Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6

18 files changed, 523 insertions, 1499 deletions

diff --git a/arch/ia64/hp/sim/simserial.c b/arch/ia64/hp/sim/simserial.c
index a346e18..27f23fa 100644
--- a/arch/ia64/hp/sim/simserial.c
+++ b/arch/ia64/hp/sim/simserial.c
@@ -167,15 +167,9 @@ static  void receive_chars(struct tty_struct *tty, struct pt_regs *regs)
 			}
 		}
 		seen_esc = 0;
-		if (tty->flip.count >= TTY_FLIPBUF_SIZE) break;
 
-		*tty->flip.char_buf_ptr = ch;
-
-		*tty->flip.flag_buf_ptr = 0;
-
-		tty->flip.flag_buf_ptr++;
-		tty->flip.char_buf_ptr++;
-		tty->flip.count++;
+		if (tty_insert_flip_char(tty, ch, TTY_NORMAL) == 0)
+			break;
 	}
 	tty_flip_buffer_push(tty);
 }
diff --git a/arch/ia64/kernel/fsys.S b/arch/ia64/kernel/fsys.S
index 2ddbac6..ce42391 100644
--- a/arch/ia64/kernel/fsys.S
+++ b/arch/ia64/kernel/fsys.S
@@ -903,5 +903,6 @@ fsyscall_table:
 	data8 0
 	data8 0
 	data8 0
+	data8 0							// 1280
 
 	.org fsyscall_table + 8*NR_syscalls	// guard against failures to increase NR_syscalls
diff --git a/arch/ia64/kernel/jprobes.S b/arch/ia64/kernel/jprobes.S
index 2323377..5cd6226 100644
--- a/arch/ia64/kernel/jprobes.S
+++ b/arch/ia64/kernel/jprobes.S
@@ -60,3 +60,30 @@ END(jprobe_break)
 GLOBAL_ENTRY(jprobe_inst_return)
 	br.call.sptk.many b0=jprobe_break
 END(jprobe_inst_return)
+
+GLOBAL_ENTRY(invalidate_stacked_regs)
+	movl r16=invalidate_restore_cfm
+	;;
+	mov b6=r16
+	;;
+	br.ret.sptk.many b6
+	;;
+invalidate_restore_cfm:
+	mov r16=ar.rsc
+	;;
+	mov ar.rsc=r0
+	;;
+	loadrs
+	;;
+	mov ar.rsc=r16
+	;;
+	br.cond.sptk.many rp
+END(invalidate_stacked_regs)
+
+GLOBAL_ENTRY(flush_register_stack)
+	// flush dirty regs to backing store (must be first in insn group)
+	flushrs
+	;;
+	br.ret.sptk.many rp
+END(flush_register_stack)
+
diff --git a/arch/ia64/kernel/kprobes.c b/arch/ia64/kernel/kprobes.c
index 346fedf..50ae8c7 100644
--- a/arch/ia64/kernel/kprobes.c
+++ b/arch/ia64/kernel/kprobes.c
@@ -766,11 +766,56 @@ int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
 	return ret;
 }
 
+struct param_bsp_cfm {
+	unsigned long ip;
+	unsigned long *bsp;
+	unsigned long cfm;
+};
+
+static void ia64_get_bsp_cfm(struct unw_frame_info *info, void *arg)
+{
+	unsigned long ip;
+	struct param_bsp_cfm *lp = arg;
+
+	do {
+		unw_get_ip(info, &ip);
+		if (ip == 0)
+			break;
+		if (ip == lp->ip) {
+			unw_get_bsp(info, (unsigned long*)&lp->bsp);
+			unw_get_cfm(info, (unsigned long*)&lp->cfm);
+			return;
+		}
+	} while (unw_unwind(info) >= 0);
+	lp->bsp = 0;
+	lp->cfm = 0;
+	return;
+}
+
 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
 {
 	struct jprobe *jp = container_of(p, struct jprobe, kp);
 	unsigned long addr = ((struct fnptr *)(jp->entry))->ip;
 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
+	struct param_bsp_cfm pa;
+	int bytes;
+
+	/*
+	 * Callee owns the argument space and could overwrite it, eg
+	 * tail call optimization. So to be absolutely safe
+	 * we save the argument space before transfering the control
+	 * to instrumented jprobe function which runs in
+	 * the process context
+	 */
+	pa.ip = regs->cr_iip;
+	unw_init_running(ia64_get_bsp_cfm, &pa);
+	bytes = (char *)ia64_rse_skip_regs(pa.bsp, pa.cfm & 0x3f)
+				- (char *)pa.bsp;
+	memcpy( kcb->jprobes_saved_stacked_regs,
+		pa.bsp,
+		bytes );
+	kcb->bsp = pa.bsp;
+	kcb->cfm = pa.cfm;
 
 	/* save architectural state */
 	kcb->jprobe_saved_regs = *regs;
@@ -792,8 +837,20 @@ int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
 {
 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
+	int bytes;
 
+	/* restoring architectural state */
 	*regs = kcb->jprobe_saved_regs;
+
+	/* restoring the original argument space */
+	flush_register_stack();
+	bytes = (char *)ia64_rse_skip_regs(kcb->bsp, kcb->cfm & 0x3f)
+				- (char *)kcb->bsp;
+	memcpy( kcb->bsp,
+		kcb->jprobes_saved_stacked_regs,
+		bytes );
+	invalidate_stacked_regs();
+
 	preempt_enable_no_resched();
 	return 1;
 }
diff --git a/arch/ia64/kernel/mca_asm.S b/arch/ia64/kernel/mca_asm.S
index db32fc1..403a80a 100644
--- a/arch/ia64/kernel/mca_asm.S
+++ b/arch/ia64/kernel/mca_asm.S
@@ -847,7 +847,7 @@ ia64_state_restore:
 	;;
 	mov cr.iim=temp3
 	mov cr.iha=temp4
-	dep r22=0,r22,62,2	// pal_min_state, physical, uncached
+	dep r22=0,r22,62,1	// pal_min_state, physical, uncached
 	mov IA64_KR(CURRENT)=r21
 	ld8 r8=[temp1]		// os_status
 	ld8 r10=[temp2]		// context
diff --git a/arch/ia64/kernel/salinfo.c b/arch/ia64/kernel/salinfo.c
index a87a162..9d5a823 100644
--- a/arch/ia64/kernel/salinfo.c
+++ b/arch/ia64/kernel/salinfo.c
@@ -3,7 +3,7 @@
  *
  * Creates entries in /proc/sal for various system features.
  *
- * Copyright (c) 2003 Silicon Graphics, Inc.  All rights reserved.
+ * Copyright (c) 2003, 2006 Silicon Graphics, Inc.  All rights reserved.
  * Copyright (c) 2003 Hewlett-Packard Co
  *	Bjorn Helgaas <bjorn.helgaas@hp.com>
  *
@@ -27,9 +27,17 @@
  *   mca.c may not pass a buffer, a NULL buffer just indicates that a new
  *   record is available in SAL.
  *   Replace some NR_CPUS by cpus_online, for hotplug cpu.
+ *
+ * Jan  5 2006        kaos@sgi.com
+ *   Handle hotplug cpus coming online.
+ *   Handle hotplug cpus going offline while they still have outstanding records.
+ *   Use the cpu_* macros consistently.
+ *   Replace the counting semaphore with a mutex and a test if the cpumask is non-empty.
+ *   Modify the locking to make the test for "work to do" an atomic operation.
  */
 
 #include <linux/capability.h>
+#include <linux/cpu.h>
 #include <linux/types.h>
 #include <linux/proc_fs.h>
 #include <linux/module.h>
@@ -132,8 +140,8 @@ enum salinfo_state {
 };
 
 struct salinfo_data {
-	volatile cpumask_t	cpu_event;	/* which cpus have outstanding events */
-	struct semaphore	sem;		/* count of cpus with outstanding events (bits set in cpu_event) */
+	cpumask_t		cpu_event;	/* which cpus have outstanding events */
+	struct semaphore	mutex;
 	u8			*log_buffer;
 	u64			log_size;
 	u8			*oemdata;	/* decoded oem data */
@@ -174,6 +182,21 @@ struct salinfo_platform_oemdata_parms {
 	int ret;
 };
 
+/* Kick the mutex that tells user space that there is work to do.  Instead of
+ * trying to track the state of the mutex across multiple cpus, in user
+ * context, interrupt context, non-maskable interrupt context and hotplug cpu,
+ * it is far easier just to grab the mutex if it is free then release it.
+ *
+ * This routine must be called with data_saved_lock held, to make the down/up
+ * operation atomic.
+ */
+static void
+salinfo_work_to_do(struct salinfo_data *data)
+{
+	down_trylock(&data->mutex);
+	up(&data->mutex);
+}
+
 static void
 salinfo_platform_oemdata_cpu(void *context)
 {
@@ -212,9 +235,9 @@ salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
 
 	BUG_ON(type >= ARRAY_SIZE(salinfo_log_name));
 
+	if (irqsafe)
+		spin_lock_irqsave(&data_saved_lock, flags);
 	if (buffer) {
-		if (irqsafe)
-			spin_lock_irqsave(&data_saved_lock, flags);
 		for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
 			if (!data_saved->buffer)
 				break;
@@ -232,13 +255,11 @@ salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
 			data_saved->size = size;
 			data_saved->buffer = buffer;
 		}
-		if (irqsafe)
-			spin_unlock_irqrestore(&data_saved_lock, flags);
 	}
-
-	if (!test_and_set_bit(smp_processor_id(), &data->cpu_event)) {
-		if (irqsafe)
-			up(&data->sem);
+	cpu_set(smp_processor_id(), data->cpu_event);
+	if (irqsafe) {
+		salinfo_work_to_do(data);
+		spin_unlock_irqrestore(&data_saved_lock, flags);
 	}
 }
 
@@ -249,20 +270,17 @@ static struct timer_list salinfo_timer;
 static void
 salinfo_timeout_check(struct salinfo_data *data)
 {
-	int i;
+	unsigned long flags;
 	if (!data->open)
 		return;
-	for_each_online_cpu(i) {
-		if (test_bit(i, &data->cpu_event)) {
-			/* double up() is not a problem, user space will see no
-			 * records for the additional "events".
-			 */
-			up(&data->sem);
-		}
+	if (!cpus_empty(data->cpu_event)) {
+		spin_lock_irqsave(&data_saved_lock, flags);
+		salinfo_work_to_do(data);
+		spin_unlock_irqrestore(&data_saved_lock, flags);
 	}
 }
 
-static void 
+static void
 salinfo_timeout (unsigned long arg)
 {
 	salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_MCA);
@@ -290,16 +308,20 @@ salinfo_event_read(struct file *file, char __user *buffer, size_t count, loff_t
 	int i, n, cpu = -1;
 
 retry:
-	if (down_trylock(&data->sem)) {
+	if (cpus_empty(data->cpu_event) && down_trylock(&data->mutex)) {
 		if (file->f_flags & O_NONBLOCK)
 			return -EAGAIN;
-		if (down_interruptible(&data->sem))
+		if (down_interruptible(&data->mutex))
 			return -EINTR;
 	}
 
 	n = data->cpu_check;
 	for (i = 0; i < NR_CPUS; i++) {
-		if (test_bit(n, &data->cpu_event) && cpu_online(n)) {
+		if (cpu_isset(n, data->cpu_event)) {
+			if (!cpu_online(n)) {
+				cpu_clear(n, data->cpu_event);
+				continue;
+			}
 			cpu = n;
 			break;
 		}
@@ -310,9 +332,6 @@ retry:
 	if (cpu == -1)
 		goto retry;
 
-	/* events are sticky until the user says "clear" */
-	up(&data->sem);
-
 	/* for next read, start checking at next CPU */
 	data->cpu_check = cpu;
 	if (++data->cpu_check == NR_CPUS)
@@ -381,10 +400,8 @@ salinfo_log_release(struct inode *inode, struct file *file)
 static void
 call_on_cpu(int cpu, void (*fn)(void *), void *arg)
 {
-	cpumask_t save_cpus_allowed, new_cpus_allowed;
-	memcpy(&save_cpus_allowed, &current->cpus_allowed, sizeof(save_cpus_allowed));
-	memset(&new_cpus_allowed, 0, sizeof(new_cpus_allowed));
-	set_bit(cpu, &new_cpus_allowed);
+	cpumask_t save_cpus_allowed = current->cpus_allowed;
+	cpumask_t new_cpus_allowed = cpumask_of_cpu(cpu);
 	set_cpus_allowed(current, new_cpus_allowed);
 	(*fn)(arg);
 	set_cpus_allowed(current, save_cpus_allowed);
@@ -433,10 +450,10 @@ retry:
 	if (!data->saved_num)
 		call_on_cpu(cpu, salinfo_log_read_cpu, data);
 	if (!data->log_size) {
-	        data->state = STATE_NO_DATA;
-	        clear_bit(cpu, &data->cpu_event);
+		data->state = STATE_NO_DATA;
+		cpu_clear(cpu, data->cpu_event);
 	} else {
-	        data->state = STATE_LOG_RECORD;
+		data->state = STATE_LOG_RECORD;
 	}
 }
 
@@ -473,27 +490,31 @@ static int
 salinfo_log_clear(struct salinfo_data *data, int cpu)
 {
 	sal_log_record_header_t *rh;
+	unsigned long flags;
+	spin_lock_irqsave(&data_saved_lock, flags);
 	data->state = STATE_NO_DATA;
-	if (!test_bit(cpu, &data->cpu_event))
+	if (!cpu_isset(cpu, data->cpu_event)) {
+		spin_unlock_irqrestore(&data_saved_lock, flags);
 		return 0;
-	down(&data->sem);
-	clear_bit(cpu, &data->cpu_event);
+	}
+	cpu_clear(cpu, data->cpu_event);
 	if (data->saved_num) {
-		unsigned long flags;
-		spin_lock_irqsave(&data_saved_lock, flags);
-		shift1_data_saved(data, data->saved_num - 1 );
+		shift1_data_saved(data, data->saved_num - 1);
 		data->saved_num = 0;
-		spin_unlock_irqrestore(&data_saved_lock, flags);
 	}
+	spin_unlock_irqrestore(&data_saved_lock, flags);
 	rh = (sal_log_record_header_t *)(data->log_buffer);
 	/* Corrected errors have already been cleared from SAL */
 	if (rh->severity != sal_log_severity_corrected)
 		call_on_cpu(cpu, salinfo_log_clear_cpu, data);
 	/* clearing a record may make a new record visible */
 	salinfo_log_new_read(cpu, data);
-	if (data->state == STATE_LOG_RECORD &&
-	    !test_and_set_bit(cpu,  &data->cpu_event))
-		up(&data->sem);
+	if (data->state == STATE_LOG_RECORD) {
+		spin_lock_irqsave(&data_saved_lock, flags);
+		cpu_set(cpu, data->cpu_event);
+		salinfo_work_to_do(data);
+		spin_unlock_irqrestore(&data_saved_lock, flags);
+	}
 	return 0;
 }
 
@@ -550,6 +571,53 @@ static struct file_operations salinfo_data_fops = {
 	.write   = salinfo_log_write,
 };
 
+#ifdef	CONFIG_HOTPLUG_CPU
+static int __devinit
+salinfo_cpu_callback(struct notifier_block *nb, unsigned long action, void *hcpu)
+{
+	unsigned int i, cpu = (unsigned long)hcpu;
+	unsigned long flags;
+	struct salinfo_data *data;
+	switch (action) {
+	case CPU_ONLINE:
+		spin_lock_irqsave(&data_saved_lock, flags);
+		for (i = 0, data = salinfo_data;
+		     i < ARRAY_SIZE(salinfo_data);
+		     ++i, ++data) {
+			cpu_set(cpu, data->cpu_event);
+			salinfo_work_to_do(data);
+		}
+		spin_unlock_irqrestore(&data_saved_lock, flags);
+		break;
+	case CPU_DEAD:
+		spin_lock_irqsave(&data_saved_lock, flags);
+		for (i = 0, data = salinfo_data;
+		     i < ARRAY_SIZE(salinfo_data);
+		     ++i, ++data) {
+			struct salinfo_data_saved *data_saved;
+			int j;
+			for (j = ARRAY_SIZE(data->data_saved) - 1, data_saved = data->data_saved + j;
+			     j >= 0;
+			     --j, --data_saved) {
+				if (data_saved->buffer && data_saved->cpu == cpu) {
+					shift1_data_saved(data, j);
+				}
+			}
+			cpu_clear(cpu, data->cpu_event);
+		}
+		spin_unlock_irqrestore(&data_saved_lock, flags);
+		break;
+	}
+	return NOTIFY_OK;
+}
+
+static struct notifier_block salinfo_cpu_notifier =
+{
+	.notifier_call = salinfo_cpu_callback,
+	.priority = 0,
+};
+#endif	/* CONFIG_HOTPLUG_CPU */
+
 static int __init
 salinfo_init(void)
 {
@@ -557,7 +625,7 @@ salinfo_init(void)
 	struct proc_dir_entry **sdir = salinfo_proc_entries; /* keeps track of every entry */
 	struct proc_dir_entry *dir, *entry;
 	struct salinfo_data *data;
-	int i, j, online;
+	int i, j;
 
 	salinfo_dir = proc_mkdir("sal", NULL);
 	if (!salinfo_dir)
@@ -572,7 +640,7 @@ salinfo_init(void)
 	for (i = 0; i < ARRAY_SIZE(salinfo_log_name); i++) {
 		data = salinfo_data + i;
 		data->type = i;
-		sema_init(&data->sem, 0);
+		init_MUTEX(&data->mutex);
 		dir = proc_mkdir(salinfo_log_name[i], salinfo_dir);
 		if (!dir)
 			continue;
@@ -592,12 +660,8 @@ salinfo_init(void)
 		*sdir++ = entry;
 
 		/* we missed any events before now */
-		online = 0;
-		for_each_online_cpu(j) {
-			set_bit(j, &data->cpu_event);
-			++online;
-		}
-		sema_init(&data->sem, online);
+		for_each_online_cpu(j)
+			cpu_set(j, data->cpu_event);
 
 		*sdir++ = dir;
 	}
@@ -609,6 +673,10 @@ salinfo_init(void)
 	salinfo_timer.function = &salinfo_timeout;
 	add_timer(&salinfo_timer);
 
+#ifdef	CONFIG_HOTPLUG_CPU
+	register_cpu_notifier(&salinfo_cpu_notifier);
+#endif
+
 	return 0;
 }
 
diff --git a/arch/ia64/kernel/traps.c b/arch/ia64/kernel/traps.c
index d3e0ecb..5539190 100644
--- a/arch/ia64/kernel/traps.c
+++ b/arch/ia64/kernel/traps.c
@@ -530,12 +530,15 @@ ia64_fault (unsigned long vector, unsigned long isr, unsigned long ifa,
 		if (fsys_mode(current, &regs)) {
 			extern char __kernel_syscall_via_break[];
 			/*
-			 * Got a trap in fsys-mode: Taken Branch Trap and Single Step trap
-			 * need special handling; Debug trap is not supposed to happen.
+			 * Got a trap in fsys-mode: Taken Branch Trap
+			 * and Single Step trap need special handling;
+			 * Debug trap is ignored (we disable it here
+			 * and re-enable it in the lower-privilege trap).
 			 */
 			if (unlikely(vector == 29)) {
-				die("Got debug trap in fsys-mode---not supposed to happen!",
-				    &regs, 0);
+				set_thread_flag(TIF_DB_DISABLED);
+				ia64_psr(&regs)->db = 0;
+				ia64_psr(&regs)->lp = 1;
 				return;
 			}
 			/* re-do the system call via break 0x100000: */
@@ -589,10 +592,19 @@ ia64_fault (unsigned long vector, unsigned long isr, unsigned long ifa,
 	      case 34:
 		if (isr & 0x2) {
 			/* Lower-Privilege Transfer Trap */
+
+			/* If we disabled debug traps during an fsyscall,
+			 * re-enable them here.
+			 */
+			if (test_thread_flag(TIF_DB_DISABLED)) {
+				clear_thread_flag(TIF_DB_DISABLED);
+				ia64_psr(&regs)->db = 1;
+			}
+
 			/*
-			 * Just clear PSR.lp and then return immediately: all the
-			 * interesting work (e.g., signal delivery is done in the kernel
-			 * exit path).
+			 * Just clear PSR.lp and then return immediately:
+			 * all the interesting work (e.g., signal delivery)
+			 * is done in the kernel exit path.
 			 */
 			ia64_psr(&regs)->lp = 0;
 			return;
diff --git a/arch/ia64/mm/tlb.c b/arch/ia64/mm/tlb.c
index 41105d4..6a4eec9 100644
--- a/arch/ia64/mm/tlb.c
+++ b/arch/ia64/mm/tlb.c
@@ -90,7 +90,7 @@ ia64_global_tlb_purge (struct mm_struct *mm, unsigned long start,
 {
 	static DEFINE_SPINLOCK(ptcg_lock);
 
-	if (mm != current->active_mm) {
+	if (mm != current->active_mm || !current->mm) {
 		flush_tlb_all();
 		return;
 	}
diff --git a/arch/ia64/sn/include/xtalk/hubdev.h b/arch/ia64/sn/include/xtalk/hubdev.h
index 71c2b27..4d417c3 100644
--- a/arch/ia64/sn/include/xtalk/hubdev.h
+++ b/arch/ia64/sn/include/xtalk/hubdev.h
@@ -26,11 +26,14 @@
 #define IIO_NUM_ITTES   7
 #define HUB_NUM_BIG_WINDOW      (IIO_NUM_ITTES - 1)
 
-struct sn_flush_device_list {
+/* This struct is shared between the PROM and the kernel.
+ * Changes to this struct will require corresponding changes to the kernel.
+ */
+struct sn_flush_device_common {
 	int sfdl_bus;
 	int sfdl_slot;
 	int sfdl_pin;
-	struct bar_list {
+	struct common_bar_list {
 		unsigned long start;
 		unsigned long end;
 	} sfdl_bar_list[6];
@@ -40,14 +43,19 @@ struct sn_flush_device_list {
 	uint32_t sfdl_persistent_busnum;
 	uint32_t sfdl_persistent_segment;
 	struct pcibus_info *sfdl_pcibus_info;
+};
+
+/* This struct is kernel only and is not used by the PROM */
+struct sn_flush_device_kernel {
 	spinlock_t sfdl_flush_lock;
+	struct sn_flush_device_common *common;
 };
 
 /*
- * **widget_p - Used as an array[wid_num][device] of sn_flush_device_list.
+ * **widget_p - Used as an array[wid_num][device] of sn_flush_device_kernel.
  */
 struct sn_flush_nasid_entry  {
-	struct sn_flush_device_list **widget_p; /* Used as a array of wid_num */
+	struct sn_flush_device_kernel **widget_p; // Used as an array of wid_num
 	uint64_t iio_itte[8];
 };
 
diff --git a/arch/ia64/sn/kernel/bte_error.c b/arch/ia64/sn/kernel/bte_error.c
index fcbc748..f1ec137 100644
--- a/arch/ia64/sn/kernel/bte_error.c
+++ b/arch/ia64/sn/kernel/bte_error.c
@@ -33,7 +33,7 @@ void bte_error_handler(unsigned long);
  * Wait until all BTE related CRBs are completed
  * and then reset the interfaces.
  */
-void shub1_bte_error_handler(unsigned long _nodepda)
+int shub1_bte_error_handler(unsigned long _nodepda)
 {
 	struct nodepda_s *err_nodepda = (struct nodepda_s *)_nodepda;
 	struct timer_list *recovery_timer = &err_nodepda->bte_recovery_timer;
@@ -53,7 +53,7 @@ void shub1_bte_error_handler(unsigned long _nodepda)
 	    (err_nodepda->bte_if[1].bh_error == BTE_SUCCESS)) {
 		BTE_PRINTK(("eh:%p:%d Nothing to do.\n", err_nodepda,
 			    smp_processor_id()));
-		return;
+		return 1;
 	}
 
 	/* Determine information about our hub */
@@ -81,7 +81,7 @@ void shub1_bte_error_handler(unsigned long _nodepda)
 		mod_timer(recovery_timer, HZ * 5);
 		BTE_PRINTK(("eh:%p:%d Marked Giving up\n", err_nodepda,
 			    smp_processor_id()));
-		return;
+		return 1;
 	}
 	if (icmr.ii_icmr_fld_s.i_crb_vld != 0) {
 
@@ -99,7 +99,7 @@ void shub1_bte_error_handler(unsigned long _nodepda)
 				BTE_PRINTK(("eh:%p:%d Valid %d, Giving up\n",
 					    err_nodepda, smp_processor_id(),
 					    i));
-				return;
+				return 1;
 			}
 		}
 	}
@@ -124,6 +124,42 @@ void shub1_bte_error_handler(unsigned long _nodepda)
 	REMOTE_HUB_S(nasid, IIO_IBCR, ibcr.ii_ibcr_regval);
 
 	del_timer(recovery_timer);
+	return 0;
+}
+
+/*
+ * Wait until all BTE related CRBs are completed
+ * and then reset the interfaces.
+ */
+int shub2_bte_error_handler(unsigned long _nodepda)
+{
+	struct nodepda_s *err_nodepda = (struct nodepda_s *)_nodepda;
+	struct timer_list *recovery_timer = &err_nodepda->bte_recovery_timer;
+	struct bteinfo_s *bte;
+	nasid_t nasid;
+	u64 status;
+	int i;
+
+	nasid = cnodeid_to_nasid(err_nodepda->bte_if[0].bte_cnode);
+
+	/*
+	 * Verify that all the BTEs are complete
+	 */
+	for (i = 0; i < BTES_PER_NODE; i++) {
+		bte = &err_nodepda->bte_if[i];
+		status = BTE_LNSTAT_LOAD(bte);
+		if ((status & IBLS_ERROR) || !(status & IBLS_BUSY))
+			continue;
+		mod_timer(recovery_timer, HZ * 5);
+		BTE_PRINTK(("eh:%p:%d Marked Giving up\n", err_nodepda,
+			    smp_processor_id()));
+		return 1;
+	}
+	if (ia64_sn_bte_recovery(nasid))
+		panic("bte_error_handler(): Fatal BTE Error");
+
+	del_timer(recovery_timer);
+	return 0;
 }
 
 /*
@@ -135,7 +171,6 @@ void bte_error_handler(unsigned long _nodepda)
 	struct nodepda_s *err_nodepda = (struct nodepda_s *)_nodepda;
 	spinlock_t *recovery_lock = &err_nodepda->bte_recovery_lock;
 	int i;
-	nasid_t nasid;
 	unsigned long irq_flags;
 	volatile u64 *notify;
 	bte_result_t bh_error;
@@ -160,12 +195,15 @@ void bte_error_handler(unsigned long _nodepda)
 	}
 
 	if (is_shub1()) {
-		shub1_bte_error_handler(_nodepda);
+		if (shub1_bte_error_handler(_nodepda)) {
+			spin_unlock_irqrestore(recovery_lock, irq_flags);
+			return;
+		}
 	} else {
-		nasid = cnodeid_to_nasid(err_nodepda->bte_if[0].bte_cnode);
-
-		if (ia64_sn_bte_recovery(nasid))
-			panic("bte_error_handler(): Fatal BTE Error");
+		if (shub2_bte_error_handler(_nodepda)) {
+			spin_unlock_irqrestore(recovery_lock, irq_flags);
+			return;
+		}
 	}
 
 	for (i = 0; i < BTES_PER_NODE; i++) {
diff --git a/arch/ia64/sn/kernel/huberror.c b/arch/ia64/sn/kernel/huberror.c
index 5c5eb01..56ab6ba 100644
--- a/arch/ia64/sn/kernel/huberror.c
+++ b/arch/ia64/sn/kernel/huberror.c
@@ -32,13 +32,14 @@ static irqreturn_t hub_eint_handler(int irq, void *arg, struct pt_regs *ep)
 	ret_stuff.v0 = 0;
 	hubdev_info = (struct hubdev_info *)arg;
 	nasid = hubdev_info->hdi_nasid;
-	SAL_CALL_NOLOCK(ret_stuff, SN_SAL_HUB_ERROR_INTERRUPT,
+
+	if (is_shub1()) {
+		SAL_CALL_NOLOCK(ret_stuff, SN_SAL_HUB_ERROR_INTERRUPT,
 			(u64) nasid, 0, 0, 0, 0, 0, 0);
 
-	if ((int)ret_stuff.v0)
-		panic("hubii_eint_handler(): Fatal TIO Error");
+		if ((int)ret_stuff.v0)
+			panic("hubii_eint_handler(): Fatal TIO Error");
 
-	if (is_shub1()) {
 		if (!(nasid & 1)) /* Not a TIO, handle CRB errors */
 			(void)hubiio_crb_error_handler(hubdev_info);
 	} else 
diff --git a/arch/ia64/sn/kernel/io_init.c b/arch/ia64/sn/kernel/io_init.c
index 318087e..258d9d7 100644
--- a/arch/ia64/sn/kernel/io_init.c
+++ b/arch/ia64/sn/kernel/io_init.c
@@ -76,11 +76,12 @@ static struct sn_pcibus_provider sn_pci_default_provider = {
 };
 
 /*
- * Retrieve the DMA Flush List given nasid.  This list is needed 
- * to implement the WAR - Flush DMA data on PIO Reads.
+ * Retrieve the DMA Flush List given nasid, widget, and device.
+ * This list is needed to implement the WAR - Flush DMA data on PIO Reads.
  */
-static inline uint64_t
-sal_get_widget_dmaflush_list(u64 nasid, u64 widget_num, u64 address)
+static inline u64
+sal_get_device_dmaflush_list(u64 nasid, u64 widget_num, u64 device_num,
+			     u64 address)
 {
 
 	struct ia64_sal_retval ret_stuff;
@@ -88,17 +89,17 @@ sal_get_widget_dmaflush_list(u64 nasid, u64 widget_num, u64 address)
 	ret_stuff.v0 = 0;
 
 	SAL_CALL_NOLOCK(ret_stuff,
-			(u64) SN_SAL_IOIF_GET_WIDGET_DMAFLUSH_LIST,
-			(u64) nasid, (u64) widget_num, (u64) address, 0, 0, 0,
-			0);
-	return ret_stuff.v0;
+			(u64) SN_SAL_IOIF_GET_DEVICE_DMAFLUSH_LIST,
+			(u64) nasid, (u64) widget_num,
+			(u64) device_num, (u64) address, 0, 0, 0);
+	return ret_stuff.status;
 
 }
 
 /*
  * Retrieve the hub device info structure for the given nasid.
  */
-static inline uint64_t sal_get_hubdev_info(u64 handle, u64 address)
+static inline u64 sal_get_hubdev_info(u64 handle, u64 address)
 {
 
 	struct ia64_sal_retval ret_stuff;
@@ -114,7 +115,7 @@ static inline uint64_t sal_get_hubdev_info(u64 handle, u64 address)
 /*
  * Retrieve the pci bus information given the bus number.
  */
-static inline uint64_t sal_get_pcibus_info(u64 segment, u64 busnum, u64 address)
+static inline u64 sal_get_pcibus_info(u64 segment, u64 busnum, u64 address)
 {
 
 	struct ia64_sal_retval ret_stuff;
@@ -130,7 +131,7 @@ static inline uint64_t sal_get_pcibus_info(u64 segment, u64 busnum, u64 address)
 /*
  * Retrieve the pci device information given the bus and device|function number.
  */
-static inline uint64_t
+static inline u64
 sal_get_pcidev_info(u64 segment, u64 bus_number, u64 devfn, u64 pci_dev, 
 			u64 sn_irq_info)
 {
@@ -170,12 +171,12 @@ sn_pcidev_info_get(struct pci_dev *dev)
  */
 static void sn_fixup_ionodes(void)
 {
-
-	struct sn_flush_device_list *sn_flush_device_list;
+	struct sn_flush_device_kernel *sn_flush_device_kernel;
+	struct sn_flush_device_kernel *dev_entry;
 	struct hubdev_info *hubdev;
-	uint64_t status;
-	uint64_t nasid;
-	int i, widget;
+	u64 status;
+	u64 nasid;
+	int i, widget, device;
 
 	/*
 	 * Get SGI Specific HUB chipset information.
@@ -186,7 +187,7 @@ static void sn_fixup_ionodes(void)
 		nasid = cnodeid_to_nasid(i);
 		hubdev->max_segment_number = 0xffffffff;
 		hubdev->max_pcibus_number = 0xff;
-		status = sal_get_hubdev_info(nasid, (uint64_t) __pa(hubdev));
+		status = sal_get_hubdev_info(nasid, (u64) __pa(hubdev));
 		if (status)
 			continue;
 
@@ -213,38 +214,49 @@ static void sn_fixup_ionodes(void)
 
 		hubdev->hdi_flush_nasid_list.widget_p =
 		    kmalloc((HUB_WIDGET_ID_MAX + 1) *
-			    sizeof(struct sn_flush_device_list *), GFP_KERNEL);
-
+			    sizeof(struct sn_flush_device_kernel *),
+			    GFP_KERNEL);
 		memset(hubdev->hdi_flush_nasid_list.widget_p, 0x0,
 		       (HUB_WIDGET_ID_MAX + 1) *
-		       sizeof(struct sn_flush_device_list *));
+		       sizeof(struct sn_flush_device_kernel *));
 
 		for (widget = 0; widget <= HUB_WIDGET_ID_MAX; widget++) {
-			sn_flush_device_list = kmalloc(DEV_PER_WIDGET *
-						       sizeof(struct
-							      sn_flush_device_list),
-						       GFP_KERNEL);
-			memset(sn_flush_device_list, 0x0,
+			sn_flush_device_kernel = kmalloc(DEV_PER_WIDGET *
+						         sizeof(struct
+						        sn_flush_device_kernel),
+						        GFP_KERNEL);
+			if (!sn_flush_device_kernel)
+				BUG();
+			memset(sn_flush_device_kernel, 0x0,
 			       DEV_PER_WIDGET *
-			       sizeof(struct sn_flush_device_list));
-
-			status =
-			    sal_get_widget_dmaflush_list(nasid, widget,
-							 (uint64_t)
-							 __pa
-							 (sn_flush_device_list));
-			if (status) {
-				kfree(sn_flush_device_list);
-				continue;
+			       sizeof(struct sn_flush_device_kernel));
+
+			dev_entry = sn_flush_device_kernel;
+			for (device = 0; device < DEV_PER_WIDGET;
+			     device++,dev_entry++) {
+				dev_entry->common = kmalloc(sizeof(struct
+					      	        sn_flush_device_common),
+					                    GFP_KERNEL);
+				if (!dev_entry->common)
+					BUG();
+				memset(dev_entry->common, 0x0, sizeof(struct
+					     	       sn_flush_device_common));
+
+				status = sal_get_device_dmaflush_list(nasid,
+									widget,
+								       	device,
+						      (u64)(dev_entry->common));
+				if (status)
+					BUG();
+
+				spin_lock_init(&dev_entry->sfdl_flush_lock);
 			}
 
-			spin_lock_init(&sn_flush_device_list->sfdl_flush_lock);
-			hubdev->hdi_flush_nasid_list.widget_p[widget] =
-			    sn_flush_device_list;
-		}
-
+			if (sn_flush_device_kernel)
+				hubdev->hdi_flush_nasid_list.widget_p[widget] =
+						       sn_flush_device_kernel;
+	        }
 	}
-
 }
 
 /*
diff --git a/arch/ia64/sn/kernel/xpc.h b/arch/ia64/sn/kernel/xpc.h
deleted file mode 100644
index 5483a9f..0000000
--- a/arch/ia64/sn/kernel/xpc.h
+++ /dev/null
@@ -1,1273 +0,0 @@
-/*
- * 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) 2004-2005 Silicon Graphics, Inc.  All Rights Reserved.
- */
-
-
-/*
- * Cross Partition Communication (XPC) structures and macros.
- */
-
-#ifndef _IA64_SN_KERNEL_XPC_H
-#define _IA64_SN_KERNEL_XPC_H
-
-
-#include <linux/config.h>
-#include <linux/interrupt.h>
-#include <linux/sysctl.h>
-#include <linux/device.h>
-#include <asm/pgtable.h>
-#include <asm/processor.h>
-#include <asm/sn/bte.h>
-#include <asm/sn/clksupport.h>
-#include <asm/sn/addrs.h>
-#include <asm/sn/mspec.h>
-#include <asm/sn/shub_mmr.h>
-#include <asm/sn/xp.h>
-
-
-/*
- * XPC Version numbers consist of a major and minor number. XPC can always
- * talk to versions with same major #, and never talk to versions with a
- * different major #.
- */
-#define _XPC_VERSION(_maj, _min)	(((_maj) << 4) | ((_min) & 0xf))
-#define XPC_VERSION_MAJOR(_v)		((_v) >> 4)
-#define XPC_VERSION_MINOR(_v)		((_v) & 0xf)
-
-
-/*
- * The next macros define word or bit representations for given
- * C-brick nasid in either the SAL provided bit array representing
- * nasids in the partition/machine or the AMO_t array used for
- * inter-partition initiation communications.
- *
- * For SN2 machines, C-Bricks are alway even numbered NASIDs.  As
- * such, some space will be saved by insisting that nasid information
- * passed from SAL always be packed for C-Bricks and the
- * cross-partition interrupts use the same packing scheme.
- */
-#define XPC_NASID_W_INDEX(_n)	(((_n) / 64) / 2)
-#define XPC_NASID_B_INDEX(_n)	(((_n) / 2) & (64 - 1))
-#define XPC_NASID_IN_ARRAY(_n, _p) ((_p)[XPC_NASID_W_INDEX(_n)] & \
-				    (1UL << XPC_NASID_B_INDEX(_n)))
-#define XPC_NASID_FROM_W_B(_w, _b) (((_w) * 64 + (_b)) * 2)
-
-#define XPC_HB_DEFAULT_INTERVAL		5	/* incr HB every x secs */
-#define XPC_HB_CHECK_DEFAULT_INTERVAL	20	/* check HB every x secs */
-
-/* define the process name of HB checker and the CPU it is pinned to */
-#define XPC_HB_CHECK_THREAD_NAME	"xpc_hb"
-#define XPC_HB_CHECK_CPU		0
-
-/* define the process name of the discovery thread */
-#define XPC_DISCOVERY_THREAD_NAME	"xpc_discovery"
-
-
-/*
- * the reserved page
- *
- *   SAL reserves one page of memory per partition for XPC. Though a full page
- *   in length (16384 bytes), its starting address is not page aligned, but it
- *   is cacheline aligned. The reserved page consists of the following:
- *
- *   reserved page header
- *
- *     The first cacheline of the reserved page contains the header
- *     (struct xpc_rsvd_page). Before SAL initialization has completed,
- *     SAL has set up the following fields of the reserved page header:
- *     SAL_signature, SAL_version, partid, and nasids_size. The other
- *     fields are set up by XPC. (xpc_rsvd_page points to the local
- *     partition's reserved page.)
- *
- *   part_nasids mask
- *   mach_nasids mask
- *
- *     SAL also sets up two bitmaps (or masks), one that reflects the actual
- *     nasids in this partition (part_nasids), and the other that reflects
- *     the actual nasids in the entire machine (mach_nasids). We're only
- *     interested in the even numbered nasids (which contain the processors
- *     and/or memory), so we only need half as many bits to represent the
- *     nasids. The part_nasids mask is located starting at the first cacheline
- *     following the reserved page header. The mach_nasids mask follows right
- *     after the part_nasids mask. The size in bytes of each mask is reflected
- *     by the reserved page header field 'nasids_size'. (Local partition's
- *     mask pointers are xpc_part_nasids and xpc_mach_nasids.)
- *
- *   vars
- *   vars part
- *
- *     Immediately following the mach_nasids mask are the XPC variables
- *     required by other partitions. First are those that are generic to all
- *     partitions (vars), followed on the next available cacheline by those
- *     which are partition specific (vars part). These are setup by XPC.
- *     (Local partition's vars pointers are xpc_vars and xpc_vars_part.)
- *
- * Note: Until vars_pa is set, the partition XPC code has not been initialized.
- */
-struct xpc_rsvd_page {
-	u64 SAL_signature;	/* SAL: unique signature */
-	u64 SAL_version;	/* SAL: version */
-	u8 partid;		/* SAL: partition ID */
-	u8 version;
-	u8 pad1[6];		/* align to next u64 in cacheline */
-	volatile u64 vars_pa;
-	struct timespec stamp;	/* time when reserved page was setup by XPC */
-	u64 pad2[9];		/* align to last u64 in cacheline */
-	u64 nasids_size;	/* SAL: size of each nasid mask in bytes */
-};
-
-#define XPC_RP_VERSION _XPC_VERSION(1,1) /* version 1.1 of the reserved page */
-
-#define XPC_SUPPORTS_RP_STAMP(_version) \
-			(_version >= _XPC_VERSION(1,1))
-
-/*
- * compare stamps - the return value is:
- *
- *	< 0,	if stamp1 < stamp2
- *	= 0,	if stamp1 == stamp2
- *	> 0,	if stamp1 > stamp2
- */
-static inline int
-xpc_compare_stamps(struct timespec *stamp1, struct timespec *stamp2)
-{
-	int ret;
-
-
-	if ((ret = stamp1->tv_sec - stamp2->tv_sec) == 0) {
-		ret = stamp1->tv_nsec - stamp2->tv_nsec;
-	}
-	return ret;
-}
-
-
-/*
- * Define the structures by which XPC variables can be exported to other
- * partitions. (There are two: struct xpc_vars and struct xpc_vars_part)
- */
-
-/*
- * The following structure describes the partition generic variables
- * needed by other partitions in order to properly initialize.
- *
- * struct xpc_vars version number also applies to struct xpc_vars_part.
- * Changes to either structure and/or related functionality should be
- * reflected by incrementing either the major or minor version numbers
- * of struct xpc_vars.
- */
-struct xpc_vars {
-	u8 version;
-	u64 heartbeat;
-	u64 heartbeating_to_mask;
-	u64 heartbeat_offline;	/* if 0, heartbeat should be changing */
-	int act_nasid;
-	int act_phys_cpuid;
-	u64 vars_part_pa;
-	u64 amos_page_pa;	/* paddr of page of AMOs from MSPEC driver */
-	AMO_t *amos_page;	/* vaddr of page of AMOs from MSPEC driver */
-};
-
-#define XPC_V_VERSION _XPC_VERSION(3,1) /* version 3.1 of the cross vars */
-
-#define XPC_SUPPORTS_DISENGAGE_REQUEST(_version) \
-			(_version >= _XPC_VERSION(3,1))
-
-
-static inline int
-xpc_hb_allowed(partid_t partid, struct xpc_vars *vars)
-{
-	return ((vars->heartbeating_to_mask & (1UL << partid)) != 0);
-}
-
-static inline void
-xpc_allow_hb(partid_t partid, struct xpc_vars *vars)
-{
-	u64 old_mask, new_mask;
-
-	do {
-		old_mask = vars->heartbeating_to_mask;
-		new_mask = (old_mask | (1UL << partid));
-	} while (cmpxchg(&vars->heartbeating_to_mask, old_mask, new_mask) !=
-							old_mask);
-}
-
-static inline void
-xpc_disallow_hb(partid_t partid, struct xpc_vars *vars)
-{
-	u64 old_mask, new_mask;
-
-	do {
-		old_mask = vars->heartbeating_to_mask;
-		new_mask = (old_mask & ~(1UL << partid));
-	} while (cmpxchg(&vars->heartbeating_to_mask, old_mask, new_mask) !=
-							old_mask);
-}
-
-
-/*
- * The AMOs page consists of a number of AMO variables which are divided into
- * four groups, The first two groups are used to identify an IRQ's sender.
- * These two groups consist of 64 and 128 AMO variables respectively. The last
- * two groups, consisting of just one AMO variable each, are used to identify
- * the remote partitions that are currently engaged (from the viewpoint of
- * the XPC running on the remote partition).
- */
-#define XPC_NOTIFY_IRQ_AMOS	   0
-#define XPC_ACTIVATE_IRQ_AMOS	   (XPC_NOTIFY_IRQ_AMOS + XP_MAX_PARTITIONS)
-#define XPC_ENGAGED_PARTITIONS_AMO (XPC_ACTIVATE_IRQ_AMOS + XP_NASID_MASK_WORDS)
-#define XPC_DISENGAGE_REQUEST_AMO  (XPC_ENGAGED_PARTITIONS_AMO + 1)
-
-
-/*
- * The following structure describes the per partition specific variables.
- *
- * An array of these structures, one per partition, will be defined. As a
- * partition becomes active XPC will copy the array entry corresponding to
- * itself from that partition. It is desirable that the size of this
- * structure evenly divide into a cacheline, such that none of the entries
- * in this array crosses a cacheline boundary. As it is now, each entry
- * occupies half a cacheline.
- */
-struct xpc_vars_part {
-	volatile u64 magic;
-
-	u64 openclose_args_pa;	/* physical address of open and close args */
-	u64 GPs_pa;		/* physical address of Get/Put values */
-
-	u64 IPI_amo_pa;		/* physical address of IPI AMO_t structure */
-	int IPI_nasid;		/* nasid of where to send IPIs */
-	int IPI_phys_cpuid;	/* physical CPU ID of where to send IPIs */
-
-	u8 nchannels;		/* #of defined channels supported */
-
-	u8 reserved[23];	/* pad to a full 64 bytes */
-};
-
-/*
- * The vars_part MAGIC numbers play a part in the first contact protocol.
- *
- * MAGIC1 indicates that the per partition specific variables for a remote
- * partition have been initialized by this partition.
- *
- * MAGIC2 indicates that this partition has pulled the remote partititions
- * per partition variables that pertain to this partition.
- */
-#define XPC_VP_MAGIC1	0x0053524156435058L  /* 'XPCVARS\0'L (little endian) */
-#define XPC_VP_MAGIC2	0x0073726176435058L  /* 'XPCvars\0'L (little endian) */
-
-
-/* the reserved page sizes and offsets */
-
-#define XPC_RP_HEADER_SIZE	L1_CACHE_ALIGN(sizeof(struct xpc_rsvd_page))
-#define XPC_RP_VARS_SIZE 	L1_CACHE_ALIGN(sizeof(struct xpc_vars))
-
-#define XPC_RP_PART_NASIDS(_rp) (u64 *) ((u8 *) _rp + XPC_RP_HEADER_SIZE)
-#define XPC_RP_MACH_NASIDS(_rp) (XPC_RP_PART_NASIDS(_rp) + xp_nasid_mask_words)
-#define XPC_RP_VARS(_rp)	((struct xpc_vars *) XPC_RP_MACH_NASIDS(_rp) + xp_nasid_mask_words)
-#define XPC_RP_VARS_PART(_rp)	(struct xpc_vars_part *) ((u8 *) XPC_RP_VARS(rp) + XPC_RP_VARS_SIZE)
-
-
-/*
- * Functions registered by add_timer() or called by kernel_thread() only
- * allow for a single 64-bit argument. The following macros can be used to
- * pack and unpack two (32-bit, 16-bit or 8-bit) arguments into or out from
- * the passed argument.
- */
-#define XPC_PACK_ARGS(_arg1, _arg2) \
-			((((u64) _arg1) & 0xffffffff) | \
-			((((u64) _arg2) & 0xffffffff) << 32))
-
-#define XPC_UNPACK_ARG1(_args)	(((u64) _args) & 0xffffffff)
-#define XPC_UNPACK_ARG2(_args)	((((u64) _args) >> 32) & 0xffffffff)
-
-
-
-/*
- * Define a Get/Put value pair (pointers) used with a message queue.
- */
-struct xpc_gp {
-	volatile s64 get;	/* Get value */
-	volatile s64 put;	/* Put value */
-};
-
-#define XPC_GP_SIZE \
-		L1_CACHE_ALIGN(sizeof(struct xpc_gp) * XPC_NCHANNELS)
-
-
-
-/*
- * Define a structure that contains arguments associated with opening and
- * closing a channel.
- */
-struct xpc_openclose_args {
-	u16 reason;		/* reason why channel is closing */
-	u16 msg_size;		/* sizeof each message entry */
-	u16 remote_nentries;	/* #of message entries in remote msg queue */
-	u16 local_nentries;	/* #of message entries in local msg queue */
-	u64 local_msgqueue_pa;	/* physical address of local message queue */
-};
-
-#define XPC_OPENCLOSE_ARGS_SIZE \
-	      L1_CACHE_ALIGN(sizeof(struct xpc_openclose_args) * XPC_NCHANNELS)
-
-
-
-/* struct xpc_msg flags */
-
-#define	XPC_M_DONE		0x01	/* msg has been received/consumed */
-#define	XPC_M_READY		0x02	/* msg is ready to be sent */
-#define	XPC_M_INTERRUPT		0x04	/* send interrupt when msg consumed */
-
-
-#define XPC_MSG_ADDRESS(_payload) \
-		((struct xpc_msg *)((u8 *)(_payload) - XPC_MSG_PAYLOAD_OFFSET))
-
-
-
-/*
- * Defines notify entry.
- *
- * This is used to notify a message's sender that their message was received
- * and consumed by the intended recipient.
- */
-struct xpc_notify {
-	struct semaphore sema;		/* notify semaphore */
-	volatile u8 type;			/* type of notification */
-
-	/* the following two fields are only used if type == XPC_N_CALL */
-	xpc_notify_func func;		/* user's notify function */
-	void *key;			/* pointer to user's key */
-};
-
-/* struct xpc_notify type of notification */
-
-#define	XPC_N_CALL		0x01	/* notify function provided by user */
-
-
-
-/*
- * Define the structure that manages all the stuff required by a channel. In
- * particular, they are used to manage the messages sent across the channel.
- *
- * This structure is private to a partition, and is NOT shared across the
- * partition boundary.
- *
- * There is an array of these structures for each remote partition. It is
- * allocated at the time a partition becomes active. The array contains one
- * of these structures for each potential channel connection to that partition.
- *
- * Each of these structures manages two message queues (circular buffers).
- * They are allocated at the time a channel connection is made. One of
- * these message queues (local_msgqueue) holds the locally created messages
- * that are destined for the remote partition. The other of these message
- * queues (remote_msgqueue) is a locally cached copy of the remote partition's
- * own local_msgqueue.
- *
- * The following is a description of the Get/Put pointers used to manage these
- * two message queues. Consider the local_msgqueue to be on one partition
- * and the remote_msgqueue to be its cached copy on another partition. A
- * description of what each of the lettered areas contains is included.
- *
- *
- *                     local_msgqueue      remote_msgqueue
- *
- *                        |/////////|      |/////////|
- *    w_remote_GP.get --> +---------+      |/////////|
- *                        |    F    |      |/////////|
- *     remote_GP.get  --> +---------+      +---------+ <-- local_GP->get
- *                        |         |      |         |
- *                        |         |      |    E    |
- *                        |         |      |         |
- *                        |         |      +---------+ <-- w_local_GP.get
- *                        |    B    |      |/////////|
- *                        |         |      |////D////|
- *                        |         |      |/////////|
- *                        |         |      +---------+ <-- w_remote_GP.put
- *                        |         |      |////C////|
- *      local_GP->put --> +---------+      +---------+ <-- remote_GP.put
- *                        |         |      |/////////|
- *                        |    A    |      |/////////|
- *                        |         |      |/////////|
- *     w_local_GP.put --> +---------+      |/////////|
- *                        |/////////|      |/////////|
- *
- *
- *	    ( remote_GP.[get|put] are cached copies of the remote
- *	      partition's local_GP->[get|put], and thus their values can
- *	      lag behind their counterparts on the remote partition. )
- *
- *
- *  A - Messages that have been allocated, but have not yet been sent to the
- *	remote partition.
- *
- *  B - Messages that have been sent, but have not yet been acknowledged by the
- *      remote partition as having been received.
- *
- *  C - Area that needs to be prepared for the copying of sent messages, by
- *	the clearing of the message flags of any previously received messages.
- *
- *  D - Area into which sent messages are to be copied from the remote
- *	partition's local_msgqueue and then delivered to their intended
- *	recipients. [ To allow for a multi-message copy, another pointer
- *	(next_msg_to_pull) has been added to keep track of the next message
- *	number needing to be copied (pulled). It chases after w_remote_GP.put.
- *	Any messages lying between w_local_GP.get and next_msg_to_pull have
- *	been copied and are ready to be delivered. ]
- *
- *  E - Messages that have been copied and delivered, but have not yet been
- *	acknowledged by the recipient as having been received.
- *
- *  F - Messages that have been acknowledged, but XPC has not yet notified the
- *	sender that the message was received by its intended recipient.
- *	This is also an area that needs to be prepared for the allocating of
- *	new messages, by the clearing of the message flags of the acknowledged
- *	messages.
- */
-struct xpc_channel {
-	partid_t partid;		/* ID of remote partition connected */
-	spinlock_t lock;		/* lock for updating this structure */
-	u32 flags;			/* general flags */
-
-	enum xpc_retval reason;		/* reason why channel is disconnect'g */
-	int reason_line;		/* line# disconnect initiated from */
-
-	u16 number;			/* channel # */
-
-	u16 msg_size;			/* sizeof each msg entry */
-	u16 local_nentries;		/* #of msg entries in local msg queue */
-	u16 remote_nentries;		/* #of msg entries in remote msg queue*/
-
-	void *local_msgqueue_base;	/* base address of kmalloc'd space */
-	struct xpc_msg *local_msgqueue;	/* local message queue */
-	void *remote_msgqueue_base;	/* base address of kmalloc'd space */
-	struct xpc_msg *remote_msgqueue;/* cached copy of remote partition's */
-					/* local message queue */
-	u64 remote_msgqueue_pa;		/* phys addr of remote partition's */
-					/* local message queue */
-
-	atomic_t references;		/* #of external references to queues */
-
-	atomic_t n_on_msg_allocate_wq;   /* #on msg allocation wait queue */
-	wait_queue_head_t msg_allocate_wq; /* msg allocation wait queue */
-
-	u8 delayed_IPI_flags;		/* IPI flags received, but delayed */
-					/* action until channel disconnected */
-
-	/* queue of msg senders who want to be notified when msg received */
-
-	atomic_t n_to_notify;		/* #of msg senders to notify */
-	struct xpc_notify *notify_queue;/* notify queue for messages sent */
-
-	xpc_channel_func func;		/* user's channel function */
-	void *key;			/* pointer to user's key */
-
-	struct semaphore msg_to_pull_sema; /* next msg to pull serialization */
-	struct semaphore wdisconnect_sema; /* wait for channel disconnect */
-
-	struct xpc_openclose_args *local_openclose_args; /* args passed on */
-					/* opening or closing of channel */
-
-	/* various flavors of local and remote Get/Put values */
-
-	struct xpc_gp *local_GP;	/* local Get/Put values */
-	struct xpc_gp remote_GP;	/* remote Get/Put values */
-	struct xpc_gp w_local_GP;	/* working local Get/Put values */
-	struct xpc_gp w_remote_GP;	/* working remote Get/Put values */
-	s64 next_msg_to_pull;		/* Put value of next msg to pull */
-
-	/* kthread management related fields */
-
-// >>> rethink having kthreads_assigned_limit and kthreads_idle_limit; perhaps
-// >>> allow the assigned limit be unbounded and let the idle limit be dynamic
-// >>> dependent on activity over the last interval of time
-	atomic_t kthreads_assigned;	/* #of kthreads assigned to channel */
-	u32 kthreads_assigned_limit; 	/* limit on #of kthreads assigned */
-	atomic_t kthreads_idle;		/* #of kthreads idle waiting for work */
-	u32 kthreads_idle_limit;	/* limit on #of kthreads idle */
-	atomic_t kthreads_active;	/* #of kthreads actively working */
-	// >>> following field is temporary
-	u32 kthreads_created;		/* total #of kthreads created */
-
-	wait_queue_head_t idle_wq;	/* idle kthread wait queue */
-
-} ____cacheline_aligned;
-
-
-/* struct xpc_channel flags */
-
-#define	XPC_C_WASCONNECTED	0x00000001 /* channel was connected */
-
-#define	XPC_C_ROPENREPLY	0x00000002 /* remote open channel reply */
-#define	XPC_C_OPENREPLY		0x00000004 /* local open channel reply */
-#define	XPC_C_ROPENREQUEST	0x00000008 /* remote open channel request */
-#define	XPC_C_OPENREQUEST	0x00000010 /* local open channel request */
-
-#define	XPC_C_SETUP		0x00000020 /* channel's msgqueues are alloc'd */
-#define	XPC_C_CONNECTCALLOUT	0x00000040 /* channel connected callout made */
-#define	XPC_C_CONNECTED		0x00000080 /* local channel is connected */
-#define	XPC_C_CONNECTING	0x00000100 /* channel is being connected */
-
-#define	XPC_C_RCLOSEREPLY	0x00000200 /* remote close channel reply */
-#define	XPC_C_CLOSEREPLY	0x00000400 /* local close channel reply */
-#define	XPC_C_RCLOSEREQUEST	0x00000800 /* remote close channel request */
-#define	XPC_C_CLOSEREQUEST	0x00001000 /* local close channel request */
-
-#define	XPC_C_DISCONNECTED	0x00002000 /* channel is disconnected */
-#define	XPC_C_DISCONNECTING	0x00004000 /* channel is being disconnected */
-#define	XPC_C_DISCONNECTCALLOUT	0x00008000 /* chan disconnected callout made */
-#define	XPC_C_WDISCONNECT	0x00010000 /* waiting for channel disconnect */
-
-
-
-/*
- * Manages channels on a partition basis. There is one of these structures
- * for each partition (a partition will never utilize the structure that
- * represents itself).
- */
-struct xpc_partition {
-
-	/* XPC HB infrastructure */
-
-	u8 remote_rp_version;		/* version# of partition's rsvd pg */
-	struct timespec remote_rp_stamp;/* time when rsvd pg was initialized */
-	u64 remote_rp_pa;		/* phys addr of partition's rsvd pg */
-	u64 remote_vars_pa;		/* phys addr of partition's vars */
-	u64 remote_vars_part_pa;	/* phys addr of partition's vars part */
-	u64 last_heartbeat;		/* HB at last read */
-	u64 remote_amos_page_pa;	/* phys addr of partition's amos page */
-	int remote_act_nasid;		/* active part's act/deact nasid */
-	int remote_act_phys_cpuid;	/* active part's act/deact phys cpuid */
-	u32 act_IRQ_rcvd;		/* IRQs since activation */
-	spinlock_t act_lock;		/* protect updating of act_state */
-	u8 act_state;			/* from XPC HB viewpoint */
-	u8 remote_vars_version;		/* version# of partition's vars */
-	enum xpc_retval reason;		/* reason partition is deactivating */
-	int reason_line;		/* line# deactivation initiated from */
-	int reactivate_nasid;		/* nasid in partition to reactivate */
-
-	unsigned long disengage_request_timeout; /* timeout in jiffies */
-	struct timer_list disengage_request_timer;
-
-
-	/* XPC infrastructure referencing and teardown control */
-
-	volatile u8 setup_state;	/* infrastructure setup state */
-	wait_queue_head_t teardown_wq;	/* kthread waiting to teardown infra */
-	atomic_t references;		/* #of references to infrastructure */
-
-
-	/*
-	 * NONE OF THE PRECEDING FIELDS OF THIS STRUCTURE WILL BE CLEARED WHEN
-	 * XPC SETS UP THE NECESSARY INFRASTRUCTURE TO SUPPORT CROSS PARTITION
-	 * COMMUNICATION. ALL OF THE FOLLOWING FIELDS WILL BE CLEARED. (THE
-	 * 'nchannels' FIELD MUST BE THE FIRST OF THE FIELDS TO BE CLEARED.)
-	 */
-
-
-	u8 nchannels;		   /* #of defined channels supported */
-	atomic_t nchannels_active; /* #of channels that are not DISCONNECTED */
-	atomic_t nchannels_engaged;/* #of channels engaged with remote part */
-	struct xpc_channel *channels;/* array of channel structures */
-
-	void *local_GPs_base;	  /* base address of kmalloc'd space */
-	struct xpc_gp *local_GPs; /* local Get/Put values */
-	void *remote_GPs_base;    /* base address of kmalloc'd space */
-	struct xpc_gp *remote_GPs;/* copy of remote partition's local Get/Put */
-				  /* values */
-	u64 remote_GPs_pa;	  /* phys address of remote partition's local */
-				  /* Get/Put values */
-
-
-	/* fields used to pass args when opening or closing a channel */
-
-	void *local_openclose_args_base;  /* base address of kmalloc'd space */
-	struct xpc_openclose_args *local_openclose_args;  /* local's args */
-	void *remote_openclose_args_base; /* base address of kmalloc'd space */
-	struct xpc_openclose_args *remote_openclose_args; /* copy of remote's */
-					  /* args */
-	u64 remote_openclose_args_pa;	  /* phys addr of remote's args */
-
-
-	/* IPI sending, receiving and handling related fields */
-
-	int remote_IPI_nasid;	    /* nasid of where to send IPIs */
-	int remote_IPI_phys_cpuid;  /* phys CPU ID of where to send IPIs */
-	AMO_t *remote_IPI_amo_va;   /* address of remote IPI AMO_t structure */
-
-	AMO_t *local_IPI_amo_va;    /* address of IPI AMO_t structure */
-	u64 local_IPI_amo;	    /* IPI amo flags yet to be handled */
-	char IPI_owner[8];	    /* IPI owner's name */
-	struct timer_list dropped_IPI_timer; /* dropped IPI timer */
-
-	spinlock_t IPI_lock;	    /* IPI handler lock */
-
-
-	/* channel manager related fields */
-
-	atomic_t channel_mgr_requests;	/* #of requests to activate chan mgr */
-	wait_queue_head_t channel_mgr_wq; /* channel mgr's wait queue */
-
-} ____cacheline_aligned;
-
-
-/* struct xpc_partition act_state values (for XPC HB) */
-
-#define	XPC_P_INACTIVE		0x00	/* partition is not active */
-#define XPC_P_ACTIVATION_REQ	0x01	/* created thread to activate */
-#define XPC_P_ACTIVATING	0x02	/* activation thread started */
-#define XPC_P_ACTIVE		0x03	/* xpc_partition_up() was called */
-#define XPC_P_DEACTIVATING	0x04	/* partition deactivation initiated */
-
-
-#define XPC_DEACTIVATE_PARTITION(_p, _reason) \
-			xpc_deactivate_partition(__LINE__, (_p), (_reason))
-
-
-/* struct xpc_partition setup_state values */
-
-#define XPC_P_UNSET		0x00	/* infrastructure was never setup */
-#define XPC_P_SETUP		0x01	/* infrastructure is setup */
-#define XPC_P_WTEARDOWN		0x02	/* waiting to teardown infrastructure */
-#define XPC_P_TORNDOWN		0x03	/* infrastructure is torndown */
-
-
-
-/*
- * struct xpc_partition IPI_timer #of seconds to wait before checking for
- * dropped IPIs. These occur whenever an IPI amo write doesn't complete until
- * after the IPI was received.
- */
-#define XPC_P_DROPPED_IPI_WAIT	(0.25 * HZ)
-
-
-/* number of seconds to wait for other partitions to disengage */
-#define XPC_DISENGAGE_REQUEST_DEFAULT_TIMELIMIT	90
-
-/* interval in seconds to print 'waiting disengagement' messages */
-#define XPC_DISENGAGE_PRINTMSG_INTERVAL		10
-
-
-#define XPC_PARTID(_p)	((partid_t) ((_p) - &xpc_partitions[0]))
-
-
-
-/* found in xp_main.c */
-extern struct xpc_registration xpc_registrations[];
-
-
-/* found in xpc_main.c */
-extern struct device *xpc_part;
-extern struct device *xpc_chan;
-extern int xpc_disengage_request_timelimit;
-extern irqreturn_t xpc_notify_IRQ_handler(int, void *, struct pt_regs *);
-extern void xpc_dropped_IPI_check(struct xpc_partition *);
-extern void xpc_activate_partition(struct xpc_partition *);
-extern void xpc_activate_kthreads(struct xpc_channel *, int);
-extern void xpc_create_kthreads(struct xpc_channel *, int);
-extern void xpc_disconnect_wait(int);
-
-
-/* found in xpc_partition.c */
-extern int xpc_exiting;
-extern struct xpc_vars *xpc_vars;
-extern struct xpc_rsvd_page *xpc_rsvd_page;
-extern struct xpc_vars_part *xpc_vars_part;
-extern struct xpc_partition xpc_partitions[XP_MAX_PARTITIONS + 1];
-extern char xpc_remote_copy_buffer[];
-extern struct xpc_rsvd_page *xpc_rsvd_page_init(void);
-extern void xpc_allow_IPI_ops(void);
-extern void xpc_restrict_IPI_ops(void);
-extern int xpc_identify_act_IRQ_sender(void);
-extern int xpc_partition_disengaged(struct xpc_partition *);
-extern enum xpc_retval xpc_mark_partition_active(struct xpc_partition *);
-extern void xpc_mark_partition_inactive(struct xpc_partition *);
-extern void xpc_discovery(void);
-extern void xpc_check_remote_hb(void);
-extern void xpc_deactivate_partition(const int, struct xpc_partition *,
-						enum xpc_retval);
-extern enum xpc_retval xpc_initiate_partid_to_nasids(partid_t, void *);
-
-
-/* found in xpc_channel.c */
-extern void xpc_initiate_connect(int);
-extern void xpc_initiate_disconnect(int);
-extern enum xpc_retval xpc_initiate_allocate(partid_t, int, u32, void **);
-extern enum xpc_retval xpc_initiate_send(partid_t, int, void *);
-extern enum xpc_retval xpc_initiate_send_notify(partid_t, int, void *,
-						xpc_notify_func, void *);
-extern void xpc_initiate_received(partid_t, int, void *);
-extern enum xpc_retval xpc_setup_infrastructure(struct xpc_partition *);
-extern enum xpc_retval xpc_pull_remote_vars_part(struct xpc_partition *);
-extern void xpc_process_channel_activity(struct xpc_partition *);
-extern void xpc_connected_callout(struct xpc_channel *);
-extern void xpc_deliver_msg(struct xpc_channel *);
-extern void xpc_disconnect_channel(const int, struct xpc_channel *,
-					enum xpc_retval, unsigned long *);
-extern void xpc_disconnecting_callout(struct xpc_channel *);
-extern void xpc_partition_going_down(struct xpc_partition *, enum xpc_retval);
-extern void xpc_teardown_infrastructure(struct xpc_partition *);
-
-
-
-static inline void
-xpc_wakeup_channel_mgr(struct xpc_partition *part)
-{
-	if (atomic_inc_return(&part->channel_mgr_requests) == 1) {
-		wake_up(&part->channel_mgr_wq);
-	}
-}
-
-
-
-/*
- * These next two inlines are used to keep us from tearing down a channel's
- * msg queues while a thread may be referencing them.
- */
-static inline void
-xpc_msgqueue_ref(struct xpc_channel *ch)
-{
-	atomic_inc(&ch->references);
-}
-
-static inline void
-xpc_msgqueue_deref(struct xpc_channel *ch)
-{
-	s32 refs = atomic_dec_return(&ch->references);
-
-	DBUG_ON(refs < 0);
-	if (refs == 0) {
-		xpc_wakeup_channel_mgr(&xpc_partitions[ch->partid]);
-	}
-}
-
-
-
-#define XPC_DISCONNECT_CHANNEL(_ch, _reason, _irqflgs) \
-		xpc_disconnect_channel(__LINE__, _ch, _reason, _irqflgs)
-
-
-/*
- * These two inlines are used to keep us from tearing down a partition's
- * setup infrastructure while a thread may be referencing it.
- */
-static inline void
-xpc_part_deref(struct xpc_partition *part)
-{
-	s32 refs = atomic_dec_return(&part->references);
-
-
-	DBUG_ON(refs < 0);
-	if (refs == 0 && part->setup_state == XPC_P_WTEARDOWN) {
-		wake_up(&part->teardown_wq);
-	}
-}
-
-static inline int
-xpc_part_ref(struct xpc_partition *part)
-{
-	int setup;
-
-
-	atomic_inc(&part->references);
-	setup = (part->setup_state == XPC_P_SETUP);
-	if (!setup) {
-		xpc_part_deref(part);
-	}
-	return setup;
-}
-
-
-
-/*
- * The following macro is to be used for the setting of the reason and
- * reason_line fields in both the struct xpc_channel and struct xpc_partition
- * structures.
- */
-#define XPC_SET_REASON(_p, _reason, _line) \
-	{ \
-		(_p)->reason = _reason; \
-		(_p)->reason_line = _line; \
-	}
-
-
-
-/*
- * This next set of inlines are used to keep track of when a partition is
- * potentially engaged in accessing memory belonging to another partition.
- */
-
-static inline void
-xpc_mark_partition_engaged(struct xpc_partition *part)
-{
-	unsigned long irq_flags;
-	AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
-				(XPC_ENGAGED_PARTITIONS_AMO * sizeof(AMO_t)));
-
-
-	local_irq_save(irq_flags);
-
-	/* set bit corresponding to our partid in remote partition's AMO */
-	FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_OR,
-						(1UL << sn_partition_id));
-	/*
-	 * We must always use the nofault function regardless of whether we
-	 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
-	 * didn't, we'd never know that the other partition is down and would
-	 * keep sending IPIs and AMOs to it until the heartbeat times out.
-	 */
-	(void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
-				variable), xp_nofault_PIOR_target));
-
-	local_irq_restore(irq_flags);
-}
-
-static inline void
-xpc_mark_partition_disengaged(struct xpc_partition *part)
-{
-	unsigned long irq_flags;
-	AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
-				(XPC_ENGAGED_PARTITIONS_AMO * sizeof(AMO_t)));
-
-
-	local_irq_save(irq_flags);
-
-	/* clear bit corresponding to our partid in remote partition's AMO */
-	FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
-						~(1UL << sn_partition_id));
-	/*
-	 * We must always use the nofault function regardless of whether we
-	 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
-	 * didn't, we'd never know that the other partition is down and would
-	 * keep sending IPIs and AMOs to it until the heartbeat times out.
-	 */
-	(void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
-				variable), xp_nofault_PIOR_target));
-
-	local_irq_restore(irq_flags);
-}
-
-static inline void
-xpc_request_partition_disengage(struct xpc_partition *part)
-{
-	unsigned long irq_flags;
-	AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
-				(XPC_DISENGAGE_REQUEST_AMO * sizeof(AMO_t)));
-
-
-	local_irq_save(irq_flags);
-
-	/* set bit corresponding to our partid in remote partition's AMO */
-	FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_OR,
-						(1UL << sn_partition_id));
-	/*
-	 * We must always use the nofault function regardless of whether we
-	 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
-	 * didn't, we'd never know that the other partition is down and would
-	 * keep sending IPIs and AMOs to it until the heartbeat times out.
-	 */
-	(void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
-				variable), xp_nofault_PIOR_target));
-
-	local_irq_restore(irq_flags);
-}
-
-static inline void
-xpc_cancel_partition_disengage_request(struct xpc_partition *part)
-{
-	unsigned long irq_flags;
-	AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
-				(XPC_DISENGAGE_REQUEST_AMO * sizeof(AMO_t)));
-
-
-	local_irq_save(irq_flags);
-
-	/* clear bit corresponding to our partid in remote partition's AMO */
-	FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
-						~(1UL << sn_partition_id));
-	/*
-	 * We must always use the nofault function regardless of whether we
-	 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
-	 * didn't, we'd never know that the other partition is down and would
-	 * keep sending IPIs and AMOs to it until the heartbeat times out.
-	 */
-	(void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
-				variable), xp_nofault_PIOR_target));
-
-	local_irq_restore(irq_flags);
-}
-
-static inline u64
-xpc_partition_engaged(u64 partid_mask)
-{
-	AMO_t *amo = xpc_vars->amos_page + XPC_ENGAGED_PARTITIONS_AMO;
-
-
-	/* return our partition's AMO variable ANDed with partid_mask */
-	return (FETCHOP_LOAD_OP(TO_AMO((u64) &amo->variable), FETCHOP_LOAD) &
-								partid_mask);
-}
-
-static inline u64
-xpc_partition_disengage_requested(u64 partid_mask)
-{
-	AMO_t *amo = xpc_vars->amos_page + XPC_DISENGAGE_REQUEST_AMO;
-
-
-	/* return our partition's AMO variable ANDed with partid_mask */
-	return (FETCHOP_LOAD_OP(TO_AMO((u64) &amo->variable), FETCHOP_LOAD) &
-								partid_mask);
-}
-
-static inline void
-xpc_clear_partition_engaged(u64 partid_mask)
-{
-	AMO_t *amo = xpc_vars->amos_page + XPC_ENGAGED_PARTITIONS_AMO;
-
-
-	/* clear bit(s) based on partid_mask in our partition's AMO */
-	FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
-								~partid_mask);
-}
-
-static inline void
-xpc_clear_partition_disengage_request(u64 partid_mask)
-{
-	AMO_t *amo = xpc_vars->amos_page + XPC_DISENGAGE_REQUEST_AMO;
-
-
-	/* clear bit(s) based on partid_mask in our partition's AMO */
-	FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
-								~partid_mask);
-}
-
-
-
-/*
- * The following set of macros and inlines are used for the sending and
- * receiving of IPIs (also known as IRQs). There are two flavors of IPIs,
- * one that is associated with partition activity (SGI_XPC_ACTIVATE) and
- * the other that is associated with channel activity (SGI_XPC_NOTIFY).
- */
-
-static inline u64
-xpc_IPI_receive(AMO_t *amo)
-{
-	return FETCHOP_LOAD_OP(TO_AMO((u64) &amo->variable), FETCHOP_CLEAR);
-}
-
-
-static inline enum xpc_retval
-xpc_IPI_send(AMO_t *amo, u64 flag, int nasid, int phys_cpuid, int vector)
-{
-	int ret = 0;
-	unsigned long irq_flags;
-
-
-	local_irq_save(irq_flags);
-
-	FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_OR, flag);
-	sn_send_IPI_phys(nasid, phys_cpuid, vector, 0);
-
-	/*
-	 * We must always use the nofault function regardless of whether we
-	 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
-	 * didn't, we'd never know that the other partition is down and would
-	 * keep sending IPIs and AMOs to it until the heartbeat times out.
-	 */
-	ret = xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->variable),
-				xp_nofault_PIOR_target));
-
-	local_irq_restore(irq_flags);
-
-	return ((ret == 0) ? xpcSuccess : xpcPioReadError);
-}
-
-
-/*
- * IPIs associated with SGI_XPC_ACTIVATE IRQ.
- */
-
-/*
- * Flag the appropriate AMO variable and send an IPI to the specified node.
- */
-static inline void
-xpc_activate_IRQ_send(u64 amos_page_pa, int from_nasid, int to_nasid,
-			int to_phys_cpuid)
-{
-	int w_index = XPC_NASID_W_INDEX(from_nasid);
-	int b_index = XPC_NASID_B_INDEX(from_nasid);
-	AMO_t *amos = (AMO_t *) __va(amos_page_pa +
-				(XPC_ACTIVATE_IRQ_AMOS * sizeof(AMO_t)));
-
-
-	(void) xpc_IPI_send(&amos[w_index], (1UL << b_index), to_nasid,
-				to_phys_cpuid, SGI_XPC_ACTIVATE);
-}
-
-static inline void
-xpc_IPI_send_activate(struct xpc_vars *vars)
-{
-	xpc_activate_IRQ_send(vars->amos_page_pa, cnodeid_to_nasid(0),
-				vars->act_nasid, vars->act_phys_cpuid);
-}
-
-static inline void
-xpc_IPI_send_activated(struct xpc_partition *part)
-{
-	xpc_activate_IRQ_send(part->remote_amos_page_pa, cnodeid_to_nasid(0),
-			part->remote_act_nasid, part->remote_act_phys_cpuid);
-}
-
-static inline void
-xpc_IPI_send_reactivate(struct xpc_partition *part)
-{
-	xpc_activate_IRQ_send(xpc_vars->amos_page_pa, part->reactivate_nasid,
-				xpc_vars->act_nasid, xpc_vars->act_phys_cpuid);
-}
-
-static inline void
-xpc_IPI_send_disengage(struct xpc_partition *part)
-{
-	xpc_activate_IRQ_send(part->remote_amos_page_pa, cnodeid_to_nasid(0),
-			part->remote_act_nasid, part->remote_act_phys_cpuid);
-}
-
-
-/*
- * IPIs associated with SGI_XPC_NOTIFY IRQ.
- */
-
-/*
- * Send an IPI to the remote partition that is associated with the
- * specified channel.
- */
-#define XPC_NOTIFY_IRQ_SEND(_ch, _ipi_f, _irq_f) \
-		xpc_notify_IRQ_send(_ch, _ipi_f, #_ipi_f, _irq_f)
-
-static inline void
-xpc_notify_IRQ_send(struct xpc_channel *ch, u8 ipi_flag, char *ipi_flag_string,
-			unsigned long *irq_flags)
-{
-	struct xpc_partition *part = &xpc_partitions[ch->partid];
-	enum xpc_retval ret;
-
-
-	if (likely(part->act_state != XPC_P_DEACTIVATING)) {
-		ret = xpc_IPI_send(part->remote_IPI_amo_va,
-					(u64) ipi_flag << (ch->number * 8),
-					part->remote_IPI_nasid,
-					part->remote_IPI_phys_cpuid,
-					SGI_XPC_NOTIFY);
-		dev_dbg(xpc_chan, "%s sent to partid=%d, channel=%d, ret=%d\n",
-			ipi_flag_string, ch->partid, ch->number, ret);
-		if (unlikely(ret != xpcSuccess)) {
-			if (irq_flags != NULL) {
-				spin_unlock_irqrestore(&ch->lock, *irq_flags);
-			}
-			XPC_DEACTIVATE_PARTITION(part, ret);
-			if (irq_flags != NULL) {
-				spin_lock_irqsave(&ch->lock, *irq_flags);
-			}
-		}
-	}
-}
-
-
-/*
- * Make it look like the remote partition, which is associated with the
- * specified channel, sent us an IPI. This faked IPI will be handled
- * by xpc_dropped_IPI_check().
- */
-#define XPC_NOTIFY_IRQ_SEND_LOCAL(_ch, _ipi_f) \
-		xpc_notify_IRQ_send_local(_ch, _ipi_f, #_ipi_f)
-
-static inline void
-xpc_notify_IRQ_send_local(struct xpc_channel *ch, u8 ipi_flag,
-				char *ipi_flag_string)
-{
-	struct xpc_partition *part = &xpc_partitions[ch->partid];
-
-
-	FETCHOP_STORE_OP(TO_AMO((u64) &part->local_IPI_amo_va->variable),
-			FETCHOP_OR, ((u64) ipi_flag << (ch->number * 8)));
-	dev_dbg(xpc_chan, "%s sent local from partid=%d, channel=%d\n",
-		ipi_flag_string, ch->partid, ch->number);
-}
-
-
-/*
- * The sending and receiving of IPIs includes the setting of an AMO variable
- * to indicate the reason the IPI was sent. The 64-bit variable is divided
- * up into eight bytes, ordered from right to left. Byte zero pertains to
- * channel 0, byte one to channel 1, and so on. Each byte is described by
- * the following IPI flags.
- */
-
-#define	XPC_IPI_CLOSEREQUEST	0x01
-#define	XPC_IPI_CLOSEREPLY	0x02
-#define	XPC_IPI_OPENREQUEST	0x04
-#define	XPC_IPI_OPENREPLY	0x08
-#define	XPC_IPI_MSGREQUEST	0x10
-
-
-/* given an AMO variable and a channel#, get its associated IPI flags */
-#define XPC_GET_IPI_FLAGS(_amo, _c)	((u8) (((_amo) >> ((_c) * 8)) & 0xff))
-#define XPC_SET_IPI_FLAGS(_amo, _c, _f)	(_amo) |= ((u64) (_f) << ((_c) * 8))
-
-#define	XPC_ANY_OPENCLOSE_IPI_FLAGS_SET(_amo) ((_amo) & 0x0f0f0f0f0f0f0f0f)
-#define XPC_ANY_MSG_IPI_FLAGS_SET(_amo)       ((_amo) & 0x1010101010101010)
-
-
-static inline void
-xpc_IPI_send_closerequest(struct xpc_channel *ch, unsigned long *irq_flags)
-{
-	struct xpc_openclose_args *args = ch->local_openclose_args;
-
-
-	args->reason = ch->reason;
-
-	XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_CLOSEREQUEST, irq_flags);
-}
-
-static inline void
-xpc_IPI_send_closereply(struct xpc_channel *ch, unsigned long *irq_flags)
-{
-	XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_CLOSEREPLY, irq_flags);
-}
-
-static inline void
-xpc_IPI_send_openrequest(struct xpc_channel *ch, unsigned long *irq_flags)
-{
-	struct xpc_openclose_args *args = ch->local_openclose_args;
-
-
-	args->msg_size = ch->msg_size;
-	args->local_nentries = ch->local_nentries;
-
-	XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_OPENREQUEST, irq_flags);
-}
-
-static inline void
-xpc_IPI_send_openreply(struct xpc_channel *ch, unsigned long *irq_flags)
-{
-	struct xpc_openclose_args *args = ch->local_openclose_args;
-
-
-	args->remote_nentries = ch->remote_nentries;
-	args->local_nentries = ch->local_nentries;
-	args->local_msgqueue_pa = __pa(ch->local_msgqueue);
-
-	XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_OPENREPLY, irq_flags);
-}
-
-static inline void
-xpc_IPI_send_msgrequest(struct xpc_channel *ch)
-{
-	XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_MSGREQUEST, NULL);
-}
-
-static inline void
-xpc_IPI_send_local_msgrequest(struct xpc_channel *ch)
-{
-	XPC_NOTIFY_IRQ_SEND_LOCAL(ch, XPC_IPI_MSGREQUEST);
-}
-
-
-/*
- * Memory for XPC's AMO variables is allocated by the MSPEC driver. These
- * pages are located in the lowest granule. The lowest granule uses 4k pages
- * for cached references and an alternate TLB handler to never provide a
- * cacheable mapping for the entire region. This will prevent speculative
- * reading of cached copies of our lines from being issued which will cause
- * a PI FSB Protocol error to be generated by the SHUB. For XPC, we need 64
- * AMO variables (based on XP_MAX_PARTITIONS) for message notification and an
- * additional 128 AMO variables (based on XP_NASID_MASK_WORDS) for partition
- * activation and 2 AMO variables for partition deactivation.
- */
-static inline AMO_t *
-xpc_IPI_init(int index)
-{
-	AMO_t *amo = xpc_vars->amos_page + index;
-
-
-	(void) xpc_IPI_receive(amo);	/* clear AMO variable */
-	return amo;
-}
-
-
-
-static inline enum xpc_retval
-xpc_map_bte_errors(bte_result_t error)
-{
-	switch (error) {
-	case BTE_SUCCESS:	return xpcSuccess;
-	case BTEFAIL_DIR:	return xpcBteDirectoryError;
-	case BTEFAIL_POISON:	return xpcBtePoisonError;
-	case BTEFAIL_WERR:	return xpcBteWriteError;
-	case BTEFAIL_ACCESS:	return xpcBteAccessError;
-	case BTEFAIL_PWERR:	return xpcBtePWriteError;
-	case BTEFAIL_PRERR:	return xpcBtePReadError;
-	case BTEFAIL_TOUT:	return xpcBteTimeOutError;
-	case BTEFAIL_XTERR:	return xpcBteXtalkError;
-	case BTEFAIL_NOTAVAIL:	return xpcBteNotAvailable;
-	default:		return xpcBteUnmappedError;
-	}
-}
-
-
-
-static inline void *
-xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
-{
-	/* see if kmalloc will give us cachline aligned memory by default */
-	*base = kmalloc(size, flags);
-	if (*base == NULL) {
-		return NULL;
-	}
-	if ((u64) *base == L1_CACHE_ALIGN((u64) *base)) {
-		return *base;
-	}
-	kfree(*base);
-
-	/* nope, we'll have to do it ourselves */
-	*base = kmalloc(size + L1_CACHE_BYTES, flags);
-	if (*base == NULL) {
-		return NULL;
-	}
-	return (void *) L1_CACHE_ALIGN((u64) *base);
-}
-
-
-/*
- * Check to see if there is any channel activity to/from the specified
- * partition.
- */
-static inline void
-xpc_check_for_channel_activity(struct xpc_partition *part)
-{
-	u64 IPI_amo;
-	unsigned long irq_flags;
-
-
-	IPI_amo = xpc_IPI_receive(part->local_IPI_amo_va);
-	if (IPI_amo == 0) {
-		return;
-	}
-
-	spin_lock_irqsave(&part->IPI_lock, irq_flags);
-	part->local_IPI_amo |= IPI_amo;
-	spin_unlock_irqrestore(&part->IPI_lock, irq_flags);
-
-	dev_dbg(xpc_chan, "received IPI from partid=%d, IPI_amo=0x%lx\n",
-		XPC_PARTID(part), IPI_amo);
-
-	xpc_wakeup_channel_mgr(part);
-}
-
-
-#endif /* _IA64_SN_KERNEL_XPC_H */
-
diff --git a/arch/ia64/sn/kernel/xpc_channel.c b/arch/ia64/sn/kernel/xpc_channel.c
index abf4fc2..0c0a689 100644
--- a/arch/ia64/sn/kernel/xpc_channel.c
+++ b/arch/ia64/sn/kernel/xpc_channel.c
@@ -3,7 +3,7 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (c) 2004-2005 Silicon Graphics, Inc.  All Rights Reserved.
+ * Copyright (c) 2004-2006 Silicon Graphics, Inc.  All Rights Reserved.
  */
 
 
@@ -24,7 +24,7 @@
 #include <linux/slab.h>
 #include <asm/sn/bte.h>
 #include <asm/sn/sn_sal.h>
-#include "xpc.h"
+#include <asm/sn/xpc.h>
 
 
 /*
@@ -779,6 +779,12 @@ xpc_process_disconnect(struct xpc_channel *ch, unsigned long *irq_flags)
 
 	/* both sides are disconnected now */
 
+	if (ch->flags & XPC_C_CONNECTCALLOUT) {
+		spin_unlock_irqrestore(&ch->lock, *irq_flags);
+		xpc_disconnect_callout(ch, xpcDisconnected);
+		spin_lock_irqsave(&ch->lock, *irq_flags);
+	}
+
 	/* it's now safe to free the channel's message queues */
 	xpc_free_msgqueues(ch);
 
@@ -1645,7 +1651,7 @@ xpc_disconnect_channel(const int line, struct xpc_channel *ch,
 
 
 void
-xpc_disconnecting_callout(struct xpc_channel *ch)
+xpc_disconnect_callout(struct xpc_channel *ch, enum xpc_retval reason)
 {
 	/*
 	 * Let the channel's registerer know that the channel is being
@@ -1654,15 +1660,13 @@ xpc_disconnecting_callout(struct xpc_channel *ch)
 	 */
 
 	if (ch->func != NULL) {
-		dev_dbg(xpc_chan, "ch->func() called, reason=xpcDisconnecting,"
-			" partid=%d, channel=%d\n", ch->partid, ch->number);
+		dev_dbg(xpc_chan, "ch->func() called, reason=%d, partid=%d, "
+			"channel=%d\n", reason, ch->partid, ch->number);
 
-		ch->func(xpcDisconnecting, ch->partid, ch->number, NULL,
-								ch->key);
+		ch->func(reason, ch->partid, ch->number, NULL, ch->key);
 
-		dev_dbg(xpc_chan, "ch->func() returned, reason="
-			"xpcDisconnecting, partid=%d, channel=%d\n",
-			ch->partid, ch->number);
+		dev_dbg(xpc_chan, "ch->func() returned, reason=%d, partid=%d, "
+			"channel=%d\n", reason, ch->partid, ch->number);
 	}
 }
 
diff --git a/arch/ia64/sn/kernel/xpc_main.c b/arch/ia64/sn/kernel/xpc_main.c
index b617236..8930586e 100644
--- a/arch/ia64/sn/kernel/xpc_main.c
+++ b/arch/ia64/sn/kernel/xpc_main.c
@@ -3,7 +3,7 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (c) 2004-2005 Silicon Graphics, Inc.  All Rights Reserved.
+ * Copyright (c) 2004-2006 Silicon Graphics, Inc.  All Rights Reserved.
  */
 
 
@@ -59,7 +59,7 @@
 #include <asm/sn/sn_sal.h>
 #include <asm/kdebug.h>
 #include <asm/uaccess.h>
-#include "xpc.h"
+#include <asm/sn/xpc.h>
 
 
 /* define two XPC debug device structures to be used with dev_dbg() et al */
@@ -82,6 +82,9 @@ struct device *xpc_part = &xpc_part_dbg_subname;
 struct device *xpc_chan = &xpc_chan_dbg_subname;
 
 
+static int xpc_kdebug_ignore;
+
+
 /* systune related variables for /proc/sys directories */
 
 static int xpc_hb_interval = XPC_HB_DEFAULT_INTERVAL;
@@ -162,6 +165,8 @@ static ctl_table xpc_sys_dir[] = {
 };
 static struct ctl_table_header *xpc_sysctl;
 
+/* non-zero if any remote partition disengage request was timed out */
+int xpc_disengage_request_timedout;
 
 /* #of IRQs received */
 static atomic_t xpc_act_IRQ_rcvd;
@@ -773,7 +778,7 @@ xpc_daemonize_kthread(void *args)
 			ch->flags |= XPC_C_DISCONNECTCALLOUT;
 			spin_unlock_irqrestore(&ch->lock, irq_flags);
 
-			xpc_disconnecting_callout(ch);
+			xpc_disconnect_callout(ch, xpcDisconnecting);
 		} else {
 			spin_unlock_irqrestore(&ch->lock, irq_flags);
 		}
@@ -921,9 +926,9 @@ static void
 xpc_do_exit(enum xpc_retval reason)
 {
 	partid_t partid;
-	int active_part_count;
+	int active_part_count, printed_waiting_msg = 0;
 	struct xpc_partition *part;
-	unsigned long printmsg_time;
+	unsigned long printmsg_time, disengage_request_timeout = 0;
 
 
 	/* a 'rmmod XPC' and a 'reboot' cannot both end up here together */
@@ -953,7 +958,8 @@ xpc_do_exit(enum xpc_retval reason)
 
 	/* wait for all partitions to become inactive */
 
-	printmsg_time = jiffies;
+	printmsg_time = jiffies + (XPC_DISENGAGE_PRINTMSG_INTERVAL * HZ);
+	xpc_disengage_request_timedout = 0;
 
 	do {
 		active_part_count = 0;
@@ -969,20 +975,39 @@ xpc_do_exit(enum xpc_retval reason)
 			active_part_count++;
 
 			XPC_DEACTIVATE_PARTITION(part, reason);
-		}
 
-		if (active_part_count == 0) {
-			break;
+			if (part->disengage_request_timeout >
+						disengage_request_timeout) {
+				disengage_request_timeout =
+						part->disengage_request_timeout;
+			}
 		}
 
-		if (jiffies >= printmsg_time) {
-			dev_info(xpc_part, "waiting for partitions to "
-				"deactivate/disengage, active count=%d, remote "
-				"engaged=0x%lx\n", active_part_count,
-				xpc_partition_engaged(1UL << partid));
-
-			printmsg_time = jiffies +
+		if (xpc_partition_engaged(-1UL)) {
+			if (time_after(jiffies, printmsg_time)) {
+				dev_info(xpc_part, "waiting for remote "
+					"partitions to disengage, timeout in "
+					"%ld seconds\n",
+					(disengage_request_timeout - jiffies)
+									/ HZ);
+				printmsg_time = jiffies +
 					(XPC_DISENGAGE_PRINTMSG_INTERVAL * HZ);
+				printed_waiting_msg = 1;
+			}
+
+		} else if (active_part_count > 0) {
+			if (printed_waiting_msg) {
+				dev_info(xpc_part, "waiting for local partition"
+					" to disengage\n");
+				printed_waiting_msg = 0;
+			}
+
+		} else {
+			if (!xpc_disengage_request_timedout) {
+				dev_info(xpc_part, "all partitions have "
+					"disengaged\n");
+			}
+			break;
 		}
 
 		/* sleep for a 1/3 of a second or so */
@@ -1000,11 +1025,13 @@ xpc_do_exit(enum xpc_retval reason)
 	del_timer_sync(&xpc_hb_timer);
 	DBUG_ON(xpc_vars->heartbeating_to_mask != 0);
 
-	/* take ourselves off of the reboot_notifier_list */
-	(void) unregister_reboot_notifier(&xpc_reboot_notifier);
+	if (reason == xpcUnloading) {
+		/* take ourselves off of the reboot_notifier_list */
+		(void) unregister_reboot_notifier(&xpc_reboot_notifier);
 
-	/* take ourselves off of the die_notifier list */
-	(void) unregister_die_notifier(&xpc_die_notifier);
+		/* take ourselves off of the die_notifier list */
+		(void) unregister_die_notifier(&xpc_die_notifier);
+	}
 
 	/* close down protections for IPI operations */
 	xpc_restrict_IPI_ops();
@@ -1020,7 +1047,35 @@ xpc_do_exit(enum xpc_retval reason)
 
 
 /*
- * Called when the system is about to be either restarted or halted.
+ * This function is called when the system is being rebooted.
+ */
+static int
+xpc_system_reboot(struct notifier_block *nb, unsigned long event, void *unused)
+{
+	enum xpc_retval reason;
+
+
+	switch (event) {
+	case SYS_RESTART:
+		reason = xpcSystemReboot;
+		break;
+	case SYS_HALT:
+		reason = xpcSystemHalt;
+		break;
+	case SYS_POWER_OFF:
+		reason = xpcSystemPoweroff;
+		break;
+	default:
+		reason = xpcSystemGoingDown;
+	}
+
+	xpc_do_exit(reason);
+	return NOTIFY_DONE;
+}
+
+
+/*
+ * Notify other partitions to disengage from all references to our memory.
  */
 static void
 xpc_die_disengage(void)
@@ -1028,7 +1083,7 @@ xpc_die_disengage(void)
 	struct xpc_partition *part;
 	partid_t partid;
 	unsigned long engaged;
-	long time, print_time, disengage_request_timeout;
+	long time, printmsg_time, disengage_request_timeout;
 
 
 	/* keep xpc_hb_checker thread from doing anything (just in case) */
@@ -1055,57 +1110,53 @@ xpc_die_disengage(void)
 		}
 	}
 
-	print_time = rtc_time();
-	disengage_request_timeout = print_time +
+	time = rtc_time();
+	printmsg_time = time +
+		(XPC_DISENGAGE_PRINTMSG_INTERVAL * sn_rtc_cycles_per_second);
+	disengage_request_timeout = time +
 		(xpc_disengage_request_timelimit * sn_rtc_cycles_per_second);
 
 	/* wait for all other partitions to disengage from us */
 
-	while ((engaged = xpc_partition_engaged(-1UL)) &&
-			(time = rtc_time()) < disengage_request_timeout) {
+	while (1) {
+		engaged = xpc_partition_engaged(-1UL);
+		if (!engaged) {
+			dev_info(xpc_part, "all partitions have disengaged\n");
+			break;
+		}
 
-		if (time >= print_time) {
+		time = rtc_time();
+		if (time >= disengage_request_timeout) {
+			for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+				if (engaged & (1UL << partid)) {
+					dev_info(xpc_part, "disengage from "
+						"remote partition %d timed "
+						"out\n", partid);
+				}
+			}
+			break;
+		}
+
+		if (time >= printmsg_time) {
 			dev_info(xpc_part, "waiting for remote partitions to "
-				"disengage, engaged=0x%lx\n", engaged);
-			print_time = time + (XPC_DISENGAGE_PRINTMSG_INTERVAL *
+				"disengage, timeout in %ld seconds\n",
+				(disengage_request_timeout - time) /
+						sn_rtc_cycles_per_second);
+			printmsg_time = time +
+					(XPC_DISENGAGE_PRINTMSG_INTERVAL *
 						sn_rtc_cycles_per_second);
 		}
 	}
-	dev_info(xpc_part, "finished waiting for remote partitions to "
-				"disengage, engaged=0x%lx\n", engaged);
-}
-
-
-/*
- * This function is called when the system is being rebooted.
- */
-static int
-xpc_system_reboot(struct notifier_block *nb, unsigned long event, void *unused)
-{
-	enum xpc_retval reason;
-
-
-	switch (event) {
-	case SYS_RESTART:
-		reason = xpcSystemReboot;
-		break;
-	case SYS_HALT:
-		reason = xpcSystemHalt;
-		break;
-	case SYS_POWER_OFF:
-		reason = xpcSystemPoweroff;
-		break;
-	default:
-		reason = xpcSystemGoingDown;
-	}
-
-	xpc_do_exit(reason);
-	return NOTIFY_DONE;
 }
 
 
 /*
- * This function is called when the system is being rebooted.
+ * This function is called when the system is being restarted or halted due
+ * to some sort of system failure. If this is the case we need to notify the
+ * other partitions to disengage from all references to our memory.
+ * This function can also be called when our heartbeater could be offlined
+ * for a time. In this case we need to notify other partitions to not worry
+ * about the lack of a heartbeat.
  */
 static int
 xpc_system_die(struct notifier_block *nb, unsigned long event, void *unused)
@@ -1115,11 +1166,25 @@ xpc_system_die(struct notifier_block *nb, unsigned long event, void *unused)
 	case DIE_MACHINE_HALT:
 		xpc_die_disengage();
 		break;
+
+	case DIE_KDEBUG_ENTER:
+		/* Should lack of heartbeat be ignored by other partitions? */
+		if (!xpc_kdebug_ignore) {
+			break;
+		}
+		/* fall through */
 	case DIE_MCA_MONARCH_ENTER:
 	case DIE_INIT_MONARCH_ENTER:
 		xpc_vars->heartbeat++;
 		xpc_vars->heartbeat_offline = 1;
 		break;
+
+	case DIE_KDEBUG_LEAVE:
+		/* Is lack of heartbeat being ignored by other partitions? */
+		if (!xpc_kdebug_ignore) {
+			break;
+		}
+		/* fall through */
 	case DIE_MCA_MONARCH_LEAVE:
 	case DIE_INIT_MONARCH_LEAVE:
 		xpc_vars->heartbeat++;
@@ -1344,3 +1409,7 @@ module_param(xpc_disengage_request_timelimit, int, 0);
 MODULE_PARM_DESC(xpc_disengage_request_timelimit, "Number of seconds to wait "
 		"for disengage request to complete.");
 
+module_param(xpc_kdebug_ignore, int, 0);
+MODULE_PARM_DESC(xpc_kdebug_ignore, "Should lack of heartbeat be ignored by "
+		"other partitions when dropping into kdebug.");
+
diff --git a/arch/ia64/sn/kernel/xpc_partition.c b/arch/ia64/sn/kernel/xpc_partition.c
index cdd6431..88a730e 100644
--- a/arch/ia64/sn/kernel/xpc_partition.c
+++ b/arch/ia64/sn/kernel/xpc_partition.c
@@ -3,7 +3,7 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (c) 2004-2005 Silicon Graphics, Inc.  All Rights Reserved.
+ * Copyright (c) 2004-2006 Silicon Graphics, Inc.  All Rights Reserved.
  */
 
 
@@ -28,7 +28,7 @@
 #include <asm/sn/sn_sal.h>
 #include <asm/sn/nodepda.h>
 #include <asm/sn/addrs.h>
-#include "xpc.h"
+#include <asm/sn/xpc.h>
 
 
 /* XPC is exiting flag */
@@ -771,7 +771,8 @@ xpc_identify_act_IRQ_req(int nasid)
 		}
 	}
 
-	if (!xpc_partition_disengaged(part)) {
+	if (part->disengage_request_timeout > 0 &&
+					!xpc_partition_disengaged(part)) {
 		/* still waiting on other side to disengage from us */
 		return;
 	}
@@ -873,6 +874,9 @@ xpc_partition_disengaged(struct xpc_partition *part)
 			 * request in a timely fashion, so assume it's dead.
 			 */
 
+			dev_info(xpc_part, "disengage from remote partition %d "
+				"timed out\n", partid);
+			xpc_disengage_request_timedout = 1;
 			xpc_clear_partition_engaged(1UL << partid);
 			disengaged = 1;
 		}
diff --git a/arch/ia64/sn/pci/pcibr/pcibr_dma.c b/arch/ia64/sn/pci/pcibr/pcibr_dma.c
index 3409347..e68332d 100644
--- a/arch/ia64/sn/pci/pcibr/pcibr_dma.c
+++ b/arch/ia64/sn/pci/pcibr/pcibr_dma.c
@@ -218,7 +218,9 @@ void sn_dma_flush(uint64_t addr)
 	uint64_t flags;
 	uint64_t itte;
 	struct hubdev_info *hubinfo;
-	volatile struct sn_flush_device_list *p;
+	volatile struct sn_flush_device_kernel *p;
+	volatile struct sn_flush_device_common *common;
+
 	struct sn_flush_nasid_entry *flush_nasid_list;
 
 	if (!sn_ioif_inited)
@@ -268,17 +270,17 @@ void sn_dma_flush(uint64_t addr)
 	p = &flush_nasid_list->widget_p[wid_num][0];
 
 	/* find a matching BAR */
-	for (i = 0; i < DEV_PER_WIDGET; i++) {
+	for (i = 0; i < DEV_PER_WIDGET; i++,p++) {
+		common = p->common;
 		for (j = 0; j < PCI_ROM_RESOURCE; j++) {
-			if (p->sfdl_bar_list[j].start == 0)
+			if (common->sfdl_bar_list[j].start == 0)
 				break;
-			if (addr >= p->sfdl_bar_list[j].start
-			    && addr <= p->sfdl_bar_list[j].end)
+			if (addr >= common->sfdl_bar_list[j].start
+			    && addr <= common->sfdl_bar_list[j].end)
 				break;
 		}
-		if (j < PCI_ROM_RESOURCE && p->sfdl_bar_list[j].start != 0)
+		if (j < PCI_ROM_RESOURCE && common->sfdl_bar_list[j].start != 0)
 			break;
-		p++;
 	}
 
 	/* if no matching BAR, return without doing anything. */
@@ -304,24 +306,24 @@ void sn_dma_flush(uint64_t addr)
 		if ((1 << XWIDGET_PART_REV_NUM_REV(revnum)) & PV907516) {
 			return;
 		} else {
-			pcireg_wrb_flush_get(p->sfdl_pcibus_info,
-					     (p->sfdl_slot - 1));
+			pcireg_wrb_flush_get(common->sfdl_pcibus_info,
+					     (common->sfdl_slot - 1));
 		}
 	} else {
-		spin_lock_irqsave(&((struct sn_flush_device_list *)p)->
-				  sfdl_flush_lock, flags);
-
-		*p->sfdl_flush_addr = 0;
+		spin_lock_irqsave((spinlock_t *)&p->sfdl_flush_lock,
+				  flags);
+		*common->sfdl_flush_addr = 0;
 
 		/* force an interrupt. */
-		*(volatile uint32_t *)(p->sfdl_force_int_addr) = 1;
+		*(volatile uint32_t *)(common->sfdl_force_int_addr) = 1;
 
 		/* wait for the interrupt to come back. */
-		while (*(p->sfdl_flush_addr) != 0x10f)
+		while (*(common->sfdl_flush_addr) != 0x10f)
 			cpu_relax();
 
 		/* okay, everything is synched up. */
-		spin_unlock_irqrestore((spinlock_t *)&p->sfdl_flush_lock, flags);
+		spin_unlock_irqrestore((spinlock_t *)&p->sfdl_flush_lock,
+				       flags);
 	}
 	return;
 }
diff --git a/arch/ia64/sn/pci/pcibr/pcibr_provider.c b/arch/ia64/sn/pci/pcibr/pcibr_provider.c
index 1f500c8..e328e94 100644
--- a/arch/ia64/sn/pci/pcibr/pcibr_provider.c
+++ b/arch/ia64/sn/pci/pcibr/pcibr_provider.c
@@ -92,7 +92,8 @@ pcibr_bus_fixup(struct pcibus_bussoft *prom_bussoft, struct pci_controller *cont
 	cnodeid_t near_cnode;
 	struct hubdev_info *hubdev_info;
 	struct pcibus_info *soft;
-	struct sn_flush_device_list *sn_flush_device_list;
+	struct sn_flush_device_kernel *sn_flush_device_kernel;
+	struct sn_flush_device_common *common;
 
 	if (! IS_PCI_BRIDGE_ASIC(prom_bussoft->bs_asic_type)) {
 		return NULL;
@@ -137,20 +138,19 @@ pcibr_bus_fixup(struct pcibus_bussoft *prom_bussoft, struct pci_controller *cont
 	hubdev_info = (struct hubdev_info *)(NODEPDA(cnode)->pdinfo);
 
 	if (hubdev_info->hdi_flush_nasid_list.widget_p) {
-		sn_flush_device_list = hubdev_info->hdi_flush_nasid_list.
+		sn_flush_device_kernel = hubdev_info->hdi_flush_nasid_list.
 		    widget_p[(int)soft->pbi_buscommon.bs_xid];
-		if (sn_flush_device_list) {
+		if (sn_flush_device_kernel) {
 			for (j = 0; j < DEV_PER_WIDGET;
-			     j++, sn_flush_device_list++) {
-				if (sn_flush_device_list->sfdl_slot == -1)
+			     j++, sn_flush_device_kernel++) {
+				common = sn_flush_device_kernel->common;
+				if (common->sfdl_slot == -1)
 					continue;
-				if ((sn_flush_device_list->
-				     sfdl_persistent_segment ==
+				if ((common->sfdl_persistent_segment ==
 				     soft->pbi_buscommon.bs_persist_segment) &&
-				     (sn_flush_device_list->
-				     sfdl_persistent_busnum ==
+				     (common->sfdl_persistent_busnum ==
 				     soft->pbi_buscommon.bs_persist_busnum))
-					sn_flush_device_list->sfdl_pcibus_info =
+					common->sfdl_pcibus_info =
 					    soft;
 			}
 		}