list etiquette [was Re: [mythtv-users] OT? /proc/pci]

[David George]

On Mon, 12 Jul 2004, Tim Tait wrote:

> Jason Kraftcheck wrote:
>
> Joel Anderson wrote:
>
> Marc-David Meijer wrote:
>
> As I understand it, PCI devices don't depend on IRQs anyway.  They use
> interrupt "channels", and IRQs are just kept around to keep legacy stuff
> happy.  You will often see many devices appear to "share" the same IRQ,
> but it is not really true since IRQ's can't really _be_ shared.
>
> Someone please comment if I haven't explained this correctly.  It's a
> simplification, but it basically means that IRQ issues shouldn't come
> into play unless you have some legacy hardware in the system also.
>
>
>
> I'm no expert either, but I think they are actually "shared".  The
> interrupt handler must poll each device by invoking the interrupt
> handler in the driver for each device sharing the interrupt.  Thus
> there's a slight performance loss when sharing interrupts, and also the
> possibility for buggy drivers to interfere with each other.
>
>
> Newer PCI devices support "Message Signalled Interrupts" which use no
> physical IRQ line, but instead write a pre-configured value to the PCI
> bridge inbound MSI register to cause a unique interrupt. I can't say for
> sure the 250MCE actually does that (or that the driver sets it up that
> way) but it is possible.  People have commented eslwhere on the list
> that it appears the 250MCE needs a PCI Rev 2.2 compliant motherboard, so
> you might want to check your  mobo docs out and see if it complies.
>
> Tim

Is there a reply here somewhere?  Please don't send html to mail lists.