[ Greg added to cc: ]

On Monday 14 of June 2004 11:58, Christoph Hellwig wrote:

On Sun, Jun 13, 2004 at 07:36:08PM +0200, Bartlomiej Zolnierkiewicz wrote:

quoted

quoted

IMHO the PCI ->probe methods should always be __devinit.  It's rather
hard to make sure they're never every hotplugged in any way, especially
with the dynamic id adding via sysfs thing.

I generally agree but IMO it makes no sense for i.e. piix.c.

Are you sure?  I've seen piix3/4 in very strange place, iirc even in
a docking station which is hotpluggable.

Do you mean that south-bridge chipset itself is hotpluggable?

AFAIK it is only ATA hotplug not PCI one.

And even if for this special hardware it's usually not doable there
are things like greg's fake hotplug pci driver.  So a non-__devinit pci
probe method is a bug, please fix them in PCI.

Greg, should I add "fake" PCI hotplug support to some IDE
drivers just to make fake hotplug PCI driver happy?

Cheers.