Re: PATCH: straighten out the IDE layer locking and add hotplug

[Alan Cox <alan@redhat.com>]

On Tue, Aug 17, 2004 at 04:30:07PM +0200, Bartlomiej Zolnierkiewicz wrote:
> You forgot about the sad fact that most host drivers can be modular
> thanks to prematuraly making them modular in 2.4. :/

If they are modular their order already changes in some case

> ide_match_hwif() checks for hwif->chipset - ordering will not be the same
> i.e. you load driver for some IDE PCI controller which doesn't have drives 
> attached to it, unload it, load some other driver - hwifs will be reused - 
> some sequence in 2.4 will possibly leave you with different ordering because 
> hwif->chipset will stay as ide_pci not ide_unknown

You can't unload them in 2.4. 

> There are other much more crazy scenerios when you consider using 
> HDIO_SCAN_HWIF nad HDIO_UNREGISTER_HWIF ioctls. ;)

Those cases yes

> > Once you have drive and controller hot plug you don't need them. Until then
> 
> Yep, please tell me how are you going to support drive hot plug?

We can do it the 2.4-ac way - that works with the locking I think. What
might be nicer if it works out is to follow the shutdown/suspend code
approach so that we actually queue the "unplug" into the command stream.

> > some laptop users rely on them. I'd prefer to ignore the issue (its a
> > privileged code path) until the hotplug is there, or patch it up by
> > allowing unregister only of SCAN_HWIF added hwifs ?
> 
> I prefer short deprecation -> removal -> forgetting about them. :)

Ditto but I need the thinkpad docking bay working somehow.

Alan