Re: [BK PATCH] PCI and PCI Hotplug changes and fixes for 2.5.70

[Dave Jones]

On Thu, Jun 05, 2003 at 10:18:35AM -0700, Greg KH wrote:

 > > The fact that a tree-wide 'cleanup' like this goes in just a few hours
 > > after its posted before chance to comment is another argument, but
 > > concentrating on the technical point here, I still think this is a
 > > step backwards.
 > 
 > Why?  I just got rid of a function (well macro) that isn't even needed
 > (as proven by replacing it with an existing function.)  That's
 > technically a good thing :)

Not when that replacement reduces readability, which in the case of
agpgart is all I care about wrt these changes.

 > Now I can agree that some of those replacements could be done with a
 > different function call, as almost none of the replacements in the
 > driver/* tree really want to walk all of the pci devices in the tree.
 > They usually just want to walk all devices of a type of pci device (be
 > it capability, or other trait.)  I'd be glad to take changes of this
 > sort in the future.

Ok, for the ones I'm interested in, (agpgart), I don't see how things
can get much cleaner than they used to be.

07 - hammer. Needs to walk the whole list, matching pci devices on
     bus 0, func 3, slots >23 & <32
	 This set of rules is so specialised, I see it hard to concieve how
	 a generic helper function for the pci layer could be written.
08 - generic. Needs to know about every AGP device on the bus.
     ok, a for_each_agp_dev may actually make life easier here,
	 but as its the only place this happens, consider it inlined,
	 using pci_for_each_dev
09 - isoch. Could also use a 'for_each_agp_dev', but to be honest,
     it shouldn't be scanning at all, but being passed what it needs.

For the time being, I'm actually tempted to hack up a 'for_each_agp_dev'
for the latter two, but 07 still bugs me.

		Dave