Merge branch 'irq-core-for-linus' of...

Merge branch 'irq-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip

* 'irq-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  x86, sparseirq: clean up Kconfig entry
  x86: turn CONFIG_SPARSE_IRQ off by default
  sparseirq: fix numa_migrate_irq_desc dependency and comments
  sparseirq: add kernel-doc notation for new member in irq_desc, -v2
  locking, irq: enclose irq_desc_lock_class in CONFIG_LOCKDEP
  sparseirq, xen: make sure irq_desc is allocated for interrupts
  sparseirq: fix !SMP building, #2
  x86, sparseirq: move irq_desc according to smp_affinity, v7
  proc: enclose desc variable of show_stat() in CONFIG_SPARSE_IRQ
  sparse irqs: add irqnr.h to the user headers list
  sparse irqs: handle !GENIRQ platforms
  sparseirq: fix !SMP && !PCI_MSI && !HT_IRQ build
  sparseirq: fix Alpha build failure
  sparseirq: fix typo in !CONFIG_IO_APIC case
  x86, MSI: pass irq_cfg and irq_desc
  x86: MSI start irq numbering from nr_irqs_gsi
  x86: use NR_IRQS_LEGACY
  sparse irq_desc[] array: core kernel and x86 changes
  genirq: record IRQ_LEVEL in irq_desc[]
  irq.h: remove padding from irq_desc on 64bits

arch/x86/Kconfig

@@ -247,6 +247,28 @@ config X86_HAS_BOOT_CPU_ID
 	def_bool y
 	depends on X86_VOYAGER
 
+config SPARSE_IRQ
+	bool "Support sparse irq numbering"
+	depends on PCI_MSI || HT_IRQ
+	help
+	  This enables support for sparse irqs. This is useful for distro
+	  kernels that want to define a high CONFIG_NR_CPUS value but still
+	  want to have low kernel memory footprint on smaller machines.
+
+	  ( Sparse IRQs can also be beneficial on NUMA boxes, as they spread
+	    out the irq_desc[] array in a more NUMA-friendly way. )
+
+	  If you don't know what to do here, say N.
+
+config NUMA_MIGRATE_IRQ_DESC
+	bool "Move irq desc when changing irq smp_affinity"
+	depends on SPARSE_IRQ && NUMA
+	default n
+	help
+	  This enables moving irq_desc to cpu/node that irq will use handled.
+
+	  If you don't know what to do here, say N.
+
 config X86_FIND_SMP_CONFIG
 	def_bool y
 	depends on X86_MPPARSE || X86_VOYAGER

arch/x86/include/asm/io_apic.h

@@ -198,17 +198,14 @@ extern void restore_IO_APIC_setup(void);
 extern void reinit_intr_remapped_IO_APIC(int);
 #endif
 
-extern int probe_nr_irqs(void);
+extern void probe_nr_irqs_gsi(void);
 
 #else  /* !CONFIG_X86_IO_APIC */
 #define io_apic_assign_pci_irqs 0
 static const int timer_through_8259 = 0;
-static inline void ioapic_init_mappings(void) { }
+static inline void ioapic_init_mappings(void)	{ }
 
-static inline int probe_nr_irqs(void)
-{
-	return NR_IRQS;
-}
+static inline void probe_nr_irqs_gsi(void)	{ }
 #endif
 
 #endif /* _ASM_X86_IO_APIC_H */

arch/x86/include/asm/irq_vectors.h

@@ -101,12 +101,23 @@
 #define LAST_VM86_IRQ		15
 #define invalid_vm86_irq(irq)	((irq) < 3 || (irq) > 15)
 
+#define NR_IRQS_LEGACY		16
+
 #if defined(CONFIG_X86_IO_APIC) && !defined(CONFIG_X86_VOYAGER)
+
+#ifndef CONFIG_SPARSE_IRQ
 # if NR_CPUS < MAX_IO_APICS
 #  define NR_IRQS (NR_VECTORS + (32 * NR_CPUS))
 # else
 #  define NR_IRQS (NR_VECTORS + (32 * MAX_IO_APICS))
 # endif
+#else
+# if (8 * NR_CPUS) > (32 * MAX_IO_APICS)
+#  define NR_IRQS (NR_VECTORS + (8 * NR_CPUS))
+# else
+#  define NR_IRQS (NR_VECTORS + (32 * MAX_IO_APICS))
+# endif
+#endif
 
 #elif defined(CONFIG_X86_VOYAGER)
 

arch/x86/kernel/io_apic.c

@@ -108,93 +108,252 @@ static int __init parse_noapic(char *str)
 early_param("noapic", parse_noapic);
 
 struct irq_pin_list;
+
+/*
+ * This is performance-critical, we want to do it O(1)
+ *
+ * the indexing order of this array favors 1:1 mappings
+ * between pins and IRQs.
+ */
+
+struct irq_pin_list {
+	int apic, pin;
+	struct irq_pin_list *next;
+};
+
+static struct irq_pin_list *get_one_free_irq_2_pin(int cpu)
+{
+	struct irq_pin_list *pin;
+	int node;
+
+	node = cpu_to_node(cpu);
+
+	pin = kzalloc_node(sizeof(*pin), GFP_ATOMIC, node);
+	printk(KERN_DEBUG "  alloc irq_2_pin on cpu %d node %d\n", cpu, node);
+
+	return pin;
+}
+
 struct irq_cfg {
-	unsigned int irq;
 	struct irq_pin_list *irq_2_pin;
 	cpumask_t domain;
 	cpumask_t old_domain;
 	unsigned move_cleanup_count;
 	u8 vector;
 	u8 move_in_progress : 1;
+#ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
+	u8 move_desc_pending : 1;
+#endif
 };
 
 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
+#ifdef CONFIG_SPARSE_IRQ
+static struct irq_cfg irq_cfgx[] = {
+#else
 static struct irq_cfg irq_cfgx[NR_IRQS] = {
-	[0]  = { .irq =  0, .domain = CPU_MASK_ALL, .vector = IRQ0_VECTOR,  },
-	[1]  = { .irq =  1, .domain = CPU_MASK_ALL, .vector = IRQ1_VECTOR,  },
-	[2]  = { .irq =  2, .domain = CPU_MASK_ALL, .vector = IRQ2_VECTOR,  },
-	[3]  = { .irq =  3, .domain = CPU_MASK_ALL, .vector = IRQ3_VECTOR,  },
-	[4]  = { .irq =  4, .domain = CPU_MASK_ALL, .vector = IRQ4_VECTOR,  },
-	[5]  = { .irq =  5, .domain = CPU_MASK_ALL, .vector = IRQ5_VECTOR,  },
-	[6]  = { .irq =  6, .domain = CPU_MASK_ALL, .vector = IRQ6_VECTOR,  },
-	[7]  = { .irq =  7, .domain = CPU_MASK_ALL, .vector = IRQ7_VECTOR,  },
-	[8]  = { .irq =  8, .domain = CPU_MASK_ALL, .vector = IRQ8_VECTOR,  },
-	[9]  = { .irq =  9, .domain = CPU_MASK_ALL, .vector = IRQ9_VECTOR,  },
-	[10] = { .irq = 10, .domain = CPU_MASK_ALL, .vector = IRQ10_VECTOR, },
-	[11] = { .irq = 11, .domain = CPU_MASK_ALL, .vector = IRQ11_VECTOR, },
-	[12] = { .irq = 12, .domain = CPU_MASK_ALL, .vector = IRQ12_VECTOR, },
-	[13] = { .irq = 13, .domain = CPU_MASK_ALL, .vector = IRQ13_VECTOR, },
-	[14] = { .irq = 14, .domain = CPU_MASK_ALL, .vector = IRQ14_VECTOR, },
-	[15] = { .irq = 15, .domain = CPU_MASK_ALL, .vector = IRQ15_VECTOR, },
+#endif
+	[0]  = { .domain = CPU_MASK_ALL, .vector = IRQ0_VECTOR,  },
+	[1]  = { .domain = CPU_MASK_ALL, .vector = IRQ1_VECTOR,  },
+	[2]  = { .domain = CPU_MASK_ALL, .vector = IRQ2_VECTOR,  },
+	[3]  = { .domain = CPU_MASK_ALL, .vector = IRQ3_VECTOR,  },
+	[4]  = { .domain = CPU_MASK_ALL, .vector = IRQ4_VECTOR,  },
+	[5]  = { .domain = CPU_MASK_ALL, .vector = IRQ5_VECTOR,  },
+	[6]  = { .domain = CPU_MASK_ALL, .vector = IRQ6_VECTOR,  },
+	[7]  = { .domain = CPU_MASK_ALL, .vector = IRQ7_VECTOR,  },
+	[8]  = { .domain = CPU_MASK_ALL, .vector = IRQ8_VECTOR,  },
+	[9]  = { .domain = CPU_MASK_ALL, .vector = IRQ9_VECTOR,  },
+	[10] = { .domain = CPU_MASK_ALL, .vector = IRQ10_VECTOR, },
+	[11] = { .domain = CPU_MASK_ALL, .vector = IRQ11_VECTOR, },
+	[12] = { .domain = CPU_MASK_ALL, .vector = IRQ12_VECTOR, },
+	[13] = { .domain = CPU_MASK_ALL, .vector = IRQ13_VECTOR, },
+	[14] = { .domain = CPU_MASK_ALL, .vector = IRQ14_VECTOR, },
+	[15] = { .domain = CPU_MASK_ALL, .vector = IRQ15_VECTOR, },
 };
 
-#define for_each_irq_cfg(irq, cfg)		\
-	for (irq = 0, cfg = irq_cfgx; irq < nr_irqs; irq++, cfg++)
+void __init arch_early_irq_init(void)
+{
+	struct irq_cfg *cfg;
+	struct irq_desc *desc;
+	int count;
+	int i;
+
+	cfg = irq_cfgx;
+	count = ARRAY_SIZE(irq_cfgx);
+
+	for (i = 0; i < count; i++) {
+		desc = irq_to_desc(i);
+		desc->chip_data = &cfg[i];
+	}
+}
 
+#ifdef CONFIG_SPARSE_IRQ
 static struct irq_cfg *irq_cfg(unsigned int irq)
 {
-	return irq < nr_irqs ? irq_cfgx + irq : NULL;
+	struct irq_cfg *cfg = NULL;
+	struct irq_desc *desc;
+
+	desc = irq_to_desc(irq);
+	if (desc)
+		cfg = desc->chip_data;
+
+	return cfg;
 }
 
-static struct irq_cfg *irq_cfg_alloc(unsigned int irq)
+static struct irq_cfg *get_one_free_irq_cfg(int cpu)
 {
-	return irq_cfg(irq);
+	struct irq_cfg *cfg;
+	int node;
+
+	node = cpu_to_node(cpu);
+
+	cfg = kzalloc_node(sizeof(*cfg), GFP_ATOMIC, node);
+	printk(KERN_DEBUG "  alloc irq_cfg on cpu %d node %d\n", cpu, node);
+
+	return cfg;
 }
 
-/*
- * Rough estimation of how many shared IRQs there are, can be changed
- * anytime.
- */
-#define MAX_PLUS_SHARED_IRQS NR_IRQS
-#define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
+void arch_init_chip_data(struct irq_desc *desc, int cpu)
+{
+	struct irq_cfg *cfg;
 
-/*
- * This is performance-critical, we want to do it O(1)
- *
- * the indexing order of this array favors 1:1 mappings
- * between pins and IRQs.
- */
+	cfg = desc->chip_data;
+	if (!cfg) {
+		desc->chip_data = get_one_free_irq_cfg(cpu);
+		if (!desc->chip_data) {
+			printk(KERN_ERR "can not alloc irq_cfg\n");
+			BUG_ON(1);
+		}
+	}
+}
 
-struct irq_pin_list {
-	int apic, pin;
-	struct irq_pin_list *next;
-};
+#ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
+
+static void
+init_copy_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg, int cpu)
+{
+	struct irq_pin_list *old_entry, *head, *tail, *entry;
+
+	cfg->irq_2_pin = NULL;
+	old_entry = old_cfg->irq_2_pin;
+	if (!old_entry)
+		return;
 
-static struct irq_pin_list irq_2_pin_head[PIN_MAP_SIZE];
-static struct irq_pin_list *irq_2_pin_ptr;
+	entry = get_one_free_irq_2_pin(cpu);
+	if (!entry)
+		return;
 
-static void __init irq_2_pin_init(void)
+	entry->apic	= old_entry->apic;
+	entry->pin	= old_entry->pin;
+	head		= entry;
+	tail		= entry;
+	old_entry	= old_entry->next;
+	while (old_entry) {
+		entry = get_one_free_irq_2_pin(cpu);
+		if (!entry) {
+			entry = head;
+			while (entry) {
+				head = entry->next;
+				kfree(entry);
+				entry = head;
+			}
+			/* still use the old one */
+			return;
+		}
+		entry->apic	= old_entry->apic;
+		entry->pin	= old_entry->pin;
+		tail->next	= entry;
+		tail		= entry;
+		old_entry	= old_entry->next;
+	}
+
+	tail->next = NULL;
+	cfg->irq_2_pin = head;
+}
+
+static void free_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg)
 {
-	struct irq_pin_list *pin = irq_2_pin_head;
-	int i;
+	struct irq_pin_list *entry, *next;
 
-	for (i = 1; i < PIN_MAP_SIZE; i++)
-		pin[i-1].next = &pin[i];
+	if (old_cfg->irq_2_pin == cfg->irq_2_pin)
+		return;
+
+	entry = old_cfg->irq_2_pin;
 
-	irq_2_pin_ptr = &pin[0];
+	while (entry) {
+		next = entry->next;
+		kfree(entry);
+		entry = next;
+	}
+	old_cfg->irq_2_pin = NULL;
 }
 
-static struct irq_pin_list *get_one_free_irq_2_pin(void)
+void arch_init_copy_chip_data(struct irq_desc *old_desc,
+				 struct irq_desc *desc, int cpu)
 {
-	struct irq_pin_list *pin = irq_2_pin_ptr;
+	struct irq_cfg *cfg;
+	struct irq_cfg *old_cfg;
 
-	if (!pin)
-		panic("can not get more irq_2_pin\n");
+	cfg = get_one_free_irq_cfg(cpu);
 
-	irq_2_pin_ptr = pin->next;
-	pin->next = NULL;
-	return pin;
+	if (!cfg)
+		return;
+
+	desc->chip_data = cfg;
+
+	old_cfg = old_desc->chip_data;
+
+	memcpy(cfg, old_cfg, sizeof(struct irq_cfg));
+
+	init_copy_irq_2_pin(old_cfg, cfg, cpu);
+}
+
+static void free_irq_cfg(struct irq_cfg *old_cfg)
+{
+	kfree(old_cfg);
+}
+
+void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
+{
+	struct irq_cfg *old_cfg, *cfg;
+
+	old_cfg = old_desc->chip_data;
+	cfg = desc->chip_data;
+
+	if (old_cfg == cfg)
+		return;
+
+	if (old_cfg) {
+		free_irq_2_pin(old_cfg, cfg);
+		free_irq_cfg(old_cfg);
+		old_desc->chip_data = NULL;
+	}
+}
+
+static void set_extra_move_desc(struct irq_desc *desc, cpumask_t mask)
+{
+	struct irq_cfg *cfg = desc->chip_data;
+
+	if (!cfg->move_in_progress) {
+		/* it means that domain is not changed */
+		if (!cpus_intersects(desc->affinity, mask))
+			cfg->move_desc_pending = 1;
+	}
 }
+#endif
+
+#else
+static struct irq_cfg *irq_cfg(unsigned int irq)
+{
+	return irq < nr_irqs ? irq_cfgx + irq : NULL;
+}
+
+#endif
+
+#ifndef CONFIG_NUMA_MIGRATE_IRQ_DESC
+static inline void set_extra_move_desc(struct irq_desc *desc, cpumask_t mask)
+{
+}
+#endif
 
 struct io_apic {
 	unsigned int index;
@@ -237,11 +396,10 @@ static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned
 	writel(value, &io_apic->data);
 }
 
-static bool io_apic_level_ack_pending(unsigned int irq)
+static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
 {
 	struct irq_pin_list *entry;
 	unsigned long flags;
-	struct irq_cfg *cfg = irq_cfg(irq);
 
 	spin_lock_irqsave(&ioapic_lock, flags);
 	entry = cfg->irq_2_pin;
@@ -323,13 +481,12 @@ static void ioapic_mask_entry(int apic, int pin)
 }
 
 #ifdef CONFIG_SMP
-static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
+static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
 {
 	int apic, pin;
-	struct irq_cfg *cfg;
 	struct irq_pin_list *entry;
+	u8 vector = cfg->vector;
 
-	cfg = irq_cfg(irq);
 	entry = cfg->irq_2_pin;
 	for (;;) {
 		unsigned int reg;
@@ -359,24 +516,27 @@ static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
 	}
 }
 
-static int assign_irq_vector(int irq, cpumask_t mask);
+static int assign_irq_vector(int irq, struct irq_cfg *cfg, cpumask_t mask);
 
-static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
+static void set_ioapic_affinity_irq_desc(struct irq_desc *desc, cpumask_t mask)
 {
 	struct irq_cfg *cfg;
 	unsigned long flags;
 	unsigned int dest;
 	cpumask_t tmp;
-	struct irq_desc *desc;
+	unsigned int irq;
 
 	cpus_and(tmp, mask, cpu_online_map);
 	if (cpus_empty(tmp))
 		return;
 
-	cfg = irq_cfg(irq);
-	if (assign_irq_vector(irq, mask))
+	irq = desc->irq;
+	cfg = desc->chip_data;
+	if (assign_irq_vector(irq, cfg, mask))
 		return;
 
+	set_extra_move_desc(desc, mask);
+
 	cpus_and(tmp, cfg->domain, mask);
 	dest = cpu_mask_to_apicid(tmp);
 	/*
@@ -384,12 +544,20 @@ static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
 	 */
 	dest = SET_APIC_LOGICAL_ID(dest);
 
-	desc = irq_to_desc(irq);
 	spin_lock_irqsave(&ioapic_lock, flags);
-	__target_IO_APIC_irq(irq, dest, cfg->vector);
+	__target_IO_APIC_irq(irq, dest, cfg);
 	desc->affinity = mask;
 	spin_unlock_irqrestore(&ioapic_lock, flags);
 }
+
+static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
+{
+	struct irq_desc *desc;
+
+	desc = irq_to_desc(irq);
+
+	set_ioapic_affinity_irq_desc(desc, mask);
+}
 #endif /* CONFIG_SMP */
 
 /*
@@ -397,16 +565,18 @@ static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
  * shared ISA-space IRQs, so we have to support them. We are super
  * fast in the common case, and fast for shared ISA-space IRQs.
  */
-static void add_pin_to_irq(unsigned int irq, int apic, int pin)
+static void add_pin_to_irq_cpu(struct irq_cfg *cfg, int cpu, int apic, int pin)
 {
-	struct irq_cfg *cfg;
 	struct irq_pin_list *entry;
 
-	/* first time to refer irq_cfg, so with new */
-	cfg = irq_cfg_alloc(irq);
 	entry = cfg->irq_2_pin;
 	if (!entry) {
-		entry = get_one_free_irq_2_pin();
+		entry = get_one_free_irq_2_pin(cpu);
+		if (!entry) {
+			printk(KERN_ERR "can not alloc irq_2_pin to add %d - %d\n",
+					apic, pin);
+			return;
+		}
 		cfg->irq_2_pin = entry;
 		entry->apic = apic;
 		entry->pin = pin;
@@ -421,7 +591,7 @@ static void add_pin_to_irq(unsigned int irq, int apic, int pin)
 		entry = entry->next;
 	}
 
-	entry->next = get_one_free_irq_2_pin();
+	entry->next = get_one_free_irq_2_pin(cpu);
 	entry = entry->next;
 	entry->apic = apic;
 	entry->pin = pin;
@@ -430,11 +600,10 @@ static void add_pin_to_irq(unsigned int irq, int apic, int pin)
 /*
  * Reroute an IRQ to a different pin.
  */
-static void __init replace_pin_at_irq(unsigned int irq,
+static void __init replace_pin_at_irq_cpu(struct irq_cfg *cfg, int cpu,
 				      int oldapic, int oldpin,
 				      int newapic, int newpin)
 {
-	struct irq_cfg *cfg = irq_cfg(irq);
 	struct irq_pin_list *entry = cfg->irq_2_pin;
 	int replaced = 0;
 
@@ -451,18 +620,16 @@ static void __init replace_pin_at_irq(unsigned int irq,
 
 	/* why? call replace before add? */
 	if (!replaced)
-		add_pin_to_irq(irq, newapic, newpin);
+		add_pin_to_irq_cpu(cfg, cpu, newapic, newpin);
 }
 
-static inline void io_apic_modify_irq(unsigned int irq,
+static inline void io_apic_modify_irq(struct irq_cfg *cfg,
 				int mask_and, int mask_or,
 				void (*final)(struct irq_pin_list *entry))
 {
 	int pin;
-	struct irq_cfg *cfg;
 	struct irq_pin_list *entry;
 
-	cfg = irq_cfg(irq);
 	for (entry = cfg->irq_2_pin; entry != NULL; entry = entry->next) {
 		unsigned int reg;
 		pin = entry->pin;
@@ -475,9 +642,9 @@ static inline void io_apic_modify_irq(unsigned int irq,
 	}
 }
 
-static void __unmask_IO_APIC_irq(unsigned int irq)
+static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
 {
-	io_apic_modify_irq(irq, ~IO_APIC_REDIR_MASKED, 0, NULL);
+	io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
 }
 
 #ifdef CONFIG_X86_64
@@ -492,47 +659,64 @@ void io_apic_sync(struct irq_pin_list *entry)
 	readl(&io_apic->data);
 }
 
-static void __mask_IO_APIC_irq(unsigned int irq)
+static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
 {
-	io_apic_modify_irq(irq, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
+	io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
 }
 #else /* CONFIG_X86_32 */
-static void __mask_IO_APIC_irq(unsigned int irq)
+static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
 {
-	io_apic_modify_irq(irq, ~0, IO_APIC_REDIR_MASKED, NULL);
+	io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, NULL);
 }
 
-static void __mask_and_edge_IO_APIC_irq(unsigned int irq)
+static void __mask_and_edge_IO_APIC_irq(struct irq_cfg *cfg)
 {
-	io_apic_modify_irq(irq, ~IO_APIC_REDIR_LEVEL_TRIGGER,
+	io_apic_modify_irq(cfg, ~IO_APIC_REDIR_LEVEL_TRIGGER,
 			IO_APIC_REDIR_MASKED, NULL);
 }
 
-static void __unmask_and_level_IO_APIC_irq(unsigned int irq)
+static void __unmask_and_level_IO_APIC_irq(struct irq_cfg *cfg)
 {
-	io_apic_modify_irq(irq, ~IO_APIC_REDIR_MASKED,
+	io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED,
 			IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
 }
 #endif /* CONFIG_X86_32 */
 
-static void mask_IO_APIC_irq (unsigned int irq)
+static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
 {
+	struct irq_cfg *cfg = desc->chip_data;
 	unsigned long flags;
 
+	BUG_ON(!cfg);
+
 	spin_lock_irqsave(&ioapic_lock, flags);
-	__mask_IO_APIC_irq(irq);
+	__mask_IO_APIC_irq(cfg);
 	spin_unlock_irqrestore(&ioapic_lock, flags);
 }
 
-static void unmask_IO_APIC_irq (unsigned int irq)
+static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
 {
+	struct irq_cfg *cfg = desc->chip_data;
 	unsigned long flags;
 
 	spin_lock_irqsave(&ioapic_lock, flags);
-	__unmask_IO_APIC_irq(irq);
+	__unmask_IO_APIC_irq(cfg);
 	spin_unlock_irqrestore(&ioapic_lock, flags);
 }
 
+static void mask_IO_APIC_irq(unsigned int irq)
+{
+	struct irq_desc *desc = irq_to_desc(irq);
+
+	mask_IO_APIC_irq_desc(desc);
+}
+static void unmask_IO_APIC_irq(unsigned int irq)
+{
+	struct irq_desc *desc = irq_to_desc(irq);
+
+	unmask_IO_APIC_irq_desc(desc);
+}
+
 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
 {
 	struct IO_APIC_route_entry entry;
@@ -809,7 +993,7 @@ EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
  */
 static int EISA_ELCR(unsigned int irq)
 {
-	if (irq < 16) {
+	if (irq < NR_IRQS_LEGACY) {
 		unsigned int port = 0x4d0 + (irq >> 3);
 		return (inb(port) >> (irq & 7)) & 1;
 	}
@@ -1034,7 +1218,7 @@ void unlock_vector_lock(void)
 	spin_unlock(&vector_lock);
 }
 
-static int __assign_irq_vector(int irq, cpumask_t mask)
+static int __assign_irq_vector(int irq, struct irq_cfg *cfg, cpumask_t mask)
 {
 	/*
 	 * NOTE! The local APIC isn't very good at handling
@@ -1050,16 +1234,13 @@ static int __assign_irq_vector(int irq, cpumask_t mask)
 	static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
 	unsigned int old_vector;
 	int cpu;
-	struct irq_cfg *cfg;
 
-	cfg = irq_cfg(irq);
+	if ((cfg->move_in_progress) || cfg->move_cleanup_count)
+		return -EBUSY;
 
 	/* Only try and allocate irqs on cpus that are present */
 	cpus_and(mask, mask, cpu_online_map);
 
-	if ((cfg->move_in_progress) || cfg->move_cleanup_count)
-		return -EBUSY;
-
 	old_vector = cfg->vector;
 	if (old_vector) {
 		cpumask_t tmp;
@@ -1113,24 +1294,22 @@ static int __assign_irq_vector(int irq, cpumask_t mask)
 	return -ENOSPC;
 }
 
-static int assign_irq_vector(int irq, cpumask_t mask)
+static int assign_irq_vector(int irq, struct irq_cfg *cfg, cpumask_t mask)
 {
 	int err;
 	unsigned long flags;
 
 	spin_lock_irqsave(&vector_lock, flags);
-	err = __assign_irq_vector(irq, mask);
+	err = __assign_irq_vector(irq, cfg, mask);
 	spin_unlock_irqrestore(&vector_lock, flags);
 	return err;
 }
 
-static void __clear_irq_vector(int irq)
+static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
 {
-	struct irq_cfg *cfg;
 	cpumask_t mask;
 	int cpu, vector;
 
-	cfg = irq_cfg(irq);
 	BUG_ON(!cfg->vector);
 
 	vector = cfg->vector;
@@ -1162,9 +1341,13 @@ void __setup_vector_irq(int cpu)
 	/* This function must be called with vector_lock held */
 	int irq, vector;
 	struct irq_cfg *cfg;
+	struct irq_desc *desc;
 
 	/* Mark the inuse vectors */
-	for_each_irq_cfg(irq, cfg) {
+	for_each_irq_desc(irq, desc) {
+		if (!desc)
+			continue;
+		cfg = desc->chip_data;
 		if (!cpu_isset(cpu, cfg->domain))
 			continue;
 		vector = cfg->vector;
@@ -1215,11 +1398,8 @@ static inline int IO_APIC_irq_trigger(int irq)
 }
 #endif
 
-static void ioapic_register_intr(int irq, unsigned long trigger)
+static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
 {
-	struct irq_desc *desc;
-
-	desc = irq_to_desc(irq);
 
 	if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
 	    trigger == IOAPIC_LEVEL)
@@ -1311,7 +1491,7 @@ static int setup_ioapic_entry(int apic, int irq,
 	return 0;
 }
 
-static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
+static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq, struct irq_desc *desc,
 			      int trigger, int polarity)
 {
 	struct irq_cfg *cfg;
@@ -1321,10 +1501,10 @@ static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
 	if (!IO_APIC_IRQ(irq))
 		return;
 
-	cfg = irq_cfg(irq);
+	cfg = desc->chip_data;
 
 	mask = TARGET_CPUS;
-	if (assign_irq_vector(irq, mask))
+	if (assign_irq_vector(irq, cfg, mask))
 		return;
 
 	cpus_and(mask, cfg->domain, mask);
@@ -1341,12 +1521,12 @@ static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
 			       cfg->vector)) {
 		printk("Failed to setup ioapic entry for ioapic  %d, pin %d\n",
 		       mp_ioapics[apic].mp_apicid, pin);
-		__clear_irq_vector(irq);
+		__clear_irq_vector(irq, cfg);
 		return;
 	}
 
-	ioapic_register_intr(irq, trigger);
-	if (irq < 16)
+	ioapic_register_intr(irq, desc, trigger);
+	if (irq < NR_IRQS_LEGACY)
 		disable_8259A_irq(irq);
 
 	ioapic_write_entry(apic, pin, entry);
@@ -1356,6 +1536,9 @@ static void __init setup_IO_APIC_irqs(void)
 {
 	int apic, pin, idx, irq;
 	int notcon = 0;
+	struct irq_desc *desc;
+	struct irq_cfg *cfg;
+	int cpu = boot_cpu_id;
 
 	apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
 
@@ -1387,9 +1570,15 @@ static void __init setup_IO_APIC_irqs(void)
 			if (multi_timer_check(apic, irq))
 				continue;
 #endif
-			add_pin_to_irq(irq, apic, pin);
+			desc = irq_to_desc_alloc_cpu(irq, cpu);
+			if (!desc) {
+				printk(KERN_INFO "can not get irq_desc for %d\n", irq);
+				continue;
+			}
+			cfg = desc->chip_data;
+			add_pin_to_irq_cpu(cfg, cpu, apic, pin);
 
-			setup_IO_APIC_irq(apic, pin, irq,
+			setup_IO_APIC_irq(apic, pin, irq, desc,
 					irq_trigger(idx), irq_polarity(idx));
 		}
 	}
@@ -1448,6 +1637,7 @@ __apicdebuginit(void) print_IO_APIC(void)
 	union IO_APIC_reg_03 reg_03;
 	unsigned long flags;
 	struct irq_cfg *cfg;
+	struct irq_desc *desc;
 	unsigned int irq;
 
 	if (apic_verbosity == APIC_QUIET)
@@ -1537,8 +1727,13 @@ __apicdebuginit(void) print_IO_APIC(void)
 	}
 	}
 	printk(KERN_DEBUG "IRQ to pin mappings:\n");
-	for_each_irq_cfg(irq, cfg) {
-		struct irq_pin_list *entry = cfg->irq_2_pin;
+	for_each_irq_desc(irq, desc) {
+		struct irq_pin_list *entry;
+
+		if (!desc)
+			continue;
+		cfg = desc->chip_data;
+		entry = cfg->irq_2_pin;
 		if (!entry)
 			continue;
 		printk(KERN_DEBUG "IRQ%d ", irq);
@@ -2022,14 +2217,16 @@ static unsigned int startup_ioapic_irq(unsigned int irq)
 {
 	int was_pending = 0;
 	unsigned long flags;
+	struct irq_cfg *cfg;
 
 	spin_lock_irqsave(&ioapic_lock, flags);
-	if (irq < 16) {
+	if (irq < NR_IRQS_LEGACY) {
 		disable_8259A_irq(irq);
 		if (i8259A_irq_pending(irq))
 			was_pending = 1;
 	}
-	__unmask_IO_APIC_irq(irq);
+	cfg = irq_cfg(irq);
+	__unmask_IO_APIC_irq(cfg);
 	spin_unlock_irqrestore(&ioapic_lock, flags);
 
 	return was_pending;
@@ -2092,35 +2289,37 @@ static DECLARE_DELAYED_WORK(ir_migration_work, ir_irq_migration);
  * as simple as edge triggered migration and we can do the irq migration
  * with a simple atomic update to IO-APIC RTE.
  */
-static void migrate_ioapic_irq(int irq, cpumask_t mask)
+static void migrate_ioapic_irq_desc(struct irq_desc *desc, cpumask_t mask)
 {
 	struct irq_cfg *cfg;
-	struct irq_desc *desc;
 	cpumask_t tmp, cleanup_mask;
 	struct irte irte;
 	int modify_ioapic_rte;
 	unsigned int dest;
 	unsigned long flags;
+	unsigned int irq;
 
 	cpus_and(tmp, mask, cpu_online_map);
 	if (cpus_empty(tmp))
 		return;
 
+	irq = desc->irq;
 	if (get_irte(irq, &irte))
 		return;
 
-	if (assign_irq_vector(irq, mask))
+	cfg = desc->chip_data;
+	if (assign_irq_vector(irq, cfg, mask))
 		return;
 
-	cfg = irq_cfg(irq);
+	set_extra_move_desc(desc, mask);
+
 	cpus_and(tmp, cfg->domain, mask);
 	dest = cpu_mask_to_apicid(tmp);
 
-	desc = irq_to_desc(irq);
 	modify_ioapic_rte = desc->status & IRQ_LEVEL;
 	if (modify_ioapic_rte) {
 		spin_lock_irqsave(&ioapic_lock, flags);
-		__target_IO_APIC_irq(irq, dest, cfg->vector);
+		__target_IO_APIC_irq(irq, dest, cfg);
 		spin_unlock_irqrestore(&ioapic_lock, flags);
 	}
 
@@ -2142,14 +2341,14 @@ static void migrate_ioapic_irq(int irq, cpumask_t mask)
 	desc->affinity = mask;
 }
 
-static int migrate_irq_remapped_level(int irq)
+static int migrate_irq_remapped_level_desc(struct irq_desc *desc)
 {
 	int ret = -1;
-	struct irq_desc *desc = irq_to_desc(irq);
+	struct irq_cfg *cfg = desc->chip_data;
 
-	mask_IO_APIC_irq(irq);
+	mask_IO_APIC_irq_desc(desc);
 
-	if (io_apic_level_ack_pending(irq)) {
+	if (io_apic_level_ack_pending(cfg)) {
 		/*
 		 * Interrupt in progress. Migrating irq now will change the
 		 * vector information in the IO-APIC RTE and that will confuse
@@ -2161,14 +2360,15 @@ static int migrate_irq_remapped_level(int irq)
 	}
 
 	/* everthing is clear. we have right of way */
-	migrate_ioapic_irq(irq, desc->pending_mask);
+	migrate_ioapic_irq_desc(desc, desc->pending_mask);
 
 	ret = 0;
 	desc->status &= ~IRQ_MOVE_PENDING;
 	cpus_clear(desc->pending_mask);
 
 unmask:
-	unmask_IO_APIC_irq(irq);
+	unmask_IO_APIC_irq_desc(desc);
+
 	return ret;
 }
 
@@ -2178,6 +2378,9 @@ static void ir_irq_migration(struct work_struct *work)
 	struct irq_desc *desc;
 
 	for_each_irq_desc(irq, desc) {
+		if (!desc)
+			continue;
+
 		if (desc->status & IRQ_MOVE_PENDING) {
 			unsigned long flags;
 
@@ -2198,18 +2401,22 @@ static void ir_irq_migration(struct work_struct *work)
 /*
  * Migrates the IRQ destination in the process context.
  */
-static void set_ir_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
+static void set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc, cpumask_t mask)
 {
-	struct irq_desc *desc = irq_to_desc(irq);
-
 	if (desc->status & IRQ_LEVEL) {
 		desc->status |= IRQ_MOVE_PENDING;
 		desc->pending_mask = mask;
-		migrate_irq_remapped_level(irq);
+		migrate_irq_remapped_level_desc(desc);
 		return;
 	}
 
-	migrate_ioapic_irq(irq, mask);
+	migrate_ioapic_irq_desc(desc, mask);
+}
+static void set_ir_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
+{
+	struct irq_desc *desc = irq_to_desc(irq);
+
+	set_ir_ioapic_affinity_irq_desc(desc, mask);
 }
 #endif
 
@@ -2228,6 +2435,9 @@ asmlinkage void smp_irq_move_cleanup_interrupt(void)
 		struct irq_cfg *cfg;
 		irq = __get_cpu_var(vector_irq)[vector];
 
+		if (irq == -1)
+			continue;
+
 		desc = irq_to_desc(irq);
 		if (!desc)
 			continue;
@@ -2249,19 +2459,40 @@ asmlinkage void smp_irq_move_cleanup_interrupt(void)
 	irq_exit();
 }
 
-static void irq_complete_move(unsigned int irq)
+static void irq_complete_move(struct irq_desc **descp)
 {
-	struct irq_cfg *cfg = irq_cfg(irq);
+	struct irq_desc *desc = *descp;
+	struct irq_cfg *cfg = desc->chip_data;
 	unsigned vector, me;
 
-	if (likely(!cfg->move_in_progress))
+	if (likely(!cfg->move_in_progress)) {
+#ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
+		if (likely(!cfg->move_desc_pending))
+			return;
+
+		/* domain has not changed, but affinity did */
+		me = smp_processor_id();
+		if (cpu_isset(me, desc->affinity)) {
+			*descp = desc = move_irq_desc(desc, me);
+			/* get the new one */
+			cfg = desc->chip_data;
+			cfg->move_desc_pending = 0;
+		}
+#endif
 		return;
+	}
 
 	vector = ~get_irq_regs()->orig_ax;
 	me = smp_processor_id();
 	if ((vector == cfg->vector) && cpu_isset(me, cfg->domain)) {
 		cpumask_t cleanup_mask;
 
+#ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
+		*descp = desc = move_irq_desc(desc, me);
+		/* get the new one */
+		cfg = desc->chip_data;
+#endif
+
 		cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
 		cfg->move_cleanup_count = cpus_weight(cleanup_mask);
 		send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
@@ -2269,8 +2500,9 @@ static void irq_complete_move(unsigned int irq)
 	}
 }
 #else
-static inline void irq_complete_move(unsigned int irq) {}
+static inline void irq_complete_move(struct irq_desc **descp) {}
 #endif
+
 #ifdef CONFIG_INTR_REMAP
 static void ack_x2apic_level(unsigned int irq)
 {
@@ -2281,11 +2513,14 @@ static void ack_x2apic_edge(unsigned int irq)
 {
 	ack_x2APIC_irq();
 }
+
 #endif
 
 static void ack_apic_edge(unsigned int irq)
 {
-	irq_complete_move(irq);
+	struct irq_desc *desc = irq_to_desc(irq);
+
+	irq_complete_move(&desc);
 	move_native_irq(irq);
 	ack_APIC_irq();
 }
@@ -2294,18 +2529,21 @@ atomic_t irq_mis_count;
 
 static void ack_apic_level(unsigned int irq)
 {
+	struct irq_desc *desc = irq_to_desc(irq);
+
 #ifdef CONFIG_X86_32
 	unsigned long v;
 	int i;
 #endif
+	struct irq_cfg *cfg;
 	int do_unmask_irq = 0;
 
-	irq_complete_move(irq);
+	irq_complete_move(&desc);
 #ifdef CONFIG_GENERIC_PENDING_IRQ
 	/* If we are moving the irq we need to mask it */
-	if (unlikely(irq_to_desc(irq)->status & IRQ_MOVE_PENDING)) {
+	if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
 		do_unmask_irq = 1;
-		mask_IO_APIC_irq(irq);
+		mask_IO_APIC_irq_desc(desc);
 	}
 #endif
 
@@ -2329,7 +2567,8 @@ static void ack_apic_level(unsigned int irq)
 	* operation to prevent an edge-triggered interrupt escaping meanwhile.
 	* The idea is from Manfred Spraul.  --macro
 	*/
-	i = irq_cfg(irq)->vector;
+	cfg = desc->chip_data;
+	i = cfg->vector;
 
 	v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
 #endif
@@ -2368,17 +2607,18 @@ static void ack_apic_level(unsigned int irq)
 		 * accurate and is causing problems then it is a hardware bug
 		 * and you can go talk to the chipset vendor about it.
 		 */
-		if (!io_apic_level_ack_pending(irq))
+		cfg = desc->chip_data;
+		if (!io_apic_level_ack_pending(cfg))
 			move_masked_irq(irq);
-		unmask_IO_APIC_irq(irq);
+		unmask_IO_APIC_irq_desc(desc);
 	}
 
 #ifdef CONFIG_X86_32
 	if (!(v & (1 << (i & 0x1f)))) {
 		atomic_inc(&irq_mis_count);
 		spin_lock(&ioapic_lock);
-		__mask_and_edge_IO_APIC_irq(irq);
-		__unmask_and_level_IO_APIC_irq(irq);
+		__mask_and_edge_IO_APIC_irq(cfg);
+		__unmask_and_level_IO_APIC_irq(cfg);
 		spin_unlock(&ioapic_lock);
 	}
 #endif
@@ -2429,20 +2669,22 @@ static inline void init_IO_APIC_traps(void)
 	 * Also, we've got to be careful not to trash gate
 	 * 0x80, because int 0x80 is hm, kind of importantish. ;)
 	 */
-	for_each_irq_cfg(irq, cfg) {
-		if (IO_APIC_IRQ(irq) && !cfg->vector) {
+	for_each_irq_desc(irq, desc) {
+		if (!desc)
+			continue;
+
+		cfg = desc->chip_data;
+		if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
 			/*
 			 * Hmm.. We don't have an entry for this,
 			 * so default to an old-fashioned 8259
 			 * interrupt if we can..
 			 */
-			if (irq < 16)
+			if (irq < NR_IRQS_LEGACY)
 				make_8259A_irq(irq);
-			else {
-				desc = irq_to_desc(irq);
+			else
 				/* Strange. Oh, well.. */
 				desc->chip = &no_irq_chip;
-			}
 		}
 	}
 }
@@ -2467,7 +2709,7 @@ static void unmask_lapic_irq(unsigned int irq)
 	apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
 }
 
-static void ack_lapic_irq (unsigned int irq)
+static void ack_lapic_irq(unsigned int irq)
 {
 	ack_APIC_irq();
 }
@@ -2479,11 +2721,8 @@ static struct irq_chip lapic_chip __read_mostly = {
 	.ack		= ack_lapic_irq,
 };
 
-static void lapic_register_intr(int irq)
+static void lapic_register_intr(int irq, struct irq_desc *desc)
 {
-	struct irq_desc *desc;
-
-	desc = irq_to_desc(irq);
 	desc->status &= ~IRQ_LEVEL;
 	set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
 				      "edge");
@@ -2587,7 +2826,9 @@ int timer_through_8259 __initdata;
  */
 static inline void __init check_timer(void)
 {
-	struct irq_cfg *cfg = irq_cfg(0);
+	struct irq_desc *desc = irq_to_desc(0);
+	struct irq_cfg *cfg = desc->chip_data;
+	int cpu = boot_cpu_id;
 	int apic1, pin1, apic2, pin2;
 	unsigned long flags;
 	unsigned int ver;
@@ -2602,7 +2843,7 @@ static inline void __init check_timer(void)
 	 * get/set the timer IRQ vector:
 	 */
 	disable_8259A_irq(0);
-	assign_irq_vector(0, TARGET_CPUS);
+	assign_irq_vector(0, cfg, TARGET_CPUS);
 
 	/*
 	 * As IRQ0 is to be enabled in the 8259A, the virtual
@@ -2653,10 +2894,10 @@ static inline void __init check_timer(void)
 		 * Ok, does IRQ0 through the IOAPIC work?
 		 */
 		if (no_pin1) {
-			add_pin_to_irq(0, apic1, pin1);
+			add_pin_to_irq_cpu(cfg, cpu, apic1, pin1);
 			setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
 		}
-		unmask_IO_APIC_irq(0);
+		unmask_IO_APIC_irq_desc(desc);
 		if (timer_irq_works()) {
 			if (nmi_watchdog == NMI_IO_APIC) {
 				setup_nmi();
@@ -2682,9 +2923,9 @@ static inline void __init check_timer(void)
 		/*
 		 * legacy devices should be connected to IO APIC #0
 		 */
-		replace_pin_at_irq(0, apic1, pin1, apic2, pin2);
+		replace_pin_at_irq_cpu(cfg, cpu, apic1, pin1, apic2, pin2);
 		setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
-		unmask_IO_APIC_irq(0);
+		unmask_IO_APIC_irq_desc(desc);
 		enable_8259A_irq(0);
 		if (timer_irq_works()) {
 			apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
@@ -2716,7 +2957,7 @@ static inline void __init check_timer(void)
 	apic_printk(APIC_QUIET, KERN_INFO
 		    "...trying to set up timer as Virtual Wire IRQ...\n");
 
-	lapic_register_intr(0);
+	lapic_register_intr(0, desc);
 	apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector);	/* Fixed mode */
 	enable_8259A_irq(0);
 
@@ -2901,22 +3142,26 @@ unsigned int create_irq_nr(unsigned int irq_want)
 	unsigned int irq;
 	unsigned int new;
 	unsigned long flags;
-	struct irq_cfg *cfg_new;
-
-	irq_want = nr_irqs - 1;
+	struct irq_cfg *cfg_new = NULL;
+	int cpu = boot_cpu_id;
+	struct irq_desc *desc_new = NULL;
 
 	irq = 0;
 	spin_lock_irqsave(&vector_lock, flags);
-	for (new = irq_want; new > 0; new--) {
+	for (new = irq_want; new < NR_IRQS; new++) {
 		if (platform_legacy_irq(new))
 			continue;
-		cfg_new = irq_cfg(new);
-		if (cfg_new && cfg_new->vector != 0)
+
+		desc_new = irq_to_desc_alloc_cpu(new, cpu);
+		if (!desc_new) {
+			printk(KERN_INFO "can not get irq_desc for %d\n", new);
 			continue;
-		/* check if need to create one */
-		if (!cfg_new)
-			cfg_new = irq_cfg_alloc(new);
-		if (__assign_irq_vector(new, TARGET_CPUS) == 0)
+		}
+		cfg_new = desc_new->chip_data;
+
+		if (cfg_new->vector != 0)
+			continue;
+		if (__assign_irq_vector(new, cfg_new, TARGET_CPUS) == 0)
 			irq = new;
 		break;
 	}
@@ -2924,15 +3169,21 @@ unsigned int create_irq_nr(unsigned int irq_want)
 
 	if (irq > 0) {
 		dynamic_irq_init(irq);
+		/* restore it, in case dynamic_irq_init clear it */
+		if (desc_new)
+			desc_new->chip_data = cfg_new;
 	}
 	return irq;
 }
 
+static int nr_irqs_gsi = NR_IRQS_LEGACY;
 int create_irq(void)
 {
+	unsigned int irq_want;
 	int irq;
 
-	irq = create_irq_nr(nr_irqs - 1);
+	irq_want = nr_irqs_gsi;
+	irq = create_irq_nr(irq_want);
 
 	if (irq == 0)
 		irq = -1;
@@ -2943,14 +3194,22 @@ int create_irq(void)
 void destroy_irq(unsigned int irq)
 {
 	unsigned long flags;
+	struct irq_cfg *cfg;
+	struct irq_desc *desc;
 
+	/* store it, in case dynamic_irq_cleanup clear it */
+	desc = irq_to_desc(irq);
+	cfg = desc->chip_data;
 	dynamic_irq_cleanup(irq);
+	/* connect back irq_cfg */
+	if (desc)
+		desc->chip_data = cfg;
 
 #ifdef CONFIG_INTR_REMAP
 	free_irte(irq);
 #endif
 	spin_lock_irqsave(&vector_lock, flags);
-	__clear_irq_vector(irq);
+	__clear_irq_vector(irq, cfg);
 	spin_unlock_irqrestore(&vector_lock, flags);
 }
 
@@ -2965,12 +3224,12 @@ static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_ms
 	unsigned dest;
 	cpumask_t tmp;
 
+	cfg = irq_cfg(irq);
 	tmp = TARGET_CPUS;
-	err = assign_irq_vector(irq, tmp);
+	err = assign_irq_vector(irq, cfg, tmp);
 	if (err)
 		return err;
 
-	cfg = irq_cfg(irq);
 	cpus_and(tmp, cfg->domain, tmp);
 	dest = cpu_mask_to_apicid(tmp);
 
@@ -3028,35 +3287,35 @@ static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_ms
 #ifdef CONFIG_SMP
 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
 {
+	struct irq_desc *desc = irq_to_desc(irq);
 	struct irq_cfg *cfg;
 	struct msi_msg msg;
 	unsigned int dest;
 	cpumask_t tmp;
-	struct irq_desc *desc;
 
 	cpus_and(tmp, mask, cpu_online_map);
 	if (cpus_empty(tmp))
 		return;
 
-	if (assign_irq_vector(irq, mask))
+	cfg = desc->chip_data;
+	if (assign_irq_vector(irq, cfg, mask))
 		return;
 
-	cfg = irq_cfg(irq);
+	set_extra_move_desc(desc, mask);
+
 	cpus_and(tmp, cfg->domain, mask);
 	dest = cpu_mask_to_apicid(tmp);
 
-	read_msi_msg(irq, &msg);
+	read_msi_msg_desc(desc, &msg);
 
 	msg.data &= ~MSI_DATA_VECTOR_MASK;
 	msg.data |= MSI_DATA_VECTOR(cfg->vector);
 	msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
 	msg.address_lo |= MSI_ADDR_DEST_ID(dest);
 
-	write_msi_msg(irq, &msg);
-	desc = irq_to_desc(irq);
+	write_msi_msg_desc(desc, &msg);
 	desc->affinity = mask;
 }
-
 #ifdef CONFIG_INTR_REMAP
 /*
  * Migrate the MSI irq to another cpumask. This migration is
@@ -3064,11 +3323,11 @@ static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
  */
 static void ir_set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
 {
+	struct irq_desc *desc = irq_to_desc(irq);
 	struct irq_cfg *cfg;
 	unsigned int dest;
 	cpumask_t tmp, cleanup_mask;
 	struct irte irte;
-	struct irq_desc *desc;
 
 	cpus_and(tmp, mask, cpu_online_map);
 	if (cpus_empty(tmp))
@@ -3077,10 +3336,12 @@ static void ir_set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
 	if (get_irte(irq, &irte))
 		return;
 
-	if (assign_irq_vector(irq, mask))
+	cfg = desc->chip_data;
+	if (assign_irq_vector(irq, cfg, mask))
 		return;
 
-	cfg = irq_cfg(irq);
+	set_extra_move_desc(desc, mask);
+
 	cpus_and(tmp, cfg->domain, mask);
 	dest = cpu_mask_to_apicid(tmp);
 
@@ -3104,9 +3365,9 @@ static void ir_set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
 		cfg->move_in_progress = 0;
 	}
 
-	desc = irq_to_desc(irq);
 	desc->affinity = mask;
 }
+
 #endif
 #endif /* CONFIG_SMP */
 
@@ -3165,7 +3426,7 @@ static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
 }
 #endif
 
-static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc, int irq)
+static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
 {
 	int ret;
 	struct msi_msg msg;
@@ -3174,7 +3435,7 @@ static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc, int irq)
 	if (ret < 0)
 		return ret;
 
-	set_irq_msi(irq, desc);
+	set_irq_msi(irq, msidesc);
 	write_msi_msg(irq, &msg);
 
 #ifdef CONFIG_INTR_REMAP
@@ -3194,26 +3455,13 @@ static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc, int irq)
 	return 0;
 }
 
-static unsigned int build_irq_for_pci_dev(struct pci_dev *dev)
-{
-	unsigned int irq;
-
-	irq = dev->bus->number;
-	irq <<= 8;
-	irq |= dev->devfn;
-	irq <<= 12;
-
-	return irq;
-}
-
-int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
+int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc)
 {
 	unsigned int irq;
 	int ret;
 	unsigned int irq_want;
 
-	irq_want = build_irq_for_pci_dev(dev) + 0x100;
-
+	irq_want = nr_irqs_gsi;
 	irq = create_irq_nr(irq_want);
 	if (irq == 0)
 		return -1;
@@ -3227,7 +3475,7 @@ int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
 		goto error;
 no_ir:
 #endif
-	ret = setup_msi_irq(dev, desc, irq);
+	ret = setup_msi_irq(dev, msidesc, irq);
 	if (ret < 0) {
 		destroy_irq(irq);
 		return ret;
@@ -3245,7 +3493,7 @@ int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 {
 	unsigned int irq;
 	int ret, sub_handle;
-	struct msi_desc *desc;
+	struct msi_desc *msidesc;
 	unsigned int irq_want;
 
 #ifdef CONFIG_INTR_REMAP
@@ -3253,10 +3501,11 @@ int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 	int index = 0;
 #endif
 
-	irq_want = build_irq_for_pci_dev(dev) + 0x100;
+	irq_want = nr_irqs_gsi;
 	sub_handle = 0;
-	list_for_each_entry(desc, &dev->msi_list, list) {
-		irq = create_irq_nr(irq_want--);
+	list_for_each_entry(msidesc, &dev->msi_list, list) {
+		irq = create_irq_nr(irq_want);
+		irq_want++;
 		if (irq == 0)
 			return -1;
 #ifdef CONFIG_INTR_REMAP
@@ -3288,7 +3537,7 @@ int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 		}
 no_ir:
 #endif
-		ret = setup_msi_irq(dev, desc, irq);
+		ret = setup_msi_irq(dev, msidesc, irq);
 		if (ret < 0)
 			goto error;
 		sub_handle++;
@@ -3309,20 +3558,22 @@ void arch_teardown_msi_irq(unsigned int irq)
 #ifdef CONFIG_SMP
 static void dmar_msi_set_affinity(unsigned int irq, cpumask_t mask)
 {
+	struct irq_desc *desc = irq_to_desc(irq);
 	struct irq_cfg *cfg;
 	struct msi_msg msg;
 	unsigned int dest;
 	cpumask_t tmp;
-	struct irq_desc *desc;
 
 	cpus_and(tmp, mask, cpu_online_map);
 	if (cpus_empty(tmp))
 		return;
 
-	if (assign_irq_vector(irq, mask))
+	cfg = desc->chip_data;
+	if (assign_irq_vector(irq, cfg, mask))
 		return;
 
-	cfg = irq_cfg(irq);
+	set_extra_move_desc(desc, mask);
+
 	cpus_and(tmp, cfg->domain, mask);
 	dest = cpu_mask_to_apicid(tmp);
 
@@ -3334,9 +3585,9 @@ static void dmar_msi_set_affinity(unsigned int irq, cpumask_t mask)
 	msg.address_lo |= MSI_ADDR_DEST_ID(dest);
 
 	dmar_msi_write(irq, &msg);
-	desc = irq_to_desc(irq);
 	desc->affinity = mask;
 }
+
 #endif /* CONFIG_SMP */
 
 struct irq_chip dmar_msi_type = {
@@ -3370,8 +3621,8 @@ int arch_setup_dmar_msi(unsigned int irq)
 #ifdef CONFIG_SMP
 static void hpet_msi_set_affinity(unsigned int irq, cpumask_t mask)
 {
+	struct irq_desc *desc = irq_to_desc(irq);
 	struct irq_cfg *cfg;
-	struct irq_desc *desc;
 	struct msi_msg msg;
 	unsigned int dest;
 	cpumask_t tmp;
@@ -3380,10 +3631,12 @@ static void hpet_msi_set_affinity(unsigned int irq, cpumask_t mask)
 	if (cpus_empty(tmp))
 		return;
 
-	if (assign_irq_vector(irq, mask))
+	cfg = desc->chip_data;
+	if (assign_irq_vector(irq, cfg, mask))
 		return;
 
-	cfg = irq_cfg(irq);
+	set_extra_move_desc(desc, mask);
+
 	cpus_and(tmp, cfg->domain, mask);
 	dest = cpu_mask_to_apicid(tmp);
 
@@ -3395,9 +3648,9 @@ static void hpet_msi_set_affinity(unsigned int irq, cpumask_t mask)
 	msg.address_lo |= MSI_ADDR_DEST_ID(dest);
 
 	hpet_msi_write(irq, &msg);
-	desc = irq_to_desc(irq);
 	desc->affinity = mask;
 }
+
 #endif /* CONFIG_SMP */
 
 struct irq_chip hpet_msi_type = {
@@ -3452,26 +3705,28 @@ static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
 
 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
 {
+	struct irq_desc *desc = irq_to_desc(irq);
 	struct irq_cfg *cfg;
 	unsigned int dest;
 	cpumask_t tmp;
-	struct irq_desc *desc;
 
 	cpus_and(tmp, mask, cpu_online_map);
 	if (cpus_empty(tmp))
 		return;
 
-	if (assign_irq_vector(irq, mask))
+	cfg = desc->chip_data;
+	if (assign_irq_vector(irq, cfg, mask))
 		return;
 
-	cfg = irq_cfg(irq);
+	set_extra_move_desc(desc, mask);
+
 	cpus_and(tmp, cfg->domain, mask);
 	dest = cpu_mask_to_apicid(tmp);
 
 	target_ht_irq(irq, dest, cfg->vector);
-	desc = irq_to_desc(irq);
 	desc->affinity = mask;
 }
+
 #endif
 
 static struct irq_chip ht_irq_chip = {
@@ -3491,13 +3746,13 @@ int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
 	int err;
 	cpumask_t tmp;
 
+	cfg = irq_cfg(irq);
 	tmp = TARGET_CPUS;
-	err = assign_irq_vector(irq, tmp);
+	err = assign_irq_vector(irq, cfg, tmp);
 	if (!err) {
 		struct ht_irq_msg msg;
 		unsigned dest;
 
-		cfg = irq_cfg(irq);
 		cpus_and(tmp, cfg->domain, tmp);
 		dest = cpu_mask_to_apicid(tmp);
 
@@ -3543,7 +3798,9 @@ int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
 	unsigned long flags;
 	int err;
 
-	err = assign_irq_vector(irq, *eligible_cpu);
+	cfg = irq_cfg(irq);
+
+	err = assign_irq_vector(irq, cfg, *eligible_cpu);
 	if (err != 0)
 		return err;
 
@@ -3552,8 +3809,6 @@ int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
 				      irq_name);
 	spin_unlock_irqrestore(&vector_lock, flags);
 
-	cfg = irq_cfg(irq);
-
 	mmr_value = 0;
 	entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
 	BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
@@ -3605,9 +3860,16 @@ int __init io_apic_get_redir_entries (int ioapic)
 	return reg_01.bits.entries;
 }
 
-int __init probe_nr_irqs(void)
+void __init probe_nr_irqs_gsi(void)
 {
-	return NR_IRQS;
+	int idx;
+	int nr = 0;
+
+	for (idx = 0; idx < nr_ioapics; idx++)
+		nr += io_apic_get_redir_entries(idx) + 1;
+
+	if (nr > nr_irqs_gsi)
+		nr_irqs_gsi = nr;
 }
 
 /* --------------------------------------------------------------------------
@@ -3706,19 +3968,31 @@ int __init io_apic_get_version(int ioapic)
 
 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
 {
+	struct irq_desc *desc;
+	struct irq_cfg *cfg;
+	int cpu = boot_cpu_id;
+
 	if (!IO_APIC_IRQ(irq)) {
 		apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
 			ioapic);
 		return -EINVAL;
 	}
 
+	desc = irq_to_desc_alloc_cpu(irq, cpu);
+	if (!desc) {
+		printk(KERN_INFO "can not get irq_desc %d\n", irq);
+		return 0;
+	}
+
 	/*
 	 * IRQs < 16 are already in the irq_2_pin[] map
 	 */
-	if (irq >= 16)
-		add_pin_to_irq(irq, ioapic, pin);
+	if (irq >= NR_IRQS_LEGACY) {
+		cfg = desc->chip_data;
+		add_pin_to_irq_cpu(cfg, cpu, ioapic, pin);
+	}
 
-	setup_IO_APIC_irq(ioapic, pin, irq, triggering, polarity);
+	setup_IO_APIC_irq(ioapic, pin, irq, desc, triggering, polarity);
 
 	return 0;
 }
@@ -3772,9 +4046,10 @@ void __init setup_ioapic_dest(void)
 			 * when you have too many devices, because at that time only boot
 			 * cpu is online.
 			 */
-			cfg = irq_cfg(irq);
+			desc = irq_to_desc(irq);
+			cfg = desc->chip_data;
 			if (!cfg->vector) {
-				setup_IO_APIC_irq(ioapic, pin, irq,
+				setup_IO_APIC_irq(ioapic, pin, irq, desc,
 						  irq_trigger(irq_entry),
 						  irq_polarity(irq_entry));
 				continue;
@@ -3784,7 +4059,6 @@ void __init setup_ioapic_dest(void)
 			/*
 			 * Honour affinities which have been set in early boot
 			 */
-			desc = irq_to_desc(irq);
 			if (desc->status &
 			    (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
 				mask = desc->affinity;
@@ -3793,10 +4067,10 @@ void __init setup_ioapic_dest(void)
 
 #ifdef CONFIG_INTR_REMAP
 			if (intr_remapping_enabled)
-				set_ir_ioapic_affinity_irq(irq, mask);
+				set_ir_ioapic_affinity_irq_desc(desc, mask);
 			else
 #endif
-				set_ioapic_affinity_irq(irq, mask);
+				set_ioapic_affinity_irq_desc(desc, mask);
 		}
 
 	}
@@ -3845,7 +4119,6 @@ void __init ioapic_init_mappings(void)
 	struct resource *ioapic_res;
 	int i;
 
-	irq_2_pin_init();
 	ioapic_res = ioapic_setup_resources();
 	for (i = 0; i < nr_ioapics; i++) {
 		if (smp_found_config) {

arch/x86/kernel/irq.c

@@ -118,6 +118,9 @@ int show_interrupts(struct seq_file *p, void *v)
 	}
 
 	desc = irq_to_desc(i);
+	if (!desc)
+		return 0;
+
 	spin_lock_irqsave(&desc->lock, flags);
 #ifndef CONFIG_SMP
 	any_count = kstat_irqs(i);

arch/x86/kernel/irq_32.c

@@ -242,6 +242,8 @@ void fixup_irqs(cpumask_t map)
 	for_each_irq_desc(irq, desc) {
 		cpumask_t mask;
 
+		if (!desc)
+			continue;
 		if (irq == 2)
 			continue;
 

arch/x86/kernel/irq_64.c

@@ -91,6 +91,8 @@ void fixup_irqs(cpumask_t map)
 		int break_affinity = 0;
 		int set_affinity = 1;
 
+		if (!desc)
+			continue;
 		if (irq == 2)
 			continue;
 

arch/x86/kernel/irqinit_32.c

@@ -68,8 +68,7 @@ void __init init_ISA_irqs (void)
 	/*
 	 * 16 old-style INTA-cycle interrupts:
 	 */
-	for (i = 0; i < 16; i++) {
-		/* first time call this irq_desc */
+	for (i = 0; i < NR_IRQS_LEGACY; i++) {
 		struct irq_desc *desc = irq_to_desc(i);
 
 		desc->status = IRQ_DISABLED;

arch/x86/kernel/irqinit_64.c

@@ -76,8 +76,7 @@ void __init init_ISA_irqs(void)
 	init_bsp_APIC();
 	init_8259A(0);
 
-	for (i = 0; i < 16; i++) {
-		/* first time call this irq_desc */
+	for (i = 0; i < NR_IRQS_LEGACY; i++) {
 		struct irq_desc *desc = irq_to_desc(i);
 
 		desc->status = IRQ_DISABLED;

arch/x86/kernel/setup.c

@@ -953,7 +953,7 @@ void __init setup_arch(char **cmdline_p)
 	ioapic_init_mappings();
 
 	/* need to wait for io_apic is mapped */
-	nr_irqs = probe_nr_irqs();
+	probe_nr_irqs_gsi();
 
 	kvm_guest_init();
 

drivers/char/random.c

@@ -558,23 +558,9 @@ struct timer_rand_state {
 	unsigned dont_count_entropy:1;
 };
 
-static struct timer_rand_state *irq_timer_state[NR_IRQS];
-
-static struct timer_rand_state *get_timer_rand_state(unsigned int irq)
-{
-	if (irq >= nr_irqs)
-		return NULL;
-
-	return irq_timer_state[irq];
-}
-
-static void set_timer_rand_state(unsigned int irq, struct timer_rand_state *state)
-{
-	if (irq >= nr_irqs)
-		return;
-
-	irq_timer_state[irq] = state;
-}
+#ifndef CONFIG_SPARSE_IRQ
+struct timer_rand_state *irq_timer_state[NR_IRQS];
+#endif
 
 static struct timer_rand_state input_timer_state;
 
@@ -933,8 +919,10 @@ void rand_initialize_irq(int irq)
 {
 	struct timer_rand_state *state;
 
+#ifndef CONFIG_SPARSE_IRQ
 	if (irq >= nr_irqs)
 		return;
+#endif
 
 	state = get_timer_rand_state(irq);
 

drivers/pci/intr_remapping.c

@@ -5,6 +5,7 @@
 #include <linux/pci.h>
 #include <linux/irq.h>
 #include <asm/io_apic.h>
+#include <asm/smp.h>
 #include <linux/intel-iommu.h>
 #include "intr_remapping.h"
 
@@ -19,17 +20,75 @@ struct irq_2_iommu {
 	u8  irte_mask;
 };
 
-static struct irq_2_iommu irq_2_iommuX[NR_IRQS];
+#ifdef CONFIG_SPARSE_IRQ
+static struct irq_2_iommu *get_one_free_irq_2_iommu(int cpu)
+{
+	struct irq_2_iommu *iommu;
+	int node;
+
+	node = cpu_to_node(cpu);
+
+	iommu = kzalloc_node(sizeof(*iommu), GFP_ATOMIC, node);
+	printk(KERN_DEBUG "alloc irq_2_iommu on cpu %d node %d\n", cpu, node);
+
+	return iommu;
+}
 
 static struct irq_2_iommu *irq_2_iommu(unsigned int irq)
 {
-	return (irq < nr_irqs) ? irq_2_iommuX + irq : NULL;
+	struct irq_desc *desc;
+
+	desc = irq_to_desc(irq);
+
+	if (WARN_ON_ONCE(!desc))
+		return NULL;
+
+	return desc->irq_2_iommu;
+}
+
+static struct irq_2_iommu *irq_2_iommu_alloc_cpu(unsigned int irq, int cpu)
+{
+	struct irq_desc *desc;
+	struct irq_2_iommu *irq_iommu;
+
+	/*
+	 * alloc irq desc if not allocated already.
+	 */
+	desc = irq_to_desc_alloc_cpu(irq, cpu);
+	if (!desc) {
+		printk(KERN_INFO "can not get irq_desc for %d\n", irq);
+		return NULL;
+	}
+
+	irq_iommu = desc->irq_2_iommu;
+
+	if (!irq_iommu)
+		desc->irq_2_iommu = get_one_free_irq_2_iommu(cpu);
+
+	return desc->irq_2_iommu;
 }
 
+static struct irq_2_iommu *irq_2_iommu_alloc(unsigned int irq)
+{
+	return irq_2_iommu_alloc_cpu(irq, boot_cpu_id);
+}
+
+#else /* !CONFIG_SPARSE_IRQ */
+
+static struct irq_2_iommu irq_2_iommuX[NR_IRQS];
+
+static struct irq_2_iommu *irq_2_iommu(unsigned int irq)
+{
+	if (irq < nr_irqs)
+		return &irq_2_iommuX[irq];
+
+	return NULL;
+}
 static struct irq_2_iommu *irq_2_iommu_alloc(unsigned int irq)
 {
 	return irq_2_iommu(irq);
 }
+#endif
 
 static DEFINE_SPINLOCK(irq_2_ir_lock);
 
@@ -86,9 +145,11 @@ int alloc_irte(struct intel_iommu *iommu, int irq, u16 count)
 	if (!count)
 		return -1;
 
+#ifndef CONFIG_SPARSE_IRQ
 	/* protect irq_2_iommu_alloc later */
 	if (irq >= nr_irqs)
 		return -1;
+#endif
 
 	/*
 	 * start the IRTE search from index 0.
@@ -130,6 +191,12 @@ int alloc_irte(struct intel_iommu *iommu, int irq, u16 count)
 		table->base[i].present = 1;
 
 	irq_iommu = irq_2_iommu_alloc(irq);
+	if (!irq_iommu) {
+		spin_unlock(&irq_2_ir_lock);
+		printk(KERN_ERR "can't allocate irq_2_iommu\n");
+		return -1;
+	}
+
 	irq_iommu->iommu = iommu;
 	irq_iommu->irte_index =  index;
 	irq_iommu->sub_handle = 0;
@@ -177,6 +244,12 @@ int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index, u16 subhandle)
 
 	irq_iommu = irq_2_iommu_alloc(irq);
 
+	if (!irq_iommu) {
+		spin_unlock(&irq_2_ir_lock);
+		printk(KERN_ERR "can't allocate irq_2_iommu\n");
+		return -1;
+	}
+
 	irq_iommu->iommu = iommu;
 	irq_iommu->irte_index = index;
 	irq_iommu->sub_handle = subhandle;

drivers/pci/msi.c

@@ -103,11 +103,11 @@ static void msix_set_enable(struct pci_dev *dev, int enable)
 	}
 }
 
-static void msix_flush_writes(unsigned int irq)
+static void msix_flush_writes(struct irq_desc *desc)
 {
 	struct msi_desc *entry;
 
-	entry = get_irq_msi(irq);
+	entry = get_irq_desc_msi(desc);
 	BUG_ON(!entry || !entry->dev);
 	switch (entry->msi_attrib.type) {
 	case PCI_CAP_ID_MSI:
@@ -135,11 +135,11 @@ static void msix_flush_writes(unsigned int irq)
  * Returns 1 if it succeeded in masking the interrupt and 0 if the device
  * doesn't support MSI masking.
  */
-static int msi_set_mask_bits(unsigned int irq, u32 mask, u32 flag)
+static int msi_set_mask_bits(struct irq_desc *desc, u32 mask, u32 flag)
 {
 	struct msi_desc *entry;
 
-	entry = get_irq_msi(irq);
+	entry = get_irq_desc_msi(desc);
 	BUG_ON(!entry || !entry->dev);
 	switch (entry->msi_attrib.type) {
 	case PCI_CAP_ID_MSI:
@@ -172,9 +172,9 @@ static int msi_set_mask_bits(unsigned int irq, u32 mask, u32 flag)
 	return 1;
 }
 
-void read_msi_msg(unsigned int irq, struct msi_msg *msg)
+void read_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
 {
-	struct msi_desc *entry = get_irq_msi(irq);
+	struct msi_desc *entry = get_irq_desc_msi(desc);
 	switch(entry->msi_attrib.type) {
 	case PCI_CAP_ID_MSI:
 	{
@@ -211,9 +211,16 @@ void read_msi_msg(unsigned int irq, struct msi_msg *msg)
 	}
 }
 
-void write_msi_msg(unsigned int irq, struct msi_msg *msg)
+void read_msi_msg(unsigned int irq, struct msi_msg *msg)
 {
-	struct msi_desc *entry = get_irq_msi(irq);
+	struct irq_desc *desc = irq_to_desc(irq);
+
+	read_msi_msg_desc(desc, msg);
+}
+
+void write_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
+{
+	struct msi_desc *entry = get_irq_desc_msi(desc);
 	switch (entry->msi_attrib.type) {
 	case PCI_CAP_ID_MSI:
 	{
@@ -252,21 +259,31 @@ void write_msi_msg(unsigned int irq, struct msi_msg *msg)
 	entry->msg = *msg;
 }
 
+void write_msi_msg(unsigned int irq, struct msi_msg *msg)
+{
+	struct irq_desc *desc = irq_to_desc(irq);
+
+	write_msi_msg_desc(desc, msg);
+}
+
 void mask_msi_irq(unsigned int irq)
 {
-	msi_set_mask_bits(irq, 1, 1);
-	msix_flush_writes(irq);
+	struct irq_desc *desc = irq_to_desc(irq);
+
+	msi_set_mask_bits(desc, 1, 1);
+	msix_flush_writes(desc);
 }
 
 void unmask_msi_irq(unsigned int irq)
 {
-	msi_set_mask_bits(irq, 1, 0);
-	msix_flush_writes(irq);
+	struct irq_desc *desc = irq_to_desc(irq);
+
+	msi_set_mask_bits(desc, 1, 0);
+	msix_flush_writes(desc);
 }
 
 static int msi_free_irqs(struct pci_dev* dev);
 
-
 static struct msi_desc* alloc_msi_entry(void)
 {
 	struct msi_desc *entry;
@@ -303,9 +320,11 @@ static void __pci_restore_msi_state(struct pci_dev *dev)
 	pci_intx_for_msi(dev, 0);
 	msi_set_enable(dev, 0);
 	write_msi_msg(dev->irq, &entry->msg);
-	if (entry->msi_attrib.maskbit)
-		msi_set_mask_bits(dev->irq, entry->msi_attrib.maskbits_mask,
+	if (entry->msi_attrib.maskbit) {
+		struct irq_desc *desc = irq_to_desc(dev->irq);
+		msi_set_mask_bits(desc, entry->msi_attrib.maskbits_mask,
 				  entry->msi_attrib.masked);
+	}
 
 	pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
 	control &= ~PCI_MSI_FLAGS_QSIZE;
@@ -327,8 +346,9 @@ static void __pci_restore_msix_state(struct pci_dev *dev)
 	msix_set_enable(dev, 0);
 
 	list_for_each_entry(entry, &dev->msi_list, list) {
+		struct irq_desc *desc = irq_to_desc(entry->irq);
 		write_msi_msg(entry->irq, &entry->msg);
-		msi_set_mask_bits(entry->irq, 1, entry->msi_attrib.masked);
+		msi_set_mask_bits(desc, 1, entry->msi_attrib.masked);
 	}
 
 	BUG_ON(list_empty(&dev->msi_list));
@@ -596,7 +616,8 @@ void pci_msi_shutdown(struct pci_dev* dev)
 	/* Return the the pci reset with msi irqs unmasked */
 	if (entry->msi_attrib.maskbit) {
 		u32 mask = entry->msi_attrib.maskbits_mask;
-		msi_set_mask_bits(dev->irq, mask, ~mask);
+		struct irq_desc *desc = irq_to_desc(dev->irq);
+		msi_set_mask_bits(desc, mask, ~mask);
 	}
 	if (!entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI)
 		return;

drivers/xen/events.c

@@ -141,8 +141,12 @@ static void init_evtchn_cpu_bindings(void)
 	int i;
 
 	/* By default all event channels notify CPU#0. */
-	for_each_irq_desc(i, desc)
+	for_each_irq_desc(i, desc) {
+		if (!desc)
+			continue;
+
 		desc->affinity = cpumask_of_cpu(0);
+	}
 #endif
 
 	memset(cpu_evtchn, 0, sizeof(cpu_evtchn));
@@ -229,15 +233,20 @@ static void unmask_evtchn(int port)
 static int find_unbound_irq(void)
 {
 	int irq;
+	struct irq_desc *desc;
 
 	/* Only allocate from dynirq range */
-	for_each_irq_nr(irq)
+	for (irq = 0; irq < nr_irqs; irq++)
 		if (irq_bindcount[irq] == 0)
 			break;
 
 	if (irq == nr_irqs)
 		panic("No available IRQ to bind to: increase nr_irqs!\n");
 
+	desc = irq_to_desc_alloc_cpu(irq, 0);
+	if (WARN_ON(desc == NULL))
+		return -1;
+
 	return irq;
 }
 
@@ -792,7 +801,7 @@ void xen_irq_resume(void)
 		mask_evtchn(evtchn);
 
 	/* No IRQ <-> event-channel mappings. */
-	for_each_irq_nr(irq)
+	for (irq = 0; irq < nr_irqs; irq++)
 		irq_info[irq].evtchn = 0; /* zap event-channel binding */
 
 	for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
@@ -824,7 +833,7 @@ void __init xen_init_IRQ(void)
 		mask_evtchn(i);
 
 	/* Dynamic IRQ space is currently unbound. Zero the refcnts. */
-	for_each_irq_nr(i)
+	for (i = 0; i < nr_irqs; i++)
 		irq_bindcount[i] = 0;
 
 	irq_ctx_init(smp_processor_id());

fs/proc/stat.c

@@ -44,10 +44,13 @@ static int show_stat(struct seq_file *p, void *v)
 		softirq = cputime64_add(softirq, kstat_cpu(i).cpustat.softirq);
 		steal = cputime64_add(steal, kstat_cpu(i).cpustat.steal);
 		guest = cputime64_add(guest, kstat_cpu(i).cpustat.guest);
-
-		for_each_irq_nr(j)
+		for_each_irq_nr(j) {
+#ifdef CONFIG_SPARSE_IRQ
+			if (!irq_to_desc(j))
+				continue;
+#endif
 			sum += kstat_irqs_cpu(j, i);
-
+		}
 		sum += arch_irq_stat_cpu(i);
 	}
 	sum += arch_irq_stat();
@@ -92,7 +95,12 @@ static int show_stat(struct seq_file *p, void *v)
 	/* sum again ? it could be updated? */
 	for_each_irq_nr(j) {
 		per_irq_sum = 0;
-
+#ifdef CONFIG_SPARSE_IRQ
+		if (!irq_to_desc(j)) {
+			seq_printf(p, " %u", per_irq_sum);
+			continue;
+		}
+#endif
 		for_each_possible_cpu(i)
 			per_irq_sum += kstat_irqs_cpu(j, i);
 

include/linux/Kbuild

@@ -313,6 +313,7 @@ unifdef-y += ptrace.h
 unifdef-y += qnx4_fs.h
 unifdef-y += quota.h
 unifdef-y += random.h
+unifdef-y += irqnr.h
 unifdef-y += reboot.h
 unifdef-y += reiserfs_fs.h
 unifdef-y += reiserfs_xattr.h

include/linux/interrupt.h

@@ -14,6 +14,8 @@
 #include <linux/irqflags.h>
 #include <linux/smp.h>
 #include <linux/percpu.h>
+#include <linux/irqnr.h>
+
 #include <asm/atomic.h>
 #include <asm/ptrace.h>
 #include <asm/system.h>

include/linux/irq.h

@@ -129,9 +129,14 @@ struct irq_chip {
 	const char	*typename;
 };
 
+struct timer_rand_state;
+struct irq_2_iommu;
 /**
  * struct irq_desc - interrupt descriptor
  * @irq:		interrupt number for this descriptor
+ * @timer_rand_state:	pointer to timer rand state struct
+ * @kstat_irqs:		irq stats per cpu
+ * @irq_2_iommu:	iommu with this irq
  * @handle_irq:		highlevel irq-events handler [if NULL, __do_IRQ()]
  * @chip:		low level interrupt hardware access
  * @msi_desc:		MSI descriptor
@@ -143,8 +148,8 @@ struct irq_chip {
  * @depth:		disable-depth, for nested irq_disable() calls
  * @wake_depth:		enable depth, for multiple set_irq_wake() callers
  * @irq_count:		stats field to detect stalled irqs
- * @irqs_unhandled:	stats field for spurious unhandled interrupts
  * @last_unhandled:	aging timer for unhandled count
+ * @irqs_unhandled:	stats field for spurious unhandled interrupts
  * @lock:		locking for SMP
  * @affinity:		IRQ affinity on SMP
  * @cpu:		cpu index useful for balancing
@@ -154,6 +159,13 @@ struct irq_chip {
  */
 struct irq_desc {
 	unsigned int		irq;
+#ifdef CONFIG_SPARSE_IRQ
+	struct timer_rand_state *timer_rand_state;
+	unsigned int            *kstat_irqs;
+# ifdef CONFIG_INTR_REMAP
+	struct irq_2_iommu      *irq_2_iommu;
+# endif
+#endif
 	irq_flow_handler_t	handle_irq;
 	struct irq_chip		*chip;
 	struct msi_desc		*msi_desc;
@@ -165,8 +177,8 @@ struct irq_desc {
 	unsigned int		depth;		/* nested irq disables */
 	unsigned int		wake_depth;	/* nested wake enables */
 	unsigned int		irq_count;	/* For detecting broken IRQs */
-	unsigned int		irqs_unhandled;
 	unsigned long		last_unhandled;	/* Aging timer for unhandled count */
+	unsigned int		irqs_unhandled;
 	spinlock_t		lock;
 #ifdef CONFIG_SMP
 	cpumask_t		affinity;
@@ -181,12 +193,51 @@ struct irq_desc {
 	const char		*name;
 } ____cacheline_internodealigned_in_smp;
 
+extern void early_irq_init(void);
+extern void arch_early_irq_init(void);
+extern void arch_init_chip_data(struct irq_desc *desc, int cpu);
+extern void arch_init_copy_chip_data(struct irq_desc *old_desc,
+					struct irq_desc *desc, int cpu);
+extern void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc);
 
+#ifndef CONFIG_SPARSE_IRQ
 extern struct irq_desc irq_desc[NR_IRQS];
 
 static inline struct irq_desc *irq_to_desc(unsigned int irq)
 {
-	return (irq < nr_irqs) ? irq_desc + irq : NULL;
+	return (irq < NR_IRQS) ? irq_desc + irq : NULL;
+}
+static inline struct irq_desc *irq_to_desc_alloc_cpu(unsigned int irq, int cpu)
+{
+	return irq_to_desc(irq);
+}
+
+#else
+
+extern struct irq_desc *irq_to_desc(unsigned int irq);
+extern struct irq_desc *irq_to_desc_alloc_cpu(unsigned int irq, int cpu);
+extern struct irq_desc *move_irq_desc(struct irq_desc *old_desc, int cpu);
+
+# define for_each_irq_desc(irq, desc)		\
+	for (irq = 0, desc = irq_to_desc(irq); irq < nr_irqs; irq++, desc = irq_to_desc(irq))
+# define for_each_irq_desc_reverse(irq, desc)                          \
+	for (irq = nr_irqs - 1, desc = irq_to_desc(irq); irq >= 0; irq--, desc = irq_to_desc(irq))
+
+#define kstat_irqs_this_cpu(DESC) \
+	((DESC)->kstat_irqs[smp_processor_id()])
+#define kstat_incr_irqs_this_cpu(irqno, DESC) \
+	((DESC)->kstat_irqs[smp_processor_id()]++)
+
+#endif
+
+static inline struct irq_desc *
+irq_remap_to_desc(unsigned int irq, struct irq_desc *desc)
+{
+#ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
+	return irq_to_desc(irq);
+#else
+	return desc;
+#endif
 }
 
 /*
@@ -380,6 +431,11 @@ extern int set_irq_msi(unsigned int irq, struct msi_desc *entry);
 #define get_irq_data(irq)	(irq_to_desc(irq)->handler_data)
 #define get_irq_msi(irq)	(irq_to_desc(irq)->msi_desc)
 
+#define get_irq_desc_chip(desc)		((desc)->chip)
+#define get_irq_desc_chip_data(desc)	((desc)->chip_data)
+#define get_irq_desc_data(desc)		((desc)->handler_data)
+#define get_irq_desc_msi(desc)		((desc)->msi_desc)
+
 #endif /* CONFIG_GENERIC_HARDIRQS */
 
 #endif /* !CONFIG_S390 */

include/linux/irqnr.h

 #ifndef _LINUX_IRQNR_H
 #define _LINUX_IRQNR_H
 
+/*
+ * Generic irq_desc iterators:
+ */
+#ifdef __KERNEL__
+
 #ifndef CONFIG_GENERIC_HARDIRQS
 #include <asm/irq.h>
 # define nr_irqs		NR_IRQS
 
 # define for_each_irq_desc(irq, desc)		\
 	for (irq = 0; irq < nr_irqs; irq++)
+
+# define for_each_irq_desc_reverse(irq, desc)                          \
+	for (irq = nr_irqs - 1; irq >= 0; irq--)
 #else
+
 extern int nr_irqs;
 
+#ifndef CONFIG_SPARSE_IRQ
+
+struct irq_desc;
 # define for_each_irq_desc(irq, desc)		\
 	for (irq = 0, desc = irq_desc; irq < nr_irqs; irq++, desc++)
-
-# define for_each_irq_desc_reverse(irq, desc)				\
-	for (irq = nr_irqs - 1, desc = irq_desc + (nr_irqs - 1);	\
-	     irq >= 0; irq--, desc--)
+# define for_each_irq_desc_reverse(irq, desc)                          \
+	for (irq = nr_irqs - 1, desc = irq_desc + (nr_irqs - 1);        \
+	    irq >= 0; irq--, desc--)
+#endif
 #endif
 
-#define for_each_irq_nr(irq)			\
-	for (irq = 0; irq < nr_irqs; irq++)
+#define for_each_irq_nr(irq)                   \
+       for (irq = 0; irq < nr_irqs; irq++)
+
+#endif /* __KERNEL__ */
 
 #endif

include/linux/kernel_stat.h

@@ -28,7 +28,9 @@ struct cpu_usage_stat {
 
 struct kernel_stat {
 	struct cpu_usage_stat	cpustat;
-	unsigned int irqs[NR_IRQS];
+#ifndef CONFIG_SPARSE_IRQ
+       unsigned int irqs[NR_IRQS];
+#endif
 };
 
 DECLARE_PER_CPU(struct kernel_stat, kstat);
@@ -39,6 +41,10 @@ DECLARE_PER_CPU(struct kernel_stat, kstat);
 
 extern unsigned long long nr_context_switches(void);
 
+#ifndef CONFIG_SPARSE_IRQ
+#define kstat_irqs_this_cpu(irq) \
+	(kstat_this_cpu.irqs[irq])
+
 struct irq_desc;
 
 static inline void kstat_incr_irqs_this_cpu(unsigned int irq,
@@ -46,11 +52,17 @@ static inline void kstat_incr_irqs_this_cpu(unsigned int irq,
 {
 	kstat_this_cpu.irqs[irq]++;
 }
+#endif
+
 
+#ifndef CONFIG_SPARSE_IRQ
 static inline unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
 {
        return kstat_cpu(cpu).irqs[irq];
 }
+#else
+extern unsigned int kstat_irqs_cpu(unsigned int irq, int cpu);
+#endif
 
 /*
  * Number of interrupts per specific IRQ source, since bootup

include/linux/lockdep.h

@@ -390,7 +390,7 @@ do {								\
 
 #endif /* CONFIG_LOCK_STAT */
 
-#if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_GENERIC_HARDIRQS)
+#ifdef CONFIG_GENERIC_HARDIRQS
 extern void early_init_irq_lock_class(void);
 #else
 static inline void early_init_irq_lock_class(void)

include/linux/msi.h

@@ -10,8 +10,11 @@ struct msi_msg {
 };
 
 /* Helper functions */
+struct irq_desc;
 extern void mask_msi_irq(unsigned int irq);
 extern void unmask_msi_irq(unsigned int irq);
+extern void read_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg);
+extern void write_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg);
 extern void read_msi_msg(unsigned int irq, struct msi_msg *msg);
 extern void write_msi_msg(unsigned int irq, struct msi_msg *msg);
 

include/linux/random.h

@@ -8,6 +8,7 @@
 #define _LINUX_RANDOM_H
 
 #include <linux/ioctl.h>
+#include <linux/irqnr.h>
 
 /* ioctl()'s for the random number generator */
 
@@ -44,6 +45,56 @@ struct rand_pool_info {
 
 extern void rand_initialize_irq(int irq);
 
+struct timer_rand_state;
+#ifndef CONFIG_SPARSE_IRQ
+
+extern struct timer_rand_state *irq_timer_state[];
+
+static inline struct timer_rand_state *get_timer_rand_state(unsigned int irq)
+{
+	if (irq >= nr_irqs)
+		return NULL;
+
+	return irq_timer_state[irq];
+}
+
+static inline void set_timer_rand_state(unsigned int irq, struct timer_rand_state *state)
+{
+	if (irq >= nr_irqs)
+		return;
+
+	irq_timer_state[irq] = state;
+}
+
+#else
+
+#include <linux/irq.h>
+static inline struct timer_rand_state *get_timer_rand_state(unsigned int irq)
+{
+	struct irq_desc *desc;
+
+	desc = irq_to_desc(irq);
+
+	if (!desc)
+		return NULL;
+
+	return desc->timer_rand_state;
+}
+
+static inline void set_timer_rand_state(unsigned int irq, struct timer_rand_state *state)
+{
+	struct irq_desc *desc;
+
+	desc = irq_to_desc(irq);
+
+	if (!desc)
+		return;
+
+	desc->timer_rand_state = state;
+}
+#endif
+
+
 extern void add_input_randomness(unsigned int type, unsigned int code,
 				 unsigned int value);
 extern void add_interrupt_randomness(int irq);

init/main.c

@@ -540,6 +540,15 @@ void __init __weak thread_info_cache_init(void)
 {
 }
 
+void __init __weak arch_early_irq_init(void)
+{
+}
+
+void __init __weak early_irq_init(void)
+{
+	arch_early_irq_init();
+}
+
 asmlinkage void __init start_kernel(void)
 {
 	char * command_line;
@@ -604,6 +613,8 @@ asmlinkage void __init start_kernel(void)
 	sort_main_extable();
 	trap_init();
 	rcu_init();
+	/* init some links before init_ISA_irqs() */
+	early_irq_init();
 	init_IRQ();
 	pidhash_init();
 	init_timers();

kernel/irq/Makefile

@@ -3,3 +3,4 @@ obj-y := handle.o manage.o spurious.o resend.o chip.o devres.o
 obj-$(CONFIG_GENERIC_IRQ_PROBE) += autoprobe.o
 obj-$(CONFIG_PROC_FS) += proc.o
 obj-$(CONFIG_GENERIC_PENDING_IRQ) += migration.o
+obj-$(CONFIG_NUMA_MIGRATE_IRQ_DESC) += numa_migrate.o

kernel/irq/autoprobe.c

@@ -40,6 +40,9 @@ unsigned long probe_irq_on(void)
 	 * flush such a longstanding irq before considering it as spurious.
 	 */
 	for_each_irq_desc_reverse(i, desc) {
+		if (!desc)
+			continue;
+
 		spin_lock_irq(&desc->lock);
 		if (!desc->action && !(desc->status & IRQ_NOPROBE)) {
 			/*
@@ -68,6 +71,9 @@ unsigned long probe_irq_on(void)
 	 * happened in the previous stage, it may have masked itself)
 	 */
 	for_each_irq_desc_reverse(i, desc) {
+		if (!desc)
+			continue;
+
 		spin_lock_irq(&desc->lock);
 		if (!desc->action && !(desc->status & IRQ_NOPROBE)) {
 			desc->status |= IRQ_AUTODETECT | IRQ_WAITING;
@@ -86,6 +92,9 @@ unsigned long probe_irq_on(void)
 	 * Now filter out any obviously spurious interrupts
 	 */
 	for_each_irq_desc(i, desc) {
+		if (!desc)
+			continue;
+
 		spin_lock_irq(&desc->lock);
 		status = desc->status;
 
@@ -124,6 +133,9 @@ unsigned int probe_irq_mask(unsigned long val)
 	int i;
 
 	for_each_irq_desc(i, desc) {
+		if (!desc)
+			continue;
+
 		spin_lock_irq(&desc->lock);
 		status = desc->status;
 
@@ -166,6 +178,9 @@ int probe_irq_off(unsigned long val)
 	unsigned int status;
 
 	for_each_irq_desc(i, desc) {
+		if (!desc)
+			continue;
+
 		spin_lock_irq(&desc->lock);
 		status = desc->status;
 

kernel/irq/chip.c

@@ -24,9 +24,10 @@
  */
 void dynamic_irq_init(unsigned int irq)
 {
-	struct irq_desc *desc = irq_to_desc(irq);
+	struct irq_desc *desc;
 	unsigned long flags;
 
+	desc = irq_to_desc(irq);
 	if (!desc) {
 		WARN(1, KERN_ERR "Trying to initialize invalid IRQ%d\n", irq);
 		return;
@@ -124,6 +125,7 @@ int set_irq_type(unsigned int irq, unsigned int type)
 		return -ENODEV;
 	}
 
+	type &= IRQ_TYPE_SENSE_MASK;
 	if (type == IRQ_TYPE_NONE)
 		return 0;
 
@@ -352,6 +354,7 @@ handle_level_irq(unsigned int irq, struct irq_desc *desc)
 
 	spin_lock(&desc->lock);
 	mask_ack_irq(desc, irq);
+	desc = irq_remap_to_desc(irq, desc);
 
 	if (unlikely(desc->status & IRQ_INPROGRESS))
 		goto out_unlock;
@@ -429,6 +432,7 @@ handle_fasteoi_irq(unsigned int irq, struct irq_desc *desc)
 	desc->status &= ~IRQ_INPROGRESS;
 out:
 	desc->chip->eoi(irq);
+	desc = irq_remap_to_desc(irq, desc);
 
 	spin_unlock(&desc->lock);
 }
@@ -465,12 +469,14 @@ handle_edge_irq(unsigned int irq, struct irq_desc *desc)
 		    !desc->action)) {
 		desc->status |= (IRQ_PENDING | IRQ_MASKED);
 		mask_ack_irq(desc, irq);
+		desc = irq_remap_to_desc(irq, desc);
 		goto out_unlock;
 	}
 	kstat_incr_irqs_this_cpu(irq, desc);
 
 	/* Start handling the irq */
 	desc->chip->ack(irq);
+	desc = irq_remap_to_desc(irq, desc);
 
 	/* Mark the IRQ currently in progress.*/
 	desc->status |= IRQ_INPROGRESS;
@@ -531,8 +537,10 @@ handle_percpu_irq(unsigned int irq, struct irq_desc *desc)
 	if (!noirqdebug)
 		note_interrupt(irq, desc, action_ret);
 
-	if (desc->chip->eoi)
+	if (desc->chip->eoi) {
 		desc->chip->eoi(irq);
+		desc = irq_remap_to_desc(irq, desc);
+	}
 }
 
 void
@@ -567,8 +575,10 @@ __set_irq_handler(unsigned int irq, irq_flow_handler_t handle, int is_chained,
 
 	/* Uninstall? */
 	if (handle == handle_bad_irq) {
-		if (desc->chip != &no_irq_chip)
+		if (desc->chip != &no_irq_chip) {
 			mask_ack_irq(desc, irq);
+			desc = irq_remap_to_desc(irq, desc);
+		}
 		desc->status |= IRQ_DISABLED;
 		desc->depth = 1;
 	}

kernel/irq/handle.c

@@ -15,9 +15,16 @@
 #include <linux/random.h>
 #include <linux/interrupt.h>
 #include <linux/kernel_stat.h>
+#include <linux/rculist.h>
+#include <linux/hash.h>
 
 #include "internals.h"
 
+/*
+ * lockdep: we want to handle all irq_desc locks as a single lock-class:
+ */
+struct lock_class_key irq_desc_lock_class;
+
 /**
  * handle_bad_irq - handle spurious and unhandled irqs
  * @irq:       the interrupt number
@@ -49,6 +56,155 @@ void handle_bad_irq(unsigned int irq, struct irq_desc *desc)
 int nr_irqs = NR_IRQS;
 EXPORT_SYMBOL_GPL(nr_irqs);
 
+void __init __attribute__((weak)) arch_early_irq_init(void)
+{
+}
+
+#ifdef CONFIG_SPARSE_IRQ
+static struct irq_desc irq_desc_init = {
+	.irq	    = -1,
+	.status	    = IRQ_DISABLED,
+	.chip	    = &no_irq_chip,
+	.handle_irq = handle_bad_irq,
+	.depth      = 1,
+	.lock       = __SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
+#ifdef CONFIG_SMP
+	.affinity   = CPU_MASK_ALL
+#endif
+};
+
+void init_kstat_irqs(struct irq_desc *desc, int cpu, int nr)
+{
+	unsigned long bytes;
+	char *ptr;
+	int node;
+
+	/* Compute how many bytes we need per irq and allocate them */
+	bytes = nr * sizeof(unsigned int);
+
+	node = cpu_to_node(cpu);
+	ptr = kzalloc_node(bytes, GFP_ATOMIC, node);
+	printk(KERN_DEBUG "  alloc kstat_irqs on cpu %d node %d\n", cpu, node);
+
+	if (ptr)
+		desc->kstat_irqs = (unsigned int *)ptr;
+}
+
+void __attribute__((weak)) arch_init_chip_data(struct irq_desc *desc, int cpu)
+{
+}
+
+static void init_one_irq_desc(int irq, struct irq_desc *desc, int cpu)
+{
+	memcpy(desc, &irq_desc_init, sizeof(struct irq_desc));
+	desc->irq = irq;
+#ifdef CONFIG_SMP
+	desc->cpu = cpu;
+#endif
+	lockdep_set_class(&desc->lock, &irq_desc_lock_class);
+	init_kstat_irqs(desc, cpu, nr_cpu_ids);
+	if (!desc->kstat_irqs) {
+		printk(KERN_ERR "can not alloc kstat_irqs\n");
+		BUG_ON(1);
+	}
+	arch_init_chip_data(desc, cpu);
+}
+
+/*
+ * Protect the sparse_irqs:
+ */
+DEFINE_SPINLOCK(sparse_irq_lock);
+
+struct irq_desc *irq_desc_ptrs[NR_IRQS] __read_mostly;
+
+static struct irq_desc irq_desc_legacy[NR_IRQS_LEGACY] __cacheline_aligned_in_smp = {
+	[0 ... NR_IRQS_LEGACY-1] = {
+		.irq	    = -1,
+		.status	    = IRQ_DISABLED,
+		.chip	    = &no_irq_chip,
+		.handle_irq = handle_bad_irq,
+		.depth	    = 1,
+		.lock	    = __SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
+#ifdef CONFIG_SMP
+		.affinity   = CPU_MASK_ALL
+#endif
+	}
+};
+
+/* FIXME: use bootmem alloc ...*/
+static unsigned int kstat_irqs_legacy[NR_IRQS_LEGACY][NR_CPUS];
+
+void __init early_irq_init(void)
+{
+	struct irq_desc *desc;
+	int legacy_count;
+	int i;
+
+	desc = irq_desc_legacy;
+	legacy_count = ARRAY_SIZE(irq_desc_legacy);
+
+	for (i = 0; i < legacy_count; i++) {
+		desc[i].irq = i;
+		desc[i].kstat_irqs = kstat_irqs_legacy[i];
+
+		irq_desc_ptrs[i] = desc + i;
+	}
+
+	for (i = legacy_count; i < NR_IRQS; i++)
+		irq_desc_ptrs[i] = NULL;
+
+	arch_early_irq_init();
+}
+
+struct irq_desc *irq_to_desc(unsigned int irq)
+{
+	return (irq < NR_IRQS) ? irq_desc_ptrs[irq] : NULL;
+}
+
+struct irq_desc *irq_to_desc_alloc_cpu(unsigned int irq, int cpu)
+{
+	struct irq_desc *desc;
+	unsigned long flags;
+	int node;
+
+	if (irq >= NR_IRQS) {
+		printk(KERN_WARNING "irq >= NR_IRQS in irq_to_desc_alloc: %d %d\n",
+				irq, NR_IRQS);
+		WARN_ON(1);
+		return NULL;
+	}
+
+	desc = irq_desc_ptrs[irq];
+	if (desc)
+		return desc;
+
+	spin_lock_irqsave(&sparse_irq_lock, flags);
+
+	/* We have to check it to avoid races with another CPU */
+	desc = irq_desc_ptrs[irq];
+	if (desc)
+		goto out_unlock;
+
+	node = cpu_to_node(cpu);
+	desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node);
+	printk(KERN_DEBUG "  alloc irq_desc for %d on cpu %d node %d\n",
+		 irq, cpu, node);
+	if (!desc) {
+		printk(KERN_ERR "can not alloc irq_desc\n");
+		BUG_ON(1);
+	}
+	init_one_irq_desc(irq, desc, cpu);
+
+	irq_desc_ptrs[irq] = desc;
+
+out_unlock:
+	spin_unlock_irqrestore(&sparse_irq_lock, flags);
+
+	return desc;
+}
+
+#else
+
 struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
 	[0 ... NR_IRQS-1] = {
 		.status = IRQ_DISABLED,
@@ -62,6 +218,8 @@ struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
 	}
 };
 
+#endif
+
 /*
  * What should we do if we get a hw irq event on an illegal vector?
  * Each architecture has to answer this themself.
@@ -179,8 +337,11 @@ unsigned int __do_IRQ(unsigned int irq)
 		/*
 		 * No locking required for CPU-local interrupts:
 		 */
-		if (desc->chip->ack)
+		if (desc->chip->ack) {
 			desc->chip->ack(irq);
+			/* get new one */
+			desc = irq_remap_to_desc(irq, desc);
+		}
 		if (likely(!(desc->status & IRQ_DISABLED))) {
 			action_ret = handle_IRQ_event(irq, desc->action);
 			if (!noirqdebug)
@@ -191,8 +352,10 @@ unsigned int __do_IRQ(unsigned int irq)
 	}
 
 	spin_lock(&desc->lock);
-	if (desc->chip->ack)
+	if (desc->chip->ack) {
 		desc->chip->ack(irq);
+		desc = irq_remap_to_desc(irq, desc);
+	}
 	/*
 	 * REPLAY is when Linux resends an IRQ that was dropped earlier
 	 * WAITING is used by probe to mark irqs that are being tested
@@ -259,19 +422,25 @@ unsigned int __do_IRQ(unsigned int irq)
 }
 #endif
 
-
-#ifdef CONFIG_TRACE_IRQFLAGS
-/*
- * lockdep: we want to handle all irq_desc locks as a single lock-class:
- */
-static struct lock_class_key irq_desc_lock_class;
-
 void early_init_irq_lock_class(void)
 {
 	struct irq_desc *desc;
 	int i;
 
-	for_each_irq_desc(i, desc)
+	for_each_irq_desc(i, desc) {
+		if (!desc)
+			continue;
+
 		lockdep_set_class(&desc->lock, &irq_desc_lock_class);
+	}
+}
+
+#ifdef CONFIG_SPARSE_IRQ
+unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
+{
+	struct irq_desc *desc = irq_to_desc(irq);
+	return desc->kstat_irqs[cpu];
 }
 #endif
+EXPORT_SYMBOL(kstat_irqs_cpu);
+

kernel/irq/internals.h

@@ -13,6 +13,11 @@ extern void compat_irq_chip_set_default_handler(struct irq_desc *desc);
 extern int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
 		unsigned long flags);
 
+extern struct lock_class_key irq_desc_lock_class;
+extern void init_kstat_irqs(struct irq_desc *desc, int cpu, int nr);
+extern spinlock_t sparse_irq_lock;
+extern struct irq_desc *irq_desc_ptrs[NR_IRQS];
+
 #ifdef CONFIG_PROC_FS
 extern void register_irq_proc(unsigned int irq, struct irq_desc *desc);
 extern void register_handler_proc(unsigned int irq, struct irqaction *action);

kernel/irq/manage.c

@@ -370,16 +370,18 @@ int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
 		return 0;
 	}
 
-	ret = chip->set_type(irq, flags & IRQF_TRIGGER_MASK);
+	/* caller masked out all except trigger mode flags */
+	ret = chip->set_type(irq, flags);
 
 	if (ret)
 		pr_err("setting trigger mode %d for irq %u failed (%pF)\n",
-				(int)(flags & IRQF_TRIGGER_MASK),
-				irq, chip->set_type);
+				(int)flags, irq, chip->set_type);
 	else {
+		if (flags & (IRQ_TYPE_LEVEL_LOW | IRQ_TYPE_LEVEL_HIGH))
+			flags |= IRQ_LEVEL;
 		/* note that IRQF_TRIGGER_MASK == IRQ_TYPE_SENSE_MASK */
-		desc->status &= ~IRQ_TYPE_SENSE_MASK;
-		desc->status |= flags & IRQ_TYPE_SENSE_MASK;
+		desc->status &= ~(IRQ_LEVEL | IRQ_TYPE_SENSE_MASK);
+		desc->status |= flags;
 	}
 
 	return ret;
@@ -459,7 +461,8 @@ __setup_irq(unsigned int irq, struct irq_desc * desc, struct irqaction *new)
 
 		/* Setup the type (level, edge polarity) if configured: */
 		if (new->flags & IRQF_TRIGGER_MASK) {
-			ret = __irq_set_trigger(desc, irq, new->flags);
+			ret = __irq_set_trigger(desc, irq,
+					new->flags & IRQF_TRIGGER_MASK);
 
 			if (ret) {
 				spin_unlock_irqrestore(&desc->lock, flags);

kernel/irq/numa_migrate.c

+/*
+ * NUMA irq-desc migration code
+ *
+ * Migrate IRQ data structures (irq_desc, chip_data, etc.) over to
+ * the new "home node" of the IRQ.
+ */
+
+#include <linux/irq.h>
+#include <linux/module.h>
+#include <linux/random.h>
+#include <linux/interrupt.h>
+#include <linux/kernel_stat.h>
+
+#include "internals.h"
+
+static void init_copy_kstat_irqs(struct irq_desc *old_desc,
+				 struct irq_desc *desc,
+				 int cpu, int nr)
+{
+	unsigned long bytes;
+
+	init_kstat_irqs(desc, cpu, nr);
+
+	if (desc->kstat_irqs != old_desc->kstat_irqs) {
+		/* Compute how many bytes we need per irq and allocate them */
+		bytes = nr * sizeof(unsigned int);
+
+		memcpy(desc->kstat_irqs, old_desc->kstat_irqs, bytes);
+	}
+}
+
+static void free_kstat_irqs(struct irq_desc *old_desc, struct irq_desc *desc)
+{
+	if (old_desc->kstat_irqs == desc->kstat_irqs)
+		return;
+
+	kfree(old_desc->kstat_irqs);
+	old_desc->kstat_irqs = NULL;
+}
+
+static void init_copy_one_irq_desc(int irq, struct irq_desc *old_desc,
+		 struct irq_desc *desc, int cpu)
+{
+	memcpy(desc, old_desc, sizeof(struct irq_desc));
+	desc->cpu = cpu;
+	lockdep_set_class(&desc->lock, &irq_desc_lock_class);
+	init_copy_kstat_irqs(old_desc, desc, cpu, nr_cpu_ids);
+	arch_init_copy_chip_data(old_desc, desc, cpu);
+}
+
+static void free_one_irq_desc(struct irq_desc *old_desc, struct irq_desc *desc)
+{
+	free_kstat_irqs(old_desc, desc);
+	arch_free_chip_data(old_desc, desc);
+}
+
+static struct irq_desc *__real_move_irq_desc(struct irq_desc *old_desc,
+						int cpu)
+{
+	struct irq_desc *desc;
+	unsigned int irq;
+	unsigned long flags;
+	int node;
+
+	irq = old_desc->irq;
+
+	spin_lock_irqsave(&sparse_irq_lock, flags);
+
+	/* We have to check it to avoid races with another CPU */
+	desc = irq_desc_ptrs[irq];
+
+	if (desc && old_desc != desc)
+			goto out_unlock;
+
+	node = cpu_to_node(cpu);
+	desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node);
+	printk(KERN_DEBUG "  move irq_desc for %d to cpu %d node %d\n",
+		 irq, cpu, node);
+	if (!desc) {
+		printk(KERN_ERR "can not get new irq_desc for moving\n");
+		/* still use old one */
+		desc = old_desc;
+		goto out_unlock;
+	}
+	init_copy_one_irq_desc(irq, old_desc, desc, cpu);
+
+	irq_desc_ptrs[irq] = desc;
+
+	/* free the old one */
+	free_one_irq_desc(old_desc, desc);
+	kfree(old_desc);
+
+out_unlock:
+	spin_unlock_irqrestore(&sparse_irq_lock, flags);
+
+	return desc;
+}
+
+struct irq_desc *move_irq_desc(struct irq_desc *desc, int cpu)
+{
+	int old_cpu;
+	int node, old_node;
+
+	/* those all static, do move them */
+	if (desc->irq < NR_IRQS_LEGACY)
+		return desc;
+
+	old_cpu = desc->cpu;
+	printk(KERN_DEBUG
+		 "try to move irq_desc from cpu %d to %d\n", old_cpu, cpu);
+	if (old_cpu != cpu) {
+		node = cpu_to_node(cpu);
+		old_node = cpu_to_node(old_cpu);
+		if (old_node != node)
+			desc = __real_move_irq_desc(desc, cpu);
+		else
+			desc->cpu = cpu;
+	}
+
+	return desc;
+}
+

kernel/irq/proc.c

@@ -243,7 +243,11 @@ void init_irq_proc(void)
 	/*
 	 * Create entries for all existing IRQs.
 	 */
-	for_each_irq_desc(irq, desc)
+	for_each_irq_desc(irq, desc) {
+		if (!desc)
+			continue;
+
 		register_irq_proc(irq, desc);
+	}
 }
 

kernel/irq/spurious.c

@@ -91,6 +91,9 @@ static int misrouted_irq(int irq)
 	int i, ok = 0;
 
 	for_each_irq_desc(i, desc) {
+		if (!desc)
+			continue;
+
 		if (!i)
 			 continue;
 
@@ -112,6 +115,8 @@ static void poll_spurious_irqs(unsigned long dummy)
 	for_each_irq_desc(i, desc) {
 		unsigned int status;
 
+		if (!desc)
+			continue;
 		if (!i)
 			 continue;