* Eric W. Biederman <[EMAIL PROTECTED]> [020522 02:31]:
> Try it and report. My gut impression is that the only way to successfully
> trim the Linux kernel to a useable level is to start with a stub with the same
> directory structure as the linux kernel and gradually add features back in.
This is the approach that was made by milo a long time ago, but this is
a very problematic one. The size is still between 300 and 600k (with x86
emulation and a pretty big bunch of drivers, user interface,...)
This can definitely be shrinked down, but the way taking binary versions
of the drivers directly and linking them into the skeleton causes
trouble with every new kernel release. Lots of "functions" used by
drivers are resolved into macros. In the end you dont have to care for
the function used by the drivers, but also a lot of other configuration
issues that are not obvious. This leads to the need to fix for every new
kernel that is released. Another chance might be to use the skeleton and
compile the drivers directly dependent on this skeleton, so you don't
fiddle with binary object files from different trees/configs. Otoh, it
might be easier to port the drivers directly then, or rather try to get
them work on as well LinuxBIOS and Linux by choosing the API they use
with some CONFIG_ ifdefs. This way efforts about back-and-forth-porting
the drivers all the time might be reduced a lot.
Stefan
--
The x86 isn't all that complex - it just doesn't make a lot of
sense. -- Mike Johnson, Leader of 80x86 Design at AMD
Microprocessor Report (1994)
