Messages in this thread | | | Date | Mon, 17 Jun 2002 11:00:26 +0200 (CEST) | From | Ingo Molnar <> | Subject | Re: [patch] 2.4.19-pre10-ac2: O(1) scheduler merge, -A3. |
| |
On Mon, 17 Jun 2002, Zwane Mwaikambo wrote:
> > > Can we add a config time option for irqbalance? I consider it extra > > > overhead for setups which can do the interrupt distribution via hardware > > > properly, [...] > > > > What x86 hardware do you have in mind? > > ye olde generic x86 SMP box, the interrupt handling imbalance came about > with the P4 and their newer APIC setup did it not? Although i am aware > that some x86 SMP boxes don't do the distribution properly too, thats > why i reckon config option would be best.
even generic x86 SMP boxes benefit from irqbalance due to better irq affinity. Actually one could hardly find a worse way to distribute interrupts than the IO-APIC + lowest priority delivery mode does... [in fact there is one, the P4's do it ;-) ]
> > > [...] also irqbalance breaks NUMAQ horribly seeing as it assumes a > > > number of things like addressing modes. > > > > exactly what does it assume that breaks NUMAQ? > > <Disclaimer> > I am not a NUMAQ expert and do not even have access to one for testing > </Disclaimer> > > The addressing mode, irq_balance assumes that the addressing mode is > logical mode (when programming the IOREDTBL entries), whilst NUMAQ uses > a completely different addressing architecture. [...]
irqbalance uses the set_ioapic_affinity() method to set affinity. The clustered APIC code is broken if it doesnt handle this properly. (i dont have such hardware so i cant tell, but it indeed doesnt appear to handle this case properly.) By wrapping around at node boundary the irqbalance code will work just fine.
> [...] Also another thing is consider this situation; > > irqbalance programs IOAPIC#0 on node0 to deliver to CPU#6 on node1 > > Will that interrupt get delivered?
i agree that this could be a problem, but set_ioapic_affinity() can be made dependent on the actual NUMA setup that is used. This is absolutely needed anyway for a proper /proc/irq/*/smp_affinity feature.
Ingo
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |