V Út, 03. 08. 2004 v 16:36, Geert Uytterhoeven píše: > > 1) kernel and ramdisk must be in the same contiguous memory (IIUIC) > > Perhaps this restriction can be relaxed?
that's the only sensible chance we have. But for that we need someone with pretty good kernel internals knowledge willing to come up with ideas and actual patches. And I can compile it and test it on the Afterburner, Didier on CT60 and I am sure I'd find someone with a TT030. BTW, the reason why bootstrap generally stays away from FastRAM is that the PMMU in TOS is set differently than in Linux so whatever you load to FastRAM in TOS disappears as soon as Linux inits the PMMU for itself. If the kernel is in FastRAM it gets killed by this immediately, that's clear. If it's ramdisk then it's simply inaccessible then. Maybe if we loaded it at one place in FastRAM but passed different values to the kernel that would match the linux PMMU setup it could work? Example: load it at 0x1400000 but since the FastRAM begins physically at 0x1000000 but is later mapped to 0x400000 in linux the bootinfo would pass the ramdisk start at 0x0x800000?? That could do it with a lot of luck perhaps. In order to further play with this idea I would have to know exactly how the kernel memory is organized. I know the physical layout and in TOS it's 1:1 mapping (on some machines; other machines do have different maping). Also Didier said the kernel expects to have the ramdisk right next to itself. Is that really a requirement or can it be placed in the RAM freely? Who knows. Petr
