On 11 December 2012 18:34, Peter Maydell <peter.maydell@xxxxxxxxxx> wrote: > On 11 December 2012 18:27, Andreas Sandberg <andreas.sandberg@xxxxxxx> wrote: >> What makes matters more complicated is that the timer can't be >> hidden from the guest since ID_PFR1 is invariant.ther. > > Yes. Forgot to say, KVM should support trapping to hyp on the ID registers (via HCR.TID[0123]) and returning values suitable for the virtual CPU we're exposing to the guest. We don't do that at the moment but will need to (for instance for A15 guest on A7 host, a combo we can't pull off at the moment). There's some interesting as-yet-unopened cans of worms here regarding how exactly we get userspace to say "I'd like a CPU with/without arch.timers/neon/vfp/etc" and the kernel to DTRT. (eg, whose job is it to make the ID regs and the exposed features match, and do we drive one off the other or leave them independent). The user-space/kernel API has all the right bits to let this happen, we just haven't implemented it for the initial version. Speaking of cans of worms and the arch.timers, an interesting question we thought of the other week is "how does VM migration work between host machines whose arch.timers run at different frequencies?" -- PMM _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
