Re: Patches for REALLY TINY 386 kernels

["H. Peter Anvin" <hpa@zytor.com>]

Andi Kleen wrote:
>>
>> However, compressed size reductions as an abstract thing is useful for
>> this market.  Just not these particular ones.  The first thing to get
>> there is probably an LZMA-based compressor instead of gzip.
> 
> That would need more memory again.
> 

Actually, even with a 64K dictionary size (for which point the
decompression runtime memory requirements are comparable to gzip) LZMA
beats both gzip -9 and bzip2 -9 quite handily:

Reference (this is an i386 kernel):

-rwxrwxr-x 1 hpa hpa 5607558 Jul 18 15:04 vmlinux.bin*
-rw-rw-r-- 1 hpa hpa 2658275 Jul 18 15:04 vmlinux.bin.bz2
-rw-rw-r-- 1 hpa hpa 2760849 Jul 18 15:04 vmlinux.bin.gz

Pure LZMA with dictionary sizes from 2^16 to 2^24:

-rw-rw-r-- 1 hpa hpa 2380983 Jul 18 15:14 d16.7z
-rw-rw-r-- 1 hpa hpa 2317458 Jul 18 15:15 d18.7z
-rw-rw-r-- 1 hpa hpa 2284746 Jul 18 15:15 d20.7z
-rw-rw-r-- 1 hpa hpa 2264001 Jul 18 15:15 d22.7z
-rw-rw-r-- 1 hpa hpa 2263185 Jul 18 15:15 d24.7z

LZMA with BCJ precompression:

-rw-rw-r-- 1 hpa hpa 2236070 Jul 18 15:14 d16bcj.7z
-rw-rw-r-- 1 hpa hpa 2167873 Jul 18 15:15 d18bcj.7z
-rw-rw-r-- 1 hpa hpa 2134541 Jul 18 15:15 d20bcj.7z
-rw-rw-r-- 1 hpa hpa 2112987 Jul 18 15:15 d22bcj.7z
-rw-rw-r-- 1 hpa hpa 2111917 Jul 18 15:15 d24bcj.7z

> Better just write less bloated code. Perhaps mandatory bloatometer
> runs during -rc*s for kernels with minimal config with public code pig shame lists
> similar to the regression lists are useful. Anyone volunteering?
> 
> I suspect there is also much more low hanging fruit of this around.

Most likely.

> I don't think eliminating cpuid is a step forward though; that's
> just madness.

Agreed, especially given the invasiveness of the patch.

	-hpa