Re: i82559 on ebsa patch

[Andrew Lunn <andrew dot lunn at ascom dot ch>]

On Fri, Oct 18, 2002 at 02:09:10PM +0100, Jonathan Larmour wrote:
> Andrew Lunn wrote:
> >On Fri, Oct 18, 2002 at 04:25:41AM +0100, Jonathan Larmour wrote:
> >>
> >>And since I can't work it out, why is DEMUX_ALL no longer needed?
> >
> >
> >It gave me a headache trying to work this out and im not sure i can
> >remember correctly. DEMUX_ALL and MUX_INTERRUPTS are mutually
> >exclusive.
> 
> <Hugo>Oh yeah!</Hugo> There's a #error check for that even.

Hi Hugo.

> 
> Presumably the distinction is as described in the comment for DEMUX_ALL:
> 
> //               others.  In this circumstance it is permitted for
> //               cyg_pci_translate_interrupt [HAL_PCI_TRANSLATE_INTERRUPT]
> //               to return invalid for 2nd and subsequent devices.
> 
> 
> > If you have one device, its still on
> >CYGNUM_HAL_INTERRUPT_PCI_IRQ so you still need MUX_INTERRUPTS.
> 
> Hmm I'd have thought that that's ok since it will just attach the normal 
> eth_isr and everything will work fine with only one device. It should 
> still have recognised PCI_IRQ as the correct vector in the PCI scan surely?

I think there is a peculiarity here on the EBSA. The PCI code always
says the devices are on interrupt X and Y, when in fact they are on
PCI_IRQ. I don't remember the details, but i think its something like
the hardware is returning the wrong info to the pci library.
 
> >For testing i actually did disable one of the devices by CDL, since i
> >only have a DHCP server on one of them and was too impatient for
> >init_all_network_interfaces() to timeout and fail on the other before
> >running the net tests.
> 
> If you have actually tested that config, then I'm not so bothered - 
> DEMUX_ALL can't be required if it recognised the second device. I'm just 
> thinking now that there might be redundant code in use for one device.

There could be a small amount of redundant code, but the #defines are
already so hairy and headache inducing, im not sure the pain is worth
the gain. The SA110 is probably a dead platform now except for some of
Ascoms legacy hardware. All our hardware has two ethernets....

    Andrew