On Thu, Aug 25, 2016 at 12:50:59PM +0200, Ralph Sennhauser wrote: > Hi Daniel, > > On Thu, 25 Aug 2016 10:53:34 +0200 > Daniel Golle <dan...@makrotopia.org> wrote: > > [snip] > > > Are you planning to switch to initramfs and then load the ubi and > > ubifs modules in order to get a persistent rootfs?! > > Remember that otherwise, these modules are needed to be built-in to > > the kernel in order to mount the rootfs during boot. > > I use an initramfs to mount root indeed. This is just an alternative to > 7 patches in generic and mvebu to do the same. > > The kernel partition is 6MB, so there is plenty space and using busybox > works fine except there is no ubiblock applet. To work around this I > build ubi as a module instead of built-in and use module parameter to > create a blockdevice from the squashfs volume for use by the overlayfs.
Does the memory used for the initramfs get freed once you switched to the 'real' rootfs? If so, this would indeed be a quite good alternative to the current way and also provide other pre-init features such as failsafe-mode. I'll also be working on improving the patch-situation, afaik it's currently 3 patches which are not yet upstream to handle mounting ubifs as well as creating ubiblock and selecting it as root_dev for squashfs. I'm planning to implement parsing the root= cmdline parameter in the same way UBIFS does, so in case of non-ubifs rootfs, a ubiblock can be auto-created based on the volume specified in the cmdline. This will improve the situation eg. for dual-boot, as the volume name to-be-mounted will no longer need to be hard-coded into the kernel. Once this is done, the remaining patches can be replaced with those fit-for-upstream versions. Cheers Daniel > > Cheers > Ralph _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel