Re: [RFC][PATCH 1/4] PCI legacy I/O port free driver - Introducepci_set_bar_mask*()

[Kenji Kaneshige]

Russell King wrote:
> On Wed, Feb 15, 2006 at 03:03:56PM +0900, Kenji Kaneshige wrote:
>
> > Andrew Morton wrote:
> >
> > > Kenji Kaneshige <kaneshige.kenji@xxxxxxxxxxxxxx> wrote:
> > >
> > >
> > > > This patch introduces a new interface pci_select_resource() for PCI
> > > > device drivers to tell kernel what resources they want to use.
> > >
> > >
> > > It'd be nice if we didn't need to introduce any new API functions for this.
> > > If we could just do:
> > >
> > > struct pci_something pci_something_table[] = {
> > > 	...
> > > 	{
> > > 		...
> > > 		.dont_allocate_io_space = 1,
> > > 		...
> > > 	},
> > > 	...
> > > };
> > >
> > > within each driver which wants it.
> > >
> > > But I can't think of a suitable per-device-id structure with which we can
> > > do that :(
> >
> > My another idea was to use pci quirks. In this approach, we don't
> > need to introduce any new API. But I gave up this idea because it
> > looked abuse of pci quirks.
> >
> > Anyway, I try to think about new ideas we don't need to introduce
> > any new API.
>
>
> What about pci_enable_device_bars() ?

Yes, it's one option (In fact, my first idea was using it),
though we need to move the following lines into
pci_enable_device_bars() from pci_enable_device().

       pci_fixup_device(pci_fixup_enable, dev);
       dev->is_enabled = 1;

Actually, IIRC, there are one or two existing drivers using it.
In addition to pci_enable_device_bars(), we can use
pci_request_region() for requesting each region instead of using
pci_request_regions().

The reason I didn't use this option was I thought we would need
bigger changes to drivers if we use pci_enable_devie_bars() and
pci_request_region(). For example, if we use pci_request_region()
for requesting the specific regions and if an error occurs
while doing that, we need to release only the regions we succeeded
in requesting. I think this is a little bit troublesome for driver
writers. In adition, though this is my personal opinion, only one
API to enable devices (e.g. pci_enable_device()) looks nice to me.
Anyway, I think it would be nice if we can solve the problem by
as small changes as possible to the existing drivers.
How do you think?

Thanks,
Kenji Kaneshige
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/