Pull vector-domain into release branch

Documentation/kernel-parameters.txt

@@ -1885,6 +1885,9 @@ and is between 256 and 4096 characters. It is defined in the file
 			vdso=1: enable VDSO (default)
 			vdso=0: disable VDSO mapping
 
+	vector=		[IA-64,SMP]
+			vector=percpu: enable percpu vector domain
+
 	video=		[FB] Frame buffer configuration
 			See Documentation/fb/modedb.txt.
 

arch/ia64/kernel/iosapic.c

@@ -118,15 +118,25 @@ static DEFINE_SPINLOCK(iosapic_lock);
  * vector.
  */
 
-struct iosapic_rte_info {
-	struct list_head rte_list;	/* node in list of RTEs sharing the
-					 * same vector */
+#define NO_REF_RTE	0
+
+static struct iosapic {
 	char __iomem	*addr;		/* base address of IOSAPIC */
-	unsigned int	gsi_base;	/* first GSI assigned to this
-					 * IOSAPIC */
+	unsigned int	gsi_base;	/* GSI base */
+	unsigned short	num_rte;	/* # of RTEs on this IOSAPIC */
+	int		rtes_inuse;	/* # of RTEs in use on this IOSAPIC */
+#ifdef CONFIG_NUMA
+	unsigned short	node;		/* numa node association via pxm */
+#endif
+	spinlock_t	lock;		/* lock for indirect reg access */
+} iosapic_lists[NR_IOSAPICS];
+
+struct iosapic_rte_info {
+	struct list_head rte_list;	/* RTEs sharing the same vector */
 	char		rte_index;	/* IOSAPIC RTE index */
 	int		refcnt;		/* reference counter */
 	unsigned int	flags;		/* flags */
+	struct iosapic	*iosapic;
 } ____cacheline_aligned;
 
 static struct iosapic_intr_info {
@@ -140,24 +150,23 @@ static struct iosapic_intr_info {
 	unsigned char 	polarity: 1;	/* interrupt polarity
 					 * (see iosapic.h) */
 	unsigned char	trigger	: 1;	/* trigger mode (see iosapic.h) */
-} iosapic_intr_info[IA64_NUM_VECTORS];
-
-static struct iosapic {
-	char __iomem	*addr;		/* base address of IOSAPIC */
-	unsigned int 	gsi_base;	/* first GSI assigned to this
-					 * IOSAPIC */
-	unsigned short 	num_rte;	/* # of RTEs on this IOSAPIC */
-	int		rtes_inuse;	/* # of RTEs in use on this IOSAPIC */
-#ifdef CONFIG_NUMA
-	unsigned short	node;		/* numa node association via pxm */
-#endif
-} iosapic_lists[NR_IOSAPICS];
+} iosapic_intr_info[NR_IRQS];
 
 static unsigned char pcat_compat __devinitdata;	/* 8259 compatibility flag */
 
 static int iosapic_kmalloc_ok;
 static LIST_HEAD(free_rte_list);
 
+static inline void
+iosapic_write(struct iosapic *iosapic, unsigned int reg, u32 val)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&iosapic->lock, flags);
+	__iosapic_write(iosapic->addr, reg, val);
+	spin_unlock_irqrestore(&iosapic->lock, flags);
+}
+
 /*
  * Find an IOSAPIC associated with a GSI
  */
@@ -175,17 +184,18 @@ find_iosapic (unsigned int gsi)
 	return -1;
 }
 
-static inline int
-_gsi_to_vector (unsigned int gsi)
+static inline int __gsi_to_irq(unsigned int gsi)
 {
+	int irq;
 	struct iosapic_intr_info *info;
 	struct iosapic_rte_info *rte;
 
-	for (info = iosapic_intr_info; info <
-		     iosapic_intr_info + IA64_NUM_VECTORS; ++info)
+	for (irq = 0; irq < NR_IRQS; irq++) {
+		info = &iosapic_intr_info[irq];
 		list_for_each_entry(rte, &info->rtes, rte_list)
-			if (rte->gsi_base + rte->rte_index == gsi)
-				return info - iosapic_intr_info;
+			if (rte->iosapic->gsi_base + rte->rte_index == gsi)
+				return irq;
+	}
 	return -1;
 }
 
@@ -196,7 +206,10 @@ _gsi_to_vector (unsigned int gsi)
 inline int
 gsi_to_vector (unsigned int gsi)
 {
-	return _gsi_to_vector(gsi);
+	int irq = __gsi_to_irq(gsi);
+	if (check_irq_used(irq) < 0)
+		return -1;
+	return irq_to_vector(irq);
 }
 
 int
@@ -204,66 +217,48 @@ gsi_to_irq (unsigned int gsi)
 {
 	unsigned long flags;
 	int irq;
-	/*
-	 * XXX fix me: this assumes an identity mapping between IA-64 vector
-	 * and Linux irq numbers...
-	 */
+
 	spin_lock_irqsave(&iosapic_lock, flags);
-	{
-		irq = _gsi_to_vector(gsi);
-	}
+	irq = __gsi_to_irq(gsi);
 	spin_unlock_irqrestore(&iosapic_lock, flags);
-
 	return irq;
 }
 
-static struct iosapic_rte_info *gsi_vector_to_rte(unsigned int gsi,
-						  unsigned int vec)
+static struct iosapic_rte_info *find_rte(unsigned int irq, unsigned int gsi)
 {
 	struct iosapic_rte_info *rte;
 
-	list_for_each_entry(rte, &iosapic_intr_info[vec].rtes, rte_list)
-		if (rte->gsi_base + rte->rte_index == gsi)
+	list_for_each_entry(rte, &iosapic_intr_info[irq].rtes, rte_list)
+		if (rte->iosapic->gsi_base + rte->rte_index == gsi)
 			return rte;
 	return NULL;
 }
 
 static void
-set_rte (unsigned int gsi, unsigned int vector, unsigned int dest, int mask)
+set_rte (unsigned int gsi, unsigned int irq, unsigned int dest, int mask)
 {
 	unsigned long pol, trigger, dmode;
 	u32 low32, high32;
-	char __iomem *addr;
 	int rte_index;
 	char redir;
 	struct iosapic_rte_info *rte;
+	ia64_vector vector = irq_to_vector(irq);
 
 	DBG(KERN_DEBUG"IOSAPIC: routing vector %d to 0x%x\n", vector, dest);
 
-	rte = gsi_vector_to_rte(gsi, vector);
+	rte = find_rte(irq, gsi);
 	if (!rte)
 		return;		/* not an IOSAPIC interrupt */
 
 	rte_index = rte->rte_index;
-	addr	= rte->addr;
-	pol     = iosapic_intr_info[vector].polarity;
-	trigger = iosapic_intr_info[vector].trigger;
-	dmode   = iosapic_intr_info[vector].dmode;
+	pol     = iosapic_intr_info[irq].polarity;
+	trigger = iosapic_intr_info[irq].trigger;
+	dmode   = iosapic_intr_info[irq].dmode;
 
 	redir = (dmode == IOSAPIC_LOWEST_PRIORITY) ? 1 : 0;
 
 #ifdef CONFIG_SMP
-	{
-		unsigned int irq;
-
-		for (irq = 0; irq < NR_IRQS; ++irq)
-			if (irq_to_vector(irq) == vector) {
-				set_irq_affinity_info(irq,
-						      (int)(dest & 0xffff),
-						      redir);
-				break;
-			}
-	}
+	set_irq_affinity_info(irq, (int)(dest & 0xffff), redir);
 #endif
 
 	low32 = ((pol << IOSAPIC_POLARITY_SHIFT) |
@@ -275,10 +270,10 @@ set_rte (unsigned int gsi, unsigned int vector, unsigned int dest, int mask)
 	/* dest contains both id and eid */
 	high32 = (dest << IOSAPIC_DEST_SHIFT);
 
-	iosapic_write(addr, IOSAPIC_RTE_HIGH(rte_index), high32);
-	iosapic_write(addr, IOSAPIC_RTE_LOW(rte_index), low32);
-	iosapic_intr_info[vector].low32 = low32;
-	iosapic_intr_info[vector].dest = dest;
+	iosapic_write(rte->iosapic, IOSAPIC_RTE_HIGH(rte_index), high32);
+	iosapic_write(rte->iosapic, IOSAPIC_RTE_LOW(rte_index), low32);
+	iosapic_intr_info[irq].low32 = low32;
+	iosapic_intr_info[irq].dest = dest;
 }
 
 static void
@@ -294,15 +289,18 @@ kexec_disable_iosapic(void)
 {
 	struct iosapic_intr_info *info;
 	struct iosapic_rte_info *rte;
-	u8 vec = 0;
-	for (info = iosapic_intr_info; info <
-			iosapic_intr_info + IA64_NUM_VECTORS; ++info, ++vec) {
+	ia64_vector vec;
+	int irq;
+
+	for (irq = 0; irq < NR_IRQS; irq++) {
+		info = &iosapic_intr_info[irq];
+		vec = irq_to_vector(irq);
 		list_for_each_entry(rte, &info->rtes,
 				rte_list) {
-			iosapic_write(rte->addr,
+			iosapic_write(rte->iosapic,
 					IOSAPIC_RTE_LOW(rte->rte_index),
 					IOSAPIC_MASK|vec);
-			iosapic_eoi(rte->addr, vec);
+			iosapic_eoi(rte->iosapic->addr, vec);
 		}
 	}
 }
@@ -311,54 +309,36 @@ kexec_disable_iosapic(void)
 static void
 mask_irq (unsigned int irq)
 {
-	unsigned long flags;
-	char __iomem *addr;
 	u32 low32;
 	int rte_index;
-	ia64_vector vec = irq_to_vector(irq);
 	struct iosapic_rte_info *rte;
 
-	if (list_empty(&iosapic_intr_info[vec].rtes))
+	if (list_empty(&iosapic_intr_info[irq].rtes))
 		return;			/* not an IOSAPIC interrupt! */
 
-	spin_lock_irqsave(&iosapic_lock, flags);
-	{
-		/* set only the mask bit */
-		low32 = iosapic_intr_info[vec].low32 |= IOSAPIC_MASK;
-		list_for_each_entry(rte, &iosapic_intr_info[vec].rtes,
-				    rte_list) {
-			addr = rte->addr;
-			rte_index = rte->rte_index;
-			iosapic_write(addr, IOSAPIC_RTE_LOW(rte_index), low32);
-		}
+	/* set only the mask bit */
+	low32 = iosapic_intr_info[irq].low32 |= IOSAPIC_MASK;
+	list_for_each_entry(rte, &iosapic_intr_info[irq].rtes, rte_list) {
+		rte_index = rte->rte_index;
+		iosapic_write(rte->iosapic, IOSAPIC_RTE_LOW(rte_index), low32);
 	}
-	spin_unlock_irqrestore(&iosapic_lock, flags);
 }
 
 static void
 unmask_irq (unsigned int irq)
 {
-	unsigned long flags;
-	char __iomem *addr;
 	u32 low32;
 	int rte_index;
-	ia64_vector vec = irq_to_vector(irq);
 	struct iosapic_rte_info *rte;
 
-	if (list_empty(&iosapic_intr_info[vec].rtes))
+	if (list_empty(&iosapic_intr_info[irq].rtes))
 		return;			/* not an IOSAPIC interrupt! */
 
-	spin_lock_irqsave(&iosapic_lock, flags);
-	{
-		low32 = iosapic_intr_info[vec].low32 &= ~IOSAPIC_MASK;
-		list_for_each_entry(rte, &iosapic_intr_info[vec].rtes,
-				    rte_list) {
-			addr = rte->addr;
-			rte_index = rte->rte_index;
-			iosapic_write(addr, IOSAPIC_RTE_LOW(rte_index), low32);
-		}
+	low32 = iosapic_intr_info[irq].low32 &= ~IOSAPIC_MASK;
+	list_for_each_entry(rte, &iosapic_intr_info[irq].rtes, rte_list) {
+		rte_index = rte->rte_index;
+		iosapic_write(rte->iosapic, IOSAPIC_RTE_LOW(rte_index), low32);
 	}
-	spin_unlock_irqrestore(&iosapic_lock, flags);
 }
 
 
@@ -366,23 +346,24 @@ static void
 iosapic_set_affinity (unsigned int irq, cpumask_t mask)
 {
 #ifdef CONFIG_SMP
-	unsigned long flags;
 	u32 high32, low32;
 	int dest, rte_index;
-	char __iomem *addr;
 	int redir = (irq & IA64_IRQ_REDIRECTED) ? 1 : 0;
-	ia64_vector vec;
 	struct iosapic_rte_info *rte;
+	struct iosapic *iosapic;
 
 	irq &= (~IA64_IRQ_REDIRECTED);
-	vec = irq_to_vector(irq);
 
+	cpus_and(mask, mask, cpu_online_map);
 	if (cpus_empty(mask))
 		return;
 
+	if (reassign_irq_vector(irq, first_cpu(mask)))
+		return;
+
 	dest = cpu_physical_id(first_cpu(mask));
 
-	if (list_empty(&iosapic_intr_info[vec].rtes))
+	if (list_empty(&iosapic_intr_info[irq].rtes))
 		return;			/* not an IOSAPIC interrupt */
 
 	set_irq_affinity_info(irq, dest, redir);
@@ -390,31 +371,24 @@ iosapic_set_affinity (unsigned int irq, cpumask_t mask)
 	/* dest contains both id and eid */
 	high32 = dest << IOSAPIC_DEST_SHIFT;
 
-	spin_lock_irqsave(&iosapic_lock, flags);
-	{
-		low32 = iosapic_intr_info[vec].low32 &
-			~(7 << IOSAPIC_DELIVERY_SHIFT);
-
-		if (redir)
-		        /* change delivery mode to lowest priority */
-			low32 |= (IOSAPIC_LOWEST_PRIORITY <<
-				  IOSAPIC_DELIVERY_SHIFT);
-		else
-		        /* change delivery mode to fixed */
-			low32 |= (IOSAPIC_FIXED << IOSAPIC_DELIVERY_SHIFT);
-
-		iosapic_intr_info[vec].low32 = low32;
-		iosapic_intr_info[vec].dest = dest;
-		list_for_each_entry(rte, &iosapic_intr_info[vec].rtes,
-				    rte_list) {
-			addr = rte->addr;
-			rte_index = rte->rte_index;
-			iosapic_write(addr, IOSAPIC_RTE_HIGH(rte_index),
-				      high32);
-			iosapic_write(addr, IOSAPIC_RTE_LOW(rte_index), low32);
-		}
+	low32 = iosapic_intr_info[irq].low32 & ~(7 << IOSAPIC_DELIVERY_SHIFT);
+	if (redir)
+		/* change delivery mode to lowest priority */
+		low32 |= (IOSAPIC_LOWEST_PRIORITY << IOSAPIC_DELIVERY_SHIFT);
+	else
+		/* change delivery mode to fixed */
+		low32 |= (IOSAPIC_FIXED << IOSAPIC_DELIVERY_SHIFT);
+	low32 &= IOSAPIC_VECTOR_MASK;
+	low32 |= irq_to_vector(irq);
+
+	iosapic_intr_info[irq].low32 = low32;
+	iosapic_intr_info[irq].dest = dest;
+	list_for_each_entry(rte, &iosapic_intr_info[irq].rtes, rte_list) {
+		iosapic = rte->iosapic;
+		rte_index = rte->rte_index;
+		iosapic_write(iosapic, IOSAPIC_RTE_HIGH(rte_index), high32);
+		iosapic_write(iosapic, IOSAPIC_RTE_LOW(rte_index), low32);
 	}
-	spin_unlock_irqrestore(&iosapic_lock, flags);
 #endif
 }
 
@@ -434,10 +408,20 @@ iosapic_end_level_irq (unsigned int irq)
 {
 	ia64_vector vec = irq_to_vector(irq);
 	struct iosapic_rte_info *rte;
+	int do_unmask_irq = 0;
 
-	move_native_irq(irq);
-	list_for_each_entry(rte, &iosapic_intr_info[vec].rtes, rte_list)
-		iosapic_eoi(rte->addr, vec);
+	if (unlikely(irq_desc[irq].status & IRQ_MOVE_PENDING)) {
+		do_unmask_irq = 1;
+		mask_irq(irq);
+	}
+
+	list_for_each_entry(rte, &iosapic_intr_info[irq].rtes, rte_list)
+		iosapic_eoi(rte->iosapic->addr, vec);
+
+	if (unlikely(do_unmask_irq)) {
+		move_masked_irq(irq);
+		unmask_irq(irq);
+	}
 }
 
 #define iosapic_shutdown_level_irq	mask_irq
@@ -519,13 +503,12 @@ iosapic_version (char __iomem *addr)
 	 *	unsigned int reserved2 : 8;
 	 * }
 	 */
-	return iosapic_read(addr, IOSAPIC_VERSION);
+	return __iosapic_read(addr, IOSAPIC_VERSION);
 }
 
-static int iosapic_find_sharable_vector (unsigned long trigger,
-					 unsigned long pol)
+static int iosapic_find_sharable_irq(unsigned long trigger, unsigned long pol)
 {
-	int i, vector = -1, min_count = -1;
+	int i, irq = -ENOSPC, min_count = -1;
 	struct iosapic_intr_info *info;
 
 	/*
@@ -533,21 +516,21 @@ static int iosapic_find_sharable_vector (unsigned long trigger,
 	 * supported yet
 	 */
 	if (trigger == IOSAPIC_EDGE)
-		return -1;
+		return -EINVAL;
 
-	for (i = IA64_FIRST_DEVICE_VECTOR; i <= IA64_LAST_DEVICE_VECTOR; i++) {
+	for (i = 0; i <= NR_IRQS; i++) {
 		info = &iosapic_intr_info[i];
 		if (info->trigger == trigger && info->polarity == pol &&
-		    (info->dmode == IOSAPIC_FIXED || info->dmode ==
-		     IOSAPIC_LOWEST_PRIORITY)) {
+		    (info->dmode == IOSAPIC_FIXED ||
+		     info->dmode == IOSAPIC_LOWEST_PRIORITY) &&
+		    can_request_irq(i, IRQF_SHARED)) {
 			if (min_count == -1 || info->count < min_count) {
-				vector = i;
+				irq = i;
 				min_count = info->count;
 			}
 		}
 	}
-
-	return vector;
+	return irq;
 }
 
 /*
@@ -555,25 +538,25 @@ static int iosapic_find_sharable_vector (unsigned long trigger,
  *  assign a new vector for the other and make the vector available
  */
 static void __init
-iosapic_reassign_vector (int vector)
+iosapic_reassign_vector (int irq)
 {
-	int new_vector;
+	int new_irq;
 
-	if (!list_empty(&iosapic_intr_info[vector].rtes)) {
-		new_vector = assign_irq_vector(AUTO_ASSIGN);
-		if (new_vector < 0)
+	if (!list_empty(&iosapic_intr_info[irq].rtes)) {
+		new_irq = create_irq();
+		if (new_irq < 0)
 			panic("%s: out of interrupt vectors!\n", __FUNCTION__);
 		printk(KERN_INFO "Reassigning vector %d to %d\n",
-		       vector, new_vector);
-		memcpy(&iosapic_intr_info[new_vector], &iosapic_intr_info[vector],
+		       irq_to_vector(irq), irq_to_vector(new_irq));
+		memcpy(&iosapic_intr_info[new_irq], &iosapic_intr_info[irq],
 		       sizeof(struct iosapic_intr_info));
-		INIT_LIST_HEAD(&iosapic_intr_info[new_vector].rtes);
-		list_move(iosapic_intr_info[vector].rtes.next,
-			  &iosapic_intr_info[new_vector].rtes);
-		memset(&iosapic_intr_info[vector], 0,
+		INIT_LIST_HEAD(&iosapic_intr_info[new_irq].rtes);
+		list_move(iosapic_intr_info[irq].rtes.next,
+			  &iosapic_intr_info[new_irq].rtes);
+		memset(&iosapic_intr_info[irq], 0,
 		       sizeof(struct iosapic_intr_info));
-		iosapic_intr_info[vector].low32 = IOSAPIC_MASK;
-		INIT_LIST_HEAD(&iosapic_intr_info[vector].rtes);
+		iosapic_intr_info[irq].low32 = IOSAPIC_MASK;
+		INIT_LIST_HEAD(&iosapic_intr_info[irq].rtes);
 	}
 }
 
@@ -610,29 +593,18 @@ static struct iosapic_rte_info *iosapic_alloc_rte (void)
 	return rte;
 }
 
-static void iosapic_free_rte (struct iosapic_rte_info *rte)
+static inline int irq_is_shared (int irq)
 {
-	if (rte->flags & RTE_PREALLOCATED)
-		list_add_tail(&rte->rte_list, &free_rte_list);
-	else
-		kfree(rte);
-}
-
-static inline int vector_is_shared (int vector)
-{
-	return (iosapic_intr_info[vector].count > 1);
+	return (iosapic_intr_info[irq].count > 1);
 }
 
 static int
-register_intr (unsigned int gsi, int vector, unsigned char delivery,
+register_intr (unsigned int gsi, int irq, unsigned char delivery,
 	       unsigned long polarity, unsigned long trigger)
 {
 	irq_desc_t *idesc;
 	struct hw_interrupt_type *irq_type;
-	int rte_index;
 	int index;
-	unsigned long gsi_base;
-	void __iomem *iosapic_address;
 	struct iosapic_rte_info *rte;
 
 	index = find_iosapic(gsi);
@@ -642,10 +614,7 @@ register_intr (unsigned int gsi, int vector, unsigned char delivery,
 		return -ENODEV;
 	}
 
-	iosapic_address = iosapic_lists[index].addr;
-	gsi_base = iosapic_lists[index].gsi_base;
-
-	rte = gsi_vector_to_rte(gsi, vector);
+	rte = find_rte(irq, gsi);
 	if (!rte) {
 		rte = iosapic_alloc_rte();
 		if (!rte) {
@@ -654,40 +623,42 @@ register_intr (unsigned int gsi, int vector, unsigned char delivery,
 			return -ENOMEM;
 		}
 
-		rte_index = gsi - gsi_base;
-		rte->rte_index	= rte_index;
-		rte->addr	= iosapic_address;
-		rte->gsi_base	= gsi_base;
+		rte->iosapic	= &iosapic_lists[index];
+		rte->rte_index	= gsi - rte->iosapic->gsi_base;
 		rte->refcnt++;
-		list_add_tail(&rte->rte_list, &iosapic_intr_info[vector].rtes);
-		iosapic_intr_info[vector].count++;
+		list_add_tail(&rte->rte_list, &iosapic_intr_info[irq].rtes);
+		iosapic_intr_info[irq].count++;
 		iosapic_lists[index].rtes_inuse++;
 	}
-	else if (vector_is_shared(vector)) {
-		struct iosapic_intr_info *info = &iosapic_intr_info[vector];
-		if (info->trigger != trigger || info->polarity != polarity) {
+	else if (rte->refcnt == NO_REF_RTE) {
+		struct iosapic_intr_info *info = &iosapic_intr_info[irq];
+		if (info->count > 0 &&
+		    (info->trigger != trigger || info->polarity != polarity)){
 			printk (KERN_WARNING
 				"%s: cannot override the interrupt\n",
 				__FUNCTION__);
 			return -EINVAL;
 		}
+		rte->refcnt++;
+		iosapic_intr_info[irq].count++;
+		iosapic_lists[index].rtes_inuse++;
 	}
 
-	iosapic_intr_info[vector].polarity = polarity;
-	iosapic_intr_info[vector].dmode    = delivery;
-	iosapic_intr_info[vector].trigger  = trigger;
+	iosapic_intr_info[irq].polarity = polarity;
+	iosapic_intr_info[irq].dmode    = delivery;
+	iosapic_intr_info[irq].trigger  = trigger;
 
 	if (trigger == IOSAPIC_EDGE)
 		irq_type = &irq_type_iosapic_edge;
 	else
 		irq_type = &irq_type_iosapic_level;
 
-	idesc = irq_desc + vector;
+	idesc = irq_desc + irq;
 	if (idesc->chip != irq_type) {
 		if (idesc->chip != &no_irq_type)
 			printk(KERN_WARNING
 			       "%s: changing vector %d from %s to %s\n",
-			       __FUNCTION__, vector,
+			       __FUNCTION__, irq_to_vector(irq),
 			       idesc->chip->name, irq_type->name);
 		idesc->chip = irq_type;
 	}
@@ -695,18 +666,19 @@ register_intr (unsigned int gsi, int vector, unsigned char delivery,
 }
 
 static unsigned int
-get_target_cpu (unsigned int gsi, int vector)
+get_target_cpu (unsigned int gsi, int irq)
 {
 #ifdef CONFIG_SMP
 	static int cpu = -1;
 	extern int cpe_vector;
+	cpumask_t domain = irq_to_domain(irq);
 
 	/*
 	 * In case of vector shared by multiple RTEs, all RTEs that
 	 * share the vector need to use the same destination CPU.
 	 */
-	if (!list_empty(&iosapic_intr_info[vector].rtes))
-		return iosapic_intr_info[vector].dest;
+	if (!list_empty(&iosapic_intr_info[irq].rtes))
+		return iosapic_intr_info[irq].dest;
 
 	/*
 	 * If the platform supports redirection via XTP, let it
@@ -723,7 +695,7 @@ get_target_cpu (unsigned int gsi, int vector)
 		return cpu_physical_id(smp_processor_id());
 
 #ifdef CONFIG_ACPI
-	if (cpe_vector > 0 && vector == IA64_CPEP_VECTOR)
+	if (cpe_vector > 0 && irq_to_vector(irq) == IA64_CPEP_VECTOR)
 		return get_cpei_target_cpu();
 #endif
 
@@ -738,7 +710,7 @@ get_target_cpu (unsigned int gsi, int vector)
 			goto skip_numa_setup;
 
 		cpu_mask = node_to_cpumask(iosapic_lists[iosapic_index].node);
-
+		cpus_and(cpu_mask, cpu_mask, domain);
 		for_each_cpu_mask(numa_cpu, cpu_mask) {
 			if (!cpu_online(numa_cpu))
 				cpu_clear(numa_cpu, cpu_mask);
@@ -749,8 +721,8 @@ get_target_cpu (unsigned int gsi, int vector)
 		if (!num_cpus)
 			goto skip_numa_setup;
 
-		/* Use vector assignment to distribute across cpus in node */
-		cpu_index = vector % num_cpus;
+		/* Use irq assignment to distribute across cpus in node */
+		cpu_index = irq % num_cpus;
 
 		for (numa_cpu = first_cpu(cpu_mask) ; i < cpu_index ; i++)
 			numa_cpu = next_cpu(numa_cpu, cpu_mask);
@@ -768,7 +740,7 @@ skip_numa_setup:
 	do {
 		if (++cpu >= NR_CPUS)
 			cpu = 0;
-	} while (!cpu_online(cpu));
+	} while (!cpu_online(cpu) || !cpu_isset(cpu, domain));
 
 	return cpu_physical_id(cpu);
 #else  /* CONFIG_SMP */
@@ -785,84 +757,72 @@ int
 iosapic_register_intr (unsigned int gsi,
 		       unsigned long polarity, unsigned long trigger)
 {
-	int vector, mask = 1, err;
+	int irq, mask = 1, err;
 	unsigned int dest;
 	unsigned long flags;
 	struct iosapic_rte_info *rte;
 	u32 low32;
-again:
+
 	/*
 	 * If this GSI has already been registered (i.e., it's a
 	 * shared interrupt, or we lost a race to register it),
 	 * don't touch the RTE.
 	 */
 	spin_lock_irqsave(&iosapic_lock, flags);
-	{
-		vector = gsi_to_vector(gsi);
-		if (vector > 0) {
-			rte = gsi_vector_to_rte(gsi, vector);
+	irq = __gsi_to_irq(gsi);
+	if (irq > 0) {
+		rte = find_rte(irq, gsi);
+		if(iosapic_intr_info[irq].count == 0) {
+			assign_irq_vector(irq);
+			dynamic_irq_init(irq);
+		} else if (rte->refcnt != NO_REF_RTE) {
 			rte->refcnt++;
-			spin_unlock_irqrestore(&iosapic_lock, flags);
-			return vector;
+			goto unlock_iosapic_lock;
 		}
-	}
-	spin_unlock_irqrestore(&iosapic_lock, flags);
+	} else
+		irq = create_irq();
 
 	/* If vector is running out, we try to find a sharable vector */
-	vector = assign_irq_vector(AUTO_ASSIGN);
-	if (vector < 0) {
-		vector = iosapic_find_sharable_vector(trigger, polarity);
-  		if (vector < 0)
-			return -ENOSPC;
+	if (irq < 0) {
+		irq = iosapic_find_sharable_irq(trigger, polarity);
+		if (irq < 0)
+			goto unlock_iosapic_lock;
 	}
 
-	spin_lock_irqsave(&irq_desc[vector].lock, flags);
-	spin_lock(&iosapic_lock);
-	{
-		if (gsi_to_vector(gsi) > 0) {
-			if (list_empty(&iosapic_intr_info[vector].rtes))
-				free_irq_vector(vector);
-			spin_unlock(&iosapic_lock);
-			spin_unlock_irqrestore(&irq_desc[vector].lock,
-					       flags);
-			goto again;
-		}
-
-		dest = get_target_cpu(gsi, vector);
-		err = register_intr(gsi, vector, IOSAPIC_LOWEST_PRIORITY,
-			      polarity, trigger);
-		if (err < 0) {
-			spin_unlock(&iosapic_lock);
-			spin_unlock_irqrestore(&irq_desc[vector].lock,
-					       flags);
-			return err;
-		}
-
-		/*
-		 * If the vector is shared and already unmasked for
-		 * other interrupt sources, don't mask it.
-		 */
-		low32 = iosapic_intr_info[vector].low32;
-		if (vector_is_shared(vector) && !(low32 & IOSAPIC_MASK))
-			mask = 0;
-		set_rte(gsi, vector, dest, mask);
+	spin_lock(&irq_desc[irq].lock);
+	dest = get_target_cpu(gsi, irq);
+	err = register_intr(gsi, irq, IOSAPIC_LOWEST_PRIORITY,
+			    polarity, trigger);
+	if (err < 0) {
+		irq = err;
+		goto unlock_all;
 	}
-	spin_unlock(&iosapic_lock);
-	spin_unlock_irqrestore(&irq_desc[vector].lock, flags);
+
+	/*
+	 * If the vector is shared and already unmasked for other
+	 * interrupt sources, don't mask it.
+	 */
+	low32 = iosapic_intr_info[irq].low32;
+	if (irq_is_shared(irq) && !(low32 & IOSAPIC_MASK))
+		mask = 0;
+	set_rte(gsi, irq, dest, mask);
 
 	printk(KERN_INFO "GSI %u (%s, %s) -> CPU %d (0x%04x) vector %d\n",
 	       gsi, (trigger == IOSAPIC_EDGE ? "edge" : "level"),
 	       (polarity == IOSAPIC_POL_HIGH ? "high" : "low"),
-	       cpu_logical_id(dest), dest, vector);
-
-	return vector;
+	       cpu_logical_id(dest), dest, irq_to_vector(irq));
+ unlock_all:
+	spin_unlock(&irq_desc[irq].lock);
+ unlock_iosapic_lock:
+	spin_unlock_irqrestore(&iosapic_lock, flags);
+	return irq;
 }
 
 void
 iosapic_unregister_intr (unsigned int gsi)
 {
 	unsigned long flags;
-	int irq, vector, index;
+	int irq, index;
 	irq_desc_t *idesc;
 	u32 low32;
 	unsigned long trigger, polarity;
@@ -881,78 +841,56 @@ iosapic_unregister_intr (unsigned int gsi)
 		WARN_ON(1);
 		return;
 	}
-	vector = irq_to_vector(irq);
 
-	idesc = irq_desc + irq;
-	spin_lock_irqsave(&idesc->lock, flags);
-	spin_lock(&iosapic_lock);
-	{
-		if ((rte = gsi_vector_to_rte(gsi, vector)) == NULL) {
-			printk(KERN_ERR
-			       "iosapic_unregister_intr(%u) unbalanced\n",
-			       gsi);
-			WARN_ON(1);
-			goto out;
-		}
+	spin_lock_irqsave(&iosapic_lock, flags);
+	if ((rte = find_rte(irq, gsi)) == NULL) {
+		printk(KERN_ERR "iosapic_unregister_intr(%u) unbalanced\n",
+		       gsi);
+		WARN_ON(1);
+		goto out;
+	}
 
-		if (--rte->refcnt > 0)
-			goto out;
+	if (--rte->refcnt > 0)
+		goto out;
 
-		/* Mask the interrupt */
-		low32 = iosapic_intr_info[vector].low32 | IOSAPIC_MASK;
-		iosapic_write(rte->addr, IOSAPIC_RTE_LOW(rte->rte_index),
-			      low32);
+	idesc = irq_desc + irq;
+	rte->refcnt = NO_REF_RTE;
 
-		/* Remove the rte entry from the list */
-		list_del(&rte->rte_list);
-		iosapic_intr_info[vector].count--;
-		iosapic_free_rte(rte);
-		index = find_iosapic(gsi);
-		iosapic_lists[index].rtes_inuse--;
-		WARN_ON(iosapic_lists[index].rtes_inuse < 0);
-
-		trigger	 = iosapic_intr_info[vector].trigger;
-		polarity = iosapic_intr_info[vector].polarity;
-		dest     = iosapic_intr_info[vector].dest;
-		printk(KERN_INFO
-		       "GSI %u (%s, %s) -> CPU %d (0x%04x)"
-		       " vector %d unregistered\n",
-		       gsi, (trigger == IOSAPIC_EDGE ? "edge" : "level"),
-		       (polarity == IOSAPIC_POL_HIGH ? "high" : "low"),
-		       cpu_logical_id(dest), dest, vector);
+	/* Mask the interrupt */
+	low32 = iosapic_intr_info[irq].low32 | IOSAPIC_MASK;
+	iosapic_write(rte->iosapic, IOSAPIC_RTE_LOW(rte->rte_index), low32);
 
-		if (list_empty(&iosapic_intr_info[vector].rtes)) {
-			/* Sanity check */
-			BUG_ON(iosapic_intr_info[vector].count);
+	iosapic_intr_info[irq].count--;
+	index = find_iosapic(gsi);
+	iosapic_lists[index].rtes_inuse--;
+	WARN_ON(iosapic_lists[index].rtes_inuse < 0);
 
-			/* Clear the interrupt controller descriptor */
-			idesc->chip = &no_irq_type;
+	trigger  = iosapic_intr_info[irq].trigger;
+	polarity = iosapic_intr_info[irq].polarity;
+	dest     = iosapic_intr_info[irq].dest;
+	printk(KERN_INFO
+	       "GSI %u (%s, %s) -> CPU %d (0x%04x) vector %d unregistered\n",
+	       gsi, (trigger == IOSAPIC_EDGE ? "edge" : "level"),
+	       (polarity == IOSAPIC_POL_HIGH ? "high" : "low"),
+	       cpu_logical_id(dest), dest, irq_to_vector(irq));
 
+	if (iosapic_intr_info[irq].count == 0) {
 #ifdef CONFIG_SMP
-			/* Clear affinity */
-			cpus_setall(idesc->affinity);
+		/* Clear affinity */
+		cpus_setall(idesc->affinity);
 #endif
-
-			/* Clear the interrupt information */
-			memset(&iosapic_intr_info[vector], 0,
-			       sizeof(struct iosapic_intr_info));
-			iosapic_intr_info[vector].low32 |= IOSAPIC_MASK;
-			INIT_LIST_HEAD(&iosapic_intr_info[vector].rtes);
-
-			if (idesc->action) {
-				printk(KERN_ERR
-				       "interrupt handlers still exist on"
-				       "IRQ %u\n", irq);
-				WARN_ON(1);
-			}
-
-			/* Free the interrupt vector */
-			free_irq_vector(vector);
-		}
+		/* Clear the interrupt information */
+		iosapic_intr_info[irq].dest = 0;
+		iosapic_intr_info[irq].dmode = 0;
+		iosapic_intr_info[irq].polarity = 0;
+		iosapic_intr_info[irq].trigger = 0;
+		iosapic_intr_info[irq].low32 |= IOSAPIC_MASK;
+
+		/* Destroy and reserve IRQ */
+		destroy_and_reserve_irq(irq);
 	}
  out:
-	spin_unlock(&iosapic_lock);
-	spin_unlock_irqrestore(&idesc->lock, flags);
+	spin_unlock_irqrestore(&iosapic_lock, flags);
 }
 
 /*
@@ -965,27 +903,30 @@ iosapic_register_platform_intr (u32 int_type, unsigned int gsi,
 {
 	static const char * const name[] = {"unknown", "PMI", "INIT", "CPEI"};
 	unsigned char delivery;
-	int vector, mask = 0;
+	int irq, vector, mask = 0;
 	unsigned int dest = ((id << 8) | eid) & 0xffff;
 
 	switch (int_type) {
 	      case ACPI_INTERRUPT_PMI:
-		vector = iosapic_vector;
+		irq = vector = iosapic_vector;
+		bind_irq_vector(irq, vector, CPU_MASK_ALL);
 		/*
 		 * since PMI vector is alloc'd by FW(ACPI) not by kernel,
 		 * we need to make sure the vector is available
 		 */
-		iosapic_reassign_vector(vector);
+		iosapic_reassign_vector(irq);
 		delivery = IOSAPIC_PMI;
 		break;
 	      case ACPI_INTERRUPT_INIT:
-		vector = assign_irq_vector(AUTO_ASSIGN);
-		if (vector < 0)
+		irq = create_irq();
+		if (irq < 0)
 			panic("%s: out of interrupt vectors!\n", __FUNCTION__);
+		vector = irq_to_vector(irq);
 		delivery = IOSAPIC_INIT;
 		break;
 	      case ACPI_INTERRUPT_CPEI:
-		vector = IA64_CPE_VECTOR;
+		irq = vector = IA64_CPE_VECTOR;
+		BUG_ON(bind_irq_vector(irq, vector, CPU_MASK_ALL));
 		delivery = IOSAPIC_LOWEST_PRIORITY;
 		mask = 1;
 		break;
@@ -995,7 +936,7 @@ iosapic_register_platform_intr (u32 int_type, unsigned int gsi,
 		return -1;
 	}
 
-	register_intr(gsi, vector, delivery, polarity, trigger);
+	register_intr(gsi, irq, delivery, polarity, trigger);
 
 	printk(KERN_INFO
 	       "PLATFORM int %s (0x%x): GSI %u (%s, %s) -> CPU %d (0x%04x)"
@@ -1005,7 +946,7 @@ iosapic_register_platform_intr (u32 int_type, unsigned int gsi,
 	       (polarity == IOSAPIC_POL_HIGH ? "high" : "low"),
 	       cpu_logical_id(dest), dest, vector);
 
-	set_rte(gsi, vector, dest, mask);
+	set_rte(gsi, irq, dest, mask);
 	return vector;
 }
 
@@ -1017,30 +958,32 @@ iosapic_override_isa_irq (unsigned int isa_irq, unsigned int gsi,
 			  unsigned long polarity,
 			  unsigned long trigger)
 {
-	int vector;
+	int vector, irq;
 	unsigned int dest = cpu_physical_id(smp_processor_id());
 
-	vector = isa_irq_to_vector(isa_irq);
-
-	register_intr(gsi, vector, IOSAPIC_LOWEST_PRIORITY, polarity, trigger);
+	irq = vector = isa_irq_to_vector(isa_irq);
+	BUG_ON(bind_irq_vector(irq, vector, CPU_MASK_ALL));
+	register_intr(gsi, irq, IOSAPIC_LOWEST_PRIORITY, polarity, trigger);
 
 	DBG("ISA: IRQ %u -> GSI %u (%s,%s) -> CPU %d (0x%04x) vector %d\n",
 	    isa_irq, gsi, trigger == IOSAPIC_EDGE ? "edge" : "level",
 	    polarity == IOSAPIC_POL_HIGH ? "high" : "low",
 	    cpu_logical_id(dest), dest, vector);
 
-	set_rte(gsi, vector, dest, 1);
+	set_rte(gsi, irq, dest, 1);
 }
 
 void __init
 iosapic_system_init (int system_pcat_compat)
 {
-	int vector;
+	int irq;
 
-	for (vector = 0; vector < IA64_NUM_VECTORS; ++vector) {
-		iosapic_intr_info[vector].low32 = IOSAPIC_MASK;
+	for (irq = 0; irq < NR_IRQS; ++irq) {
+		iosapic_intr_info[irq].low32 = IOSAPIC_MASK;
 		/* mark as unused */
-		INIT_LIST_HEAD(&iosapic_intr_info[vector].rtes);
+		INIT_LIST_HEAD(&iosapic_intr_info[irq].rtes);
+
+		iosapic_intr_info[irq].count = 0;
 	}
 
 	pcat_compat = system_pcat_compat;
@@ -1108,31 +1051,35 @@ iosapic_init (unsigned long phys_addr, unsigned int gsi_base)
 	unsigned long flags;
 
 	spin_lock_irqsave(&iosapic_lock, flags);
-	{
-		addr = ioremap(phys_addr, 0);
-		ver = iosapic_version(addr);
+	index = find_iosapic(gsi_base);
+	if (index >= 0) {
+		spin_unlock_irqrestore(&iosapic_lock, flags);
+		return -EBUSY;
+	}
 
-		if ((err = iosapic_check_gsi_range(gsi_base, ver))) {
-			iounmap(addr);
-			spin_unlock_irqrestore(&iosapic_lock, flags);
-			return err;
-		}
+	addr = ioremap(phys_addr, 0);
+	ver = iosapic_version(addr);
+	if ((err = iosapic_check_gsi_range(gsi_base, ver))) {
+		iounmap(addr);
+		spin_unlock_irqrestore(&iosapic_lock, flags);
+		return err;
+	}
 
-		/*
-		 * The MAX_REDIR register holds the highest input pin
-		 * number (starting from 0).
-		 * We add 1 so that we can use it for number of pins (= RTEs)
-		 */
-		num_rte = ((ver >> 16) & 0xff) + 1;
+	/*
+	 * The MAX_REDIR register holds the highest input pin number
+	 * (starting from 0).  We add 1 so that we can use it for
+	 * number of pins (= RTEs)
+	 */
+	num_rte = ((ver >> 16) & 0xff) + 1;
 
-		index = iosapic_alloc();
-		iosapic_lists[index].addr = addr;
-		iosapic_lists[index].gsi_base = gsi_base;
-		iosapic_lists[index].num_rte = num_rte;
+	index = iosapic_alloc();
+	iosapic_lists[index].addr = addr;
+	iosapic_lists[index].gsi_base = gsi_base;
+	iosapic_lists[index].num_rte = num_rte;
 #ifdef CONFIG_NUMA
-		iosapic_lists[index].node = MAX_NUMNODES;
+	iosapic_lists[index].node = MAX_NUMNODES;
 #endif
-	}
+	spin_lock_init(&iosapic_lists[index].lock);
 	spin_unlock_irqrestore(&iosapic_lock, flags);
 
 	if ((gsi_base == 0) && pcat_compat) {
@@ -1157,25 +1104,22 @@ iosapic_remove (unsigned int gsi_base)
 	unsigned long flags;
 
 	spin_lock_irqsave(&iosapic_lock, flags);
-	{
-		index = find_iosapic(gsi_base);
-		if (index < 0) {
-			printk(KERN_WARNING "%s: No IOSAPIC for GSI base %u\n",
-			       __FUNCTION__, gsi_base);
-			goto out;
-		}
-
-		if (iosapic_lists[index].rtes_inuse) {
-			err = -EBUSY;
-			printk(KERN_WARNING
-			       "%s: IOSAPIC for GSI base %u is busy\n",
-			       __FUNCTION__, gsi_base);
-			goto out;
-		}
+	index = find_iosapic(gsi_base);
+	if (index < 0) {
+		printk(KERN_WARNING "%s: No IOSAPIC for GSI base %u\n",
+		       __FUNCTION__, gsi_base);
+		goto out;
+	}
 
-		iounmap(iosapic_lists[index].addr);
-		iosapic_free(index);
+	if (iosapic_lists[index].rtes_inuse) {
+		err = -EBUSY;
+		printk(KERN_WARNING "%s: IOSAPIC for GSI base %u is busy\n",
+		       __FUNCTION__, gsi_base);
+		goto out;
 	}
+
+	iounmap(iosapic_lists[index].addr);
+	iosapic_free(index);
  out:
 	spin_unlock_irqrestore(&iosapic_lock, flags);
 	return err;

arch/ia64/kernel/irq.c

@@ -35,7 +35,7 @@ void ack_bad_irq(unsigned int irq)
 #ifdef CONFIG_IA64_GENERIC
 unsigned int __ia64_local_vector_to_irq (ia64_vector vec)
 {
-	return (unsigned int) vec;
+	return __get_cpu_var(vector_irq)[vec];
 }
 #endif
 

arch/ia64/kernel/irq_ia64.c

@@ -46,6 +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;
@@ -54,6 +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.
  */
@@ -64,46 +72,269 @@ __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);
+
+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] = IA64_SPURIOUS_INT_VECTOR
+};
+
+static cpumask_t vector_table[IA64_MAX_DEVICE_VECTORS] = {
+	[0 ... IA64_MAX_DEVICE_VECTORS - 1] = CPU_MASK_NONE
+};
+
+static int irq_status[NR_IRQS] = {
+	[0 ... NR_IRQS -1] = IRQ_UNUSED
+};
+
+int check_irq_used(int irq)
+{
+	if (irq_status[irq] == IRQ_USED)
+		return 1;
+
+	return -1;
+}
+
+static void reserve_irq(unsigned int irq)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&vector_lock, flags);
+	irq_status[irq] = IRQ_RSVD;
+	spin_unlock_irqrestore(&vector_lock, flags);
+}
+
+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;
+
+	cpus_and(mask, domain, cpu_online_map);
+	if (cpus_empty(mask))
+		return -EINVAL;
+
+	for (pos = 0; pos < IA64_NUM_DEVICE_VECTORS; pos++) {
+		cpus_and(mask, domain, vector_table[pos]);
+		if (!cpus_empty(mask))
+			continue;
+		return IA64_FIRST_DEVICE_VECTOR + pos;
+	}
+	return -ENOSPC;
+}
+
+static int __bind_irq_vector(int irq, int vector, cpumask_t domain)
+{
+	cpumask_t mask;
+	int cpu, pos;
+	struct irq_cfg *cfg = &irq_cfg[irq];
+
+	cpus_and(mask, domain, cpu_online_map);
+	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;
+	pos = vector - IA64_FIRST_DEVICE_VECTOR;
+	cpus_or(vector_table[pos], vector_table[pos], 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, pos;
+	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;
+	cpus_and(mask, cfg->domain, cpu_online_map);
+	for_each_cpu_mask(cpu, mask)
+		per_cpu(vector_irq, cpu)[vector] = IA64_SPURIOUS_INT_VECTOR;
+	cfg->vector = IRQ_VECTOR_UNASSIGNED;
+	cfg->domain = CPU_MASK_NONE;
+	irq_status[irq] = IRQ_UNUSED;
+	pos = vector - IA64_FIRST_DEVICE_VECTOR;
+	cpus_andnot(vector_table[pos], vector_table[pos], 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
 assign_irq_vector (int irq)
 {
-	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;
+	unsigned long flags;
+	int vector, cpu;
+	cpumask_t domain;
+
+	vector = -ENOSPC;
+
+	spin_lock_irqsave(&vector_lock, flags);
+	if (irq < 0) {
+		goto out;
+	}
+	for_each_online_cpu(cpu) {
+		domain = vector_allocation_domain(cpu);
+		vector = find_unassigned_vector(domain);
+		if (vector >= 0)
+			break;
+	}
+	if (vector < 0)
+		goto out;
+	BUG_ON(__bind_irq_vector(irq, vector, domain));
+ out:
+	spin_unlock_irqrestore(&vector_lock, flags);
 	return vector;
 }
 
 void
 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);
+}
 
-	pos = vector - IA64_FIRST_DEVICE_VECTOR;
-	return test_and_set_bit(pos, ia64_vector_mask);
+/*
+ * 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] = IA64_SPURIOUS_INT_VECTOR;
+	/* 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 __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 1;
+}
+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)
+{
+	dynamic_irq_cleanup(irq);
+
+	clear_irq_vector(irq);
+	reserve_irq(irq);
+}
+
+static int __reassign_irq_vector(int irq, int cpu)
+{
+	struct irq_cfg *cfg = &irq_cfg[irq];
+	int vector;
+	cpumask_t domain;
+
+	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;
+	__clear_irq_vector(irq);
+	BUG_ON(__bind_irq_vector(irq, vector, domain));
+	return 0;
+}
+
+int reassign_irq_vector(int irq, int cpu)
+{
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&vector_lock, flags);
+	ret = __reassign_irq_vector(irq, cpu);
+	spin_unlock_irqrestore(&vector_lock, flags);
+	return ret;
 }
 
 /*
@@ -111,18 +342,35 @@ reserve_irq_vector (int vector)
  */
 int create_irq(void)
 {
-	int vector = assign_irq_vector(AUTO_ASSIGN);
-
-	if (vector >= 0)
-		dynamic_irq_init(vector);
-
-	return vector;
+	unsigned long flags;
+	int irq, vector, cpu;
+	cpumask_t domain;
+
+	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)
+		dynamic_irq_init(irq);
+	return irq;
 }
 
 void destroy_irq(unsigned int irq)
 {
 	dynamic_irq_cleanup(irq);
-	free_irq_vector(irq);
+	clear_irq_vector(irq);
 }
 
 #ifdef CONFIG_SMP
@@ -301,14 +549,13 @@ 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));
+	desc = irq_desc + irq;
+	desc->status |= IRQ_PER_CPU;
+	desc->chip = &irq_type_ia64_lsapic;
+	if (action)
+		setup_irq(irq, action);
 }
 
 void __init

arch/ia64/kernel/msi_ia64.c

@@ -13,6 +13,7 @@
 
 #define MSI_DATA_VECTOR_SHIFT		0
 #define	    MSI_DATA_VECTOR(v)		(((u8)v) << MSI_DATA_VECTOR_SHIFT)
+#define MSI_DATA_VECTOR_MASK		0xffffff00
 
 #define MSI_DATA_DELIVERY_SHIFT		8
 #define     MSI_DATA_DELIVERY_FIXED	(0 << MSI_DATA_DELIVERY_SHIFT)
@@ -50,17 +51,29 @@ static struct irq_chip	ia64_msi_chip;
 static void ia64_set_msi_irq_affinity(unsigned int irq, cpumask_t cpu_mask)
 {
 	struct msi_msg msg;
-	u32 addr;
+	u32 addr, data;
+	int cpu = first_cpu(cpu_mask);
+
+	if (!cpu_online(cpu))
+		return;
+
+	if (reassign_irq_vector(irq, cpu))
+		return;
 
 	read_msi_msg(irq, &msg);
 
 	addr = msg.address_lo;
 	addr &= MSI_ADDR_DESTID_MASK;
-	addr |= MSI_ADDR_DESTID_CPU(cpu_physical_id(first_cpu(cpu_mask)));
+	addr |= MSI_ADDR_DESTID_CPU(cpu_physical_id(cpu));
 	msg.address_lo = addr;
 
+	data = msg.data;
+	data &= MSI_DATA_VECTOR_MASK;
+	data |= MSI_DATA_VECTOR(irq_to_vector(irq));
+	msg.data = data;
+
 	write_msi_msg(irq, &msg);
-	irq_desc[irq].affinity = cpu_mask;
+	irq_desc[irq].affinity = cpumask_of_cpu(cpu);
 }
 #endif /* CONFIG_SMP */
 
@@ -69,13 +82,15 @@ int ia64_setup_msi_irq(struct pci_dev *pdev, struct msi_desc *desc)
 	struct msi_msg	msg;
 	unsigned long	dest_phys_id;
 	int	irq, vector;
+	cpumask_t mask;
 
 	irq = create_irq();
 	if (irq < 0)
 		return irq;
 
 	set_irq_msi(irq, desc);
-	dest_phys_id = cpu_physical_id(first_cpu(cpu_online_map));
+	cpus_and(mask, irq_to_domain(irq), cpu_online_map);
+	dest_phys_id = cpu_physical_id(first_cpu(mask));
 	vector = irq_to_vector(irq);
 
 	msg.address_hi = 0;

arch/ia64/kernel/smpboot.c

@@ -395,9 +395,13 @@ smp_callin (void)
 	fix_b0_for_bsp();
 
 	lock_ipi_calllock();
+	spin_lock(&vector_lock);
+	/* Setup the per cpu irq handling data structures */
+	__setup_vector_irq(cpuid);
 	cpu_set(cpuid, cpu_online_map);
 	unlock_ipi_calllock();
 	per_cpu(cpu_state, cpuid) = CPU_ONLINE;
+	spin_unlock(&vector_lock);
 
 	smp_setup_percpu_timer();
 

include/asm-ia64/hw_irq.h

@@ -90,13 +90,27 @@ enum {
 extern __u8 isa_irq_to_vector_map[16];
 #define isa_irq_to_vector(x)	isa_irq_to_vector_map[(x)]
 
+struct irq_cfg {
+	ia64_vector vector;
+	cpumask_t domain;
+};
+extern spinlock_t vector_lock;
+extern struct irq_cfg irq_cfg[NR_IRQS];
+#define irq_to_domain(x)	irq_cfg[(x)].domain
+DECLARE_PER_CPU(int[IA64_NUM_VECTORS], vector_irq);
+
 extern struct hw_interrupt_type irq_type_ia64_lsapic;	/* CPU-internal interrupt controller */
 
+extern int bind_irq_vector(int irq, int vector, cpumask_t domain);
 extern int assign_irq_vector (int irq);	/* allocate a free vector */
 extern void free_irq_vector (int vector);
 extern int reserve_irq_vector (int vector);
+extern void __setup_vector_irq(int cpu);
+extern int reassign_irq_vector(int irq, int cpu);
 extern void ia64_send_ipi (int cpu, int vector, int delivery_mode, int redirect);
 extern void register_percpu_irq (ia64_vector vec, struct irqaction *action);
+extern int check_irq_used (int irq);
+extern void destroy_and_reserve_irq (unsigned int irq);
 
 static inline void ia64_resend_irq(unsigned int vector)
 {
@@ -113,7 +127,7 @@ extern irq_desc_t irq_desc[NR_IRQS];
 static inline unsigned int
 __ia64_local_vector_to_irq (ia64_vector vec)
 {
-	return (unsigned int) vec;
+	return __get_cpu_var(vector_irq)[vec];
 }
 #endif
 
@@ -131,7 +145,7 @@ __ia64_local_vector_to_irq (ia64_vector vec)
 static inline ia64_vector
 irq_to_vector (int irq)
 {
-	return (ia64_vector) irq;
+	return irq_cfg[irq].vector;
 }
 
 /*

include/asm-ia64/iosapic.h

@@ -47,19 +47,21 @@
 #define	IOSAPIC_MASK_SHIFT		16
 #define	IOSAPIC_MASK			(1<<IOSAPIC_MASK_SHIFT)
 
+#define IOSAPIC_VECTOR_MASK		0xffffff00
+
 #ifndef __ASSEMBLY__
 
 #ifdef CONFIG_IOSAPIC
 
 #define NR_IOSAPICS			256
 
-static inline unsigned int iosapic_read(char __iomem *iosapic, unsigned int reg)
+static inline unsigned int __iosapic_read(char __iomem *iosapic, unsigned int reg)
 {
 	writel(reg, iosapic + IOSAPIC_REG_SELECT);
 	return readl(iosapic + IOSAPIC_WINDOW);
 }
 
-static inline void iosapic_write(char __iomem *iosapic, unsigned int reg, u32 val)
+static inline void __iosapic_write(char __iomem *iosapic, unsigned int reg, u32 val)
 {
 	writel(reg, iosapic + IOSAPIC_REG_SELECT);
 	writel(val, iosapic + IOSAPIC_WINDOW);

include/asm-ia64/irq.h

@@ -14,8 +14,13 @@
 #include <linux/types.h>
 #include <linux/cpumask.h>
 
-#define NR_IRQS		256
-#define NR_IRQ_VECTORS	NR_IRQS
+#define NR_VECTORS	256
+
+#if (NR_VECTORS + 32 * NR_CPUS) < 1024
+#define NR_IRQS (NR_VECTORS + 32 * NR_CPUS)
+#else
+#define NR_IRQS 1024
+#endif
 
 static __inline__ int
 irq_canonicalize (int irq)