On Mon, Feb 17, 2025 at 05:07:28PM +0100, Borislav Petkov wrote: > On Sun, Feb 16, 2025 at 09:59:59PM -0800, Yosry Ahmed wrote: > > If 1-2% is the cost for keeping the MSR enabled at all times, I wonder > > if we should just do that for simplicitly, and have it its own > > mitigation option (chosen by the cmdline). > > Yes, that's what David and I think we should do initially. > > Then we can chase some more involved scheme like setting the bit before the > first VM runs and then clearing it when the last one exits. I *think* I've > seen something like that in KVM but I need to stare in detail. > > > - Upon return to userspace (similar to your previous proposal). In this > > case we run userspace with the MSR cleared, and only perform an IBPB > > in the exit to userspace pace. > > You want to IBPB before clearing the MSR as otherwise host kernel will be > running with the mistrained gunk from the guest. I meant IBPB + MSR clear before going to userspace, or IBPB + MSR clear before a context switch. > > > Any thoughts? > > Yes, let's keep it simple and do anything more involved *only* if it is really > necessary. > > -- > Regards/Gruss, > Boris. > > https://people.kernel.org/tglx/notes-about-netiquette
