On Tue, Nov 23, 1999 at 04:11:01PM +0100, Andrea Arcangeli wrote:
> 2) in the pci code the sysdata must be copied on
> all PCI devices and bridges before playing with
> the device. This also allowed me to simplify
> the alpha code that was taking the sysdata stored in
> an ugly static variable to handle the first pcibios opeartions
> (probing_hose). (on i386 this make no difference
> as the sysdata is NULL there)
If you havn't already, you'll soon get angry mail from DaveM
about this, as this breaks Sparc. I of course tried for that
very change when the code was first written.
> +/* The IO address space is larger than 0xffff */
> +#define TSUNAMI_IO_SPACE (TSUNAMI_CONF(0) - TSUNAMI_IO(0))
> +#define TSUNAMI_MEM_SPACE (_TSUNAMI_IACK_SC(0) - TSUNAMI_MEM(0))
Why this change? The IO address space may _be_ larger than 64k, but
we prefer not to use more than 64k. The fear being that there are
almost certainly broken PCI cards out there that only implement 16 bits.
Given the number that don't implement all 32 bits for dma or other such
memory space widgetry, this is not an unfounded concern.
r~
