pc, pc64, xen: rewrite interrupt handling code

This implements proper intrdisable() support for all
interrupt controllers.

For enable, (*arch->intrassign)(Vctl*) fills in the
Vctl.enable and Vctl.disable pointers with the
appropriate routines and returns the assigned
vector number.

Once the Vctl struct has been linked to its vector
chain, Vctl.enable(Vctl*, shared) gets called with a
flag if the vector has been already enabled (shared).

This order is important here as enabling the interrupt
on the controller before we have linked the chain can
cause spurious interrupts, expecially on mp system
where the interrupt can target a different cpu than
the caller of intrenable().

The intrdisable() case is the other way around.
We first disable the interrupt on the controller
and after that unlink the Vctl from the chain.
On a multiprocessor, the xfree() of the Vctl struct
is delayed to avoid freeing it while it is still
in use by another cpu.

The xen port now also uses pc/irq.c which has been
made generic enougth to handle xen's irq scheme.
Also, archgeneric is now a separate file to avoid
pulling in dependencies from the 8259 interrupt
controller code.

sys/src/9/pc/archacpi.c

@@ -256,6 +256,7 @@ Foundbus:
 	ai->intr = pi;
 	ai->apic = a;
 	ai->next = bus->aintr;
+	ai->bus = bus;
 	bus->aintr = ai;
 }
 
@@ -733,13 +734,15 @@ acpireset(void)
 }
 
 static int identify(void);
+extern int i8259irqno(int, int);
 
 PCArch archacpi = {
 .id=		"ACPI",	
 .ident=		identify,
 .reset=		acpireset,
 .intrinit=	acpiinit,
-.intrenable=	mpintrenable,
+.intrassign=	mpintrassign,
+.intrirqno=	i8259irqno,
 .intron=	lapicintron,
 .introff=	lapicintroff,
 .fastclock=	i8253read,

sys/src/9/pc/archgeneric.c

@@ -0,0 +1,59 @@
+#include "u.h"
+#include "../port/lib.h"
+#include "mem.h"
+#include "dat.h"
+#include "fns.h"
+#include "io.h"
+#include "ureg.h"
+#include "../port/error.h"
+
+extern int i8259assign(Vctl*);
+extern int i8259irqno(int, int);
+extern void i8259init(void);
+extern int i8259isr(int);
+extern void i8259on(void);
+extern void i8259off(void);
+extern int i8259vecno(int);
+
+void
+archreset(void)
+{
+	i8042reset();
+
+	/*
+	 * Often the BIOS hangs during restart if a conventional 8042
+	 * warm-boot sequence is tried. The following is Intel specific and
+	 * seems to perform a cold-boot, but at least it comes back.
+	 * And sometimes there is no keyboard...
+	 *
+	 * The reset register (0xcf9) is usually in one of the bridge
+	 * chips. The actual location and sequence could be extracted from
+	 * ACPI but why bother, this is the end of the line anyway.
+	 */
+	print("Takes a licking and keeps on ticking...\n");
+	*(ushort*)KADDR(0x472) = 0x1234;	/* BIOS warm-boot flag */
+	outb(0xcf9, 0x02);
+	outb(0xcf9, 0x06);
+
+	print("can't reset\n");
+	for(;;)
+		idle();
+}
+
+PCArch archgeneric = {
+.id=		"generic",
+.ident=		0,
+.reset=		archreset,
+
+.intrinit=	i8259init,
+.intrassign=	i8259assign,
+.intrirqno=	i8259irqno,
+.intrvecno=	i8259vecno,
+.intrspurious=	i8259isr,
+.intron=	i8259on,
+.introff=	i8259off,
+
+.clockenable=	i8253enable,
+.fastclock=	i8253read,
+.timerset=	i8253timerset,
+};

sys/src/9/pc/archmp.c

@@ -194,6 +194,7 @@ mkiointr(PCMPintr* p)
 	}
 	aintr->apic = mpioapic[p->apicno];
 	aintr->next = bus->aintr;
+	aintr->bus = bus;
 	bus->aintr = aintr;
 
 	return aintr;
@@ -366,13 +367,15 @@ mpreset(void)
 }
 
 static int identify(void);
+extern int i8259irqno(int, int);
 
 PCArch archmp = {
 .id=		"_MP_",	
 .ident=		identify,
 .reset=		mpreset,
 .intrinit=	pcmpinit,
-.intrenable=	mpintrenable,
+.intrassign=	mpintrassign,
+.intrirqno=	i8259irqno,
 .intron=	lapicintron,
 .introff=	lapicintroff,
 .fastclock=	i8253read,

sys/src/9/pc/dat.h

@@ -280,13 +280,12 @@ struct PCArch
 	char*	id;
 	int	(*ident)(void);		/* this should be in the model */
 	void	(*reset)(void);		/* this should be in the model */
-	int	(*serialpower)(int);	/* 1 == on, 0 == off */
-	int	(*modempower)(int);	/* 1 == on, 0 == off */
 
 	void	(*intrinit)(void);
-	int	(*intrenable)(Vctl*);
+	int	(*intrassign)(Vctl*);
+	int	(*intrirqno)(int, int);
+	int	(*intrspurious)(int);
 	int	(*intrvecno)(int);
-	int	(*intrdisable)(int);
 	void	(*introff)(void);
 	void	(*intron)(void);
 

sys/src/9/pc/devarch.c

@@ -324,31 +324,6 @@ nop(void)
 {
 }
 
-void
-archreset(void)
-{
-	i8042reset();
-
-	/*
-	 * Often the BIOS hangs during restart if a conventional 8042
-	 * warm-boot sequence is tried. The following is Intel specific and
-	 * seems to perform a cold-boot, but at least it comes back.
-	 * And sometimes there is no keyboard...
-	 *
-	 * The reset register (0xcf9) is usually in one of the bridge
-	 * chips. The actual location and sequence could be extracted from
-	 * ACPI but why bother, this is the end of the line anyway.
-	 */
-	print("Takes a licking and keeps on ticking...\n");
-	*(ushort*)KADDR(0x472) = 0x1234;	/* BIOS warm-boot flag */
-	outb(0xcf9, 0x02);
-	outb(0xcf9, 0x06);
-
-	print("can't reset\n");
-	for(;;)
-		idle();
-}
-
 /*
  * 386 has no compare-and-swap instruction.
  * Run it with interrupts turned off instead.
@@ -380,25 +355,6 @@ int (*cmpswap)(long*, long, long) = cmpswap386;
 PCArch* arch;
 extern PCArch* knownarch[];
 
-PCArch archgeneric = {
-.id=		"generic",
-.ident=		0,
-.reset=		archreset,
-.serialpower=	unimplemented,
-.modempower=	unimplemented,
-
-.intrinit=	i8259init,
-.intrenable=	i8259enable,
-.intrvecno=	i8259vecno,
-.intrdisable=	i8259disable,
-.intron=	i8259on,
-.introff=	i8259off,
-
-.clockenable=	i8253enable,
-.fastclock=	i8253read,
-.timerset=	i8253timerset,
-};
-
 typedef struct X86type X86type;
 struct X86type {
 	int	family;
@@ -912,26 +868,22 @@ archinit(void)
 {
 	PCArch **p;
 
-	arch = &archgeneric;
+	arch = knownarch[0];
 	for(p = knownarch; *p != nil; p++){
 		if((*p)->ident != nil && (*p)->ident() == 0){
 			arch = *p;
 			break;
 		}
 	}
-	if(arch != &archgeneric){
+	if(arch != knownarch[0]){
 		if(arch->id == nil)
-			arch->id = archgeneric.id;
+			arch->id = knownarch[0]->id;
 		if(arch->reset == nil)
-			arch->reset = archgeneric.reset;
+			arch->reset = knownarch[0]->reset;
-		if(arch->serialpower == nil)
-			arch->serialpower = archgeneric.serialpower;
-		if(arch->modempower == nil)
-			arch->modempower = archgeneric.modempower;
 		if(arch->intrinit == nil)
-			arch->intrinit = archgeneric.intrinit;
+			arch->intrinit = knownarch[0]->intrinit;
-		if(arch->intrenable == nil)
+		if(arch->intrassign == nil)
-			arch->intrenable = archgeneric.intrenable;
+			arch->intrassign = knownarch[0]->intrassign;
 	}
 
 	/*
@@ -1100,6 +1052,33 @@ dumpmcregs(void)
 	}
 }
 
+static void
+nmihandler(Ureg *ureg, void*)
+{
+	iprint("cpu%d: nmi PC %#p, status %ux\n",
+		m->machno, ureg->pc, inb(0x61));
+	while(m->machno != 0)
+		;
+}
+
+void
+nmienable(void)
+{
+	int x;
+
+	trapenable(VectorNMI, nmihandler, nil, "nmi");
+
+	/*
+	 * Hack: should be locked with NVRAM access.
+	 */
+	outb(0x70, 0x80);		/* NMI latch clear */
+	outb(0x70, 0);
+
+	x = inb(0x61) & 0x07;		/* Enable NMI */
+	outb(0x61, 0x0C|x);
+	outb(0x61, x);
+}
+
 void
 setupwatchpts(Proc *pr, Watchpt *wp, int nwp)
 {

sys/src/9/pc/fns.h

@@ -65,13 +65,6 @@ void	i8253init(void);
 void	i8253reset(void);
 uvlong	i8253read(uvlong*);
 void	i8253timerset(uvlong);
-int	i8259disable(int);
-int	i8259enable(Vctl*);
-void	i8259init(void);
-int	i8259isr(int);
-void	i8259on(void);
-void	i8259off(void);
-int	i8259vecno(int);
 void	idle(void);
 void	idlehands(void);
 int	inb(int);
@@ -111,6 +104,7 @@ char*	mtrr(uvlong, uvlong, char *);
 void	mtrrclock(void);
 int	mtrrprint(char *, long);
 void	mtrrsync(void);
+void	nmienable(void);
 uchar	nvramread(int);
 void	nvramwrite(int, uchar);
 void	outb(int, int);

sys/src/9/pc/i8259.c

@@ -127,8 +127,38 @@ i8259isr(int vno)
 	return isr & (1<<irq);
 }
 
+static int
+irqenable(Vctl *v, int shared)
+{
+	if(shared)
+		return 0;
+	ilock(&i8259lock);
+	i8259mask &= ~(1<<v->irq);
+	if(v->irq < 8)
+		outb(Int0aux, i8259mask & 0xFF);
+	else
+		outb(Int1aux, (i8259mask>>8) & 0xFF);
+	iunlock(&i8259lock);
+	return 0;
+}
+
+static int
+irqdisable(Vctl *v, int shared)
+{
+	if(shared)
+		return 0;
+	ilock(&i8259lock);
+	i8259mask |= 1<<v->irq;
+	if(v->irq < 8)
+		outb(Int0aux, i8259mask & 0xFF);
+	else
+		outb(Int1aux, (i8259mask>>8) & 0xFF);
+	iunlock(&i8259lock);
+	return 0;
+}
+
 int
-i8259enable(Vctl* v)
+i8259assign(Vctl *v)
 {
 	int irq, irqbit;
 
@@ -150,54 +180,42 @@ i8259enable(Vctl* v)
 		iunlock(&i8259lock);
 		return -1;
 	}
-	i8259mask &= ~irqbit;
+	iunlock(&i8259lock);
-	if(irq < 8)
-		outb(Int0aux, i8259mask & 0xFF);
-	else
-		outb(Int1aux, (i8259mask>>8) & 0xFF);
 
 	if(i8259elcr & irqbit)
 		v->eoi = i8259isr;
 	else
 		v->isr = i8259isr;
-	iunlock(&i8259lock);
+
+	v->enable = irqenable;
+	v->disable = irqdisable;
 
 	return VectorPIC+irq;
 }
 
+int
+i8259irqno(int irq, int tbdf)
+{
+	if(tbdf != BUSUNKNOWN && (irq == 0xff || irq == 0))
+		return -1;
+
+	/*
+	 * IRQ2 doesn't really exist, it's used to gang the interrupt
+	 * controllers together. A device set to IRQ2 will appear on
+	 * the second interrupt controller as IRQ9.
+	 */
+	if(irq == 2)
+		irq = 9;
+
+	return irq;
+}
+
 int
 i8259vecno(int irq)
 {
 	return VectorPIC+irq;
 }
 
-int
-i8259disable(int irq)
-{
-	int irqbit;
-
-	/*
-	 * Given an IRQ, disable the corresponding interrupt
-	 * in the 8259.
-	 */
-	if(irq < 0 || irq > MaxIrqPIC){
-		print("i8259disable: irq %d out of range\n", irq);
-		return -1;
-	}
-	irqbit = 1<<irq;
-
-	ilock(&i8259lock);
-	if(!(i8259mask & irqbit)){
-		i8259mask |= irqbit;
-		if(irq < 8)
-			outb(Int0aux, i8259mask & 0xFF);
-		else
-			outb(Int1aux, (i8259mask>>8) & 0xFF);
-	}
-	iunlock(&i8259lock);
-	return 0;
-}
-
 void
 i8259on(void)
 {

sys/src/9/pc/io.h

@@ -46,19 +46,23 @@ enum {
 };
 
 typedef struct Vctl {
-	Vctl*	next;			/* handlers on this vector */
+	Vctl	*next;			/* handlers on this vector */
 
 	void	(*f)(Ureg*, void*);	/* handler to call */
-	void*	a;			/* argument to call it with */
+	void	*a;			/* argument to call it with */
-
-	int	isintr;			/* interrupt or fault/trap */
 
 	int	(*isr)(int);		/* get isr bit for this irq */
 	int	(*eoi)(int);		/* eoi */
 
-	void	(*disable)(Vctl*);
+	int	(*enable)(Vctl*, int);
+	int	(*disable)(Vctl*, int);
+	void	*aux;
+
 	int	irq;
 	int	tbdf;
+	int	vno;
+	int	cpu;
+	int	local;
 
 	char	name[KNAMELEN];		/* of driver */
 } Vctl;

sys/src/9/pc/irq.c

@@ -9,7 +9,7 @@
 #include	"../port/error.h"
 
 static Lock vctllock;
-static Vctl *vctl[256];
+static Vctl *vclock, *vctl[256];
 
 enum
 {
@@ -46,86 +46,55 @@ irqhandled(Ureg *ureg, int vno)
 	Vctl *ctl, *v;
 	int i;
 
-	if(ctl = vctl[vno]){
+	ctl = vctl[vno];
-		if(ctl->isintr){
+	if(ctl != nil){
-			m->perf.intrts = perfticks();
+		if(vno < VectorPIC){
-			m->intr++;
+			(*ctl->f)(ureg, ctl->a);
-			if(vno >= VectorPIC)
+			return 1;
-				m->lastintr = ctl->irq;
 		}
-		if(ctl->isr)
-			ctl->isr(vno);
-		for(v = ctl; v != nil; v = v->next){
-			if(v->f)
-				v->f(ureg, v->a);
-		}
-		if(ctl->eoi)
-			ctl->eoi(vno);
 
-		if(ctl->isintr){
+		m->perf.intrts = perfticks();
-			intrtime(m, vno);
+		m->intr++;
+		m->lastintr = ctl->irq;
+		if(ctl->isr != nil)
+			(*ctl->isr)(vno);
+		for(v = ctl; v != nil; v = v->next)
+			(*v->f)(ureg, v->a);
+		if(ctl->eoi != nil)
+			(*ctl->eoi)(vno);
+		intrtime(m, vno);
 
-			if(up){
+		if(up != nil){
-				if(ctl->irq == IrqCLOCK || ctl->irq == IrqTIMER){
+			if(ctl == vclock){
-					/* delaysched set because we held a lock or because our quantum ended */
+				/* delaysched set because we held a lock or because our quantum ended */
-					if(up->delaysched)
+				if(up->delaysched)
-						sched();
+					sched();
-				} else {
+			} else {
-					preempted();
+				preempted();
-				}
 			}
 		}
 		return 1;
 	}
 
-	if(vno < VectorPIC)
+	if(vno < VectorPIC || vno == VectorSYSCALL)
 		return 0;
 
-	/*
-	 * An unknown interrupt.
-	 * Check for a default IRQ7. This can happen when
-	 * the IRQ input goes away before the acknowledge.
-	 * In this case, a 'default IRQ7' is generated, but
-	 * the corresponding bit in the ISR isn't set.
-	 * In fact, just ignore all such interrupts.
-	 */
-
-	/* call all interrupt routines, just in case */
-	for(i = VectorPIC; i <= MaxIrqLAPIC; i++){
-		ctl = vctl[i];
-		if(ctl == nil)
-			continue;
-		if(!ctl->isintr)
-			continue;
-		for(v = ctl; v != nil; v = v->next){
-			if(v->f)
-				v->f(ureg, v->a);
-		}
-		/* should we do this? */
-		if(ctl->eoi)
-			ctl->eoi(i);
-	}
-
-	/* clear the interrupt */
-	i8259isr(vno);
-
-	if(0)print("cpu%d: spurious interrupt %d, last %d\n",
-		m->machno, vno, m->lastintr);
-	if(0)if(conf.nmach > 1){
-		for(i = 0; i < MAXMACH; i++){
-			Mach *mach;
-
-			if(active.machs[i] == 0)
-				continue;
-			mach = MACHP(i);
-			if(m->machno == mach->machno)
-				continue;
-			print(" cpu%d: last %d",
-				mach->machno, mach->lastintr);
-		}
-		print("\n");
-	}
 	m->spuriousintr++;
+	if(arch->intrspurious != nil && (*arch->intrspurious)(vno) == 0)
+		return 1;	/* false alarm */
+
+	iprint("cpu%d: spurious interrupt %d, last %d\n",
+		m->machno, vno, m->lastintr);
+
+	/* call all non-local interrupt routines, just in case */
+	for(i = VectorPIC; i < nelem(vctl); i++){
+		ctl = vctl[i];
+		if(ctl == nil || ctl == vclock || ctl->local)
+			continue;
+		for(v = ctl; v != nil; v = v->next)
+			(*v->f)(ureg, v->a);
+	}
+
 	return -1;
 }
 
@@ -134,28 +103,59 @@ trapenable(int vno, void (*f)(Ureg*, void*), void* a, char *name)
 {
 	Vctl *v;
 
-	if(vno < 0 || vno >= VectorPIC)
+	if(f == nil){
-		panic("trapenable: vno %d", vno);
+		print("trapenable: nil handler for %d, for %s\n",
+			vno, name);
+		return;
+	}
+
+	if(vno < 0 || vno >= VectorPIC){
+		print("trapenable: vno %d out of range", vno);
+		return;
+	}
+
 	if((v = xalloc(sizeof(Vctl))) == nil)
 		panic("trapenable: out of memory");
-	v->tbdf = BUSUNKNOWN;
+
 	v->f = f;
 	v->a = a;
+
+	v->tbdf = BUSUNKNOWN;
+	v->irq = -1;
+	v->vno = vno;
+	v->cpu = -1;
+
 	strncpy(v->name, name, KNAMELEN-1);
 	v->name[KNAMELEN-1] = 0;
 
 	ilock(&vctllock);
-	if(vctl[vno])
+	if(vctl[vno] != nil){
-		v->next = vctl[vno]->next;
+		print("trapenable: vno %d assigned twice: %s %s\n",
+			vno, vctl[vno]->name, v->name);
+		iunlock(&vctllock);
+		xfree(v);
+		return;
+	}
 	vctl[vno] = v;
 	iunlock(&vctllock);
 }
 
+static Vctl*
+delayfree(Vctl *v)
+{
+	static Vctl *q[4];
+	static uint x;
+	Vctl *r;
+
+	r = q[x], q[x] = v;
+	x = (x+1) % nelem(q);
+	return r;
+}
+
 void
 intrenable(int irq, void (*f)(Ureg*, void*), void* a, int tbdf, char *name)
 {
-	int vno;
+	Vctl **pv, *v;
-	Vctl *v;
 
 	if(f == nil){
 		print("intrenable: nil handler for %d, tbdf 0x%uX for %s\n",
@@ -163,45 +163,63 @@ intrenable(int irq, void (*f)(Ureg*, void*), void* a, int tbdf, char *name)
 		return;
 	}
 
-	if(tbdf != BUSUNKNOWN && (irq == 0xff || irq == 0))
+	if(arch->intrirqno != nil)
-		irq = -1;
+		irq = (*arch->intrirqno)(irq, tbdf);
-
-
-	/*
-	 * IRQ2 doesn't really exist, it's used to gang the interrupt
-	 * controllers together. A device set to IRQ2 will appear on
-	 * the second interrupt controller as IRQ9.
-	 */
-	if(irq == 2)
-		irq = 9;
 
 	if((v = xalloc(sizeof(Vctl))) == nil)
 		panic("intrenable: out of memory");
-	v->isintr = 1;
+
-	v->irq = irq;
-	v->tbdf = tbdf;
 	v->f = f;
 	v->a = a;
+
+	v->tbdf = tbdf;
+	v->irq = irq;
+	v->vno = -1;
+	v->cpu = -1;
+
 	strncpy(v->name, name, KNAMELEN-1);
 	v->name[KNAMELEN-1] = 0;
 
 	ilock(&vctllock);
-	vno = arch->intrenable(v);
+	v->vno = (*arch->intrassign)(v);
-	if(vno == -1){
+	if(v->vno < VectorPIC || v->vno >= nelem(vctl)){
-		iunlock(&vctllock);
+		print("intrenable: couldn't assign irq %d, tbdf 0x%uX for %s\n",
-		print("intrenable: couldn't enable irq %d, tbdf 0x%uX for %s\n",
 			irq, tbdf, v->name);
-		xfree(v);
+Unlockandfree:
+		iunlock(&vctllock);
+		if(v != nil)
+			xfree(v);
 		return;
 	}
-	if(vctl[vno]){
+	pv = &vctl[v->vno];
-		if(vctl[vno]->isr != v->isr || vctl[vno]->eoi != v->eoi)
+	if(*pv != nil){
-			panic("intrenable: handler: %s %s %#p %#p %#p %#p",
+		if((*pv)->isr != v->isr || (*pv)->eoi != v->eoi){
-				vctl[vno]->name, v->name,
+			print("intrenable: incompatible handler: %s %s %#p %#p %#p %#p\n",
-				vctl[vno]->isr, v->isr, vctl[vno]->eoi, v->eoi);
+				(*pv)->name, v->name,
-		v->next = vctl[vno];
+				(*pv)->isr, v->isr,
+				(*pv)->eoi, v->eoi);
+			goto Unlockandfree;
+		}
+		if(*pv == vclock)
+			pv = &vclock->next;
+		v->next = *pv;
+	}
+	if(strcmp(name, "clock") == 0)
+		vclock = v;
+	*pv = v;
+	if(v->enable != nil){
+		coherence();
+		if((*v->enable)(v, pv != &vctl[v->vno] || v->next != nil) < 0){
+			print("intrenable: couldn't enable irq %d, tbdf 0x%uX for %s\n",
+				irq, tbdf, v->name);
+			*pv = v->next;
+			if(v == vclock)
+				vclock = nil;
+			if(conf.nmach > 1)
+				v = delayfree(v);
+			goto Unlockandfree;
+		}
 	}
-	vctl[vno] = v;
 	iunlock(&vctllock);
 }
 
@@ -211,40 +229,47 @@ intrdisable(int irq, void (*f)(Ureg *, void *), void *a, int tbdf, char *name)
 	Vctl **pv, *v;
 	int vno;
 
-	if(irq == 2)
+	if(arch->intrirqno != nil)
-		irq = 9;
+		irq = (*arch->intrirqno)(irq, tbdf);
-	if(arch->intrvecno == nil || (tbdf != BUSUNKNOWN && (irq == 0xff || irq == 0))){
+
-		/*
+	if(irq != -1 && arch->intrvecno != nil) {
-		 * on APIC machine, irq is pretty meaningless
+		vno = (*arch->intrvecno)(irq);
-		 * and disabling a the vector is not implemented.
+		if(vno < VectorPIC || vno >= nelem(vctl)){
-		 * however, we still want to remove the matching
+			irq = -1;
-		 * Vctl entry to prevent calling Vctl.f() with a
+			vno = VectorPIC;
-		 * stale Vctl.a pointer.
+		}
-		 */
+	} else {
 		irq = -1;
 		vno = VectorPIC;
-	} else {
-		vno = arch->intrvecno(irq);
 	}
+
 	ilock(&vctllock);
 	do {
 		for(pv = &vctl[vno]; (v = *pv) != nil; pv = &v->next){
-			if(v->isintr && (v->irq == irq || irq == -1)
+			if((v->irq == irq || irq == -1)
 			&& v->tbdf == tbdf && v->f == f && v->a == a
 			&& strcmp(v->name, name) == 0)
 				break;
 		}
 		if(v != nil){
-			if(v->disable != nil)
+			if(v->disable != nil){
-				(*v->disable)(v);
+				if((*v->disable)(v, pv != &vctl[vno] || v->next != nil) < 0){
+					print("intrdisable: couldn't disable irq %d, tbdf 0x%uX for %s\n",
+						irq, tbdf, name);
+				}
+				coherence();
+			}
 			*pv = v->next;
-			xfree(v);
+			if(v == vclock)
-
+				vclock = nil;
-			if(irq != -1 && vctl[vno] == nil && arch->intrdisable != nil)
+			if(conf.nmach > 1)
-				arch->intrdisable(irq);
+				v = delayfree(v);
-			break;
+			iunlock(&vctllock);
+			if(v != nil)
+				xfree(v);
+			return;
 		}
-	} while(irq == -1 && ++vno <= MaxVectorAPIC);
+	} while(irq == -1 && ++vno < nelem(vctl));
 	iunlock(&vctllock);
 }
 
@@ -258,8 +283,8 @@ irqallocread(Chan*, void *a, long n, vlong offset)
 	if(n < 0 || offset < 0)
 		error(Ebadarg);
 
-	for(vno=0; vno<nelem(vctl); vno++){
+	for(vno = 0; vno < nelem(vctl); vno++){
-		for(v=vctl[vno]; v; v=v->next){
+		for(v = vctl[vno]; v != nil; v = v->next){
 			m = snprint(buf, sizeof(buf), "%11d %11d %.*s\n", vno, v->irq, KNAMELEN, v->name);
 			offset -= m;
 			if(offset >= 0)
@@ -274,40 +299,8 @@ irqallocread(Chan*, void *a, long n, vlong offset)
 	return 0;
 }
 
-static void
-nmienable(void)
-{
-	int x;
-
-	/*
-	 * Hack: should be locked with NVRAM access.
-	 */
-	outb(0x70, 0x80);		/* NMI latch clear */
-	outb(0x70, 0);
-
-	x = inb(0x61) & 0x07;		/* Enable NMI */
-	outb(0x61, 0x0C|x);
-	outb(0x61, x);
-}
-
-static void
-nmihandler(Ureg *ureg, void*)
-{
-	/*
-	 * Don't re-enable, it confuses the crash dumps.
-	nmienable();
-	 */
-	iprint("cpu%d: nmi PC %#p, status %ux\n",
-		m->machno, ureg->pc, inb(0x61));
-	while(m->machno != 0)
-		;
-}
-
 void
 irqinit(void)
 {
 	addarchfile("irqalloc", 0444, irqallocread, nil);
-
-	trapenable(VectorNMI, nmihandler, nil, "nmi");
-	nmienable();
 }

sys/src/9/pc/mp.c

@@ -10,6 +10,8 @@
 #include "mp.h"
 #include "apbootstrap.i"
 
+extern void i8259init(void);
+
 /* filled in by pcmpinit or acpiinit */
 Bus* mpbus;
 Bus* mpbuslast;
@@ -291,23 +293,51 @@ allocvector(void)
 }
 
 static int
-mpintrenablex(Vctl* v, int tbdf)
+ioapicirqenable(Vctl *v, int shared)
+{
+	Aintr *aintr = v->aux;
+	int lo, hi;
+
+	if(shared)
+		return 0;
+	hi = v->cpu<<24;
+	lo = mpintrinit(aintr->bus, aintr->intr, v->vno, v->irq);
+	lo |= ApicPHYSICAL;			/* no-op */
+ 	ioapicrdtw(aintr->apic, aintr->intr->intin, hi, lo);
+	return 0;
+}
+
+static int
+ioapicirqdisable(Vctl *v, int shared)
+{
+	Aintr *aintr = v->aux;
+	int lo, hi;
+
+	if(shared)
+		return 0;
+	hi = 0;
+	lo = ApicIMASK;
+ 	ioapicrdtw(aintr->apic, aintr->intr->intin, hi, lo);
+	return 0;
+}
+
+static int
+mpintrassignx(Vctl* v, int tbdf)
 {
 	Bus *bus;
+	Pcidev *pci;
 	Aintr *aintr;
-	Apic *apic;
+	int bno, dno, pin, irq, type, lo, hi, n;
-	Pcidev *pcidev;
-	int bno, dno, pin, hi, irq, lo, n, type, vno;
 
 	type = BUSTYPE(tbdf);
 	bno = BUSBNO(tbdf);
 	dno = BUSDNO(tbdf);
 
 	pin = 0;
-	pcidev = nil;
+	pci = nil;
 	if(type == BusPCI){
-		if(pcidev = pcimatchtbdf(tbdf))
+		if((pci = pcimatchtbdf(tbdf)) != nil)
-			pin = pcicfgr8(pcidev, PciINTP);
+			pin = pcicfgr8(pci, PciINTP);
 	} else if(type == BusISA)
 		bno = mpisabus;
 
@@ -325,14 +355,14 @@ Findbus:
 		 * by the MP or ACPI tables then walk up the bus translating interrupt
 		 * pin to parent bus.
 		 */
-		if(pcidev && pcidev->parent && pin > 0){
+		if(pci != nil && pci->parent != nil && pin > 0){
 			pin = ((dno+(pin-1))%4)+1;
-			pcidev = pcidev->parent;
+			pci = pci->parent;
-			bno = BUSBNO(pcidev->tbdf);
+			bno = BUSBNO(pci->tbdf);
-			dno = BUSDNO(pcidev->tbdf);
+			dno = BUSDNO(pci->tbdf);
 			goto Findbus;
 		}
-		print("mpintrenable: can't find bus type %d, number %d\n", type, bno);
+		print("mpintrassign: can't find bus type %d, number %d\n", type, bno);
 		return -1;
 	}
 
@@ -346,65 +376,56 @@ Findbus:
 			irq = (dno<<2)|(pin-1);
 		else
 			irq = -1;
-	}
+	} else
-	else
 		irq = v->irq;
 
 	/*
 	 * Find a matching interrupt entry from the list of interrupts
 	 * attached to this bus.
 	 */
-	for(aintr = bus->aintr; aintr; aintr = aintr->next){
+	for(aintr = bus->aintr; aintr != nil; aintr = aintr->next){
 		if(aintr->intr->irq != irq)
 			continue;
-		if(0){
+
-			PCMPintr* p = aintr->intr;
-	   	 	print("mpintrenablex: bus %d intin %d irq %d\n",
-				p->busno, p->intin, p->irq);
-		}
 		/*
 		 * Check if already enabled. Multifunction devices may share
 		 * INT[A-D]# so, if already enabled, check the polarity matches
 		 * and the trigger is level.
-		 *
-		 * Should check the devices differ only in the function number,
-		 * but that can wait for the planned enable/disable rewrite.
-		 * The RDT read here is safe for now as currently interrupts
-		 * are never disabled once enabled.
 		 */
-		apic = aintr->apic;
+		ioapicrdtr(aintr->apic, aintr->intr->intin, &hi, &lo);
-		ioapicrdtr(apic, aintr->intr->intin, 0, &lo);
-		if(!(lo & ApicIMASK)){
-			vno = lo & 0xFF;
-			if(0) print("%s vector %d (!imask)\n", v->name, vno);
-			n = mpintrinit(bus, aintr->intr, vno, v->irq);
-			n |= ApicPHYSICAL;		/* no-op */
-			lo &= ~(ApicRemoteIRR|ApicDELIVS);
-			if(n != lo){
-				print("mpintrenable: multiple botch irq %d, tbdf %uX, lo %8.8uX, n %8.8uX\n",
-					v->irq, tbdf, lo, n);
-				return -1;
-			}
-			v->isr = lapicisr;
-			v->eoi = lapiceoi;
-			return vno;
-		}
-
-		vno = allocvector();
-		hi = mpintrcpu()<<24;
-		lo = mpintrinit(bus, aintr->intr, vno, v->irq);
-		lo |= ApicPHYSICAL;			/* no-op */
 		if(lo & ApicIMASK){
-			print("mpintrenable: disabled irq %d, tbdf %uX, lo %8.8uX, hi %8.8uX\n",
+			v->vno = allocvector();
-				v->irq, tbdf, lo, hi);
+			v->cpu = mpintrcpu();
-			return -1;
+			lo = mpintrinit(aintr->bus, aintr->intr, v->vno, v->irq);
+			lo |= ApicPHYSICAL;			/* no-op */
+			if(lo & ApicIMASK){
+				print("mpintrassign: disabled irq %d, tbdf %uX, lo %8.8uX, hi %8.8uX\n",
+					v->irq, v->tbdf, lo, hi);
+				break;
+			}
+		} else {
+			v->vno = lo & 0xFF;
+			v->cpu = hi >> 24;
+			lo &= ~(ApicRemoteIRR|ApicDELIVS);
+			n = mpintrinit(aintr->bus, aintr->intr, v->vno, v->irq);
+			n |= ApicPHYSICAL;			/* no-op */
+			if(lo != n){
+				print("mpintrassign: multiple botch irq %d, tbdf %uX, lo %8.8uX, n %8.8uX\n",
+					v->irq, v->tbdf, lo, n);
+				break;
+			}
 		}
-		if((apic->flags & PcmpEN) && apic->type == PcmpIOAPIC)
- 			ioapicrdtw(apic, aintr->intr->intin, hi, lo);
 
 		v->isr = lapicisr;
 		v->eoi = lapiceoi;
-		return vno;
+
+		if((aintr->apic->flags & PcmpEN) && aintr->apic->type == PcmpIOAPIC){
+			v->aux = aintr;
+			v->enable = ioapicirqenable;
+			v->disable = ioapicirqdisable;
+		}
+
+		return v->vno;
 	}
 
 	return -1;
@@ -469,23 +490,29 @@ htmsienable(Pcidev *pdev)
 	return -1;
 }
 
-static void
+static int
-msiintrdisable(Vctl *v)
+msiirqenable(Vctl *v, int)
 {
-	Pcidev *pci;
+	Pcidev *pci = v->aux;
+	return pcimsienable(pci, 0xFEE00000ULL | (v->cpu << 12), v->vno | (1<<14));
+}
 
-	if((pci = pcimatchtbdf(v->tbdf)) != nil)
+static int
-		pcimsidisable(pci);
+msiirqdisable(Vctl *v, int)
+{
+	Pcidev *pci = v->aux;
+	return pcimsidisable(pci);
 }
 
 static int
 msiintrenable(Vctl *v)
 {
-	int tbdf, vno, cpu;
 	Pcidev *pci;
+	int tbdf;
 
 	if(getconf("*nomsi") != nil)
 		return -1;
+
 	tbdf = v->tbdf;
 	if(tbdf == BUSUNKNOWN || BUSTYPE(tbdf) != BusPCI)
 		return -1;
@@ -498,18 +525,21 @@ msiintrenable(Vctl *v)
 		return -1;
 	if(pcimsidisable(pci) < 0)
 		return -1;
-	vno = allocvector();
+
-	cpu = mpintrcpu();
+	v->vno = allocvector();
-	if(pcimsienable(pci, 0xFEE00000ULL | (cpu << 12), vno | (1<<14)) < 0)
+	v->cpu = mpintrcpu();
-		return -1;
-	v->disable = msiintrdisable;
 	v->isr = lapicisr;
 	v->eoi = lapiceoi;
-	return vno;
+
+	v->aux = pci;
+	v->enable = msiirqenable;
+	v->disable = msiirqdisable;
+
+	return v->vno;
 }
 
 int
-mpintrenable(Vctl* v)
+mpintrassign(Vctl* v)
 {
 	int irq, tbdf, vno;
 
@@ -524,17 +554,18 @@ mpintrenable(Vctl* v)
 	 * breakpoint and page-fault).
 	 */
 	tbdf = v->tbdf;
-	if(tbdf != BUSUNKNOWN && (vno = mpintrenablex(v, tbdf)) != -1)
+	if(tbdf != BUSUNKNOWN && (vno = mpintrassignx(v, tbdf)) != -1)
 		return vno;
 
 	irq = v->irq;
 	if(irq >= IrqLINT0 && irq <= MaxIrqLAPIC){
+		v->local = 1;
 		if(irq != IrqSPURIOUS)
 			v->isr = lapiceoi;
 		return VectorPIC+irq;
 	}
 	if(irq < 0 || irq > MaxIrqPIC){
-		print("mpintrenable: irq %d out of range\n", irq);
+		print("mpintrassign: irq %d out of range\n", irq);
 		return -1;
 	}
 
@@ -550,16 +581,16 @@ mpintrenable(Vctl* v)
 	 * be compatible with ISA.
 	 */
 	if(mpeisabus != -1){
-		vno = mpintrenablex(v, MKBUS(BusEISA, 0, 0, 0));
+		vno = mpintrassignx(v, MKBUS(BusEISA, 0, 0, 0));
 		if(vno != -1)
 			return vno;
 	}
 	if(mpisabus != -1){
-		vno = mpintrenablex(v, MKBUS(BusISA, 0, 0, 0));
+		vno = mpintrassignx(v, MKBUS(BusISA, 0, 0, 0));
 		if(vno != -1)
 			return vno;
 	}
-	print("mpintrenable: out of choices eisa %d isa %d tbdf %uX irq %d\n",
+	print("mpintrassign: out of choices eisa %d isa %d tbdf %uX irq %d\n",
 		mpeisabus, mpisabus, v->tbdf, v->irq);
 	return -1;
 }

sys/src/9/pc/mp.h

@@ -168,6 +168,7 @@ typedef struct Bus {
 typedef struct Aintr {
 	PCMPintr* intr;
 	Apic*	apic;
+	Bus*	bus;
 	Aintr*	next;
 };
 
@@ -249,7 +250,7 @@ extern void lapictimerset(uvlong);
 
 extern int mpintrinit(Bus*, PCMPintr*, int, int);
 extern void mpinit(void);
-extern int mpintrenable(Vctl*);
+extern int mpintrassign(Vctl*);
 extern void mpshutdown(void);
 extern void mpstartap(Apic*);
 

sys/src/9/pc/pc

@@ -96,6 +96,7 @@ link
 misc
 	pci		pcipc
 
+	archgeneric	devkbd i8259 i8253
 	archacpi	mp apic squidboy ec
 	archmp		mp apic squidboy
 	mtrr

sys/src/9/pc/trap.c

@@ -66,6 +66,8 @@ trapinit(void)
 {
 	irqinit();
 
+	nmienable();
+
 	/*
 	 * Special traps.
 	 * Syscall() is called directly without going through trap().

sys/src/9/pc64/dat.h

@@ -254,13 +254,12 @@ struct PCArch
 	char*	id;
 	int	(*ident)(void);		/* this should be in the model */
 	void	(*reset)(void);		/* this should be in the model */
-	int	(*serialpower)(int);	/* 1 == on, 0 == off */
-	int	(*modempower)(int);	/* 1 == on, 0 == off */
 
 	void	(*intrinit)(void);
-	int	(*intrenable)(Vctl*);
+	int	(*intrassign)(Vctl*);
+	int	(*intrirqno)(int, int);
 	int	(*intrvecno)(int);
-	int	(*intrdisable)(int);
+	int	(*intrspurious)(int);
 	void	(*introff)(void);
 	void	(*intron)(void);
 

sys/src/9/pc64/fns.h

@@ -63,13 +63,6 @@ void	i8253init(void);
 void	i8253reset(void);
 uvlong	i8253read(uvlong*);
 void	i8253timerset(uvlong);
-int	i8259disable(int);
-int	i8259enable(Vctl*);
-void	i8259init(void);
-int	i8259isr(int);
-void	i8259on(void);
-void	i8259off(void);
-int	i8259vecno(int);
 void	idle(void);
 void	idlehands(void);
 int	inb(int);
@@ -109,6 +102,7 @@ char*	mtrr(uvlong, uvlong, char *);
 void	mtrrclock(void);
 int	mtrrprint(char *, long);
 void	mtrrsync(void);
+void	nmienable(void);
 void	noteret(void);
 uchar	nvramread(int);
 void	nvramwrite(int, uchar);

sys/src/9/pc64/pc64

@@ -93,6 +93,7 @@ link
 
 misc
 	pci		pcipc
+	archgeneric	devkbd i8259 i8253
 	archacpi	mp apic squidboy ec
 	archmp		mp apic squidboy
 	mtrr

sys/src/9/pc64/trap.c

@@ -67,6 +67,8 @@ trapinit(void)
 {
 	irqinit();
 
+	nmienable();
+
 	/*
 	 * Special traps.
 	 * Syscall() is called directly without going through trap().

sys/src/9/xen/archxen.c

@@ -52,9 +52,7 @@ shutdown(void)
 	HYPERVISOR_shutdown(1);
 }
 
-int xenintrenable(Vctl *v);
+int xenintrassign(Vctl *v);
-int xenintrvecno(int irq);
-int xenintrdisable(int irq);
 void	xentimerenable(void);
 uvlong	xentimerread(uvlong*);
 void	xentimerset(uvlong);
@@ -64,16 +62,14 @@ PCArch archxen = {
 .ident=		identify,
 .reset=		shutdown,
 .intrinit=	intrinit,
-.intrenable=	xenintrenable,
+.intrassign=	xenintrassign,
-.intrvecno=	xenintrvecno,
-.intrdisable=	xenintrdisable,
 .clockenable=	xentimerenable,
 .fastclock=	xentimerread,
 .timerset=	xentimerset,
 };
 
 /*
- * Placeholders to satisfy external references in generic devarch.c
+ * Placeholders to satisfy external references in devarch.c
  */
 ulong	getcr4(void)	{ return 0; }
 void	putcr4(ulong)	{}
@@ -83,19 +79,7 @@ ulong	inl(int)	{ return 0; }
 void	outb(int, int)	{}
 void	outs(int, ushort)	{}
 void	outl(int, ulong)	{}
-void	i8042reset(void)	{}
+
-void	i8253enable(void)	{}
-void	i8253init(void)	{}
-void	i8253link(void)	{}
-uvlong	i8253read(uvlong*)	{ return 0; }
-void	i8253timerset(uvlong)	{}
-int	i8259disable(int)	{ return 0; }
-int	i8259enable(Vctl*)	{ return 0; }
-void	i8259init(void)	{}
-int	i8259isr(int)	{ return 0; }
-void	i8259on(void)	{}
-void	i8259off(void)	{}
-int	i8259vecno(int)	{ return 0; }
 int	mtrrprint(char*, long) { return 0; }
 char*	mtrr(uvlong, uvlong, char *) { return nil; }
 void	mtrrsync(void) {}

sys/src/9/xen/fns.h

@@ -39,13 +39,6 @@ void	i8253init(void);
 void	i8253link(void);
 uvlong	i8253read(uvlong*);
 void	i8253timerset(uvlong);
-int	i8259disable(int);
-int	i8259enable(Vctl*);
-void	i8259init(void);
-int	i8259isr(int);
-void	i8259on(void);
-void	i8259off(void);
-int	i8259vecno(int);
 void	idle(void);
 void	idlehands(void);
 int	inb(int);

sys/src/9/xen/mkfile

@@ -71,6 +71,7 @@ OBJ=\
 	mmu.$O\
 	random.$O\
 	rdb.$O\
+	irq.$O\
 	trap.$O\
 	$CONF.root.$O\
 	$CONF.rootc.$O\

sys/src/9/xen/trap.c

@@ -28,6 +28,9 @@ enum {
   
 void	noted(Ureg*, ulong);
 
+extern void irqinit(void);
+extern int irqhandled(Ureg*, int);
+
 static void debugbpt(Ureg*, void*);
 static void fault386(Ureg*, void*);
 static void safe_fault386(Ureg*, void*);
@@ -35,148 +38,6 @@ static void doublefault(Ureg*, void*);
 static void unexpected(Ureg*, void*);
 static void _dumpstack(Ureg*);
 
-static Lock vctllock;
-static Vctl *vctl[256];
-
-enum
-{
-	Ntimevec = 20		/* number of time buckets for each intr */
-};
-ulong intrtimes[256][Ntimevec];
-
-void
-intrenable(int irq, void (*f)(Ureg*, void*), void* a, int tbdf, char *name)
-{
-	int vno;
-	Vctl *v;
-
-/**/
-	SETUPLOG(dprint("intrenable: irq %d, f %p, a %p, tbdf 0x%x, name %s\n", 
-			irq, f, a, tbdf, name);)
-/**/
-	if(f == nil){
-		print("intrenable: nil handler for %d, tbdf 0x%uX for %s\n",
-			irq, tbdf, name);
-		return;
-	}
-
-	v = xalloc(sizeof(Vctl));
-	v->isintr = 1;
-	v->irq = irq;
-	v->tbdf = tbdf;
-	v->f = f;
-	v->a = a;
-	strncpy(v->name, name, KNAMELEN-1);
-	v->name[KNAMELEN-1] = 0;
-
-	ilock(&vctllock);
-	vno = arch->intrenable(v);
-	if(vno == -1){
-		iunlock(&vctllock);
-		print("intrenable: couldn't enable irq %d, tbdf 0x%uX for %s\n",
-			irq, tbdf, v->name);
-		xfree(v);
-		return;
-	}
-	if(vctl[vno]){
-		if(vctl[vno]->isr != v->isr || vctl[vno]->eoi != v->eoi)
-			panic("intrenable: handler: %s %s %p %p %p %p\n",
-				vctl[vno]->name, v->name,
-				vctl[vno]->isr, v->isr, vctl[vno]->eoi, v->eoi);
-		v->next = vctl[vno];
-	}
-	vctl[vno] = v;
-	SETUPLOG(dprint("INTRENABLE: vctl[%d] is %p\n", vno, vctl[vno]);)
-	iunlock(&vctllock);
-}
-
-void
-intrdisable(int irq, void (*f)(Ureg *, void *), void *a, int tbdf, char *name)
-{
-	Vctl **pv, *v;
-	int vno;
-
-	vno = arch->intrvecno(irq);
-	ilock(&vctllock);
-	for(pv = &vctl[vno]; (v = *pv) != nil; pv = &v->next){
-		if(v->isintr && v->irq == irq
-		&& v->tbdf == tbdf && v->f == f && v->a == a
-		&& strcmp(v->name, name) == 0){
-			*pv = v->next;
-			xfree(v);
-
-			if(vctl[vno] == nil && arch->intrdisable != nil)
-				arch->intrdisable(irq);
-			break;
-		}
-	}
-	iunlock(&vctllock);
-}
-
-static long
-irqallocread(Chan*, void *a, long n, vlong offset)
-{
-	char buf[2*(11+1)+KNAMELEN+1+1];
-	int vno, m;
-	Vctl *v;
-
-	if(n < 0 || offset < 0)
-		error(Ebadarg);
-
-	for(vno=0; vno<nelem(vctl); vno++){
-		for(v=vctl[vno]; v; v=v->next){
-			m = snprint(buf, sizeof(buf), "%11d %11d %.*s\n", vno, v->irq, KNAMELEN, v->name);
-			offset -= m;
-			if(offset >= 0)
-				continue;
-			if(n > -offset)
-				n = -offset;
-			offset += m;
-			memmove(a, buf+offset, n);
-			return n;
-		}
-	}
-	return 0;
-}
-
-void
-trapenable(int vno, void (*f)(Ureg*, void*), void* a, char *name)
-{
-	Vctl *v;
-
-	if(vno < 0 || vno >= VectorPIC)
-		panic("trapenable: vno %d\n", vno);
-	v = xalloc(sizeof(Vctl));
-	v->tbdf = BUSUNKNOWN;
-	v->f = f;
-	v->a = a;
-	strncpy(v->name, name, KNAMELEN);
-	v->name[KNAMELEN-1] = 0;
-
-	lock(&vctllock);
-	if(vctl[vno])
-		v->next = vctl[vno]->next;
-	vctl[vno] = v;
-	unlock(&vctllock);
-}
-
-static void
-nmienable(void)
-{
-	/* leave this here in case plan 9 ever makes it to dom0 */
-#ifdef NOWAY
-	/*
-	 * Hack: should be locked with NVRAM access.
-	 */
-	outb(0x70, 0x80);		/* NMI latch clear */
-	outb(0x70, 0);
-
-	x = inb(0x61) & 0x07;		/* Enable NMI */
-	outb(0x61, 0x08|x);
-	outb(0x61, x);
-#endif
-}
-
 /* we started out doing the 'giant bulk init' for all traps. 
   * we're going to do them one-by-one since error analysis is 
   * so much easier that way.
@@ -212,6 +73,8 @@ trapinit(void)
 		vaddr += 6;
 	}
 
+	irqinit();
+
 	/*
 	 * Special traps.
 	 * Syscall() is called directly without going through trap().
@@ -220,9 +83,6 @@ trapinit(void)
 	trapenable(VectorPF, fault386, 0, "fault386");
 	trapenable(Vector2F, doublefault, 0, "doublefault");
 	trapenable(Vector15, unexpected, 0, "unexpected");
-
-	nmienable();
-	addarchfile("irqalloc", 0444, irqallocread, nil);
 }
 
 static char* excname[32] = {
@@ -260,27 +120,18 @@ static char* excname[32] = {
 	"31 (reserved)",
 };
 
-/*
+static int
- *  keep histogram of interrupt service times
+usertrap(int vno)
- */
-void
-intrtime(Mach*, int vno)
 {
-	ulong diff;
+	char buf[ERRMAX];
-	ulong x;
 
-	x = perfticks();
+	if(vno < nelem(excname)){
-	diff = x - m->perf.intrts;
+		spllo();
-	m->perf.intrts = x;
+		sprint(buf, "sys: trap: %s", excname[vno]);
-
+		postnote(up, 1, buf, NDebug);
-	m->perf.inintr += diff;
+		return 1;
-	if(up == nil && m->perf.inidle > diff)
+	}
-		m->perf.inidle -= diff;
+	return 0;
-
-	diff /= m->cpumhz*100;	// quantum = 100µsec
-	if(diff >= Ntimevec)
-		diff = Ntimevec-1;
-	intrtimes[vno][diff]++;
 }
 
 /* go to user space */
@@ -304,120 +155,27 @@ kexit(Ureg*)
  *  rather than directly vectoring the handler.  However, this avoids a
  *  lot of code duplication and possible bugs.  The only exception is
  *  VectorSYSCALL.
- *  Trap is called with interrupts (and events) disabled via interrupt-gates.
+ *  Trap is called with interrupts disabled via interrupt-gates.
  */
 void
 trap(Ureg* ureg)
 {
-	int clockintr, i, vno, user;
+	int vno, user;
-	char buf[ERRMAX];
-	Vctl *ctl, *v;
-	Mach *mach;
 
-	TRAPLOG(dprint("trap ureg %lux %lux\n", (ulong*)ureg, ureg->trap);)
+	user = userureg(ureg);
-	m->perf.intrts = perfticks();
-	user = (ureg->cs & 0xFFFF) == UESEL;
 	if(user){
 		up->dbgreg = ureg;
 		cycles(&up->kentry);
 	}
 
-	clockintr = 0;
-
 	vno = ureg->trap;
-	if(vno < 0 || vno >= 256)
+	if(!irqhandled(ureg, vno) && (!user || !usertrap(vno))){
-		panic("bad interrupt number %d\n", vno);
+		if(!user){
-	TRAPLOG(dprint("trap: vno is 0x%x, vctl[%d] is %p\n", vno, vno, vctl[vno]);)
+			/* early fault before trapinit() */
-	if(ctl = vctl[vno]){
+			if(vno == VectorPF)
-		INTRLOG(dprint("ctl is %p, isintr is %d\n", ctl, ctl->isintr);)
+				fault386(ureg, 0);
-		if(ctl->isintr){
-			m->intr++;
-			if(vno >= VectorPIC && vno != VectorSYSCALL)
-				m->lastintr = ctl->irq;
 		}
 
-		INTRLOG(dprint("ctl %p, isr %p\n", ctl, ctl->isr);)
-		if(ctl->isr)
-			ctl->isr(vno);
-		for(v = ctl; v != nil; v = v->next){
-			INTRLOG(dprint("ctl %p, f is %p\n", v, v->f);)
-			if(v->f)
-				v->f(ureg, v->a);
-		}
-		INTRLOG(dprint("ctl %p, eoi %p\n", ctl, ctl->eoi);)
-		if(ctl->eoi)
-			ctl->eoi(vno);
-
-		if(ctl->isintr){
-			intrtime(m, vno);
-
-			//if(ctl->irq == IrqCLOCK || ctl->irq == IrqTIMER)
-			if (ctl->tbdf != BUSUNKNOWN && ctl->irq == VIRQ_TIMER)
-				clockintr = 1;
-
-			if(up && !clockintr)
-				preempted();
-		}
-	}
-	else if(vno <= nelem(excname) && user){
-		spllo();
-		sprint(buf, "sys: trap: %s", excname[vno]);
-		postnote(up, 1, buf, NDebug);
-	}
-	else if(vno >= VectorPIC && vno != VectorSYSCALL){
-		/*
-		 * An unknown interrupt.
-		 * Check for a default IRQ7. This can happen when
-		 * the IRQ input goes away before the acknowledge.
-		 * In this case, a 'default IRQ7' is generated, but
-		 * the corresponding bit in the ISR isn't set.
-		 * In fact, just ignore all such interrupts.
-		 */
-
-		/* call all interrupt routines, just in case */
-		for(i = VectorPIC; i <= MaxIrqLAPIC; i++){
-			ctl = vctl[i];
-			if(ctl == nil)
-				continue;
-			if(!ctl->isintr)
-				continue;
-			for(v = ctl; v != nil; v = v->next){
-				if(v->f)
-					v->f(ureg, v->a);
-			}
-			/* should we do this? */
-			if(ctl->eoi)
-				ctl->eoi(i);
-		}
-
-		iprint("cpu%d: spurious interrupt %d, last %d\n",
-			m->machno, vno, m->lastintr);
-		if(0)if(conf.nmach > 1){
-			for(i = 0; i < MAXMACH; i++){
-				if(active.machs[i] == 0)
-					continue;
-				mach = MACHP(i);
-				if(m->machno == mach->machno)
-					continue;
-				print(" cpu%d: last %d",
-					mach->machno, mach->lastintr);
-			}
-			print("\n");
-		}
-		m->spuriousintr++;
-		if(user)
-			kexit(ureg);
-		return;
-	}
-	else{
-		if(vno == VectorNMI){
-			nmienable();
-			if(m->machno != 0){
-				print("cpu%d: PC %8.8luX\n",
-					m->machno, ureg->pc);
-				for(;;);
-			}
-		}
 		dumpregs(ureg);
 		if(!user){
 			ureg->sp = (ulong)&ureg->sp;
@@ -425,18 +183,10 @@ trap(Ureg* ureg)
 		}
 		if(vno < nelem(excname))
 			panic("%s", excname[vno]);
-		panic("unknown trap/intr: %d\n", vno);
+		panic("unknown trap/intr: %d", vno);
 	}
 	splhi();
 
-	/* delaysched set because we held a lock or because our quantum ended */
-	if(up && up->delaysched && clockintr){
-		INTRLOG(dprint("calling sched in trap? \n");)
-		sched();
-		INTRLOG(dprint("Back from calling sched in trap?\n");)
-		splhi();
-	}
-
 	if(user){
 		if(up->procctl || up->nnote)
 			notify(ureg);

sys/src/9/xen/xensystem.c

@@ -389,6 +389,26 @@ xenupcall(Ureg *ureg)
 	
 }
 
+static int
+xenirqenable(Vctl *v, int shared)
+{
+	if(!shared){
+		uint port = v->vno-100;
+		HYPERVISOR_shared_info->evtchn_mask[port/32] &= ~(1<<(port%32));
+	}
+	return 0;
+}
+
+static int
+xenirqdisable(Vctl *v, int shared)
+{
+	if(!shared){
+		uint port = v->vno-100;
+		HYPERVISOR_shared_info->evtchn_mask[port/32] |= (1<<(port%32));
+	}
+	return 0;
+}
+
 /*
  * tbdf field is abused to distinguish virqs from channels:
  *
@@ -396,36 +416,29 @@ xenupcall(Ureg *ureg)
  * tbdf=0 -> irq is a channel number
  */
 int
-xenintrenable(Vctl *v)
+xenintrassign(Vctl *v)
 {
 	evtchn_op_t op;
 	uint port;
 
-	/* XXX locking? */
 	if (v->tbdf != BUSUNKNOWN) {
 		op.cmd = EVTCHNOP_bind_virq;
 		op.u.bind_virq.virq = v->irq;
 		op.u.bind_virq.vcpu = m->machno;
-		if(HYPERVISOR_event_channel_op(&op) != 0)
+		if(HYPERVISOR_event_channel_op(&op) != 0){
-			panic("xenintrenable: bind %d failed", v->irq);
+			print("xenintrenable: bind %d failed", v->irq);
+			return -1;
+		}
 		port = op.u.bind_virq.port;
 	} else
 		port = v->irq;
 	if (port > 155)
 		return -1;
-	HYPERVISOR_shared_info->evtchn_mask[port/32] &= ~(1<<(port%32));
+	v->enable = xenirqenable;
-	if(0)print("xenintrenable %s: irq %d port %d mask[%d] = %#lux\n", v->name, v->irq, port, port/32, HYPERVISOR_shared_info->evtchn_mask[port/32]);
+	v->disable = xenirqdisable;
 	return 100+port;
 }
 
-int
-xenintrdisable(int irq)
-{
-	USED(irq);
-	panic("xenintrdisable notyet\n");
-	return 0;
-}
-
 int
 xenintrvecno(int irq)
 {

sys/src/9/xen/xentimer.c

@@ -68,7 +68,7 @@ xentimerclock(Ureg* ureg, void*)
 void
 xentimerenable(void)
 {
-	intrenable(VIRQ_TIMER, xentimerclock, nil, 0, "Xen Timer");
+	intrenable(VIRQ_TIMER, xentimerclock, nil, 0, "clock");
 }
 
 uvlong