1 files changed, 437 insertions, 58 deletions

diff --git a/arch/ia64/kernel/irq_ia64.c b/arch/ia64/kernel/irq_ia64.c
index a041367f043..03ea78ed64a 100644
--- a/arch/ia64/kernel/irq_ia64.c
+++ b/arch/ia64/kernel/irq_ia64.c
@@ -1,5 +1,5 @@
 /*
- * linux/arch/ia64/kernel/irq.c
+ * linux/arch/ia64/kernel/irq_ia64.c
  *
  * Copyright (C) 1998-2001 Hewlett-Packard Co
  *	Stephane Eranian <eranian@hpl.hp.com>
@@ -22,14 +22,15 @@
 #include <linux/interrupt.h>
 #include <linux/ioport.h>
 #include <linux/kernel_stat.h>
-#include <linux/slab.h>
 #include <linux/ptrace.h>
-#include <linux/random.h>	/* for rand_initialize_irq() */
 #include <linux/signal.h>
 #include <linux/smp.h>
-#include <linux/smp_lock.h>
 #include <linux/threads.h>
 #include <linux/bitops.h>
+#include <linux/irq.h>
+#include <linux/ratelimit.h>
+#include <linux/acpi.h>
+#include <linux/sched.h>
 
 #include <asm/delay.h>
 #include <asm/intrinsics.h>
@@ -37,7 +38,7 @@
 #include <asm/hw_irq.h>
 #include <asm/machvec.h>
 #include <asm/pgtable.h>
-#include <asm/system.h>
+#include <asm/tlbflush.h>
 
 #ifdef CONFIG_PERFMON
 # include <asm/perfmon.h>
@@ -45,6 +46,12 @@
 
 #define IRQ_DEBUG	0
 
+#define IRQ_VECTOR_UNASSIGNED	(0)
+
+#define IRQ_UNUSED		(0)
+#define IRQ_USED		(1)
+#define IRQ_RSVD		(2)
+
 /* These can be overridden in platform_irq_init */
 int ia64_first_device_vector = IA64_DEF_FIRST_DEVICE_VECTOR;
 int ia64_last_device_vector = IA64_DEF_LAST_DEVICE_VECTOR;
@@ -53,6 +60,8 @@ int ia64_last_device_vector = IA64_DEF_LAST_DEVICE_VECTOR;
 void __iomem *ipi_base_addr = ((void __iomem *)
 			       (__IA64_UNCACHED_OFFSET | IA64_IPI_DEFAULT_BASE_ADDR));
 
+static cpumask_t vector_allocation_domain(int cpu);
+
 /*
  * Legacy IRQ to IA-64 vector translation table.
  */
@@ -63,52 +72,365 @@ __u8 isa_irq_to_vector_map[16] = {
 };
 EXPORT_SYMBOL(isa_irq_to_vector_map);
 
-static unsigned long ia64_vector_mask[BITS_TO_LONGS(IA64_MAX_DEVICE_VECTORS)];
+DEFINE_SPINLOCK(vector_lock);
 
-int
-assign_irq_vector (int irq)
+struct irq_cfg irq_cfg[NR_IRQS] __read_mostly = {
+	[0 ... NR_IRQS - 1] = {
+		.vector = IRQ_VECTOR_UNASSIGNED,
+		.domain = CPU_MASK_NONE
+	}
+};
+
+DEFINE_PER_CPU(int[IA64_NUM_VECTORS], vector_irq) = {
+	[0 ... IA64_NUM_VECTORS - 1] = -1
+};
+
+static cpumask_t vector_table[IA64_NUM_VECTORS] = {
+	[0 ... IA64_NUM_VECTORS - 1] = CPU_MASK_NONE
+};
+
+static int irq_status[NR_IRQS] = {
+	[0 ... NR_IRQS -1] = IRQ_UNUSED
+};
+
+static inline int find_unassigned_irq(void)
 {
+	int irq;
+
+	for (irq = IA64_FIRST_DEVICE_VECTOR; irq < NR_IRQS; irq++)
+		if (irq_status[irq] == IRQ_UNUSED)
+			return irq;
+	return -ENOSPC;
+}
+
+static inline int find_unassigned_vector(cpumask_t domain)
+{
+	cpumask_t mask;
 	int pos, vector;
- again:
-	pos = find_first_zero_bit(ia64_vector_mask, IA64_NUM_DEVICE_VECTORS);
-	vector = IA64_FIRST_DEVICE_VECTOR + pos;
-	if (vector > IA64_LAST_DEVICE_VECTOR)
-		return -ENOSPC;
-	if (test_and_set_bit(pos, ia64_vector_mask))
-		goto again;
+
+	cpumask_and(&mask, &domain, cpu_online_mask);
+	if (cpus_empty(mask))
+		return -EINVAL;
+
+	for (pos = 0; pos < IA64_NUM_DEVICE_VECTORS; pos++) {
+		vector = IA64_FIRST_DEVICE_VECTOR + pos;
+		cpus_and(mask, domain, vector_table[vector]);
+		if (!cpus_empty(mask))
+			continue;
+		return vector;
+	}
+	return -ENOSPC;
+}
+
+static int __bind_irq_vector(int irq, int vector, cpumask_t domain)
+{
+	cpumask_t mask;
+	int cpu;
+	struct irq_cfg *cfg = &irq_cfg[irq];
+
+	BUG_ON((unsigned)irq >= NR_IRQS);
+	BUG_ON((unsigned)vector >= IA64_NUM_VECTORS);
+
+	cpumask_and(&mask, &domain, cpu_online_mask);
+	if (cpus_empty(mask))
+		return -EINVAL;
+	if ((cfg->vector == vector) && cpus_equal(cfg->domain, domain))
+		return 0;
+	if (cfg->vector != IRQ_VECTOR_UNASSIGNED)
+		return -EBUSY;
+	for_each_cpu_mask(cpu, mask)
+		per_cpu(vector_irq, cpu)[vector] = irq;
+	cfg->vector = vector;
+	cfg->domain = domain;
+	irq_status[irq] = IRQ_USED;
+	cpus_or(vector_table[vector], vector_table[vector], domain);
+	return 0;
+}
+
+int bind_irq_vector(int irq, int vector, cpumask_t domain)
+{
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&vector_lock, flags);
+	ret = __bind_irq_vector(irq, vector, domain);
+	spin_unlock_irqrestore(&vector_lock, flags);
+	return ret;
+}
+
+static void __clear_irq_vector(int irq)
+{
+	int vector, cpu;
+	cpumask_t mask;
+	cpumask_t domain;
+	struct irq_cfg *cfg = &irq_cfg[irq];
+
+	BUG_ON((unsigned)irq >= NR_IRQS);
+	BUG_ON(cfg->vector == IRQ_VECTOR_UNASSIGNED);
+	vector = cfg->vector;
+	domain = cfg->domain;
+	cpumask_and(&mask, &cfg->domain, cpu_online_mask);
+	for_each_cpu_mask(cpu, mask)
+		per_cpu(vector_irq, cpu)[vector] = -1;
+	cfg->vector = IRQ_VECTOR_UNASSIGNED;
+	cfg->domain = CPU_MASK_NONE;
+	irq_status[irq] = IRQ_UNUSED;
+	cpus_andnot(vector_table[vector], vector_table[vector], domain);
+}
+
+static void clear_irq_vector(int irq)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&vector_lock, flags);
+	__clear_irq_vector(irq);
+	spin_unlock_irqrestore(&vector_lock, flags);
+}
+
+int
+ia64_native_assign_irq_vector (int irq)
+{
+	unsigned long flags;
+	int vector, cpu;
+	cpumask_t domain = CPU_MASK_NONE;
+
+	vector = -ENOSPC;
+
+	spin_lock_irqsave(&vector_lock, flags);
+	for_each_online_cpu(cpu) {
+		domain = vector_allocation_domain(cpu);
+		vector = find_unassigned_vector(domain);
+		if (vector >= 0)
+			break;
+	}
+	if (vector < 0)
+		goto out;
+	if (irq == AUTO_ASSIGN)
+		irq = vector;
+	BUG_ON(__bind_irq_vector(irq, vector, domain));
+ out:
+	spin_unlock_irqrestore(&vector_lock, flags);
 	return vector;
 }
 
 void
-free_irq_vector (int vector)
+ia64_native_free_irq_vector (int vector)
 {
-	int pos;
-
-	if (vector < IA64_FIRST_DEVICE_VECTOR || vector > IA64_LAST_DEVICE_VECTOR)
+	if (vector < IA64_FIRST_DEVICE_VECTOR ||
+	    vector > IA64_LAST_DEVICE_VECTOR)
 		return;
-
-	pos = vector - IA64_FIRST_DEVICE_VECTOR;
-	if (!test_and_clear_bit(pos, ia64_vector_mask))
-		printk(KERN_WARNING "%s: double free!\n", __FUNCTION__);
+	clear_irq_vector(vector);
 }
 
 int
 reserve_irq_vector (int vector)
 {
-	int pos;
-
 	if (vector < IA64_FIRST_DEVICE_VECTOR ||
 	    vector > IA64_LAST_DEVICE_VECTOR)
 		return -EINVAL;
+	return !!bind_irq_vector(vector, vector, CPU_MASK_ALL);
+}
+
+/*
+ * Initialize vector_irq on a new cpu. This function must be called
+ * with vector_lock held.
+ */
+void __setup_vector_irq(int cpu)
+{
+	int irq, vector;
+
+	/* Clear vector_irq */
+	for (vector = 0; vector < IA64_NUM_VECTORS; ++vector)
+		per_cpu(vector_irq, cpu)[vector] = -1;
+	/* Mark the inuse vectors */
+	for (irq = 0; irq < NR_IRQS; ++irq) {
+		if (!cpu_isset(cpu, irq_cfg[irq].domain))
+			continue;
+		vector = irq_to_vector(irq);
+		per_cpu(vector_irq, cpu)[vector] = irq;
+	}
+}
+
+#if defined(CONFIG_SMP) && (defined(CONFIG_IA64_GENERIC) || defined(CONFIG_IA64_DIG))
+
+static enum vector_domain_type {
+	VECTOR_DOMAIN_NONE,
+	VECTOR_DOMAIN_PERCPU
+} vector_domain_type = VECTOR_DOMAIN_NONE;
+
+static cpumask_t vector_allocation_domain(int cpu)
+{
+	if (vector_domain_type == VECTOR_DOMAIN_PERCPU)
+		return cpumask_of_cpu(cpu);
+	return CPU_MASK_ALL;
+}
+
+static int __irq_prepare_move(int irq, int cpu)
+{
+	struct irq_cfg *cfg = &irq_cfg[irq];
+	int vector;
+	cpumask_t domain;
+
+	if (cfg->move_in_progress || cfg->move_cleanup_count)
+		return -EBUSY;
+	if (cfg->vector == IRQ_VECTOR_UNASSIGNED || !cpu_online(cpu))
+		return -EINVAL;
+	if (cpu_isset(cpu, cfg->domain))
+		return 0;
+	domain = vector_allocation_domain(cpu);
+	vector = find_unassigned_vector(domain);
+	if (vector < 0)
+		return -ENOSPC;
+	cfg->move_in_progress = 1;
+	cfg->old_domain = cfg->domain;
+	cfg->vector = IRQ_VECTOR_UNASSIGNED;
+	cfg->domain = CPU_MASK_NONE;
+	BUG_ON(__bind_irq_vector(irq, vector, domain));
+	return 0;
+}
+
+int irq_prepare_move(int irq, int cpu)
+{
+	unsigned long flags;
+	int ret;
 
-	pos = vector - IA64_FIRST_DEVICE_VECTOR;
-	return test_and_set_bit(pos, ia64_vector_mask);
+	spin_lock_irqsave(&vector_lock, flags);
+	ret = __irq_prepare_move(irq, cpu);
+	spin_unlock_irqrestore(&vector_lock, flags);
+	return ret;
+}
+
+void irq_complete_move(unsigned irq)
+{
+	struct irq_cfg *cfg = &irq_cfg[irq];
+	cpumask_t cleanup_mask;
+	int i;
+
+	if (likely(!cfg->move_in_progress))
+		return;
+
+	if (unlikely(cpu_isset(smp_processor_id(), cfg->old_domain)))
+		return;
+
+	cpumask_and(&cleanup_mask, &cfg->old_domain, cpu_online_mask);
+	cfg->move_cleanup_count = cpus_weight(cleanup_mask);
+	for_each_cpu_mask(i, cleanup_mask)
+		platform_send_ipi(i, IA64_IRQ_MOVE_VECTOR, IA64_IPI_DM_INT, 0);
+	cfg->move_in_progress = 0;
+}
+
+static irqreturn_t smp_irq_move_cleanup_interrupt(int irq, void *dev_id)
+{
+	int me = smp_processor_id();
+	ia64_vector vector;
+	unsigned long flags;
+
+	for (vector = IA64_FIRST_DEVICE_VECTOR;
+	     vector < IA64_LAST_DEVICE_VECTOR; vector++) {
+		int irq;
+		struct irq_desc *desc;
+		struct irq_cfg *cfg;
+		irq = __get_cpu_var(vector_irq)[vector];
+		if (irq < 0)
+			continue;
+
+		desc = irq_to_desc(irq);
+		cfg = irq_cfg + irq;
+		raw_spin_lock(&desc->lock);
+		if (!cfg->move_cleanup_count)
+			goto unlock;
+
+		if (!cpu_isset(me, cfg->old_domain))
+			goto unlock;
+
+		spin_lock_irqsave(&vector_lock, flags);
+		__get_cpu_var(vector_irq)[vector] = -1;
+		cpu_clear(me, vector_table[vector]);
+		spin_unlock_irqrestore(&vector_lock, flags);
+		cfg->move_cleanup_count--;
+	unlock:
+		raw_spin_unlock(&desc->lock);
+	}
+	return IRQ_HANDLED;
+}
+
+static struct irqaction irq_move_irqaction = {
+	.handler =	smp_irq_move_cleanup_interrupt,
+	.name =		"irq_move"
+};
+
+static int __init parse_vector_domain(char *arg)
+{
+	if (!arg)
+		return -EINVAL;
+	if (!strcmp(arg, "percpu")) {
+		vector_domain_type = VECTOR_DOMAIN_PERCPU;
+		no_int_routing = 1;
+	}
+	return 0;
+}
+early_param("vector", parse_vector_domain);
+#else
+static cpumask_t vector_allocation_domain(int cpu)
+{
+	return CPU_MASK_ALL;
+}
+#endif
+
+
+void destroy_and_reserve_irq(unsigned int irq)
+{
+	unsigned long flags;
+
+	irq_init_desc(irq);
+	spin_lock_irqsave(&vector_lock, flags);
+	__clear_irq_vector(irq);
+	irq_status[irq] = IRQ_RSVD;
+	spin_unlock_irqrestore(&vector_lock, flags);
+}
+
+/*
+ * Dynamic irq allocate and deallocation for MSI
+ */
+int create_irq(void)
+{
+	unsigned long flags;
+	int irq, vector, cpu;
+	cpumask_t domain = CPU_MASK_NONE;
+
+	irq = vector = -ENOSPC;
+	spin_lock_irqsave(&vector_lock, flags);
+	for_each_online_cpu(cpu) {
+		domain = vector_allocation_domain(cpu);
+		vector = find_unassigned_vector(domain);
+		if (vector >= 0)
+			break;
+	}
+	if (vector < 0)
+		goto out;
+	irq = find_unassigned_irq();
+	if (irq < 0)
+		goto out;
+	BUG_ON(__bind_irq_vector(irq, vector, domain));
+ out:
+	spin_unlock_irqrestore(&vector_lock, flags);
+	if (irq >= 0)
+		irq_init_desc(irq);
+	return irq;
+}
+
+void destroy_irq(unsigned int irq)
+{
+	irq_init_desc(irq);
+	clear_irq_vector(irq);
 }
 
 #ifdef CONFIG_SMP
 #	define IS_RESCHEDULE(vec)	(vec == IA64_IPI_RESCHEDULE)
+#	define IS_LOCAL_TLB_FLUSH(vec)	(vec == IA64_IPI_LOCAL_TLB_FLUSH)
 #else
 #	define IS_RESCHEDULE(vec)	(0)
+#	define IS_LOCAL_TLB_FLUSH(vec)	(0)
 #endif
 /*
  * That's where the IVT branches when we get an external
@@ -118,6 +440,7 @@ reserve_irq_vector (int vector)
 void
 ia64_handle_irq (ia64_vector vector, struct pt_regs *regs)
 {
+	struct pt_regs *old_regs = set_irq_regs(regs);
 	unsigned long saved_tpr;
 
 #if IRQ_DEBUG
@@ -135,13 +458,9 @@ ia64_handle_irq (ia64_vector vector, struct pt_regs *regs)
 		sp = ia64_getreg(_IA64_REG_SP);
 
 		if ((sp - bsp) < 1024) {
-			static unsigned char count;
-			static long last_time;
+			static DEFINE_RATELIMIT_STATE(ratelimit, 5 * HZ, 5);
 
-			if (jiffies - last_time > 5*HZ)
-				count = 0;
-			if (++count < 5) {
-				last_time = jiffies;
+			if (__ratelimit(&ratelimit)) {
 				printk("ia64_handle_irq: DANGER: less than "
 				       "1KB of free stack space!!\n"
 				       "(bsp=0x%lx, sp=%lx)\n", bsp, sp);
@@ -159,11 +478,25 @@ ia64_handle_irq (ia64_vector vector, struct pt_regs *regs)
 	saved_tpr = ia64_getreg(_IA64_REG_CR_TPR);
 	ia64_srlz_d();
 	while (vector != IA64_SPURIOUS_INT_VECTOR) {
-		if (!IS_RESCHEDULE(vector)) {
+		int irq = local_vector_to_irq(vector);
+
+		if (unlikely(IS_LOCAL_TLB_FLUSH(vector))) {
+			smp_local_flush_tlb();
+			kstat_incr_irq_this_cpu(irq);
+		} else if (unlikely(IS_RESCHEDULE(vector))) {
+			scheduler_ipi();
+			kstat_incr_irq_this_cpu(irq);
+		} else {
 			ia64_setreg(_IA64_REG_CR_TPR, vector);
 			ia64_srlz_d();
 
-			__do_IRQ(local_vector_to_irq(vector), regs);
+			if (unlikely(irq < 0)) {
+				printk(KERN_ERR "%s: Unexpected interrupt "
+				       "vector %d on CPU %d is not mapped "
+				       "to any IRQ!\n", __func__, vector,
+				       smp_processor_id());
+			} else
+				generic_handle_irq(irq);
 
 			/*
 			 * Disable interrupts and send EOI:
@@ -180,6 +513,7 @@ ia64_handle_irq (ia64_vector vector, struct pt_regs *regs)
 	 * come through until ia64_eoi() has been done.
 	 */
 	irq_exit();
+	set_irq_regs(old_regs);
 }
 
 #ifdef CONFIG_HOTPLUG_CPU
@@ -195,15 +529,24 @@ void ia64_process_pending_intr(void)
 
 	vector = ia64_get_ivr();
 
-	 irq_enter();
-	 saved_tpr = ia64_getreg(_IA64_REG_CR_TPR);
-	 ia64_srlz_d();
+	irq_enter();
+	saved_tpr = ia64_getreg(_IA64_REG_CR_TPR);
+	ia64_srlz_d();
 
 	 /*
 	  * Perform normal interrupt style processing
 	  */
 	while (vector != IA64_SPURIOUS_INT_VECTOR) {
-		if (!IS_RESCHEDULE(vector)) {
+		int irq = local_vector_to_irq(vector);
+
+		if (unlikely(IS_LOCAL_TLB_FLUSH(vector))) {
+			smp_local_flush_tlb();
+			kstat_incr_irq_this_cpu(irq);
+		} else if (unlikely(IS_RESCHEDULE(vector))) {
+			kstat_incr_irq_this_cpu(irq);
+		} else {
+			struct pt_regs *old_regs = set_irq_regs(NULL);
+
 			ia64_setreg(_IA64_REG_CR_TPR, vector);
 			ia64_srlz_d();
 
@@ -213,8 +556,16 @@ void ia64_process_pending_intr(void)
 			 * it will work. I hope it works!.
 			 * Probably could shared code.
 			 */
-			vectors_in_migration[local_vector_to_irq(vector)]=0;
-			__do_IRQ(local_vector_to_irq(vector), NULL);
+			if (unlikely(irq < 0)) {
+				printk(KERN_ERR "%s: Unexpected interrupt "
+				       "vector %d on CPU %d not being mapped "
+				       "to any IRQ!!\n", __func__, vector,
+				       smp_processor_id());
+			} else {
+				vectors_in_migration[irq]=0;
+				generic_handle_irq(irq);
+			}
+			set_irq_regs(old_regs);
 
 			/*
 			 * Disable interrupts and send EOI
@@ -231,37 +582,69 @@ void ia64_process_pending_intr(void)
 
 
 #ifdef CONFIG_SMP
-extern irqreturn_t handle_IPI (int irq, void *dev_id, struct pt_regs *regs);
+
+static irqreturn_t dummy_handler (int irq, void *dev_id)
+{
+	BUG();
+}
 
 static struct irqaction ipi_irqaction = {
 	.handler =	handle_IPI,
-	.flags =	IRQF_DISABLED,
 	.name =		"IPI"
 };
+
+/*
+ * KVM uses this interrupt to force a cpu out of guest mode
+ */
+static struct irqaction resched_irqaction = {
+	.handler =	dummy_handler,
+	.name =		"resched"
+};
+
+static struct irqaction tlb_irqaction = {
+	.handler =	dummy_handler,
+	.name =		"tlb_flush"
+};
+
 #endif
 
 void
-register_percpu_irq (ia64_vector vec, struct irqaction *action)
+ia64_native_register_percpu_irq (ia64_vector vec, struct irqaction *action)
 {
-	irq_desc_t *desc;
 	unsigned int irq;
 
-	for (irq = 0; irq < NR_IRQS; ++irq)
-		if (irq_to_vector(irq) == vec) {
-			desc = irq_desc + irq;
-			desc->status |= IRQ_PER_CPU;
-			desc->chip = &irq_type_ia64_lsapic;
-			if (action)
-				setup_irq(irq, action);
-		}
+	irq = vec;
+	BUG_ON(bind_irq_vector(irq, vec, CPU_MASK_ALL));
+	irq_set_status_flags(irq, IRQ_PER_CPU);
+	irq_set_chip(irq, &irq_type_ia64_lsapic);
+	if (action)
+		setup_irq(irq, action);
+	irq_set_handler(irq, handle_percpu_irq);
+}
+
+void __init
+ia64_native_register_ipi(void)
+{
+#ifdef CONFIG_SMP
+	register_percpu_irq(IA64_IPI_VECTOR, &ipi_irqaction);
+	register_percpu_irq(IA64_IPI_RESCHEDULE, &resched_irqaction);
+	register_percpu_irq(IA64_IPI_LOCAL_TLB_FLUSH, &tlb_irqaction);
+#endif
 }
 
 void __init
 init_IRQ (void)
 {
+#ifdef CONFIG_ACPI
+	acpi_boot_init();
+#endif
+	ia64_register_ipi();
 	register_percpu_irq(IA64_SPURIOUS_INT_VECTOR, NULL);
 #ifdef CONFIG_SMP
-	register_percpu_irq(IA64_IPI_VECTOR, &ipi_irqaction);
+#if defined(CONFIG_IA64_GENERIC) || defined(CONFIG_IA64_DIG)
+	if (vector_domain_type != VECTOR_DOMAIN_NONE)
+		register_percpu_irq(IA64_IRQ_MOVE_VECTOR, &irq_move_irqaction);
+#endif
 #endif
 #ifdef CONFIG_PERFMON
 	pfm_init_percpu();
@@ -276,11 +659,7 @@ ia64_send_ipi (int cpu, int vector, int delivery_mode, int redirect)
 	unsigned long ipi_data;
 	unsigned long phys_cpu_id;
 
-#ifdef CONFIG_SMP
 	phys_cpu_id = cpu_physical_id(cpu);
-#else
-	phys_cpu_id = (ia64_getreg(_IA64_REG_CR_LID) >> 16) & 0xffff;
-#endif
 
 	/*
 	 * cpu number is in 8bit ID and 8bit EID