Ram initialization and small c.

[GNUOrder]

Would it be easier to design this into gcc as a separate architecture or 
would that be too unmanagable?  It may or may not be of benefit to other 
embedded projects as well so there may be more help keeping it up to date if 
it was in the gcc tree.  Then again, someone could modify it for their own 
needs and make it unusable for what it was intended.

GO

On Thursday 13 February 2003 00:29, Eric W. Biederman wrote:
> Years ago when C was young and implementations were not readily
> available people compilers for subsets of C and called then small C
> compilers.
>
> Currently one of the greatest challenges in LinuxBIOS is to write
> ram initialization code.  On x86 this is all done in 8 general purpose
> registers.  And it takes about a 1000 lines of assembly per memory
> controller, to do a good job and handle all of the auto-configuration
> cases.  Debugging the assembly is bad as frequently adding the debug
> code breaks the code we are trying to debug.  Additionally there is no
> code reuse between the two binaries that make up LinuxBIOS as one is
> in C and the other in assembly.
>
> By this point I have enough practice I am good at writing the assembly
> code, that I think it hardly matters.  But I have seen several
> proficient coders just grind to a halt when confronted with miles of
> assembly.
>
> So what I propose is to write a small C compiler for LinuxBIOS.
> Various reports from ages past put it at about a man-month of effort
> to get the first version going.  And it may be better than that as
> there are a couple of places I can copy code from. tcc, lcc, gcc, and
> other small c projects.
>
> Up until this point all C compilers have had an assumption that you
> can have a stack.  And most simple C compilers have kept none of the
> data structures needed to do inline and similar optimizations that
> are necessary when you don't have ram.
>
> So I think it is easiest to build up something very simple from
> scratch.  That is the route I am going to try.  No promises, but
> I should start a trial implementation soon.
>
> Ron.  I am going to need to do a branch soon as I do the Hammer
> port and integrate a small C compiler.  And I want to make some
> clean fixes and remove some of the backwards compatibility cruft, and
> the linuxbios 1.0 code base is an inappropriate place to target.
> It should be possible to roll in most of the improvements into the 1.0
> codebase, but that will just increase the clutter of the codebase.
>
> Eric
>
> _______________________________________________
> Linuxbios mailing list
> Linuxbios at clustermatic.org
> http://www.clustermatic.org/mailman/listinfo/linuxbios