Re: [PATCH] pci: change msi-x vector to 32bit

[James Bottomley <James.Bottomley@HansenPartnership.com>]

On Mon, 2008-08-18 at 14:45 -0700, Eric W. Biederman wrote:
> > Yes, agree with the above except the static irq number ... although once
> > that's hidden from the user, I suppose I really don't care any more.
> 
> There is no point at all in having an irq number if it is not visible
> to user space.
> 
> Reaching a point where we don't export irq numbers to user space is
> a hard problem.  We have the irq balancer in user space (gag) and as
> well as interfaces like /proc/interrupts, and for old ISA devices
> there is no way to autodetect the configuration.  So it requires a lot
> of work.

Sure, but you have 16 (or whatever) legacy interrupts.  You still call
them 1-16 (or ISA-1 through ISA-16).  By the time we reach this stage,
we're essentially doing string table lookups for the interrupts, so
there's no need to pre-allocate them (except as a possible arch
implementation detail).

> The obvious first step is still to remove the architecture dependence
> on irq number, and introduce another way of talking about irqs.  Then
> we can proceed to change the driver and the user space interfaces.
> 
> I completely agree that irq number 99.9% of the time should be a completely
> abstract token.

Sure, although one nice reason for doing the abstraction first is that
it stops people imposing fragile numbering schemes on irq ...

James