Re: [RFC][AGPGART]Allow multiple backends to be initialized

[Christoph Hellwig <hch@infradead.org>]

> -struct agp_bridge_data agp_bridge_dummy = { .type = NOT_SUPPORTED };
> -struct agp_bridge_data *agp_bridge = &agp_bridge_dummy;
> +agp_bridge_data_p agp_bridge;

Please don't intoroduce new new typedefs, especially not for pointer
types.

> +LIST_HEAD(agp_bridges);
>  EXPORT_SYMBOL(agp_bridge);
> -
> +EXPORT_SYMBOL(agp_bridges);

> -int agp_backend_acquire(void)
> +int agp_backend_acquire(struct pci_dev *pdev, agp_bridge_data_p 
> *acquired_bridge)
>  {
> -	if (agp_bridge->type == NOT_SUPPORTED)
> -		return -EINVAL;
> -	if (atomic_read(&agp_bridge->agp_in_use))
> +	agp_bridge_data_p bridge;
> +
> +	if (!pdev) {
> +		if (!agp_bridge)
> +			return -ENODEV;
> +		bridge = agp_bridge;
> +	} else {
> +        	list_for_each_entry(bridge, &agp_bridges, list) {
> +			int match=0;
> +			switch(pdev->class) {
> +				/* Standard bridges have a valid pci_dev */
> +				case PCI_CLASS_BRIDGE_HOST:
> +					if (bridge->dev==pdev)
> +						match=1;
> +					break;
> +				/* Non-standard bridges can use a devices pci_dev */
> +				default:
> +                			if (bridge->dev->bus==pdev->bus)
> +						match=1;
> +					break;
> +			}
> +                        if (match)
> +				break;
> +        	}
> +	}

Wrong interface.  Please pass in the pci_dev of the grpahics cards and
add a new method for lowlevel drivers to find the bridge.  For the normal
bridges (aka everything but the hp, alpha and your new driver) you'd do
a generic helper that just walks down the parent pointers until it finds
a class.

Also I'd suggest returning the found bridges as return value of the
function.

> -	printk(KERN_INFO PFX "AGP aperture is %dM @ 0x%lx\n",
> -	       size_value, bridge->gart_bus_addr);
> +	printk(KERN_INFO PFX "agp_bridge = 0x%lx: AGP aperture is %dM @ 0x%lx\n",
> +	       (unsigned long)bridge, size_value, bridge->gart_bus_addr);

pointers should be printed using %p, but this isn't a place where you should
print it at all.