Stack backtracing and live updates

vomlehn — Thu, 09 Apr 2015 02:02:17 +0000

It turns out I've spent a whole lot of time on backtracing, mostly on MIPS processors. And it is generally not even *possible* to do this in all cases. So, no, nothing that needs to be 100% reliable should be based on stack backtracing.

I've also spent a lot of time working on hot swapping kernels. Ten years is probably about right as an estimate, which is why I'm not working on such a thing now. I think the Containers folks have the only really viable approach. However, tou don't want to take a huge performance hit all at once so you need an approach where you run two kernels and gradually migrate userspace processes and device drivers from one to the other. It's quite a project but I think it is theoretically doable.

A rough patch for live patching

wenhua.yu — Mon, 09 Mar 2015 07:46:16 +0000

不错,就是不知道稳定的版本啥时候能出来

LWN: Comments on "A rough patch for live patching"

Stack backtracing and live updates

A rough patch for live patching