Re: [tip:x86/apic] x86: Fix out of order gsi -- add remap_ioapic_gsi_to_irq()

[ebiederm@xmission.com (Eric W. Biederman)]

Yinghai Lu <yinghai@kernel.org> writes:

> On 02/27/2010 01:30 PM, Eric W. Biederman wrote:
> then will have all irq_desc for ioapic stay with BSP node.

This is what I am thinking....

static void __init setup_IO_APIC_irqs(void)
{
	int apic_id, pin, idx, irq;
	int notcon = 0;
	struct irq_desc *desc;
	struct irq_cfg *cfg;
	int node = cpu_to_node(boot_cpu_id);

	apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");

	for(apic_id = 0; apic_id < nr_ioapics; apic_id++) {
		for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
			idx = find_irq_entry(apic_id, pin, mp_INT);
			if (idx == -1) {
				if (!notcon) {
					notcon = 1;
					apic_printk(APIC_VERBOSE,
						KERN_DEBUG " %d-%d",
						mp_ioapics[apic_id].apicid, pin);
				} else
					apic_printk(APIC_VERBOSE, " %d-%d",
						mp_ioapics[apic_id].apicid, pin);
				continue;
			}
			if (notcon) {
				apic_printk(APIC_VERBOSE,
					" (apicid-pin) not connected\n");
				notcon = 0;
			}
			
			irq = pin_2_irq(idx, apic_id, pin);

+			if ((apic_id > 0) && (irq > 16))
+				continue;

			
			/*
			 * Skip the timer IRQ if there's a quirk handler
			 * installed and if it returns 1:
			 */
			if (apic->multi_timer_check &&
				apic->multi_timer_check(apic_id, irq))
				continue;
			
			desc = irq_to_desc_alloc_node(irq, node);
			if (!desc) {
				printk(KERN_INFO "can not get irq_desc for %d\n", irq);
				continue;
			}
			cfg = desc->chip_data;
			add_pin_to_irq_node(cfg, node, apic_id, pin);
			/*
			 * don't mark it in pin_programmed, so later acpi could
			 * set it correctly when irq < 16
			 */
			setup_IO_APIC_irq(apic_id, pin, irq, desc,
				irq_trigger(idx), irq_polarity(idx));
		}
	}
	
	if (notcon)
		apic_printk(APIC_VERBOSE,
			" (apicid-pin) not connected\n");
}


>> YH your current remapping patch looks like a pretty horrible hack
>> instead of real solution to the problem.  I honestly think starting
>> with it will just obscure what is going and make it harder to
>> introduce a clean gsi_to_irq/irq_to_gsi.
>
> good, the mapping looks much clear.

Thanks.  This code is just too convoluted to introduce many more
hacks and not have it fall over.

Eric