- Add the interrupt vector number to intr_event_create so MI code can

   lookup hard interrupt events by number.  Ignore the irq# for soft intrs.
 - Add support to cpuset for binding hardware interrupts.  This has the
   side effect of binding any ithread associated with the hard interrupt.
   As per restrictions imposed by MD code we can only bind interrupts to
   a single cpu presently.  Interrupts can be 'unbound' by binding them
   to all cpus.

Reviewed by:	jhb
Sponsored by:	Nokia

11 files changed, 127 insertions, 69 deletions

diff --git a/sys/amd64/amd64/intr_machdep.c b/sys/amd64/amd64/intr_machdep.c
index 0f57958..43d4032 100644
--- a/sys/amd64/amd64/intr_machdep.c
+++ b/sys/amd64/amd64/intr_machdep.c
@@ -139,7 +139,7 @@ intr_register_source(struct intsrc *isrc)
 	vector = isrc->is_pic->pic_vector(isrc);
 	if (interrupt_sources[vector] != NULL)
 		return (EEXIST);
-	error = intr_event_create(&isrc->is_event, isrc, 0,
+	error = intr_event_create(&isrc->is_event, isrc, 0, vector,
 	    intr_disable_src, (mask_fn)isrc->is_pic->pic_enable_source,
 	    (mask_fn)isrc->is_pic->pic_eoi_source, intr_assign_cpu, "irq%d:",
 	    vector);
diff --git a/sys/arm/arm/intr.c b/sys/arm/arm/intr.c
index 5fbdce7..ba30bef 100644
--- a/sys/arm/arm/intr.c
+++ b/sys/arm/arm/intr.c
@@ -70,7 +70,7 @@ arm_setup_irqhandler(const char *name, driver_filter_t *filt,
 		return;
 	event = intr_events[irq];
 	if (event == NULL) {
-		error = intr_event_create(&event, (void *)irq, 0,
+		error = intr_event_create(&event, (void *)irq, 0, irq,
 		    (mask_fn)arm_mask_irq, (mask_fn)arm_unmask_irq,
 		    NULL, NULL, "intr%d:", irq);
 		if (error)
diff --git a/sys/i386/i386/intr_machdep.c b/sys/i386/i386/intr_machdep.c
index a2b8a1d..1c04560 100644
--- a/sys/i386/i386/intr_machdep.c
+++ b/sys/i386/i386/intr_machdep.c
@@ -130,7 +130,7 @@ intr_register_source(struct intsrc *isrc)
 	vector = isrc->is_pic->pic_vector(isrc);
 	if (interrupt_sources[vector] != NULL)
 		return (EEXIST);
-	error = intr_event_create(&isrc->is_event, isrc, 0,
+	error = intr_event_create(&isrc->is_event, isrc, 0, vector,
 	    intr_disable_src, (mask_fn)isrc->is_pic->pic_enable_source,
 	    (mask_fn)isrc->is_pic->pic_eoi_source, intr_assign_cpu, "irq%d:",
 	    vector);
diff --git a/sys/ia64/ia64/interrupt.c b/sys/ia64/ia64/interrupt.c
index c0e4ae3..81ad524 100644
--- a/sys/ia64/ia64/interrupt.c
+++ b/sys/ia64/ia64/interrupt.c
@@ -340,7 +340,7 @@ ia64_setup_intr(const char *name, int irq, driver_filter_t filter,
 			return (ENOMEM);
 
 		error = intr_event_create(&i->event, (void *)(uintptr_t)vector,
-		    0, ia64_intr_mask, ia64_intr_unmask, ia64_intr_eoi,
+		    0, vector, ia64_intr_mask, ia64_intr_unmask, ia64_intr_eoi,
 		    NULL, "irq%u:", irq);
 		if (error) {
 			free(i, M_DEVBUF);
diff --git a/sys/kern/kern_cpuset.c b/sys/kern/kern_cpuset.c
index 23d0510..22b6677 100644
--- a/sys/kern/kern_cpuset.c
+++ b/sys/kern/kern_cpuset.c
@@ -412,6 +412,8 @@ cpuset_which(cpuwhich_t which, id_t id, struct proc **pp, struct thread **tdp,
 			return (0);
 		}
 		return (ESRCH);
+	case CPU_WHICH_IRQ:
+		return (0);
 	default:
 		return (EINVAL);
 	}
@@ -760,6 +762,8 @@ cpuset_getid(struct thread *td, struct cpuset_getid_args *uap)
 		break;
 	case CPU_WHICH_CPUSET:
 		break;
+	case CPU_WHICH_IRQ:
+		return (EINVAL);
 	}
 	switch (uap->level) {
 	case CPU_LEVEL_ROOT:
@@ -820,6 +824,9 @@ cpuset_getaffinity(struct thread *td, struct cpuset_getaffinity_args *uap)
 			break;
 		case CPU_WHICH_CPUSET:
 			break;
+		case CPU_WHICH_IRQ:
+			error = EINVAL;
+			goto out;
 		}
 		if (uap->level == CPU_LEVEL_ROOT)
 			nset = cpuset_refroot(set);
@@ -845,6 +852,9 @@ cpuset_getaffinity(struct thread *td, struct cpuset_getaffinity_args *uap)
 		case CPU_WHICH_CPUSET:
 			CPU_COPY(&set->cs_mask, mask);
 			break;
+		case CPU_WHICH_IRQ:
+			error = intr_getaffinity(uap->id, mask);
+			break;
 		}
 		break;
 	default:
@@ -921,6 +931,9 @@ cpuset_setaffinity(struct thread *td, struct cpuset_setaffinity_args *uap)
 			break;
 		case CPU_WHICH_CPUSET:
 			break;
+		case CPU_WHICH_IRQ:
+			error = EINVAL;
+			goto out;
 		}
 		if (uap->level == CPU_LEVEL_ROOT)
 			nset = cpuset_refroot(set);
@@ -946,6 +959,9 @@ cpuset_setaffinity(struct thread *td, struct cpuset_setaffinity_args *uap)
 				cpuset_rel(set);
 			}
 			break;
+		case CPU_WHICH_IRQ:
+			error = intr_setaffinity(uap->id, mask);
+			break;
 		default:
 			error = EINVAL;
 			break;
diff --git a/sys/kern/kern_intr.c b/sys/kern/kern_intr.c
index 16bd42a..e626988 100644
--- a/sys/kern/kern_intr.c
+++ b/sys/kern/kern_intr.c
@@ -28,11 +28,11 @@
 __FBSDID("$FreeBSD$");
 
 #include "opt_ddb.h"
-#include "opt_intr_filter.h"
 
 #include <sys/param.h>
 #include <sys/bus.h>
 #include <sys/conf.h>
+#include <sys/cpuset.h>
 #include <sys/rtprio.h>
 #include <sys/systm.h>
 #include <sys/interrupt.h>
@@ -92,6 +92,8 @@ SYSCTL_INT(_hw, OID_AUTO, intr_storm_threshold, CTLFLAG_RW,
     "Number of consecutive interrupts before storm protection is enabled");
 static TAILQ_HEAD(, intr_event) event_list =
     TAILQ_HEAD_INITIALIZER(event_list);
+static struct mtx event_lock;
+MTX_SYSINIT(intr_event_list, &event_lock, "intr event list", MTX_DEF);
 
 static void	intr_event_update(struct intr_event *ie);
 #ifdef INTR_FILTER
@@ -244,7 +246,7 @@ intr_event_update(struct intr_event *ie)
 }
 
 int
-intr_event_create(struct intr_event **event, void *source,int flags,
+intr_event_create(struct intr_event **event, void *source,int flags, int irq,
     void (*pre_ithread)(void *), void (*post_ithread)(void *),
     void (*post_filter)(void *), int (*assign_cpu)(void *, u_char),
     const char *fmt, ...)
@@ -262,6 +264,7 @@ intr_event_create(struct intr_event **event, void *source,int flags,
 	ie->ie_post_filter = post_filter;
 	ie->ie_assign_cpu = assign_cpu;
 	ie->ie_flags = flags;
+	ie->ie_irq = irq;
 	ie->ie_cpu = NOCPU;
 	TAILQ_INIT(&ie->ie_handlers);
 	mtx_init(&ie->ie_lock, "intr event", NULL, MTX_DEF);
@@ -270,9 +273,9 @@ intr_event_create(struct intr_event **event, void *source,int flags,
 	vsnprintf(ie->ie_name, sizeof(ie->ie_name), fmt, ap);
 	va_end(ap);
 	strlcpy(ie->ie_fullname, ie->ie_name, sizeof(ie->ie_fullname));
-	mtx_pool_lock(mtxpool_sleep, &event_list);
+	mtx_lock(&event_lock);
 	TAILQ_INSERT_TAIL(&event_list, ie, ie_list);
-	mtx_pool_unlock(mtxpool_sleep, &event_list);
+	mtx_unlock(&event_lock);
 	if (event != NULL)
 		*event = ie;
 	CTR2(KTR_INTR, "%s: created %s", __func__, ie->ie_name);
@@ -290,7 +293,8 @@ intr_event_create(struct intr_event **event, void *source,int flags,
 int
 intr_event_bind(struct intr_event *ie, u_char cpu)
 {
-	struct thread *td;
+	cpuset_t mask;
+	lwpid_t id;
 	int error;
 
 	/* Need a CPU to bind to. */
@@ -299,28 +303,97 @@ intr_event_bind(struct intr_event *ie, u_char cpu)
 
 	if (ie->ie_assign_cpu == NULL)
 		return (EOPNOTSUPP);
-
-	/* Don't allow a bind request if the interrupt is already bound. */
+	/*
+	 * If we have any ithreads try to set their mask first since this
+	 * can fail.
+	 */
 	mtx_lock(&ie->ie_lock);
-	if (ie->ie_cpu != NOCPU && cpu != NOCPU) {
+	if (ie->ie_thread != NULL) {
+		CPU_ZERO(&mask);
+		if (cpu == NOCPU)
+			CPU_COPY(cpuset_root, &mask);
+		else
+			CPU_SET(cpu, &mask);
+		id = ie->ie_thread->it_thread->td_tid;
+		mtx_unlock(&ie->ie_lock);
+		error = cpuset_setthread(id, &mask);
+		if (error)
+			return (error);
+	} else
 		mtx_unlock(&ie->ie_lock);
-		return (EBUSY);
-	}
-	mtx_unlock(&ie->ie_lock);
-
 	error = ie->ie_assign_cpu(ie->ie_source, cpu);
 	if (error)
 		return (error);
 	mtx_lock(&ie->ie_lock);
-	if (ie->ie_thread != NULL)
-		td = ie->ie_thread->it_thread;
-	else
-		td = NULL;
-	if (td != NULL)
-		thread_lock(td);
 	ie->ie_cpu = cpu;
-	if (td != NULL)
-		thread_unlock(td);
+	mtx_unlock(&ie->ie_lock);
+
+	return (error);
+}
+
+static struct intr_event *
+intr_lookup(int irq)
+{
+	struct intr_event *ie;
+
+	mtx_lock(&event_lock);
+	TAILQ_FOREACH(ie, &event_list, ie_list)
+		if (ie->ie_irq == irq &&
+		    (ie->ie_flags & IE_SOFT) == 0 &&
+		    TAILQ_FIRST(&ie->ie_handlers) != NULL)
+			break;
+	mtx_unlock(&event_lock);
+	return (ie);
+}
+
+int
+intr_setaffinity(int irq, void *m)
+{
+	struct intr_event *ie;
+	cpuset_t *mask;
+	u_char cpu;
+	int error;
+	int n;
+
+	mask = m;
+	error = 0;
+	cpu = NOCPU;
+	/*
+	 * If we're setting all cpus we can unbind.  Otherwise make sure
+	 * only one cpu is in the set.
+	 */
+	if (CPU_CMP(cpuset_root, mask)) {
+		for (n = 0; n < CPU_SETSIZE; n++) {
+			if (!CPU_ISSET(n, mask))
+				continue;
+			if (cpu != NOCPU)
+				return (EINVAL);
+			cpu = (u_char)n;
+		}
+	}
+	ie = intr_lookup(irq);
+	if (ie == NULL)
+		return (ESRCH);
+	intr_event_bind(ie, cpu);
+	return (error);
+}
+
+int
+intr_getaffinity(int irq, void *m)
+{
+	struct intr_event *ie;
+	cpuset_t *mask;
+
+	mask = m;
+	ie = intr_lookup(irq);
+	if (ie == NULL)
+		return (ESRCH);
+	CPU_ZERO(mask);
+	mtx_lock(&ie->ie_lock);
+	if (ie->ie_cpu == NOCPU)
+		CPU_COPY(cpuset_root, mask);
+	else
+		CPU_SET(ie->ie_cpu, mask);
 	mtx_unlock(&ie->ie_lock);
 	return (0);
 }
@@ -329,14 +402,14 @@ int
 intr_event_destroy(struct intr_event *ie)
 {
 
+	mtx_lock(&event_lock);
 	mtx_lock(&ie->ie_lock);
 	if (!TAILQ_EMPTY(&ie->ie_handlers)) {
 		mtx_unlock(&ie->ie_lock);
+		mtx_unlock(&event_lock);
 		return (EBUSY);
 	}
-	mtx_pool_lock(mtxpool_sleep, &event_list);
 	TAILQ_REMOVE(&event_list, ie, ie_list);
-	mtx_pool_unlock(mtxpool_sleep, &event_list);
 #ifndef notyet
 	if (ie->ie_thread != NULL) {
 		ithread_destroy(ie->ie_thread);
@@ -344,6 +417,7 @@ intr_event_destroy(struct intr_event *ie)
 	}
 #endif
 	mtx_unlock(&ie->ie_lock);
+	mtx_unlock(&event_lock);
 	mtx_destroy(&ie->ie_lock);
 	free(ie, M_ITHREAD);
 	return (0);
@@ -914,7 +988,7 @@ swi_add(struct intr_event **eventp, const char *name, driver_intr_t handler,
 		if (!(ie->ie_flags & IE_SOFT))
 			return (EINVAL);
 	} else {
-		error = intr_event_create(&ie, NULL, IE_SOFT,
+		error = intr_event_create(&ie, NULL, IE_SOFT, 0,
 		    NULL, NULL, NULL, NULL, "swi%d:", pri);
 		if (error)
 			return (error);
@@ -1106,7 +1180,6 @@ ithread_loop(void *arg)
 	struct intr_event *ie;
 	struct thread *td;
 	struct proc *p;
-	u_char cpu;
 
 	td = curthread;
 	p = td->td_proc;
@@ -1115,7 +1188,6 @@ ithread_loop(void *arg)
 	    ("%s: ithread and proc linkage out of sync", __func__));
 	ie = ithd->it_event;
 	ie->ie_count = 0;
-	cpu = NOCPU;
 
 	/*
 	 * As long as we have interrupts outstanding, go through the
@@ -1161,21 +1233,6 @@ ithread_loop(void *arg)
 			ie->ie_count = 0;
 			mi_switch(SW_VOL, NULL);
 		}
-
-#ifdef SMP
-		/*
-		 * Ensure we are bound to the correct CPU.  We can't
-		 * move ithreads until SMP is running however, so just
-		 * leave interrupts on the boor CPU during boot.
-		 */
-		if (ie->ie_cpu != cpu && smp_started) {
-			cpu = ie->ie_cpu;
-			if (cpu == NOCPU)
-				sched_unbind(td);
-			else
-				sched_bind(td, cpu);
-		}
-#endif
 		thread_unlock(td);
 	}
 }
@@ -1276,7 +1333,6 @@ ithread_loop(void *arg)
 	struct thread *td;
 	struct proc *p;
 	int priv;
-	u_char cpu;
 
 	td = curthread;
 	p = td->td_proc;
@@ -1287,7 +1343,6 @@ ithread_loop(void *arg)
 	    ("%s: ithread and proc linkage out of sync", __func__));
 	ie = ithd->it_event;
 	ie->ie_count = 0;
-	cpu = NOCPU;
 
 	/*
 	 * As long as we have interrupts outstanding, go through the
@@ -1336,21 +1391,6 @@ ithread_loop(void *arg)
 			ie->ie_count = 0;
 			mi_switch(SW_VOL, NULL);
 		}
-
-#ifdef SMP
-		/*
-		 * Ensure we are bound to the correct CPU.  We can't
-		 * move ithreads until SMP is running however, so just
-		 * leave interrupts on the boor CPU during boot.
-		 */
-		if (!priv && ie->ie_cpu != cpu && smp_started) {
-			cpu = ie->ie_cpu;
-			if (cpu == NOCPU)
-				sched_unbind(td);
-			else
-				sched_bind(td, cpu);
-		}
-#endif
 		thread_unlock(td);
 	}
 }
@@ -1580,8 +1620,6 @@ db_dump_intr_event(struct intr_event *ie, int handlers)
 		db_printf("(pid %d)", it->it_thread->td_proc->p_pid);
 	else
 		db_printf("(no thread)");
-	if (ie->ie_cpu != NOCPU)
-		db_printf(" (CPU %d)", ie->ie_cpu);
 	if ((ie->ie_flags & (IE_SOFT | IE_ENTROPY | IE_ADDING_THREAD)) != 0 ||
 	    (it != NULL && it->it_need)) {
 		db_printf(" {");
diff --git a/sys/powerpc/powerpc/intr_machdep.c b/sys/powerpc/powerpc/intr_machdep.c
index 5f1a651..d9d7591 100644
--- a/sys/powerpc/powerpc/intr_machdep.c
+++ b/sys/powerpc/powerpc/intr_machdep.c
@@ -226,7 +226,7 @@ powerpc_setup_intr(const char *name, u_int irq, driver_filter_t filter,
 		return (ENOMEM);
 
 	if (i->event == NULL) {
-		error = intr_event_create(&i->event, (void *)irq, 0,
+		error = intr_event_create(&i->event, (void *)irq, 0, irq,
 		    powerpc_intr_mask, powerpc_intr_unmask, powerpc_intr_eoi,
 		    NULL, "irq%u:", irq);
 		if (error)
diff --git a/sys/sparc64/sparc64/intr_machdep.c b/sys/sparc64/sparc64/intr_machdep.c
index f2dfa48..c771b6a 100644
--- a/sys/sparc64/sparc64/intr_machdep.c
+++ b/sys/sparc64/sparc64/intr_machdep.c
@@ -282,7 +282,7 @@ intr_controller_register(int vec, const struct intr_controller *ic,
 	 * CPU as long as the source of a level sensitive interrupt is
 	 * not cleared.
 	 */
-	error = intr_event_create(&ie, iv, 0, NULL, intr_enable_eoi,
+	error = intr_event_create(&ie, iv, 0, vec, NULL, intr_enable_eoi,
 	    intr_enable_eoi, NULL, "vec%d:", vec);
 	if (error != 0)
 		return (error);
diff --git a/sys/sun4v/sun4v/intr_machdep.c b/sys/sun4v/sun4v/intr_machdep.c
index 9dcfd3e..4192e0b 100644
--- a/sys/sun4v/sun4v/intr_machdep.c
+++ b/sys/sun4v/sun4v/intr_machdep.c
@@ -307,8 +307,8 @@ inthand_add(const char *name, int vec, driver_filter_t *filt,
 	ie = iv->iv_event;
 	mtx_unlock_spin(&intr_table_lock);
 	if (ie == NULL) {
-		errcode = intr_event_create(&ie, (void *)(intptr_t)vec, 0, NULL,
-		    intr_enable, intr_enable, NULL, "vec%d:", vec);
+		errcode = intr_event_create(&ie, (void *)(intptr_t)vec, 0, vec,
+		    NULL, intr_enable, intr_enable, NULL, "vec%d:", vec);
 		if (errcode)
 			return (errcode);
 		mtx_lock_spin(&intr_table_lock);
diff --git a/sys/sys/cpuset.h b/sys/sys/cpuset.h
index 581a1e7..c4bda75 100644
--- a/sys/sys/cpuset.h
+++ b/sys/sys/cpuset.h
@@ -131,6 +131,7 @@ typedef	struct _cpuset {
 #define	CPU_WHICH_TID		1	/* Specifies a thread id. */
 #define	CPU_WHICH_PID		2	/* Specifies a process id. */
 #define	CPU_WHICH_CPUSET	3	/* Specifies a set id. */
+#define	CPU_WHICH_IRQ		4	/* Specifies an irq #. */
 
 /*
  * Reserved cpuset identifiers.
diff --git a/sys/sys/interrupt.h b/sys/sys/interrupt.h
index f47bb26..be5d0f4 100644
--- a/sys/sys/interrupt.h
+++ b/sys/sys/interrupt.h
@@ -107,6 +107,7 @@ struct intr_event {
 	int		ie_count;	/* Loop counter. */
 	int		ie_warncnt;	/* Rate-check interrupt storm warns. */
 	struct timeval	ie_warntm;
+	int		ie_irq;		/* Physical irq number if !SOFT. */
 	u_char		ie_cpu;		/* CPU this event is bound to. */
 };
 
@@ -151,14 +152,16 @@ int	intr_event_add_handler(struct intr_event *ie, const char *name,
 	    u_char pri, enum intr_type flags, void **cookiep);	    
 int	intr_event_bind(struct intr_event *ie, u_char cpu);
 int	intr_event_create(struct intr_event **event, void *source,
-	    int flags, void (*pre_ithread)(void *),
+	    int flags, int irq, void (*pre_ithread)(void *),
 	    void (*post_ithread)(void *), void (*post_filter)(void *),
 	    int (*assign_cpu)(void *, u_char), const char *fmt, ...)
-	    __printflike(8, 9);
+	    __printflike(9, 10);
 int	intr_event_destroy(struct intr_event *ie);
 int	intr_event_handle(struct intr_event *ie, struct trapframe *frame);
 int	intr_event_remove_handler(void *cookie);
+int	intr_getaffinity(int irq, void *mask);
 void	*intr_handler_source(void *cookie);
+int	intr_setaffinity(int irq, void *mask);
 int	swi_add(struct intr_event **eventp, const char *name,
 	    driver_intr_t handler, void *arg, int pri, enum intr_type flags,
 	    void **cookiep);