Re: [PATCH] KVM: VMX: Stop context switching MSR_IA32_UMWAIT_CONTROL

[Paolo Bonzini <pbonzini@redhat.com>]

On 23/06/20 02:51, Sean Christopherson wrote:
> Remove support for context switching between the guest's and host's
> desired UMWAIT_CONTROL.  Propagating the guest's value to hardware isn't
> required for correct functionality, e.g. KVM intercepts reads and writes
> to the MSR, and the latency effects of the settings controlled by the
> MSR are not architecturally visible.
> 
> As a general rule, KVM should not allow the guest to control power
> management settings unless explicitly enabled by userspace, e.g. see
> KVM_CAP_X86_DISABLE_EXITS.  E.g. Intel's SDM explicitly states that C0.2
> can improve the performance of SMT siblings.  A devious guest could
> disable C0.2 so as to improve the performance of their workloads at the
> detriment to workloads running in the host or on other VMs.
> 
> Wholesale removal of UMWAIT_CONTROL context switching also fixes a race
> condition where updates from the host may cause KVM to enter the guest
> with the incorrect value.  Because updates are are propagated to all
> CPUs via IPI (SMP function callback), the value in hardware may be
> stale with respect to the cached value and KVM could enter the guest
> with the wrong value in hardware.  As above, the guest can't observe the
> bad value, but it's a weird and confusing wart in the implementation.
> 
> Removal also fixes the unnecessary usage of VMX's atomic load/store MSR
> lists.  Using the lists is only necessary for MSRs that are required for
> correct functionality immediately upon VM-Enter/VM-Exit, e.g. EFER on
> old hardware, or for MSRs that need to-the-uop precision, e.g. perf
> related MSRs.  For UMWAIT_CONTROL, the effects are only visible in the
> kernel via TPAUSE/delay(), and KVM doesn't do any form of delay in
> vcpu_vmx_run().  Using the atomic lists is undesirable as they are more
> expensive than direct RDMSR/WRMSR.
> 
> Furthermore, even if giving the guest control of the MSR is legitimate,
> e.g. in pass-through scenarios, it's not clear that the benefits would
> outweigh the overhead.  E.g. saving and restoring an MSR across a VMX
> roundtrip costs ~250 cycles, and if the guest diverged from the host
> that cost would be paid on every run of the guest.  In other words, if
> there is a legitimate use case then it should be enabled by a new
> per-VM capability.
> 
> Note, KVM still needs to emulate MSR_IA32_UMWAIT_CONTROL so that it can
> correctly expose other WAITPKG features to the guest, e.g. TPAUSE,
> UMWAIT and UMONITOR.
> 
> Fixes: 6e3ba4abcea56 ("KVM: vmx: Emulate MSR IA32_UMWAIT_CONTROL")
> Cc: stable@vger.kernel.org
> Cc: Jingqi Liu <jingqi.liu@intel.com>
> Cc: Tao Xu <tao3.xu@intel.com>
> Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
> ---
>  arch/x86/include/asm/mwait.h |  2 --
>  arch/x86/kernel/cpu/umwait.c |  6 ------
>  arch/x86/kvm/vmx/vmx.c       | 18 ------------------
>  3 files changed, 26 deletions(-)
> 
> diff --git a/arch/x86/include/asm/mwait.h b/arch/x86/include/asm/mwait.h
> index 73d997aa2966..e039a933aca3 100644
> --- a/arch/x86/include/asm/mwait.h
> +++ b/arch/x86/include/asm/mwait.h
> @@ -25,8 +25,6 @@
>  #define TPAUSE_C01_STATE		1
>  #define TPAUSE_C02_STATE		0
>  
> -u32 get_umwait_control_msr(void);
> -
>  static inline void __monitor(const void *eax, unsigned long ecx,
>  			     unsigned long edx)
>  {
> diff --git a/arch/x86/kernel/cpu/umwait.c b/arch/x86/kernel/cpu/umwait.c
> index 300e3fd5ade3..ec8064c0ae03 100644
> --- a/arch/x86/kernel/cpu/umwait.c
> +++ b/arch/x86/kernel/cpu/umwait.c
> @@ -18,12 +18,6 @@
>   */
>  static u32 umwait_control_cached = UMWAIT_CTRL_VAL(100000, UMWAIT_C02_ENABLE);
>  
> -u32 get_umwait_control_msr(void)
> -{
> -	return umwait_control_cached;
> -}
> -EXPORT_SYMBOL_GPL(get_umwait_control_msr);
> -
>  /*
>   * Cache the original IA32_UMWAIT_CONTROL MSR value which is configured by
>   * hardware or BIOS before kernel boot.
> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> index 08e26a9518c2..b2447c1ee362 100644
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -6606,23 +6606,6 @@ static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
>  					msrs[i].host, false);
>  }
>  
> -static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
> -{
> -	u32 host_umwait_control;
> -
> -	if (!vmx_has_waitpkg(vmx))
> -		return;
> -
> -	host_umwait_control = get_umwait_control_msr();
> -
> -	if (vmx->msr_ia32_umwait_control != host_umwait_control)
> -		add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
> -			vmx->msr_ia32_umwait_control,
> -			host_umwait_control, false);
> -	else
> -		clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
> -}
> -
>  static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
>  {
>  	struct vcpu_vmx *vmx = to_vmx(vcpu);
> @@ -6730,7 +6713,6 @@ static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu)
>  
>  	if (vcpu_to_pmu(vcpu)->version)
>  		atomic_switch_perf_msrs(vmx);
> -	atomic_switch_umwait_control_msr(vmx);
>  
>  	if (enable_preemption_timer)
>  		vmx_update_hv_timer(vcpu);
> 

Queued, thanks.

Paolo