Add the rt linux 4.1.3-rt3 as base

Import the rt linux 4.1.3-rt3 as OPNFV kvm base.

It's from git://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-rt-devel.git linux-4.1.y-rt and
the base is:

commit 0917f823c59692d751951bf5ea699a2d1e2f26a2
Author: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Date:   Sat Jul 25 12:13:34 2015 +0200

    Prepare v4.1.3-rt3

    Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

We lose all the git history this way and it's not good. We
should apply another opnfv project repo in future.

Change-Id: I87543d81c9df70d99c5001fbdf646b202c19f423
Signed-off-by: Yunhong Jiang <yunhong.jiang@intel.com>

24 files changed, 5121 insertions, 0 deletions

diff --git a/kernel/arch/ia64/hp/common/Makefile b/kernel/arch/ia64/hp/common/Makefile
new file mode 100644
index 000000000..9e179dd06
--- /dev/null
+++ b/kernel/arch/ia64/hp/common/Makefile
@@ -0,0 +1,11 @@
+#
+# ia64/platform/hp/common/Makefile
+#
+# Copyright (C) 2002 Hewlett Packard
+# Copyright (C) Alex Williamson (alex_williamson@hp.com)
+#
+
+obj-y := sba_iommu.o
+obj-$(CONFIG_IA64_HP_ZX1_SWIOTLB) += hwsw_iommu.o
+obj-$(CONFIG_IA64_GENERIC) += hwsw_iommu.o
+obj-$(CONFIG_IA64_HP_AML_NFW) += aml_nfw.o
diff --git a/kernel/arch/ia64/hp/common/aml_nfw.c b/kernel/arch/ia64/hp/common/aml_nfw.c
new file mode 100644
index 000000000..84715fcbb
--- /dev/null
+++ b/kernel/arch/ia64/hp/common/aml_nfw.c
@@ -0,0 +1,235 @@
+/*
+ * OpRegion handler to allow AML to call native firmware
+ *
+ * (c) Copyright 2007 Hewlett-Packard Development Company, L.P.
+ *	Bjorn Helgaas <bjorn.helgaas@hp.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This driver implements HP Open Source Review Board proposal 1842,
+ * which was approved on 9/20/2006.
+ *
+ * For technical documentation, see the HP SPPA Firmware EAS, Appendix F.
+ *
+ * ACPI does not define a mechanism for AML methods to call native firmware
+ * interfaces such as PAL or SAL.  This OpRegion handler adds such a mechanism.
+ * After the handler is installed, an AML method can call native firmware by
+ * storing the arguments and firmware entry point to specific offsets in the
+ * OpRegion.  When AML reads the "return value" offset from the OpRegion, this
+ * handler loads up the arguments, makes the firmware call, and returns the
+ * result.
+ */
+
+#include <linux/module.h>
+#include <linux/acpi.h>
+#include <asm/sal.h>
+
+MODULE_AUTHOR("Bjorn Helgaas <bjorn.helgaas@hp.com>");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("ACPI opregion handler for native firmware calls");
+
+static bool force_register;
+module_param_named(force, force_register, bool, 0);
+MODULE_PARM_DESC(force, "Install opregion handler even without HPQ5001 device");
+
+#define AML_NFW_SPACE		0xA1
+
+struct ia64_pdesc {
+	void *ip;
+	void *gp;
+};
+
+/*
+ * N.B.  The layout of this structure is defined in the HP SPPA FW EAS, and
+ *	 the member offsets are embedded in AML methods.
+ */
+struct ia64_nfw_context {
+	u64 arg[8];
+	struct ia64_sal_retval ret;
+	u64 ip;
+	u64 gp;
+	u64 pad[2];
+};
+
+static void *virt_map(u64 address)
+{
+	if (address & (1UL << 63))
+		return (void *) (__IA64_UNCACHED_OFFSET | address);
+
+	return __va(address);
+}
+
+static void aml_nfw_execute(struct ia64_nfw_context *c)
+{
+	struct ia64_pdesc virt_entry;
+	ia64_sal_handler entry;
+
+	virt_entry.ip = virt_map(c->ip);
+	virt_entry.gp = virt_map(c->gp);
+
+	entry = (ia64_sal_handler) &virt_entry;
+
+	IA64_FW_CALL(entry, c->ret,
+		     c->arg[0], c->arg[1], c->arg[2], c->arg[3],
+		     c->arg[4], c->arg[5], c->arg[6], c->arg[7]);
+}
+
+static void aml_nfw_read_arg(u8 *offset, u32 bit_width, u64 *value)
+{
+	switch (bit_width) {
+	case 8:
+		*value = *(u8 *)offset;
+		break;
+	case 16:
+		*value = *(u16 *)offset;
+		break;
+	case 32:
+		*value = *(u32 *)offset;
+		break;
+	case 64:
+		*value = *(u64 *)offset;
+		break;
+	}
+}
+
+static void aml_nfw_write_arg(u8 *offset, u32 bit_width, u64 *value)
+{
+	switch (bit_width) {
+	case 8:
+		*(u8 *) offset = *value;
+		break;
+	case 16:
+		*(u16 *) offset = *value;
+		break;
+	case 32:
+		*(u32 *) offset = *value;
+		break;
+	case 64:
+		*(u64 *) offset = *value;
+		break;
+	}
+}
+
+static acpi_status aml_nfw_handler(u32 function, acpi_physical_address address,
+	u32 bit_width, u64 *value, void *handler_context,
+	void *region_context)
+{
+	struct ia64_nfw_context *context = handler_context;
+	u8 *offset = (u8 *) context + address;
+
+	if (bit_width !=  8 && bit_width != 16 &&
+	    bit_width != 32 && bit_width != 64)
+		return AE_BAD_PARAMETER;
+
+	if (address + (bit_width >> 3) > sizeof(struct ia64_nfw_context))
+		return AE_BAD_PARAMETER;
+
+	switch (function) {
+	case ACPI_READ:
+		if (address == offsetof(struct ia64_nfw_context, ret))
+			aml_nfw_execute(context);
+		aml_nfw_read_arg(offset, bit_width, value);
+		break;
+	case ACPI_WRITE:
+		aml_nfw_write_arg(offset, bit_width, value);
+		break;
+	}
+
+	return AE_OK;
+}
+
+static struct ia64_nfw_context global_context;
+static int global_handler_registered;
+
+static int aml_nfw_add_global_handler(void)
+{
+	acpi_status status;
+
+	if (global_handler_registered)
+		return 0;
+
+	status = acpi_install_address_space_handler(ACPI_ROOT_OBJECT,
+		AML_NFW_SPACE, aml_nfw_handler, NULL, &global_context);
+	if (ACPI_FAILURE(status))
+		return -ENODEV;
+
+	global_handler_registered = 1;
+	printk(KERN_INFO "Global 0x%02X opregion handler registered\n",
+		AML_NFW_SPACE);
+	return 0;
+}
+
+static int aml_nfw_remove_global_handler(void)
+{
+	acpi_status status;
+
+	if (!global_handler_registered)
+		return 0;
+
+	status = acpi_remove_address_space_handler(ACPI_ROOT_OBJECT,
+		AML_NFW_SPACE, aml_nfw_handler);
+	if (ACPI_FAILURE(status))
+		return -ENODEV;
+
+	global_handler_registered = 0;
+	printk(KERN_INFO "Global 0x%02X opregion handler removed\n",
+		AML_NFW_SPACE);
+	return 0;
+}
+
+static int aml_nfw_add(struct acpi_device *device)
+{
+	/*
+	 * We would normally allocate a new context structure and install
+	 * the address space handler for the specific device we found.
+	 * But the HP-UX implementation shares a single global context
+	 * and always puts the handler at the root, so we'll do the same.
+	 */
+	return aml_nfw_add_global_handler();
+}
+
+static int aml_nfw_remove(struct acpi_device *device)
+{
+	return aml_nfw_remove_global_handler();
+}
+
+static const struct acpi_device_id aml_nfw_ids[] = {
+	{"HPQ5001", 0},
+	{"", 0}
+};
+
+static struct acpi_driver acpi_aml_nfw_driver = {
+	.name = "native firmware",
+	.ids = aml_nfw_ids,
+	.ops = {
+		.add = aml_nfw_add,
+		.remove = aml_nfw_remove,
+		},
+};
+
+static int __init aml_nfw_init(void)
+{
+	int result;
+
+	if (force_register)
+		aml_nfw_add_global_handler();
+
+	result = acpi_bus_register_driver(&acpi_aml_nfw_driver);
+	if (result < 0) {
+		aml_nfw_remove_global_handler();
+		return result;
+	}
+
+	return 0;
+}
+
+static void __exit aml_nfw_exit(void)
+{
+	acpi_bus_unregister_driver(&acpi_aml_nfw_driver);
+	aml_nfw_remove_global_handler();
+}
+
+module_init(aml_nfw_init);
+module_exit(aml_nfw_exit);
diff --git a/kernel/arch/ia64/hp/common/hwsw_iommu.c b/kernel/arch/ia64/hp/common/hwsw_iommu.c
new file mode 100644
index 000000000..1e4cae5ae
--- /dev/null
+++ b/kernel/arch/ia64/hp/common/hwsw_iommu.c
@@ -0,0 +1,59 @@
+/*
+ * Copyright (c) 2004 Hewlett-Packard Development Company, L.P.
+ *   Contributed by David Mosberger-Tang <davidm@hpl.hp.com>
+ *
+ * This is a pseudo I/O MMU which dispatches to the hardware I/O MMU
+ * whenever possible.  We assume that the hardware I/O MMU requires
+ * full 32-bit addressability, as is the case, e.g., for HP zx1-based
+ * systems (there, the I/O MMU window is mapped at 3-4GB).  If a
+ * device doesn't provide full 32-bit addressability, we fall back on
+ * the sw I/O TLB.  This is good enough to let us support broken
+ * hardware such as soundcards which have a DMA engine that can
+ * address only 28 bits.
+ */
+
+#include <linux/device.h>
+#include <linux/dma-mapping.h>
+#include <linux/swiotlb.h>
+#include <linux/export.h>
+#include <asm/machvec.h>
+
+extern struct dma_map_ops sba_dma_ops, swiotlb_dma_ops;
+
+/* swiotlb declarations & definitions: */
+extern int swiotlb_late_init_with_default_size (size_t size);
+
+/*
+ * Note: we need to make the determination of whether or not to use
+ * the sw I/O TLB based purely on the device structure.  Anything else
+ * would be unreliable or would be too intrusive.
+ */
+static inline int use_swiotlb(struct device *dev)
+{
+	return dev && dev->dma_mask &&
+		!sba_dma_ops.dma_supported(dev, *dev->dma_mask);
+}
+
+struct dma_map_ops *hwsw_dma_get_ops(struct device *dev)
+{
+	if (use_swiotlb(dev))
+		return &swiotlb_dma_ops;
+	return &sba_dma_ops;
+}
+EXPORT_SYMBOL(hwsw_dma_get_ops);
+
+void __init
+hwsw_init (void)
+{
+	/* default to a smallish 2MB sw I/O TLB */
+	if (swiotlb_late_init_with_default_size (2 * (1<<20)) != 0) {
+#ifdef CONFIG_IA64_GENERIC
+		/* Better to have normal DMA than panic */
+		printk(KERN_WARNING "%s: Failed to initialize software I/O TLB,"
+		       " reverting to hpzx1 platform vector\n", __func__);
+		machvec_init("hpzx1");
+#else
+		panic("Unable to initialize software I/O TLB services");
+#endif
+	}
+}
diff --git a/kernel/arch/ia64/hp/common/sba_iommu.c b/kernel/arch/ia64/hp/common/sba_iommu.c
new file mode 100644
index 000000000..344387a55
--- /dev/null
+++ b/kernel/arch/ia64/hp/common/sba_iommu.c
@@ -0,0 +1,2241 @@
+/*
+**  IA64 System Bus Adapter (SBA) I/O MMU manager
+**
+**	(c) Copyright 2002-2005 Alex Williamson
+**	(c) Copyright 2002-2003 Grant Grundler
+**	(c) Copyright 2002-2005 Hewlett-Packard Company
+**
+**	Portions (c) 2000 Grant Grundler (from parisc I/O MMU code)
+**	Portions (c) 1999 Dave S. Miller (from sparc64 I/O MMU code)
+**
+**	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 module initializes the IOC (I/O Controller) found on HP
+** McKinley machines and their successors.
+**
+*/
+
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+#include <linux/init.h>
+#include <linux/mm.h>
+#include <linux/string.h>
+#include <linux/pci.h>
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
+#include <linux/acpi.h>
+#include <linux/efi.h>
+#include <linux/nodemask.h>
+#include <linux/bitops.h>         /* hweight64() */
+#include <linux/crash_dump.h>
+#include <linux/iommu-helper.h>
+#include <linux/dma-mapping.h>
+#include <linux/prefetch.h>
+
+#include <asm/delay.h>		/* ia64_get_itc() */
+#include <asm/io.h>
+#include <asm/page.h>		/* PAGE_OFFSET */
+#include <asm/dma.h>
+
+#include <asm/acpi-ext.h>
+
+extern int swiotlb_late_init_with_default_size (size_t size);
+
+#define PFX "IOC: "
+
+/*
+** Enabling timing search of the pdir resource map.  Output in /proc.
+** Disabled by default to optimize performance.
+*/
+#undef PDIR_SEARCH_TIMING
+
+/*
+** This option allows cards capable of 64bit DMA to bypass the IOMMU.  If
+** not defined, all DMA will be 32bit and go through the TLB.
+** There's potentially a conflict in the bio merge code with us
+** advertising an iommu, but then bypassing it.  Since I/O MMU bypassing
+** appears to give more performance than bio-level virtual merging, we'll
+** do the former for now.  NOTE: BYPASS_SG also needs to be undef'd to
+** completely restrict DMA to the IOMMU.
+*/
+#define ALLOW_IOV_BYPASS
+
+/*
+** This option specifically allows/disallows bypassing scatterlists with
+** multiple entries.  Coalescing these entries can allow better DMA streaming
+** and in some cases shows better performance than entirely bypassing the
+** IOMMU.  Performance increase on the order of 1-2% sequential output/input
+** using bonnie++ on a RAID0 MD device (sym2 & mpt).
+*/
+#undef ALLOW_IOV_BYPASS_SG
+
+/*
+** If a device prefetches beyond the end of a valid pdir entry, it will cause
+** a hard failure, ie. MCA.  Version 3.0 and later of the zx1 LBA should
+** disconnect on 4k boundaries and prevent such issues.  If the device is
+** particularly aggressive, this option will keep the entire pdir valid such
+** that prefetching will hit a valid address.  This could severely impact
+** error containment, and is therefore off by default.  The page that is
+** used for spill-over is poisoned, so that should help debugging somewhat.
+*/
+#undef FULL_VALID_PDIR
+
+#define ENABLE_MARK_CLEAN
+
+/*
+** The number of debug flags is a clue - this code is fragile.  NOTE: since
+** tightening the use of res_lock the resource bitmap and actual pdir are no
+** longer guaranteed to stay in sync.  The sanity checking code isn't going to
+** like that.
+*/
+#undef DEBUG_SBA_INIT
+#undef DEBUG_SBA_RUN
+#undef DEBUG_SBA_RUN_SG
+#undef DEBUG_SBA_RESOURCE
+#undef ASSERT_PDIR_SANITY
+#undef DEBUG_LARGE_SG_ENTRIES
+#undef DEBUG_BYPASS
+
+#if defined(FULL_VALID_PDIR) && defined(ASSERT_PDIR_SANITY)
+#error FULL_VALID_PDIR and ASSERT_PDIR_SANITY are mutually exclusive
+#endif
+
+#define SBA_INLINE	__inline__
+/* #define SBA_INLINE */
+
+#ifdef DEBUG_SBA_INIT
+#define DBG_INIT(x...)	printk(x)
+#else
+#define DBG_INIT(x...)
+#endif
+
+#ifdef DEBUG_SBA_RUN
+#define DBG_RUN(x...)	printk(x)
+#else
+#define DBG_RUN(x...)
+#endif
+
+#ifdef DEBUG_SBA_RUN_SG
+#define DBG_RUN_SG(x...)	printk(x)
+#else
+#define DBG_RUN_SG(x...)
+#endif
+
+
+#ifdef DEBUG_SBA_RESOURCE
+#define DBG_RES(x...)	printk(x)
+#else
+#define DBG_RES(x...)
+#endif
+
+#ifdef DEBUG_BYPASS
+#define DBG_BYPASS(x...)	printk(x)
+#else
+#define DBG_BYPASS(x...)
+#endif
+
+#ifdef ASSERT_PDIR_SANITY
+#define ASSERT(expr) \
+        if(!(expr)) { \
+                printk( "\n" __FILE__ ":%d: Assertion " #expr " failed!\n",__LINE__); \
+                panic(#expr); \
+        }
+#else
+#define ASSERT(expr)
+#endif
+
+/*
+** The number of pdir entries to "free" before issuing
+** a read to PCOM register to flush out PCOM writes.
+** Interacts with allocation granularity (ie 4 or 8 entries
+** allocated and free'd/purged at a time might make this
+** less interesting).
+*/
+#define DELAYED_RESOURCE_CNT	64
+
+#define PCI_DEVICE_ID_HP_SX2000_IOC	0x12ec
+
+#define ZX1_IOC_ID	((PCI_DEVICE_ID_HP_ZX1_IOC << 16) | PCI_VENDOR_ID_HP)
+#define ZX2_IOC_ID	((PCI_DEVICE_ID_HP_ZX2_IOC << 16) | PCI_VENDOR_ID_HP)
+#define REO_IOC_ID	((PCI_DEVICE_ID_HP_REO_IOC << 16) | PCI_VENDOR_ID_HP)
+#define SX1000_IOC_ID	((PCI_DEVICE_ID_HP_SX1000_IOC << 16) | PCI_VENDOR_ID_HP)
+#define SX2000_IOC_ID	((PCI_DEVICE_ID_HP_SX2000_IOC << 16) | PCI_VENDOR_ID_HP)
+
+#define ZX1_IOC_OFFSET	0x1000	/* ACPI reports SBA, we want IOC */
+
+#define IOC_FUNC_ID	0x000
+#define IOC_FCLASS	0x008	/* function class, bist, header, rev... */
+#define IOC_IBASE	0x300	/* IO TLB */
+#define IOC_IMASK	0x308
+#define IOC_PCOM	0x310
+#define IOC_TCNFG	0x318
+#define IOC_PDIR_BASE	0x320
+
+#define IOC_ROPE0_CFG	0x500
+#define   IOC_ROPE_AO	  0x10	/* Allow "Relaxed Ordering" */
+
+
+/* AGP GART driver looks for this */
+#define ZX1_SBA_IOMMU_COOKIE	0x0000badbadc0ffeeUL
+
+/*
+** The zx1 IOC supports 4/8/16/64KB page sizes (see TCNFG register)
+**
+** Some IOCs (sx1000) can run at the above pages sizes, but are
+** really only supported using the IOC at a 4k page size.
+**
+** iovp_size could only be greater than PAGE_SIZE if we are
+** confident the drivers really only touch the next physical
+** page iff that driver instance owns it.
+*/
+static unsigned long iovp_size;
+static unsigned long iovp_shift;
+static unsigned long iovp_mask;
+
+struct ioc {
+	void __iomem	*ioc_hpa;	/* I/O MMU base address */
+	char		*res_map;	/* resource map, bit == pdir entry */
+	u64		*pdir_base;	/* physical base address */
+	unsigned long	ibase;		/* pdir IOV Space base */
+	unsigned long	imask;		/* pdir IOV Space mask */
+
+	unsigned long	*res_hint;	/* next avail IOVP - circular search */
+	unsigned long	dma_mask;
+	spinlock_t	res_lock;	/* protects the resource bitmap, but must be held when */
+					/* clearing pdir to prevent races with allocations. */
+	unsigned int	res_bitshift;	/* from the RIGHT! */
+	unsigned int	res_size;	/* size of resource map in bytes */
+#ifdef CONFIG_NUMA
+	unsigned int	node;		/* node where this IOC lives */
+#endif
+#if DELAYED_RESOURCE_CNT > 0
+	spinlock_t	saved_lock;	/* may want to try to get this on a separate cacheline */
+					/* than res_lock for bigger systems. */
+	int		saved_cnt;
+	struct sba_dma_pair {
+		dma_addr_t	iova;
+		size_t		size;
+	} saved[DELAYED_RESOURCE_CNT];
+#endif
+
+#ifdef PDIR_SEARCH_TIMING
+#define SBA_SEARCH_SAMPLE	0x100
+	unsigned long avg_search[SBA_SEARCH_SAMPLE];
+	unsigned long avg_idx;	/* current index into avg_search */
+#endif
+
+	/* Stuff we don't need in performance path */
+	struct ioc	*next;		/* list of IOC's in system */
+	acpi_handle	handle;		/* for multiple IOC's */
+	const char 	*name;
+	unsigned int	func_id;
+	unsigned int	rev;		/* HW revision of chip */
+	u32		iov_size;
+	unsigned int	pdir_size;	/* in bytes, determined by IOV Space size */
+	struct pci_dev	*sac_only_dev;
+};
+
+static struct ioc *ioc_list, *ioc_found;
+static int reserve_sba_gart = 1;
+
+static SBA_INLINE void sba_mark_invalid(struct ioc *, dma_addr_t, size_t);
+static SBA_INLINE void sba_free_range(struct ioc *, dma_addr_t, size_t);
+
+#define sba_sg_address(sg)	sg_virt((sg))
+
+#ifdef FULL_VALID_PDIR
+static u64 prefetch_spill_page;
+#endif
+
+#ifdef CONFIG_PCI
+# define GET_IOC(dev)	((dev_is_pci(dev))						\
+			 ? ((struct ioc *) PCI_CONTROLLER(to_pci_dev(dev))->iommu) : NULL)
+#else
+# define GET_IOC(dev)	NULL
+#endif
+
+/*
+** DMA_CHUNK_SIZE is used by the SCSI mid-layer to break up
+** (or rather not merge) DMAs into manageable chunks.
+** On parisc, this is more of the software/tuning constraint
+** rather than the HW. I/O MMU allocation algorithms can be
+** faster with smaller sizes (to some degree).
+*/
+#define DMA_CHUNK_SIZE  (BITS_PER_LONG*iovp_size)
+
+#define ROUNDUP(x,y) ((x + ((y)-1)) & ~((y)-1))
+
+/************************************
+** SBA register read and write support
+**
+** BE WARNED: register writes are posted.
+**  (ie follow writes which must reach HW with a read)
+**
+*/
+#define READ_REG(addr)       __raw_readq(addr)
+#define WRITE_REG(val, addr) __raw_writeq(val, addr)
+
+#ifdef DEBUG_SBA_INIT
+
+/**
+ * sba_dump_tlb - debugging only - print IOMMU operating parameters
+ * @hpa: base address of the IOMMU
+ *
+ * Print the size/location of the IO MMU PDIR.
+ */
+static void
+sba_dump_tlb(char *hpa)
+{
+	DBG_INIT("IO TLB at 0x%p\n", (void *)hpa);
+	DBG_INIT("IOC_IBASE    : %016lx\n", READ_REG(hpa+IOC_IBASE));
+	DBG_INIT("IOC_IMASK    : %016lx\n", READ_REG(hpa+IOC_IMASK));
+	DBG_INIT("IOC_TCNFG    : %016lx\n", READ_REG(hpa+IOC_TCNFG));
+	DBG_INIT("IOC_PDIR_BASE: %016lx\n", READ_REG(hpa+IOC_PDIR_BASE));
+	DBG_INIT("\n");
+}
+#endif
+
+
+#ifdef ASSERT_PDIR_SANITY
+
+/**
+ * sba_dump_pdir_entry - debugging only - print one IOMMU PDIR entry
+ * @ioc: IO MMU structure which owns the pdir we are interested in.
+ * @msg: text to print ont the output line.
+ * @pide: pdir index.
+ *
+ * Print one entry of the IO MMU PDIR in human readable form.
+ */
+static void
+sba_dump_pdir_entry(struct ioc *ioc, char *msg, uint pide)
+{
+	/* start printing from lowest pde in rval */
+	u64 *ptr = &ioc->pdir_base[pide  & ~(BITS_PER_LONG - 1)];
+	unsigned long *rptr = (unsigned long *) &ioc->res_map[(pide >>3) & -sizeof(unsigned long)];
+	uint rcnt;
+
+	printk(KERN_DEBUG "SBA: %s rp %p bit %d rval 0x%lx\n",
+		 msg, rptr, pide & (BITS_PER_LONG - 1), *rptr);
+
+	rcnt = 0;
+	while (rcnt < BITS_PER_LONG) {
+		printk(KERN_DEBUG "%s %2d %p %016Lx\n",
+		       (rcnt == (pide & (BITS_PER_LONG - 1)))
+		       ? "    -->" : "       ",
+		       rcnt, ptr, (unsigned long long) *ptr );
+		rcnt++;
+		ptr++;
+	}
+	printk(KERN_DEBUG "%s", msg);
+}
+
+
+/**
+ * sba_check_pdir - debugging only - consistency checker
+ * @ioc: IO MMU structure which owns the pdir we are interested in.
+ * @msg: text to print ont the output line.
+ *
+ * Verify the resource map and pdir state is consistent
+ */
+static int
+sba_check_pdir(struct ioc *ioc, char *msg)
+{
+	u64 *rptr_end = (u64 *) &(ioc->res_map[ioc->res_size]);
+	u64 *rptr = (u64 *) ioc->res_map;	/* resource map ptr */
+	u64 *pptr = ioc->pdir_base;	/* pdir ptr */
+	uint pide = 0;
+
+	while (rptr < rptr_end) {
+		u64 rval;
+		int rcnt; /* number of bits we might check */
+
+		rval = *rptr;
+		rcnt = 64;
+
+		while (rcnt) {
+			/* Get last byte and highest bit from that */
+			u32 pde = ((u32)((*pptr >> (63)) & 0x1));
+			if ((rval & 0x1) ^ pde)
+			{
+				/*
+				** BUMMER!  -- res_map != pdir --
+				** Dump rval and matching pdir entries
+				*/
+				sba_dump_pdir_entry(ioc, msg, pide);
+				return(1);
+			}
+			rcnt--;
+			rval >>= 1;	/* try the next bit */
+			pptr++;
+			pide++;
+		}
+		rptr++;	/* look at next word of res_map */
+	}
+	/* It'd be nice if we always got here :^) */
+	return 0;
+}
+
+
+/**
+ * sba_dump_sg - debugging only - print Scatter-Gather list
+ * @ioc: IO MMU structure which owns the pdir we are interested in.
+ * @startsg: head of the SG list
+ * @nents: number of entries in SG list
+ *
+ * print the SG list so we can verify it's correct by hand.
+ */
+static void
+sba_dump_sg( struct ioc *ioc, struct scatterlist *startsg, int nents)
+{
+	while (nents-- > 0) {
+		printk(KERN_DEBUG " %d : DMA %08lx/%05x CPU %p\n", nents,
+		       startsg->dma_address, startsg->dma_length,
+		       sba_sg_address(startsg));
+		startsg = sg_next(startsg);
+	}
+}
+
+static void
+sba_check_sg( struct ioc *ioc, struct scatterlist *startsg, int nents)
+{
+	struct scatterlist *the_sg = startsg;
+	int the_nents = nents;
+
+	while (the_nents-- > 0) {
+		if (sba_sg_address(the_sg) == 0x0UL)
+			sba_dump_sg(NULL, startsg, nents);
+		the_sg = sg_next(the_sg);
+	}
+}
+
+#endif /* ASSERT_PDIR_SANITY */
+
+
+
+
+/**************************************************************
+*
+*   I/O Pdir Resource Management
+*
+*   Bits set in the resource map are in use.
+*   Each bit can represent a number of pages.
+*   LSbs represent lower addresses (IOVA's).
+*
+***************************************************************/
+#define PAGES_PER_RANGE 1	/* could increase this to 4 or 8 if needed */
+
+/* Convert from IOVP to IOVA and vice versa. */
+#define SBA_IOVA(ioc,iovp,offset) ((ioc->ibase) | (iovp) | (offset))
+#define SBA_IOVP(ioc,iova) ((iova) & ~(ioc->ibase))
+
+#define PDIR_ENTRY_SIZE	sizeof(u64)
+
+#define PDIR_INDEX(iovp)   ((iovp)>>iovp_shift)
+
+#define RESMAP_MASK(n)    ~(~0UL << (n))
+#define RESMAP_IDX_MASK   (sizeof(unsigned long) - 1)
+
+
+/**
+ * For most cases the normal get_order is sufficient, however it limits us
+ * to PAGE_SIZE being the minimum mapping alignment and TC flush granularity.
+ * It only incurs about 1 clock cycle to use this one with the static variable
+ * and makes the code more intuitive.
+ */
+static SBA_INLINE int
+get_iovp_order (unsigned long size)
+{
+	long double d = size - 1;
+	long order;
+
+	order = ia64_getf_exp(d);
+	order = order - iovp_shift - 0xffff + 1;
+	if (order < 0)
+		order = 0;
+	return order;
+}
+
+static unsigned long ptr_to_pide(struct ioc *ioc, unsigned long *res_ptr,
+				 unsigned int bitshiftcnt)
+{
+	return (((unsigned long)res_ptr - (unsigned long)ioc->res_map) << 3)
+		+ bitshiftcnt;
+}
+
+/**
+ * sba_search_bitmap - find free space in IO PDIR resource bitmap
+ * @ioc: IO MMU structure which owns the pdir we are interested in.
+ * @bits_wanted: number of entries we need.
+ * @use_hint: use res_hint to indicate where to start looking
+ *
+ * Find consecutive free bits in resource bitmap.
+ * Each bit represents one entry in the IO Pdir.
+ * Cool perf optimization: search for log2(size) bits at a time.
+ */
+static SBA_INLINE unsigned long
+sba_search_bitmap(struct ioc *ioc, struct device *dev,
+		  unsigned long bits_wanted, int use_hint)
+{
+	unsigned long *res_ptr;
+	unsigned long *res_end = (unsigned long *) &(ioc->res_map[ioc->res_size]);
+	unsigned long flags, pide = ~0UL, tpide;
+	unsigned long boundary_size;
+	unsigned long shift;
+	int ret;
+
+	ASSERT(((unsigned long) ioc->res_hint & (sizeof(unsigned long) - 1UL)) == 0);
+	ASSERT(res_ptr < res_end);
+
+	boundary_size = (unsigned long long)dma_get_seg_boundary(dev) + 1;
+	boundary_size = ALIGN(boundary_size, 1ULL << iovp_shift) >> iovp_shift;
+
+	BUG_ON(ioc->ibase & ~iovp_mask);
+	shift = ioc->ibase >> iovp_shift;
+
+	spin_lock_irqsave(&ioc->res_lock, flags);
+
+	/* Allow caller to force a search through the entire resource space */
+	if (likely(use_hint)) {
+		res_ptr = ioc->res_hint;
+	} else {
+		res_ptr = (ulong *)ioc->res_map;
+		ioc->res_bitshift = 0;
+	}
+
+	/*
+	 * N.B.  REO/Grande defect AR2305 can cause TLB fetch timeouts
+	 * if a TLB entry is purged while in use.  sba_mark_invalid()
+	 * purges IOTLB entries in power-of-two sizes, so we also
+	 * allocate IOVA space in power-of-two sizes.
+	 */
+	bits_wanted = 1UL << get_iovp_order(bits_wanted << iovp_shift);
+
+	if (likely(bits_wanted == 1)) {
+		unsigned int bitshiftcnt;
+		for(; res_ptr < res_end ; res_ptr++) {
+			if (likely(*res_ptr != ~0UL)) {
+				bitshiftcnt = ffz(*res_ptr);
+				*res_ptr |= (1UL << bitshiftcnt);
+				pide = ptr_to_pide(ioc, res_ptr, bitshiftcnt);
+				ioc->res_bitshift = bitshiftcnt + bits_wanted;
+				goto found_it;
+			}
+		}
+		goto not_found;
+
+	}
+	
+	if (likely(bits_wanted <= BITS_PER_LONG/2)) {
+		/*
+		** Search the resource bit map on well-aligned values.
+		** "o" is the alignment.
+		** We need the alignment to invalidate I/O TLB using
+		** SBA HW features in the unmap path.
+		*/
+		unsigned long o = 1 << get_iovp_order(bits_wanted << iovp_shift);
+		uint bitshiftcnt = ROUNDUP(ioc->res_bitshift, o);
+		unsigned long mask, base_mask;
+
+		base_mask = RESMAP_MASK(bits_wanted);
+		mask = base_mask << bitshiftcnt;
+
+		DBG_RES("%s() o %ld %p", __func__, o, res_ptr);
+		for(; res_ptr < res_end ; res_ptr++)
+		{ 
+			DBG_RES("    %p %lx %lx\n", res_ptr, mask, *res_ptr);
+			ASSERT(0 != mask);
+			for (; mask ; mask <<= o, bitshiftcnt += o) {
+				tpide = ptr_to_pide(ioc, res_ptr, bitshiftcnt);
+				ret = iommu_is_span_boundary(tpide, bits_wanted,
+							     shift,
+							     boundary_size);
+				if ((0 == ((*res_ptr) & mask)) && !ret) {
+					*res_ptr |= mask;     /* mark resources busy! */
+					pide = tpide;
+					ioc->res_bitshift = bitshiftcnt + bits_wanted;
+					goto found_it;
+				}
+			}
+
+			bitshiftcnt = 0;
+			mask = base_mask;
+
+		}
+
+	} else {
+		int qwords, bits, i;
+		unsigned long *end;
+
+		qwords = bits_wanted >> 6; /* /64 */
+		bits = bits_wanted - (qwords * BITS_PER_LONG);
+
+		end = res_end - qwords;
+
+		for (; res_ptr < end; res_ptr++) {
+			tpide = ptr_to_pide(ioc, res_ptr, 0);
+			ret = iommu_is_span_boundary(tpide, bits_wanted,
+						     shift, boundary_size);
+			if (ret)
+				goto next_ptr;
+			for (i = 0 ; i < qwords ; i++) {
+				if (res_ptr[i] != 0)
+					goto next_ptr;
+			}
+			if (bits && res_ptr[i] && (__ffs(res_ptr[i]) < bits))
+				continue;
+
+			/* Found it, mark it */
+			for (i = 0 ; i < qwords ; i++)
+				res_ptr[i] = ~0UL;
+			res_ptr[i] |= RESMAP_MASK(bits);
+
+			pide = tpide;
+			res_ptr += qwords;
+			ioc->res_bitshift = bits;
+			goto found_it;
+next_ptr:
+			;
+		}
+	}
+
+not_found:
+	prefetch(ioc->res_map);
+	ioc->res_hint = (unsigned long *) ioc->res_map;
+	ioc->res_bitshift = 0;
+	spin_unlock_irqrestore(&ioc->res_lock, flags);
+	return (pide);
+
+found_it:
+	ioc->res_hint = res_ptr;
+	spin_unlock_irqrestore(&ioc->res_lock, flags);
+	return (pide);
+}
+
+
+/**
+ * sba_alloc_range - find free bits and mark them in IO PDIR resource bitmap
+ * @ioc: IO MMU structure which owns the pdir we are interested in.
+ * @size: number of bytes to create a mapping for
+ *
+ * Given a size, find consecutive unmarked and then mark those bits in the
+ * resource bit map.
+ */
+static int
+sba_alloc_range(struct ioc *ioc, struct device *dev, size_t size)
+{
+	unsigned int pages_needed = size >> iovp_shift;
+#ifdef PDIR_SEARCH_TIMING
+	unsigned long itc_start;
+#endif
+	unsigned long pide;
+
+	ASSERT(pages_needed);
+	ASSERT(0 == (size & ~iovp_mask));
+
+#ifdef PDIR_SEARCH_TIMING
+	itc_start = ia64_get_itc();
+#endif
+	/*
+	** "seek and ye shall find"...praying never hurts either...
+	*/
+	pide = sba_search_bitmap(ioc, dev, pages_needed, 1);
+	if (unlikely(pide >= (ioc->res_size << 3))) {
+		pide = sba_search_bitmap(ioc, dev, pages_needed, 0);
+		if (unlikely(pide >= (ioc->res_size << 3))) {
+#if DELAYED_RESOURCE_CNT > 0
+			unsigned long flags;
+
+			/*
+			** With delayed resource freeing, we can give this one more shot.  We're
+			** getting close to being in trouble here, so do what we can to make this
+			** one count.
+			*/
+			spin_lock_irqsave(&ioc->saved_lock, flags);
+			if (ioc->saved_cnt > 0) {
+				struct sba_dma_pair *d;
+				int cnt = ioc->saved_cnt;
+
+				d = &(ioc->saved[ioc->saved_cnt - 1]);
+
+				spin_lock(&ioc->res_lock);
+				while (cnt--) {
+					sba_mark_invalid(ioc, d->iova, d->size);
+					sba_free_range(ioc, d->iova, d->size);
+					d--;
+				}
+				ioc->saved_cnt = 0;
+				READ_REG(ioc->ioc_hpa+IOC_PCOM);	/* flush purges */
+				spin_unlock(&ioc->res_lock);
+			}
+			spin_unlock_irqrestore(&ioc->saved_lock, flags);
+
+			pide = sba_search_bitmap(ioc, dev, pages_needed, 0);
+			if (unlikely(pide >= (ioc->res_size << 3))) {
+				printk(KERN_WARNING "%s: I/O MMU @ %p is"
+				       "out of mapping resources, %u %u %lx\n",
+				       __func__, ioc->ioc_hpa, ioc->res_size,
+				       pages_needed, dma_get_seg_boundary(dev));
+				return -1;
+			}
+#else
+			printk(KERN_WARNING "%s: I/O MMU @ %p is"
+			       "out of mapping resources, %u %u %lx\n",
+			       __func__, ioc->ioc_hpa, ioc->res_size,
+			       pages_needed, dma_get_seg_boundary(dev));
+			return -1;
+#endif
+		}
+	}
+
+#ifdef PDIR_SEARCH_TIMING
+	ioc->avg_search[ioc->avg_idx++] = (ia64_get_itc() - itc_start) / pages_needed;
+	ioc->avg_idx &= SBA_SEARCH_SAMPLE - 1;
+#endif
+
+	prefetchw(&(ioc->pdir_base[pide]));
+
+#ifdef ASSERT_PDIR_SANITY
+	/* verify the first enable bit is clear */
+	if(0x00 != ((u8 *) ioc->pdir_base)[pide*PDIR_ENTRY_SIZE + 7]) {
+		sba_dump_pdir_entry(ioc, "sba_search_bitmap() botched it?", pide);
+	}
+#endif
+
+	DBG_RES("%s(%x) %d -> %lx hint %x/%x\n",
+		__func__, size, pages_needed, pide,
+		(uint) ((unsigned long) ioc->res_hint - (unsigned long) ioc->res_map),
+		ioc->res_bitshift );
+
+	return (pide);
+}
+
+
+/**
+ * sba_free_range - unmark bits in IO PDIR resource bitmap
+ * @ioc: IO MMU structure which owns the pdir we are interested in.
+ * @iova: IO virtual address which was previously allocated.
+ * @size: number of bytes to create a mapping for
+ *
+ * clear bits in the ioc's resource map
+ */
+static SBA_INLINE void
+sba_free_range(struct ioc *ioc, dma_addr_t iova, size_t size)
+{
+	unsigned long iovp = SBA_IOVP(ioc, iova);
+	unsigned int pide = PDIR_INDEX(iovp);
+	unsigned int ridx = pide >> 3;	/* convert bit to byte address */
+	unsigned long *res_ptr = (unsigned long *) &((ioc)->res_map[ridx & ~RESMAP_IDX_MASK]);
+	int bits_not_wanted = size >> iovp_shift;
+	unsigned long m;
+
+	/* Round up to power-of-two size: see AR2305 note above */
+	bits_not_wanted = 1UL << get_iovp_order(bits_not_wanted << iovp_shift);
+	for (; bits_not_wanted > 0 ; res_ptr++) {
+		
+		if (unlikely(bits_not_wanted > BITS_PER_LONG)) {
+
+			/* these mappings start 64bit aligned */
+			*res_ptr = 0UL;
+			bits_not_wanted -= BITS_PER_LONG;
+			pide += BITS_PER_LONG;
+
+		} else {
+
+			/* 3-bits "bit" address plus 2 (or 3) bits for "byte" == bit in word */
+			m = RESMAP_MASK(bits_not_wanted) << (pide & (BITS_PER_LONG - 1));
+			bits_not_wanted = 0;
+
+			DBG_RES("%s( ,%x,%x) %x/%lx %x %p %lx\n", __func__, (uint) iova, size,
+			        bits_not_wanted, m, pide, res_ptr, *res_ptr);
+
+			ASSERT(m != 0);
+			ASSERT(bits_not_wanted);
+			ASSERT((*res_ptr & m) == m); /* verify same bits are set */
+			*res_ptr &= ~m;
+		}
+	}
+}
+
+
+/**************************************************************
+*
+*   "Dynamic DMA Mapping" support (aka "Coherent I/O")
+*
+***************************************************************/
+
+/**
+ * sba_io_pdir_entry - fill in one IO PDIR entry
+ * @pdir_ptr:  pointer to IO PDIR entry
+ * @vba: Virtual CPU address of buffer to map
+ *
+ * SBA Mapping Routine
+ *
+ * Given a virtual address (vba, arg1) sba_io_pdir_entry()
+ * loads the I/O PDIR entry pointed to by pdir_ptr (arg0).
+ * Each IO Pdir entry consists of 8 bytes as shown below
+ * (LSB == bit 0):
+ *
+ *  63                    40                                 11    7        0
+ * +-+---------------------+----------------------------------+----+--------+
+ * |V|        U            |            PPN[39:12]            | U  |   FF   |
+ * +-+---------------------+----------------------------------+----+--------+
+ *
+ *  V  == Valid Bit
+ *  U  == Unused
+ * PPN == Physical Page Number
+ *
+ * The physical address fields are filled with the results of virt_to_phys()
+ * on the vba.
+ */
+
+#if 1
+#define sba_io_pdir_entry(pdir_ptr, vba) *pdir_ptr = ((vba & ~0xE000000000000FFFULL)	\
+						      | 0x8000000000000000ULL)
+#else
+void SBA_INLINE
+sba_io_pdir_entry(u64 *pdir_ptr, unsigned long vba)
+{
+	*pdir_ptr = ((vba & ~0xE000000000000FFFULL) | 0x80000000000000FFULL);
+}
+#endif
+
+#ifdef ENABLE_MARK_CLEAN
+/**
+ * Since DMA is i-cache coherent, any (complete) pages that were written via
+ * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
+ * flush them when they get mapped into an executable vm-area.
+ */
+static void
+mark_clean (void *addr, size_t size)
+{
+	unsigned long pg_addr, end;
+
+	pg_addr = PAGE_ALIGN((unsigned long) addr);
+	end = (unsigned long) addr + size;
+	while (pg_addr + PAGE_SIZE <= end) {
+		struct page *page = virt_to_page((void *)pg_addr);
+		set_bit(PG_arch_1, &page->flags);
+		pg_addr += PAGE_SIZE;
+	}
+}
+#endif
+
+/**
+ * sba_mark_invalid - invalidate one or more IO PDIR entries
+ * @ioc: IO MMU structure which owns the pdir we are interested in.
+ * @iova:  IO Virtual Address mapped earlier
+ * @byte_cnt:  number of bytes this mapping covers.
+ *
+ * Marking the IO PDIR entry(ies) as Invalid and invalidate
+ * corresponding IO TLB entry. The PCOM (Purge Command Register)
+ * is to purge stale entries in the IO TLB when unmapping entries.
+ *
+ * The PCOM register supports purging of multiple pages, with a minium
+ * of 1 page and a maximum of 2GB. Hardware requires the address be
+ * aligned to the size of the range being purged. The size of the range
+ * must be a power of 2. The "Cool perf optimization" in the
+ * allocation routine helps keep that true.
+ */
+static SBA_INLINE void
+sba_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt)
+{
+	u32 iovp = (u32) SBA_IOVP(ioc,iova);
+
+	int off = PDIR_INDEX(iovp);
+
+	/* Must be non-zero and rounded up */
+	ASSERT(byte_cnt > 0);
+	ASSERT(0 == (byte_cnt & ~iovp_mask));
+
+#ifdef ASSERT_PDIR_SANITY
+	/* Assert first pdir entry is set */
+	if (!(ioc->pdir_base[off] >> 60)) {
+		sba_dump_pdir_entry(ioc,"sba_mark_invalid()", PDIR_INDEX(iovp));
+	}
+#endif
+
+	if (byte_cnt <= iovp_size)
+	{
+		ASSERT(off < ioc->pdir_size);
+
+		iovp |= iovp_shift;     /* set "size" field for PCOM */
+
+#ifndef FULL_VALID_PDIR
+		/*
+		** clear I/O PDIR entry "valid" bit
+		** Do NOT clear the rest - save it for debugging.
+		** We should only clear bits that have previously
+		** been enabled.
+		*/
+		ioc->pdir_base[off] &= ~(0x80000000000000FFULL);
+#else
+		/*
+  		** If we want to maintain the PDIR as valid, put in
+		** the spill page so devices prefetching won't
+		** cause a hard fail.
+		*/
+		ioc->pdir_base[off] = (0x80000000000000FFULL | prefetch_spill_page);
+#endif
+	} else {
+		u32 t = get_iovp_order(byte_cnt) + iovp_shift;
+
+		iovp |= t;
+		ASSERT(t <= 31);   /* 2GB! Max value of "size" field */
+
+		do {
+			/* verify this pdir entry is enabled */
+			ASSERT(ioc->pdir_base[off]  >> 63);
+#ifndef FULL_VALID_PDIR
+			/* clear I/O Pdir entry "valid" bit first */
+			ioc->pdir_base[off] &= ~(0x80000000000000FFULL);
+#else
+			ioc->pdir_base[off] = (0x80000000000000FFULL | prefetch_spill_page);
+#endif
+			off++;
+			byte_cnt -= iovp_size;
+		} while (byte_cnt > 0);
+	}
+
+	WRITE_REG(iovp | ioc->ibase, ioc->ioc_hpa+IOC_PCOM);
+}
+
+/**
+ * sba_map_single_attrs - map one buffer and return IOVA for DMA
+ * @dev: instance of PCI owned by the driver that's asking.
+ * @addr:  driver buffer to map.
+ * @size:  number of bytes to map in driver buffer.
+ * @dir:  R/W or both.
+ * @attrs: optional dma attributes
+ *
+ * See Documentation/DMA-API-HOWTO.txt
+ */
+static dma_addr_t sba_map_page(struct device *dev, struct page *page,
+			       unsigned long poff, size_t size,
+			       enum dma_data_direction dir,
+			       struct dma_attrs *attrs)
+{
+	struct ioc *ioc;
+	void *addr = page_address(page) + poff;
+	dma_addr_t iovp;
+	dma_addr_t offset;
+	u64 *pdir_start;
+	int pide;
+#ifdef ASSERT_PDIR_SANITY
+	unsigned long flags;
+#endif
+#ifdef ALLOW_IOV_BYPASS
+	unsigned long pci_addr = virt_to_phys(addr);
+#endif
+
+#ifdef ALLOW_IOV_BYPASS
+	ASSERT(to_pci_dev(dev)->dma_mask);
+	/*
+ 	** Check if the PCI device can DMA to ptr... if so, just return ptr
+ 	*/
+	if (likely((pci_addr & ~to_pci_dev(dev)->dma_mask) == 0)) {
+		/*
+ 		** Device is bit capable of DMA'ing to the buffer...
+		** just return the PCI address of ptr
+ 		*/
+		DBG_BYPASS("sba_map_single_attrs() bypass mask/addr: "
+			   "0x%lx/0x%lx\n",
+		           to_pci_dev(dev)->dma_mask, pci_addr);
+		return pci_addr;
+	}
+#endif
+	ioc = GET_IOC(dev);
+	ASSERT(ioc);
+
+	prefetch(ioc->res_hint);
+
+	ASSERT(size > 0);
+	ASSERT(size <= DMA_CHUNK_SIZE);
+
+	/* save offset bits */
+	offset = ((dma_addr_t) (long) addr) & ~iovp_mask;
+
+	/* round up to nearest iovp_size */
+	size = (size + offset + ~iovp_mask) & iovp_mask;
+
+#ifdef ASSERT_PDIR_SANITY
+	spin_lock_irqsave(&ioc->res_lock, flags);
+	if (sba_check_pdir(ioc,"Check before sba_map_single_attrs()"))
+		panic("Sanity check failed");
+	spin_unlock_irqrestore(&ioc->res_lock, flags);
+#endif
+
+	pide = sba_alloc_range(ioc, dev, size);
+	if (pide < 0)
+		return 0;
+
+	iovp = (dma_addr_t) pide << iovp_shift;
+
+	DBG_RUN("%s() 0x%p -> 0x%lx\n", __func__, addr, (long) iovp | offset);
+
+	pdir_start = &(ioc->pdir_base[pide]);
+
+	while (size > 0) {
+		ASSERT(((u8 *)pdir_start)[7] == 0); /* verify availability */
+		sba_io_pdir_entry(pdir_start, (unsigned long) addr);
+
+		DBG_RUN("     pdir 0x%p %lx\n", pdir_start, *pdir_start);
+
+		addr += iovp_size;
+		size -= iovp_size;
+		pdir_start++;
+	}
+	/* force pdir update */
+	wmb();
+
+	/* form complete address */
+#ifdef ASSERT_PDIR_SANITY
+	spin_lock_irqsave(&ioc->res_lock, flags);
+	sba_check_pdir(ioc,"Check after sba_map_single_attrs()");
+	spin_unlock_irqrestore(&ioc->res_lock, flags);
+#endif
+	return SBA_IOVA(ioc, iovp, offset);
+}
+
+static dma_addr_t sba_map_single_attrs(struct device *dev, void *addr,
+				       size_t size, enum dma_data_direction dir,
+				       struct dma_attrs *attrs)
+{
+	return sba_map_page(dev, virt_to_page(addr),
+			    (unsigned long)addr & ~PAGE_MASK, size, dir, attrs);
+}
+
+#ifdef ENABLE_MARK_CLEAN
+static SBA_INLINE void
+sba_mark_clean(struct ioc *ioc, dma_addr_t iova, size_t size)
+{
+	u32	iovp = (u32) SBA_IOVP(ioc,iova);
+	int	off = PDIR_INDEX(iovp);
+	void	*addr;
+
+	if (size <= iovp_size) {
+		addr = phys_to_virt(ioc->pdir_base[off] &
+		                    ~0xE000000000000FFFULL);
+		mark_clean(addr, size);
+	} else {
+		do {
+			addr = phys_to_virt(ioc->pdir_base[off] &
+			                    ~0xE000000000000FFFULL);
+			mark_clean(addr, min(size, iovp_size));
+			off++;
+			size -= iovp_size;
+		} while (size > 0);
+	}
+}
+#endif
+
+/**
+ * sba_unmap_single_attrs - unmap one IOVA and free resources
+ * @dev: instance of PCI owned by the driver that's asking.
+ * @iova:  IOVA of driver buffer previously mapped.
+ * @size:  number of bytes mapped in driver buffer.
+ * @dir:  R/W or both.
+ * @attrs: optional dma attributes
+ *
+ * See Documentation/DMA-API-HOWTO.txt
+ */
+static void sba_unmap_page(struct device *dev, dma_addr_t iova, size_t size,
+			   enum dma_data_direction dir, struct dma_attrs *attrs)
+{
+	struct ioc *ioc;
+#if DELAYED_RESOURCE_CNT > 0
+	struct sba_dma_pair *d;
+#endif
+	unsigned long flags;
+	dma_addr_t offset;
+
+	ioc = GET_IOC(dev);
+	ASSERT(ioc);
+
+#ifdef ALLOW_IOV_BYPASS
+	if (likely((iova & ioc->imask) != ioc->ibase)) {
+		/*
+		** Address does not fall w/in IOVA, must be bypassing
+		*/
+		DBG_BYPASS("sba_unmap_single_attrs() bypass addr: 0x%lx\n",
+			   iova);
+
+#ifdef ENABLE_MARK_CLEAN
+		if (dir == DMA_FROM_DEVICE) {
+			mark_clean(phys_to_virt(iova), size);
+		}
+#endif
+		return;
+	}
+#endif
+	offset = iova & ~iovp_mask;
+
+	DBG_RUN("%s() iovp 0x%lx/%x\n", __func__, (long) iova, size);
+
+	iova ^= offset;        /* clear offset bits */
+	size += offset;
+	size = ROUNDUP(size, iovp_size);
+
+#ifdef ENABLE_MARK_CLEAN
+	if (dir == DMA_FROM_DEVICE)
+		sba_mark_clean(ioc, iova, size);
+#endif
+
+#if DELAYED_RESOURCE_CNT > 0
+	spin_lock_irqsave(&ioc->saved_lock, flags);
+	d = &(ioc->saved[ioc->saved_cnt]);
+	d->iova = iova;
+	d->size = size;
+	if (unlikely(++(ioc->saved_cnt) >= DELAYED_RESOURCE_CNT)) {
+		int cnt = ioc->saved_cnt;
+		spin_lock(&ioc->res_lock);
+		while (cnt--) {
+			sba_mark_invalid(ioc, d->iova, d->size);
+			sba_free_range(ioc, d->iova, d->size);
+			d--;
+		}
+		ioc->saved_cnt = 0;
+		READ_REG(ioc->ioc_hpa+IOC_PCOM);	/* flush purges */
+		spin_unlock(&ioc->res_lock);
+	}
+	spin_unlock_irqrestore(&ioc->saved_lock, flags);
+#else /* DELAYED_RESOURCE_CNT == 0 */
+	spin_lock_irqsave(&ioc->res_lock, flags);
+	sba_mark_invalid(ioc, iova, size);
+	sba_free_range(ioc, iova, size);
+	READ_REG(ioc->ioc_hpa+IOC_PCOM);	/* flush purges */
+	spin_unlock_irqrestore(&ioc->res_lock, flags);
+#endif /* DELAYED_RESOURCE_CNT == 0 */
+}
+
+void sba_unmap_single_attrs(struct device *dev, dma_addr_t iova, size_t size,
+			    enum dma_data_direction dir, struct dma_attrs *attrs)
+{
+	sba_unmap_page(dev, iova, size, dir, attrs);
+}
+
+/**
+ * sba_alloc_coherent - allocate/map shared mem for DMA
+ * @dev: instance of PCI owned by the driver that's asking.
+ * @size:  number of bytes mapped in driver buffer.
+ * @dma_handle:  IOVA of new buffer.
+ *
+ * See Documentation/DMA-API-HOWTO.txt
+ */
+static void *
+sba_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle,
+		   gfp_t flags, struct dma_attrs *attrs)
+{
+	struct ioc *ioc;
+	void *addr;
+
+	ioc = GET_IOC(dev);
+	ASSERT(ioc);
+
+#ifdef CONFIG_NUMA
+	{
+		int node = ioc->node;
+		struct page *page;
+
+		if (node == NUMA_NO_NODE)
+			node = numa_node_id();
+
+		page = alloc_pages_exact_node(node, flags, get_order(size));
+		if (unlikely(!page))
+			return NULL;
+
+		addr = page_address(page);
+	}
+#else
+	addr = (void *) __get_free_pages(flags, get_order(size));
+#endif
+	if (unlikely(!addr))
+		return NULL;
+
+	memset(addr, 0, size);
+	*dma_handle = virt_to_phys(addr);
+
+#ifdef ALLOW_IOV_BYPASS
+	ASSERT(dev->coherent_dma_mask);
+	/*
+ 	** Check if the PCI device can DMA to ptr... if so, just return ptr
+ 	*/
+	if (likely((*dma_handle & ~dev->coherent_dma_mask) == 0)) {
+		DBG_BYPASS("sba_alloc_coherent() bypass mask/addr: 0x%lx/0x%lx\n",
+		           dev->coherent_dma_mask, *dma_handle);
+
+		return addr;
+	}
+#endif
+
+	/*
+	 * If device can't bypass or bypass is disabled, pass the 32bit fake
+	 * device to map single to get an iova mapping.
+	 */
+	*dma_handle = sba_map_single_attrs(&ioc->sac_only_dev->dev, addr,
+					   size, 0, NULL);
+
+	return addr;
+}
+
+
+/**
+ * sba_free_coherent - free/unmap shared mem for DMA
+ * @dev: instance of PCI owned by the driver that's asking.
+ * @size:  number of bytes mapped in driver buffer.
+ * @vaddr:  virtual address IOVA of "consistent" buffer.
+ * @dma_handler:  IO virtual address of "consistent" buffer.
+ *
+ * See Documentation/DMA-API-HOWTO.txt
+ */
+static void sba_free_coherent(struct device *dev, size_t size, void *vaddr,
+			      dma_addr_t dma_handle, struct dma_attrs *attrs)
+{
+	sba_unmap_single_attrs(dev, dma_handle, size, 0, NULL);
+	free_pages((unsigned long) vaddr, get_order(size));
+}
+
+
+/*
+** Since 0 is a valid pdir_base index value, can't use that
+** to determine if a value is valid or not. Use a flag to indicate
+** the SG list entry contains a valid pdir index.
+*/
+#define PIDE_FLAG 0x1UL
+
+#ifdef DEBUG_LARGE_SG_ENTRIES
+int dump_run_sg = 0;
+#endif
+
+
+/**
+ * sba_fill_pdir - write allocated SG entries into IO PDIR
+ * @ioc: IO MMU structure which owns the pdir we are interested in.
+ * @startsg:  list of IOVA/size pairs
+ * @nents: number of entries in startsg list
+ *
+ * Take preprocessed SG list and write corresponding entries
+ * in the IO PDIR.
+ */
+
+static SBA_INLINE int
+sba_fill_pdir(
+	struct ioc *ioc,
+	struct scatterlist *startsg,
+	int nents)
+{
+	struct scatterlist *dma_sg = startsg;	/* pointer to current DMA */
+	int n_mappings = 0;
+	u64 *pdirp = NULL;
+	unsigned long dma_offset = 0;
+
+	while (nents-- > 0) {
+		int     cnt = startsg->dma_length;
+		startsg->dma_length = 0;
+
+#ifdef DEBUG_LARGE_SG_ENTRIES
+		if (dump_run_sg)
+			printk(" %2d : %08lx/%05x %p\n",
+				nents, startsg->dma_address, cnt,
+				sba_sg_address(startsg));
+#else
+		DBG_RUN_SG(" %d : %08lx/%05x %p\n",
+				nents, startsg->dma_address, cnt,
+				sba_sg_address(startsg));
+#endif
+		/*
+		** Look for the start of a new DMA stream
+		*/
+		if (startsg->dma_address & PIDE_FLAG) {
+			u32 pide = startsg->dma_address & ~PIDE_FLAG;
+			dma_offset = (unsigned long) pide & ~iovp_mask;
+			startsg->dma_address = 0;
+			if (n_mappings)
+				dma_sg = sg_next(dma_sg);
+			dma_sg->dma_address = pide | ioc->ibase;
+			pdirp = &(ioc->pdir_base[pide >> iovp_shift]);
+			n_mappings++;
+		}
+
+		/*
+		** Look for a VCONTIG chunk
+		*/
+		if (cnt) {
+			unsigned long vaddr = (unsigned long) sba_sg_address(startsg);
+			ASSERT(pdirp);
+
+			/* Since multiple Vcontig blocks could make up
+			** one DMA stream, *add* cnt to dma_len.
+			*/
+			dma_sg->dma_length += cnt;
+			cnt += dma_offset;
+			dma_offset=0;	/* only want offset on first chunk */
+			cnt = ROUNDUP(cnt, iovp_size);
+			do {
+				sba_io_pdir_entry(pdirp, vaddr);
+				vaddr += iovp_size;
+				cnt -= iovp_size;
+				pdirp++;
+			} while (cnt > 0);
+		}
+		startsg = sg_next(startsg);
+	}
+	/* force pdir update */
+	wmb();
+
+#ifdef DEBUG_LARGE_SG_ENTRIES
+	dump_run_sg = 0;
+#endif
+	return(n_mappings);
+}
+
+
+/*
+** Two address ranges are DMA contiguous *iff* "end of prev" and
+** "start of next" are both on an IOV page boundary.
+**
+** (shift left is a quick trick to mask off upper bits)
+*/
+#define DMA_CONTIG(__X, __Y) \
+	(((((unsigned long) __X) | ((unsigned long) __Y)) << (BITS_PER_LONG - iovp_shift)) == 0UL)
+
+
+/**
+ * sba_coalesce_chunks - preprocess the SG list
+ * @ioc: IO MMU structure which owns the pdir we are interested in.
+ * @startsg:  list of IOVA/size pairs
+ * @nents: number of entries in startsg list
+ *
+ * First pass is to walk the SG list and determine where the breaks are
+ * in the DMA stream. Allocates PDIR entries but does not fill them.
+ * Returns the number of DMA chunks.
+ *
+ * Doing the fill separate from the coalescing/allocation keeps the
+ * code simpler. Future enhancement could make one pass through
+ * the sglist do both.
+ */
+static SBA_INLINE int
+sba_coalesce_chunks(struct ioc *ioc, struct device *dev,
+	struct scatterlist *startsg,
+	int nents)
+{
+	struct scatterlist *vcontig_sg;    /* VCONTIG chunk head */
+	unsigned long vcontig_len;         /* len of VCONTIG chunk */
+	unsigned long vcontig_end;
+	struct scatterlist *dma_sg;        /* next DMA stream head */
+	unsigned long dma_offset, dma_len; /* start/len of DMA stream */
+	int n_mappings = 0;
+	unsigned int max_seg_size = dma_get_max_seg_size(dev);
+	int idx;
+
+	while (nents > 0) {
+		unsigned long vaddr = (unsigned long) sba_sg_address(startsg);
+
+		/*
+		** Prepare for first/next DMA stream
+		*/
+		dma_sg = vcontig_sg = startsg;
+		dma_len = vcontig_len = vcontig_end = startsg->length;
+		vcontig_end +=  vaddr;
+		dma_offset = vaddr & ~iovp_mask;
+
+		/* PARANOID: clear entries */
+		startsg->dma_address = startsg->dma_length = 0;
+
+		/*
+		** This loop terminates one iteration "early" since
+		** it's always looking one "ahead".
+		*/
+		while (--nents > 0) {
+			unsigned long vaddr;	/* tmp */
+
+			startsg = sg_next(startsg);
+
+			/* PARANOID */
+			startsg->dma_address = startsg->dma_length = 0;
+
+			/* catch brokenness in SCSI layer */
+			ASSERT(startsg->length <= DMA_CHUNK_SIZE);
+
+			/*
+			** First make sure current dma stream won't
+			** exceed DMA_CHUNK_SIZE if we coalesce the
+			** next entry.
+			*/
+			if (((dma_len + dma_offset + startsg->length + ~iovp_mask) & iovp_mask)
+			    > DMA_CHUNK_SIZE)
+				break;
+
+			if (dma_len + startsg->length > max_seg_size)
+				break;
+
+			/*
+			** Then look for virtually contiguous blocks.
+			**
+			** append the next transaction?
+			*/
+			vaddr = (unsigned long) sba_sg_address(startsg);
+			if  (vcontig_end == vaddr)
+			{
+				vcontig_len += startsg->length;
+				vcontig_end += startsg->length;
+				dma_len     += startsg->length;
+				continue;
+			}
+
+#ifdef DEBUG_LARGE_SG_ENTRIES
+			dump_run_sg = (vcontig_len > iovp_size);
+#endif
+
+			/*
+			** Not virtually contiguous.
+			** Terminate prev chunk.
+			** Start a new chunk.
+			**
+			** Once we start a new VCONTIG chunk, dma_offset
+			** can't change. And we need the offset from the first
+			** chunk - not the last one. Ergo Successive chunks
+			** must start on page boundaries and dove tail
+			** with it's predecessor.
+			*/
+			vcontig_sg->dma_length = vcontig_len;
+
+			vcontig_sg = startsg;
+			vcontig_len = startsg->length;
+
+			/*
+			** 3) do the entries end/start on page boundaries?
+			**    Don't update vcontig_end until we've checked.
+			*/
+			if (DMA_CONTIG(vcontig_end, vaddr))
+			{
+				vcontig_end = vcontig_len + vaddr;
+				dma_len += vcontig_len;
+				continue;
+			} else {
+				break;
+			}
+		}
+
+		/*
+		** End of DMA Stream
+		** Terminate last VCONTIG block.
+		** Allocate space for DMA stream.
+		*/
+		vcontig_sg->dma_length = vcontig_len;
+		dma_len = (dma_len + dma_offset + ~iovp_mask) & iovp_mask;
+		ASSERT(dma_len <= DMA_CHUNK_SIZE);
+		idx = sba_alloc_range(ioc, dev, dma_len);
+		if (idx < 0) {
+			dma_sg->dma_length = 0;
+			return -1;
+		}
+		dma_sg->dma_address = (dma_addr_t)(PIDE_FLAG | (idx << iovp_shift)
+						   | dma_offset);
+		n_mappings++;
+	}
+
+	return n_mappings;
+}
+
+static void sba_unmap_sg_attrs(struct device *dev, struct scatterlist *sglist,
+			       int nents, enum dma_data_direction dir,
+			       struct dma_attrs *attrs);
+/**
+ * sba_map_sg - map Scatter/Gather list
+ * @dev: instance of PCI owned by the driver that's asking.
+ * @sglist:  array of buffer/length pairs
+ * @nents:  number of entries in list
+ * @dir:  R/W or both.
+ * @attrs: optional dma attributes
+ *
+ * See Documentation/DMA-API-HOWTO.txt
+ */
+static int sba_map_sg_attrs(struct device *dev, struct scatterlist *sglist,
+			    int nents, enum dma_data_direction dir,
+			    struct dma_attrs *attrs)
+{
+	struct ioc *ioc;
+	int coalesced, filled = 0;
+#ifdef ASSERT_PDIR_SANITY
+	unsigned long flags;
+#endif
+#ifdef ALLOW_IOV_BYPASS_SG
+	struct scatterlist *sg;
+#endif
+
+	DBG_RUN_SG("%s() START %d entries\n", __func__, nents);
+	ioc = GET_IOC(dev);
+	ASSERT(ioc);
+
+#ifdef ALLOW_IOV_BYPASS_SG
+	ASSERT(to_pci_dev(dev)->dma_mask);
+	if (likely((ioc->dma_mask & ~to_pci_dev(dev)->dma_mask) == 0)) {
+		for_each_sg(sglist, sg, nents, filled) {
+			sg->dma_length = sg->length;
+			sg->dma_address = virt_to_phys(sba_sg_address(sg));
+		}
+		return filled;
+	}
+#endif
+	/* Fast path single entry scatterlists. */
+	if (nents == 1) {
+		sglist->dma_length = sglist->length;
+		sglist->dma_address = sba_map_single_attrs(dev, sba_sg_address(sglist), sglist->length, dir, attrs);
+		return 1;
+	}
+
+#ifdef ASSERT_PDIR_SANITY
+	spin_lock_irqsave(&ioc->res_lock, flags);
+	if (sba_check_pdir(ioc,"Check before sba_map_sg_attrs()"))
+	{
+		sba_dump_sg(ioc, sglist, nents);
+		panic("Check before sba_map_sg_attrs()");
+	}
+	spin_unlock_irqrestore(&ioc->res_lock, flags);
+#endif
+
+	prefetch(ioc->res_hint);
+
+	/*
+	** First coalesce the chunks and allocate I/O pdir space
+	**
+	** If this is one DMA stream, we can properly map using the
+	** correct virtual address associated with each DMA page.
+	** w/o this association, we wouldn't have coherent DMA!
+	** Access to the virtual address is what forces a two pass algorithm.
+	*/
+	coalesced = sba_coalesce_chunks(ioc, dev, sglist, nents);
+	if (coalesced < 0) {
+		sba_unmap_sg_attrs(dev, sglist, nents, dir, attrs);
+		return 0;
+	}
+
+	/*
+	** Program the I/O Pdir
+	**
+	** map the virtual addresses to the I/O Pdir
+	** o dma_address will contain the pdir index
+	** o dma_len will contain the number of bytes to map
+	** o address contains the virtual address.
+	*/
+	filled = sba_fill_pdir(ioc, sglist, nents);
+
+#ifdef ASSERT_PDIR_SANITY
+	spin_lock_irqsave(&ioc->res_lock, flags);
+	if (sba_check_pdir(ioc,"Check after sba_map_sg_attrs()"))
+	{
+		sba_dump_sg(ioc, sglist, nents);
+		panic("Check after sba_map_sg_attrs()\n");
+	}
+	spin_unlock_irqrestore(&ioc->res_lock, flags);
+#endif
+
+	ASSERT(coalesced == filled);
+	DBG_RUN_SG("%s() DONE %d mappings\n", __func__, filled);
+
+	return filled;
+}
+
+/**
+ * sba_unmap_sg_attrs - unmap Scatter/Gather list
+ * @dev: instance of PCI owned by the driver that's asking.
+ * @sglist:  array of buffer/length pairs
+ * @nents:  number of entries in list
+ * @dir:  R/W or both.
+ * @attrs: optional dma attributes
+ *
+ * See Documentation/DMA-API-HOWTO.txt
+ */
+static void sba_unmap_sg_attrs(struct device *dev, struct scatterlist *sglist,
+			       int nents, enum dma_data_direction dir,
+			       struct dma_attrs *attrs)
+{
+#ifdef ASSERT_PDIR_SANITY
+	struct ioc *ioc;
+	unsigned long flags;
+#endif
+
+	DBG_RUN_SG("%s() START %d entries,  %p,%x\n",
+		   __func__, nents, sba_sg_address(sglist), sglist->length);
+
+#ifdef ASSERT_PDIR_SANITY
+	ioc = GET_IOC(dev);
+	ASSERT(ioc);
+
+	spin_lock_irqsave(&ioc->res_lock, flags);
+	sba_check_pdir(ioc,"Check before sba_unmap_sg_attrs()");
+	spin_unlock_irqrestore(&ioc->res_lock, flags);
+#endif
+
+	while (nents && sglist->dma_length) {
+
+		sba_unmap_single_attrs(dev, sglist->dma_address,
+				       sglist->dma_length, dir, attrs);
+		sglist = sg_next(sglist);
+		nents--;
+	}
+
+	DBG_RUN_SG("%s() DONE (nents %d)\n", __func__,  nents);
+
+#ifdef ASSERT_PDIR_SANITY
+	spin_lock_irqsave(&ioc->res_lock, flags);
+	sba_check_pdir(ioc,"Check after sba_unmap_sg_attrs()");
+	spin_unlock_irqrestore(&ioc->res_lock, flags);
+#endif
+
+}
+
+/**************************************************************
+*
+*   Initialization and claim
+*
+***************************************************************/
+
+static void
+ioc_iova_init(struct ioc *ioc)
+{
+	int tcnfg;
+	int agp_found = 0;
+	struct pci_dev *device = NULL;
+#ifdef FULL_VALID_PDIR
+	unsigned long index;
+#endif
+
+	/*
+	** Firmware programs the base and size of a "safe IOVA space"
+	** (one that doesn't overlap memory or LMMIO space) in the
+	** IBASE and IMASK registers.
+	*/
+	ioc->ibase = READ_REG(ioc->ioc_hpa + IOC_IBASE) & ~0x1UL;
+	ioc->imask = READ_REG(ioc->ioc_hpa + IOC_IMASK) | 0xFFFFFFFF00000000UL;
+
+	ioc->iov_size = ~ioc->imask + 1;
+
+	DBG_INIT("%s() hpa %p IOV base 0x%lx mask 0x%lx (%dMB)\n",
+		__func__, ioc->ioc_hpa, ioc->ibase, ioc->imask,
+		ioc->iov_size >> 20);
+
+	switch (iovp_size) {
+		case  4*1024: tcnfg = 0; break;
+		case  8*1024: tcnfg = 1; break;
+		case 16*1024: tcnfg = 2; break;
+		case 64*1024: tcnfg = 3; break;
+		default:
+			panic(PFX "Unsupported IOTLB page size %ldK",
+				iovp_size >> 10);
+			break;
+	}
+	WRITE_REG(tcnfg, ioc->ioc_hpa + IOC_TCNFG);
+
+	ioc->pdir_size = (ioc->iov_size / iovp_size) * PDIR_ENTRY_SIZE;
+	ioc->pdir_base = (void *) __get_free_pages(GFP_KERNEL,
+						   get_order(ioc->pdir_size));
+	if (!ioc->pdir_base)
+		panic(PFX "Couldn't allocate I/O Page Table\n");
+
+	memset(ioc->pdir_base, 0, ioc->pdir_size);
+
+	DBG_INIT("%s() IOV page size %ldK pdir %p size %x\n", __func__,
+		iovp_size >> 10, ioc->pdir_base, ioc->pdir_size);
+
+	ASSERT(ALIGN((unsigned long) ioc->pdir_base, 4*1024) == (unsigned long) ioc->pdir_base);
+	WRITE_REG(virt_to_phys(ioc->pdir_base), ioc->ioc_hpa + IOC_PDIR_BASE);
+
+	/*
+	** If an AGP device is present, only use half of the IOV space
+	** for PCI DMA.  Unfortunately we can't know ahead of time
+	** whether GART support will actually be used, for now we
+	** can just key on an AGP device found in the system.
+	** We program the next pdir index after we stop w/ a key for
+	** the GART code to handshake on.
+	*/
+	for_each_pci_dev(device)	
+		agp_found |= pci_find_capability(device, PCI_CAP_ID_AGP);
+
+	if (agp_found && reserve_sba_gart) {
+		printk(KERN_INFO PFX "reserving %dMb of IOVA space at 0x%lx for agpgart\n",
+		      ioc->iov_size/2 >> 20, ioc->ibase + ioc->iov_size/2);
+		ioc->pdir_size /= 2;
+		((u64 *)ioc->pdir_base)[PDIR_INDEX(ioc->iov_size/2)] = ZX1_SBA_IOMMU_COOKIE;
+	}
+#ifdef FULL_VALID_PDIR
+	/*
+  	** Check to see if the spill page has been allocated, we don't need more than
+	** one across multiple SBAs.
+	*/
+	if (!prefetch_spill_page) {
+		char *spill_poison = "SBAIOMMU POISON";
+		int poison_size = 16;
+		void *poison_addr, *addr;
+
+		addr = (void *)__get_free_pages(GFP_KERNEL, get_order(iovp_size));
+		if (!addr)
+			panic(PFX "Couldn't allocate PDIR spill page\n");
+
+		poison_addr = addr;
+		for ( ; (u64) poison_addr < addr + iovp_size; poison_addr += poison_size)
+			memcpy(poison_addr, spill_poison, poison_size);
+
+		prefetch_spill_page = virt_to_phys(addr);
+
+		DBG_INIT("%s() prefetch spill addr: 0x%lx\n", __func__, prefetch_spill_page);
+	}
+	/*
+  	** Set all the PDIR entries valid w/ the spill page as the target
+	*/
+	for (index = 0 ; index < (ioc->pdir_size / PDIR_ENTRY_SIZE) ; index++)
+		((u64 *)ioc->pdir_base)[index] = (0x80000000000000FF | prefetch_spill_page);
+#endif
+
+	/* Clear I/O TLB of any possible entries */
+	WRITE_REG(ioc->ibase | (get_iovp_order(ioc->iov_size) + iovp_shift), ioc->ioc_hpa + IOC_PCOM);
+	READ_REG(ioc->ioc_hpa + IOC_PCOM);
+
+	/* Enable IOVA translation */
+	WRITE_REG(ioc->ibase | 1, ioc->ioc_hpa + IOC_IBASE);
+	READ_REG(ioc->ioc_hpa + IOC_IBASE);
+}
+
+static void __init
+ioc_resource_init(struct ioc *ioc)
+{
+	spin_lock_init(&ioc->res_lock);
+#if DELAYED_RESOURCE_CNT > 0
+	spin_lock_init(&ioc->saved_lock);
+#endif
+
+	/* resource map size dictated by pdir_size */
+	ioc->res_size = ioc->pdir_size / PDIR_ENTRY_SIZE; /* entries */
+	ioc->res_size >>= 3;  /* convert bit count to byte count */
+	DBG_INIT("%s() res_size 0x%x\n", __func__, ioc->res_size);
+
+	ioc->res_map = (char *) __get_free_pages(GFP_KERNEL,
+						 get_order(ioc->res_size));
+	if (!ioc->res_map)
+		panic(PFX "Couldn't allocate resource map\n");
+
+	memset(ioc->res_map, 0, ioc->res_size);
+	/* next available IOVP - circular search */
+	ioc->res_hint = (unsigned long *) ioc->res_map;
+
+#ifdef ASSERT_PDIR_SANITY
+	/* Mark first bit busy - ie no IOVA 0 */
+	ioc->res_map[0] = 0x1;
+	ioc->pdir_base[0] = 0x8000000000000000ULL | ZX1_SBA_IOMMU_COOKIE;
+#endif
+#ifdef FULL_VALID_PDIR
+	/* Mark the last resource used so we don't prefetch beyond IOVA space */
+	ioc->res_map[ioc->res_size - 1] |= 0x80UL; /* res_map is chars */
+	ioc->pdir_base[(ioc->pdir_size / PDIR_ENTRY_SIZE) - 1] = (0x80000000000000FF
+							      | prefetch_spill_page);
+#endif
+
+	DBG_INIT("%s() res_map %x %p\n", __func__,
+		 ioc->res_size, (void *) ioc->res_map);
+}
+
+static void __init
+ioc_sac_init(struct ioc *ioc)
+{
+	struct pci_dev *sac = NULL;
+	struct pci_controller *controller = NULL;
+
+	/*
+	 * pci_alloc_coherent() must return a DMA address which is
+	 * SAC (single address cycle) addressable, so allocate a
+	 * pseudo-device to enforce that.
+	 */
+	sac = kzalloc(sizeof(*sac), GFP_KERNEL);
+	if (!sac)
+		panic(PFX "Couldn't allocate struct pci_dev");
+
+	controller = kzalloc(sizeof(*controller), GFP_KERNEL);
+	if (!controller)
+		panic(PFX "Couldn't allocate struct pci_controller");
+
+	controller->iommu = ioc;
+	sac->sysdata = controller;
+	sac->dma_mask = 0xFFFFFFFFUL;
+#ifdef CONFIG_PCI
+	sac->dev.bus = &pci_bus_type;
+#endif
+	ioc->sac_only_dev = sac;
+}
+
+static void __init
+ioc_zx1_init(struct ioc *ioc)
+{
+	unsigned long rope_config;
+	unsigned int i;
+
+	if (ioc->rev < 0x20)
+		panic(PFX "IOC 2.0 or later required for IOMMU support\n");
+
+	/* 38 bit memory controller + extra bit for range displaced by MMIO */
+	ioc->dma_mask = (0x1UL << 39) - 1;
+
+	/*
+	** Clear ROPE(N)_CONFIG AO bit.
+	** Disables "NT Ordering" (~= !"Relaxed Ordering")
+	** Overrides bit 1 in DMA Hint Sets.
+	** Improves netperf UDP_STREAM by ~10% for tg3 on bcm5701.
+	*/
+	for (i=0; i<(8*8); i+=8) {
+		rope_config = READ_REG(ioc->ioc_hpa + IOC_ROPE0_CFG + i);
+		rope_config &= ~IOC_ROPE_AO;
+		WRITE_REG(rope_config, ioc->ioc_hpa + IOC_ROPE0_CFG + i);
+	}
+}
+
+typedef void (initfunc)(struct ioc *);
+
+struct ioc_iommu {
+	u32 func_id;
+	char *name;
+	initfunc *init;
+};
+
+static struct ioc_iommu ioc_iommu_info[] __initdata = {
+	{ ZX1_IOC_ID, "zx1", ioc_zx1_init },
+	{ ZX2_IOC_ID, "zx2", NULL },
+	{ SX1000_IOC_ID, "sx1000", NULL },
+	{ SX2000_IOC_ID, "sx2000", NULL },
+};
+
+static void ioc_init(unsigned long hpa, struct ioc *ioc)
+{
+	struct ioc_iommu *info;
+
+	ioc->next = ioc_list;
+	ioc_list = ioc;
+
+	ioc->ioc_hpa = ioremap(hpa, 0x1000);
+
+	ioc->func_id = READ_REG(ioc->ioc_hpa + IOC_FUNC_ID);
+	ioc->rev = READ_REG(ioc->ioc_hpa + IOC_FCLASS) & 0xFFUL;
+	ioc->dma_mask = 0xFFFFFFFFFFFFFFFFUL;	/* conservative */
+
+	for (info = ioc_iommu_info; info < ioc_iommu_info + ARRAY_SIZE(ioc_iommu_info); info++) {
+		if (ioc->func_id == info->func_id) {
+			ioc->name = info->name;
+			if (info->init)
+				(info->init)(ioc);
+		}
+	}
+
+	iovp_size = (1 << iovp_shift);
+	iovp_mask = ~(iovp_size - 1);
+
+	DBG_INIT("%s: PAGE_SIZE %ldK, iovp_size %ldK\n", __func__,
+		PAGE_SIZE >> 10, iovp_size >> 10);
+
+	if (!ioc->name) {
+		ioc->name = kmalloc(24, GFP_KERNEL);
+		if (ioc->name)
+			sprintf((char *) ioc->name, "Unknown (%04x:%04x)",
+				ioc->func_id & 0xFFFF, (ioc->func_id >> 16) & 0xFFFF);
+		else
+			ioc->name = "Unknown";
+	}
+
+	ioc_iova_init(ioc);
+	ioc_resource_init(ioc);
+	ioc_sac_init(ioc);
+
+	if ((long) ~iovp_mask > (long) ia64_max_iommu_merge_mask)
+		ia64_max_iommu_merge_mask = ~iovp_mask;
+
+	printk(KERN_INFO PFX
+		"%s %d.%d HPA 0x%lx IOVA space %dMb at 0x%lx\n",
+		ioc->name, (ioc->rev >> 4) & 0xF, ioc->rev & 0xF,
+		hpa, ioc->iov_size >> 20, ioc->ibase);
+}
+
+
+
+/**************************************************************************
+**
+**   SBA initialization code (HW and SW)
+**
+**   o identify SBA chip itself
+**   o FIXME: initialize DMA hints for reasonable defaults
+**
+**************************************************************************/
+
+#ifdef CONFIG_PROC_FS
+static void *
+ioc_start(struct seq_file *s, loff_t *pos)
+{
+	struct ioc *ioc;
+	loff_t n = *pos;
+
+	for (ioc = ioc_list; ioc; ioc = ioc->next)
+		if (!n--)
+			return ioc;
+
+	return NULL;
+}
+
+static void *
+ioc_next(struct seq_file *s, void *v, loff_t *pos)
+{
+	struct ioc *ioc = v;
+
+	++*pos;
+	return ioc->next;
+}
+
+static void
+ioc_stop(struct seq_file *s, void *v)
+{
+}
+
+static int
+ioc_show(struct seq_file *s, void *v)
+{
+	struct ioc *ioc = v;
+	unsigned long *res_ptr = (unsigned long *)ioc->res_map;
+	int i, used = 0;
+
+	seq_printf(s, "Hewlett Packard %s IOC rev %d.%d\n",
+		ioc->name, ((ioc->rev >> 4) & 0xF), (ioc->rev & 0xF));
+#ifdef CONFIG_NUMA
+	if (ioc->node != NUMA_NO_NODE)
+		seq_printf(s, "NUMA node       : %d\n", ioc->node);
+#endif
+	seq_printf(s, "IOVA size       : %ld MB\n", ((ioc->pdir_size >> 3) * iovp_size)/(1024*1024));
+	seq_printf(s, "IOVA page size  : %ld kb\n", iovp_size/1024);
+
+	for (i = 0; i < (ioc->res_size / sizeof(unsigned long)); ++i, ++res_ptr)
+		used += hweight64(*res_ptr);
+
+	seq_printf(s, "PDIR size       : %d entries\n", ioc->pdir_size >> 3);
+	seq_printf(s, "PDIR used       : %d entries\n", used);
+
+#ifdef PDIR_SEARCH_TIMING
+	{
+		unsigned long i = 0, avg = 0, min, max;
+		min = max = ioc->avg_search[0];
+		for (i = 0; i < SBA_SEARCH_SAMPLE; i++) {
+			avg += ioc->avg_search[i];
+			if (ioc->avg_search[i] > max) max = ioc->avg_search[i];
+			if (ioc->avg_search[i] < min) min = ioc->avg_search[i];
+		}
+		avg /= SBA_SEARCH_SAMPLE;
+		seq_printf(s, "Bitmap search   : %ld/%ld/%ld (min/avg/max CPU Cycles/IOVA page)\n",
+		           min, avg, max);
+	}
+#endif
+#ifndef ALLOW_IOV_BYPASS
+	 seq_printf(s, "IOVA bypass disabled\n");
+#endif
+	return 0;
+}
+
+static const struct seq_operations ioc_seq_ops = {
+	.start = ioc_start,
+	.next  = ioc_next,
+	.stop  = ioc_stop,
+	.show  = ioc_show
+};
+
+static int
+ioc_open(struct inode *inode, struct file *file)
+{
+	return seq_open(file, &ioc_seq_ops);
+}
+
+static const struct file_operations ioc_fops = {
+	.open    = ioc_open,
+	.read    = seq_read,
+	.llseek  = seq_lseek,
+	.release = seq_release
+};
+
+static void __init
+ioc_proc_init(void)
+{
+	struct proc_dir_entry *dir;
+
+	dir = proc_mkdir("bus/mckinley", NULL);
+	if (!dir)
+		return;
+
+	proc_create(ioc_list->name, 0, dir, &ioc_fops);
+}
+#endif
+
+static void
+sba_connect_bus(struct pci_bus *bus)
+{
+	acpi_handle handle, parent;
+	acpi_status status;
+	struct ioc *ioc;
+
+	if (!PCI_CONTROLLER(bus))
+		panic(PFX "no sysdata on bus %d!\n", bus->number);
+
+	if (PCI_CONTROLLER(bus)->iommu)
+		return;
+
+	handle = acpi_device_handle(PCI_CONTROLLER(bus)->companion);
+	if (!handle)
+		return;
+
+	/*
+	 * The IOC scope encloses PCI root bridges in the ACPI
+	 * namespace, so work our way out until we find an IOC we
+	 * claimed previously.
+	 */
+	do {
+		for (ioc = ioc_list; ioc; ioc = ioc->next)
+			if (ioc->handle == handle) {
+				PCI_CONTROLLER(bus)->iommu = ioc;
+				return;
+			}
+
+		status = acpi_get_parent(handle, &parent);
+		handle = parent;
+	} while (ACPI_SUCCESS(status));
+
+	printk(KERN_WARNING "No IOC for PCI Bus %04x:%02x in ACPI\n", pci_domain_nr(bus), bus->number);
+}
+
+static void __init
+sba_map_ioc_to_node(struct ioc *ioc, acpi_handle handle)
+{
+#ifdef CONFIG_NUMA
+	unsigned int node;
+
+	node = acpi_get_node(handle);
+	if (node != NUMA_NO_NODE && !node_online(node))
+		node = NUMA_NO_NODE;
+
+	ioc->node = node;
+#endif
+}
+
+static void acpi_sba_ioc_add(struct ioc *ioc)
+{
+	acpi_handle handle = ioc->handle;
+	acpi_status status;
+	u64 hpa, length;
+	struct acpi_device_info *adi;
+
+	ioc_found = ioc->next;
+	status = hp_acpi_csr_space(handle, &hpa, &length);
+	if (ACPI_FAILURE(status))
+		goto err;
+
+	status = acpi_get_object_info(handle, &adi);
+	if (ACPI_FAILURE(status))
+		goto err;
+
+	/*
+	 * For HWP0001, only SBA appears in ACPI namespace.  It encloses the PCI
+	 * root bridges, and its CSR space includes the IOC function.
+	 */
+	if (strncmp("HWP0001", adi->hardware_id.string, 7) == 0) {
+		hpa += ZX1_IOC_OFFSET;
+		/* zx1 based systems default to kernel page size iommu pages */
+		if (!iovp_shift)
+			iovp_shift = min(PAGE_SHIFT, 16);
+	}
+	kfree(adi);
+
+	/*
+	 * default anything not caught above or specified on cmdline to 4k
+	 * iommu page size
+	 */
+	if (!iovp_shift)
+		iovp_shift = 12;
+
+	ioc_init(hpa, ioc);
+	/* setup NUMA node association */
+	sba_map_ioc_to_node(ioc, handle);
+	return;
+
+ err:
+	kfree(ioc);
+}
+
+static const struct acpi_device_id hp_ioc_iommu_device_ids[] = {
+	{"HWP0001", 0},
+	{"HWP0004", 0},
+	{"", 0},
+};
+
+static int acpi_sba_ioc_attach(struct acpi_device *device,
+			       const struct acpi_device_id *not_used)
+{
+	struct ioc *ioc;
+
+	ioc = kzalloc(sizeof(*ioc), GFP_KERNEL);
+	if (!ioc)
+		return -ENOMEM;
+
+	ioc->next = ioc_found;
+	ioc_found = ioc;
+	ioc->handle = device->handle;
+	return 1;
+}
+
+
+static struct acpi_scan_handler acpi_sba_ioc_handler = {
+	.ids	= hp_ioc_iommu_device_ids,
+	.attach	= acpi_sba_ioc_attach,
+};
+
+static int __init acpi_sba_ioc_init_acpi(void)
+{
+	return acpi_scan_add_handler(&acpi_sba_ioc_handler);
+}
+/* This has to run before acpi_scan_init(). */
+arch_initcall(acpi_sba_ioc_init_acpi);
+
+extern struct dma_map_ops swiotlb_dma_ops;
+
+static int __init
+sba_init(void)
+{
+	if (!ia64_platform_is("hpzx1") && !ia64_platform_is("hpzx1_swiotlb"))
+		return 0;
+
+#if defined(CONFIG_IA64_GENERIC)
+	/* If we are booting a kdump kernel, the sba_iommu will
+	 * cause devices that were not shutdown properly to MCA
+	 * as soon as they are turned back on.  Our only option for
+	 * a successful kdump kernel boot is to use the swiotlb.
+	 */
+	if (is_kdump_kernel()) {
+		dma_ops = &swiotlb_dma_ops;
+		if (swiotlb_late_init_with_default_size(64 * (1<<20)) != 0)
+			panic("Unable to initialize software I/O TLB:"
+				  " Try machvec=dig boot option");
+		machvec_init("dig");
+		return 0;
+	}
+#endif
+
+	/*
+	 * ioc_found should be populated by the acpi_sba_ioc_handler's .attach()
+	 * routine, but that only happens if acpi_scan_init() has already run.
+	 */
+	while (ioc_found)
+		acpi_sba_ioc_add(ioc_found);
+
+	if (!ioc_list) {
+#ifdef CONFIG_IA64_GENERIC
+		/*
+		 * If we didn't find something sba_iommu can claim, we
+		 * need to setup the swiotlb and switch to the dig machvec.
+		 */
+		dma_ops = &swiotlb_dma_ops;
+		if (swiotlb_late_init_with_default_size(64 * (1<<20)) != 0)
+			panic("Unable to find SBA IOMMU or initialize "
+			      "software I/O TLB: Try machvec=dig boot option");
+		machvec_init("dig");
+#else
+		panic("Unable to find SBA IOMMU: Try a generic or DIG kernel");
+#endif
+		return 0;
+	}
+
+#if defined(CONFIG_IA64_GENERIC) || defined(CONFIG_IA64_HP_ZX1_SWIOTLB)
+	/*
+	 * hpzx1_swiotlb needs to have a fairly small swiotlb bounce
+	 * buffer setup to support devices with smaller DMA masks than
+	 * sba_iommu can handle.
+	 */
+	if (ia64_platform_is("hpzx1_swiotlb")) {
+		extern void hwsw_init(void);
+
+		hwsw_init();
+	}
+#endif
+
+#ifdef CONFIG_PCI
+	{
+		struct pci_bus *b = NULL;
+		while ((b = pci_find_next_bus(b)) != NULL)
+			sba_connect_bus(b);
+	}
+#endif
+
+#ifdef CONFIG_PROC_FS
+	ioc_proc_init();
+#endif
+	return 0;
+}
+
+subsys_initcall(sba_init); /* must be initialized after ACPI etc., but before any drivers... */
+
+static int __init
+nosbagart(char *str)
+{
+	reserve_sba_gart = 0;
+	return 1;
+}
+
+static int sba_dma_supported (struct device *dev, u64 mask)
+{
+	/* make sure it's at least 32bit capable */
+	return ((mask & 0xFFFFFFFFUL) == 0xFFFFFFFFUL);
+}
+
+static int sba_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
+{
+	return 0;
+}
+
+__setup("nosbagart", nosbagart);
+
+static int __init
+sba_page_override(char *str)
+{
+	unsigned long page_size;
+
+	page_size = memparse(str, &str);
+	switch (page_size) {
+		case 4096:
+		case 8192:
+		case 16384:
+		case 65536:
+			iovp_shift = ffs(page_size) - 1;
+			break;
+		default:
+			printk("%s: unknown/unsupported iommu page size %ld\n",
+			       __func__, page_size);
+	}
+
+	return 1;
+}
+
+__setup("sbapagesize=",sba_page_override);
+
+struct dma_map_ops sba_dma_ops = {
+	.alloc			= sba_alloc_coherent,
+	.free			= sba_free_coherent,
+	.map_page		= sba_map_page,
+	.unmap_page		= sba_unmap_page,
+	.map_sg			= sba_map_sg_attrs,
+	.unmap_sg		= sba_unmap_sg_attrs,
+	.sync_single_for_cpu	= machvec_dma_sync_single,
+	.sync_sg_for_cpu	= machvec_dma_sync_sg,
+	.sync_single_for_device	= machvec_dma_sync_single,
+	.sync_sg_for_device	= machvec_dma_sync_sg,
+	.dma_supported		= sba_dma_supported,
+	.mapping_error		= sba_dma_mapping_error,
+};
+
+void sba_dma_init(void)
+{
+	dma_ops = &sba_dma_ops;
+}
diff --git a/kernel/arch/ia64/hp/sim/Kconfig b/kernel/arch/ia64/hp/sim/Kconfig
new file mode 100644
index 000000000..d84707d55
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/Kconfig
@@ -0,0 +1,22 @@
+
+menu "HP Simulator drivers"
+	depends on IA64_HP_SIM || IA64_GENERIC
+
+config HP_SIMETH
+	bool "Simulated Ethernet "
+	depends on NET
+
+config HP_SIMSERIAL
+	bool "Simulated serial driver support"
+	depends on TTY
+
+config HP_SIMSERIAL_CONSOLE
+	bool "Console for HP simulator"
+	depends on HP_SIMSERIAL
+
+config HP_SIMSCSI
+	bool "Simulated SCSI disk"
+	depends on SCSI=y
+
+endmenu
+
diff --git a/kernel/arch/ia64/hp/sim/Makefile b/kernel/arch/ia64/hp/sim/Makefile
new file mode 100644
index 000000000..d10da4793
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/Makefile
@@ -0,0 +1,16 @@
+#
+# ia64/platform/hp/sim/Makefile
+#
+# Copyright (C) 2002 Hewlett-Packard Co.
+#	David Mosberger-Tang <davidm@hpl.hp.com>
+# Copyright (C) 1999 Silicon Graphics, Inc.
+# Copyright (C) Srinivasa Thirumalachar (sprasad@engr.sgi.com)
+#
+
+obj-y := hpsim_irq.o hpsim_setup.o hpsim.o
+obj-$(CONFIG_IA64_GENERIC) += hpsim_machvec.o
+
+obj-$(CONFIG_HP_SIMETH)	+= simeth.o
+obj-$(CONFIG_HP_SIMSERIAL) += simserial.o
+obj-$(CONFIG_HP_SIMSERIAL_CONSOLE) += hpsim_console.o
+obj-$(CONFIG_HP_SIMSCSI) += simscsi.o
diff --git a/kernel/arch/ia64/hp/sim/boot/Makefile b/kernel/arch/ia64/hp/sim/boot/Makefile
new file mode 100644
index 000000000..2e805e0cc
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/boot/Makefile
@@ -0,0 +1,37 @@
+#
+# ia64/boot/Makefile
+#
+# This file is subject to the terms and conditions of the GNU General Public
+# License.  See the file "COPYING" in the main directory of this archive
+# for more details.
+#
+# Copyright (C) 1998, 2003 by David Mosberger-Tang <davidm@hpl.hp.com>
+#
+
+targets-$(CONFIG_IA64_HP_SIM)  += bootloader
+targets := vmlinux.bin vmlinux.gz $(targets-y)
+
+quiet_cmd_cptotop = LN      $@
+      cmd_cptotop = ln -f $< $@
+
+vmlinux.gz: $(obj)/vmlinux.gz $(addprefix $(obj)/,$(targets-y))
+	$(call cmd,cptotop)
+	@echo '  Kernel: $@ is ready'
+
+boot: bootloader
+
+bootloader: $(obj)/bootloader
+	$(call cmd,cptotop)
+
+$(obj)/vmlinux.gz: $(obj)/vmlinux.bin FORCE
+	$(call if_changed,gzip)
+
+$(obj)/vmlinux.bin: vmlinux FORCE
+	$(call if_changed,objcopy)
+
+
+LDFLAGS_bootloader = -static -T
+
+$(obj)/bootloader: $(src)/bootloader.lds $(obj)/bootloader.o $(obj)/boot_head.o $(obj)/fw-emu.o \
+                   lib/lib.a arch/ia64/lib/built-in.o arch/ia64/lib/lib.a FORCE
+	$(call if_changed,ld)
diff --git a/kernel/arch/ia64/hp/sim/boot/boot_head.S b/kernel/arch/ia64/hp/sim/boot/boot_head.S
new file mode 100644
index 000000000..880856549
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/boot/boot_head.S
@@ -0,0 +1,164 @@
+/*
+ * Copyright (C) 1998-2003 Hewlett-Packard Co
+ *	David Mosberger-Tang <davidm@hpl.hp.com>
+ */
+
+#include <asm/asmmacro.h>
+#include <asm/pal.h>
+
+	.bss
+	.align 16
+stack_mem:
+	.skip 16834
+
+	.text
+
+/* This needs to be defined because lib/string.c:strlcat() calls it in case of error... */
+GLOBAL_ENTRY(printk)
+	break 0
+END(printk)
+
+GLOBAL_ENTRY(_start)
+	.prologue
+	.save rp, r0
+	.body
+	movl gp = __gp
+	movl sp = stack_mem+16384-16
+	bsw.1
+	br.call.sptk.many rp=start_bootloader
+0:	nop 0		  /* dummy nop to make unwinding work */
+END(_start)
+
+/*
+ * Set a break point on this function so that symbols are available to set breakpoints in
+ * the kernel being debugged.
+ */
+GLOBAL_ENTRY(debug_break)
+	br.ret.sptk.many b0
+END(debug_break)
+
+GLOBAL_ENTRY(ssc)
+	.regstk 5,0,0,0
+	mov r15=in4
+	break 0x80001
+	br.ret.sptk.many b0
+END(ssc)
+
+GLOBAL_ENTRY(jmp_to_kernel)
+	.regstk 2,0,0,0
+	mov r28=in0
+	mov b7=in1
+	br.sptk.few b7
+END(jmp_to_kernel)
+
+/*
+ * r28 contains the index of the PAL function
+ * r29--31 the args
+ * Return values in ret0--3 (r8--11)
+ */
+GLOBAL_ENTRY(pal_emulator_static)
+	mov r8=-1
+	mov r9=256
+	;;
+	cmp.gtu p6,p7=r9,r28		/* r28 <= 255? */
+(p6)	br.cond.sptk.few static
+	;;
+	mov r9=512
+	;;
+	cmp.gtu p6,p7=r9,r28
+(p6)	br.cond.sptk.few stacked
+	;;
+static:	cmp.eq p6,p7=PAL_PTCE_INFO,r28
+(p7)	br.cond.sptk.few 1f
+	;;
+	mov r8=0			/* status = 0 */
+	movl r9=0x100000000		/* tc.base */
+	movl r10=0x0000000200000003	/* count[0], count[1] */
+	movl r11=0x1000000000002000	/* stride[0], stride[1] */
+	br.cond.sptk.few rp
+1:	cmp.eq p6,p7=PAL_FREQ_RATIOS,r28
+(p7)	br.cond.sptk.few 1f
+	mov r8=0			/* status = 0 */
+	movl r9 =0x100000064		/* proc_ratio (1/100) */
+	movl r10=0x100000100		/* bus_ratio<<32 (1/256) */
+	movl r11=0x100000064		/* itc_ratio<<32 (1/100) */
+	;;
+1:	cmp.eq p6,p7=PAL_RSE_INFO,r28
+(p7)	br.cond.sptk.few 1f
+	mov r8=0			/* status = 0 */
+	mov r9=96			/* num phys stacked */
+	mov r10=0			/* hints */
+	mov r11=0
+	br.cond.sptk.few rp
+1:	cmp.eq p6,p7=PAL_CACHE_FLUSH,r28		/* PAL_CACHE_FLUSH */
+(p7)	br.cond.sptk.few 1f
+	mov r9=ar.lc
+	movl r8=524288			/* flush 512k million cache lines (16MB) */
+	;;
+	mov ar.lc=r8
+	movl r8=0xe000000000000000
+	;;
+.loop:	fc r8
+	add r8=32,r8
+	br.cloop.sptk.few .loop
+	sync.i
+	;;
+	srlz.i
+	;;
+	mov ar.lc=r9
+	mov r8=r0
+	;;
+1:	cmp.eq p6,p7=PAL_PERF_MON_INFO,r28
+(p7)	br.cond.sptk.few 1f
+	mov r8=0			/* status = 0 */
+	movl r9 =0x08122f04		/* generic=4 width=47 retired=8 cycles=18 */
+	mov r10=0			/* reserved */
+	mov r11=0			/* reserved */
+	mov r16=0xffff			/* implemented PMC */
+	mov r17=0x3ffff			/* implemented PMD */
+	add r18=8,r29			/* second index */
+	;;
+	st8 [r29]=r16,16		/* store implemented PMC */
+	st8 [r18]=r0,16			/* clear remaining bits  */
+	;;
+	st8 [r29]=r0,16			/* clear remaining bits  */
+	st8 [r18]=r0,16			/* clear remaining bits  */
+	;;
+	st8 [r29]=r17,16		/* store implemented PMD */
+	st8 [r18]=r0,16			/* clear remaining bits  */
+	mov r16=0xf0			/* cycles count capable PMC */
+	;;
+	st8 [r29]=r0,16			/* clear remaining bits  */
+	st8 [r18]=r0,16			/* clear remaining bits  */
+	mov r17=0xf0			/* retired bundles capable PMC */
+	;;
+	st8 [r29]=r16,16		/* store cycles capable */
+	st8 [r18]=r0,16			/* clear remaining bits  */
+	;;
+	st8 [r29]=r0,16			/* clear remaining bits  */
+	st8 [r18]=r0,16			/* clear remaining bits  */
+	;;
+	st8 [r29]=r17,16		/* store retired bundle capable */
+	st8 [r18]=r0,16			/* clear remaining bits  */
+	;;
+	st8 [r29]=r0,16			/* clear remaining bits  */
+	st8 [r18]=r0,16			/* clear remaining bits  */
+	;;
+1:	cmp.eq p6,p7=PAL_VM_SUMMARY,r28
+(p7)	br.cond.sptk.few 1f
+	mov	r8=0			/* status = 0  */
+	movl	r9=0x2044040020F1865	/* num_tc_levels=2, num_unique_tcs=4 */
+					/* max_itr_entry=64, max_dtr_entry=64 */
+					/* hash_tag_id=2, max_pkr=15 */
+					/* key_size=24, phys_add_size=50, vw=1 */
+	movl	r10=0x183C		/* rid_size=24, impl_va_msb=60 */
+	;;
+1:	cmp.eq p6,p7=PAL_MEM_ATTRIB,r28
+(p7)	br.cond.sptk.few 1f
+	mov	r8=0			/* status = 0 */
+	mov	r9=0x80|0x01		/* NatPage|WB */
+	;;
+1:	br.cond.sptk.few rp
+stacked:
+	br.ret.sptk.few rp
+END(pal_emulator_static)
diff --git a/kernel/arch/ia64/hp/sim/boot/bootloader.c b/kernel/arch/ia64/hp/sim/boot/bootloader.c
new file mode 100644
index 000000000..28f4b230b
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/boot/bootloader.c
@@ -0,0 +1,174 @@
+/*
+ * arch/ia64/hp/sim/boot/bootloader.c
+ *
+ * Loads an ELF kernel.
+ *
+ * Copyright (C) 1998-2003 Hewlett-Packard Co
+ *	David Mosberger-Tang <davidm@hpl.hp.com>
+ *	Stephane Eranian <eranian@hpl.hp.com>
+ *
+ * 01/07/99 S.Eranian modified to pass command line arguments to kernel
+ */
+struct task_struct;	/* forward declaration for elf.h */
+
+#include <linux/elf.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+
+#include <asm/elf.h>
+#include <asm/intrinsics.h>
+#include <asm/pal.h>
+#include <asm/pgtable.h>
+#include <asm/sal.h>
+
+#include "ssc.h"
+
+struct disk_req {
+	unsigned long addr;
+	unsigned len;
+};
+
+struct disk_stat {
+	int fd;
+	unsigned count;
+};
+
+extern void jmp_to_kernel (unsigned long bp, unsigned long e_entry);
+extern struct ia64_boot_param *sys_fw_init (const char *args, int arglen);
+extern void debug_break (void);
+
+static void
+cons_write (const char *buf)
+{
+	unsigned long ch;
+
+	while ((ch = *buf++) != '\0') {
+		ssc(ch, 0, 0, 0, SSC_PUTCHAR);
+		if (ch == '\n')
+		  ssc('\r', 0, 0, 0, SSC_PUTCHAR);
+	}
+}
+
+#define MAX_ARGS 32
+
+void
+start_bootloader (void)
+{
+	static char mem[4096];
+	static char buffer[1024];
+	unsigned long off;
+	int fd, i;
+	struct disk_req req;
+	struct disk_stat stat;
+	struct elfhdr *elf;
+	struct elf_phdr *elf_phdr;	/* program header */
+	unsigned long e_entry, e_phoff, e_phnum;
+	register struct ia64_boot_param *bp;
+	char *kpath, *args;
+	long arglen = 0;
+
+	ssc(0, 0, 0, 0, SSC_CONSOLE_INIT);
+
+	/*
+	 * S.Eranian: extract the commandline argument from the simulator
+	 *
+	 * The expected format is as follows:
+         *
+	 *	kernelname args...
+	 *
+	 * Both are optional but you can't have the second one without the first.
+	 */
+	arglen = ssc((long) buffer, 0, 0, 0, SSC_GET_ARGS);
+
+	kpath = "vmlinux";
+	args = buffer;
+	if (arglen > 0) {
+		kpath = buffer;
+		while (*args != ' ' && *args != '\0')
+			++args, --arglen;
+		if (*args == ' ')
+			*args++ = '\0', --arglen;
+	}
+
+	if (arglen <= 0) {
+		args = "";
+		arglen = 1;
+	}
+
+	fd = ssc((long) kpath, 1, 0, 0, SSC_OPEN);
+
+	if (fd < 0) {
+		cons_write(kpath);
+		cons_write(": file not found, reboot now\n");
+		for(;;);
+	}
+	stat.fd = fd;
+	off = 0;
+
+	req.len = sizeof(mem);
+	req.addr = (long) mem;
+	ssc(fd, 1, (long) &req, off, SSC_READ);
+	ssc((long) &stat, 0, 0, 0, SSC_WAIT_COMPLETION);
+
+	elf = (struct elfhdr *) mem;
+	if (elf->e_ident[0] == 0x7f && strncmp(elf->e_ident + 1, "ELF", 3) != 0) {
+		cons_write("not an ELF file\n");
+		return;
+	}
+	if (elf->e_type != ET_EXEC) {
+		cons_write("not an ELF executable\n");
+		return;
+	}
+	if (!elf_check_arch(elf)) {
+		cons_write("kernel not for this processor\n");
+		return;
+	}
+
+	e_entry = elf->e_entry;
+	e_phnum = elf->e_phnum;
+	e_phoff = elf->e_phoff;
+
+	cons_write("loading ");
+	cons_write(kpath);
+	cons_write("...\n");
+
+	for (i = 0; i < e_phnum; ++i) {
+		req.len = sizeof(*elf_phdr);
+		req.addr = (long) mem;
+		ssc(fd, 1, (long) &req, e_phoff, SSC_READ);
+		ssc((long) &stat, 0, 0, 0, SSC_WAIT_COMPLETION);
+		if (stat.count != sizeof(*elf_phdr)) {
+			cons_write("failed to read phdr\n");
+			return;
+		}
+		e_phoff += sizeof(*elf_phdr);
+
+		elf_phdr = (struct elf_phdr *) mem;
+
+		if (elf_phdr->p_type != PT_LOAD)
+			continue;
+
+		req.len = elf_phdr->p_filesz;
+		req.addr = __pa(elf_phdr->p_paddr);
+		ssc(fd, 1, (long) &req, elf_phdr->p_offset, SSC_READ);
+		ssc((long) &stat, 0, 0, 0, SSC_WAIT_COMPLETION);
+		memset((char *)__pa(elf_phdr->p_paddr) + elf_phdr->p_filesz, 0,
+		       elf_phdr->p_memsz - elf_phdr->p_filesz);
+	}
+	ssc(fd, 0, 0, 0, SSC_CLOSE);
+
+	cons_write("starting kernel...\n");
+
+	/* fake an I/O base address: */
+	ia64_setreg(_IA64_REG_AR_KR0, 0xffffc000000UL);
+
+	bp = sys_fw_init(args, arglen);
+
+	ssc(0, (long) kpath, 0, 0, SSC_LOAD_SYMBOLS);
+
+	debug_break();
+	jmp_to_kernel((unsigned long) bp, e_entry);
+
+	cons_write("kernel returned!\n");
+	ssc(-1, 0, 0, 0, SSC_EXIT);
+}
diff --git a/kernel/arch/ia64/hp/sim/boot/bootloader.lds b/kernel/arch/ia64/hp/sim/boot/bootloader.lds
new file mode 100644
index 000000000..3977f25a1
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/boot/bootloader.lds
@@ -0,0 +1,66 @@
+OUTPUT_FORMAT("elf64-ia64-little")
+OUTPUT_ARCH(ia64)
+ENTRY(_start)
+SECTIONS
+{
+  /* Read-only sections, merged into text segment: */
+  . = 0x100000;
+
+  _text = .;
+  .text : { *(__ivt_section) *(.text) }
+  _etext = .;
+
+  /* Global data */
+  _data = .;
+  .rodata : { *(.rodata) *(.rodata.*) }
+  .data    : { *(.data) *(.gnu.linkonce.d*) CONSTRUCTORS }
+  __gp = ALIGN (8) + 0x200000;
+  .got           : { *(.got.plt) *(.got) }
+  /* We want the small data sections together, so single-instruction offsets
+     can access them all, and initialized data all before uninitialized, so
+     we can shorten the on-disk segment size.  */
+  .sdata     : { *(.sdata) }
+  _edata  =  .;
+
+  __bss_start = .;
+  .sbss      : { *(.sbss) *(.scommon) }
+  .bss       : { *(.bss) *(COMMON) }
+  . = ALIGN(64 / 8);
+  __bss_stop = .;
+  _end = . ;
+
+  /* Stabs debugging sections.  */
+  .stab 0 : { *(.stab) }
+  .stabstr 0 : { *(.stabstr) }
+  .stab.excl 0 : { *(.stab.excl) }
+  .stab.exclstr 0 : { *(.stab.exclstr) }
+  .stab.index 0 : { *(.stab.index) }
+  .stab.indexstr 0 : { *(.stab.indexstr) }
+  .comment 0 : { *(.comment) }
+  /* DWARF debug sections.
+     Symbols in the DWARF debugging sections are relative to the beginning
+     of the section so we begin them at 0.  */
+  /* DWARF 1 */
+  .debug          0 : { *(.debug) }
+  .line           0 : { *(.line) }
+  /* GNU DWARF 1 extensions */
+  .debug_srcinfo  0 : { *(.debug_srcinfo) }
+  .debug_sfnames  0 : { *(.debug_sfnames) }
+  /* DWARF 1.1 and DWARF 2 */
+  .debug_aranges  0 : { *(.debug_aranges) }
+  .debug_pubnames 0 : { *(.debug_pubnames) }
+  /* DWARF 2 */
+  .debug_info     0 : { *(.debug_info) }
+  .debug_abbrev   0 : { *(.debug_abbrev) }
+  .debug_line     0 : { *(.debug_line) }
+  .debug_frame    0 : { *(.debug_frame) }
+  .debug_str      0 : { *(.debug_str) }
+  .debug_loc      0 : { *(.debug_loc) }
+  .debug_macinfo  0 : { *(.debug_macinfo) }
+  /* SGI/MIPS DWARF 2 extensions */
+  .debug_weaknames 0 : { *(.debug_weaknames) }
+  .debug_funcnames 0 : { *(.debug_funcnames) }
+  .debug_typenames 0 : { *(.debug_typenames) }
+  .debug_varnames  0 : { *(.debug_varnames) }
+  /* These must appear regardless of  .  */
+}
diff --git a/kernel/arch/ia64/hp/sim/boot/fw-emu.c b/kernel/arch/ia64/hp/sim/boot/fw-emu.c
new file mode 100644
index 000000000..87bf9ad8c
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/boot/fw-emu.c
@@ -0,0 +1,373 @@
+/*
+ * PAL & SAL emulation.
+ *
+ * Copyright (C) 1998-2001 Hewlett-Packard Co
+ *	David Mosberger-Tang <davidm@hpl.hp.com>
+ */
+
+#ifdef CONFIG_PCI
+# include <linux/pci.h>
+#endif
+
+#include <linux/efi.h>
+#include <asm/io.h>
+#include <asm/pal.h>
+#include <asm/sal.h>
+#include <asm/setup.h>
+
+#include "ssc.h"
+
+#define MB	(1024*1024UL)
+
+#define SIMPLE_MEMMAP	1
+
+#if SIMPLE_MEMMAP
+# define NUM_MEM_DESCS	4
+#else
+# define NUM_MEM_DESCS	16
+#endif
+
+static char fw_mem[(  sizeof(struct ia64_boot_param)
+		    + sizeof(efi_system_table_t)
+		    + sizeof(efi_runtime_services_t)
+		    + 1*sizeof(efi_config_table_t)
+		    + sizeof(struct ia64_sal_systab)
+		    + sizeof(struct ia64_sal_desc_entry_point)
+		    + NUM_MEM_DESCS*(sizeof(efi_memory_desc_t))
+		    + 1024)] __attribute__ ((aligned (8)));
+
+#define SECS_PER_HOUR   (60 * 60)
+#define SECS_PER_DAY    (SECS_PER_HOUR * 24)
+
+/* Compute the `struct tm' representation of *T,
+   offset OFFSET seconds east of UTC,
+   and store year, yday, mon, mday, wday, hour, min, sec into *TP.
+   Return nonzero if successful.  */
+int
+offtime (unsigned long t, efi_time_t *tp)
+{
+	const unsigned short int __mon_yday[2][13] =
+	{
+		/* Normal years.  */
+		{ 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334, 365 },
+		/* Leap years.  */
+		{ 0, 31, 60, 91, 121, 152, 182, 213, 244, 274, 305, 335, 366 }
+	};
+	long int days, rem, y;
+	const unsigned short int *ip;
+
+	days = t / SECS_PER_DAY;
+	rem = t % SECS_PER_DAY;
+	while (rem < 0) {
+		rem += SECS_PER_DAY;
+		--days;
+	}
+	while (rem >= SECS_PER_DAY) {
+		rem -= SECS_PER_DAY;
+		++days;
+	}
+	tp->hour = rem / SECS_PER_HOUR;
+	rem %= SECS_PER_HOUR;
+	tp->minute = rem / 60;
+	tp->second = rem % 60;
+	/* January 1, 1970 was a Thursday.  */
+	y = 1970;
+
+#	define DIV(a, b) ((a) / (b) - ((a) % (b) < 0))
+#	define LEAPS_THRU_END_OF(y) (DIV (y, 4) - DIV (y, 100) + DIV (y, 400))
+#	define __isleap(year) \
+	  ((year) % 4 == 0 && ((year) % 100 != 0 || (year) % 400 == 0))
+
+	while (days < 0 || days >= (__isleap (y) ? 366 : 365)) {
+		/* Guess a corrected year, assuming 365 days per year.  */
+		long int yg = y + days / 365 - (days % 365 < 0);
+
+		/* Adjust DAYS and Y to match the guessed year.  */
+		days -= ((yg - y) * 365 + LEAPS_THRU_END_OF (yg - 1)
+			 - LEAPS_THRU_END_OF (y - 1));
+		y = yg;
+	}
+	tp->year = y;
+	ip = __mon_yday[__isleap(y)];
+	for (y = 11; days < (long int) ip[y]; --y)
+		continue;
+	days -= ip[y];
+	tp->month = y + 1;
+	tp->day = days + 1;
+	return 1;
+}
+
+extern void pal_emulator_static (void);
+
+/* Macro to emulate SAL call using legacy IN and OUT calls to CF8, CFC etc.. */
+
+#define BUILD_CMD(addr)		((0x80000000 | (addr)) & ~3)
+
+#define REG_OFFSET(addr)	(0x00000000000000FF & (addr))
+#define DEVICE_FUNCTION(addr)	(0x000000000000FF00 & (addr))
+#define BUS_NUMBER(addr)	(0x0000000000FF0000 & (addr))
+
+static efi_status_t
+fw_efi_get_time (efi_time_t *tm, efi_time_cap_t *tc)
+{
+#if defined(CONFIG_IA64_HP_SIM) || defined(CONFIG_IA64_GENERIC)
+	struct {
+		int tv_sec;	/* must be 32bits to work */
+		int tv_usec;
+	} tv32bits;
+
+	ssc((unsigned long) &tv32bits, 0, 0, 0, SSC_GET_TOD);
+
+	memset(tm, 0, sizeof(*tm));
+	offtime(tv32bits.tv_sec, tm);
+
+	if (tc)
+		memset(tc, 0, sizeof(*tc));
+#else
+#	error Not implemented yet...
+#endif
+	return EFI_SUCCESS;
+}
+
+static void
+efi_reset_system (int reset_type, efi_status_t status, unsigned long data_size, efi_char16_t *data)
+{
+#if defined(CONFIG_IA64_HP_SIM) || defined(CONFIG_IA64_GENERIC)
+	ssc(status, 0, 0, 0, SSC_EXIT);
+#else
+#	error Not implemented yet...
+#endif
+}
+
+static efi_status_t
+efi_unimplemented (void)
+{
+	return EFI_UNSUPPORTED;
+}
+
+static struct sal_ret_values
+sal_emulator (long index, unsigned long in1, unsigned long in2,
+	      unsigned long in3, unsigned long in4, unsigned long in5,
+	      unsigned long in6, unsigned long in7)
+{
+	long r9  = 0;
+	long r10 = 0;
+	long r11 = 0;
+	long status;
+
+	/*
+	 * Don't do a "switch" here since that gives us code that
+	 * isn't self-relocatable.
+	 */
+	status = 0;
+	if (index == SAL_FREQ_BASE) {
+		if (in1 == SAL_FREQ_BASE_PLATFORM)
+			r9 = 200000000;
+		else if (in1 == SAL_FREQ_BASE_INTERVAL_TIMER) {
+			/*
+			 * Is this supposed to be the cr.itc frequency
+			 * or something platform specific?  The SAL
+			 * doc ain't exactly clear on this...
+			 */
+			r9 = 700000000;
+		} else if (in1 == SAL_FREQ_BASE_REALTIME_CLOCK)
+			r9 = 1;
+		else
+			status = -1;
+	} else if (index == SAL_SET_VECTORS) {
+		;
+	} else if (index == SAL_GET_STATE_INFO) {
+		;
+	} else if (index == SAL_GET_STATE_INFO_SIZE) {
+		;
+	} else if (index == SAL_CLEAR_STATE_INFO) {
+		;
+	} else if (index == SAL_MC_RENDEZ) {
+		;
+	} else if (index == SAL_MC_SET_PARAMS) {
+		;
+	} else if (index == SAL_CACHE_FLUSH) {
+		;
+	} else if (index == SAL_CACHE_INIT) {
+		;
+#ifdef CONFIG_PCI
+	} else if (index == SAL_PCI_CONFIG_READ) {
+		/*
+		 * in1 contains the PCI configuration address and in2
+		 * the size of the read.  The value that is read is
+		 * returned via the general register r9.
+		 */
+                outl(BUILD_CMD(in1), 0xCF8);
+                if (in2 == 1)                           /* Reading byte  */
+                        r9 = inb(0xCFC + ((REG_OFFSET(in1) & 3)));
+                else if (in2 == 2)                      /* Reading word  */
+                        r9 = inw(0xCFC + ((REG_OFFSET(in1) & 2)));
+                else                                    /* Reading dword */
+                        r9 = inl(0xCFC);
+                status = PCIBIOS_SUCCESSFUL;
+	} else if (index == SAL_PCI_CONFIG_WRITE) {
+	      	/*
+		 * in1 contains the PCI configuration address, in2 the
+		 * size of the write, and in3 the actual value to be
+		 * written out.
+		 */
+                outl(BUILD_CMD(in1), 0xCF8);
+                if (in2 == 1)                           /* Writing byte  */
+                        outb(in3, 0xCFC + ((REG_OFFSET(in1) & 3)));
+                else if (in2 == 2)                      /* Writing word  */
+                        outw(in3, 0xCFC + ((REG_OFFSET(in1) & 2)));
+                else                                    /* Writing dword */
+                        outl(in3, 0xCFC);
+                status = PCIBIOS_SUCCESSFUL;
+#endif /* CONFIG_PCI */
+	} else if (index == SAL_UPDATE_PAL) {
+		;
+	} else {
+		status = -1;
+	}
+	return ((struct sal_ret_values) {status, r9, r10, r11});
+}
+
+struct ia64_boot_param *
+sys_fw_init (const char *args, int arglen)
+{
+	efi_system_table_t *efi_systab;
+	efi_runtime_services_t *efi_runtime;
+	efi_config_table_t *efi_tables;
+	struct ia64_sal_systab *sal_systab;
+	efi_memory_desc_t *efi_memmap, *md;
+	unsigned long *pal_desc, *sal_desc;
+	struct ia64_sal_desc_entry_point *sal_ed;
+	struct ia64_boot_param *bp;
+	unsigned char checksum = 0;
+	char *cp, *cmd_line;
+	int i = 0;
+#	define MAKE_MD(typ, attr, start, end)		\
+	do {						\
+		md = efi_memmap + i++;			\
+		md->type = typ;				\
+		md->pad = 0;				\
+		md->phys_addr = start;			\
+		md->virt_addr = 0;			\
+		md->num_pages = (end - start) >> 12;	\
+		md->attribute = attr;			\
+	} while (0)
+
+	memset(fw_mem, 0, sizeof(fw_mem));
+
+	pal_desc = (unsigned long *) &pal_emulator_static;
+	sal_desc = (unsigned long *) &sal_emulator;
+
+	cp = fw_mem;
+	efi_systab  = (void *) cp; cp += sizeof(*efi_systab);
+	efi_runtime = (void *) cp; cp += sizeof(*efi_runtime);
+	efi_tables  = (void *) cp; cp += sizeof(*efi_tables);
+	sal_systab  = (void *) cp; cp += sizeof(*sal_systab);
+	sal_ed      = (void *) cp; cp += sizeof(*sal_ed);
+	efi_memmap  = (void *) cp; cp += NUM_MEM_DESCS*sizeof(*efi_memmap);
+	bp	    = (void *) cp; cp += sizeof(*bp);
+	cmd_line    = (void *) cp;
+
+	if (args) {
+		if (arglen >= 1024)
+			arglen = 1023;
+		memcpy(cmd_line, args, arglen);
+	} else {
+		arglen = 0;
+	}
+	cmd_line[arglen] = '\0';
+
+	memset(efi_systab, 0, sizeof(*efi_systab));
+	efi_systab->hdr.signature = EFI_SYSTEM_TABLE_SIGNATURE;
+	efi_systab->hdr.revision  = ((1 << 16) | 00);
+	efi_systab->hdr.headersize = sizeof(efi_systab->hdr);
+	efi_systab->fw_vendor = __pa("H\0e\0w\0l\0e\0t\0t\0-\0P\0a\0c\0k\0a\0r\0d\0\0");
+	efi_systab->fw_revision = 1;
+	efi_systab->runtime = (void *) __pa(efi_runtime);
+	efi_systab->nr_tables = 1;
+	efi_systab->tables = __pa(efi_tables);
+
+	efi_runtime->hdr.signature = EFI_RUNTIME_SERVICES_SIGNATURE;
+	efi_runtime->hdr.revision = EFI_RUNTIME_SERVICES_REVISION;
+	efi_runtime->hdr.headersize = sizeof(efi_runtime->hdr);
+	efi_runtime->get_time = (void *)__pa(&fw_efi_get_time);
+	efi_runtime->set_time = (void *)__pa(&efi_unimplemented);
+	efi_runtime->get_wakeup_time = (void *)__pa(&efi_unimplemented);
+	efi_runtime->set_wakeup_time = (void *)__pa(&efi_unimplemented);
+	efi_runtime->set_virtual_address_map = (void *)__pa(&efi_unimplemented);
+	efi_runtime->get_variable = (void *)__pa(&efi_unimplemented);
+	efi_runtime->get_next_variable = (void *)__pa(&efi_unimplemented);
+	efi_runtime->set_variable = (void *)__pa(&efi_unimplemented);
+	efi_runtime->get_next_high_mono_count = (void *)__pa(&efi_unimplemented);
+	efi_runtime->reset_system = (void *)__pa(&efi_reset_system);
+
+	efi_tables->guid = SAL_SYSTEM_TABLE_GUID;
+	efi_tables->table = __pa(sal_systab);
+
+	/* fill in the SAL system table: */
+	memcpy(sal_systab->signature, "SST_", 4);
+	sal_systab->size = sizeof(*sal_systab);
+	sal_systab->sal_rev_minor = 1;
+	sal_systab->sal_rev_major = 0;
+	sal_systab->entry_count = 1;
+
+#ifdef CONFIG_IA64_GENERIC
+        strcpy(sal_systab->oem_id, "Generic");
+        strcpy(sal_systab->product_id, "IA-64 system");
+#endif
+
+#ifdef CONFIG_IA64_HP_SIM
+	strcpy(sal_systab->oem_id, "Hewlett-Packard");
+	strcpy(sal_systab->product_id, "HP-simulator");
+#endif
+
+	/* fill in an entry point: */
+	sal_ed->type = SAL_DESC_ENTRY_POINT;
+	sal_ed->pal_proc = __pa(pal_desc[0]);
+	sal_ed->sal_proc = __pa(sal_desc[0]);
+	sal_ed->gp = __pa(sal_desc[1]);
+
+	for (cp = (char *) sal_systab; cp < (char *) efi_memmap; ++cp)
+		checksum += *cp;
+
+	sal_systab->checksum = -checksum;
+
+#if SIMPLE_MEMMAP
+	/* simulate free memory at physical address zero */
+	MAKE_MD(EFI_BOOT_SERVICES_DATA,		EFI_MEMORY_WB,    0*MB,    1*MB);
+	MAKE_MD(EFI_PAL_CODE,			EFI_MEMORY_WB,    1*MB,    2*MB);
+	MAKE_MD(EFI_CONVENTIONAL_MEMORY,	EFI_MEMORY_WB,    2*MB,  130*MB);
+	MAKE_MD(EFI_CONVENTIONAL_MEMORY,	EFI_MEMORY_WB, 4096*MB, 4128*MB);
+#else
+	MAKE_MD( 4,		   0x9, 0x0000000000000000, 0x0000000000001000);
+	MAKE_MD( 7,		   0x9, 0x0000000000001000, 0x000000000008a000);
+	MAKE_MD( 4,		   0x9, 0x000000000008a000, 0x00000000000a0000);
+	MAKE_MD( 5, 0x8000000000000009, 0x00000000000c0000, 0x0000000000100000);
+	MAKE_MD( 7,		   0x9, 0x0000000000100000, 0x0000000004400000);
+	MAKE_MD( 2,		   0x9, 0x0000000004400000, 0x0000000004be5000);
+	MAKE_MD( 7,		   0x9, 0x0000000004be5000, 0x000000007f77e000);
+	MAKE_MD( 6, 0x8000000000000009, 0x000000007f77e000, 0x000000007fb94000);
+	MAKE_MD( 6, 0x8000000000000009, 0x000000007fb94000, 0x000000007fb95000);
+	MAKE_MD( 6, 0x8000000000000009, 0x000000007fb95000, 0x000000007fc00000);
+	MAKE_MD(13, 0x8000000000000009, 0x000000007fc00000, 0x000000007fc3a000);
+	MAKE_MD( 7,		   0x9, 0x000000007fc3a000, 0x000000007fea0000);
+	MAKE_MD( 5, 0x8000000000000009, 0x000000007fea0000, 0x000000007fea8000);
+	MAKE_MD( 7,		   0x9, 0x000000007fea8000, 0x000000007feab000);
+	MAKE_MD( 5, 0x8000000000000009, 0x000000007feab000, 0x000000007ffff000);
+	MAKE_MD( 7,		   0x9, 0x00000000ff400000, 0x0000000104000000);
+#endif
+
+	bp->efi_systab = __pa(&fw_mem);
+	bp->efi_memmap = __pa(efi_memmap);
+	bp->efi_memmap_size = NUM_MEM_DESCS*sizeof(efi_memory_desc_t);
+	bp->efi_memdesc_size = sizeof(efi_memory_desc_t);
+	bp->efi_memdesc_version = 1;
+	bp->command_line = __pa(cmd_line);
+	bp->console_info.num_cols = 80;
+	bp->console_info.num_rows = 25;
+	bp->console_info.orig_x = 0;
+	bp->console_info.orig_y = 24;
+	bp->fpswa = 0;
+
+	return bp;
+}
diff --git a/kernel/arch/ia64/hp/sim/boot/ssc.h b/kernel/arch/ia64/hp/sim/boot/ssc.h
new file mode 100644
index 000000000..3b94c03e4
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/boot/ssc.h
@@ -0,0 +1,35 @@
+/*
+ * Copyright (C) 1998-2003 Hewlett-Packard Co
+ *	David Mosberger-Tang <davidm@hpl.hp.com>
+ *	Stephane Eranian <eranian@hpl.hp.com>
+ */
+#ifndef ssc_h
+#define ssc_h
+
+/* Simulator system calls: */
+
+#define SSC_CONSOLE_INIT		20
+#define SSC_GETCHAR			21
+#define SSC_PUTCHAR			31
+#define SSC_OPEN			50
+#define SSC_CLOSE			51
+#define SSC_READ			52
+#define SSC_WRITE			53
+#define SSC_GET_COMPLETION		54
+#define SSC_WAIT_COMPLETION		55
+#define SSC_CONNECT_INTERRUPT		58
+#define SSC_GENERATE_INTERRUPT		59
+#define SSC_SET_PERIODIC_INTERRUPT	60
+#define SSC_GET_RTC			65
+#define SSC_EXIT			66
+#define SSC_LOAD_SYMBOLS		69
+#define SSC_GET_TOD			74
+
+#define SSC_GET_ARGS			75
+
+/*
+ * Simulator system call.
+ */
+extern long ssc (long arg0, long arg1, long arg2, long arg3, int nr);
+
+#endif /* ssc_h */
diff --git a/kernel/arch/ia64/hp/sim/hpsim.S b/kernel/arch/ia64/hp/sim/hpsim.S
new file mode 100644
index 000000000..ff16e8a85
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/hpsim.S
@@ -0,0 +1,10 @@
+#include <asm/asmmacro.h>
+
+/*
+ * Simulator system call.
+ */
+GLOBAL_ENTRY(ia64_ssc)
+	mov r15=r36
+	break 0x80001
+	br.ret.sptk.many rp
+END(ia64_ssc)
diff --git a/kernel/arch/ia64/hp/sim/hpsim_console.c b/kernel/arch/ia64/hp/sim/hpsim_console.c
new file mode 100644
index 000000000..01663bc42
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/hpsim_console.c
@@ -0,0 +1,76 @@
+/*
+ * Platform dependent support for HP simulator.
+ *
+ * Copyright (C) 1998, 1999, 2002 Hewlett-Packard Co
+ *	David Mosberger-Tang <davidm@hpl.hp.com>
+ * Copyright (C) 1999 Vijay Chander <vijay@engr.sgi.com>
+ */
+
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/param.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/tty.h>
+#include <linux/kdev_t.h>
+#include <linux/console.h>
+
+#include <asm/delay.h>
+#include <asm/irq.h>
+#include <asm/pal.h>
+#include <asm/machvec.h>
+#include <asm/pgtable.h>
+#include <asm/sal.h>
+#include <asm/hpsim.h>
+
+#include "hpsim_ssc.h"
+
+static int simcons_init (struct console *, char *);
+static void simcons_write (struct console *, const char *, unsigned);
+static struct tty_driver *simcons_console_device (struct console *, int *);
+
+static struct console hpsim_cons = {
+	.name =		"simcons",
+	.write =	simcons_write,
+	.device =	simcons_console_device,
+	.setup =	simcons_init,
+	.flags =	CON_PRINTBUFFER,
+	.index =	-1,
+};
+
+static int
+simcons_init (struct console *cons, char *options)
+{
+	return 0;
+}
+
+static void
+simcons_write (struct console *cons, const char *buf, unsigned count)
+{
+	unsigned long ch;
+
+	while (count-- > 0) {
+		ch = *buf++;
+		ia64_ssc(ch, 0, 0, 0, SSC_PUTCHAR);
+		if (ch == '\n')
+		  ia64_ssc('\r', 0, 0, 0, SSC_PUTCHAR);
+	}
+}
+
+static struct tty_driver *simcons_console_device (struct console *c, int *index)
+{
+	*index = c->index;
+	return hp_simserial_driver;
+}
+
+int simcons_register(void)
+{
+	if (!ia64_platform_is("hpsim"))
+		return 1;
+
+	if (hpsim_cons.flags & CON_ENABLED)
+		return 1;
+
+	register_console(&hpsim_cons);
+	return 0;
+}
diff --git a/kernel/arch/ia64/hp/sim/hpsim_irq.c b/kernel/arch/ia64/hp/sim/hpsim_irq.c
new file mode 100644
index 000000000..0aa70ebda
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/hpsim_irq.c
@@ -0,0 +1,75 @@
+/*
+ * Platform dependent support for HP simulator.
+ *
+ * Copyright (C) 1998-2001 Hewlett-Packard Co
+ * Copyright (C) 1998-2001 David Mosberger-Tang <davidm@hpl.hp.com>
+ */
+
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/irq.h>
+
+#include "hpsim_ssc.h"
+
+static unsigned int
+hpsim_irq_startup(struct irq_data *data)
+{
+	return 0;
+}
+
+static void
+hpsim_irq_noop(struct irq_data *data)
+{
+}
+
+static int
+hpsim_set_affinity_noop(struct irq_data *d, const struct cpumask *b, bool f)
+{
+	return 0;
+}
+
+static struct irq_chip irq_type_hp_sim = {
+	.name =			"hpsim",
+	.irq_startup =		hpsim_irq_startup,
+	.irq_shutdown =		hpsim_irq_noop,
+	.irq_enable =		hpsim_irq_noop,
+	.irq_disable =		hpsim_irq_noop,
+	.irq_ack =		hpsim_irq_noop,
+	.irq_set_affinity =	hpsim_set_affinity_noop,
+};
+
+static void hpsim_irq_set_chip(int irq)
+{
+	struct irq_chip *chip = irq_get_chip(irq);
+
+	if (chip == &no_irq_chip)
+		irq_set_chip(irq, &irq_type_hp_sim);
+}
+
+static void hpsim_connect_irq(int intr, int irq)
+{
+	ia64_ssc(intr, irq, 0, 0, SSC_CONNECT_INTERRUPT);
+}
+
+int hpsim_get_irq(int intr)
+{
+	int irq = assign_irq_vector(AUTO_ASSIGN);
+
+	if (irq >= 0) {
+		hpsim_irq_set_chip(irq);
+		irq_set_handler(irq, handle_simple_irq);
+		hpsim_connect_irq(intr, irq);
+	}
+
+	return irq;
+}
+
+void __init
+hpsim_irq_init (void)
+{
+	int i;
+
+	for_each_active_irq(i)
+		hpsim_irq_set_chip(i);
+}
diff --git a/kernel/arch/ia64/hp/sim/hpsim_machvec.c b/kernel/arch/ia64/hp/sim/hpsim_machvec.c
new file mode 100644
index 000000000..c21419359
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/hpsim_machvec.c
@@ -0,0 +1,3 @@
+#define MACHVEC_PLATFORM_NAME		hpsim
+#define MACHVEC_PLATFORM_HEADER		<asm/machvec_hpsim.h>
+#include <asm/machvec_init.h>
diff --git a/kernel/arch/ia64/hp/sim/hpsim_setup.c b/kernel/arch/ia64/hp/sim/hpsim_setup.c
new file mode 100644
index 000000000..664a5402a
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/hpsim_setup.c
@@ -0,0 +1,40 @@
+/*
+ * Platform dependent support for HP simulator.
+ *
+ * Copyright (C) 1998, 1999, 2002 Hewlett-Packard Co
+ *	David Mosberger-Tang <davidm@hpl.hp.com>
+ * Copyright (C) 1999 Vijay Chander <vijay@engr.sgi.com>
+ */
+#include <linux/console.h>
+#include <linux/init.h>
+#include <linux/kdev_t.h>
+#include <linux/kernel.h>
+#include <linux/major.h>
+#include <linux/param.h>
+#include <linux/root_dev.h>
+#include <linux/string.h>
+#include <linux/types.h>
+
+#include <asm/delay.h>
+#include <asm/irq.h>
+#include <asm/pal.h>
+#include <asm/machvec.h>
+#include <asm/pgtable.h>
+#include <asm/sal.h>
+#include <asm/hpsim.h>
+
+#include "hpsim_ssc.h"
+
+void
+ia64_ctl_trace (long on)
+{
+	ia64_ssc(on, 0, 0, 0, SSC_CTL_TRACE);
+}
+
+void __init
+hpsim_setup (char **cmdline_p)
+{
+	ROOT_DEV = Root_SDA1;		/* default to first SCSI drive */
+
+	simcons_register();
+}
diff --git a/kernel/arch/ia64/hp/sim/hpsim_ssc.h b/kernel/arch/ia64/hp/sim/hpsim_ssc.h
new file mode 100644
index 000000000..bfa390627
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/hpsim_ssc.h
@@ -0,0 +1,36 @@
+/*
+ * Platform dependent support for HP simulator.
+ *
+ * Copyright (C) 1998, 1999 Hewlett-Packard Co
+ * Copyright (C) 1998, 1999 David Mosberger-Tang <davidm@hpl.hp.com>
+ * Copyright (C) 1999 Vijay Chander <vijay@engr.sgi.com>
+ */
+#ifndef _IA64_PLATFORM_HPSIM_SSC_H
+#define _IA64_PLATFORM_HPSIM_SSC_H
+
+/* Simulator system calls: */
+
+#define SSC_CONSOLE_INIT		20
+#define SSC_GETCHAR			21
+#define SSC_PUTCHAR			31
+#define SSC_CONNECT_INTERRUPT		58
+#define SSC_GENERATE_INTERRUPT		59
+#define SSC_SET_PERIODIC_INTERRUPT	60
+#define SSC_GET_RTC			65
+#define SSC_EXIT			66
+#define SSC_LOAD_SYMBOLS		69
+#define SSC_GET_TOD			74
+#define SSC_CTL_TRACE			76
+
+#define SSC_NETDEV_PROBE		100
+#define SSC_NETDEV_SEND			101
+#define SSC_NETDEV_RECV			102
+#define SSC_NETDEV_ATTACH		103
+#define SSC_NETDEV_DETACH		104
+
+/*
+ * Simulator system call.
+ */
+extern long ia64_ssc (long arg0, long arg1, long arg2, long arg3, int nr);
+
+#endif /* _IA64_PLATFORM_HPSIM_SSC_H */
diff --git a/kernel/arch/ia64/hp/sim/simeth.c b/kernel/arch/ia64/hp/sim/simeth.c
new file mode 100644
index 000000000..d1b04c4c9
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/simeth.c
@@ -0,0 +1,509 @@
+/*
+ * Simulated Ethernet Driver
+ *
+ * Copyright (C) 1999-2001, 2003 Hewlett-Packard Co
+ *	Stephane Eranian <eranian@hpl.hp.com>
+ */
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/types.h>
+#include <linux/in.h>
+#include <linux/string.h>
+#include <linux/init.h>
+#include <linux/errno.h>
+#include <linux/interrupt.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/inetdevice.h>
+#include <linux/if_ether.h>
+#include <linux/if_arp.h>
+#include <linux/skbuff.h>
+#include <linux/notifier.h>
+#include <linux/bitops.h>
+#include <asm/irq.h>
+#include <asm/hpsim.h>
+
+#include "hpsim_ssc.h"
+
+#define SIMETH_RECV_MAX	10
+
+/*
+ * Maximum possible received frame for Ethernet.
+ * We preallocate an sk_buff of that size to avoid costly
+ * memcpy for temporary buffer into sk_buff. We do basically
+ * what's done in other drivers, like eepro with a ring.
+ * The difference is, of course, that we don't have real DMA !!!
+ */
+#define SIMETH_FRAME_SIZE	ETH_FRAME_LEN
+
+
+#define NETWORK_INTR			8
+
+struct simeth_local {
+	struct net_device_stats stats;
+	int 			simfd;	 /* descriptor in the simulator */
+};
+
+static int simeth_probe1(void);
+static int simeth_open(struct net_device *dev);
+static int simeth_close(struct net_device *dev);
+static int simeth_tx(struct sk_buff *skb, struct net_device *dev);
+static int simeth_rx(struct net_device *dev);
+static struct net_device_stats *simeth_get_stats(struct net_device *dev);
+static irqreturn_t simeth_interrupt(int irq, void *dev_id);
+static void set_multicast_list(struct net_device *dev);
+static int simeth_device_event(struct notifier_block *this,unsigned long event, void *ptr);
+
+static char *simeth_version="0.3";
+
+/*
+ * This variable is used to establish a mapping between the Linux/ia64 kernel
+ * and the host linux kernel.
+ *
+ * As of today, we support only one card, even though most of the code
+ * is ready for many more. The mapping is then:
+ *	linux/ia64 -> linux/x86
+ * 	   eth0    -> eth1
+ *
+ * In the future, we some string operations, we could easily support up
+ * to 10 cards (0-9).
+ *
+ * The default mapping can be changed on the kernel command line by
+ * specifying simeth=ethX (or whatever string you want).
+ */
+static char *simeth_device="eth0";	 /* default host interface to use */
+
+
+
+static volatile unsigned int card_count; /* how many cards "found" so far */
+static int simeth_debug;		/* set to 1 to get debug information */
+
+/*
+ * Used to catch IFF_UP & IFF_DOWN events
+ */
+static struct notifier_block simeth_dev_notifier = {
+	simeth_device_event,
+	NULL
+};
+
+
+/*
+ * Function used when using a kernel command line option.
+ *
+ * Format: simeth=interface_name (like eth0)
+ */
+static int __init
+simeth_setup(char *str)
+{
+	simeth_device = str;
+	return 1;
+}
+
+__setup("simeth=", simeth_setup);
+
+/*
+ * Function used to probe for simeth devices when not installed
+ * as a loadable module
+ */
+
+int __init
+simeth_probe (void)
+{
+	int r;
+
+	printk(KERN_INFO "simeth: v%s\n", simeth_version);
+
+	r = simeth_probe1();
+
+	if (r == 0) register_netdevice_notifier(&simeth_dev_notifier);
+
+	return r;
+}
+
+static inline int
+netdev_probe(char *name, unsigned char *ether)
+{
+	return ia64_ssc(__pa(name), __pa(ether), 0,0, SSC_NETDEV_PROBE);
+}
+
+
+static inline int
+netdev_attach(int fd, int irq, unsigned int ipaddr)
+{
+	/* this puts the host interface in the right mode (start interrupting) */
+	return ia64_ssc(fd, ipaddr, 0,0, SSC_NETDEV_ATTACH);
+}
+
+
+static inline int
+netdev_detach(int fd)
+{
+	/*
+	 * inactivate the host interface (don't interrupt anymore) */
+	return ia64_ssc(fd, 0,0,0, SSC_NETDEV_DETACH);
+}
+
+static inline int
+netdev_send(int fd, unsigned char *buf, unsigned int len)
+{
+	return ia64_ssc(fd, __pa(buf), len, 0, SSC_NETDEV_SEND);
+}
+
+static inline int
+netdev_read(int fd, unsigned char *buf, unsigned int len)
+{
+	return ia64_ssc(fd, __pa(buf), len, 0, SSC_NETDEV_RECV);
+}
+
+static const struct net_device_ops simeth_netdev_ops = {
+	.ndo_open		= simeth_open,
+	.ndo_stop		= simeth_close,
+	.ndo_start_xmit		= simeth_tx,
+	.ndo_get_stats		= simeth_get_stats,
+	.ndo_set_rx_mode	= set_multicast_list, /* not yet used */
+
+};
+
+/*
+ * Function shared with module code, so cannot be in init section
+ *
+ * So far this function "detects" only one card (test_&_set) but could
+ * be extended easily.
+ *
+ * Return:
+ * 	- -ENODEV is no device found
+ *	- -ENOMEM is no more memory
+ *	- 0 otherwise
+ */
+static int
+simeth_probe1(void)
+{
+	unsigned char mac_addr[ETH_ALEN];
+	struct simeth_local *local;
+	struct net_device *dev;
+	int fd, err, rc;
+
+	/*
+	 * XXX Fix me
+	 * let's support just one card for now
+	 */
+	if (test_and_set_bit(0, &card_count))
+		return -ENODEV;
+
+	/*
+	 * check with the simulator for the device
+	 */
+	fd = netdev_probe(simeth_device, mac_addr);
+	if (fd == -1)
+		return -ENODEV;
+
+	dev = alloc_etherdev(sizeof(struct simeth_local));
+	if (!dev)
+		return -ENOMEM;
+
+	memcpy(dev->dev_addr, mac_addr, sizeof(mac_addr));
+
+	local = netdev_priv(dev);
+	local->simfd = fd; /* keep track of underlying file descriptor */
+
+	dev->netdev_ops = &simeth_netdev_ops;
+
+	err = register_netdev(dev);
+	if (err) {
+		free_netdev(dev);
+		return err;
+	}
+
+	/*
+	 * attach the interrupt in the simulator, this does enable interrupts
+	 * until a netdev_attach() is called
+	 */
+	if ((rc = hpsim_get_irq(NETWORK_INTR)) < 0)
+		panic("%s: out of interrupt vectors!\n", __func__);
+	dev->irq = rc;
+
+	printk(KERN_INFO "%s: hosteth=%s simfd=%d, HwAddr=%pm, IRQ %d\n",
+	       dev->name, simeth_device, local->simfd, dev->dev_addr, dev->irq);
+
+	return 0;
+}
+
+/*
+ * actually binds the device to an interrupt vector
+ */
+static int
+simeth_open(struct net_device *dev)
+{
+	if (request_irq(dev->irq, simeth_interrupt, 0, "simeth", dev)) {
+		printk(KERN_WARNING "simeth: unable to get IRQ %d.\n", dev->irq);
+		return -EAGAIN;
+	}
+
+	netif_start_queue(dev);
+
+	return 0;
+}
+
+/* copied from lapbether.c */
+static __inline__ int dev_is_ethdev(struct net_device *dev)
+{
+       return ( dev->type == ARPHRD_ETHER && strncmp(dev->name, "dummy", 5));
+}
+
+
+/*
+ * Handler for IFF_UP or IFF_DOWN
+ *
+ * The reason for that is that we don't want to be interrupted when the
+ * interface is down. There is no way to unconnect in the simualtor. Instead
+ * we use this function to shutdown packet processing in the frame filter
+ * in the simulator. Thus no interrupts are generated
+ *
+ *
+ * That's also the place where we pass the IP address of this device to the
+ * simulator so that that we can start filtering packets for it
+ *
+ * There may be a better way of doing this, but I don't know which yet.
+ */
+static int
+simeth_device_event(struct notifier_block *this,unsigned long event, void *ptr)
+{
+	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
+	struct simeth_local *local;
+	struct in_device *in_dev;
+	struct in_ifaddr **ifap = NULL;
+	struct in_ifaddr *ifa = NULL;
+	int r;
+
+
+	if ( ! dev ) {
+		printk(KERN_WARNING "simeth_device_event dev=0\n");
+		return NOTIFY_DONE;
+	}
+
+	if (dev_net(dev) != &init_net)
+		return NOTIFY_DONE;
+
+	if ( event != NETDEV_UP && event != NETDEV_DOWN ) return NOTIFY_DONE;
+
+	/*
+	 * Check whether or not it's for an ethernet device
+	 *
+	 * XXX Fixme: This works only as long as we support one
+	 * type of ethernet device.
+	 */
+	if ( !dev_is_ethdev(dev) ) return NOTIFY_DONE;
+
+	if ((in_dev=dev->ip_ptr) != NULL) {
+		for (ifap=&in_dev->ifa_list; (ifa=*ifap) != NULL; ifap=&ifa->ifa_next)
+			if (strcmp(dev->name, ifa->ifa_label) == 0) break;
+	}
+	if ( ifa == NULL ) {
+		printk(KERN_ERR "simeth_open: can't find device %s's ifa\n", dev->name);
+		return NOTIFY_DONE;
+	}
+
+	printk(KERN_INFO "simeth_device_event: %s ipaddr=0x%x\n",
+	       dev->name, ntohl(ifa->ifa_local));
+
+	/*
+	 * XXX Fix me
+	 * if the device was up, and we're simply reconfiguring it, not sure
+	 * we get DOWN then UP.
+	 */
+
+	local = netdev_priv(dev);
+	/* now do it for real */
+	r = event == NETDEV_UP ?
+		netdev_attach(local->simfd, dev->irq, ntohl(ifa->ifa_local)):
+		netdev_detach(local->simfd);
+
+	printk(KERN_INFO "simeth: netdev_attach/detach: event=%s ->%d\n",
+	       event == NETDEV_UP ? "attach":"detach", r);
+
+	return NOTIFY_DONE;
+}
+
+static int
+simeth_close(struct net_device *dev)
+{
+	netif_stop_queue(dev);
+
+	free_irq(dev->irq, dev);
+
+	return 0;
+}
+
+/*
+ * Only used for debug
+ */
+static void
+frame_print(unsigned char *from, unsigned char *frame, int len)
+{
+	int i;
+
+	printk("%s: (%d) %02x", from, len, frame[0] & 0xff);
+	for(i=1; i < 6; i++ ) {
+		printk(":%02x", frame[i] &0xff);
+	}
+	printk(" %2x", frame[6] &0xff);
+	for(i=7; i < 12; i++ ) {
+		printk(":%02x", frame[i] &0xff);
+	}
+	printk(" [%02x%02x]\n", frame[12], frame[13]);
+
+	for(i=14; i < len; i++ ) {
+		printk("%02x ", frame[i] &0xff);
+		if ( (i%10)==0) printk("\n");
+	}
+	printk("\n");
+}
+
+
+/*
+ * Function used to transmit of frame, very last one on the path before
+ * going to the simulator.
+ */
+static int
+simeth_tx(struct sk_buff *skb, struct net_device *dev)
+{
+	struct simeth_local *local = netdev_priv(dev);
+
+#if 0
+	/* ensure we have at least ETH_ZLEN bytes (min frame size) */
+	unsigned int length = ETH_ZLEN < skb->len ? skb->len : ETH_ZLEN;
+	/* Where do the extra padding bytes comes from inthe skbuff ? */
+#else
+	/* the real driver in the host system is going to take care of that
+	 * or maybe it's the NIC itself.
+	 */
+	unsigned int length = skb->len;
+#endif
+
+	local->stats.tx_bytes += skb->len;
+	local->stats.tx_packets++;
+
+
+	if (simeth_debug > 5) frame_print("simeth_tx", skb->data, length);
+
+	netdev_send(local->simfd, skb->data, length);
+
+	/*
+	 * we are synchronous on write, so we don't simulate a
+	 * trasnmit complete interrupt, thus we don't need to arm a tx
+	 */
+
+	dev_kfree_skb(skb);
+	return NETDEV_TX_OK;
+}
+
+static inline struct sk_buff *
+make_new_skb(struct net_device *dev)
+{
+	struct sk_buff *nskb;
+
+	/*
+	 * The +2 is used to make sure that the IP header is nicely
+	 * aligned (on 4byte boundary I assume 14+2=16)
+	 */
+	nskb = dev_alloc_skb(SIMETH_FRAME_SIZE + 2);
+	if ( nskb == NULL ) {
+		printk(KERN_NOTICE "%s: memory squeeze. dropping packet.\n", dev->name);
+		return NULL;
+	}
+
+	skb_reserve(nskb, 2);	/* Align IP on 16 byte boundaries */
+
+	skb_put(nskb,SIMETH_FRAME_SIZE);
+
+	return nskb;
+}
+
+/*
+ * called from interrupt handler to process a received frame
+ */
+static int
+simeth_rx(struct net_device *dev)
+{
+	struct simeth_local	*local;
+	struct sk_buff		*skb;
+	int			len;
+	int			rcv_count = SIMETH_RECV_MAX;
+
+	local = netdev_priv(dev);
+	/*
+	 * the loop concept has been borrowed from other drivers
+	 * looks to me like it's a throttling thing to avoid pushing to many
+	 * packets at one time into the stack. Making sure we can process them
+	 * upstream and make forward progress overall
+	 */
+	do {
+		if ( (skb=make_new_skb(dev)) == NULL ) {
+			printk(KERN_NOTICE "%s: memory squeeze. dropping packet.\n", dev->name);
+			local->stats.rx_dropped++;
+			return 0;
+		}
+		/*
+		 * Read only one frame at a time
+		 */
+		len = netdev_read(local->simfd, skb->data, SIMETH_FRAME_SIZE);
+		if ( len == 0 ) {
+			if ( simeth_debug > 0 ) printk(KERN_WARNING "%s: count=%d netdev_read=0\n",
+						       dev->name, SIMETH_RECV_MAX-rcv_count);
+			break;
+		}
+#if 0
+		/*
+		 * XXX Fix me
+		 * Should really do a csum+copy here
+		 */
+		skb_copy_to_linear_data(skb, frame, len);
+#endif
+		skb->protocol = eth_type_trans(skb, dev);
+
+		if ( simeth_debug > 6 ) frame_print("simeth_rx", skb->data, len);
+
+		/*
+		 * push the packet up & trigger software interrupt
+		 */
+		netif_rx(skb);
+
+		local->stats.rx_packets++;
+		local->stats.rx_bytes += len;
+
+	} while ( --rcv_count );
+
+	return len; /* 0 = nothing left to read, otherwise, we can try again */
+}
+
+/*
+ * Interrupt handler (Yes, we can do it too !!!)
+ */
+static irqreturn_t
+simeth_interrupt(int irq, void *dev_id)
+{
+	struct net_device *dev = dev_id;
+
+	/*
+	 * very simple loop because we get interrupts only when receiving
+	 */
+	while (simeth_rx(dev));
+	return IRQ_HANDLED;
+}
+
+static struct net_device_stats *
+simeth_get_stats(struct net_device *dev)
+{
+	struct simeth_local *local = netdev_priv(dev);
+
+	return &local->stats;
+}
+
+/* fake multicast ability */
+static void
+set_multicast_list(struct net_device *dev)
+{
+	printk(KERN_WARNING "%s: set_multicast_list called\n", dev->name);
+}
+
+__initcall(simeth_probe);
diff --git a/kernel/arch/ia64/hp/sim/simscsi.c b/kernel/arch/ia64/hp/sim/simscsi.c
new file mode 100644
index 000000000..3a428f19a
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/simscsi.c
@@ -0,0 +1,380 @@
+/*
+ * Simulated SCSI driver.
+ *
+ * Copyright (C) 1999, 2001-2003 Hewlett-Packard Co
+ *	David Mosberger-Tang <davidm@hpl.hp.com>
+ *	Stephane Eranian <eranian@hpl.hp.com>
+ *
+ * 02/01/15 David Mosberger	Updated for v2.5.1
+ * 99/12/18 David Mosberger	Added support for READ10/WRITE10 needed by linux v2.3.33
+ */
+#include <linux/blkdev.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/kernel.h>
+#include <linux/timer.h>
+#include <asm/irq.h>
+#include "hpsim_ssc.h"
+
+#include <scsi/scsi.h>
+#include <scsi/scsi_cmnd.h>
+#include <scsi/scsi_device.h>
+#include <scsi/scsi_host.h>
+
+#define DEBUG_SIMSCSI	0
+
+#define SIMSCSI_REQ_QUEUE_LEN	64
+#define DEFAULT_SIMSCSI_ROOT	"/var/ski-disks/sd"
+
+/* Simulator system calls: */
+
+#define SSC_OPEN			50
+#define SSC_CLOSE			51
+#define SSC_READ			52
+#define SSC_WRITE			53
+#define SSC_GET_COMPLETION		54
+#define SSC_WAIT_COMPLETION		55
+
+#define SSC_WRITE_ACCESS		2
+#define SSC_READ_ACCESS			1
+
+#if DEBUG_SIMSCSI
+  int simscsi_debug;
+# define DBG	simscsi_debug
+#else
+# define DBG	0
+#endif
+
+static struct Scsi_Host *host;
+
+static void simscsi_interrupt (unsigned long val);
+static DECLARE_TASKLET(simscsi_tasklet, simscsi_interrupt, 0);
+
+struct disk_req {
+	unsigned long addr;
+	unsigned len;
+};
+
+struct disk_stat {
+	int fd;
+	unsigned count;
+};
+
+static int desc[16] = {
+	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
+};
+
+static struct queue_entry {
+	struct scsi_cmnd *sc;
+} queue[SIMSCSI_REQ_QUEUE_LEN];
+
+static int rd, wr;
+static atomic_t num_reqs = ATOMIC_INIT(0);
+
+/* base name for default disks */
+static char *simscsi_root = DEFAULT_SIMSCSI_ROOT;
+
+#define MAX_ROOT_LEN	128
+
+/*
+ * used to setup a new base for disk images
+ * to use /foo/bar/disk[a-z] as disk images
+ * you have to specify simscsi=/foo/bar/disk on the command line
+ */
+static int __init
+simscsi_setup (char *s)
+{
+	/* XXX Fix me we may need to strcpy() ? */
+	if (strlen(s) > MAX_ROOT_LEN) {
+		printk(KERN_ERR "simscsi_setup: prefix too long---using default %s\n",
+		       simscsi_root);
+	} else
+		simscsi_root = s;
+	return 1;
+}
+
+__setup("simscsi=", simscsi_setup);
+
+static void
+simscsi_interrupt (unsigned long val)
+{
+	struct scsi_cmnd *sc;
+
+	while ((sc = queue[rd].sc) != NULL) {
+		atomic_dec(&num_reqs);
+		queue[rd].sc = NULL;
+		if (DBG)
+			printk("simscsi_interrupt: done with %ld\n", sc->serial_number);
+		(*sc->scsi_done)(sc);
+		rd = (rd + 1) % SIMSCSI_REQ_QUEUE_LEN;
+	}
+}
+
+static int
+simscsi_biosparam (struct scsi_device *sdev, struct block_device *n,
+		sector_t capacity, int ip[])
+{
+	ip[0] = 64;		/* heads */
+	ip[1] = 32;		/* sectors */
+	ip[2] = capacity >> 11;	/* cylinders */
+	return 0;
+}
+
+static void
+simscsi_sg_readwrite (struct scsi_cmnd *sc, int mode, unsigned long offset)
+{
+	int i;
+	struct scatterlist *sl;
+	struct disk_stat stat;
+	struct disk_req req;
+
+	stat.fd = desc[sc->device->id];
+
+	scsi_for_each_sg(sc, sl, scsi_sg_count(sc), i) {
+		req.addr = __pa(sg_virt(sl));
+		req.len  = sl->length;
+		if (DBG)
+			printk("simscsi_sg_%s @ %lx (off %lx) use_sg=%d len=%d\n",
+			       mode == SSC_READ ? "read":"write", req.addr, offset,
+			       scsi_sg_count(sc) - i, sl->length);
+		ia64_ssc(stat.fd, 1, __pa(&req), offset, mode);
+		ia64_ssc(__pa(&stat), 0, 0, 0, SSC_WAIT_COMPLETION);
+
+		/* should not happen in our case */
+		if (stat.count != req.len) {
+			sc->result = DID_ERROR << 16;
+			return;
+		}
+		offset +=  sl->length;
+	}
+	sc->result = GOOD;
+}
+
+/*
+ * function handling both READ_6/WRITE_6 (non-scatter/gather mode)
+ * commands.
+ * Added 02/26/99 S.Eranian
+ */
+static void
+simscsi_readwrite6 (struct scsi_cmnd *sc, int mode)
+{
+	unsigned long offset;
+
+	offset = (((sc->cmnd[1] & 0x1f) << 16) | (sc->cmnd[2] << 8) | sc->cmnd[3])*512;
+	simscsi_sg_readwrite(sc, mode, offset);
+}
+
+static size_t
+simscsi_get_disk_size (int fd)
+{
+	struct disk_stat stat;
+	size_t bit, sectors = 0;
+	struct disk_req req;
+	char buf[512];
+
+	/*
+	 * This is a bit kludgey: the simulator doesn't provide a
+	 * direct way of determining the disk size, so we do a binary
+	 * search, assuming a maximum disk size of 128GB.
+	 */
+	for (bit = (128UL << 30)/512; bit != 0; bit >>= 1) {
+		req.addr = __pa(&buf);
+		req.len = sizeof(buf);
+		ia64_ssc(fd, 1, __pa(&req), ((sectors | bit) - 1)*512, SSC_READ);
+		stat.fd = fd;
+		ia64_ssc(__pa(&stat), 0, 0, 0, SSC_WAIT_COMPLETION);
+		if (stat.count == sizeof(buf))
+			sectors |= bit;
+	}
+	return sectors - 1;	/* return last valid sector number */
+}
+
+static void
+simscsi_readwrite10 (struct scsi_cmnd *sc, int mode)
+{
+	unsigned long offset;
+
+	offset = (((unsigned long)sc->cmnd[2] << 24) 
+		| ((unsigned long)sc->cmnd[3] << 16)
+		| ((unsigned long)sc->cmnd[4] <<  8) 
+		| ((unsigned long)sc->cmnd[5] <<  0))*512UL;
+	simscsi_sg_readwrite(sc, mode, offset);
+}
+
+static int
+simscsi_queuecommand_lck (struct scsi_cmnd *sc, void (*done)(struct scsi_cmnd *))
+{
+	unsigned int target_id = sc->device->id;
+	char fname[MAX_ROOT_LEN+16];
+	size_t disk_size;
+	char *buf;
+	char localbuf[36];
+#if DEBUG_SIMSCSI
+	register long sp asm ("sp");
+
+	if (DBG)
+		printk("simscsi_queuecommand: target=%d,cmnd=%u,sc=%lu,sp=%lx,done=%p\n",
+		       target_id, sc->cmnd[0], sc->serial_number, sp, done);
+#endif
+
+	sc->result = DID_BAD_TARGET << 16;
+	sc->scsi_done = done;
+	if (target_id <= 15 && sc->device->lun == 0) {
+		switch (sc->cmnd[0]) {
+		      case INQUIRY:
+			if (scsi_bufflen(sc) < 35) {
+				break;
+			}
+			sprintf (fname, "%s%c", simscsi_root, 'a' + target_id);
+			desc[target_id] = ia64_ssc(__pa(fname), SSC_READ_ACCESS|SSC_WRITE_ACCESS,
+						   0, 0, SSC_OPEN);
+			if (desc[target_id] < 0) {
+				/* disk doesn't exist... */
+				break;
+			}
+			buf = localbuf;
+			buf[0] = 0;	/* magnetic disk */
+			buf[1] = 0;	/* not a removable medium */
+			buf[2] = 2;	/* SCSI-2 compliant device */
+			buf[3] = 2;	/* SCSI-2 response data format */
+			buf[4] = 31;	/* additional length (bytes) */
+			buf[5] = 0;	/* reserved */
+			buf[6] = 0;	/* reserved */
+			buf[7] = 0;	/* various flags */
+			memcpy(buf + 8, "HP      SIMULATED DISK  0.00",  28);
+			scsi_sg_copy_from_buffer(sc, buf, 36);
+			sc->result = GOOD;
+			break;
+
+		      case TEST_UNIT_READY:
+			sc->result = GOOD;
+			break;
+
+		      case READ_6:
+			if (desc[target_id] < 0 )
+				break;
+			simscsi_readwrite6(sc, SSC_READ);
+			break;
+
+		      case READ_10:
+			if (desc[target_id] < 0 )
+				break;
+			simscsi_readwrite10(sc, SSC_READ);
+			break;
+
+		      case WRITE_6:
+			if (desc[target_id] < 0)
+				break;
+			simscsi_readwrite6(sc, SSC_WRITE);
+			break;
+
+		      case WRITE_10:
+			if (desc[target_id] < 0)
+				break;
+			simscsi_readwrite10(sc, SSC_WRITE);
+			break;
+
+		      case READ_CAPACITY:
+			if (desc[target_id] < 0 || scsi_bufflen(sc) < 8) {
+				break;
+			}
+			buf = localbuf;
+			disk_size = simscsi_get_disk_size(desc[target_id]);
+
+			buf[0] = (disk_size >> 24) & 0xff;
+			buf[1] = (disk_size >> 16) & 0xff;
+			buf[2] = (disk_size >>  8) & 0xff;
+			buf[3] = (disk_size >>  0) & 0xff;
+			/* set block size of 512 bytes: */
+			buf[4] = 0;
+			buf[5] = 0;
+			buf[6] = 2;
+			buf[7] = 0;
+			scsi_sg_copy_from_buffer(sc, buf, 8);
+			sc->result = GOOD;
+			break;
+
+		      case MODE_SENSE:
+		      case MODE_SENSE_10:
+			/* sd.c uses this to determine whether disk does write-caching. */
+			scsi_sg_copy_from_buffer(sc, (char *)empty_zero_page,
+						 PAGE_SIZE);
+			sc->result = GOOD;
+			break;
+
+		      case START_STOP:
+			printk(KERN_ERR "START_STOP\n");
+			break;
+
+		      default:
+			panic("simscsi: unknown SCSI command %u\n", sc->cmnd[0]);
+		}
+	}
+	if (sc->result == DID_BAD_TARGET) {
+		sc->result |= DRIVER_SENSE << 24;
+		sc->sense_buffer[0] = 0x70;
+		sc->sense_buffer[2] = 0x00;
+	}
+	if (atomic_read(&num_reqs) >= SIMSCSI_REQ_QUEUE_LEN) {
+		panic("Attempt to queue command while command is pending!!");
+	}
+	atomic_inc(&num_reqs);
+	queue[wr].sc = sc;
+	wr = (wr + 1) % SIMSCSI_REQ_QUEUE_LEN;
+
+	tasklet_schedule(&simscsi_tasklet);
+	return 0;
+}
+
+static DEF_SCSI_QCMD(simscsi_queuecommand)
+
+static int
+simscsi_host_reset (struct scsi_cmnd *sc)
+{
+	printk(KERN_ERR "simscsi_host_reset: not implemented\n");
+	return 0;
+}
+
+static struct scsi_host_template driver_template = {
+	.name			= "simulated SCSI host adapter",
+	.proc_name		= "simscsi",
+	.queuecommand		= simscsi_queuecommand,
+	.eh_host_reset_handler	= simscsi_host_reset,
+	.bios_param		= simscsi_biosparam,
+	.can_queue		= SIMSCSI_REQ_QUEUE_LEN,
+	.this_id		= -1,
+	.sg_tablesize		= SG_ALL,
+	.max_sectors		= 1024,
+	.cmd_per_lun		= SIMSCSI_REQ_QUEUE_LEN,
+	.use_clustering		= DISABLE_CLUSTERING,
+};
+
+static int __init
+simscsi_init(void)
+{
+	int error;
+
+	host = scsi_host_alloc(&driver_template, 0);
+	if (!host)
+		return -ENOMEM;
+
+	error = scsi_add_host(host, NULL);
+	if (error)
+		goto free_host;
+	scsi_scan_host(host);
+	return 0;
+
+ free_host:
+	scsi_host_put(host);
+	return error;
+}
+
+static void __exit
+simscsi_exit(void)
+{
+	scsi_remove_host(host);
+	scsi_host_put(host);
+}
+
+module_init(simscsi_init);
+module_exit(simscsi_exit);
diff --git a/kernel/arch/ia64/hp/sim/simserial.c b/kernel/arch/ia64/hp/sim/simserial.c
new file mode 100644
index 000000000..e70cadec7
--- /dev/null
+++ b/kernel/arch/ia64/hp/sim/simserial.c
@@ -0,0 +1,545 @@
+/*
+ * Simulated Serial Driver (fake serial)
+ *
+ * This driver is mostly used for bringup purposes and will go away.
+ * It has a strong dependency on the system console. All outputs
+ * are rerouted to the same facility as the one used by printk which, in our
+ * case means sys_sim.c console (goes via the simulator).
+ *
+ * Copyright (C) 1999-2000, 2002-2003 Hewlett-Packard Co
+ *	Stephane Eranian <eranian@hpl.hp.com>
+ *	David Mosberger-Tang <davidm@hpl.hp.com>
+ */
+
+#include <linux/init.h>
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/tty.h>
+#include <linux/tty_flip.h>
+#include <linux/major.h>
+#include <linux/fcntl.h>
+#include <linux/mm.h>
+#include <linux/seq_file.h>
+#include <linux/slab.h>
+#include <linux/capability.h>
+#include <linux/circ_buf.h>
+#include <linux/console.h>
+#include <linux/irq.h>
+#include <linux/module.h>
+#include <linux/serial.h>
+#include <linux/sysrq.h>
+#include <linux/uaccess.h>
+
+#include <asm/hpsim.h>
+
+#include "hpsim_ssc.h"
+
+#undef SIMSERIAL_DEBUG	/* define this to get some debug information */
+
+#define KEYBOARD_INTR	3	/* must match with simulator! */
+
+#define NR_PORTS	1	/* only one port for now */
+
+struct serial_state {
+	struct tty_port port;
+	struct circ_buf xmit;
+	int irq;
+	int x_char;
+};
+
+static struct serial_state rs_table[NR_PORTS];
+
+struct tty_driver *hp_simserial_driver;
+
+static struct console *console;
+
+static void receive_chars(struct tty_port *port)
+{
+	unsigned char ch;
+	static unsigned char seen_esc = 0;
+
+	while ( (ch = ia64_ssc(0, 0, 0, 0, SSC_GETCHAR)) ) {
+		if (ch == 27 && seen_esc == 0) {
+			seen_esc = 1;
+			continue;
+		} else if (seen_esc == 1 && ch == 'O') {
+			seen_esc = 2;
+			continue;
+		} else if (seen_esc == 2) {
+			if (ch == 'P') /* F1 */
+				show_state();
+#ifdef CONFIG_MAGIC_SYSRQ
+			if (ch == 'S') { /* F4 */
+				do {
+					ch = ia64_ssc(0, 0, 0, 0, SSC_GETCHAR);
+				} while (!ch);
+				handle_sysrq(ch);
+			}
+#endif
+			seen_esc = 0;
+			continue;
+		}
+		seen_esc = 0;
+
+		if (tty_insert_flip_char(port, ch, TTY_NORMAL) == 0)
+			break;
+	}
+	tty_flip_buffer_push(port);
+}
+
+/*
+ * This is the serial driver's interrupt routine for a single port
+ */
+static irqreturn_t rs_interrupt_single(int irq, void *dev_id)
+{
+	struct serial_state *info = dev_id;
+
+	receive_chars(&info->port);
+
+	return IRQ_HANDLED;
+}
+
+/*
+ * -------------------------------------------------------------------
+ * Here ends the serial interrupt routines.
+ * -------------------------------------------------------------------
+ */
+
+static int rs_put_char(struct tty_struct *tty, unsigned char ch)
+{
+	struct serial_state *info = tty->driver_data;
+	unsigned long flags;
+
+	if (!info->xmit.buf)
+		return 0;
+
+	local_irq_save(flags);
+	if (CIRC_SPACE(info->xmit.head, info->xmit.tail, SERIAL_XMIT_SIZE) == 0) {
+		local_irq_restore(flags);
+		return 0;
+	}
+	info->xmit.buf[info->xmit.head] = ch;
+	info->xmit.head = (info->xmit.head + 1) & (SERIAL_XMIT_SIZE-1);
+	local_irq_restore(flags);
+	return 1;
+}
+
+static void transmit_chars(struct tty_struct *tty, struct serial_state *info,
+		int *intr_done)
+{
+	int count;
+	unsigned long flags;
+
+	local_irq_save(flags);
+
+	if (info->x_char) {
+		char c = info->x_char;
+
+		console->write(console, &c, 1);
+
+		info->x_char = 0;
+
+		goto out;
+	}
+
+	if (info->xmit.head == info->xmit.tail || tty->stopped) {
+#ifdef SIMSERIAL_DEBUG
+		printk("transmit_chars: head=%d, tail=%d, stopped=%d\n",
+		       info->xmit.head, info->xmit.tail, tty->stopped);
+#endif
+		goto out;
+	}
+	/*
+	 * We removed the loop and try to do it in to chunks. We need
+	 * 2 operations maximum because it's a ring buffer.
+	 *
+	 * First from current to tail if possible.
+	 * Then from the beginning of the buffer until necessary
+	 */
+
+	count = min(CIRC_CNT(info->xmit.head, info->xmit.tail, SERIAL_XMIT_SIZE),
+		    SERIAL_XMIT_SIZE - info->xmit.tail);
+	console->write(console, info->xmit.buf+info->xmit.tail, count);
+
+	info->xmit.tail = (info->xmit.tail+count) & (SERIAL_XMIT_SIZE-1);
+
+	/*
+	 * We have more at the beginning of the buffer
+	 */
+	count = CIRC_CNT(info->xmit.head, info->xmit.tail, SERIAL_XMIT_SIZE);
+	if (count) {
+		console->write(console, info->xmit.buf, count);
+		info->xmit.tail += count;
+	}
+out:
+	local_irq_restore(flags);
+}
+
+static void rs_flush_chars(struct tty_struct *tty)
+{
+	struct serial_state *info = tty->driver_data;
+
+	if (info->xmit.head == info->xmit.tail || tty->stopped ||
+			!info->xmit.buf)
+		return;
+
+	transmit_chars(tty, info, NULL);
+}
+
+static int rs_write(struct tty_struct * tty,
+		    const unsigned char *buf, int count)
+{
+	struct serial_state *info = tty->driver_data;
+	int	c, ret = 0;
+	unsigned long flags;
+
+	if (!info->xmit.buf)
+		return 0;
+
+	local_irq_save(flags);
+	while (1) {
+		c = CIRC_SPACE_TO_END(info->xmit.head, info->xmit.tail, SERIAL_XMIT_SIZE);
+		if (count < c)
+			c = count;
+		if (c <= 0) {
+			break;
+		}
+		memcpy(info->xmit.buf + info->xmit.head, buf, c);
+		info->xmit.head = ((info->xmit.head + c) &
+				   (SERIAL_XMIT_SIZE-1));
+		buf += c;
+		count -= c;
+		ret += c;
+	}
+	local_irq_restore(flags);
+	/*
+	 * Hey, we transmit directly from here in our case
+	 */
+	if (CIRC_CNT(info->xmit.head, info->xmit.tail, SERIAL_XMIT_SIZE) &&
+			!tty->stopped)
+		transmit_chars(tty, info, NULL);
+
+	return ret;
+}
+
+static int rs_write_room(struct tty_struct *tty)
+{
+	struct serial_state *info = tty->driver_data;
+
+	return CIRC_SPACE(info->xmit.head, info->xmit.tail, SERIAL_XMIT_SIZE);
+}
+
+static int rs_chars_in_buffer(struct tty_struct *tty)
+{
+	struct serial_state *info = tty->driver_data;
+
+	return CIRC_CNT(info->xmit.head, info->xmit.tail, SERIAL_XMIT_SIZE);
+}
+
+static void rs_flush_buffer(struct tty_struct *tty)
+{
+	struct serial_state *info = tty->driver_data;
+	unsigned long flags;
+
+	local_irq_save(flags);
+	info->xmit.head = info->xmit.tail = 0;
+	local_irq_restore(flags);
+
+	tty_wakeup(tty);
+}
+
+/*
+ * This function is used to send a high-priority XON/XOFF character to
+ * the device
+ */
+static void rs_send_xchar(struct tty_struct *tty, char ch)
+{
+	struct serial_state *info = tty->driver_data;
+
+	info->x_char = ch;
+	if (ch) {
+		/*
+		 * I guess we could call console->write() directly but
+		 * let's do that for now.
+		 */
+		transmit_chars(tty, info, NULL);
+	}
+}
+
+/*
+ * ------------------------------------------------------------
+ * rs_throttle()
+ *
+ * This routine is called by the upper-layer tty layer to signal that
+ * incoming characters should be throttled.
+ * ------------------------------------------------------------
+ */
+static void rs_throttle(struct tty_struct * tty)
+{
+	if (I_IXOFF(tty))
+		rs_send_xchar(tty, STOP_CHAR(tty));
+
+	printk(KERN_INFO "simrs_throttle called\n");
+}
+
+static void rs_unthrottle(struct tty_struct * tty)
+{
+	struct serial_state *info = tty->driver_data;
+
+	if (I_IXOFF(tty)) {
+		if (info->x_char)
+			info->x_char = 0;
+		else
+			rs_send_xchar(tty, START_CHAR(tty));
+	}
+	printk(KERN_INFO "simrs_unthrottle called\n");
+}
+
+static int rs_ioctl(struct tty_struct *tty, unsigned int cmd, unsigned long arg)
+{
+	if ((cmd != TIOCGSERIAL) && (cmd != TIOCSSERIAL) &&
+	    (cmd != TIOCSERCONFIG) && (cmd != TIOCSERGSTRUCT) &&
+	    (cmd != TIOCMIWAIT)) {
+		if (tty->flags & (1 << TTY_IO_ERROR))
+		    return -EIO;
+	}
+
+	switch (cmd) {
+	case TIOCGSERIAL:
+	case TIOCSSERIAL:
+	case TIOCSERGSTRUCT:
+	case TIOCMIWAIT:
+		return 0;
+	case TIOCSERCONFIG:
+	case TIOCSERGETLSR: /* Get line status register */
+		return -EINVAL;
+	case TIOCSERGWILD:
+	case TIOCSERSWILD:
+		/* "setserial -W" is called in Debian boot */
+		printk (KERN_INFO "TIOCSER?WILD ioctl obsolete, ignored.\n");
+		return 0;
+	}
+	return -ENOIOCTLCMD;
+}
+
+#define RELEVANT_IFLAG(iflag) (iflag & (IGNBRK|BRKINT|IGNPAR|PARMRK|INPCK))
+
+/*
+ * This routine will shutdown a serial port; interrupts are disabled, and
+ * DTR is dropped if the hangup on close termio flag is on.
+ */
+static void shutdown(struct tty_port *port)
+{
+	struct serial_state *info = container_of(port, struct serial_state,
+			port);
+	unsigned long flags;
+
+	local_irq_save(flags);
+	if (info->irq)
+		free_irq(info->irq, info);
+
+	if (info->xmit.buf) {
+		free_page((unsigned long) info->xmit.buf);
+		info->xmit.buf = NULL;
+	}
+	local_irq_restore(flags);
+}
+
+static void rs_close(struct tty_struct *tty, struct file * filp)
+{
+	struct serial_state *info = tty->driver_data;
+
+	tty_port_close(&info->port, tty, filp);
+}
+
+static void rs_hangup(struct tty_struct *tty)
+{
+	struct serial_state *info = tty->driver_data;
+
+	rs_flush_buffer(tty);
+	tty_port_hangup(&info->port);
+}
+
+static int activate(struct tty_port *port, struct tty_struct *tty)
+{
+	struct serial_state *state = container_of(port, struct serial_state,
+			port);
+	unsigned long flags, page;
+	int retval = 0;
+
+	page = get_zeroed_page(GFP_KERNEL);
+	if (!page)
+		return -ENOMEM;
+
+	local_irq_save(flags);
+
+	if (state->xmit.buf)
+		free_page(page);
+	else
+		state->xmit.buf = (unsigned char *) page;
+
+	if (state->irq) {
+		retval = request_irq(state->irq, rs_interrupt_single, 0,
+				"simserial", state);
+		if (retval)
+			goto errout;
+	}
+
+	state->xmit.head = state->xmit.tail = 0;
+
+	/*
+	 * Set up the tty->alt_speed kludge
+	 */
+	if ((port->flags & ASYNC_SPD_MASK) == ASYNC_SPD_HI)
+		tty->alt_speed = 57600;
+	if ((port->flags & ASYNC_SPD_MASK) == ASYNC_SPD_VHI)
+		tty->alt_speed = 115200;
+	if ((port->flags & ASYNC_SPD_MASK) == ASYNC_SPD_SHI)
+		tty->alt_speed = 230400;
+	if ((port->flags & ASYNC_SPD_MASK) == ASYNC_SPD_WARP)
+		tty->alt_speed = 460800;
+
+errout:
+	local_irq_restore(flags);
+	return retval;
+}
+
+
+/*
+ * This routine is called whenever a serial port is opened.  It
+ * enables interrupts for a serial port, linking in its async structure into
+ * the IRQ chain.   It also performs the serial-specific
+ * initialization for the tty structure.
+ */
+static int rs_open(struct tty_struct *tty, struct file * filp)
+{
+	struct serial_state *info = rs_table + tty->index;
+	struct tty_port *port = &info->port;
+
+	tty->driver_data = info;
+	port->low_latency = (port->flags & ASYNC_LOW_LATENCY) ? 1 : 0;
+
+	/*
+	 * figure out which console to use (should be one already)
+	 */
+	console = console_drivers;
+	while (console) {
+		if ((console->flags & CON_ENABLED) && console->write) break;
+		console = console->next;
+	}
+
+	return tty_port_open(port, tty, filp);
+}
+
+/*
+ * /proc fs routines....
+ */
+
+static int rs_proc_show(struct seq_file *m, void *v)
+{
+	int i;
+
+	seq_printf(m, "simserinfo:1.0\n");
+	for (i = 0; i < NR_PORTS; i++)
+		seq_printf(m, "%d: uart:16550 port:3F8 irq:%d\n",
+		       i, rs_table[i].irq);
+	return 0;
+}
+
+static int rs_proc_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, rs_proc_show, NULL);
+}
+
+static const struct file_operations rs_proc_fops = {
+	.owner		= THIS_MODULE,
+	.open		= rs_proc_open,
+	.read		= seq_read,
+	.llseek		= seq_lseek,
+	.release	= single_release,
+};
+
+static const struct tty_operations hp_ops = {
+	.open = rs_open,
+	.close = rs_close,
+	.write = rs_write,
+	.put_char = rs_put_char,
+	.flush_chars = rs_flush_chars,
+	.write_room = rs_write_room,
+	.chars_in_buffer = rs_chars_in_buffer,
+	.flush_buffer = rs_flush_buffer,
+	.ioctl = rs_ioctl,
+	.throttle = rs_throttle,
+	.unthrottle = rs_unthrottle,
+	.send_xchar = rs_send_xchar,
+	.hangup = rs_hangup,
+	.proc_fops = &rs_proc_fops,
+};
+
+static const struct tty_port_operations hp_port_ops = {
+	.activate = activate,
+	.shutdown = shutdown,
+};
+
+static int __init simrs_init(void)
+{
+	struct serial_state *state;
+	int retval;
+
+	if (!ia64_platform_is("hpsim"))
+		return -ENODEV;
+
+	hp_simserial_driver = alloc_tty_driver(NR_PORTS);
+	if (!hp_simserial_driver)
+		return -ENOMEM;
+
+	printk(KERN_INFO "SimSerial driver with no serial options enabled\n");
+
+	/* Initialize the tty_driver structure */
+
+	hp_simserial_driver->driver_name = "simserial";
+	hp_simserial_driver->name = "ttyS";
+	hp_simserial_driver->major = TTY_MAJOR;
+	hp_simserial_driver->minor_start = 64;
+	hp_simserial_driver->type = TTY_DRIVER_TYPE_SERIAL;
+	hp_simserial_driver->subtype = SERIAL_TYPE_NORMAL;
+	hp_simserial_driver->init_termios = tty_std_termios;
+	hp_simserial_driver->init_termios.c_cflag =
+		B9600 | CS8 | CREAD | HUPCL | CLOCAL;
+	hp_simserial_driver->flags = TTY_DRIVER_REAL_RAW;
+	tty_set_operations(hp_simserial_driver, &hp_ops);
+
+	state = rs_table;
+	tty_port_init(&state->port);
+	state->port.ops = &hp_port_ops;
+	state->port.close_delay = 0; /* XXX really 0? */
+
+	retval = hpsim_get_irq(KEYBOARD_INTR);
+	if (retval < 0) {
+		printk(KERN_ERR "%s: out of interrupt vectors!\n",
+				__func__);
+		goto err_free_tty;
+	}
+
+	state->irq = retval;
+
+	/* the port is imaginary */
+	printk(KERN_INFO "ttyS0 at 0x03f8 (irq = %d) is a 16550\n", state->irq);
+
+	tty_port_link_device(&state->port, hp_simserial_driver, 0);
+	retval = tty_register_driver(hp_simserial_driver);
+	if (retval) {
+		printk(KERN_ERR "Couldn't register simserial driver\n");
+		goto err_free_tty;
+	}
+
+	return 0;
+err_free_tty:
+	put_tty_driver(hp_simserial_driver);
+	tty_port_destroy(&state->port);
+	return retval;
+}
+
+#ifndef MODULE
+__initcall(simrs_init);
+#endif
diff --git a/kernel/arch/ia64/hp/zx1/Makefile b/kernel/arch/ia64/hp/zx1/Makefile
new file mode 100644
index 000000000..61e878729
--- /dev/null
+++ b/kernel/arch/ia64/hp/zx1/Makefile
@@ -0,0 +1,8 @@
+#
+# ia64/hp/zx1/Makefile
+#
+# Copyright (C) 2002 Hewlett Packard
+# Copyright (C) Alex Williamson (alex_williamson@hp.com)
+#
+
+obj-$(CONFIG_IA64_GENERIC) += hpzx1_machvec.o hpzx1_swiotlb_machvec.o
diff --git a/kernel/arch/ia64/hp/zx1/hpzx1_machvec.c b/kernel/arch/ia64/hp/zx1/hpzx1_machvec.c
new file mode 100644
index 000000000..32518b0f9
--- /dev/null
+++ b/kernel/arch/ia64/hp/zx1/hpzx1_machvec.c
@@ -0,0 +1,3 @@
+#define MACHVEC_PLATFORM_NAME		hpzx1
+#define MACHVEC_PLATFORM_HEADER		<asm/machvec_hpzx1.h>
+#include <asm/machvec_init.h>
diff --git a/kernel/arch/ia64/hp/zx1/hpzx1_swiotlb_machvec.c b/kernel/arch/ia64/hp/zx1/hpzx1_swiotlb_machvec.c
new file mode 100644
index 000000000..4392a96b3
--- /dev/null
+++ b/kernel/arch/ia64/hp/zx1/hpzx1_swiotlb_machvec.c
@@ -0,0 +1,3 @@
+#define MACHVEC_PLATFORM_NAME		hpzx1_swiotlb
+#define MACHVEC_PLATFORM_HEADER		<asm/machvec_hpzx1_swiotlb.h>
+#include <asm/machvec_init.h>