On Wed, 2017-12-13 at 15:56 +0100, Bjoern Walk wrote:
Why do we cleanup here and abandon the rest of the information? Since the information in /proc/cpuinfo is kind of volatile in its format, shouldn't we be liberal in what we accept? If we can't parse it, we just report mhz = 0, but gathering the rest of the information in this function is still valuable.
Most functions in libvirt either perform all tasks succesfully or return a failure, so failing here is in line both with that and with the existing behavior.
So for example for S390 we have introduced CPU frequency information in /proc/cpuinfo only recently. That means that depending on your kernel version, you either read freq. info and the rest of the stuff or you discard the whole CPU. I found this highly unintuitive.
That's actually a very good reason! I ended up liking your approach to refactoring more than mine after all. So I made a couple of very small tweaks to it and I'm going to include it mostly as-is in my v2, coming in a minute. If you have a problem with any of my tweaks, just let me know. -- Andrea Bolognani / Red Hat / Virtualization