MIPS: Octeon: Rewrite interrupt handling code.

This includes conversion to new style irq_chip functions, and
correctly enabling/disabling per-CPU interrupts.

The hardware interrupt bit to irq number mapping is now done with a
flexible map, instead of by bit twiddling the irq number.

[ tglx: Adjusted to new irq_cpu_on/offline callbacks and
        __irq_set_affinity_lock ]

Signed-off-by: David Daney <ddaney@caviumnetworks.com>
Cc: linux-mips@linux-mips.org
Cc: ralf@linux-mips.org
LKML-Reference: <1301081931-11240-5-git-send-email-ddaney@caviumnetworks.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

6 files changed, 921 insertions, 817 deletions

diff --git a/arch/mips/cavium-octeon/octeon-irq.c b/arch/mips/cavium-octeon/octeon-irq.c
index ce7500cdf5b7..03c081da3df4 100644
--- a/arch/mips/cavium-octeon/octeon-irq.c
+++ b/arch/mips/cavium-octeon/octeon-irq.c
@@ -3,10 +3,13 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (C) 2004-2008, 2009, 2010 Cavium Networks
+ * Copyright (C) 2004-2008, 2009, 2010, 2011 Cavium Networks
  */
-#include <linux/irq.h>
+
 #include <linux/interrupt.h>
+#include <linux/bitops.h>
+#include <linux/percpu.h>
+#include <linux/irq.h>
 #include <linux/smp.h>
 
 #include <asm/octeon/octeon.h>
@@ -14,6 +17,47 @@
 static DEFINE_RAW_SPINLOCK(octeon_irq_ciu0_lock);
 static DEFINE_RAW_SPINLOCK(octeon_irq_ciu1_lock);
 
+static DEFINE_PER_CPU(unsigned long, octeon_irq_ciu0_en_mirror);
+static DEFINE_PER_CPU(unsigned long, octeon_irq_ciu1_en_mirror);
+
+static __read_mostly u8 octeon_irq_ciu_to_irq[8][64];
+
+union octeon_ciu_chip_data {
+	void *p;
+	unsigned long l;
+	struct {
+		unsigned int line:6;
+		unsigned int bit:6;
+	} s;
+};
+
+struct octeon_core_chip_data {
+	struct mutex core_irq_mutex;
+	bool current_en;
+	bool desired_en;
+	u8 bit;
+};
+
+#define MIPS_CORE_IRQ_LINES 8
+
+static struct octeon_core_chip_data octeon_irq_core_chip_data[MIPS_CORE_IRQ_LINES];
+
+static void __init octeon_irq_set_ciu_mapping(int irq, int line, int bit,
+					      struct irq_chip *chip,
+					      irq_flow_handler_t handler)
+{
+	union octeon_ciu_chip_data cd;
+
+	irq_set_chip_and_handler(irq, chip, handler);
+
+	cd.l = 0;
+	cd.s.line = line;
+	cd.s.bit = bit;
+
+	irq_set_chip_data(irq, cd.p);
+	octeon_irq_ciu_to_irq[line][bit] = irq;
+}
+
 static int octeon_coreid_for_cpu(int cpu)
 {
 #ifdef CONFIG_SMP
@@ -23,9 +67,20 @@ static int octeon_coreid_for_cpu(int cpu)
 #endif
 }
 
-static void octeon_irq_core_ack(unsigned int irq)
+static int octeon_cpu_for_coreid(int coreid)
+{
+#ifdef CONFIG_SMP
+	return cpu_number_map(coreid);
+#else
+	return smp_processor_id();
+#endif
+}
+
+static void octeon_irq_core_ack(struct irq_data *data)
 {
-	unsigned int bit = irq - OCTEON_IRQ_SW0;
+	struct octeon_core_chip_data *cd = irq_data_get_irq_chip_data(data);
+	unsigned int bit = cd->bit;
+
 	/*
 	 * We don't need to disable IRQs to make these atomic since
 	 * they are already disabled earlier in the low level
@@ -37,131 +92,133 @@ static void octeon_irq_core_ack(unsigned int irq)
 		clear_c0_cause(0x100 << bit);
 }
 
-static void octeon_irq_core_eoi(unsigned int irq)
+static void octeon_irq_core_eoi(struct irq_data *data)
 {
-	struct irq_desc *desc = irq_to_desc(irq);
-	unsigned int bit = irq - OCTEON_IRQ_SW0;
-	/*
-	 * If an IRQ is being processed while we are disabling it the
-	 * handler will attempt to unmask the interrupt after it has
-	 * been disabled.
-	 */
-	if ((unlikely(desc->status & IRQ_DISABLED)))
-		return;
+	struct octeon_core_chip_data *cd = irq_data_get_irq_chip_data(data);
+
 	/*
 	 * We don't need to disable IRQs to make these atomic since
 	 * they are already disabled earlier in the low level
 	 * interrupt code.
 	 */
-	set_c0_status(0x100 << bit);
+	set_c0_status(0x100 << cd->bit);
 }
 
-static void octeon_irq_core_enable(unsigned int irq)
+static void octeon_irq_core_set_enable_local(void *arg)
 {
-	unsigned long flags;
-	unsigned int bit = irq - OCTEON_IRQ_SW0;
+	struct irq_data *data = arg;
+	struct octeon_core_chip_data *cd = irq_data_get_irq_chip_data(data);
+	unsigned int mask = 0x100 << cd->bit;
 
 	/*
-	 * We need to disable interrupts to make sure our updates are
-	 * atomic.
+	 * Interrupts are already disabled, so these are atomic.
 	 */
-	local_irq_save(flags);
-	set_c0_status(0x100 << bit);
-	local_irq_restore(flags);
+	if (cd->desired_en)
+		set_c0_status(mask);
+	else
+		clear_c0_status(mask);
+
 }
 
-static void octeon_irq_core_disable_local(unsigned int irq)
+static void octeon_irq_core_disable(struct irq_data *data)
 {
-	unsigned long flags;
-	unsigned int bit = irq - OCTEON_IRQ_SW0;
-	/*
-	 * We need to disable interrupts to make sure our updates are
-	 * atomic.
-	 */
-	local_irq_save(flags);
-	clear_c0_status(0x100 << bit);
-	local_irq_restore(flags);
+	struct octeon_core_chip_data *cd = irq_data_get_irq_chip_data(data);
+	cd->desired_en = false;
 }
 
-static void octeon_irq_core_disable(unsigned int irq)
+static void octeon_irq_core_enable(struct irq_data *data)
 {
-#ifdef CONFIG_SMP
-	on_each_cpu((void (*)(void *)) octeon_irq_core_disable_local,
-		    (void *) (long) irq, 1);
-#else
-	octeon_irq_core_disable_local(irq);
-#endif
+	struct octeon_core_chip_data *cd = irq_data_get_irq_chip_data(data);
+	cd->desired_en = true;
 }
 
-static struct irq_chip octeon_irq_chip_core = {
-	.name = "Core",
-	.enable = octeon_irq_core_enable,
-	.disable = octeon_irq_core_disable,
-	.ack = octeon_irq_core_ack,
-	.eoi = octeon_irq_core_eoi,
-};
+static void octeon_irq_core_bus_lock(struct irq_data *data)
+{
+	struct octeon_core_chip_data *cd = irq_data_get_irq_chip_data(data);
 
+	mutex_lock(&cd->core_irq_mutex);
+}
 
-static void octeon_irq_ciu0_ack(unsigned int irq)
+static void octeon_irq_core_bus_sync_unlock(struct irq_data *data)
 {
-	switch (irq) {
-	case OCTEON_IRQ_GMX_DRP0:
-	case OCTEON_IRQ_GMX_DRP1:
-	case OCTEON_IRQ_IPD_DRP:
-	case OCTEON_IRQ_KEY_ZERO:
-	case OCTEON_IRQ_TIMER0:
-	case OCTEON_IRQ_TIMER1:
-	case OCTEON_IRQ_TIMER2:
-	case OCTEON_IRQ_TIMER3:
-	{
-		int index = cvmx_get_core_num() * 2;
-		u64 mask = 1ull << (irq - OCTEON_IRQ_WORKQ0);
-		/*
-		 * CIU timer type interrupts must be acknoleged by
-		 * writing a '1' bit to their sum0 bit.
-		 */
-		cvmx_write_csr(CVMX_CIU_INTX_SUM0(index), mask);
-		break;
-	}
-	default:
-		break;
+	struct octeon_core_chip_data *cd = irq_data_get_irq_chip_data(data);
+
+	if (cd->desired_en != cd->current_en) {
+		on_each_cpu(octeon_irq_core_set_enable_local, data, 1);
+
+		cd->current_en = cd->desired_en;
 	}
 
-	/*
-	 * In order to avoid any locking accessing the CIU, we
-	 * acknowledge CIU interrupts by disabling all of them.  This
-	 * way we can use a per core register and avoid any out of
-	 * core locking requirements.  This has the side affect that
-	 * CIU interrupts can't be processed recursively.
-	 *
-	 * We don't need to disable IRQs to make these atomic since
-	 * they are already disabled earlier in the low level
-	 * interrupt code.
-	 */
-	clear_c0_status(0x100 << 2);
+	mutex_unlock(&cd->core_irq_mutex);
 }
 
-static void octeon_irq_ciu0_eoi(unsigned int irq)
+
+static void octeon_irq_core_cpu_online(struct irq_data *data)
 {
-	/*
-	 * Enable all CIU interrupts again.  We don't need to disable
-	 * IRQs to make these atomic since they are already disabled
-	 * earlier in the low level interrupt code.
-	 */
-	set_c0_status(0x100 << 2);
+	if (irqd_irq_disabled(data))
+		octeon_irq_core_eoi(data);
+}
+
+static void octeon_irq_core_cpu_offline(struct irq_data *data)
+{
+	if (irqd_irq_disabled(data))
+		octeon_irq_core_ack(data);
 }
 
-static int next_coreid_for_irq(struct irq_desc *desc)
+static struct irq_chip octeon_irq_chip_core = {
+	.name = "Core",
+	.irq_enable = octeon_irq_core_enable,
+	.irq_disable = octeon_irq_core_disable,
+	.irq_ack = octeon_irq_core_ack,
+	.irq_eoi = octeon_irq_core_eoi,
+	.irq_bus_lock = octeon_irq_core_bus_lock,
+	.irq_bus_sync_unlock = octeon_irq_core_bus_sync_unlock,
+
+	.irq_cpu_online = octeon_irq_core_cpu_online,
+	.irq_cpu_offline = octeon_irq_core_cpu_offline,
+};
+
+static void __init octeon_irq_init_core(void)
+{
+	int i;
+	int irq;
+	struct octeon_core_chip_data *cd;
+
+	for (i = 0; i < MIPS_CORE_IRQ_LINES; i++) {
+		cd = &octeon_irq_core_chip_data[i];
+		cd->current_en = false;
+		cd->desired_en = false;
+		cd->bit = i;
+		mutex_init(&cd->core_irq_mutex);
+
+		irq = OCTEON_IRQ_SW0 + i;
+		switch (irq) {
+		case OCTEON_IRQ_TIMER:
+		case OCTEON_IRQ_SW0:
+		case OCTEON_IRQ_SW1:
+		case OCTEON_IRQ_5:
+		case OCTEON_IRQ_PERF:
+			irq_set_chip_data(irq, cd);
+			irq_set_chip_and_handler(irq, &octeon_irq_chip_core,
+						 handle_percpu_irq);
+			break;
+		default:
+			break;
+		}
+	}
+}
+
+static int next_cpu_for_irq(struct irq_data *data)
 {
 
 #ifdef CONFIG_SMP
-	int coreid;
-	int weight = cpumask_weight(desc->affinity);
+	int cpu;
+	int weight = cpumask_weight(data->affinity);
 
 	if (weight > 1) {
-		int cpu = smp_processor_id();
+		cpu = smp_processor_id();
 		for (;;) {
-			cpu = cpumask_next(cpu, desc->affinity);
+			cpu = cpumask_next(cpu, data->affinity);
 			if (cpu >= nr_cpu_ids) {
 				cpu = -1;
 				continue;
@@ -169,83 +226,175 @@ static int next_coreid_for_irq(struct irq_desc *desc)
 				break;
 			}
 		}
-		coreid = octeon_coreid_for_cpu(cpu);
 	} else if (weight == 1) {
-		coreid = octeon_coreid_for_cpu(cpumask_first(desc->affinity));
+		cpu = cpumask_first(data->affinity);
 	} else {
-		coreid = cvmx_get_core_num();
+		cpu = smp_processor_id();
 	}
-	return coreid;
+	return cpu;
 #else
-	return cvmx_get_core_num();
+	return smp_processor_id();
 #endif
 }
 
-static void octeon_irq_ciu0_enable(unsigned int irq)
+static void octeon_irq_ciu_enable(struct irq_data *data)
 {
-	struct irq_desc *desc = irq_to_desc(irq);
-	int coreid = next_coreid_for_irq(desc);
+	int cpu = next_cpu_for_irq(data);
+	int coreid = octeon_coreid_for_cpu(cpu);
+	unsigned long *pen;
 	unsigned long flags;
-	uint64_t en0;
-	int bit = irq - OCTEON_IRQ_WORKQ0;	/* Bit 0-63 of EN0 */
+	union octeon_ciu_chip_data cd;
+
+	cd.p = irq_data_get_irq_chip_data(data);
 
-	raw_spin_lock_irqsave(&octeon_irq_ciu0_lock, flags);
-	en0 = cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2));
-	en0 |= 1ull << bit;
-	cvmx_write_csr(CVMX_CIU_INTX_EN0(coreid * 2), en0);
-	cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2));
-	raw_spin_unlock_irqrestore(&octeon_irq_ciu0_lock, flags);
+	if (cd.s.line == 0) {
+		raw_spin_lock_irqsave(&octeon_irq_ciu0_lock, flags);
+		pen = &per_cpu(octeon_irq_ciu0_en_mirror, cpu);
+		set_bit(cd.s.bit, pen);
+		cvmx_write_csr(CVMX_CIU_INTX_EN0(coreid * 2), *pen);
+		raw_spin_unlock_irqrestore(&octeon_irq_ciu0_lock, flags);
+	} else {
+		raw_spin_lock_irqsave(&octeon_irq_ciu1_lock, flags);
+		pen = &per_cpu(octeon_irq_ciu1_en_mirror, cpu);
+		set_bit(cd.s.bit, pen);
+		cvmx_write_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1), *pen);
+		raw_spin_unlock_irqrestore(&octeon_irq_ciu1_lock, flags);
+	}
 }
 
-static void octeon_irq_ciu0_enable_mbox(unsigned int irq)
+static void octeon_irq_ciu_enable_local(struct irq_data *data)
+{
+	unsigned long *pen;
+	unsigned long flags;
+	union octeon_ciu_chip_data cd;
+
+	cd.p = irq_data_get_irq_chip_data(data);
+
+	if (cd.s.line == 0) {
+		raw_spin_lock_irqsave(&octeon_irq_ciu0_lock, flags);
+		pen = &__get_cpu_var(octeon_irq_ciu0_en_mirror);
+		set_bit(cd.s.bit, pen);
+		cvmx_write_csr(CVMX_CIU_INTX_EN0(cvmx_get_core_num() * 2), *pen);
+		raw_spin_unlock_irqrestore(&octeon_irq_ciu0_lock, flags);
+	} else {
+		raw_spin_lock_irqsave(&octeon_irq_ciu1_lock, flags);
+		pen = &__get_cpu_var(octeon_irq_ciu1_en_mirror);
+		set_bit(cd.s.bit, pen);
+		cvmx_write_csr(CVMX_CIU_INTX_EN1(cvmx_get_core_num() * 2 + 1), *pen);
+		raw_spin_unlock_irqrestore(&octeon_irq_ciu1_lock, flags);
+	}
+}
+
+static void octeon_irq_ciu_disable_local(struct irq_data *data)
+{
+	unsigned long *pen;
+	unsigned long flags;
+	union octeon_ciu_chip_data cd;
+
+	cd.p = irq_data_get_irq_chip_data(data);
+
+	if (cd.s.line == 0) {
+		raw_spin_lock_irqsave(&octeon_irq_ciu0_lock, flags);
+		pen = &__get_cpu_var(octeon_irq_ciu0_en_mirror);
+		clear_bit(cd.s.bit, pen);
+		cvmx_write_csr(CVMX_CIU_INTX_EN0(cvmx_get_core_num() * 2), *pen);
+		raw_spin_unlock_irqrestore(&octeon_irq_ciu0_lock, flags);
+	} else {
+		raw_spin_lock_irqsave(&octeon_irq_ciu1_lock, flags);
+		pen = &__get_cpu_var(octeon_irq_ciu1_en_mirror);
+		clear_bit(cd.s.bit, pen);
+		cvmx_write_csr(CVMX_CIU_INTX_EN1(cvmx_get_core_num() * 2 + 1), *pen);
+		raw_spin_unlock_irqrestore(&octeon_irq_ciu1_lock, flags);
+	}
+}
+
+static void octeon_irq_ciu_disable_all(struct irq_data *data)
 {
-	int coreid = cvmx_get_core_num();
 	unsigned long flags;
-	uint64_t en0;
-	int bit = irq - OCTEON_IRQ_WORKQ0;	/* Bit 0-63 of EN0 */
+	unsigned long *pen;
+	int cpu;
+	union octeon_ciu_chip_data cd;
+
+	wmb(); /* Make sure flag changes arrive before register updates. */
 
-	raw_spin_lock_irqsave(&octeon_irq_ciu0_lock, flags);
-	en0 = cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2));
-	en0 |= 1ull << bit;
-	cvmx_write_csr(CVMX_CIU_INTX_EN0(coreid * 2), en0);
-	cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2));
-	raw_spin_unlock_irqrestore(&octeon_irq_ciu0_lock, flags);
+	cd.p = irq_data_get_irq_chip_data(data);
+
+	if (cd.s.line == 0) {
+		raw_spin_lock_irqsave(&octeon_irq_ciu0_lock, flags);
+		for_each_online_cpu(cpu) {
+			int coreid = octeon_coreid_for_cpu(cpu);
+			pen = &per_cpu(octeon_irq_ciu0_en_mirror, cpu);
+			clear_bit(cd.s.bit, pen);
+			cvmx_write_csr(CVMX_CIU_INTX_EN0(coreid * 2), *pen);
+		}
+		raw_spin_unlock_irqrestore(&octeon_irq_ciu0_lock, flags);
+	} else {
+		raw_spin_lock_irqsave(&octeon_irq_ciu1_lock, flags);
+		for_each_online_cpu(cpu) {
+			int coreid = octeon_coreid_for_cpu(cpu);
+			pen = &per_cpu(octeon_irq_ciu1_en_mirror, cpu);
+			clear_bit(cd.s.bit, pen);
+			cvmx_write_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1), *pen);
+		}
+		raw_spin_unlock_irqrestore(&octeon_irq_ciu1_lock, flags);
+	}
 }
 
-static void octeon_irq_ciu0_disable(unsigned int irq)
+static void octeon_irq_ciu_enable_all(struct irq_data *data)
 {
-	int bit = irq - OCTEON_IRQ_WORKQ0;	/* Bit 0-63 of EN0 */
 	unsigned long flags;
-	uint64_t en0;
+	unsigned long *pen;
 	int cpu;
-	raw_spin_lock_irqsave(&octeon_irq_ciu0_lock, flags);
-	for_each_online_cpu(cpu) {
-		int coreid = octeon_coreid_for_cpu(cpu);
-		en0 = cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2));
-		en0 &= ~(1ull << bit);
-		cvmx_write_csr(CVMX_CIU_INTX_EN0(coreid * 2), en0);
+	union octeon_ciu_chip_data cd;
+
+	cd.p = irq_data_get_irq_chip_data(data);
+
+	if (cd.s.line == 0) {
+		raw_spin_lock_irqsave(&octeon_irq_ciu0_lock, flags);
+		for_each_online_cpu(cpu) {
+			int coreid = octeon_coreid_for_cpu(cpu);
+			pen = &per_cpu(octeon_irq_ciu0_en_mirror, cpu);
+			set_bit(cd.s.bit, pen);
+			cvmx_write_csr(CVMX_CIU_INTX_EN0(coreid * 2), *pen);
+		}
+		raw_spin_unlock_irqrestore(&octeon_irq_ciu0_lock, flags);
+	} else {
+		raw_spin_lock_irqsave(&octeon_irq_ciu1_lock, flags);
+		for_each_online_cpu(cpu) {
+			int coreid = octeon_coreid_for_cpu(cpu);
+			pen = &per_cpu(octeon_irq_ciu1_en_mirror, cpu);
+			set_bit(cd.s.bit, pen);
+			cvmx_write_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1), *pen);
+		}
+		raw_spin_unlock_irqrestore(&octeon_irq_ciu1_lock, flags);
 	}
-	/*
-	 * We need to do a read after the last update to make sure all
-	 * of them are done.
-	 */
-	cvmx_read_csr(CVMX_CIU_INTX_EN0(cvmx_get_core_num() * 2));
-	raw_spin_unlock_irqrestore(&octeon_irq_ciu0_lock, flags);
 }
 
 /*
  * Enable the irq on the next core in the affinity set for chips that
  * have the EN*_W1{S,C} registers.
  */
-static void octeon_irq_ciu0_enable_v2(unsigned int irq)
+static void octeon_irq_ciu_enable_v2(struct irq_data *data)
 {
-	int index;
-	u64 mask = 1ull << (irq - OCTEON_IRQ_WORKQ0);
-	struct irq_desc *desc = irq_to_desc(irq);
+	u64 mask;
+	int cpu = next_cpu_for_irq(data);
+	union octeon_ciu_chip_data cd;
 
-	if ((desc->status & IRQ_DISABLED) == 0) {
-		index = next_coreid_for_irq(desc) * 2;
+	cd.p = irq_data_get_irq_chip_data(data);
+	mask = 1ull << (cd.s.bit);
+
+	/*
+	 * Called under the desc lock, so these should never get out
+	 * of sync.
+	 */
+	if (cd.s.line == 0) {
+		int index = octeon_coreid_for_cpu(cpu) * 2;
+		set_bit(cd.s.bit, &per_cpu(octeon_irq_ciu0_en_mirror, cpu));
 		cvmx_write_csr(CVMX_CIU_INTX_EN0_W1S(index), mask);
+	} else {
+		int index = octeon_coreid_for_cpu(cpu) * 2 + 1;
+		set_bit(cd.s.bit, &per_cpu(octeon_irq_ciu1_en_mirror, cpu));
+		cvmx_write_csr(CVMX_CIU_INTX_EN1_W1S(index), mask);
 	}
 }
 
@@ -253,83 +402,180 @@ static void octeon_irq_ciu0_enable_v2(unsigned int irq)
  * Enable the irq on the current CPU for chips that
  * have the EN*_W1{S,C} registers.
  */
-static void octeon_irq_ciu0_enable_mbox_v2(unsigned int irq)
+static void octeon_irq_ciu_enable_local_v2(struct irq_data *data)
 {
-	int index;
-	u64 mask = 1ull << (irq - OCTEON_IRQ_WORKQ0);
+	u64 mask;
+	union octeon_ciu_chip_data cd;
+
+	cd.p = irq_data_get_irq_chip_data(data);
+	mask = 1ull << (cd.s.bit);
 
-	index = cvmx_get_core_num() * 2;
-	cvmx_write_csr(CVMX_CIU_INTX_EN0_W1S(index), mask);
+	if (cd.s.line == 0) {
+		int index = cvmx_get_core_num() * 2;
+		set_bit(cd.s.bit, &__get_cpu_var(octeon_irq_ciu0_en_mirror));
+		cvmx_write_csr(CVMX_CIU_INTX_EN0_W1S(index), mask);
+	} else {
+		int index = cvmx_get_core_num() * 2 + 1;
+		set_bit(cd.s.bit, &__get_cpu_var(octeon_irq_ciu1_en_mirror));
+		cvmx_write_csr(CVMX_CIU_INTX_EN1_W1S(index), mask);
+	}
+}
+
+static void octeon_irq_ciu_disable_local_v2(struct irq_data *data)
+{
+	u64 mask;
+	union octeon_ciu_chip_data cd;
+
+	cd.p = irq_data_get_irq_chip_data(data);
+	mask = 1ull << (cd.s.bit);
+
+	if (cd.s.line == 0) {
+		int index = cvmx_get_core_num() * 2;
+		clear_bit(cd.s.bit, &__get_cpu_var(octeon_irq_ciu0_en_mirror));
+		cvmx_write_csr(CVMX_CIU_INTX_EN0_W1C(index), mask);
+	} else {
+		int index = cvmx_get_core_num() * 2 + 1;
+		clear_bit(cd.s.bit, &__get_cpu_var(octeon_irq_ciu1_en_mirror));
+		cvmx_write_csr(CVMX_CIU_INTX_EN1_W1C(index), mask);
+	}
 }
 
 /*
- * Disable the irq on the current core for chips that have the EN*_W1{S,C}
- * registers.
+ * Write to the W1C bit in CVMX_CIU_INTX_SUM0 to clear the irq.
  */
-static void octeon_irq_ciu0_ack_v2(unsigned int irq)
-{
-	int index = cvmx_get_core_num() * 2;
-	u64 mask = 1ull << (irq - OCTEON_IRQ_WORKQ0);
-
-	switch (irq) {
-	case OCTEON_IRQ_GMX_DRP0:
-	case OCTEON_IRQ_GMX_DRP1:
-	case OCTEON_IRQ_IPD_DRP:
-	case OCTEON_IRQ_KEY_ZERO:
-	case OCTEON_IRQ_TIMER0:
-	case OCTEON_IRQ_TIMER1:
-	case OCTEON_IRQ_TIMER2:
-	case OCTEON_IRQ_TIMER3:
-		/*
-		 * CIU timer type interrupts must be acknoleged by
-		 * writing a '1' bit to their sum0 bit.
-		 */
+static void octeon_irq_ciu_ack(struct irq_data *data)
+{
+	u64 mask;
+	union octeon_ciu_chip_data cd;
+
+	cd.p = data->chip_data;
+	mask = 1ull << (cd.s.bit);
+
+	if (cd.s.line == 0) {
+		int index = cvmx_get_core_num() * 2;
 		cvmx_write_csr(CVMX_CIU_INTX_SUM0(index), mask);
-		break;
-	default:
-		break;
+	} else {
+		cvmx_write_csr(CVMX_CIU_INT_SUM1, mask);
 	}
-
-	cvmx_write_csr(CVMX_CIU_INTX_EN0_W1C(index), mask);
 }
 
 /*
- * Enable the irq on the current core for chips that have the EN*_W1{S,C}
+ * Disable the irq on the all cores for chips that have the EN*_W1{S,C}
  * registers.
  */
-static void octeon_irq_ciu0_eoi_mbox_v2(unsigned int irq)
+static void octeon_irq_ciu_disable_all_v2(struct irq_data *data)
 {
-	struct irq_desc *desc = irq_to_desc(irq);
-	int index = cvmx_get_core_num() * 2;
-	u64 mask = 1ull << (irq - OCTEON_IRQ_WORKQ0);
+	int cpu;
+	u64 mask;
+	union octeon_ciu_chip_data cd;
 
-	if (likely((desc->status & IRQ_DISABLED) == 0))
-		cvmx_write_csr(CVMX_CIU_INTX_EN0_W1S(index), mask);
+	wmb(); /* Make sure flag changes arrive before register updates. */
+
+	cd.p = data->chip_data;
+	mask = 1ull << (cd.s.bit);
+
+	if (cd.s.line == 0) {
+		for_each_online_cpu(cpu) {
+			int index = octeon_coreid_for_cpu(cpu) * 2;
+			clear_bit(cd.s.bit, &per_cpu(octeon_irq_ciu0_en_mirror, cpu));
+			cvmx_write_csr(CVMX_CIU_INTX_EN0_W1C(index), mask);
+		}
+	} else {
+		for_each_online_cpu(cpu) {
+			int index = octeon_coreid_for_cpu(cpu) * 2 + 1;
+			clear_bit(cd.s.bit, &per_cpu(octeon_irq_ciu1_en_mirror, cpu));
+			cvmx_write_csr(CVMX_CIU_INTX_EN1_W1C(index), mask);
+		}
+	}
 }
 
 /*
- * Disable the irq on the all cores for chips that have the EN*_W1{S,C}
+ * Enable the irq on the all cores for chips that have the EN*_W1{S,C}
  * registers.
  */
-static void octeon_irq_ciu0_disable_all_v2(unsigned int irq)
+static void octeon_irq_ciu_enable_all_v2(struct irq_data *data)
 {
-	u64 mask = 1ull << (irq - OCTEON_IRQ_WORKQ0);
-	int index;
 	int cpu;
-	for_each_online_cpu(cpu) {
-		index = octeon_coreid_for_cpu(cpu) * 2;
-		cvmx_write_csr(CVMX_CIU_INTX_EN0_W1C(index), mask);
+	u64 mask;
+	union octeon_ciu_chip_data cd;
+
+	cd.p = data->chip_data;
+	mask = 1ull << (cd.s.bit);
+
+	if (cd.s.line == 0) {
+		for_each_online_cpu(cpu) {
+			int index = octeon_coreid_for_cpu(cpu) * 2;
+			set_bit(cd.s.bit, &per_cpu(octeon_irq_ciu0_en_mirror, cpu));
+			cvmx_write_csr(CVMX_CIU_INTX_EN0_W1S(index), mask);
+		}
+	} else {
+		for_each_online_cpu(cpu) {
+			int index = octeon_coreid_for_cpu(cpu) * 2 + 1;
+			set_bit(cd.s.bit, &per_cpu(octeon_irq_ciu1_en_mirror, cpu));
+			cvmx_write_csr(CVMX_CIU_INTX_EN1_W1S(index), mask);
+		}
 	}
 }
 
+static void octeon_irq_cpu_online_mbox(struct irq_data *data)
+{
+	if (irqd_irq_disabled(data))
+		octeon_irq_ciu_enable_local(data);
+}
+
+static void octeon_irq_cpu_online_mbox_v2(struct irq_data *data)
+{
+	if (irqd_irq_disabled(data))
+		octeon_irq_ciu_enable_local_v2(data);
+}
+
+static void octeon_irq_cpu_offline_mbox(struct irq_data *data)
+{
+	if (irqd_irq_disabled(data))
+		octeon_irq_ciu_disable_local(data);
+}
+
+static void octeon_irq_cpu_offline_mbox_v2(struct irq_data *data)
+{
+	if (irqd_irq_disabled(data))
+		octeon_irq_ciu_disable_local_v2(data);
+}
+
 #ifdef CONFIG_SMP
-static int octeon_irq_ciu0_set_affinity(unsigned int irq, const struct cpumask *dest)
+
+static void octeon_irq_cpu_offline_ciu(struct irq_data *data)
+{
+	int cpu = smp_processor_id();
+	cpumask_t new_affinity;
+
+	if (!cpumask_test_cpu(cpu, data->affinity))
+		return;
+
+	if (cpumask_weight(data->affinity) > 1) {
+		/*
+		 * It has multi CPU affinity, just remove this CPU
+		 * from the affinity set.
+		 */
+		cpumask_copy(&new_affinity, data->affinity);
+		cpumask_clear_cpu(cpu, &new_affinity);
+	} else {
+		/* Otherwise, put it on lowest numbered online CPU. */
+		cpumask_clear(&new_affinity);
+		cpumask_set_cpu(cpumask_first(cpu_online_mask), &new_affinity);
+	}
+	__irq_set_affinity_locked(data, &new_affinity);
+}
+
+static int octeon_irq_ciu_set_affinity(struct irq_data *data,
+				       const struct cpumask *dest, bool force)
 {
 	int cpu;
-	struct irq_desc *desc = irq_to_desc(irq);
+	struct irq_desc *desc = irq_to_desc(data->irq);
 	int enable_one = (desc->status & IRQ_DISABLED) == 0;
 	unsigned long flags;
-	int bit = irq - OCTEON_IRQ_WORKQ0;	/* Bit 0-63 of EN0 */
+	union octeon_ciu_chip_data cd;
+
+	cd.p = data->chip_data;
 
 	/*
 	 * For non-v2 CIU, we will allow only single CPU affinity.
@@ -339,26 +585,40 @@ static int octeon_irq_ciu0_set_affinity(unsigned int irq, const struct cpumask *
 	if (cpumask_weight(dest) != 1)
 		return -EINVAL;
 
-	raw_spin_lock_irqsave(&octeon_irq_ciu0_lock, flags);
-	for_each_online_cpu(cpu) {
-		int coreid = octeon_coreid_for_cpu(cpu);
-		uint64_t en0 =
-			cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2));
-		if (cpumask_test_cpu(cpu, dest) && enable_one) {
-			enable_one = 0;
-			en0 |= 1ull << bit;
-		} else {
-			en0 &= ~(1ull << bit);
+	if (desc->status & IRQ_DISABLED)
+		return 0;
+
+	if (cd.s.line == 0) {
+		raw_spin_lock_irqsave(&octeon_irq_ciu0_lock, flags);
+		for_each_online_cpu(cpu) {
+			int coreid = octeon_coreid_for_cpu(cpu);
+			unsigned long *pen = &per_cpu(octeon_irq_ciu0_en_mirror, cpu);
+
+			if (cpumask_test_cpu(cpu, dest) && enable_one) {
+				enable_one = 0;
+				set_bit(cd.s.bit, pen);
+			} else {
+				clear_bit(cd.s.bit, pen);
+			}
+			cvmx_write_csr(CVMX_CIU_INTX_EN0(coreid * 2), *pen);
+		}
+		raw_spin_unlock_irqrestore(&octeon_irq_ciu0_lock, flags);
+	} else {
+		raw_spin_lock_irqsave(&octeon_irq_ciu1_lock, flags);
+		for_each_online_cpu(cpu) {
+			int coreid = octeon_coreid_for_cpu(cpu);
+			unsigned long *pen = &per_cpu(octeon_irq_ciu1_en_mirror, cpu);
+
+			if (cpumask_test_cpu(cpu, dest) && enable_one) {
+				enable_one = 0;
+				set_bit(cd.s.bit, pen);
+			} else {
+				clear_bit(cd.s.bit, pen);
+			}
+			cvmx_write_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1), *pen);
 		}
-		cvmx_write_csr(CVMX_CIU_INTX_EN0(coreid * 2), en0);
+		raw_spin_unlock_irqrestore(&octeon_irq_ciu1_lock, flags);
 	}
-	/*
-	 * We need to do a read after the last update to make sure all
-	 * of them are done.
-	 */
-	cvmx_read_csr(CVMX_CIU_INTX_EN0(cvmx_get_core_num() * 2));
-	raw_spin_unlock_irqrestore(&octeon_irq_ciu0_lock, flags);
-
 	return 0;
 }
 
@@ -366,22 +626,47 @@ static int octeon_irq_ciu0_set_affinity(unsigned int irq, const struct cpumask *
  * Set affinity for the irq for chips that have the EN*_W1{S,C}
  * registers.
  */
-static int octeon_irq_ciu0_set_affinity_v2(unsigned int irq,
-					   const struct cpumask *dest)
+static int octeon_irq_ciu_set_affinity_v2(struct irq_data *data,
+					  const struct cpumask *dest,
+					  bool force)
 {
 	int cpu;
-	int index;
-	struct irq_desc *desc = irq_to_desc(irq);
+	struct irq_desc *desc = irq_to_desc(data->irq);
 	int enable_one = (desc->status & IRQ_DISABLED) == 0;
-	u64 mask = 1ull << (irq - OCTEON_IRQ_WORKQ0);
-
-	for_each_online_cpu(cpu) {
-		index = octeon_coreid_for_cpu(cpu) * 2;
-		if (cpumask_test_cpu(cpu, dest) && enable_one) {
-			enable_one = 0;
-			cvmx_write_csr(CVMX_CIU_INTX_EN0_W1S(index), mask);
-		} else {
-			cvmx_write_csr(CVMX_CIU_INTX_EN0_W1C(index), mask);
+	u64 mask;
+	union octeon_ciu_chip_data cd;
+
+	if (desc->status & IRQ_DISABLED)
+		return 0;
+
+	cd.p = data->chip_data;
+	mask = 1ull << cd.s.bit;
+
+	if (cd.s.line == 0) {
+		for_each_online_cpu(cpu) {
+			unsigned long *pen = &per_cpu(octeon_irq_ciu0_en_mirror, cpu);
+			int index = octeon_coreid_for_cpu(cpu) * 2;
+			if (cpumask_test_cpu(cpu, dest) && enable_one) {
+				enable_one = 0;
+				set_bit(cd.s.bit, pen);
+				cvmx_write_csr(CVMX_CIU_INTX_EN0_W1S(index), mask);
+			} else {
+				clear_bit(cd.s.bit, pen);
+				cvmx_write_csr(CVMX_CIU_INTX_EN0_W1C(index), mask);
+			}
+		}
+	} else {
+		for_each_online_cpu(cpu) {
+			unsigned long *pen = &per_cpu(octeon_irq_ciu1_en_mirror, cpu);
+			int index = octeon_coreid_for_cpu(cpu) * 2 + 1;
+			if (cpumask_test_cpu(cpu, dest) && enable_one) {
+				enable_one = 0;
+				set_bit(cd.s.bit, pen);
+				cvmx_write_csr(CVMX_CIU_INTX_EN1_W1S(index), mask);
+			} else {
+				clear_bit(cd.s.bit, pen);
+				cvmx_write_csr(CVMX_CIU_INTX_EN1_W1C(index), mask);
+			}
 		}
 	}
 	return 0;
@@ -389,80 +674,101 @@ static int octeon_irq_ciu0_set_affinity_v2(unsigned int irq,
 #endif
 
 /*
+ * The v1 CIU code already masks things, so supply a dummy version to
+ * the core chip code.
+ */
+static void octeon_irq_dummy_mask(struct irq_data *data)
+{
+	return;
+}
+
+/*
  * Newer octeon chips have support for lockless CIU operation.
  */
-static struct irq_chip octeon_irq_chip_ciu0_v2 = {
-	.name = "CIU0",
-	.enable = octeon_irq_ciu0_enable_v2,
-	.disable = octeon_irq_ciu0_disable_all_v2,
-	.eoi = octeon_irq_ciu0_enable_v2,
+static struct irq_chip octeon_irq_chip_ciu_v2 = {
+	.name = "CIU",
+	.irq_enable = octeon_irq_ciu_enable_v2,
+	.irq_disable = octeon_irq_ciu_disable_all_v2,
+	.irq_mask = octeon_irq_ciu_disable_local_v2,
+	.irq_unmask = octeon_irq_ciu_enable_v2,
 #ifdef CONFIG_SMP
-	.set_affinity = octeon_irq_ciu0_set_affinity_v2,
+	.irq_set_affinity = octeon_irq_ciu_set_affinity_v2,
+	.irq_cpu_offline = octeon_irq_cpu_offline_ciu,
 #endif
 };
 
-static struct irq_chip octeon_irq_chip_ciu0 = {
-	.name = "CIU0",
-	.enable = octeon_irq_ciu0_enable,
-	.disable = octeon_irq_ciu0_disable,
-	.eoi = octeon_irq_ciu0_eoi,
+static struct irq_chip octeon_irq_chip_ciu_edge_v2 = {
+	.name = "CIU-E",
+	.irq_enable = octeon_irq_ciu_enable_v2,
+	.irq_disable = octeon_irq_ciu_disable_all_v2,
+	.irq_ack = octeon_irq_ciu_ack,
+	.irq_mask = octeon_irq_ciu_disable_local_v2,
+	.irq_unmask = octeon_irq_ciu_enable_v2,
 #ifdef CONFIG_SMP
-	.set_affinity = octeon_irq_ciu0_set_affinity,
+	.irq_set_affinity = octeon_irq_ciu_set_affinity_v2,
+	.irq_cpu_offline = octeon_irq_cpu_offline_ciu,
 #endif
 };
 
-/* The mbox versions don't do any affinity or round-robin. */
-static struct irq_chip octeon_irq_chip_ciu0_mbox_v2 = {
-	.name = "CIU0-M",
-	.enable = octeon_irq_ciu0_enable_mbox_v2,
-	.disable = octeon_irq_ciu0_disable,
-	.eoi = octeon_irq_ciu0_eoi_mbox_v2,
+static struct irq_chip octeon_irq_chip_ciu = {
+	.name = "CIU",
+	.irq_enable = octeon_irq_ciu_enable,
+	.irq_disable = octeon_irq_ciu_disable_all,
+	.irq_mask = octeon_irq_dummy_mask,
+#ifdef CONFIG_SMP
+	.irq_set_affinity = octeon_irq_ciu_set_affinity,
+	.irq_cpu_offline = octeon_irq_cpu_offline_ciu,
+#endif
 };
 
-static struct irq_chip octeon_irq_chip_ciu0_mbox = {
-	.name = "CIU0-M",
-	.enable = octeon_irq_ciu0_enable_mbox,
-	.disable = octeon_irq_ciu0_disable,
-	.eoi = octeon_irq_ciu0_eoi,
+static struct irq_chip octeon_irq_chip_ciu_edge = {
+	.name = "CIU-E",
+	.irq_enable = octeon_irq_ciu_enable,
+	.irq_disable = octeon_irq_ciu_disable_all,
+	.irq_mask = octeon_irq_dummy_mask,
+	.irq_ack = octeon_irq_ciu_ack,
+#ifdef CONFIG_SMP
+	.irq_set_affinity = octeon_irq_ciu_set_affinity,
+	.irq_cpu_offline = octeon_irq_cpu_offline_ciu,
+#endif
 };
 
-static void octeon_irq_ciu1_ack(unsigned int irq)
-{
-	/*
-	 * In order to avoid any locking accessing the CIU, we
-	 * acknowledge CIU interrupts by disabling all of them.  This
-	 * way we can use a per core register and avoid any out of
-	 * core locking requirements.  This has the side affect that
-	 * CIU interrupts can't be processed recursively.  We don't
-	 * need to disable IRQs to make these atomic since they are
-	 * already disabled earlier in the low level interrupt code.
-	 */
-	clear_c0_status(0x100 << 3);
-}
+/* The mbox versions don't do any affinity or round-robin. */
+static struct irq_chip octeon_irq_chip_ciu_mbox_v2 = {
+	.name = "CIU-M",
+	.irq_enable = octeon_irq_ciu_enable_all_v2,
+	.irq_disable = octeon_irq_ciu_disable_all_v2,
+	.irq_ack = octeon_irq_ciu_disable_local_v2,
+	.irq_eoi = octeon_irq_ciu_enable_local_v2,
+
+	.irq_cpu_online = octeon_irq_cpu_online_mbox_v2,
+	.irq_cpu_offline = octeon_irq_cpu_offline_mbox_v2,
+};
 
-static void octeon_irq_ciu1_eoi(unsigned int irq)
-{
-	/*
-	 * Enable all CIU interrupts again.  We don't need to disable
-	 * IRQs to make these atomic since they are already disabled
-	 * earlier in the low level interrupt code.
-	 */
-	set_c0_status(0x100 << 3);
-}
+static struct irq_chip octeon_irq_chip_ciu_mbox = {
+	.name = "CIU-M",
+	.irq_enable = octeon_irq_ciu_enable_all,
+	.irq_disable = octeon_irq_ciu_disable_all,
 
-static void octeon_irq_ciu1_enable(unsigned int irq)
+	.irq_cpu_online = octeon_irq_cpu_online_mbox,
+	.irq_cpu_offline = octeon_irq_cpu_offline_mbox,
+};
+
+/*
+ * Watchdog interrupts are special.  They are associated with a single
+ * core, so we hardwire the affinity to that core.
+ */
+static void octeon_irq_ciu_wd_enable(struct irq_data *data)
 {
-	struct irq_desc *desc = irq_to_desc(irq);
-	int coreid = next_coreid_for_irq(desc);
 	unsigned long flags;
-	uint64_t en1;
-	int bit = irq - OCTEON_IRQ_WDOG0;	/* Bit 0-63 of EN1 */
+	unsigned long *pen;
+	int coreid = data->irq - OCTEON_IRQ_WDOG0;	/* Bit 0-63 of EN1 */
+	int cpu = octeon_cpu_for_coreid(coreid);
 
 	raw_spin_lock_irqsave(&octeon_irq_ciu1_lock, flags);
-	en1 = cvmx_read_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1));
-	en1 |= 1ull << bit;
-	cvmx_write_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1), en1);
-	cvmx_read_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1));
+	pen = &per_cpu(octeon_irq_ciu1_en_mirror, cpu);
+	set_bit(coreid, pen);
+	cvmx_write_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1), *pen);
 	raw_spin_unlock_irqrestore(&octeon_irq_ciu1_lock, flags);
 }
 
@@ -470,286 +776,281 @@ static void octeon_irq_ciu1_enable(unsigned int irq)
  * Watchdog interrupts are special.  They are associated with a single
  * core, so we hardwire the affinity to that core.
  */
-static void octeon_irq_ciu1_wd_enable(unsigned int irq)
+static void octeon_irq_ciu1_wd_enable_v2(struct irq_data *data)
 {
-	unsigned long flags;
-	uint64_t en1;
-	int bit = irq - OCTEON_IRQ_WDOG0;	/* Bit 0-63 of EN1 */
-	int coreid = bit;
+	int coreid = data->irq - OCTEON_IRQ_WDOG0;
+	int cpu = octeon_cpu_for_coreid(coreid);
 
-	raw_spin_lock_irqsave(&octeon_irq_ciu1_lock, flags);
-	en1 = cvmx_read_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1));
-	en1 |= 1ull << bit;
-	cvmx_write_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1), en1);
-	cvmx_read_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1));
-	raw_spin_unlock_irqrestore(&octeon_irq_ciu1_lock, flags);
+