On Wed, 2017-11-15 at 14:53 +0800, quan.xu04@xxxxxxxxx wrote: > > On 2017/11/15 05:54, riel@xxxxxxxxxx wrote: > > From: Rik van Riel <riel@xxxxxxxxxx> > > > > Currently, every time a VCPU is scheduled out, the host kernel will > > first save the guest FPU/xstate context, then load the qemu > > userspace > > FPU context, only to then immediately save the qemu userspace FPU > > context back to memory. When scheduling in a VCPU, the same > > extraneous > > FPU loads and saves are done. > > Rik, be careful with VM migration. with you patch, I don't think you > could load fpu/xstate > context accurately after VM migration. Can you explain why you believe that? Getting the guest FPU or XSTATE is done under the vcpu->mutex. This patch switches out guest and userspace FPU/XSTATE under the vcpu->mutex, and switches it back before releasing the vcpu->mutex. By the time a KVM_GET_FPU has obtained the vcpu->mutex, the guest FPU state will be in vcpu->arch.guest_fpu.state, where you expect it to be. What am I missing?
