Jonathan Bennett wrote: > This patch seems like a really good idea for the ar71xx family. What > sorts of problems are preventing this from being implemented? As small > as it sounds, an extra 192k would be really useful on quite a few > routers.
I don't know of any actual problems with the patch. When I originally proposed the patch on this list, Dave Taht was concerned that it might not leave enough headroom to field-update the kernel with a larger one (https://lists.openwrt.org/pipermail/openwrt-devel/2011-August/012026.html), although my position is that this is what the configuration-preserving sysupgrade feature is for (https://lists.openwrt.org/pipermail/openwrt-devel/2011-August/012033.html). My patch is specific to WNDR3700 (and derivatives including WNDR3700v2 and WNDR3800), not the entire ar71xx family. I've been keeping the patch up-to-date (https://dev.openwrt.org/ticket/8781) and have been using it in all of my images for nearly a year. I know that others have been using this patch as well, and it's present in at least one community binary build. In current trunk images, the savings is only 128kB. In the time since I originally wrote this patch, the kernel's grown into another erase block. Someone else suggested detecting the beginning of the rootfs as is done on another target (https://dev.openwrt.org/browser/trunk/target/linux/lantiq/patches/201-owrt-mtd_uimage_split.patch), although I'm not very excited about scanning for a magic number in this manner. It's too fragile. It may be possible to determine the kernel size by other means, which would make it possible for the kernel to calculate the kernel-rootfs split at runtime more reliably. Unfortunately, detecting the kernel partition size in this manner would be board-specific, because different boards load the kernel in different ways. _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
