Re: Double Interrupt with HT

[john stultz <johnstul@us.ibm.com>]

On Tue, 2003-12-16 at 14:55, Miroslaw KLABA wrote:
> I had the problem with 2.4.22, 2.4.22-ac4, 2.4.23 and 2.4.24-pre1.

Ok, so its been around awhile. Do you remember what was the last 2.4
kernel where you did not see this problem?

> The problem is that all the kernel is working "twice the speed".
> The command "while true; do date; sleep 1; done;" shows that the date is growing
> 2 seconds per second... :/
> I found a patch for irqbalance for 2.4.23, and now I don't have the problem 
> anymore with the clock.
> http://www.hardrock.org/kernel/2.4.23/irqbalance-2.4.23-jb.patch

Hmm. Just skimming that patch, I notice it won't work on clustered apic
systems. They've dropped the following chunk from set_ioapic_affinity
and forgot to re-add it.

-	/* pick a single cpu for clustered xapics */
-	if(clustered_apic_mode == CLUSTERED_APIC_XAPIC){
-		int cpu = ffs(mask)-1;
-		mask = cpu_to_physical_apicid(cpu);
-	}

Further I can't see how it fixes the problem, but it may just be working
around the issue. I'd be interested in what the patch author thinks. 

> I think it is a bug with the via chipset, but I'm not able to get deeper in the
> kernel code.

Could be, but I suspect interrupt routing isn't happening properly at
boot time. The irqbalance code just forces it to be readjusted correctly
once your up and running. 

thanks
-john