21 files changed, 992 insertions, 394 deletions

diff --git a/arch/sparc64/kernel/Makefile b/arch/sparc64/kernel/Makefile
index 18e31a8db01..d8d19093d12 100644
--- a/arch/sparc64/kernel/Makefile
+++ b/arch/sparc64/kernel/Makefile
@@ -8,11 +8,11 @@ EXTRA_CFLAGS := -Werror
 extra-y		:= head.o init_task.o vmlinux.lds
 
 obj-y		:= process.o setup.o cpu.o idprom.o \
-		   traps.o devices.o auxio.o una_asm.o \
+		   traps.o auxio.o una_asm.o \
 		   irq.o ptrace.o time.o sys_sparc.o signal.o \
 		   unaligned.o central.o pci.o starfire.o semaphore.o \
 		   power.o sbus.o iommu_common.o sparc64_ksyms.o chmc.o \
-		   visemul.o prom.o of_device.o hvapi.o sstate.o
+		   visemul.o prom.o of_device.o hvapi.o sstate.o mdesc.o
 
 obj-$(CONFIG_STACKTRACE) += stacktrace.o
 obj-$(CONFIG_PCI)	 += ebus.o isa.o pci_common.o pci_iommu.o \
diff --git a/arch/sparc64/kernel/devices.c b/arch/sparc64/kernel/devices.c
deleted file mode 100644
index 0e03c8e218c..00000000000
--- a/arch/sparc64/kernel/devices.c
+++ /dev/null
@@ -1,196 +0,0 @@
-/* devices.c: Initial scan of the prom device tree for important
- *            Sparc device nodes which we need to find.
- *
- * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
- */
-
-#include <linux/kernel.h>
-#include <linux/threads.h>
-#include <linux/init.h>
-#include <linux/ioport.h>
-#include <linux/string.h>
-#include <linux/spinlock.h>
-#include <linux/errno.h>
-#include <linux/bootmem.h>
-
-#include <asm/page.h>
-#include <asm/oplib.h>
-#include <asm/system.h>
-#include <asm/smp.h>
-#include <asm/spitfire.h>
-#include <asm/timer.h>
-#include <asm/cpudata.h>
-
-/* Used to synchronize accesses to NatSemi SUPER I/O chip configure
- * operations in asm/ns87303.h
- */
-DEFINE_SPINLOCK(ns87303_lock);
-
-extern void cpu_probe(void);
-extern void central_probe(void);
-
-static const char *cpu_mid_prop(void)
-{
-	if (tlb_type == spitfire)
-		return "upa-portid";
-	return "portid";
-}
-
-static int get_cpu_mid(struct device_node *dp)
-{
-	struct property *prop;
-
-	if (tlb_type == hypervisor) {
-		struct linux_prom64_registers *reg;
-		int len;
-
-		prop = of_find_property(dp, "cpuid", &len);
-		if (prop && len == 4)
-			return *(int *) prop->value;
-
-		prop = of_find_property(dp, "reg", NULL);
-		reg = prop->value;
-		return (reg[0].phys_addr >> 32) & 0x0fffffffUL;
-	} else {
-		const char *prop_name = cpu_mid_prop();
-
-		prop = of_find_property(dp, prop_name, NULL);
-		if (prop)
-			return *(int *) prop->value;
-		return 0;
-	}
-}
-
-static int check_cpu_node(struct device_node *dp, int *cur_inst,
-			  int (*compare)(struct device_node *, int, void *),
-			  void *compare_arg,
-			  struct device_node **dev_node, int *mid)
-{
-	if (!compare(dp, *cur_inst, compare_arg)) {
-		if (dev_node)
-			*dev_node = dp;
-		if (mid)
-			*mid = get_cpu_mid(dp);
-		return 0;
-	}
-
-	(*cur_inst)++;
-
-	return -ENODEV;
-}
-
-static int __cpu_find_by(int (*compare)(struct device_node *, int, void *),
-			 void *compare_arg,
-			 struct device_node **dev_node, int *mid)
-{
-	struct device_node *dp;
-	int cur_inst;
-
-	cur_inst = 0;
-	for_each_node_by_type(dp, "cpu") {
-		int err = check_cpu_node(dp, &cur_inst,
-					 compare, compare_arg,
-					 dev_node, mid);
-		if (err == 0)
-			return 0;
-	}
-
-	return -ENODEV;
-}
-
-static int cpu_instance_compare(struct device_node *dp, int instance, void *_arg)
-{
-	int desired_instance = (int) (long) _arg;
-
-	if (instance == desired_instance)
-		return 0;
-	return -ENODEV;
-}
-
-int cpu_find_by_instance(int instance, struct device_node **dev_node, int *mid)
-{
-	return __cpu_find_by(cpu_instance_compare, (void *)(long)instance,
-			     dev_node, mid);
-}
-
-static int cpu_mid_compare(struct device_node *dp, int instance, void *_arg)
-{
-	int desired_mid = (int) (long) _arg;
-	int this_mid;
-
-	this_mid = get_cpu_mid(dp);
-	if (this_mid == desired_mid)
-		return 0;
-	return -ENODEV;
-}
-
-int cpu_find_by_mid(int mid, struct device_node **dev_node)
-{
-	return __cpu_find_by(cpu_mid_compare, (void *)(long)mid,
-			     dev_node, NULL);
-}
-
-void __init device_scan(void)
-{
-	/* FIX ME FAST... -DaveM */
-	ioport_resource.end = 0xffffffffffffffffUL;
-
-	prom_printf("Booting Linux...\n");
-
-#ifndef CONFIG_SMP
-	{
-		struct device_node *dp;
-		int err, def;
-
-		err = cpu_find_by_instance(0, &dp, NULL);
-		if (err) {
-			prom_printf("No cpu nodes, cannot continue\n");
-			prom_halt();
-		}
-		cpu_data(0).clock_tick =
-			of_getintprop_default(dp, "clock-frequency", 0);
-
-		def = ((tlb_type == hypervisor) ?
-		       (8 * 1024) :
-		       (16 * 1024));
-		cpu_data(0).dcache_size = of_getintprop_default(dp,
-								"dcache-size",
-								def);
-
-		def = 32;
-		cpu_data(0).dcache_line_size =
-			of_getintprop_default(dp, "dcache-line-size", def);
-
-		def = 16 * 1024;
-		cpu_data(0).icache_size = of_getintprop_default(dp,
-								"icache-size",
-								def);
-
-		def = 32;
-		cpu_data(0).icache_line_size =
-			of_getintprop_default(dp, "icache-line-size", def);
-
-		def = ((tlb_type == hypervisor) ?
-		       (3 * 1024 * 1024) :
-		       (4 * 1024 * 1024));
-		cpu_data(0).ecache_size = of_getintprop_default(dp,
-								"ecache-size",
-								def);
-
-		def = 64;
-		cpu_data(0).ecache_line_size =
-			of_getintprop_default(dp, "ecache-line-size", def);
-		printk("CPU[0]: Caches "
-		       "D[sz(%d):line_sz(%d)] "
-		       "I[sz(%d):line_sz(%d)] "
-		       "E[sz(%d):line_sz(%d)]\n",
-		       cpu_data(0).dcache_size, cpu_data(0).dcache_line_size,
-		       cpu_data(0).icache_size, cpu_data(0).icache_line_size,
-		       cpu_data(0).ecache_size, cpu_data(0).ecache_line_size);
-	}
-#endif
-
-	central_probe();
-
-	cpu_probe();
-}
diff --git a/arch/sparc64/kernel/entry.S b/arch/sparc64/kernel/entry.S
index b5dbd570915..f8cc3c0731c 100644
--- a/arch/sparc64/kernel/entry.S
+++ b/arch/sparc64/kernel/entry.S
@@ -1949,3 +1949,12 @@ sun4v_mach_set_soft_state:
 	ta	HV_FAST_TRAP
 	retl
 	 nop
+
+	.globl	sun4v_mach_desc
+sun4v_mach_desc:
+	mov	%o2, %o4
+	mov	HV_FAST_MACH_DESC, %o5
+	ta	HV_FAST_TRAP
+	stx	%o1, [%o4]
+	retl
+	 nop
diff --git a/arch/sparc64/kernel/irq.c b/arch/sparc64/kernel/irq.c
index 3edc18e1b81..a36f8dd0c02 100644
--- a/arch/sparc64/kernel/irq.c
+++ b/arch/sparc64/kernel/irq.c
@@ -171,8 +171,6 @@ skip:
 	return 0;
 }
 
-extern unsigned long real_hard_smp_processor_id(void);
-
 static unsigned int sun4u_compute_tid(unsigned long imap, unsigned long cpuid)
 {
 	unsigned int tid;
@@ -694,9 +692,20 @@ void init_irqwork_curcpu(void)
 	trap_block[cpu].irq_worklist = 0;
 }
 
-static void __cpuinit register_one_mondo(unsigned long paddr, unsigned long type)
+/* Please be very careful with register_one_mondo() and
+ * sun4v_register_mondo_queues().
+ *
+ * On SMP this gets invoked from the CPU trampoline before
+ * the cpu has fully taken over the trap table from OBP,
+ * and it's kernel stack + %g6 thread register state is
+ * not fully cooked yet.
+ *
+ * Therefore you cannot make any OBP calls, not even prom_printf,
+ * from these two routines.
+ */
+static void __cpuinit register_one_mondo(unsigned long paddr, unsigned long type, unsigned long qmask)
 {
-	unsigned long num_entries = 128;
+	unsigned long num_entries = (qmask + 1) / 64;
 	unsigned long status;
 
 	status = sun4v_cpu_qconf(type, paddr, num_entries);
@@ -711,44 +720,58 @@ static void __cpuinit sun4v_register_mondo_queues(int this_cpu)
 {
 	struct trap_per_cpu *tb = &trap_block[this_cpu];
 
-	register_one_mondo(tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO);
-	register_one_mondo(tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO);
-	register_one_mondo(tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR);
-	register_one_mondo(tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR);
+	register_one_mondo(tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO,
+			   tb->cpu_mondo_qmask);
+	register_one_mondo(tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO,
+			   tb->dev_mondo_qmask);
+	register_one_mondo(tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR,
+			   tb->resum_qmask);
+	register_one_mondo(tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR,
+			   tb->nonresum_qmask);
 }
 
-static void __cpuinit alloc_one_mondo(unsigned long *pa_ptr, int use_bootmem)
+static void __cpuinit alloc_one_mondo(unsigned long *pa_ptr, unsigned long qmask, int use_bootmem)
 {
-	void *page;
+	unsigned long size = PAGE_ALIGN(qmask + 1);
+	unsigned long order = get_order(size);
+	void *p = NULL;
 
-	if (use_bootmem)
-		page = alloc_bootmem_low_pages(PAGE_SIZE);
-	else
-		page = (void *) get_zeroed_page(GFP_ATOMIC);
+	if (use_bootmem) {
+		p = __alloc_bootmem_low(size, size, 0);
+	} else {
+		struct page *page = alloc_pages(GFP_ATOMIC | __GFP_ZERO, order);
+		if (page)
+			p = page_address(page);
+	}
 
-	if (!page) {
+	if (!p) {
 		prom_printf("SUN4V: Error, cannot allocate mondo queue.\n");
 		prom_halt();
 	}
 
-	*pa_ptr = __pa(page);
+	*pa_ptr = __pa(p);
 }
 
-static void __cpuinit alloc_one_kbuf(unsigned long *pa_ptr, int use_bootmem)
+static void __cpuinit alloc_one_kbuf(unsigned long *pa_ptr, unsigned long qmask, int use_bootmem)
 {
-	void *page;
+	unsigned long size = PAGE_ALIGN(qmask + 1);
+	unsigned long order = get_order(size);
+	void *p = NULL;
 
-	if (use_bootmem)
-		page = alloc_bootmem_low_pages(PAGE_SIZE);
-	else
-		page = (void *) get_zeroed_page(GFP_ATOMIC);
+	if (use_bootmem) {
+		p = __alloc_bootmem_low(size, size, 0);
+	} else {
+		struct page *page = alloc_pages(GFP_ATOMIC | __GFP_ZERO, order);
+		if (page)
+			p = page_address(page);
+	}
 
-	if (!page) {
+	if (!p) {
 		prom_printf("SUN4V: Error, cannot allocate kbuf page.\n");
 		prom_halt();
 	}
 
-	*pa_ptr = __pa(page);
+	*pa_ptr = __pa(p);
 }
 
 static void __cpuinit init_cpu_send_mondo_info(struct trap_per_cpu *tb, int use_bootmem)
@@ -779,12 +802,12 @@ void __cpuinit sun4v_init_mondo_queues(int use_bootmem, int cpu, int alloc, int
 	struct trap_per_cpu *tb = &trap_block[cpu];
 
 	if (alloc) {
-		alloc_one_mondo(&tb->cpu_mondo_pa, use_bootmem);
-		alloc_one_mondo(&tb->dev_mondo_pa, use_bootmem);
-		alloc_one_mondo(&tb->resum_mondo_pa, use_bootmem);
-		alloc_one_kbuf(&tb->resum_kernel_buf_pa, use_bootmem);
-		alloc_one_mondo(&tb->nonresum_mondo_pa, use_bootmem);
-		alloc_one_kbuf(&tb->nonresum_kernel_buf_pa, use_bootmem);
+		alloc_one_mondo(&tb->cpu_mondo_pa, tb->cpu_mondo_qmask, use_bootmem);
+		alloc_one_mondo(&tb->dev_mondo_pa, tb->dev_mondo_qmask, use_bootmem);
+		alloc_one_mondo(&tb->resum_mondo_pa, tb->resum_qmask, use_bootmem);
+		alloc_one_kbuf(&tb->resum_kernel_buf_pa, tb->resum_qmask, use_bootmem);
+		alloc_one_mondo(&tb->nonresum_mondo_pa, tb->nonresum_qmask, use_bootmem);
+		alloc_one_kbuf(&tb->nonresum_kernel_buf_pa, tb->nonresum_qmask, use_bootmem);
 
 		init_cpu_send_mondo_info(tb, use_bootmem);
 	}
diff --git a/arch/sparc64/kernel/mdesc.c b/arch/sparc64/kernel/mdesc.c
new file mode 100644
index 00000000000..9246c2cf957
--- /dev/null
+++ b/arch/sparc64/kernel/mdesc.c
@@ -0,0 +1,619 @@
+/* mdesc.c: Sun4V machine description handling.
+ *
+ * Copyright (C) 2007 David S. Miller <davem@davemloft.net>
+ */
+#include <linux/kernel.h>
+#include <linux/types.h>
+#include <linux/bootmem.h>
+#include <linux/log2.h>
+
+#include <asm/hypervisor.h>
+#include <asm/mdesc.h>
+#include <asm/prom.h>
+#include <asm/oplib.h>
+#include <asm/smp.h>
+
+/* Unlike the OBP device tree, the machine description is a full-on
+ * DAG.  An arbitrary number of ARCs are possible from one
+ * node to other nodes and thus we can't use the OBP device_node
+ * data structure to represent these nodes inside of the kernel.
+ *
+ * Actually, it isn't even a DAG, because there are back pointers
+ * which create cycles in the graph.
+ *
+ * mdesc_hdr and mdesc_elem describe the layout of the data structure
+ * we get from the Hypervisor.
+ */
+struct mdesc_hdr {
+	u32	version; /* Transport version */
+	u32	node_sz; /* node block size */
+	u32	name_sz; /* name block size */
+	u32	data_sz; /* data block size */
+};
+
+struct mdesc_elem {
+	u8	tag;
+#define MD_LIST_END	0x00
+#define MD_NODE		0x4e
+#define MD_NODE_END	0x45
+#define MD_NOOP		0x20
+#define MD_PROP_ARC	0x61
+#define MD_PROP_VAL	0x76
+#define MD_PROP_STR	0x73
+#define MD_PROP_DATA	0x64
+	u8	name_len;
+	u16	resv;
+	u32	name_offset;
+	union {
+		struct {
+			u32	data_len;
+			u32	data_offset;
+		} data;
+		u64	val;
+	} d;
+};
+
+static struct mdesc_hdr *main_mdesc;
+static struct mdesc_node *allnodes;
+
+static struct mdesc_node *allnodes_tail;
+static unsigned int unique_id;
+
+static struct mdesc_node **mdesc_hash;
+static unsigned int mdesc_hash_size;
+
+static inline unsigned int node_hashfn(u64 node)
+{
+	return ((unsigned int) (node ^ (node >> 8) ^ (node >> 16)))
+		& (mdesc_hash_size - 1);
+}
+
+static inline void hash_node(struct mdesc_node *mp)
+{
+	struct mdesc_node **head = &mdesc_hash[node_hashfn(mp->node)];
+
+	mp->hash_next = *head;
+	*head = mp;
+
+	if (allnodes_tail) {
+		allnodes_tail->allnodes_next = mp;
+		allnodes_tail = mp;
+	} else {
+		allnodes = allnodes_tail = mp;
+	}
+}
+
+static struct mdesc_node *find_node(u64 node)
+{
+	struct mdesc_node *mp = mdesc_hash[node_hashfn(node)];
+
+	while (mp) {
+		if (mp->node == node)
+			return mp;
+
+		mp = mp->hash_next;
+	}
+	return NULL;
+}
+
+struct property *md_find_property(const struct mdesc_node *mp,
+				  const char *name,
+				  int *lenp)
+{
+	struct property *pp;
+
+	for (pp = mp->properties; pp != 0; pp = pp->next) {
+		if (strcasecmp(pp->name, name) == 0) {
+			if (lenp)
+				*lenp = pp->length;
+			break;
+		}
+	}
+	return pp;
+}
+EXPORT_SYMBOL(md_find_property);
+
+/*
+ * Find a property with a given name for a given node
+ * and return the value.
+ */
+const void *md_get_property(const struct mdesc_node *mp, const char *name,
+			    int *lenp)
+{
+	struct property *pp = md_find_property(mp, name, lenp);
+	return pp ? pp->value : NULL;
+}
+EXPORT_SYMBOL(md_get_property);
+
+struct mdesc_node *md_find_node_by_name(struct mdesc_node *from,
+					const char *name)
+{
+	struct mdesc_node *mp;
+
+	mp = from ? from->allnodes_next : allnodes;
+	for (; mp != NULL; mp = mp->allnodes_next) {
+		if (strcmp(mp->name, name) == 0)
+			break;
+	}
+	return mp;
+}
+EXPORT_SYMBOL(md_find_node_by_name);
+
+static unsigned int mdesc_early_allocated;
+
+static void * __init mdesc_early_alloc(unsigned long size)
+{
+	void *ret;
+
+	ret = __alloc_bootmem(size, SMP_CACHE_BYTES, 0UL);
+	if (ret == NULL) {
+		prom_printf("MDESC: alloc of %lu bytes failed.\n", size);
+		prom_halt();
+	}
+
+	memset(ret, 0, size);
+
+	mdesc_early_allocated += size;
+
+	return ret;
+}
+
+static unsigned int __init count_arcs(struct mdesc_elem *ep)
+{
+	unsigned int ret = 0;
+
+	ep++;
+	while (ep->tag != MD_NODE_END) {
+		if (ep->tag == MD_PROP_ARC)
+			ret++;
+		ep++;
+	}
+	return ret;
+}
+
+static void __init mdesc_node_alloc(u64 node, struct mdesc_elem *ep, const char *names)
+{
+	unsigned int num_arcs = count_arcs(ep);
+	struct mdesc_node *mp;
+
+	mp = mdesc_early_alloc(sizeof(*mp) +
+			       (num_arcs * sizeof(struct mdesc_arc)));
+	mp->name = names + ep->name_offset;
+	mp->node = node;
+	mp->unique_id = unique_id++;
+	mp->num_arcs = num_arcs;
+
+	hash_node(mp);
+}
+
+static inline struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
+{
+	return (struct mdesc_elem *) (mdesc + 1);
+}
+
+static inline void *name_block(struct mdesc_hdr *mdesc)
+{
+	return ((void *) node_block(mdesc)) + mdesc->node_sz;
+}
+
+static inline void *data_block(struct mdesc_hdr *mdesc)
+{
+	return ((void *) name_block(mdesc)) + mdesc->name_sz;
+}
+
+/* In order to avoid recursion (the graph can be very deep) we use a
+ * two pass algorithm.  First we allocate all the nodes and hash them.
+ * Then we iterate over each node, filling in the arcs and properties.
+ */
+static void __init build_all_nodes(struct mdesc_hdr *mdesc)
+{
+	struct mdesc_elem *start, *ep;
+	struct mdesc_node *mp;
+	const char *names;
+	void *data;
+	u64 last_node;
+
+	start = ep = node_block(mdesc);
+	last_node = mdesc->node_sz / 16;
+
+	names = name_block(mdesc);
+
+	while (1) {
+		u64 node = ep - start;
+
+		if (ep->tag == MD_LIST_END)
+			break;
+
+		if (ep->tag != MD_NODE) {
+			prom_printf("MDESC: Inconsistent element list.\n");
+			prom_halt();
+		}
+
+		mdesc_node_alloc(node, ep, names);
+
+		if (ep->d.val >= last_node) {
+			printk("MDESC: Warning, early break out of node scan.\n");
+			printk("MDESC: Next node [%lu] last_node [%lu].\n",
+			       node, last_node);
+			break;
+		}
+
+		ep = start + ep->d.val;
+	}
+
+	data = data_block(mdesc);
+	for (mp = allnodes; mp; mp = mp->allnodes_next) {
+		struct mdesc_elem *ep = start + mp->node;
+		struct property **link = &mp->properties;
+		unsigned int this_arc = 0;
+
+		ep++;
+		while (ep->tag != MD_NODE_END) {
+			switch (ep->tag) {
+			case MD_PROP_ARC: {
+				struct mdesc_node *target;
+
+				if (this_arc >= mp->num_arcs) {
+					prom_printf("MDESC: ARC overrun [%u:%u]\n",
+						    this_arc, mp->num_arcs);
+					prom_halt();
+				}
+				target = find_node(ep->d.val);
+				if (!target) {
+					printk("MDESC: Warning, arc points to "
+					       "missing node, ignoring.\n");
+					break;
+				}
+				mp->arcs[this_arc].name =
+					(names + ep->name_offset);
+				mp->arcs[this_arc].arc = target;
+				this_arc++;
+				break;
+			}
+
+			case MD_PROP_VAL:
+			case MD_PROP_STR:
+			case MD_PROP_DATA: {
+				struct property *p = mdesc_early_alloc(sizeof(*p));
+
+				p->unique_id = unique_id++;
+				p->name = (char *) names + ep->name_offset;
+				if (ep->tag == MD_PROP_VAL) {
+					p->value = &ep->d.val;
+					p->length = 8;
+				} else {
+					p->value = data + ep->d.data.data_offset;
+					p->length = ep->d.data.data_len;
+				}
+				*link = p;
+				link = &p->next;
+				break;
+			}
+
+			case MD_NOOP:
+				break;
+
+			default:
+				printk("MDESC: Warning, ignoring unknown tag type %02x\n",
+				       ep->tag);
+			}
+			ep++;
+		}
+	}
+}
+
+static unsigned int __init count_nodes(struct mdesc_hdr *mdesc)
+{
+	struct mdesc_elem *ep = node_block(mdesc);
+	struct mdesc_elem *end;
+	unsigned int cnt = 0;
+
+	end = ((void *)ep) + mdesc->node_sz;
+	while (ep < end) {
+		if (ep->tag == MD_NODE)
+			cnt++;
+		ep++;
+	}
+	return cnt;
+}
+
+static void __init report_platform_properties(void)
+{
+	struct mdesc_node *pn = md_find_node_by_name(NULL, "platform");
+	const char *s;
+	const u64 *v;
+
+	if (!pn) {
+		prom_printf("No platform node in machine-description.\n");
+		prom_halt();
+	}
+
+	s = md_get_property(pn, "banner-name", NULL);
+	printk("PLATFORM: banner-name [%s]\n", s);
+	s = md_get_property(pn, "name", NULL);
+	printk("PLATFORM: name [%s]\n", s);
+
+	v = md_get_property(pn, "hostid", NULL);
+	if (v)
+		printk("PLATFORM: hostid [%08lx]\n", *v);
+	v = md_get_property(pn, "serial#", NULL);
+	if (v)
+		printk("PLATFORM: serial# [%08lx]\n", *v);
+	v = md_get_property(pn, "stick-frequency", NULL);
+	printk("PLATFORM: stick-frequency [%08lx]\n", *v);
+	v = md_get_property(pn, "mac-address", NULL);
+	if (v)
+		printk("PLATFORM: mac-address [%lx]\n", *v);
+	v = md_get_property(pn, "watchdog-resolution", NULL);
+	if (v)
+		printk("PLATFORM: watchdog-resolution [%lu ms]\n", *v);
+	v = md_get_property(pn, "watchdog-max-timeout", NULL);
+	if (v)
+		printk("PLATFORM: watchdog-max-timeout [%lu ms]\n", *v);
+	v = md_get_property(pn, "max-cpus", NULL);
+	if (v)
+		printk("PLATFORM: max-cpus [%lu]\n", *v);
+}
+
+static int inline find_in_proplist(const char *list, const char *match, int len)
+{
+	while (len > 0) {
+		int l;
+
+		if (!strcmp(list, match))
+			return 1;
+		l = strlen(list) + 1;
+		list += l;
+		len -= l;
+	}
+	return 0;
+}
+
+static void __init fill_in_one_cache(cpuinfo_sparc *c, struct mdesc_node *mp)
+{
+	const u64 *level = md_get_property(mp, "level", NULL);
+	const u64 *size = md_get_property(mp, "size", NULL);
+	const u64 *line_size = md_get_property(mp, "line-size", NULL);
+	const char *type;
+	int type_len;
+
+	type = md_get_property(mp, "type", &type_len);
+
+	switch (*level) {
+	case 1:
+		if (find_in_proplist(type, "instn", type_len)) {
+			c->icache_size = *size;
+			c->icache_line_size = *line_size;
+		} else if (find_in_proplist(type, "data", type_len)) {
+			c->dcache_size = *size;
+			c->dcache_line_size = *line_size;
+		}
+		break;
+
+	case 2:
+		c->ecache_size = *size;
+		c->ecache_line_size = *line_size;
+		break;
+
+	default:
+		break;
+	}
+
+	if (*level == 1) {
+		unsigned int i;
+
+		for (i = 0; i < mp->num_arcs; i++) {
+			struct mdesc_node *t = mp->arcs[i].arc;
+
+			if (strcmp(mp->arcs[i].name, "fwd"))
+				continue;
+
+			if (!strcmp(t->name, "cache"))
+				fill_in_one_cache(c, t);
+		}
+	}
+}
+
+static void __init mark_core_ids(struct mdesc_node *mp, int core_id)
+{
+	unsigned int i;
+
+	for (i = 0; i < mp->num_arcs; i++) {
+		struct mdesc_node *t = mp->arcs[i].arc;
+		const u64 *id;
+
+		if (strcmp(mp->arcs[i].name, "back"))
+			continue;
+
+		if (!strcmp(t->name, "cpu")) {
+			id = md_get_property(t, "id", NULL);
+			if (*id < NR_CPUS)
+				cpu_data(*id).core_id = core_id;
+		} else {
+			unsigned int j;
+
+			for (j = 0; j < t->num_arcs; j++) {
+				struct mdesc_node *n = t->arcs[j].arc;
+
+				if (strcmp(t->arcs[j].name, "back"))
+					continue;
+
+				if (strcmp(n->name, "cpu"))
+					continue;
+
+				id = md_get_property(n, "id", NULL);
+				if (*id < NR_CPUS)
+					cpu_data(*id).core_id = core_id;
+			}
+		}
+	}
+}
+
+static void __init set_core_ids(void)
+{
+	struct mdesc_node *mp;
+	int idx;
+
+	idx = 1;
+	md_for_each_node_by_name(mp, "cache") {
+		const u64 *level = md_get_property(mp, "level", NULL);
+		const char *type;
+		int len;
+
+		if (*level != 1)
+			continue;
+
+		type = md_get_property(mp, "type", &len);
+		if (!find_in_proplist(type, "instn", len))
+			continue;
+
+		mark_core_ids(mp, idx);
+
+		idx++;
+	}
+}
+
+static void __init get_one_mondo_bits(const u64 *p, unsigned int *mask, unsigned char def)
+{
+	u64 val;
+
+	if (!p)
+		goto use_default;
+	val = *p;
+
+	if (!val || val >= 64)
+		goto use_default;
+
+	*mask = ((1U << val) * 64U) - 1U;
+	return;
+
+use_default:
+	*mask = ((1U << def) * 64U) - 1U;
+}
+
+static void __init get_mondo_data(struct mdesc_node *mp, struct trap_per_cpu *tb)
+{
+	const u64 *val;
+
+	val = md_get_property(mp, "q-cpu-mondo-#bits", NULL);
+	get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7);
+
+	val = md_get_property(mp, "q-dev-mondo-#bits", NULL);
+	get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7);
+
+	val = md_get_property(mp, "q-resumable-#bits", NULL);
+	get_one_mondo_bits(val, &tb->resum_qmask, 6);
+
+	val = md_get_property(mp, "q-nonresumable-#bits", NULL);
+	get_one_mondo_bits(val, &tb->nonresum_qmask, 2);
+}
+
+static void __init mdesc_fill_in_cpu_data(void)
+{
+	struct mdesc_node *mp;
+
+	ncpus_probed = 0;
+	md_for_each_node_by_name(mp, "cpu") {
+		const u64 *id = md_get_property(mp, "id", NULL);
+		const u64 *cfreq = md_get_property(mp, "clock-frequency", NULL);
+		struct trap_per_cpu *tb;
+		cpuinfo_sparc *c;
+		unsigned int i;
+		int cpuid;
+
+		ncpus_probed++;
+
+		cpuid = *id;
+
+#ifdef CONFIG_SMP
+		if (cpuid >= NR_CPUS)
+			continue;
+#else
+		/* On uniprocessor we only want the values for the
+		 * real physical cpu the kernel booted onto, however
+		 * cpu_data() only has one entry at index 0.
+		 */
+		if (cpuid != real_hard_smp_processor_id())
+			continue;
+		cpuid = 0;
+#endif
+
+		c = &cpu_data(cpuid);
+		c->clock_tick = *cfreq;
+
+		tb = &trap_block[cpuid];
+		get_mondo_data(mp, tb);
+
+		for (i = 0; i < mp->num_arcs; i++) {
+			struct mdesc_node *t = mp->arcs[i].arc;
+			unsigned int j;
+
+			if (strcmp(mp->arcs[i].name, "fwd"))
+				continue;
+
+			if (!strcmp(t->name, "cache")) {
+				fill_in_one_cache(c, t);
+				continue;
+			}
+
+			for (j = 0; j < t->num_arcs; j++) {
+				struct mdesc_node *n;
+
+				n = t->arcs[j].arc;
+				if (strcmp(t->arcs[j].name, "fwd"))
+					continue;
+
+				if (!strcmp(n->name, "cache"))
+					fill_in_one_cache(c, n);
+			}
+		}
+
+#ifdef CONFIG_SMP
+		cpu_set(cpuid, cpu_present_map);
+		cpu_set(cpuid, phys_cpu_present_map);
+#endif
+
+		c->core_id = 0;
+	}
+
+	set_core_ids();
+
+	smp_fill_in_sib_core_maps();
+}
+
+void __init sun4v_mdesc_init(void)
+{
+	unsigned long len, real_len, status;
+
+	(void) sun4v_mach_desc(0UL, 0UL, &len);
+
+	printk("MDESC: Size is %lu bytes.\n", len);
+
+	main_mdesc = mdesc_early_alloc(len);
+
+	status = sun4v_mach_desc(__pa(main_mdesc), len, &real_len);
+	if (status != HV_EOK || real_len > len) {
+		prom_printf("sun4v_mach_desc fails, err(%lu), "
+			    "len(%lu), real_len(%lu)\n",
+			    status, len, real_len);
+		prom_halt();
+	}
+
+	len = count_nodes(main_mdesc);
+	printk("MDESC: %lu nodes.\n", len);
+
+	len = roundup_pow_of_two(len);
+
+	mdesc_hash = mdesc_early_alloc(len * sizeof(struct mdesc_node *));
+	mdesc_hash_size = len;
+
+	printk("MDESC: Hash size %lu entries.\n", len);
+
+	build_all_nodes(main_mdesc);
+
+	printk("MDESC: Built graph with %u bytes of memory.\n",
+	       mdesc_early_allocated);
+
+	report_platform_properties();
+	mdesc_fill_in_cpu_data();
+}
diff --git a/arch/sparc64/kernel/pci_sabre.c b/arch/sparc64/kernel/pci_sabre.c
index e2377796de8..323d6c27851 100644
--- a/arch/sparc64/kernel/pci_sabre.c
+++ b/arch/sparc64/kernel/pci_sabre.c
@@ -762,9 +762,10 @@ void sabre_init(struct device_node *dp, char *model_name)
 			/* Of course, Sun has to encode things a thousand
 			 * different ways, inconsistently.
 			 */
-			cpu_find_by_instance(0, &dp, NULL);
-			if (!strcmp(dp->name, "SUNW,UltraSPARC-IIe"))
-				hummingbird_p = 1;
+			for_each_node_by_type(dp, "cpu") {
+				if (!strcmp(dp->name, "SUNW,UltraSPARC-IIe"))
+					hummingbird_p = 1;
+			}
 		}
 	}
 
diff --git a/arch/sparc64/kernel/prom.c b/arch/sparc64/kernel/prom.c
index 02830e4671f..dad4b3ba705 100644
--- a/arch/sparc64/kernel/prom.c
+++ b/arch/sparc64/kernel/prom.c
@@ -28,6 +28,7 @@
 #include <asm/irq.h>
 #include <asm/asi.h>
 #include <asm/upa.h>
+#include <asm/smp.h>
 
 static struct device_node *allnodes;
 
@@ -1665,6 +1666,150 @@ static struct device_node * __init build_tree(struct device_node *parent, phandl
 	return ret;
 }
 
+static const char *get_mid_prop(void)
+{
+	return (tlb_type == spitfire ? "upa-portid" : "portid");
+}
+
+struct device_node *of_find_node_by_cpuid(int cpuid)
+{
+	struct device_node *dp;
+	const char *mid_prop = get_mid_prop();
+
+	for_each_node_by_type(dp, "cpu") {
+		int id = of_getintprop_default(dp, mid_prop, -1);
+		const char *this_mid_prop = mid_prop;
+
+		if (id < 0) {
+			this_mid_prop = "cpuid";
+			id = of_getintprop_default(dp, this_mid_prop, -1);
+		}
+
+		if (id < 0) {
+			prom_printf("OF: Serious problem, cpu lacks "
+				    "%s property", this_mid_prop);
+			prom_halt();
+		}
+		if (cpuid == id)
+			return dp;
+	}
+	return NULL;
+}
+
+static void __init of_fill_in_cpu_data(void)
+{
+	struct device_node *dp;
+	const char *mid_prop = get_mid_prop();
+
+	ncpus_probed = 0;
+	for_each_node_by_type(dp, "cpu") {
+		int cpuid = of_getintprop_default(dp, mid_prop, -1);
+		const char *this_mid_prop = mid_prop;
+		struct device_node *portid_parent;
+		int portid = -1;
+
+		portid_parent = NULL;
+		if (cpuid < 0) {
+			this_mid_prop = "cpuid";
+			cpuid = of_getintprop_default(dp, this_mid_prop, -1);
+			if (cpuid >= 0) {
+				int limit = 2;
+
+				portid_parent = dp;
+				while (limit--) {
+					portid_parent = portid_parent->parent;
+					if (!portid_parent)
+						break;
+					portid = of_getintprop_default(portid_parent,
+								       "portid", -1);
+					if (portid >= 0)
+						break;
+				}
+			}
+		}
+
+		if (cpuid < 0) {
+			prom_printf("OF: Serious problem, cpu lacks "
+				    "%s property", this_mid_prop);
+			prom_halt();
+		}
+
+		ncpus_probed++;
+
+#ifdef CONFIG_SMP
+		if (cpuid >= NR_CPUS)
+			continue;
+#else
+		/* On uniprocessor we only want the values for the
+		 * real physical cpu the kernel booted onto, however
+		 * cpu_data() only has one entry at index 0.
+		 */
+		if (cpuid != real_hard_smp_processor_id())
+			continue;
+		cpuid = 0;
+#endif
+
+		cpu_data(cpuid).clock_tick =
+			of_getintprop_default(dp, "clock-frequency", 0);
+
+		if (portid_parent) {
+			cpu_data(cpuid).dcache_size =
+				of_getintprop_default(dp, "l1-dcache-size",
+						      16 * 1024);
+			cpu_data(cpuid).dcache_line_size =
+				of_getintprop_default(dp, "l1-dcache-line-size",
+						      32);
+			cpu_data(cpuid).icache_size =
+				of_getintprop_default(dp, "l1-icache-size",
+						      8 * 1024);
+			cpu_data(cpuid).icache_line_size =
+				of_getintprop_default(dp, "l1-icache-line-size",
+						      32);
+			cpu_data(cpuid).ecache_size =
+				of_getintprop_default(dp, "l2-cache-size", 0);
+			cpu_data(cpuid).ecache_line_size =
+				of_getintprop_default(dp, "l2-cache-line-size", 0);
+			if (!cpu_data(cpuid).ecache_size ||
+			    !cpu_data(cpuid).ecache_line_size) {
+				cpu_data(cpuid).ecache_size =
+					of_getintprop_default(portid_parent,
+							      "l2-cache-size",
+							      (4 * 1024 * 1024));
+				cpu_data(cpuid).ecache_line_size =
+					of_getintprop_default(portid_parent,
+							      "l2-cache-line-size", 64);
+			}
+
+			cpu_data(cpuid).core_id = portid + 1;
+		} else {
+			cpu_data(cpuid).dcache_size =
+				of_getintprop_default(dp, "dcache-size", 16 * 1024);
+			cpu_data(cpuid).dcache_line_size =
+				of_getintprop_default(dp, "dcache-line-size", 32);
+
+			cpu_data(cpuid).icache_size =
+				of_getintprop_default(dp, "icache-size", 16 * 1024);
+			cpu_data(cpuid).icache_line_size =
+				of_getintprop_default(dp, "icache-line-size", 32);
+
+			cpu_data(cpuid).ecache_size =
+				of_getintprop_default(dp, "ecache-size",
+						      (4 * 1024 * 1024));
+			cpu_data(cpuid).ecache_line_size =
+				of_getintprop_default(dp, "ecache-line-size", 64);
+
+			cpu_data(cpuid).core_id = 0;
+		}
+
+#ifdef CONFIG_SMP
+		cpu_set(cpuid, cpu_present_map);
+		cpu_set(cpuid, phys_cpu_present_map);
+#endif
+	}
+
+	smp_fill_in_sib_core_maps();
+}
+
 void __init prom_build_devicetree(void)
 {
 	struct device_node **nextp;
@@ -1679,4 +1824,7 @@ void __init prom_build_devicetree(void)
 				     &nextp);
 	printk("PROM: Built device tree with %u bytes of memory.\n",
 	       prom_early_allocated);
+
+	if (tlb_type != hypervisor)
+		of_fill_in_cpu_data();
 }
diff --git a/arch/sparc64/kernel/setup.c b/arch/sparc64/kernel/setup.c
index dea9c3c9ec5..de9b4c13f1c 100644
--- a/arch/sparc64/kernel/setup.c
+++ b/arch/sparc64/kernel/setup.c
@@ -46,11 +46,17 @@
 #include <asm/sections.h>
 #include <asm/setup.h>
 #include <asm/mmu.h>
+#include <asm/ns87303.h>
 
 #ifdef CONFIG_IP_PNP
 #include <net/ipconfig.h>
 #endif
 
+/* Used to synchronize accesses to NatSemi SUPER I/O chip configure
+ * operations in asm/ns87303.h
+ */
+DEFINE_SPINLOCK(ns87303_lock);
+
 struct screen_info screen_info = {
 	0, 0,			/* orig-x, orig-y */
 	0,			/* unused */
@@ -370,8 +376,6 @@ void __init setup_arch(char **cmdline_p)
 	init_cur_cpu_trap(current_thread_info());
 
 	paging_init();
-
-	smp_setup_cpu_possible_map();
 }
 
 static int __init set_preferred_console(void)
@@ -424,7 +428,7 @@ extern void mmu_info(struct seq_file *);
 unsigned int dcache_parity_tl1_occurred;
 unsigned int icache_parity_tl1_occurred;
 
-static int ncpus_probed;
+int ncpus_probed;
 
 static int show_cpuinfo(struct seq_file *m, void *__unused)
 {
@@ -516,14 +520,6 @@ static int __init topology_init(void)
 
 	err = -ENOMEM;
 
-	/* Count the number of physically present processors in
-	 * the machine, even on uniprocessor, so that /proc/cpuinfo
-	 * output is consistent with 2.4.x
-	 */
-	ncpus_probed = 0;
-	while (!cpu_find_by_instance(ncpus_probed, NULL, NULL))
-		ncpus_probed++;
-
 	for_each_possible_cpu(i) {
 		struct cpu *p = kzalloc(sizeof(*p), GFP_KERNEL);
 		if (p) {
diff --git a/arch/sparc64/kernel/smp.c b/arch/sparc64/kernel/smp.c
index 24fdf1d0adc..f7fa873c800 100644
--- a/arch/sparc64/kernel/smp.c
+++ b/arch/sparc64/kernel/smp.c
@@ -40,6 +40,7 @@
 #include <asm/tlb.h>
 #include <asm/sections.h>
 #include <asm/prom.h>
+#include <asm/mdesc.h>
 
 extern void calibrate_delay(void);
 
@@ -75,53 +76,6 @@ void smp_bogo(struct seq_file *m)
 			   i, cpu_data(i).clock_tick);
 }
 
-void __init smp_store_cpu_info(int id)
-{
-	struct device_node *dp;
-	int def;
-
-	cpu_data(id).udelay_val			= loops_per_jiffy;
-
-	cpu_find_by_mid(id, &dp);
-	cpu_data(id).clock_tick =
-		of_getintprop_default(dp, "clock-frequency", 0);
-
-	def = ((tlb_type == hypervisor) ? (8 * 1024) : (16 * 1024));
-	cpu_data(id).dcache_size =
-		of_getintprop_default(dp, "dcache-size", def);
-
-	def = 32;
-	cpu_data(id).dcache_line_size =
-		of_getintprop_default(dp, "dcache-line-size", def);
-
-	def = 16 * 1024;
-	cpu_data(id).icache_size =
-		of_getintprop_default(dp, "icache-size", def);
-
-	def = 32;
-	cpu_data(id).icache_line_size =
-		of_getintprop_default(dp, "icache-line-size", def);
-
-	def = ((tlb_type == hypervisor) ?
-	       (3 * 1024 * 1024) :
-	       (4 * 1024 * 1024));
-	cpu_data(id).ecache_size =
-		of_getintprop_default(dp, "ecache-size", def);
-
-	def = 64;
-	cpu_data(id).ecache_line_size =
-		of_getintprop_default(dp, "ecache-line-size", def);
-
-	printk("CPU[%d]: Caches "
-	       "D[sz(%d):line_sz(%d)] "
-	       "I[sz(%d):line_sz(%d)] "
-	       "E[sz(%d):line_sz(%d)]\n",
-	       id,
-	       cpu_data(id).dcache_size, cpu_data(id).dcache_line_size,
-	       cpu_data(id).icache_size, cpu_data(id).icache_line_size,
-	       cpu_data(id).ecache_size, cpu_data(id).ecache_line_size);
-}
-
 extern void setup_sparc64_timer(void);
 
 static volatile unsigned long callin_flag = 0;
@@ -145,7 +99,7 @@ void __init smp_callin(void)
 	local_irq_enable();
 
 	calibrate_delay();
-	smp_store_cpu_info(cpuid);
+	cpu_data(cpuid).udelay_val = loops_per_jiffy;
 	callin_flag = 1;
 	__asm__ __volatile__("membar #Sync\n\t"
 			     "flush  %%g6" : : : "memory");
@@ -340,9 +294,8 @@ static int __devinit smp_boot_one_cpu(unsigned int cpu)
 
 		prom_startcpu_cpuid(cpu, entry, cookie);
 	} else {
-		struct device_node *dp;
+		struct device_node *dp = of_find_node_by_cpuid(cpu);
 
-		cpu_find_by_mid(cpu, &dp);
 		prom_startcpu(dp->node, entry, cookie);
 	}
 
@@ -1191,23 +1144,14 @@ int setup_profiling_timer(unsigned int multiplier)
 
 static void __init smp_tune_scheduling(void)
 {
-	struct device_node *dp;
-	int instance;
-	unsigned int def, smallest = ~0U;
-
-	def = ((tlb_type == hypervisor) ?
-	       (3 * 1024 * 1024) :
-	       (4 * 1024 * 1024));
+	unsigned int smallest = ~0U;
+	int i;
 
-	instance = 0;
-	while (!cpu_find_by_instance(instance, &dp, NULL)) {
-		unsigned int val;
+	for (i = 0; i < NR_CPUS; i++) {
+		unsigned int val = cpu_data(i).ecache_size;
 
-		val = of_getintprop_default(dp, "ecache-size", def);
-		if (val < smallest)
+		if (val && val < smallest)
 			smallest = val;
-
-		instance++;
 	}
 
 	/* Any value less than 256K is nonsense.  */
@@ -1230,58 +1174,42 @@ void __init smp_prepare_cpus(unsigned int max_cpus)
 	int i;
 
 	if (num_possible_cpus() > max_cpus) {
-		int instance, mid;
-
-		instance = 0;
-		while (!cpu_find_by_instance(instance, NULL, &mid)) {
-			if (mid != boot_cpu_id) {
-				cpu_clear(mid, phys_cpu_present_map);
-				cpu_clear(mid, cpu_present_map);
+		for_each_possible_cpu(i) {
+			if (i != boot_cpu_id) {
+				cpu_clear(i, phys_cpu_present_map);
+				cpu_clear(i, cpu_present_map);
 				if (num_possible_cpus() <= max_cpus)
 					break;
 			}
-			instance++;
-		}
-	}
-
-	for_each_possible_cpu(i) {
-		if (tlb_type == hypervisor) {
-			int j;
-
-			/* XXX get this mapping from machine description */
-			for_each_possible_cpu(j) {
-				if ((j >> 2) == (i >> 2))
-					cpu_set(j, cpu_sibling_map[i]);
-			}
-		} else {
-			cpu_set(i, cpu_sibling_map[i]);
 		}
 	}
 
-	smp_store_cpu_info(boot_cpu_id);
+	cpu_data(boot_cpu_id).udelay_val = loops_per_jiffy;
 	smp_tune_scheduling();
 }
 
-/* Set this up early so that things like the scheduler can init
- * properly.  We use the same cpu mask for both the present and
- * possible cpu map.
- */
-void __init smp_setup_cpu_possible_map(void)
+void __devinit smp_prepare_boot_cpu(void)
 {
-	int instance, mid;
-
-	instance = 0;
-	while (!cpu_find_by_instance(instance, NULL, &mid)) {
-		if (mid < NR_CPUS) {
-			cpu_set(mid, phys_cpu_present_map);
-			cpu_set(mid, cpu_present_map);
-		}
-		instance++;
-	}
 }
 
-void __devinit smp_prepare_boot_cpu(void)
+void __devinit smp_fill_in_sib_core_maps(void)
 {
+	unsigned int i;
+
+	for_each_possible_cpu(i) {
+		unsigned int j;
+
+		if (cpu_data(i).core_id == 0) {
+			cpu_set(i, cpu_sibling_map[i]);
+			continue;
+		}
+
+		for_each_possible_cpu(j) {
+			if (cpu_data(i).core_id ==
+			    cpu_data(j).core_id)
+				cpu_set(j, cpu_sibling_map[i]);
+		}
+	}
 }
 
 int __cpuinit __cpu_up(unsigned int cpu)
@@ -1337,7 +1265,7 @@ unsigned long __per_cpu_shift __read_mostly;
 EXPORT_SYMBOL(__per_cpu_base);
 EXPORT_SYMBOL(__per_cpu_shift);
 
-void __init setup_per_cpu_areas(void)
+void __init real_setup_per_cpu_areas(void)
 {
 	unsigned long goal, size, i;
 	char *ptr;
diff --git a/arch/sparc64/kernel/sun4v_ivec.S b/arch/sparc64/kernel/sun4v_ivec.S
index 405855dd886..574bc248bca 100644
--- a/arch/sparc64/kernel/sun4v_ivec.S
+++ b/arch/sparc64/kernel/sun4v_ivec.S
@@ -22,12 +22,12 @@ sun4v_cpu_mondo:
 	be,pn	%xcc, sun4v_cpu_mondo_queue_empty
 	 nop
 
-	/* Get &trap_block[smp_processor_id()] into %g3.  */
-	ldxa	[%g0] ASI_SCRATCHPAD, %g3
-	sub	%g3, TRAP_PER_CPU_FAULT_INFO, %g3
+	/* Get &trap_block[smp_processor_id()] into %g4.  */
+	ldxa	[%g0] ASI_SCRATCHPAD, %g4
+	sub	%g4, TRAP_PER_CPU_FAULT_INFO, %g4
 
 	/* Get CPU mondo queue base phys address into %g7.  */
-	ldx	[%g3 + TRAP_PER_CPU_CPU_MONDO_PA], %g7
+	ldx	[%g4 + TRAP_PER_CPU_CPU_MONDO_PA], %g7
 
 	/* Now get the cross-call arguments and handler PC, same
 	 * layout as sun4u:
@@ -47,8 +47,7 @@ sun4v_cpu_mondo:
 	add	%g2, 0x40 - 0x8 - 0x8, %g2
 
 	/* Update queue head pointer.  */
-	sethi	%hi(8192 - 1), %g4
-	or	%g4, %lo(8192 - 1), %g4
+	lduw	[%g4 + TRAP_PER_CPU_CPU_MONDO_QMASK], %g4
 	and	%g2, %g4, %g2
 
 	mov	INTRQ_CPU_MONDO_HEAD, %g4
@@ -71,12 +70,12 @@ sun4v_dev_mondo:
 	be,pn	%xcc, sun4v_dev_mondo_queue_empty
 	 nop
 
-	/* Get &trap_block[smp_processor_id()] into %g3.  */
-	ldxa	[%g0] ASI_SCRATCHPAD, %g3
-	sub	%g3, TRAP_PER_CPU_FAULT_INFO, %g3
+	/* Get &trap_block[smp_processor_id()] into %g4.  */
+	ldxa	[%g0] ASI_SCRATCHPAD, %g4
+	sub	%g4, TRAP_PER_CPU_FAULT_INFO, %g4
 
 	/* Get DEV mondo queue base phys address into %g5.  */
-	ldx	[%g3 + TRAP_PER_CPU_DEV_MONDO_PA], %g5
+	ldx	[%g4 + TRAP_PER_CPU_DEV_MONDO_PA], %g5
 
 	/* Load IVEC into %g3.  */
 	ldxa	[%g5 + %g2] ASI_PHYS_USE_EC, %g3
@@ -90,8 +89,7 @@ sun4v_dev_mondo:
 	 */
 
 	/* Update queue head pointer, this frees up some registers.  */
-	sethi	%hi(8192 - 1), %g4
-	or	%g4, %lo(8192 - 1), %g4
+	lduw	[%g4 + TRAP_PER_CPU_DEV_MONDO_QMASK], %g4
 	and	%g2, %g4, %g2
 
 	mov	INTRQ_DEVICE_MONDO_HEAD, %g4
@@ -143,6 +141,8 @@ sun4v_res_mondo:
 	brnz,pn	%g1, sun4v_res_mondo_queue_full
 	 nop
 
+	lduw	[%g3 + TRAP_PER_CPU_RESUM_QMASK], %g4
+
 	/* Remember this entry's offset in %g1.  */
 	mov	%g2, %g1
 
@@ -173,8 +173,6 @@ sun4v_res_mondo:
 	add	%g2, 0x08, %g2
 
 	/* Update queue head pointer.  */
-	sethi	%hi(8192 - 1), %g4
-	or	%g4, %lo(8192 - 1), %g4
 	and	%g2, %g4, %g2
 
 	mov	INTRQ_RESUM_MONDO_HEAD, %g4
@@ -254,6 +252,8 @@ sun4v_nonres_mondo:
 	brnz,pn	%g1, sun4v_nonres_mondo_queue_full
 	 nop
 
+	lduw	[%g3 + TRAP_PER_CPU_NONRESUM_QMASK], %g4
+
 	/* Remember this entry's offset in %g1.  */
 	mov	%g2, %g1
 
@@ -284,8 +284,6 @@ sun4v_nonres_mondo:
 	add	%g2, 0x08, %g2
 
 	/* Update queue head pointer.  */
-	sethi	%hi(8192 - 1), %g4
-	or	%g4, %lo(8192 - 1), %g4
 	and	%g2, %g4, %g2
 
 	mov	INTRQ_NONRESUM_MONDO_HEAD, %g4
diff --git a/arch/sparc64/kernel/time.c b/arch/sparc64/kernel/time.c
index 2d63d768996..0f62ea82953 100644
--- a/arch/sparc64/kernel/time.c
+++ b/arch/sparc64/kernel/time.c
@@ -862,7 +862,6 @@ fs_initcall(clock_init);
 static unsigned long sparc64_init_timers(void)
 {
 	struct device_node *dp;
-	struct property *prop;
 	unsigned long clock;
 #ifdef CONFIG_SMP
 	extern void smp_tick_init(void);
@@ -879,17 +878,15 @@ static unsigned long sparc64_init_timers(void)
 		if (manuf == 0x17 && impl == 0x13) {
 			/* Hummingbird, aka Ultra-IIe */
 			tick_ops = &hbtick_operations;
-			prop = of_find_property(dp, "stick-frequency", NULL);
+			clock = of_getintprop_default(dp, "stick-frequency", 0);
 		} else {
 			tick_ops = &tick_operations;
-			cpu_find_by_instance(0, &dp, NULL);
-			prop = of_find_property(dp, "clock-frequency", NULL);
+			clock = local_cpu_data().clock_tick;
 		}
 	} else {
 		tick_ops = &stick_operations;
-		prop = of_find_property(dp, "stick-frequency", NULL);
+		clock = of_getintprop_default(dp, "stick-frequency", 0);
 	}
-	clock = *(unsigned int *) prop->value;
 
 #ifdef CONFIG_SMP
 	smp_tick_init();
diff --git a/arch/sparc64/kernel/traps.c b/arch/sparc64/kernel/traps.c
index d0fde36395b..00a9e3286c8 100644
--- a/arch/sparc64/kernel/traps.c
+++ b/arch/sparc64/kernel/traps.c
@@ -795,8 +795,7 @@ extern unsigned int cheetah_deferred_trap_vector[], cheetah_deferred_trap_vector
 void __init cheetah_ecache_flush_init(void)
 {
 	unsigned long largest_size, smallest_linesize, order, ver;
-	struct device_node *dp;
-	int i, instance, sz;
+	int i, sz;
 
 	/* Scan all cpu device tree nodes, note two values:
 	 * 1) largest E-cache size
@@ -805,18 +804,20 @@ void __init cheetah_ecache_flush_init(void)
 	largest_size = 0UL;
 	smallest_linesize = ~0UL;
 
-	instance = 0;
-	while (!cpu_find_by_instance(instance, &dp, NULL)) {
+	for (i = 0; i < NR_CPUS; i++) {
 		unsigned long val;
 
-		val = of_getintprop_default(dp, "ecache-size",
-					    (2 * 1024 * 1024));
+		val = cpu_data(i).ecache_size;
+		if (!val)
+			continue;
+
 		if (val > largest_size)
 			largest_size = val;
-		val = of_getintprop_default(dp, "ecache-line-size", 64);
+
+		val = cpu_data(i).ecache_line_size;
 		if (val < smallest_linesize)
 			smallest_linesize = val;
-		instance++;
+
 	}
 
 	if (largest_size == 0UL || smallest_linesize == ~0UL) {
@@ -2564,7 +2565,15 @@ void __init trap_init(void)
 	    (TRAP_PER_CPU_TSB_HUGE_TEMP !=
 	     offsetof(struct trap_per_cpu, tsb_huge_temp)) ||
 	    (TRAP_PER_CPU_IRQ_WORKLIST !=
-	     offsetof(struct trap_per_cpu, irq_worklist)))
+	     offsetof(struct trap_per_cpu, irq_worklist)) ||
+	    (TRAP_PER_CPU_CPU_MONDO_QMASK !=
+	     offsetof(struct trap_per_cpu, cpu_mondo_qmask)) ||
+	    (TRAP_PER_CPU_DEV_MONDO_QMASK !=
+	     offsetof(struct trap_per_cpu, dev_mondo_qmask)) ||
+	    (TRAP_PER_CPU_RESUM_QMASK !=
+	     offsetof(struct trap_per_cpu, resum_qmask)) ||
+	    (TRAP_PER_CPU_NONRESUM_QMASK !=
+	     offsetof(struct trap_per_cpu, nonresum_qmask)))
 		trap_per_cpu_offsets_are_bolixed_dave();
 
 	if ((TSB_CONFIG_TSB !=
diff --git a/arch/sparc64/mm/init.c b/arch/sparc64/mm/init.c
index 0c9995c3b8e..977698269d3 100644
--- a/arch/sparc64/mm/init.c
+++ b/arch/sparc64/mm/init.c
@@ -23,6 +23,7 @@
 #include <linux/kprobes.h>
 #include <linux/cache.h>
 #include <linux/sort.h>
+#include <linux/percpu.h>
 
 #include <asm/head.h>
 #include <asm/system.h>
@@ -44,8 +45,7 @@
 #include <asm/hypervisor.h>
 #include <asm/prom.h>
 #include <asm/sstate.h>
-
-extern void device_scan(void);
+#include <asm/mdesc.h>
 
 #define MAX_PHYS_ADDRESS	(1UL << 42UL)
 #define KPTE_BITMAP_CHUNK_SZ	(256UL * 1024UL * 1024UL)
@@ -1335,6 +1335,9 @@ void __cpuinit sun4v_ktsb_register(void)
 extern void cheetah_ecache_flush_init(void);
 extern void sun4v_patch_tlb_handlers(void);
 
+extern void cpu_probe(void);
+extern void central_probe(void);
+
 static unsigned long last_valid_pfn;
 pgd_t swapper_pg_dir[2048];
 
@@ -1419,8 +1422,13 @@ void __init paging_init(void)
 
 	kernel_physical_mapping_init();
 
+	real_setup_per_cpu_areas();
+
 	prom_build_devicetree();
 
+	if (tlb_type == hypervisor)
+		sun4v_mdesc_init();
+
 	{
 		unsigned long zones_size[MAX_NR_ZONES];
 		unsigned long zholes_size[MAX_NR_ZONES];
@@ -1437,7 +1445,10 @@ void __init paging_init(void)
 				    zholes_size);
 	}
 
-	device_scan();
+	prom_printf("Booting Linux...\n");
+
+	central_probe();
+	cpu_probe();
 }
 
 static void __init taint_real_pages(void)
diff --git a/include/asm-sparc64/cpudata.h b/include/asm-sparc64/cpudata.h
index e89922d6718..f321b1d2122 100644
--- a/include/asm-sparc64/cpudata.h
+++ b/include/asm-sparc64/cpudata.h
@@ -17,11 +17,11 @@
 typedef struct {
 	/* Dcache line 1 */
 	unsigned int	__softirq_pending; /* must be 1st, see rtrap.S */
-	unsigned int	__pad0_1;
-	unsigned int	__pad0_2;
-	unsigned int	__pad1;
+	unsigned int	__pad0;
 	unsigned long	clock_tick;	/* %tick's per second */
 	unsigned long	udelay_val;
+	unsigned int	__pad1;
+	unsigned int	__pad2;
 
 	/* Dcache line 2, rarely used */
 	unsigned int	dcache_size;
@@ -30,8 +30,8 @@ typedef struct {
 	unsigned int	icache_line_size;
 	unsigned int	ecache_size;
 	unsigned int	ecache_line_size;
+	int		core_id;
 	unsigned int	__pad3;
-	unsigned int	__pad4;
 } cpuinfo_sparc;
 
 DECLARE_PER_CPU(cpuinfo_sparc, __cpu_data);
@@ -76,12 +76,18 @@ struct trap_per_cpu {
 
 /* Dcache line 8: IRQ work list, and keep trap_block a power-of-2 in size.  */
 	unsigned int		irq_worklist;
-	unsigned int		__pad1;
-	unsigned long		__pad2[3];
+	unsigned int		cpu_mondo_qmask;
+	unsigned int		dev_mondo_qmask;
+	unsigned int		resum_qmask;
+	unsigned int		nonresum_qmask;
+	unsigned int		__pad2[3];
 } __attribute__((aligned(64)));
 extern struct trap_per_cpu trap_block[NR_CPUS];
 extern void init_cur_cpu_trap(struct thread_info *);
 extern void setup_tba(void);
+extern int ncpus_probed;
+
+extern unsigned long real_hard_smp_processor_id(void);
 
 struct cpuid_patch_entry {
 	unsigned int	addr;
@@ -122,6 +128,10 @@ extern struct sun4v_2insn_patch_entry __sun4v_2insn_patch,
 #define TRAP_PER_CPU_TSB_HUGE		0xd0
 #define TRAP_PER_CPU_TSB_HUGE_TEMP	0xd8
 #define TRAP_PER_CPU_IRQ_WORKLIST	0xe0
+#define TRAP_PER_CPU_CPU_MONDO_QMASK	0xe4
+#define TRAP_PER_CPU_DEV_MONDO_QMASK	0xe8
+#define TRAP_PER_CPU_RESUM_QMASK	0xec
+#define TRAP_PER_CPU_NONRESUM_QMASK	0xf0
 
 #define TRAP_BLOCK_SZ_SHIFT		8
 
diff --git a/include/asm-sparc64/hypervisor.h b/include/asm-sparc64/hypervisor.h
index 17b233b64af..0a241c82fc7 100644
--- a/include/asm-sparc64/hypervisor.h
+++ b/include/asm-sparc64/hypervisor.h
@@ -120,6 +120,11 @@
  */
 #define HV_FAST_MACH_DESC		0x01
 
+#ifndef __ASSEMBLY__
+extern unsigned long sun4v_mach_desc(unsigned long buffer_pa, unsigned long buf_len,
+				     unsigned long *real_buf_len);
+#endif
+
 /* mach_exit()
  * TRAP:	HV_FAST_TRAP
  * FUNCTION:	HV_FAST_MACH_SIR
diff --git a/include/asm-sparc64/mdesc.h b/include/asm-sparc64/mdesc.h
new file mode 100644
index 00000000000..124eb8ca237
--- /dev/null
+++ b/include/asm-sparc64/mdesc.h
@@ -0,0 +1,39 @@
+#ifndef _SPARC64_MDESC_H
+#define _SPARC64_MDESC_H
+
+#include <linux/types.h>
+#include <asm/prom.h>
+
+struct mdesc_node;
+struct mdesc_arc {
+	const char		*name;
+	struct mdesc_node	*arc;
+};
+
+struct mdesc_node {
+	const char		*name;
+	u64			node;
+	unsigned int		unique_id;
+	unsigned int		num_arcs;
+	struct property		*properties;
+	struct mdesc_node	*hash_next;
+	struct mdesc_node	*allnodes_next;
+	struct mdesc_arc	arcs[0];
+};
+
+extern struct mdesc_node *md_find_node_by_name(struct mdesc_node *from,
+					       const char *name);
+#define md_for_each_node_by_name(__mn, __name) \
+	for (__mn = md_find_node_by_name(NULL, __name); __mn; \
+	     __mn = md_find_node_by_name(__mn, __name))
+
+extern struct property *md_find_property(const struct mdesc_node *mp,
+					 const char *name,
+					 int *lenp);
+extern const void *md_get_property(const struct mdesc_node *mp,
+				   const char *name,
+				   int *lenp);
+
+extern void sun4v_mdesc_init(void);
+
+#endif
diff --git a/include/asm-sparc64/oplib.h b/include/asm-sparc64/oplib.h
index 07275e2366a..992f9f7a476 100644
--- a/include/asm-sparc64/oplib.h
+++ b/include/asm-sparc64/oplib.h
@@ -319,11 +319,6 @@ extern int prom_inst2pkg(int);
 extern int prom_service_exists(const char *service_name);
 extern void prom_sun4v_guest_soft_state(void);
 
-/* CPU probing helpers.  */
-struct device_node;
-int cpu_find_by_instance(int instance, struct device_node **dev_node, int *mid);
-int cpu_find_by_mid(int mid, struct device_node **prom_node);
-
 /* Client interface level routines. */
 extern void prom_set_trap_table(unsigned long tba);
 extern void prom_set_trap_table_sun4v(unsigned long tba, unsigned long mmfsa);
diff --git a/include/asm-sparc64/percpu.h b/include/asm-sparc64/percpu.h
index ced8cbde046..88db872ce2f 100644
--- a/include/asm-sparc64/percpu.h
+++ b/include/asm-sparc64/percpu.h
@@ -5,7 +5,8 @@
 
 #ifdef CONFIG_SMP
 
-extern void setup_per_cpu_areas(void);
+#define setup_per_cpu_areas()			do { } while (0)
+extern void real_setup_per_cpu_areas(void);
 
 extern unsigned long __per_cpu_base;
 extern unsigned long __per_cpu_shift;
@@ -34,6 +35,7 @@ do {								\
 } while (0)
 #else /* ! SMP */
 
+#define real_setup_per_cpu_areas()		do { } while (0)
 #define DEFINE_PER_CPU(type, name) \
     __typeof__(type) per_cpu__##name
 
diff --git a/include/asm-sparc64/prom.h b/include/asm-sparc64/prom.h
index ddad5f99ac7..b4df3042add 100644
--- a/include/asm-sparc64/prom.h
+++ b/include/asm-sparc64/prom.h
@@ -90,6 +90,7 @@ extern struct device_node *of_find_compatible_node(struct device_node *from,
 	const char *type, const char *compat);
 extern struct device_node *of_find_node_by_path(const char *path);
 extern struct device_node *of_find_node_by_phandle(phandle handle);
+extern struct device_node *of_find_node_by_cpuid(int cpuid);
 extern struct device_node *of_get_parent(const struct device_node *node);
 extern struct device_node *of_get_next_child(const struct device_node *node,
 					     struct device_node *prev);
diff --git a/include/asm-sparc64/smp.h b/include/asm-sparc64/smp.h
index 869d16fb907..f76e1492add 100644
--- a/include/asm-sparc64/smp.h
+++ b/include/asm-sparc64/smp.h
@@ -41,7 +41,7 @@ extern cpumask_t cpu_sibling_map[NR_CPUS];
 extern int hard_smp_processor_id(void);
 #define raw_smp_processor_id() (current_thread_info()->cpu)
 
-extern void smp_setup_cpu_possible_map(void);
+extern void smp_fill_in_sib_core_maps(void);
 extern unsigned char boot_cpu_id;
 
 #endif /* !(__ASSEMBLY__) */
@@ -49,7 +49,7 @@ extern unsigned char boot_cpu_id;
 #else
 
 #define hard_smp_processor_id()		0
-#define smp_setup_cpu_possible_map() do { } while (0)
+#define smp_fill_in_sib_core_maps() do { } while (0)
 #define boot_cpu_id	(0)
 
 #endif /* !(CONFIG_SMP) */
diff --git a/include/asm-sparc64/topology.h b/include/asm-sparc64/topology.h
index 98a6c613589..e0d450d600e 100644
--- a/include/asm-sparc64/topology.h
+++ b/include/asm-sparc64/topology.h
@@ -6,4 +6,7 @@
 
 #include <asm-generic/topology.h>
 
+#define topology_core_id(cpu)			(cpu_data(cpu).core_id)
+#define topology_thread_siblings(cpu)		(cpu_sibling_map[cpu])
+
 #endif /* _ASM_SPARC64_TOPOLOGY_H */