Re: IRQ Routing. [A 2.6 question about smp irq balancing]

[Martin J. Bligh]

>> > Are their any linux tools to allow the user to view irq routing details, 
>> > and maybe change the routing after boot ?
>> > 
>> > This might be useful in special cases.
>> 
>> Yeah, "cat /proc/interrupts" and 
>> "echo <cpu_bitmask> > /proc/irq/<number>/smp_affinity"
>> 
>> M.
> 
> 	Hi;
> 
> 	Here is my /proc/interrupts for 2.6.0-test6:
> 
>            CPU0       CPU1       
>   0:      10442 1795758405    IO-APIC-edge  timer
>   1:     737209          0    IO-APIC-edge  i8042
>   2:          0          0          XT-PIC  cascade
>   3:        517          1    IO-APIC-edge  serial
>   4:  100488033          1    IO-APIC-edge  serial
>   5:          0          0    IO-APIC-edge  eth0
>   8:          2          0    IO-APIC-edge  rtc
>  12:    1626560          1    IO-APIC-edge  i8042
>  14:    8988748          8    IO-APIC-edge  ide0
>  15:     377275          1    IO-APIC-edge  ide1
>  16:    1364704          2   IO-APIC-level  eth1
>  17:     857789     144630   IO-APIC-level  eth2
>  19:     165579          1   IO-APIC-level  Ensoniq AudioPCI
> NMI: 1795768424 1795763795 
> LOC: 1795877980 1795878044 
> ERR:          0
> MIS:        351
> 
> 	Now, on 2.4 kernels, both columns are about even.
> Upon noticing this, I tried to get more interrupts on cpu1.
> You can see I got some on eth2, by running many 'ping -f's at the
> same time. I havent been able to get anything more out of ide[01]
> for cpu1, even though Ive run many -j4 compiles of huge packages,
> which coincided with other things that bring my load up to 8+ over
> the course of 20 days of uptime. /proc/irq/<num>/smp_affinity
> is 3 (ie. both cpus) for all <num>. I was able to induce interrupts
> on cpu1 for ide0 by setting it to 2.
> 	Ok, my question is, is this how it is supposed to be?
> I ask because I am seeing a throughput decrease on 2.6 v 2.4
> on hdparm -t, bonnie, kernel compile benchmarks, though interactivity
> seems to be better.

I doubt that's caused by interrupt load, but try this:

diff -urN linux-2.6.0-test4-clean/arch/i386/kernel/io_apic.c
linux-2.6.0-test4-div10/arch/i386/kernel/io_apic.c
--- linux-2.6.0-test4-clean/arch/i386/kernel/io_apic.c 2003-09-15
06:46:02.000000000 -0700
+++ linux-2.6.0-test4-div10/arch/i386/kernel/io_apic.c 2003-09-15
06:47:17.000000000 -0700
@@ -393,7 +393,7 @@
        unsigned long max_cpu_irq = 0, min_cpu_irq = (~0);
        unsigned long move_this_load = 0;
        int max_loaded = 0, min_loaded = 0;
-       unsigned long useful_load_threshold = balanced_irq_interval +
10;
+       unsigned long useful_load_threshold = (balanced_irq_interval +
10) / 10;
        int selected_irq;
        int tmp_loaded, first_attempt = 1;
        unsigned long tmp_cpu_irq;



-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/