Skip to contentSkip to conversation

2.5.34-bk: ide interfaces sharing same irq broken...

PATCH? Petr Vandrovec 2 participants 2 messages Sep 15, 2002 – Sep 16, 2002

lore
Share & tools
lore
#0 lore raw

Hi Jens,
can you look at attached patch? Without this patch my machine dies while
probing IDE with endless stream of 'unexpected irq'.

Problem is both primary and secondary channels of my PDC20265 use irq 10.
When we probe first interface, everything is fine, as irq 10 is disabled, and so irq assertion is silently ignored (but it would die if I'll start using USB which also wants irq 10...). But when secondary channel was probed, irq 10 was asserted, and things went really wrong... So I made disable_irq() unconditional: I see no reason why it should depend on ack_intr. Any code which will cause IRQ assertion, and does not install either IRQ handler, or which does not disable that IRQ, is buggy.

Another problem I noticed is that actual probe code in probe_hwif between
disable_irq() and enable_irq() could change hwif->irq, and so we could enable irq we did not disable (it happened to me after I removed ack_intr
test: irq is 0 on entry when probing onboard VIA, but on exit it is 14... so we enabled irq 14 source without previously disabling it, and it caused bad things to happen).

This patch makes disabling IRQ during probe unconditional, and makes sure
that we enable same irq we disabled.

Of course that moving init_irq call before probe would be nicer, and
it would solve also pending edge-triggered IRQ, but unfortunately as code is currently written, we do not know IRQ until after probe, and at this point we already asserted IRQ, even on PCI devices which can share IRQ with someone else.

After this patch, and after commenting out "ide_intr: unexpected interrupt"
printk in ide.c:ide_intr system works finally fine.
Thanks,
Petr Vandrovec [email protected]

P.S.: I set 'Cc' list to my best knowledge. If you feel that you are missing (Andre & Alan, I believe that you both work on 2.4 only), add yourself to the "IDE DRIVER [GENERAL]" entry in MAINTAINERS. There is still Martin listed there.

drivers/ide/ide-probe.c +11 -3
diff -urdN linux/drivers/ide/ide-probe.c linux/drivers/ide/ide-probe.c
@@ -592,6 +592,7 @@
 {
 	unsigned int unit;
 	unsigned long flags;
+	unsigned int irqd;
 
 	if (hwif->noprobe)
 		return;
@@ -623,8 +624,14 @@
 		return;	
 	}
 
-	if (hwif->hw.ack_intr && hwif->irq)
+	/* We must always disable IRQ, as probe_for_drive will assert IRQ, but
+	   we'll install our IRQ driver much later... */
+	if ((1 || hwif->hw.ack_intr) && hwif->irq) {
 		disable_irq(hwif->irq);
+		irqd = hwif->irq;
+	} else {
+		irqd = 0;	/* Use 0... IDE does not support using irq=0 at all... */
+	}
 
 	local_irq_set(flags);
 	/*
@@ -659,8 +666,9 @@
 
 	}
 	local_irq_restore(flags);
-	if (hwif->hw.ack_intr && hwif->irq)
-		enable_irq(hwif->irq);
+	/* Use cached IRQ number. It might be (and is...) changed by probe code above */
+	if (irqd)
+		enable_irq(irqd);
 
 	for (unit = 0; unit < MAX_DRIVES; ++unit) {
 		ide_drive_t *drive = &hwif->drives[unit];
#1 lore raw
Quoted 23 lines · Petr Vandrovec

Hi Jens,
can you look at attached patch? Without this patch my machine dies while
probing IDE with endless stream of 'unexpected irq'.

Problem is both primary and secondary channels of my PDC20265 use irq 10.
When we probe first interface, everything is fine, as irq 10 is disabled, and so irq assertion is silently ignored (but it would die if I'll start using USB which also wants irq 10...). But when secondary channel was probed, irq 10 was asserted, and things went really wrong... So I made disable_irq() unconditional: I see no reason why it should depend on ack_intr. Any code which will cause IRQ assertion, and does not install either IRQ handler, or which does not disable that IRQ, is buggy.

Another problem I noticed is that actual probe code in probe_hwif between
disable_irq() and enable_irq() could change hwif->irq, and so we could enable irq we did not disable (it happened to me after I removed ack_intr
test: irq is 0 on entry when probing onboard VIA, but on exit it is 14... so we enabled irq 14 source without previously disabling it, and it caused bad things to happen).

This patch makes disabling IRQ during probe unconditional, and makes sure
that we enable same irq we disabled.

Quoted 8 lines

Of course that moving init_irq call before probe would be nicer, and
it would solve also pending edge-triggered IRQ, but unfortunately as code is currently written, we do not know IRQ until after probe, and at this point we already asserted IRQ, even on PCI devices which can share IRQ with someone else.

After this patch, and after commenting out "ide_intr: unexpected interrupt"
printk in ide.c:ide_intr system works finally fine.

Thanks Petr, very good spotting! I'll pass it along with the unexpected intr removal, that it seems Linus lost (it was sent). Ed, could you see if this solves your problem as well?

ide/ide.c renamed from opt/kernel/linux-2.5.35/drivers/ide/ide.c +0 -1
@@ -1376,7 +1376,6 @@
 
 	if ((handler = hwgroup->handler) == NULL ||
 	    hwgroup->poll_timeout != 0) {
-		printk("ide_intr: unexpected interrupt!\n");
 		/*
 		 * Not expecting an interrupt from this drive.
 		 * That means this could be:
ide/ide-probe.c renamed from opt/kernel/linux-2.5.35/drivers/ide/ide-probe.c +13 -3
@@ -592,6 +592,7 @@
 {
 	unsigned int unit;
 	unsigned long flags;
+	unsigned int irqd;
 
 	if (hwif->noprobe)
 		return;
@@ -623,7 +624,12 @@
 		return;	
 	}
 
-	if (hwif->hw.ack_intr && hwif->irq)
+	/*
+	 * We must always disable IRQ, as probe_for_drive will assert IRQ, but
+	 * we'll install our IRQ driver much later...
+	 */
+	irqd = hwif->irq;
+	if (irqd)
 		disable_irq(hwif->irq);
 
 	local_irq_set(flags);
@@ -659,8 +665,12 @@
 
 	}
 	local_irq_restore(flags);
-	if (hwif->hw.ack_intr && hwif->irq)
-		enable_irq(hwif->irq);
+	/*
+	 * Use cached IRQ number. It might be (and is...) changed by probe
+	 * code above
+	 */
+	if (irqd)
+		enable_irq(irqd);
 
 	for (unit = 0; unit < MAX_DRIVES; ++unit) {
 		ide_drive_t *drive = &hwif->drives[unit];
Signature · 1 line
Jens Axboe