Wrapping up things for the upcoming Yocto 1.1 release tends to keep us all heads down hammering away. Let me take a momemnt to share what I've beeing working on. We are fortunate to have a base kernel version alignment between Yocto and the PREEMPT_RT tree. Yocto 1.1 will include a refreshed real-time kernel and I have expanded real-time support to all the meta-intel BSPs as well as several of those from oe-core and meta-yocto.
Beyond the release, I have been trying to get my teeth into building truly small and fast images. While our current core-image-minimal image is by no means heavy-weight by current standards, those handful of megabytes receive heavy scrutiny from the folks in SoC land where DRAM cells are the shopping malls of die real-estate. Cutting the kernel down in size can be largely accomplished with a carefully crafted Linux kernel .config. Managing these for more than a few machines however, is tedious... at best. Fortunately, others are interested in cutting back on defconfig bloat in the kernel and I helped lead a Linux kernel config fragment management BoF session at the recent Linux Plumbers Conference. John Stultz has posted his merge_config.sh script for review on LKML following that session. Better fragment management will be an important tool for maintaining .config files across multiple BSPs.
As a sibling project to reduced image size, I'm looking at bringing down image boot time. Kernel size and features can have a significant impact on your boot time, especially on slower media, like MMC over SDIO which is so common in the embedded space. Of course, userspace plays a big part in boot time as well. Fortunately, my extended team is also looking into that. I'll have more details for everyone in October. I better, they already put me on the schedule at ELCE!