Hi Jean-Christophe,
Thanks for sharing your notes; we actually had a call the 26th of August on Barebox; I don't think the notes from the call were shared, so I'm sharing them below.
I think much more would have to be discussed, so certainly I suspect there will be more calls. Grant is probably the guy who will make that happen.
Cheers,
== Attendees ==
Arnd Bergmann Shaju Abraham Matt Waddel Paul McKenney Loïc Minier Robert Schwebel John Rigby Grant Likely Steve Sakoman Mounir Bsaibes Dave Martin Nicolas Pitre (mid-call)
== Minutes ==
* Quick mention of what Linaro is for Robert * Grant working on defining a boot architecture * Telling manufacturers what firmware should be doing, where it stops * Not trying to cover the whole world, but more the fairly common case * Predictable set of interfaces * Barebox updates * Sascha Hauer main maintainer * Biggest problem was u-boot's architecture; old, full of ifdefs, not well designed structure * Pengutronix making a lot of bootloaders for industrial systems * Trying to be a mini-kernel, with the same coding style, modules etc. BUT trying to keep it to a minimum * Why not evolve u-boot instead of forking? * Wolfgang pushed back when introducing Kconfig, modules etc. * Not much feedback on the new ideas * Upstream policy would prevent breakage of old boards; but priority was fixing the architecture rather than keeping support for hundreds of boards which they didn't have access to * Was a bad idea to use the new u-boot * Is Barebox trying to replace u-boot, or will the two exist together? * Pengutronix using it because they see it as useful * Presented at conferences, trying to gain more community around the project * Why not strip down the kernel down to be a bootloader? * Some SH4 people do this * Problem is that it takes longer to boot the kernel * Will likely remain too slow and too large to be a production bootloader, since drivers rely on too much functionality * If the kernel is the bootloader, no need for two sets of drivers * Need to communicate on upgrade risks in the field, more than how the bootloader is laid out * Trying to write common device drivers * Is there effort to be 100% compatible at the source level with the kernel drivers? * Problem is getting rid off the possibility of running user programs? * Current kexecbooting needs some way of loading the initial kernel from NAND or whatever * U-Boot and Barebox don't ship in production; why? * Pengutronix has industry and automotive customers, and they do ship Barebox * RedBoot has the same issue -- but there's the copyright assignment as well * Real devices don't really care about loading the kernel from different places, or even replacing the kernel * Size and hence price of flash is also a pressure * In terms of boot time, managed to boot an init in 326ms on imx35 w/Barebox; can also strip down 30 kB * Apparently, phone manufacturers are really concerned about speed of boot, and GPL (even v2) because they don't want to share some changes * Size impacts speed, so is an indirect constraint * e.g. allows running on SRAM instead of regular RAM on OMAP * Good idea to allow upgrading the device tree and the kernel, but not exposing the bootloader * Would be nice to unify into a common tree the first and second stage bootloader * Is it possible to build a single barebox which supports multiple boards? Is it relocatable? * Might be some limitations, Sascha would know