[PATCH] genirq: x86_64 irq: make vector_irq per cpu

This refactors the irq handling code to make the vectors a per cpu resource so
the same vector number can be simultaneously used on multiple cpus for
different irqs.

This should make systems that were hitting limits on the total number of irqs
much more livable.

[akpm@osdl.org: build fix]
[akpm@osdl.org: __target_IO_APIC_irq is unneeded on UP]
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Rajesh Shah <rajesh.shah@intel.com>
Cc: Andi Kleen <ak@muc.de>
Cc: "Protasevich, Natalie" <Natalie.Protasevich@UNISYS.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

5 files changed, 144 insertions, 53 deletions

diff --git a/arch/x86_64/kernel/i8259.c b/arch/x86_64/kernel/i8259.c
index bddde431871a..0612a33bb896 100644
--- a/arch/x86_64/kernel/i8259.c
+++ b/arch/x86_64/kernel/i8259.c
@@ -394,7 +394,7 @@ device_initcall(i8259A_init_sysfs);
  */
 
 static struct irqaction irq2 = { no_action, 0, CPU_MASK_NONE, "cascade", NULL, NULL};
-int vector_irq[NR_VECTORS] __read_mostly = {
+DEFINE_PER_CPU(vector_irq_t, vector_irq) = {
 	[0 ... FIRST_EXTERNAL_VECTOR - 1] = -1,
 	[FIRST_EXTERNAL_VECTOR + 0] = 0,
 	[FIRST_EXTERNAL_VECTOR + 1] = 1,
@@ -520,7 +520,7 @@ void __init init_IRQ(void)
 	 * IRQ0 must be given a fixed assignment and initialized,
 	 * because it's used before the IO-APIC is set up.
 	 */
-	vector_irq[FIRST_DEVICE_VECTOR] = 0;
+	__get_cpu_var(vector_irq)[FIRST_DEVICE_VECTOR] = 0;
 
 	/*
 	 * The reschedule interrupt is a CPU-to-CPU reschedule-helper
diff --git a/arch/x86_64/kernel/io_apic.c b/arch/x86_64/kernel/io_apic.c
index 9f849492c296..7d53f66394a7 100644
--- a/arch/x86_64/kernel/io_apic.c
+++ b/arch/x86_64/kernel/io_apic.c
@@ -44,7 +44,7 @@
 #include <asm/nmi.h>
 #include <asm/msidef.h>
 
-static int assign_irq_vector(int irq);
+static int assign_irq_vector(int irq, cpumask_t mask);
 
 #define __apicdebuginit  __init
 
@@ -135,11 +135,35 @@ static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
 }
 
 #ifdef CONFIG_SMP
+static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
+{
+	int apic, pin;
+	struct irq_pin_list *entry = irq_2_pin + irq;
+
+	BUG_ON(irq >= NR_IRQS);
+	for (;;) {
+		unsigned int reg;
+		apic = entry->apic;
+		pin = entry->pin;
+		if (pin == -1)
+			break;
+		io_apic_write(apic, 0x11 + pin*2, dest);
+		reg = io_apic_read(apic, 0x10 + pin*2);
+		reg &= ~0x000000ff;
+		reg |= vector;
+		io_apic_modify(apic, reg);
+		if (!entry->next)
+			break;
+		entry = irq_2_pin + entry->next;
+	}
+}
+
 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
 {
 	unsigned long flags;
 	unsigned int dest;
 	cpumask_t tmp;
+	int vector;
 
 	cpus_and(tmp, mask, cpu_online_map);
 	if (cpus_empty(tmp))
@@ -147,7 +171,13 @@ static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
 
 	cpus_and(mask, tmp, CPU_MASK_ALL);
 
-	dest = cpu_mask_to_apicid(mask);
+	vector = assign_irq_vector(irq, mask);
+	if (vector < 0)
+		return;
+
+	cpus_clear(tmp);
+	cpu_set(vector >> 8, tmp);
+	dest = cpu_mask_to_apicid(tmp);
 
 	/*
 	 * Only the high 8 bits are valid.
@@ -155,7 +185,7 @@ static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
 	dest = SET_APIC_LOGICAL_ID(dest);
 
 	spin_lock_irqsave(&ioapic_lock, flags);
-	__DO_ACTION(1, = dest, )
+	__target_IO_APIC_irq(irq, dest, vector & 0xff);
 	set_native_irq_info(irq, mask);
 	spin_unlock_irqrestore(&ioapic_lock, flags);
 }
@@ -512,7 +542,7 @@ int gsi_irq_sharing(int gsi)
 
 	tries = NR_IRQS;
   try_again:
-	vector = assign_irq_vector(gsi);
+	vector = assign_irq_vector(gsi, TARGET_CPUS);
 
 	/*
 	 * Sharing vectors means sharing IRQs, so scan irq_vectors for previous
@@ -591,45 +621,77 @@ static inline int IO_APIC_irq_trigger(int irq)
 }
 
 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
-u8 irq_vector[NR_IRQ_VECTORS] __read_mostly = { FIRST_DEVICE_VECTOR , 0 };
+unsigned int irq_vector[NR_IRQ_VECTORS] __read_mostly = { FIRST_EXTERNAL_VECTOR, 0 };
 
-static int __assign_irq_vector(int irq)
+static int __assign_irq_vector(int irq, cpumask_t mask)
 {
-	static int current_vector = FIRST_DEVICE_VECTOR, offset = 0;
-	int vector;
+	/*
+	 * NOTE! The local APIC isn't very good at handling
+	 * multiple interrupts at the same interrupt level.
+	 * As the interrupt level is determined by taking the
+	 * vector number and shifting that right by 4, we
+	 * want to spread these out a bit so that they don't
+	 * all fall in the same interrupt level.
+	 *
+	 * Also, we've got to be careful not to trash gate
+	 * 0x80, because int 0x80 is hm, kind of importantish. ;)
+	 */
+	static struct {
+		int vector;
+		int offset;
+	} pos[NR_CPUS] = { [ 0 ... NR_CPUS - 1] = {FIRST_DEVICE_VECTOR, 0} };
+	int old_vector = -1;
+	int cpu;
 
 	BUG_ON((unsigned)irq >= NR_IRQ_VECTORS);
 
-	if (IO_APIC_VECTOR(irq) > 0) {
-		return IO_APIC_VECTOR(irq);
+	if (IO_APIC_VECTOR(irq) > 0)
+		old_vector = IO_APIC_VECTOR(irq);
+	if ((old_vector > 0) && cpu_isset(old_vector >> 8, mask)) {
+		return old_vector;
 	}
+
+	for_each_cpu_mask(cpu, mask) {
+		int vector, offset;
+		vector = pos[cpu].vector;
+		offset = pos[cpu].offset;
 next:
-	current_vector += 8;
-	if (current_vector == IA32_SYSCALL_VECTOR)
-		goto next;
-
-	if (current_vector >= FIRST_SYSTEM_VECTOR) {
-		/* If we run out of vectors on large boxen, must share them. */
-		offset = (offset + 1) % 8;
-		current_vector = FIRST_DEVICE_VECTOR + offset;
+		vector += 8;
+		if (vector >= FIRST_SYSTEM_VECTOR) {
+			/* If we run out of vectors on large boxen, must share them. */
+			offset = (offset + 1) % 8;
+			vector = FIRST_DEVICE_VECTOR + offset;
+		}
+		if (unlikely(pos[cpu].vector == vector))
+			continue;
+		if (vector == IA32_SYSCALL_VECTOR)
+			goto next;
+		if (per_cpu(vector_irq, cpu)[vector] != -1)
+			goto next;
+		/* Found one! */
+		pos[cpu].vector = vector;
+		pos[cpu].offset = offset;
+		if (old_vector >= 0) {
+			int old_cpu = old_vector >> 8;
+			old_vector &= 0xff;
+			per_cpu(vector_irq, old_cpu)[old_vector] = -1;
+		}
+		per_cpu(vector_irq, cpu)[vector] = irq;
+		vector |= cpu << 8;
+		IO_APIC_VECTOR(irq) = vector;
+		return vector;
 	}
-
-	vector = current_vector;
-	vector_irq[vector] = irq;
-	IO_APIC_VECTOR(irq) = vector;
-
-	return vector;
+	return -ENOSPC;
 }
 
-static int assign_irq_vector(int irq)
+static int assign_irq_vector(int irq, cpumask_t mask)
 {
 	int vector;
 	unsigned long flags;
 
 	spin_lock_irqsave(&vector_lock, flags);
-	vector = __assign_irq_vector(irq);
+	vector = __assign_irq_vector(irq, mask);
 	spin_unlock_irqrestore(&vector_lock, flags);
-
 	return vector;
 }
 
@@ -699,8 +761,15 @@ static void __init setup_IO_APIC_irqs(void)
 			continue;
 
 		if (IO_APIC_IRQ(irq)) {
-			vector = assign_irq_vector(irq);
-			entry.vector = vector;
+			cpumask_t mask;
+			vector = assign_irq_vector(irq, TARGET_CPUS);
+			if (vector < 0)
+				continue;
+
+			cpus_clear(mask);
+			cpu_set(vector >> 8, mask);
+			entry.dest.logical.logical_dest = cpu_mask_to_apicid(mask);
+			entry.vector = vector & 0xff;
 
 			ioapic_register_intr(irq, vector, IOAPIC_AUTO);
 			if (!apic && (irq < 16))
@@ -1197,7 +1266,14 @@ static unsigned int startup_ioapic_irq(unsigned int irq)
 
 static int ioapic_retrigger_irq(unsigned int irq)
 {
-	send_IPI_self(IO_APIC_VECTOR(irq));
+	cpumask_t mask;
+	unsigned vector;
+
+	vector = irq_vector[irq];
+	cpus_clear(mask);
+	cpu_set(vector >> 8, mask);
+
+	send_IPI_mask(mask, vector & 0xff);
 
 	return 1;
 }
@@ -1419,7 +1495,7 @@ static inline void check_timer(void)
 	 * get/set the timer IRQ vector:
 	 */
 	disable_8259A_irq(0);
-	vector = assign_irq_vector(0);
+	vector = assign_irq_vector(0, TARGET_CPUS);
 
 	/*
 	 * Subtle, code in do_timer_interrupt() expects an AEOI
@@ -1662,7 +1738,7 @@ int create_irq(void)
 			continue;
 		if (irq_vector[new] != 0)
 			continue;
-		vector = __assign_irq_vector(new);
+		vector = __assign_irq_vector(new, TARGET_CPUS);
 		if (likely(vector > 0))
 			irq = new;
 		break;
@@ -1698,12 +1774,12 @@ static int msi_msg_setup(struct pci_dev *pdev, unsigned int irq, struct msi_msg
 	int vector;
 	unsigned dest;
 
-	vector = assign_irq_vector(irq);
+	vector = assign_irq_vector(irq, TARGET_CPUS);
 	if (vector >= 0) {
 		cpumask_t tmp;
 
 		cpus_clear(tmp);
-		cpu_set(first_cpu(cpu_online_map), tmp);
+		cpu_set(vector >> 8, tmp);
 		dest = cpu_mask_to_apicid(tmp);
 
 		msg->address_hi = MSI_ADDR_BASE_HI;
@@ -1738,9 +1814,13 @@ static void msi_msg_set_affinity(unsigned int irq, cpumask_t mask, struct msi_ms
 	int vector;
 	unsigned dest;
 
-	vector = assign_irq_vector(irq);
+	vector = assign_irq_vector(irq, mask);
 	if (vector > 0) {
-		dest = cpu_mask_to_apicid(mask);
+		cpumask_t tmp;
+
+		cpus_clear(tmp);
+		cpu_set(vector >> 8, tmp);
+		dest = cpu_mask_to_apicid(tmp);
 
 		msg->data &= ~MSI_DATA_VECTOR_MASK;
 		msg->data |= MSI_DATA_VECTOR(vector);
@@ -1783,6 +1863,8 @@ int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int p
 {
 	struct IO_APIC_route_entry entry;
 	unsigned long flags;
+	int vector;
+	cpumask_t mask;
 
 	if (!IO_APIC_IRQ(irq)) {
 		apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
@@ -1790,6 +1872,21 @@ int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int p
 		return -EINVAL;
 	}
 
+	irq = gsi_irq_sharing(irq);
+	/*
+	 * IRQs < 16 are already in the irq_2_pin[] map
+	 */
+	if (irq >= 16)
+		add_pin_to_irq(irq, ioapic, pin);
+
+
+	vector = assign_irq_vector(irq, TARGET_CPUS);
+	if (vector < 0)
+		return vector;
+
+	cpus_clear(mask);
+	cpu_set(vector >> 8, mask);
+
 	/*
 	 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
 	 * Note that we mask (disable) IRQs now -- these get enabled when the
@@ -1800,19 +1897,11 @@ int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int p
 
 	entry.delivery_mode = INT_DELIVERY_MODE;
 	entry.dest_mode = INT_DEST_MODE;
-	entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
+	entry.dest.logical.logical_dest = cpu_mask_to_apicid(mask);
 	entry.trigger = triggering;
 	entry.polarity = polarity;
 	entry.mask = 1;					 /* Disabled (masked) */
-
-	irq = gsi_irq_sharing(irq);
-	/*
-	 * IRQs < 16 are already in the irq_2_pin[] map
-	 */
-	if (irq >= 16)
-		add_pin_to_irq(irq, ioapic, pin);
-
-	entry.vector = assign_irq_vector(irq);
+	entry.vector = vector & 0xff;
 
 	apic_printk(APIC_VERBOSE,KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry (%d-%d -> 0x%x -> "
 		"IRQ %d Mode:%i Active:%i)\n", ioapic, 
diff --git a/arch/x86_64/kernel/irq.c b/arch/x86_64/kernel/irq.c
index 4542fb031994..506f27c85ca5 100644
--- a/arch/x86_64/kernel/irq.c
+++ b/arch/x86_64/kernel/irq.c
@@ -110,7 +110,7 @@ asmlinkage unsigned int do_IRQ(struct pt_regs *regs)
 
 	exit_idle();
 	irq_enter();
-	irq = vector_irq[vector];
+	irq = __get_cpu_var(vector_irq)[vector];
 
 	if (unlikely(irq >= NR_IRQS)) {
 		printk(KERN_EMERG "%s: cannot handle IRQ %d\n",
diff --git a/include/asm-x86_64/hw_irq.h b/include/asm-x86_64/hw_irq.h
index 9f6a0bfed727..53d0d9fd10d6 100644
--- a/include/asm-x86_64/hw_irq.h
+++ b/include/asm-x86_64/hw_irq.h
@@ -19,6 +19,7 @@
 #include <asm/irq.h>
 #include <linux/profile.h>
 #include <linux/smp.h>
+#include <linux/percpu.h>
 #endif
 
 #define NMI_VECTOR		0x02
@@ -73,8 +74,9 @@
 
 
 #ifndef __ASSEMBLY__
-extern u8 irq_vector[NR_IRQ_VECTORS];
-extern int vector_irq[NR_VECTORS];
+extern unsigned int irq_vector[NR_IRQ_VECTORS];
+typedef int vector_irq_t[NR_VECTORS];
+DECLARE_PER_CPU(vector_irq_t, vector_irq);
 #define IO_APIC_VECTOR(irq)	(irq_vector[irq])
 
 /*
diff --git a/include/asm-x86_64/irq.h b/include/asm-x86_64/irq.h
index b8f87287847f..5006c6e75656 100644
--- a/include/asm-x86_64/irq.h
+++ b/include/asm-x86_64/irq.h
@@ -31,8 +31,8 @@
 
 #define FIRST_SYSTEM_VECTOR	0xef   /* duplicated in hw_irq.h */
 
-#define NR_IRQS 224
-#define NR_IRQ_VECTORS (32 * NR_CPUS)
+#define NR_IRQS (NR_VECTORS + (32 *NR_CPUS))
+#define NR_IRQ_VECTORS NR_IRQS
 
 static __inline__ int irq_canonicalize(int irq)
 {