On 10/09/18 04:48, David Gibson wrote: > On Sat, Sep 08, 2018 at 10:08:20AM +0100, Mark Cave-Ayland wrote: >> Whilst the PReP specification describes how all PCI IRQs are routed via IRQ >> 15 on the interrupt controller, the real 40p machine has routing quirk in >> that the LSI SCSI device is routed to IRQ 13. >> >> This is implemented using a little hack: the existing IRQ routing code uses >> (irq_num + (pci_dev->devfn >> 3)) & 1 to give the PCI interrupt pin, where >> the "& 1" ensures that the only pins A and B (0 and 1) will ever be used. >> >> Rather than fix the mask to "& 3" we leave the existing routing above as-is >> and then force the LSI SCSI device to use pin C (2). This enables us to >> route pin 2 permanantly to IRQ 13 since the LSI SCSI device will be its >> only user. >> >> Signed-off-by: Mark Cave-Ayland <mark.cave-ayl...@ilande.co.uk> > > I don't think this is really the right approach. > > As noted in an earlier mail, it's really common for on-board devices > on pre-express PCI boards to have their LSIs wired specially, rather > than via the PCI irq pins which are used for "slotted" devices. > > I think we should explicitly model it like that: wiring the SCSI > device's irq directly to system IRQ 13, rather than wiring it via a > PCI LSI pin. > > Wiring the SCSI to an otherwise unused pin, then routing that > specially to the system irq is confusing. It also might be incorrect > if we tried to add a "slotted" device that actually used PINC (rare, > I'll grant you).
Okay. I was hoping to get away with not altering the LSI SCSI device itself, but after some experimentation I managed to come up with a solution that is not too invasive and certainly more readable than this attempt. Let me tidy up and post a v3. ATB, Mark.
