On 04/05/07 02:09 +0200, Peter Stuge wrote:
On Thu, May 03, 2007 at 09:39:27AM -0600, Jordan Crouse wrote:
Would it make sense to move parts or all of the VSM code into the operating system?
Thats what OFW does for OLPC - and its fraught with danger. Case in point? A few days ago, the amount of video memory was doubled in OFW, to 16MB. But since the PCI spoofing happens in the kernel, it didn't get the message that the video memory changed, and the PCI BAR was still sized to 8MB, and X goes boom.
Oh, no, that's not what I meant.
I meant that the code in the kernel would be completely native to the hardware. No spoofing.
So what would handle the several dozen MSRs needed to set up routing - some of which are for essential devices that don't have drivers? And we would have to invent a new bus just for Geode devices (some of which share code with other entities - amd74xxx IDE for example), because without PCI, you wouldn't be able to match vendor/device IDs, or control the devices, or assign I/O or memory resources in virtual space, at least not easily.
So basically it would take the Geode from a pretty well supported platform to one that would be completely custom and on the fringe, and certainly not suitable for the mainstream kernel. I think the SMM handler is a small price to pay. :)
Jordan