[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [Patch] x86/HVM: Fix RTC interrupt modelling

On 02/10/2014 04:34 PM, Jan Beulich wrote:
   * The results from XenRT suggest that the new emulation is better than the
"Better" in the sense of the limited set of uses of the virtual hardware
by whatever selection of guest OSes is being run there. But very
likely not "better" in the sense on matching up with how the respective
hardware specification would require it to behave.

The context of the above sentence was in a justification for including it in 4.4. Obviously "occasionally gets stuck during boot" is a pretty bad bug that we'd like to see fix. But given the tricky nature of this whole area, there's a risk that this will cause regressions in *other* situations or operating systems. What I understand Andy to be saying is that with the patch, the RTC appears to cause less problems than without it.

What your analysis is missing, Andy, is what the effects might be if there were a bug. Obviously other guests might hang during boot; but what else? Might they hang at some point much later, perhaps when being pounded with interrupts due to heavy network traffic? Might the clock begin to drift or jump around? Would the XenRT testing catch that if it happened? And, would those potential bugs be worse than what we have now?

There's a reason for trying to go through the whole exercise, particularly in bugs like this. I do have a lot of faith in our intuition to consider hundreds of individual factors and come up with a reasonably good judgement of the probabilities -- but only if it is guided properly. We are all very prone to only consider the things we happen to be thinking about, and to completely ignore all the things we don't happen to be looking at. My own temptation, looking at this bug, is to say, "Random hangs during boot, yeah, that's pretty bad; we should take it." But then I'm only looking at the positives of the patch: I'm not really making a balance of the positives versus the negatives.

The goal of going through the "worst-case-scenario" exercise is to bring to our minds the potential outcomes we are prone to ignore. Only then can we reasonably trust our intuition to make a properly informed judgement.


Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.