Messages in this thread | | | Subject | Re: [PATCH 1/2] x86,kvm: move qemu/guest FPU switching out to vcpu_run | From | Quan Xu <> | Date | Thu, 16 Nov 2017 21:35:53 +0800 |
| |
On 2017-11-16 20:18, Paolo Bonzini wrote: > On 16/11/2017 13:12, Quan Xu wrote: >> However it seems there is still some gap.. >> >> as Rik said, "at context switch time, the context switch code will save >> the guest FPU state to current->thread.fpu when the VCPU thread is scheduled out." > By "guest FPU state" Rik means "guest FPU with host PKRU". :( actually it is host_pkru, just with different names..
> Guest PKRU is always stored in vcpu->arch.pkru rather than in the guest > FPU state, so guest PKRU will never be in current->thread.fpu.state either. > > KVM_GET_XSAVE will the guest FPU state with vcpu->arch.pkru and > migration will work properly. agreed, I fix it.. that's why I concern.. there are so much methods to write PKRU with host_pkru/guest_pkru..
after migration, KVM_GET|SET_XSAVE restore the right pkru.. but we introduce another method:
-- When the VCPU thread is scheduled back in, the context switch code will restore current->thread.fpu to the FPU registers.
there is still a window to restore current->thread.fpu to the FPU registers before enter guest mode and
preempt_disable().
on target machine, after migration, the pkru value is source_host_pkru in current->thread.fpu.
in case,
source_host_pkru_value != guest_pkru_value, target_host_pkru_value == guest_pkru_value..
source_host_pkru_value may be restored to PKRU.. make pkru status inconsistent..
thanks
Quan Alibaba Cloud > Thanks, > > Paolo > >> after preempt_enable() in vcpu_enter_guest(), the vcpu thread is >> scheduled out, >> in kvm_x86_ops->run, the PKRU has been restored with host_pkru (IF >> guest_pkru != host_pkru).. >> instead of guest_pkru.. >> >> then the PKRU is host_pkru, how to save guest_pkru current->thread.fpu? >> >> as mentioned, all this _assumes_ that the pkru is in >> current->thread.fpu.state as well. >> >> >> thanks, >> >> Quan >> Alibaba Cloud >> >>> Context switching will only happen in vcpu_enter_guest() after >>> preempt_enable() for a preemptible kernel, or in vcpu_run via >>> cond_resched() for a non-preemptible kernel. >>> >>> Thanks, >>> >>> Paolo >>> >>>> VM migration again, in case, >>>> source_host_pkru_value != guest_pkru_value, >>>> target_host_pkru_value == guest_pkru_value.. >>>> >>>> the pkru status would be inconsistent.. >
| |