diff options
Diffstat (limited to 'package/linux/kernel-source/arch/mips/brcm-boards/bcm947xx/compressed/README.mjn3')
| -rw-r--r-- | package/linux/kernel-source/arch/mips/brcm-boards/bcm947xx/compressed/README.mjn3 | 20 |
1 files changed, 20 insertions, 0 deletions
diff --git a/package/linux/kernel-source/arch/mips/brcm-boards/bcm947xx/compressed/README.mjn3 b/package/linux/kernel-source/arch/mips/brcm-boards/bcm947xx/compressed/README.mjn3 new file mode 100644 index 0000000000..0e93e55f90 --- /dev/null +++ b/package/linux/kernel-source/arch/mips/brcm-boards/bcm947xx/compressed/README.mjn3 @@ -0,0 +1,20 @@ + +FIRST... See the dedication in the decompress_bunzip2.c file as it applies +here too. Donations to hospice in Toni's memory would be appreciated. + +As far as the code goes... the bzImage is just a bzip'd kernel image with +a small gzip'd decompressor/loader stuck on front. CFE ungzip's the loader +app which then relocates the bunzip decompressor into higher memory and +bunzip's the compressed kernel directly from flash. Then the instruction +cache is flushed (to remove traces of the loader) and the kernel is executed. + +Of course, a better approach would be to add bunzip decompression to CFE. + +Notes... + 1) Instruction cache size and linesize are hardcoded (see the #warning). + 2) Currently assumes at least 16M or ram. + 3) Thanks to Mike Baker mbm at alt dot org for bouncing ideas back + and forth as well as diagnosing the last (icache) bug. + +Manuel Novoa III <mjn3@codepoet.org> May 30, 2004 + |