Merge http://lia64.bkbits.net/linux-ia64-release-2.6.9

into ppc970.osdl.org:/home/torvalds/v2.6/linux

9 files changed, 905 insertions, 9 deletions

diff --git a/arch/ia64/Kconfig b/arch/ia64/Kconfig
index aa742c601da2..2478ea8892e5 100644
--- a/arch/ia64/Kconfig
+++ b/arch/ia64/Kconfig
@@ -76,6 +76,12 @@ config IA64_HP_ZX1
 
 config IA64_SGI_SN2
 	bool "SGI-SN2"
+	help
+	  Selecting this option will optimize the kernel for use on sn2 based
+	  systems, but the resulting kernel binary will not run on other
+	  types of ia64 systems.  If you have an SGI Altix system, it's safe
+	  to select this option.  If in doubt, select ia64 generic support
+	  instead.
 
 config IA64_HP_SIM
 	bool "Ski-simulator"
diff --git a/arch/ia64/kernel/minstate.h b/arch/ia64/kernel/minstate.h
index 540301f6efaa..b1460dcc3e64 100644
--- a/arch/ia64/kernel/minstate.h
+++ b/arch/ia64/kernel/minstate.h
@@ -54,8 +54,7 @@
 (pUStk)	mov ar.rsc=0x3;		/* set eager mode, pl 0, little-endian, loadrs=0 */		\
 
 #define MINSTATE_END_SAVE_MIN_PHYS								\
-	or r12=r12,r14;		/* make sp a kernel virtual address */				\
-	or r13=r13,r14;		/* make `current' a kernel virtual address */			\
+	dep r12=-1,r12,61,3;		/* make sp a kernel virtual address */			\
 	;;
 
 #ifdef MINSTATE_VIRT
@@ -65,7 +64,7 @@
 #endif
 
 #ifdef MINSTATE_PHYS
-# define MINSTATE_GET_CURRENT(reg)	mov reg=IA64_KR(CURRENT);; dep reg=0,reg,61,3
+# define MINSTATE_GET_CURRENT(reg)	mov reg=IA64_KR(CURRENT);; tpa reg=reg
 # define MINSTATE_START_SAVE_MIN	MINSTATE_START_SAVE_MIN_PHYS
 # define MINSTATE_END_SAVE_MIN		MINSTATE_END_SAVE_MIN_PHYS
 #endif
@@ -172,7 +171,6 @@
 	;;											\
 .mem.offset 0,0; st8.spill [r16]=r15,16;							\
 .mem.offset 8,0; st8.spill [r17]=r14,16;							\
-	dep r14=-1,r0,61,3;									\
 	;;											\
 .mem.offset 0,0; st8.spill [r16]=r2,16;								\
 .mem.offset 8,0; st8.spill [r17]=r3,16;								\
diff --git a/arch/ia64/sn/kernel/sn2/Makefile b/arch/ia64/sn/kernel/sn2/Makefile
index 0143b8182778..170bde4549da 100644
--- a/arch/ia64/sn/kernel/sn2/Makefile
+++ b/arch/ia64/sn/kernel/sn2/Makefile
@@ -10,4 +10,4 @@
 #
 
 obj-y += cache.o io.o ptc_deadlock.o sn2_smp.o sn_proc_fs.o \
-	 prominfo_proc.o timer.o timer_interrupt.o
+	 prominfo_proc.o timer.o timer_interrupt.o sn_hwperf.o
diff --git a/arch/ia64/sn/kernel/sn2/prominfo_proc.c b/arch/ia64/sn/kernel/sn2/prominfo_proc.c
index d9228584d8cc..e8dc684a14c0 100644
--- a/arch/ia64/sn/kernel/sn2/prominfo_proc.c
+++ b/arch/ia64/sn/kernel/sn2/prominfo_proc.c
@@ -222,7 +222,7 @@ dump_version(char *page, unsigned long *fit)
 {
 	int nentries;
 	int fentry;
-	unsigned long qw;
+	unsigned long qw = 0;
 	int len;
 	nasid_t nasid = NASID_GET(fit);
 
diff --git a/arch/ia64/sn/kernel/sn2/sn_hwperf.c b/arch/ia64/sn/kernel/sn2/sn_hwperf.c
new file mode 100644
index 000000000000..8a810d339e2f
--- /dev/null
+++ b/arch/ia64/sn/kernel/sn2/sn_hwperf.c
@@ -0,0 +1,652 @@
+/* 
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 2004 Silicon Graphics, Inc. All rights reserved.
+ *
+ * SGI Altix topology and hardware performance monitoring API.
+ * Mark Goodwin <markgw@sgi.com>. 
+ *
+ * Creates /proc/sgi_sn/sn_topology (read-only) to export
+ * info about Altix nodes, routers, CPUs and NumaLink
+ * interconnection/topology.
+ *
+ * Also creates a dynamic misc device named "sn_hwperf"
+ * that supports an ioctl interface to call down into SAL
+ * to discover hw objects, topology and to read/write
+ * memory mapped registers, e.g. for performance monitoring.
+ * The "sn_hwperf" device is registered only after the procfs
+ * file is first opened, i.e. only if/when it's needed. 
+ *
+ * This API is used by SGI Performance Co-Pilot and other
+ * tools, see http://oss.sgi.com/projects/pcp
+ */
+
+#include <linux/fs.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/seq_file.h>
+#include <linux/miscdevice.h>
+#include <linux/cpumask.h>
+#include <linux/smp_lock.h>
+#include <asm/processor.h>
+#include <asm/topology.h>
+#include <asm/smp.h>
+#include <asm/semaphore.h>
+#include <asm/segment.h>
+#include <asm/uaccess.h>
+#include <asm-ia64/sal.h>
+#include <asm-ia64/sn/sn_sal.h>
+#include <asm-ia64/sn/sn2/sn_hwperf.h>
+
+static void *sn_hwperf_salheap = NULL;
+static int sn_hwperf_obj_cnt = 0;
+static nasid_t sn_hwperf_master_nasid = INVALID_NASID;
+static int sn_hwperf_init(void);
+static DECLARE_MUTEX(sn_hwperf_init_mutex);
+
+static int sn_hwperf_enum_objects(int *nobj, struct sn_hwperf_object_info **ret)
+{
+	int e;
+	u64 sz;
+	struct sn_hwperf_object_info *objbuf = NULL;
+
+	if ((e = sn_hwperf_init()) < 0) {
+		printk("sn_hwperf_init failed: err %d\n", e);
+		goto out;
+	}
+
+	sz = sn_hwperf_obj_cnt * sizeof(struct sn_hwperf_object_info);
+	if ((objbuf = (struct sn_hwperf_object_info *) vmalloc(sz)) == NULL) {
+		printk("sn_hwperf_enum_objects: vmalloc(%d) failed\n", (int)sz);
+		e = -ENOMEM;
+		goto out;
+	}
+
+	e = ia64_sn_hwperf_op(sn_hwperf_master_nasid, SN_HWPERF_ENUM_OBJECTS,
+		0, sz, (u64) objbuf, 0, 0, NULL);
+	if (e != SN_HWPERF_OP_OK) {
+		e = -EINVAL;
+		vfree(objbuf);
+	}
+
+out:
+	*nobj = sn_hwperf_obj_cnt;
+	*ret = objbuf;
+	return e;
+}
+
+static int sn_hwperf_geoid_to_cnode(char *location)
+{
+	int cnode;
+	int mod, slot, slab;
+	int cmod, cslot, cslab;
+
+	if (sscanf(location, "%03dc%02d#%d", &mod, &slot, &slab) != 3)
+		return -1;
+	for (cnode = 0; cnode < numnodes; cnode++) {
+		/* XXX: need a better way than this ... */
+		if (sscanf(NODEPDA(cnode)->hwg_node_name,
+		   "hw/module/%03dc%02d/slab/%d", &cmod, &cslot, &cslab) == 3) {
+			if (mod == cmod && slot == cslot && slab == cslab)
+				break;
+		}
+	}
+
+	return cnode < numnodes ? cnode : -1;
+}
+
+static int sn_hwperf_obj_to_cnode(struct sn_hwperf_object_info * obj)
+{
+	if (!obj->sn_hwp_this_part)
+		return -1;
+	return sn_hwperf_geoid_to_cnode(obj->location);
+}
+
+static int sn_hwperf_generic_ordinal(struct sn_hwperf_object_info *obj,
+				struct sn_hwperf_object_info *objs)
+{
+	int ordinal;
+	struct sn_hwperf_object_info *p;
+
+	for (ordinal=0, p=objs; p != obj; p++) {
+		if (SN_HWPERF_FOREIGN(p))
+			continue;
+		if (p->location[3] == obj->location[3])
+			ordinal++;
+	}
+
+	return ordinal;
+}
+
+#ifndef MODULE_IOBRICK 
+/* this will be available when ioif TIO support is added */
+#define MODULE_IOBRICK (MODULE_OPUSBRICK+1)
+#endif
+
+static const char *sn_hwperf_get_brickname(struct sn_hwperf_object_info *obj,
+				struct sn_hwperf_object_info *objs, int *ordinal)
+{
+	int i;
+	const char *objtype = NULL;
+
+	for (i=0; i < MAX_BRICK_TYPES; i++) {
+		if (brick_types[i] != obj->location[3])
+			continue;
+		switch (i) {
+		case MODULE_CBRICK:
+		    objtype = "node";
+		    *ordinal = sn_hwperf_obj_to_cnode(obj); /* cnodeid */
+		    break;
+
+		case MODULE_RBRICK:
+		    objtype = "router";
+		    *ordinal = sn_hwperf_generic_ordinal(obj, objs);
+		    break;
+
+		case MODULE_IOBRICK:
+		    objtype = "ionode";
+		    *ordinal = sn_hwperf_generic_ordinal(obj, objs);
+		    break;
+		}
+		break;
+	}
+
+	if (i == MAX_BRICK_TYPES) {
+		objtype = "other";
+		*ordinal = sn_hwperf_generic_ordinal(obj, objs);
+	}
+
+	return objtype;
+}
+
+static int sn_topology_show(struct seq_file *s, void *d)
+{
+	int sz;
+	int pt;
+	int e;
+	int i;
+	int j;
+	const char *brickname;
+	int ordinal;
+	cpumask_t cpumask;
+	char slice;
+	struct cpuinfo_ia64 *c;
+	struct sn_hwperf_port_info *ptdata;
+	struct sn_hwperf_object_info *p;
+	struct sn_hwperf_object_info *obj = d;	/* this object */
+	struct sn_hwperf_object_info *objs = s->private; /* all objects */
+
+	if (obj == objs) {
+		seq_printf(s, "# sn_topology version 1\n");
+		seq_printf(s, "# objtype ordinal location partition"
+			" [attribute value [, ...]]\n");
+	}
+
+	if (SN_HWPERF_FOREIGN(obj)) {
+		/* private in another partition: not interesting */
+		return 0;
+	}
+
+	for (i = 0; obj->name[i]; i++) {
+		if (obj->name[i] == ' ')
+			obj->name[i] = '_';
+	}
+
+	brickname = sn_hwperf_get_brickname(obj, objs, &ordinal);
+	seq_printf(s, "%s %d %s %s asic %s", brickname, ordinal, obj->location,
+		obj->sn_hwp_this_part ? "local" : "shared", obj->name);
+
+	if (obj->location[3] != 'c')
+		seq_putc(s, '\n');
+	else {
+		seq_printf(s, ", nasid 0x%x", cnodeid_to_nasid(ordinal));
+		for (i=0; i < numnodes; i++) {
+			seq_printf(s, i ? ":%d" : ", dist %d",
+				node_distance(ordinal, i));
+		}
+		seq_putc(s, '\n');
+
+		/*
+		 * CPUs on this node
+		 */
+		cpumask = node_to_cpumask(ordinal);
+		for_each_online_cpu(i) {
+			if (cpu_isset(i, cpumask)) {
+				slice = 'a' + cpuid_to_slice(i);
+				c = cpu_data(i);
+				seq_printf(s, "cpu %d %s%c local"
+					" freq %luMHz, arch ia64",
+					i, obj->location, slice,
+					c->proc_freq / 1000000);
+				for_each_online_cpu(j) {
+					seq_printf(s, j ? ":%d" : ", dist %d",
+						node_distance(
+						    cpuid_to_cnodeid(i),
+						    cpuid_to_cnodeid(j)));
+				}
+				seq_putc(s, '\n');
+			}
+		}
+	}
+
+	if (obj->ports) {
+		/*
+		 * numalink ports
+		 */
+		sz = obj->ports * sizeof(struct sn_hwperf_port_info);
+		if ((ptdata = vmalloc(sz)) == NULL)
+			return -ENOMEM;
+		e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
+				      SN_HWPERF_ENUM_PORTS, obj->id, sz,
+				      (u64) ptdata, 0, 0, NULL);
+		if (e != SN_HWPERF_OP_OK)
+			return -EINVAL;
+		for (ordinal=0, p=objs; p != obj; p++) {
+			if (!SN_HWPERF_FOREIGN(p))
+				ordinal += p->ports;
+		}
+		for (pt = 0; pt < obj->ports; pt++) {
+			for (p = objs, i = 0; i < sn_hwperf_obj_cnt; i++, p++) {
+				if (ptdata[pt].conn_id == p->id) {
+					break;
+				}
+			}
+			if (i >= sn_hwperf_obj_cnt)
+				continue;
+			seq_printf(s, "numalink %d %s-%d",
+			    ordinal+pt, obj->location, ptdata[pt].port);
+
+			if (obj->sn_hwp_this_part && p->sn_hwp_this_part)
+				/* both ends local to this partition */
+				seq_puts(s, " local");
+			else if (!obj->sn_hwp_this_part && !p->sn_hwp_this_part)
+				/* both ends of the link in foreign partiton */
+				seq_puts(s, " foreign");
+			else
+				/* link straddles a partition */
+				seq_puts(s, " shared");
+
+			/*
+			 * Unlikely, but strictly should query the LLP config
+			 * registers because an NL4R can be configured to run
+			 * NL3 protocol, even when not talking to an NL3 router.
+			 * Ditto for node-node.
+			 */
+			seq_printf(s, " endpoint %s-%d, protocol %s\n",
+				p->location, ptdata[pt].conn_port,
+				strcmp(obj->name, "NL3Router") == 0 ||
+				strcmp(p->name, "NL3Router") == 0 ?
+				"LLP3" : "LLP4");
+		}
+		vfree(ptdata);
+	}
+
+	return 0;
+}
+
+static void *sn_topology_start(struct seq_file *s, loff_t * pos)
+{
+	struct sn_hwperf_object_info *objs = s->private;
+
+	if (*pos < sn_hwperf_obj_cnt)
+		return (void *)(objs + *pos);
+
+	return NULL;
+}
+
+static void *sn_topology_next(struct seq_file *s, void *v, loff_t * pos)
+{
+	++*pos;
+	return sn_topology_start(s, pos);
+}
+
+static void sn_topology_stop(struct seq_file *m, void *v)
+{
+	return;
+}
+
+/*
+ * /proc/sgi_sn/sn_topology, read-only using seq_file
+ */
+static struct seq_operations sn_topology_seq_ops = {
+	.start = sn_topology_start,
+	.next = sn_topology_next,
+	.stop = sn_topology_stop,
+	.show = sn_topology_show
+};
+
+struct sn_hwperf_op_info {
+	u64 op;
+	struct sn_hwperf_ioctl_args *a;
+	void *p;
+	int *v0;
+	int ret;
+};
+
+static void sn_hwperf_call_sal(void *info)
+{
+	struct sn_hwperf_op_info *op_info = info;
+	int r;
+
+	r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op_info->op,
+		      op_info->a->arg, op_info->a->sz,
+		      (u64) op_info->p, 0, 0, op_info->v0);
+	op_info->ret = r;
+}
+
+static int sn_hwperf_op_cpu(struct sn_hwperf_op_info *op_info)
+{
+	u32 cpu;
+	u32 use_ipi;
+	int r = 0;
+	cpumask_t save_allowed;
+	
+	cpu = (op_info->a->arg & SN_HWPERF_ARG_CPU_MASK) >> 32;
+	use_ipi = op_info->a->arg & SN_HWPERF_ARG_USE_IPI_MASK;
+	op_info->a->arg &= SN_HWPERF_ARG_OBJID_MASK;
+
+	if (cpu != SN_HWPERF_ARG_ANY_CPU) {
+		if (cpu >= num_online_cpus() || !cpu_online(cpu)) {
+			r = -EINVAL;
+			goto out;
+		}
+	}
+
+	if (cpu == SN_HWPERF_ARG_ANY_CPU || cpu == get_cpu()) {
+		/* don't care, or already on correct cpu */
+		sn_hwperf_call_sal(op_info);
+	}
+	else {
+		if (use_ipi) {
+			/* use an interprocessor interrupt to call SAL */
+			smp_call_function_single(cpu, sn_hwperf_call_sal,
+				op_info, 1, 1);
+		}
+		else {
+			/* migrate the task before calling SAL */ 
+			save_allowed = current->cpus_allowed;
+			set_cpus_allowed(current, cpumask_of_cpu(cpu));
+			sn_hwperf_call_sal(op_info);
+			set_cpus_allowed(current, save_allowed);
+		}
+	}
+	r = op_info->ret;
+
+out:
+	return r;
+}
+
+/*
+ * ioctl for "sn_hwperf" misc device
+ */
+static int
+sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg)
+{
+	struct sn_hwperf_ioctl_args a;
+	struct cpuinfo_ia64 *cdata;
+	struct sn_hwperf_object_info *objs;
+	struct sn_hwperf_object_info *cpuobj;
+	struct sn_hwperf_op_info op_info;
+	void *p = NULL;
+	int nobj;
+	char slice;
+	int node;
+	int r;
+	int v0;
+	int i;
+	int j;
+
+	unlock_kernel();
+
+	/* only user requests are allowed here */
+	if ((op & SN_HWPERF_OP_MASK) < 10) {
+		r = -EINVAL;
+		goto error;
+	}
+	r = copy_from_user(&a, (const void *)arg,
+		sizeof(struct sn_hwperf_ioctl_args));
+	if (r != 0) {
+		r = -EFAULT;
+		goto error;
+	}
+
+	/*
+	 * Allocate memory to hold a kernel copy of the user buffer. The
+	 * buffer contents are either copied in or out (or both) of user
+	 * space depending on the flags encoded in the requested operation.
+	 */
+	if (a.ptr) {
+		p = vmalloc(a.sz);
+		if (!p) {
+			r = -ENOMEM;
+			goto error;
+		}
+	}
+
+	if (op & SN_HWPERF_OP_MEM_COPYIN) {
+		r = copy_from_user(p, (const void *)a.ptr, a.sz);
+		if (r != 0) {
+			r = -EFAULT;
+			goto error;
+		}
+	}
+
+	switch (op) {
+	case SN_HWPERF_GET_CPU_INFO:
+		if (a.sz == sizeof(u64)) {
+			/* special case to get size needed */
+			*(u64 *) p = (u64) num_online_cpus() *
+				sizeof(struct sn_hwperf_object_info);
+		} else
+		if (a.sz < num_online_cpus() * sizeof(struct sn_hwperf_object_info)) {
+			r = -ENOMEM;
+			goto error;
+		} else
+		if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
+			memset(p, 0, a.sz);
+			for (i = 0; i < nobj; i++) {
+				node = sn_hwperf_obj_to_cnode(objs + i);
+				for_each_online_cpu(j) {
+					if (node != cpu_to_node(j))
+						continue;
+					cpuobj = (struct sn_hwperf_object_info *) p + j;
+					slice = 'a' + cpuid_to_slice(j);
+					cdata = cpu_data(j);
+					cpuobj->id = j;
+					snprintf(cpuobj->name,
+						 sizeof(cpuobj->name),
+						 "CPU %luMHz %s",
+						 cdata->proc_freq / 1000000,
+						 cdata->vendor);
+					snprintf(cpuobj->location,
+						 sizeof(cpuobj->location),
+						 "%s%c", objs[i].location,
+						 slice);
+				}
+			}
+
+			vfree(objs);
+		}
+		break;
+
+	case SN_HWPERF_GET_NODE_NASID:
+		if (a.sz != sizeof(u64) ||
+		   (node = a.arg) < 0 || node >= numnodes) {
+			r = -EINVAL;
+			goto error;
+		}
+		*(u64 *)p = (u64)cnodeid_to_nasid(node);
+		break;
+
+	case SN_HWPERF_GET_OBJ_NODE:
+		if (a.sz != sizeof(u64) || a.arg < 0) {
+			r = -EINVAL;
+			goto error;
+		}
+		if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
+			if (a.arg >= nobj) {
+				r = -EINVAL;
+				vfree(objs);
+				goto error;
+			}
+			if (objs[(i = a.arg)].id != a.arg) {
+				for (i = 0; i < nobj; i++) {
+					if (objs[i].id == a.arg)
+						break;
+				}
+			}
+			if (i == nobj) {
+				r = -EINVAL;
+				vfree(objs);
+				goto error;
+			}
+			*(u64 *)p = (u64)sn_hwperf_obj_to_cnode(objs + i);
+			vfree(objs);
+		}
+		break;
+
+	case SN_HWPERF_GET_MMRS:
+	case SN_HWPERF_SET_MMRS:
+	case SN_HWPERF_OBJECT_DISTANCE:
+		op_info.p = p;
+		op_info.a = &a;
+		op_info.v0 = &v0;
+		op_info.op = op;
+		r = sn_hwperf_op_cpu(&op_info);
+		break;
+
+	default:
+		/* all other ops are a direct SAL call */
+		r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op,
+			      a.arg, a.sz, (u64) p, 0, 0, &v0);
+		a.v0 = v0;
+		break;
+	}
+
+	if (op & SN_HWPERF_OP_MEM_COPYOUT) {
+		r = copy_to_user((void *)a.ptr, p, a.sz);
+		if (r != 0) {
+			r = -EFAULT;
+			goto error;
+		}
+	}
+
+error:
+	if (p)
+		vfree(p);
+
+	lock_kernel();
+	return r;
+}
+
+static struct file_operations sn_hwperf_fops = {
+	.ioctl = sn_hwperf_ioctl,
+};
+
+static struct miscdevice sn_hwperf_dev = {
+	MISC_DYNAMIC_MINOR,
+	"sn_hwperf",
+	&sn_hwperf_fops
+};
+
+static int sn_hwperf_init(void)
+{
+	u64 v;
+	int salr;
+	int e = 0;
+
+	/* single threaded, once-only initialization */
+	down(&sn_hwperf_init_mutex);
+	if (sn_hwperf_salheap) {
+		up(&sn_hwperf_init_mutex);
+		return e;
+	}
+
+	/*
+	 * The PROM code needs a fixed reference node. For convenience the
+	 * same node as the console I/O is used.
+	 */
+	sn_hwperf_master_nasid = (nasid_t) ia64_sn_get_console_nasid();
+
+	/*
+	 * Request the needed size and install the PROM scratch area.
+	 * The PROM keeps various tracking bits in this memory area.
+	 */
+	salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
+				 (u64) SN_HWPERF_GET_HEAPSIZE, 0,
+				 (u64) sizeof(u64), (u64) &v, 0, 0, NULL);
+	if (salr != SN_HWPERF_OP_OK) {
+		e = -EINVAL;
+		goto out;
+	}
+
+	if ((sn_hwperf_salheap = vmalloc(v)) == NULL) {
+		e = -ENOMEM;
+		goto out;
+	}
+	salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
+				 SN_HWPERF_INSTALL_HEAP, 0, v,
+				 (u64) sn_hwperf_salheap, 0, 0, NULL);
+	if (salr != SN_HWPERF_OP_OK) {
+		e = -EINVAL;
+		goto out;
+	}
+
+	salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
+				 SN_HWPERF_OBJECT_COUNT, 0,
+				 sizeof(u64), (u64) &v, 0, 0, NULL);
+	if (salr != SN_HWPERF_OP_OK) {
+		e = -EINVAL;
+		goto out;
+	}
+	sn_hwperf_obj_cnt = (int)v;
+
+out:
+	if (e < 0 && sn_hwperf_salheap) {
+		vfree(sn_hwperf_salheap);
+		sn_hwperf_salheap = NULL;
+		sn_hwperf_obj_cnt = 0;
+	}
+
+	if (!e) {
+		/*
+		 * Register a dynamic misc device for ioctl. Platforms
+		 * supporting hotplug will create /dev/sn_hwperf, else
+		 * user can to look up the minor number in /proc/misc.
+		 */
+		if ((e = misc_register(&sn_hwperf_dev)) != 0) {
+			printk(KERN_ERR "sn_hwperf_init: misc register "
+			       "for \"sn_hwperf\" failed, err %d\n", e);
+		}
+	}
+
+	up(&sn_hwperf_init_mutex);
+	return e;
+}
+
+int sn_topology_open(struct inode *inode, struct file *file)
+{
+	int e;
+	struct seq_file *seq;
+	struct sn_hwperf_object_info *objbuf;
+	int nobj;
+
+	if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
+		e = seq_open(file, &sn_topology_seq_ops);
+		seq = file->private_data;
+		seq->private = objbuf;
+	}
+
+	return e;
+}
+
+int sn_topology_release(struct inode *inode, struct file *file)
+{
+	struct seq_file *seq = file->private_data;
+
+	if (seq->private)
+		vfree(seq->private);
+	return seq_release(inode, file);
+}
diff --git a/arch/ia64/sn/kernel/sn2/sn_proc_fs.c b/arch/ia64/sn/kernel/sn2/sn_proc_fs.c
index aa98abf43f5e..55c1065e9ef6 100644
--- a/arch/ia64/sn/kernel/sn2/sn_proc_fs.c
+++ b/arch/ia64/sn/kernel/sn2/sn_proc_fs.c
@@ -107,6 +107,10 @@ static struct proc_dir_entry *sn_procfs_create_entry(
 	return e;
 }
 
+/* /proc/sgi_sn/sn_topology uses seq_file, see sn_hwperf.c */
+extern int sn_topology_open(struct inode *, struct file *);
+extern int sn_topology_release(struct inode *, struct file *);
+
 void register_sn_procfs(void)
 {
 	static struct proc_dir_entry *sgi_proc_dir = NULL;
@@ -132,6 +136,9 @@ void register_sn_procfs(void)
 
 	sn_procfs_create_entry("coherence_id", sgi_proc_dir, 
 		coherence_id_open, single_release);
+	
+	sn_procfs_create_entry("sn_topology", sgi_proc_dir,
+		sn_topology_open, sn_topology_release);
 }
 
 #endif /* CONFIG_PROC_FS */
diff --git a/include/asm-ia64/mca_asm.h b/include/asm-ia64/mca_asm.h
index 2aa8437dcf75..55deddd7e5bc 100644
--- a/include/asm-ia64/mca_asm.h
+++ b/include/asm-ia64/mca_asm.h
@@ -162,9 +162,6 @@
 	;;							\
 	srlz.d;							\
 	mov	r13 = ar.k6;					\
-	;;							\
-	DATA_PA_TO_VA(r13,temp1);				\
-	;;							\
 	mov	temp2 = ar.bspstore;				\
 	;;							\
 	DATA_PA_TO_VA(temp2,temp1);				\
diff --git a/include/asm-ia64/sn/sn2/sn_hwperf.h b/include/asm-ia64/sn/sn2/sn_hwperf.h
new file mode 100644
index 000000000000..2036382edd6b
--- /dev/null
+++ b/include/asm-ia64/sn/sn2/sn_hwperf.h
@@ -0,0 +1,218 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 2004 Silicon Graphics, Inc. All rights reserved.
+ *
+ * Data types used by the SN_SAL_HWPERF_OP SAL call for monitoring
+ * SGI Altix node and router hardware
+ *
+ * Mark Goodwin <markgw@sgi.com> Mon Aug 30 12:23:46 EST 2004
+ */
+
+#ifndef SN_HWPERF_H
+#define SN_HWPERF_H
+
+/*
+ * object structure. SN_HWPERF_ENUM_OBJECTS and SN_HWPERF_GET_CPU_INFO
+ * return an array of these. Do not change this without also
+ * changing the corresponding SAL code.
+ */
+#define SN_HWPERF_MAXSTRING		128
+struct sn_hwperf_object_info {
+	u32 id;
+	union {
+		struct {
+			u64 this_part:1;
+			u64 is_shared:1;
+		} fields;
+		struct {
+			u64 flags;
+			u64 reserved;
+		} b;
+	} f;
+	char name[SN_HWPERF_MAXSTRING];
+	char location[SN_HWPERF_MAXSTRING];
+	u32 ports;
+};
+
+#define sn_hwp_this_part	f.fields.this_part
+#define sn_hwp_is_shared	f.fields.is_shared
+#define sn_hwp_flags		f.b.flags
+
+#define SN_HWPERF_FOREIGN(x)	(!(x)->sn_hwp_this_part && !(x)->sn_hwp_is_shared)
+
+/* numa port structure, SN_HWPERF_ENUM_PORTS returns an array of these */
+struct sn_hwperf_port_info {
+	u32 port;
+	u32 conn_id;
+	u32 conn_port;
+};
+
+/* for HWPERF_{GET,SET}_MMRS */
+struct sn_hwperf_data {
+	u64 addr;
+	u64 data;
+};
+
+/* user ioctl() argument, see below */
+struct sn_hwperf_ioctl_args {
+        u64 arg;		/* argument, usually an object id */
+        u64 sz;                 /* size of transfer */
+        void *ptr;              /* pointer to source/target */
+        u32 v0;			/* second return value */
+};
+
+/*
+ * For SN_HWPERF_{GET,SET}_MMRS and SN_HWPERF_OBJECT_DISTANCE,
+ * sn_hwperf_ioctl_args.arg can be used to specify a CPU on which
+ * to call SAL, and whether to use an interprocessor interrupt
+ * or task migration in order to do so. If the CPU specified is
+ * SN_HWPERF_ARG_ANY_CPU, then the current CPU will be used.
+ */
+#define SN_HWPERF_ARG_ANY_CPU		0x7fffffffUL
+#define SN_HWPERF_ARG_CPU_MASK		0x7fffffff00000000ULL
+#define SN_HWPERF_ARG_USE_IPI_MASK	0x8000000000000000ULL
+#define SN_HWPERF_ARG_OBJID_MASK	0x00000000ffffffffULL
+
+/* 
+ * ioctl requests on the "sn_hwperf" misc device that call SAL.
+ */
+#define SN_HWPERF_OP_MEM_COPYIN		0x1000
+#define SN_HWPERF_OP_MEM_COPYOUT	0x2000
+#define SN_HWPERF_OP_MASK		0x0fff
+
+/*
+ * Determine mem requirement.
+ * arg	don't care
+ * sz	8
+ * p	pointer to u64 integer
+ */
+#define	SN_HWPERF_GET_HEAPSIZE		1
+
+/*
+ * Install mem for SAL drvr
+ * arg	don't care
+ * sz	sizeof buffer pointed to by p
+ * p	pointer to buffer for scratch area
+ */
+#define SN_HWPERF_INSTALL_HEAP		2
+
+/*
+ * Determine number of objects
+ * arg	don't care
+ * sz	8
+ * p	pointer to u64 integer
+ */
+#define SN_HWPERF_OBJECT_COUNT		(10|SN_HWPERF_OP_MEM_COPYOUT)
+
+/*
+ * Determine object "distance", relative to a cpu. This operation can
+ * execute on a designated logical cpu number, using either an IPI or
+ * via task migration. If the cpu number is SN_HWPERF_ANY_CPU, then
+ * the current CPU is used. See the SN_HWPERF_ARG_* macros above.
+ *
+ * arg	bitmap of IPI flag, cpu number and object id
+ * sz	8
+ * p	pointer to u64 integer
+ */
+#define SN_HWPERF_OBJECT_DISTANCE	(11|SN_HWPERF_OP_MEM_COPYOUT)
+
+/*
+ * Enumerate objects. Special case if sz == 8, returns the required
+ * buffer size.
+ * arg	don't care
+ * sz	sizeof buffer pointed to by p
+ * p	pointer to array of struct sn_hwperf_object_info
+ */
+#define SN_HWPERF_ENUM_OBJECTS		(12|SN_HWPERF_OP_MEM_COPYOUT)
+
+/*
+ * Enumerate NumaLink ports for an object. Special case if sz == 8,
+ * returns the required buffer size.
+ * arg	object id
+ * sz	sizeof buffer pointed to by p
+ * p	pointer to array of struct sn_hwperf_port_info
+ */
+#define SN_HWPERF_ENUM_PORTS		(13|SN_HWPERF_OP_MEM_COPYOUT)
+
+/*
+ * SET/GET memory mapped registers. These operations can execute
+ * on a designated logical cpu number, using either an IPI or via
+ * task migration. If the cpu number is SN_HWPERF_ANY_CPU, then
+ * the current CPU is used. See the SN_HWPERF_ARG_* macros above.
+ *
+ * arg	bitmap of ipi flag, cpu number and object id
+ * sz	sizeof buffer pointed to by p
+ * p	pointer to array of struct sn_hwperf_data
+ */
+#define SN_HWPERF_SET_MMRS		(14|SN_HWPERF_OP_MEM_COPYIN)
+#define SN_HWPERF_GET_MMRS		(15|SN_HWPERF_OP_MEM_COPYOUT| \
+					    SN_HWPERF_OP_MEM_COPYIN)
+/*
+ * Lock a shared object
+ * arg	object id
+ * sz	don't care
+ * p	don't care
+ */
+#define SN_HWPERF_ACQUIRE		16
+
+/*
+ * Unlock a shared object
+ * arg	object id
+ * sz	don't care
+ * p	don't care
+ */
+#define SN_HWPERF_RELEASE		17
+
+/*
+ * Break a lock on a shared object
+ * arg	object id
+ * sz	don't care
+ * p	don't care
+ */
+#define SN_HWPERF_FORCE_RELEASE		18
+
+/*
+ * ioctl requests on "sn_hwperf" that do not call SAL
+ */
+
+/*
+ * get cpu info as an array of hwperf_object_info_t. 
+ * id is logical CPU number, name is description, location
+ * is geoid (e.g. 001c04#1c). Special case if sz == 8,
+ * returns the required buffer size.
+ *
+ * arg	don't care
+ * sz	sizeof buffer pointed to by p
+ * p	pointer to array of struct sn_hwperf_object_info
+ */
+#define SN_HWPERF_GET_CPU_INFO		(100|SN_HWPERF_OP_MEM_COPYOUT)
+
+/*
+ * Given an object id, return it's node number (aka cnode).
+ * arg	object id
+ * sz	8
+ * p	pointer to u64 integer
+ */
+#define SN_HWPERF_GET_OBJ_NODE		(101|SN_HWPERF_OP_MEM_COPYOUT)
+
+/*
+ * Given a node number (cnode), return it's nasid.
+ * arg	ordinal node number (aka cnodeid)
+ * sz	8
+ * p	pointer to u64 integer
+ */
+#define SN_HWPERF_GET_NODE_NASID	(102|SN_HWPERF_OP_MEM_COPYOUT)
+
+/* return codes */
+#define SN_HWPERF_OP_OK			0
+#define SN_HWPERF_OP_NOMEM		1
+#define SN_HWPERF_OP_NO_PERM		2
+#define SN_HWPERF_OP_IO_ERROR		3
+#define SN_HWPERF_OP_BUSY		4
+#define SN_HWPERF_OP_RECONFIGURE	253
+#define SN_HWPERF_OP_INVAL		254
+
+#endif				/* SN_HWPERF_H */
diff --git a/include/asm-ia64/sn/sn_sal.h b/include/asm-ia64/sn/sn_sal.h
index 8f47f4f34f5a..5a0490b4cda1 100644
--- a/include/asm-ia64/sn/sn_sal.h
+++ b/include/asm-ia64/sn/sn_sal.h
@@ -61,6 +61,7 @@
 
 #define  SN_SAL_SYSCTL_IOBRICK_PCI_OP		   0x02000042	// reentrant
 #define	 SN_SAL_IROUTER_OP			   0x02000043
+#define  SN_SAL_HWPERF_OP			   0x02000050   // lock
 
 /*
  * Service-specific constants
@@ -844,4 +845,21 @@ ia64_sn_irtr_init(nasid_t nasid, void *buf, int len)
 	return (int) rv.status;
 }
 
+/*
+ * This is the access point to the Altix PROM hardware performance
+ * and status monitoring interface. For info on using this, see
+ * include/asm-ia64/sn/sn2/sn_hwperf.h
+ */
+static inline int
+ia64_sn_hwperf_op(nasid_t nasid, u64 opcode, u64 a0, u64 a1, u64 a2,
+                  u64 a3, u64 a4, int *v0)
+{
+	struct ia64_sal_retval rv;
+	SAL_CALL_NOLOCK(rv, SN_SAL_HWPERF_OP, (u64)nasid,
+		opcode, a0, a1, a2, a3, a4);
+	if (v0)
+		*v0 = (int) rv.v0;
+	return (int) rv.status;
+}
+
 #endif /* _ASM_IA64_SN_SN_SAL_H */