Auto-update from upstream

17 files changed, 343 insertions, 241 deletions

diff --git a/arch/ia64/Kconfig b/arch/ia64/Kconfig
index cd2051f5c9ce..18318749884b 100644
--- a/arch/ia64/Kconfig
+++ b/arch/ia64/Kconfig
@@ -77,6 +77,7 @@ choice
 config IA64_GENERIC
 	bool "generic"
 	select ACPI
+	select PCI
 	select NUMA
 	select ACPI_NUMA
 	help
@@ -273,7 +274,6 @@ config HOTPLUG_CPU
 config SCHED_SMT
 	bool "SMT scheduler support"
 	depends on SMP
-	default off
 	help
 	  Improves the CPU scheduler's decision making when dealing with
 	  Intel IA64 chips with MultiThreading at a cost of slightly increased
diff --git a/arch/ia64/hp/common/sba_iommu.c b/arch/ia64/hp/common/sba_iommu.c
index bdccd0b1eb60..5825ddee58d6 100644
--- a/arch/ia64/hp/common/sba_iommu.c
+++ b/arch/ia64/hp/common/sba_iommu.c
@@ -1958,7 +1958,7 @@ sba_map_ioc_to_node(struct ioc *ioc, acpi_handle handle)
 	if (pxm < 0)
 		return;
 
-	node = pxm_to_nid_map[pxm];
+	node = pxm_to_node(pxm);
 
 	if (node >= MAX_NUMNODES || !node_online(node))
 		return;
@@ -1999,7 +1999,7 @@ acpi_sba_ioc_add(struct acpi_device *device)
 		if (!iovp_shift)
 			iovp_shift = min(PAGE_SHIFT, 16);
 	}
-	ACPI_MEM_FREE(dev_info);
+	kfree(dev_info);
 
 	/*
 	 * default anything not caught above or specified on cmdline to 4k
diff --git a/arch/ia64/kernel/acpi.c b/arch/ia64/kernel/acpi.c
index 58c93a30348c..6ea642beaaee 100644
--- a/arch/ia64/kernel/acpi.c
+++ b/arch/ia64/kernel/acpi.c
@@ -68,8 +68,6 @@ EXPORT_SYMBOL(pm_power_off);
 unsigned char acpi_kbd_controller_present = 1;
 unsigned char acpi_legacy_devices;
 
-static unsigned int __initdata acpi_madt_rev;
-
 unsigned int acpi_cpei_override;
 unsigned int acpi_cpei_phys_cpuid;
 
@@ -243,6 +241,8 @@ acpi_parse_iosapic(acpi_table_entry_header * header, const unsigned long end)
 	return iosapic_init(iosapic->address, iosapic->global_irq_base);
 }
 
+static unsigned int __initdata acpi_madt_rev;
+
 static int __init
 acpi_parse_plat_int_src(acpi_table_entry_header * header,
 			const unsigned long end)
@@ -415,9 +415,6 @@ static int __initdata srat_num_cpus;	/* number of cpus */
 static u32 __devinitdata pxm_flag[PXM_FLAG_LEN];
 #define pxm_bit_set(bit)	(set_bit(bit,(void *)pxm_flag))
 #define pxm_bit_test(bit)	(test_bit(bit,(void *)pxm_flag))
-/* maps to convert between proximity domain and logical node ID */
-int __devinitdata pxm_to_nid_map[MAX_PXM_DOMAINS];
-int __initdata nid_to_pxm_map[MAX_NUMNODES];
 static struct acpi_table_slit __initdata *slit_table;
 
 static int get_processor_proximity_domain(struct acpi_table_processor_affinity *pa)
@@ -533,22 +530,17 @@ void __init acpi_numa_arch_fixup(void)
 	 * MCD - This can probably be dropped now.  No need for pxm ID to node ID
 	 * mapping with sparse node numbering iff MAX_PXM_DOMAINS <= MAX_NUMNODES.
 	 */
-	/* calculate total number of nodes in system from PXM bitmap */
-	memset(pxm_to_nid_map, -1, sizeof(pxm_to_nid_map));
-	memset(nid_to_pxm_map, -1, sizeof(nid_to_pxm_map));
 	nodes_clear(node_online_map);
 	for (i = 0; i < MAX_PXM_DOMAINS; i++) {
 		if (pxm_bit_test(i)) {
-			int nid = num_online_nodes();
-			pxm_to_nid_map[i] = nid;
-			nid_to_pxm_map[nid] = i;
+			int nid = acpi_map_pxm_to_node(i);
 			node_set_online(nid);
 		}
 	}
 
 	/* set logical node id in memory chunk structure */
 	for (i = 0; i < num_node_memblks; i++)
-		node_memblk[i].nid = pxm_to_nid_map[node_memblk[i].nid];
+		node_memblk[i].nid = pxm_to_node(node_memblk[i].nid);
 
 	/* assign memory bank numbers for each chunk on each node */
 	for_each_online_node(i) {
@@ -562,7 +554,7 @@ void __init acpi_numa_arch_fixup(void)
 
 	/* set logical node id in cpu structure */
 	for (i = 0; i < srat_num_cpus; i++)
-		node_cpuid[i].nid = pxm_to_nid_map[node_cpuid[i].nid];
+		node_cpuid[i].nid = pxm_to_node(node_cpuid[i].nid);
 
 	printk(KERN_INFO "Number of logical nodes in system = %d\n",
 	       num_online_nodes());
@@ -575,11 +567,11 @@ void __init acpi_numa_arch_fixup(void)
 	for (i = 0; i < slit_table->localities; i++) {
 		if (!pxm_bit_test(i))
 			continue;
-		node_from = pxm_to_nid_map[i];
+		node_from = pxm_to_node(i);
 		for (j = 0; j < slit_table->localities; j++) {
 			if (!pxm_bit_test(j))
 				continue;
-			node_to = pxm_to_nid_map[j];
+			node_to = pxm_to_node(j);
 			node_distance(node_from, node_to) =
 			    slit_table->entry[i * slit_table->localities + j];
 		}
@@ -785,9 +777,9 @@ int acpi_map_cpu2node(acpi_handle handle, int cpu, long physid)
 
 	/*
 	 * Assuming that the container driver would have set the proximity
-	 * domain and would have initialized pxm_to_nid_map[pxm_id] && pxm_flag
+	 * domain and would have initialized pxm_to_node(pxm_id) && pxm_flag
 	 */
-	node_cpuid[cpu].nid = (pxm_id < 0) ? 0 : pxm_to_nid_map[pxm_id];
+	node_cpuid[cpu].nid = (pxm_id < 0) ? 0 : pxm_to_node(pxm_id);
 
 	node_cpuid[cpu].phys_id = physid;
 #endif
@@ -966,7 +958,7 @@ acpi_map_iosapic(acpi_handle handle, u32 depth, void *context, void **ret)
 	if (pxm < 0)
 		return AE_OK;
 
-	node = pxm_to_nid_map[pxm];
+	node = pxm_to_node(pxm);
 
 	if (node >= MAX_NUMNODES || !node_online(node) ||
 	    cpus_empty(node_to_cpumask(node)))
diff --git a/arch/ia64/kernel/entry.S b/arch/ia64/kernel/entry.S
index bcb80ca5cf40..32c999f58d12 100644
--- a/arch/ia64/kernel/entry.S
+++ b/arch/ia64/kernel/entry.S
@@ -1584,7 +1584,7 @@ sys_call_table:
 	data8 sys_keyctl
 	data8 sys_ioprio_set
 	data8 sys_ioprio_get			// 1275
-	data8 sys_ni_syscall
+	data8 sys_move_pages
 	data8 sys_inotify_init
 	data8 sys_inotify_add_watch
 	data8 sys_inotify_rm_watch
diff --git a/arch/ia64/kernel/irq_ia64.c b/arch/ia64/kernel/irq_ia64.c
index 6c4d59fd0364..ef9a2b49307a 100644
--- a/arch/ia64/kernel/irq_ia64.c
+++ b/arch/ia64/kernel/irq_ia64.c
@@ -46,6 +46,10 @@
 
 #define IRQ_DEBUG	0
 
+/* These can be overridden in platform_irq_init */
+int ia64_first_device_vector = IA64_DEF_FIRST_DEVICE_VECTOR;
+int ia64_last_device_vector = IA64_DEF_LAST_DEVICE_VECTOR;
+
 /* default base addr of IPI table */
 void __iomem *ipi_base_addr = ((void __iomem *)
 			       (__IA64_UNCACHED_OFFSET | IA64_IPI_DEFAULT_BASE_ADDR));
@@ -60,7 +64,7 @@ __u8 isa_irq_to_vector_map[16] = {
 };
 EXPORT_SYMBOL(isa_irq_to_vector_map);
 
-static unsigned long ia64_vector_mask[BITS_TO_LONGS(IA64_NUM_DEVICE_VECTORS)];
+static unsigned long ia64_vector_mask[BITS_TO_LONGS(IA64_MAX_DEVICE_VECTORS)];
 
 int
 assign_irq_vector (int irq)
@@ -89,6 +93,19 @@ free_irq_vector (int vector)
 		printk(KERN_WARNING "%s: double free!\n", __FUNCTION__);
 }
 
+int
+reserve_irq_vector (int vector)
+{
+	int pos;
+
+	if (vector < IA64_FIRST_DEVICE_VECTOR ||
+	    vector > IA64_LAST_DEVICE_VECTOR)
+		return -EINVAL;
+
+	pos = vector - IA64_FIRST_DEVICE_VECTOR;
+	return test_and_set_bit(pos, ia64_vector_mask);
+}
+
 #ifdef CONFIG_SMP
 #	define IS_RESCHEDULE(vec)	(vec == IA64_IPI_RESCHEDULE)
 #else
diff --git a/arch/ia64/kernel/perfmon.c b/arch/ia64/kernel/perfmon.c
index 077f21216b65..6d7bc8ff7b3a 100644
--- a/arch/ia64/kernel/perfmon.c
+++ b/arch/ia64/kernel/perfmon.c
@@ -532,7 +532,6 @@ static ctl_table pfm_sysctl_root[] = {
 static struct ctl_table_header *pfm_sysctl_header;
 
 static int pfm_context_unload(pfm_context_t *ctx, void *arg, int count, struct pt_regs *regs);
-static int pfm_flush(struct file *filp);
 
 #define pfm_get_cpu_var(v)		__ia64_per_cpu_var(v)
 #define pfm_get_cpu_data(a,b)		per_cpu(a, b)
@@ -595,10 +594,11 @@ pfm_get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
 }
 
 
-static struct super_block *
-pfmfs_get_sb(struct file_system_type *fs_type, int flags, const char *dev_name, void *data)
+static int
+pfmfs_get_sb(struct file_system_type *fs_type, int flags, const char *dev_name, void *data,
+	     struct vfsmount *mnt)
 {
-	return get_sb_pseudo(fs_type, "pfm:", NULL, PFMFS_MAGIC);
+	return get_sb_pseudo(fs_type, "pfm:", NULL, PFMFS_MAGIC, mnt);
 }
 
 static struct file_system_type pfm_fs_type = {
@@ -1773,7 +1773,7 @@ pfm_syswide_cleanup_other_cpu(pfm_context_t *ctx)
  * When caller is self-monitoring, the context is unloaded.
  */
 static int
-pfm_flush(struct file *filp)
+pfm_flush(struct file *filp, fl_owner_t id)
 {
 	pfm_context_t *ctx;
 	struct task_struct *task;
diff --git a/arch/ia64/kernel/uncached.c b/arch/ia64/kernel/uncached.c
index fcd2bad0286f..5f03b9e524dd 100644
--- a/arch/ia64/kernel/uncached.c
+++ b/arch/ia64/kernel/uncached.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2001-2005 Silicon Graphics, Inc.  All rights reserved.
+ * Copyright (C) 2001-2006 Silicon Graphics, Inc.  All rights reserved.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms of version 2 of the GNU General Public License
@@ -29,15 +29,8 @@
 #include <asm/tlbflush.h>
 #include <asm/sn/arch.h>
 
-#define DEBUG	0
 
-#if DEBUG
-#define dprintk			printk
-#else
-#define dprintk(x...)		do { } while (0)
-#endif
-
-void __init efi_memmap_walk_uc (efi_freemem_callback_t callback);
+extern void __init efi_memmap_walk_uc(efi_freemem_callback_t, void *);
 
 #define MAX_UNCACHED_GRANULES	5
 static int allocated_granules;
@@ -60,6 +53,7 @@ static void uncached_ipi_visibility(void *data)
 static void uncached_ipi_mc_drain(void *data)
 {
 	int status;
+
 	status = ia64_pal_mc_drain();
 	if (status)
 		printk(KERN_WARNING "ia64_pal_mc_drain() failed with %i on "
@@ -67,30 +61,35 @@ static void uncached_ipi_mc_drain(void *data)
 }
 
 
-static unsigned long
-uncached_get_new_chunk(struct gen_pool *poolp)
+/*
+ * Add a new chunk of uncached memory pages to the specified pool.
+ *
+ * @pool: pool to add new chunk of uncached memory to
+ * @nid: node id of node to allocate memory from, or -1
+ *
+ * This is accomplished by first allocating a granule of cached memory pages
+ * and then converting them to uncached memory pages.
+ */
+static int uncached_add_chunk(struct gen_pool *pool, int nid)
 {
 	struct page *page;
-	void *tmp;
 	int status, i;
-	unsigned long addr, node;
+	unsigned long c_addr, uc_addr;
 
 	if (allocated_granules >= MAX_UNCACHED_GRANULES)
-		return 0;
+		return -1;
+
+	/* attempt to allocate a granule's worth of cached memory pages */
 
-	node = poolp->private;
-	page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO,
+	page = alloc_pages_node(nid, GFP_KERNEL | __GFP_ZERO,
 				IA64_GRANULE_SHIFT-PAGE_SHIFT);
+	if (!page)
+		return -1;
 
-	dprintk(KERN_INFO "get_new_chunk page %p, addr %lx\n",
-		page, (unsigned long)(page-vmem_map) << PAGE_SHIFT);
+	/* convert the memory pages from cached to uncached */
 
-	/*
-	 * Do magic if no mem on local node! XXX
-	 */
-	if (!page)
-		return 0;
-	tmp = page_address(page);
+	c_addr = (unsigned long)page_address(page);
+	uc_addr = c_addr - PAGE_OFFSET + __IA64_UNCACHED_OFFSET;
 
 	/*
 	 * There's a small race here where it's possible for someone to
@@ -100,76 +99,90 @@ uncached_get_new_chunk(struct gen_pool *poolp)
 	for (i = 0; i < (IA64_GRANULE_SIZE / PAGE_SIZE); i++)
 		SetPageUncached(&page[i]);
 
-	flush_tlb_kernel_range(tmp, tmp + IA64_GRANULE_SIZE);
+	flush_tlb_kernel_range(uc_addr, uc_adddr + IA64_GRANULE_SIZE);
 
 	status = ia64_pal_prefetch_visibility(PAL_VISIBILITY_PHYSICAL);
-
-	dprintk(KERN_INFO "pal_prefetch_visibility() returns %i on cpu %i\n",
-		status, raw_smp_processor_id());
-
 	if (!status) {
 		status = smp_call_function(uncached_ipi_visibility, NULL, 0, 1);
 		if (status)
-			printk(KERN_WARNING "smp_call_function failed for "
-			       "uncached_ipi_visibility! (%i)\n", status);
+			goto failed;
 	}
 
+	preempt_disable();
+
 	if (ia64_platform_is("sn2"))
-		sn_flush_all_caches((unsigned long)tmp, IA64_GRANULE_SIZE);
+		sn_flush_all_caches(uc_addr, IA64_GRANULE_SIZE);
 	else
-		flush_icache_range((unsigned long)tmp,
-				   (unsigned long)tmp+IA64_GRANULE_SIZE);
+		flush_icache_range(uc_addr, uc_addr + IA64_GRANULE_SIZE);
+
+	/* flush the just introduced uncached translation from the TLB */
+	local_flush_tlb_all();
+
+	preempt_enable();
 
 	ia64_pal_mc_drain();
 	status = smp_call_function(uncached_ipi_mc_drain, NULL, 0, 1);
 	if (status)
-		printk(KERN_WARNING "smp_call_function failed for "
-		       "uncached_ipi_mc_drain! (%i)\n", status);
+		goto failed;
 
-	addr = (unsigned long)tmp - PAGE_OFFSET + __IA64_UNCACHED_OFFSET;
+	/*
+	 * The chunk of memory pages has been converted to uncached so now we
+	 * can add it to the pool.
+	 */
+	status = gen_pool_add(pool, uc_addr, IA64_GRANULE_SIZE, nid);
+	if (status)
+		goto failed;
 
 	allocated_granules++;
-	return addr;
+	return 0;
+
+	/* failed to convert or add the chunk so give it back to the kernel */
+failed:
+	for (i = 0; i < (IA64_GRANULE_SIZE / PAGE_SIZE); i++)
+		ClearPageUncached(&page[i]);
+
+	free_pages(c_addr, IA64_GRANULE_SHIFT-PAGE_SHIFT);
+	return -1;
 }
 
 
 /*
  * uncached_alloc_page
  *
+ * @starting_nid: node id of node to start with, or -1
+ *
  * Allocate 1 uncached page. Allocates on the requested node. If no
  * uncached pages are available on the requested node, roundrobin starting
- * with higher nodes.
+ * with the next higher node.
  */
-unsigned long
-uncached_alloc_page(int nid)
+unsigned long uncached_alloc_page(int starting_nid)
 {
-	unsigned long maddr;
+	unsigned long uc_addr;
+	struct gen_pool *pool;
+	int nid;
 
-	maddr = gen_pool_alloc(uncached_pool[nid], PAGE_SIZE);
+	if (unlikely(starting_nid >= MAX_NUMNODES))
+		return 0;
 
-	dprintk(KERN_DEBUG "uncached_alloc_page returns %lx on node %i\n",
-		maddr, nid);
+	if (starting_nid < 0)
+		starting_nid = numa_node_id();
+	nid = starting_nid;
 
-	/*
-	 * If no memory is availble on our local node, try the
-	 * remaining nodes in the system.
-	 */
-	if (!maddr) {
-		int i;
-
-		for (i = MAX_NUMNODES - 1; i >= 0; i--) {
-			if (i == nid || !node_online(i))
-				continue;
-			maddr = gen_pool_alloc(uncached_pool[i], PAGE_SIZE);
-			dprintk(KERN_DEBUG "uncached_alloc_page alternate search "
-				"returns %lx on node %i\n", maddr, i);
-			if (maddr) {
-				break;
-			}
-		}
-	}
+	do {
+		if (!node_online(nid))
+			continue;
+		pool = uncached_pool[nid];
+		if (pool == NULL)
+			continue;
+		do {
+			uc_addr = gen_pool_alloc(pool, PAGE_SIZE);
+			if (uc_addr != 0)
+				return uc_addr;
+		} while (uncached_add_chunk(pool, nid) == 0);
+
+	} while ((nid = (nid + 1) % MAX_NUMNODES) != starting_nid);
 
-	return maddr;
+	return 0;
 }
 EXPORT_SYMBOL(uncached_alloc_page);
 
@@ -177,21 +190,22 @@ EXPORT_SYMBOL(uncached_alloc_page);
 /*
  * uncached_free_page
  *
+ * @uc_addr: uncached address of page to free
+ *
  * Free a single uncached page.
  */
-void
-uncached_free_page(unsigned long maddr)
+void uncached_free_page(unsigned long uc_addr)
 {
-	int node;
-
-	node = paddr_to_nid(maddr - __IA64_UNCACHED_OFFSET);
+	int nid = paddr_to_nid(uc_addr - __IA64_UNCACHED_OFFSET);
+	struct gen_pool *pool = uncached_pool[nid];
 
-	dprintk(KERN_DEBUG "uncached_free_page(%lx) on node %i\n", maddr, node);
+	if (unlikely(pool == NULL))
+		return;
 
-	if ((maddr & (0XFUL << 60)) != __IA64_UNCACHED_OFFSET)
-		panic("uncached_free_page invalid address %lx\n", maddr);
+	if ((uc_addr & (0XFUL << 60)) != __IA64_UNCACHED_OFFSET)
+		panic("uncached_free_page invalid address %lx\n", uc_addr);
 
-	gen_pool_free(uncached_pool[node], maddr, PAGE_SIZE);
+	gen_pool_free(pool, uc_addr, PAGE_SIZE);
 }
 EXPORT_SYMBOL(uncached_free_page);
 
@@ -199,43 +213,39 @@ EXPORT_SYMBOL(uncached_free_page);
 /*
  * uncached_build_memmap,
  *
+ * @uc_start: uncached starting address of a chunk of uncached memory
+ * @uc_end: uncached ending address of a chunk of uncached memory
+ * @arg: ignored, (NULL argument passed in on call to efi_memmap_walk_uc())
+ *
  * Called at boot time to build a map of pages that can be used for
  * memory special operations.
  */
-static int __init
-uncached_build_memmap(unsigned long start, unsigned long end, void *arg)
+static int __init uncached_build_memmap(unsigned long uc_start,
+					unsigned long uc_end, void *arg)
 {
-	long length = end - start;
-	int node;
-
-	dprintk(KERN_ERR "uncached_build_memmap(%lx %lx)\n", start, end);
+	int nid = paddr_to_nid(uc_start - __IA64_UNCACHED_OFFSET);
+	struct gen_pool *pool = uncached_pool[nid];
+	size_t size = uc_end - uc_start;
 
 	touch_softlockup_watchdog();
-	memset((char *)start, 0, length);
 
-	node = paddr_to_nid(start - __IA64_UNCACHED_OFFSET);
-
-	for (; start < end ; start += PAGE_SIZE) {
-		dprintk(KERN_INFO "sticking %lx into the pool!\n", start);
-		gen_pool_free(uncached_pool[node], start, PAGE_SIZE);
+	if (pool != NULL) {
+		memset((char *)uc_start, 0, size);
+		(void) gen_pool_add(pool, uc_start, size, nid);
 	}
-
 	return 0;
 }
 
 
-static int __init uncached_init(void) {
-	int i;
+static int __init uncached_init(void)
+{
+	int nid;
 
-	for (i = 0; i < MAX_NUMNODES; i++) {
-		if (!node_online(i))
-			continue;
-		uncached_pool[i] = gen_pool_create(0, IA64_GRANULE_SHIFT,
-						   &uncached_get_new_chunk, i);
+	for_each_online_node(nid) {
+		uncached_pool[nid] = gen_pool_create(PAGE_SHIFT, nid);
 	}
 
-	efi_memmap_walk_uc(uncached_build_memmap);
-
+	efi_memmap_walk_uc(uncached_build_memmap, NULL);
 	return 0;
 }
 
diff --git a/arch/ia64/mm/init.c b/arch/ia64/mm/init.c
index cafa8776a53d..11f08001f8c2 100644
--- a/arch/ia64/mm/init.c
+++ b/arch/ia64/mm/init.c
@@ -671,9 +671,11 @@ int add_memory(u64 start, u64 size)
 
 	return ret;
 }
+EXPORT_SYMBOL_GPL(add_memory);
 
 int remove_memory(u64 start, u64 size)
 {
 	return -EINVAL;
 }
+EXPORT_SYMBOL_GPL(remove_memory);
 #endif
diff --git a/arch/ia64/pci/pci.c b/arch/ia64/pci/pci.c
index 30d148f34042..61dd8608da4f 100644
--- a/arch/ia64/pci/pci.c
+++ b/arch/ia64/pci/pci.c
@@ -352,7 +352,7 @@ pci_acpi_scan_root(struct acpi_device *device, int domain, int bus)
 	pxm = acpi_get_pxm(controller->acpi_handle);
 #ifdef CONFIG_NUMA
 	if (pxm >= 0)
-		controller->node = pxm_to_nid_map[pxm];
+		controller->node = pxm_to_node(pxm);
 #endif
 
 	acpi_walk_resources(device->handle, METHOD_NAME__CRS, count_window,
diff --git a/arch/ia64/sn/kernel/io_init.c b/arch/ia64/sn/kernel/io_init.c
index 5101ac462643..dc09a6a28a37 100644
--- a/arch/ia64/sn/kernel/io_init.c
+++ b/arch/ia64/sn/kernel/io_init.c
@@ -58,7 +58,7 @@ static int max_pcibus_number = 255;	/* Default highest pci bus number */
  */
 
 static dma_addr_t
-sn_default_pci_map(struct pci_dev *pdev, unsigned long paddr, size_t size)
+sn_default_pci_map(struct pci_dev *pdev, unsigned long paddr, size_t size, int type)
 {
 	return 0;
 }
@@ -457,13 +457,6 @@ void sn_pci_fixup_slot(struct pci_dev *dev)
 		pcidev_info->pdi_sn_irq_info = NULL;
 		kfree(sn_irq_info);
 	}
-
-	/*
-	 * MSI currently not supported on altix.  Remove this when
-	 * the MSI abstraction patches are integrated into the kernel
-	 * (sometime after 2.6.16 releases)
-	 */
-	dev->no_msi = 1;
 }
 
 /*
diff --git a/arch/ia64/sn/kernel/irq.c b/arch/ia64/sn/kernel/irq.c
index c265e02f5036..dc8e2b696713 100644
--- a/arch/ia64/sn/kernel/irq.c
+++ b/arch/ia64/sn/kernel/irq.c
@@ -26,11 +26,11 @@ static void unregister_intr_pda(struct sn_irq_info *sn_irq_info);
 
 int sn_force_interrupt_flag = 1;
 extern int sn_ioif_inited;
-static struct list_head **sn_irq_lh;
+struct list_head **sn_irq_lh;
 static spinlock_t sn_irq_info_lock = SPIN_LOCK_UNLOCKED; /* non-IRQ lock */
 
-static inline u64 sn_intr_alloc(nasid_t local_nasid, int local_widget,
-				     u64 sn_irq_info,
+u64 sn_intr_alloc(nasid_t local_nasid, int local_widget,
+				     struct sn_irq_info *sn_irq_info,
 				     int req_irq, nasid_t req_nasid,
 				     int req_slice)
 {
@@ -40,12 +40,13 @@ static inline u64 sn_intr_alloc(nasid_t local_nasid, int local_widget,
 
 	SAL_CALL_NOLOCK(ret_stuff, (u64) SN_SAL_IOIF_INTERRUPT,
 			(u64) SAL_INTR_ALLOC, (u64) local_nasid,
-			(u64) local_widget, (u64) sn_irq_info, (u64) req_irq,
+			(u64) local_widget, __pa(sn_irq_info), (u64) req_irq,
 			(u64) req_nasid, (u64) req_slice);
+
 	return ret_stuff.status;
 }
 
-static inline void sn_intr_free(nasid_t local_nasid, int local_widget,
+void sn_intr_free(nasid_t local_nasid, int local_widget,
 				struct sn_irq_info *sn_irq_info)
 {
 	struct ia64_sal_retval ret_stuff;
@@ -112,73 +113,91 @@ static void sn_end_irq(unsigned int irq)
 
 static void sn_irq_info_free(struct rcu_head *head);
 
-static void sn_set_affinity_irq(unsigned int irq, cpumask_t mask)
+struct sn_irq_info *sn_retarget_vector(struct sn_irq_info *sn_irq_info,
+				       nasid_t nasid, int slice)
 {
-	struct sn_irq_info *sn_irq_info, *sn_irq_info_safe;
-	int cpuid, cpuphys;
+	int vector;
+	int cpuphys;
+	int64_t bridge;
+	int local_widget, status;
+	nasid_t local_nasid;
+	struct sn_irq_info *new_irq_info;
+	struct sn_pcibus_provider *pci_provider;
 
-	cpuid = first_cpu(mask);
-	cpuphys = cpu_physical_id(cpuid);
+	new_irq_info = kmalloc(sizeof(struct sn_irq_info), GFP_ATOMIC);
+	if (new_irq_info == NULL)
+		return NULL;
 
-	list_for_each_entry_safe(sn_irq_info, sn_irq_info_safe,
-				 sn_irq_lh[irq], list) {
-		u64 bridge;
-		int local_widget, status;
-		nasid_t local_nasid;
-		struct sn_irq_info *new_irq_info;
-		struct sn_pcibus_provider *pci_provider;
-
-		new_irq_info = kmalloc(sizeof(struct sn_irq_info), GFP_ATOMIC);
-		if (new_irq_info == NULL)
-			break;
-		memcpy(new_irq_info, sn_irq_info, sizeof(struct sn_irq_info));
-
-		bridge = (u64) new_irq_info->irq_bridge;
-		if (!bridge) {
-			kfree(new_irq_info);
-			break; /* irq is not a device interrupt */
-		}
+	memcpy(new_irq_info, sn_irq_info, sizeof(struct sn_irq_info));
+
+	bridge = (u64) new_irq_info->irq_bridge;
+	if (!bridge) {
+		kfree(new_irq_info);
+		return NULL; /* irq is not a device interrupt */
+	}
 
-		local_nasid = NASID_GET(bridge);
+	local_nasid = NASID_GET(bridge);
 
-		if (local_nasid & 1)
-			local_widget = TIO_SWIN_WIDGETNUM(bridge);
-		else
-			local_widget = SWIN_WIDGETNUM(bridge);
+	if (local_nasid & 1)
+		local_widget = TIO_SWIN_WIDGETNUM(bridge);
+	else
+		local_widget = SWIN_WIDGETNUM(bridge);
 
-		/* Free the old PROM new_irq_info structure */
-		sn_intr_free(local_nasid, local_widget, new_irq_info);
-		/* Update kernels new_irq_info with new target info */
-		unregister_intr_pda(new_irq_info);
+	vector = sn_irq_info->irq_irq;
+	/* Free the old PROM new_irq_info structure */
+	sn_intr_free(local_nasid, local_widget, new_irq_info);
+	/* Update kernels new_irq_info with new target info */
+	unregister_intr_pda(new_irq_info);
 
-		/* allocate a new PROM new_irq_info struct */
-		status = sn_intr_alloc(local_nasid, local_widget,
-				       __pa(new_irq_info), irq,
-				       cpuid_to_nasid(cpuid),
-				       cpuid_to_slice(cpuid));
+	/* allocate a new PROM new_irq_info struct */
+	status = sn_intr_alloc(local_nasid, local_widget,
+			       new_irq_info, vector,
+			       nasid, slice);
 
-		/* SAL call failed */
-		if (status) {
-			kfree(new_irq_info);
-			break;
-		}
+	/* SAL call failed */
+	if (status) {
+		kfree(new_irq_info);
+		return NULL;
+	}
 
-		new_irq_info->irq_cpuid = cpuid;
-		register_intr_pda(new_irq_info);
+	cpuphys = nasid_slice_to_cpuid(nasid, slice);
+	new_irq_info->irq_cpuid = cpuphys;
+	register_intr_pda(new_irq_info);
 
-		pci_provider = sn_pci_provider[new_irq_info->irq_bridge_type];
-		if (pci_provider && pci_provider->target_interrupt)
-			(pci_provider->target_interrupt)(new_irq_info);
+	pci_provider = sn_pci_provider[new_irq_info->irq_bridge_type];
 
-		spin_lock(&sn_irq_info_lock);
-		list_replace_rcu(&sn_irq_info->list, &new_irq_info->list);
-		spin_unlock(&sn_irq_info_lock);
-		call_rcu(&sn_irq_info->rcu, sn_irq_info_free);
+	/*
+	 * If this represents a line interrupt, target it.  If it's
+	 * an msi (irq_int_bit < 0), it's already targeted.
+	 */
+	if (new_irq_info->irq_int_bit >= 0 &&
+	    pci_provider && pci_provider->target_interrupt)
+		(pci_provider->target_interrupt)(new_irq_info);
+
+	spin_lock(&sn_irq_info_lock);
+	list_replace_rcu(&sn_irq_info->list, &new_irq_info->list);
+	spin_unlock(&sn_irq_info_lock);
+	call_rcu(&sn_irq_info->rcu, sn_irq_info_free);
 
 #ifdef CONFIG_SMP
-		set_irq_affinity_info((irq & 0xff), cpuphys, 0);
+	set_irq_affinity_info((vector & 0xff), cpuphys, 0);
 #endif
-	}
+
+	return new_irq_info;
+}
+
+static void sn_set_affinity_irq(unsigned int irq, cpumask_t mask)
+{
+	struct sn_irq_info *sn_irq_info, *sn_irq_info_safe;
+	nasid_t nasid;
+	int slice;
+
+	nasid = cpuid_to_nasid(first_cpu(mask));
+	slice = cpuid_to_slice(first_cpu(mask));
+
+	list_for_each_entry_safe(sn_irq_info, sn_irq_info_safe,
+				 sn_irq_lh[irq], list)
+		(void)sn_retarget_vector(sn_irq_info, nasid, slice);
 }
 
 struct hw_interrupt_type irq_type_sn = {
@@ -202,6 +221,9 @@ void sn_irq_init(void)
 	int i;
 	irq_desc_t *base_desc = irq_desc;
 
+	ia64_first_device_vector = IA64_SN2_FIRST_DEVICE_VECTOR;
+	ia64_last_device_vector = IA64_SN2_LAST_DEVICE_VECTOR;
+
 	for (i = 0; i < NR_IRQS; i++) {
 		if (base_desc[i].handler == &no_irq_type) {
 			base_desc[i].handler = &irq_type_sn;
@@ -285,6 +307,7 @@ void sn_irq_fixup(struct pci_dev *pci_dev, struct sn_irq_info *sn_irq_info)
 	/* link it into the sn_irq[irq] list */
 	spin_lock(&sn_irq_info_lock);
 	list_add_rcu(&sn_irq_info->list, sn_irq_lh[sn_irq_info->irq_irq]);
+	reserve_irq_vector(sn_irq_info->irq_irq);
 	spin_unlock(&sn_irq_info_lock);
 
 	register_intr_pda(sn_irq_info);
@@ -310,8 +333,11 @@ void sn_irq_unfixup(struct pci_dev *pci_dev)
 	spin_lock(&sn_irq_info_lock);
 	list_del_rcu(&sn_irq_info->list);
 	spin_unlock(&sn_irq_info_lock);
+	if (list_empty(sn_irq_lh[sn_irq_info->irq_irq]))
+		free_irq_vector(sn_irq_info->irq_irq);
 	call_rcu(&sn_irq_info->rcu, sn_irq_info_free);
 	pci_dev_put(pci_dev);
+
 }
 
 static inline void
diff --git a/arch/ia64/sn/kernel/setup.c b/arch/ia64/sn/kernel/setup.c
index 30988dfbddff..93577abae36d 100644
--- a/arch/ia64/sn/kernel/setup.c
+++ b/arch/ia64/sn/kernel/setup.c
@@ -139,7 +139,7 @@ static int __init pxm_to_nasid(int pxm)
 	int i;
 	int nid;
 
-	nid = pxm_to_nid_map[pxm];
+	nid = pxm_to_node(pxm);
 	for (i = 0; i < num_node_memblks; i++) {
 		if (node_memblk[i].nid == nid) {
 			return NASID_GET(node_memblk[i].start_paddr);
@@ -704,7 +704,7 @@ void __init build_cnode_tables(void)
 	 * cnode == node for all C & M bricks.
 	 */
 	for_each_online_node(node) {
-		nasid = pxm_to_nasid(nid_to_pxm_map[node]);
+		nasid = pxm_to_nasid(node_to_pxm(node));
 		sn_cnodeid_to_nasid[node] = nasid;
 		physical_node_map[nasid] = node;
 	}
diff --git a/arch/ia64/sn/kernel/sn2/cache.c b/arch/ia64/sn/kernel/sn2/cache.c
index bc3cfa17cd0f..2862cb33026d 100644
--- a/arch/ia64/sn/kernel/sn2/cache.c
+++ b/arch/ia64/sn/kernel/sn2/cache.c
@@ -3,11 +3,12 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  * 
- * Copyright (C) 2001-2003 Silicon Graphics, Inc. All rights reserved.
+ * Copyright (C) 2001-2003, 2006 Silicon Graphics, Inc. All rights reserved.
  *
  */
 #include <linux/module.h>
 #include <asm/pgalloc.h>
+#include <asm/sn/arch.h>
 
 /**
  * sn_flush_all_caches - flush a range of address from all caches (incl. L4)
@@ -17,18 +18,24 @@
  * Flush a range of addresses from all caches including L4. 
  * All addresses fully or partially contained within 
  * @flush_addr to @flush_addr + @bytes are flushed
- * from the all caches.
+ * from all caches.
  */
 void
 sn_flush_all_caches(long flush_addr, long bytes)
 {
-	flush_icache_range(flush_addr, flush_addr+bytes);
+	unsigned long addr = flush_addr;
+
+	/* SHub1 requires a cached address */
+	if (is_shub1() && (addr & RGN_BITS) == RGN_BASE(RGN_UNCACHED))
+		addr = (addr - RGN_BASE(RGN_UNCACHED)) + RGN_BASE(RGN_KERNEL);
+
+	flush_icache_range(addr, addr + bytes);
 	/*
 	 * The last call may have returned before the caches
 	 * were actually flushed, so we call it again to make
 	 * sure.
 	 */
-	flush_icache_range(flush_addr, flush_addr+bytes);
+	flush_icache_range(addr, addr + bytes);
 	mb();
 }
 EXPORT_SYMBOL(sn_flush_all_caches);
diff --git a/arch/ia64/sn/pci/pci_dma.c b/arch/ia64/sn/pci/pci_dma.c
index b4b84c269210..7a291a271511 100644
--- a/arch/ia64/sn/pci/pci_dma.c
+++ b/arch/ia64/sn/pci/pci_dma.c
@@ -11,7 +11,7 @@
 
 #include <linux/module.h>
 #include <asm/dma.h>
-#include <asm/sn/pcibr_provider.h>
+#include <asm/sn/intr.h>
 #include <asm/sn/pcibus_provider_defs.h>
 #include <asm/sn/pcidev.h>
 #include <asm/sn/sn_sal.h>
@@ -113,7 +113,8 @@ void *sn_dma_alloc_coherent(struct device *dev, size_t size,
 	 * resources.
 	 */
 
-	*dma_handle = provider->dma_map_consistent(pdev, phys_addr, size);
+	*dma_handle = provider->dma_map_consistent(pdev, phys_addr, size,
+						   SN_DMA_ADDR_PHYS);
 	if (!*dma_handle) {
 		printk(KERN_ERR "%s: out of ATEs\n", __FUNCTION__);
 		free_pages((unsigned long)cpuaddr, get_order(size));
@@ -176,7 +177,7 @@ dma_addr_t sn_dma_map_single(struct device *dev, void *cpu_addr, size_t size,
 	BUG_ON(dev->bus != &pci_bus_type);
 
 	phys_addr = __pa(cpu_addr);
-	dma_addr = provider->dma_map(pdev, phys_addr, size);
+	dma_addr = provider->dma_map(pdev, phys_addr, size, SN_DMA_ADDR_PHYS);
 	if (!dma_addr) {
 		printk(KERN_ERR "%s: out of ATEs\n", __FUNCTION__);
 		return 0;
@@ -260,7 +261,8 @@ int sn_dma_map_sg(struct device *dev, struct scatterlist *sg, int nhwentries,
 	for (i = 0; i < nhwentries; i++, sg++) {
 		phys_addr = SG_ENT_PHYS_ADDRESS(sg);
 		sg->dma_address = provider->dma_map(pdev,
-						    phys_addr, sg->length);
+						    phys_addr, sg->length,
+						    SN_DMA_ADDR_PHYS);
 
 		if (!sg->dma_address) {
 			printk(KERN_ERR "%s: out of ATEs\n", __FUNCTION__);
diff --git a/arch/ia64/sn/pci/pcibr/pcibr_dma.c b/arch/ia64/sn/pci/pcibr/pcibr_dma.c
index 9f86bb6519aa..a86c7b945962 100644
--- a/arch/ia64/sn/pci/pcibr/pcibr_dma.c
+++ b/arch/ia64/sn/pci/pcibr/pcibr_dma.c
@@ -41,7 +41,7 @@ extern int sn_ioif_inited;
 
 static dma_addr_t
 pcibr_dmamap_ate32(struct pcidev_info *info,
-		   u64 paddr, size_t req_size, u64 flags)
+		   u64 paddr, size_t req_size, u64 flags, int dma_flags)
 {
 
 	struct pcidev_info *pcidev_info = info->pdi_host_pcidev_info;
@@ -81,9 +81,12 @@ pcibr_dmamap_ate32(struct pcidev_info *info,
 	if (IS_PCIX(pcibus_info))
 		ate_flags &= ~(PCI32_ATE_PREF);
 
-	xio_addr =
-	    IS_PIC_SOFT(pcibus_info) ? PHYS_TO_DMA(paddr) :
-	    PHYS_TO_TIODMA(paddr);
+	if (SN_DMA_ADDRTYPE(dma_flags == SN_DMA_ADDR_PHYS))
+		xio_addr = IS_PIC_SOFT(pcibus_info) ? PHYS_TO_DMA(paddr) :
+	    					      PHYS_TO_TIODMA(paddr);
+	else
+		xio_addr = paddr;
+
 	offset = IOPGOFF(xio_addr);
 	ate = ate_flags | (xio_addr - offset);
 
@@ -91,6 +94,13 @@ pcibr_dmamap_ate32(struct pcidev_info *info,
 	if (IS_PIC_SOFT(pcibus_info)) {
 		ate |= (pcibus_info->pbi_hub_xid << PIC_ATE_TARGETID_SHFT);
 	}
+
+	/*
+	 * If we're mapping for MSI, set the MSI bit in the ATE
+	 */
+	if (dma_flags & SN_DMA_MSI)
+		ate |= PCI32_ATE_MSI;
+
 	ate_write(pcibus_info, ate_index, ate_count, ate);
 
 	/*
@@ -105,20 +115,27 @@ pcibr_dmamap_ate32(struct pcidev_info *info,
 	if (pcibus_info->pbi_devreg[internal_device] & PCIBR_DEV_SWAP_DIR)
 		ATE_SWAP_ON(pci_addr);
 
+
 	return pci_addr;
 }
 
 static dma_addr_t
 pcibr_dmatrans_direct64(struct pcidev_info * info, u64 paddr,
-			u64 dma_attributes)
+			u64 dma_attributes, int dma_flags)
 {
 	struct pcibus_info *pcibus_info = (struct pcibus_info *)
 	    ((info->pdi_host_pcidev_info)->pdi_pcibus_info);
 	u64 pci_addr;
 
 	/* Translate to Crosstalk View of Physical Address */
-	pci_addr = (IS_PIC_SOFT(pcibus_info) ? PHYS_TO_DMA(paddr) :
-		    PHYS_TO_TIODMA(paddr)) | dma_attributes;
+	if (SN_DMA_ADDRTYPE(dma_flags) == SN_DMA_ADDR_PHYS)
+		pci_addr = IS_PIC_SOFT(pcibus_info) ?
+				PHYS_TO_DMA(paddr) :
+		    		PHYS_TO_TIODMA(paddr) | dma_attributes;
+	else
+		pci_addr = IS_PIC_SOFT(pcibus_info) ?
+				paddr :
+				paddr | dma_attributes;
 
 	/* Handle Bus mode */
 	if (IS_PCIX(pcibus_info))
@@ -130,7 +147,9 @@ pcibr_dmatrans_direct64(struct pcidev_info * info, u64 paddr,
 		    ((u64) pcibus_info->
 		     pbi_hub_xid << PIC_PCI64_ATTR_TARG_SHFT);
 	} else
-		pci_addr |= TIOCP_PCI64_CMDTYPE_MEM;
+		pci_addr |= (dma_flags & SN_DMA_MSI) ?
+				TIOCP_PCI64_CMDTYPE_MSI :
+				TIOCP_PCI64_CMDTYPE_MEM;
 
 	/* If PCI mode, func zero uses VCHAN0, every other func uses VCHAN1 */
 	if (!IS_PCIX(pcibus_info) && PCI_FUNC(info->pdi_linux_pcidev->devfn))
@@ -141,7 +160,7 @@ pcibr_dmatrans_direct64(struct pcidev_info * info, u64 paddr,
 
 static dma_addr_t
 pcibr_dmatrans_direct32(struct pcidev_info * info,
-			u64 paddr, size_t req_size, u64 flags)
+			u64 paddr, size_t req_size, u64 flags, int dma_flags)
 {
 	struct pcidev_info *pcidev_info = info->pdi_host_pcidev_info;
 	struct pcibus_info *pcibus_info = (struct pcibus_info *)pcidev_info->
@@ -156,8 +175,14 @@ pcibr_dmatrans_direct32(struct pcidev_info * info,
 		return 0;
 	}
 
-	xio_addr = IS_PIC_SOFT(pcibus_info) ? PHYS_TO_DMA(paddr) :
-	    PHYS_TO_TIODMA(paddr);
+	if (dma_flags & SN_DMA_MSI)
+		return 0;
+
+	if (SN_DMA_ADDRTYPE(dma_flags) == SN_DMA_ADDR_PHYS)
+		xio_addr = IS_PIC_SOFT(pcibus_info) ? PHYS_TO_DMA(paddr) :
+	    					      PHYS_TO_TIODMA(paddr);
+	else
+		xio_addr = paddr;
 
 	xio_base = pcibus_info->pbi_dir_xbase;
 	offset = xio_addr - xio_base;
@@ -327,7 +352,7 @@ void sn_dma_flush(u64 addr)
  */
 
 dma_addr_t
-pcibr_dma_map(struct pci_dev * hwdev, unsigned long phys_addr, size_t size)
+pcibr_dma_map(struct pci_dev * hwdev, unsigned long phys_addr, size_t size, int dma_flags)
 {
 	dma_addr_t dma_handle;
 	struct pcidev_info *pcidev_info = SN_PCIDEV_INFO(hwdev);
@@ -344,11 +369,11 @@ pcibr_dma_map(struct pci_dev * hwdev, unsigned long phys_addr, size_t size)
 		 */
 
 		dma_handle = pcibr_dmatrans_direct64(pcidev_info, phys_addr,
-						     PCI64_ATTR_PREF);
+						     PCI64_ATTR_PREF, dma_flags);
 	} else {
 		/* Handle 32-63 bit cards via direct mapping */
 		dma_handle = pcibr_dmatrans_direct32(pcidev_info, phys_addr,
-						     size, 0);
+						     size, 0, dma_flags);
 		if (!dma_handle) {
 			/*
 			 * It is a 32 bit card and we cannot do direct mapping,
@@ -356,7 +381,8 @@ pcibr_dma_map(struct pci_dev * hwdev, unsigned long phys_addr, size_t size)
 			 */
 
 			dma_handle = pcibr_dmamap_ate32(pcidev_info, phys_addr,
-							size, PCI32_ATE_PREF);
+							size, PCI32_ATE_PREF,
+							dma_flags);
 		}
 	}
 
@@ -365,18 +391,18 @@ pcibr_dma_map(struct pci_dev * hwdev, unsigned long phys_addr, size_t size)
 
 dma_addr_t
 pcibr_dma_map_consistent(struct pci_dev * hwdev, unsigned long phys_addr,
-			 size_t size)
+			 size_t size, int dma_flags)
 {
 	dma_addr_t dma_handle;
 	struct pcidev_info *pcidev_info = SN_PCIDEV_INFO(hwdev);
 
 	if (hwdev->dev.coherent_dma_mask == ~0UL) {
 		dma_handle = pcibr_dmatrans_direct64(pcidev_info, phys_addr,
-					    PCI64_ATTR_BAR);
+					    PCI64_ATTR_BAR, dma_flags);
 	} else {
 		dma_handle = (dma_addr_t) pcibr_dmamap_ate32(pcidev_info,
 						    phys_addr, size,
-						    PCI32_ATE_BAR);
+						    PCI32_ATE_BAR, dma_flags);
 	}
 
 	return dma_handle;
diff --git a/arch/ia64/sn/pci/tioca_provider.c b/arch/ia64/sn/pci/tioca_provider.c
index be0176912968..20de72791b97 100644
--- a/arch/ia64/sn/pci/tioca_provider.c
+++ b/arch/ia64/sn/pci/tioca_provider.c
@@ -515,11 +515,17 @@ tioca_dma_unmap(struct pci_dev *pdev, dma_addr_t bus_addr, int dir)
  * use the GART mapped mode.
  */
 static u64
-tioca_dma_map(struct pci_dev *pdev, u64 paddr, size_t byte_count)
+tioca_dma_map(struct pci_dev *pdev, u64 paddr, size_t byte_count, int dma_flags)
 {
 	u64 mapaddr;
 
 	/*
+	 * Not supported for now ...
+	 */
+	if (dma_flags & SN_DMA_MSI)
+		return 0;
+
+	/*
 	 * If card is 64 or 48 bit addresable, use a direct mapping.  32
 	 * bit direct is so restrictive w.r.t. where the memory resides that
 	 * we don't use it even though CA has some support.
diff --git a/arch/ia64/sn/pci/tioce_provider.c b/arch/ia64/sn/pci/tioce_provider.c
index 85f3b3d4c606..2d7948567ebc 100644
--- a/arch/ia64/sn/pci/tioce_provider.c
+++ b/arch/ia64/sn/pci/tioce_provider.c
@@ -170,7 +170,8 @@ tioce_mmr_war_post(struct tioce_kernel *kern, void *mmr_addr)
 	(ATE_PAGE((start)+(len)-1, pagesize) - ATE_PAGE(start, pagesize) + 1)
 
 #define ATE_VALID(ate)	((ate) & (1UL << 63))
-#define ATE_MAKE(addr, ps) (((addr) & ~ATE_PAGEMASK(ps)) | (1UL << 63))
+#define ATE_MAKE(addr, ps, msi) \
+	(((addr) & ~ATE_PAGEMASK(ps)) | (1UL << 63) | ((msi)?(1UL << 62):0))
 
 /*
  * Flavors of ate-based mapping supported by tioce_alloc_map()
@@ -196,15 +197,17 @@ tioce_mmr_war_post(struct tioce_kernel *kern, void *mmr_addr)
  *
  * 63    - must be 1 to indicate d64 mode to CE hardware
  * 62    - barrier bit ... controlled with tioce_dma_barrier()
- * 61    - 0 since this is not an MSI transaction
+ * 61    - msi bit ... specified through dma_flags
  * 60:54 - reserved, MBZ
  */
 static u64
-tioce_dma_d64(unsigned long ct_addr)
+tioce_dma_d64(unsigned long ct_addr, int dma_flags)
 {
 	u64 bus_addr;
 
 	bus_addr = ct_addr | (1UL << 63);
+	if (dma_flags & SN_DMA_MSI)
+		bus_addr |= (1UL << 61);
 
 	return bus_addr;
 }
@@ -261,7 +264,7 @@ pcidev_to_tioce(struct pci_dev *pdev, struct tioce **base,
  */
 static u64
 tioce_alloc_map(struct tioce_kernel *ce_kern, int type, int port,
-		u64 ct_addr, int len)
+		u64 ct_addr, int len, int dma_flags)
 {
 	int i;
 	int j;
@@ -270,6 +273,7 @@ tioce_alloc_map(struct tioce_kernel *ce_kern, int type, int port,
 	int entries;
 	int nates;
 	u64 pagesize;
+	int msi_capable, msi_wanted;
 	u64 *ate_shadow;
 	u64 *ate_reg;
 	u64 addr;
@@ -291,6 +295,7 @@ tioce_alloc_map(struct tioce_kernel *ce_kern, int type, int port,
 		ate_reg = ce_mmr->ce_ure_ate3240;
 		pagesize = ce_kern->ce_ate3240_pagesize;
 		bus_base = TIOCE_M32_MIN;
+		msi_capable = 1;
 		break;
 	case TIOCE_ATE_M40:
 		first = 0;
@@ -299,6 +304,7 @@ tioce_alloc_map(struct tioce_kernel *ce_kern, int type, int port,
 		ate_reg = ce_mmr->ce_ure_ate40;
 		pagesize = MB(64);
 		bus_base = TIOCE_M40_MIN;
+		msi_capable = 0;
 		break;
 	case TIOCE_ATE_M40S:
 		/*
@@ -311,11 +317,16 @@ tioce_alloc_map(struct tioce_kernel *ce_kern, int type, int port,
 		ate_reg = ce_mmr->ce_ure_ate3240;
 		pagesize = GB(16);
 		bus_base = TIOCE_M40S_MIN;
+		msi_capable = 0;
 		break;
 	default:
 		return 0;
 	}
 
+	msi_wanted = dma_flags & SN_DMA_MSI;
+	if (msi_wanted && !msi_capable)
+		return 0;
+
 	nates = ATE_NPAGES(ct_addr, len, pagesize);
 	if (nates > entries)
 		return 0;
@@ -344,7 +355,7 @@ tioce_alloc_map(struct tioce_kernel *ce_kern, int type, int port,
 	for (j = 0; j < nates; j++) {
 		u64 ate;
 
-		ate = ATE_MAKE(addr, pagesize);
+		ate = ATE_MAKE(addr, pagesize, msi_wanted);
 		ate_shadow[i + j] = ate;
 		tioce_mmr_storei(ce_kern, &ate_reg[i + j], ate);
 		addr += pagesize;
@@ -371,7 +382,7 @@ tioce_alloc_map(struct tioce_kernel *ce_kern, int type, int port,
  * Map @paddr into 32-bit bus space of the CE associated with @pcidev_info.
  */
 static u64
-tioce_dma_d32(struct pci_dev *pdev, u64 ct_addr)
+tioce_dma_d32(struct pci_dev *pdev, u64 ct_addr, int dma_flags)
 {
 	int dma_ok;
 	int port;
@@ -381,6 +392,9 @@ tioce_dma_d32(struct pci_dev *pdev, u64 ct_addr)
 	u64 ct_lower;
 	dma_addr_t bus_addr;
 
+	if (dma_flags & SN_DMA_MSI)
+		return 0;
+
 	ct_upper = ct_addr & ~0x3fffffffUL;
 	ct_lower = ct_addr & 0x3fffffffUL;
 
@@ -507,7 +521,7 @@ tioce_dma_unmap(struct pci_dev *pdev, dma_addr_t bus_addr, int dir)
  */
 static u64
 tioce_do_dma_map(struct pci_dev *pdev, u64 paddr, size_t byte_count,
-		 int barrier)
+		 int barrier, int dma_flags)
 {
 	unsigned long flags;
 	u64 ct_addr;
@@ -523,15 +537,18 @@ tioce_do_dma_map(struct pci_dev *pdev, u64 paddr, size_t byte_count,
 	if (dma_mask < 0x7fffffffUL)
 		return 0;
 
-	ct_addr = PHYS_TO_TIODMA(paddr);
+	if (SN_DMA_ADDRTYPE(dma_flags) == SN_DMA_ADDR_PHYS)
+		ct_addr = PHYS_TO_TIODMA(paddr);
+	else
+		ct_addr = paddr;
 
 	/*
 	 * If the device can generate 64 bit addresses, create a D64 map.
-	 * Since this should never fail, bypass the rest of the checks.
 	 */
 	if (dma_mask == ~0UL) {
-		mapaddr = tioce_dma_d64(ct_addr);
-		goto dma_map_done;
+		mapaddr = tioce_dma_d64(ct_addr, dma_flags);
+		if (mapaddr)
+			goto dma_map_done;
 	}
 
 	pcidev_to_tioce(pdev, NULL, &ce_kern, &port);
@@ -574,18 +591,22 @@ tioce_do_dma_map(struct pci_dev *pdev, u64 paddr, size_t byte_count,
 
 		if (byte_count > MB(64)) {
 			mapaddr = tioce_alloc_map(ce_kern, TIOCE_ATE_M40S,
-						  port, ct_addr, byte_count);
+						  port, ct_addr, byte_count,
+						  dma_flags);
 			if (!mapaddr)
 				mapaddr =
 				    tioce_alloc_map(ce_kern, TIOCE_ATE_M40, -1,
-						    ct_addr, byte_count);
+						    ct_addr, byte_count,
+						    dma_flags);
 		} else {
 			mapaddr = tioce_alloc_map(ce_kern, TIOCE_ATE_M40, -1,
-						  ct_addr, byte_count);
+						  ct_addr, byte_count,
+						  dma_flags);
 			if (!mapaddr)
 				mapaddr =
 				    tioce_alloc_map(ce_kern, TIOCE_ATE_M40S,
-						    port, ct_addr, byte_count);
+						    port, ct_addr, byte_count,
+						    dma_flags);
 		}
 	}
 
@@ -593,7 +614,7 @@ tioce_do_dma_map(struct pci_dev *pdev, u64 paddr, size_t byte_count,
 	 * 32-bit direct is the next mode to try
 	 */
 	if (!mapaddr && dma_mask >= 0xffffffffUL)
-		mapaddr = tioce_dma_d32(pdev, ct_addr);
+		mapaddr = tioce_dma_d32(pdev, ct_addr, dma_flags);
 
 	/*
 	 * Last resort, try 32-bit ATE-based map.
@@ -601,7 +622,7 @@ tioce_do_dma_map(struct pci_dev *pdev, u64 paddr, size_t byte_count,
 	if (!mapaddr)
 		mapaddr =
 		    tioce_alloc_map(ce_kern, TIOCE_ATE_M32, -1, ct_addr,
-				    byte_count);
+				    byte_count, dma_flags);
 
 	spin_unlock_irqrestore(&ce_kern->ce_lock, flags);
 
@@ -622,9 +643,9 @@ dma_map_done:
  * in the address.
  */
 static u64
-tioce_dma(struct pci_dev *pdev, u64 paddr, size_t byte_count)
+tioce_dma(struct pci_dev *pdev, u64 paddr, size_t byte_count, int dma_flags)
 {
-	return tioce_do_dma_map(pdev, paddr, byte_count, 0);
+	return tioce_do_dma_map(pdev, paddr, byte_count, 0, dma_flags);
 }
 
 /**
@@ -636,9 +657,9 @@ tioce_dma(struct pci_dev *pdev, u64 paddr, size_t byte_count)
  * Simply call tioce_do_dma_map() to create a map with the barrier bit set
  * in the address.
  */ static u64
-tioce_dma_consistent(struct pci_dev *pdev, u64 paddr, size_t byte_count)
+tioce_dma_consistent(struct pci_dev *pdev, u64 paddr, size_t byte_count, int dma_flags)
 {
-	return tioce_do_dma_map(pdev, paddr, byte_count, 1);
+	return tioce_do_dma_map(pdev, paddr, byte_count, 1, dma_flags);
 }
 
 /**
@@ -696,7 +717,7 @@ tioce_reserve_m32(struct tioce_kernel *ce_kern, u64 base, u64 limit)
 	while (ate_index <= last_ate) {
 		u64 ate;
 
-		ate = ATE_MAKE(0xdeadbeef, ps);
+		ate = ATE_MAKE(0xdeadbeef, ps, 0);
 		ce_kern->ce_ate3240_shadow[ate_index] = ate;
 		tioce_mmr_storei(ce_kern, &ce_mmr->ce_ure_ate3240[ate_index],
 				 ate);