[PATCH] fix kirq code for clustered mode

[PATCH] fix kirq code for clustered mode

[Dave Hansen]

[-- Attachment #1: Type: text/plain, Size: 196 bytes --]

The new kirq code breaks clustered apic mode.  This 2-liner fixes it.
It should compile down to the same thing, unless you're using a
clustered apic sub-arch.

-- 
Dave Hansen
haveblue@us.ibm.com

[-- Attachment #2: kirq-apicid-fix-2.5.62-2.patch --]
[-- Type: text/plain, Size: 1016 bytes --]

diff -ru linux-2.5.62-clean/arch/i386/kernel/io_apic.c linux-2.5.62-kirqfix/arch/i386/kernel/io_apic.c
--- linux-2.5.62-clean/arch/i386/kernel/io_apic.c	Mon Feb 17 14:56:10 2003
+++ linux-2.5.62-kirqfix/arch/i386/kernel/io_apic.c	Tue Feb 18 09:44:22 2003
@@ -441,7 +441,7 @@
 		Dprintk("irq = %d moved to cpu = %d\n", selected_irq, min_loaded);
 		/* mark for change destination */
 		spin_lock(&desc->lock);
-		irq_balance_mask[selected_irq] = target_cpu_mask;
+		irq_balance_mask[selected_irq] = cpu_to_logical_apicid(min_loaded);
 		spin_unlock(&desc->lock);
 		/* Since we made a change, come back sooner to 
 		 * check for more variation.
@@ -515,7 +515,7 @@
 	
 	/* push everything to CPU 0 to give us a starting point.  */
 	for (i = 0 ; i < NR_IRQS ; i++)
-		irq_balance_mask[i] = 1 << 0;
+		irq_balance_mask[i] = cpu_to_logical_apicid(0);
 	for (;;) {
 		set_current_state(TASK_INTERRUPTIBLE);
 		time_remaining = schedule_timeout(time_remaining);
Only in linux-2.5.62-kirqfix/arch/i386/kernel: io_apic.c~

Re: [PATCH] fix kirq code for clustered mode

[William Lee Irwin III]

On Tue, Feb 18, 2003 at 09:51:28AM -0800, Dave Hansen wrote:
> The new kirq code breaks clustered apic mode.  This 2-liner fixes it.
> It should compile down to the same thing, unless you're using a
> clustered apic sub-arch.

This isn't quite enough:


static int irq_affinity_write_proc (struct file *file, const char *buffer,
					unsigned long count, void *data)
{
	int irq = (long) data, full_count = count, err;
	unsigned long new_value;

	if (!irq_desc[irq].handler->set_affinity)
		return -EIO;

	err = parse_hex_value(buffer, count, &new_value);

	/*
	 * Do not allow disabling IRQs completely - it's a too easy
	 * way to make the system unusable accidentally :-) At least
	 * one online CPU still has to be targeted.
	 */
	if (!(new_value & cpu_online_map))
		return -EINVAL;

	irq_affinity[irq] = new_value;
	irq_desc[irq].handler->set_affinity(irq, new_value);

	return full_count;
}

This is a bitmask and it's being handed directly to set_ioapic_affinity().
Or, at least, it's inconsistent, being &'d with cpu_online_map.


-- wli

Re: [PATCH] fix kirq code for clustered mode

[William Lee Irwin III]

On Tue, Feb 18, 2003 at 09:51:28AM -0800, Dave Hansen wrote:
>> The new kirq code breaks clustered apic mode.  This 2-liner fixes it.
>> It should compile down to the same thing, unless you're using a
>> clustered apic sub-arch.

On Thu, Feb 20, 2003 at 04:19:17AM -0800, William Lee Irwin III wrote:
> This isn't quite enough:

Also, even with your patch, I get:

Uhhuh. NMI received for unknown reason 25 on CPU 3.
Uhhuh. NMI received for unknown reason 35 on CPU 2.
Dazed and confused, but trying to continue
...

Hmm... something is going seriously wrong here. I suspect someone
is clobbering my RTE's even though no_irq_balance is #defined to 1.
Or, worse yet, NMI interrupt gate boogie is going seriously wrong.


-- wli