[PATCH v3 1/2] powerpc/PCI: move DMA & IRQ init to device_add() notification path

Bjorn Helgaas bhelgaas at google.com
Tue Jun 19 07:06:41 EST 2012

I'm trying to make some progress on these patches, but I'm concerned
about this bit:

> diff --git a/drivers/pcmcia/cardbus.c b/drivers/pcmcia/cardbus.c
> index 24caeaf..a980691 100644
> --- a/drivers/pcmcia/cardbus.c
> +++ b/drivers/pcmcia/cardbus.c
> @@ -85,7 +84,6 @@ int __ref cb_alloc(struct pcmcia_socket *s)
>         */
>        pci_bus_size_bridges(bus);
>        pci_bus_assign_resources(bus);
> -       cardbus_config_irq_and_cls(bus, s->pci_irq);
>        /* socket specific tune function */
>        if (s->tune_bridge)
> @@ -93,6 +91,7 @@ int __ref cb_alloc(struct pcmcia_socket *s)
>        pci_enable_bridges(bus);
>        pci_bus_add_devices(bus);
> +       cardbus_config_irq_and_cls(bus, s->pci_irq);
>        return 0;
>  }

We're moving the CardBus IRQ config from before pci_bus_add_devices()
to after.  I see why you did that: we're proposing to do the powerpc
DMA & IRQ setup in pci_bus_add_devices(), so we don't want to have the
powerpc IRQ init clobber the CardBus IRQ config.

But a driver can claim the device as soon as we call
pci_bus_add_devices(), so we're potentially changing dev->irq after a
driver has already looked at it, which sounds like a bug.

There are only five possibilities for powerpc pci_irq_fixup:


If these were normal PCI header quirks instead, they could run
earlier, and we wouldn't need to move this
cardbus_config_irq_and_cls() call.  Is it possible to make these
quirks, Ben?


More information about the Linuxppc-dev mailing list