RE: Message Signalled Interrupt Support

[Lame Brooks-G14738 <brooks.lame@motorola.com>]

> Could anyone please explain how the MSI is supposed to work??

First, _read_ PCI 2.2, section 6.8.  It explains that instead of a PCI
master device signaling a shared interrupt vector via a pin, the device can
signal a unique vector via a write transaction to a system address.
Presumably, the PnP OS will configure each MSI capable device with a system
MSI receiving address to write to (probably unique per MSI device) and
allocate 1 to 32 unique 'vectors' (the message data) per device.  Then, the
vector data written to the MSI receiving address will be handed off to the
device driver.  There's a lot of flexibility in there, so each platform may
handle the specifics a little differently, but I think that's about right.
This is way cooler than four shared interrupt pins!  Of course only master
devices can use MSI; slave devices must stick with the int# pins.  Comments
from system software implementers?
-- BrooksL

> ----- Original Message -----
> From: Carl Jackson <jackson@rsn.hp.com>
> To: PCI SIG email reflector <pci-sig@znyx.com>
> Sent: Wednesday, February 23, 2000 4:52 AM
> Subject: Re: Message Signalled Interrupt Support
> 
> 
> > At 05:29 PM 2/20/00 , Brad Hosler wrote:
> > >PCI 2.2 defines a Message Signalled Interrupt (MSI) 
> feature, to allow the
> card
> > >to issue an interrupt as a PCI write message.
> > >
> > >Its presenting a "who-comes-first" scenario.  There are SW 
> developers
> that
> > >want to support this feature; however, they want to know 
> if there are
> IHV's
> > >who are adding support to their cards.  If there is no IHV 
> interest,
> > >SW support
> > >may be pulled due to lack of validation.
> > >
> > >Are there any cards available or that will be available 
> that will support
> > >the MSI capability?
> >
> >
> > PCI-X requires devices that generate interrupts to support 
> MSI (with a
> > 64 bit message address).  MSI is still optional for platforms and
> > system software but the required availability should make support in
> > systems much more likely.
> >
> > --
> > Carl Jackson                      Hewlett-Packard Company
> > voice: (972)497-4268              High Performance Systems Lab
> > fax:   (972)497-4245              3000 Waterview Parkway
> > mailto:jackson@rsn.hp.com         Richardson, TX 75080
> >
> >
>