On Mon, Oct 28, 2024 at 09:43:45AM -0700, Raghavendra Rao Ananta wrote: > On Sat, Oct 26, 2024 at 7:53 AM Oliver Upton <oliver.upton@xxxxxxxxx> wrote: > > On Sat, Oct 26, 2024 at 08:43:21AM +0100, Marc Zyngier wrote: > > > I think this would fix the problem you're seeing without changing the > > > userspace view of an erroneous configuration. It would also pave the > > > way for the complete removal of the interrupt notification to > > > userspace, which I claim has no user and is just a shit idea. > > > > Yeah, looks like this ought to get it done. > > > > Even with a fix for this particular issue I do wonder if we should > > categorically harden against late initialization failures and un-init > > the vCPU (or bug VM, where necessary) to avoid dealing with half-baked > > vCPUs/VMs across our UAPI surfaces. > > > > A sane userspace will probably crash when KVM_RUN returns EINVAL anyway. > > Thanks for the suggestion. Sure, I'll take another look at the > possible things that we can uninitialize and try to re-spin the patch. > > Marc, > > If you feel userspace_irqchip_in_use is not necessary anymore, and as > a quick fix to this issue, we can get rid of that independent of the > un-init effort. It's a good cleanup to begin with, even better that it fixes a genuine bug. Raghu, could you please test Marc's diff and send it as a patch (w/ correct attribution) if it works? I'm willing to bet that we have more init/uninit bugs lurking, so we can still follow up w/ robustness improvements once we're happy w/ the shape of them. -- Thanks, Oliver
