Burl Nyswonger burl@watchguard.com wrote:
- high-level (ie: C and Forth) -- It would be nice if there could be some way to abstract the chipset "advanced" features into the OpenFirmware device tree. If not, this is still where I think the "performance" related chipset initialization should happen.
Ugh, no. I spent some time moving the PCI optimisation from the kernel to userspace (see http://linux.powertweak.com). Moving this to BIOS space would not make much sense. The BIOS should contain enough code to get the system into a sane enough state to boot a kernel, and nothing more.
Another problem with PCI tweaking, is that some options are motherboard specific. You have no way of detecting if that option is going to work. When a motherboard manufacturer licenses a BIOS from a vendor, they get source with #defines to tell them how certain parts of the board work. How the data/address lines are decoded for example..
Asides from the mobo manufacturer, no-one is going to know how a memory bank has been wired on their board, so they won't know the order in which certain things have to be performed. The only way you can offer this to users is in a 'try and see' approach. Such a tactic does NOT belong in BIOS land.
regards,