What's the alternative now to the yocto-bsp tool? How would you explain to someone who did not see Yocto before what's the way to create a BSP e.g. with a multi_v7_defconfig? Create everything from scratch manually? I am not after a replacement, but after something which makes it easier (less error prone) for someone who is not into bitbake/oe/yocto, but has a custom kernel to add this kernel to the YP. As I said, a starting point would be upstream kernel + multi_v7_defconfig + patches (if needed).
We found the tool to be error prone, a big maintenance problem and giving users false expectations. There is the BSP developers guide, https://www.yoctoproject.org/docs/2.5/bsp-guide/bsp-guide.html which contains much of the information a user would need. If there is specific information missing we'd be happy to have an enhancement bug where we addressed these issues but we need specific issues to address and help with documentation, i.e. suggestions of the missing information we should include. If you're willing to step up and help create/maintain some kind of replacement for yocto-bsp we could discuss that too but right now we don't have enough momentum of people behind this.
Even if I throw the kernel tooling back into sumo it seems to works for what I want it. I would like to give people a template for a BSP as a starting point. Think about being a Yocto/OE newbie, but someone who knows Embedded Linux. I believe creating a BSP from scratch is non trivial. Maybe we don't even need a tool to create such templates. It's a bit messy if you look around the various BSPs floating around and some "sane" templates could help. Many chip- and board manufacturers have BSPs, but they usually include also some special software stack around them as well and this makes it even more messy. What do you think about this? Do you think we could gather somewhere BSP templates which work with poky in the complexity (and not more complex) of what's produced by the BSP tool? Does this make sense?
I'm trying to think of how these templates would work, but yes, there is probably a way to do such a thing without tooling. Perhaps in the style of the skeleton kernel module example ? or we also have some microservice templates in meta-overc that we use for containers, a variation of that technique might work here. I'm thinking of something that is a BSP framework, that wouldn't parse if used directly, but if you filled in the right variables, it would parse and inherit the existing BSPs (or create something new). Thoughts ? (In reply to comment #2) > Even if I throw the kernel tooling back into sumo it seems to works for what > I want it. > > I would like to give people a template for a BSP as a starting point. Think > about being a Yocto/OE newbie, but someone who knows Embedded Linux. I > believe creating a BSP from scratch is non trivial. > > Maybe we don't even need a tool to create such templates. > > It's a bit messy if you look around the various BSPs floating around and > some "sane" templates could help. > > Many chip- and board manufacturers have BSPs, but they usually include also > some special software stack around them as well and this makes it even more > messy. > > What do you think about this? > > Do you think we could gather somewhere BSP templates which work with poky in > the complexity (and not more complex) of what's produced by the BSP tool? > > Does this make sense?