Ronald G Minnich <[EMAIL PROTECTED]> writes: > On 13 Mar 2002, Eric W. Biederman wrote: > > > Though the more I think about this the more I want to put this in ELFBios > > (which may sit in the LinuxBIOS tree) and not in the core. There > > are so many weird issues. Having the bootloader do the work may > > be the best thing. > > yes. I agree. I thought about this too. Jeff is right again.
Possibly. If we didn't have the size issues everything would be a different matter. And it is worth noting at this point I am in serious brain storm mode. And as such I can hold 4 or 5 simultaneously conflicting ideas until something is implemented that actually solves the problem. > I hate it when he does that :-) > > I think we still let linuxbios do some PCI setup so that some basic sanity > prevails. It doesn't cost much and it is useful for just about everything > I've seen. Then for systems like Plan 9 we have another intermediate > system that does the ugly stuff. At this point I really think evolving the LinuxBIOS tree so it can like oskit. I will probably work from 3 directions right now. Evolving the LinuxBIOS core so it can do more. Evolving Etherboot so it can do more. Evolving the linux kernel so it can take better advantage of etherboot. And moving the all in more or less the same direction. And if I ever figure out the code sharing issues with redboot I will probably play with that one to. The question is not one of what can be done. But rather what is maintainable and small enough in the long term. > Plan 9 will catch up eventually -- they'll have to address hot plug > someday, and once they do, I think our problems will be solved. Well there is still FreeDOS, NetBSD, FreeBSD, Mach, and .... The solution to the general problem is to assign a working value to all of the device resources (including irqs). It doesn't take any significant time so it should be a problem to do. Eric
