On 24/05/2016 16:20, Alan Jenkins wrote: > On 24/05/16 13:51, Paolo Bonzini wrote: >> On 23/05/2016 16:30, Alan Jenkins wrote: >>> https://github.com/torvalds/linux/compare/master...sourcejedi:kvm_lockup_fix?expand=1 >>> >> Here the comparison (and the expired_advance_ns field) is only used to >> issue the warning. > No, because expired_advance_ns is used as the limit for the busy-wait. Right, but I expected the limit to go away together with the warning in the final patch, if the root cause can be fixed. >> Could you change the patch to use __delay instead of >> ndelay and > > No, because the limit is in ns, not host TSC ticks. Also, we were > passing _guest_ TSC ticks to __delay(), which looks like a second > potential cause of lockups because guest TSC ticks are scaled. This is correct. It's missing a call to kvm_scale_tsc if you keep the __delay. > The problem was I didn't want to make a user-triggerable WARN_ON. But if > we don't want WARN_ON at all, then maybe it's ok to compare the module > parameter racily, so we don't need the extra field. There's no existing > documentation which I could update, to say that reducing > lapic_timer_advance_ns is not supported, right? (Because even with > checks, it means the guest deadline timer can fire when the TSC is > showing a time prior to the deadline, so it could break guest assumptions). I'd just make lapic_timer_advance_ns read-only, I think. Thanks, Paolo -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
