On Wed, 2017-11-15 at 12:33 +0800, Wanpeng Li wrote: > 2017-11-15 11:03 GMT+08:00 Rik van Riel <riel@xxxxxxxxxx>: > > On Wed, 2017-11-15 at 08:47 +0800, Wanpeng Li wrote: > > > 2017-11-15 5:54 GMT+08:00 <riel@xxxxxxxxxx>: > > > > 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. > > > > > > > > This could be avoided by moving from a model where the guest > > > > FPU is > > > > loaded and stored with preemption disabled, to a model where > > > > the > > > > qemu userspace FPU is swapped out for the guest FPU context for > > > > the duration of the KVM_RUN ioctl. > > > > > > What will happen if CONFIG_PREEMPT is enabled? > > > > The scheduler will save the guest FPU context when a > > VCPU thread is preempted, and restore it when it is > > scheduled back in. > > I mean all the involved processes will use fpu. Before patch if > kernel > preempt occur: > > context_switch > -> prepare_task_switch > -> fire_sched_out_preempt_notifiers > -> kvm_sched_out > -> kvm_arch_vcpu_put > -> kvm_put_guest_fpu > -> copy_fpregs_to_fpstate(&vcpu- > >arch.guest_fpu) > save xsave area to guest fpu > buffer > -> __kernel_fpu_end > -> > copy_kernel_to_fpregs(¤t->thread.fpu.state) > restore prev vCPU qemu > userspace FPU to the xsave area > -> switch_to > -> __switch_to > -> switch_fpu_prepare > -> copy_fpregs_to_fpstate => save xsave area to > prev > vCPU qemu userspace FPU > -> switch_fpu_finish > -> copy_kernel_to_fpgregs => restore next task FPU > to xsave area > > > After the patch: > > context_switch > -> prepare_task_switch > -> fire_sched_out_preempt_notifiers > -> kvm_sched_out > > -> switch_to > -> __switch_to > -> switch_fpu_prepare > -> copy_fpregs_to_fpstate => Oops > save xsave area to prev vCPU qemu userspace FPU, > actually the guest FPU buffer is loaded in xsave area, you transmit > guest FPU in xsave area into the prev vCPU qemu userspace FPU When entering kvm_arch_vcpu_ioctl_run we save the qemu userspace FPU context in &vcpu->arch.user_fpu, and we restore that before leaving kvm_arch_vcpu_ioctl_run. Userspace should always see the userspace FPU context, no? Am I overlooking anything?
