Re: [patch 2.6.10-rc3 1/4] agpgart: allow multiple backends to be initialized

From: Mike Werner
Date: Sat Dec 18 2004 - 12:18:27 EST

On Saturday 18 December 2004 06:48, Christoph Hellwig wrote:
> On Fri, Dec 17, 2004 at 12:55:59PM -0800, Mike Werner wrote:
> > This new version reduces the number of changes required by users of the agpgart
> > such as drm to support the new api for multiple agp bridges.
> > The first patch doesn't touch any platform specific files and all current platform
> > gart drivers will just work the same as they do now since the global
> > agp_bridge is still supported as the default bridge.
> The agp_bridge_find function pointer is bogus, that way you can only support
> one backend at a time.
Obviously you mean one type of backend here.
I have tried to simplify this patch as much as possible so that it only tries to do one thing
and that is just the api change. I think the searching for valid bridges is a separate issue
since none of the currently supported hardware needs it. The only possible platform
that I assumed might is amd64 and Andi Kleen specifically told me it doesn't.

> Most other bits of the patch are fine, but in either case you first need to
> change the agp bridge driver API to take a struct agp_bridge_data in every
> method, else all these changes don't make sense at all.
I don't agree that you *must* pass the agp_bridge_data pointer for every method.
You don't need it for bind/unbind/free if you associate each memory region
allocated using agp_allocate_memory(bridge,...) with a particular bridge
which is what the patch does. That is, agp_memory knows which bridge it belongs to.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at