Re: [PATCH] sparse_irq aka dyn_irq v13

[Mike Travis <travis@sgi.com>]

David Miller wrote:
> From: Mike Travis <travis@sgi.com>
> Date: Thu, 13 Nov 2008 16:15:12 -0800
> 
>> David Miller wrote:
>>> From: Mike Travis <travis@sgi.com>
>>> Date: Thu, 13 Nov 2008 15:11:29 -0800
>>>
>>> We use a value of 256 and I've been booting linux on 128 cpu sparc64
>>> systems with lots of PCI-E host controllers (and others have booted it
>>> on even larger ones).  All of which have several NUMA domains.
>>>
>>> It's not an issue.
>> Are you saying that having a fixed count of IRQ's is not an issue?  With
>> NR_CPUS=4096 what would you fix it to?  (Currently it's NR_CPUS * 32
>> but that might not be sufficient.)  Would NR_CPUS=16384 make it an issue?
> 
> Nope, and nope.  I frequently run kernels with NR_CPUS set to huge
> values.
> 
> It seems that the issue of x86 is that it has it's IRQ count tied to
> the number of cpus, that's not very intelligent.  Perhaps that part
> should be rearranged somehow?

Yes, you're probably right but it is what it is.  Most of the irq vectors
have more to do with cpus than with i/o devices (the system vectors, ipi,
kdb and gru [a uv thing] interrupt vectors come first to mind.)  These do
by necessity need to grow with NR_CPUS, if you're fixing the total IRQ count.

There's been a couple of different proposals to attempt to disassociate
i/o and system vectors though even attempting to guess at the number of
i/o devices is tricky.  Every one of the 512 nodes on a UV system *may*
have a number of i/o devices attached to them though practically this
will be rare.