Re: CVS commit: pkgsrc

[christos%zoulas.com@localhost (Christos Zoulas)]

On Dec 18,  3:50pm, M.Drochner%fz-juelich.de@localhost (Matthias Drochner) 
wrote:
-- Subject: Re: CVS commit: pkgsrc

| 
| christos%netbsd.org@localhost said:
| > - Make doscmd_loader use loadfile() [Tested only with ELF] 
| 
| Is this really necessary? I don't know details of "doscmd", but
| I'd guess this can be done easier and more portable if the
| binary to load is prepared by "objcopy --output-target=binary".

That is a great idea. Unfortunately I did not think about it :-(
I hated having to use loadfile, but I did not want to write another
ELF loader.  I'll try to get it working soon with objcopy...

| I was somehow hoping that the i386 "installboot" doesn't need
| "loadfile" anymore once the bootstrap gets another stage, so
| this "#ifdef _STANDALONE" mess could be nuked one day. It's
| just too much black magic for my taste.
| (Even now it could use "objcopy", but this wouldn't be nice
| for install floppies.)
| 
| I'd propose a slight interface change to loadfile anyway:
| the md code should get its hands at the file and the marks[]
| array before the final load is done. As things are now,
| the way load addresses are calculated doesn't work too well
| on i386, in particular if we thing of kernels with different
| KERNBASE and/or different load addresses -- that's too much
| for a single macro in loadfile_machdep.h. Besides this, the
| kernel identification should be md. (Our loader will happily
| try to boot a Linux kernel - unsuccessfully of course.)
 
Well, you have a perfectly good keyboard I suppose, so fix it.

christos