Linux-2.6.12-rc2v2.6.12-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!

6 files changed, 3657 insertions, 0 deletions

diff --git a/arch/i386/mach-voyager/Makefile b/arch/i386/mach-voyager/Makefile
new file mode 100644
index 000000000000..f24d29651318
--- /dev/null
+++ b/arch/i386/mach-voyager/Makefile
@@ -0,0 +1,8 @@
+#
+# Makefile for the linux kernel.
+#
+
+EXTRA_CFLAGS	+= -I../kernel
+obj-y			:= setup.o voyager_basic.o voyager_thread.o
+
+obj-$(CONFIG_SMP)	+= voyager_smp.o voyager_cat.o
diff --git a/arch/i386/mach-voyager/setup.c b/arch/i386/mach-voyager/setup.c
new file mode 100644
index 000000000000..df123fc487bb
--- /dev/null
+++ b/arch/i386/mach-voyager/setup.c
@@ -0,0 +1,48 @@
+/*
+ *	Machine specific setup for generic
+ */
+
+#include <linux/config.h>
+#include <linux/init.h>
+#include <linux/irq.h>
+#include <linux/interrupt.h>
+#include <asm/acpi.h>
+#include <asm/arch_hooks.h>
+
+void __init pre_intr_init_hook(void)
+{
+	init_ISA_irqs();
+}
+
+/*
+ * IRQ2 is cascade interrupt to second interrupt controller
+ */
+static struct irqaction irq2 = { no_action, 0, CPU_MASK_NONE, "cascade", NULL, NULL};
+
+void __init intr_init_hook(void)
+{
+#ifdef CONFIG_SMP
+	smp_intr_init();
+#endif
+
+	if (!acpi_ioapic)
+		setup_irq(2, &irq2);
+}
+
+void __init pre_setup_arch_hook(void)
+{
+	/* Voyagers run their CPUs from independent clocks, so disable
+	 * the TSC code because we can't sync them */
+	tsc_disable = 1;
+}
+
+void __init trap_init_hook(void)
+{
+}
+
+static struct irqaction irq0  = { timer_interrupt, SA_INTERRUPT, CPU_MASK_NONE, "timer", NULL, NULL};
+
+void __init time_init_hook(void)
+{
+	setup_irq(0, &irq0);
+}
diff --git a/arch/i386/mach-voyager/voyager_basic.c b/arch/i386/mach-voyager/voyager_basic.c
new file mode 100644
index 000000000000..602aea240e9b
--- /dev/null
+++ b/arch/i386/mach-voyager/voyager_basic.c
@@ -0,0 +1,325 @@
+/* Copyright (C) 1999,2001 
+ *
+ * Author: J.E.J.Bottomley@HansenPartnership.com
+ *
+ * linux/arch/i386/kernel/voyager.c
+ *
+ * This file contains all the voyager specific routines for getting
+ * initialisation of the architecture to function.  For additional
+ * features see:
+ *
+ *	voyager_cat.c - Voyager CAT bus interface
+ *	voyager_smp.c - Voyager SMP hal (emulates linux smp.c)
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/sched.h>
+#include <linux/ptrace.h>
+#include <linux/ioport.h>
+#include <linux/interrupt.h>
+#include <linux/init.h>
+#include <linux/delay.h>
+#include <linux/reboot.h>
+#include <linux/sysrq.h>
+#include <asm/io.h>
+#include <asm/voyager.h>
+#include <asm/vic.h>
+#include <linux/pm.h>
+#include <linux/irq.h>
+#include <asm/tlbflush.h>
+#include <asm/arch_hooks.h>
+
+/*
+ * Power off function, if any
+ */
+void (*pm_power_off)(void);
+
+int voyager_level = 0;
+
+struct voyager_SUS *voyager_SUS = NULL;
+
+#ifdef CONFIG_SMP
+static void
+voyager_dump(int dummy1, struct pt_regs *dummy2, struct tty_struct *dummy3)
+{
+	/* get here via a sysrq */
+	voyager_smp_dump();
+}
+
+static struct sysrq_key_op sysrq_voyager_dump_op = {
+	.handler	= voyager_dump,
+	.help_msg	= "Voyager",
+	.action_msg	= "Dump Voyager Status",
+};
+#endif
+
+void
+voyager_detect(struct voyager_bios_info *bios)
+{
+	if(bios->len != 0xff) {
+		int class = (bios->class_1 << 8) 
+			| (bios->class_2 & 0xff);
+
+		printk("Voyager System detected.\n"
+		       "        Class %x, Revision %d.%d\n",
+		       class, bios->major, bios->minor);
+		if(class == VOYAGER_LEVEL4) 
+			voyager_level = 4;
+		else if(class < VOYAGER_LEVEL5_AND_ABOVE)
+			voyager_level = 3;
+		else
+			voyager_level = 5;
+		printk("        Architecture Level %d\n", voyager_level);
+		if(voyager_level < 4)
+			printk("\n**WARNING**: Voyager HAL only supports Levels 4 and 5 Architectures at the moment\n\n");
+		/* install the power off handler */
+		pm_power_off = voyager_power_off;
+#ifdef CONFIG_SMP
+		register_sysrq_key('v', &sysrq_voyager_dump_op);
+#endif
+	} else {
+		printk("\n\n**WARNING**: No Voyager Subsystem Found\n");
+	}
+}
+
+void
+voyager_system_interrupt(int cpl, void *dev_id, struct pt_regs *regs)
+{
+	printk("Voyager: detected system interrupt\n");
+}
+
+/* Routine to read information from the extended CMOS area */
+__u8
+voyager_extended_cmos_read(__u16 addr)
+{
+	outb(addr & 0xff, 0x74);
+	outb((addr >> 8) & 0xff, 0x75);
+	return inb(0x76);
+}
+
+/* internal definitions for the SUS Click Map of memory */
+
+#define CLICK_ENTRIES	16
+#define CLICK_SIZE	4096	/* click to byte conversion for Length */
+
+typedef struct ClickMap {
+	struct Entry {
+		__u32	Address;
+		__u32	Length;
+	} Entry[CLICK_ENTRIES];
+} ClickMap_t;
+
+
+/* This routine is pretty much an awful hack to read the bios clickmap by
+ * mapping it into page 0.  There are usually three regions in the map:
+ * 	Base Memory
+ * 	Extended Memory
+ *	zero length marker for end of map
+ *
+ * Returns are 0 for failure and 1 for success on extracting region.
+ */
+int __init
+voyager_memory_detect(int region, __u32 *start, __u32 *length)
+{
+	int i;
+	int retval = 0;
+	__u8 cmos[4];
+	ClickMap_t *map;
+	unsigned long map_addr;
+	unsigned long old;
+
+	if(region >= CLICK_ENTRIES) {
+		printk("Voyager: Illegal ClickMap region %d\n", region);
+		return 0;
+	}
+
+	for(i = 0; i < sizeof(cmos); i++)
+		cmos[i] = voyager_extended_cmos_read(VOYAGER_MEMORY_CLICKMAP + i);
+
+	map_addr = *(unsigned long *)cmos;
+
+	/* steal page 0 for this */
+	old = pg0[0];
+	pg0[0] = ((map_addr & PAGE_MASK) | _PAGE_RW | _PAGE_PRESENT);
+	local_flush_tlb();
+	/* now clear everything out but page 0 */
+	map = (ClickMap_t *)(map_addr & (~PAGE_MASK));
+
+	/* zero length is the end of the clickmap */
+	if(map->Entry[region].Length != 0) {
+		*length = map->Entry[region].Length * CLICK_SIZE;
+		*start = map->Entry[region].Address;
+		retval = 1;
+	}
+
+	/* replace the mapping */
+	pg0[0] = old;
+	local_flush_tlb();
+	return retval;
+}
+
+/* voyager specific handling code for timer interrupts.  Used to hand
+ * off the timer tick to the SMP code, since the VIC doesn't have an
+ * internal timer (The QIC does, but that's another story). */
+void
+voyager_timer_interrupt(struct pt_regs *regs)
+{
+	if((jiffies & 0x3ff) == 0) {
+
+		/* There seems to be something flaky in either
+		 * hardware or software that is resetting the timer 0
+		 * count to something much higher than it should be
+		 * This seems to occur in the boot sequence, just
+		 * before root is mounted.  Therefore, every 10
+		 * seconds or so, we sanity check the timer zero count
+		 * and kick it back to where it should be.
+		 *
+		 * FIXME: This is the most awful hack yet seen.  I
+		 * should work out exactly what is interfering with
+		 * the timer count settings early in the boot sequence
+		 * and swiftly introduce it to something sharp and
+		 * pointy.  */
+		__u16 val;
+		extern spinlock_t i8253_lock;
+
+		spin_lock(&i8253_lock);
+		
+		outb_p(0x00, 0x43);
+		val = inb_p(0x40);
+		val |= inb(0x40) << 8;
+		spin_unlock(&i8253_lock);
+
+		if(val > LATCH) {
+			printk("\nVOYAGER: countdown timer value too high (%d), resetting\n\n", val);
+			spin_lock(&i8253_lock);
+			outb(0x34,0x43);
+			outb_p(LATCH & 0xff , 0x40);	/* LSB */
+			outb(LATCH >> 8 , 0x40);	/* MSB */
+			spin_unlock(&i8253_lock);
+		}
+	}
+#ifdef CONFIG_SMP
+	smp_vic_timer_interrupt(regs);
+#endif
+}
+
+void
+voyager_power_off(void)
+{
+	printk("VOYAGER Power Off\n");
+
+	if(voyager_level == 5) {
+		voyager_cat_power_off();
+	} else if(voyager_level == 4) {
+		/* This doesn't apparently work on most L4 machines,
+		 * but the specs say to do this to get automatic power
+		 * off.  Unfortunately, if it doesn't power off the
+		 * machine, it ends up doing a cold restart, which
+		 * isn't really intended, so comment out the code */
+#if 0
+		int port;
+
+	  
+		/* enable the voyager Configuration Space */
+		outb((inb(VOYAGER_MC_SETUP) & 0xf0) | 0x8, 
+		     VOYAGER_MC_SETUP);
+		/* the port for the power off flag is an offset from the
+		   floating base */
+		port = (inb(VOYAGER_SSPB_RELOCATION_PORT) << 8) + 0x21;
+		/* set the power off flag */
+		outb(inb(port) | 0x1, port);
+#endif
+	}
+	/* and wait for it to happen */
+	for(;;) {
+		__asm("cli");
+		__asm("hlt");
+	}
+}
+
+/* copied from process.c */
+static inline void
+kb_wait(void)
+{
+	int i;
+
+	for (i=0; i<0x10000; i++)
+		if ((inb_p(0x64) & 0x02) == 0)
+			break;
+}
+
+void
+machine_restart(char *cmd)
+{
+	printk("Voyager Warm Restart\n");
+	kb_wait();
+
+	if(voyager_level == 5) {
+		/* write magic values to the RTC to inform system that
+		 * shutdown is beginning */
+		outb(0x8f, 0x70);
+		outb(0x5 , 0x71);
+		
+		udelay(50);
+		outb(0xfe,0x64);         /* pull reset low */
+	} else if(voyager_level == 4) {
+		__u16 catbase = inb(VOYAGER_SSPB_RELOCATION_PORT)<<8;
+		__u8 basebd = inb(VOYAGER_MC_SETUP);
+		
+		outb(basebd | 0x08, VOYAGER_MC_SETUP);
+		outb(0x02, catbase + 0x21);
+	}
+	for(;;) {
+		asm("cli");
+		asm("hlt");
+	}
+}
+
+EXPORT_SYMBOL(machine_restart);
+
+void
+mca_nmi_hook(void)
+{
+	__u8 dumpval __attribute__((unused)) = inb(0xf823);
+	__u8 swnmi __attribute__((unused)) = inb(0xf813);
+
+	/* FIXME: assume dump switch pressed */
+	/* check to see if the dump switch was pressed */
+	VDEBUG(("VOYAGER: dumpval = 0x%x, swnmi = 0x%x\n", dumpval, swnmi));
+	/* clear swnmi */
+	outb(0xff, 0xf813);
+	/* tell SUS to ignore dump */
+	if(voyager_level == 5 && voyager_SUS != NULL) {
+		if(voyager_SUS->SUS_mbox == VOYAGER_DUMP_BUTTON_NMI) {
+			voyager_SUS->kernel_mbox = VOYAGER_NO_COMMAND;
+			voyager_SUS->kernel_flags |= VOYAGER_OS_IN_PROGRESS;
+			udelay(1000);
+			voyager_SUS->kernel_mbox = VOYAGER_IGNORE_DUMP;
+			voyager_SUS->kernel_flags &= ~VOYAGER_OS_IN_PROGRESS;
+		}
+	}
+	printk(KERN_ERR "VOYAGER: Dump switch pressed, printing CPU%d tracebacks\n", smp_processor_id());
+	show_stack(NULL, NULL);
+	show_state();
+}
+
+
+
+void
+machine_halt(void)
+{
+	/* treat a halt like a power off */
+	machine_power_off();
+}
+
+EXPORT_SYMBOL(machine_halt);
+
+void machine_power_off(void)
+{
+	if (pm_power_off)
+		pm_power_off();
+}
+
+EXPORT_SYMBOL(machine_power_off);
diff --git a/arch/i386/mach-voyager/voyager_cat.c b/arch/i386/mach-voyager/voyager_cat.c
new file mode 100644
index 000000000000..23967fe658d3
--- /dev/null
+++ b/arch/i386/mach-voyager/voyager_cat.c
@@ -0,0 +1,1178 @@
+/* -*- mode: c; c-basic-offset: 8 -*- */
+
+/* Copyright (C) 1999,2001
+ *
+ * Author: J.E.J.Bottomley@HansenPartnership.com
+ *
+ * linux/arch/i386/kernel/voyager_cat.c
+ *
+ * This file contains all the logic for manipulating the CAT bus
+ * in a level 5 machine.
+ *
+ * The CAT bus is a serial configuration and test bus.  Its primary
+ * uses are to probe the initial configuration of the system and to
+ * diagnose error conditions when a system interrupt occurs.  The low
+ * level interface is fairly primitive, so most of this file consists
+ * of bit shift manipulations to send and receive packets on the
+ * serial bus */
+
+#include <linux/config.h>
+#include <linux/types.h>
+#include <linux/completion.h>
+#include <linux/sched.h>
+#include <asm/voyager.h>
+#include <asm/vic.h>
+#include <linux/ioport.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/delay.h>
+#include <asm/io.h>
+
+#ifdef VOYAGER_CAT_DEBUG
+#define CDEBUG(x)	printk x
+#else
+#define CDEBUG(x)
+#endif
+
+/* the CAT command port */
+#define CAT_CMD		(sspb + 0xe)
+/* the CAT data port */
+#define CAT_DATA	(sspb + 0xd)
+
+/* the internal cat functions */
+static void cat_pack(__u8 *msg, __u16 start_bit, __u8 *data, 
+		     __u16 num_bits);
+static void cat_unpack(__u8 *msg, __u16 start_bit, __u8 *data,
+		       __u16 num_bits);
+static void cat_build_header(__u8 *header, const __u16 len, 
+			     const __u16 smallest_reg_bits,
+			     const __u16 longest_reg_bits);
+static int cat_sendinst(voyager_module_t *modp, voyager_asic_t *asicp,
+			__u8 reg, __u8 op);
+static int cat_getdata(voyager_module_t *modp, voyager_asic_t *asicp,
+		       __u8 reg, __u8 *value);
+static int cat_shiftout(__u8 *data, __u16 data_bytes, __u16 header_bytes,
+			__u8 pad_bits);
+static int cat_write(voyager_module_t *modp, voyager_asic_t *asicp, __u8 reg,
+		     __u8 value);
+static int cat_read(voyager_module_t *modp, voyager_asic_t *asicp, __u8 reg,
+		    __u8 *value);
+static int cat_subread(voyager_module_t *modp, voyager_asic_t *asicp,
+		       __u16 offset, __u16 len, void *buf);
+static int cat_senddata(voyager_module_t *modp, voyager_asic_t *asicp,
+			__u8 reg, __u8 value);
+static int cat_disconnect(voyager_module_t *modp, voyager_asic_t *asicp);
+static int cat_connect(voyager_module_t *modp, voyager_asic_t *asicp);
+
+static inline const char *
+cat_module_name(int module_id)
+{
+	switch(module_id) {
+	case 0x10:
+		return "Processor Slot 0";
+	case 0x11:
+		return "Processor Slot 1";
+	case 0x12:
+		return "Processor Slot 2";
+	case 0x13:
+		return "Processor Slot 4";
+	case 0x14:
+		return "Memory Slot 0";
+	case 0x15:
+		return "Memory Slot 1";
+	case 0x18:
+		return "Primary Microchannel";
+	case 0x19:
+		return "Secondary Microchannel";
+	case 0x1a:
+		return "Power Supply Interface";
+	case 0x1c:
+		return "Processor Slot 5";
+	case 0x1d:
+		return "Processor Slot 6";
+	case 0x1e:
+		return "Processor Slot 7";
+	case 0x1f:
+		return "Processor Slot 8";
+	default:
+		return "Unknown Module";
+	}
+}
+
+static int sspb = 0;		/* stores the super port location */
+int voyager_8slot = 0;		/* set to true if a 51xx monster */
+
+voyager_module_t *voyager_cat_list;
+
+/* the I/O port assignments for the VIC and QIC */
+static struct resource vic_res = {
+	"Voyager Interrupt Controller", 0xFC00, 0xFC6F };
+static struct resource qic_res = {
+	"Quad Interrupt Controller", 0xFC70, 0xFCFF };
+
+/* This function is used to pack a data bit stream inside a message.
+ * It writes num_bits of the data buffer in msg starting at start_bit.
+ * Note: This function assumes that any unused bit in the data stream
+ * is set to zero so that the ors will work correctly */
+#define BITS_PER_BYTE 8
+static void
+cat_pack(__u8 *msg, const __u16 start_bit, __u8 *data, const __u16 num_bits)
+{
+	/* compute initial shift needed */
+	const __u16 offset = start_bit % BITS_PER_BYTE;
+	__u16 len = num_bits / BITS_PER_BYTE;
+	__u16 byte = start_bit / BITS_PER_BYTE;
+	__u16 residue = (num_bits % BITS_PER_BYTE) + offset;
+	int i;
+
+	/* adjust if we have more than a byte of residue */
+	if(residue >= BITS_PER_BYTE) {
+		residue -= BITS_PER_BYTE;
+		len++;
+	}
+
+	/* clear out the bits.  We assume here that if len==0 then
+	 * residue >= offset.  This is always true for the catbus
+	 * operations */
+	msg[byte] &= 0xff << (BITS_PER_BYTE - offset); 
+	msg[byte++] |= data[0] >> offset;
+	if(len == 0)
+		return;
+	for(i = 1; i < len; i++)
+		msg[byte++] = (data[i-1] << (BITS_PER_BYTE - offset))
+			| (data[i] >> offset);
+	if(residue != 0) {
+		__u8 mask = 0xff >> residue;
+		__u8 last_byte = data[i-1] << (BITS_PER_BYTE - offset)
+			| (data[i] >> offset);
+		
+		last_byte &= ~mask;
+		msg[byte] &= mask;
+		msg[byte] |= last_byte;
+	}
+	return;
+}
+/* unpack the data again (same arguments as cat_pack()). data buffer
+ * must be zero populated.
+ *
+ * Function: given a message string move to start_bit and copy num_bits into
+ * data (starting at bit 0 in data).
+ */
+static void
+cat_unpack(__u8 *msg, const __u16 start_bit, __u8 *data, const __u16 num_bits)
+{
+	/* compute initial shift needed */
+	const __u16 offset = start_bit % BITS_PER_BYTE;
+	__u16 len = num_bits / BITS_PER_BYTE;
+	const __u8 last_bits = num_bits % BITS_PER_BYTE;
+	__u16 byte = start_bit / BITS_PER_BYTE;
+	int i;
+
+	if(last_bits != 0)
+		len++;
+
+	/* special case: want < 8 bits from msg and we can get it from
+	 * a single byte of the msg */
+	if(len == 0 && BITS_PER_BYTE - offset >= num_bits) {
+		data[0] = msg[byte] << offset;
+		data[0] &= 0xff >> (BITS_PER_BYTE - num_bits);
+		return;
+	}
+	for(i = 0; i < len; i++) {
+		/* this annoying if has to be done just in case a read of
+		 * msg one beyond the array causes a panic */
+		if(offset != 0) {
+			data[i] = msg[byte++] << offset;
+			data[i] |= msg[byte] >> (BITS_PER_BYTE - offset);
+		}
+		else {
+			data[i] = msg[byte++];
+		}
+	}
+	/* do we need to truncate the final byte */
+	if(last_bits != 0) {
+		data[i-1] &= 0xff << (BITS_PER_BYTE - last_bits);
+	}
+	return;
+}
+
+static void
+cat_build_header(__u8 *header, const __u16 len, const __u16 smallest_reg_bits,
+		 const __u16 longest_reg_bits)
+{
+	int i;
+	__u16 start_bit = (smallest_reg_bits - 1) % BITS_PER_BYTE;
+	__u8 *last_byte = &header[len - 1];
+
+	if(start_bit == 0)
+		start_bit = 1;	/* must have at least one bit in the hdr */
+	
+	for(i=0; i < len; i++)
+		header[i] = 0;
+
+	for(i = start_bit; i > 0; i--)
+		*last_byte = ((*last_byte) << 1) + 1;
+
+}
+
+static int
+cat_sendinst(voyager_module_t *modp, voyager_asic_t *asicp, __u8 reg, __u8 op)
+{
+	__u8 parity, inst, inst_buf[4] = { 0 };
+	__u8 iseq[VOYAGER_MAX_SCAN_PATH], hseq[VOYAGER_MAX_REG_SIZE];
+	__u16 ibytes, hbytes, padbits;
+	int i;
+	
+	/* 
+	 * Parity is the parity of the register number + 1 (READ_REGISTER
+	 * and WRITE_REGISTER always add '1' to the number of bits == 1)
+	 */
+	parity = (__u8)(1 + (reg & 0x01) +
+	         ((__u8)(reg & 0x02) >> 1) +
+	         ((__u8)(reg & 0x04) >> 2) +
+	         ((__u8)(reg & 0x08) >> 3)) % 2;
+
+	inst = ((parity << 7) | (reg << 2) | op);
+
+	outb(VOYAGER_CAT_IRCYC, CAT_CMD);
+	if(!modp->scan_path_connected) {
+		if(asicp->asic_id != VOYAGER_CAT_ID) {
+			printk("**WARNING***: cat_sendinst has disconnected scan path not to CAT asic\n");
+			return 1;
+		}
+		outb(VOYAGER_CAT_HEADER, CAT_DATA);
+		outb(inst, CAT_DATA);
+		if(inb(CAT_DATA) != VOYAGER_CAT_HEADER) {
+			CDEBUG(("VOYAGER CAT: cat_sendinst failed to get CAT_HEADER\n"));
+			return 1;
+		}
+		return 0;
+	}
+	ibytes = modp->inst_bits / BITS_PER_BYTE;
+	if((padbits = modp->inst_bits % BITS_PER_BYTE) != 0) {
+		padbits = BITS_PER_BYTE - padbits;
+		ibytes++;
+	}
+	hbytes = modp->largest_reg / BITS_PER_BYTE;
+	if(modp->largest_reg % BITS_PER_BYTE)
+		hbytes++;
+	CDEBUG(("cat_sendinst: ibytes=%d, hbytes=%d\n", ibytes, hbytes));
+	/* initialise the instruction sequence to 0xff */
+	for(i=0; i < ibytes + hbytes; i++)
+		iseq[i] = 0xff;
+	cat_build_header(hseq, hbytes, modp->smallest_reg, modp->largest_reg);
+	cat_pack(iseq, modp->inst_bits, hseq, hbytes * BITS_PER_BYTE);
+	inst_buf[0] = inst;
+	inst_buf[1] = 0xFF >> (modp->largest_reg % BITS_PER_BYTE);
+	cat_pack(iseq, asicp->bit_location, inst_buf, asicp->ireg_length);
+#ifdef VOYAGER_CAT_DEBUG
+	printk("ins = 0x%x, iseq: ", inst);
+	for(i=0; i< ibytes + hbytes; i++)
+		printk("0x%x ", iseq[i]);
+	printk("\n");
+#endif
+	if(cat_shiftout(iseq, ibytes, hbytes, padbits)) {
+		CDEBUG(("VOYAGER CAT: cat_sendinst: cat_shiftout failed\n"));
+		return 1;
+	}
+	CDEBUG(("CAT SHIFTOUT DONE\n"));
+	return 0;
+}
+
+static int
+cat_getdata(voyager_module_t *modp, voyager_asic_t *asicp, __u8 reg, 
+	    __u8 *value)
+{
+	if(!modp->scan_path_connected) {
+		if(asicp->asic_id != VOYAGER_CAT_ID) {
+			CDEBUG(("VOYAGER CAT: ERROR: cat_getdata to CAT asic with scan path connected\n"));
+			return 1;
+		}
+		if(reg > VOYAGER_SUBADDRHI) 
+			outb(VOYAGER_CAT_RUN, CAT_CMD);
+		outb(VOYAGER_CAT_DRCYC, CAT_CMD);
+		outb(VOYAGER_CAT_HEADER, CAT_DATA);
+		*value = inb(CAT_DATA);
+		outb(0xAA, CAT_DATA);
+		if(inb(CAT_DATA) != VOYAGER_CAT_HEADER) {
+			CDEBUG(("cat_getdata: failed to get VOYAGER_CAT_HEADER\n"));
+			return 1;
+		}
+		return 0;
+	}
+	else {
+		__u16 sbits = modp->num_asics -1 + asicp->ireg_length;
+		__u16 sbytes = sbits / BITS_PER_BYTE;
+		__u16 tbytes;
+		__u8 string[VOYAGER_MAX_SCAN_PATH], trailer[VOYAGER_MAX_REG_SIZE];
+		__u8 padbits;
+		int i;
+		
+		outb(VOYAGER_CAT_DRCYC, CAT_CMD);
+
+		if((padbits = sbits % BITS_PER_BYTE) != 0) {
+			padbits = BITS_PER_BYTE - padbits;
+			sbytes++;
+		}
+		tbytes = asicp->ireg_length / BITS_PER_BYTE;
+		if(asicp->ireg_length % BITS_PER_BYTE)
+			tbytes++;
+		CDEBUG(("cat_getdata: tbytes = %d, sbytes = %d, padbits = %d\n",
+			tbytes,	sbytes, padbits));
+		cat_build_header(trailer, tbytes, 1, asicp->ireg_length);
+
+		
+		for(i = tbytes - 1; i >= 0; i--) {
+			outb(trailer[i], CAT_DATA);
+			string[sbytes + i] = inb(CAT_DATA);
+		}
+
+		for(i = sbytes - 1; i >= 0; i--) {
+			outb(0xaa, CAT_DATA);
+			string[i] = inb(CAT_DATA);
+		}
+		*value = 0;
+		cat_unpack(string, padbits + (tbytes * BITS_PER_BYTE) + asicp->asic_location, value, asicp->ireg_length);
+#ifdef VOYAGER_CAT_DEBUG
+		printk("value=0x%x, string: ", *value);
+		for(i=0; i< tbytes+sbytes; i++)
+			printk("0x%x ", string[i]);
+		printk("\n");
+#endif
+		
+		/* sanity check the rest of the return */
+		for(i=0; i < tbytes; i++) {
+			__u8 input = 0;
+
+			cat_unpack(string, padbits + (i * BITS_PER_BYTE), &input, BITS_PER_BYTE);
+			if(trailer[i] != input) {
+				CDEBUG(("cat_getdata: failed to sanity check rest of ret(%d) 0x%x != 0x%x\n", i, input, trailer[i]));
+				return 1;
+			}
+		}
+		CDEBUG(("cat_getdata DONE\n"));
+		return 0;
+	}
+}
+
+static int
+cat_shiftout(__u8 *data, __u16 data_bytes, __u16 header_bytes, __u8 pad_bits)
+{
+	int i;
+	
+	for(i = data_bytes + header_bytes - 1; i >= header_bytes; i--)
+		outb(data[i], CAT_DATA);
+
+	for(i = header_bytes - 1; i >= 0; i--) {
+		__u8 header = 0;
+		__u8 input;
+
+		outb(data[i], CAT_DATA);
+		input = inb(CAT_DATA);
+		CDEBUG(("cat_shiftout: returned 0x%x\n", input));
+		cat_unpack(data, ((data_bytes + i) * BITS_PER_BYTE) - pad_bits,
+			   &header, BITS_PER_BYTE);
+		if(input != header) {
+			CDEBUG(("VOYAGER CAT: cat_shiftout failed to return header 0x%x != 0x%x\n", input, header));
+			return 1;
+		}
+	}
+	return 0;
+}
+
+static int
+cat_senddata(voyager_module_t *modp, voyager_asic_t *asicp, 
+	     __u8 reg, __u8 value)
+{
+	outb(VOYAGER_CAT_DRCYC, CAT_CMD);
+	if(!modp->scan_path_connected) {
+		if(asicp->asic_id != VOYAGER_CAT_ID) {
+			CDEBUG(("VOYAGER CAT: ERROR: scan path disconnected when asic != CAT\n"));
+			return 1;
+		}
+		outb(VOYAGER_CAT_HEADER, CAT_DATA);
+		outb(value, CAT_DATA);
+		if(inb(CAT_DATA) != VOYAGER_CAT_HEADER) {
+			CDEBUG(("cat_senddata: failed to get correct header response to sent data\n"));
+			return 1;
+		}
+		if(reg > VOYAGER_SUBADDRHI) {
+			outb(VOYAGER_CAT_RUN, CAT_CMD);
+			outb(VOYAGER_CAT_END, CAT_CMD);
+			outb(VOYAGER_CAT_RUN, CAT_CMD);
+		}
+		
+		return 0;
+	}
+	else {
+		__u16 hbytes = asicp->ireg_length / BITS_PER_BYTE;
+		__u16 dbytes = (modp->num_asics - 1 + asicp->ireg_length)/BITS_PER_BYTE;
+		__u8 padbits, dseq[VOYAGER_MAX_SCAN_PATH], 
+			hseq[VOYAGER_MAX_REG_SIZE];
+		int i;
+
+		if((padbits = (modp->num_asics - 1 
+			       + asicp->ireg_length) % BITS_PER_BYTE) != 0) {
+			padbits = BITS_PER_BYTE - padbits;
+			dbytes++;
+		}
+		if(asicp->ireg_length % BITS_PER_BYTE)
+			hbytes++;
+		
+		cat_build_header(hseq, hbytes, 1, asicp->ireg_length);
+		
+		for(i = 0; i < dbytes + hbytes; i++)
+			dseq[i] = 0xff;
+		CDEBUG(("cat_senddata: dbytes=%d, hbytes=%d, padbits=%d\n",
+			dbytes, hbytes, padbits));
+		cat_pack(dseq, modp->num_asics - 1 + asicp->ireg_length,
+			 hseq, hbytes * BITS_PER_BYTE);
+		cat_pack(dseq, asicp->asic_location, &value, 
+			 asicp->ireg_length);
+#ifdef VOYAGER_CAT_DEBUG
+		printk("dseq ");
+		for(i=0; i<hbytes+dbytes; i++) {
+			printk("0x%x ", dseq[i]);
+		}
+		printk("\n");
+#endif
+		return cat_shiftout(dseq, dbytes, hbytes, padbits);
+	}
+}
+
+static int
+cat_write(voyager_module_t *modp, voyager_asic_t *asicp, __u8 reg,
+	 __u8 value)
+{
+	if(cat_sendinst(modp, asicp, reg, VOYAGER_WRITE_CONFIG))
+		return 1;
+	return cat_senddata(modp, asicp, reg, value);
+}
+
+static int
+cat_read(voyager_module_t *modp, voyager_asic_t *asicp, __u8 reg,
+	 __u8 *value)
+{
+	if(cat_sendinst(modp, asicp, reg, VOYAGER_READ_CONFIG))
+		return 1;
+	return cat_getdata(modp, asicp, reg, value);
+}
+
+static int
+cat_subaddrsetup(voyager_module_t *modp, voyager_asic_t *asicp, __u16 offset,
+		 __u16 len)
+{
+	__u8 val;
+
+	if(len > 1) {
+		/* set auto increment */
+		__u8 newval;
+		
+		if(cat_read(modp, asicp, VOYAGER_AUTO_INC_REG, &val)) {
+			CDEBUG(("cat_subaddrsetup: read of VOYAGER_AUTO_INC_REG failed\n"));
+			return 1;
+		}
+		CDEBUG(("cat_subaddrsetup: VOYAGER_AUTO_INC_REG = 0x%x\n", val));
+		newval = val | VOYAGER_AUTO_INC;
+		if(newval != val) {
+			if(cat_write(modp, asicp, VOYAGER_AUTO_INC_REG, val)) {
+				CDEBUG(("cat_subaddrsetup: write to VOYAGER_AUTO_INC_REG failed\n"));
+				return 1;
+			}
+		}
+	}
+	if(cat_write(modp, asicp, VOYAGER_SUBADDRLO, (__u8)(offset &0xff))) {
+		CDEBUG(("cat_subaddrsetup: write to SUBADDRLO failed\n"));
+		return 1;
+	}
+	if(asicp->subaddr > VOYAGER_SUBADDR_LO) {
+		if(cat_write(modp, asicp, VOYAGER_SUBADDRHI, (__u8)(offset >> 8))) {
+			CDEBUG(("cat_subaddrsetup: write to SUBADDRHI failed\n"));
+			return 1;
+		}
+		cat_read(modp, asicp, VOYAGER_SUBADDRHI, &val);
+		CDEBUG(("cat_subaddrsetup: offset = %d, hi = %d\n", offset, val));
+	}
+	cat_read(modp, asicp, VOYAGER_SUBADDRLO, &val);
+	CDEBUG(("cat_subaddrsetup: offset = %d, lo = %d\n", offset, val));
+	return 0;
+}
+		
+static int
+cat_subwrite(voyager_module_t *modp, voyager_asic_t *asicp, __u16 offset,
+	    __u16 len, void *buf)
+{
+	int i, retval;
+
+	/* FIXME: need special actions for VOYAGER_CAT_ID here */
+	if(asicp->asic_id == VOYAGER_CAT_ID) {
+		CDEBUG(("cat_subwrite: ATTEMPT TO WRITE TO CAT ASIC\n"));
+		/* FIXME -- This is supposed to be handled better
+		 * There is a problem writing to the cat asic in the
+		 * PSI.  The 30us delay seems to work, though */
+		udelay(30);
+	}
+		
+	if((retval = cat_subaddrsetup(modp, asicp, offset, len)) != 0) {
+		printk("cat_subwrite: cat_subaddrsetup FAILED\n");
+		return retval;
+	}
+	
+	if(cat_sendinst(modp, asicp, VOYAGER_SUBADDRDATA, VOYAGER_WRITE_CONFIG)) {
+		printk("cat_subwrite: cat_sendinst FAILED\n");
+		return 1;
+	}
+	for(i = 0; i < len; i++) {
+		if(cat_senddata(modp, asicp, 0xFF, ((__u8 *)buf)[i])) {
+			printk("cat_subwrite: cat_sendata element at %d FAILED\n", i);
+			return 1;
+		}
+	}
+	return 0;
+}
+static int
+cat_subread(voyager_module_t *modp, voyager_asic_t *asicp, __u16 offset,
+	    __u16 len, void *buf)
+{
+	int i, retval;
+
+	if((retval = cat_subaddrsetup(modp, asicp, offset, len)) != 0) {
+		CDEBUG(("cat_subread: cat_subaddrsetup FAILED\n"));
+		return retval;
+	}
+
+	if(cat_sendinst(modp, asicp, VOYAGER_SUBADDRDATA, VOYAGER_READ_CONFIG)) {
+		CDEBUG(("cat_subread: cat_sendinst failed\n"));
+		return 1;
+	}
+	for(i = 0; i < len; i++) {
+		if(cat_getdata(modp, asicp, 0xFF,
+			       &((__u8 *)buf)[i])) {
+			CDEBUG(("cat_subread: cat_getdata element %d failed\n", i));
+			return 1;
+		}
+	}
+	return 0;
+}
+
+
+/* buffer for storing EPROM data read in during initialisation */
+static __initdata __u8 eprom_buf[0xFFFF];
+static voyager_module_t *voyager_initial_module;
+
+/* Initialise the cat bus components.  We assume this is called by the
+ * boot cpu *after* all memory initialisation has been done (so we can
+ * use kmalloc) but before smp initialisation, so we can probe the SMP
+ * configuration and pick up necessary information.  */
+void
+voyager_cat_init(void)
+{
+	voyager_module_t **modpp = &voyager_initial_module;
+	voyager_asic_t **asicpp;
+	voyager_asic_t *qabc_asic = NULL;
+	int i, j;
+	unsigned long qic_addr = 0;
+	__u8 qabc_data[0x20];
+	__u8 num_submodules, val;
+	voyager_eprom_hdr_t *eprom_hdr = (voyager_eprom_hdr_t *)&eprom_buf[0];
+	
+	__u8 cmos[4];
+	unsigned long addr;
+	
+	/* initiallise the SUS mailbox */
+	for(i=0; i<sizeof(cmos); i++)
+		cmos[i] = voyager_extended_cmos_read(VOYAGER_DUMP_LOCATION + i);
+	addr = *(unsigned long *)cmos;
+	if((addr & 0xff000000) != 0xff000000) {
+		printk(KERN_ERR "Voyager failed to get SUS mailbox (addr = 0x%lx\n", addr);
+	} else {
+		static struct resource res;
+		
+		res.name = "voyager SUS";
+		res.start = addr;
+		res.end = addr+0x3ff;
+		
+		request_resource(&iomem_resource, &res);
+		voyager_SUS = (struct voyager_SUS *)
+			ioremap(addr, 0x400);
+		printk(KERN_NOTICE "Voyager SUS mailbox version 0x%x\n",
+		       voyager_SUS->SUS_version);
+		voyager_SUS->kernel_version = VOYAGER_MAILBOX_VERSION;
+		voyager_SUS->kernel_flags = VOYAGER_OS_HAS_SYSINT;
+	}
+
+	/* clear the processor counts */
+	voyager_extended_vic_processors = 0;
+	voyager_quad_processors = 0;
+
+
+
+	printk("VOYAGER: beginning CAT bus probe\n");
+	/* set up the SuperSet Port Block which tells us where the
+	 * CAT communication port is */
+	sspb = inb(VOYAGER_SSPB_RELOCATION_PORT) * 0x100;
+	VDEBUG(("VOYAGER DEBUG: sspb = 0x%x\n", sspb));
+
+	/* now find out if were 8 slot or normal */
+	if((inb(VIC_PROC_WHO_AM_I) & EIGHT_SLOT_IDENTIFIER)
+	   == EIGHT_SLOT_IDENTIFIER) {
+		voyager_8slot = 1;
+		printk(KERN_NOTICE "Voyager: Eight slot 51xx configuration detected\n");
+	}
+
+	for(i = VOYAGER_MIN_MODULE;
+	    i <= VOYAGER_MAX_MODULE; i++) {
+		__u8 input;
+		int asic;
+		__u16 eprom_size;
+		__u16 sp_offset;
+
+		outb(VOYAGER_CAT_DESELECT, VOYAGER_CAT_CONFIG_PORT);
+		outb(i, VOYAGER_CAT_CONFIG_PORT);
+
+		/* check the presence of the module */
+		outb(VOYAGER_CAT_RUN, CAT_CMD);
+		outb(VOYAGER_CAT_IRCYC, CAT_CMD);
+		outb(VOYAGER_CAT_HEADER, CAT_DATA);
+		/* stream series of alternating 1's and 0's to stimulate
+		 * response */
+		outb(0xAA, CAT_DATA);
+		input = inb(CAT_DATA);
+		outb(VOYAGER_CAT_END, CAT_CMD);
+		if(input != VOYAGER_CAT_HEADER) {
+			continue;
+		}
+		CDEBUG(("VOYAGER DEBUG: found module id 0x%x, %s\n", i,
+			cat_module_name(i)));
+		*modpp = kmalloc(sizeof(voyager_module_t), GFP_KERNEL); /*&voyager_module_storage[cat_count++];*/
+		if(*modpp == NULL) {
+			printk("**WARNING** kmalloc failure in cat_init\n");
+			continue;
+		}
+		memset(*modpp, 0, sizeof(voyager_module_t));
+		/* need temporary asic for cat_subread.  It will be
+		 * filled in correctly later */
+		(*modpp)->asic = kmalloc(sizeof(voyager_asic_t), GFP_KERNEL); /*&voyager_asic_storage[asic_count];*/
+		if((*modpp)->asic == NULL) {
+			printk("**WARNING** kmalloc failure in cat_init\n");
+			continue;
+		}
+		memset((*modpp)->asic, 0, sizeof(voyager_asic_t));
+		(*modpp)->asic->asic_id = VOYAGER_CAT_ID;
+		(*modpp)->asic->subaddr = VOYAGER_SUBADDR_HI;
+		(*modpp)->module_addr = i;
+		(*modpp)->scan_path_connected = 0;
+		if(i == VOYAGER_PSI) {
+			/* Exception leg for modules with no EEPROM */
+			printk("Module \"%s\"\n", cat_module_name(i));
+			continue;
+		}
+			       
+		CDEBUG(("cat_init: Reading eeprom for module 0x%x at offset %d\n", i, VOYAGER_XSUM_END_OFFSET));
+		outb(VOYAGER_CAT_RUN, CAT_CMD);
+		cat_disconnect(*modpp, (*modpp)->asic);
+		if(cat_subread(*modpp, (*modpp)->asic,
+			       VOYAGER_XSUM_END_OFFSET, sizeof(eprom_size),
+			       &eprom_size)) {
+			printk("**WARNING**: Voyager couldn't read EPROM size for module 0x%x\n", i);
+			outb(VOYAGER_CAT_END, CAT_CMD);
+			continue;
+		}
+		if(eprom_size > sizeof(eprom_buf)) {
+			printk("**WARNING**: Voyager insufficient size to read EPROM data, module 0x%x.  Need %d\n", i, eprom_size);
+			outb(VOYAGER_CAT_END, CAT_CMD);
+			continue;
+		}
+		outb(VOYAGER_CAT_END, CAT_CMD);
+		outb(VOYAGER_CAT_RUN, CAT_CMD);
+		CDEBUG(("cat_init: module 0x%x, eeprom_size %d\n", i, eprom_size));
+		if(cat_subread(*modpp, (*modpp)->asic, 0, 
+			       eprom_size, eprom_buf)) {
+			outb(VOYAGER_CAT_END, CAT_CMD);
+			continue;
+		}
+		outb(VOYAGER_CAT_END, CAT_CMD);
+		printk("Module \"%s\", version 0x%x, tracer 0x%x, asics %d\n",
+		       cat_module_name(i), eprom_hdr->version_id,
+		       *((__u32 *)eprom_hdr->tracer),  eprom_hdr->num_asics);
+		(*modpp)->ee_size = eprom_hdr->ee_size;
+		(*modpp)->num_asics = eprom_hdr->num_asics;
+		asicpp = &((*modpp)->asic);
+		sp_offset = eprom_hdr->scan_path_offset;
+		/* All we really care about are the Quad cards.  We
+                 * identify them because they are in a processor slot
+                 * and have only four asics */
+		if((i < 0x10 || (i>=0x14 && i < 0x1c) || i>0x1f)) {
+			modpp = &((*modpp)->next);
+			continue;
+		}
+		/* Now we know it's in a processor slot, does it have
+		 * a quad baseboard submodule */
+		outb(VOYAGER_CAT_RUN, CAT_CMD);
+		cat_read(*modpp, (*modpp)->asic, VOYAGER_SUBMODPRESENT,
+			 &num_submodules);
+		/* lowest two bits, active low */
+		num_submodules = ~(0xfc | num_submodules);
+		CDEBUG(("VOYAGER CAT: %d submodules present\n", num_submodules));
+		if(num_submodules == 0) {
+			/* fill in the dyadic extended processors */
+			__u8 cpu = i & 0x07;
+
+			printk("Module \"%s\": Dyadic Processor Card\n",
+			       cat_module_name(i));
+			voyager_extended_vic_processors |= (1<<cpu);
+			cpu += 4;
+			voyager_extended_vic_processors |= (1<<cpu);
+			outb(VOYAGER_CAT_END, CAT_CMD);
+			continue;
+		}
+
+		/* now we want to read the asics on the first submodule,
+		 * which should be the quad base board */
+
+		cat_read(*modpp, (*modpp)->asic, VOYAGER_SUBMODSELECT, &val);
+		CDEBUG(("cat_init: SUBMODSELECT value = 0x%x\n", val));
+		val = (val & 0x7c) | VOYAGER_QUAD_BASEBOARD;
+		cat_write(*modpp, (*modpp)->asic, VOYAGER_SUBMODSELECT, val);
+
+		outb(VOYAGER_CAT_END, CAT_CMD);
+			 
+
+		CDEBUG(("cat_init: Reading eeprom for module 0x%x at offset %d\n", i, VOYAGER_XSUM_END_OFFSET));
+		outb(VOYAGER_CAT_RUN, CAT_CMD);
+		cat_disconnect(*modpp, (*modpp)->asic);
+		if(cat_subread(*modpp, (*modpp)->asic,
+			       VOYAGER_XSUM_END_OFFSET, sizeof(eprom_size),
+			       &eprom_size)) {
+			printk("**WARNING**: Voyager couldn't read EPROM size for module 0x%x\n", i);
+			outb(VOYAGER_CAT_END, CAT_CMD);
+			continue;
+		}
+		if(eprom_size > sizeof(eprom_buf)) {
+			printk("**WARNING**: Voyager insufficient size to read EPROM data, module 0x%x.  Need %d\n", i, eprom_size);
+			outb(VOYAGER_CAT_END, CAT_CMD);
+			continue;
+		}
+		outb(VOYAGER_CAT_END, CAT_CMD);
+		outb(VOYAGER_CAT_RUN, CAT_CMD);
+		CDEBUG(("cat_init: module 0x%x, eeprom_size %d\n", i, eprom_size));
+		if(cat_subread(*modpp, (*modpp)->asic, 0, 
+			       eprom_size, eprom_buf)) {
+			outb(VOYAGER_CAT_END, CAT_CMD);
+			continue;
+		}
+		outb(VOYAGER_CAT_END, CAT_CMD);
+		/* Now do everything for the QBB submodule 1 */
+		(*modpp)->ee_size = eprom_hdr->ee_size;
+		(*modpp)->num_asics = eprom_hdr->num_asics;
+		asicpp = &((*modpp)->asic);
+		sp_offset = eprom_hdr->scan_path_offset;
+		/* get rid of the dummy CAT asic and read the real one */
+		kfree((*modpp)->asic);
+		for(asic=0; asic < (*modpp)->num_asics; asic++) {
+			int j;
+			voyager_asic_t *asicp = *asicpp 
+				= kmalloc(sizeof(voyager_asic_t), GFP_KERNEL); /*&voyager_asic_storage[asic_count++];*/
+			voyager_sp_table_t *sp_table;
+			voyager_at_t *asic_table;
+			voyager_jtt_t *jtag_table;
+
+			if(asicp == NULL) {
+				printk("**WARNING** kmalloc failure in cat_init\n");
+				continue;
+			}
+			memset(asicp, 0, sizeof(voyager_asic_t));
+			asicpp = &(asicp->next);
+			asicp->asic_location = asic;
+			sp_table = (voyager_sp_table_t *)(eprom_buf + sp_offset);
+			asicp->asic_id = sp_table->asic_id;
+			asic_table = (voyager_at_t *)(eprom_buf + sp_table->asic_data_offset);
+			for(j=0; j<4; j++)
+				asicp->jtag_id[j] = asic_table->jtag_id[j];
+			jtag_table = (voyager_jtt_t *)(eprom_buf + asic_table->jtag_offset);
+			asicp->ireg_length = jtag_table->ireg_len;
+			asicp->bit_location = (*modpp)->inst_bits;
+			(*modpp)->inst_bits += asicp->ireg_length;
+			if(asicp->ireg_length > (*modpp)->largest_reg)
+				(*modpp)->largest_reg = asicp->ireg_length;
+			if (asicp->ireg_length < (*modpp)->smallest_reg ||
+			    (*modpp)->smallest_reg == 0)
+				(*modpp)->smallest_reg = asicp->ireg_length;
+			CDEBUG(("asic 0x%x, ireg_length=%d, bit_location=%d\n",
+				asicp->asic_id, asicp->ireg_length,
+				asicp->bit_location));
+			if(asicp->asic_id == VOYAGER_QUAD_QABC) {
+				CDEBUG(("VOYAGER CAT: QABC ASIC found\n"));
+				qabc_asic = asicp;
+			}
+			sp_offset += sizeof(voyager_sp_table_t);
+		}
+		CDEBUG(("Module inst_bits = %d, largest_reg = %d, smallest_reg=%d\n",
+			(*modpp)->inst_bits, (*modpp)->largest_reg,
+			(*modpp)->smallest_reg));
+		/* OK, now we have the QUAD ASICs set up, use them.
+		 * we need to:
+		 *
+		 * 1. Find the Memory area for the Quad CPIs.
+		 * 2. Find the Extended VIC processor
+		 * 3. Configure a second extended VIC processor (This
+		 *    cannot be done for the 51xx.
+		 * */
+		outb(VOYAGER_CAT_RUN, CAT_CMD);
+		cat_connect(*modpp, (*modpp)->asic);
+		CDEBUG(("CAT CONNECTED!!\n"));
+		cat_subread(*modpp, qabc_asic, 0, sizeof(qabc_data), qabc_data);
+		qic_addr = qabc_data[5] << 8;
+		qic_addr = (qic_addr | qabc_data[6]) << 8;
+		qic_addr = (qic_addr | qabc_data[7]) << 8;
+		printk("Module \"%s\": Quad Processor Card; CPI 0x%lx, SET=0x%x\n",
+		       cat_module_name(i), qic_addr, qabc_data[8]);
+#if 0				/* plumbing fails---FIXME */
+		if((qabc_data[8] & 0xf0) == 0) {
+			/* FIXME: 32 way 8 CPU slot monster cannot be
+			 * plumbed this way---need to check for it */
+
+			printk("Plumbing second Extended Quad Processor\n");
+			/* second VIC line hardwired to Quad CPU 1 */
+			qabc_data[8] |= 0x20;
+			cat_subwrite(*modpp, qabc_asic, 8, 1, &qabc_data[8]);
+#ifdef VOYAGER_CAT_DEBUG
+			/* verify plumbing */
+			cat_subread(*modpp, qabc_asic, 8, 1, &qabc_data[8]);
+			if((qabc_data[8] & 0xf0) == 0) {
+				CDEBUG(("PLUMBING FAILED: 0x%x\n", qabc_data[8]));
+			}
+#endif
+		}
+#endif
+
+		{
+			struct resource *res = kmalloc(sizeof(struct resource),GFP_KERNEL);
+			memset(res, 0, sizeof(struct resource));
+			res->name = kmalloc(128, GFP_KERNEL);
+			sprintf((char *)res->name, "Voyager %s Quad CPI", cat_module_name(i));
+			res->start = qic_addr;
+			res->end = qic_addr + 0x3ff;
+			request_resource(&iomem_resource, res);
+		}
+
+		qic_addr = (unsigned long)ioremap(qic_addr, 0x400);
+				
+		for(j = 0; j < 4; j++) {
+			__u8 cpu;
+
+			if(voyager_8slot) {
+				/* 8 slot has a different mapping,
+				 * each slot has only one vic line, so
+				 * 1 cpu in each slot must be < 8 */
+				cpu = (i & 0x07) + j*8;
+			} else {
+				cpu = (i & 0x03) + j*4;
+			}
+			if( (qabc_data[8] & (1<<j))) {
+				voyager_extended_vic_processors |= (1<<cpu);
+			}
+			if(qabc_data[8] & (1<<(j+4)) ) {
+				/* Second SET register plumbed: Quad
+				 * card has two VIC connected CPUs.
+				 * Secondary cannot be booted as a VIC
+				 * CPU */
+				voyager_extended_vic_processors |= (1<<cpu);
+				voyager_allowed_boot_processors &= (~(1<<cpu));
+			}
+
+			voyager_quad_processors |= (1<<cpu);
+			voyager_quad_cpi_addr[cpu] = (struct voyager_qic_cpi *)
+				(qic_addr+(j<<8));
+			CDEBUG(("CPU%d: CPI address 0x%lx\n", cpu,
+				(unsigned long)voyager_quad_cpi_addr[cpu]));
+		}
+		outb(VOYAGER_CAT_END, CAT_CMD);
+
+		
+		
+		*asicpp = NULL;
+		modpp = &((*modpp)->next);
+	}
+	*modpp = NULL;
+	printk("CAT Bus Initialisation finished: extended procs 0x%x, quad procs 0x%x, allowed vic boot = 0x%x\n", voyager_extended_vic_processors, voyager_quad_processors, voyager_allowed_boot_processors);
+	request_resource(&ioport_resource, &vic_res);
+	if(voyager_quad_processors)
+		request_resource(&ioport_resource, &qic_res);
+	/* set up the front power switch */
+}
+
+int
+voyager_cat_readb(__u8 module, __u8 asic, int reg)
+{
+	return 0;
+}
+
+static int
+cat_disconnect(voyager_module_t *modp, voyager_asic_t *asicp) 
+{
+	__u8 val;
+	int err = 0;
+
+	if(!modp->scan_path_connected)
+		return 0;
+	if(asicp->asic_id != VOYAGER_CAT_ID) {
+		CDEBUG(("cat_disconnect: ASIC is not CAT\n"));
+		return 1;
+	}
+	err = cat_read(modp, asicp, VOYAGER_SCANPATH, &val);
+	if(err) {
+		CDEBUG(("cat_disconnect: failed to read SCANPATH\n"));
+		return err;
+	}
+	val &= VOYAGER_DISCONNECT_ASIC;
+	err = cat_write(modp, asicp, VOYAGER_SCANPATH, val);
+	if(err) {
+		CDEBUG(("cat_disconnect: failed to write SCANPATH\n"));
+		return err;
+	}
+	outb(VOYAGER_CAT_END, CAT_CMD);
+	outb(VOYAGER_CAT_RUN, CAT_CMD);
+	modp->scan_path_connected = 0;
+
+	return 0;
+}
+
+static int
+cat_connect(voyager_module_t *modp, voyager_asic_t *asicp) 
+{
+	__u8 val;
+	int err = 0;
+
+	if(modp->scan_path_connected)
+		return 0;
+	if(asicp->asic_id != VOYAGER_CAT_ID) {
+		CDEBUG(("cat_connect: ASIC is not CAT\n"));
+		return 1;
+	}
+
+	err = cat_read(modp, asicp, VOYAGER_SCANPATH, &val);
+	if(err) {
+		CDEBUG(("cat_connect: failed to read SCANPATH\n"));
+		return err;
+	}
+	val |= VOYAGER_CONNECT_ASIC;
+	err = cat_write(modp, asicp, VOYAGER_SCANPATH, val);
+	if(err) {
+		CDEBUG(("cat_connect: failed to write SCANPATH\n"));
+		return err;
+	}
+	outb(VOYAGER_CAT_END, CAT_CMD);
+	outb(VOYAGER_CAT_RUN, CAT_CMD);
+	modp->scan_path_connected = 1;
+
+	return 0;
+}
+
+void
+voyager_cat_power_off(void)
+{
+	/* Power the machine off by writing to the PSI over the CAT
+         * bus */
+	__u8 data;
+	voyager_module_t psi = { 0 };
+	voyager_asic_t psi_asic = { 0 };
+
+	psi.asic = &psi_asic;
+	psi.asic->asic_id = VOYAGER_CAT_ID;
+	psi.asic->subaddr = VOYAGER_SUBADDR_HI;
+	psi.module_addr = VOYAGER_PSI;
+	psi.scan_path_connected = 0;
+
+	outb(VOYAGER_CAT_END, CAT_CMD);
+	/* Connect the PSI to the CAT Bus */
+	outb(VOYAGER_CAT_DESELECT, VOYAGER_CAT_CONFIG_PORT);
+	outb(VOYAGER_PSI, VOYAGER_CAT_CONFIG_PORT);
+	outb(VOYAGER_CAT_RUN, CAT_CMD);
+	cat_disconnect(&psi, &psi_asic);
+	/* Read the status */
+	cat_subread(&psi, &psi_asic, VOYAGER_PSI_GENERAL_REG, 1, &data);
+	outb(VOYAGER_CAT_END, CAT_CMD);
+	CDEBUG(("PSI STATUS 0x%x\n", data));
+	/* These two writes are power off prep and perform */
+	data = PSI_CLEAR;
+	outb(VOYAGER_CAT_RUN, CAT_CMD);
+	cat_subwrite(&psi, &psi_asic, VOYAGER_PSI_GENERAL_REG, 1, &data);
+	outb(VOYAGER_CAT_END, CAT_CMD);
+	data = PSI_POWER_DOWN;
+	outb(VOYAGER_CAT_RUN, CAT_CMD);
+	cat_subwrite(&psi, &psi_asic, VOYAGER_PSI_GENERAL_REG, 1, &data);
+	outb(VOYAGER_CAT_END, CAT_CMD);
+}
+
+struct voyager_status voyager_status = { 0 };
+
+void
+voyager_cat_psi(__u8 cmd, __u16 reg, __u8 *data)
+{
+	voyager_module_t psi = { 0 };
+	voyager_asic_t psi_asic = { 0 };
+
+	psi.asic = &psi_asic;
+	psi.asic->asic_id = VOYAGER_CAT_ID;
+	psi.asic->subaddr = VOYAGER_SUBADDR_HI;
+	psi.module_addr = VOYAGER_PSI;
+	psi.scan_path_connected = 0;
+
+	outb(VOYAGER_CAT_END, CAT_CMD);
+	/* Connect the PSI to the CAT Bus */
+	outb(VOYAGER_CAT_DESELECT, VOYAGER_CAT_CONFIG_PORT);
+	outb(VOYAGER_PSI, VOYAGER_CAT_CONFIG_PORT);
+	outb(VOYAGER_CAT_RUN, CAT_CMD);
+	cat_disconnect(&psi, &psi_asic);
+	switch(cmd) {
+	case VOYAGER_PSI_READ:
+		cat_read(&psi, &psi_asic, reg, data);
+		break;
+	case VOYAGER_PSI_WRITE:
+		cat_write(&psi, &psi_asic, reg, *data);
+		break;
+	case VOYAGER_PSI_SUBREAD:
+		cat_subread(&psi, &psi_asic, reg, 1, data);
+		break;
+	case VOYAGER_PSI_SUBWRITE:
+		cat_subwrite(&psi, &psi_asic, reg, 1, data);
+		break;
+	default:
+		printk(KERN_ERR "Voyager PSI, unrecognised command %d\n", cmd);
+		break;
+	}
+	outb(VOYAGER_CAT_END, CAT_CMD);
+}
+
+void
+voyager_cat_do_common_interrupt(void)
+{
+	/* This is caused either by a memory parity error or something
+	 * in the PSI */
+	__u8 data;
+	voyager_module_t psi = { 0 };
+	voyager_asic_t psi_asic = { 0 };
+	struct voyager_psi psi_reg;
+	int i;
+ re_read:
+	psi.asic = &psi_asic;
+	psi.asic->asic_id = VOYAGER_CAT_ID;
+	psi.asic->subaddr = VOYAGER_SUBADDR_HI;
+	psi.module_addr = VOYAGER_PSI;
+	psi.scan_path_connected = 0;
+
+	outb(VOYAGER_CAT_END, CAT_CMD);
+	/* Connect the PSI to the CAT Bus */
+	outb(VOYAGER_CAT_DESELECT, VOYAGER_CAT_CONFIG_PORT);
+	outb(VOYAGER_PSI, VOYAGER_CAT_CONFIG_PORT);
+	outb(VOYAGER_CAT_RUN, CAT_CMD);
+	cat_disconnect(&psi, &psi_asic);
+	/* Read the status.  NOTE: Need to read *all* the PSI regs here
+	 * otherwise the cmn int will be reasserted */
+	for(i = 0; i < sizeof(psi_reg.regs); i++) {
+		cat_read(&psi, &psi_asic, i, &((__u8 *)&psi_reg.regs)[i]);
+	}
+	outb(VOYAGER_CAT_END, CAT_CMD);
+	if((psi_reg.regs.checkbit & 0x02) == 0) {
+		psi_reg.regs.checkbit |= 0x02;
+		cat_write(&psi, &psi_asic, 5, psi_reg.regs.checkbit);
+		printk("VOYAGER RE-READ PSI\n");
+		goto re_read;
+	}
+	outb(VOYAGER_CAT_RUN, CAT_CMD);
+	for(i = 0; i < sizeof(psi_reg.subregs); i++) {
+		/* This looks strange, but the PSI doesn't do auto increment
+		 * correctly */
+		cat_subread(&psi, &psi_asic, VOYAGER_PSI_SUPPLY_REG + i, 
+			    1, &((__u8 *)&psi_reg.subregs)[i]); 
+	}
+	outb(VOYAGER_CAT_END, CAT_CMD);
+#ifdef VOYAGER_CAT_DEBUG
+	printk("VOYAGER PSI: ");
+	for(i=0; i<sizeof(psi_reg.regs); i++)
+		printk("%02x ", ((__u8 *)&psi_reg.regs)[i]);
+	printk("\n           ");
+	for(i=0; i<sizeof(psi_reg.subregs); i++)
+		printk("%02x ", ((__u8 *)&psi_reg.subregs)[i]);
+	printk("\n");
+#endif
+	if(psi_reg.regs.intstatus & PSI_MON) {
+		/* switch off or power fail */
+
+		if(psi_reg.subregs.supply & PSI_SWITCH_OFF) {
+			if(voyager_status.switch_off) {
+				printk(KERN_ERR "Voyager front panel switch turned off again---Immediate power off!\n");
+				voyager_cat_power_off();
+				/* not reached */
+			} else {
+				printk(KERN_ERR "Voyager front panel switch turned off\n");
+				voyager_status.switch_off = 1;
+				voyager_status.request_from_kernel = 1;
+				up(&kvoyagerd_sem);
+			}
+			/* Tell the hardware we're taking care of the
+			 * shutdown, otherwise it will power the box off
+			 * within 3 seconds of the switch being pressed and,
+			 * which is much more important to us, continue to 
+			 * assert the common interrupt */
+			data = PSI_CLR_SWITCH_OFF;
+			outb(VOYAGER_CAT_RUN, CAT_CMD);
+			cat_subwrite(&psi, &psi_asic, VOYAGER_PSI_SUPPLY_REG,
+				     1, &data);
+			outb(VOYAGER_CAT_END, CAT_CMD);
+		} else {
+
+			VDEBUG(("Voyager ac fail reg 0x%x\n",
+				psi_reg.subregs.ACfail));
+			if((psi_reg.subregs.ACfail & AC_FAIL_STAT_CHANGE) == 0) {
+				/* No further update */
+				return;
+			}
+#if 0
+			/* Don't bother trying to find out who failed.
+			 * FIXME: This probably makes the code incorrect on
+			 * anything other than a 345x */
+			for(i=0; i< 5; i++) {
+				if( psi_reg.subregs.ACfail &(1<<i)) {
+					break;
+				}
+			}
+			printk(KERN_NOTICE "AC FAIL IN SUPPLY %d\n", i);
+#endif
+			/* DON'T do this: it shuts down the AC PSI 
+			outb(VOYAGER_CAT_RUN, CAT_CMD);
+			data = PSI_MASK_MASK | i;
+			cat_subwrite(&psi, &psi_asic, VOYAGER_PSI_MASK,
+				     1, &data);
+			outb(VOYAGER_CAT_END, CAT_CMD);
+			*/
+			printk(KERN_ERR "Voyager AC power failure\n");
+			outb(VOYAGER_CAT_RUN, CAT_CMD);
+			data = PSI_COLD_START;
+			cat_subwrite(&psi, &psi_asic, VOYAGER_PSI_GENERAL_REG,
+				     1, &data);
+			outb(VOYAGER_CAT_END, CAT_CMD);
+			voyager_status.power_fail = 1;
+			voyager_status.request_from_kernel = 1;
+			up(&kvoyagerd_sem);
+		}
+		
+		
+	} else if(psi_reg.regs.intstatus & PSI_FAULT) {
+		/* Major fault! */
+		printk(KERN_ERR "Voyager PSI Detected major fault, immediate power off!\n");
+		voyager_cat_power_off();
+		/* not reached */
+	} else if(psi_reg.regs.intstatus & (PSI_DC_FAIL | PSI_ALARM
+					    | PSI_CURRENT | PSI_DVM
+					    | PSI_PSCFAULT | PSI_STAT_CHG)) {
+		/* other psi fault */
+
+		printk(KERN_WARNING "Voyager PSI status 0x%x\n", data);
+		/* clear the PSI fault */
+		outb(VOYAGER_CAT_RUN, CAT_CMD);
+		cat_write(&psi, &psi_asic, VOYAGER_PSI_STATUS_REG, 0);
+		outb(VOYAGER_CAT_END, CAT_CMD);
+	}
+}
diff --git a/arch/i386/mach-voyager/voyager_smp.c b/arch/i386/mach-voyager/voyager_smp.c
new file mode 100644
index 000000000000..903d739ca74a
--- /dev/null
+++ b/arch/i386/mach-voyager/voyager_smp.c
@@ -0,0 +1,1931 @@
+/* -*- mode: c; c-basic-offset: 8 -*- */
+
+/* Copyright (C) 1999,2001
+ *
+ * Author: J.E.J.Bottomley@HansenPartnership.com
+ *
+ * linux/arch/i386/kernel/voyager_smp.c
+ *
+ * This file provides all the same external entries as smp.c but uses
+ * the voyager hal to provide the functionality
+ */
+#include <linux/config.h>
+#include <linux/mm.h>
+#include <linux/kernel_stat.h>
+#include <linux/delay.h>
+#include <linux/mc146818rtc.h>
+#include <linux/cache.h>
+#include <linux/interrupt.h>
+#include <linux/smp_lock.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/bootmem.h>
+#include <linux/completion.h>
+#include <asm/desc.h>
+#include <asm/voyager.h>
+#include <asm/vic.h>
+#include <asm/mtrr.h>
+#include <asm/pgalloc.h>
+#include <asm/tlbflush.h>
+#include <asm/arch_hooks.h>
+
+#include <linux/irq.h>
+
+/* TLB state -- visible externally, indexed physically */
+DEFINE_PER_CPU(struct tlb_state, cpu_tlbstate) ____cacheline_aligned = { &init_mm, 0 };
+
+/* CPU IRQ affinity -- set to all ones initially */
+static unsigned long cpu_irq_affinity[NR_CPUS] __cacheline_aligned = { [0 ... NR_CPUS-1]  = ~0UL };
+
+/* per CPU data structure (for /proc/cpuinfo et al), visible externally
+ * indexed physically */
+struct cpuinfo_x86 cpu_data[NR_CPUS] __cacheline_aligned;
+
+/* physical ID of the CPU used to boot the system */
+unsigned char boot_cpu_id;
+
+/* The memory line addresses for the Quad CPIs */
+struct voyager_qic_cpi *voyager_quad_cpi_addr[NR_CPUS] __cacheline_aligned;
+
+/* The masks for the Extended VIC processors, filled in by cat_init */
+__u32 voyager_extended_vic_processors = 0;
+
+/* Masks for the extended Quad processors which cannot be VIC booted */
+__u32 voyager_allowed_boot_processors = 0;
+
+/* The mask for the Quad Processors (both extended and non-extended) */
+__u32 voyager_quad_processors = 0;
+
+/* Total count of live CPUs, used in process.c to display
+ * the CPU information and in irq.c for the per CPU irq
+ * activity count.  Finally exported by i386_ksyms.c */
+static int voyager_extended_cpus = 1;
+
+/* Have we found an SMP box - used by time.c to do the profiling
+   interrupt for timeslicing; do not set to 1 until the per CPU timer
+   interrupt is active */
+int smp_found_config = 0;
+
+/* Used for the invalidate map that's also checked in the spinlock */
+static volatile unsigned long smp_invalidate_needed;
+
+/* Bitmask of currently online CPUs - used by setup.c for
+   /proc/cpuinfo, visible externally but still physical */
+cpumask_t cpu_online_map = CPU_MASK_NONE;
+
+/* Bitmask of CPUs present in the system - exported by i386_syms.c, used
+ * by scheduler but indexed physically */
+cpumask_t phys_cpu_present_map = CPU_MASK_NONE;
+
+
+/* The internal functions */
+static void send_CPI(__u32 cpuset, __u8 cpi);
+static void ack_CPI(__u8 cpi);
+static int ack_QIC_CPI(__u8 cpi);
+static void ack_special_QIC_CPI(__u8 cpi);
+static void ack_VIC_CPI(__u8 cpi);
+static void send_CPI_allbutself(__u8 cpi);
+static void enable_vic_irq(unsigned int irq);
+static void disable_vic_irq(unsigned int irq);
+static unsigned int startup_vic_irq(unsigned int irq);
+static void enable_local_vic_irq(unsigned int irq);
+static void disable_local_vic_irq(unsigned int irq);
+static void before_handle_vic_irq(unsigned int irq);
+static void after_handle_vic_irq(unsigned int irq);
+static void set_vic_irq_affinity(unsigned int irq, cpumask_t mask);
+static void ack_vic_irq(unsigned int irq);
+static void vic_enable_cpi(void);
+static void do_boot_cpu(__u8 cpuid);
+static void do_quad_bootstrap(void);
+static inline void wrapper_smp_local_timer_interrupt(struct pt_regs *);
+
+int hard_smp_processor_id(void);
+
+/* Inline functions */
+static inline void
+send_one_QIC_CPI(__u8 cpu, __u8 cpi)
+{
+	voyager_quad_cpi_addr[cpu]->qic_cpi[cpi].cpi =
+		(smp_processor_id() << 16) + cpi;
+}
+
+static inline void
+send_QIC_CPI(__u32 cpuset, __u8 cpi)
+{
+	int cpu;
+
+	for_each_online_cpu(cpu) {
+		if(cpuset & (1<<cpu)) {
+#ifdef VOYAGER_DEBUG
+			if(!cpu_isset(cpu, cpu_online_map))
+				VDEBUG(("CPU%d sending cpi %d to CPU%d not in cpu_online_map\n", hard_smp_processor_id(), cpi, cpu));
+#endif
+			send_one_QIC_CPI(cpu, cpi - QIC_CPI_OFFSET);
+		}
+	}
+}
+
+static inline void
+send_one_CPI(__u8 cpu, __u8 cpi)
+{
+	if(voyager_quad_processors & (1<<cpu))
+		send_one_QIC_CPI(cpu, cpi - QIC_CPI_OFFSET);
+	else
+		send_CPI(1<<cpu, cpi);
+}
+
+static inline void
+send_CPI_allbutself(__u8 cpi)
+{
+	__u8 cpu = smp_processor_id();
+	__u32 mask = cpus_addr(cpu_online_map)[0] & ~(1 << cpu);
+	send_CPI(mask, cpi);
+}
+
+static inline int
+is_cpu_quad(void)
+{
+	__u8 cpumask = inb(VIC_PROC_WHO_AM_I);
+	return ((cpumask & QUAD_IDENTIFIER) == QUAD_IDENTIFIER);
+}
+
+static inline int
+is_cpu_extended(void)
+{
+	__u8 cpu = hard_smp_processor_id();
+
+	return(voyager_extended_vic_processors & (1<<cpu));
+}
+
+static inline int
+is_cpu_vic_boot(void)
+{
+	__u8 cpu = hard_smp_processor_id();
+
+	return(voyager_extended_vic_processors
+	       & voyager_allowed_boot_processors & (1<<cpu));
+}
+
+
+static inline void
+ack_CPI(__u8 cpi)
+{
+	switch(cpi) {
+	case VIC_CPU_BOOT_CPI:
+		if(is_cpu_quad() && !is_cpu_vic_boot())
+			ack_QIC_CPI(cpi);
+		else
+			ack_VIC_CPI(cpi);
+		break;
+	case VIC_SYS_INT:
+	case VIC_CMN_INT: 
+		/* These are slightly strange.  Even on the Quad card,
+		 * They are vectored as VIC CPIs */
+		if(is_cpu_quad())
+			ack_special_QIC_CPI(cpi);
+		else
+			ack_VIC_CPI(cpi);
+		break;
+	default:
+		printk("VOYAGER ERROR: CPI%d is in common CPI code\n", cpi);
+		break;
+	}
+}
+
+/* local variables */
+
+/* The VIC IRQ descriptors -- these look almost identical to the
+ * 8259 IRQs except that masks and things must be kept per processor
+ */
+static struct hw_interrupt_type vic_irq_type = {
+	.typename = "VIC-level",
+	.startup = startup_vic_irq,
+	.shutdown = disable_vic_irq,
+	.enable = enable_vic_irq,
+	.disable = disable_vic_irq,
+	.ack = before_handle_vic_irq,
+	.end = after_handle_vic_irq,
+	.set_affinity = set_vic_irq_affinity,
+};
+
+/* used to count up as CPUs are brought on line (starts at 0) */
+static int cpucount = 0;
+
+/* steal a page from the bottom of memory for the trampoline and
+ * squirrel its address away here.  This will be in kernel virtual
+ * space */
+static __u32 trampoline_base;
+
+/* The per cpu profile stuff - used in smp_local_timer_interrupt */
+static DEFINE_PER_CPU(int, prof_multiplier) = 1;
+static DEFINE_PER_CPU(int, prof_old_multiplier) = 1;
+static DEFINE_PER_CPU(int, prof_counter) =  1;
+
+/* the map used to check if a CPU has booted */
+static __u32 cpu_booted_map;
+
+/* the synchronize flag used to hold all secondary CPUs spinning in
+ * a tight loop until the boot sequence is ready for them */
+static cpumask_t smp_commenced_mask = CPU_MASK_NONE;
+
+/* This is for the new dynamic CPU boot code */
+cpumask_t cpu_callin_map = CPU_MASK_NONE;
+cpumask_t cpu_callout_map = CPU_MASK_NONE;
+
+/* The per processor IRQ masks (these are usually kept in sync) */
+static __u16 vic_irq_mask[NR_CPUS] __cacheline_aligned;
+
+/* the list of IRQs to be enabled by the VIC_ENABLE_IRQ_CPI */
+static __u16 vic_irq_enable_mask[NR_CPUS] __cacheline_aligned = { 0 };
+
+/* Lock for enable/disable of VIC interrupts */
+static  __cacheline_aligned DEFINE_SPINLOCK(vic_irq_lock);
+
+/* The boot processor is correctly set up in PC mode when it 
+ * comes up, but the secondaries need their master/slave 8259
+ * pairs initializing correctly */
+
+/* Interrupt counters (per cpu) and total - used to try to
+ * even up the interrupt handling routines */
+static long vic_intr_total = 0;
+static long vic_intr_count[NR_CPUS] __cacheline_aligned = { 0 };
+static unsigned long vic_tick[NR_CPUS] __cacheline_aligned = { 0 };
+
+/* Since we can only use CPI0, we fake all the other CPIs */
+static unsigned long vic_cpi_mailbox[NR_CPUS] __cacheline_aligned;
+
+/* debugging routine to read the isr of the cpu's pic */
+static inline __u16
+vic_read_isr(void)
+{
+	__u16 isr;
+
+	outb(0x0b, 0xa0);
+	isr = inb(0xa0) << 8;
+	outb(0x0b, 0x20);
+	isr |= inb(0x20);
+
+	return isr;
+}
+
+static __init void
+qic_setup(void)
+{
+	if(!is_cpu_quad()) {
+		/* not a quad, no setup */
+		return;
+	}
+	outb(QIC_DEFAULT_MASK0, QIC_MASK_REGISTER0);
+	outb(QIC_CPI_ENABLE, QIC_MASK_REGISTER1);
+	
+	if(is_cpu_extended()) {
+		/* the QIC duplicate of the VIC base register */
+		outb(VIC_DEFAULT_CPI_BASE, QIC_VIC_CPI_BASE_REGISTER);
+		outb(QIC_DEFAULT_CPI_BASE, QIC_CPI_BASE_REGISTER);
+
+		/* FIXME: should set up the QIC timer and memory parity
+		 * error vectors here */
+	}
+}
+
+static __init void
+vic_setup_pic(void)
+{
+	outb(1, VIC_REDIRECT_REGISTER_1);
+	/* clear the claim registers for dynamic routing */
+	outb(0, VIC_CLAIM_REGISTER_0);
+	outb(0, VIC_CLAIM_REGISTER_1);
+
+	outb(0, VIC_PRIORITY_REGISTER);
+	/* Set the Primary and Secondary Microchannel vector
+	 * bases to be the same as the ordinary interrupts
+	 *
+	 * FIXME: This would be more efficient using separate
+	 * vectors. */
+	outb(FIRST_EXTERNAL_VECTOR, VIC_PRIMARY_MC_BASE);
+	outb(FIRST_EXTERNAL_VECTOR, VIC_SECONDARY_MC_BASE);
+	/* Now initiallise the master PIC belonging to this CPU by
+	 * sending the four ICWs */
+
+	/* ICW1: level triggered, ICW4 needed */
+	outb(0x19, 0x20);
+
+	/* ICW2: vector base */
+	outb(FIRST_EXTERNAL_VECTOR, 0x21);
+
+	/* ICW3: slave at line 2 */
+	outb(0x04, 0x21);
+
+	/* ICW4: 8086 mode */
+	outb(0x01, 0x21);
+
+	/* now the same for the slave PIC */
+
+	/* ICW1: level trigger, ICW4 needed */
+	outb(0x19, 0xA0);
+
+	/* ICW2: slave vector base */
+	outb(FIRST_EXTERNAL_VECTOR + 8, 0xA1);
+	
+	/* ICW3: slave ID */
+	outb(0x02, 0xA1);
+
+	/* ICW4: 8086 mode */
+	outb(0x01, 0xA1);
+}
+
+static void
+do_quad_bootstrap(void)
+{
+	if(is_cpu_quad() && is_cpu_vic_boot()) {
+		int i;
+		unsigned long flags;
+		__u8 cpuid = hard_smp_processor_id();
+
+		local_irq_save(flags);
+
+		for(i = 0; i<4; i++) {
+			/* FIXME: this would be >>3 &0x7 on the 32 way */
+			if(((cpuid >> 2) & 0x03) == i)
+				/* don't lower our own mask! */
+				continue;
+
+			/* masquerade as local Quad CPU */
+			outb(QIC_CPUID_ENABLE | i, QIC_PROCESSOR_ID);
+			/* enable the startup CPI */
+			outb(QIC_BOOT_CPI_MASK, QIC_MASK_REGISTER1);
+			/* restore cpu id */
+			outb(0, QIC_PROCESSOR_ID);
+		}
+		local_irq_restore(flags);
+	}
+}
+
+
+/* Set up all the basic stuff: read the SMP config and make all the
+ * SMP information reflect only the boot cpu.  All others will be
+ * brought on-line later. */
+void __init 
+find_smp_config(void)
+{
+	int i;
+
+	boot_cpu_id = hard_smp_processor_id();
+
+	printk("VOYAGER SMP: Boot cpu is %d\n", boot_cpu_id);
+
+	/* initialize the CPU structures (moved from smp_boot_cpus) */
+	for(i=0; i<NR_CPUS; i++) {
+		cpu_irq_affinity[i] = ~0;
+	}
+	cpu_online_map = cpumask_of_cpu(boot_cpu_id);
+
+	/* The boot CPU must be extended */
+	voyager_extended_vic_processors = 1<<boot_cpu_id;
+	/* initially, all of the first 8 cpu's can boot */
+	voyager_allowed_boot_processors = 0xff;
+	/* set up everything for just this CPU, we can alter
+	 * this as we start the other CPUs later */
+	/* now get the CPU disposition from the extended CMOS */
+	cpus_addr(phys_cpu_present_map)[0] = voyager_extended_cmos_read(VOYAGER_PROCESSOR_PRESENT_MASK);
+	cpus_addr(phys_cpu_present_map)[0] |= voyager_extended_cmos_read(VOYAGER_PROCESSOR_PRESENT_MASK + 1) << 8;
+	cpus_addr(phys_cpu_present_map)[0] |= voyager_extended_cmos_read(VOYAGER_PROCESSOR_PRESENT_MASK + 2) << 16;
+	cpus_addr(phys_cpu_present_map)[0] |= voyager_extended_cmos_read(VOYAGER_PROCESSOR_PRESENT_MASK + 3) << 24;
+	printk("VOYAGER SMP: phys_cpu_present_map = 0x%lx\n", cpus_addr(phys_cpu_present_map)[0]);
+	/* Here we set up the VIC to enable SMP */
+	/* enable the CPIs by writing the base vector to their register */
+	outb(VIC_DEFAULT_CPI_BASE, VIC_CPI_BASE_REGISTER);
+	outb(1, VIC_REDIRECT_REGISTER_1);
+	/* set the claim registers for static routing --- Boot CPU gets
+	 * all interrupts untill all other CPUs started */
+	outb(0xff, VIC_CLAIM_REGISTER_0);
+	outb(0xff, VIC_CLAIM_REGISTER_1);
+	/* Set the Primary and Secondary Microchannel vector
+	 * bases to be the same as the ordinary interrupts
+	 *
+	 * FIXME: This would be more efficient using separate
+	 * vectors. */
+	outb(FIRST_EXTERNAL_VECTOR, VIC_PRIMARY_MC_BASE);
+	outb(FIRST_EXTERNAL_VECTOR, VIC_SECONDARY_MC_BASE);
+
+	/* Finally tell the firmware that we're driving */
+	outb(inb(VOYAGER_SUS_IN_CONTROL_PORT) | VOYAGER_IN_CONTROL_FLAG,
+	     VOYAGER_SUS_IN_CONTROL_PORT);
+
+	current_thread_info()->cpu = boot_cpu_id;
+}
+
+/*
+ *	The bootstrap kernel entry code has set these up. Save them
+ *	for a given CPU, id is physical */
+void __init
+smp_store_cpu_info(int id)
+{
+	struct cpuinfo_x86 *c=&cpu_data[id];
+
+	*c = boot_cpu_data;
+
+	identify_cpu(c);
+}
+
+/* set up the trampoline and return the physical address of the code */
+static __u32 __init
+setup_trampoline(void)
+{
+	/* these two are global symbols in trampoline.S */
+	extern __u8 trampoline_end[];
+	extern __u8 trampoline_data[];
+
+	memcpy((__u8 *)trampoline_base, trampoline_data,
+	       trampoline_end - trampoline_data);
+	return virt_to_phys((__u8 *)trampoline_base);
+}
+
+/* Routine initially called when a non-boot CPU is brought online */
+static void __init
+start_secondary(void *unused)
+{
+	__u8 cpuid = hard_smp_processor_id();
+	/* external functions not defined in the headers */
+	extern void calibrate_delay(void);
+
+	cpu_init();
+
+	/* OK, we're in the routine */
+	ack_CPI(VIC_CPU_BOOT_CPI);
+
+	/* setup the 8259 master slave pair belonging to this CPU ---
+         * we won't actually receive any until the boot CPU
+         * relinquishes it's static routing mask */
+	vic_setup_pic();
+
+	qic_setup();
+
+	if(is_cpu_quad() && !is_cpu_vic_boot()) {
+		/* clear the boot CPI */
+		__u8 dummy;
+
+		dummy = voyager_quad_cpi_addr[cpuid]->qic_cpi[VIC_CPU_BOOT_CPI].cpi;
+		printk("read dummy %d\n", dummy);
+	}
+
+	/* lower the mask to receive CPIs */
+	vic_enable_cpi();
+
+	VDEBUG(("VOYAGER SMP: CPU%d, stack at about %p\n", cpuid, &cpuid));
+
+	/* enable interrupts */
+	local_irq_enable();
+
+	/* get our bogomips */
+	calibrate_delay();
+
+	/* save our processor parameters */
+	smp_store_cpu_info(cpuid);
+
+	/* if we're a quad, we may need to bootstrap other CPUs */
+	do_quad_bootstrap();
+
+	/* FIXME: this is rather a poor hack to prevent the CPU
+	 * activating softirqs while it's supposed to be waiting for
+	 * permission to proceed.  Without this, the new per CPU stuff
+	 * in the softirqs will fail */
+	local_irq_disable();
+	cpu_set(cpuid, cpu_callin_map);
+
+	/* signal that we're done */
+	cpu_booted_map = 1;
+
+	while (!cpu_isset(cpuid, smp_commenced_mask))
+		rep_nop();
+	local_irq_enable();
+
+	local_flush_tlb();
+
+	cpu_set(cpuid, cpu_online_map);
+	wmb();
+	cpu_idle();
+}
+
+
+/* Routine to kick start the given CPU and wait for it to report ready
+ * (or timeout in startup).  When this routine returns, the requested
+ * CPU is either fully running and configured or known to be dead.
+ *
+ * We call this routine sequentially 1 CPU at a time, so no need for
+ * locking */
+
+static void __init
+do_boot_cpu(__u8 cpu)
+{
+	struct task_struct *idle;
+	int timeout;
+	unsigned long flags;
+	int quad_boot = (1<<cpu) & voyager_quad_processors 
+		& ~( voyager_extended_vic_processors
+		     & voyager_allowed_boot_processors);
+
+	/* For the 486, we can't use the 4Mb page table trick, so
+	 * must map a region of memory */
+#ifdef CONFIG_M486
+	int i;
+	unsigned long *page_table_copies = (unsigned long *)
+		__get_free_page(GFP_KERNEL);
+#endif
+	pgd_t orig_swapper_pg_dir0;
+
+	/* This is an area in head.S which was used to set up the
+	 * initial kernel stack.  We need to alter this to give the
+	 * booting CPU a new stack (taken from its idle process) */
+	extern struct {
+		__u8 *esp;
+		unsigned short ss;
+	} stack_start;
+	/* This is the format of the CPI IDT gate (in real mode) which
+	 * we're hijacking to boot the CPU */
+	union 	IDTFormat {
+		struct seg {
+			__u16	Offset;
+			__u16	Segment;
+		} idt;
+		__u32 val;
+	} hijack_source;
+
+	__u32 *hijack_vector;
+	__u32 start_phys_address = setup_trampoline();
+
+	/* There's a clever trick to this: The linux trampoline is
+	 * compiled to begin at absolute location zero, so make the
+	 * address zero but have the data segment selector compensate
+	 * for the actual address */
+	hijack_source.idt.Offset = start_phys_address & 0x000F;
+	hijack_source.idt.Segment = (start_phys_address >> 4) & 0xFFFF;
+
+	cpucount++;
+	idle = fork_idle(cpu);
+	if(IS_ERR(idle))
+		panic("failed fork for CPU%d", cpu);
+	idle->thread.eip = (unsigned long) start_secondary;
+	/* init_tasks (in sched.c) is indexed logically */
+	stack_start.esp = (void *) idle->thread.esp;
+
+	irq_ctx_init(cpu);
+
+	/* Note: Don't modify initial ss override */
+	VDEBUG(("VOYAGER SMP: Booting CPU%d at 0x%lx[%x:%x], stack %p\n", cpu, 
+		(unsigned long)hijack_source.val, hijack_source.idt.Segment,
+		hijack_source.idt.Offset, stack_start.esp));
+	/* set the original swapper_pg_dir[0] to map 0 to 4Mb transparently
+	 * (so that the booting CPU can find start_32 */
+	orig_swapper_pg_dir0 = swapper_pg_dir[0];
+#ifdef CONFIG_M486
+	if(page_table_copies == NULL)
+		panic("No free memory for 486 page tables\n");
+	for(i = 0; i < PAGE_SIZE/sizeof(unsigned long); i++)
+		page_table_copies[i] = (i * PAGE_SIZE) 
+			| _PAGE_RW | _PAGE_USER | _PAGE_PRESENT;
+
+	((unsigned long *)swapper_pg_dir)[0] = 
+		((virt_to_phys(page_table_copies)) & PAGE_MASK)
+		| _PAGE_RW | _PAGE_USER | _PAGE_PRESENT;
+#else
+	((unsigned long *)swapper_pg_dir)[0] = 
+		(virt_to_phys(pg0) & PAGE_MASK)
+		| _PAGE_RW | _PAGE_USER | _PAGE_PRESENT;
+#endif
+
+	if(quad_boot) {
+		printk("CPU %d: non extended Quad boot\n", cpu);
+		hijack_vector = (__u32 *)phys_to_virt((VIC_CPU_BOOT_CPI + QIC_DEFAULT_CPI_BASE)*4);
+		*hijack_vector = hijack_source.val;
+	} else {
+		printk("CPU%d: extended VIC boot\n", cpu);
+		hijack_vector = (__u32 *)phys_to_virt((VIC_CPU_BOOT_CPI + VIC_DEFAULT_CPI_BASE)*4);
+		*hijack_vector = hijack_source.val;
+		/* VIC errata, may also receive interrupt at this address */
+		hijack_vector = (__u32 *)phys_to_virt((VIC_CPU_BOOT_ERRATA_CPI + VIC_DEFAULT_CPI_BASE)*4);
+		*hijack_vector = hijack_source.val;
+	}
+	/* All non-boot CPUs start with interrupts fully masked.  Need
+	 * to lower the mask of the CPI we're about to send.  We do
+	 * this in the VIC by masquerading as the processor we're
+	 * about to boot and lowering its interrupt mask */
+	local_irq_save(flags);
+	if(quad_boot) {
+		send_one_QIC_CPI(cpu, VIC_CPU_BOOT_CPI);
+	} else {
+		outb(VIC_CPU_MASQUERADE_ENABLE | cpu, VIC_PROCESSOR_ID);
+		/* here we're altering registers belonging to `cpu' */
+		
+		outb(VIC_BOOT_INTERRUPT_MASK, 0x21);
+		/* now go back to our original identity */
+		outb(boot_cpu_id, VIC_PROCESSOR_ID);
+
+		/* and boot the CPU */
+
+		send_CPI((1<<cpu), VIC_CPU_BOOT_CPI);
+	}
+	cpu_booted_map = 0;
+	local_irq_restore(flags);
+
+	/* now wait for it to become ready (or timeout) */
+	for(timeout = 0; timeout < 50000; timeout++) {
+		if(cpu_booted_map)
+			break;
+		udelay(100);
+	}
+	/* reset the page table */
+	swapper_pg_dir[0] = orig_swapper_pg_dir0;
+	local_flush_tlb();
+#ifdef CONFIG_M486
+	free_page((unsigned long)page_table_copies);
+#endif
+	  
+	if (cpu_booted_map) {
+		VDEBUG(("CPU%d: Booted successfully, back in CPU %d\n",
+			cpu, smp_processor_id()));
+	
+		printk("CPU%d: ", cpu);
+		print_cpu_info(&cpu_data[cpu]);
+		wmb();
+		cpu_set(cpu, cpu_callout_map);
+	}
+	else {
+		printk("CPU%d FAILED TO BOOT: ", cpu);
+		if (*((volatile unsigned char *)phys_to_virt(start_phys_address))==0xA5)
+			printk("Stuck.\n");
+		else
+			printk("Not responding.\n");
+		
+		cpucount--;
+	}
+}
+
+void __init
+smp_boot_cpus(void)
+{
+	int i;
+
+	/* CAT BUS initialisation must be done after the memory */
+	/* FIXME: The L4 has a catbus too, it just needs to be
+	 * accessed in a totally different way */
+	if(voyager_level == 5) {
+		voyager_cat_init();
+
+		/* now that the cat has probed the Voyager System Bus, sanity
+		 * check the cpu map */
+		if( ((voyager_quad_processors | voyager_extended_vic_processors)
+		     & cpus_addr(phys_cpu_present_map)[0]) != cpus_addr(phys_cpu_present_map)[0]) {
+			/* should panic */
+			printk("\n\n***WARNING*** Sanity check of CPU present map FAILED\n");
+		}
+	} else if(voyager_level == 4)
+		voyager_extended_vic_processors = cpus_addr(phys_cpu_present_map)[0];
+
+	/* this sets up the idle task to run on the current cpu */
+	voyager_extended_cpus = 1;
+	/* Remove the global_irq_holder setting, it triggers a BUG() on
+	 * schedule at the moment */
+	//global_irq_holder = boot_cpu_id;
+
+	/* FIXME: Need to do something about this but currently only works
+	 * on CPUs with a tsc which none of mine have. 
+	smp_tune_scheduling();
+	 */
+	smp_store_cpu_info(boot_cpu_id);
+	printk("CPU%d: ", boot_cpu_id);
+	print_cpu_info(&cpu_data[boot_cpu_id]);
+
+	if(is_cpu_quad()) {
+		/* booting on a Quad CPU */
+		printk("VOYAGER SMP: Boot CPU is Quad\n");
+		qic_setup();
+		do_quad_bootstrap();
+	}
+
+	/* enable our own CPIs */
+	vic_enable_cpi();
+
+	cpu_set(boot_cpu_id, cpu_online_map);
+	cpu_set(boot_cpu_id, cpu_callout_map);
+	
+	/* loop over all the extended VIC CPUs and boot them.  The 
+	 * Quad CPUs must be bootstrapped by their extended VIC cpu */
+	for(i = 0; i < NR_CPUS; i++) {
+		if(i == boot_cpu_id || !cpu_isset(i, phys_cpu_present_map))
+			continue;
+		do_boot_cpu(i);
+		/* This udelay seems to be needed for the Quad boots
+		 * don't remove unless you know what you're doing */
+		udelay(1000);
+	}
+	/* we could compute the total bogomips here, but why bother?,
+	 * Code added from smpboot.c */
+	{
+		unsigned long bogosum = 0;
+		for (i = 0; i < NR_CPUS; i++)
+			if (cpu_isset(i, cpu_online_map))
+				bogosum += cpu_data[i].loops_per_jiffy;
+		printk(KERN_INFO "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
+			cpucount+1,
+			bogosum/(500000/HZ),
+			(bogosum/(5000/HZ))%100);
+	}
+	voyager_extended_cpus = hweight32(voyager_extended_vic_processors);
+	printk("VOYAGER: Extended (interrupt handling CPUs): %d, non-extended: %d\n", voyager_extended_cpus, num_booting_cpus() - voyager_extended_cpus);
+	/* that's it, switch to symmetric mode */
+	outb(0, VIC_PRIORITY_REGISTER);
+	outb(0, VIC_CLAIM_REGISTER_0);
+	outb(0, VIC_CLAIM_REGISTER_1);
+	
+	VDEBUG(("VOYAGER SMP: Booted with %d CPUs\n", num_booting_cpus()));
+}
+
+/* Reload the secondary CPUs task structure (this function does not
+ * return ) */
+void __init 
+initialize_secondary(void)
+{
+#if 0
+	// AC kernels only
+	set_current(hard_get_current());
+#endif
+
+	/*
+	 * We don't actually need to load the full TSS,
+	 * basically just the stack pointer and the eip.
+	 */
+
+	asm volatile(
+		"movl %0,%%esp\n\t"
+		"jmp *%1"
+		:
+		:"r" (current->thread.esp),"r" (current->thread.eip));
+}
+
+/* handle a Voyager SYS_INT -- If we don't, the base board will
+ * panic the system.
+ *
+ * System interrupts occur because some problem was detected on the
+ * various busses.  To find out what you have to probe all the
+ * hardware via the CAT bus.  FIXME: At the moment we do nothing. */
+fastcall void
+smp_vic_sys_interrupt(struct pt_regs *regs)
+{
+	ack_CPI(VIC_SYS_INT);
+	printk("Voyager SYSTEM INTERRUPT\n");
+}
+
+/* Handle a voyager CMN_INT; These interrupts occur either because of
+ * a system status change or because a single bit memory error
+ * occurred.  FIXME: At the moment, ignore all this. */
+fastcall void
+smp_vic_cmn_interrupt(struct pt_regs *regs)
+{
+	static __u8 in_cmn_int = 0;
+	static DEFINE_SPINLOCK(cmn_int_lock);
+
+	/* common ints are broadcast, so make sure we only do this once */
+	_raw_spin_lock(&cmn_int_lock);
+	if(in_cmn_int)
+		goto unlock_end;
+
+	in_cmn_int++;
+	_raw_spin_unlock(&cmn_int_lock);
+
+	VDEBUG(("Voyager COMMON INTERRUPT\n"));
+
+	if(voyager_level == 5)
+		voyager_cat_do_common_interrupt();
+
+	_raw_spin_lock(&cmn_int_lock);
+	in_cmn_int = 0;
+ unlock_end:
+	_raw_spin_unlock(&cmn_int_lock);
+	ack_CPI(VIC_CMN_INT);
+}
+
+/*
+ * Reschedule call back. Nothing to do, all the work is done
+ * automatically when we return from the interrupt.  */
+static void
+smp_reschedule_interrupt(void)
+{
+	/* do nothing */
+}
+
+static struct mm_struct * flush_mm;
+static unsigned long flush_va;
+static DEFINE_SPINLOCK(tlbstate_lock);
+#define FLUSH_ALL	0xffffffff
+
+/*
+ * We cannot call mmdrop() because we are in interrupt context, 
+ * instead update mm->cpu_vm_mask.
+ *
+ * We need to reload %cr3 since the page tables may be going
+ * away from under us..
+ */
+static inline void
+leave_mm (unsigned long cpu)
+{
+	if (per_cpu(cpu_tlbstate, cpu).state == TLBSTATE_OK)
+		BUG();
+	cpu_clear(cpu, per_cpu(cpu_tlbstate, cpu).active_mm->cpu_vm_mask);
+	load_cr3(swapper_pg_dir);
+}
+
+
+/*
+ * Invalidate call-back
+ */
+static void 
+smp_invalidate_interrupt(void)
+{
+	__u8 cpu = smp_processor_id();
+
+	if (!test_bit(cpu, &smp_invalidate_needed))
+		return;
+	/* This will flood messages.  Don't uncomment unless you see
+	 * Problems with cross cpu invalidation
+	VDEBUG(("VOYAGER SMP: CPU%d received INVALIDATE_CPI\n",
+		smp_processor_id()));
+	*/
+
+	if (flush_mm == per_cpu(cpu_tlbstate, cpu).active_mm) {
+		if (per_cpu(cpu_tlbstate, cpu).state == TLBSTATE_OK) {
+			if (flush_va == FLUSH_ALL)
+				local_flush_tlb();
+			else
+				__flush_tlb_one(flush_va);
+		} else
+			leave_mm(cpu);
+	}
+	smp_mb__before_clear_bit();
+	clear_bit(cpu, &smp_invalidate_needed);
+	smp_mb__after_clear_bit();
+}
+
+/* All the new flush operations for 2.4 */
+
+
+/* This routine is called with a physical cpu mask */
+static void
+flush_tlb_others (unsigned long cpumask, struct mm_struct *mm,
+						unsigned long va)
+{
+	int stuck = 50000;
+
+	if (!cpumask)
+		BUG();
+	if ((cpumask & cpus_addr(cpu_online_map)[0]) != cpumask)
+		BUG();
+	if (cpumask & (1 << smp_processor_id()))
+		BUG();
+	if (!mm)
+		BUG();
+
+	spin_lock(&tlbstate_lock);
+	
+	flush_mm = mm;
+	flush_va = va;
+	atomic_set_mask(cpumask, &smp_invalidate_needed);
+	/*
+	 * We have to send the CPI only to
+	 * CPUs affected.
+	 */
+	send_CPI(cpumask, VIC_INVALIDATE_CPI);
+
+	while (smp_invalidate_needed) {
+		mb();
+		if(--stuck == 0) {
+			printk("***WARNING*** Stuck doing invalidate CPI (CPU%d)\n", smp_processor_id());
+			break;
+		}
+	}
+
+	/* Uncomment only to debug invalidation problems
+	VDEBUG(("VOYAGER SMP: Completed invalidate CPI (CPU%d)\n", cpu));
+	*/
+
+	flush_mm = NULL;
+	flush_va = 0;
+	spin_unlock(&tlbstate_lock);
+}
+
+void
+flush_tlb_current_task(void)
+{
+	struct mm_struct *mm = current->mm;
+	unsigned long cpu_mask;
+
+	preempt_disable();
+
+	cpu_mask = cpus_addr(mm->cpu_vm_mask)[0] & ~(1 << smp_processor_id());
+	local_flush_tlb();
+	if (cpu_mask)
+		flush_tlb_others(cpu_mask, mm, FLUSH_ALL);
+
+	preempt_enable();
+}
+
+
+void
+flush_tlb_mm (struct mm_struct * mm)
+{
+	unsigned long cpu_mask;
+
+	preempt_disable();
+
+	cpu_mask = cpus_addr(mm->cpu_vm_mask)[0] & ~(1 << smp_processor_id());
+
+	if (current->active_mm == mm) {
+		if (current->mm)
+			local_flush_tlb();
+		else
+			leave_mm(smp_processor_id());
+	}
+	if (cpu_mask)
+		flush_tlb_others(cpu_mask, mm, FLUSH_ALL);
+
+	preempt_enable();
+}
+
+void flush_tlb_page(struct vm_area_struct * vma, unsigned long va)
+{
+	struct mm_struct *mm = vma->vm_mm;
+	unsigned long cpu_mask;
+
+	preempt_disable();
+
+	cpu_mask = cpus_addr(mm->cpu_vm_mask)[0] & ~(1 << smp_processor_id());
+	if (current->active_mm == mm) {
+		if(current->mm)
+			__flush_tlb_one(va);
+		 else
+		 	leave_mm(smp_processor_id());
+	}
+
+	if (cpu_mask)
+		flush_tlb_others(cpu_mask, mm, va);
+
+	preempt_enable();
+}
+
+/* enable the requested IRQs */
+static void
+smp_enable_irq_interrupt(void)
+{
+	__u8 irq;
+	__u8 cpu = get_cpu();
+
+	VDEBUG(("VOYAGER SMP: CPU%d enabling irq mask 0x%x\n", cpu,
+	       vic_irq_enable_mask[cpu]));
+
+	spin_lock(&vic_irq_lock);
+	for(irq = 0; irq < 16; irq++) {
+		if(vic_irq_enable_mask[cpu] & (1<<irq))
+			enable_local_vic_irq(irq);
+	}
+	vic_irq_enable_mask[cpu] = 0;
+	spin_unlock(&vic_irq_lock);
+
+	put_cpu_no_resched();
+}
+	
+/*
+ *	CPU halt call-back
+ */
+static void
+smp_stop_cpu_function(void *dummy)
+{
+	VDEBUG(("VOYAGER SMP: CPU%d is STOPPING\n", smp_processor_id()));
+	cpu_clear(smp_processor_id(), cpu_online_map);
+	local_irq_disable();
+	for(;;)
+	       __asm__("hlt");
+}
+
+static DEFINE_SPINLOCK(call_lock);
+
+struct call_data_struct {
+	void (*func) (void *info);
+	void *info;
+	volatile unsigned long started;
+	volatile unsigned long finished;
+	int wait;
+};
+
+static struct call_data_struct * call_data;
+
+/* execute a thread on a new CPU.  The function to be called must be
+ * previously set up.  This is used to schedule a function for
+ * execution on all CPU's - set up the function then broadcast a
+ * function_interrupt CPI to come here on each CPU */
+static void
+smp_call_function_interrupt(void)
+{
+	void (*func) (void *info) = call_data->func;
+	void *info = call_data->info;
+	/* must take copy of wait because call_data may be replaced
+	 * unless the function is waiting for us to finish */
+	int wait = call_data->wait;
+	__u8 cpu = smp_processor_id();
+
+	/*
+	 * Notify initiating CPU that I've grabbed the data and am
+	 * about to execute the function
+	 */
+	mb();
+	if(!test_and_clear_bit(cpu, &call_data->started)) {
+		/* If the bit wasn't set, this could be a replay */
+		printk(KERN_WARNING "VOYAGER SMP: CPU %d received call funtion with no call pending\n", cpu);
+		return;
+	}
+	/*
+	 * At this point the info structure may be out of scope unless wait==1
+	 */
+	irq_enter();
+	(*func)(info);
+	irq_exit();
+	if (wait) {
+		mb();
+		clear_bit(cpu, &call_data->finished);
+	}
+}
+
+/* Call this function on all CPUs using the function_interrupt above 
+    <func> The function to run. This must be fast and non-blocking.
+    <info> An arbitrary pointer to pass to the function.
+    <retry> If true, keep retrying until ready.
+    <wait> If true, wait until function has completed on other CPUs.
+    [RETURNS] 0 on success, else a negative status code. Does not return until
+    remote CPUs are nearly ready to execute <<func>> or are or have executed.
+*/
+int
+smp_call_function (void (*func) (void *info), void *info, int retry,
+		   int wait)
+{
+	struct call_data_struct data;
+	__u32 mask = cpus_addr(cpu_online_map)[0];
+
+	mask &= ~(1<<smp_processor_id());
+
+	if (!mask)
+		return 0;
+
+	/* Can deadlock when called with interrupts disabled */
+	WARN_ON(irqs_disabled());
+
+	data.func = func;
+	data.info = info;
+	data.started = mask;
+	data.wait = wait;
+	if (wait)
+		data.finished = mask;
+
+	spin_lock(&call_lock);
+	call_data = &data;
+	wmb();
+	/* Send a message to all other CPUs and wait for them to respond */
+	send_CPI_allbutself(VIC_CALL_FUNCTION_CPI);
+
+	/* Wait for response */
+	while (data.started)
+		barrier();
+
+	if (wait)
+		while (data.finished)
+			barrier();
+
+	spin_unlock(&call_lock);
+
+	return 0;
+}
+
+/* Sorry about the name.  In an APIC based system, the APICs
+ * themselves are programmed to send a timer interrupt.  This is used
+ * by linux to reschedule the processor.  Voyager doesn't have this,
+ * so we use the system clock to interrupt one processor, which in
+ * turn, broadcasts a timer CPI to all the others --- we receive that
+ * CPI here.  We don't use this actually for counting so losing
+ * ticks doesn't matter 
+ *
+ * FIXME: For those CPU's which actually have a local APIC, we could
+ * try to use it to trigger this interrupt instead of having to
+ * broadcast the timer tick.  Unfortunately, all my pentium DYADs have
+ * no local APIC, so I can't do this
+ *
+ * This function is currently a placeholder and is unused in the code */
+fastcall void 
+smp_apic_timer_interrupt(struct pt_regs *regs)
+{
+	wrapper_smp_local_timer_interrupt(regs);
+}
+
+/* All of the QUAD interrupt GATES */
+fastcall void
+smp_qic_timer_interrupt(struct pt_regs *regs)
+{
+	ack_QIC_CPI(QIC_TIMER_CPI);
+	wrapper_smp_local_timer_interrupt(regs);
+}
+
+fastcall void
+smp_qic_invalidate_interrupt(struct pt_regs *regs)
+{
+	ack_QIC_CPI(QIC_INVALIDATE_CPI);
+	smp_invalidate_interrupt();
+}
+
+fastcall void
+smp_qic_reschedule_interrupt(struct pt_regs *regs)
+{
+	ack_QIC_CPI(QIC_RESCHEDULE_CPI);
+	smp_reschedule_interrupt();
+}
+
+fastcall void
+smp_qic_enable_irq_interrupt(struct pt_regs *regs)
+{
+	ack_QIC_CPI(QIC_ENABLE_IRQ_CPI);
+	smp_enable_irq_interrupt();
+}
+
+fastcall void
+smp_qic_call_function_interrupt(struct pt_regs *regs)
+{
+	ack_QIC_CPI(QIC_CALL_FUNCTION_CPI);
+	smp_call_function_interrupt();
+}
+
+fastcall void
+smp_vic_cpi_interrupt(struct pt_regs *regs)
+{
+	__u8 cpu = smp_processor_id();
+
+	if(is_cpu_quad())
+		ack_QIC_CPI(VIC_CPI_LEVEL0);
+	else
+		ack_VIC_CPI(VIC_CPI_LEVEL0);
+
+	if(test_and_clear_bit(VIC_TIMER_CPI, &vic_cpi_mailbox[cpu]))
+		wrapper_smp_local_timer_interrupt(regs);
+	if(test_and_clear_bit(VIC_INVALIDATE_CPI, &vic_cpi_mailbox[cpu]))
+		smp_invalidate_interrupt();
+	if(test_and_clear_bit(VIC_RESCHEDULE_CPI, &vic_cpi_mailbox[cpu]))
+		smp_reschedule_interrupt();
+	if(test_and_clear_bit(VIC_ENABLE_IRQ_CPI, &vic_cpi_mailbox[cpu]))
+		smp_enable_irq_interrupt();
+	if(test_and_clear_bit(VIC_CALL_FUNCTION_CPI, &vic_cpi_mailbox[cpu]))
+		smp_call_function_interrupt();
+}
+
+static void
+do_flush_tlb_all(void* info)
+{
+	unsigned long cpu = smp_processor_id();
+
+	__flush_tlb_all();
+	if (per_cpu(cpu_tlbstate, cpu).state == TLBSTATE_LAZY)
+		leave_mm(cpu);
+}
+
+
+/* flush the TLB of every active CPU in the system */
+void
+flush_tlb_all(void)
+{
+	on_each_cpu(do_flush_tlb_all, 0, 1, 1);
+}
+
+/* used to set up the trampoline for other CPUs when the memory manager
+ * is sorted out */
+void __init
+smp_alloc_memory(void)
+{
+	trampoline_base = (__u32)alloc_bootmem_low_pages(PAGE_SIZE);
+	if(__pa(trampoline_base) >= 0x93000)
+		BUG();
+}
+
+/* send a reschedule CPI to one CPU by physical CPU number*/
+void
+smp_send_reschedule(int cpu)
+{
+	send_one_CPI(cpu, VIC_RESCHEDULE_CPI);
+}
+
+
+int
+hard_smp_processor_id(void)
+{
+	__u8 i;
+	__u8 cpumask = inb(VIC_PROC_WHO_AM_I);
+	if((cpumask & QUAD_IDENTIFIER) == QUAD_IDENTIFIER)
+		return cpumask & 0x1F;
+
+	for(i = 0; i < 8; i++) {
+		if(cpumask & (1<<i))
+			return i;
+	}
+	printk("** WARNING ** Illegal cpuid returned by VIC: %d", cpumask);
+	return 0;
+}
+
+/* broadcast a halt to all other CPUs */
+void
+smp_send_stop(void)
+{
+	smp_call_function(smp_stop_cpu_function, NULL, 1, 1);
+}
+
+/* this function is triggered in time.c when a clock tick fires
+ * we need to re-broadcast the tick to all CPUs */
+void
+smp_vic_timer_interrupt(struct pt_regs *regs)
+{
+	send_CPI_allbutself(VIC_TIMER_CPI);
+	smp_local_timer_interrupt(regs);
+}
+
+static inline void
+wrapper_smp_local_timer_interrupt(struct pt_regs *regs)
+{
+	irq_enter();
+	smp_local_timer_interrupt(regs);
+	irq_exit();
+}
+
+/* local (per CPU) timer interrupt.  It does both profiling and
+ * process statistics/rescheduling.
+ *
+ * We do profiling in every local tick, statistics/rescheduling
+ * happen only every 'profiling multiplier' ticks. The default
+ * multiplier is 1 and it can be changed by writing the new multiplier
+ * value into /proc/profile.
+ */
+void
+smp_local_timer_interrupt(struct pt_regs * regs)
+{
+	int cpu = smp_processor_id();
+	long weight;
+
+	profile_tick(CPU_PROFILING, regs);
+	if (--per_cpu(prof_counter, cpu) <= 0) {
+		/*
+		 * The multiplier may have changed since the last time we got
+		 * to this point as a result of the user writing to
+		 * /proc/profile. In this case we need to adjust the APIC
+		 * timer accordingly.
+		 *
+		 * Interrupts are already masked off at this point.
+		 */
+		per_cpu(prof_counter,cpu) = per_cpu(prof_multiplier, cpu);
+		if (per_cpu(prof_counter, cpu) !=
+					per_cpu(prof_old_multiplier, cpu)) {
+			/* FIXME: need to update the vic timer tick here */
+			per_cpu(prof_old_multiplier, cpu) =
+						per_cpu(prof_counter, cpu);
+		}
+
+		update_process_times(user_mode(regs));
+	}
+
+	if( ((1<<cpu) & voyager_extended_vic_processors) == 0)
+		/* only extended VIC processors participate in
+		 * interrupt distribution */
+		return;
+
+	/*
+	 * We take the 'long' return path, and there every subsystem
+	 * grabs the apropriate locks (kernel lock/ irq lock).
+	 *
+	 * we might want to decouple profiling from the 'long path',
+	 * and do the profiling totally in assembly.
+	 *
+	 * Currently this isn't too much of an issue (performance wise),
+	 * we can take more than 100K local irqs per second on a 100 MHz P5.
+	 */
+
+	if((++vic_tick[cpu] & 0x7) != 0)
+		return;
+	/* get here every 16 ticks (about every 1/6 of a second) */
+
+	/* Change our priority to give someone else a chance at getting
+         * the IRQ. The algorithm goes like this:
+	 *
+	 * In the VIC, the dynamically routed interrupt is always
+	 * handled by the lowest priority eligible (i.e. receiving
+	 * interrupts) CPU.  If >1 eligible CPUs are equal lowest, the
+	 * lowest processor number gets it.
+	 *
+	 * The priority of a CPU is controlled by a special per-CPU
+	 * VIC priority register which is 3 bits wide 0 being lowest
+	 * and 7 highest priority..
+	 *
+	 * Therefore we subtract the average number of interrupts from
+	 * the number we've fielded.  If this number is negative, we
+	 * lower the activity count and if it is positive, we raise
+	 * it.
+	 *
+	 * I'm afraid this still leads to odd looking interrupt counts:
+	 * the totals are all roughly equal, but the individual ones
+	 * look rather skewed.
+	 *
+	 * FIXME: This algorithm is total crap when mixed with SMP
+	 * affinity code since we now try to even up the interrupt
+	 * counts when an affinity binding is keeping them on a
+	 * particular CPU*/
+	weight = (vic_intr_count[cpu]*voyager_extended_cpus
+		  - vic_intr_total) >> 4;
+	weight += 4;
+	if(weight > 7)
+		weight = 7;
+	if(weight < 0)
+		weight = 0;
+	
+	outb((__u8)weight, VIC_PRIORITY_REGISTER);
+
+#ifdef VOYAGER_DEBUG
+	if((vic_tick[cpu] & 0xFFF) == 0) {
+		/* print this message roughly every 25 secs */
+		printk("VOYAGER SMP: vic_tick[%d] = %lu, weight = %ld\n",
+		       cpu, vic_tick[cpu], weight);
+	}
+#endif
+}
+
+/* setup the profiling timer */
+int 
+setup_profiling_timer(unsigned int multiplier)
+{
+	int i;
+
+	if ( (!multiplier))
+		return -EINVAL;
+
+	/* 
+	 * Set the new multiplier for each CPU. CPUs don't start using the
+	 * new values until the next timer interrupt in which they do process
+	 * accounting.
+	 */
+	for (i = 0; i < NR_CPUS; ++i)
+		per_cpu(prof_multiplier, i) = multiplier;
+
+	return 0;
+}
+
+
+/*  The CPIs are handled in the per cpu 8259s, so they must be
+ *  enabled to be received: FIX: enabling the CPIs in the early
+ *  boot sequence interferes with bug checking; enable them later
+ *  on in smp_init */
+#define VIC_SET_GATE(cpi, vector) \
+	set_intr_gate((cpi) + VIC_DEFAULT_CPI_BASE, (vector))
+#define QIC_SET_GATE(cpi, vector) \
+	set_intr_gate((cpi) + QIC_DEFAULT_CPI_BASE, (vector))
+
+void __init
+smp_intr_init(void)
+{
+	int i;
+
+	/* initialize the per cpu irq mask to all disabled */
+	for(i = 0; i < NR_CPUS; i++)
+		vic_irq_mask[i] = 0xFFFF;
+
+	VIC_SET_GATE(VIC_CPI_LEVEL0, vic_cpi_interrupt);
+
+	VIC_SET_GATE(VIC_SYS_INT, vic_sys_interrupt);
+	VIC_SET_GATE(VIC_CMN_INT, vic_cmn_interrupt);
+
+	QIC_SET_GATE(QIC_TIMER_CPI, qic_timer_interrupt);
+	QIC_SET_GATE(QIC_INVALIDATE_CPI, qic_invalidate_interrupt);
+	QIC_SET_GATE(QIC_RESCHEDULE_CPI, qic_reschedule_interrupt);
+	QIC_SET_GATE(QIC_ENABLE_IRQ_CPI, qic_enable_irq_interrupt);
+	QIC_SET_GATE(QIC_CALL_FUNCTION_CPI, qic_call_function_interrupt);
+	
+
+	/* now put the VIC descriptor into the first 48 IRQs 
+	 *
+	 * This is for later: first 16 correspond to PC IRQs; next 16
+	 * are Primary MC IRQs and final 16 are Secondary MC IRQs */
+	for(i = 0; i < 48; i++)
+		irq_desc[i].handler = &vic_irq_type;
+}
+
+/* send a CPI at level cpi to a set of cpus in cpuset (set 1 bit per
+ * processor to receive CPI */
+static void
+send_CPI(__u32 cpuset, __u8 cpi)
+{
+	int cpu;
+	__u32 quad_cpuset = (cpuset & voyager_quad_processors);
+
+	if(cpi < VIC_START_FAKE_CPI) {
+		/* fake CPI are only used for booting, so send to the 
+		 * extended quads as well---Quads must be VIC booted */
+		outb((__u8)(cpuset), VIC_CPI_Registers[cpi]);
+		return;
+	}
+	if(quad_cpuset)
+		send_QIC_CPI(quad_cpuset, cpi);
+	cpuset &= ~quad_cpuset;
+	cpuset &= 0xff;		/* only first 8 CPUs vaild for VIC CPI */
+	if(cpuset == 0)
+		return;
+	for_each_online_cpu(cpu) {
+		if(cpuset & (1<<cpu))
+			set_bit(cpi, &vic_cpi_mailbox[cpu]);
+	}
+	if(cpuset)
+		outb((__u8)cpuset, VIC_CPI_Registers[VIC_CPI_LEVEL0]);
+}
+
+/* Acknowledge receipt of CPI in the QIC, clear in QIC hardware and
+ * set the cache line to shared by reading it.
+ *
+ * DON'T make this inline otherwise the cache line read will be
+ * optimised away
+ * */
+static int
+ack_QIC_CPI(__u8 cpi) {
+	__u8 cpu = hard_smp_processor_id();
+
+	cpi &= 7;
+
+	outb(1<<cpi, QIC_INTERRUPT_CLEAR1);
+	return voyager_quad_cpi_addr[cpu]->qic_cpi[cpi].cpi;
+}
+
+static void
+ack_special_QIC_CPI(__u8 cpi)
+{
+	switch(cpi) {
+	case VIC_CMN_INT:
+		outb(QIC_CMN_INT, QIC_INTERRUPT_CLEAR0);
+		break;
+	case VIC_SYS_INT:
+		outb(QIC_SYS_INT, QIC_INTERRUPT_CLEAR0);
+		break;
+	}
+	/* also clear at the VIC, just in case (nop for non-extended proc) */
+	ack_VIC_CPI(cpi);
+}
+
+/* Acknowledge receipt of CPI in the VIC (essentially an EOI) */
+static void
+ack_VIC_CPI(__u8 cpi)
+{
+#ifdef VOYAGER_DEBUG
+	unsigned long flags;
+	__u16 isr;
+	__u8 cpu = smp_processor_id();
+
+	local_irq_save(flags);
+	isr = vic_read_isr();
+	if((isr & (1<<(cpi &7))) == 0) {
+		printk("VOYAGER SMP: CPU%d lost CPI%d\n", cpu, cpi);
+	}
+#endif
+	/* send specific EOI; the two system interrupts have
+	 * bit 4 set for a separate vector but behave as the
+	 * corresponding 3 bit intr */
+	outb_p(0x60|(cpi & 7),0x20);
+
+#ifdef VOYAGER_DEBUG
+	if((vic_read_isr() & (1<<(cpi &7))) != 0) {
+		printk("VOYAGER SMP: CPU%d still asserting CPI%d\n", cpu, cpi);
+	}
+	local_irq_restore(flags);
+#endif
+}
+
+/* cribbed with thanks from irq.c */
+#define __byte(x,y) 	(((unsigned char *)&(y))[x])
+#define cached_21(cpu)	(__byte(0,vic_irq_mask[cpu]))
+#define cached_A1(cpu)	(__byte(1,vic_irq_mask[cpu]))
+
+static unsigned int
+startup_vic_irq(unsigned int irq)
+{
+	enable_vic_irq(irq);
+
+	return 0;
+}
+
+/* The enable and disable routines.  This is where we run into
+ * conflicting architectural philosophy.  Fundamentally, the voyager
+ * architecture does not expect to have to disable interrupts globally
+ * (the IRQ controllers belong to each CPU).  The processor masquerade
+ * which is used to start the system shouldn't be used in a running OS
+ * since it will cause great confusion if two separate CPUs drive to
+ * the same IRQ controller (I know, I've tried it).
+ *
+ * The solution is a variant on the NCR lazy SPL design:
+ *
+ * 1) To disable an interrupt, do nothing (other than set the
+ *    IRQ_DISABLED flag).  This dares the interrupt actually to arrive.
+ *
+ * 2) If the interrupt dares to come in, raise the local mask against
+ *    it (this will result in all the CPU masks being raised
+ *    eventually).
+ *
+ * 3) To enable the interrupt, lower the mask on the local CPU and
+ *    broadcast an Interrupt enable CPI which causes all other CPUs to
+ *    adjust their masks accordingly.  */
+
+static void
+enable_vic_irq(unsigned int irq)
+{
+	/* linux doesn't to processor-irq affinity, so enable on
+	 * all CPUs we know about */
+	int cpu = smp_processor_id(), real_cpu;
+	__u16 mask = (1<<irq);
+	__u32 processorList = 0;
+	unsigned long flags;
+
+	VDEBUG(("VOYAGER: enable_vic_irq(%d) CPU%d affinity 0x%lx\n",
+		irq, cpu, cpu_irq_affinity[cpu]));
+	spin_lock_irqsave(&vic_irq_lock, flags);
+	for_each_online_cpu(real_cpu) {
+		if(!(voyager_extended_vic_processors & (1<<real_cpu)))
+			continue;
+		if(!(cpu_irq_affinity[real_cpu] & mask)) {
+			/* irq has no affinity for this CPU, ignore */
+			continue;
+		}
+		if(real_cpu == cpu) {
+			enable_local_vic_irq(irq);
+		}
+		else if(vic_irq_mask[real_cpu] & mask) {
+			vic_irq_enable_mask[real_cpu] |= mask;
+			processorList |= (1<<real_cpu);
+		}
+	}
+	spin_unlock_irqrestore(&vic_irq_lock, flags);
+	if(processorList)
+		send_CPI(processorList, VIC_ENABLE_IRQ_CPI);
+}
+
+static void
+disable_vic_irq(unsigned int irq)
+{
+	/* lazy disable, do nothing */
+}
+
+static void
+enable_local_vic_irq(unsigned int irq)
+{
+	__u8 cpu = smp_processor_id();
+	__u16 mask = ~(1 << irq);
+	__u16 old_mask = vic_irq_mask[cpu];
+
+	vic_irq_mask[cpu] &= mask;
+	if(vic_irq_mask[cpu] == old_mask)
+		return;
+
+	VDEBUG(("VOYAGER DEBUG: Enabling irq %d in hardware on CPU %d\n",
+		irq, cpu));
+
+	if (irq & 8) {
+		outb_p(cached_A1(cpu),0xA1);
+		(void)inb_p(0xA1);
+	}
+	else {
+		outb_p(cached_21(cpu),0x21);
+		(void)inb_p(0x21);
+	}
+}
+
+static void
+disable_local_vic_irq(unsigned int irq)
+{
+	__u8 cpu = smp_processor_id();
+	__u16 mask = (1 << irq);
+	__u16 old_mask = vic_irq_mask[cpu];
+
+	if(irq == 7)
+		return;
+
+	vic_irq_mask[cpu] |= mask;
+	if(old_mask == vic_irq_mask[cpu])
+		return;
+
+	VDEBUG(("VOYAGER DEBUG: Disabling irq %d in hardware on CPU %d\n",
+		irq, cpu));
+
+	if (irq & 8) {
+		outb_p(cached_A1(cpu),0xA1);
+		(void)inb_p(0xA1);
+	}
+	else {
+		outb_p(cached_21(cpu),0x21);
+		(void)inb_p(0x21);
+	}
+}
+
+/* The VIC is level triggered, so the ack can only be issued after the
+ * interrupt completes.  However, we do Voyager lazy interrupt
+ * handling here: It is an extremely expensive operation to mask an
+ * interrupt in the vic, so we merely set a flag (IRQ_DISABLED).  If
+ * this interrupt actually comes in, then we mask and ack here to push
+ * the interrupt off to another CPU */
+static void
+before_handle_vic_irq(unsigned int irq)
+{
+	irq_desc_t *desc = irq_desc + irq;
+	__u8 cpu = smp_processor_id();
+
+	_raw_spin_lock(&vic_irq_lock);
+	vic_intr_total++;
+	vic_intr_count[cpu]++;
+
+	if(!(cpu_irq_affinity[cpu] & (1<<irq))) {
+		/* The irq is not in our affinity mask, push it off
+		 * onto another CPU */
+		VDEBUG(("VOYAGER DEBUG: affinity triggered disable of irq %d on cpu %d\n",
+			irq, cpu));
+		disable_local_vic_irq(irq);
+		/* set IRQ_INPROGRESS to prevent the handler in irq.c from
+		 * actually calling the interrupt routine */
+		desc->status |= IRQ_REPLAY | IRQ_INPROGRESS;
+	} else if(desc->status & IRQ_DISABLED) {
+		/* Damn, the interrupt actually arrived, do the lazy
+		 * disable thing. The interrupt routine in irq.c will
+		 * not handle a IRQ_DISABLED interrupt, so nothing more
+		 * need be done here */
+		VDEBUG(("VOYAGER DEBUG: lazy disable of irq %d on CPU %d\n",
+			irq, cpu));
+		disable_local_vic_irq(irq);
+		desc->status |= IRQ_REPLAY;
+	} else {
+		desc->status &= ~IRQ_REPLAY;
+	}
+
+	_raw_spin_unlock(&vic_irq_lock);
+}
+
+/* Finish the VIC interrupt: basically mask */
+static void
+after_handle_vic_irq(unsigned int irq)
+{
+	irq_desc_t *desc = irq_desc + irq;
+
+	_raw_spin_lock(&vic_irq_lock);
+	{
+		unsigned int status = desc->status & ~IRQ_INPROGRESS;
+#ifdef VOYAGER_DEBUG
+		__u16 isr;
+#endif
+
+		desc->status = status;
+		if ((status & IRQ_DISABLED))
+			disable_local_vic_irq(irq);
+#ifdef VOYAGER_DEBUG
+		/* DEBUG: before we ack, check what's in progress */
+		isr = vic_read_isr();
+		if((isr & (1<<irq) && !(status & IRQ_REPLAY)) == 0) {
+			int i;
+			__u8 cpu = smp_processor_id();
+			__u8 real_cpu;
+			int mask; /* Um... initialize me??? --RR */
+
+			printk("VOYAGER SMP: CPU%d lost interrupt %d\n",
+			       cpu, irq);
+			for_each_cpu(real_cpu, mask) {
+
+				outb(VIC_CPU_MASQUERADE_ENABLE | real_cpu,
+				     VIC_PROCESSOR_ID);
+				isr = vic_read_isr();
+				if(isr & (1<<irq)) {
+					printk("VOYAGER SMP: CPU%d ack irq %d\n",
+					       real_cpu, irq);
+					ack_vic_irq(irq);
+				}
+				outb(cpu, VIC_PROCESSOR_ID);
+			}
+		}
+#endif /* VOYAGER_DEBUG */
+		/* as soon as we ack, the interrupt is eligible for
+		 * receipt by another CPU so everything must be in
+		 * order here  */
+		ack_vic_irq(irq);
+		if(status & IRQ_REPLAY) {
+			/* replay is set if we disable the interrupt
+			 * in the before_handle_vic_irq() routine, so
+			 * clear the in progress bit here to allow the
+			 * next CPU to handle this correctly */
+			desc->status &= ~(IRQ_REPLAY | IRQ_INPROGRESS);
+		}
+#ifdef VOYAGER_DEBUG
+		isr = vic_read_isr();
+		if((isr & (1<<irq)) != 0)
+			printk("VOYAGER SMP: after_handle_vic_irq() after ack irq=%d, isr=0x%x\n",
+			       irq, isr);
+#endif /* VOYAGER_DEBUG */
+	}
+	_raw_spin_unlock(&vic_irq_lock);
+
+	/* All code after this point is out of the main path - the IRQ
+	 * may be intercepted by another CPU if reasserted */
+}
+
+
+/* Linux processor - interrupt affinity manipulations.
+ *
+ * For each processor, we maintain a 32 bit irq affinity mask.
+ * Initially it is set to all 1's so every processor accepts every
+ * interrupt.  In this call, we change the processor's affinity mask:
+ *
+ * Change from enable to disable:
+ *
+ * If the interrupt ever comes in to the processor, we will disable it
+ * and ack it to push it off to another CPU, so just accept the mask here.
+ *
+ * Change from disable to enable:
+ *
+ * change the mask and then do an interrupt enable CPI to re-enable on
+ * the selected processors */
+
+void
+set_vic_irq_affinity(unsigned int irq, cpumask_t mask)
+{
+	/* Only extended processors handle interrupts */
+	unsigned long real_mask;
+	unsigned long irq_mask = 1 << irq;
+	int cpu;
+
+	real_mask = cpus_addr(mask)[0] & voyager_extended_vic_processors;
+	
+	if(cpus_addr(mask)[0] == 0)
+		/* can't have no cpu's to accept the interrupt -- extremely
+		 * bad things will happen */
+		return;
+
+	if(irq == 0)
+		/* can't change the affinity of the timer IRQ.  This
+		 * is due to the constraint in the voyager
+		 * architecture that the CPI also comes in on and IRQ
+		 * line and we have chosen IRQ0 for this.  If you
+		 * raise the mask on this interrupt, the processor
+		 * will no-longer be able to accept VIC CPIs */
+		return;
+
+	if(irq >= 32) 
+		/* You can only have 32 interrupts in a voyager system
+		 * (and 32 only if you have a secondary microchannel
+		 * bus) */
+		return;
+
+	for_each_online_cpu(cpu) {
+		unsigned long cpu_mask = 1 << cpu;
+		
+		if(cpu_mask & real_mask) {
+			/* enable the interrupt for this cpu */
+			cpu_irq_affinity[cpu] |= irq_mask;
+		} else {
+			/* disable the interrupt for this cpu */
+			cpu_irq_affinity[cpu] &= ~irq_mask;
+		}
+	}
+	/* this is magic, we now have the correct affinity maps, so
+	 * enable the interrupt.  This will send an enable CPI to
+	 * those cpu's who need to enable it in their local masks,
+	 * causing them to correct for the new affinity . If the
+	 * interrupt is currently globally disabled, it will simply be
+	 * disabled again as it comes in (voyager lazy disable).  If
+	 * the affinity map is tightened to disable the interrupt on a
+	 * cpu, it will be pushed off when it comes in */
+	enable_vic_irq(irq);
+}
+
+static void
+ack_vic_irq(unsigned int irq)
+{
+	if (irq & 8) {
+		outb(0x62,0x20);	/* Specific EOI to cascade */
+		outb(0x60|(irq & 7),0xA0);
+	} else {
+		outb(0x60 | (irq & 7),0x20);
+	}
+}
+
+/* enable the CPIs.  In the VIC, the CPIs are delivered by the 8259
+ * but are not vectored by it.  This means that the 8259 mask must be
+ * lowered to receive them */
+static __init void
+vic_enable_cpi(void)
+{
+	__u8 cpu = smp_processor_id();
+	
+	/* just take a copy of the current mask (nop for boot cpu) */
+	vic_irq_mask[cpu] = vic_irq_mask[boot_cpu_id];
+
+	enable_local_vic_irq(VIC_CPI_LEVEL0);
+	enable_local_vic_irq(VIC_CPI_LEVEL1);
+	/* for sys int and cmn int */
+	enable_local_vic_irq(7);
+
+	if(is_cpu_quad()) {
+		outb(QIC_DEFAULT_MASK0, QIC_MASK_REGISTER0);
+		outb(QIC_CPI_ENABLE, QIC_MASK_REGISTER1);
+		VDEBUG(("VOYAGER SMP: QIC ENABLE CPI: CPU%d: MASK 0x%x\n",
+			cpu, QIC_CPI_ENABLE));
+	}
+
+	VDEBUG(("VOYAGER SMP: ENABLE CPI: CPU%d: MASK 0x%x\n",
+		cpu, vic_irq_mask[cpu]));
+}
+
+void
+voyager_smp_dump()
+{
+	int old_cpu = smp_processor_id(), cpu;
+
+	/* dump the interrupt masks of each processor */
+	for_each_online_cpu(cpu) {
+		__u16 imr, isr, irr;
+		unsigned long flags;
+
+		local_irq_save(flags);
+		outb(VIC_CPU_MASQUERADE_ENABLE | cpu, VIC_PROCESSOR_ID);
+		imr = (inb(0xa1) << 8) | inb(0x21);
+		outb(0x0a, 0xa0);
+		irr = inb(0xa0) << 8;
+		outb(0x0a, 0x20);
+		irr |= inb(0x20);
+		outb(0x0b, 0xa0);
+		isr = inb(0xa0) << 8;
+		outb(0x0b, 0x20);
+		isr |= inb(0x20);
+		outb(old_cpu, VIC_PROCESSOR_ID);
+		local_irq_restore(flags);
+		printk("\tCPU%d: mask=0x%x, IMR=0x%x, IRR=0x%x, ISR=0x%x\n",
+		       cpu, vic_irq_mask[cpu], imr, irr, isr);
+#if 0
+		/* These lines are put in to try to unstick an un ack'd irq */
+		if(isr != 0) {
+			int irq;
+			for(irq=0; irq<16; irq++) {
+				if(isr & (1<<irq)) {
+					printk("\tCPU%d: ack irq %d\n",
+					       cpu, irq);
+					local_irq_save(flags);
+					outb(VIC_CPU_MASQUERADE_ENABLE | cpu,
+					     VIC_PROCESSOR_ID);
+					ack_vic_irq(irq);
+					outb(old_cpu, VIC_PROCESSOR_ID);
+					local_irq_restore(flags);
+				}
+			}
+		}
+#endif
+	}
+}
+
+void
+smp_voyager_power_off(void *dummy)
+{
+	if(smp_processor_id() == boot_cpu_id) 
+		voyager_power_off();
+	else
+		smp_stop_cpu_function(NULL);
+}
+
+void __init
+smp_prepare_cpus(unsigned int max_cpus)
+{
+	/* FIXME: ignore max_cpus for now */
+	smp_boot_cpus();
+}
+
+void __devinit smp_prepare_boot_cpu(void)
+{
+	cpu_set(smp_processor_id(), cpu_online_map);
+	cpu_set(smp_processor_id(), cpu_callout_map);
+}
+
+int __devinit
+__cpu_up(unsigned int cpu)
+{
+	/* This only works at boot for x86.  See "rewrite" above. */
+	if (cpu_isset(cpu, smp_commenced_mask))
+		return -ENOSYS;
+
+	/* In case one didn't come up */
+	if (!cpu_isset(cpu, cpu_callin_map))
+		return -EIO;
+	/* Unleash the CPU! */
+	cpu_set(cpu, smp_commenced_mask);
+	while (!cpu_isset(cpu, cpu_online_map))
+		mb();
+	return 0;
+}
+
+void __init 
+smp_cpus_done(unsigned int max_cpus)
+{
+	zap_low_mappings();
+}
diff --git a/arch/i386/mach-voyager/voyager_thread.c b/arch/i386/mach-voyager/voyager_thread.c
new file mode 100644
index 000000000000..9980eef31fda
--- /dev/null
+++ b/arch/i386/mach-voyager/voyager_thread.c
@@ -0,0 +1,167 @@
+/* -*- mode: c; c-basic-offset: 8 -*- */
+
+/* Copyright (C) 2001
+ *
+ * Author: J.E.J.Bottomley@HansenPartnership.com
+ *
+ * linux/arch/i386/kernel/voyager_thread.c
+ *
+ * This module provides the machine status monitor thread for the
+ * voyager architecture.  This allows us to monitor the machine
+ * environment (temp, voltage, fan function) and the front panel and
+ * internal UPS.  If a fault is detected, this thread takes corrective
+ * action (usually just informing init)
+ * */
+
+#include <linux/module.h>
+#include <linux/config.h>
+#include <linux/mm.h>
+#include <linux/kernel_stat.h>
+#include <linux/delay.h>
+#include <linux/mc146818rtc.h>
+#include <linux/smp_lock.h>
+#include <linux/init.h>
+#include <linux/bootmem.h>
+#include <linux/kmod.h>
+#include <linux/completion.h>
+#include <linux/sched.h>
+#include <asm/desc.h>
+#include <asm/voyager.h>
+#include <asm/vic.h>
+#include <asm/mtrr.h>
+#include <asm/msr.h>
+
+#include <linux/irq.h>
+
+#define THREAD_NAME "kvoyagerd"
+
+/* external variables */
+int kvoyagerd_running = 0;
+DECLARE_MUTEX_LOCKED(kvoyagerd_sem);
+
+static int thread(void *);
+
+static __u8 set_timeout = 0;
+
+/* Start the machine monitor thread.  Return 1 if OK, 0 if fail */
+static int __init
+voyager_thread_start(void)
+{
+	if(kernel_thread(thread, NULL, CLONE_KERNEL) < 0) {
+		/* This is serious, but not fatal */
+		printk(KERN_ERR "Voyager: Failed to create system monitor thread!!!\n");
+		return 1;
+	}
+	return 0;
+}
+
+static int
+execute(const char *string)
+{
+	int ret;
+
+	char *envp[] = {
+		"HOME=/",
+		"TERM=linux",
+		"PATH=/sbin:/usr/sbin:/bin:/usr/bin",
+		NULL,
+	};
+	char *argv[] = {
+		"/bin/bash",
+		"-c",
+		(char *)string,
+		NULL,
+	};
+
+	if ((ret = call_usermodehelper(argv[0], argv, envp, 1)) != 0) {
+		printk(KERN_ERR "Voyager failed to run \"%s\": %i\n",
+		       string, ret);
+	}
+	return ret;
+}
+
+static void
+check_from_kernel(void)
+{
+	if(voyager_status.switch_off) {
+		
+		/* FIXME: This should be configureable via proc */
+		execute("umask 600; echo 0 > /etc/initrunlvl; kill -HUP 1");
+	} else if(voyager_status.power_fail) {
+		VDEBUG(("Voyager daemon detected AC power failure\n"));
+		
+		/* FIXME: This should be configureable via proc */
+		execute("umask 600; echo F > /etc/powerstatus; kill -PWR 1");
+		set_timeout = 1;
+	}
+}
+
+static void
+check_continuing_condition(void)
+{
+	if(voyager_status.power_fail) {
+		__u8 data;
+		voyager_cat_psi(VOYAGER_PSI_SUBREAD, 
+				VOYAGER_PSI_AC_FAIL_REG, &data);
+		if((data & 0x1f) == 0) {
+			/* all power restored */
+			printk(KERN_NOTICE "VOYAGER AC power restored, cancelling shutdown\n");
+			/* FIXME: should be user configureable */
+			execute("umask 600; echo O > /etc/powerstatus; kill -PWR 1");
+			set_timeout = 0;
+		}
+	}
+}
+
+static void
+wakeup(unsigned long unused)
+{
+	up(&kvoyagerd_sem);
+}
+
+static int
+thread(void *unused)
+{
+	struct timer_list wakeup_timer;
+
+	kvoyagerd_running = 1;
+
+	reparent_to_init();
+	daemonize(THREAD_NAME);
+
+	set_timeout = 0;
+
+	init_timer(&wakeup_timer);
+
+	sigfillset(&current->blocked);
+	current->signal->tty = NULL;
+
+	printk(KERN_NOTICE "Voyager starting monitor thread\n");
+
+	for(;;) {
+		down_interruptible(&kvoyagerd_sem);
+		VDEBUG(("Voyager Daemon awoken\n"));
+		if(voyager_status.request_from_kernel == 0) {
+			/* probably awoken from timeout */
+			check_continuing_condition();
+		} else {
+			check_from_kernel();
+			voyager_status.request_from_kernel = 0;
+		}
+		if(set_timeout) {
+			del_timer(&wakeup_timer);
+			wakeup_timer.expires = HZ + jiffies;
+			wakeup_timer.function = wakeup;
+			add_timer(&wakeup_timer);
+		}
+	}
+}
+
+static void __exit
+voyager_thread_stop(void)
+{
+	/* FIXME: do nothing at the moment */
+}
+
+module_init(voyager_thread_start);
+//module_exit(voyager_thread_stop);