Konstantin Zhidkov <[EMAIL PROTECTED]> writes: > --- Ronald G Minnich <[EMAIL PROTECTED]> wrote: > > On Wed, 20 Feb 2002, Konstantin Zhidkov wrote: > > > > > 1. Is there a way to tell the kernel that A0000-FFFFF region is > > RAM? > > > > yes, is it really absolutely ram? This can be done. > > Almost(?) all modern chipsets can set up c0000-fffff region as RAM, > both writable and cacheable, reducing the black hole to 128K at A0000. > Boot ROM, when needs to be burn, usually can be found at the end of 4G > space. So, why to waste 192K, if OS does not need "any stinking BIOS"?
Currently the Linux kernel does not believe a pcbios that tells it there is ram there. I have played with this and usually I setup ram there, but I don't yet have a good kernel patch for this. In some instances it is very convinient to have the alias at 0xF0000 during initial setup, but LinuxBIOS currently can be setup to not use it. > AMD SC400/410 MCU's (and, I hope, many others) can also map A0000-BFFFF > to DRAM address space. This is not yet done, but I'm working... There are a few chips that handle that. > > > 2. How does LinuxBIOS reports RAM size to the kernel? > > Northsouthbridge > > > driver reports 4MB RAM. Then LinuxBIOS displays "totalram: 4M". > > Then > > > the kernel reports 16MB. Currently I fixed this by adding "mem=" > > > parameter to the kernel command line, but I want to detect amount > > of > > > memory dynamically (it may be extended to 8MB on future boards). > > > > > > Hmm. If linuxbios says 4M, then I think it reports 4M to the kernel. > > Is > > the kernel doing something strange? > > Seems to be my fault: I disabled PCI initialization in hardwaremain, > but did not revised write_tables. What needs to happen to write_tables? I suspect LinuxBIOS needs a few significant changes to support a 486. And I'd like to help ensure they go in cleanly. Eric
