linux/arch/powerpc/sysdev/xive/common.c

/*
 * Copyright 2016,2017 IBM Corporation.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; either version
 * 2 of the License, or (at your option) any later version.
 */

#define pr_fmt(fmt) "xive: " fmt

#include <linux/types.h>
#include <linux/threads.h>
#include <linux/kernel.h>
#include <linux/irq.h>
#include <linux/debugfs.h>
#include <linux/smp.h>
#include <linux/interrupt.h>
#include <linux/seq_file.h>
#include <linux/init.h>
#include <linux/cpu.h>
#include <linux/of.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/msi.h>

#include <asm/prom.h>
#include <asm/io.h>
#include <asm/smp.h>
#include <asm/machdep.h>
#include <asm/irq.h>
#include <asm/errno.h>
#include <asm/xive.h>
#include <asm/xive-regs.h>
#include <asm/xmon.h>

#include "xive-internal.h"

#undef DEBUG_FLUSH
#undef DEBUG_ALL

#ifdef DEBUG_ALL
#define DBG_VERBOSE(fmt...)	pr_devel(fmt)
#else
#define DBG_VERBOSE(fmt...)	do { } while(0)
#endif

bool __xive_enabled;
EXPORT_SYMBOL_GPL(__xive_enabled);
bool xive_cmdline_disabled;

/* We use only one priority for now */
static u8 xive_irq_priority;

/* TIMA exported to KVM */
void __iomem *xive_tima;
EXPORT_SYMBOL_GPL(xive_tima);
u32 xive_tima_offset;

/* Backend ops */
static const struct xive_ops *xive_ops;

/* Our global interrupt domain */
static struct irq_domain *xive_irq_domain;

#ifdef CONFIG_SMP
/* The IPIs all use the same logical irq number */
static u32 xive_ipi_irq;
#endif

/* Xive state for each CPU */
static DEFINE_PER_CPU(struct xive_cpu *, xive_cpu);

/*
 * A "disabled" interrupt should never fire, to catch problems
 * we set its logical number to this
 */
#define XIVE_BAD_IRQ		0x7fffffff
#define XIVE_MAX_IRQ		(XIVE_BAD_IRQ - 1)

/* An invalid CPU target */
#define XIVE_INVALID_TARGET	(-1)

/*
 * Read the next entry in a queue, return its content if it's valid
 * or 0 if there is no new entry.
 *
 * The queue pointer is moved forward unless "just_peek" is set
 */
static u32 xive_read_eq(struct xive_q *q, bool just_peek)
{
	u32 cur;

	if (!q->qpage)
		return 0;
	cur = be32_to_cpup(q->qpage + q->idx);

	/* Check valid bit (31) vs current toggle polarity */
	if ((cur >> 31) == q->toggle)
		return 0;

	/* If consuming from the queue ... */
	if (!just_peek) {
		/* Next entry */
		q->idx = (q->idx + 1) & q->msk;

		/* Wrap around: flip valid toggle */
		if (q->idx == 0)
			q->toggle ^= 1;
	}
	/* Mask out the valid bit (31) */
	return cur & 0x7fffffff;
}

/*
 * Scans all the queue that may have interrupts in them
 * (based on "pending_prio") in priority order until an
 * interrupt is found or all the queues are empty.
 *
 * Then updates the CPPR (Current Processor Priority
 * Register) based on the most favored interrupt found
 * (0xff if none) and return what was found (0 if none).
 *
 * If just_peek is set, return the most favored pending
 * interrupt if any but don't update the queue pointers.
 *
 * Note: This function can operate generically on any number
 * of queues (up to 8). The current implementation of the XIVE
 * driver only uses a single queue however.
 *
 * Note2: This will also "flush" "the pending_count" of a queue
 * into the "count" when that queue is observed to be empty.
 * This is used to keep track of the amount of interrupts
 * targetting a queue. When an interrupt is moved away from
 * a queue, we only decrement that queue count once the queue
 * has been observed empty to avoid races.
 */
static u32 xive_scan_interrupts(struct xive_cpu *xc, bool just_peek)
{
	u32 irq = 0;
	u8 prio;

	/* Find highest pending priority */
	while (xc->pending_prio != 0) {
		struct xive_q *q;

		prio = ffs(xc->pending_prio) - 1;
		DBG_VERBOSE("scan_irq: trying prio %d\n", prio);

		/* Try to fetch */
		irq = xive_read_eq(&xc->queue[prio], just_peek);

		/* Found something ? That's it */
		if (irq)
			break;

		/* Clear pending bits */
		xc->pending_prio &= ~(1 << prio);

		/*
		 * Check if the queue count needs adjusting due to
		 * interrupts being moved away. See description of
		 * xive_dec_target_count()
		 */
		q = &xc->queue[prio];
		if (atomic_read(&q->pending_count)) {
			int p = atomic_xchg(&q->pending_count, 0);
			if (p) {
				WARN_ON(p > atomic_read(&q->count));
				atomic_sub(p, &q->count);
			}
		}
	}

	/* If nothing was found, set CPPR to 0xff */
	if (irq == 0)
		prio = 0xff;

	/* Update HW CPPR to match if necessary */
	if (prio != xc->cppr) {
		DBG_VERBOSE("scan_irq: adjusting CPPR to %d\n", prio);
		xc->cppr = prio;
		out_8(xive_tima + xive_tima_offset + TM_CPPR, prio);
	}

	return irq;
}

/*
 * This is used to perform the magic loads from an ESB
 * described in xive.h
 */
static notrace u8 xive_poke_esb(struct xive_irq_data *xd, u32 offset)
{
	u64 val;

	/* Handle HW errata */
	if (xd->flags & XIVE_IRQ_FLAG_SHIFT_BUG)
		offset |= offset << 4;

	val = in_be64(xd->eoi_mmio + offset);

	return (u8)val;
}

#ifdef CONFIG_XMON
static notrace void xive_dump_eq(const char *name, struct xive_q *q)
{
	u32 i0, i1, idx;

	if (!q->qpage)
		return;
	idx = q->idx;
	i0 = be32_to_cpup(q->qpage + idx);
	idx = (idx + 1) & q->msk;
	i1 = be32_to_cpup(q->qpage + idx);
	xmon_printf("  %s Q T=%d %08x %08x ...\n", name,
		    q->toggle, i0, i1);
}

notrace void xmon_xive_do_dump(int cpu)
{
	struct xive_cpu *xc = per_cpu(xive_cpu, cpu);

	xmon_printf("XIVE state for CPU %d:\n", cpu);
	xmon_printf("  pp=%02x cppr=%02x\n", xc->pending_prio, xc->cppr);
	xive_dump_eq("IRQ", &xc->queue[xive_irq_priority]);
#ifdef CONFIG_SMP
	{
		u64 val = xive_poke_esb(&xc->ipi_data, XIVE_ESB_GET);
		xmon_printf("  IPI state: %x:%c%c\n", xc->hw_ipi,
			val & XIVE_ESB_VAL_P ? 'P' : 'p',
			val & XIVE_ESB_VAL_P ? 'Q' : 'q');
	}
#endif
}
#endif /* CONFIG_XMON */

static unsigned int xive_get_irq(void)
{
	struct xive_cpu *xc = __this_cpu_read(xive_cpu);
	u32 irq;

	/*
	 * This can be called either as a result of a HW interrupt or
	 * as a "replay" because EOI decided there was still something
	 * in one of the queues.
	 *
	 * First we perform an ACK cycle in order to update our mask
	 * of pending priorities. This will also have the effect of
	 * updating the CPPR to the most favored pending interrupts.
	 *
	 * In the future, if we have a way to differenciate a first
	 * entry (on HW interrupt) from a replay triggered by EOI,
	 * we could skip this on replays unless we soft-mask tells us
	 * that a new HW interrupt occurred.
	 */
	xive_ops->update_pending(xc);

	DBG_VERBOSE("get_irq: pending=%02x\n", xc->pending_prio);

	/* Scan our queue(s) for interrupts */
	irq = xive_scan_interrupts(xc, false);

	DBG_VERBOSE("get_irq: got irq 0x%x, new pending=0x%02x\n",
	    irq, xc->pending_prio);

	/* Return pending interrupt if any */
	if (irq == XIVE_BAD_IRQ)
		return 0;
	return irq;
}

/*
 * After EOI'ing an interrupt, we need to re-check the queue
 * to see if another interrupt is pending since multiple
 * interrupts can coalesce into a single notification to the
 * CPU.
 *
 * If we find that there is indeed more in there, we call
 * force_external_irq_replay() to make Linux synthetize an
 * external interrupt on the next call to local_irq_restore().
 */
static void xive_do_queue_eoi(struct xive_cpu *xc)
{
	if (xive_scan_interrupts(xc, true) != 0) {
		DBG_VERBOSE("eoi: pending=0x%02x\n", xc->pending_prio);
		force_external_irq_replay();
	}
}

/*
 * EOI an interrupt at the source. There are several methods
 * to do this depending on the HW version and source type
 */
void xive_do_source_eoi(u32 hw_irq, struct xive_irq_data *xd)
{
	/* If the XIVE supports the new "store EOI facility, use it */
	if (xd->flags & XIVE_IRQ_FLAG_STORE_EOI)
		out_be64(xd->eoi_mmio + XIVE_ESB_STORE_EOI, 0);
	else if (hw_irq && xd->flags & XIVE_IRQ_FLAG_EOI_FW) {
		/*
		 * The FW told us to call it. This happens for some
		 * interrupt sources that need additional HW whacking
		 * beyond the ESB manipulation. For example LPC interrupts
		 * on P9 DD1.0 need a latch to be clared in the LPC bridge
		 * itself. The Firmware will take care of it.
		 */
		if (WARN_ON_ONCE(!xive_ops->eoi))
			return;
		xive_ops->eoi(hw_irq);
	} else {
		u8 eoi_val;

		/*
		 * Otherwise for EOI, we use the special MMIO that does
		 * a clear of both P and Q and returns the old Q,
		 * except for LSIs where we use the "EOI cycle" special
		 * load.
		 *
		 * This allows us to then do a re-trigger if Q was set
		 * rather than synthesizing an interrupt in software
		 *
		 * For LSIs, using the HW EOI cycle works around a problem
		 * on P9 DD1 PHBs where the other ESB accesses don't work
		 * properly.
		 */
		if (xd->flags & XIVE_IRQ_FLAG_LSI)
			in_be64(xd->eoi_mmio);
		else {
			eoi_val = xive_poke_esb(xd, XIVE_ESB_SET_PQ_00);
			DBG_VERBOSE("eoi_val=%x\n", offset, eoi_val);

			/* Re-trigger if needed */
			if ((eoi_val & XIVE_ESB_VAL_Q) && xd->trig_mmio)
				out_be64(xd->trig_mmio, 0);
		}
	}
}

/* irq_chip eoi callback */
static void xive_irq_eoi(struct irq_data *d)
{
	struct xive_irq_data *xd = irq_data_get_irq_handler_data(d);
	struct xive_cpu *xc = __this_cpu_read(xive_cpu);

	DBG_VERBOSE("eoi_irq: irq=%d [0x%lx] pending=%02x\n",
		    d->irq, irqd_to_hwirq(d), xc->pending_prio);

	/*
	 * EOI the source if it hasn't been disabled and hasn't
	 * been passed-through to a KVM guest
	 */
	if (!irqd_irq_disabled(d) && !irqd_is_forwarded_to_vcpu(d))
		xive_do_source_eoi(irqd_to_hwirq(d), xd);

	/*
	 * Clear saved_p to indicate that it's no longer occupying
	 * a queue slot on the target queue
	 */
	xd->saved_p = false;

	/* Check for more work in the queue */
	xive_do_queue_eoi(xc);
}

/*
 * Helper used to mask and unmask an interrupt source. This
 * is only called for normal interrupts that do not require
 * masking/unmasking via firmware.
 */
static void xive_do_source_set_mask(struct xive_irq_data *xd,
				    bool mask)
{
	u64 val;

	/*
	 * If the interrupt had P set, it may be in a queue.
	 *
	 * We need to make sure we don't re-enable it until it
	 * has been fetched from that queue and EOId. We keep
	 * a copy of that P state and use it to restore the
	 * ESB accordingly on unmask.
	 */
	if (mask) {
		val = xive_poke_esb(xd, XIVE_ESB_SET_PQ_01);
		xd->saved_p = !!(val & XIVE_ESB_VAL_P);
	} else if (xd->saved_p)
		xive_poke_esb(xd, XIVE_ESB_SET_PQ_10);
	else
		xive_poke_esb(xd, XIVE_ESB_SET_PQ_00);
}

/*
 * Try to chose "cpu" as a new interrupt target. Increments
 * the queue accounting for that target if it's not already
 * full.
 */
static bool xive_try_pick_target(int cpu)
{
	struct xive_cpu *xc = per_cpu(xive_cpu, cpu);
	struct xive_q *q = &xc->queue[xive_irq_priority];
	int max;

	/*
	 * Calculate max number of interrupts in that queue.
	 *
	 * We leave a gap of 1 just in case...
	 */
	max = (q->msk + 1) - 1;
	return !!atomic_add_unless(&q->count, 1, max);
}

/*
 * Un-account an interrupt for a target CPU. We don't directly
 * decrement q->count since the interrupt might still be present
 * in the queue.
 *
 * Instead increment a separate counter "pending_count" which
 * will be substracted from "count" later when that CPU observes
 * the queue to be empty.
 */
static void xive_dec_target_count(int cpu)
{
	struct xive_cpu *xc = per_cpu(xive_cpu, cpu);
	struct xive_q *q = &xc->queue[xive_irq_priority];

	if (unlikely(WARN_ON(cpu < 0 || !xc))) {
		pr_err("%s: cpu=%d xc=%p\n", __func__, cpu, xc);
		return;
	}

	/*
	 * We increment the "pending count" which will be used
	 * to decrement the target queue count whenever it's next
	 * processed and found empty. This ensure that we don't
	 * decrement while we still have the interrupt there
	 * occupying a slot.
	 */
	atomic_inc(&q->pending_count);
}

/* Find a tentative CPU target in a CPU mask */
static int xive_find_target_in_mask(const struct cpumask *mask,
				    unsigned int fuzz)
{
	int cpu, first, num, i;

	/* Pick up a starting point CPU in the mask based on  fuzz */
	num = min_t(int, cpumask_weight(mask), nr_cpu_ids);
	first = fuzz % num;

	/* Locate it */
	cpu = cpumask_first(mask);
	for (i = 0; i < first && cpu < nr_cpu_ids; i++)
		cpu = cpumask_next(cpu, mask);

	/* Sanity check */
	if (WARN_ON(cpu >= nr_cpu_ids))
		cpu = cpumask_first(cpu_online_mask);

	/* Remember first one to handle wrap-around */
	first = cpu;

	/*
	 * Now go through the entire mask until we find a valid
	 * target.
	 */
	for (;;) {
		/*
		 * We re-check online as the fallback case passes us
		 * an untested affinity mask
		 */
		if (cpu_online(cpu) && xive_try_pick_target(cpu))
			return cpu;
		cpu = cpumask_next(cpu, mask);
		if (cpu == first)
			break;
		/* Wrap around */
		if (cpu >= nr_cpu_ids)
			cpu = cpumask_first(mask);
	}
	return -1;
}

/*
 * Pick a target CPU for an interrupt. This is done at
 * startup or if the affinity is changed in a way that
 * invalidates the current target.
 */
static int xive_pick_irq_target(struct irq_data *d,
				const struct cpumask *affinity)
{
	static unsigned int fuzz;
	struct xive_irq_data *xd = irq_data_get_irq_handler_data(d);
	cpumask_var_t mask;
	int cpu = -1;

	/*
	 * If we have chip IDs, first we try to build a mask of
	 * CPUs matching the CPU and find a target in there
	 */
	if (xd->src_chip != XIVE_INVALID_CHIP_ID &&
		zalloc_cpumask_var(&mask, GFP_ATOMIC)) {
		/* Build a mask of matching chip IDs */
		for_each_cpu_and(cpu, affinity, cpu_online_mask) {
			struct xive_cpu *xc = per_cpu(xive_cpu, cpu);
			if (xc->chip_id == xd->src_chip)
				cpumask_set_cpu(cpu, mask);
		}
		/* Try to find a target */
		if (cpumask_empty(mask))
			cpu = -1;
		else
			cpu = xive_find_target_in_mask(mask, fuzz++);
		free_cpumask_var(mask);
		if (cpu >= 0)
			return cpu;
		fuzz--;
	}

	/* No chip IDs, fallback to using the affinity mask */
	return xive_find_target_in_mask(affinity, fuzz++);
}

static unsigned int xive_irq_startup(struct irq_data *d)
{
	struct xive_irq_data *xd = irq_data_get_irq_handler_data(d);
	unsigned int hw_irq = (unsigned int)irqd_to_hwirq(d);
	int target, rc;

	pr_devel("xive_irq_startup: irq %d [0x%x] data @%p\n",
		 d->irq, hw_irq, d);

#ifdef CONFIG_PCI_MSI
	/*
	 * The generic MSI code returns with the interrupt disabled on the
	 * card, using the MSI mask bits. Firmware doesn't appear to unmask
	 * at that level, so we do it here by hand.
	 */
	if (irq_data_get_msi_desc(d))
		pci_msi_unmask_irq(d);
#endif

	/* Pick a target */
	target = xive_pick_irq_target(d, irq_data_get_affinity_mask(d));
	if (target == XIVE_INVALID_TARGET) {
		/* Try again breaking affinity */
		target = xive_pick_irq_target(d, cpu_online_mask);
		if (target == XIVE_INVALID_TARGET)
			return -ENXIO;
		pr_warn("irq %d started with broken affinity\n", d->irq);
	}

	/* Sanity check */
	if (WARN_ON(target == XIVE_INVALID_TARGET ||
		    target >= nr_cpu_ids))
		target = smp_processor_id();

	xd->target = target;

	/*
	 * Configure the logical number to be the Linux IRQ number
	 * and set the target queue
	 */
	rc = xive_ops->configure_irq(hw_irq,
				     get_hard_smp_processor_id(target),
				     xive_irq_priority, d->irq);
	if (rc)
		return rc;

	/* Unmask the ESB */
	xive_do_source_set_mask(xd, false);

	return 0;
}

static void xive_irq_shutdown(struct irq_data *d)
{
	struct xive_irq_data *xd = irq_data_get_irq_handler_data(d);
	unsigned int hw_irq = (unsigned int)irqd_to_hwirq(d);

	pr_devel("xive_irq_shutdown: irq %d [0x%x] data @%p\n",
		 d->irq, hw_irq, d);

	if (WARN_ON(xd->target == XIVE_INVALID_TARGET))
		return;

	/* Mask the interrupt at the source */
	xive_do_source_set_mask(xd, true);

	/*
	 * The above may have set saved_p. We clear it otherwise it
	 * will prevent re-enabling later on. It is ok to forget the
	 * fact that the interrupt might be in a queue because we are
	 * accounting that already in xive_dec_target_count() and will
	 * be re-routing it to a new queue with proper accounting when
	 * it's started up again
	 */
	xd->saved_p = false;

	/*
	 * Mask the interrupt in HW in the IVT/EAS and set the number
	 * to be the "bad" IRQ number
	 */
	xive_ops->configure_irq(hw_irq,
				get_hard_smp_processor_id(xd->target),
				0xff, XIVE_BAD_IRQ);

	xive_dec_target_count(xd->target);
	xd->target = XIVE_INVALID_TARGET;
}

static void xive_irq_unmask(struct irq_data *d)
{
	struct xive_irq_data *xd = irq_data_get_irq_handler_data(d);

	pr_devel("xive_irq_unmask: irq %d data @%p\n", d->irq, xd);

	/*
	 * This is a workaround for PCI LSI problems on P9, for
	 * these, we call FW to set the mask. The problems might
	 * be fixed by P9 DD2.0, if that is the case, firmware
	 * will no longer set that flag.
	 */
	if (xd->flags & XIVE_IRQ_FLAG_MASK_FW) {
		unsigned int hw_irq = (unsigned int)irqd_to_hwirq(d);
		xive_ops->configure_irq(hw_irq,
					get_hard_smp_processor_id(xd->target),
					xive_irq_priority, d->irq);
		return;
	}

	xive_do_source_set_mask(xd, false);
}

static void xive_irq_mask(struct irq_data *d)
{
	struct xive_irq_data *xd = irq_data_get_irq_handler_data(d);

	pr_devel("xive_irq_mask: irq %d data @%p\n", d->irq, xd);

	/*
	 * This is a workaround for PCI LSI problems on P9, for
	 * these, we call OPAL to set the mask. The problems might
	 * be fixed by P9 DD2.0, if that is the case, firmware
	 * will no longer set that flag.
	 */
	if (xd->flags & XIVE_IRQ_FLAG_MASK_FW) {
		unsigned int hw_irq = (unsigned int)irqd_to_hwirq(d);
		xive_ops->configure_irq(hw_irq,
					get_hard_smp_processor_id(xd->target),
					0xff, d->irq);
		return;
	}

	xive_do_source_set_mask(xd, true);
}

static int xive_irq_set_affinity(struct irq_data *d,
				 const struct cpumask *cpumask,
				 bool force)
{
	struct xive_irq_data *xd = irq_data_get_irq_handler_data(d);
	unsigned int hw_irq = (unsigned int)irqd_to_hwirq(d);
	u32 target, old_target;
	int rc = 0;

	pr_devel("xive_irq_set_affinity: irq %d\n", d->irq);

	/* Is this valid ? */
	if (cpumask_any_and(cpumask, cpu_online_mask) >= nr_cpu_ids)
		return -EINVAL;

	/* Don't do anything if the interrupt isn't started */
	if (!irqd_is_started(d))
		return IRQ_SET_MASK_OK;

	/*
	 * If existing target is already in the new mask, and is
	 * online then do nothing.
	 */
	if (xd->target != XIVE_INVALID_TARGET &&
	    cpu_online(xd->target) &&
	    cpumask_test_cpu(xd->target, cpumask))
		return IRQ_SET_MASK_OK;

	/* Pick a new target */
	target = xive_pick_irq_target(d, cpumask);

	/* No target found */
	if (target == XIVE_INVALID_TARGET)
		return -ENXIO;

	/* Sanity check */
	if (WARN_ON(target >= nr_cpu_ids))
		target = smp_processor_id();

	old_target = xd->target;

	/*
	 * Only configure the irq if it's not currently passed-through to
	 * a KVM guest
	 */
	if (!irqd_is_forwarded_to_vcpu(d))
		rc = xive_ops->configure_irq(hw_irq,
					     get_hard_smp_processor_id(target),
					     xive_irq_priority, d->irq);
	if (rc < 0) {
		pr_err("Error %d reconfiguring irq %d\n", rc, d->irq);
		return rc;
	}

	pr_devel("  target: 0x%x\n", target);
	xd->target = target;

	/* Give up previous target */
	if (old_target != XIVE_INVALID_TARGET)
	    xive_dec_target_count(old_target);

	return IRQ_SET_MASK_OK;
}

static int xive_irq_set_type(struct irq_data *d, unsigned int flow_type)
{
	struct xive_irq_data *xd = irq_data_get_irq_handler_data(d);

	/*
	 * We only support these. This has really no effect other than setting
	 * the corresponding descriptor bits mind you but those will in turn
	 * affect the resend function when re-enabling an edge interrupt.
	 *
	 * Set set the default to edge as explained in map().
	 */
	if (flow_type == IRQ_TYPE_DEFAULT || flow_type == IRQ_TYPE_NONE)
		flow_type = IRQ_TYPE_EDGE_RISING;

	if (flow_type != IRQ_TYPE_EDGE_RISING &&
	    flow_type != IRQ_TYPE_LEVEL_LOW)
		return -EINVAL;

	irqd_set_trigger_type(d, flow_type);

	/*
	 * Double check it matches what the FW thinks
	 *
	 * NOTE: We don't know yet if the PAPR interface will provide
	 * the LSI vs MSI information apart from the device-tree so
	 * this check might have to move into an optional backend call
	 * that is specific to the native backend
	 */
	if ((flow_type == IRQ_TYPE_LEVEL_LOW) !=
	    !!(xd->flags & XIVE_IRQ_FLAG_LSI)) {
		pr_warn("Interrupt %d (HW 0x%x) type mismatch, Linux says %s, FW says %s\n",
			d->irq, (u32)irqd_to_hwirq(d),
			(flow_type == IRQ_TYPE_LEVEL_LOW) ? "Level" : "Edge",
			(xd->flags & XIVE_IRQ_FLAG_LSI) ? "Level" : "Edge");
	}

	return IRQ_SET_MASK_OK_NOCOPY;
}

static int xive_irq_retrigger(struct irq_data *d)
{
	struct xive_irq_data *xd = irq_data_get_irq_handler_data(d);

	/* This should be only for MSIs */
	if (WARN_ON(xd->flags & XIVE_IRQ_FLAG_LSI))
		return 0;

	/*
	 * To perform a retrigger, we first set the PQ bits to
	 * 11, then perform an EOI.
	 */
	xive_poke_esb(xd, XIVE_ESB_SET_PQ_11);

	/*
	 * Note: We pass "0" to the hw_irq argument in order to
	 * avoid calling into the backend EOI code which we don't
	 * want to do in the case of a re-trigger. Backends typically
	 * only do EOI for LSIs anyway.
	 */
	xive_do_source_eoi(0, xd);

	return 1;
}

static int xive_irq_set_vcpu_affinity(struct irq_data *d, void *state)
{
	struct xive_irq_data *xd = irq_data_get_irq_handler_data(d);
	unsigned int hw_irq = (unsigned int)irqd_to_hwirq(d);
	int rc;
	u8 pq;

	/*
	 * We only support this on interrupts that do not require
	 * firmware calls for masking and unmasking
	 */
	if (xd->flags & XIVE_IRQ_FLAG_MASK_FW)
		return -EIO;

	/*
	 * This is called by KVM with state non-NULL for enabling
	 * pass-through or NULL for disabling it
	 */
	if (state) {
		irqd_set_forwarded_to_vcpu(d);

		/* Set it to PQ=10 state to prevent further sends */
		pq = xive_poke_esb(xd, XIVE_ESB_SET_PQ_10);

		/* No target ? nothing to do */
		if (xd->target == XIVE_INVALID_TARGET) {
			/*
			 * An untargetted interrupt should have been
			 * also masked at the source
			 */
			WARN_ON(pq & 2);

			return 0;
		}

		/*
		 * If P was set, adjust state to PQ=11 to indicate
		 * that a resend is needed for the interrupt to reach
		 * the guest. Also remember the value of P.
		 *
		 * This also tells us that it's in flight to a host queue
		 * or has already been fetched but hasn't been EOIed yet
		 * by the host. This it's potentially using up a host
		 * queue slot. This is important to know because as long
		 * as this is the case, we must not hard-unmask it when
		 * "returning" that interrupt to the host.
		 *
		 * This saved_p is cleared by the host EOI, when we know
		 * for sure the queue slot is no longer in use.
		 */
		if (pq & 2) {
			pq = xive_poke_esb(xd, XIVE_ESB_SET_PQ_11);
			xd->saved_p = true;

			/*
			 * Sync the XIVE source HW to ensure the interrupt
			 * has gone through the EAS before we change its
			 * target to the guest. That should guarantee us
			 * that we *will* eventually get an EOI for it on
			 * the host. Otherwise there would be a small window
			 * for P to be seen here but the interrupt going
			 * to the guest queue.
			 */
			if (xive_ops->sync_source)
				xive_ops->sync_source(hw_irq);
		} else
			xd->saved_p = false;
	} else {
		irqd_clr_forwarded_to_vcpu(d);

		/* No host target ? hard mask and return */
		if (xd->target == XIVE_INVALID_TARGET) {
			xive_do_source_set_mask(xd, true);
			return 0;
		}

		/*
		 * Sync the XIVE source HW to ensure the interrupt
		 * has gone through the EAS before we change its
		 * target to the host.
		 */
		if (xive_ops->sync_source)
			xive_ops->sync_source(hw_irq);

		/*
		 * By convention we are called with the interrupt in
		 * a PQ=10 or PQ=11 state, ie, it won't fire and will
		 * have latched in Q whether there's a pending HW
		 * interrupt or not.
		 *
		 * First reconfigure the target.
		 */
		rc = xive_ops->configure_irq(hw_irq,
					     get_hard_smp_processor_id(xd->target),
					     xive_irq_priority, d->irq);
		if (rc)
			return rc;

		/*
		 * Then if saved_p is not set, effectively re-enable the
		 * interrupt with an EOI. If it is set, we know there is
		 * still a message in a host queue somewhere that will be
		 * EOId eventually.
		 *
		 * Note: We don't check irqd_irq_disabled(). Effectively,
		 * we *will* let the irq get through even if masked if the
		 * HW is still firing it in order to deal with the whole
		 * saved_p business properly. If the interrupt triggers
		 * while masked, the generic code will re-mask it anyway.
		 */
		if (!xd->saved_p)
			xive_do_source_eoi(hw_irq, xd);

	}
	return 0;
}

static struct irq_chip xive_irq_chip = {
	.name = "XIVE-IRQ",
	.irq_startup = xive_irq_startup,
	.irq_shutdown = xive_irq_shutdown,
	.irq_eoi = xive_irq_eoi,
	.irq_mask = xive_irq_mask,
	.irq_unmask = xive_irq_unmask,
	.irq_set_affinity = xive_irq_set_affinity,
	.irq_set_type = xive_irq_set_type,
	.irq_retrigger = xive_irq_retrigger,
	.irq_set_vcpu_affinity = xive_irq_set_vcpu_affinity,
};

bool is_xive_irq(struct irq_chip *chip)
{
	return chip == &xive_irq_chip;
}
EXPORT_SYMBOL_GPL(is_xive_irq);

void xive_cleanup_irq_data(struct xive_irq_data *xd)
{
	if (xd->eoi_mmio) {
		iounmap(xd->eoi_mmio);
		if (xd->eoi_mmio == xd->trig_mmio)
			xd->trig_mmio = NULL;
		xd->eoi_mmio = NULL;
	}
	if (xd->trig_mmio) {
		iounmap(xd->trig_mmio);
		xd->trig_mmio = NULL;
	}
}
EXPORT_SYMBOL_GPL(xive_cleanup_irq_data);

static int xive_irq_alloc_data(unsigned int virq, irq_hw_number_t hw)
{
	struct xive_irq_data *xd;
	int rc;

	xd = kzalloc(sizeof(struct xive_irq_data), GFP_KERNEL);
	if (!xd)
		return -ENOMEM;
	rc = xive_ops->populate_irq_data(hw, xd);
	if (rc) {
		kfree(xd);
		return rc;
	}
	xd->target = XIVE_INVALID_TARGET;
	irq_set_handler_data(virq, xd);

	return 0;
}

static void xive_irq_free_data(unsigned int virq)
{
	struct xive_irq_data *xd = irq_get_handler_data(virq);

	if (!xd)
		return;
	irq_set_handler_data(virq, NULL);
	xive_cleanup_irq_data(xd);
	kfree(xd);
}

#ifdef CONFIG_SMP

static void xive_cause_ipi(int cpu)
{
	struct xive_cpu *xc;
	struct xive_irq_data *xd;

	xc = per_cpu(xive_cpu, cpu);

	DBG_VERBOSE("IPI CPU %d -> %d (HW IRQ 0x%x)\n",
		    smp_processor_id(), cpu, xc->hw_ipi);

	xd = &xc->ipi_data;
	if (WARN_ON(!xd->trig_mmio))
		return;
	out_be64(xd->trig_mmio, 0);
}

static irqreturn_t xive_muxed_ipi_action(int irq, void *dev_id)
{
	return smp_ipi_demux();
}

static void xive_ipi_eoi(struct irq_data *d)
{
	struct xive_cpu *xc = __this_cpu_read(xive_cpu);

	/* Handle possible race with unplug and drop stale IPIs */
	if (!xc)
		return;
	xive_do_source_eoi(xc->hw_ipi, &xc->ipi_data);
	xive_do_queue_eoi(xc);
}

static void xive_ipi_do_nothing(struct irq_data *d)
{
	/*
	 * Nothing to do, we never mask/unmask IPIs, but the callback
	 * has to exist for the struct irq_chip.
	 */
}

static struct irq_chip xive_ipi_chip = {
	.name = "XIVE-IPI",
	.irq_eoi = xive_ipi_eoi,
	.irq_mask = xive_ipi_do_nothing,
	.irq_unmask = xive_ipi_do_nothing,
};

static void __init xive_request_ipi(void)
{
	unsigned int virq;

	/*
	 * Initialization failed, move on, we might manage to
	 * reach the point where we display our errors before
	 * the system falls appart
	 */
	if (!xive_irq_domain)
		return;

	/* Initialize it */
	virq = irq_create_mapping(xive_irq_domain, 0);
	xive_ipi_irq = virq;

	WARN_ON(request_irq(virq, xive_muxed_ipi_action,
			    IRQF_PERCPU | IRQF_NO_THREAD, "IPI", NULL));
}

static int xive_setup_cpu_ipi(unsigned int cpu)
{
	struct xive_cpu *xc;
	int rc;

	pr_debug("Setting up IPI for CPU %d\n", cpu);

	xc = per_cpu(xive_cpu, cpu);

	/* Check if we are already setup */
	if (xc->hw_ipi != 0)
		return 0;

	/* Grab an IPI from the backend, this will populate xc->hw_ipi */
	if (xive_ops->get_ipi(cpu, xc))
		return -EIO;

	/*
	 * Populate the IRQ data in the xive_cpu structure and
	 * configure the HW / enable the IPIs.
	 */
	rc = xive_ops->populate_irq_data(xc->hw_ipi, &xc->ipi_data);
	if (rc) {
		pr_err("Failed to populate IPI data on CPU %d\n", cpu);
		return -EIO;
	}
	rc = xive_ops->configure_irq(xc->hw_ipi,
				     get_hard_smp_processor_id(cpu),
				     xive_irq_priority, xive_ipi_irq);
	if (rc) {
		pr_err("Failed to map IPI CPU %d\n", cpu);
		return -EIO;
	}
	pr_devel("CPU %d HW IPI %x, virq %d, trig_mmio=%p\n", cpu,
	    xc->hw_ipi, xive_ipi_irq, xc->ipi_data.trig_mmio);

	/* Unmask it */
	xive_do_source_set_mask(&xc->ipi_data, false);

	return 0;
}

static void xive_cleanup_cpu_ipi(unsigned int cpu, struct xive_cpu *xc)
{
	/* Disable the IPI and free the IRQ data */

	/* Already cleaned up ? */
	if (xc->hw_ipi == 0)
		return;

	/* Mask the IPI */
	xive_do_source_set_mask(&xc->ipi_data, true);

	/*
	 * Note: We don't call xive_cleanup_irq_data() to free
	 * the mappings as this is called from an IPI on kexec
	 * which is not a safe environment to call iounmap()
	 */

	/* Deconfigure/mask in the backend */
	xive_ops->configure_irq(xc->hw_ipi, hard_smp_processor_id(),
				0xff, xive_ipi_irq);

	/* Free the IPIs in the backend */
	xive_ops->put_ipi(cpu, xc);
}

void __init xive_smp_probe(void)
{
	smp_ops->cause_ipi = xive_cause_ipi;

	/* Register the IPI */
	xive_request_ipi();

	/* Allocate and setup IPI for the boot CPU */
	xive_setup_cpu_ipi(smp_processor_id());
}

#endif /* CONFIG_SMP */

static int xive_irq_domain_map(struct irq_domain *h, unsigned int virq,
			       irq_hw_number_t hw)
{
	int rc;

	/*
	 * Mark interrupts as edge sensitive by default so that resend
	 * actually works. Will fix that up below if needed.
	 */
	irq_clear_status_flags(virq, IRQ_LEVEL);

#ifdef CONFIG_SMP
	/* IPIs are special and come up with HW number 0 */
	if (hw == 0) {
		/*
		 * IPIs are marked per-cpu. We use separate HW interrupts under
		 * the hood but associated with the same "linux" interrupt
		 */
		irq_set_chip_and_handler(virq, &xive_ipi_chip,
					 handle_percpu_irq);
		return 0;
	}
#endif

	rc = xive_irq_alloc_data(virq, hw);
	if (rc)
		return rc;

	irq_set_chip_and_handler(virq, &xive_irq_chip, handle_fasteoi_irq);

	return 0;
}

static void xive_irq_domain_unmap(struct irq_domain *d, unsigned int virq)
{
	struct irq_data *data = irq_get_irq_data(virq);
	unsigned int hw_irq;

	/* XXX Assign BAD number */
	if (!data)
		return;
	hw_irq = (unsigned int)irqd_to_hwirq(data);
	if (hw_irq)
		xive_irq_free_data(virq);
}

static int xive_irq_domain_xlate(struct irq_domain *h, struct device_node *ct,
				 const u32 *intspec, unsigned int intsize,
				 irq_hw_number_t *out_hwirq, unsigned int *out_flags)

{
	*out_hwirq = intspec[0];

	/*
	 * If intsize is at least 2, we look for the type in the second cell,
	 * we assume the LSB indicates a level interrupt.
	 */
	if (intsize > 1) {
		if (intspec[1] & 1)
			*out_flags = IRQ_TYPE_LEVEL_LOW;
		else
			*out_flags = IRQ_TYPE_EDGE_RISING;
	} else
		*out_flags = IRQ_TYPE_LEVEL_LOW;

	return 0;
}

static int xive_irq_domain_match(struct irq_domain *h, struct device_node *node,
				 enum irq_domain_bus_token bus_token)
{
	return xive_ops->match(node);
}

static const struct irq_domain_ops xive_irq_domain_ops = {
	.match = xive_irq_domain_match,
	.map = xive_irq_domain_map,
	.unmap = xive_irq_domain_unmap,
	.xlate = xive_irq_domain_xlate,
};

static void __init xive_init_host(void)
{
	xive_irq_domain = irq_domain_add_nomap(NULL, XIVE_MAX_IRQ,
					       &xive_irq_domain_ops, NULL);
	if (WARN_ON(xive_irq_domain == NULL))
		return;
	irq_set_default_host(xive_irq_domain);
}

static void xive_cleanup_cpu_queues(unsigned int cpu, struct xive_cpu *xc)
{
	if (xc->queue[xive_irq_priority].qpage)
		xive_ops->cleanup_queue(cpu, xc, xive_irq_priority);
}

static int xive_setup_cpu_queues(unsigned int cpu, struct xive_cpu *xc)
{
	int rc = 0;

	/* We setup 1 queues for now with a 64k page */
	if (!xc->queue[xive_irq_priority].qpage)
		rc = xive_ops->setup_queue(cpu, xc, xive_irq_priority);

	return rc;
}

static int xive_prepare_cpu(unsigned int cpu)
{
	struct xive_cpu *xc;

	xc = per_cpu(xive_cpu, cpu);
	if (!xc) {
		struct device_node *np;

		xc = kzalloc_node(sizeof(struct xive_cpu),
				  GFP_KERNEL, cpu_to_node(cpu));
		if (!xc)
			return -ENOMEM;
		np = of_get_cpu_node(cpu, NULL);
		if (np)
			xc->chip_id = of_get_ibm_chip_id(np);
		of_node_put(np);

		per_cpu(xive_cpu, cpu) = xc;
	}

	/* Setup EQs if not already */
	return xive_setup_cpu_queues(cpu, xc);
}

static void xive_setup_cpu(void)
{
	struct xive_cpu *xc = __this_cpu_read(xive_cpu);

	/* Debug: Dump the TM state */
	pr_devel("CPU %d [HW 0x%02x] VT=%02x\n",
	    smp_processor_id(), hard_smp_processor_id(),
	    in_8(xive_tima + xive_tima_offset + TM_WORD2));

	/* The backend might have additional things to do */
	if (xive_ops->setup_cpu)
		xive_ops->setup_cpu(smp_processor_id(), xc);

	/* Set CPPR to 0xff to enable flow of interrupts */
	xc->cppr = 0xff;
	out_8(xive_tima + xive_tima_offset + TM_CPPR, 0xff);
}

#ifdef CONFIG_SMP
void xive_smp_setup_cpu(void)
{
	pr_devel("SMP setup CPU %d\n", smp_processor_id());

	/* This will have already been done on the boot CPU */
	if (smp_processor_id() != boot_cpuid)
		xive_setup_cpu();

}

int xive_smp_prepare_cpu(unsigned int cpu)
{
	int rc;

	/* Allocate per-CPU data and queues */
	rc = xive_prepare_cpu(cpu);
	if (rc)
		return rc;

	/* Allocate and setup IPI for the new CPU */
	return xive_setup_cpu_ipi(cpu);
}

#ifdef CONFIG_HOTPLUG_CPU
static void xive_flush_cpu_queue(unsigned int cpu, struct xive_cpu *xc)
{
	u32 irq;

	/* We assume local irqs are disabled */
	WARN_ON(!irqs_disabled());

	/* Check what's already in the CPU queue */
	while ((irq = xive_scan_interrupts(xc, false)) != 0) {
		/*
		 * We need to re-route that interrupt to its new destination.
		 * First get and lock the descriptor
		 */
		struct irq_desc *desc = irq_to_desc(irq);
		struct irq_data *d = irq_desc_get_irq_data(desc);
		struct xive_irq_data *xd;
		unsigned int hw_irq = (unsigned int)irqd_to_hwirq(d);

		/*
		 * Ignore anything that isn't a XIVE irq and ignore
		 * IPIs, so can just be dropped.
		 */
		if (d->domain != xive_irq_domain || hw_irq == 0)
			continue;

		/*
		 * The IRQ should have already been re-routed, it's just a
		 * stale in the old queue, so re-trigger it in order to make
		 * it reach is new destination.
		 */
#ifdef DEBUG_FLUSH
		pr_info("CPU %d: Got irq %d while offline, re-sending...\n",
			cpu, irq);
#endif
		raw_spin_lock(&desc->lock);
		xd = irq_desc_get_handler_data(desc);

		/*
		 * For LSIs, we EOI, this will cause a resend if it's
		 * still asserted. Otherwise do an MSI retrigger.
		 */
		if (xd->flags & XIVE_IRQ_FLAG_LSI)
			xive_do_source_eoi(irqd_to_hwirq(d), xd);
		else
			xive_irq_retrigger(d);

		raw_spin_unlock(&desc->lock);
	}
}

void xive_smp_disable_cpu(void)
{
	struct xive_cpu *xc = __this_cpu_read(xive_cpu);
	unsigned int cpu = smp_processor_id();

	/* Migrate interrupts away from the CPU */
	irq_migrate_all_off_this_cpu();

	/* Set CPPR to 0 to disable flow of interrupts */
	xc->cppr = 0;
	out_8(xive_tima + xive_tima_offset + TM_CPPR, 0);

	/* Flush everything still in the queue */
	xive_flush_cpu_queue(cpu, xc);

	/* Re-enable CPPR  */
	xc->cppr = 0xff;
	out_8(xive_tima + xive_tima_offset + TM_CPPR, 0xff);
}

void xive_flush_interrupt(void)
{
	struct xive_cpu *xc = __this_cpu_read(xive_cpu);
	unsigned int cpu = smp_processor_id();

	/* Called if an interrupt occurs while the CPU is hot unplugged */
	xive_flush_cpu_queue(cpu, xc);
}

#endif /* CONFIG_HOTPLUG_CPU */

#endif /* CONFIG_SMP */

void xive_teardown_cpu(void)
{
	struct xive_cpu *xc = __this_cpu_read(xive_cpu);
	unsigned int cpu = smp_processor_id();

	/* Set CPPR to 0 to disable flow of interrupts */
	xc->cppr = 0;
	out_8(xive_tima + xive_tima_offset + TM_CPPR, 0);

	if (xive_ops->teardown_cpu)
		xive_ops->teardown_cpu(cpu, xc);
}

void xive_kexec_teardown_cpu(int secondary)
{
	struct xive_cpu *xc = __this_cpu_read(xive_cpu);
	unsigned int cpu = smp_processor_id();

	/* Set CPPR to 0 to disable flow of interrupts */
	xc->cppr = 0;
	out_8(xive_tima + xive_tima_offset + TM_CPPR, 0);

	/* Backend cleanup if any */
	if (xive_ops->teardown_cpu)
		xive_ops->teardown_cpu(cpu, xc);

#ifdef CONFIG_SMP
	/* Get rid of IPI */
	xive_cleanup_cpu_ipi(cpu, xc);
#endif

	/* Disable and free the queues */
	xive_cleanup_cpu_queues(cpu, xc);
}

void xive_shutdown(void)
{
	xive_ops->shutdown();
}

bool __init xive_core_init(const struct xive_ops *ops, void __iomem *area, u32 offset,
			   u8 max_prio)
{
	xive_tima = area;
	xive_tima_offset = offset;
	xive_ops = ops;
	xive_irq_priority = max_prio;

	ppc_md.get_irq = xive_get_irq;
	__xive_enabled = true;

	pr_devel("Initializing host..\n");
	xive_init_host();

	pr_devel("Initializing boot CPU..\n");

	/* Allocate per-CPU data and queues */
	xive_prepare_cpu(smp_processor_id());

	/* Get ready for interrupts */
	xive_setup_cpu();

	pr_info("Interrupt handling initialized with %s backend\n",
		xive_ops->name);
	pr_info("Using priority %d for all interrupts\n", max_prio);

	return true;
}

__be32 *xive_queue_page_alloc(unsigned int cpu, u32 queue_shift)
{
	unsigned int alloc_order;
	struct page *pages;
	__be32 *qpage;

	alloc_order = xive_alloc_order(queue_shift);
	pages = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, alloc_order);
	if (!pages)
		return ERR_PTR(-ENOMEM);
	qpage = (__be32 *)page_address(pages);
	memset(qpage, 0, 1 << queue_shift);

	return qpage;
}

static int __init xive_off(char *arg)
{
	xive_cmdline_disabled = true;
	return 0;
}
__setup("xive=off", xive_off);