Re: [PATCH 02/04] x86: add get_irq_cfg in io_apic_64.c

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

Mike Travis <travis@sgi.com> writes:


>> arch/x86/kernel/io_apic_64.c | 181 +++++++++++++++++++++++++++++++++----------
>>  1 file changed, 141 insertions(+), 40 deletions(-)
>> 
>> Index: linux-2.6/arch/x86/kernel/io_apic_64.c
>> ===================================================================
>> --- linux-2.6.orig/arch/x86/kernel/io_apic_64.c
>> +++ linux-2.6/arch/x86/kernel/io_apic_64.c
>> @@ -57,7 +57,11 @@
>>  
>>  #define __apicdebuginit(type) static type __init
>>  
>> +struct irq_cfg;
>> +
>>  struct irq_cfg {
>> +	unsigned int irq;
>> +	struct irq_cfg *next;
>>  	cpumask_t domain;
>>  	cpumask_t old_domain;
>         ^^^^^^^^^
> One thought here... most interrupts cannot be serviced by any cpu in
> the system, but instead need to be serviced by the cpu attached to
> the ioapic or on the local node.  So defining some subset of cpumask_t
> would save a lot of space.  For example:
>
> 	nodecpumask_t {
> 		int	node;
> 		DEFINE_BITMAP(..., MAX_CPUS_PER_NODE);
> 	};
>
> And of course, providing some utilities to convert nodecpumask_t <==>
> cpumask_t.
>
> ("node" might not be the proper abstraction... maybe "irqcpumask_t"?

I agree this is someplace we could optimize.  In practice we seem
to have 3 choices on x86.

1) A single cpu.
2) lowest priority interrupt delivery to a set of possibly 8 cpus.
3) A class of interrupt that is delivered locally to each individual
   cpu.

If you have a true NUMA system it should still be possible to handle
interrupts on the wrong Node just prohibitively expensive.

Eric