Martin Michlmayr wrote: > * John Reiser <[EMAIL PROTECTED]> [2008-08-23 13:16]: > >>Kernels as recent as vmlinuz-2.6.25-2-ixp4xx (1365048 bytes) still use >>gzip. Changing to upx/nrv ... would save an additional 10% in size ... >>Changing to lzma ... would save an additional 20% in size ...
> Is this supported in the mainline kernel? (The next question is > whether our kernel build system copes with it.) It works as a post-processing step that takes the final compressed vmlinux output of the mainline kernel build process, analyzes the file, finds the gzip compressed part, decompresses, re-compresses using upx/nrv or lzma, adds the new decompressor, and writes a new vmlinux file. This has been operational on i386 since late 2000, seven years ago. The keys to making it work are good documentation of the execution environment in which the boot-time decompressor must operate (including cache flushing, for instance) and a debugging harness which supports single-stepping. As recently as kernel 2.6.19, there have been proof-of-concept patches to replace the gzip portion of the mainline kernel build process. The patches were not pursued due to inertia. The additional information provided by i386 boot protocol 2.08 (compressed part is full ELF image) reduces the benefits of tighter integration, but not completely. -- -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
