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!

10 files changed, 2323 insertions, 0 deletions

diff --git a/arch/parisc/lib/Makefile b/arch/parisc/lib/Makefile
new file mode 100644
index 00000000000..7bf70567629
--- /dev/null
+++ b/arch/parisc/lib/Makefile
@@ -0,0 +1,9 @@
+#
+# Makefile for parisc-specific library files
+#
+
+lib-y	:= lusercopy.o bitops.o checksum.o io.o memset.o fixup.o memcpy.o
+
+obj-y	:= iomap.o
+
+lib-$(CONFIG_SMP) += debuglocks.o
diff --git a/arch/parisc/lib/bitops.c b/arch/parisc/lib/bitops.c
new file mode 100644
index 00000000000..2de182f6fe8
--- /dev/null
+++ b/arch/parisc/lib/bitops.c
@@ -0,0 +1,84 @@
+/*
+ * bitops.c: atomic operations which got too long to be inlined all over
+ *      the place.
+ * 
+ * Copyright 1999 Philipp Rumpf (prumpf@tux.org)
+ * Copyright 2000 Grant Grundler (grundler@cup.hp.com)
+ */
+
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/spinlock.h>
+#include <asm/system.h>
+#include <asm/atomic.h>
+
+#ifdef CONFIG_SMP
+spinlock_t __atomic_hash[ATOMIC_HASH_SIZE] __lock_aligned = {
+	[0 ... (ATOMIC_HASH_SIZE-1)]  = SPIN_LOCK_UNLOCKED
+};
+#endif
+
+#ifdef __LP64__
+unsigned long __xchg64(unsigned long x, unsigned long *ptr)
+{
+	unsigned long temp, flags;
+
+	_atomic_spin_lock_irqsave(ptr, flags);
+	temp = *ptr;
+	*ptr = x;
+	_atomic_spin_unlock_irqrestore(ptr, flags);
+	return temp;
+}
+#endif
+
+unsigned long __xchg32(int x, int *ptr)
+{
+	unsigned long flags;
+	long temp;
+
+	_atomic_spin_lock_irqsave(ptr, flags);
+	temp = (long) *ptr;	/* XXX - sign extension wanted? */
+	*ptr = x;
+	_atomic_spin_unlock_irqrestore(ptr, flags);
+	return (unsigned long)temp;
+}
+
+
+unsigned long __xchg8(char x, char *ptr)
+{
+	unsigned long flags;
+	long temp;
+
+	_atomic_spin_lock_irqsave(ptr, flags);
+	temp = (long) *ptr;	/* XXX - sign extension wanted? */
+	*ptr = x;
+	_atomic_spin_unlock_irqrestore(ptr, flags);
+	return (unsigned long)temp;
+}
+
+
+#ifdef __LP64__
+unsigned long __cmpxchg_u64(volatile unsigned long *ptr, unsigned long old, unsigned long new)
+{
+	unsigned long flags;
+	unsigned long prev;
+
+	_atomic_spin_lock_irqsave(ptr, flags);
+	if ((prev = *ptr) == old)
+		*ptr = new;
+	_atomic_spin_unlock_irqrestore(ptr, flags);
+	return prev;
+}
+#endif
+
+unsigned long __cmpxchg_u32(volatile unsigned int *ptr, unsigned int old, unsigned int new)
+{
+	unsigned long flags;
+	unsigned int prev;
+
+	_atomic_spin_lock_irqsave(ptr, flags);
+	if ((prev = *ptr) == old)
+		*ptr = new;
+	_atomic_spin_unlock_irqrestore(ptr, flags);
+	return (unsigned long)prev;
+}
diff --git a/arch/parisc/lib/checksum.c b/arch/parisc/lib/checksum.c
new file mode 100644
index 00000000000..8a1e08068e7
--- /dev/null
+++ b/arch/parisc/lib/checksum.c
@@ -0,0 +1,148 @@
+/*
+ * INET		An implementation of the TCP/IP protocol suite for the LINUX
+ *		operating system.  INET is implemented using the  BSD Socket
+ *		interface as the means of communication with the user level.
+ *
+ *		MIPS specific IP/TCP/UDP checksumming routines
+ *
+ * Authors:	Ralf Baechle, <ralf@waldorf-gmbh.de>
+ *		Lots of code moved from tcp.c and ip.c; see those files
+ *		for more names.
+ *
+ *		This program is free software; you can redistribute it and/or
+ *		modify it under the terms of the GNU General Public License
+ *		as published by the Free Software Foundation; either version
+ *		2 of the License, or (at your option) any later version.
+ *
+ * $Id: checksum.c,v 1.3 1997/12/01 17:57:34 ralf Exp $
+ */
+#include <linux/module.h>
+#include <linux/types.h>
+
+#include <net/checksum.h>
+#include <asm/byteorder.h>
+#include <asm/string.h>
+#include <asm/uaccess.h>
+
+#define addc(_t,_r)                     \
+	__asm__ __volatile__ (          \
+"       add             %0, %1, %0\n"   \
+"       addc            %0, %%r0, %0\n" \
+	: "=r"(_t)                      \
+	: "r"(_r), "0"(_t));
+
+static inline unsigned short from32to16(unsigned int x)
+{
+	/* 32 bits --> 16 bits + carry */
+	x = (x & 0xffff) + (x >> 16);
+	/* 16 bits + carry --> 16 bits including carry */
+	x = (x & 0xffff) + (x >> 16);
+	return (unsigned short)x;
+}
+
+static inline unsigned int do_csum(const unsigned char * buff, int len)
+{
+	int odd, count;
+	unsigned int result = 0;
+
+	if (len <= 0)
+		goto out;
+	odd = 1 & (unsigned long) buff;
+	if (odd) {
+		result = be16_to_cpu(*buff);
+		len--;
+		buff++;
+	}
+	count = len >> 1;		/* nr of 16-bit words.. */
+	if (count) {
+		if (2 & (unsigned long) buff) {
+			result += *(unsigned short *) buff;
+			count--;
+			len -= 2;
+			buff += 2;
+		}
+		count >>= 1;		/* nr of 32-bit words.. */
+		if (count) {
+			while (count >= 4) {
+				unsigned int r1, r2, r3, r4;
+				r1 = *(unsigned int *)(buff + 0);
+				r2 = *(unsigned int *)(buff + 4);
+				r3 = *(unsigned int *)(buff + 8);
+				r4 = *(unsigned int *)(buff + 12);
+				addc(result, r1);
+				addc(result, r2);
+				addc(result, r3);
+				addc(result, r4);
+				count -= 4;
+				buff += 16;
+			}
+			while (count) {
+				unsigned int w = *(unsigned int *) buff;
+				count--;
+				buff += 4;
+				addc(result, w);
+			}
+			result = (result & 0xffff) + (result >> 16);
+		}
+		if (len & 2) {
+			result += *(unsigned short *) buff;
+			buff += 2;
+		}
+	}
+	if (len & 1)
+		result += le16_to_cpu(*buff);
+	result = from32to16(result);
+	if (odd)
+		result = swab16(result);
+out:
+	return result;
+}
+
+/*
+ * computes a partial checksum, e.g. for TCP/UDP fragments
+ */
+unsigned int csum_partial(const unsigned char *buff, int len, unsigned int sum)
+{
+	unsigned int result = do_csum(buff, len);
+	addc(result, sum);
+	return from32to16(result);
+}
+
+EXPORT_SYMBOL(csum_partial);
+
+/*
+ * copy while checksumming, otherwise like csum_partial
+ */
+unsigned int csum_partial_copy_nocheck(const unsigned char *src, unsigned char *dst,
+				       int len, unsigned int sum)
+{
+	/*
+	 * It's 2:30 am and I don't feel like doing it real ...
+	 * This is lots slower than the real thing (tm)
+	 */
+	sum = csum_partial(src, len, sum);
+	memcpy(dst, src, len);
+
+	return sum;
+}
+EXPORT_SYMBOL(csum_partial_copy_nocheck);
+
+/*
+ * Copy from userspace and compute checksum.  If we catch an exception
+ * then zero the rest of the buffer.
+ */
+unsigned int csum_partial_copy_from_user(const unsigned char __user *src,
+					unsigned char *dst, int len,
+					unsigned int sum, int *err_ptr)
+{
+	int missing;
+
+	missing = copy_from_user(dst, src, len);
+	if (missing) {
+		memset(dst + len - missing, 0, missing);
+		*err_ptr = -EFAULT;
+	}
+		
+	return csum_partial(dst, len, sum);
+}
+EXPORT_SYMBOL(csum_partial_copy_from_user);
diff --git a/arch/parisc/lib/debuglocks.c b/arch/parisc/lib/debuglocks.c
new file mode 100644
index 00000000000..1b33fe6e5b7
--- /dev/null
+++ b/arch/parisc/lib/debuglocks.c
@@ -0,0 +1,277 @@
+/* 
+ *    Debugging versions of SMP locking primitives.
+ *
+ *    Copyright (C) 2004 Thibaut VARENE <varenet@parisc-linux.org>
+ *
+ *    Some code stollen from alpha & sparc64 ;)
+ *
+ *    This program is free software; you can redistribute it and/or modify
+ *    it under the terms of the GNU General Public License as published by
+ *    the Free Software Foundation; either version 2 of the License, or
+ *    (at your option) any later version.
+ *
+ *    This program is distributed in the hope that 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
+ *
+ *    We use pdc_printf() throughout the file for all output messages, to avoid
+ *    losing messages because of disabled interrupts. Since we're using these
+ *    messages for debugging purposes, it makes sense not to send them to the
+ *    linux console.
+ */
+
+
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/spinlock.h>
+#include <linux/hardirq.h>	/* in_interrupt() */
+#include <asm/system.h>
+#include <asm/hardirq.h>	/* in_interrupt() */
+#include <asm/pdc.h>
+
+#undef INIT_STUCK
+#define INIT_STUCK 1L << 30
+
+#ifdef CONFIG_DEBUG_SPINLOCK
+
+
+void _dbg_spin_lock(spinlock_t * lock, const char *base_file, int line_no)
+{
+	volatile unsigned int *a;
+	long stuck = INIT_STUCK;
+	void *inline_pc = __builtin_return_address(0);
+	unsigned long started = jiffies;
+	int printed = 0;
+	int cpu = smp_processor_id();
+
+try_again:
+
+	/* Do the actual locking */
+	/* <T-Bone> ggg: we can't get stuck on the outter loop?
+	 * <ggg> T-Bone: We can hit the outer loop
+	 *	alot if multiple CPUs are constantly racing for a lock
+	 *	and the backplane is NOT fair about which CPU sees
+	 *	the update first. But it won't hang since every failed
+	 *	attempt will drop us back into the inner loop and
+	 *	decrement `stuck'.
+	 * <ggg> K-class and some of the others are NOT fair in the HW
+	 * 	implementation so we could see false positives.
+	 * 	But fixing the lock contention is easier than
+	 * 	fixing the HW to be fair.
+	 * <tausq> __ldcw() returns 1 if we get the lock; otherwise we
+	 * 	spin until the value of the lock changes, or we time out.
+	 */
+	mb();
+	a = __ldcw_align(lock);
+	while (stuck && (__ldcw(a) == 0))
+		while ((*a == 0) && --stuck);
+	mb();
+
+	if (unlikely(stuck <= 0)) {
+		pdc_printf(
+			"%s:%d: spin_lock(%s/%p) stuck in %s at %p(%d)"
+			" owned by %s:%d in %s at %p(%d)\n",
+			base_file, line_no, lock->module, lock,
+			current->comm, inline_pc, cpu,
+			lock->bfile, lock->bline, lock->task->comm,
+			lock->previous, lock->oncpu);
+		stuck = INIT_STUCK;
+		printed = 1;
+		goto try_again;
+	}
+
+	/* Exiting.  Got the lock.  */
+	lock->oncpu = cpu;
+	lock->previous = inline_pc;
+	lock->task = current;
+	lock->bfile = (char *)base_file;
+	lock->bline = line_no;
+
+	if (unlikely(printed)) {
+		pdc_printf(
+			"%s:%d: spin_lock grabbed in %s at %p(%d) %ld ticks\n",
+			base_file, line_no, current->comm, inline_pc,
+			cpu, jiffies - started);
+	}
+}
+
+void _dbg_spin_unlock(spinlock_t * lock, const char *base_file, int line_no)
+{
+	CHECK_LOCK(lock);
+	volatile unsigned int *a;
+	mb();
+	a = __ldcw_align(lock);
+	if (unlikely((*a != 0) && lock->babble)) {
+		lock->babble--;
+		pdc_printf(
+			"%s:%d: spin_unlock(%s:%p) not locked\n",
+			base_file, line_no, lock->module, lock);
+	}
+	*a = 1;	
+	mb();
+}
+
+int _dbg_spin_trylock(spinlock_t * lock, const char *base_file, int line_no)
+{
+	int ret;
+	volatile unsigned int *a;
+	mb();
+	a = __ldcw_align(lock);
+	ret = (__ldcw(a) != 0);
+	mb();
+	if (ret) {
+		lock->oncpu = smp_processor_id();
+		lock->previous = __builtin_return_address(0);
+		lock->task = current;
+	} else {
+		lock->bfile = (char *)base_file;
+		lock->bline = line_no;
+	}
+	return ret;
+}
+
+#endif /* CONFIG_DEBUG_SPINLOCK */
+
+#ifdef CONFIG_DEBUG_RWLOCK
+
+/* Interrupts trouble detailed explanation, thx Grant:
+ *
+ * o writer (wants to modify data) attempts to acquire the rwlock
+ * o He gets the write lock.
+ * o Interupts are still enabled, we take an interrupt with the
+ *   write still holding the lock.
+ * o interrupt handler tries to acquire the rwlock for read.
+ * o deadlock since the writer can't release it at this point.
+ * 
+ * In general, any use of spinlocks that competes between "base"
+ * level and interrupt level code will risk deadlock. Interrupts
+ * need to be disabled in the base level routines to avoid it.
+ * Or more precisely, only the IRQ the base level routine
+ * is competing with for the lock.  But it's more efficient/faster
+ * to just disable all interrupts on that CPU to guarantee
+ * once it gets the lock it can release it quickly too.
+ */
+ 
+void _dbg_write_lock(rwlock_t *rw, const char *bfile, int bline)
+{
+	void *inline_pc = __builtin_return_address(0);
+	unsigned long started = jiffies;
+	long stuck = INIT_STUCK;
+	int printed = 0;
+	int cpu = smp_processor_id();
+	
+	if(unlikely(in_interrupt())) {	/* acquiring write lock in interrupt context, bad idea */
+		pdc_printf("write_lock caller: %s:%d, IRQs enabled,\n", bfile, bline);
+		BUG();
+	}
+
+	/* Note: if interrupts are disabled (which is most likely), the printk
+	will never show on the console. We might need a polling method to flush
+	the dmesg buffer anyhow. */
+	
+retry:
+	_raw_spin_lock(&rw->lock);
+
+	if(rw->counter != 0) {
+		/* this basically never happens */
+		_raw_spin_unlock(&rw->lock);
+		
+		stuck--;
+		if ((unlikely(stuck <= 0)) && (rw->counter < 0)) {
+			pdc_printf(
+				"%s:%d: write_lock stuck on writer"
+				" in %s at %p(%d) %ld ticks\n",
+				bfile, bline, current->comm, inline_pc,
+				cpu, jiffies - started);
+			stuck = INIT_STUCK;
+			printed = 1;
+		}
+		else if (unlikely(stuck <= 0)) {
+			pdc_printf(
+				"%s:%d: write_lock stuck on reader"
+				" in %s at %p(%d) %ld ticks\n",
+				bfile, bline, current->comm, inline_pc,
+				cpu, jiffies - started);
+			stuck = INIT_STUCK;
+			printed = 1;
+		}
+		
+		while(rw->counter != 0);
+
+		goto retry;
+	}
+
+	/* got it.  now leave without unlocking */
+	rw->counter = -1; /* remember we are locked */
+
+	if (unlikely(printed)) {
+		pdc_printf(
+			"%s:%d: write_lock grabbed in %s at %p(%d) %ld ticks\n",
+			bfile, bline, current->comm, inline_pc,
+			cpu, jiffies - started);
+	}
+}
+
+int _dbg_write_trylock(rwlock_t *rw, const char *bfile, int bline)
+{
+#if 0
+	void *inline_pc = __builtin_return_address(0);
+	int cpu = smp_processor_id();
+#endif
+	
+	if(unlikely(in_interrupt())) {	/* acquiring write lock in interrupt context, bad idea */
+		pdc_printf("write_lock caller: %s:%d, IRQs enabled,\n", bfile, bline);
+		BUG();
+	}
+
+	/* Note: if interrupts are disabled (which is most likely), the printk
+	will never show on the console. We might need a polling method to flush
+	the dmesg buffer anyhow. */
+	
+	_raw_spin_lock(&rw->lock);
+
+	if(rw->counter != 0) {
+		/* this basically never happens */
+		_raw_spin_unlock(&rw->lock);
+		return 0;
+	}
+
+	/* got it.  now leave without unlocking */
+	rw->counter = -1; /* remember we are locked */
+#if 0
+	pdc_printf("%s:%d: try write_lock grabbed in %s at %p(%d)\n",
+		   bfile, bline, current->comm, inline_pc, cpu);
+#endif
+	return 1;
+}
+
+void _dbg_read_lock(rwlock_t * rw, const char *bfile, int bline)
+{
+#if 0
+	void *inline_pc = __builtin_return_address(0);
+	unsigned long started = jiffies;
+	int cpu = smp_processor_id();
+#endif
+	unsigned long flags;
+
+	local_irq_save(flags);
+	_raw_spin_lock(&rw->lock); 
+
+	rw->counter++;
+#if 0
+	pdc_printf(
+		"%s:%d: read_lock grabbed in %s at %p(%d) %ld ticks\n",
+		bfile, bline, current->comm, inline_pc,
+		cpu, jiffies - started);
+#endif
+	_raw_spin_unlock(&rw->lock);
+	local_irq_restore(flags);
+}
+
+#endif /* CONFIG_DEBUG_RWLOCK */
diff --git a/arch/parisc/lib/fixup.S b/arch/parisc/lib/fixup.S
new file mode 100644
index 00000000000..134f0cd240f
--- /dev/null
+++ b/arch/parisc/lib/fixup.S
@@ -0,0 +1,89 @@
+/*
+ * Linux/PA-RISC Project (http://www.parisc-linux.org/)
+ *
+ *  Copyright (C) 2004  Randolph Chung <tausq@debian.org>
+ *
+ *    This program is free software; you can redistribute it and/or modify
+ *    it under the terms of the GNU General Public License as published by
+ *    the Free Software Foundation; either version 2, or (at your option)
+ *    any later version.
+ *
+ *    This program is distributed in the hope that 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., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * 
+ * Fixup routines for kernel exception handling.
+ */
+#include <linux/config.h>
+#include <asm/offsets.h>
+#include <asm/assembly.h>
+#include <asm/errno.h>
+
+#ifdef CONFIG_SMP
+	.macro  get_fault_ip t1 t2
+	addil LT%__per_cpu_offset,%r27
+	LDREG RT%__per_cpu_offset(%r1),\t1
+	/* t2 = smp_processor_id() */
+	mfctl 30,\t2
+	ldw TI_CPU(\t2),\t2
+#ifdef __LP64__
+	extrd,u \t2,63,32,\t2
+#endif
+	/* t2 = &__per_cpu_offset[smp_processor_id()]; */
+	LDREG,s \t2(\t1),\t2 
+	addil LT%per_cpu__exception_data,%r27
+	LDREG RT%per_cpu__exception_data(%r1),\t1
+	/* t1 = &__get_cpu_var(exception_data) */
+	add,l \t1,\t2,\t1
+	/* t1 = t1->fault_ip */
+	LDREG EXCDATA_IP(\t1), \t1
+	.endm
+#else
+	.macro  get_fault_ip t1 t2
+	/* t1 = &__get_cpu_var(exception_data) */
+	addil LT%per_cpu__exception_data,%r27
+	LDREG RT%per_cpu__exception_data(%r1),\t2
+	/* t1 = t2->fault_ip */
+	LDREG EXCDATA_IP(\t2), \t1
+	.endm
+#endif
+
+	.text
+	.section .fixup, "ax"
+
+	/* get_user() fixups, store -EFAULT in r8, and 0 in r9 */
+	.export fixup_get_user_skip_1
+fixup_get_user_skip_1:
+	get_fault_ip %r1,%r8
+	ldo 4(%r1), %r1
+	ldi -EFAULT, %r8
+	bv %r0(%r1)
+	copy %r0, %r9
+
+	.export fixup_get_user_skip_2
+fixup_get_user_skip_2:
+	get_fault_ip %r1,%r8
+	ldo 8(%r1), %r1
+	ldi -EFAULT, %r8
+	bv %r0(%r1)
+	copy %r0, %r9
+
+	/* put_user() fixups, store -EFAULT in r8 */
+	.export fixup_put_user_skip_1
+fixup_put_user_skip_1:
+	get_fault_ip %r1,%r8
+	ldo 4(%r1), %r1
+	bv %r0(%r1)
+	ldi -EFAULT, %r8
+
+	.export fixup_put_user_skip_2
+fixup_put_user_skip_2:
+	get_fault_ip %r1,%r8
+	ldo 8(%r1), %r1
+	bv %r0(%r1)
+	ldi -EFAULT, %r8
diff --git a/arch/parisc/lib/io.c b/arch/parisc/lib/io.c
new file mode 100644
index 00000000000..7c1406ff825
--- /dev/null
+++ b/arch/parisc/lib/io.c
@@ -0,0 +1,488 @@
+/*
+ * arch/parisc/lib/io.c
+ *
+ * Copyright (c) Matthew Wilcox 2001 for Hewlett-Packard
+ * Copyright (c) Randolph Chung 2001 <tausq@debian.org>
+ *
+ * IO accessing functions which shouldn't be inlined because they're too big
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <asm/io.h>
+
+/* Copies a block of memory to a device in an efficient manner.
+ * Assumes the device can cope with 32-bit transfers.  If it can't,
+ * don't use this function.
+ */
+void memcpy_toio(volatile void __iomem *dst, const void *src, int count)
+{
+	if (((unsigned long)dst & 3) != ((unsigned long)src & 3))
+		goto bytecopy;
+	while ((unsigned long)dst & 3) {
+		writeb(*(char *)src, dst++);
+		src++;
+		count--;
+	}
+	while (count > 3) {
+		__raw_writel(*(u32 *)src, dst);
+		src += 4;
+		dst += 4;
+		count -= 4;
+	}
+ bytecopy:
+	while (count--) {
+		writeb(*(char *)src, dst++);
+		src++;
+	}
+}
+
+/*
+** Copies a block of memory from a device in an efficient manner.
+** Assumes the device can cope with 32-bit transfers.  If it can't,
+** don't use this function.
+**
+** CR16 counts on C3000 reading 256 bytes from Symbios 896 RAM:
+**	27341/64    = 427 cyc per int
+**	61311/128   = 478 cyc per short
+**	122637/256  = 479 cyc per byte
+** Ergo bus latencies dominant (not transfer size).
+**      Minimize total number of transfers at cost of CPU cycles.
+**	TODO: only look at src alignment and adjust the stores to dest.
+*/
+void memcpy_fromio(void *dst, const volatile void __iomem *src, int count)
+{
+	/* first compare alignment of src/dst */ 
+	if ( (((unsigned long)dst ^ (unsigned long)src) & 1) || (count < 2) )
+		goto bytecopy;
+
+	if ( (((unsigned long)dst ^ (unsigned long)src) & 2) || (count < 4) )
+		goto shortcopy;
+
+	/* Then check for misaligned start address */
+	if ((unsigned long)src & 1) {
+		*(u8 *)dst = readb(src);
+		src++;
+		dst++;
+		count--;
+		if (count < 2) goto bytecopy;
+	}
+
+	if ((unsigned long)src & 2) {
+		*(u16 *)dst = __raw_readw(src);
+		src += 2;
+		dst += 2;
+		count -= 2;
+	}
+
+	while (count > 3) {
+		*(u32 *)dst = __raw_readl(src);
+		dst += 4;
+		src += 4;
+		count -= 4;
+	}
+
+ shortcopy:
+	while (count > 1) {
+		*(u16 *)dst = __raw_readw(src);
+		src += 2;
+		dst += 2;
+		count -= 2;
+	}
+
+ bytecopy:
+	while (count--) {
+		*(char *)dst = readb(src);
+		src++;
+		dst++;
+	}
+}
+
+/* Sets a block of memory on a device to a given value.
+ * Assumes the device can cope with 32-bit transfers.  If it can't,
+ * don't use this function.
+ */
+void memset_io(volatile void __iomem *addr, unsigned char val, int count)
+{
+	u32 val32 = (val << 24) | (val << 16) | (val << 8) | val;
+	while ((unsigned long)addr & 3) {
+		writeb(val, addr++);
+		count--;
+	}
+	while (count > 3) {
+		__raw_writel(val32, addr);
+		addr += 4;
+		count -= 4;
+	}
+	while (count--) {
+		writeb(val, addr++);
+	}
+}
+
+/*
+ * Read COUNT 8-bit bytes from port PORT into memory starting at
+ * SRC.
+ */
+void insb (unsigned long port, void *dst, unsigned long count)
+{
+	unsigned char *p;
+
+	p = (unsigned char *)dst;
+
+	while (((unsigned long)p) & 0x3) {
+		if (!count)
+			return;
+		count--;
+		*p = inb(port);
+		p++;
+	}
+
+	while (count >= 4) {
+		unsigned int w;
+		count -= 4;
+		w = inb(port) << 24;
+		w |= inb(port) << 16;
+		w |= inb(port) << 8;
+		w |= inb(port);
+		*(unsigned int *) p = w;
+		p += 4;
+	}
+
+	while (count) {
+		--count;
+		*p = inb(port);
+		p++;
+	}
+}
+
+
+/*
+ * Read COUNT 16-bit words from port PORT into memory starting at
+ * SRC.  SRC must be at least short aligned.  This is used by the
+ * IDE driver to read disk sectors.  Performance is important, but
+ * the interfaces seems to be slow: just using the inlined version
+ * of the inw() breaks things.
+ */
+void insw (unsigned long port, void *dst, unsigned long count)
+{
+	unsigned int l = 0, l2;
+	unsigned char *p;
+
+	p = (unsigned char *)dst;
+	
+	if (!count)
+		return;
+	
+	switch (((unsigned long)p) & 0x3)
+	{
+	 case 0x00:			/* Buffer 32-bit aligned */
+		while (count>=2) {
+			
+			count -= 2;
+			l = cpu_to_le16(inw(port)) << 16;
+			l |= cpu_to_le16(inw(port));
+			*(unsigned int *)p = l;
+			p += 4;
+		}
+		if (count) {
+			*(unsigned short *)p = cpu_to_le16(inw(port));
+		}
+		break;
+	
+	 case 0x02:			/* Buffer 16-bit aligned */
+		*(unsigned short *)p = cpu_to_le16(inw(port));
+		p += 2;
+		count--;
+		while (count>=2) {
+			
+			count -= 2;
+			l = cpu_to_le16(inw(port)) << 16;
+			l |= cpu_to_le16(inw(port));
+			*(unsigned int *)p = l;
+			p += 4;
+		}
+		if (count) {
+			*(unsigned short *)p = cpu_to_le16(inw(port));
+		}
+		break;
+		
+	 case 0x01:			/* Buffer 8-bit aligned */
+	 case 0x03:
+		/* I don't bother with 32bit transfers
+		 * in this case, 16bit will have to do -- DE */
+		--count;
+		
+		l = cpu_to_le16(inw(port));
+		*p = l >> 8;
+		p++;
+		while (count--)
+		{
+			l2 = cpu_to_le16(inw(port));
+			*(unsigned short *)p = (l & 0xff) << 8 | (l2 >> 8);
+			p += 2;
+			l = l2;
+		}
+		*p = l & 0xff;
+		break;
+	}
+}
+
+
+
+/*
+ * Read COUNT 32-bit words from port PORT into memory starting at
+ * SRC. Now works with any alignment in SRC. Performance is important,
+ * but the interfaces seems to be slow: just using the inlined version
+ * of the inl() breaks things.
+ */
+void insl (unsigned long port, void *dst, unsigned long count)
+{
+	unsigned int l = 0, l2;
+	unsigned char *p;
+
+	p = (unsigned char *)dst;
+	
+	if (!count)
+		return;
+	
+	switch (((unsigned long) dst) & 0x3)
+	{
+	 case 0x00:			/* Buffer 32-bit aligned */
+		while (count--)
+		{
+			*(unsigned int *)p = cpu_to_le32(inl(port));
+			p += 4;
+		}
+		break;
+	
+	 case 0x02:			/* Buffer 16-bit aligned */
+		--count;
+		
+		l = cpu_to_le32(inl(port));
+		*(unsigned short *)p = l >> 16;
+		p += 2;
+		
+		while (count--)
+		{
+			l2 = cpu_to_le32(inl(port));
+			*(unsigned int *)p = (l & 0xffff) << 16 | (l2 >> 16);
+			p += 4;
+			l = l2;
+		}
+		*(unsigned short *)p = l & 0xffff;
+		break;
+	 case 0x01:			/* Buffer 8-bit aligned */
+		--count;
+		
+		l = cpu_to_le32(inl(port));
+		*(unsigned char *)p = l >> 24;
+		p++;
+		*(unsigned short *)p = (l >> 8) & 0xffff;
+		p += 2;
+		while (count--)
+		{
+			l2 = cpu_to_le32(inl(port));
+			*(unsigned int *)p = (l & 0xff) << 24 | (l2 >> 8);
+			p += 4;
+			l = l2;
+		}
+		*p = l & 0xff;
+		break;
+	 case 0x03:			/* Buffer 8-bit aligned */
+		--count;
+		
+		l = cpu_to_le32(inl(port));
+		*p = l >> 24;
+		p++;
+		while (count--)
+		{
+			l2 = cpu_to_le32(inl(port));
+			*(unsigned int *)p = (l & 0xffffff) << 8 | l2 >> 24;
+			p += 4;
+			l = l2;
+		}
+		*(unsigned short *)p = (l >> 8) & 0xffff;
+		p += 2;
+		*p = l & 0xff;
+		break;
+	}
+}
+
+
+/*
+ * Like insb but in the opposite direction.
+ * Don't worry as much about doing aligned memory transfers:
+ * doing byte reads the "slow" way isn't nearly as slow as
+ * doing byte writes the slow way (no r-m-w cycle).
+ */
+void outsb(unsigned long port, const void * src, unsigned long count)
+{
+	const unsigned char *p;
+
+	p = (const unsigned char *)src;
+	while (count) {
+		count--;
+		outb(*p, port);
+		p++;
+	}
+}
+
+/*
+ * Like insw but in the opposite direction.  This is used by the IDE
+ * driver to write disk sectors.  Performance is important, but the
+ * interfaces seems to be slow: just using the inlined version of the
+ * outw() breaks things.
+ */
+void outsw (unsigned long port, const void *src, unsigned long count)
+{
+	unsigned int l = 0, l2;
+	const unsigned char *p;
+
+	p = (const unsigned char *)src;
+	
+	if (!count)
+		return;
+	
+	switch (((unsigned long)p) & 0x3)
+	{
+	 case 0x00:			/* Buffer 32-bit aligned */
+		while (count>=2) {
+			count -= 2;
+			l = *(unsigned int *)p;
+			p += 4;
+			outw(le16_to_cpu(l >> 16), port);
+			outw(le16_to_cpu(l & 0xffff), port);
+		}
+		if (count) {
+			outw(le16_to_cpu(*(unsigned short*)p), port);
+		}
+		break;
+	
+	 case 0x02:			/* Buffer 16-bit aligned */
+		
+		outw(le16_to_cpu(*(unsigned short*)p), port);
+		p += 2;
+		count--;
+		
+		while (count>=2) {
+			count -= 2;
+			l = *(unsigned int *)p;
+			p += 4;
+			outw(le16_to_cpu(l >> 16), port);
+			outw(le16_to_cpu(l & 0xffff), port);
+		}
+		if (count) {
+			outw(le16_to_cpu(*(unsigned short *)p), port);
+		}
+		break;
+		
+	 case 0x01:			/* Buffer 8-bit aligned */	
+		/* I don't bother with 32bit transfers
+		 * in this case, 16bit will have to do -- DE */
+		
+		l  = *p << 8;
+		p++;
+		count--;
+		while (count)
+		{
+			count--;
+			l2 = *(unsigned short *)p;
+			p += 2;
+			outw(le16_to_cpu(l | l2 >> 8), port);
+		        l = l2 << 8;
+		}
+		l2 = *(unsigned char *)p;
+		outw (le16_to_cpu(l | l2>>8), port);
+		break;
+	
+	}
+}
+
+
+/*
+ * Like insl but in the opposite direction.  This is used by the IDE
+ * driver to write disk sectors.  Works with any alignment in SRC.
+ *  Performance is important, but the interfaces seems to be slow:
+ * just using the inlined version of the outl() breaks things.
+ */
+void outsl (unsigned long port, const void *src, unsigned long count)
+{
+	unsigned int l = 0, l2;
+	const unsigned char *p;
+
+	p = (const unsigned char *)src;
+	
+	if (!count)
+		return;
+	
+	switch (((unsigned long)p) & 0x3)
+	{
+	 case 0x00:			/* Buffer 32-bit aligned */
+		while (count--)
+		{
+			outl(le32_to_cpu(*(unsigned int *)p), port);
+			p += 4;
+		}
+		break;
+	
+	 case 0x02:			/* Buffer 16-bit aligned */
+		--count;
+		
+		l = *(unsigned short *)p;
+		p += 2;
+		
+		while (count--)
+		{
+			l2 = *(unsigned int *)p;
+			p += 4;
+			outl (le32_to_cpu(l << 16 | l2 >> 16), port);
+			l = l2;
+		}
+		l2 = *(unsigned short *)p;
+		outl (le32_to_cpu(l << 16 | l2), port);
+		break;
+	 case 0x01:			/* Buffer 8-bit aligned */
+		--count;
+
+		l = *p << 24;
+		p++;
+		l |= *(unsigned short *)p << 8;
+		p += 2;
+
+		while (count--)
+		{
+			l2 = *(unsigned int *)p;
+			p += 4;
+			outl (le32_to_cpu(l | l2 >> 24), port);
+			l = l2 << 8;
+		}
+		l2 = *p;
+		outl (le32_to_cpu(l | l2), port);
+		break;
+	 case 0x03:			/* Buffer 8-bit aligned */
+		--count;
+		
+		l = *p << 24;
+		p++;
+
+		while (count--)
+		{
+			l2 = *(unsigned int *)p;
+			p += 4;
+			outl (le32_to_cpu(l | l2 >> 8), port);
+			l = l2 << 24;
+		}
+		l2 = *(unsigned short *)p << 16;
+		p += 2;
+		l2 |= *p;
+		outl (le32_to_cpu(l | l2), port);
+		break;
+	}
+}
+
+EXPORT_SYMBOL(insb);
+EXPORT_SYMBOL(insw);
+EXPORT_SYMBOL(insl);
+EXPORT_SYMBOL(outsb);
+EXPORT_SYMBOL(outsw);
+EXPORT_SYMBOL(outsl);
diff --git a/arch/parisc/lib/iomap.c b/arch/parisc/lib/iomap.c
new file mode 100644
index 00000000000..290a62e7120
--- /dev/null
+++ b/arch/parisc/lib/iomap.c
@@ -0,0 +1,422 @@
+/*
+ * iomap.c - Implement iomap interface for PA-RISC
+ * Copyright (c) 2004 Matthew Wilcox
+ */
+
+#include <linux/ioport.h>
+#include <linux/pci.h>
+#include <asm/io.h>
+
+/*
+ * The iomap space on 32-bit PA-RISC is intended to look like this:
+ * 00000000-7fffffff virtual mapped IO
+ * 80000000-8fffffff ISA/EISA port space that can't be virtually mapped
+ * 90000000-9fffffff Dino port space