Re: [PATCH 0/8] git pull request for tip/tracing/core

[Jack Steiner]

On Wed, Feb 11, 2009 at 09:57:00AM -0800, Luck, Tony wrote:
> > > Before we go and make the change, Peter brought up a good point on IRC. Is
> > > there any reason that ia64 needs 1 << 14 IRQs?  That's 16384!
> > >
> > > Perhaps the better solution wolud be (if possible), to simply lower the
> > > number of bits.
> >
> > i'm the wrong person to be asked about that. (Cc:-ed the right people)
> 
> People build some pretty big systems on ia64.  SGI's largest has 4096
> cpus ... so 16384 IRQs is only 4 per cpu.  That doesn't sound like very
> many to me.

As long as per-cpu interrupts (ex., timer, cpe_hndlr, cmc_hndlr, etc) are not counted
in the IRQ count, SGI is ok with  256 IRQs. We had problems overflowing the old
limit of 256 but have worked around it.

If per-cpu interrupts ARE counted in the total IRQs, then we currently use 5
per-cpu interrupts for each cpu. A 4k cpu config will need 20k IRQs + additional
IRQs for real devices.

I know the IRQ infrastructure has changed recently so this may no longer apply.


--- jack
--
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/