)
+			continue;
+
+		/* Send the event announcing the panic. */
+		intf->handlers->set_run_to_completion(intf->send_info, 1);
+		i_ipmi_request(NULL,
+			       intf,
+			       &addr,
+			       0,
+			       &msg,
+			       NULL,
+			       &smi_msg,
+			       &recv_msg,
+			       0,
+			       intf->my_address,
+			       intf->my_lun,
+			       0, 1); /* Don't retry, and don't wait. */
+	}
+
+#ifdef CONFIG_IPMI_PANIC_STRING
+	/* On every interface, dump a bunch of OEM event holding the
+	   string. */
+	if (!str) 
+		return;
+
+	for (i=0; i<MAX_IPMI_INTERFACES; i++) {
+		char                  *p = str;
+		struct ipmi_ipmb_addr *ipmb;
+		int                   j;
+
+		intf = ipmi_interfaces[i];
+		if (intf == NULL)
+			continue;
+
+		/* First job here is to figure out where to send the
+		   OEM events.  There's no way in IPMI to send OEM
+		   events using an event send command, so we have to
+		   find the SEL to put them in and stick them in
+		   there. */
+
+		/* Get capabilities from the get device id. */
+		intf->local_sel_device = 0;
+		intf->local_event_generator = 0;
+		intf->event_receiver = 0;
+
+		/* Request the device info from the local MC. */
+		msg.netfn = IPMI_NETFN_APP_REQUEST;
+		msg.cmd = IPMI_GET_DEVICE_ID_CMD;
+		msg.data = NULL;
+		msg.data_len = 0;
+		intf->null_user_handler = device_id_fetcher;
+		i_ipmi_request(NULL,
+			       intf,
+			       &addr,
+			       0,
+			       &msg,
+			       NULL,
+			       &smi_msg,
+			       &recv_msg,
+			       0,
+			       intf->my_address,
+			       intf->my_lun,
+			       0, 1); /* Don't retry, and don't wait. */
+
+		if (intf->local_event_generator) {
+			/* Request the event receiver from the local MC. */
+			msg.netfn = IPMI_NETFN_SENSOR_EVENT_REQUEST;
+			msg.cmd = IPMI_GET_EVENT_RECEIVER_CMD;
+			msg.data = NULL;
+			msg.data_len = 0;
+			intf->null_user_handler = event_receiver_fetcher;
+			i_ipmi_request(NULL,
+				       intf,
+				       &addr,
+				       0,
+				       &msg,
+				       NULL,
+				       &smi_msg,
+				       &recv_msg,
+				       0,
+				       intf->my_address,
+				       intf->my_lun,
+				       0, 1); /* no retry, and no wait. */
+		}
+		intf->null_user_handler = NULL;
+
+		/* Validate the event receiver.  The low bit must not
+		   be 1 (it must be a valid IPMB address), it cannot
+		   be zero, and it must not be my address. */
+                if (((intf->event_receiver & 1) == 0)
+		    && (intf->event_receiver != 0)
+		    && (intf->event_receiver != intf->my_address))
+		{
+			/* The event receiver is valid, send an IPMB
+			   message. */
+			ipmb = (struct ipmi_ipmb_addr *) &addr;
+			ipmb->addr_type = IPMI_IPMB_ADDR_TYPE;
+			ipmb->channel = 0; /* FIXME - is this right? */
+			ipmb->lun = intf->event_receiver_lun;
+			ipmb->slave_addr = intf->event_receiver;
+		} else if (intf->local_sel_device) {
+			/* The event receiver was not valid (or was
+			   me), but I am an SEL device, just dump it
+			   in my SEL. */
+			si = (struct ipmi_system_interface_addr *) &addr;
+			si->addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
+			si->channel = IPMI_BMC_CHANNEL;
+			si->lun = 0;
+		} else
+			continue; /* No where to send the event. */
+
+		
+		msg.netfn = IPMI_NETFN_STORAGE_REQUEST; /* Storage. */
+		msg.cmd = IPMI_ADD_SEL_ENTRY_CMD;
+		msg.data = data;
+		msg.data_len = 16;
+
+		j = 0;
+		while (*p) {
+			int size = strlen(p);
+
+			if (size > 11)
+				size = 11;
+			data[0] = 0;
+			data[1] = 0;
+			data[2] = 0xf0; /* OEM event without timestamp. */
+			data[3] = intf->my_address;
+			data[4] = j++; /* sequence # */
+			/* Always give 11 bytes, so strncpy will fill
+			   it with zeroes for me. */
+			strncpy(data+5, p, 11);
+			p += size;
+
+			i_ipmi_request(NULL,
+				       intf,
+				       &addr,
+				       0,
+				       &msg,
+				       NULL,
+				       &smi_msg,
+				       &recv_msg,
+				       0,
+				       intf->my_address,
+				       intf->my_lun,
+				       0, 1); /* no retry, and no wait. */
+		}
+	}	
+#endif /* CONFIG_IPMI_PANIC_STRING */
+}
+#endif /* CONFIG_IPMI_PANIC_EVENT */
+
+static int has_paniced = 0;
+
+static int panic_event(struct notifier_block *this,
+		       unsigned long         event,
+                       void                  *ptr)
+{
+	int        i;
+	ipmi_smi_t intf;
+
+	if (has_paniced)
+		return NOTIFY_DONE;
+	has_paniced = 1;
+
+	/* For every registered interface, set it to run to completion. */
+	for (i=0; i<MAX_IPMI_INTERFACES; i++) {
+		intf = ipmi_interfaces[i];
+		if (intf == NULL)
+			continue;
+
+		intf->handlers->set_run_to_completion(intf->send_info, 1);
+	}
+
+#ifdef CONFIG_IPMI_PANIC_EVENT
+	send_panic_events(ptr);
+#endif
+
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block panic_block = {
+	.notifier_call	= panic_event,
+	.next		= NULL,
+	.priority	= 200	/* priority: INT_MAX >= x >= 0 */
+};
+
+static int ipmi_init_msghandler(void)
+{
+	int i;
+
+	if (initialized)
+		return 0;
+
+	printk(KERN_INFO "ipmi message handler version "
+	       IPMI_MSGHANDLER_VERSION "\n");
+
+	for (i=0; i<MAX_IPMI_INTERFACES; i++) {
+		ipmi_interfaces[i] = NULL;
+	}
+
+	proc_ipmi_root = proc_mkdir("ipmi", NULL);
+	if (!proc_ipmi_root) {
+	    printk(KERN_ERR PFX "Unable to create IPMI proc dir");
+	    return -ENOMEM;
+	}
+
+	proc_ipmi_root->owner = THIS_MODULE;
+
+	init_timer(&ipmi_timer);
+	ipmi_timer.data = 0;
+	ipmi_timer.function = ipmi_timeout;
+	ipmi_timer.expires = jiffies + IPMI_TIMEOUT_JIFFIES;
+	add_timer(&ipmi_timer);
+
+	notifier_chain_register(&panic_notifier_list, &panic_block);
+
+	initialized = 1;
+
+	return 0;
+}
+
+static __init int ipmi_init_msghandler_mod(void)
+{
+	ipmi_init_msghandler();
+	return 0;
+}
+
+static __exit void cleanup_ipmi(void)
+{
+	int count;
+
+	if (!initialized)
+		return;
+
+	notifier_chain_unregister(&panic_notifier_list, &panic_block);
+
+	/* This can't be called if any interfaces exist, so no worry about
+	   shutting down the interfaces. */
+
+	/* Tell the timer to stop, then wait for it to stop.  This avoids
+	   problems with race conditions removing the timer here. */
+	stop_operation = 1;
+	while (!timer_stopped) {
+		set_current_state(TASK_UNINTERRUPTIBLE);
+		schedule_timeout(1);
+	}
+
+	remove_proc_entry(proc_ipmi_root->name, &proc_root);
+
+	initialized = 0;
+
+	/* Check for buffer leaks. */
+	count = atomic_read(&smi_msg_inuse_count);
+	if (count != 0)
+		printk(KERN_WARNING PFX "SMI message count %d at exit\n",
+		       count);
+	count = atomic_read(&recv_msg_inuse_count);
+	if (count != 0)
+		printk(KERN_WARNING PFX "recv message count %d at exit\n",
+		       count);
+}
+module_exit(cleanup_ipmi);
+
+module_init(ipmi_init_msghandler_mod);
+MODULE_LICENSE("GPL");
+
+EXPORT_SYMBOL(ipmi_create_user);
+EXPORT_SYMBOL(ipmi_destroy_user);
+EXPORT_SYMBOL(ipmi_get_version);
+EXPORT_SYMBOL(ipmi_request_settime);
+EXPORT_SYMBOL(ipmi_request_supply_msgs);
+EXPORT_SYMBOL(ipmi_register_smi);
+EXPORT_SYMBOL(ipmi_unregister_smi);
+EXPORT_SYMBOL(ipmi_register_for_cmd);
+EXPORT_SYMBOL(ipmi_unregister_for_cmd);
+EXPORT_SYMBOL(ipmi_smi_msg_received);
+EXPORT_SYMBOL(ipmi_smi_watchdog_pretimeout);
+EXPORT_SYMBOL(ipmi_alloc_smi_msg);
+EXPORT_SYMBOL(ipmi_addr_length);
+EXPORT_SYMBOL(ipmi_validate_addr);
+EXPORT_SYMBOL(ipmi_set_gets_events);
+EXPORT_SYMBOL(ipmi_smi_watcher_register);
+EXPORT_SYMBOL(ipmi_smi_watcher_unregister);
+EXPORT_SYMBOL(ipmi_set_my_address);
+EXPORT_SYMBOL(ipmi_get_my_address);
+EXPORT_SYMBOL(ipmi_set_my_LUN);
+EXPORT_SYMBOL(ipmi_get_my_LUN);
+EXPORT_SYMBOL(ipmi_smi_add_proc_entry);
+EXPORT_SYMBOL(ipmi_user_set_run_to_completion);
diff --git a/drivers/char/ipmi/ipmi_poweroff.c b/drivers/char/ipmi/ipmi_poweroff.c
new file mode 100644
index 000000000000..cb5cdc6f14bf
--- /dev/null
+++ b/drivers/char/ipmi/ipmi_poweroff.c
@@ -0,0 +1,549 @@
+/*
+ * ipmi_poweroff.c
+ *
+ * MontaVista IPMI Poweroff extension to sys_reboot
+ *
+ * Author: MontaVista Software, Inc.
+ *         Steven Dake <sdake@mvista.com>
+ *         Corey Minyard <cminyard@mvista.com>
+ *         source@mvista.com
+ *
+ * Copyright 2002,2004 MontaVista Software Inc.
+ *
+ *  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.
+ */
+#include <asm/semaphore.h>
+#include <linux/kdev_t.h>
+#include <linux/module.h>
+#include <linux/string.h>
+#include <linux/ipmi.h>
+#include <linux/ipmi_smi.h>
+
+#define PFX "IPMI poweroff: "
+#define IPMI_POWEROFF_VERSION	"v33"
+
+/* Where to we insert our poweroff function? */
+extern void (*pm_power_off)(void);
+
+/* Stuff from the get device id command. */
+static unsigned int mfg_id;
+static unsigned int prod_id;
+static unsigned char capabilities;
+
+/* We use our own messages for this operation, we don't let the system
+   allocate them, since we may be in a panic situation.  The whole
+   thing is single-threaded, anyway, so multiple messages are not
+   required. */
+static void dummy_smi_free(struct ipmi_smi_msg *msg)
+{
+}
+static void dummy_recv_free(struct ipmi_recv_msg *msg)
+{
+}
+static struct ipmi_smi_msg halt_smi_msg =
+{
+	.done = dummy_smi_free
+};
+static struct ipmi_recv_msg halt_recv_msg =
+{
+	.done = dummy_recv_free
+};
+
+
+/*
+ * Code to send a message and wait for the reponse.
+ */
+
+static void receive_handler(struct ipmi_recv_msg *recv_msg, void *handler_data)
+{
+	struct semaphore *sem = recv_msg->user_msg_data;
+
+	if (sem)
+		up(sem);
+}
+
+static struct ipmi_user_hndl ipmi_poweroff_handler =
+{
+	.ipmi_recv_hndl = receive_handler
+};
+
+
+static int ipmi_request_wait_for_response(ipmi_user_t            user,
+					  struct ipmi_addr       *addr,
+					  struct kernel_ipmi_msg *send_msg)
+{
+	int              rv;
+	struct semaphore sem;
+
+	sema_init (&sem, 0);
+
+	rv = ipmi_request_supply_msgs(user, addr, 0, send_msg, &sem,
+				      &halt_smi_msg, &halt_recv_msg, 0);
+	if (rv)
+		return rv;
+
+	down (&sem);
+
+	return halt_recv_msg.msg.data[0];
+}
+
+/* We are in run-to-completion mode, no semaphore is desired. */
+static int ipmi_request_in_rc_mode(ipmi_user_t            user,
+				   struct ipmi_addr       *addr,
+				   struct kernel_ipmi_msg *send_msg)
+{
+	int              rv;
+
+	rv = ipmi_request_supply_msgs(user, addr, 0, send_msg, NULL,
+				      &halt_smi_msg, &halt_recv_msg, 0);
+	if (rv)
+		return rv;
+
+	return halt_recv_msg.msg.data[0];
+}
+
+/*
+ * ATCA Support
+ */
+
+#define IPMI_NETFN_ATCA			0x2c
+#define IPMI_ATCA_SET_POWER_CMD		0x11
+#define IPMI_ATCA_GET_ADDR_INFO_CMD	0x01
+#define IPMI_PICMG_ID			0
+
+static int ipmi_atca_detect (ipmi_user_t user)
+{
+	struct ipmi_system_interface_addr smi_addr;
+	struct kernel_ipmi_msg            send_msg;
+	int                               rv;
+	unsigned char                     data[1];
+
+        /*
+         * Configure IPMI address for local access
+         */
+        smi_addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
+        smi_addr.channel = IPMI_BMC_CHANNEL;
+        smi_addr.lun = 0;
+
+	/*
+	 * Use get address info to check and see if we are ATCA
+	 */
+	send_msg.netfn = IPMI_NETFN_ATCA;
+	send_msg.cmd = IPMI_ATCA_GET_ADDR_INFO_CMD;
+	data[0] = IPMI_PICMG_ID;
+	send_msg.data = data;
+	send_msg.data_len = sizeof(data);
+	rv = ipmi_request_wait_for_response(user,
+					    (struct ipmi_addr *) &smi_addr,
+					    &send_msg);
+	return !rv;
+}
+
+static void ipmi_poweroff_atca (ipmi_user_t user)
+{
+	struct ipmi_system_interface_addr smi_addr;
+	struct kernel_ipmi_msg            send_msg;
+	int                               rv;
+	unsigned char                     data[4];
+
+        /*
+         * Configure IPMI address for local access
+         */
+        smi_addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
+        smi_addr.channel = IPMI_BMC_CHANNEL;
+        smi_addr.lun = 0;
+
+	printk(KERN_INFO PFX "Powering down via ATCA power command\n");
+
+	/*
+	 * Power down
+	 */
+	send_msg.netfn = IPMI_NETFN_ATCA;
+	send_msg.cmd = IPMI_ATCA_SET_POWER_CMD;
+	data[0] = IPMI_PICMG_ID;
+	data[1] = 0; /* FRU id */
+	data[2] = 0; /* Power Level */
+	data[3] = 0; /* Don't change saved presets */
+	send_msg.data = data;
+	send_msg.data_len = sizeof (data);
+	rv = ipmi_request_in_rc_mode(user,
+				     (struct ipmi_addr *) &smi_addr,
+				     &send_msg);
+	if (rv) {
+		printk(KERN_ERR PFX "Unable to send ATCA powerdown message,"
+		       " IPMI error 0x%x\n", rv);
+		goto out;
+	}
+
+ out:
+	return;
+}
+
+/*
+ * CPI1 Support
+ */
+
+#define IPMI_NETFN_OEM_1				0xf8
+#define OEM_GRP_CMD_SET_RESET_STATE		0x84
+#define OEM_GRP_CMD_SET_POWER_STATE		0x82
+#define IPMI_NETFN_OEM_8				0xf8
+#define OEM_GRP_CMD_REQUEST_HOTSWAP_CTRL	0x80
+#define OEM_GRP_CMD_GET_SLOT_GA			0xa3
+#define IPMI_NETFN_SENSOR_EVT			0x10
+#define IPMI_CMD_GET_EVENT_RECEIVER		0x01
+
+#define IPMI_CPI1_PRODUCT_ID		0x000157
+#define IPMI_CPI1_MANUFACTURER_ID	0x0108
+
+static int ipmi_cpi1_detect (ipmi_user_t user)
+{
+	return ((mfg_id == IPMI_CPI1_MANUFACTURER_ID)
+		&& (prod_id == IPMI_CPI1_PRODUCT_ID));
+}
+
+static void ipmi_poweroff_cpi1 (ipmi_user_t user)
+{
+	struct ipmi_system_interface_addr smi_addr;
+	struct ipmi_ipmb_addr             ipmb_addr;
+	struct kernel_ipmi_msg            send_msg;
+	int                               rv;
+	unsigned char                     data[1];
+	int                               slot;
+	unsigned char                     hotswap_ipmb;
+	unsigned char                     aer_addr;
+	unsigned char                     aer_lun;
+
+        /*
+         * Configure IPMI address for local access
+         */
+        smi_addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
+        smi_addr.channel = IPMI_BMC_CHANNEL;
+        smi_addr.lun = 0;
+
+	printk(KERN_INFO PFX "Powering down via CPI1 power command\n");
+
+	/*
+	 * Get IPMI ipmb address
+	 */
+	send_msg.netfn = IPMI_NETFN_OEM_8 >> 2;
+	send_msg.cmd = OEM_GRP_CMD_GET_SLOT_GA;
+	send_msg.data = NULL;
+	send_msg.data_len = 0;
+	rv = ipmi_request_in_rc_mode(user,
+				     (struct ipmi_addr *) &smi_addr,
+				     &send_msg);
+	if (rv)
+		goto out;
+	slot = halt_recv_msg.msg.data[1];
+	hotswap_ipmb = (slot > 9) ? (0xb0 + 2 * slot) : (0xae + 2 * slot);
+
+	/*
+	 * Get active event receiver
+	 */
+	send_msg.netfn = IPMI_NETFN_SENSOR_EVT >> 2;
+	send_msg.cmd = IPMI_CMD_GET_EVENT_RECEIVER;
+	send_msg.data = NULL;
+	send_msg.data_len = 0;
+	rv = ipmi_request_in_rc_mode(user,
+				     (struct ipmi_addr *) &smi_addr,
+				     &send_msg);
+	if (rv)
+		goto out;
+	aer_addr = halt_recv_msg.msg.data[1];
+	aer_lun = halt_recv_msg.msg.data[2];
+
+	/*
+	 * Setup IPMB address target instead of local target
+	 */
+	ipmb_addr.addr_type = IPMI_IPMB_ADDR_TYPE;
+	ipmb_addr.channel = 0;
+	ipmb_addr.slave_addr = aer_addr;
+	ipmb_addr.lun = aer_lun;
+
+	/*
+	 * Send request hotswap control to remove blade from dpv
+	 */
+	send_msg.netfn = IPMI_NETFN_OEM_8 >> 2;
+	send_msg.cmd = OEM_GRP_CMD_REQUEST_HOTSWAP_CTRL;
+	send_msg.data = &hotswap_ipmb;
+	send_msg.data_len = 1;
+	ipmi_request_in_rc_mode(user,
+				(struct ipmi_addr *) &ipmb_addr,
+				&send_msg);
+
+	/*
+	 * Set reset asserted
+	 */
+	send_msg.netfn = IPMI_NETFN_OEM_1 >> 2;
+	send_msg.cmd = OEM_GRP_CMD_SET_RESET_STATE;
+	send_msg.data = data;
+	data[0] = 1; /* Reset asserted state */
+	send_msg.data_len = 1;
+	rv = ipmi_request_in_rc_mode(user,
+				     (struct ipmi_addr *) &smi_addr,
+				     &send_msg);
+	if (rv)
+		goto out;
+
+	/*
+	 * Power down
+	 */
+	send_msg.netfn = IPMI_NETFN_OEM_1 >> 2;
+	send_msg.cmd = OEM_GRP_CMD_SET_POWER_STATE;
+	send_msg.data = data;
+	data[0] = 1; /* Power down state */
+	send_msg.data_len = 1;
+	rv = ipmi_request_in_rc_mode(user,
+				     (struct ipmi_addr *) &smi_addr,
+				     &send_msg);
+	if (rv)
+		goto out;
+
+ out:
+	return;
+}
+
+/*
+ * Standard chassis support
+ */
+
+#define IPMI_NETFN_CHASSIS_REQUEST	0
+#define IPMI_CHASSIS_CONTROL_CMD	0x02
+
+static int ipmi_chassis_detect (ipmi_user_t user)
+{
+	/* Chassis support, use it. */
+	return (capabilities & 0x80);
+}
+
+static void ipmi_poweroff_chassis (ipmi_user_t user)
+{
+	struct ipmi_system_interface_addr smi_addr;
+	struct kernel_ipmi_msg            send_msg;
+	int                               rv;
+	unsigned char                     data[1];
+
+        /*
+         * Configure IPMI address for local access
+         */
+        smi_addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
+        smi_addr.channel = IPMI_BMC_CHANNEL;
+        smi_addr.lun = 0;
+
+	printk(KERN_INFO PFX "Powering down via IPMI chassis control command\n");
+
+	/*
+	 * Power down
+	 */
+	send_msg.netfn = IPMI_NETFN_CHASSIS_REQUEST;
+	send_msg.cmd = IPMI_CHASSIS_CONTROL_CMD;
+	data[0] = 0; /* Power down */
+	send_msg.data = data;
+	send_msg.data_len = sizeof(data);
+	rv = ipmi_request_in_rc_mode(user,
+				     (struct ipmi_addr *) &smi_addr,
+				     &send_msg);
+	if (rv) {
+		printk(KERN_ERR PFX "Unable to send chassis powerdown message,"
+		       " IPMI error 0x%x\n", rv);
+		goto out;
+	}
+
+ out:
+	return;
+}
+
+
+/* Table of possible power off functions. */
+struct poweroff_function {
+	char *platform_type;
+	int  (*detect)(ipmi_user_t user);
+	void (*poweroff_func)(ipmi_user_t user);
+};
+
+static struct poweroff_function poweroff_functions[] = {
+	{ .platform_type	= "ATCA",
+	  .detect		= ipmi_atca_detect,
+	  .poweroff_func	= ipmi_poweroff_atca },
+	{ .platform_type	= "CPI1",
+	  .detect		= ipmi_cpi1_detect,
+	  .poweroff_func	= ipmi_poweroff_cpi1 },
+	/* Chassis should generally be last, other things should override
+	   it. */
+	{ .platform_type	= "chassis",
+	  .detect		= ipmi_chassis_detect,
+	  .poweroff_func	= ipmi_poweroff_chassis },
+};
+#define NUM_PO_FUNCS (sizeof(poweroff_functions) \
+		      / sizeof(struct poweroff_function))
+
+
+/* Our local state. */
+static int ready = 0;
+static ipmi_user_t ipmi_user;
+static void (*specific_poweroff_func)(ipmi_user_t user) = NULL;
+
+/* Holds the old poweroff function so we can restore it on removal. */
+static void (*old_poweroff_func)(void);
+
+
+/* Called on a powerdown request. */
+static void ipmi_poweroff_function (void)
+{
+	if (!ready)
+		return;
+
+	/* Use run-to-completion mode, since interrupts may be off. */
+	ipmi_user_set_run_to_completion(ipmi_user, 1);
+	specific_poweroff_func(ipmi_user);
+	ipmi_user_set_run_to_completion(ipmi_user, 0);
+}
+
+/* Wait for an IPMI interface to be installed, the first one installed
+   will be grabbed by this code and used to perform the powerdown. */
+static void ipmi_po_new_smi(int if_num)
+{
+	struct ipmi_system_interface_addr smi_addr;
+	struct kernel_ipmi_msg            send_msg;
+	int                               rv;
+	int                               i;
+
+	if (ready)
+		return;
+
+	rv = ipmi_create_user(if_num, &ipmi_poweroff_handler, NULL, &ipmi_user);
+	if (rv) {
+		printk(KERN_ERR PFX "could not create IPMI user, error %d\n",
+		       rv);
+		return;
+	}
+
+        /*
+         * Do a get device ide and store some results, since this is
+	 * used by several functions.
+         */
+        smi_addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
+        smi_addr.channel = IPMI_BMC_CHANNEL;
+        smi_addr.lun = 0;
+
+	send_msg.netfn = IPMI_NETFN_APP_REQUEST;
+	send_msg.cmd = IPMI_GET_DEVICE_ID_CMD;
+	send_msg.data = NULL;
+	send_msg.data_len = 0;
+	rv = ipmi_request_wait_for_response(ipmi_user,
+					    (struct ipmi_addr *) &smi_addr,
+					    &send_msg);
+	if (rv) {
+		printk(KERN_ERR PFX "Unable to send IPMI get device id info,"
+		       " IPMI error 0x%x\n", rv);
+		goto out_err;
+	}
+
+	if (halt_recv_msg.msg.data_len < 12) {
+		printk(KERN_ERR PFX "(chassis) IPMI get device id info too,"
+		       " short, was %d bytes, needed %d bytes\n",
+		       halt_recv_msg.msg.data_len, 12);
+		goto out_err;
+	}
+
+	mfg_id = (halt_recv_msg.msg.data[7]
+		  | (halt_recv_msg.msg.data[8] << 8)
+		  | (halt_recv_msg.msg.data[9] << 16));
+	prod_id = (halt_recv_msg.msg.data[10]
+		   | (halt_recv_msg.msg.data[11] << 8));
+	capabilities = halt_recv_msg.msg.data[6];
+
+
+	/* Scan for a poweroff method */
+	for (i=0; i<NUM_PO_FUNCS; i++) {
+		if (poweroff_functions[i].detect(ipmi_user))
+			goto found;
+	}
+
+ out_err:
+	printk(KERN_ERR PFX "Unable to find a poweroff function that"
+	       " will work, giving up\n");
+	ipmi_destroy_user(ipmi_user);
+	return;
+
+ found:
+	printk(KERN_INFO PFX "Found a %s style poweroff function\n",
+	       poweroff_functions[i].platform_type);
+	specific_poweroff_func = poweroff_functions[i].poweroff_func;
+	old_poweroff_func = pm_power_off;
+	pm_power_off = ipmi_poweroff_function;
+	ready = 1;
+}
+
+static void ipmi_po_smi_gone(int if_num)
+{
+	/* This can never be called, because once poweroff driver is
+	   registered, the interface can't go away until the power
+	   driver is unregistered. */
+}
+
+static struct ipmi_smi_watcher smi_watcher =
+{
+	.owner    = THIS_MODULE,
+	.new_smi  = ipmi_po_new_smi,
+	.smi_gone = ipmi_po_smi_gone
+};
+
+
+/*
+ * Startup and shutdown functions.
+ */
+static int ipmi_poweroff_init (void)
+{
+	int rv;
+
+	printk ("Copyright (C) 2004 MontaVista Software -"
+		" IPMI Powerdown via sys_reboot version "
+		IPMI_POWEROFF_VERSION ".\n");
+
+	rv = ipmi_smi_watcher_register(&smi_watcher);
+	if (rv)
+		printk(KERN_ERR PFX "Unable to register SMI watcher: %d\n", rv);
+
+	return rv;
+}
+
+#ifdef MODULE
+static __exit void ipmi_poweroff_cleanup(void)
+{
+	int rv;
+
+	ipmi_smi_watcher_unregister(&smi_watcher);
+
+	if (ready) {
+		rv = ipmi_destroy_user(ipmi_user);
+		if (rv)
+			printk(KERN_ERR PFX "could not cleanup the IPMI"
+			       " user: 0x%x\n", rv);
+		pm_power_off = old_poweroff_func;
+	}
+}
+module_exit(ipmi_poweroff_cleanup);
+#endif
+
+module_init(ipmi_poweroff_init);
+MODULE_LICENSE("GPL");
diff --git a/drivers/char/ipmi/ipmi_si_intf.c b/drivers/char/ipmi/ipmi_si_intf.c
new file mode 100644
index 000000000000..29de259a981e
--- /dev/null
+++ b/drivers/char/ipmi/ipmi_si_intf.c
@@ -0,0 +1,2359 @@
+/*
+ * 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.
+ *
+ *  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/config.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <asm/system.h>
+#include <linux/sched.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 <asm/irq.h>
+#ifdef CONFIG_HIGH_RES_TIMERS
+#include <linux/hrtime.h>
+# if defined(schedule_next_int)
+/* Old high-res timer code, do translations. */
+#  define get_arch_cycles(a) quick_update_jiffies_sub(a)
+#  define arch_cycles_per_jiffy cycles_per_jiffies
+# endif
+static inline void add_usec_to_timer(struct timer_list *t, long v)
+{
+	t->sub_expires += nsec_to_arch_cycle(v * 1000);
+	while (t->sub_expires >= arch_cycles_per_jiffy)
+	{
+		t->expires++;
+		t->sub_expires -= arch_cycles_per_jiffy;
+	}
+}
+#endif
+#include <linux/interrupt.h>
+#include <linux/rcupdate.h>
+#include <linux/ipmi_smi.h>
+#include <asm/io.h>
+#include "ipmi_si_sm.h"
+#include <linux/init.h>
+
+#define IPMI_SI_VERSION "v33"
+
+/* 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_CLEARING_FLAGS_THEN_SET_IRQ,
+	SI_GETTING_MESSAGES,
+	SI_ENABLE_INTERRUPTS1,
+	SI_ENABLE_INTERRUPTS2
+	/* FIXME - add watchdog stuff. */
+};
+
+enum si_type {
+    SI_KCS, SI_SMIC, SI_BT
+};
+
+struct smi_info
+{
+	ipmi_smi_t             intf;
+	struct si_sm_data      *si_sm;
+	struct si_sm_handlers  *handlers;
+	enum si_type           si_type;
+	spinlock_t             si_lock;
+	spinlock_t             msg_lock;
+	struct list_head       xmit_msgs;
+	struct list_head       hp_xmit_msgs;
+	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;
+
+	/* 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
+	unsigned char       msg_flags;
+
+	/* 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. */
+	int                 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;
+
+	/* The time (in jiffies) the last timeout occurred at. */
+	unsigned long       last_timeout_jiffies;
+
+	/* Used to gracefully stop the timer without race conditions. */
+	volatile int        stop_operation;
+	volatile int        timer_stopped;
+
+	/* 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. */
+	int interrupt_disabled;
+
+	unsigned char ipmi_si_dev_rev;
+	unsigned char ipmi_si_fw_rev_major;
+	unsigned char ipmi_si_fw_rev_minor;
+	unsigned char ipmi_version_major;
+	unsigned char ipmi_version_minor;
+
+	/* Slave address, could be reported from DMI. */
+	unsigned char slave_addr;
+
+	/* Counters and things for the proc filesystem. */
+	spinlock_t count_lock;
+	unsigned long short_timeouts;
+	unsigned long long_timeouts;
+	unsigned long timeout_restarts;
+	unsigned long idles;
+	unsigned long interrupts;
+	unsigned long attentions;
+	unsigned long flag_fetches;
+	unsigned long hosed_count;
+	unsigned long complete_transactions;
+	unsigned long events;
+	unsigned long watchdog_pretimeouts;
+	unsigned long incoming_messages;
+};
+
+static void si_restart_short_timer(struct smi_info *smi_info);
+
+static void deliver_recv_msg(struct smi_info *smi_info,
+			     struct ipmi_smi_msg *msg)
+{
+	/* Deliver the message to the upper layer with the lock
+           released. */
+	spin_unlock(&(smi_info->si_lock));
+	ipmi_smi_msg_received(smi_info->intf, msg);
+	spin_lock(&(smi_info->si_lock));
+}
+
+static void return_hosed_msg(struct smi_info *smi_info)
+{
+	struct ipmi_smi_msg *msg = smi_info->curr_msg;
+
+	/* Make it a reponse */
+	msg->rsp[0] = msg->data[0] | 4;
+	msg->rsp[1] = msg->data[1];
+	msg->rsp[2] = 0xFF; /* Unknown error. */
+	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;
+	struct list_head *entry = NULL;
+#ifdef DEBUG_TIMING
+	struct timeval t;
+#endif
+
+	/* No need to save flags, we aleady have interrupts off and we
+	   already hold the SMI lock. */
+	spin_lock(&(smi_info->msg_lock));
+
+	/* Pick the high priority queue first. */
+	if (! list_empty(&(smi_info->hp_xmit_msgs))) {
+		entry = smi_info->hp_xmit_msgs.next;
+	} else if (! list_empty(&(smi_info->xmit_msgs))) {
+		entry = smi_info->xmit_msgs.next;
+	}
+
+	if (!entry) {
+		smi_info->curr_msg = NULL;
+		rv = SI_SM_IDLE;
+	} else {
+		int err;
+
+		list_del(entry);
+		smi_info->curr_msg = list_entry(entry,
+						struct ipmi_smi_msg,
+						link);
+#ifdef DEBUG_TIMING
+		do_gettimeofday(&t);
+		printk("**Start2: %d.%9.9d\n", t.tv_sec, t.tv_usec);
+#endif
+		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);
+		}
+
+		rv = SI_SM_CALL_WITHOUT_DELAY;
+	}
+	spin_unlock(&(smi_info->msg_lock));
+
+	return rv;
+}
+
+static void start_enable_irq(struct smi_info *smi_info)
+{
+	unsigned char msg[2];
+
+	/* If we are enabling interrupts, we have to tell the
+	   BMC to use them. */
+	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_ENABLE_INTERRUPTS1;
+}
+
+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;
+}
+
+/* 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. */
+static inline void disable_si_irq(struct smi_info *smi_info)
+{
+	if ((smi_info->irq) && (!smi_info->interrupt_disabled)) {
+		disable_irq_nosync(smi_info->irq);
+		smi_info->interrupt_disabled = 1;
+	}
+}
+
+static inline void enable_si_irq(struct smi_info *smi_info)
+{
+	if ((smi_info->irq) && (smi_info->interrupt_disabled)) {
+		enable_irq(smi_info->irq);
+		smi_info->interrupt_disabled = 0;
+	}
+}
+
+static void handle_flags(struct smi_info *smi_info)
+{
+	if (smi_info->msg_flags & WDT_PRE_TIMEOUT_INT) {
+		/* Watchdog pre-timeout */
+		spin_lock(&smi_info->count_lock);
+		smi_info->watchdog_pretimeouts++;
+		spin_unlock(&smi_info->count_lock);
+
+		start_clear_flags(smi_info);
+		smi_info->msg_flags &= ~WDT_PRE_TIMEOUT_INT;
+		spin_unlock(&(smi_info->si_lock));
+		ipmi_smi_watchdog_pretimeout(smi_info->intf);
+		spin_lock(&(smi_info->si_lock));
+	} else if (smi_info->msg_flags & RECEIVE_MSG_AVAIL) {
+		/* Messages available. */
+		smi_info->curr_msg = ipmi_alloc_smi_msg();
+		if (!smi_info->curr_msg) {
+			disable_si_irq(smi_info);
+			smi_info->si_state = SI_NORMAL;
+			return;
+		}
+		enable_si_irq(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;
+	} else if (smi_info->msg_flags & EVENT_MSG_BUFFER_FULL) {
+		/* Events available. */
+		smi_info->curr_msg = ipmi_alloc_smi_msg();
+		if (!smi_info->curr_msg) {
+			disable_si_irq(smi_info);
+			smi_info->si_state = SI_NORMAL;
+			return;
+		}
+		enable_si_irq(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;
+	} else {
+		smi_info->si_state = SI_NORMAL;
+	}
+}
+
+static void handle_transaction_done(struct smi_info *smi_info)
+{
+	struct ipmi_smi_msg *msg;
+#ifdef DEBUG_TIMING
+	struct timeval t;
+
+	do_gettimeofday(&t);
+	printk("**Done: %d.%9.9d\n", t.tv_sec, t.tv_usec);
+#endif
+	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:
+	case SI_CLEARING_FLAGS_THEN_SET_IRQ:
+	{
+		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 */
+			printk(KERN_WARNING
+			       "ipmi_si: Error clearing flags: %2.2x\n",
+			       msg[2]);
+		}
+		if (smi_info->si_state == SI_CLEARING_FLAGS_THEN_SET_IRQ)
+			start_enable_irq(smi_info);
+		else
+			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 {
+			spin_lock(&smi_info->count_lock);
+			smi_info->events++;
+			spin_unlock(&smi_info->count_lock);
+
+			/* 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 {
+			spin_lock(&smi_info->count_lock);
+			smi_info->incoming_messages++;
+			spin_unlock(&smi_info->count_lock);
+
+			/* 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_ENABLE_INTERRUPTS1:
+	{
+		unsigned char msg[4];
+
+		/* 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) {
+			printk(KERN_WARNING
+			       "ipmi_si: Could not enable interrupts"
+			       ", failed get, using polled mode.\n");
+			smi_info->si_state = SI_NORMAL;
+		} else {
+			msg[0] = (IPMI_NETFN_APP_REQUEST << 2);
+			msg[1] = IPMI_SET_BMC_GLOBAL_ENABLES_CMD;
+			msg[2] = msg[3] | 1; /* enable msg queue int */
+			smi_info->handlers->start_transaction(
+				smi_info->si_sm, msg, 3);
+			smi_info->si_state = SI_ENABLE_INTERRUPTS2;
+		}
+		break;
+	}
+
+	case SI_ENABLE_INTERRUPTS2:
+	{
+		unsigned char msg[4];
+
+		/* 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) {
+			printk(KERN_WARNING
+			       "ipmi_si: Could not enable interrupts"
+			       ", failed set, using polled mode.\n");
+		}
+		smi_info->si_state = SI_NORMAL;
+		break;
+	}
+	}
+}
+
+/* Called on timeouts and events.  Timeouts should pass the elapsed
+   time, interrupts should pass in zero. */
+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)
+	{
+		spin_lock(&smi_info->count_lock);
+		smi_info->complete_transactions++;
+		spin_unlock(&smi_info->count_lock);
+
+		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)
+	{
+		spin_lock(&smi_info->count_lock);
+		smi_info->hosed_count++;
+		spin_unlock(&smi_info->count_lock);
+
+		/* 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);
+		}
+		si_sm_result = smi_info->handlers->event(smi_info->si_sm, 0);
+	}
+
+	/* We prefer handling attn over new messages. */
+	if (si_sm_result == SI_SM_ATTN)
+	{
+		unsigned char msg[2];
+
+		spin_lock(&smi_info->count_lock);
+		smi_info->attentions++;
+		spin_unlock(&smi_info->count_lock);
+
+		/* 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) {
+		spin_lock(&smi_info->count_lock);
+		smi_info->idles++;
+		spin_unlock(&smi_info->count_lock);
+
+		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. */
+		unsigned char msg[2];
+
+		spin_lock(&smi_info->count_lock);
+		smi_info->flag_fetches++;
+		spin_unlock(&smi_info->count_lock);
+
+		atomic_set(&smi_info->req_events, 0);
+		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;
+	}
+
+	return si_sm_result;
+}
+
+static void sender(void                *send_info,
+		   struct ipmi_smi_msg *msg,
+		   int                 priority)
+{
+	struct smi_info   *smi_info = send_info;
+	enum si_sm_result result;
+	unsigned long     flags;
+#ifdef DEBUG_TIMING
+	struct timeval    t;
+#endif
+
+	spin_lock_irqsave(&(smi_info->msg_lock), flags);
+#ifdef DEBUG_TIMING
+	do_gettimeofday(&t);
+	printk("**Enqueue: %d.%9.9d\n", t.tv_sec, t.tv_usec);
+#endif
+
+	if (smi_info->run_to_completion) {
+		/* If we are running to completion, then throw it in
+		   the list and run transactions until everything is
+		   clear.  Priority doesn't matter here. */
+		list_add_tail(&(msg->link), &(smi_info->xmit_msgs));
+
+		/* We have to release the msg lock and claim the smi
+		   lock in this case, because of race conditions. */
+		spin_unlock_irqrestore(&(smi_info->msg_lock), flags);
+
+		spin_lock_irqsave(&(smi_info->si_lock), flags);
+		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);
+		}
+		spin_unlock_irqrestore(&(smi_info->si_lock), flags);
+		return;
+	} else {
+		if (priority > 0) {
+			list_add_tail(&(msg->link), &(smi_info->hp_xmit_msgs));
+		} else {
+			list_add_tail(&(msg->link), &(smi_info->xmit_msgs));
+		}
+	}
+	spin_unlock_irqrestore(&(smi_info->msg_lock), flags);
+
+	spin_lock_irqsave(&(smi_info->si_lock), flags);
+	if ((smi_info->si_state == SI_NORMAL)
+	    && (smi_info->curr_msg == NULL))
+	{
+		start_next_msg(smi_info);
+		si_restart_short_timer(smi_info);
+	}
+	spin_unlock_irqrestore(&(smi_info->si_lock), flags);
+}
+
+static void set_run_to_completion(void *send_info, int i_run_to_completion)
+{
+	struct smi_info   *smi_info = send_info;
+	enum si_sm_result result;
+	unsigned long     flags;
+
+	spin_lock_irqsave(&(smi_info->si_lock), flags);
+
+	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);
+		}
+	}
+
+	spin_unlock_irqrestore(&(smi_info->si_lock), flags);
+}
+
+static void poll(void *send_info)
+{
+	struct smi_info *smi_info = send_info;
+
+	smi_event_handler(smi_info, 0);
+}
+
+static void request_events(void *send_info)
+{
+	struct smi_info *smi_info = send_info;
+
+	atomic_set(&smi_info->req_events, 1);
+}
+
+static int initialized = 0;
+
+/* Must be called with interrupts off and with the si_lock held. */
+static void si_restart_short_timer(struct smi_info *smi_info)
+{
+#if defined(CONFIG_HIGH_RES_TIMERS)
+	unsigned long flags;
+	unsigned long jiffies_now;
+
+	if (del_timer(&(smi_info->si_timer))) {
+		/* If we don't delete the timer, then it will go off
+		   immediately, anyway.  So we only process if we
+		   actually delete the timer. */
+
+		/* We already have irqsave on, so no need for it
+                   here. */
+		read_lock(&xtime_lock);
+		jiffies_now = jiffies;
+		smi_info->si_timer.expires = jiffies_now;
+		smi_info->si_timer.sub_expires = get_arch_cycles(jiffies_now);
+
+		add_usec_to_timer(&smi_info->si_timer, SI_SHORT_TIMEOUT_USEC);
+
+		add_timer(&(smi_info->si_timer));
+		spin_lock_irqsave(&smi_info->count_lock, flags);
+		smi_info->timeout_restarts++;
+		spin_unlock_irqrestore(&smi_info->count_lock, flags);
+	}
+#endif
+}
+
+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;
+	unsigned long     time_diff;
+#ifdef DEBUG_TIMING
+	struct timeval    t;
+#endif
+
+	if (smi_info->stop_operation) {
+		smi_info->timer_stopped = 1;
+		return;
+	}
+
+	spin_lock_irqsave(&(smi_info->si_lock), flags);
+#ifdef DEBUG_TIMING
+	do_gettimeofday(&t);
+	printk("**Timer: %d.%9.9d\n", t.tv_sec, t.tv_usec);
+#endif
+	jiffies_now = jiffies;
+	time_diff = ((jiffies_now - smi_info->last_timeout_jiffies)
+		     * SI_USEC_PER_JIFFY);
+	smi_result = smi_event_handler(smi_info, time_diff);
+
+	spin_unlock_irqrestore(&(smi_info->si_lock), flags);
+
+	smi_info->last_timeout_jiffies = jiffies_now;
+
+	if ((smi_info->irq) && (! smi_info->interrupt_disabled)) {
+		/* Running with interrupts, only do long timeouts. */
+		smi_info->si_timer.expires = jiffies + SI_TIMEOUT_JIFFIES;
+		spin_lock_irqsave(&smi_info->count_lock, flags);
+		smi_info->long_timeouts++;
+		spin_unlock_irqrestore(&smi_info->count_lock, flags);
+		goto do_add_timer;
+	}
+
+	/* If the state machine asks for a short delay, then shorten
+           the timer timeout. */
+	if (smi_result == SI_SM_CALL_WITH_DELAY) {
+		spin_lock_irqsave(&smi_info->count_lock, flags);
+		smi_info->short_timeouts++;
+		spin_unlock_irqrestore(&smi_info->count_lock, flags);
+#if defined(CONFIG_HIGH_RES_TIMERS)
+		read_lock(&xtime_lock);
+                smi_info->si_timer.expires = jiffies;
+                smi_info->si_timer.sub_expires
+                        = get_arch_cycles(smi_info->si_timer.expires);
+                read_unlock(&xtime_lock);
+		add_usec_to_timer(&smi_info->si_timer, SI_SHORT_TIMEOUT_USEC);
+#else
+		smi_info->si_timer.expires = jiffies + 1;
+#endif
+	} else {
+		spin_lock_irqsave(&smi_info->count_lock, flags);
+		smi_info->long_timeouts++;
+		spin_unlock_irqrestore(&smi_info->count_lock, flags);
+		smi_info->si_timer.expires = jiffies + SI_TIMEOUT_JIFFIES;
+#if defined(CONFIG_HIGH_RES_TIMERS)
+		smi_info->si_timer.sub_expires = 0;
+#endif
+	}
+
+ do_add_timer:
+	add_timer(&(smi_info->si_timer));
+}
+
+static irqreturn_t si_irq_handler(int irq, void *data, struct pt_regs *regs)
+{
+	struct smi_info *smi_info = data;
+	unsigned long   flags;
+#ifdef DEBUG_TIMING
+	struct timeval  t;
+#endif
+
+	spin_lock_irqsave(&(smi_info->si_lock), flags);
+
+	spin_lock(&smi_info->count_lock);
+	smi_info->interrupts++;
+	spin_unlock(&smi_info->count_lock);
+
+	if (smi_info->stop_operation)
+		goto out;
+
+#ifdef DEBUG_TIMING
+	do_gettimeofday(&t);
+	printk("**Interrupt: %d.%9.9d\n", t.tv_sec, t.tv_usec);
+#endif
+	smi_event_handler(smi_info, 0);
+ out:
+	spin_unlock_irqrestore(&(smi_info->si_lock), flags);
+	return IRQ_HANDLED;
+}
+
+static struct ipmi_smi_handlers handlers =
+{
+	.owner                  = THIS_MODULE,
+	.sender			= sender,
+	.request_events		= request_events,
+	.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 */
+
+#define SI_MAX_PARMS 4
+#define SI_MAX_DRIVERS ((SI_MAX_PARMS * 2) + 2)
+static struct smi_info *smi_infos[SI_MAX_DRIVERS] =
+{ NULL, NULL, NULL, NULL };
+
+#define DEVICE_NAME "ipmi_si"
+
+#define DEFAULT_KCS_IO_PORT	0xca2
+#define DEFAULT_SMIC_IO_PORT	0xca9
+#define DEFAULT_BT_IO_PORT	0xe4
+#define DEFAULT_REGSPACING	1
+
+static int           si_trydefaults = 1;
+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 int num_addrs;
+static unsigned int  ports[SI_MAX_PARMS];
+static int num_ports;
+static int           irqs[SI_MAX_PARMS];
+static int num_irqs;
+static int           regspacings[SI_MAX_PARMS];
+static int num_regspacings = 0;
+static int           regsizes[SI_MAX_PARMS];
+static int num_regsizes = 0;
+static int           regshifts[SI_MAX_PARMS];
+static int num_regshifts = 0;
+static int slave_addrs[SI_MAX_PARMS];
+static int num_slave_addrs = 0;
+
+
+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, long, &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, int, &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.");
+
+
+#define IPMI_MEM_ADDR_SPACE 1
+#define IPMI_IO_ADDR_SPACE  2
+
+#if defined(CONFIG_ACPI_INTERPRETER) || defined(CONFIG_X86) || defined(CONFIG_PCI)
+static int is_new_interface(int intf, u8 addr_space, unsigned long base_addr)
+{
+	int i;
+
+	for (i = 0; i < SI_MAX_PARMS; ++i) {
+		/* Don't check our address. */
+		if (i == intf)
+			continue;
+		if (si_type[i] != NULL) {
+			if ((addr_space == IPMI_MEM_ADDR_SPACE &&
+			     base_addr == addrs[i]) ||
+			    (addr_space == IPMI_IO_ADDR_SPACE &&
+			     base_addr == ports[i]))
+				return 0;
+		}
+		else
+			break;
+	}
+
+	return 1;
+}
+#endif
+
+static int std_irq_setup(struct smi_info *info)
+{
+	int rv;
+
+	if (!info->irq)
+		return 0;
+
+	rv = request_irq(info->irq,
+			 si_irq_handler,
+			 SA_INTERRUPT,
+			 DEVICE_NAME,
+			 info);
+	if (rv) {
+		printk(KERN_WARNING
+		       "ipmi_si: %s unable to claim interrupt %d,"
+		       " running polled\n",
+		       DEVICE_NAME, info->irq);
+		info->irq = 0;
+	} else {
+		printk("  Using irq %d\n", info->irq);
+	}
+
+	return rv;
+}
+
+static void std_irq_cleanup(struct smi_info *info)
+{
+	if (!info->irq)
+		return;
+
+	free_irq(info->irq, info);
+}
+
+static unsigned char port_inb(struct si_sm_io *io, unsigned int offset)
+{
+	unsigned int *addr = io->info;
+
+	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->info;
+
+	outb(b, (*addr)+(offset * io->regspacing));
+}
+
+static unsigned char port_inw(struct si_sm_io *io, unsigned int offset)
+{
+	unsigned int *addr = io->info;
+
+	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->info;
+
+	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->info;
+
+	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->info;
+
+	outl(b << io->regshift, (*addr)+(offset * io->regspacing));
+}
+
+static void port_cleanup(struct smi_info *info)
+{
+	unsigned int *addr = info->io.info;
+	int           mapsize;
+
+	if (addr && (*addr)) {
+		mapsize = ((info->io_size * info->io.regspacing)
+			   - (info->io.regspacing - info->io.regsize));
+
+		release_region (*addr, mapsize);
+	}
+	kfree(info);
+}
+
+static int port_setup(struct smi_info *info)
+{
+	unsigned int *addr = info->io.info;
+	int           mapsize;
+
+	if (!addr || (!*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:
+		printk("ipmi_si: 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_region(*addr, mapsize, DEVICE_NAME) == NULL)
+		return -EIO;
+	return 0;
+}
+
+static int try_init_port(int intf_num, struct smi_info **new_info)
+{
+	struct smi_info *info;
+
+	if (!ports[intf_num])
+		return -ENODEV;
+
+	if (!is_new_interface(intf_num, IPMI_IO_ADDR_SPACE,
+			      ports[intf_num]))
+		return -ENODEV;
+
+	info = kmalloc(sizeof(*info), GFP_KERNEL);
+	if (!info) {
+		printk(KERN_ERR "ipmi_si: Could not allocate SI data (1)\n");
+		return -ENOMEM;
+	}
+	memset(info, 0, sizeof(*info));
+
+	info->io_setup = port_setup;
+	info->io.info = &(ports[intf_num]);
+	info->io.addr = NULL;
+	info->io.regspacing = regspacings[intf_num];
+	if (!info->io.regspacing)
+		info->io.regspacing = DEFAULT_REGSPACING;
+	info->io.regsize = regsizes[intf_num];
+	if (!info->io.regsize)
+		info->io.regsize = DEFAULT_REGSPACING;
+	info->io.regshift = regshifts[intf_num];
+	info->irq = 0;
+	info->irq_setup = NULL;
+	*new_info = info;
+
+	if (si_type[intf_num] == NULL)
+		si_type[intf_num] = "kcs";
+
+	printk("ipmi_si: Trying \"%s\" at I/O port 0x%x\n",
+	       si_type[intf_num], ports[intf_num]);
+	return 0;
+}
+
+static unsigned char mem_inb(struct si_sm_io *io, unsigned int offset)
+{
+	return readb((io->addr)+(offset * io->regspacing));
+}
+
+static void mem_outb(struct si_sm_io *io, unsigned int offset,
+		     unsigned char b)
+{
+	writeb(b, (io->addr)+(offset * io->regspacing));
+}
+
+static unsigned char mem_inw(struct si_sm_io *io, unsigned int offset)
+{
+	return (readw((io->addr)+(offset * io->regspacing)) >> io->regshift)
+		&& 0xff;
+}
+
+static void 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 mem_inl(struct si_sm_io *io, unsigned int offset)
+{
+	return (readl((io->addr)+(offset * io->regspacing)) >> io->regshift)
+		&& 0xff;
+}
+
+static void 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.info;
+	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);
+	}
+	kfree(info);
+}
+
+static int mem_setup(struct smi_info *info)
+{
+	unsigned long *addr = info->io.info;
+	int           mapsize;
+
+	if (!addr || (!*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 = mem_inb;
+		info->io.outputb = mem_outb;
+		break;
+	case 2:
+		info->io.inputb = mem_inw;
+		info->io.outputb = mem_outw;
+		break;
+	case 4:
+		info->io.inputb = mem_inl;
+		info->io.outputb = mem_outl;
+		break;
+#ifdef readq
+	case 8:
+		info->io.inputb = mem_inq;
+		info->io.outputb = mem_outq;
+		break;
+#endif
+	default:
+		printk("ipmi_si: 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;
+}
+
+static int try_init_mem(int intf_num, struct smi_info **new_info)
+{
+	struct smi_info *info;
+
+	if (!addrs[intf_num])
+		return -ENODEV;
+
+	if (!is_new_interface(intf_num, IPMI_MEM_ADDR_SPACE,
+			      addrs[intf_num]))
+		return -ENODEV;
+
+	info = kmalloc(sizeof(*info), GFP_KERNEL);
+	if (!info) {
+		printk(KERN_ERR "ipmi_si: Could not allocate SI data (2)\n");
+		return -ENOMEM;
+	}
+	memset(info, 0, sizeof(*info));
+
+	info->io_setup = mem_setup;
+	info->io.info = &addrs[intf_num];
+	info->io.addr = NULL;
+	info->io.regspacing = regspacings[intf_num];
+	if (!info->io.regspacing)
+		info->io.regspacing = DEFAULT_REGSPACING;
+	info->io.regsize = regsizes[intf_num];
+	if (!info->io.regsize)
+		info->io.regsize = DEFAULT_REGSPACING;
+	info->io.regshift = regshifts[intf_num];
+	info->irq = 0;
+	info->irq_setup = NULL;
+	*new_info = info;
+
+	if (si_type[intf_num] == NULL)
+		si_type[intf_num] = "kcs";
+
+	printk("ipmi_si: Trying \"%s\" at memory address 0x%lx\n",
+	       si_type[intf_num], addrs[intf_num]);
+	return 0;
+}
+
+
+#ifdef CONFIG_ACPI_INTERPRETER
+
+#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 = 0;
+
+/* For GPE-type interrupts. */
+static u32 ipmi_acpi_gpe(void *context)
+{
+	struct smi_info *smi_info = context;
+	unsigned long   flags;
+#ifdef DEBUG_TIMING
+	struct timeval t;
+#endif
+
+	spin_lock_irqsave(&(smi_info->si_lock), flags);
+
+	spin_lock(&smi_info->count_lock);
+	smi_info->interrupts++;
+	spin_unlock(&smi_info->count_lock);
+
+	if (smi_info->stop_operation)
+		goto out;
+
+#ifdef DEBUG_TIMING
+	do_gettimeofday(&t);
+	printk("**ACPI_GPE: %d.%9.9d\n", t.tv_sec, t.tv_usec);
+#endif
+	smi_event_handler(smi_info, 0);
+ out:
+	spin_unlock_irqrestore(&(smi_info->si_lock), flags);
+
+	return ACPI_INTERRUPT_HANDLED;
+}
+
+static int acpi_gpe_irq_setup(struct smi_info *info)
+{
+	acpi_status status;
+
+	if (!info->irq)
+		return 0;
+
+	/* FIXME - is level triggered right? */
+	status = acpi_install_gpe_handler(NULL,
+					  info->irq,
+					  ACPI_GPE_LEVEL_TRIGGERED,
+					  &ipmi_acpi_gpe,
+					  info);
+	if (status != AE_OK) {
+		printk(KERN_WARNING
+		       "ipmi_si: %s unable to claim ACPI GPE %d,"
+		       " running polled\n",
+		       DEVICE_NAME, info->irq);
+		info->irq = 0;
+		return -EINVAL;
+	} else {
+		printk("  Using ACPI GPE %d\n", info->irq);
+		return 0;
+	}
+}
+
+static void acpi_gpe_irq_cleanup(struct smi_info *info)
+{
+	if (!info->irq)
+		return;
+
+	acpi_remove_gpe_handler(NULL, info->irq, &ipmi_acpi_gpe);
+}
+
+/*
+ * Defined at
+ * http://h21007.www2.hp.com/dspp/files/unprotected/devresource/Docs/TechPapers/IA64/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_acpi(int intf_num, struct smi_info **new_info)
+{
+	struct smi_info  *info;
+	acpi_status      status;
+	struct SPMITable *spmi;
+	char             *io_type;
+	u8 		 addr_space;
+
+	if (acpi_failure)
+		return -ENODEV;
+
+	status = acpi_get_firmware_table("SPMI", intf_num+1,
+					 ACPI_LOGICAL_ADDRESSING,
+					 (struct acpi_table_header **) &spmi);
+	if (status != AE_OK) {
+		acpi_failure = 1;
+		return -ENODEV;
+	}
+
+	if (spmi->IPMIlegacy != 1) {
+	    printk(KERN_INFO "IPMI: Bad SPMI legacy %d\n", spmi->IPMIlegacy);
+  	    return -ENODEV;
+	}
+
+	if (spmi->addr.address_space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
+		addr_space = IPMI_MEM_ADDR_SPACE;
+	else
+		addr_space = IPMI_IO_ADDR_SPACE;
+	if (!is_new_interface(-1, addr_space, spmi->addr.address))
+		return -ENODEV;
+
+	if (!spmi->addr.register_bit_width) {
+		acpi_failure = 1;
+		return -ENODEV;
+	}
+
+	/* Figure out the interface type. */
+	switch (spmi->InterfaceType)
+	{
+	case 1:	/* KCS */
+		si_type[intf_num] = "kcs";
+		break;
+
+	case 2:	/* SMIC */
+		si_type[intf_num] = "smic";
+		break;
+
+	case 3:	/* BT */
+		si_type[intf_num] = "bt";
+		break;
+
+	default:
+		printk(KERN_INFO "ipmi_si: Unknown ACPI/SPMI SI type %d\n",
+			spmi->InterfaceType);
+		return -EIO;
+	}
+
+	info = kmalloc(sizeof(*info), GFP_KERNEL);
+	if (!info) {
+		printk(KERN_ERR "ipmi_si: Could not allocate SI data (3)\n");
+		return -ENOMEM;
+	}
+	memset(info, 0, sizeof(*info));
+
+	if (spmi->InterruptType & 1) {
+		/* We've got a GPE interrupt. */
+		info->irq = spmi->GPE;
+		info->irq_setup = acpi_gpe_irq_setup;
+		info->irq_cleanup = acpi_gpe_irq_cleanup;
+	} else if (spmi->InterruptType & 2) {
+		/* We've got an APIC/SAPIC interrupt. */
+		info->irq = spmi->GlobalSystemInterrupt;
+		info->irq_setup = std_irq_setup;
+		info->irq_cleanup = std_irq_cleanup;
+	} else {
+		/* Use the default interrupt setting. */
+		info->irq = 0;
+		info->irq_setup = NULL;
+	}
+
+	regspacings[intf_num] = spmi->addr.register_bit_width / 8;
+	info->io.regspacing = spmi->addr.register_bit_width / 8;
+	regsizes[intf_num] = regspacings[intf_num];
+	info->io.regsize = regsizes[intf_num];
+	regshifts[intf_num] = spmi->addr.register_bit_offset;
+	info->io.regshift = regshifts[intf_num];
+
+	if (spmi->addr.address_space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
+		io_type = "memory";
+		info->io_setup = mem_setup;
+		addrs[intf_num] = spmi->addr.address;
+		info->io.info = &(addrs[intf_num]);
+	} else if (spmi->addr.address_space_id == ACPI_ADR_SPACE_SYSTEM_IO) {
+		io_type = "I/O";
+		info->io_setup = port_setup;
+		ports[intf_num] = spmi->addr.address;
+		info->io.info = &(ports[intf_num]);
+	} else {
+		kfree(info);
+		printk("ipmi_si: Unknown ACPI I/O Address type\n");
+		return -EIO;
+	}
+
+	*new_info = info;
+
+	printk("ipmi_si: ACPI/SPMI specifies \"%s\" %s SI @ 0x%lx\n",
+	       si_type[intf_num], io_type, (unsigned long) spmi->addr.address);
+	return 0;
+}
+#endif
+
+#ifdef CONFIG_X86
+typedef struct dmi_ipmi_data
+{
+	u8   		type;
+	u8   		addr_space;
+	unsigned long	base_addr;
+	u8   		irq;
+	u8              offset;
+	u8              slave_addr;
+} dmi_ipmi_data_t;
+
+static dmi_ipmi_data_t dmi_data[SI_MAX_DRIVERS];
+static int dmi_data_entries;
+
+typedef struct dmi_header
+{
+	u8	type;
+	u8	length;
+	u16	handle;
+} dmi_header_t;
+
+static int decode_dmi(dmi_header_t *dm, int intf_num)
+{
+	u8		*data = (u8 *)dm;
+	unsigned long  	base_addr;
+	u8		reg_spacing;
+	u8              len = dm->length;
+	dmi_ipmi_data_t *ipmi_data = dmi_data+intf_num;
+
+	ipmi_data->type = data[4];
+
+	memcpy(&base_addr, data+8, sizeof(unsigned long));
+	if (len >= 0x11) {
+		if (base_addr & 1) {
+			/* I/O */
+			base_addr &= 0xFFFE;
+			ipmi_data->addr_space = IPMI_IO_ADDR_SPACE;
+		}
+		else {
+			/* Memory */
+			ipmi_data->addr_space = IPMI_MEM_ADDR_SPACE;
+		}
+		/* If bit 4 of byte 0x10 is set, then the lsb for the address
+		   is odd. */
+		ipmi_data->base_addr = base_addr | ((data[0x10] & 0x10) >> 4);
+
+		ipmi_data->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 */
+		    ipmi_data->offset = 1;
+		    break;
+		case 0x01: /* 32-bit boundaries */
+		    ipmi_data->offset = 4;
+		    break;
+		case 0x02: /* 16-byte boundaries */
+		    ipmi_data->offset = 16;
+		    break;
+		default:
+		    /* Some other interface, just ignore it. */
+		    return -EIO;
+		}
+	} else {
+		/* Old DMI spec. */
+		ipmi_data->base_addr = base_addr;
+		ipmi_data->addr_space = IPMI_IO_ADDR_SPACE;
+		ipmi_data->offset = 1;
+	}
+
+	ipmi_data->slave_addr = data[6];
+
+	if (is_new_interface(-1, ipmi_data->addr_space,ipmi_data->base_addr)) {
+		dmi_data_entries++;
+		return 0;
+	}
+
+	memset(ipmi_data, 0, sizeof(dmi_ipmi_data_t));
+
+	return -1;
+}
+
+static int dmi_table(u32 base, int len, int num)
+{
+	u8 		  *buf;
+	struct dmi_header *dm;
+	u8 		  *data;
+	int 		  i=1;
+	int		  status=-1;
+	int               intf_num = 0;
+
+	buf = ioremap(base, len);
+	if(buf==NULL)
+		return -1;
+
+	data = buf;
+
+	while(i<num && (data - buf) < len)
+	{
+		dm=(dmi_header_t *)data;
+
+		if((data-buf+dm->length) >= len)
+        		break;
+
+		if (dm->type == 38) {
+			if (decode_dmi(dm, intf_num) == 0) {
+				intf_num++;
+				if (intf_num >= SI_MAX_DRIVERS)
+					break;
+			}
+		}
+
+	        data+=dm->length;
+		while((data-buf) < len && (*data || data[1]))
+			data++;
+		data+=2;
+		i++;
+	}
+	iounmap(buf);
+
+	return status;
+}
+
+inline static int dmi_checksum(u8 *buf)
+{
+	u8   sum=0;
+	int  a;
+
+	for(a=0; a<15; a++)
+		sum+=buf[a];
+	return (sum==0);
+}
+
+static int dmi_decode(void)
+{
+	u8   buf[15];
+	u32  fp=0xF0000;
+
+#ifdef CONFIG_SIMNOW
+	return -1;
+#endif
+
+	while(fp < 0xFFFFF)
+	{
+		isa_memcpy_fromio(buf, fp, 15);
+		if(memcmp(buf, "_DMI_", 5)==0 && dmi_checksum(buf))
+		{
+			u16 num=buf[13]<<8|buf[12];
+			u16 len=buf[7]<<8|buf[6];
+			u32 base=buf[11]<<24|buf[10]<<16|buf[9]<<8|buf[8];
+
+			if(dmi_table(base, len, num) == 0)
+				return 0;
+		}
+		fp+=16;
+	}
+
+	return -1;
+}
+
+static int try_init_smbios(int intf_num, struct smi_info **new_info)
+{
+	struct smi_info   *info;
+	dmi_ipmi_data_t   *ipmi_data = dmi_data+intf_num;
+	char              *io_type;
+
+	if (intf_num >= dmi_data_entries)
+		return -ENODEV;
+
+	switch(ipmi_data->type) {
+		case 0x01: /* KCS */
+			si_type[intf_num] = "kcs";
+			break;
+		case 0x02: /* SMIC */
+			si_type[intf_num] = "smic";
+			break;
+		case 0x03: /* BT */
+			si_type[intf_num] = "bt";
+			break;
+		default:
+			return -EIO;
+	}
+
+	info = kmalloc(sizeof(*info), GFP_KERNEL);
+	if (!info) {
+		printk(KERN_ERR "ipmi_si: Could not allocate SI data (4)\n");
+		return -ENOMEM;
+	}
+	memset(info, 0, sizeof(*info));
+
+	if (ipmi_data->addr_space == 1) {
+		io_type = "memory";
+		info->io_setup = mem_setup;
+		addrs[intf_num] = ipmi_data->base_addr;
+		info->io.info = &(addrs[intf_num]);
+	} else if (ipmi_data->addr_space == 2) {
+		io_type = "I/O";
+		info->io_setup = port_setup;
+		ports[intf_num] = ipmi_data->base_addr;
+		info->io.info = &(ports[intf_num]);
+	} else {
+		kfree(info);
+		printk("ipmi_si: Unknown SMBIOS I/O Address type.\n");
+		return -EIO;
+	}
+
+	regspacings[intf_num] = ipmi_data->offset;
+	info->io.regspacing = regspacings[intf_num];
+	if (!info->io.regspacing)
+		info->io.regspacing = DEFAULT_REGSPACING;
+	info->io.regsize = DEFAULT_REGSPACING;
+	info->io.regshift = regshifts[intf_num];
+
+	info->slave_addr = ipmi_data->slave_addr;
+
+	irqs[intf_num] = ipmi_data->irq;
+
+	*new_info = info;
+
+	printk("ipmi_si: Found SMBIOS-specified state machine at %s"
+	       " address 0x%lx, slave address 0x%x\n",
+	       io_type, (unsigned long)ipmi_data->base_addr,
+	       ipmi_data->slave_addr);
+	return 0;
+}
+#endif /* CONFIG_X86 */
+
+#ifdef CONFIG_PCI
+
+#define PCI_ERMC_CLASSCODE  0x0C0700
+#define PCI_HP_VENDOR_ID    0x103C
+#define PCI_MMC_DEVICE_ID   0x121A
+#define PCI_MMC_ADDR_CW     0x10
+
+/* Avoid more than one attempt to probe pci smic. */
+static int pci_smic_checked = 0;
+
+static int find_pci_smic(int intf_num, struct smi_info **new_info)
+{
+	struct smi_info  *info;
+	int              error;
+	struct pci_dev   *pci_dev = NULL;
+	u16    		 base_addr;
+	int              fe_rmc = 0;
+
+	if (pci_smic_checked)
+		return -ENODEV;
+
+	pci_smic_checked = 1;
+
+	if ((pci_dev = pci_get_device(PCI_HP_VENDOR_ID, PCI_MMC_DEVICE_ID,
+				       NULL)))
+		;
+	else if ((pci_dev = pci_get_class(PCI_ERMC_CLASSCODE, NULL)) &&
+		 pci_dev->subsystem_vendor == PCI_HP_VENDOR_ID)
+		fe_rmc = 1;
+	else
+		return -ENODEV;
+
+	error = pci_read_config_word(pci_dev, PCI_MMC_ADDR_CW, &base_addr);
+	if (error)
+	{
+		pci_dev_put(pci_dev);
+		printk(KERN_ERR
+		       "ipmi_si: pci_read_config_word() failed (%d).\n",
+		       error);
+		return -ENODEV;
+	}
+
+	/* Bit 0: 1 specifies programmed I/O, 0 specifies memory mapped I/O */
+	if (!(base_addr & 0x0001))
+	{
+		pci_dev_put(pci_dev);
+		printk(KERN_ERR
+		       "ipmi_si: memory mapped I/O not supported for PCI"
+		       " smic.\n");
+		return -ENODEV;
+	}
+
+	base_addr &= 0xFFFE;
+	if (!fe_rmc)
+		/* Data register starts at base address + 1 in eRMC */
+		++base_addr;
+
+	if (!is_new_interface(-1, IPMI_IO_ADDR_SPACE, base_addr)) {
+		pci_dev_put(pci_dev);
+		return -ENODEV;
+	}
+
+	info = kmalloc(sizeof(*info), GFP_KERNEL);
+	if (!info) {
+		pci_dev_put(pci_dev);
+		printk(KERN_ERR "ipmi_si: Could not allocate SI data (5)\n");
+		return -ENOMEM;
+	}
+	memset(info, 0, sizeof(*info));
+
+	info->io_setup = port_setup;
+	ports[intf_num] = base_addr;
+	info->io.info = &(ports[intf_num]);
+	info->io.regspacing = regspacings[intf_num];
+	if (!info->io.regspacing)
+		info->io.regspacing = DEFAULT_REGSPACING;
+	info->io.regsize = DEFAULT_REGSPACING;
+	info->io.regshift = regshifts[intf_num];
+
+	*new_info = info;
+
+	irqs[intf_num] = pci_dev->irq;
+	si_type[intf_num] = "smic";
+
+	printk("ipmi_si: Found PCI SMIC at I/O address 0x%lx\n",
+		(long unsigned int) base_addr);
+
+	pci_dev_put(pci_dev);
+	return 0;
+}
+#endif /* CONFIG_PCI */
+
+static int try_init_plug_and_play(int intf_num, struct smi_info **new_info)
+{
+#ifdef CONFIG_PCI
+	if (find_pci_smic(intf_num, new_info)==0)
+		return 0;
+#endif
+	/* Include other methods here. */
+
+	return -ENODEV;
+}
+
+
+static int try_get_dev_id(struct smi_info *smi_info)
+{
+	unsigned char      msg[2];
+	unsigned char      *resp;
+	unsigned long      resp_len;
+	enum si_sm_result smi_result;
+	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);
+
+	smi_result = smi_info->handlers->event(smi_info->si_sm, 0);
+	for (;;)
+	{
+		if (smi_result == SI_SM_CALL_WITH_DELAY) {
+			set_current_state(TASK_UNINTERRUPTIBLE);
+			schedule_timeout(1);
+			smi_result = smi_info->handlers->event(
+				smi_info->si_sm, 100);
+		}
+		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. */
+		rv = -ENODEV;
+		goto out;
+	}
+
+	/* Otherwise, we got some data. */
+	resp_len = smi_info->handlers->get_result(smi_info->si_sm,
+						  resp, IPMI_MAX_MSG_LENGTH);
+	if (resp_len < 6) {
+		/* That's odd, it should be longer. */
+		rv = -EINVAL;
+		goto out;
+	}
+
+	if ((resp[1] != IPMI_GET_DEVICE_ID_CMD) || (resp[2] != 0)) {
+		/* That's odd, it shouldn't be able to fail. */
+		rv = -EINVAL;
+		goto out;
+	}
+
+	/* Record info from the get device id, in case we need it. */
+	smi_info->ipmi_si_dev_rev = resp[4] & 0xf;
+	smi_info->ipmi_si_fw_rev_major = resp[5] & 0x7f;
+	smi_info->ipmi_si_fw_rev_minor = resp[6];
+	smi_info->ipmi_version_major = resp[7] & 0xf;
+	smi_info->ipmi_version_minor = resp[7] >> 4;
+
+ out:
+	kfree(resp);
+	return rv;
+}
+
+static int type_file_read_proc(char *page, char **start, off_t off,
+			       int count, int *eof, void *data)
+{
+	char            *out = (char *) page;
+	struct smi_info *smi = data;
+
+	switch (smi->si_type) {
+	    case SI_KCS:
+		return sprintf(out, "kcs\n");
+	    case SI_SMIC:
+		return sprintf(out, "smic\n");
+	    case SI_BT:
+		return sprintf(out, "bt\n");
+	    default:
+		return 0;
+	}
+}
+
+static int stat_file_read_proc(char *page, char **start, off_t off,
+			       int count, int *eof, void *data)
+{
+	char            *out = (char *) page;
+	struct smi_info *smi = data;
+
+	out += sprintf(out, "interrupts_enabled:    %d\n",
+		       smi->irq && !smi->interrupt_disabled);
+	out += sprintf(out, "short_timeouts:        %ld\n",
+		       smi->short_timeouts);
+	out += sprintf(out, "long_timeouts:         %ld\n",
+		       smi->long_timeouts);
+	out += sprintf(out, "timeout_restarts:      %ld\n",
+		       smi->timeout_restarts);
+	out += sprintf(out, "idles:                 %ld\n",
+		       smi->idles);
+	out += sprintf(out, "interrupts:            %ld\n",
+		       smi->interrupts);
+	out += sprintf(out, "attentions:            %ld\n",
+		       smi->attentions);
+	out += sprintf(out, "flag_fetches:          %ld\n",
+		       smi->flag_fetches);
+	out += sprintf(out, "hosed_count:           %ld\n",
+		       smi->hosed_count);
+	out += sprintf(out, "complete_transactions: %ld\n",
+		       smi->complete_transactions);
+	out += sprintf(out, "events:                %ld\n",
+		       smi->events);
+	out += sprintf(out, "watchdog_pretimeouts:  %ld\n",
+		       smi->watchdog_pretimeouts);
+	out += sprintf(out, "incoming_messages:     %ld\n",
+		       smi->incoming_messages);
+
+	return (out - ((char *) page));
+}
+
+/* Returns 0 if initialized, or negative on an error. */
+static int init_one_smi(int intf_num, struct smi_info **smi)
+{
+	int		rv;
+	struct smi_info *new_smi;
+
+
+	rv = try_init_mem(intf_num, &new_smi);
+	if (rv)
+		rv = try_init_port(intf_num, &new_smi);
+#ifdef CONFIG_ACPI_INTERPRETER
+	if ((rv) && (si_trydefaults)) {
+		rv = try_init_acpi(intf_num, &new_smi);
+	}
+#endif
+#ifdef CONFIG_X86
+	if ((rv) && (si_trydefaults)) {
+		rv = try_init_smbios(intf_num, &new_smi);
+        }
+#endif
+	if ((rv) && (si_trydefaults)) {
+		rv = try_init_plug_and_play(intf_num, &new_smi);
+	}
+
+
+	if (rv)
+		return rv;
+
+	/* 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;
+
+	if (!new_smi->irq_setup) {
+		new_smi->irq = irqs[intf_num];
+		new_smi->irq_setup = std_irq_setup;
+		new_smi->irq_cleanup = std_irq_cleanup;
+	}
+
+	/* Default to KCS if no type is specified. */
+	if (si_type[intf_num] == NULL) {
+		if (si_trydefaults)
+			si_type[intf_num] = "kcs";
+		else {
+			rv = -EINVAL;
+			goto out_err;
+		}
+	}
+
+	/* Set up the state machine to use. */
+	if (strcmp(si_type[intf_num], "kcs") == 0) {
+		new_smi->handlers = &kcs_smi_handlers;
+		new_smi->si_type = SI_KCS;
+	} else if (strcmp(si_type[intf_num], "smic") == 0) {
+		new_smi->handlers = &smic_smi_handlers;
+		new_smi->si_type = SI_SMIC;
+	} else if (strcmp(si_type[intf_num], "bt") == 0) {
+		new_smi->handlers = &bt_smi_handlers;
+		new_smi->si_type = SI_BT;
+	} else {
+		/* 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(" 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(" Could not set up I/O space\n");
+		goto out_err;
+	}
+
+	spin_lock_init(&(new_smi->si_lock));
+	spin_lock_init(&(new_smi->msg_lock));
+	spin_lock_init(&(new_smi->count_lock));
+
+	/* Do low-level detection first. */
+	if (new_smi->handlers->detect(new_smi->si_sm)) {
+		rv = -ENODEV;
+		goto out_err;
+	}
+
+	/* Attempt a get device id command.  If it fails, we probably
+           don't have a SMI here. */
+	rv = try_get_dev_id(new_smi);
+	if (rv)
+		goto out_err;
+
+	/* Try to claim any interrupts. */
+	new_smi->irq_setup(new_smi);
+
+	INIT_LIST_HEAD(&(new_smi->xmit_msgs));
+	INIT_LIST_HEAD(&(new_smi->hp_xmit_msgs));
+	new_smi->curr_msg = NULL;
+	atomic_set(&new_smi->req_events, 0);
+	new_smi->run_to_completion = 0;
+
+	new_smi->interrupt_disabled = 0;
+	new_smi->timer_stopped = 0;
+	new_smi->stop_operation = 0;
+
+	/* 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. */
+	if (new_smi->irq)
+		new_smi->si_state = SI_CLEARING_FLAGS_THEN_SET_IRQ;
+
+	/* The ipmi_register_smi() code does some operations to
+	   determine the channel information, so we must be ready to
+	   handle operations before it is called.  This means we have
+	   to stop the timer if we get an error after this point. */
+	init_timer(&(new_smi->si_timer));
+	new_smi->si_timer.data = (long) new_smi;
+	new_smi->si_timer.function = smi_timeout;
+	new_smi->last_timeout_jiffies = jiffies;
+	new_smi->si_timer.expires = jiffies + SI_TIMEOUT_JIFFIES;
+	add_timer(&(new_smi->si_timer));
+
+	rv = ipmi_register_smi(&handlers,
+			       new_smi,
+			       new_smi->ipmi_version_major,
+			       new_smi->ipmi_version_minor,
+			       new_smi->slave_addr,
+			       &(new_smi->intf));
+	if (rv) {
+		printk(KERN_ERR
+		       "ipmi_si: Unable to register device: error %d\n",
+		       rv);
+		goto out_err_stop_timer;
+	}
+
+	rv = ipmi_smi_add_proc_entry(new_smi->intf, "type",
+				     type_file_read_proc, NULL,
+				     new_smi, THIS_MODULE);
+	if (rv) {
+		printk(KERN_ERR
+		       "ipmi_si: Unable to create proc entry: %d\n",
+		       rv);
+		goto out_err_stop_timer;
+	}
+
+	rv = ipmi_smi_add_proc_entry(new_smi->intf, "si_stats",
+				     stat_file_read_proc, NULL,
+				     new_smi, THIS_MODULE);
+	if (rv) {
+		printk(KERN_ERR
+		       "ipmi_si: Unable to create proc entry: %d\n",
+		       rv);
+		goto out_err_stop_timer;
+	}
+
+	*smi = new_smi;
+
+	printk(" IPMI %s interface initialized\n", si_type[intf_num]);
+
+	return 0;
+
+ out_err_stop_timer:
+	new_smi->stop_operation = 1;
+
+	/* Wait for the timer to stop.  This avoids problems with race
+	   conditions removing the timer here. */
+	while (!new_smi->timer_stopped) {
+		set_current_state(TASK_UNINTERRUPTIBLE);
+		schedule_timeout(1);
+	}
+
+ out_err:
+	if (new_smi->intf)
+		ipmi_unregister_smi(new_smi->intf);
+
+	new_smi->irq_cleanup(new_smi);
+
+	/* Wait until we know that we are out of any interrupt
+	   handlers might have been running before we freed the
+	   interrupt. */
+	synchronize_kernel();
+
+	if (new_smi->si_sm) {
+		if (new_smi->handlers)
+			new_smi->handlers->cleanup(new_smi->si_sm);
+		kfree(new_smi->si_sm);
+	}
+	new_smi->io_cleanup(new_smi);
+
+	return rv;
+}
+
+static __init int init_ipmi_si(void)
+{
+	int  rv = 0;
+	int  pos = 0;
+	int  i;
+	char *str;
+
+	if (initialized)
+		return 0;
+	initialized = 1;
+
+	/* 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 version "
+	       IPMI_SI_VERSION);
+	if (kcs_smi_handlers.version)
+		printk(", KCS version %s", kcs_smi_handlers.version);
+	if (smic_smi_handlers.version)
+		printk(", SMIC version %s", smic_smi_handlers.version);
+	if (bt_smi_handlers.version)
+   	        printk(", BT version %s", bt_smi_handlers.version);
+	printk("\n");
+
+#ifdef CONFIG_X86
+	dmi_decode();
+#endif
+
+	rv = init_one_smi(0, &(smi_infos[pos]));
+	if (rv && !ports[0] && si_trydefaults) {
+		/* If we are trying defaults and the initial port is
+                   not set, then set it. */
+		si_type[0] = "kcs";
+		ports[0] = DEFAULT_KCS_IO_PORT;
+		rv = init_one_smi(0, &(smi_infos[pos]));
+		if (rv) {
+			/* No KCS - try SMIC */
+			si_type[0] = "smic";
+			ports[0] = DEFAULT_SMIC_IO_PORT;
+			rv = init_one_smi(0, &(smi_infos[pos]));
+		}
+		if (rv) {
+			/* No SMIC - try BT */
+			si_type[0] = "bt";
+			ports[0] = DEFAULT_BT_IO_PORT;
+			rv = init_one_smi(0, &(smi_infos[pos]));
+		}
+	}
+	if (rv == 0)
+		pos++;
+
+	for (i=1; i < SI_MAX_PARMS; i++) {
+		rv = init_one_smi(i, &(smi_infos[pos]));
+		if (rv == 0)
+			pos++;
+	}
+
+	if (smi_infos[0] == NULL) {
+		printk("ipmi_si: Unable to find any System Interface(s)\n");
+		return -ENODEV;
+	}
+
+	return 0;
+}
+module_init(init_ipmi_si);
+
+static void __exit cleanup_one_si(struct smi_info *to_clean)
+{
+	int           rv;
+	unsigned long flags;
+
+	if (! to_clean)
+		return;
+
+	/* Tell the timer and interrupt handlers that we are shutting
+	   down. */
+	spin_lock_irqsave(&(to_clean->si_lock), flags);
+	spin_lock(&(to_clean->msg_lock));
+
+	to_clean->stop_operation = 1;
+
+	to_clean->irq_cleanup(to_clean);
+
+	spin_unlock(&(to_clean->msg_lock));
+	spin_unlock_irqrestore(&(to_clean->si_lock), flags);
+
+	/* Wait until we know that we are out of any interrupt
+	   handlers might have been running before we freed the
+	   interrupt. */
+	synchronize_kernel();
+
+	/* Wait for the timer to stop.  This avoids problems with race
+	   conditions removing the timer here. */
+	while (!to_clean->timer_stopped) {
+		set_current_state(TASK_UNINTERRUPTIBLE);
+		schedule_timeout(1);
+	}
+
+	/* Interrupts and timeouts are stopped, now make sure the
+	   interface is in a clean state. */
+	while ((to_clean->curr_msg) || (to_clean->si_state != SI_NORMAL)) {
+		poll(to_clean);
+		set_current_state(TASK_UNINTERRUPTIBLE);
+		schedule_timeout(1);
+	}
+
+	rv = ipmi_unregister_smi(to_clean->intf);
+	if (rv) {
+		printk(KERN_ERR
+		       "ipmi_si: Unable to unregister device: errno=%d\n",
+		       rv);
+	}
+
+	to_clean->handlers->cleanup(to_clean->si_sm);
+
+	kfree(to_clean->si_sm);
+
+	to_clean->io_cleanup(to_clean);
+}
+
+static __exit void cleanup_ipmi_si(void)
+{
+	int i;
+
+	if (!initialized)
+		return;
+
+	for (i=0; i<SI_MAX_DRIVERS; i++) {
+		cleanup_one_si(smi_infos[i]);
+	}
+}
+module_exit(cleanup_ipmi_si);
+
+MODULE_LICENSE("GPL");
diff --git a/drivers/char/ipmi/ipmi_si_sm.h b/drivers/char/ipmi/ipmi_si_sm.h
new file mode 100644
index 000000000000..a0212b004016
--- /dev/null
+++ b/drivers/char/ipmi/ipmi_si_sm.h
@@ -0,0 +1,120 @@
+/*
+ * ipmi_si_sm.h
+ *
+ * State machine interface for low-level IPMI system management
+ * interface state machines.  This code is the interface between
+ * the ipmi_smi code (that handles the policy of a KCS, SMIC, or
+ * BT interface) and the actual low-level state machine.
+ *
+ * Author: MontaVista Software, Inc.
+ *         Corey Minyard <minyard@mvista.com>
+ *         source@mvista.com
+ *
+ * Copyright 2002 MontaVista Software Inc.
+ *
+ *  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 is defined by the state machines themselves, it is an opaque
+   data type for them to use. */
+struct si_sm_data;
+
+/* The structure for doing I/O in the state machine.  The state
+   machine doesn't have the actual I/O routines, they are done through
+   this interface. */
+struct si_sm_io
+{
+	unsigned char (*inputb)(struct si_sm_io *io, unsigned int offset);
+	void (*outputb)(struct si_sm_io *io,
+			unsigned int  offset,
+			unsigned char b);
+
+	/* Generic info used by the actual handling routines, the
+           state machine shouldn't touch these. */
+	void *info;
+	void *addr;
+	int  regspacing;
+	int  regsize;
+	int  regshift;
+};
+
+/* Results of SMI events. */
+enum si_sm_result
+{
+	SI_SM_CALL_WITHOUT_DELAY, /* Call the driver again immediately */
+	SI_SM_CALL_WITH_DELAY,	/* Delay some before calling again. */
+	SI_SM_TRANSACTION_COMPLETE, /* A transaction is finished. */
+	SI_SM_IDLE,		/* The SM is in idle state. */
+	SI_SM_HOSED,		/* The hardware violated the state machine. */
+	SI_SM_ATTN		/* The hardware is asserting attn and the
+				   state machine is idle. */
+};
+
+/* Handlers for the SMI state machine. */
+struct si_sm_handlers
+{
+	/* Put the version number of the state machine here so the
+           upper layer can print it. */
+	char *version;
+
+	/* Initialize the data and return the amount of I/O space to
+           reserve for the space. */
+	unsigned int (*init_data)(struct si_sm_data *smi,
+				  struct si_sm_io   *io);
+
+	/* Start a new transaction in the state machine.  This will
+	   return -2 if the state machine is not idle, -1 if the size
+	   is invalid (to large or too small), or 0 if the transaction
+	   is successfully completed. */
+	int (*start_transaction)(struct si_sm_data *smi,
+				 unsigned char *data, unsigned int size);
+
+	/* Return the results after the transaction.  This will return
+	   -1 if the buffer is too small, zero if no transaction is
+	   present, or the actual length of the result data. */
+	int (*get_result)(struct si_sm_data *smi,
+			  unsigned char *data, unsigned int length);
+
+	/* Call this periodically (for a polled interface) or upon
+	   receiving an interrupt (for a interrupt-driven interface).
+	   If interrupt driven, you should probably poll this
+	   periodically when not in idle state.  This should be called
+	   with the time that passed since the last call, if it is
+	   significant.  Time is in microseconds. */
+	enum si_sm_result (*event)(struct si_sm_data *smi, long time);
+
+	/* Attempt to detect an SMI.  Returns 0 on success or nonzero
+           on failure. */
+	int (*detect)(struct si_sm_data *smi);
+
+	/* The interface is shutting down, so clean it up. */
+	void (*cleanup)(struct si_sm_data *smi);
+
+	/* Return the size of the SMI structure in bytes. */
+	int (*size)(void);
+};
+
+/* Current state machines that we can use. */
+extern struct si_sm_handlers kcs_smi_handlers;
+extern struct si_sm_handlers smic_smi_handlers;
+extern struct si_sm_handlers bt_smi_handlers;
+
diff --git a/drivers/char/ipmi/ipmi_smic_sm.c b/drivers/char/ipmi/ipmi_smic_sm.c
new file mode 100644
index 000000000000..ae18747e670b
--- /dev/null
+++ b/drivers/char/ipmi/ipmi_smic_sm.c
@@ -0,0 +1,599 @@
+/*
+ * ipmi_smic_sm.c
+ *
+ * The state-machine driver for an IPMI SMIC driver
+ *
+ * It started as a copy of Corey Minyard's driver for the KSC interface
+ * and the kernel patch "mmcdev-patch-245" by HP
+ *
+ * modified by:	Hannes Schulz <schulz@schwaar.com>
+ *		ipmi@schwaar.com
+ *
+ *
+ * Corey Minyard's driver for the KSC interface has the following
+ * copyright notice:
+ *   Copyright 2002 MontaVista Software Inc.
+ *
+ * the kernel patch "mmcdev-patch-245" by HP has the following
+ * copyright notice:
+ * (c) Copyright 2001 Grant Grundler (c) Copyright
+ * 2001 Hewlett-Packard Company
+ *
+ *
+ *  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.  */
+
+#include <linux/kernel.h> /* For printk. */
+#include <linux/string.h>
+#include <linux/ipmi_msgdefs.h>		/* for completion codes */
+#include "ipmi_si_sm.h"
+
+#define IPMI_SMIC_VERSION "v33"
+
+/* smic_debug is a bit-field
+ *	SMIC_DEBUG_ENABLE -	turned on for now
+ *	SMIC_DEBUG_MSG -	commands and their responses
+ *	SMIC_DEBUG_STATES -	state machine
+*/
+#define SMIC_DEBUG_STATES	4
+#define SMIC_DEBUG_MSG		2
+#define	SMIC_DEBUG_ENABLE	1
+
+static int smic_debug = 1;
+
+enum smic_states {
+	SMIC_IDLE,
+	SMIC_START_OP,
+	SMIC_OP_OK,
+	SMIC_WRITE_START,
+	SMIC_WRITE_NEXT,
+	SMIC_WRITE_END,
+	SMIC_WRITE2READ,
+	SMIC_READ_START,
+	SMIC_READ_NEXT,
+	SMIC_READ_END,
+	SMIC_HOSED
+};
+
+#define MAX_SMIC_READ_SIZE 80
+#define MAX_SMIC_WRITE_SIZE 80
+#define SMIC_MAX_ERROR_RETRIES 3
+
+/* Timeouts in microseconds. */
+#define SMIC_RETRY_TIMEOUT 100000
+
+/* SMIC Flags Register Bits */
+#define SMIC_RX_DATA_READY	0x80
+#define SMIC_TX_DATA_READY	0x40
+#define SMIC_SMI		0x10
+#define SMIC_EVM_DATA_AVAIL	0x08
+#define SMIC_SMS_DATA_AVAIL	0x04
+#define SMIC_FLAG_BSY		0x01
+
+/* SMIC Error Codes */
+#define	EC_NO_ERROR		0x00
+#define	EC_ABORTED		0x01
+#define	EC_ILLEGAL_CONTROL	0x02
+#define	EC_NO_RESPONSE		0x03
+#define	EC_ILLEGAL_COMMAND	0x04
+#define	EC_BUFFER_FULL		0x05
+
+struct si_sm_data
+{
+	enum smic_states state;
+	struct si_sm_io *io;
+        unsigned char	 write_data[MAX_SMIC_WRITE_SIZE];
+        int		 write_pos;
+        int		 write_count;
+        int		 orig_write_count;
+        unsigned char	 read_data[MAX_SMIC_READ_SIZE];
+        int		 read_pos;
+        int		 truncated;
+        unsigned int	 error_retries;
+        long		 smic_timeout;
+};
+
+static unsigned int init_smic_data (struct si_sm_data *smic,
+				    struct si_sm_io *io)
+{
+	smic->state = SMIC_IDLE;
+	smic->io = io;
+	smic->write_pos = 0;
+	smic->write_count = 0;
+	smic->orig_write_count = 0;
+	smic->read_pos = 0;
+	smic->error_retries = 0;
+	smic->truncated = 0;
+	smic->smic_timeout = SMIC_RETRY_TIMEOUT;
+
+	/* We use 3 bytes of I/O. */
+	return 3;
+}
+
+static int start_smic_transaction(struct si_sm_data *smic,
+				  unsigned char *data, unsigned int size)
+{
+	unsigned int i;
+
+	if ((size < 2) || (size > MAX_SMIC_WRITE_SIZE)) {
+		return -1;
+	}
+	if ((smic->state != SMIC_IDLE) && (smic->state != SMIC_HOSED)) {
+		return -2;
+	}
+	if (smic_debug & SMIC_DEBUG_MSG) {
+		printk(KERN_INFO "start_smic_transaction -");
+		for (i = 0; i < size; i ++) {
+			printk (" %02x", (unsigned char) (data [i]));
+		}
+		printk ("\n");
+	}
+	smic->error_retries = 0;
+	memcpy(smic->write_data, data, size);
+	smic->write_count = size;
+	smic->orig_write_count = size;
+	smic->write_pos = 0;
+	smic->read_pos = 0;
+	smic->state = SMIC_START_OP;
+	smic->smic_timeout = SMIC_RETRY_TIMEOUT;
+	return 0;
+}
+
+static int smic_get_result(struct si_sm_data *smic,
+			   unsigned char *data, unsigned int length)
+{
+	int i;
+
+	if (smic_debug & SMIC_DEBUG_MSG) {
+		printk (KERN_INFO "smic_get result -");
+		for (i = 0; i < smic->read_pos; i ++) {
+			printk (" %02x", (smic->read_data [i]));
+		}
+		printk ("\n");
+	}
+	if (length < smic->read_pos) {
+		smic->read_pos = length;
+		smic->truncated = 1;
+	}
+	memcpy(data, smic->read_data, smic->read_pos);
+
+	if ((length >= 3) && (smic->read_pos < 3)) {
+		data[2] = IPMI_ERR_UNSPECIFIED;
+		smic->read_pos = 3;
+	}
+	if (smic->truncated) {
+		data[2] = IPMI_ERR_MSG_TRUNCATED;
+		smic->truncated = 0;
+	}
+	return smic->read_pos;
+}
+
+static inline unsigned char read_smic_flags(struct si_sm_data *smic)
+{
+	return smic->io->inputb(smic->io, 2);
+}
+
+static inline unsigned char read_smic_status(struct si_sm_data *smic)
+{
+	return smic->io->inputb(smic->io, 1);
+}
+
+static inline unsigned char read_smic_data(struct si_sm_data *smic)
+{
+	return smic->io->inputb(smic->io, 0);
+}
+
+static inline void write_smic_flags(struct si_sm_data *smic,
+				    unsigned char   flags)
+{
+	smic->io->outputb(smic->io, 2, flags);
+}
+
+static inline void write_smic_control(struct si_sm_data *smic,
+				      unsigned char   control)
+{
+	smic->io->outputb(smic->io, 1, control);
+}
+
+static inline void write_si_sm_data (struct si_sm_data *smic,
+				   unsigned char   data)
+{
+	smic->io->outputb(smic->io, 0, data);
+}
+
+static inline void start_error_recovery(struct si_sm_data *smic, char *reason)
+{
+	(smic->error_retries)++;
+	if (smic->error_retries > SMIC_MAX_ERROR_RETRIES) {
+		if (smic_debug & SMIC_DEBUG_ENABLE) {
+			printk(KERN_WARNING
+			       "ipmi_smic_drv: smic hosed: %s\n", reason);
+		}
+		smic->state = SMIC_HOSED;
+	} else {
+		smic->write_count = smic->orig_write_count;
+		smic->write_pos = 0;
+		smic->read_pos = 0;
+		smic->state = SMIC_START_OP;
+		smic->smic_timeout = SMIC_RETRY_TIMEOUT;
+	}
+}
+
+static inline void write_next_byte(struct si_sm_data *smic)
+{
+	write_si_sm_data(smic, smic->write_data[smic->write_pos]);
+	(smic->write_pos)++;
+	(smic->write_count)--;
+}
+
+static inline void read_next_byte (struct si_sm_data *smic)
+{
+	if (smic->read_pos >= MAX_SMIC_READ_SIZE) {
+		read_smic_data (smic);
+		smic->truncated = 1;
+	} else {
+		smic->read_data[smic->read_pos] = read_smic_data(smic);
+		(smic->read_pos)++;
+	}
+}
+
+/*  SMIC Control/Status Code Components */
+#define	SMIC_GET_STATUS		0x00	/* Control form's name */
+#define	SMIC_READY		0x00	/* Status  form's name */
+#define	SMIC_WR_START		0x01	/* Unified Control/Status names... */
+#define	SMIC_WR_NEXT		0x02
+#define	SMIC_WR_END		0x03
+#define	SMIC_RD_START		0x04
+#define	SMIC_RD_NEXT		0x05
+#define	SMIC_RD_END		0x06
+#define	SMIC_CODE_MASK		0x0f
+
+#define	SMIC_CONTROL		0x00
+#define	SMIC_STATUS		0x80
+#define	SMIC_CS_MASK		0x80
+
+#define	SMIC_SMS		0x40
+#define	SMIC_SMM		0x60
+#define	SMIC_STREAM_MASK	0x60
+
+/*  SMIC Control Codes */
+#define	SMIC_CC_SMS_GET_STATUS	(SMIC_CONTROL|SMIC_SMS|SMIC_GET_STATUS)
+#define	SMIC_CC_SMS_WR_START	(SMIC_CONTROL|SMIC_SMS|SMIC_WR_START)
+#define	SMIC_CC_SMS_WR_NEXT	(SMIC_CONTROL|SMIC_SMS|SMIC_WR_NEXT)
+#define	SMIC_CC_SMS_WR_END	(SMIC_CONTROL|SMIC_SMS|SMIC_WR_END)
+#define	SMIC_CC_SMS_RD_START	(SMIC_CONTROL|SMIC_SMS|SMIC_RD_START)
+#define	SMIC_CC_SMS_RD_NEXT	(SMIC_CONTROL|SMIC_SMS|SMIC_RD_NEXT)
+#define	SMIC_CC_SMS_RD_END	(SMIC_CONTROL|SMIC_SMS|SMIC_RD_END)
+
+#define	SMIC_CC_SMM_GET_STATUS	(SMIC_CONTROL|SMIC_SMM|SMIC_GET_STATUS)
+#define	SMIC_CC_SMM_WR_START	(SMIC_CONTROL|SMIC_SMM|SMIC_WR_START)
+#define	SMIC_CC_SMM_WR_NEXT	(SMIC_CONTROL|SMIC_SMM|SMIC_WR_NEXT)
+#define	SMIC_CC_SMM_WR_END	(SMIC_CONTROL|SMIC_SMM|SMIC_WR_END)
+#define	SMIC_CC_SMM_RD_START	(SMIC_CONTROL|SMIC_SMM|SMIC_RD_START)
+#define	SMIC_CC_SMM_RD_NEXT	(SMIC_CONTROL|SMIC_SMM|SMIC_RD_NEXT)
+#define	SMIC_CC_SMM_RD_END	(SMIC_CONTROL|SMIC_SMM|SMIC_RD_END)
+
+/*  SMIC Status Codes */
+#define	SMIC_SC_SMS_READY	(SMIC_STATUS|SMIC_SMS|SMIC_READY)
+#define	SMIC_SC_SMS_WR_START	(SMIC_STATUS|SMIC_SMS|SMIC_WR_START)
+#define	SMIC_SC_SMS_WR_NEXT	(SMIC_STATUS|SMIC_SMS|SMIC_WR_NEXT)
+#define	SMIC_SC_SMS_WR_END	(SMIC_STATUS|SMIC_SMS|SMIC_WR_END)
+#define	SMIC_SC_SMS_RD_START	(SMIC_STATUS|SMIC_SMS|SMIC_RD_START)
+#define	SMIC_SC_SMS_RD_NEXT	(SMIC_STATUS|SMIC_SMS|SMIC_RD_NEXT)
+#define	SMIC_SC_SMS_RD_END	(SMIC_STATUS|SMIC_SMS|SMIC_RD_END)
+
+#define	SMIC_SC_SMM_READY	(SMIC_STATUS|SMIC_SMM|SMIC_READY)
+#define	SMIC_SC_SMM_WR_START	(SMIC_STATUS|SMIC_SMM|SMIC_WR_START)
+#define	SMIC_SC_SMM_WR_NEXT	(SMIC_STATUS|SMIC_SMM|SMIC_WR_NEXT)
+#define	SMIC_SC_SMM_WR_END	(SMIC_STATUS|SMIC_SMM|SMIC_WR_END)
+#define	SMIC_SC_SMM_RD_START	(SMIC_STATUS|SMIC_SMM|SMIC_RD_START)
+#define	SMIC_SC_SMM_RD_NEXT	(SMIC_STATUS|SMIC_SMM|SMIC_RD_NEXT)
+#define	SMIC_SC_SMM_RD_END	(SMIC_STATUS|SMIC_SMM|SMIC_RD_END)
+
+/* these are the control/status codes we actually use
+	SMIC_CC_SMS_GET_STATUS	0x40
+	SMIC_CC_SMS_WR_START	0x41
+	SMIC_CC_SMS_WR_NEXT	0x42
+	SMIC_CC_SMS_WR_END	0x43
+	SMIC_CC_SMS_RD_START	0x44
+	SMIC_CC_SMS_RD_NEXT	0x45
+	SMIC_CC_SMS_RD_END	0x46
+
+	SMIC_SC_SMS_READY	0xC0
+	SMIC_SC_SMS_WR_START	0xC1
+	SMIC_SC_SMS_WR_NEXT	0xC2
+	SMIC_SC_SMS_WR_END	0xC3
+	SMIC_SC_SMS_RD_START	0xC4
+	SMIC_SC_SMS_RD_NEXT	0xC5
+	SMIC_SC_SMS_RD_END	0xC6
+*/
+
+static enum si_sm_result smic_event (struct si_sm_data *smic, long time)
+{
+	unsigned char status;
+	unsigned char flags;
+	unsigned char data;
+
+	if (smic->state == SMIC_HOSED) {
+		init_smic_data(smic, smic->io);
+		return SI_SM_HOSED;
+	}
+	if (smic->state != SMIC_IDLE) {
+		if (smic_debug & SMIC_DEBUG_STATES) {
+			printk(KERN_INFO
+			       "smic_event - smic->smic_timeout = %ld,"
+			       " time = %ld\n",
+			       smic->smic_timeout, time);
+		}
+/* FIXME: smic_event is sometimes called with time > SMIC_RETRY_TIMEOUT */
+		if (time < SMIC_RETRY_TIMEOUT) {
+			smic->smic_timeout -= time;
+			if (smic->smic_timeout < 0) {
+				start_error_recovery(smic, "smic timed out.");
+				return SI_SM_CALL_WITH_DELAY;
+			}
+		}
+	}
+	flags = read_smic_flags(smic);
+	if (flags & SMIC_FLAG_BSY)
+		return SI_SM_CALL_WITH_DELAY;
+
+	status = read_smic_status (smic);
+	if (smic_debug & SMIC_DEBUG_STATES)
+		printk(KERN_INFO
+		       "smic_event - state = %d, flags = 0x%02x,"
+		       " status = 0x%02x\n",
+		       smic->state, flags, status);
+
+	switch (smic->state) {
+	case SMIC_IDLE:
+		/* in IDLE we check for available messages */
+		if (flags & (SMIC_SMI |
+			     SMIC_EVM_DATA_AVAIL | SMIC_SMS_DATA_AVAIL))
+		{
+			return SI_SM_ATTN;
+		}
+		return SI_SM_IDLE;
+
+	case SMIC_START_OP:
+		/* sanity check whether smic is really idle */
+		write_smic_control(smic, SMIC_CC_SMS_GET_STATUS);
+		write_smic_flags(smic, flags | SMIC_FLAG_BSY);
+		smic->state = SMIC_OP_OK;
+		break;
+
+	case SMIC_OP_OK:
+		if (status != SMIC_SC_SMS_READY) {
+				/* this should not happen */
+			start_error_recovery(smic,
+					     "state = SMIC_OP_OK,"
+					     " status != SMIC_SC_SMS_READY");
+			return SI_SM_CALL_WITH_DELAY;
+		}
+		/* OK so far; smic is idle let us start ... */
+		write_smic_control(smic, SMIC_CC_SMS_WR_START);
+		write_next_byte(smic);
+		write_smic_flags(smic, flags | SMIC_FLAG_BSY);
+		smic->state = SMIC_WRITE_START;
+		break;
+
+	case SMIC_WRITE_START:
+		if (status != SMIC_SC_SMS_WR_START) {
+			start_error_recovery(smic,
+					     "state = SMIC_WRITE_START, "
+					     "status != SMIC_SC_SMS_WR_START");
+			return SI_SM_CALL_WITH_DELAY;
+		}
+		/* we must not issue WR_(NEXT|END) unless
+                   TX_DATA_READY is set */
+		if (flags & SMIC_TX_DATA_READY) {
+			if (smic->write_count == 1) {
+				/* last byte */
+				write_smic_control(smic, SMIC_CC_SMS_WR_END);
+				smic->state = SMIC_WRITE_END;
+			} else {
+				write_smic_control(smic, SMIC_CC_SMS_WR_NEXT);
+				smic->state = SMIC_WRITE_NEXT;
+			}
+			write_next_byte(smic);
+			write_smic_flags(smic, flags | SMIC_FLAG_BSY);
+		}
+		else {
+			return SI_SM_CALL_WITH_DELAY;
+		}
+		break;
+
+	case SMIC_WRITE_NEXT:
+		if (status != SMIC_SC_SMS_WR_NEXT) {
+			start_error_recovery(smic,
+					     "state = SMIC_WRITE_NEXT, "
+					     "status != SMIC_SC_SMS_WR_NEXT");
+			return SI_SM_CALL_WITH_DELAY;
+		}
+		/* this is the same code as in SMIC_WRITE_START */
+		if (flags & SMIC_TX_DATA_READY) {
+			if (smic->write_count == 1) {
+				write_smic_control(smic, SMIC_CC_SMS_WR_END);
+				smic->state = SMIC_WRITE_END;
+			}
+			else {
+				write_smic_control(smic, SMIC_CC_SMS_WR_NEXT);
+				smic->state = SMIC_WRITE_NEXT;
+			}
+			write_next_byte(smic);
+			write_smic_flags(smic, flags | SMIC_FLAG_BSY);
+		}
+		else {
+			return SI_SM_CALL_WITH_DELAY;
+		}
+		break;
+
+	case SMIC_WRITE_END:
+		if (status != SMIC_SC_SMS_WR_END) {
+			start_error_recovery (smic,
+					      "state = SMIC_WRITE_END, "
+					      "status != SMIC_SC_SMS_WR_END");
+			return SI_SM_CALL_WITH_DELAY;
+		}
+		/* data register holds an error code */
+		data = read_smic_data(smic);
+		if (data != 0) {
+			if (smic_debug & SMIC_DEBUG_ENABLE) {
+				printk(KERN_INFO
+				       "SMIC_WRITE_END: data = %02x\n", data);
+			}
+			start_error_recovery(smic,
+					     "state = SMIC_WRITE_END, "
+					     "data != SUCCESS");
+			return SI_SM_CALL_WITH_DELAY;
+		} else {
+			smic->state = SMIC_WRITE2READ;
+		}
+		break;
+
+	case SMIC_WRITE2READ:
+		/* we must wait for RX_DATA_READY to be set before we
+                   can continue */
+		if (flags & SMIC_RX_DATA_READY) {
+			write_smic_control(smic, SMIC_CC_SMS_RD_START);
+			write_smic_flags(smic, flags | SMIC_FLAG_BSY);
+			smic->state = SMIC_READ_START;
+		} else {
+			return SI_SM_CALL_WITH_DELAY;
+		}
+		break;
+
+	case SMIC_READ_START:
+		if (status != SMIC_SC_SMS_RD_START) {
+			start_error_recovery(smic,
+					     "state = SMIC_READ_START, "
+					     "status != SMIC_SC_SMS_RD_START");
+			return SI_SM_CALL_WITH_DELAY;
+		}
+		if (flags & SMIC_RX_DATA_READY) {
+			read_next_byte(smic);
+			write_smic_control(smic, SMIC_CC_SMS_RD_NEXT);
+			write_smic_flags(smic, flags | SMIC_FLAG_BSY);
+			smic->state = SMIC_READ_NEXT;
+		} else {
+			return SI_SM_CALL_WITH_DELAY;
+		}
+		break;
+
+	case SMIC_READ_NEXT:
+		switch (status) {
+		/* smic tells us that this is the last byte to be read
+                   --> clean up */
+		case SMIC_SC_SMS_RD_END:
+			read_next_byte(smic);
+			write_smic_control(smic, SMIC_CC_SMS_RD_END);
+			write_smic_flags(smic, flags | SMIC_FLAG_BSY);
+			smic->state = SMIC_READ_END;
+			break;
+		case SMIC_SC_SMS_RD_NEXT:
+			if (flags & SMIC_RX_DATA_READY) {
+				read_next_byte(smic);
+				write_smic_control(smic, SMIC_CC_SMS_RD_NEXT);
+				write_smic_flags(smic, flags | SMIC_FLAG_BSY);
+				smic->state = SMIC_READ_NEXT;
+			} else {
+				return SI_SM_CALL_WITH_DELAY;
+			}
+			break;
+		default:
+			start_error_recovery(
+				smic,
+				"state = SMIC_READ_NEXT, "
+				"status != SMIC_SC_SMS_RD_(NEXT|END)");
+			return SI_SM_CALL_WITH_DELAY;
+		}
+		break;
+
+	case SMIC_READ_END:
+		if (status != SMIC_SC_SMS_READY) {
+			start_error_recovery(smic,
+					     "state = SMIC_READ_END, "
+					     "status != SMIC_SC_SMS_READY");
+			return SI_SM_CALL_WITH_DELAY;
+		}
+		data = read_smic_data(smic);
+		/* data register holds an error code */
+		if (data != 0) {
+			if (smic_debug & SMIC_DEBUG_ENABLE) {
+				printk(KERN_INFO
+				       "SMIC_READ_END: data = %02x\n", data);
+			}
+			start_error_recovery(smic,
+					     "state = SMIC_READ_END, "
+					     "data != SUCCESS");
+			return SI_SM_CALL_WITH_DELAY;
+		} else {
+			smic->state = SMIC_IDLE;
+			return SI_SM_TRANSACTION_COMPLETE;
+		}
+
+	case SMIC_HOSED:
+		init_smic_data(smic, smic->io);
+		return SI_SM_HOSED;
+
+	default:
+		if (smic_debug & SMIC_DEBUG_ENABLE) {
+			printk(KERN_WARNING "smic->state = %d\n", smic->state);
+			start_error_recovery(smic, "state = UNKNOWN");
+			return SI_SM_CALL_WITH_DELAY;
+		}
+	}
+	smic->smic_timeout = SMIC_RETRY_TIMEOUT;
+	return SI_SM_CALL_WITHOUT_DELAY;
+}
+
+static int smic_detect(struct si_sm_data *smic)
+{
+	/* It's impossible for the SMIC fnags register to be all 1's,
+	   (assuming a properly functioning, self-initialized BMC)
+	   but that's what you get from reading a bogus address, so we
+	   test that first. */
+	if (read_smic_flags(smic) == 0xff)
+		return 1;
+
+	return 0;
+}
+
+static void smic_cleanup(struct si_sm_data *kcs)
+{
+}
+
+static int smic_size(void)
+{
+	return sizeof(struct si_sm_data);
+}
+
+struct si_sm_handlers smic_smi_handlers =
+{
+	.version           = IPMI_SMIC_VERSION,
+	.init_data         = init_smic_data,
+	.start_transaction = start_smic_transaction,
+	.get_result        = smic_get_result,
+	.event             = smic_event,
+	.detect            = smic_detect,
+	.cleanup           = smic_cleanup,
+	.size              = smic_size,
+};
diff --git a/drivers/char/ipmi/ipmi_watchdog.c b/drivers/char/ipmi/ipmi_watchdog.c
new file mode 100644
index 000000000000..fd7093879c66
--- /dev/null
+++ b/drivers/char/ipmi/ipmi_watchdog.c
@@ -0,0 +1,1068 @@
+/*
+ * ipmi_watchdog.c
+ *
+ * A watchdog timer based upon the IPMI interface.
+ *
+ * Author: MontaVista Software, Inc.
+ *         Corey Minyard <minyard@mvista.com>
+ *         source@mvista.com
+ *
+ * Copyright 2002 MontaVista Software Inc.
+ *
+ *  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.
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/ipmi.h>
+#include <linux/ipmi_smi.h>
+#include <linux/watchdog.h>
+#include <linux/miscdevice.h>
+#include <linux/init.h>
+#include <linux/rwsem.h>
+#include <linux/errno.h>
+#include <asm/uaccess.h>
+#include <linux/notifier.h>
+#include <linux/nmi.h>
+#include <linux/reboot.h>
+#include <linux/wait.h>
+#include <linux/poll.h>
+#ifdef CONFIG_X86_LOCAL_APIC
+#include <asm/apic.h>
+#endif
+
+#define	PFX "IPMI Watchdog: "
+
+#define IPMI_WATCHDOG_VERSION "v33"
+
+/*
+ * The IPMI command/response information for the watchdog timer.
+ */
+
+/* values for byte 1 of the set command, byte 2 of the get response. */
+#define WDOG_DONT_LOG		(1 << 7)
+#define WDOG_DONT_STOP_ON_SET	(1 << 6)
+#define WDOG_SET_TIMER_USE(byte, use) \
+	byte = ((byte) & 0xf8) | ((use) & 0x7)
+#define WDOG_GET_TIMER_USE(byte) ((byte) & 0x7)
+#define WDOG_TIMER_USE_BIOS_FRB2	1
+#define WDOG_TIMER_USE_BIOS_POST	2
+#define WDOG_TIMER_USE_OS_LOAD		3
+#define WDOG_TIMER_USE_SMS_OS		4
+#define WDOG_TIMER_USE_OEM		5
+
+/* values for byte 2 of the set command, byte 3 of the get response. */
+#define WDOG_SET_PRETIMEOUT_ACT(byte, use) \
+	byte = ((byte) & 0x8f) | (((use) & 0x7) << 4)
+#define WDOG_GET_PRETIMEOUT_ACT(byte) (((byte) >> 4) & 0x7)
+#define WDOG_PRETIMEOUT_NONE		0
+#define WDOG_PRETIMEOUT_SMI		1
+#define WDOG_PRETIMEOUT_NMI		2
+#define WDOG_PRETIMEOUT_MSG_INT		3
+
+/* Operations that can be performed on a pretimout. */
+#define WDOG_PREOP_NONE		0
+#define WDOG_PREOP_PANIC	1
+#define WDOG_PREOP_GIVE_DATA	2 /* Cause data to be available to
+                                     read.  Doesn't work in NMI
+                                     mode. */
+
+/* Actions to perform on a full timeout. */
+#define WDOG_SET_TIMEOUT_ACT(byte, use) \
+	byte = ((byte) & 0xf8) | ((use) & 0x7)
+#define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7)
+#define WDOG_TIMEOUT_NONE		0
+#define WDOG_TIMEOUT_RESET		1
+#define WDOG_TIMEOUT_POWER_DOWN		2
+#define WDOG_TIMEOUT_POWER_CYCLE	3
+
+/* Byte 3 of the get command, byte 4 of the get response is the
+   pre-timeout in seconds. */
+
+/* Bits for setting byte 4 of the set command, byte 5 of the get response. */
+#define WDOG_EXPIRE_CLEAR_BIOS_FRB2	(1 << 1)
+#define WDOG_EXPIRE_CLEAR_BIOS_POST	(1 << 2)
+#define WDOG_EXPIRE_CLEAR_OS_LOAD	(1 << 3)
+#define WDOG_EXPIRE_CLEAR_SMS_OS	(1 << 4)
+#define WDOG_EXPIRE_CLEAR_OEM		(1 << 5)
+
+/* Setting/getting the watchdog timer value.  This is for bytes 5 and
+   6 (the timeout time) of the set command, and bytes 6 and 7 (the
+   timeout time) and 8 and 9 (the current countdown value) of the
+   response.  The timeout value is given in seconds (in the command it
+   is 100ms intervals). */
+#define WDOG_SET_TIMEOUT(byte1, byte2, val) \
+	(byte1) = (((val) * 10) & 0xff), (byte2) = (((val) * 10) >> 8)
+#define WDOG_GET_TIMEOUT(byte1, byte2) \
+	(((byte1) | ((byte2) << 8)) / 10)
+
+#define IPMI_WDOG_RESET_TIMER		0x22
+#define IPMI_WDOG_SET_TIMER		0x24
+#define IPMI_WDOG_GET_TIMER		0x25
+
+/* These are here until the real ones get into the watchdog.h interface. */
+#ifndef WDIOC_GETTIMEOUT
+#define	WDIOC_GETTIMEOUT        _IOW(WATCHDOG_IOCTL_BASE, 20, int)
+#endif
+#ifndef WDIOC_SET_PRETIMEOUT
+#define	WDIOC_SET_PRETIMEOUT     _IOW(WATCHDOG_IOCTL_BASE, 21, int)
+#endif
+#ifndef WDIOC_GET_PRETIMEOUT
+#define	WDIOC_GET_PRETIMEOUT     _IOW(WATCHDOG_IOCTL_BASE, 22, int)
+#endif
+
+#ifdef CONFIG_WATCHDOG_NOWAYOUT
+static int nowayout = 1;
+#else
+static int nowayout;
+#endif
+
+static ipmi_user_t watchdog_user = NULL;
+
+/* Default the timeout to 10 seconds. */
+static int timeout = 10;
+
+/* The pre-timeout is disabled by default. */
+static int pretimeout = 0;
+
+/* Default action is to reset the board on a timeout. */
+static unsigned char action_val = WDOG_TIMEOUT_RESET;
+
+static char action[16] = "reset";
+
+static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE;
+
+static char preaction[16] = "pre_none";
+
+static unsigned char preop_val = WDOG_PREOP_NONE;
+
+static char preop[16] = "preop_none";
+static DEFINE_SPINLOCK(ipmi_read_lock);
+static char data_to_read = 0;
+static DECLARE_WAIT_QUEUE_HEAD(read_q);
+static struct fasync_struct *fasync_q = NULL;
+static char pretimeout_since_last_heartbeat = 0;
+static char expect_close;
+
+/* If true, the driver will start running as soon as it is configured
+   and ready. */
+static int start_now = 0;
+
+module_param(timeout, int, 0);
+MODULE_PARM_DESC(timeout, "Timeout value in seconds.");
+module_param(pretimeout, int, 0);
+MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds.");
+module_param_string(action, action, sizeof(action), 0);
+MODULE_PARM_DESC(action, "Timeout action. One of: "
+		 "reset, none, power_cycle, power_off.");
+module_param_string(preaction, preaction, sizeof(preaction), 0);
+MODULE_PARM_DESC(preaction, "Pretimeout action.  One of: "
+		 "pre_none, pre_smi, pre_nmi, pre_int.");
+module_param_string(preop, preop, sizeof(preop), 0);
+MODULE_PARM_DESC(preop, "Pretimeout driver operation.  One of: "
+		 "preop_none, preop_panic, preop_give_data.");
+module_param(start_now, int, 0);
+MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as"
+		 "soon as the driver is loaded.");
+module_param(nowayout, int, 0);
+MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started (default=CONFIG_WATCHDOG_NOWAYOUT)");
+
+/* Default state of the timer. */
+static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
+
+/* If shutting down via IPMI, we ignore the heartbeat. */
+static int ipmi_ignore_heartbeat = 0;
+
+/* Is someone using the watchdog?  Only one user is allowed. */
+static unsigned long ipmi_wdog_open = 0;
+
+/* If set to 1, the heartbeat command will set the state to reset and
+   start the timer.  The timer doesn't normally run when the driver is
+   first opened until the heartbeat is set the first time, this
+   variable is used to accomplish this. */
+static int ipmi_start_timer_on_heartbeat = 0;
+
+/* IPMI version of the BMC. */
+static unsigned char ipmi_version_major;
+static unsigned char ipmi_version_minor;
+
+
+static int ipmi_heartbeat(void);
+static void panic_halt_ipmi_heartbeat(void);
+
+
+/* We use a semaphore to make sure that only one thing can send a set
+   timeout at one time, because we only have one copy of the data.
+   The semaphore is claimed when the set_timeout is sent and freed
+   when both messages are free. */
+static atomic_t set_timeout_tofree = ATOMIC_INIT(0);
+static DECLARE_MUTEX(set_timeout_lock);
+static void set_timeout_free_smi(struct ipmi_smi_msg *msg)
+{
+    if (atomic_dec_and_test(&set_timeout_tofree))
+	    up(&set_timeout_lock);
+}
+static void set_timeout_free_recv(struct ipmi_recv_msg *msg)
+{
+    if (atomic_dec_and_test(&set_timeout_tofree))
+	    up(&set_timeout_lock);
+}
+static struct ipmi_smi_msg set_timeout_smi_msg =
+{
+	.done = set_timeout_free_smi
+};
+static struct ipmi_recv_msg set_timeout_recv_msg =
+{
+	.done = set_timeout_free_recv
+};
+ 
+static int i_ipmi_set_timeout(struct ipmi_smi_msg  *smi_msg,
+			      struct ipmi_recv_msg *recv_msg,
+			      int                  *send_heartbeat_now)
+{
+	struct kernel_ipmi_msg            msg;
+	unsigned char                     data[6];
+	int                               rv;
+	struct ipmi_system_interface_addr addr;
+	int                               hbnow = 0;
+
+
+	data[0] = 0;
+	WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS);
+
+	if ((ipmi_version_major > 1)
+	    || ((ipmi_version_major == 1) && (ipmi_version_minor >= 5)))
+	{
+		/* This is an IPMI 1.5-only feature. */
+		data[0] |= WDOG_DONT_STOP_ON_SET;
+	} else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
+		/* In ipmi 1.0, setting the timer stops the watchdog, we
+		   need to start it back up again. */
+		hbnow = 1;
+	}
+
+	data[1] = 0;
+	WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state);
+	if (pretimeout > 0) {
+	    WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val);
+	    data[2] = pretimeout;
+	} else {
+	    WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE);
+	    data[2] = 0; /* No pretimeout. */
+	}
+	data[3] = 0;
+	WDOG_SET_TIMEOUT(data[4], data[5], timeout);
+
+	addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
+	addr.channel = IPMI_BMC_CHANNEL;
+	addr.lun = 0;
+
+	msg.netfn = 0x06;
+	msg.cmd = IPMI_WDOG_SET_TIMER;
+	msg.data = data;
+	msg.data_len = sizeof(data);
+	rv = ipmi_request_supply_msgs(watchdog_user,
+				      (struct ipmi_addr *) &addr,
+				      0,
+				      &msg,
+				      NULL,
+				      smi_msg,
+				      recv_msg,
+				      1);
+	if (rv) {
+		printk(KERN_WARNING PFX "set timeout error: %d\n",
+		       rv);
+	}
+
+	if (send_heartbeat_now)
+	    *send_heartbeat_now = hbnow;
+
+	return rv;
+}
+
+/* Parameters to ipmi_set_timeout */
+#define IPMI_SET_TIMEOUT_NO_HB			0
+#define IPMI_SET_TIMEOUT_HB_IF_NECESSARY	1
+#define IPMI_SET_TIMEOUT_FORCE_HB		2
+
+static int ipmi_set_timeout(int do_heartbeat)
+{
+	int send_heartbeat_now;
+	int rv;
+
+
+	/* We can only send one of these at a time. */
+	down(&set_timeout_lock);
+
+	atomic_set(&set_timeout_tofree, 2);
+
+	rv = i_ipmi_set_timeout(&set_timeout_smi_msg,
+				&set_timeout_recv_msg,
+				&send_heartbeat_now);
+	if (rv) {
+		up(&set_timeout_lock);
+	} else {
+		if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB)
+		    || ((send_heartbeat_now)
+			&& (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY)))
+		{
+			rv = ipmi_heartbeat();
+		}
+	}
+
+	return rv;
+}
+
+static void dummy_smi_free(struct ipmi_smi_msg *msg)
+{
+}
+static void dummy_recv_free(struct ipmi_recv_msg *msg)
+{
+}
+static struct ipmi_smi_msg panic_halt_smi_msg =
+{
+	.done = dummy_smi_free
+};
+static struct ipmi_recv_msg panic_halt_recv_msg =
+{
+	.done = dummy_recv_free
+};
+
+/* Special call, doesn't claim any locks.  This is only to be called
+   at panic or halt time, in run-to-completion mode, when the caller
+   is the only CPU and the only thing that will be going is these IPMI
+   calls. */
+static void panic_halt_ipmi_set_timeout(void)
+{
+	int send_heartbeat_now;
+	int rv;
+
+	rv = i_ipmi_set_timeout(&panic_halt_smi_msg,
+				&panic_halt_recv_msg,
+				&send_heartbeat_now);
+	if (!rv) {
+		if (send_heartbeat_now)
+			panic_halt_ipmi_heartbeat();
+	}
+}
+
+/* We use a semaphore to make sure that only one thing can send a
+   heartbeat at one time, because we only have one copy of the data.
+   The semaphore is claimed when the set_timeout is sent and freed
+   when both messages are free. */
+static atomic_t heartbeat_tofree = ATOMIC_INIT(0);
+static DECLARE_MUTEX(heartbeat_lock);
+static DECLARE_MUTEX_LOCKED(heartbeat_wait_lock);
+static void heartbeat_free_smi(struct ipmi_smi_msg *msg)
+{
+    if (atomic_dec_and_test(&heartbeat_tofree))
+	    up(&heartbeat_wait_lock);
+}
+static void heartbeat_free_recv(struct ipmi_recv_msg *msg)
+{
+    if (atomic_dec_and_test(&heartbeat_tofree))
+	    up(&heartbeat_wait_lock);
+}
+static struct ipmi_smi_msg heartbeat_smi_msg =
+{
+	.done = heartbeat_free_smi
+};
+static struct ipmi_recv_msg heartbeat_recv_msg =
+{
+	.done = heartbeat_free_recv
+};
+ 
+static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg =
+{
+	.done = dummy_smi_free
+};
+static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg =
+{
+	.done = dummy_recv_free
+};
+ 
+static int ipmi_heartbeat(void)
+{
+	struct kernel_ipmi_msg            msg;
+	int                               rv;
+	struct ipmi_system_interface_addr addr;
+
+	if (ipmi_ignore_heartbeat) {
+		return 0;
+	}
+
+	if (ipmi_start_timer_on_heartbeat) {
+		ipmi_start_timer_on_heartbeat = 0;
+		ipmi_watchdog_state = action_val;
+		return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
+	} else if (pretimeout_since_last_heartbeat) {
+		/* A pretimeout occurred, make sure we set the timeout.
+		   We don't want to set the action, though, we want to
+		   leave that alone (thus it can't be combined with the
+		   above operation. */
+		pretimeout_since_last_heartbeat = 0;
+		return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
+	}
+
+	down(&heartbeat_lock);
+
+	atomic_set(&heartbeat_tofree, 2);
+
+	/* Don't reset the timer if we have the timer turned off, that
+           re-enables the watchdog. */
+	if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) {
+		up(&heartbeat_lock);
+		return 0;
+	}
+
+	addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
+	addr.channel = IPMI_BMC_CHANNEL;
+	addr.lun = 0;
+
+	msg.netfn = 0x06;
+	msg.cmd = IPMI_WDOG_RESET_TIMER;
+	msg.data = NULL;
+	msg.data_len = 0;
+	rv = ipmi_request_supply_msgs(watchdog_user,
+				      (struct ipmi_addr *) &addr,
+				      0,
+				      &msg,
+				      NULL,
+				      &heartbeat_smi_msg,
+				      &heartbeat_recv_msg,
+				      1);
+	if (rv) {
+		up(&heartbeat_lock);
+		printk(KERN_WARNING PFX "heartbeat failure: %d\n",
+		       rv);
+		return rv;
+	}
+
+	/* Wait for the heartbeat to be sent. */
+	down(&heartbeat_wait_lock);
+
+	if (heartbeat_recv_msg.msg.data[0] != 0) {
+	    /* Got an error in the heartbeat response.  It was already
+	       reported in ipmi_wdog_msg_handler, but we should return
+	       an error here. */
+	    rv = -EINVAL;
+	}
+
+	up(&heartbeat_lock);
+
+	return rv;
+}
+
+static void panic_halt_ipmi_heartbeat(void)
+{
+	struct kernel_ipmi_msg             msg;
+	struct ipmi_system_interface_addr addr;
+
+
+	/* Don't reset the timer if we have the timer turned off, that
+           re-enables the watchdog. */
+	if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
+		return;
+
+	addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
+	addr.channel = IPMI_BMC_CHANNEL;
+	addr.lun = 0;
+
+	msg.netfn = 0x06;
+	msg.cmd = IPMI_WDOG_RESET_TIMER;
+	msg.data = NULL;
+	msg.data_len = 0;
+	ipmi_request_supply_msgs(watchdog_user,
+				 (struct ipmi_addr *) &addr,
+				 0,
+				 &msg,
+				 NULL,
+				 &panic_halt_heartbeat_smi_msg,
+				 &panic_halt_heartbeat_recv_msg,
+				 1);
+}
+
+static struct watchdog_info ident=
+{
+	.options	= 0,	/* WDIOF_SETTIMEOUT, */
+	.firmware_version = 1,
+	.identity	= "IPMI"
+};
+
+static int ipmi_ioctl(struct inode *inode, struct file *file,
+		      unsigned int cmd, unsigned long arg)
+{
+	void __user *argp = (void __user *)arg;
+	int i;
+	int val;
+
+	switch(cmd) {
+	case WDIOC_GETSUPPORT:
+		i = copy_to_user(argp, &ident, sizeof(ident));
+		return i ? -EFAULT : 0;
+
+	case WDIOC_SETTIMEOUT:
+		i = copy_from_user(&val, argp, sizeof(int));
+		if (i)
+			return -EFAULT;
+		timeout = val;
+		return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
+
+	case WDIOC_GETTIMEOUT:
+		i = copy_to_user(argp, &timeout, sizeof(timeout));
+		if (i)
+			return -EFAULT;
+		return 0;
+
+	case WDIOC_SET_PRETIMEOUT:
+		i = copy_from_user(&val, argp, sizeof(int));
+		if (i)
+			return -EFAULT;
+		pretimeout = val;
+		return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
+
+	case WDIOC_GET_PRETIMEOUT:
+		i = copy_to_user(argp, &pretimeout, sizeof(pretimeout));
+		if (i)
+			return -EFAULT;
+		return 0;
+
+	case WDIOC_KEEPALIVE:
+		return ipmi_heartbeat();
+
+	case WDIOC_SETOPTIONS:
+		i = copy_from_user(&val, argp, sizeof(int));
+		if (i)
+			return -EFAULT;
+		if (val & WDIOS_DISABLECARD)
+		{
+			ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
+			ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
+			ipmi_start_timer_on_heartbeat = 0;
+		}
+
+		if (val & WDIOS_ENABLECARD)
+		{
+			ipmi_watchdog_state = action_val;
+			ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
+		}
+		return 0;
+
+	case WDIOC_GETSTATUS:
+		val = 0;
+		i = copy_to_user(argp, &val, sizeof(val));
+		if (i)
+			return -EFAULT;
+		return 0;
+
+	default:
+		return -ENOIOCTLCMD;
+	}
+}
+
+static ssize_t ipmi_write(struct file *file,
+			  const char  __user *buf,
+			  size_t      len,
+			  loff_t      *ppos)
+{
+	int rv;
+
+	if (len) {
+	    	if (!nowayout) {
+		    	size_t i;
+
+			/* In case it was set long ago */
+			expect_close = 0;
+
+    			for (i = 0; i != len; i++) {
+				char c;
+
+				if (get_user(c, buf + i))
+					return -EFAULT;
+				if (c == 'V')
+					expect_close = 42;
+			}
+		}
+		rv = ipmi_heartbeat();
+		if (rv)
+			return rv;
+		return 1;
+	}
+	return 0;
+}
+
+static ssize_t ipmi_read(struct file *file,
+			 char        __user *buf,
+			 size_t      count,
+			 loff_t      *ppos)
+{
+	int          rv = 0;
+	wait_queue_t wait;
+
+	if (count <= 0)
+		return 0;
+
+	/* Reading returns if the pretimeout has gone off, and it only does
+	   it once per pretimeout. */
+	spin_lock(&ipmi_read_lock);
+	if (!data_to_read) {
+		if (file->f_flags & O_NONBLOCK) {
+			rv = -EAGAIN;
+			goto out;
+		}
+		
+		init_waitqueue_entry(&wait, current);
+		add_wait_queue(&read_q, &wait);
+		while (!data_to_read) {
+			set_current_state(TASK_INTERRUPTIBLE);
+			spin_unlock(&ipmi_read_lock);
+			schedule();
+			spin_lock(&ipmi_read_lock);
+		}
+		remove_wait_queue(&read_q, &wait);
+	    
+		if (signal_pending(current)) {
+			rv = -ERESTARTSYS;
+			goto out;
+		}
+	}
+	data_to_read = 0;
+
+ out:
+	spin_unlock(&ipmi_read_lock);
+
+	if (rv == 0) {
+		if (copy_to_user(buf, &data_to_read, 1))
+			rv = -EFAULT;
+		else
+			rv = 1;
+	}
+
+	return rv;
+}
+
+static int ipmi_open(struct inode *ino, struct file *filep)
+{
+        switch (iminor(ino))
+        {
+                case WATCHDOG_MINOR:
+		    if(test_and_set_bit(0, &ipmi_wdog_open))
+                        return -EBUSY;
+
+		    /* Don't start the timer now, let it start on the
+		       first heartbeat. */
+		    ipmi_start_timer_on_heartbeat = 1;
+                    return nonseekable_open(ino, filep);
+
+                default:
+                    return (-ENODEV);
+        }
+}
+
+static unsigned int ipmi_poll(struct file *file, poll_table *wait)
+{
+	unsigned int mask = 0;
+	
+	poll_wait(file, &read_q, wait);
+
+	spin_lock(&ipmi_read_lock);
+	if (data_to_read)
+		mask |= (POLLIN | POLLRDNORM);
+	spin_unlock(&ipmi_read_lock);
+
+	return mask;
+}
+
+static int ipmi_fasync(int fd, struct file *file, int on)
+{
+	int result;
+
+	result = fasync_helper(fd, file, on, &fasync_q);
+
+	return (result);
+}
+
+static int ipmi_close(struct inode *ino, struct file *filep)
+{
+	if (iminor(ino)==WATCHDOG_MINOR)
+	{
+		if (expect_close == 42) {
+			ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
+			ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
+			clear_bit(0, &ipmi_wdog_open);
+		} else {
+			printk(KERN_CRIT PFX "Unexpected close, not stopping watchdog!\n");
+			ipmi_heartbeat();
+		}
+	}
+
+	ipmi_fasync (-1, filep, 0);
+	expect_close = 0;
+
+	return 0;
+}
+
+static struct file_operations ipmi_wdog_fops = {
+	.owner   = THIS_MODULE,
+	.read    = ipmi_read,
+	.poll    = ipmi_poll,
+	.write   = ipmi_write,
+	.ioctl   = ipmi_ioctl,
+	.open    = ipmi_open,
+	.release = ipmi_close,
+	.fasync  = ipmi_fasync,
+};
+
+static struct miscdevice ipmi_wdog_miscdev = {
+	.minor		= WATCHDOG_MINOR,
+	.name		= "watchdog",
+	.fops		= &ipmi_wdog_fops
+};
+
+static DECLARE_RWSEM(register_sem);
+
+static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg,
+				  void                 *handler_data)
+{
+	if (msg->msg.data[0] != 0) {
+		printk(KERN_ERR PFX "response: Error %x on cmd %x\n",
+		       msg->msg.data[0],
+		       msg->msg.cmd);
+	}
+	
+	ipmi_free_recv_msg(msg);
+}
+
+static void ipmi_wdog_pretimeout_handler(void *handler_data)
+{
+	if (preaction_val != WDOG_PRETIMEOUT_NONE) {
+		if (preop_val == WDOG_PREOP_PANIC)
+			panic("Watchdog pre-timeout");
+		else if (preop_val == WDOG_PREOP_GIVE_DATA) {
+			spin_lock(&ipmi_read_lock);
+			data_to_read = 1;
+			wake_up_interruptible(&read_q);
+			kill_fasync(&fasync_q, SIGIO, POLL_IN);
+
+			spin_unlock(&ipmi_read_lock);
+		}
+	}
+
+	/* On some machines, the heartbeat will give
+	   an error and not work unless we re-enable
+	   the timer.   So do so. */
+	pretimeout_since_last_heartbeat = 1;
+}
+
+static struct ipmi_user_hndl ipmi_hndlrs =
+{
+	.ipmi_recv_hndl           = ipmi_wdog_msg_handler,
+	.ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler
+};
+
+static void ipmi_register_watchdog(int ipmi_intf)
+{
+	int rv = -EBUSY;
+
+	down_write(&register_sem);
+	if (watchdog_user)
+		goto out;
+
+	rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user);
+	if (rv < 0) {
+		printk(KERN_CRIT PFX "Unable to register with ipmi\n");
+		goto out;
+	}
+
+	ipmi_get_version(watchdog_user,
+			 &ipmi_version_major,
+			 &ipmi_version_minor);
+
+	rv = misc_register(&ipmi_wdog_miscdev);
+	if (rv < 0) {
+		ipmi_destroy_user(watchdog_user);
+		watchdog_user = NULL;
+		printk(KERN_CRIT PFX "Unable to register misc device\n");
+	}
+
+ out:
+	up_write(&register_sem);
+
+	if ((start_now) && (rv == 0)) {
+		/* Run from startup, so start the timer now. */
+		start_now = 0; /* Disable this function after first startup. */
+		ipmi_watchdog_state = action_val;
+		ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
+		printk(KERN_INFO PFX "Starting now!\n");
+	}
+}
+
+#ifdef HAVE_NMI_HANDLER
+static int
+ipmi_nmi(void *dev_id, struct pt_regs *regs, int cpu, int handled)
+{
+	/* If no one else handled the NMI, we assume it was the IPMI
+           watchdog. */
+	if ((!handled) && (preop_val == WDOG_PREOP_PANIC))
+		panic(PFX "pre-timeout");
+
+	/* On some machines, the heartbeat will give
+	   an error and not work unless we re-enable
+	   the timer.   So do so. */
+	pretimeout_since_last_heartbeat = 1;
+
+	return NOTIFY_DONE;
+}
+
+static struct nmi_handler ipmi_nmi_handler =
+{
+	.link     = LIST_HEAD_INIT(ipmi_nmi_handler.link),
+	.dev_name = "ipmi_watchdog",
+	.dev_id   = NULL,
+	.handler  = ipmi_nmi,
+	.priority = 0, /* Call us last. */
+};
+#endif
+
+static int wdog_reboot_handler(struct notifier_block *this,
+			       unsigned long         code,
+			       void                  *unused)
+{
+	static int reboot_event_handled = 0;
+
+	if ((watchdog_user) && (!reboot_event_handled)) {
+		/* Make sure we only do this once. */
+		reboot_event_handled = 1;
+
+		if (code == SYS_DOWN || code == SYS_HALT) {
+			/* Disable the WDT if we are shutting down. */
+			ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
+			panic_halt_ipmi_set_timeout();
+		} else {
+			/* Set a long timer to let the reboot happens, but
+			   reboot if it hangs. */
+			timeout = 120;
+			pretimeout = 0;
+			ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
+			panic_halt_ipmi_set_timeout();
+		}
+	}
+	return NOTIFY_OK;
+}
+
+static struct notifier_block wdog_reboot_notifier = {
+	.notifier_call	= wdog_reboot_handler,
+	.next		= NULL,
+	.priority	= 0
+};
+
+static int wdog_panic_handler(struct notifier_block *this,
+			      unsigned long         event,
+			      void                  *unused)
+{
+	static int panic_event_handled = 0;
+
+	/* On a panic, if we have a panic timeout, make sure that the thing
+	   reboots, even if it hangs during that panic. */
+	if (watchdog_user && !panic_event_handled) {
+		/* Make sure the panic doesn't hang, and make sure we
+		   do this only once. */
+		panic_event_handled = 1;
+	    
+		timeout = 255;
+		pretimeout = 0;
+		ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
+		panic_halt_ipmi_set_timeout();
+	}
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block wdog_panic_notifier = {
+	.notifier_call	= wdog_panic_handler,
+	.next		= NULL,
+	.priority	= 150	/* priority: INT_MAX >= x >= 0 */
+};
+
+
+static void ipmi_new_smi(int if_num)
+{
+	ipmi_register_watchdog(if_num);
+}
+
+static void ipmi_smi_gone(int if_num)
+{
+	/* This can never be called, because once the watchdog is
+	   registered, the interface can't go away until the watchdog
+	   is unregistered. */
+}
+
+static struct ipmi_smi_watcher smi_watcher =
+{
+	.owner    = THIS_MODULE,
+	.new_smi  = ipmi_new_smi,
+	.smi_gone = ipmi_smi_gone
+};
+
+static int __init ipmi_wdog_init(void)
+{
+	int rv;
+
+	printk(KERN_INFO PFX "driver version "
+	       IPMI_WATCHDOG_VERSION "\n");
+
+	if (strcmp(action, "reset") == 0) {
+		action_val = WDOG_TIMEOUT_RESET;
+	} else if (strcmp(action, "none") == 0) {
+		action_val = WDOG_TIMEOUT_NONE;
+	} else if (strcmp(action, "power_cycle") == 0) {
+		action_val = WDOG_TIMEOUT_POWER_CYCLE;
+	} else if (strcmp(action, "power_off") == 0) {
+		action_val = WDOG_TIMEOUT_POWER_DOWN;
+	} else {
+		action_val = WDOG_TIMEOUT_RESET;
+		printk(KERN_INFO PFX "Unknown action '%s', defaulting to"
+		       " reset\n", action);
+	}
+
+	if (strcmp(preaction, "pre_none") == 0) {
+		preaction_val = WDOG_PRETIMEOUT_NONE;
+	} else if (strcmp(preaction, "pre_smi") == 0) {
+		preaction_val = WDOG_PRETIMEOUT_SMI;
+#ifdef HAVE_NMI_HANDLER
+	} else if (strcmp(preaction, "pre_nmi") == 0) {
+		preaction_val = WDOG_PRETIMEOUT_NMI;
+#endif
+	} else if (strcmp(preaction, "pre_int") == 0) {
+		preaction_val = WDOG_PRETIMEOUT_MSG_INT;
+	} else {
+		preaction_val = WDOG_PRETIMEOUT_NONE;
+		printk(KERN_INFO PFX "Unknown preaction '%s', defaulting to"
+		       " none\n", preaction);
+	}
+
+	if (strcmp(preop, "preop_none") == 0) {
+		preop_val = WDOG_PREOP_NONE;
+	} else if (strcmp(preop, "preop_panic") == 0) {
+		preop_val = WDOG_PREOP_PANIC;
+	} else if (strcmp(preop, "preop_give_data") == 0) {
+		preop_val = WDOG_PREOP_GIVE_DATA;
+	} else {
+		preop_val = WDOG_PREOP_NONE;
+		printk(KERN_INFO PFX "Unknown preop '%s', defaulting to"
+		       " none\n", preop);
+	}
+
+#ifdef HAVE_NMI_HANDLER
+	if (preaction_val == WDOG_PRETIMEOUT_NMI) {
+		if (preop_val == WDOG_PREOP_GIVE_DATA) {
+			printk(KERN_WARNING PFX "Pretimeout op is to give data"
+			       " but NMI pretimeout is enabled, setting"
+			       " pretimeout op to none\n");
+			preop_val = WDOG_PREOP_NONE;
+		}
+#ifdef CONFIG_X86_LOCAL_APIC
+		if (nmi_watchdog == NMI_IO_APIC) {
+			printk(KERN_WARNING PFX "nmi_watchdog is set to IO APIC"
+			       " mode (value is %d), that is incompatible"
+			       " with using NMI in the IPMI watchdog."
+			       " Disabling IPMI nmi pretimeout.\n",
+			       nmi_watchdog);
+			preaction_val = WDOG_PRETIMEOUT_NONE;
+		} else {
+#endif
+		rv = request_nmi(&ipmi_nmi_handler);
+		if (rv) {
+			printk(KERN_WARNING PFX "Can't register nmi handler\n");
+			return rv;
+		}
+#ifdef CONFIG_X86_LOCAL_APIC
+		}
+#endif
+	}
+#endif
+
+	rv = ipmi_smi_watcher_register(&smi_watcher);
+	if (rv) {
+#ifdef HAVE_NMI_HANDLER
+		if (preaction_val == WDOG_PRETIMEOUT_NMI)
+			release_nmi(&ipmi_nmi_handler);
+#endif
+		printk(KERN_WARNING PFX "can't register smi watcher\n");
+		return rv;
+	}
+
+	register_reboot_notifier(&wdog_reboot_notifier);
+	notifier_chain_register(&panic_notifier_list, &wdog_panic_notifier);
+
+	return 0;
+}
+
+static __exit void ipmi_unregister_watchdog(void)
+{
+	int rv;
+
+	down_write(&register_sem);
+
+#ifdef HAVE_NMI_HANDLER
+	if (preaction_val == WDOG_PRETIMEOUT_NMI)
+		release_nmi(&ipmi_nmi_handler);
+#endif
+
+	notifier_chain_unregister(&panic_notifier_list, &wdog_panic_notifier);
+	unregister_reboot_notifier(&wdog_reboot_notifier);
+
+	if (! watchdog_user)
+		goto out;
+
+	/* Make sure no one can call us any more. */
+	misc_deregister(&ipmi_wdog_miscdev);
+
+	/* Wait to make sure the message makes it out.  The lower layer has
+	   pointers to our buffers, we want to make sure they are done before
+	   we release our memory. */
+	while (atomic_read(&set_timeout_tofree)) {
+		set_current_state(TASK_UNINTERRUPTIBLE);
+		schedule_timeout(1);
+	}
+
+	/* Disconnect from IPMI. */
+	rv = ipmi_destroy_user(watchdog_user);
+	if (rv) {
+		printk(KERN_WARNING PFX "error unlinking from IPMI: %d\n",
+		       rv);
+	}
+	watchdog_user = NULL;
+
+ out:
+	up_write(&register_sem);
+}
+
+static void __exit ipmi_wdog_exit(void)
+{
+	ipmi_smi_watcher_unregister(&smi_watcher);
+	ipmi_unregister_watchdog();
+}
+module_exit(ipmi_wdog_exit);
+module_init(ipmi_wdog_init);
+MODULE_LICENSE("GPL");