v3 IRQ table strangeness

[Carl-Daniel Hailfinger]

Hi,

is it intentional that irq_tables.h in all v3 AMD GeodeLX/CS5536 targets
lists the "NatSemi CS5535 ISA Bridge" at 00:0f.0 as interrupt router?

By the way, the dbe61 patch Mart sent tonight changes that for dbe61 to
the "CS5536 GeodeLink PCI South Bridge" at 00:12.0 as interrupt router.

Regards,
Carl-Daniel

[ron minnich]

On Wed, May 7, 2008 at 7:09 PM, Carl-Daniel Hailfinger
 wrote:
> Hi,
>
> is it intentional that irq_tables.h in all v3 AMD GeodeLX/CS5536 targets
> lists the "NatSemi CS5535 ISA Bridge" at 00:0f.0 as interrupt router?
>
> By the way, the dbe61 patch Mart sent tonight changes that for dbe61 to
> the "CS5536 GeodeLink PCI South Bridge" at 00:12.0 as interrupt router.

we can change it, the old name is historical

ron

[Carl-Daniel Hailfinger]

On 08.05.2008 05:50, ron minnich wrote:
> On Wed, May 7, 2008 at 7:09 PM, Carl-Daniel Hailfinger
>  wrote:
> > Hi,
> >
> > is it intentional that irq_tables.h in all v3 AMD GeodeLX/CS5536 targets
> > lists the "NatSemi CS5535 ISA Bridge" at 00:0f.0 as interrupt router?
> >
> > By the way, the dbe61 patch Mart sent tonight changes that for dbe61 to
> > the "CS5536 GeodeLink PCI South Bridge" at 00:12.0 as interrupt router.
>
> we can change it, the old name is historical

So PCI IDs and bus location of the interrupt router in the PIRQ table
have no meaning at all for Geode?

Regards,
Carl-Daniel

[ron minnich]

On Thu, May 8, 2008 at 3:30 PM, Carl-Daniel Hailfinger <
> So PCI IDs and bus location of the interrupt router in the PIRQ table
> have no meaning at all for Geode?

No, because linux (and who knows what else) don't even know what it is.

At present, the PIRQ is only used by coreboot!

thanks

ron

[Tom Sylla]

ron minnich wrote:
> On Thu, May 8, 2008 at 3:30 PM, Carl-Daniel Hailfinger <
> > So PCI IDs and bus location of the interrupt router in the PIRQ table
> > have no meaning at all for Geode?
>
> No, because linux (and who knows what else) don't even know what it is.
>
> At present, the PIRQ is only used by coreboot!

Well, Linux *could* know about it, with not too much work. 5536 and 5536 
have the same router interface as 5530 and therefore 5520 (the emulated 
in vsa stuff from an earlier thread). Linux knows about the 5520 and how 
to use it. The goal of the "Compatible PCI Interrupt Router" field in 
the tables is so an OS can have one driver, and use it for multiple 
routers. Linux does not actually use that field, it looks like it just 
checks device ID. 5535 and 5536 could be added in the same way as 5530 
was in this patch:

http://www.pengutronix.de/software/ptxdist/temporary-src/references/geode-55...

which is linked from coreboot's wiki.

So, you could do it pretty easily. I don't know what advantage there is 
to letting Linux steer IRQs on a geode, though. Plus, 5536 v3 coreboot 
PIRQ tables have one IRQ in their bitmaps, and no PCI exclusive (as Marc 
pointed out), so the router doesn't matter anyway.

[ron minnich]

On Thu, May 8, 2008 at 6:15 PM, Tom Sylla  wrote:
> So, you could do it pretty easily. I don't know what advantage there is to
> letting Linux steer IRQs on a geode, though. Plus, 5536 v3 coreboot PIRQ
> tables have one IRQ in their bitmaps, and no PCI exclusive (as Marc pointed
> out), so the router doesn't matter anyway.

It's why I did not bother. I didn't see anything useful for LInux to do.

Except stop saying "Guessed interrupt" when it realizes that there is
nothing it can do :-)

ron