More AP / SP bits for the 34K, the Malta bits and things. Still wants

a little polishing.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

6 files changed, 1890 insertions, 17 deletions

diff --git a/arch/mips/kernel/Makefile b/arch/mips/kernel/Makefile
index 0867417032f2..0f527063a8a8 100644
--- a/arch/mips/kernel/Makefile
+++ b/arch/mips/kernel/Makefile
@@ -34,12 +34,16 @@ obj-$(CONFIG_CPU_R6000)		+= r6000_fpu.o r4k_switch.o
 
 obj-$(CONFIG_SMP)		+= smp.o
 
+obj-$(CONFIG_MIPS_VPE_LOADER)	+= vpe.o
+obj-$(CONFIG_MIPS_VPE_APSP_API)	+= rtlx.o
+
 obj-$(CONFIG_NO_ISA)		+= dma-no-isa.o
 obj-$(CONFIG_I8259)		+= i8259.o
 obj-$(CONFIG_IRQ_CPU)		+= irq_cpu.o
 obj-$(CONFIG_IRQ_CPU_RM7K)	+= irq-rm7000.o
 obj-$(CONFIG_IRQ_CPU_RM9K)	+= irq-rm9000.o
 obj-$(CONFIG_IRQ_MV64340)	+= irq-mv6434x.o
+obj-$(CONFIG_MIPS_BOARDS_GEN)	+= irq-msc01.o
 
 obj-$(CONFIG_32BIT)		+= scall32-o32.o
 obj-$(CONFIG_64BIT)		+= scall64-64.o
diff --git a/arch/mips/kernel/genex.S b/arch/mips/kernel/genex.S
index 349ec301168f..fd904d1e4190 100644
--- a/arch/mips/kernel/genex.S
+++ b/arch/mips/kernel/genex.S
@@ -148,6 +148,38 @@ NESTED(except_vec_ejtag_debug, 0, sp)
 	__FINIT
 
 /*
+ * Vectored interrupt handler.
+ * This prototype is copied to ebase + n*IntCtl.VS and patched
+ * to invoke the handler
+ */
+NESTED(except_vec_vi, 0, sp)
+	SAVE_SOME
+	SAVE_AT
+	.set	push
+	.set	noreorder
+EXPORT(except_vec_vi_lui)
+	lui	v0, 0		/* Patched */
+	j	except_vec_vi_handler
+EXPORT(except_vec_vi_ori)
+	 ori	v0, 0		/* Patched */
+	.set	pop
+	END(except_vec_vi)
+EXPORT(except_vec_vi_end)
+
+/*
+ * Common Vectored Interrupt code
+ * Complete the register saves and invoke the handler which is passed in $v0
+ */
+NESTED(except_vec_vi_handler, 0, sp)
+	SAVE_TEMP
+	SAVE_STATIC
+	CLI
+	move	a0, sp
+	jalr	v0
+	j	ret_from_irq
+	END(except_vec_vi_handler)
+
+/*
  * EJTAG debug exception handler.
  */
 NESTED(ejtag_debug_handler, PT_SIZE, sp)
diff --git a/arch/mips/kernel/irq-msc01.c b/arch/mips/kernel/irq-msc01.c
index bf759e33c5ef..3f653c7cfbf3 100644
--- a/arch/mips/kernel/irq-msc01.c
+++ b/arch/mips/kernel/irq-msc01.c
@@ -74,7 +74,7 @@ static void disable_msc_irq(unsigned int irq)
 static void level_mask_and_ack_msc_irq(unsigned int irq)
 {
 	mask_msc_irq(irq);
-	if (!cpu_has_ei)
+	if (!cpu_has_veic)
 		MSCIC_WRITE(MSC01_IC_EOI, 0);
 }
 
@@ -84,7 +84,7 @@ static void level_mask_and_ack_msc_irq(unsigned int irq)
 static void edge_mask_and_ack_msc_irq(unsigned int irq)
 {
 	mask_msc_irq(irq);
-	if (!cpu_has_ei)
+	if (!cpu_has_veic)
 		MSCIC_WRITE(MSC01_IC_EOI, 0);
 	else {
 		u32 r;
@@ -166,14 +166,14 @@ void __init init_msc_irqs(unsigned int base, msc_irqmap_t *imp, int nirq)
 		switch (imp->im_type) {
 		case MSC01_IRQ_EDGE:
 			irq_desc[base+n].handler = &msc_edgeirq_type;
-			if (cpu_has_ei)
+			if (cpu_has_veic)
 				MSCIC_WRITE(MSC01_IC_SUP+n*8, MSC01_IC_SUP_EDGE_BIT);
 			else
 				MSCIC_WRITE(MSC01_IC_SUP+n*8, MSC01_IC_SUP_EDGE_BIT | imp->im_lvl);
 			break;
 		case MSC01_IRQ_LEVEL:
 			irq_desc[base+n].handler = &msc_levelirq_type;
-			if (cpu_has_ei)
+			if (cpu_has_veic)
 				MSCIC_WRITE(MSC01_IC_SUP+n*8, 0);
 			else
 				MSCIC_WRITE(MSC01_IC_SUP+n*8, imp->im_lvl);
diff --git a/arch/mips/kernel/rtlx.c b/arch/mips/kernel/rtlx.c
new file mode 100644
index 000000000000..8c81f3cb4e2d
--- /dev/null
+++ b/arch/mips/kernel/rtlx.c
@@ -0,0 +1,341 @@
+/*
+ * Copyright (C) 2005 MIPS Technologies, Inc.  All rights reserved.
+ *
+ *  This program is free software; you can distribute it and/or modify it
+ *  under the terms of the GNU General Public License (Version 2) as
+ *  published by the Free Software Foundation.
+ *
+ *  This program is distributed in the hope it will be useful, but WITHOUT
+ *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ *  for more details.
+ *
+ *  You should have received a copy of the GNU General Public License along
+ *  with this program; if not, write to the Free Software Foundation, Inc.,
+ *  59 Temple Place - Suite 330, Boston MA 02111-1307, USA.
+ *
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/init.h>
+#include <asm/uaccess.h>
+#include <linux/slab.h>
+#include <linux/list.h>
+#include <linux/vmalloc.h>
+#include <linux/elf.h>
+#include <linux/seq_file.h>
+#include <linux/syscalls.h>
+#include <linux/moduleloader.h>
+#include <linux/interrupt.h>
+#include <linux/poll.h>
+#include <linux/sched.h>
+#include <linux/wait.h>
+#include <asm/mipsmtregs.h>
+#include <asm/cacheflush.h>
+#include <asm/atomic.h>
+#include <asm/cpu.h>
+#include <asm/processor.h>
+#include <asm/system.h>
+#include <asm/rtlx.h>
+
+#define RTLX_MAJOR 64
+#define RTLX_TARG_VPE 1
+
+struct rtlx_info *rtlx;
+static int major;
+static char module_name[] = "rtlx";
+static inline int spacefree(int read, int write, int size);
+
+static struct chan_waitqueues {
+	wait_queue_head_t rt_queue;
+	wait_queue_head_t lx_queue;
+} channel_wqs[RTLX_CHANNELS];
+
+static struct irqaction irq;
+static int irq_num;
+
+extern void *vpe_get_shared(int index);
+
+static void rtlx_dispatch(struct pt_regs *regs)
+{
+	do_IRQ(MIPSCPU_INT_BASE + MIPS_CPU_RTLX_IRQ, regs);
+}
+
+irqreturn_t rtlx_interrupt(int irq, void *dev_id, struct pt_regs *regs)
+{
+	irqreturn_t r = IRQ_HANDLED;
+	int i;
+
+	for (i = 0; i < RTLX_CHANNELS; i++) {
+		struct rtlx_channel *chan = &rtlx->channel[i];
+
+		if (chan->lx_read != chan->lx_write)
+			wake_up_interruptible(&channel_wqs[i].lx_queue);
+	}
+
+	return r;
+}
+
+void dump_rtlx(void)
+{
+	int i;
+
+	printk("id 0x%lx state %d\n", rtlx->id, rtlx->state);
+
+	for (i = 0; i < RTLX_CHANNELS; i++) {
+		struct rtlx_channel *chan = &rtlx->channel[i];
+
+		printk(" rt_state %d lx_state %d buffer_size %d\n",
+		       chan->rt_state, chan->lx_state, chan->buffer_size);
+
+		printk(" rt_read %d rt_write %d\n",
+		       chan->rt_read, chan->rt_write);
+
+		printk(" lx_read %d lx_write %d\n",
+		       chan->lx_read, chan->lx_write);
+
+		printk(" rt_buffer <%s>\n", chan->rt_buffer);
+		printk(" lx_buffer <%s>\n", chan->lx_buffer);
+	}
+}
+
+/* call when we have the address of the shared structure from the SP side. */
+static int rtlx_init(struct rtlx_info *rtlxi)
+{
+	int i;
+
+	if (rtlxi->id != RTLX_ID) {
+		printk(KERN_WARNING "no valid RTLX id at 0x%p\n", rtlxi);
+		return (-ENOEXEC);
+	}
+
+	/* initialise the wait queues */
+	for (i = 0; i < RTLX_CHANNELS; i++) {
+		init_waitqueue_head(&channel_wqs[i].rt_queue);
+		init_waitqueue_head(&channel_wqs[i].lx_queue);
+	}
+
+	/* set up for interrupt handling */
+	memset(&irq, 0, sizeof(struct irqaction));
+
+	if (cpu_has_vint) {
+		set_vi_handler(MIPS_CPU_RTLX_IRQ, rtlx_dispatch);
+	}
+
+	irq_num = MIPSCPU_INT_BASE + MIPS_CPU_RTLX_IRQ;
+	irq.handler = rtlx_interrupt;
+	irq.flags = SA_INTERRUPT;
+	irq.name = "RTLX";
+	irq.dev_id = rtlx;
+	setup_irq(irq_num, &irq);
+
+	rtlx = rtlxi;
+	return (0);
+}
+
+/* only allow one open process at a time to open each channel */
+static int rtlx_open(struct inode *inode, struct file *filp)
+{
+	int minor, ret;
+	struct rtlx_channel *chan;
+
+	/* assume only 1 device at the mo. */
+	minor = MINOR(inode->i_rdev);
+
+	if (rtlx == NULL) {
+		struct rtlx_info **p;
+		if( (p = vpe_get_shared(RTLX_TARG_VPE)) == NULL) {
+			printk(" vpe_get_shared is NULL. Has an SP program been loaded?\n");
+			return (-EFAULT);
+		}
+
+		if (*p == NULL) {
+			printk(" vpe_shared %p %p\n", p, *p);
+			return (-EFAULT);
+		}
+
+		if ((ret = rtlx_init(*p)) < 0)
+			return (ret);
+	}
+
+	chan = &rtlx->channel[minor];
+
+	/* already open? */
+	if (chan->lx_state == RTLX_STATE_OPENED)
+		return (-EBUSY);
+
+	chan->lx_state = RTLX_STATE_OPENED;
+	return (0);
+}
+
+static int rtlx_release(struct inode *inode, struct file *filp)
+{
+	int minor;
+
+	minor = MINOR(inode->i_rdev);
+	rtlx->channel[minor].lx_state = RTLX_STATE_UNUSED;
+	return (0);
+}
+
+static unsigned int rtlx_poll(struct file *file, poll_table * wait)
+{
+	int minor;
+	unsigned int mask = 0;
+	struct rtlx_channel *chan;
+
+	minor = MINOR(file->f_dentry->d_inode->i_rdev);
+	chan = &rtlx->channel[minor];
+
+	poll_wait(file, &channel_wqs[minor].rt_queue, wait);
+	poll_wait(file, &channel_wqs[minor].lx_queue, wait);
+
+	/* data available to read? */
+	if (chan->lx_read != chan->lx_write)
+		mask |= POLLIN | POLLRDNORM;
+
+	/* space to write */
+	if (spacefree(chan->rt_read, chan->rt_write, chan->buffer_size))
+		mask |= POLLOUT | POLLWRNORM;
+
+	return (mask);
+}
+
+static ssize_t rtlx_read(struct file *file, char __user * buffer, size_t count,
+			 loff_t * ppos)
+{
+	size_t fl = 0L;
+	int minor;
+	struct rtlx_channel *lx;
+	DECLARE_WAITQUEUE(wait, current);
+
+	minor = MINOR(file->f_dentry->d_inode->i_rdev);
+	lx = &rtlx->channel[minor];
+
+	/* data available? */
+	if (lx->lx_write == lx->lx_read) {
+		if (file->f_flags & O_NONBLOCK)
+			return (0);	// -EAGAIN makes cat whinge
+
+		/* go to sleep */
+		add_wait_queue(&channel_wqs[minor].lx_queue, &wait);
+		set_current_state(TASK_INTERRUPTIBLE);
+
+		while (lx->lx_write == lx->lx_read)
+			schedule();
+
+		set_current_state(TASK_RUNNING);
+		remove_wait_queue(&channel_wqs[minor].lx_queue, &wait);
+
+		/* back running */
+	}
+
+	/* find out how much in total */
+	count = min( count,
+		     (size_t)(lx->lx_write + lx->buffer_size - lx->lx_read) % lx->buffer_size);
+
+	/* then how much from the read pointer onwards */
+	fl = min( count, (size_t)lx->buffer_size - lx->lx_read);
+
+	copy_to_user (buffer, &lx->lx_buffer[lx->lx_read], fl);
+
+	/* and if there is anything left at the beginning of the buffer */
+	if ( count - fl )
+		copy_to_user (buffer + fl, lx->lx_buffer, count - fl);
+
+	/* update the index */
+	lx->lx_read += count;
+	lx->lx_read %= lx->buffer_size;
+
+	return (count);
+}
+
+static inline int spacefree(int read, int write, int size)
+{
+	if (read == write) {
+		/* never fill the buffer completely, so indexes are always equal if empty
+		   and only empty, or !equal if data available */
+		return (size - 1);
+	}
+
+	return ((read + size - write) % size) - 1;
+}
+
+static ssize_t rtlx_write(struct file *file, const char __user * buffer,
+			  size_t count, loff_t * ppos)
+{
+	int minor;
+	struct rtlx_channel *rt;
+	size_t fl;
+	DECLARE_WAITQUEUE(wait, current);
+
+	minor = MINOR(file->f_dentry->d_inode->i_rdev);
+	rt = &rtlx->channel[minor];
+
+	/* any space left... */
+	if (!spacefree(rt->rt_read, rt->rt_write, rt->buffer_size)) {
+
+		if (file->f_flags & O_NONBLOCK)
+			return (-EAGAIN);
+
+		add_wait_queue(&channel_wqs[minor].rt_queue, &wait);
+		set_current_state(TASK_INTERRUPTIBLE);
+
+		while (!spacefree(rt->rt_read, rt->rt_write, rt->buffer_size))
+			schedule();
+
+		set_current_state(TASK_RUNNING);
+		remove_wait_queue(&channel_wqs[minor].rt_queue, &wait);
+	}
+
+	/* total number of bytes to copy */
+	count = min( count, (size_t)spacefree(rt->rt_read, rt->rt_write, rt->buffer_size) );
+
+	/* first bit from write pointer to the end of the buffer, or count */
+	fl = min(count, (size_t) rt->buffer_size - rt->rt_write);
+
+	copy_from_user(&rt->rt_buffer[rt->rt_write], buffer, fl);
+
+	/* if there's any left copy to the beginning of the buffer */
+	if( count - fl )
+		copy_from_user(rt->rt_buffer, buffer + fl, count - fl);
+
+	rt->rt_write += count;
+	rt->rt_write %= rt->buffer_size;
+
+	return(count);
+}
+
+static struct file_operations rtlx_fops = {
+	.owner = THIS_MODULE,
+	.open = rtlx_open,
+	.release = rtlx_release,
+	.write = rtlx_write,
+	.read = rtlx_read,
+	.poll = rtlx_poll
+};
+
+static int rtlx_module_init(void)
+{
+	if ((major = register_chrdev(RTLX_MAJOR, module_name, &rtlx_fops)) < 0) {
+		printk("rtlx_module_init: unable to register device\n");
+		return (-EBUSY);
+	}
+
+	if (major == 0)
+		major = RTLX_MAJOR;
+
+	return (0);
+}
+
+static void rtlx_module_exit(void)
+{
+	unregister_chrdev(major, module_name);
+}
+
+module_init(rtlx_module_init);
+module_exit(rtlx_module_exit);
+MODULE_DESCRIPTION("MIPS RTLX");
+MODULE_AUTHOR("Elizabeth Clarke, MIPS Technologies, Inc");
+MODULE_LICENSE("GPL");
diff --git a/arch/mips/kernel/traps.c b/arch/mips/kernel/traps.c
index b2fa607eeeb8..0a3969aa8dc6 100644
--- a/arch/mips/kernel/traps.c
+++ b/arch/mips/kernel/traps.c
@@ -20,6 +20,7 @@
 #include <linux/smp_lock.h>
 #include <linux/spinlock.h>
 #include <linux/kallsyms.h>
+#include <linux/bootmem.h>
 
 #include <asm/bootinfo.h>
 #include <asm/branch.h>
@@ -64,6 +65,9 @@ extern int fpu_emulator_cop1Handler(int xcptno, struct pt_regs *xcp,
 
 void (*board_be_init)(void);
 int (*board_be_handler)(struct pt_regs *regs, int is_fixup);
+void (*board_nmi_handler_setup)(void);
+void (*board_ejtag_handler_setup)(void);
+void (*board_bind_eic_interrupt)(int irq, int regset);
 
 /*
  * These constant is for searching for possible module text segments.
@@ -813,6 +817,12 @@ asmlinkage void do_reserved(struct pt_regs *regs)
 	      (regs->cp0_cause & 0x7f) >> 2);
 }
 
+asmlinkage void do_default_vi(struct pt_regs *regs)
+{
+	show_regs(regs);
+	panic("Caught unexpected vectored interrupt.");
+}
+
 /*
  * Some MIPS CPUs can enable/disable for cache parity detection, but do
  * it different ways.
@@ -921,7 +931,11 @@ void nmi_exception_handler(struct pt_regs *regs)
 	while(1) ;
 }
 
+#define VECTORSPACING 0x100	/* for EI/VI mode */
+
+unsigned long ebase;
 unsigned long exception_handlers[32];
+unsigned long vi_handlers[64];
 
 /*
  * As a side effect of the way this is implemented we're limited
@@ -935,13 +949,156 @@ void *set_except_vector(int n, void *addr)
 
 	exception_handlers[n] = handler;
 	if (n == 0 && cpu_has_divec) {
-		*(volatile u32 *)(CAC_BASE + 0x200) = 0x08000000 |
+		*(volatile u32 *)(ebase + 0x200) = 0x08000000 |
 		                                 (0x03ffffff & (handler >> 2));
-		flush_icache_range(CAC_BASE + 0x200, CAC_BASE + 0x204);
+		flush_icache_range(ebase + 0x200, ebase + 0x204);
+	}
+	return (void *)old_handler;
+}
+
+#ifdef CONFIG_CPU_MIPSR2
+/*
+ * Shadow register allocation
+ * FIXME: SMP...
+ */
+
+/* MIPSR2 shadow register sets */
+struct shadow_registers {
+	spinlock_t sr_lock;	/*  */
+	int sr_supported;	/* Number of shadow register sets supported */
+	int sr_allocated;	/* Bitmap of allocated shadow registers */
+} shadow_registers;
+
+void mips_srs_init(void)
+{
+#ifdef CONFIG_CPU_MIPSR2_SRS
+	shadow_registers.sr_supported = ((read_c0_srsctl() >> 26) & 0x0f) + 1;
+	printk ("%d MIPSR2 register sets available\n", shadow_registers.sr_supported);
+#else
+	shadow_registers.sr_supported = 1;
+#endif
+	shadow_registers.sr_allocated = 1;	/* Set 0 used by kernel */
+	spin_lock_init(&shadow_registers.sr_lock);
+}
+
+int mips_srs_max(void)
+{
+	return shadow_registers.sr_supported;
+}
+
+int mips_srs_alloc (void)
+{
+	struct shadow_registers *sr = &shadow_registers;
+	unsigned long flags;
+	int set;
+
+	spin_lock_irqsave(&sr->sr_lock, flags);
+
+	for (set = 0; set < sr->sr_supported; set++) {
+		if ((sr->sr_allocated & (1 << set)) == 0) {
+			sr->sr_allocated |= 1 << set;
+			spin_unlock_irqrestore(&sr->sr_lock, flags);
+			return set;
+		}
+	}
+
+	/* None available */
+	spin_unlock_irqrestore(&sr->sr_lock, flags);
+	return -1;
+}
+
+void mips_srs_free (int set)
+{
+	struct shadow_registers *sr = &shadow_registers;
+	unsigned long flags;
+
+	spin_lock_irqsave(&sr->sr_lock, flags);
+	sr->sr_allocated &= ~(1 << set);
+	spin_unlock_irqrestore(&sr->sr_lock, flags);
+}
+
+void *set_vi_srs_handler (int n, void *addr, int srs)
+{
+	unsigned long handler;
+	unsigned long old_handler = vi_handlers[n];
+	u32 *w;
+	unsigned char *b;
+
+	if (!cpu_has_veic && !cpu_has_vint)
+		BUG();
+
+	if (addr == NULL) {
+		handler = (unsigned long) do_default_vi;
+		srs = 0;
+	}
+	else
+		handler = (unsigned long) addr;
+	vi_handlers[n] = (unsigned long) addr;
+
+	b = (unsigned char *)(ebase + 0x200 + n*VECTORSPACING);
+
+	if (srs >= mips_srs_max())
+		panic("Shadow register set %d not supported", srs);
+
+	if (cpu_has_veic) {
+		if (board_bind_eic_interrupt)
+			board_bind_eic_interrupt (n, srs);
+	}
+	else if (cpu_has_vint) {
+		/* SRSMap is only defined if shadow sets are implemented */
+		if (mips_srs_max() > 1)
+			change_c0_srsmap (0xf << n*4, srs << n*4);
+	}
+
+	if (srs == 0) {
+		/*
+		 * If no shadow set is selected then use the default handler
+		 * that does normal register saving and a standard interrupt exit
+		 */
+
+		extern char except_vec_vi, except_vec_vi_lui;
+		extern char except_vec_vi_ori, except_vec_vi_end;
+		const int handler_len = &except_vec_vi_end - &except_vec_vi;
+		const int lui_offset = &except_vec_vi_lui - &except_vec_vi;
+		const int ori_offset = &except_vec_vi_ori - &except_vec_vi;
+
+		if (handler_len > VECTORSPACING) {
+			/*
+			 * Sigh... panicing won't help as the console
+			 * is probably not configured :(
+			 */
+			panic ("VECTORSPACING too small");
+		}
+
+		memcpy (b, &except_vec_vi, handler_len);
+		w = (u32 *)(b + lui_offset);
+		*w = (*w & 0xffff0000) | (((u32)handler >> 16) & 0xffff);
+		w = (u32 *)(b + ori_offset);
+		*w = (*w & 0xffff0000) | ((u32)handler & 0xffff);
+		flush_icache_range((unsigned long)b, (unsigned long)(b+handler_len));
+	}
+	else {
+		/*
+		 * In other cases jump directly to the interrupt handler
+		 *
+		 * It is the handlers responsibility to save registers if required
+		 * (eg hi/lo) and return from the exception using "eret"
+		 */
+		w = (u32 *)b;
+		*w++ = 0x08000000 | (((u32)handler >> 2) & 0x03fffff); /* j handler */
+		*w = 0;
+		flush_icache_range((unsigned long)b, (unsigned long)(b+8));
 	}
+
 	return (void *)old_handler;
 }
 
+void *set_vi_handler (int n, void *addr)
+{
+	return set_vi_srs_handler (n, addr, 0);
+}
+#endif
+
 /*
  * This is used by native signal handling
  */
@@ -1016,10 +1173,18 @@ void __init per_cpu_trap_init(void)
 	if (cpu_has_dsp)
 		set_c0_status(ST0_MX);
 
+#ifdef CONFIG_CPU_MIPSR2
+	write_c0_hwrena (0x0000000f); /* Allow rdhwr to all registers */
+#endif
+
 	/*
-	 * Some MIPS CPUs have a dedicated interrupt vector which reduces the
-	 * interrupt processing overhead.  Use it where available.
+	 * Interrupt handling.
 	 */
+	if (cpu_has_veic || cpu_has_vint) {
+		write_c0_ebase (ebase);
+		/* Setting vector spacing enables EI/VI mode  */
+		change_c0_intctl (0x3e0, VECTORSPACING);
+	}
 	if (cpu_has_divec)
 		set_c0_cause(CAUSEF_IV);
 
@@ -1035,13 +1200,41 @@ void __init per_cpu_trap_init(void)
 	tlb_init();
 }
 
+/* Install CPU exception handler */
+void __init set_handler (unsigned long offset, void *addr, unsigned long size)
+{
+	memcpy((void *)(ebase + offset), addr, size);
+	flush_icache_range(ebase + offset, ebase + offset + size);
+}
+
+/* Install uncached CPU exception handler */
+void __init set_uncached_handler (unsigned long offset, void *addr, unsigned long size)
+{
+#ifdef CONFIG_32BIT
+	unsigned long uncached_ebase = KSEG1ADDR(ebase);
+#endif
+#ifdef CONFIG_64BIT
+	unsigned long uncached_ebase = TO_UNCAC(ebase);
+#endif
+
+	memcpy((void *)(uncached_ebase + offset), addr, size);
+}
+
 void __init trap_init(void)
 {
 	extern char except_vec3_generic, except_vec3_r4000;
-	extern char except_vec_ejtag_debug;
 	extern char except_vec4;
 	unsigned long i;
 
+	if (cpu_has_veic || cpu_has_vint)
+		ebase = (unsigned long) alloc_bootmem_low_pages (0x200 + VECTORSPACING*64);
+	else
+		ebase = CAC_BASE;
+
+#ifdef CONFIG_CPU_MIPSR2
+	mips_srs_init();
+#endif
+
 	per_cpu_trap_init();
 
 	/*
@@ -1049,7 +1242,7 @@ void __init trap_init(void)
 	 * This will be overriden later as suitable for a particular
 	 * configuration.
 	 */
-	memcpy((void *)(CAC_BASE + 0x180), &except_vec3_generic, 0x80);
+	set_handler(0x180, &except_vec3_generic, 0x80);
 
 	/*
 	 * Setup default vectors
@@ -1061,8 +1254,8 @@ void __init trap_init(void)
 	 * Copy the EJTAG debug exception vector handler code to it's final
 	 * destination.
 	 */
-	if (cpu_has_ejtag)
-		memcpy((void *)(CAC_BASE + 0x300), &except_vec_ejtag_debug, 0x80);
+	if (cpu_has_ejtag && board_ejtag_handler_setup)
+		board_ejtag_handler_setup ();
 
 	/*
 	 * Only some CPUs have the watch exceptions.
@@ -1071,11 +1264,15 @@ void __init trap_init(void)
 		set_except_vector(23, handle_watch);
 
 	/*
-	 * Some MIPS CPUs have a dedicated interrupt vector which reduces the
-	 * interrupt processing overhead.  Use it where available.
+	 * Initialise interrupt handlers
 	 */
-	if (cpu_has_divec)
-		memcpy((void *)(CAC_BASE + 0x200), &except_vec4, 0x8);
+	if (cpu_has_veic || cpu_has_vint) {
+		int nvec = cpu_has_veic ? 64 : 8;
+		for (i = 0; i < nvec; i++)
+			set_vi_handler (i, NULL);
+	}
+	else if (cpu_has_divec)
+		set_handler(0x200, &except_vec4, 0x8);
 
 	/*
 	 * Some CPUs can enable/disable for cache parity detection, but does
@@ -1122,6 +1319,10 @@ void __init trap_init(void)
 		//set_except_vector(15, handle_ndc);
 	}
 
+
+	if (board_nmi_handler_setup)
+		board_nmi_handler_setup();
+
 	if (cpu_has_fpu && !cpu_has_nofpuex)
 		set_except_vector(15, handle_fpe);
 
@@ -1146,5 +1347,5 @@ void __init trap_init(void)
 	signal32_init();
 #endif
 
-	flush_icache_range(CAC_BASE, CAC_BASE + 0x400);
+	flush_icache_range(ebase, ebase + 0x400);
 }
diff --git a/arch/mips/kernel/vpe.c b/arch/mips/kernel/vpe.c
new file mode 100644
index 000000000000..6bf42ba08f09
--- /dev/null
+++ b/arch/mips/kernel/vpe.c
@@ -0,0 +1,1295 @@
+/*
+ * Copyright (C) 2004, 2005 MIPS Technologies, Inc.  All rights reserved.
+ *
+ *  This program is free software; you can distribute it and/or modify it
+ *  under the terms of the GNU General Public License (Version 2) as
+ *  published by the Free Software Foundation.
+ *
+ *  This program is distributed in the hope it will be useful, but WITHOUT
+ *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ *  for more details.
+ *
+ *  You should have received a copy of the GNU General Public License along
+ *  with this program; if not, write to the Free Software Foundation, Inc.,
+ *  59 Temple Place - Suite 330, Boston MA 02111-1307, USA.
+ *
+ */
+
+/*
+ * VPE support module
+ *
+ * Provides support for loading a MIPS SP program on VPE1.
+ * The SP enviroment is rather simple, no tlb's.  It needs to be relocatable
+ * (or partially linked). You should initialise your stack in the startup
+ * code. This loader looks for the symbol __start and sets up
+ * execution to resume from there. The MIPS SDE kit contains suitable examples.
+ *
+ * To load and run, simply cat a SP 'program file' to /dev/vpe1.
+ * i.e cat spapp >/dev/vpe1.
+ *
+ * You'll need to have the following device files.
+ * mknod /dev/vpe0 c 63 0
+ * mknod /dev/vpe1 c 63 1
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/init.h>
+#include <asm/uaccess.h>
+#include <linux/slab.h>
+#include <linux/list.h>
+#include <linux/vmalloc.h>
+#include <linux/elf.h>
+#include <linux/seq_file.h>
+#include <linux/syscalls.h>
+#include <linux/moduleloader.h>
+#include <linux/interrupt.h>
+#include <linux/poll.h>
+#include <linux/bootmem.h>
+#include <asm/mipsregs.h>
+#include <asm/cacheflush.h>
+#include <asm/atomic.h>
+#include <asm/cpu.h>
+#include <asm/processor.h>
+#include <asm/system.h>
+
+typedef void *vpe_handle;
+
+// defined here because the kernel module loader doesn't have
+// anything to do with it.
+#define SHN_MIPS_SCOMMON 0xff03
+
+#ifndef ARCH_SHF_SMALL
+#define ARCH_SHF_SMALL 0
+#endif
+
+/* If this is set, the section belongs in the init part of the module */
+#define INIT_OFFSET_MASK (1UL << (BITS_PER_LONG-1))
+
+// temp number,
+#define VPE_MAJOR 63
+
+static char module_name[] = "vpe";
+static int major = 0;
+
+/* grab the likely amount of memory we will need. */
+#ifdef CONFIG_MIPS_VPE_LOADER_TOM
+#define P_SIZE (2 * 1024 * 1024)
+#else
+/* add an overhead to the max kmalloc size for non-striped symbols/etc */
+#define P_SIZE (256 * 1024)
+#endif
+
+#define MAX_VPES 16
+
+enum vpe_state {
+	VPE_STATE_UNUSED = 0,
+	VPE_STATE_INUSE,
+	VPE_STATE_RUNNING
+};
+
+enum tc_state {
+	TC_STATE_UNUSED = 0,
+	TC_STATE_INUSE,
+	TC_STATE_RUNNING,
+	TC_STATE_DYNAMIC
+};
+
+struct vpe;
+typedef struct tc {
+	enum tc_state state;
+	int index;
+
+	/* parent VPE */
+	struct vpe *pvpe;
+
+	/* The list of TC's with this VPE */
+	struct list_head tc;
+
+	/* The global list of tc's */
+	struct list_head list;
+} tc_t;
+
+typedef struct vpe {
+	enum vpe_state state;
+
+	/* (device) minor associated with this vpe */
+	int minor;
+
+	/* elfloader stuff */
+	void *load_addr;
+	u32 len;
+	char *pbuffer;
+	u32 plen;
+
+	unsigned long __start;
+
+	/* tc's associated with this vpe */
+	struct list_head tc;
+
+	/* The list of vpe's */
+	struct list_head list;
+
+	/* shared symbol address */
+	void *shared_ptr;
+} vpe_t;
+
+struct vpecontrol_ {
+	/* Virtual processing elements */
+	struct list_head vpe_list;
+
+	/* Thread contexts */
+	struct list_head tc_list;
+} vpecontrol;
+
+static void release_progmem(void *ptr);
+static void dump_vpe(vpe_t * v);
+extern void save_gp_address(unsigned int secbase, unsigned int rel);
+
+/* get the vpe associated with this minor */
+struct vpe *get_vpe(int minor)
+{
+	struct vpe *v;
+
+	list_for_each_entry(v, &vpecontrol.vpe_list, list) {
+		if (v->minor == minor)
+			return v;
+	}
+
+	printk(KERN_DEBUG "VPE: get_vpe minor %d not found\n", minor);
+	return NULL;
+}
+
+/* get the vpe associated with this minor */
+struct tc *get_tc(int index)
+{
+	struct tc *t;
+
+	list_for_each_entry(t, &vpecontrol.tc_list, list) {
+		if (t->index == index)
+			return t;
+	}
+
+	printk(KERN_DEBUG "VPE: get_tc index %d not found\n", index);
+
+	return NULL;
+}
+
+struct tc *get_tc_unused(void)
+{
+	struct tc *t;
+
+	list_for_each_entry(t, &vpecontrol.tc_list, list) {
+		if (t->state == TC_STATE_UNUSED)
+			return t;
+	}
+
+	printk(KERN_DEBUG "VPE: All TC's are in use\n");
+
+	return NULL;
+}
+
+/* allocate a vpe and associate it with this minor (or index) */
+struct vpe *alloc_vpe(int minor)
+{
+	struct vpe *v;
+
+	if ((v = kmalloc(sizeof(struct vpe), GFP_KERNEL)) == NULL) {
+		printk(KERN_WARNING "VPE: alloc_vpe no mem\n");
+		return NULL;
+	}
+
+	memset(v, 0, sizeof(struct vpe));
+
+	INIT_LIST_HEAD(&v->tc);
+	list_add_tail(&v->list, &vpecontrol.vpe_list);
+
+	v->minor = minor;
+	return v;
+}
+
+/* allocate a tc. At startup only tc0 is running, all other can be halted. */
+struct tc *alloc_tc(int index)
+{
+	struct tc *t;
+
+	if ((t = kmalloc(sizeof(struct tc), GFP_KERNEL)) == NULL) {
+		printk(KERN_WARNING "VPE: alloc_tc no mem\n");
+		return NULL;
+	}
+
+	memset(t, 0, sizeof(struct tc));
+
+	INIT_LIST_HEAD(&t->tc);
+	list_add_tail(&t->list, &vpecontrol.tc_list);
+
+	t->index = index;
+
+	return t;
+}
+
+/* clean up and free everything */
+void release_vpe(struct vpe *v)
+{
+	list_del(&v->list);
+	if (v->load_addr)
+		release_progmem(v);
+	kfree(v);
+}
+
+void dump_mtregs(void)
+{
+	unsigned long val;
+
+	val = read_c0_config3();
+	printk("config3 0x%lx MT %ld\n", val,
+	       (val & CONFIG3_MT) >> CONFIG3_MT_SHIFT);
+
+	val = read_c0_mvpconf0();
+	printk("mvpconf0 0x%lx, PVPE %ld PTC %ld M %ld\n", val,
+	       (val & MVPCONF0_PVPE) >> MVPCONF0_PVPE_SHIFT,
+	       val & MVPCONF0_PTC, (val & MVPCONF0_M) >> MVPCONF0_M_SHIFT);
+
+	val = read_c0_mvpcontrol();
+	printk("MVPControl 0x%lx, STLB %ld VPC %ld EVP %ld\n", val,
+	       (val & MVPCONTROL_STLB) >> MVPCONTROL_STLB_SHIFT,
+	       (val & MVPCONTROL_VPC) >> MVPCONTROL_VPC_SHIFT,
+	       (val & MVPCONTROL_EVP));
+
+	val = read_c0_vpeconf0();
+	printk("VPEConf0 0x%lx MVP %ld\n", val,
+	       (val & VPECONF0_MVP) >> VPECONF0_MVP_SHIFT);
+}
+
+/* Find some VPE program space  */
+static void *alloc_progmem(u32 len)
+{
+#ifdef CONFIG_MIPS_VPE_LOADER_TOM
+	/* this means you must tell linux to use less memory than you physically have */
+	return (void *)((max_pfn * PAGE_SIZE) + KSEG0);
+#else
+	// simple grab some mem for now
+	return kmalloc(len, GFP_KERNEL);
+#endif
+}
+
+static void release_progmem(void *ptr)
+{
+#ifndef CONFIG_MIPS_VPE_LOADER_TOM
+	kfree(ptr);
+#endif
+}
+
+/* Update size with this section: return offset. */
+static long get_offset(unsigned long *size, Elf_Shdr * sechdr)
+{
+	long ret;
+
+	ret = ALIGN(*size, sechdr->sh_addralign ? : 1);
+	*size = ret + sechdr->sh_size;
+	return ret;
+}
+
+/* Lay out the SHF_ALLOC sections in a way not dissimilar to how ld
+   might -- code, read-only data, read-write data, small data.  Tally
+   sizes, and place the offsets into sh_entsize fields: high bit means it
+   belongs in init. */
+static void layout_sections(struct module *mod, const Elf_Ehdr * hdr,
+			    Elf_Shdr * sechdrs, const char *secstrings)
+{
+	static unsigned long const masks[][2] = {
+		/* NOTE: all executable code must be the first section
+		 * in this array; otherwise modify the text_size
+		 * finder in the two loops below */
+		{SHF_EXECINSTR | SHF_ALLOC, ARCH_SHF_SMALL},
+		{SHF_ALLOC, SHF_WRITE | ARCH_SHF_SMALL},
+		{SHF_WRITE | SHF_ALLOC, ARCH_SHF_SMALL},
+		{ARCH_SHF_SMALL | SHF_ALLOC, 0}
+	};
+	unsigned int m, i;
+
+	for (i = 0; i < hdr->e_shnum; i++)
+		sechdrs[i].sh_entsize = ~0UL;
+
+	for (m = 0; m < ARRAY_SIZE(masks); ++m) {
+		for (i = 0; i < hdr->e_shnum; ++i) {
+			Elf_Shdr *s = &sechdrs[i];
+
+			//  || strncmp(secstrings + s->sh_name, ".init", 5) == 0)
+			if ((s->sh_flags & masks[m][0]) != masks[m][0]
+			    || (s->sh_flags & masks[m][1])
+			    || s->sh_entsize != ~0UL)
+				continue;
+			s->sh_entsize = get_offset(&mod->core_size, s);
+		}
+
+		if (m == 0)
+			mod->core_text_size = mod->core_size;
+
+	}
+}
+
+
+/* from module-elf32.c, but subverted a little */
+
+struct mips_hi16 {
+	struct mips_hi16 *next;
+	Elf32_Addr *addr;
+	Elf32_Addr value;
+};
+
+static struct mips_hi16 *mips_hi16_list;
+static unsigned int gp_offs, gp_addr;
+
+static int apply_r_mips_none(struct module *me, uint32_t *location,
+			     Elf32_Addr v)
+{
+	return 0;
+}
+
+static int apply_r_mips_gprel16(struct module *me, uint32_t *location,
+				Elf32_Addr v)
+{
+	int rel;
+
+	if( !(*location & 0xffff) ) {
+		rel = (int)v - gp_addr;
+	}
+	else {
+		/* .sbss + gp(relative) + offset */
+		/* kludge! */
+		rel =  (int)(short)((int)v + gp_offs +
+				    (int)(short)(*location & 0xffff) - gp_addr);
+	}
+
+	if( (rel > 32768) || (rel < -32768) ) {
+		printk(KERN_ERR
+		       "apply_r_mips_gprel16: relative address out of range 0x%x %d\n",
+		       rel, rel);
+		return -ENOEXEC;
+	}
+
+	*location = (*location & 0xffff0000) | (rel & 0xffff);
+
+	return 0;
+}
+
+static int apply_r_mips_pc16(struct module *me, uint32_t *location,
+			     Elf32_Addr v)
+{
+	int rel;
+	rel = (((unsigned int)v - (unsigned int)location));
+	rel >>= 2;		// because the offset is in _instructions_ not bytes.
+	rel -= 1;		// and one instruction less due to the branch delay slot.
+
+	if( (rel > 32768) || (rel < -32768) ) {
+		printk(KERN_ERR
+		       "apply_r_mips_pc16: relative address out of range 0x%x\n", rel);
+		return -ENOEXEC;
+	}
+
+	*location = (*location & 0xffff0000) | (rel & 0xffff);
+
+	return 0;
+}
+
+static int apply_r_mips_32(struct module *me, uint32_t *location,
+			   Elf32_Addr v)
+{
+	*location += v;
+
+	return 0;
+}
+
+static int apply_r_mips_26(struct module *me, uint32_t *location,
+			   Elf32_Addr v)
+{
+	if (v % 4) {
+		printk(KERN_ERR "module %s: dangerous relocation mod4\n", me->name);
+		return -ENOEXEC;
+	}
+
+/* Not desperately convinced this is a good check of an overflow condition
+   anyway. But it gets in the way of handling undefined weak symbols which
+   we want to set to zero.
+   if ((v & 0xf0000000) != (((unsigned long)location + 4) & 0xf0000000)) {
+   printk(KERN_ERR
+   "module %s: relocation overflow\n",
+   me->name);
+   return -ENOEXEC;
+   }
+*/
+
+	*location = (*location & ~0x03ffffff) |
+		((*location + (v >> 2)) & 0x03ffffff);
+	return 0;
+}
+
+static int apply_r_mips_hi16(struct module *me, uint32_t *location,
+			     Elf32_Addr v)
+{
+	struct mips_hi16 *n;
+
+	/*
+	 * We cannot relocate this one now because we don't know the value of
+	 * the carry we need to add.  Save the information, and let LO16 do the
+	 * actual relocation.
+	 */
+	n = kmalloc(sizeof *n, GFP_KERNEL);
+	if (!n)
+		return -ENOMEM;
+
+	n->addr = location;
+	n->value = v;
+	n->next = mips_hi16_list;
+	mips_hi16_list = n;
+
+	return 0;
+}
+
+static int apply_r_mips_lo16(struct module *me, uint32_t *location,
+			     Elf32_Addr v)
+{
+	unsigned long insnlo = *location;
+	Elf32_Addr val, vallo;
+
+	/* Sign extend the addend we extract from the lo insn.  */
+	vallo = ((insnlo & 0xffff) ^ 0x8000) - 0x8000;
+
+	if (mips_hi16_list != NULL) {
+		struct mips_hi16 *l;
+
+		l = mips_hi16_list;
+		while (l != NULL) {
+			struct mips_hi16 *next;
+			unsigned long insn;
+
+			/*
+			 * The value for the HI16 had best be the same.
+			 */
+			if (v != l->value) {
+				printk("%d != %d\n", v, l->value);
+				goto out_danger;
+			}
+
+
+			/*
+			 * Do the HI16 relocation.  Note that we actually don't
+			 * need to know anything about the LO16 itself, except
+			 * where to find the low 16 bits of the addend needed
+			 * by the LO16.
+			 */
+			insn = *l->addr;
+			val = ((insn & 0xffff) << 16) + vallo;
+			val += v;
+
+			/*
+			 * Account for the sign extension that will happen in
+			 * the low bits.
+			 */
+			val = ((val >> 16) + ((val & 0x8000) != 0)) & 0xffff;
+
+			insn = (insn & ~0xffff) | val;
+			*l->addr = insn;
+
+			next = l->next;
+			kfree(l);
+			l = next;
+		}
+
+		mips_hi16_list = NULL;
+	}
+
+	/*
+	 * Ok, we're done with the HI16 relocs.  Now deal with the LO16.
+	 */
+	val = v + vallo;
+	insnlo = (insnlo & ~0xffff) | (val & 0xffff);
+	*location = insnlo;
+
+	return 0;
+
+out_danger:
+	printk(KERN_ERR "module %s: dangerous " "relocation\n", me->name);
+
+	return -ENOEXEC;
+}
+
+static int (*reloc_handlers[]) (struct module *me, uint32_t *location,
+				Elf32_Addr v) = {
+	[R_MIPS_NONE]	= apply_r_mips_none,
+	[R_MIPS_32]	= apply_r_mips_32,
+	[R_MIPS_26]	= apply_r_mips_26,
+	[R_MIPS_HI16]	= apply_r_mips_hi16,
+	[R_MIPS_LO16]	= apply_r_mips_lo16,
+	[R_MIPS_GPREL16] = apply_r_mips_gprel16,
+	[R_MIPS_PC16] = apply_r_mips_pc16
+};
+
+
+int apply_relocations(Elf32_Shdr *sechdrs,
+		      const char *strtab,
+		      unsigned int symindex,
+		      unsigned int relsec,
+		      struct module *me)
+{
+	Elf32_Rel *rel = (void *) sechdrs[relsec].sh_addr;
+	Elf32_Sym *sym;
+	uint32_t *location;
+	unsigned int i;
+	Elf32_Addr v;
+	int res;
+
+	for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rel); i++) {
+		Elf32_Word r_info = rel[i].r_info;
+
+		/* This is where to make the change */
+		location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
+			+ rel[i].r_offset;
+		/* This is the symbol it is referring to */
+		sym = (Elf32_Sym *)sechdrs[symindex].sh_addr
+			+ ELF32_R_SYM(r_info);
+
+		if (!sym->st_value) {
+			printk(KERN_DEBUG "%s: undefined weak symbol %s\n",
+			       me->name, strtab + sym->st_name);
+			/* just print the warning, dont barf */
+		}
+
+		v = sym->st_value;
+
+		res = reloc_handlers[ELF32_R_TYPE(r_info)](me, location, v);
+		if( res ) {
+			printk(KERN_DEBUG
+			       "relocation error 0x%x sym refer <%s> value 0x%x "
+			       "type 0x%x r_info 0x%x\n",
+			       (unsigned int)location, strtab + sym->st_name, v,
+			       r_info, ELF32_R_TYPE(r_info));
+		}
+
+		if (res)
+			return res;
+	}
+
+	return 0;
+}
+
+void save_gp_address(unsigned int secbase, unsigned int rel)
+{
+	gp_addr = secbase + rel;
+	gp_offs = gp_addr - (secbase & 0xffff0000);
+}
+/* end module-elf32.c */
+
+
+
+/* Change all symbols so that sh_value encodes the pointer directly. */
+static int simplify_symbols(Elf_Shdr * sechdrs,
+			    unsigned int symindex,
+			    const char *strtab,
+			    const char *secstrings,
+			    unsigned int nsecs, struct module *mod)
+{
+	Elf_Sym *sym = (void *)sechdrs[symindex].sh_addr;
+	unsigned long secbase, bssbase = 0;
+	unsigned int i, n = sechdrs[symindex].sh_size / sizeof(Elf_Sym);
+	int ret = 0, size;
+
+	/* find the .bss section for COMMON symbols */
+	for (i = 0; i < nsecs; i++) {
+		if (strncmp(secstrings + sechdrs[i].sh_name, ".bss", 4) == 0)
+			bssbase = sechdrs[i].sh_addr;
+	}
+
+	for (i = 1; i < n; i++) {
+		switch (sym[i].st_shndx) {
+		case SHN_COMMON:
+			/* Allocate space for the symbol in the .bss section. st_value is currently size.
+			   We want it to have the address of the symbol. */
+
+			size = sym[i].st_value;
+			sym[i].st_value = bssbase;
+
+			bssbase += size;
+			break;
+
+		case SHN_ABS:
+			/* Don't need to do anything */
+			break;
+
+		case SHN_UNDEF:
+			/* ret = -ENOENT; */
+			break;
+
+		case SHN_MIPS_SCOMMON:
+
+			printk(KERN_DEBUG
+			       "simplify_symbols: ignoring SHN_MIPS_SCOMMON symbol <%s> st_shndx %d\n",
+			       strtab + sym[i].st_name, sym[i].st_shndx);
+
+			// .sbss section
+			break;
+
+		default:
+			secbase = sechdrs[sym[i].st_shndx].sh_addr;
+
+			if (strncmp(strtab + sym[i].st_name, "_gp", 3) == 0) {
+				save_gp_address(secbase, sym[i].st_value);
+			}
+
+			sym[i].st_value += secbase;
+			break;
+		}
+
+	}
+
+	return ret;
+}
+
+#ifdef DEBUG_ELFLOADER
+static void dump_elfsymbols(Elf_Shdr * sechdrs, unsigned int symindex,
+			    const char *strtab, struct module *mod)
+{
+	Elf_Sym *sym = (void *)sechdrs[symindex].sh_addr;
+	unsigned int i, n = sechdrs[symindex].sh_size / sizeof(Elf_Sym);
+
+	printk(KERN_DEBUG "dump_elfsymbols: n %d\n", n);
+	for (i = 1; i < n; i++) {
+		printk(KERN_DEBUG " i %d name <%s> 0x%x\n", i,
+		       strtab + sym[i].st_name, sym[i].st_value);
+	}
+}
+#endif
+
+static void dump_tc(struct tc *t)
+{
+	printk(KERN_WARNING "VPE: TC index %d TCStatus 0x%lx halt 0x%lx\n",
+	       t->index, read_tc_c0_tcstatus(), read_tc_c0_tchalt());
+	printk(KERN_WARNING "VPE: tcrestart 0x%lx\n", read_tc_c0_tcrestart());
+}
+
+static void dump_tclist(void)
+{
+	struct tc *t;
+
+	list_for_each_entry(t, &vpecontrol.tc_list, list) {
+		dump_tc(t);
+	}
+}
+
+/* We are prepared so configure and start the VPE... */
+int vpe_run(vpe_t * v)
+{
+	unsigned long val;
+	struct tc *t;
+
+	/* check we are the Master VPE */
+	val = read_c0_vpeconf0();
+	if (!(val & VPECONF0_MVP)) {
+		printk(KERN_WARNING
+		       "VPE: only Master VPE's are allowed to configure MT\n");
+		return -1;
+	}
+
+	/* disable MT (using dvpe) */
+	dvpe();
+
+	/* Put MVPE's into 'configuration state' */
+	write_c0_mvpcontrol(read_c0_mvpcontrol() | MVPCONTROL_VPC);
+
+	if (!list_empty(&v->tc)) {
+		if ((t = list_entry(v->tc.next, struct tc, tc)) == NULL) {
+			printk(KERN_WARNING "VPE: TC %d is already in use.\n",
+			       t->index);
+			return -ENOEXEC;
+		}
+	} else {
+		printk(KERN_WARNING "VPE: No TC's associated with VPE %d\n",
+		       v->minor);
+		return -ENOEXEC;
+	}
+
+	settc(t->index);
+
+	val = read_vpe_c0_vpeconf0();
+
+	/* should check it is halted, and not activated */
+	if ((read_tc_c0_tcstatus() & TCSTATUS_A) || !(read_tc_c0_tchalt() & TCHALT_H)) {
+		printk(KERN_WARNING "VPE: TC %d is already doing something!\n",
+		       t->index);
+
+		dump_tclist();
+		return -ENOEXEC;
+	}
+
+	/* Write the address we want it to start running from in the TCPC register. */
+	write_tc_c0_tcrestart((unsigned long)v->__start);
+
+	/* write the sivc_info address to tccontext */
+	write_tc_c0_tccontext((unsigned long)0);
+
+	/* Set up the XTC bit in vpeconf0 to point at our tc */
+	write_vpe_c0_vpeconf0(read_vpe_c0_vpeconf0() | (t->index << VPECONF0_XTC_SHIFT));
+
+	/* mark the TC as activated, not interrupt exempt and not dynamically allocatable */
+	val = read_tc_c0_tcstatus();
+	val = (val & ~(TCSTATUS_DA | TCSTATUS_IXMT)) | TCSTATUS_A;
+	write_tc_c0_tcstatus(val);
+
+	write_tc_c0_tchalt(read_tc_c0_tchalt() & ~TCHALT_H);
+
+	/* set up VPE1 */
+	write_vpe_c0_vpecontrol(read_vpe_c0_vpecontrol() & ~VPECONTROL_TE);	// no multiple TC's
+	write_vpe_c0_vpeconf0(read_vpe_c0_vpeconf0() | VPECONF0_VPA);	// enable this VPE
+
+	/*
+	 * The sde-kit passes 'memsize' to __start in $a3, so set something
+	 * here...
+	 * Or set $a3 (register 7) to zero and define DFLT_STACK_SIZE and
+	 * DFLT_HEAP_SIZE when you compile your program
+	 */
+
+	mttgpr(7, 0);
+
+	/* set config to be the same as vpe0, particularly kseg0 coherency alg */
+	write_vpe_c0_config(read_c0_config());
+
+	/* clear out any left overs from a previous program */
+	write_vpe_c0_cause(0);
+
+	/* take system out of configuration state */
+	write_c0_mvpcontrol(read_c0_mvpcontrol() & ~MVPCONTROL_VPC);
+
+	/* clear interrupts enabled IE, ERL, EXL, and KSU from c0 status */
+	write_vpe_c0_status(read_vpe_c0_status() & ~(ST0_ERL | ST0_KSU | ST0_IE | ST0_EXL));
+
+	/* set it running */
+	evpe(EVPE_ENABLE);
+
+	return 0;
+}
+
+static unsigned long find_vpe_symbols(vpe_t * v, Elf_Shdr * sechdrs,
+				      unsigned int symindex, const char *strtab,
+				      struct module *mod)
+{
+	Elf_Sym *sym = (void *)sechdrs[symindex].sh_addr;
+	unsigned int i, n = sechdrs[symindex].sh_size / sizeof(Elf_Sym);
+
+	for (i = 1; i < n; i++) {
+		if (strcmp(strtab + sym[i].st_name, "__start") == 0) {
+			v->__start = sym[i].st_value;
+		}
+
+		if (strcmp(strtab + sym[i].st_name, "vpe_shared") == 0) {
+			v->shared_ptr = (void *)sym[i].st_value;
+		}
+	}
+
+	return 0;
+}
+
+/* Allocates a VPE with some program code space(the load address), copies the contents
+   of the program (p)buffer performing relocatations/etc, free's it when finished.
+*/
+int vpe_elfload(vpe_t * v)
+{
+	Elf_Ehdr *hdr;
+	Elf_Shdr *sechdrs;
+	long err = 0;
+	char *secstrings, *strtab = NULL;
+	unsigned int len, i, symindex = 0, strindex = 0;
+
+	struct module mod;	// so we can re-use the relocations code
+
+	memset(&mod, 0, sizeof(struct module));
+	strcpy(mod.name, "VPE dummy prog module");
+
+	hdr = (Elf_Ehdr *) v->pbuffer;
+	len = v->plen;
+
+	/* Sanity checks against insmoding binaries or wrong arch,
+	   weird elf version */
+	if (memcmp(hdr->e_ident, ELFMAG, 4) != 0
+	    || hdr->e_type != ET_REL || !elf_check_arch(hdr)
+	    || hdr->e_shentsize != sizeof(*sechdrs)) {
+		printk(KERN_WARNING
+		       "VPE program, wrong arch or weird elf version\n");
+
+		return -ENOEXEC;
+	}
+
+	if (len < hdr->e_shoff + hdr->e_shnum * sizeof(Elf_Shdr)) {
+		printk(KERN_ERR "VPE program length %u truncated\n", len);
+		return -ENOEXEC;
+	}
+
+	/* Convenience variables */
+	sechdrs = (void *)hdr + hdr->e_shoff;
+	secstrings = (void *)hdr + sechdrs[hdr->e_shstrndx].sh_offset;
+	sechdrs[0].sh_addr = 0;
+
+	/* And these should exist, but gcc whinges if we don't init them */
+	symindex = strindex = 0;
+
+	for (i = 1; i < hdr->e_shnum; i++) {
+
+		if (sechdrs[i].sh_type != SHT_NOBITS
+		    && len < sechdrs[i].sh_offset + sechdrs[i].sh_size) {
+			printk(KERN_ERR "VPE program length %u truncated\n",
+			       len);
+			return -ENOEXEC;
+		}
+
+		/* Mark all sections sh_addr with their address in the
+		   temporary image. */
+		sechdrs[i].sh_addr = (size_t) hdr + sechdrs[i].sh_offset;
+
+		/* Internal symbols and strings. */
+		if (sechdrs[i].sh_type == SHT_SYMTAB) {
+			symindex = i;
+			strindex = sechdrs[i].sh_link;
+			strtab = (char *)hdr + sechdrs[strindex].sh_offset;
+		}
+	}
+
+	layout_sections(&mod, hdr, sechdrs, secstrings);
+
+	v->load_addr = alloc_progmem(mod.core_size);
+	memset(v->load_addr, 0, mod.core_size);
+
+	printk("VPE elf_loader: loading to %p\n", v->load_addr);
+
+	for (i = 0; i < hdr->e_shnum; i++) {
+		void *dest;
+
+		if (!(sechdrs[i].sh_flags & SHF_ALLOC))
+			continue;
+
+		dest = v->load_addr + sechdrs[i].sh_entsize;
+
+		if (sechdrs[i].sh_type != SHT_NOBITS)
+			memcpy(dest, (void *)sechdrs[i].sh_addr,
+			       sechdrs[i].sh_size);
+		/* Update sh_addr to point to copy in image. */
+		sechdrs[i].sh_addr = (unsigned long)dest;
+	}
+
+	/* Fix up syms, so that st_value is a pointer to location. */
+	err =
+		simplify_symbols(sechdrs, symindex, strtab, secstrings,
+				 hdr->e_shnum, &mod);
+	if (err < 0) {
+		printk(KERN_WARNING "VPE: unable to simplify symbols\n");
+		goto cleanup;
+	}
+
+	/* Now do relocations. */
+	for (i = 1; i < hdr->e_shnum; i++) {
+		const char *strtab = (char *)sechdrs[strindex].sh_addr;
+		unsigned int info = sechdrs[i].sh_info;
+
+		/* Not a valid relocation section? */
+		if (info >= hdr->e_shnum)
+			continue;
+
+		/* Don't bother with non-allocated sections */
+		if (!(sechdrs[info].sh_flags & SHF_ALLOC))
+			continue;
+
+		if (sechdrs[i].sh_type == SHT_REL)
+			err =
+				apply_relocations(sechdrs, strtab, symindex, i, &mod);
+		else if (sechdrs[i].sh_type == SHT_RELA)
+			err = apply_relocate_add(sechdrs, strtab, symindex, i,
+						 &mod);
+		if (err < 0) {
+			printk(KERN_WARNING
+			       "vpe_elfload: error in relocations err %ld\n",
+			       err);
+			goto cleanup;
+		}
+	}
+
+	/* make sure it's physically written out */
+	flush_icache_range((unsigned long)v->load_addr,
+			   (unsigned long)v->load_addr + v->len);
+
+	if ((find_vpe_symbols(v, sechdrs, symindex, strtab, &mod)) < 0) {
+
+		printk(KERN_WARNING
+		       "VPE: program doesn't contain __start or vpe_shared symbols\n");
+		err = -ENOEXEC;
+	}
+
+	printk(" elf loaded\n");
+
+cleanup:
+	return err;
+}
+
+static void dump_vpe(vpe_t * v)
+{
+	struct tc *t;
+
+	printk(KERN_DEBUG "VPEControl 0x%lx\n", read_vpe_c0_vpecontrol());
+	printk(KERN_DEBUG "VPEConf0 0x%lx\n", read_vpe_c0_vpeconf0());
+
+	list_for_each_entry(t, &vpecontrol.tc_list, list) {
+		dump_tc(t);
+	}
+}
+
+/* checks for VPE is unused and gets ready to load program	 */
+static int vpe_open(struct inode *inode, struct file *filp)
+{
+	int minor;
+	vpe_t *v;
+
+	/* assume only 1 device at the mo. */
+	if ((minor = MINOR(inode->i_rdev)) != 1) {
+		printk(KERN_WARNING "VPE: only vpe1 is supported\n");
+		return -ENODEV;
+	}
+
+	if ((v = get_vpe(minor)) == NULL) {
+		printk(KERN_WARNING "VPE: unable to get vpe\n");
+		return -ENODEV;
+	}
+
+	if (v->state != VPE_STATE_UNUSED) {
+		unsigned long tmp;
+		struct tc *t;
+
+		printk(KERN_WARNING "VPE: device %d already in use\n", minor);
+
+		dvpe();
+		dump_vpe(v);
+
+		printk(KERN_WARNING "VPE: re-initialising %d\n", minor);
+
+		release_progmem(v->load_addr);
+
+		t = get_tc(minor);
+		settc(minor);
+		tmp = read_tc_c0_tcstatus();
+
+		/* mark not allocated and not dynamically allocatable */
+		tmp &= ~(TCSTATUS_A | TCSTATUS_DA);
+		tmp |= TCSTATUS_IXMT;	/* interrupt exempt */
+		write_tc_c0_tcstatus(tmp);
+
+		write_tc_c0_tchalt(TCHALT_H);
+
+	}
+
+	// allocate it so when we get write ops we know it's expected.
+	v->state = VPE_STATE_INUSE;
+
+	/* this of-course trashes what was there before... */
+	v->pbuffer = vmalloc(P_SIZE);
+	v->plen = P_SIZE;
+	v->load_addr = NULL;
+	v->len = 0;
+
+	return 0;
+}
+
+static int vpe_release(struct inode *inode, struct file *filp)
+{
+	int minor, ret = 0;
+	vpe_t *v;
+	Elf_Ehdr *hdr;
+
+	minor = MINOR(inode->i_rdev);
+	if ((v = get_vpe(minor)) == NULL)
+		return -ENODEV;
+
+	// simple case of fire and forget, so tell the VPE to run...
+
+	hdr = (Elf_Ehdr *) v->pbuffer;
+	if (memcmp(hdr->e_ident, ELFMAG, 4) == 0) {
+		if (vpe_elfload(v) >= 0)
+			vpe_run(v);
+		else {
+			printk(KERN_WARNING "VPE: ELF load failed.\n");
+			ret = -ENOEXEC;
+		}
+	} else {
+		printk(KERN_WARNING "VPE: only elf files are supported\n");
+		ret = -ENOEXEC;
+	}
+
+	// cleanup any temp buffers
+	if (v->pbuffer)
+		vfree(v->pbuffer);
+	v->plen = 0;
+	return ret;
+}
+
+static ssize_t vpe_write(struct file *file, const char __user * buffer,
+			 size_t count, loff_t * ppos)
+{
+	int minor;
+	size_t ret = count;
+	vpe_t *v;
+
+	minor = MINOR(file->f_dentry->d_inode->i_rdev);
+	if ((v = get_vpe(minor)) == NULL)
+		return -ENODEV;
+
+	if (v->pbuffer == NULL) {
+		printk(KERN_ERR "vpe_write: no pbuffer\n");
+		return -ENOMEM;
+	}
+
+	if ((count + v->len) > v->plen) {
+		printk(KERN_WARNING
+		       "VPE Loader: elf size too big. Perhaps strip uneeded symbols\n");
+		return -ENOMEM;
+	}
+
+	count -= copy_from_user(v->pbuffer + v->len, buffer, count);
+	if (!count) {
+		printk("vpe_write: copy_to_user failed\n");
+		return -EFAULT;
+	}
+
+	v->len += count;
+	return ret;
+}
+
+static struct file_operations vpe_fops = {
+	.owner = THIS_MODULE,
+	.open = vpe_open,
+	.release = vpe_release,
+	.write = vpe_write
+};
+
+/* module wrapper entry points */
+/* give me a vpe */
+vpe_handle vpe_alloc(void)
+{
+	int i;
+	struct vpe *v;
+
+	/* find a vpe */
+	for (i = 1; i < MAX_VPES; i++) {
+		if ((v = get_vpe(i)) != NULL) {
+			v->state = VPE_STATE_INUSE;
+			return v;
+		}
+	}
+	return NULL;
+}
+
+EXPORT_SYMBOL(vpe_alloc);
+
+/* start running from here */
+int vpe_start(vpe_handle vpe, unsigned long start)
+{
+	struct vpe *v = vpe;
+
+	v->__start = start;
+	return vpe_run(v);
+}
+
+EXPORT_SYMBOL(vpe_start);
+
+/* halt it for now */
+int vpe_stop(vpe_handle vpe)
+{
+	struct vpe *v = vpe;
+	struct tc *t;
+	unsigned int evpe_flags;
+
+	evpe_flags = dvpe();
+
+	if ((t = list_entry(v->tc.next, struct tc, tc)) != NULL) {
+
+		settc(t->index);
+		write_vpe_c0_vpeconf0(read_vpe_c0_vpeconf0() & ~VPECONF0_VPA);
+	}
+
+	evpe(evpe_flags);
+
+	return 0;
+}
+
+EXPORT_SYMBOL(vpe_stop);
+
+/* I've done with it thank you */
+int vpe_free(vpe_handle vpe)
+{
+	struct vpe *v = vpe;
+	struct tc *t;
+	unsigned int evpe_flags;
+
+	if ((t = list_entry(v->tc.next, struct tc, tc)) == NULL) {
+		return -ENOEXEC;
+	}
+
+	evpe_flags = dvpe();
+
+	/* Put MVPE's into 'configuration state' */
+	write_c0_mvpcontrol(read_c0_mvpcontrol() | MVPCONTROL_VPC);
+
+	settc(t->index);
+	write_vpe_c0_vpeconf0(read_vpe_c0_vpeconf0() & ~VPECONF0_VPA);
+
+	/* mark the TC unallocated and halt'ed */
+	write_tc_c0_tcstatus(read_tc_c0_tcstatus() & ~TCSTATUS_A);
+	write_tc_c0_tchalt(TCHALT_H);
+
+	v->state = VPE_STATE_UNUSED;
+
+	write_c0_mvpcontrol(read_c0_mvpcontrol() & ~MVPCONTROL_VPC);
+	evpe(evpe_flags);
+
+	return 0;
+}
+
+EXPORT_SYMBOL(vpe_free);
+
+void *vpe_get_shared(int index)
+{
+	struct vpe *v;
+
+	if ((v = get_vpe(index)) == NULL) {
+		printk(KERN_WARNING "vpe: invalid vpe index %d\n", index);
+		return NULL;
+	}
+
+	return v->shared_ptr;
+}
+
+EXPORT_SYMBOL(vpe_get_shared);
+
+static int __init vpe_module_init(void)
+{
+	struct vpe *v = NULL;
+	struct tc *t;
+	unsigned long val;
+	int i;
+
+	if (!cpu_has_mipsmt) {
+		printk("VPE loader: not a MIPS MT capable processor\n");
+		return -ENODEV;
+	}
+
+	if ((major = register_chrdev(VPE_MAJOR, module_name, &vpe_fops) < 0)) {
+		printk("VPE loader: unable to register character device\n");
+		return -EBUSY;
+	}
+
+	if (major == 0)
+		major = VPE_MAJOR;
+
+	dmt();
+	dvpe();
+
+	/* Put MVPE's into 'configuration state' */
+	write_c0_mvpcontrol(read_c0_mvpcontrol() | MVPCONTROL_VPC);
+
+	/* dump_mtregs(); */
+
+	INIT_LIST_HEAD(&vpecontrol.vpe_list);
+	INIT_LIST_HEAD(&vpecontrol.tc_list);
+
+	val = read_c0_mvpconf0();
+	for (i = 0; i < ((val & MVPCONF0_PTC) + 1); i++) {
+		t = alloc_tc(i);
+
+		/* VPE's */
+		if (i < ((val & MVPCONF0_PVPE) >> MVPCONF0_PVPE_SHIFT) + 1) {
+			settc(i);
+
+			if ((v = alloc_vpe(i)) == NULL) {
+				printk(KERN_WARNING "VPE: unable to allocate VPE\n");
+				return -ENODEV;
+			}
+
+			list_add(&t->tc, &v->tc);	/* add the tc to the list of this vpe's tc's. */
+
+			/* deactivate all but vpe0 */
+			if (i != 0) {
+				unsigned long tmp = read_vpe_c0_vpeconf0();
+
+				tmp &= ~VPECONF0_VPA;
+
+				/* master VPE */
+				tmp |= VPECONF0_MVP;
+				write_vpe_c0_vpeconf0(tmp);
+			}
+
+			/* disable multi-threading with TC's */
+			write_vpe_c0_vpecontrol(read_vpe_c0_vpecontrol() & ~VPECONTROL_TE);
+
+			if (i != 0) {
+				write_vpe_c0_status((read_c0_status() &
+						     ~(ST0_IM | ST0_IE | ST0_KSU))
+						    | ST0_CU0);
+
+				/* set config to be the same as vpe0, particularly kseg0 coherency alg */
+				write_vpe_c0_config(read_c0_config());
+			}
+
+		}
+
+		/* TC's */
+		t->pvpe = v;	/* set the parent vpe */
+
+		if (i != 0) {
+			unsigned long tmp;
+
+			/* tc 0 will of course be running.... */
+			if (i == 0)
+				t->state = TC_STATE_RUNNING;
+
+			settc(i);
+
+			/* bind a TC to each VPE, May as well put all excess TC's
+			   on the last VPE */
+			if (i >= (((val & MVPCONF0_PVPE) >> MVPCONF0_PVPE_SHIFT) + 1))
+				write_tc_c0_tcbind(read_tc_c0_tcbind() |
+						   ((val & MVPCONF0_PVPE) >> MVPCONF0_PVPE_SHIFT));
+			else
+				write_tc_c0_tcbind(read_tc_c0_tcbind() | i);
+
+			tmp = read_tc_c0_tcstatus();
+
+			/* mark not allocated and not dynamically allocatable */
+			tmp &= ~(TCSTATUS_A | TCSTATUS_DA);
+			tmp |= TCSTATUS_IXMT;	/* interrupt exempt */
+			write_tc_c0_tcstatus(tmp);
+
+			write_tc_c0_tchalt(TCHALT_H);
+		}
+	}
+
+	/* release config state */
+	write_c0_mvpcontrol(read_c0_mvpcontrol() & ~MVPCONTROL_VPC);
+
+	return 0;
+}
+
+static void __exit vpe_module_exit(void)
+{
+	struct vpe *v, *n;
+
+	list_for_each_entry_safe(v, n, &vpecontrol.vpe_list, list) {
+		if (v->state != VPE_STATE_UNUSED) {
+			release_vpe(v);
+		}
+	}
+
+	unregister_chrdev(major, module_name);
+}
+
+module_init(vpe_module_init);
+module_exit(vpe_module_exit);
+MODULE_DESCRIPTION("MIPS VPE Loader");
+MODULE_AUTHOR("Elizabeth Clarke, MIPS Technologies, Inc");
+MODULE_LICENSE("GPL");