[OLPC-devel] Re: suggested requirement for new kernels and initrds: must boot under qemu and/or kexec as kernel and initrd before release as a stream

[Ronald G Minnich]

Jordan Crouse wrote:

> So my personal opinion is that that all of this overrides the annoyance of two
> binary files - at least for now.

initramfs is not working very well anyway. Aside from the bugs in the 
build system, when I'm done I get this:
-rw-rw-r--  1 rminnich rminnich 466368 Jul  4 16:38 initramfs_data.cpio.gz

whereas the same identical rootfs, as an initrd, is:
-rw-rw-r--  1 rminnich rminnich 232879 Jul  4 22:36 
../../buildrom/deploy/initrd


pretty bad.

Also, you can't really kexec an initramfs-based kernel. This one is 
really weird, but when I kexec this kernel, the malloc pool for some 
reason starts at top of memory, and very early on, a kmalloc'ed item is 
at c7800000. This only happens with kexec'ing an initramfs-based kernel 
-- it boots under qemu just fine.

So:
1) bugs in build system
2) initramfs file is way too big (turns out it's another bug in the 
build system -- it does everything twice)
3) won't work under kexec

three strikes, you're out, as the saying goes.

so, back to trying to see why the initrd fails so badly under qemu/kexec.

thanks

ron