x86: Move uv to platform

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Mike Travis <travis@sgi.com>

7 files changed, 2662 insertions, 0 deletions

diff --git a/arch/x86/platform/Makefile b/arch/x86/platform/Makefile
index 06761ed5374..8519b01f1ac 100644
--- a/arch/x86/platform/Makefile
+++ b/arch/x86/platform/Makefile
@@ -4,3 +4,4 @@ obj-y	+= mrst/
 obj-y	+= scx200/
 obj-y	+= sfi/
 obj-y	+= visws/
+obj-y	+= uv/
diff --git a/arch/x86/platform/uv/Makefile b/arch/x86/platform/uv/Makefile
new file mode 100644
index 00000000000..6c40995fefb
--- /dev/null
+++ b/arch/x86/platform/uv/Makefile
@@ -0,0 +1 @@
+obj-$(CONFIG_X86_UV)		+= tlb_uv.o bios_uv.o uv_irq.o uv_sysfs.o uv_time.o
diff --git a/arch/x86/platform/uv/bios_uv.c b/arch/x86/platform/uv/bios_uv.c
new file mode 100644
index 00000000000..8bc57baaa9a
--- /dev/null
+++ b/arch/x86/platform/uv/bios_uv.c
@@ -0,0 +1,215 @@
+/*
+ * BIOS run time interface routines.
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, write to the Free Software
+ *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
+ *
+ *  Copyright (c) 2008-2009 Silicon Graphics, Inc.  All Rights Reserved.
+ *  Copyright (c) Russ Anderson <rja@sgi.com>
+ */
+
+#include <linux/efi.h>
+#include <asm/efi.h>
+#include <linux/io.h>
+#include <asm/uv/bios.h>
+#include <asm/uv/uv_hub.h>
+
+static struct uv_systab uv_systab;
+
+s64 uv_bios_call(enum uv_bios_cmd which, u64 a1, u64 a2, u64 a3, u64 a4, u64 a5)
+{
+	struct uv_systab *tab = &uv_systab;
+	s64 ret;
+
+	if (!tab->function)
+		/*
+		 * BIOS does not support UV systab
+		 */
+		return BIOS_STATUS_UNIMPLEMENTED;
+
+	ret = efi_call6((void *)__va(tab->function), (u64)which,
+			a1, a2, a3, a4, a5);
+	return ret;
+}
+EXPORT_SYMBOL_GPL(uv_bios_call);
+
+s64 uv_bios_call_irqsave(enum uv_bios_cmd which, u64 a1, u64 a2, u64 a3,
+					u64 a4, u64 a5)
+{
+	unsigned long bios_flags;
+	s64 ret;
+
+	local_irq_save(bios_flags);
+	ret = uv_bios_call(which, a1, a2, a3, a4, a5);
+	local_irq_restore(bios_flags);
+
+	return ret;
+}
+
+s64 uv_bios_call_reentrant(enum uv_bios_cmd which, u64 a1, u64 a2, u64 a3,
+					u64 a4, u64 a5)
+{
+	s64 ret;
+
+	preempt_disable();
+	ret = uv_bios_call(which, a1, a2, a3, a4, a5);
+	preempt_enable();
+
+	return ret;
+}
+
+
+long sn_partition_id;
+EXPORT_SYMBOL_GPL(sn_partition_id);
+long sn_coherency_id;
+EXPORT_SYMBOL_GPL(sn_coherency_id);
+long sn_region_size;
+EXPORT_SYMBOL_GPL(sn_region_size);
+long system_serial_number;
+EXPORT_SYMBOL_GPL(system_serial_number);
+int uv_type;
+EXPORT_SYMBOL_GPL(uv_type);
+
+
+s64 uv_bios_get_sn_info(int fc, int *uvtype, long *partid, long *coher,
+		long *region, long *ssn)
+{
+	s64 ret;
+	u64 v0, v1;
+	union partition_info_u part;
+
+	ret = uv_bios_call_irqsave(UV_BIOS_GET_SN_INFO, fc,
+				(u64)(&v0), (u64)(&v1), 0, 0);
+	if (ret != BIOS_STATUS_SUCCESS)
+		return ret;
+
+	part.val = v0;
+	if (uvtype)
+		*uvtype = part.hub_version;
+	if (partid)
+		*partid = part.partition_id;
+	if (coher)
+		*coher = part.coherence_id;
+	if (region)
+		*region = part.region_size;
+	if (ssn)
+		*ssn = v1;
+	return ret;
+}
+EXPORT_SYMBOL_GPL(uv_bios_get_sn_info);
+
+int
+uv_bios_mq_watchlist_alloc(unsigned long addr, unsigned int mq_size,
+			   unsigned long *intr_mmr_offset)
+{
+	u64 watchlist;
+	s64 ret;
+
+	/*
+	 * bios returns watchlist number or negative error number.
+	 */
+	ret = (int)uv_bios_call_irqsave(UV_BIOS_WATCHLIST_ALLOC, addr,
+			mq_size, (u64)intr_mmr_offset,
+			(u64)&watchlist, 0);
+	if (ret < BIOS_STATUS_SUCCESS)
+		return ret;
+
+	return watchlist;
+}
+EXPORT_SYMBOL_GPL(uv_bios_mq_watchlist_alloc);
+
+int
+uv_bios_mq_watchlist_free(int blade, int watchlist_num)
+{
+	return (int)uv_bios_call_irqsave(UV_BIOS_WATCHLIST_FREE,
+				blade, watchlist_num, 0, 0, 0);
+}
+EXPORT_SYMBOL_GPL(uv_bios_mq_watchlist_free);
+
+s64
+uv_bios_change_memprotect(u64 paddr, u64 len, enum uv_memprotect perms)
+{
+	return uv_bios_call_irqsave(UV_BIOS_MEMPROTECT, paddr, len,
+					perms, 0, 0);
+}
+EXPORT_SYMBOL_GPL(uv_bios_change_memprotect);
+
+s64
+uv_bios_reserved_page_pa(u64 buf, u64 *cookie, u64 *addr, u64 *len)
+{
+	s64 ret;
+
+	ret = uv_bios_call_irqsave(UV_BIOS_GET_PARTITION_ADDR, (u64)cookie,
+					(u64)addr, buf, (u64)len, 0);
+	return ret;
+}
+EXPORT_SYMBOL_GPL(uv_bios_reserved_page_pa);
+
+s64 uv_bios_freq_base(u64 clock_type, u64 *ticks_per_second)
+{
+	return uv_bios_call(UV_BIOS_FREQ_BASE, clock_type,
+			   (u64)ticks_per_second, 0, 0, 0);
+}
+EXPORT_SYMBOL_GPL(uv_bios_freq_base);
+
+/*
+ * uv_bios_set_legacy_vga_target - Set Legacy VGA I/O Target
+ * @decode: true to enable target, false to disable target
+ * @domain: PCI domain number
+ * @bus: PCI bus number
+ *
+ * Returns:
+ *    0: Success
+ *    -EINVAL: Invalid domain or bus number
+ *    -ENOSYS: Capability not available
+ *    -EBUSY: Legacy VGA I/O cannot be retargeted at this time
+ */
+int uv_bios_set_legacy_vga_target(bool decode, int domain, int bus)
+{
+	return uv_bios_call(UV_BIOS_SET_LEGACY_VGA_TARGET,
+				(u64)decode, (u64)domain, (u64)bus, 0, 0);
+}
+EXPORT_SYMBOL_GPL(uv_bios_set_legacy_vga_target);
+
+
+#ifdef CONFIG_EFI
+void uv_bios_init(void)
+{
+	struct uv_systab *tab;
+
+	if ((efi.uv_systab == EFI_INVALID_TABLE_ADDR) ||
+	    (efi.uv_systab == (unsigned long)NULL)) {
+		printk(KERN_CRIT "No EFI UV System Table.\n");
+		uv_systab.function = (unsigned long)NULL;
+		return;
+	}
+
+	tab = (struct uv_systab *)ioremap(efi.uv_systab,
+					sizeof(struct uv_systab));
+	if (strncmp(tab->signature, "UVST", 4) != 0)
+		printk(KERN_ERR "bad signature in UV system table!");
+
+	/*
+	 * Copy table to permanent spot for later use.
+	 */
+	memcpy(&uv_systab, tab, sizeof(struct uv_systab));
+	iounmap(tab);
+
+	printk(KERN_INFO "EFI UV System Table Revision %d\n",
+					uv_systab.revision);
+}
+#else	/* !CONFIG_EFI */
+
+void uv_bios_init(void) { }
+#endif
diff --git a/arch/x86/platform/uv/tlb_uv.c b/arch/x86/platform/uv/tlb_uv.c
new file mode 100644
index 00000000000..20ea20a39e2
--- /dev/null
+++ b/arch/x86/platform/uv/tlb_uv.c
@@ -0,0 +1,1661 @@
+/*
+ *	SGI UltraViolet TLB flush routines.
+ *
+ *	(c) 2008-2010 Cliff Wickman <cpw@sgi.com>, SGI.
+ *
+ *	This code is released under the GNU General Public License version 2 or
+ *	later.
+ */
+#include <linux/seq_file.h>
+#include <linux/proc_fs.h>
+#include <linux/debugfs.h>
+#include <linux/kernel.h>
+#include <linux/slab.h>
+
+#include <asm/mmu_context.h>
+#include <asm/uv/uv.h>
+#include <asm/uv/uv_mmrs.h>
+#include <asm/uv/uv_hub.h>
+#include <asm/uv/uv_bau.h>
+#include <asm/apic.h>
+#include <asm/idle.h>
+#include <asm/tsc.h>
+#include <asm/irq_vectors.h>
+#include <asm/timer.h>
+
+/* timeouts in nanoseconds (indexed by UVH_AGING_PRESCALE_SEL urgency7 30:28) */
+static int timeout_base_ns[] = {
+		20,
+		160,
+		1280,
+		10240,
+		81920,
+		655360,
+		5242880,
+		167772160
+};
+static int timeout_us;
+static int nobau;
+static int baudisabled;
+static spinlock_t disable_lock;
+static cycles_t congested_cycles;
+
+/* tunables: */
+static int max_bau_concurrent = MAX_BAU_CONCURRENT;
+static int max_bau_concurrent_constant = MAX_BAU_CONCURRENT;
+static int plugged_delay = PLUGGED_DELAY;
+static int plugsb4reset = PLUGSB4RESET;
+static int timeoutsb4reset = TIMEOUTSB4RESET;
+static int ipi_reset_limit = IPI_RESET_LIMIT;
+static int complete_threshold = COMPLETE_THRESHOLD;
+static int congested_response_us = CONGESTED_RESPONSE_US;
+static int congested_reps = CONGESTED_REPS;
+static int congested_period = CONGESTED_PERIOD;
+static struct dentry *tunables_dir;
+static struct dentry *tunables_file;
+
+static int __init setup_nobau(char *arg)
+{
+	nobau = 1;
+	return 0;
+}
+early_param("nobau", setup_nobau);
+
+/* base pnode in this partition */
+static int uv_partition_base_pnode __read_mostly;
+/* position of pnode (which is nasid>>1): */
+static int uv_nshift __read_mostly;
+static unsigned long uv_mmask __read_mostly;
+
+static DEFINE_PER_CPU(struct ptc_stats, ptcstats);
+static DEFINE_PER_CPU(struct bau_control, bau_control);
+static DEFINE_PER_CPU(cpumask_var_t, uv_flush_tlb_mask);
+
+/*
+ * Determine the first node on a uvhub. 'Nodes' are used for kernel
+ * memory allocation.
+ */
+static int __init uvhub_to_first_node(int uvhub)
+{
+	int node, b;
+
+	for_each_online_node(node) {
+		b = uv_node_to_blade_id(node);
+		if (uvhub == b)
+			return node;
+	}
+	return -1;
+}
+
+/*
+ * Determine the apicid of the first cpu on a uvhub.
+ */
+static int __init uvhub_to_first_apicid(int uvhub)
+{
+	int cpu;
+
+	for_each_present_cpu(cpu)
+		if (uvhub == uv_cpu_to_blade_id(cpu))
+			return per_cpu(x86_cpu_to_apicid, cpu);
+	return -1;
+}
+
+/*
+ * Free a software acknowledge hardware resource by clearing its Pending
+ * bit. This will return a reply to the sender.
+ * If the message has timed out, a reply has already been sent by the
+ * hardware but the resource has not been released. In that case our
+ * clear of the Timeout bit (as well) will free the resource. No reply will
+ * be sent (the hardware will only do one reply per message).
+ */
+static inline void uv_reply_to_message(struct msg_desc *mdp,
+				       struct bau_control *bcp)
+{
+	unsigned long dw;
+	struct bau_payload_queue_entry *msg;
+
+	msg = mdp->msg;
+	if (!msg->canceled) {
+		dw = (msg->sw_ack_vector << UV_SW_ACK_NPENDING) |
+						msg->sw_ack_vector;
+		uv_write_local_mmr(
+				UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS, dw);
+	}
+	msg->replied_to = 1;
+	msg->sw_ack_vector = 0;
+}
+
+/*
+ * Process the receipt of a RETRY message
+ */
+static inline void uv_bau_process_retry_msg(struct msg_desc *mdp,
+					    struct bau_control *bcp)
+{
+	int i;
+	int cancel_count = 0;
+	int slot2;
+	unsigned long msg_res;
+	unsigned long mmr = 0;
+	struct bau_payload_queue_entry *msg;
+	struct bau_payload_queue_entry *msg2;
+	struct ptc_stats *stat;
+
+	msg = mdp->msg;
+	stat = bcp->statp;
+	stat->d_retries++;
+	/*
+	 * cancel any message from msg+1 to the retry itself
+	 */
+	for (msg2 = msg+1, i = 0; i < DEST_Q_SIZE; msg2++, i++) {
+		if (msg2 > mdp->va_queue_last)
+			msg2 = mdp->va_queue_first;
+		if (msg2 == msg)
+			break;
+
+		/* same conditions for cancellation as uv_do_reset */
+		if ((msg2->replied_to == 0) && (msg2->canceled == 0) &&
+		    (msg2->sw_ack_vector) && ((msg2->sw_ack_vector &
+			msg->sw_ack_vector) == 0) &&
+		    (msg2->sending_cpu == msg->sending_cpu) &&
+		    (msg2->msg_type != MSG_NOOP)) {
+			slot2 = msg2 - mdp->va_queue_first;
+			mmr = uv_read_local_mmr
+				(UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE);
+			msg_res = msg2->sw_ack_vector;
+			/*
+			 * This is a message retry; clear the resources held
+			 * by the previous message only if they timed out.
+			 * If it has not timed out we have an unexpected
+			 * situation to report.
+			 */
+			if (mmr & (msg_res << UV_SW_ACK_NPENDING)) {
+				/*
+				 * is the resource timed out?
+				 * make everyone ignore the cancelled message.
+				 */
+				msg2->canceled = 1;
+				stat->d_canceled++;
+				cancel_count++;
+				uv_write_local_mmr(
+				    UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS,
+					(msg_res << UV_SW_ACK_NPENDING) |
+					 msg_res);
+			}
+		}
+	}
+	if (!cancel_count)
+		stat->d_nocanceled++;
+}
+
+/*
+ * Do all the things a cpu should do for a TLB shootdown message.
+ * Other cpu's may come here at the same time for this message.
+ */
+static void uv_bau_process_message(struct msg_desc *mdp,
+				   struct bau_control *bcp)
+{
+	int msg_ack_count;
+	short socket_ack_count = 0;
+	struct ptc_stats *stat;
+	struct bau_payload_queue_entry *msg;
+	struct bau_control *smaster = bcp->socket_master;
+
+	/*
+	 * This must be a normal message, or retry of a normal message
+	 */
+	msg = mdp->msg;
+	stat = bcp->statp;
+	if (msg->address == TLB_FLUSH_ALL) {
+		local_flush_tlb();
+		stat->d_alltlb++;
+	} else {
+		__flush_tlb_one(msg->address);
+		stat->d_onetlb++;
+	}
+	stat->d_requestee++;
+
+	/*
+	 * One cpu on each uvhub has the additional job on a RETRY
+	 * of releasing the resource held by the message that is
+	 * being retried.  That message is identified by sending
+	 * cpu number.
+	 */
+	if (msg->msg_type == MSG_RETRY && bcp == bcp->uvhub_master)
+		uv_bau_process_retry_msg(mdp, bcp);
+
+	/*
+	 * This is a sw_ack message, so we have to reply to it.
+	 * Count each responding cpu on the socket. This avoids
+	 * pinging the count's cache line back and forth between
+	 * the sockets.
+	 */
+	socket_ack_count = atomic_add_short_return(1, (struct atomic_short *)
+			&smaster->socket_acknowledge_count[mdp->msg_slot]);
+	if (socket_ack_count == bcp->cpus_in_socket) {
+		/*
+		 * Both sockets dump their completed count total into
+		 * the message's count.
+		 */
+		smaster->socket_acknowledge_count[mdp->msg_slot] = 0;
+		msg_ack_count = atomic_add_short_return(socket_ack_count,
+				(struct atomic_short *)&msg->acknowledge_count);
+
+		if (msg_ack_count == bcp->cpus_in_uvhub) {
+			/*
+			 * All cpus in uvhub saw it; reply
+			 */
+			uv_reply_to_message(mdp, bcp);
+		}
+	}
+
+	return;
+}
+
+/*
+ * Determine the first cpu on a uvhub.
+ */
+static int uvhub_to_first_cpu(int uvhub)
+{
+	int cpu;
+	for_each_present_cpu(cpu)
+		if (uvhub == uv_cpu_to_blade_id(cpu))
+			return cpu;
+	return -1;
+}
+
+/*
+ * Last resort when we get a large number of destination timeouts is
+ * to clear resources held by a given cpu.
+ * Do this with IPI so that all messages in the BAU message queue
+ * can be identified by their nonzero sw_ack_vector field.
+ *
+ * This is entered for a single cpu on the uvhub.
+ * The sender want's this uvhub to free a specific message's
+ * sw_ack resources.
+ */
+static void
+uv_do_reset(void *ptr)
+{
+	int i;
+	int slot;
+	int count = 0;
+	unsigned long mmr;
+	unsigned long msg_res;
+	struct bau_control *bcp;
+	struct reset_args *rap;
+	struct bau_payload_queue_entry *msg;
+	struct ptc_stats *stat;
+
+	bcp = &per_cpu(bau_control, smp_processor_id());
+	rap = (struct reset_args *)ptr;
+	stat = bcp->statp;
+	stat->d_resets++;
+
+	/*
+	 * We're looking for the given sender, and
+	 * will free its sw_ack resource.
+	 * If all cpu's finally responded after the timeout, its
+	 * message 'replied_to' was set.
+	 */
+	for (msg = bcp->va_queue_first, i = 0; i < DEST_Q_SIZE; msg++, i++) {
+		/* uv_do_reset: same conditions for cancellation as
+		   uv_bau_process_retry_msg() */
+		if ((msg->replied_to == 0) &&
+		    (msg->canceled == 0) &&
+		    (msg->sending_cpu == rap->sender) &&
+		    (msg->sw_ack_vector) &&
+		    (msg->msg_type != MSG_NOOP)) {
+			/*
+			 * make everyone else ignore this message
+			 */
+			msg->canceled = 1;
+			slot = msg - bcp->va_queue_first;
+			count++;
+			/*
+			 * only reset the resource if it is still pending
+			 */
+			mmr = uv_read_local_mmr
+					(UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE);
+			msg_res = msg->sw_ack_vector;
+			if (mmr & msg_res) {
+				stat->d_rcanceled++;
+				uv_write_local_mmr(
+				    UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS,
+					(msg_res << UV_SW_ACK_NPENDING) |
+					 msg_res);
+			}
+		}
+	}
+	return;
+}
+
+/*
+ * Use IPI to get all target uvhubs to release resources held by
+ * a given sending cpu number.
+ */
+static void uv_reset_with_ipi(struct bau_target_uvhubmask *distribution,
+			      int sender)
+{
+	int uvhub;
+	int cpu;
+	cpumask_t mask;
+	struct reset_args reset_args;
+
+	reset_args.sender = sender;
+
+	cpus_clear(mask);
+	/* find a single cpu for each uvhub in this distribution mask */
+	for (uvhub = 0;
+		    uvhub < sizeof(struct bau_target_uvhubmask) * BITSPERBYTE;
+		    uvhub++) {
+		if (!bau_uvhub_isset(uvhub, distribution))
+			continue;
+		/* find a cpu for this uvhub */
+		cpu = uvhub_to_first_cpu(uvhub);
+		cpu_set(cpu, mask);
+	}
+	/* IPI all cpus; Preemption is already disabled */
+	smp_call_function_many(&mask, uv_do_reset, (void *)&reset_args, 1);
+	return;
+}
+
+static inline unsigned long
+cycles_2_us(unsigned long long cyc)
+{
+	unsigned long long ns;
+	unsigned long us;
+	ns =  (cyc * per_cpu(cyc2ns, smp_processor_id()))
+						>> CYC2NS_SCALE_FACTOR;
+	us = ns / 1000;
+	return us;
+}
+
+/*
+ * wait for all cpus on this hub to finish their sends and go quiet
+ * leaves uvhub_quiesce set so that no new broadcasts are started by
+ * bau_flush_send_and_wait()
+ */
+static inline void
+quiesce_local_uvhub(struct bau_control *hmaster)
+{
+	atomic_add_short_return(1, (struct atomic_short *)
+		 &hmaster->uvhub_quiesce);
+}
+
+/*
+ * mark this quiet-requestor as done
+ */
+static inline void
+end_uvhub_quiesce(struct bau_control *hmaster)
+{
+	atomic_add_short_return(-1, (struct atomic_short *)
+		&hmaster->uvhub_quiesce);
+}
+
+/*
+ * Wait for completion of a broadcast software ack message
+ * return COMPLETE, RETRY(PLUGGED or TIMEOUT) or GIVEUP
+ */
+static int uv_wait_completion(struct bau_desc *bau_desc,
+	unsigned long mmr_offset, int right_shift, int this_cpu,
+	struct bau_control *bcp, struct bau_control *smaster, long try)
+{
+	unsigned long descriptor_status;
+	cycles_t ttime;
+	struct ptc_stats *stat = bcp->statp;
+	struct bau_control *hmaster;
+
+	hmaster = bcp->uvhub_master;
+
+	/* spin on the status MMR, waiting for it to go idle */
+	while ((descriptor_status = (((unsigned long)
+		uv_read_local_mmr(mmr_offset) >>
+			right_shift) & UV_ACT_STATUS_MASK)) !=
+			DESC_STATUS_IDLE) {
+		/*
+		 * Our software ack messages may be blocked because there are
+		 * no swack resources available.  As long as none of them
+		 * has timed out hardware will NACK our message and its
+		 * state will stay IDLE.
+		 */
+		if (descriptor_status == DESC_STATUS_SOURCE_TIMEOUT) {
+			stat->s_stimeout++;
+			return FLUSH_GIVEUP;
+		} else if (descriptor_status ==
+					DESC_STATUS_DESTINATION_TIMEOUT) {
+			stat->s_dtimeout++;
+			ttime = get_cycles();
+
+			/*
+			 * Our retries may be blocked by all destination
+			 * swack resources being consumed, and a timeout
+			 * pending.  In that case hardware returns the
+			 * ERROR that looks like a destination timeout.
+			 */
+			if (cycles_2_us(ttime - bcp->send_message) <
+							timeout_us) {
+				bcp->conseccompletes = 0;
+				return FLUSH_RETRY_PLUGGED;
+			}
+
+			bcp->conseccompletes = 0;
+			return FLUSH_RETRY_TIMEOUT;
+		} else {
+			/*
+			 * descriptor_status is still BUSY
+			 */
+			cpu_relax();
+		}
+	}
+	bcp->conseccompletes++;
+	return FLUSH_COMPLETE;
+}
+
+static inline cycles_t
+sec_2_cycles(unsigned long sec)
+{
+	unsigned long ns;
+	cycles_t cyc;
+
+	ns = sec * 1000000000;
+	cyc = (ns << CYC2NS_SCALE_FACTOR)/(per_cpu(cyc2ns, smp_processor_id()));
+	return cyc;
+}
+
+/*
+ * conditionally add 1 to *v, unless *v is >= u
+ * return 0 if we cannot add 1 to *v because it is >= u
+ * return 1 if we can add 1 to *v because it is < u
+ * the add is atomic
+ *
+ * This is close to atomic_add_unless(), but this allows the 'u' value
+ * to be lowered below the current 'v'.  atomic_add_unless can only stop
+ * on equal.
+ */
+static inline int atomic_inc_unless_ge(spinlock_t *lock, atomic_t *v, int u)
+{
+	spin_lock(lock);
+	if (atomic_read(v) >= u) {
+		spin_unlock(lock);
+		return 0;
+	}
+	atomic_inc(v);
+	spin_unlock(lock);
+	return 1;
+}
+
+/*
+ * Our retries are blocked by all destination swack resources being
+ * in use, and a timeout is pending. In that case hardware immediately
+ * returns the ERROR that looks like a destination timeout.
+ */
+static void
+destination_plugged(struct bau_desc *bau_desc, struct bau_control *bcp,
+			struct bau_control *hmaster, struct ptc_stats *stat)
+{
+	udelay(bcp->plugged_delay);
+	bcp->plugged_tries++;
+	if (bcp->plugged_tries >= bcp->plugsb4reset) {
+		bcp->plugged_tries = 0;
+		quiesce_local_uvhub(hmaster);
+		spin_lock(&hmaster->queue_lock);
+		uv_reset_with_ipi(&bau_desc->distribution, bcp->cpu);
+		spin_unlock(&hmaster->queue_lock);
+		end_uvhub_quiesce(hmaster);
+		bcp->ipi_attempts++;
+		stat->s_resets_plug++;
+	}
+}
+
+static void
+destination_timeout(struct bau_desc *bau_desc, struct bau_control *bcp,
+			struct bau_control *hmaster, struct ptc_stats *stat)
+{
+	hmaster->max_bau_concurrent = 1;
+	bcp->timeout_tries++;
+	if (bcp->timeout_tries >= bcp->timeoutsb4reset) {
+		bcp->timeout_tries = 0;
+		quiesce_local_uvhub(hmaster);
+		spin_lock(&hmaster->queue_lock);
+		uv_reset_with_ipi(&bau_desc->distribution, bcp->cpu);
+		spin_unlock(&hmaster->queue_lock);
+		end_uvhub_quiesce(hmaster);
+		bcp->ipi_attempts++;
+		stat->s_resets_timeout++;
+	}
+}
+
+/*
+ * Completions are taking a very long time due to a congested numalink
+ * network.
+ */
+static void
+disable_for_congestion(struct bau_control *bcp, struct ptc_stats *stat)
+{
+	int tcpu;
+	struct bau_control *tbcp;
+
+	/* let only one cpu do this disabling */
+	spin_lock(&disable_lock);
+	if (!baudisabled && bcp->period_requests &&
+	    ((bcp->period_time / bcp->period_requests) > congested_cycles)) {
+		/* it becomes this cpu's job to turn on the use of the
+		   BAU again */
+		baudisabled = 1;
+		bcp->set_bau_off = 1;
+		bcp->set_bau_on_time = get_cycles() +
+			sec_2_cycles(bcp->congested_period);
+		stat->s_bau_disabled++;
+		for_each_present_cpu(tcpu) {
+			tbcp = &per_cpu(bau_control, tcpu);
+				tbcp->baudisabled = 1;
+		}
+	}
+	spin_unlock(&disable_lock);
+}
+
+/**
+ * uv_flush_send_and_wait
+ *
+ * Send a broadcast and wait for it to complete.
+ *
+ * The flush_mask contains the cpus the broadcast is to be sent to including
+ * cpus that are on the local uvhub.
+ *
+ * Returns 0 if all flushing represented in the mask was done.
+ * Returns 1 if it gives up entirely and the original cpu mask is to be
+ * returned to the kernel.
+ */
+int uv_flush_send_and_wait(struct bau_desc *bau_desc,
+			   struct cpumask *flush_mask, struct bau_control *bcp)
+{
+	int right_shift;
+	int completion_status = 0;
+	int seq_number = 0;
+	long try = 0;
+	int cpu = bcp->uvhub_cpu;
+	int this_cpu = bcp->cpu;
+	unsigned long mmr_offset;
+	unsigned long index;
+	cycles_t time1;
+	cycles_t time2;
+	cycles_t elapsed;
+	struct ptc_stats *stat = bcp->statp;
+	struct bau_control *smaster = bcp->socket_master;
+	struct bau_control *hmaster = bcp->uvhub_master;
+
+	if (!atomic_inc_unless_ge(&hmaster->uvhub_lock,
+			&hmaster->active_descriptor_count,
+			hmaster->max_bau_concurrent)) {
+		stat->s_throttles++;
+		do {
+			cpu_relax();
+		} while (!atomic_inc_unless_ge(&hmaster->uvhub_lock,
+			&hmaster->active_descriptor_count,
+			hmaster->max_bau_concurrent));
+	}
+	while (hmaster->uvhub_quiesce)
+		cpu_relax();
+
+	if (cpu < UV_CPUS_PER_ACT_STATUS) {
+		mmr_offset = UVH_LB_BAU_SB_ACTIVATION_STATUS_0;
+		right_shift = cpu * UV_ACT_STATUS_SIZE;
+	} else {
+		mmr_offset = UVH_LB_BAU_SB_ACTIVATION_STATUS_1;
+		right_shift =
+		    ((cpu - UV_CPUS_PER_ACT_STATUS) * UV_ACT_STATUS_SIZE);
+	}
+	time1 = get_cycles();
+	do {
+		if (try == 0) {
+			bau_desc->header.msg_type = MSG_REGULAR;
+			seq_number = bcp->message_number++;
+		} else {
+			bau_desc->header.msg_type = MSG_RETRY;
+			stat->s_retry_messages++;
+		}
+		bau_desc->header.sequence = seq_number;
+		index = (1UL << UVH_LB_BAU_SB_ACTIVATION_CONTROL_PUSH_SHFT) |
+			bcp->uvhub_cpu;
+		bcp->send_message = get_cycles();
+		uv_write_local_mmr(UVH_LB_BAU_SB_ACTIVATION_CONTROL, index);
+		try++;
+		completion_status = uv_wait_completion(bau_desc, mmr_offset,
+			right_shift, this_cpu, bcp, smaster, try);
+
+		if (completion_status == FLUSH_RETRY_PLUGGED) {
+			destination_plugged(bau_desc, bcp, hmaster, stat);
+		} else if (completion_status == FLUSH_RETRY_TIMEOUT) {
+			destination_timeout(bau_desc, bcp, hmaster, stat);
+		}
+		if (bcp->ipi_attempts >= bcp->ipi_reset_limit) {
+			bcp->ipi_attempts = 0;
+			completion_status = FLUSH_GIVEUP;
+			break;
+		}
+		cpu_relax();
+	} while ((completion_status == FLUSH_RETRY_PLUGGED) ||
+		 (completion_status == FLUSH_RETRY_TIMEOUT));
+	time2 = get_cycles();
+	bcp->plugged_tries = 0;
+	bcp->timeout_tries = 0;
+	if ((completion_status == FLUSH_COMPLETE) &&
+	    (bcp->conseccompletes > bcp->complete_threshold) &&
+	    (hmaster->max_bau_concurrent <
+					hmaster->max_bau_concurrent_constant))
+			hmaster->max_bau_concurrent++;
+	while (hmaster->uvhub_quiesce)
+		cpu_relax();
+	atomic_dec(&hmaster->active_descriptor_count);
+	if (time2 > time1) {
+		elapsed = time2 - time1;
+		stat->s_time += elapsed;
+		if ((completion_status == FLUSH_COMPLETE) && (try == 1)) {
+			bcp->period_requests++;
+			bcp->period_time += elapsed;
+			if ((elapsed > congested_cycles) &&
+			    (bcp->period_requests > bcp->congested_reps)) {
+				disable_for_congestion(bcp, stat);
+			}
+		}
+	} else
+		stat->s_requestor--;
+	if (completion_status == FLUSH_COMPLETE && try > 1)
+		stat->s_retriesok++;
+	else if (completion_status == FLUSH_GIVEUP) {
+		stat->s_giveup++;
+		return 1;
+	}
+	return 0;
+}
+
+/**
+ * uv_flush_tlb_others - globally purge translation cache of a virtual
+ * address or all TLB's
+ * @cpumask: mask of all cpu's in which the address is to be removed
+ * @mm: mm_struct containing virtual address range
+ * @va: virtual address to be removed (or TLB_FLUSH_ALL for all TLB's on cpu)
+ * @cpu: the current cpu
+ *
+ * This is the entry point for initiating any UV global TLB shootdown.
+ *
+ * Purges the translation caches of all specified processors of the given
+ * virtual address, or purges all TLB's on specified processors.
+ *
+ * The caller has derived the cpumask from the mm_struct.  This function
+ * is called only if there are bits set in the mask. (e.g. flush_tlb_page())
+ *
+ * The cpumask is converted into a uvhubmask of the uvhubs containing
+ * those cpus.
+ *
+ * Note that this function should be called with preemption disabled.
+ *
+ * Returns NULL if all remote flushing was done.
+ * Returns pointer to cpumask if some remote flushing remains to be
+ * done.  The returned pointer is valid till preemption is re-enabled.
+ */
+const struct cpumask *uv_flush_tlb_others(const struct cpumask *cpumask,
+					  struct mm_struct *mm,
+					  unsigned long va, unsigned int cpu)
+{
+	int tcpu;
+	int uvhub;
+	int locals = 0;
+	int remotes = 0;
+	int hubs = 0;
+	struct bau_desc *bau_desc;
+	struct cpumask *flush_mask;
+	struct ptc_stats *stat;
+	struct bau_control *bcp;
+	struct bau_control *tbcp;
+
+	/* kernel was booted 'nobau' */
+	if (nobau)
+		return cpumask;
+
+	bcp = &per_cpu(bau_control, cpu);
+	stat = bcp->statp;
+
+	/* bau was disabled due to slow response */
+	if (bcp->baudisabled) {
+		/* the cpu that disabled it must re-enable it */
+		if (bcp->set_bau_off) {
+			if (get_cycles() >= bcp->set_bau_on_time) {
+				stat->s_bau_reenabled++;
+				baudisabled = 0;
+				for_each_present_cpu(tcpu) {
+					tbcp = &per_cpu(bau_control, tcpu);
+					tbcp->baudisabled = 0;
+					tbcp->period_requests = 0;
+					tbcp->period_time = 0;
+				}
+			}
+		}
+		return cpumask;
+	}
+
+	/*
+	 * Each sending cpu has a per-cpu mask which it fills from the caller's
+	 * cpu mask.  All cpus are converted to uvhubs and copied to the
+	 * activation descriptor.
+	 */
+	flush_mask = (struct cpumask *)per_cpu(uv_flush_tlb_mask, cpu);
+	/* don't actually do a shootdown of the local cpu */
+	cpumask_andnot(flush_mask, cpumask, cpumask_of(cpu));
+	if (cpu_isset(cpu, *cpumask))
+		stat->s_ntargself++;
+
+	bau_desc = bcp->descriptor_base;
+	bau_desc += UV_ITEMS_PER_DESCRIPTOR * bcp->uvhub_cpu;
+	bau_uvhubs_clear(&bau_desc->distribution, UV_DISTRIBUTION_SIZE);
+
+	/* cpu statistics */
+	for_each_cpu(tcpu, flush_mask) {
+		uvhub = uv_cpu_to_blade_id(tcpu);
+		bau_uvhub_set(uvhub, &bau_desc->distribution);
+		if (uvhub == bcp->uvhub)
+			locals++;
+		else
+			remotes++;
+	}
+	if ((locals + remotes) == 0)
+		return NULL;
+	stat->s_requestor++;
+	stat->s_ntargcpu += remotes + locals;
+	stat->s_ntargremotes += remotes;
+	stat->s_ntarglocals += locals;
+	remotes = bau_uvhub_weight(&bau_desc->distribution);
+
+	/* uvhub statistics */
+	hubs = bau_uvhub_weight(&bau_desc->distribution);
+	if (locals) {
+		stat->s_ntarglocaluvhub++;
+		stat->s_ntargremoteuvhub += (hubs - 1);
+	} else
+		stat->s_ntargremoteuvhub += hubs;
+	stat->s_ntarguvhub += hubs;
+	if (hubs >= 16)
+		stat->s_ntarguvhub16++;
+	else if (hubs >= 8)
+		stat->s_ntarguvhub8++;
+	else if (hubs >= 4)
+		stat->s_ntarguvhub4++;
+	else if (hubs >= 2)
+		stat->s_ntarguvhub2++;
+	else
+		stat->s_ntarguvhub1++;
+
+	bau_desc->payload.address = va;
+	bau_desc->payload.sending_cpu = cpu;
+
+	/*
+	 * uv_flush_send_and_wait returns 0 if all cpu's were messaged,
+	 * or 1 if it gave up and the original cpumask should be returned.
+	 */
+	if (!uv_flush_send_and_wait(bau_desc, flush_mask, bcp))
+		return NULL;
+	else
+		return cpumask;
+}
+
+/*
+ * The BAU message interrupt comes here. (registered by set_intr_gate)
+ * See entry_64.S
+ *
+ * We received a broadcast assist message.
+ *
+ * Interrupts are disabled; this interrupt could represent
+ * the receipt of several messages.
+ *
+ * All cores/threads on this hub get this interrupt.
+ * The last one to see it does the software ack.
+ * (the resource will not be freed until noninterruptable cpus see this
+ *  interrupt; hardware may timeout the s/w ack and reply ERROR)
+ */
+void uv_bau_message_interrupt(struct pt_regs *regs)
+{
+	int count = 0;
+	cycles_t time_start;
+	struct bau_payload_queue_entry *msg;
+	struct bau_control *bcp;
+	struct ptc_stats *stat;
+	struct msg_desc msgdesc;
+
+	time_start = get_cycles();
+	bcp = &per_cpu(bau_control, smp_processor_id());
+	stat = bcp->statp;
+	msgdesc.va_queue_first = bcp->va_queue_first;
+	msgdesc.va_queue_last = bcp->va_queue_last;
+	msg = bcp->bau_msg_head;
+	while (msg->sw_ack_vector) {
+		count++;
+		msgdesc.msg_slot = msg - msgdesc.va_queue_first;
+		msgdesc.sw_ack_slot = ffs(msg->sw_ack_vector) - 1;
+		msgdesc.msg = msg;
+		uv_bau_process_message(&msgdesc, bcp);
+		msg++;
+		if (msg > msgdesc.va_queue_last)
+			msg = msgdesc.va_queue_first;
+		bcp->bau_msg_head = msg;
+	}
+	stat->d_time += (get_cycles() - time_start);
+	if (!count)
+		stat->d_nomsg++;
+	else if (count > 1)
+		stat->d_multmsg++;
+	ack_APIC_irq();
+}
+
+/*
+ * uv_enable_timeouts
+ *
+ * Each target uvhub (i.e. a uvhub that has no cpu's) needs to have
+ * shootdown message timeouts enabled.  The timeout does not cause
+ * an interrupt, but causes an error message to be returned to
+ * the sender.
+ */
+static void uv_enable_timeouts(void)
+{
+	int uvhub;
+	int nuvhubs;
+	int pnode;
+	unsigned long mmr_image;
+
+	nuvhubs = uv_num_possible_blades();
+
+	for (uvhub = 0; uvhub < nuvhubs; uvhub++) {
+		if (!uv_blade_nr_possible_cpus(uvhub))
+			continue;
+
+		pnode = uv_blade_to_pnode(uvhub);
+		mmr_image =
+		    uv_read_global_mmr64(pnode, UVH_LB_BAU_MISC_CONTROL);
+		/*
+		 * Set the timeout period and then lock it in, in three
+		 * steps; captures and locks in the period.
+		 *
+		 * To program the period, the SOFT_ACK_MODE must be off.
+		 */
+		mmr_image &= ~((unsigned long)1 <<
+		    UVH_LB_BAU_MISC_CONTROL_ENABLE_INTD_SOFT_ACK_MODE_SHFT);
+		uv_write_global_mmr64
+		    (pnode, UVH_LB_BAU_MISC_CONTROL, mmr_image);
+		/*
+		 * Set the 4-bit period.
+		 */
+		mmr_image &= ~((unsigned long)0xf <<
+		     UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHFT);
+		mmr_image |= (UV_INTD_SOFT_ACK_TIMEOUT_PERIOD <<
+		     UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHFT);
+		uv_write_global_mmr64
+		    (pnode, UVH_LB_BAU_MISC_CONTROL, mmr_image);
+		/*
+		 * Subsequent reversals of the timebase bit (3) cause an
+		 * immediate timeout of one or all INTD resources as
+		 * indicated in bits 2:0 (7 causes all of them to timeout).
+		 */
+		mmr_image |= ((unsigned long)1 <<
+		    UVH_LB_BAU_MISC_CONTROL_ENABLE_INTD_SOFT_ACK_MODE_SHFT);
+		uv_write_global_mmr64
+		    (pnode, UVH_LB_BAU_MISC_CONTROL, mmr_image);
+	}
+}
+
+static void *uv_ptc_seq_start(struct seq_file *file, loff_t *offset)
+{
+	if (*offset < num_possible_cpus())
+		return offset;
+	return NULL;
+}
+
+static void *uv_ptc_seq_next(struct seq_file *file, void *data, loff_t *offset)
+{
+	(*offset)++;
+	if (*offset < num_possible_cpus())
+		return offset;
+	return NULL;
+}
+
+static void uv_ptc_seq_stop(struct seq_file *file, void *data)
+{
+}
+
+static inline unsigned long long
+microsec_2_cycles(unsigned long microsec)
+{
+	unsigned long ns;
+	unsigned long long cyc;
+
+	ns = microsec * 1000;
+	cyc = (ns << CYC2NS_SCALE_FACTOR)/(per_cpu(cyc2ns, smp_processor_id()));
+	return cyc;
+}
+
+/*
+ * Display the statistics thru /proc.
+ * 'data' points to the cpu number
+ */
+static int uv_ptc_seq_show(struct seq_file *file, void *data)
+{
+	struct ptc_stats *stat;
+	int cpu;
+
+	cpu = *(loff_t *)data;
+
+	if (!cpu) {
+		seq_printf(file,
+			"# cpu sent stime self locals remotes ncpus localhub ");
+		seq_printf(file,
+			"remotehub numuvhubs numuvhubs16 numuvhubs8 ");
+		seq_printf(file,
+			"numuvhubs4 numuvhubs2 numuvhubs1 dto ");
+		seq_printf(file,
+			"retries rok resetp resett giveup sto bz throt ");
+		seq_printf(file,
+			"sw_ack recv rtime all ");
+		seq_printf(file,
+			"one mult none retry canc nocan reset rcan ");
+		seq_printf(file,
+			"disable enable\n");
+	}
+	if (cpu < num_possible_cpus() && cpu_online(cpu)) {
+		stat = &per_cpu(ptcstats, cpu);
+		/* source side statistics */
+		seq_printf(file,
+			"cpu %d %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld ",
+			   cpu, stat->s_requestor, cycles_2_us(stat->s_time),
+			   stat->s_ntargself, stat->s_ntarglocals,
+			   stat->s_ntargremotes, stat->s_ntargcpu,
+			   stat->s_ntarglocaluvhub, stat->s_ntargremoteuvhub,
+			   stat->s_ntarguvhub, stat->s_ntarguvhub16);
+		seq_printf(file, "%ld %ld %ld %ld %ld ",
+			   stat->s_ntarguvhub8, stat->s_ntarguvhub4,
+			   stat->s_ntarguvhub2, stat->s_ntarguvhub1,
+			   stat->s_dtimeout);
+		seq_printf(file, "%ld %ld %ld %ld %ld %ld %ld %ld ",
+			   stat->s_retry_messages, stat->s_retriesok,
+			   stat->s_resets_plug, stat->s_resets_timeout,
+			   stat->s_giveup, stat->s_stimeout,
+			   stat->s_busy, stat->s_throttles);
+
+		/* destination side statistics */
+		seq_printf(file,
+			   "%lx %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld ",
+			   uv_read_global_mmr64(uv_cpu_to_pnode(cpu),
+					UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE),
+			   stat->d_requestee, cycles_2_us(stat->d_time),
+			   stat->d_alltlb, stat->d_onetlb, stat->d_multmsg,
+			   stat->d_nomsg, stat->d_retries, stat->d_canceled,
+			   stat->d_nocanceled, stat->d_resets,
+			   stat->d_rcanceled);
+		seq_printf(file, "%ld %ld\n",
+			stat->s_bau_disabled, stat->s_bau_reenabled);
+	}
+
+	return 0;
+}
+
+/*
+ * Display the tunables thru debugfs
+ */
+static ssize_t tunables_read(struct file *file, char __user *userbuf,
+						size_t count, loff_t *ppos)
+{
+	char *buf;
+	int ret;
+
+	buf = kasprintf(GFP_KERNEL, "%s %s %s\n%d %d %d %d %d %d %d %d %d\n",
+		"max_bau_concurrent plugged_delay plugsb4reset",
+		"timeoutsb4reset ipi_reset_limit complete_threshold",
+		"congested_response_us congested_reps congested_period",
+		max_bau_concurrent, plugged_delay, plugsb4reset,
+		timeoutsb4reset, ipi_reset_limit, complete_threshold,
+		congested_response_us, congested_reps, congested_period);
+
+	if (!buf)
+		return -ENOMEM;
+
+	ret = simple_read_from_buffer(userbuf, count, ppos, buf, strlen(buf));
+	kfree(buf);
+	return ret;
+}
+
+/*
+ * -1: resetf the statistics
+ *  0: display meaning of the statistics
+ */
+static ssize_t uv_ptc_proc_write(struct file *file, const char __user *user,
+				 size_t count, loff_t *data)
+{
+	int cpu;
+	long input_arg;
+	char optstr[64];
+	struct ptc_stats *stat;
+
+	if (count == 0 || count > sizeof(optstr))
+		return -EINVAL;
+	if (copy_from_user(optstr, user, count))
+		return -EFAULT;
+	optstr[count - 1] = '\0';
+	if (strict_strtol(optstr, 10, &input_arg) < 0) {
+		printk(KERN_DEBUG "%s is invalid\n", optstr);
+		return -EINVAL;
+	}
+
+	if (input_arg == 0) {
+		printk(KERN_DEBUG "# cpu:      cpu number\n");
+		printk(KERN_DEBUG "Sender statistics:\n");
+		printk(KERN_DEBUG
+		"sent:     number of shootdown messages sent\n");
+		printk(KERN_DEBUG
+		"stime:    time spent sending messages\n");
+		printk(KERN_DEBUG
+		"numuvhubs: number of hubs targeted with shootdown\n");
+		printk(KERN_DEBUG
+		"numuvhubs16: number times 16 or more hubs targeted\n");
+		printk(KERN_DEBUG
+		"numuvhubs8: number times 8 or more hubs targeted\n");
+		printk(KERN_DEBUG
+		"numuvhubs4: number times 4 or more hubs targeted\n");
+		printk(KERN_DEBUG
+		"numuvhubs2: number times 2 or more hubs targeted\n");
+		printk(KERN_DEBUG
+		"numuvhubs1: number times 1 hub targeted\n");
+		printk(KERN_DEBUG
+		"numcpus:  number of cpus targeted with shootdown\n");
+		printk(KERN_DEBUG
+		"dto:      number of destination timeouts\n");
+		printk(KERN_DEBUG
+		"retries:  destination timeout retries sent\n");
+		printk(KERN_DEBUG
+		"rok:   :  destination timeouts successfully retried\n");
+		printk(KERN_DEBUG
+		"resetp:   ipi-style resource resets for plugs\n");
+		printk(KERN_DEBUG
+		"resett:   ipi-style resource resets for timeouts\n");
+		printk(KERN_DEBUG
+		"giveup:   fall-backs to ipi-style shootdowns\n");
+		printk(KERN_DEBUG
+		"sto:      number of source timeouts\n");
+		printk(KERN_DEBUG
+		"bz:       number of stay-busy's\n");
+		printk(KERN_DEBUG
+		"throt:    number times spun in throttle\n");
+		printk(KERN_DEBUG "Destination side statistics:\n");
+		printk(KERN_DEBUG
+		"sw_ack:   image of UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE\n");
+		printk(KERN_DEBUG
+		"recv:     shootdown messages received\n");
+		printk(KERN_DEBUG
+		"rtime:    time spent processing messages\n");
+		printk(KERN_DEBUG
+		"all:      shootdown all-tlb messages\n");
+		printk(KERN_DEBUG
+		"one:      shootdown one-tlb messages\n");
+		printk(KERN_DEBUG
+		"mult:     interrupts that found multiple messages\n");
+		printk(KERN_DEBUG
+		"none:     interrupts that found no messages\n");
+		printk(KERN_DEBUG
+		"retry:    number of retry messages processed\n");
+		printk(KERN_DEBUG
+		"canc:     number messages canceled by retries\n");
+		printk(KERN_DEBUG
+		"nocan:    number retries that found nothing to cancel\n");
+		printk(KERN_DEBUG
+		"reset:    number of ipi-style reset requests processed\n");
+		printk(KERN_DEBUG
+		"rcan:     number messages canceled by reset requests\n");
+		printk(KERN_DEBUG
+		"disable:  number times use of the BAU was disabled\n");
+		printk(KERN_DEBUG
+		"enable:   number times use of the BAU was re-enabled\n");
+	} else if (input_arg == -1) {
+		for_each_present_cpu(cpu) {
+			stat = &per_cpu(ptcstats, cpu);
+			memset(stat, 0, sizeof(struct ptc_stats));
+		}
+	}
+
+	return count;
+}
+
+static int local_atoi(const char *name)
+{
+	int val = 0;
+
+	for (;; name++) {
+		switch (*name) {
+		case '0' ... '9':
+			val = 10*val+(*name-'0');
+			break;
+		default:
+			return val;
+		}
+	}
+}
+
+/*
+ * set the tunables
+ * 0 values reset them to defaults
+ */
+static ssize_t tunables_write(struct file *file, const char __user *user,
+				 size_t count, loff_t *data)
+{
+	int cpu;
+	int cnt = 0;
+	int val;
+	char *p;
+	char *q;
+	char instr[64];
+	struct bau_control *bcp;
+
+	if (count == 0 || count > sizeof(instr)-1)
+		return -EINVAL;
+	if (copy_from_user(instr, user, count))
+		return -EFAULT;
+
+	instr[count] = '\0';
+	/* count the fields */
+	p = instr + strspn(instr, WHITESPACE);
+	q = p;
+	for (; *p; p = q + strspn(q, WHITESPACE)) {
+		q = p + strcspn(p, WHITESPACE);
+		cnt++;
+		if (q == p)
+			break;
+	}
+	if (cnt != 9) {
+		printk(KERN_INFO "bau tunable error: should be 9 numbers\n");
+		return -EINVAL;
+	}
+
+	p = instr + strspn(instr, WHITESPACE);
+	q = p;
+	for (cnt = 0; *p; p = q + strspn(q, WHITESPACE), cnt++) {
+		q = p + strcspn(p, WHITESPACE);
+		val = local_atoi(p);
+		switch (cnt) {
+		case 0:
+			if (val == 0) {
+				max_bau_concurrent = MAX_BAU_CONCURRENT;
+				max_bau_concurrent_constant =
+							MAX_BAU_CONCURRENT;
+				continue;
+			}
+			bcp = &per_cpu(bau_control, smp_processor_id());
+			if (val < 1 || val > bcp->cpus_in_uvhub) {
+				printk(KERN_DEBUG
+				"Error: BAU max concurrent %d is invalid\n",
+				val);
+				return -EINVAL;
+			}
+			max_bau_concurrent = val;
+			max_bau_concurrent_constant = val;
+			continue;
+		case 1:
+			if (val == 0)
+				plugged_delay = PLUGGED_DELAY;
+			else
+				plugged_delay = val;
+			continue;
+		case 2:
+			if (val == 0)
+				plugsb4reset = PLUGSB4RESET;
+			else
+				plugsb4reset = val;
+			continue;
+		case 3:
+			if (val == 0)
+				timeoutsb4reset = TIMEOUTSB4RESET;
+			else
+				timeoutsb4reset = val;
+			continue;
+		case 4:
+			if (val == 0)
+				ipi_reset_limit = IPI_RESET_LIMIT;
+			else
+				ipi_reset_limit = val;
+			continue;
+		case 5:
+			if (val == 0)
+				complete_threshold = COMPLETE_THRESHOLD;
+			else
+				complete_threshold = val;
+			continue;
+		case 6:
+			if (val == 0)
+				congested_response_us = CONGESTED_RESPONSE_US;
+			else
+				congested_response_us = val;
+			continue;
+		case 7:
+			if (val == 0)
+				congested_reps = CONGESTED_REPS;
+			else
+				congested_reps = val;
+			continue;
+		case 8:
+			if (val == 0)
+				congested_period = CONGESTED_PERIOD;
+			else
+				congested_period = val;
+			continue;
+		}
+		if (q == p)
+			break;
+	}
+	for_each_present_cpu(cpu) {
+		bcp = &per_cpu(bau_control, cpu);
+		bcp->max_bau_concurrent = max_bau_concurrent;
+		bcp->max_bau_concurrent_constant = max_bau_concurrent;
+		bcp->plugged_delay = plugged_delay;
+		bcp->plugsb4reset = plugsb4reset;
+		bcp->timeoutsb4reset = timeoutsb4reset;
+		bcp->ipi_reset_limit = ipi_reset_limit;
+		bcp->complete_threshold = complete_threshold;
+		bcp->congested_response_us = congested_response_us;
+		bcp->congested_reps = congested_reps;
+		bcp->congested_period = congested_period;
+	}
+	return count;
+}
+
+static const struct seq_operations uv_ptc_seq_ops = {
+	.start		= uv_ptc_seq_start,
+	.next		= uv_ptc_seq_next,
+	.stop		= uv_ptc_seq_stop,
+	.show		= uv_ptc_seq_show
+};
+
+static int uv_ptc_proc_open(struct inode *inode, struct file *file)
+{
+	return seq_open(file, &uv_ptc_seq_ops);
+}
+
+static int tunables_open(struct inode *inode, struct file *file)
+{
+	return 0;
+}
+
+static const struct file_operations proc_uv_ptc_operations = {
+	.open		= uv_ptc_proc_open,
+	.read		= seq_read,
+	.write		= uv_ptc_proc_write,
+	.llseek		= seq_lseek,
+	.release	= seq_release,
+};
+
+static const struct file_operations tunables_fops = {
+	.open		= tunables_open,
+	.read		= tunables_read,
+	.write		= tunables_write,
+	.llseek		= default_llseek,
+};
+
+static int __init uv_ptc_init(void)
+{
+	struct proc_dir_entry *proc_uv_ptc;
+
+	if (!is_uv_system())
+		return 0;
+
+	proc_uv_ptc = proc_create(UV_PTC_BASENAME, 0444, NULL,
+				  &proc_uv_ptc_operations);
+	if (!proc_uv_ptc) {
+		printk(KERN_ERR "unable to create %s proc entry\n",
+		       UV_PTC_BASENAME);
+		return -EINVAL;
+	}
+
+	tunables_dir = debugfs_create_dir(UV_BAU_TUNABLES_DIR, NULL);
+	if (!tunables_dir) {
+		printk(KERN_ERR "unable to create debugfs directory %s\n",
+		       UV_BAU_TUNABLES_DIR);
+		return -EINVAL;
+	}
+	tunables_file = debugfs_create_file(UV_BAU_TUNABLES_FILE, 0600,
+			tunables_dir, NULL, &tunables_fops);
+	if (!tunables_file) {
+		printk(KERN_ERR "unable to create debugfs file %s\n",
+		       UV_BAU_TUNABLES_FILE);
+		return -EINVAL;
+	}
+	return 0;
+}
+
+/*
+ * initialize the sending side's sending buffers
+ */
+static void
+uv_activation_descriptor_init(int node, int pnode)
+{
+	int i;
+	int cpu;
+	unsigned long pa;
+	unsigned long m;
+	unsigned long n;
+	struct bau_desc *bau_desc;
+	struct bau_desc *bd2;
+	struct bau_control *bcp;
+
+	/*
+	 * each bau_desc is 64 bytes; there are 8 (UV_ITEMS_PER_DESCRIPTOR)
+	 * per cpu; and up to 32 (UV_ADP_SIZE) cpu's per uvhub
+	 */
+	bau_desc = (struct bau_desc *)kmalloc_node(sizeof(struct bau_desc)*
+		UV_ADP_SIZE*UV_ITEMS_PER_DESCRIPTOR, GFP_KERNEL, node);
+	BUG_ON(!bau_desc);
+
+	pa = uv_gpa(bau_desc); /* need the real nasid*/
+	n = pa >> uv_nshift;
+	m = pa & uv_mmask;
+
+	uv_write_global_mmr64(pnode, UVH_LB_BAU_SB_DESCRIPTOR_BASE,
+			      (n << UV_DESC_BASE_PNODE_SHIFT | m));
+
+	/*
+	 * initializing all 8 (UV_ITEMS_PER_DESCRIPTOR) descriptors for each
+	 * cpu even though we only use the first one; one descriptor can
+	 * describe a broadcast to 256 uv hubs.
+	 */
+	for (i = 0, bd2 = bau_desc; i < (UV_ADP_SIZE*UV_ITEMS_PER_DESCRIPTOR);
+		i++, bd2++) {
+		memset(bd2, 0, sizeof(struct bau_desc));
+		bd2->header.sw_ack_flag = 1;
+		/*
+		 * base_dest_nodeid is the nasid (pnode<<1) of the first uvhub
+		 * in the partition. The bit map will indicate uvhub numbers,
+		 * which are 0-N in a partition. Pnodes are unique system-wide.
+		 */
+		bd2->header.base_dest_nodeid = uv_partition_base_pnode << 1;
+		bd2->header.dest_subnodeid = 0x10; /* the LB */
+		bd2->header.command = UV_NET_ENDPOINT_INTD;
+		bd2->header.int_both = 1;
+		/*
+		 * all others need to be set to zero:
+		 *   fairness chaining multilevel count replied_to
+		 */
+	}
+	for_each_present_cpu(cpu) {
+		if (pnode != uv_blade_to_pnode(uv_cpu_to_blade_id(cpu)))
+			continue;
+		bcp = &per_cpu(bau_control, cpu);
+		bcp->descriptor_base = bau_desc;
+	}
+}
+
+/*
+ * initialize the destination side's receiving buffers
+ * entered for each uvhub in the partition
+ * - node is first node (kernel memory notion) on the uvhub
+ * - pnode is the uvhub's physical identifier
+ */
+static void
+uv_payload_queue_init(int node, int pnode)
+{
+	int pn;
+	int cpu;
+	char *cp;
+	unsigned long pa;
+	struct bau_payload_queue_entry *pqp;
+	struct bau_payload_queue_entry *pqp_malloc;
+	struct bau_control *bcp;
+
+	pqp = (struct bau_payload_queue_entry *) kmalloc_node(
+		(DEST_Q_SIZE + 1) * sizeof(struct bau_payload_queue_entry),
+		GFP_KERNEL, node);
+	BUG_ON(!pqp);
+	pqp_malloc = pqp;
+
+	cp = (char *)pqp + 31;
+	pqp = (struct bau_payload_queue_entry *)(((unsigned long)cp >> 5) << 5);
+
+	for_each_present_cpu(cpu) {
+		if (pnode != uv_cpu_to_pnode(cpu))
+			continue;
+		/* for every cpu on this pnode: */
+		bcp = &per_cpu(bau_control, cpu);
+		bcp->va_queue_first = pqp;
+		bcp->bau_msg_head = pqp;
+		bcp->va_queue_last = pqp + (DEST_Q_SIZE - 1);
+	}
+	/*
+	 * need the pnode of where the memory was really allocated
+	 */
+	pa = uv_gpa(pqp);
+	pn = pa >> uv_nshift;
+	uv_write_global_mmr64(pnode,
+			      UVH_LB_BAU_INTD_PAYLOAD_QUEUE_FIRST,
+			      ((unsigned long)pn << UV_PAYLOADQ_PNODE_SHIFT) |
+			      uv_physnodeaddr(pqp));
+	uv_write_global_mmr64(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_TAIL,
+			      uv_physnodeaddr(pqp));
+	uv_write_global_mmr64(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_LAST,
+			      (unsigned long)
+			      uv_physnodeaddr(pqp + (DEST_Q_SIZE - 1)));
+	/* in effect, all msg_type's are set to MSG_NOOP */
+	memset(pqp, 0, sizeof(struct bau_payload_queue_entry) * DEST_Q_SIZE);
+}
+
+/*
+ * Initialization of each UV hub's structures
+ */
+static void __init uv_init_uvhub(int uvhub, int vector)
+{
+	int node;
+	int pnode;
+	unsigned long apicid;
+
+	node = uvhub_to_first_node(uvhub);
+	pnode = uv_blade_to_pnode(uvhub);
+	uv_activation_descriptor_init(node, pnode);
+	uv_payload_queue_init(node, pnode);
+	/*
+	 * the below initialization can't be in firmware because the
+	 * messaging IRQ will be determined by the OS
+	 */
+	apicid = uvhub_to_first_apicid(uvhub);
+	uv_write_global_mmr64(pnode, UVH_BAU_DATA_CONFIG,
+				      ((apicid << 32) | vector));
+}
+
+/*
+ * We will set BAU_MISC_CONTROL with a timeout period.
+ * But the BIOS has set UVH_AGING_PRESCALE_SEL and UVH_TRANSACTION_TIMEOUT.
+ * So the destination timeout period has be be calculated from them.
+ */
+static int
+calculate_destination_timeout(void)
+{
+	unsigned long mmr_image;
+	int mult1;
+	int mult2;
+	int index;
+	int base;
+	int ret;
+	unsigned long ts_ns;
+
+	mult1 = UV_INTD_SOFT_ACK_TIMEOUT_PERIOD & BAU_MISC_CONTROL_MULT_MASK;
+	mmr_image = uv_read_local_mmr(UVH_AGING_PRESCALE_SEL);
+	index = (mmr_image >> BAU_URGENCY_7_SHIFT) & BAU_URGENCY_7_MASK;
+	mmr_image = uv_read_local_mmr(UVH_TRANSACTION_TIMEOUT);
+	mult2 = (mmr_image >> BAU_TRANS_SHIFT) & BAU_TRANS_MASK;
+	base = timeout_base_ns[index];
+	ts_ns = base * mult1 * mult2;
+	ret = ts_ns / 1000;
+	return ret;
+}
+
+/*
+ * initialize the bau_control structure for each cpu
+ */
+static void __init uv_init_per_cpu(int nuvhubs)
+{
+	int i;
+	int cpu;
+	int pnode;
+	int uvhub;
+	int have_hmaster;
+	short socket = 0;
+	unsigned short socket_mask;
+	unsigned char *uvhub_mask;
+	struct bau_control *bcp;
+	struct uvhub_desc *bdp;
+	struct socket_desc *sdp;
+	struct bau_control *hmaster = NULL;
+	struct bau_control *smaster = NULL;
+	struct socket_desc {
+		short num_cpus;
+		short cpu_number[16];
+	};
+	struct uvhub_desc {
+		unsigned short socket_mask;
+		short num_cpus;
+		short uvhub;
+		short pnode;
+		struct socket_desc socket[2];
+	};
+	struct uvhub_desc *uvhub_descs;
+
+	timeout_us = calculate_destination_timeout();
+
+	uvhub_descs = (struct uvhub_desc *)
+		kmalloc(nuvhubs * sizeof(struct uvhub_desc), GFP_KERNEL);
+	memset(uvhub_descs, 0, nuvhubs * sizeof(struct uvhub_desc));
+	uvhub_mask = kzalloc((nuvhubs+7)/8, GFP_KERNEL);
+	for_each_present_cpu(cpu) {
+		bcp = &per_cpu(bau_control, cpu);
+		memset(bcp, 0, sizeof(struct bau_control));
+		pnode = uv_cpu_hub_info(cpu)->pnode;
+		uvhub = uv_cpu_hub_info(cpu)->numa_blade_id;
+		*(uvhub_mask + (uvhub/8)) |= (1 << (uvhub%8));
+		bdp = &uvhub_descs[uvhub];
+		bdp->num_cpus++;
+		bdp->uvhub = uvhub;
+		bdp->pnode = pnode;
+		/* kludge: 'assuming' one node per socket, and assuming that
+		   disabling a socket just leaves a gap in node numbers */
+		socket = (cpu_to_node(cpu) & 1);
+		bdp->socket_mask |= (1 << socket);
+		sdp = &bdp->socket[socket];
+		sdp->cpu_number[sdp->num_cpus] = cpu;
+		sdp->num_cpus++;
+	}
+	for (uvhub = 0; uvhub < nuvhubs; uvhub++) {
+		if (!(*(uvhub_mask + (uvhub/8)) & (1 << (uvhub%8))))
+			continue;
+		have_hmaster = 0;
+		bdp = &uvhub_descs[uvhub];
+		socket_mask = bdp->socket_mask;
+		socket = 0;
+		while (socket_mask) {
+			if (!(socket_mask & 1))
+				goto nextsocket;
+			sdp = &bdp->socket[socket];
+			for (i = 0; i < sdp->num_cpus; i++) {
+				cpu = sdp->cpu_number[i];
+				bcp = &per_cpu(bau_control, cpu);
+				bcp->cpu = cpu;
+				if (i == 0) {
+					smaster = bcp;
+					if (!have_hmaster) {
+						have_hmaster++;
+						hmaster = bcp;
+					}
+				}
+				bcp->cpus_in_uvhub = bdp->num_cpus;
+				bcp->cpus_in_socket = sdp->num_cpus;
+				bcp->socket_master = smaster;
+				bcp->uvhub = bdp->uvhub;
+				bcp->uvhub_master = hmaster;
+				bcp->uvhub_cpu = uv_cpu_hub_info(cpu)->
+						blade_processor_id;
+			}
+nextsocket:
+			socket++;
+			socket_mask = (socket_mask >> 1);
+		}
+	}
+	kfree(uvhub_descs);
+	kfree(uvhub_mask);
+	for_each_present_cpu(cpu) {
+		bcp = &per_cpu(bau_control, cpu);
+		bcp->baudisabled = 0;
+		bcp->statp = &per_cpu(ptcstats, cpu);
+		/* time interval to catch a hardware stay-busy bug */
+		bcp->timeout_interval = microsec_2_cycles(2*timeout_us);
+		bcp->max_bau_concurrent = max_bau_concurrent;
+		bcp->max_bau_concurrent_constant = max_bau_concurrent;
+		bcp->plugged_delay = plugged_delay;
+		bcp->plugsb4reset = plugsb4reset;
+		bcp->timeoutsb4reset = timeoutsb4reset;
+		bcp->ipi_reset_limit = ipi_reset_limit;
+		bcp->complete_threshold = complete_threshold;
+		bcp->congested_response_us = congested_response_us;
+		bcp->congested_reps = congested_reps;
+		bcp->congested_period = congested_period;
+	}
+}
+
+/*
+ * Initialization of BAU-related structures
+ */
+static int __init uv_bau_init(void)
+{
+	int uvhub;
+	int pnode;
+	int nuvhubs;
+	int cur_cpu;
+	int vector;
+	unsigned long mmr;
+
+	if (!is_uv_system())
+		return 0;
+
+	if (nobau)
+		return 0;
+
+	for_each_possible_cpu(cur_cpu)
+		zalloc_cpumask_var_node(&per_cpu(uv_flush_tlb_mask, cur_cpu),
+				       GFP_KERNEL, cpu_to_node(cur_cpu));
+
+	uv_nshift = uv_hub_info->m_val;
+	uv_mmask = (1UL << uv_hub_info->m_val) - 1;
+	nuvhubs = uv_num_possible_blades();
+	spin_lock_init(&disable_lock);
+	congested_cycles = microsec_2_cycles(congested_response_us);
+
+	uv_init_per_cpu(nuvhubs);
+
+	uv_partition_base_pnode = 0x7fffffff;
+	for (uvhub = 0; uvhub < nuvhubs; uvhub++)
+		if (uv_blade_nr_possible_cpus(uvhub) &&
+			(uv_blade_to_pnode(uvhub) < uv_partition_base_pnode))
+			uv_partition_base_pnode = uv_blade_to_pnode(uvhub);
+
+	vector = UV_BAU_MESSAGE;
+	for_each_possible_blade(uvhub)
+		if (uv_blade_nr_possible_cpus(uvhub))
+			uv_init_uvhub(uvhub, vector);
+
+	uv_enable_timeouts();
+	alloc_intr_gate(vector, uv_bau_message_intr1);
+
+	for_each_possible_blade(uvhub) {
+		if (uv_blade_nr_possible_cpus(uvhub)) {
+			pnode = uv_blade_to_pnode(uvhub);
+			/* INIT the bau */
+			uv_write_global_mmr64(pnode,
+					UVH_LB_BAU_SB_ACTIVATION_CONTROL,
+					((unsigned long)1 << 63));
+			mmr = 1; /* should be 1 to broadcast to both sockets */
+			uv_write_global_mmr64(pnode, UVH_BAU_DATA_BROADCAST,
+						mmr);
+		}
+	}
+
+	return 0;
+}
+core_initcall(uv_bau_init);
+fs_initcall(uv_ptc_init);
diff --git a/arch/x86/platform/uv/uv_irq.c b/arch/x86/platform/uv/uv_irq.c
new file mode 100644
index 00000000000..7b24460917d
--- /dev/null
+++ b/arch/x86/platform/uv/uv_irq.c
@@ -0,0 +1,285 @@
+/*
+ * 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.
+ *
+ * SGI UV IRQ functions
+ *
+ * Copyright (C) 2008 Silicon Graphics, Inc. All rights reserved.
+ */
+
+#include <linux/module.h>
+#include <linux/rbtree.h>
+#include <linux/slab.h>
+#include <linux/irq.h>
+
+#include <asm/apic.h>
+#include <asm/uv/uv_irq.h>
+#include <asm/uv/uv_hub.h>
+
+/* MMR offset and pnode of hub sourcing interrupts for a given irq */
+struct uv_irq_2_mmr_pnode{
+	struct rb_node		list;
+	unsigned long		offset;
+	int			pnode;
+	int			irq;
+};
+
+static spinlock_t		uv_irq_lock;
+static struct rb_root		uv_irq_root;
+
+static int uv_set_irq_affinity(struct irq_data *, const struct cpumask *, bool);
+
+static void uv_noop(struct irq_data *data) { }
+
+static void uv_ack_apic(struct irq_data *data)
+{
+	ack_APIC_irq();
+}
+
+static struct irq_chip uv_irq_chip = {
+	.name			= "UV-CORE",
+	.irq_mask		= uv_noop,
+	.irq_unmask		= uv_noop,
+	.irq_eoi		= uv_ack_apic,
+	.irq_set_affinity	= uv_set_irq_affinity,
+};
+
+/*
+ * Add offset and pnode information of the hub sourcing interrupts to the
+ * rb tree for a specific irq.
+ */
+static int uv_set_irq_2_mmr_info(int irq, unsigned long offset, unsigned blade)
+{
+	struct rb_node **link = &uv_irq_root.rb_node;
+	struct rb_node *parent = NULL;
+	struct uv_irq_2_mmr_pnode *n;
+	struct uv_irq_2_mmr_pnode *e;
+	unsigned long irqflags;
+
+	n = kmalloc_node(sizeof(struct uv_irq_2_mmr_pnode), GFP_KERNEL,
+				uv_blade_to_memory_nid(blade));
+	if (!n)
+		return -ENOMEM;
+
+	n->irq = irq;
+	n->offset = offset;
+	n->pnode = uv_blade_to_pnode(blade);
+	spin_lock_irqsave(&uv_irq_lock, irqflags);
+	/* Find the right place in the rbtree: */
+	while (*link) {
+		parent = *link;
+		e = rb_entry(parent, struct uv_irq_2_mmr_pnode, list);
+
+		if (unlikely(irq == e->irq)) {
+			/* irq entry exists */
+			e->pnode = uv_blade_to_pnode(blade);
+			e->offset = offset;
+			spin_unlock_irqrestore(&uv_irq_lock, irqflags);
+			kfree(n);
+			return 0;
+		}
+
+		if (irq < e->irq)
+			link = &(*link)->rb_left;
+		else
+			link = &(*link)->rb_right;
+	}
+
+	/* Insert the node into the rbtree. */
+	rb_link_node(&n->list, parent, link);
+	rb_insert_color(&n->list, &uv_irq_root);
+
+	spin_unlock_irqrestore(&uv_irq_lock, irqflags);
+	return 0;
+}
+
+/* Retrieve offset and pnode information from the rb tree for a specific irq */
+int uv_irq_2_mmr_info(int irq, unsigned long *offset, int *pnode)
+{
+	struct uv_irq_2_mmr_pnode *e;
+	struct rb_node *n;
+	unsigned long irqflags;
+
+	spin_lock_irqsave(&uv_irq_lock, irqflags);
+	n = uv_irq_root.rb_node;
+	while (n) {
+		e = rb_entry(n, struct uv_irq_2_mmr_pnode, list);
+
+		if (e->irq == irq) {
+			*offset = e->offset;
+			*pnode = e->pnode;
+			spin_unlock_irqrestore(&uv_irq_lock, irqflags);
+			return 0;
+		}
+
+		if (irq < e->irq)
+			n = n->rb_left;
+		else
+			n = n->rb_right;
+	}
+	spin_unlock_irqrestore(&uv_irq_lock, irqflags);
+	return -1;
+}
+
+/*
+ * Re-target the irq to the specified CPU and enable the specified MMR located
+ * on the specified blade to allow the sending of MSIs to the specified CPU.
+ */
+static int
+arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
+		       unsigned long mmr_offset, int limit)
+{
+	const struct cpumask *eligible_cpu = cpumask_of(cpu);
+	struct irq_cfg *cfg = get_irq_chip_data(irq);
+	unsigned long mmr_value;
+	struct uv_IO_APIC_route_entry *entry;
+	int mmr_pnode, err;
+
+	BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) !=
+			sizeof(unsigned long));
+
+	err = assign_irq_vector(irq, cfg, eligible_cpu);
+	if (err != 0)
+		return err;
+
+	if (limit == UV_AFFINITY_CPU)
+		irq_set_status_flags(irq, IRQ_NO_BALANCING);
+	else
+		irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
+
+	set_irq_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq,
+				      irq_name);
+
+	mmr_value = 0;
+	entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
+	entry->vector		= cfg->vector;
+	entry->delivery_mode	= apic->irq_delivery_mode;
+	entry->dest_mode	= apic->irq_dest_mode;
+	entry->polarity		= 0;
+	entry->trigger		= 0;
+	entry->mask		= 0;
+	entry->dest		= apic->cpu_mask_to_apicid(eligible_cpu);
+
+	mmr_pnode = uv_blade_to_pnode(mmr_blade);
+	uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
+
+	if (cfg->move_in_progress)
+		send_cleanup_vector(cfg);
+
+	return irq;
+}
+
+/*
+ * Disable the specified MMR located on the specified blade so that MSIs are
+ * longer allowed to be sent.
+ */
+static void arch_disable_uv_irq(int mmr_pnode, unsigned long mmr_offset)
+{
+	unsigned long mmr_value;
+	struct uv_IO_APIC_route_entry *entry;
+
+	BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) !=
+			sizeof(unsigned long));
+
+	mmr_value = 0;
+	entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
+	entry->mask = 1;
+
+	uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
+}
+
+static int
+uv_set_irq_affinity(struct irq_data *data, const struct cpumask *mask,
+		    bool force)
+{
+	struct irq_cfg *cfg = data->chip_data;
+	unsigned int dest;
+	unsigned long mmr_value, mmr_offset;
+	struct uv_IO_APIC_route_entry *entry;
+	int mmr_pnode;
+
+	if (__ioapic_set_affinity(data, mask, &dest))
+		return -1;
+
+	mmr_value = 0;
+	entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
+
+	entry->vector		= cfg->vector;
+	entry->delivery_mode	= apic->irq_delivery_mode;
+	entry->dest_mode	= apic->irq_dest_mode;
+	entry->polarity		= 0;
+	entry->trigger		= 0;
+	entry->mask		= 0;
+	entry->dest		= dest;
+
+	/* Get previously stored MMR and pnode of hub sourcing interrupts */
+	if (uv_irq_2_mmr_info(data->irq, &mmr_offset, &mmr_pnode))
+		return -1;
+
+	uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
+
+	if (cfg->move_in_progress)
+		send_cleanup_vector(cfg);
+
+	return 0;
+}
+
+/*
+ * Set up a mapping of an available irq and vector, and enable the specified
+ * MMR that defines the MSI that is to be sent to the specified CPU when an
+ * interrupt is raised.
+ */
+int uv_setup_irq(char *irq_name, int cpu, int mmr_blade,
+		 unsigned long mmr_offset, int limit)
+{
+	int irq, ret;
+
+	irq = create_irq_nr(NR_IRQS_LEGACY, uv_blade_to_memory_nid(mmr_blade));
+
+	if (irq <= 0)
+		return -EBUSY;
+
+	ret = arch_enable_uv_irq(irq_name, irq, cpu, mmr_blade, mmr_offset,
+		limit);
+	if (ret == irq)
+		uv_set_irq_2_mmr_info(irq, mmr_offset, mmr_blade);
+	else
+		destroy_irq(irq);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(uv_setup_irq);
+
+/*
+ * Tear down a mapping of an irq and vector, and disable the specified MMR that
+ * defined the MSI that was to be sent to the specified CPU when an interrupt
+ * was raised.
+ *
+ * Set mmr_blade and mmr_offset to what was passed in on uv_setup_irq().
+ */
+void uv_teardown_irq(unsigned int irq)
+{
+	struct uv_irq_2_mmr_pnode *e;
+	struct rb_node *n;
+	unsigned long irqflags;
+
+	spin_lock_irqsave(&uv_irq_lock, irqflags);
+	n = uv_irq_root.rb_node;
+	while (n) {
+		e = rb_entry(n, struct uv_irq_2_mmr_pnode, list);
+		if (e->irq == irq) {
+			arch_disable_uv_irq(e->pnode, e->offset);
+			rb_erase(n, &uv_irq_root);
+			kfree(e);
+			break;
+		}
+		if (irq < e->irq)
+			n = n->rb_left;
+		else
+			n = n->rb_right;
+	}
+	spin_unlock_irqrestore(&uv_irq_lock, irqflags);
+	destroy_irq(irq);
+}
+EXPORT_SYMBOL_GPL(uv_teardown_irq);
diff --git a/arch/x86/platform/uv/uv_sysfs.c b/arch/x86/platform/uv/uv_sysfs.c
new file mode 100644
index 00000000000..309c70fb775
--- /dev/null
+++ b/arch/x86/platform/uv/uv_sysfs.c
@@ -0,0 +1,76 @@
+/*
+ * This file supports the /sys/firmware/sgi_uv interfaces for SGI UV.
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, write to the Free Software
+ *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
+ *
+ *  Copyright (c) 2008 Silicon Graphics, Inc.  All Rights Reserved.
+ *  Copyright (c) Russ Anderson
+ */
+
+#include <linux/sysdev.h>
+#include <asm/uv/bios.h>
+#include <asm/uv/uv.h>
+
+struct kobject *sgi_uv_kobj;
+
+static ssize_t partition_id_show(struct kobject *kobj,
+			struct kobj_attribute *attr, char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%ld\n", sn_partition_id);
+}
+
+static ssize_t coherence_id_show(struct kobject *kobj,
+			struct kobj_attribute *attr, char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%ld\n", partition_coherence_id());
+}
+
+static struct kobj_attribute partition_id_attr =
+	__ATTR(partition_id, S_IRUGO, partition_id_show, NULL);
+
+static struct kobj_attribute coherence_id_attr =
+	__ATTR(coherence_id, S_IRUGO, coherence_id_show, NULL);
+
+
+static int __init sgi_uv_sysfs_init(void)
+{
+	unsigned long ret;
+
+	if (!is_uv_system())
+		return -ENODEV;
+
+	if (!sgi_uv_kobj)
+		sgi_uv_kobj = kobject_create_and_add("sgi_uv", firmware_kobj);
+	if (!sgi_uv_kobj) {
+		printk(KERN_WARNING "kobject_create_and_add sgi_uv failed\n");
+		return -EINVAL;
+	}
+
+	ret = sysfs_create_file(sgi_uv_kobj, &partition_id_attr.attr);
+	if (ret) {
+		printk(KERN_WARNING "sysfs_create_file partition_id failed\n");
+		return ret;
+	}
+
+	ret = sysfs_create_file(sgi_uv_kobj, &coherence_id_attr.attr);
+	if (ret) {
+		printk(KERN_WARNING "sysfs_create_file coherence_id failed\n");
+		return ret;
+	}
+
+	return 0;
+}
+
+device_initcall(sgi_uv_sysfs_init);
diff --git a/arch/x86/platform/uv/uv_time.c b/arch/x86/platform/uv/uv_time.c
new file mode 100644
index 00000000000..56e421bc379
--- /dev/null
+++ b/arch/x86/platform/uv/uv_time.c
@@ -0,0 +1,423 @@
+/*
+ * SGI RTC clock/timer routines.
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, write to the Free Software
+ *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
+ *
+ *  Copyright (c) 2009 Silicon Graphics, Inc.  All Rights Reserved.
+ *  Copyright (c) Dimitri Sivanich
+ */
+#include <linux/clockchips.h>
+#include <linux/slab.h>
+
+#include <asm/uv/uv_mmrs.h>
+#include <asm/uv/uv_hub.h>
+#include <asm/uv/bios.h>
+#include <asm/uv/uv.h>
+#include <asm/apic.h>
+#include <asm/cpu.h>
+
+#define RTC_NAME		"sgi_rtc"
+
+static cycle_t uv_read_rtc(struct clocksource *cs);
+static int uv_rtc_next_event(unsigned long, struct clock_event_device *);
+static void uv_rtc_timer_setup(enum clock_event_mode,
+				struct clock_event_device *);
+
+static struct clocksource clocksource_uv = {
+	.name		= RTC_NAME,
+	.rating		= 400,
+	.read		= uv_read_rtc,
+	.mask		= (cycle_t)UVH_RTC_REAL_TIME_CLOCK_MASK,
+	.shift		= 10,
+	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
+};
+
+static struct clock_event_device clock_event_device_uv = {
+	.name		= RTC_NAME,
+	.features	= CLOCK_EVT_FEAT_ONESHOT,
+	.shift		= 20,
+	.rating		= 400,
+	.irq		= -1,
+	.set_next_event	= uv_rtc_next_event,
+	.set_mode	= uv_rtc_timer_setup,
+	.event_handler	= NULL,
+};
+
+static DEFINE_PER_CPU(struct clock_event_device, cpu_ced);
+
+/* There is one of these allocated per node */
+struct uv_rtc_timer_head {
+	spinlock_t	lock;
+	/* next cpu waiting for timer, local node relative: */
+	int		next_cpu;
+	/* number of cpus on this node: */
+	int		ncpus;
+	struct {
+		int	lcpu;		/* systemwide logical cpu number */
+		u64	expires;	/* next timer expiration for this cpu */
+	} cpu[1];
+};
+
+/*
+ * Access to uv_rtc_timer_head via blade id.
+ */
+static struct uv_rtc_timer_head		**blade_info __read_mostly;
+
+static int				uv_rtc_evt_enable;
+
+/*
+ * Hardware interface routines
+ */
+
+/* Send IPIs to another node */
+static void uv_rtc_send_IPI(int cpu)
+{
+	unsigned long apicid, val;
+	int pnode;
+
+	apicid = cpu_physical_id(cpu);
+	pnode = uv_apicid_to_pnode(apicid);
+	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
+	      (apicid << UVH_IPI_INT_APIC_ID_SHFT) |
+	      (X86_PLATFORM_IPI_VECTOR << UVH_IPI_INT_VECTOR_SHFT);
+
+	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
+}
+
+/* Check for an RTC interrupt pending */
+static int uv_intr_pending(int pnode)
+{
+	return uv_read_global_mmr64(pnode, UVH_EVENT_OCCURRED0) &
+		UVH_EVENT_OCCURRED0_RTC1_MASK;
+}
+
+/* Setup interrupt and return non-zero if early expiration occurred. */
+static int uv_setup_intr(int cpu, u64 expires)
+{
+	u64 val;
+	int pnode = uv_cpu_to_pnode(cpu);
+
+	uv_write_global_mmr64(pnode, UVH_RTC1_INT_CONFIG,
+		UVH_RTC1_INT_CONFIG_M_MASK);
+	uv_write_global_mmr64(pnode, UVH_INT_CMPB, -1L);
+
+	uv_write_global_mmr64(pnode, UVH_EVENT_OCCURRED0_ALIAS,
+		UVH_EVENT_OCCURRED0_RTC1_MASK);
+
+	val = (X86_PLATFORM_IPI_VECTOR << UVH_RTC1_INT_CONFIG_VECTOR_SHFT) |
+		((u64)cpu_physical_id(cpu) << UVH_RTC1_INT_CONFIG_APIC_ID_SHFT);
+
+	/* Set configuration */
+	uv_write_global_mmr64(pnode, UVH_RTC1_INT_CONFIG, val);
+	/* Initialize comparator value */
+	uv_write_global_mmr64(pnode, UVH_INT_CMPB, expires);
+
+	if (uv_read_rtc(NULL) <= expires)
+		return 0;
+
+	return !uv_intr_pending(pnode);
+}
+
+/*
+ * Per-cpu timer tracking routines
+ */
+
+static __init void uv_rtc_deallocate_timers(void)
+{
+	int bid;
+
+	for_each_possible_blade(bid) {
+		kfree(blade_info[bid]);
+	}
+	kfree(blade_info);
+}
+
+/* Allocate per-node list of cpu timer expiration times. */
+static __init int uv_rtc_allocate_timers(void)
+{
+	int cpu;
+
+	blade_info = kmalloc(uv_possible_blades * sizeof(void *), GFP_KERNEL);
+	if (!blade_info)
+		return -ENOMEM;
+	memset(blade_info, 0, uv_possible_blades * sizeof(void *));
+
+	for_each_present_cpu(cpu) {
+		int nid = cpu_to_node(cpu);
+		int bid = uv_cpu_to_blade_id(cpu);
+		int bcpu = uv_cpu_hub_info(cpu)->blade_processor_id;
+		struct uv_rtc_timer_head *head = blade_info[bid];
+
+		if (!head) {
+			head = kmalloc_node(sizeof(struct uv_rtc_timer_head) +
+				(uv_blade_nr_possible_cpus(bid) *
+					2 * sizeof(u64)),
+				GFP_KERNEL, nid);
+			if (!head) {
+				uv_rtc_deallocate_timers();
+				return -ENOMEM;
+			}
+			spin_lock_init(&head->lock);
+			head->ncpus = uv_blade_nr_possible_cpus(bid);
+			head->next_cpu = -1;
+			blade_info[bid] = head;
+		}
+
+		head->cpu[bcpu].lcpu = cpu;
+		head->cpu[bcpu].expires = ULLONG_MAX;
+	}
+
+	return 0;
+}
+
+/* Find and set the next expiring timer.  */
+static void uv_rtc_find_next_timer(struct uv_rtc_timer_head *head, int pnode)
+{
+	u64 lowest = ULLONG_MAX;
+	int c, bcpu = -1;
+
+	head->next_cpu = -1;
+	for (c = 0; c < head->ncpus; c++) {
+		u64 exp = head->cpu[c].expires;
+		if (exp < lowest) {
+			bcpu = c;
+			lowest = exp;
+		}
+	}
+	if (bcpu >= 0) {
+		head->next_cpu = bcpu;
+		c = head->cpu[bcpu].lcpu;
+		if (uv_setup_intr(c, lowest))
+			/* If we didn't set it up in time, trigger */
+			uv_rtc_send_IPI(c);
+	} else {
+		uv_write_global_mmr64(pnode, UVH_RTC1_INT_CONFIG,
+			UVH_RTC1_INT_CONFIG_M_MASK);
+	}
+}
+
+/*
+ * Set expiration time for current cpu.
+ *
+ * Returns 1 if we missed the expiration time.
+ */
+static int uv_rtc_set_timer(int cpu, u64 expires)
+{
+	int pnode = uv_cpu_to_pnode(cpu);
+	int bid = uv_cpu_to_blade_id(cpu);
+	struct uv_rtc_timer_head *head = blade_info[bid];
+	int bcpu = uv_cpu_hub_info(cpu)->blade_processor_id;
+	u64 *t = &head->cpu[bcpu].expires;
+	unsigned long flags;
+	int next_cpu;
+
+	spin_lock_irqsave(&head->lock, flags);
+
+	next_cpu = head->next_cpu;
+	*t = expires;
+
+	/* Will this one be next to go off? */
+	if (next_cpu < 0 || bcpu == next_cpu ||
+			expires < head->cpu[next_cpu].expires) {
+		head->next_cpu = bcpu;
+		if (uv_setup_intr(cpu, expires)) {
+			*t = ULLONG_MAX;
+			uv_rtc_find_next_timer(head, pnode);
+			spin_unlock_irqrestore(&head->lock, flags);
+			return -ETIME;
+		}
+	}
+
+	spin_unlock_irqrestore(&head->lock, flags);
+	return 0;
+}
+
+/*
+ * Unset expiration time for current cpu.
+ *
+ * Returns 1 if this timer was pending.
+ */
+static int uv_rtc_unset_timer(int cpu, int force)
+{
+	int pnode = uv_cpu_to_pnode(cpu);
+	int bid = uv_cpu_to_blade_id(cpu);
+	struct uv_rtc_timer_head *head = blade_info[bid];
+	int bcpu = uv_cpu_hub_info(cpu)->blade_processor_id;
+	u64 *t = &head->cpu[bcpu].expires;
+	unsigned long flags;
+	int rc = 0;
+
+	spin_lock_irqsave(&head->lock, flags);
+
+	if ((head->next_cpu == bcpu && uv_read_rtc(NULL) >= *t) || force)
+		rc = 1;
+
+	if (rc) {
+		*t = ULLONG_MAX;
+		/* Was the hardware setup for this timer? */
+		if (head->next_cpu == bcpu)
+			uv_rtc_find_next_timer(head, pnode);
+	}
+
+	spin_unlock_irqrestore(&head->lock, flags);
+
+	return rc;
+}
+
+
+/*
+ * Kernel interface routines.
+ */
+
+/*
+ * Read the RTC.
+ *
+ * Starting with HUB rev 2.0, the UV RTC register is replicated across all
+ * cachelines of it's own page.  This allows faster simultaneous reads
+ * from a given socket.
+ */
+static cycle_t uv_read_rtc(struct clocksource *cs)
+{
+	unsigned long offset;
+
+	if (uv_get_min_hub_revision_id() == 1)
+		offset = 0;
+	else
+		offset = (uv_blade_processor_id() * L1_CACHE_BYTES) % PAGE_SIZE;
+
+	return (cycle_t)uv_read_local_mmr(UVH_RTC | offset);
+}
+
+/*
+ * Program the next event, relative to now
+ */
+static int uv_rtc_next_event(unsigned long delta,
+			     struct clock_event_device *ced)
+{
+	int ced_cpu = cpumask_first(ced->cpumask);
+
+	return uv_rtc_set_timer(ced_cpu, delta + uv_read_rtc(NULL));
+}
+
+/*
+ * Setup the RTC timer in oneshot mode
+ */
+static void uv_rtc_timer_setup(enum clock_event_mode mode,
+			       struct clock_event_device *evt)
+{
+	int ced_cpu = cpumask_first(evt->cpumask);
+
+	switch (mode) {
+	case CLOCK_EVT_MODE_PERIODIC:
+	case CLOCK_EVT_MODE_ONESHOT:
+	case CLOCK_EVT_MODE_RESUME:
+		/* Nothing to do here yet */
+		break;
+	case CLOCK_EVT_MODE_UNUSED:
+	case CLOCK_EVT_MODE_SHUTDOWN:
+		uv_rtc_unset_timer(ced_cpu, 1);
+		break;
+	}
+}
+
+static void uv_rtc_interrupt(void)
+{
+	int cpu = smp_processor_id();
+	struct clock_event_device *ced = &per_cpu(cpu_ced, cpu);
+
+	if (!ced || !ced->event_handler)
+		return;
+
+	if (uv_rtc_unset_timer(cpu, 0) != 1)
+		return;
+
+	ced->event_handler(ced);
+}
+
+static int __init uv_enable_evt_rtc(char *str)
+{
+	uv_rtc_evt_enable = 1;
+
+	return 1;
+}
+__setup("uvrtcevt", uv_enable_evt_rtc);
+
+static __init void uv_rtc_register_clockevents(struct work_struct *dummy)
+{
+	struct clock_event_device *ced = &__get_cpu_var(cpu_ced);
+
+	*ced = clock_event_device_uv;
+	ced->cpumask = cpumask_of(smp_processor_id());
+	clockevents_register_device(ced);
+}
+
+static __init int uv_rtc_setup_clock(void)
+{
+	int rc;
+
+	if (!is_uv_system())
+		return -ENODEV;
+
+	clocksource_uv.mult = clocksource_hz2mult(sn_rtc_cycles_per_second,
+				clocksource_uv.shift);
+
+	/* If single blade, prefer tsc */
+	if (uv_num_possible_blades() == 1)
+		clocksource_uv.rating = 250;
+
+	rc = clocksource_register(&clocksource_uv);
+	if (rc)
+		printk(KERN_INFO "UV RTC clocksource failed rc %d\n", rc);
+	else
+		printk(KERN_INFO "UV RTC clocksource registered freq %lu MHz\n",
+			sn_rtc_cycles_per_second/(unsigned long)1E6);
+
+	if (rc || !uv_rtc_evt_enable || x86_platform_ipi_callback)
+		return rc;
+
+	/* Setup and register clockevents */
+	rc = uv_rtc_allocate_timers();
+	if (rc)
+		goto error;
+
+	x86_platform_ipi_callback = uv_rtc_interrupt;
+
+	clock_event_device_uv.mult = div_sc(sn_rtc_cycles_per_second,
+				NSEC_PER_SEC, clock_event_device_uv.shift);
+
+	clock_event_device_uv.min_delta_ns = NSEC_PER_SEC /
+						sn_rtc_cycles_per_second;
+
+	clock_event_device_uv.max_delta_ns = clocksource_uv.mask *
+				(NSEC_PER_SEC / sn_rtc_cycles_per_second);
+
+	rc = schedule_on_each_cpu(uv_rtc_register_clockevents);
+	if (rc) {
+		x86_platform_ipi_callback = NULL;
+		uv_rtc_deallocate_timers();
+		goto error;
+	}
+
+	printk(KERN_INFO "UV RTC clockevents registered\n");
+
+	return 0;
+
+error:
+	clocksource_unregister(&clocksource_uv);
+	printk(KERN_INFO "UV RTC clockevents failed rc %d\n", rc);
+
+	return rc;
+}
+arch_initcall(uv_rtc_setup_clock);