[PARISC] Allow nested interrupts

Our prior mode of operation didn't allow nested interrupts
because it makes the interrupt code much simpler.  However,
nested interrupts are better for latency.

This code uses the EIEM register to simulate level interrupts
and thus achieve nesting.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Signed-off-by: Kyle McMartin <kyle@parisc-linux.org>

4 files changed, 93 insertions, 64 deletions

diff --git a/arch/parisc/kernel/irq.c b/arch/parisc/kernel/irq.c
index c53bfeb4bf94a3..9bdd0197ceb777 100644
--- a/arch/parisc/kernel/irq.c
+++ b/arch/parisc/kernel/irq.c
@@ -45,6 +45,17 @@ extern irqreturn_t ipi_interrupt(int, void *, struct pt_regs *);
 */
 static volatile unsigned long cpu_eiem = 0;
 
+/*
+** ack bitmap ... habitually set to 1, but reset to zero
+** between ->ack() and ->end() of the interrupt to prevent
+** re-interruption of a processing interrupt.
+*/
+static volatile unsigned long global_ack_eiem = ~0UL;
+/*
+** Local bitmap, same as above but for per-cpu interrupts
+*/
+static DEFINE_PER_CPU(unsigned long, local_ack_eiem) = ~0UL;
+
 static void cpu_disable_irq(unsigned int irq)
 {
 	unsigned long eirr_bit = EIEM_MASK(irq);
@@ -62,13 +73,6 @@ static void cpu_enable_irq(unsigned int irq)
 
 	cpu_eiem |= eirr_bit;
 
-	/* FIXME: while our interrupts aren't nested, we cannot reset
-	 * the eiem mask if we're already in an interrupt.  Once we
-	 * implement nested interrupts, this can go away
-	 */
-	if (!in_interrupt())
-		set_eiem(cpu_eiem);
-
 	/* This is just a simple NOP IPI.  But what it does is cause
 	 * all the other CPUs to do a set_eiem(cpu_eiem) at the end
 	 * of the interrupt handler */
@@ -84,13 +88,45 @@ static unsigned int cpu_startup_irq(unsigned int irq)
 void no_ack_irq(unsigned int irq) { }
 void no_end_irq(unsigned int irq) { }
 
+void cpu_ack_irq(unsigned int irq)
+{
+	unsigned long mask = EIEM_MASK(irq);
+	int cpu = smp_processor_id();
+
+	/* Clear in EIEM so we can no longer process */
+	if (CHECK_IRQ_PER_CPU(irq_desc[irq].status))
+		per_cpu(local_ack_eiem, cpu) &= ~mask;
+	else
+		global_ack_eiem &= ~mask;
+
+	/* disable the interrupt */
+	set_eiem(cpu_eiem & global_ack_eiem & per_cpu(local_ack_eiem, cpu));
+	/* and now ack it */
+	mtctl(mask, 23);
+}
+
+void cpu_end_irq(unsigned int irq)
+{
+	unsigned long mask = EIEM_MASK(irq);
+	int cpu = smp_processor_id();
+
+	/* set it in the eiems---it's no longer in process */
+	if (CHECK_IRQ_PER_CPU(irq_desc[irq].status))
+		per_cpu(local_ack_eiem, cpu) |= mask;
+	else
+		global_ack_eiem |= mask;
+
+	/* enable the interrupt */
+	set_eiem(cpu_eiem & global_ack_eiem & per_cpu(local_ack_eiem, cpu));
+}
+
 #ifdef CONFIG_SMP
 int cpu_check_affinity(unsigned int irq, cpumask_t *dest)
 {
 	int cpu_dest;
 
 	/* timer and ipi have to always be received on all CPUs */
-	if (irq == TIMER_IRQ || irq == IPI_IRQ) {
+	if (CHECK_IRQ_PER_CPU(irq)) {
 		/* Bad linux design decision.  The mask has already
 		 * been set; we must reset it */
 		irq_desc[irq].affinity = CPU_MASK_ALL;
@@ -119,8 +155,8 @@ static struct hw_interrupt_type cpu_interrupt_type = {
 	.shutdown	= cpu_disable_irq,
 	.enable		= cpu_enable_irq,
 	.disable	= cpu_disable_irq,
-	.ack		= no_ack_irq,
-	.end		= no_end_irq,
+	.ack		= cpu_ack_irq,
+	.end		= cpu_end_irq,
 #ifdef CONFIG_SMP
 	.set_affinity	= cpu_set_affinity_irq,
 #endif
@@ -298,82 +334,69 @@ unsigned int txn_alloc_data(unsigned int virt_irq)
 	return virt_irq - CPU_IRQ_BASE;
 }
 
+static inline int eirr_to_irq(unsigned long eirr)
+{
+#ifdef CONFIG_64BIT
+	int bit = fls64(eirr);
+#else
+	int bit = fls(eirr);
+#endif
+	return (BITS_PER_LONG - bit) + TIMER_IRQ;
+}
+
 /* ONLY called from entry.S:intr_extint() */
 void do_cpu_irq_mask(struct pt_regs *regs)
 {
 	unsigned long eirr_val;
-
-	irq_enter();
-
-	/*
-	 * Don't allow TIMER or IPI nested interrupts.
-	 * Allowing any single interrupt to nest can lead to that CPU
-	 * handling interrupts with all enabled interrupts unmasked.
-	 */
-	set_eiem(0UL);
-
-	/* 1) only process IRQs that are enabled/unmasked (cpu_eiem)
-	 * 2) We loop here on EIRR contents in order to avoid
-	 *    nested interrupts or having to take another interrupt
-	 *    when we could have just handled it right away.
-	 */
-	for (;;) {
-		unsigned long bit = (1UL << (BITS_PER_LONG - 1));
-		unsigned int irq;
-		eirr_val = mfctl(23) & cpu_eiem;
-		if (!eirr_val)
-			break;
-
-		mtctl(eirr_val, 23); /* reset bits we are going to process */
-
-		/* Work our way from MSb to LSb...same order we alloc EIRs */
-		for (irq = TIMER_IRQ; eirr_val && bit; bit>>=1, irq++) {
+	int irq, cpu = smp_processor_id();
 #ifdef CONFIG_SMP
-			cpumask_t dest = irq_desc[irq].affinity;
+	cpumask_t dest;
 #endif
-			if (!(bit & eirr_val))
-				continue;
 
-			/* clear bit in mask - can exit loop sooner */
-			eirr_val &= ~bit;
+	local_irq_disable();
+	irq_enter();
 
-#ifdef CONFIG_SMP
-			/* FIXME: because generic set affinity mucks
-			 * with the affinity before sending it to us
-			 * we can get the situation where the affinity is
-			 * wrong for our CPU type interrupts */
-			if (irq != TIMER_IRQ && irq != IPI_IRQ &&
-			    !cpu_isset(smp_processor_id(), dest)) {
-				int cpu = first_cpu(dest);
-
-				printk(KERN_DEBUG "redirecting irq %d from CPU %d to %d\n",
-				       irq, smp_processor_id(), cpu);
-				gsc_writel(irq + CPU_IRQ_BASE,
-					   cpu_data[cpu].hpa);
-				continue;
-			}
-#endif
+	eirr_val = mfctl(23) & cpu_eiem & global_ack_eiem &
+		per_cpu(local_ack_eiem, cpu);
+	if (!eirr_val)
+		goto set_out;
+	irq = eirr_to_irq(eirr_val);
 
-			__do_IRQ(irq, regs);
-		}
+#ifdef CONFIG_SMP
+	dest = irq_desc[irq].affinity;
+	if (CHECK_IRQ_PER_CPU(irq_desc[irq].status) &&
+	    !cpu_isset(smp_processor_id(), dest)) {
+		int cpu = first_cpu(dest);
+
+		printk(KERN_DEBUG "redirecting irq %d from CPU %d to %d\n",
+		       irq, smp_processor_id(), cpu);
+		gsc_writel(irq + CPU_IRQ_BASE,
+			   cpu_data[cpu].hpa);
+		goto set_out;
 	}
+#endif
+	__do_IRQ(irq, regs);
 
-	set_eiem(cpu_eiem);	/* restore original mask */
+ out:
 	irq_exit();
-}
+	return;
 
+ set_out:
+	set_eiem(cpu_eiem & global_ack_eiem & per_cpu(local_ack_eiem, cpu));
+	goto out;
+}
 
 static struct irqaction timer_action = {
 	.handler = timer_interrupt,
 	.name = "timer",
-	.flags = IRQF_DISABLED,
+	.flags = IRQF_DISABLED | IRQF_TIMER | IRQF_PERCPU,
 };
 
 #ifdef CONFIG_SMP
 static struct irqaction ipi_action = {
 	.handler = ipi_interrupt,
 	.name = "IPI",
-	.flags = IRQF_DISABLED,
+	.flags = IRQF_DISABLED | IRQF_PERCPU,
 };
 #endif
 
diff --git a/arch/parisc/kernel/smp.c b/arch/parisc/kernel/smp.c
index f33e8de438dca4..faad338f310ed1 100644
--- a/arch/parisc/kernel/smp.c
+++ b/arch/parisc/kernel/smp.c
@@ -262,6 +262,9 @@ ipi_interrupt(int irq, void *dev_id, struct pt_regs *regs)
 					this_cpu, which);
 				return IRQ_NONE;
 			} /* Switch */
+		/* let in any pending interrupts */
+		local_irq_enable();
+		local_irq_disable();
 		} /* while (ops) */
 	}
 	return IRQ_HANDLED;
diff --git a/drivers/parisc/iosapic.c b/drivers/parisc/iosapic.c
index 1fbda77cefc29b..90489ade632e08 100644
--- a/drivers/parisc/iosapic.c
+++ b/drivers/parisc/iosapic.c
@@ -692,6 +692,7 @@ static void iosapic_end_irq(unsigned int irq)
 	DBG(KERN_DEBUG "end_irq(%d): eoi(%p, 0x%x)\n", irq,
 			vi->eoi_addr, vi->eoi_data);
 	iosapic_eoi(vi->eoi_addr, vi->eoi_data);
+	cpu_end_irq(irq);
 }
 
 static unsigned int iosapic_startup_irq(unsigned int irq)
@@ -728,7 +729,7 @@ static struct hw_interrupt_type iosapic_interrupt_type = {
 	.shutdown =	iosapic_disable_irq,
 	.enable =	iosapic_enable_irq,
 	.disable =	iosapic_disable_irq,
-	.ack =		no_ack_irq,
+	.ack =		cpu_ack_irq,
 	.end =		iosapic_end_irq,
 #ifdef CONFIG_SMP
 	.set_affinity =	iosapic_set_affinity_irq,
diff --git a/include/asm-parisc/irq.h b/include/asm-parisc/irq.h
index 6e29cfa2812d84..399c81981ed5a2 100644
--- a/include/asm-parisc/irq.h
+++ b/include/asm-parisc/irq.h
@@ -39,6 +39,8 @@ struct irq_chip;
  */
 void no_ack_irq(unsigned int irq);
 void no_end_irq(unsigned int irq);
+void cpu_ack_irq(unsigned int irq);
+void cpu_end_irq(unsigned int irq);
 
 extern int txn_alloc_irq(unsigned int nbits);
 extern int txn_claim_irq(int);