xref: /src/sys/arch/arm/pic/pic.c

/*	$NetBSD: pic.c,v 1.65 2021/02/16 22:12:49 jmcneill Exp $	*/
/*-
 * Copyright (c) 2008 The NetBSD Foundation, Inc.
 * All rights reserved.
 *
 * This code is derived from software contributed to The NetBSD Foundation
 * by Matt Thomas.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

#define _INTR_PRIVATE
#include "opt_ddb.h"
#include "opt_multiprocessor.h"

#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: pic.c,v 1.65 2021/02/16 22:12:49 jmcneill Exp $");

#include <sys/param.h>
#include <sys/atomic.h>
#include <sys/cpu.h>
#include <sys/evcnt.h>
#include <sys/interrupt.h>
#include <sys/intr.h>
#include <sys/ipi.h>
#include <sys/kernel.h>
#include <sys/kmem.h>
#include <sys/mutex.h>
#include <sys/once.h>
#include <sys/xcall.h>

#include <arm/armreg.h>
#include <arm/cpufunc.h>
#include <arm/locore.h>	/* for compat aarch64 */

#ifdef DDB
#include <arm/db_machdep.h>
#endif

#include <arm/pic/picvar.h>

#if defined(__HAVE_PIC_PENDING_INTRS)

bool pic_pending_used __read_mostly = false;

/*
 * This implementation of pending interrupts on a MULTIPROCESSOR system makes
 * the assumption that a PIC (pic_softc) shall only have all its interrupts
 * come from the same CPU.  In other words, interrupts from a single PIC will
 * not be distributed among multiple CPUs.
 */
struct pic_pending {
	volatile uint32_t blocked_pics;
	volatile uint32_t pending_pics;
	volatile uint32_t pending_ipls;
};
static uint32_t
	pic_find_pending_irqs_by_ipl(struct pic_softc *, size_t, uint32_t, int);
static struct pic_softc *
	pic_list_find_pic_by_pending_ipl(struct pic_pending *, uint32_t);
static void
	pic_deliver_irqs(struct pic_pending *, struct pic_softc *, int, void *);
static void
	pic_list_deliver_irqs(struct pic_pending *, register_t, int, void *);

#ifdef MULTIPROCESSOR
percpu_t *pic_pending_percpu;
static struct pic_pending *
pic_pending_get(void)
{
	return percpu_getref(pic_pending_percpu);
}
static void
pic_pending_put(struct pic_pending *pend)
{
	percpu_putref(pic_pending_percpu);
}
#else
struct pic_pending pic_pending;
#define	pic_pending_get()	(&pic_pending)
#define	pic_pending_put(pend)	__nothing
#endif /* MULTIPROCESSOR */
#endif /* __HAVE_PIC_PENDING_INTRS */

struct pic_softc *pic_list[PIC_MAXPICS];
#if PIC_MAXPICS > 32
#error PIC_MAXPICS > 32 not supported
#endif
struct intrsource *pic_sources[PIC_MAXMAXSOURCES];
struct intrsource *pic__iplsources[PIC_MAXMAXSOURCES];
struct intrsource **pic_iplsource[NIPL] = {
	[0 ... NIPL-1] = pic__iplsources,
};
size_t pic_ipl_offset[NIPL+1];

static kmutex_t pic_lock;
static size_t pic_sourcebase;
static int pic_lastbase;
static struct evcnt pic_deferral_ev =
    EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, "deferred", "intr");
EVCNT_ATTACH_STATIC(pic_deferral_ev);

static int pic_init(void);

#ifdef __HAVE_PIC_SET_PRIORITY
void
pic_set_priority(struct cpu_info *ci, int newipl)
{
	register_t psw = cpsid(I32_bit);
	if (pic_list[0] != NULL)
		(pic_list[0]->pic_ops->pic_set_priority)(pic_list[0], newipl);
	ci->ci_cpl = newipl;
	if ((psw & I32_bit) == 0)
		cpsie(I32_bit);
}

void
pic_set_priority_psw(struct cpu_info *ci, register_t psw, int newipl)
{
	if ((psw & I32_bit) == 0) {
		DISABLE_INTERRUPT();
	}
	if (pic_list[0] != NULL) {
		(pic_list[0]->pic_ops->pic_set_priority)(pic_list[0], newipl);
	}
	ci->ci_cpl = newipl;
	if ((psw & I32_bit) == 0) {
		ENABLE_INTERRUPT();
	}
}
#endif

#ifdef MULTIPROCESSOR
int
pic_ipi_ast(void *arg)
{
	setsoftast(curcpu());
	return 1;
}

int
pic_ipi_nop(void *arg)
{
	/* do nothing */
	return 1;
}

int
pic_ipi_xcall(void *arg)
{
	xc_ipi_handler();
	return 1;
}

int
pic_ipi_generic(void *arg)
{
	ipi_cpu_handler();
	return 1;
}

#ifdef DDB
int
pic_ipi_ddb(void *arg)
{
//	printf("%s: %s: tf=%p\n", __func__, curcpu()->ci_cpuname, arg);
	kdb_trap(-1, arg);
	return 1;
}
#endif /* DDB */

#ifdef __HAVE_PREEMPTION
int
pic_ipi_kpreempt(void *arg)
{
	atomic_or_uint(&curcpu()->ci_astpending, __BIT(1));
	return 1;
}
#endif /* __HAVE_PREEMPTION */

void
intr_cpu_init(struct cpu_info *ci)
{
	for (size_t slot = 0; slot < PIC_MAXPICS; slot++) {
		struct pic_softc * const pic = pic_list[slot];
		if (pic != NULL && pic->pic_ops->pic_cpu_init != NULL) {
			(*pic->pic_ops->pic_cpu_init)(pic, ci);
		}
	}
}

typedef void (*pic_ipi_send_func_t)(struct pic_softc *, u_long);

void
intr_ipi_send(const kcpuset_t *kcp, u_long ipi)
{
	struct cpu_info * const ci = curcpu();
	KASSERT(ipi < NIPI);
	KASSERT(kcp == NULL || kcpuset_countset(kcp) == 1);
	bool __diagused sent_p = false;
	for (size_t slot = 0; slot < PIC_MAXPICS; slot++) {
		struct pic_softc * const pic = pic_list[slot];
		if (pic == NULL || pic->pic_cpus == NULL)
			continue;
		if (kcp == NULL || kcpuset_intersecting_p(kcp, pic->pic_cpus)) {
			/*
			 * Never send to ourself.
			 *
			 * This test uses pointer comparison for systems
			 * that have a pic per cpu, e.g. RPI[23].  GIC sets
			 * pic_cpus to kcpuset_running and handles "not for
			 * self" internally.
			 */
			if (pic->pic_cpus == ci->ci_kcpuset)
				continue;

			(*pic->pic_ops->pic_ipi_send)(pic, kcp, ipi);

			/*
			 * If we were targeting a single CPU or this pic
			 * handles all cpus, we're done.
			 */
			if (kcp != NULL || pic->pic_cpus == kcpuset_running)
				return;
			sent_p = true;
		}
	}
	KASSERTMSG(cold || sent_p || ncpu <= 1, "cold %d sent_p %d ncpu %d",
	    cold, sent_p, ncpu);
}
#endif /* MULTIPROCESSOR */

#ifdef __HAVE_PIC_FAST_SOFTINTS
int
pic_handle_softint(void *arg)
{
	void softint_switch(lwp_t *, int);
	struct cpu_info * const ci = curcpu();
	const size_t softint = (size_t) arg;
	int s = splhigh();
	ci->ci_intr_depth--;	// don't count these as interrupts
	softint_switch(ci->ci_softlwps[softint], s);
	ci->ci_intr_depth++;
	splx(s);
	return 1;
}
#endif

int
pic_handle_intr(void *arg)
{
	struct pic_softc * const pic = arg;
	int rv;

	rv = (*pic->pic_ops->pic_find_pending_irqs)(pic);

	return rv > 0;
}

#if defined(__HAVE_PIC_PENDING_INTRS)
void
pic_mark_pending_source(struct pic_softc *pic, struct intrsource *is)
{
	const uint32_t ipl_mask = __BIT(is->is_ipl);

	if (!pic_pending_used)
		pic_pending_used = true;

	atomic_or_32(&pic->pic_pending_irqs[is->is_irq >> 5],
	    __BIT(is->is_irq & 0x1f));

	atomic_or_32(&pic->pic_pending_ipls, ipl_mask);
	struct pic_pending *pend = pic_pending_get();
	atomic_or_32(&pend->pending_ipls, ipl_mask);
	atomic_or_32(&pend->pending_pics, __BIT(pic->pic_id));
	pic_pending_put(pend);
}

void
pic_mark_pending(struct pic_softc *pic, int irq)
{
	struct intrsource * const is = pic->pic_sources[irq];

	KASSERT(irq < pic->pic_maxsources);
	KASSERT(is != NULL);

	pic_mark_pending_source(pic, is);
}

uint32_t
pic_mark_pending_sources(struct pic_softc *pic, size_t irq_base,
	uint32_t pending)
{
	struct intrsource ** const isbase = &pic->pic_sources[irq_base];
	struct intrsource *is;
	volatile uint32_t *ipending = &pic->pic_pending_irqs[irq_base >> 5];
	uint32_t ipl_mask = 0;

	if (!pic_pending_used)
		pic_pending_used = true;

	if (pending == 0)
		return ipl_mask;

	KASSERT((irq_base & 31) == 0);

	(*pic->pic_ops->pic_block_irqs)(pic, irq_base, pending);

	atomic_or_32(ipending, pending);
	while (pending != 0) {
		int n = ffs(pending);
		if (n-- == 0)
			break;
		is = isbase[n];
		KASSERT(is != NULL);
		KASSERT(irq_base <= is->is_irq && is->is_irq < irq_base + 32);
		pending &= ~__BIT(n);
		ipl_mask |= __BIT(is->is_ipl);
	}

	atomic_or_32(&pic->pic_pending_ipls, ipl_mask);
	struct pic_pending *pend = pic_pending_get();
	atomic_or_32(&pend->pending_ipls, ipl_mask);
	atomic_or_32(&pend->pending_pics, __BIT(pic->pic_id));
	pic_pending_put(pend);
	return ipl_mask;
}

uint32_t
pic_find_pending_irqs_by_ipl(struct pic_softc *pic, size_t irq_base,
	uint32_t pending, int ipl)
{
	uint32_t ipl_irq_mask = 0;
	uint32_t irq_mask;

	for (;;) {
		int irq = ffs(pending);
		if (irq-- == 0)
			return ipl_irq_mask;

		irq_mask = __BIT(irq);
#if 1
    		KASSERTMSG(pic->pic_sources[irq_base + irq] != NULL,
		   "%s: irq_base %zu irq %d\n", __func__, irq_base, irq);
#else
		if (pic->pic_sources[irq_base + irq] == NULL) {
			aprint_error("stray interrupt? irq_base=%zu irq=%d\n",
			    irq_base, irq);
		} else
#endif
		if (pic->pic_sources[irq_base + irq]->is_ipl == ipl)
			ipl_irq_mask |= irq_mask;

		pending &= ~irq_mask;
	}
}
#endif /* __HAVE_PIC_PENDING_INTRS */

void
pic_dispatch(struct intrsource *is, void *frame)
{
	int (*func)(void *) = is->is_func;
	void *arg = is->is_arg;

	if (__predict_false(arg == NULL)) {
		if (__predict_false(frame == NULL)) {
			pic_deferral_ev.ev_count++;
			return;
		}
		arg = frame;
	}

#ifdef MULTIPROCESSOR
	if (!is->is_mpsafe) {
		KERNEL_LOCK(1, NULL);
		const u_int ci_blcnt __diagused = curcpu()->ci_biglock_count;
		const u_int l_blcnt __diagused = curlwp->l_blcnt;
		(void)(*func)(arg);
		KASSERT(ci_blcnt == curcpu()->ci_biglock_count);
		KASSERT(l_blcnt == curlwp->l_blcnt);
		KERNEL_UNLOCK_ONE(NULL);
	} else
#endif
		(void)(*func)(arg);

	struct pic_percpu * const pcpu = percpu_getref(is->is_pic->pic_percpu);
	KASSERT(pcpu->pcpu_magic == PICPERCPU_MAGIC);
	pcpu->pcpu_evs[is->is_irq].ev_count++;
	percpu_putref(is->is_pic->pic_percpu);
}

#if defined(__HAVE_PIC_PENDING_INTRS)
void
pic_deliver_irqs(struct pic_pending *pend, struct pic_softc *pic, int ipl,
    void *frame)
{
	const uint32_t ipl_mask = __BIT(ipl);
	struct intrsource *is;
	volatile uint32_t *ipending = pic->pic_pending_irqs;
	volatile uint32_t *iblocked = pic->pic_blocked_irqs;
	size_t irq_base;
#if PIC_MAXSOURCES > 32
	size_t irq_count;
	int poi = 0;		/* Possibility of interrupting */
#endif
	uint32_t pending_irqs;
	uint32_t blocked_irqs;
	int irq;
	bool progress __diagused = false;

	KASSERT(pic->pic_pending_ipls & ipl_mask);

	irq_base = 0;
#if PIC_MAXSOURCES > 32
	irq_count = 0;
#endif

	for (;;) {
		pending_irqs = pic_find_pending_irqs_by_ipl(pic, irq_base,
		    *ipending, ipl);
		KASSERT((pending_irqs & *ipending) == pending_irqs);
		KASSERT((pending_irqs & ~(*ipending)) == 0);
		if (pending_irqs == 0) {
#if PIC_MAXSOURCES > 32
			irq_count += 32;
			if (__predict_true(irq_count >= pic->pic_maxsources)) {
				if (!poi)
					/*Interrupt at this level was handled.*/
					break;
				irq_base = 0;
				irq_count = 0;
				poi = 0;
				ipending = pic->pic_pending_irqs;
				iblocked = pic->pic_blocked_irqs;
			} else {
				irq_base += 32;
				ipending++;
				iblocked++;
				KASSERT(irq_base <= pic->pic_maxsources);
			}
			continue;
#else
			break;
#endif
		}
		progress = true;
		blocked_irqs = 0;
		do {
			irq = ffs(pending_irqs) - 1;
			KASSERT(irq >= 0);

			atomic_and_32(ipending, ~__BIT(irq));
			is = pic->pic_sources[irq_base + irq];
			if (is != NULL) {
				ENABLE_INTERRUPT();
				pic_dispatch(is, frame);
				DISABLE_INTERRUPT();
#if PIC_MAXSOURCES > 32
				/*
				 * There is a possibility of interrupting
				 * from ENABLE_INTERRUPT() to
				 * DISABLE_INTERRUPT().
				 */
				poi = 1;
#endif
				blocked_irqs |= __BIT(irq);
			} else {
				KASSERT(0);
			}
			pending_irqs = pic_find_pending_irqs_by_ipl(pic,
			    irq_base, *ipending, ipl);
		} while (pending_irqs);
		if (blocked_irqs) {
			atomic_or_32(iblocked, blocked_irqs);
			atomic_or_32(&pend->blocked_pics, __BIT(pic->pic_id));
		}
	}

	KASSERT(progress);
	/*
	 * Since interrupts are disabled, we don't have to be too careful
	 * about these.
	 */
	if (atomic_and_32_nv(&pic->pic_pending_ipls, ~ipl_mask) == 0)
		atomic_and_32(&pend->pending_pics, ~__BIT(pic->pic_id));
}

static void
pic_list_unblock_irqs(struct pic_pending *pend)
{
	uint32_t blocked_pics = pend->blocked_pics;

	pend->blocked_pics = 0;

	for (;;) {
		struct pic_softc *pic;
#if PIC_MAXSOURCES > 32
		volatile uint32_t *iblocked;
		uint32_t blocked;
		size_t irq_base;
#endif

		int pic_id = ffs(blocked_pics);
		if (pic_id-- == 0)
			return;

		pic = pic_list[pic_id];
		KASSERT(pic != NULL);
#if PIC_MAXSOURCES > 32
		for (irq_base = 0, iblocked = pic->pic_blocked_irqs;
		     irq_base < pic->pic_maxsources;
		     irq_base += 32, iblocked++) {
			if ((blocked = *iblocked) != 0) {
				(*pic->pic_ops->pic_unblock_irqs)(pic,
				    irq_base, blocked);
				atomic_and_32(iblocked, ~blocked);
			}
		}
#else
		KASSERT(pic->pic_blocked_irqs[0] != 0);
		(*pic->pic_ops->pic_unblock_irqs)(pic,
		    0, pic->pic_blocked_irqs[0]);
		pic->pic_blocked_irqs[0] = 0;
#endif
		blocked_pics &= ~__BIT(pic_id);
	}
}

struct pic_softc *
pic_list_find_pic_by_pending_ipl(struct pic_pending *pend, uint32_t ipl_mask)
{
	uint32_t pending_pics = pend->pending_pics;
	struct pic_softc *pic;

	for (;;) {
		int pic_id = ffs(pending_pics);
		if (pic_id-- == 0)
			return NULL;

		pic = pic_list[pic_id];
		KASSERT(pic != NULL);
		if (pic->pic_pending_ipls & ipl_mask)
			return pic;
		pending_pics &= ~__BIT(pic_id);
	}
}

void
pic_list_deliver_irqs(struct pic_pending *pend, register_t psw, int ipl,
    void *frame)
{
	const uint32_t ipl_mask = __BIT(ipl);
	struct pic_softc *pic;

	while ((pic = pic_list_find_pic_by_pending_ipl(pend, ipl_mask)) != NULL) {
		pic_deliver_irqs(pend, pic, ipl, frame);
		KASSERT((pic->pic_pending_ipls & ipl_mask) == 0);
	}
	atomic_and_32(&pend->pending_ipls, ~ipl_mask);
}
#endif /* __HAVE_PIC_PENDING_INTRS */

void
pic_do_pending_ints(register_t psw, int newipl, void *frame)
{
	struct cpu_info * const ci = curcpu();
	if (__predict_false(newipl == IPL_HIGH)) {
		KASSERTMSG(ci->ci_cpl == IPL_HIGH, "cpl %d", ci->ci_cpl);
		return;
	}
#if defined(__HAVE_PIC_PENDING_INTRS)
	struct pic_pending *pend = pic_pending_get();
	while ((pend->pending_ipls & ~__BIT(newipl)) > __BIT(newipl)) {
		KASSERT(pend->pending_ipls < __BIT(NIPL));
		for (;;) {
			int ipl = 31 - __builtin_clz(pend->pending_ipls);
			KASSERT(ipl < NIPL);
			if (ipl <= newipl)
				break;

			pic_set_priority_psw(ci, psw, ipl);
			pic_list_deliver_irqs(pend, psw, ipl, frame);
			pic_list_unblock_irqs(pend);
		}
	}
	pic_pending_put(pend);
#endif /* __HAVE_PIC_PENDING_INTRS */
#ifdef __HAVE_PREEMPTION
	if (newipl == IPL_NONE && (ci->ci_astpending & __BIT(1))) {
		pic_set_priority_psw(ci, psw, IPL_SCHED);
		kpreempt(0);
	}
#endif
	if (ci->ci_cpl != newipl)
		pic_set_priority_psw(ci, psw, newipl);
}

static void
pic_percpu_allocate(void *v0, void *v1, struct cpu_info *ci)
{
	struct pic_percpu * const pcpu = v0;
	struct pic_softc * const pic = v1;

	pcpu->pcpu_evs = kmem_zalloc(pic->pic_maxsources * sizeof(pcpu->pcpu_evs[0]),
	    KM_SLEEP);
	KASSERT(pcpu->pcpu_evs != NULL);

#define	PCPU_NAMELEN	32
#ifdef DIAGNOSTIC
	const size_t namelen = strlen(pic->pic_name) + 4 + strlen(ci->ci_data.cpu_name);
#endif

	KASSERT(namelen < PCPU_NAMELEN);
	pcpu->pcpu_name = kmem_alloc(PCPU_NAMELEN, KM_SLEEP);
#ifdef MULTIPROCESSOR
	snprintf(pcpu->pcpu_name, PCPU_NAMELEN,
	    "%s (%s)", pic->pic_name, ci->ci_data.cpu_name);
#else
	strlcpy(pcpu->pcpu_name, pic->pic_name, PCPU_NAMELEN);
#endif
	pcpu->pcpu_magic = PICPERCPU_MAGIC;
#if 0
	printf("%s: %s %s: <%s>\n",
	    __func__, ci->ci_data.cpu_name, pic->pic_name,
	    pcpu->pcpu_name);
#endif
}

static int
pic_init(void)
{

	mutex_init(&pic_lock, MUTEX_DEFAULT, IPL_HIGH);

	return 0;
}

int
pic_add(struct pic_softc *pic, int irqbase)
{
	int slot, maybe_slot = -1;
	size_t sourcebase;
	static ONCE_DECL(pic_once);

	RUN_ONCE(&pic_once, pic_init);

	KASSERT(strlen(pic->pic_name) > 0);

#if defined(__HAVE_PIC_PENDING_INTRS) && defined(MULTIPROCESSOR)
	if (__predict_false(pic_pending_percpu == NULL))
		pic_pending_percpu = percpu_alloc(sizeof(struct pic_pending));
#endif /* __HAVE_PIC_PENDING_INTRS && MULTIPROCESSOR */

	mutex_enter(&pic_lock);
	if (irqbase == PIC_IRQBASE_ALLOC) {
		irqbase = pic_lastbase;
	}
	for (slot = 0; slot < PIC_MAXPICS; slot++) {
		struct pic_softc * const xpic = pic_list[slot];
		if (xpic == NULL) {
			if (maybe_slot < 0)
				maybe_slot = slot;
			if (irqbase < 0)
				break;
			continue;
		}
		if (irqbase < 0 || xpic->pic_irqbase < 0)
			continue;
		if (irqbase >= xpic->pic_irqbase + xpic->pic_maxsources)
			continue;
		if (irqbase + pic->pic_maxsources <= xpic->pic_irqbase)
			continue;
		panic("pic_add: pic %s (%zu sources @ irq %u) conflicts"
		    " with pic %s (%zu sources @ irq %u)",
		    pic->pic_name, pic->pic_maxsources, irqbase,
		    xpic->pic_name, xpic->pic_maxsources, xpic->pic_irqbase);
	}
	slot = maybe_slot;
#if 0
	printf("%s: pic_sourcebase=%zu pic_maxsources=%zu\n",
	    pic->pic_name, pic_sourcebase, pic->pic_maxsources);
#endif
	KASSERTMSG(pic->pic_maxsources <= PIC_MAXSOURCES, "%zu",
	    pic->pic_maxsources);
	KASSERT(pic_sourcebase + pic->pic_maxsources <= PIC_MAXMAXSOURCES);
	sourcebase = pic_sourcebase;
	pic_sourcebase += pic->pic_maxsources;
        if (pic_lastbase < irqbase + pic->pic_maxsources)
                pic_lastbase = irqbase + pic->pic_maxsources;
	mutex_exit(&pic_lock);

	/*
	 * Allocate a pointer to each cpu's evcnts and then, for each cpu,
	 * allocate its evcnts and then attach an evcnt for each pin.
	 * We can't allocate the evcnt structures directly since
	 * percpu will move the contents of percpu memory around and
	 * corrupt the pointers in the evcnts themselves.  Remember, any
	 * problem can be solved with sufficient indirection.
	 */
	pic->pic_percpu = percpu_create(sizeof(struct pic_percpu),
	    pic_percpu_allocate, NULL, pic);

	pic->pic_sources = &pic_sources[sourcebase];
	pic->pic_irqbase = irqbase;
	pic->pic_id = slot;
#ifdef __HAVE_PIC_SET_PRIORITY
	KASSERT((slot == 0) == (pic->pic_ops->pic_set_priority != NULL));
#endif
#ifdef MULTIPROCESSOR
	KASSERT((pic->pic_cpus != NULL) == (pic->pic_ops->pic_ipi_send != NULL));
#endif
	pic_list[slot] = pic;

	return irqbase;
}

int
pic_alloc_irq(struct pic_softc *pic)
{
	int irq;

	for (irq = 0; irq < pic->pic_maxsources; irq++) {
		if (pic->pic_sources[irq] == NULL)
			return irq;
	}

	return -1;
}

static void
pic_percpu_evcnt_attach(void *v0, void *v1, struct cpu_info *ci)
{
	struct pic_percpu * const pcpu = v0;
	struct intrsource * const is = v1;

	KASSERT(pcpu->pcpu_magic == PICPERCPU_MAGIC);
	evcnt_attach_dynamic(&pcpu->pcpu_evs[is->is_irq], EVCNT_TYPE_INTR, NULL,
	    pcpu->pcpu_name, is->is_source);
}

void *
pic_establish_intr(struct pic_softc *pic, int irq, int ipl, int type,
	int (*func)(void *), void *arg, const char *xname)
{
	struct intrsource *is;
	int off, nipl;

	if (pic->pic_sources[irq]) {
		printf("pic_establish_intr: pic %s irq %d already present\n",
		    pic->pic_name, irq);
		return NULL;
	}

	is = kmem_zalloc(sizeof(*is), KM_SLEEP);
	is->is_pic = pic;
	is->is_irq = irq;
	is->is_ipl = ipl;
	is->is_type = type & 0xff;
	is->is_func = func;
	is->is_arg = arg;
#ifdef MULTIPROCESSOR
	is->is_mpsafe = (type & IST_MPSAFE) || ipl != IPL_VM;
#endif

	if (pic->pic_ops->pic_source_name)
		(*pic->pic_ops->pic_source_name)(pic, irq, is->is_source,
		    sizeof(is->is_source));
	else
		snprintf(is->is_source, sizeof(is->is_source), "irq %d", irq);

	/*
	 * Now attach the per-cpu evcnts.
	 */
	percpu_foreach(pic->pic_percpu, pic_percpu_evcnt_attach, is);

	pic->pic_sources[irq] = is;

	/*
	 * First try to use an existing slot which is empty.
	 */
	for (off = pic_ipl_offset[ipl]; off < pic_ipl_offset[ipl+1]; off++) {
		if (pic__iplsources[off] == NULL) {
			is->is_iplidx = off - pic_ipl_offset[ipl];
			pic__iplsources[off] = is;
			goto unblock;
		}
	}

	/*
	 * Move up all the sources by one.
 	 */
	if (ipl < NIPL) {
		off = pic_ipl_offset[ipl+1];
		memmove(&pic__iplsources[off+1], &pic__iplsources[off],
		    sizeof(pic__iplsources[0]) * (pic_ipl_offset[NIPL] - off));
	}

	/*
	 * Advance the offset of all IPLs higher than this.  Include an
	 * extra one as well.  Thus the number of sources per ipl is
	 * pic_ipl_offset[ipl+1] - pic_ipl_offset[ipl].
	 */
	for (nipl = ipl + 1; nipl <= NIPL; nipl++)
		pic_ipl_offset[nipl]++;

	/*
	 * Insert into the previously made position at the end of this IPL's
	 * sources.
	 */
	off = pic_ipl_offset[ipl + 1] - 1;
	is->is_iplidx = off - pic_ipl_offset[ipl];
	pic__iplsources[off] = is;

	(*pic->pic_ops->pic_establish_irq)(pic, is);

unblock:
	(*pic->pic_ops->pic_unblock_irqs)(pic, is->is_irq & ~0x1f,
	    __BIT(is->is_irq & 0x1f));

	if (xname) {
		if (is->is_xname == NULL)
			is->is_xname = kmem_zalloc(INTRDEVNAMEBUF, KM_SLEEP);
		if (is->is_xname[0] != '\0')
			strlcat(is->is_xname, ", ", INTRDEVNAMEBUF);
		strlcat(is->is_xname, xname, INTRDEVNAMEBUF);
	}

	/* We're done. */
	return is;
}

static void
pic_percpu_evcnt_deattach(void *v0, void *v1, struct cpu_info *ci)
{
	struct pic_percpu * const pcpu = v0;
	struct intrsource * const is = v1;

	KASSERT(pcpu->pcpu_magic == PICPERCPU_MAGIC);
	evcnt_detach(&pcpu->pcpu_evs[is->is_irq]);
}

void
pic_disestablish_source(struct intrsource *is)
{
	struct pic_softc * const pic = is->is_pic;
	const int irq = is->is_irq;

	KASSERT(is == pic->pic_sources[irq]);

	(*pic->pic_ops->pic_block_irqs)(pic, irq & ~0x1f, __BIT(irq & 0x1f));
	pic->pic_sources[irq] = NULL;
	pic__iplsources[pic_ipl_offset[is->is_ipl] + is->is_iplidx] = NULL;
	if (is->is_xname != NULL) {
		kmem_free(is->is_xname, INTRDEVNAMEBUF);
		is->is_xname = NULL;
	}
	/*
	 * Now detach the per-cpu evcnts.
	 */
	percpu_foreach(pic->pic_percpu, pic_percpu_evcnt_deattach, is);

	kmem_free(is, sizeof(*is));
}

void *
intr_establish(int irq, int ipl, int type, int (*func)(void *), void *arg)
{
	return intr_establish_xname(irq, ipl, type, func, arg, NULL);
}

void *
intr_establish_xname(int irq, int ipl, int type, int (*func)(void *), void *arg,
    const char *xname)
{
	KASSERT(!cpu_intr_p());
	KASSERT(!cpu_softintr_p());

	for (size_t slot = 0; slot < PIC_MAXPICS; slot++) {
		struct pic_softc * const pic = pic_list[slot];
		if (pic == NULL || pic->pic_irqbase < 0)
			continue;
		if (pic->pic_irqbase <= irq
		    && irq < pic->pic_irqbase + pic->pic_maxsources) {
			return pic_establish_intr(pic, irq - pic->pic_irqbase,
			    ipl, type, func, arg, xname);
		}
	}

	return NULL;
}

void
intr_disestablish(void *ih)
{
	struct intrsource * const is = ih;

	KASSERT(!cpu_intr_p());
	KASSERT(!cpu_softintr_p());

	pic_disestablish_source(is);
}

void
intr_mask(void *ih)
{
	struct intrsource * const is = ih;
	struct pic_softc * const pic = is->is_pic;
	const int irq = is->is_irq;

	if (atomic_inc_32_nv(&is->is_mask_count) == 1)
		(*pic->pic_ops->pic_block_irqs)(pic, irq & ~0x1f, __BIT(irq & 0x1f));
}

void
intr_unmask(void *ih)
{
	struct intrsource * const is = ih;
	struct pic_softc * const pic = is->is_pic;
	const int irq = is->is_irq;

	if (atomic_dec_32_nv(&is->is_mask_count) == 0)
		(*pic->pic_ops->pic_unblock_irqs)(pic, irq & ~0x1f, __BIT(irq & 0x1f));
}

const char *
intr_string(intr_handle_t irq, char *buf, size_t len)
{
	for (size_t slot = 0; slot < PIC_MAXPICS; slot++) {
		struct pic_softc * const pic = pic_list[slot];
		if (pic == NULL || pic->pic_irqbase < 0)
			continue;
		if (pic->pic_irqbase <= irq
		    && irq < pic->pic_irqbase + pic->pic_maxsources) {
			struct intrsource * const is = pic->pic_sources[irq - pic->pic_irqbase];
			snprintf(buf, len, "%s %s", pic->pic_name, is->is_source);
			return buf;
		}
	}

	return NULL;
}

static struct intrsource *
intr_get_source(const char *intrid)
{
	struct intrsource *is;
	intrid_t buf;
	size_t slot;
	int irq;

	KASSERT(mutex_owned(&cpu_lock));

	for (slot = 0; slot < PIC_MAXPICS; slot++) {
		struct pic_softc * const pic = pic_list[slot];
		if (pic == NULL || pic->pic_irqbase < 0)
			continue;
		for (irq = 0; irq < pic->pic_maxsources; irq++) {
			is = pic->pic_sources[irq];
			if (is == NULL || is->is_source[0] == '\0')
				continue;

			snprintf(buf, sizeof(buf), "%s %s", pic->pic_name, is->is_source);
			if (strcmp(buf, intrid) == 0)
				return is;
		}
	}

	return NULL;
}

struct intrids_handler *
interrupt_construct_intrids(const kcpuset_t *cpuset)
{
	struct intrids_handler *iih;
	struct intrsource *is;
	int count, irq, n;
	size_t slot;

	if (kcpuset_iszero(cpuset))
		return NULL;

	count = 0;
	for (slot = 0; slot < PIC_MAXPICS; slot++) {
		struct pic_softc * const pic = pic_list[slot];
		if (pic != NULL && pic->pic_irqbase >= 0) {
			for (irq = 0; irq < pic->pic_maxsources; irq++) {
				is = pic->pic_sources[irq];
				if (is && is->is_source[0] != '\0')
					count++;
			}
		}
	}

	iih = kmem_zalloc(sizeof(int) + sizeof(intrid_t) * count, KM_SLEEP);
	iih->iih_nids = count;

	for (n = 0, slot = 0; n < count && slot < PIC_MAXPICS; slot++) {
		struct pic_softc * const pic = pic_list[slot];
		if (pic == NULL || pic->pic_irqbase < 0)
			continue;
		for (irq = 0; irq < pic->pic_maxsources; irq++) {
			is = pic->pic_sources[irq];
			if (is == NULL || is->is_source[0] == '\0')
				continue;

			snprintf(iih->iih_intrids[n++], sizeof(intrid_t), "%s %s",
			    pic->pic_name, is->is_source);
		}
	}

	return iih;
}

void
interrupt_destruct_intrids(struct intrids_handler *iih)
{
	if (iih == NULL)
		return;

	kmem_free(iih, sizeof(int) + sizeof(intrid_t) * iih->iih_nids);
}

void
interrupt_get_available(kcpuset_t *cpuset)
{
	CPU_INFO_ITERATOR cii;
	struct cpu_info *ci;

	kcpuset_zero(cpuset);

	mutex_enter(&cpu_lock);
	for (CPU_INFO_FOREACH(cii, ci)) {
		if ((ci->ci_schedstate.spc_flags & SPCF_NOINTR) == 0)
			kcpuset_set(cpuset, cpu_index(ci));
	}
	mutex_exit(&cpu_lock);
}

void
interrupt_get_devname(const char *intrid, char *buf, size_t len)
{
	struct intrsource *is;

	mutex_enter(&cpu_lock);
	is = intr_get_source(intrid);
	if (is == NULL || is->is_xname == NULL)
		buf[0] = '\0';
	else
		strlcpy(buf, is->is_xname, len);
	mutex_exit(&cpu_lock);
}

struct interrupt_get_count_arg {
	struct intrsource *is;
	uint64_t count;
	u_int cpu_idx;
};

static void
interrupt_get_count_cb(void *v0, void *v1, struct cpu_info *ci)
{
	struct pic_percpu * const pcpu = v0;
	struct interrupt_get_count_arg * const arg = v1;

	if (arg->cpu_idx != cpu_index(ci))
		return;

	arg->count = pcpu->pcpu_evs[arg->is->is_irq].ev_count;
}

uint64_t
interrupt_get_count(const char *intrid, u_int cpu_idx)
{
	struct interrupt_get_count_arg arg;
	struct intrsource *is;
	uint64_t count;

	count = 0;

	mutex_enter(&cpu_lock);
	is = intr_get_source(intrid);
	if (is != NULL && is->is_pic != NULL) {
		arg.is = is;
		arg.count = 0;
		arg.cpu_idx = cpu_idx;
		percpu_foreach(is->is_pic->pic_percpu, interrupt_get_count_cb, &arg);
		count = arg.count;
	}
	mutex_exit(&cpu_lock);

	return count;
}

#ifdef MULTIPROCESSOR
void
interrupt_get_assigned(const char *intrid, kcpuset_t *cpuset)
{
	struct intrsource *is;
	struct pic_softc *pic;

	kcpuset_zero(cpuset);

	mutex_enter(&cpu_lock);
	is = intr_get_source(intrid);
	if (is != NULL) {
		pic = is->is_pic;
		if (pic && pic->pic_ops->pic_get_affinity)
			pic->pic_ops->pic_get_affinity(pic, is->is_irq, cpuset);
	}
	mutex_exit(&cpu_lock);
}

int
interrupt_distribute_handler(const char *intrid, const kcpuset_t *newset,
    kcpuset_t *oldset)
{
	struct intrsource *is;
	int error;

	mutex_enter(&cpu_lock);
	is = intr_get_source(intrid);
	if (is == NULL) {
		error = ENOENT;
	} else {
		error = interrupt_distribute(is, newset, oldset);
	}
	mutex_exit(&cpu_lock);

	return error;
}

int
interrupt_distribute(void *ih, const kcpuset_t *newset, kcpuset_t *oldset)
{
	struct intrsource * const is = ih;
	struct pic_softc * const pic = is->is_pic;

	if (pic == NULL)
		return EOPNOTSUPP;
	if (pic->pic_ops->pic_set_affinity == NULL ||
	    pic->pic_ops->pic_get_affinity == NULL)
		return EOPNOTSUPP;

	if (!is->is_mpsafe)
		return EINVAL;

	if (oldset != NULL)
		pic->pic_ops->pic_get_affinity(pic, is->is_irq, oldset);

	return pic->pic_ops->pic_set_affinity(pic, is->is_irq, newset);
}
#endif