Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Mon, 17 May 2021 17:51:16 +0000
From:   Sean Christopherson <seanjc@google.com>
To:     Wanpeng Li <kernellwp@gmail.com>
Cc:     linux-kernel@vger.kernel.org, kvm@vger.kernel.org,
        Paolo Bonzini <pbonzini@redhat.com>,
        Vitaly Kuznetsov <vkuznets@redhat.com>,
        Wanpeng Li <wanpengli@tencent.com>,
        Jim Mattson <jmattson@google.com>,
        Joerg Roedel <joro@8bytes.org>
Subject: Re: [PATCH v3 5/5] KVM: LAPIC: Narrow the timer latency between
 wait_lapic_expire and world switch
Message-ID: <YKKtFOl3oklFp1lW@google.com>
References: <1621260028-6467-1-git-send-email-wanpengli@tencent.com>
 <1621260028-6467-5-git-send-email-wanpengli@tencent.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1621260028-6467-5-git-send-email-wanpengli@tencent.com>
Precedence: bulk

On Mon, May 17, 2021, Wanpeng Li wrote:
> From: Wanpeng Li <wanpengli@tencent.com>
> 
> Let's treat lapic_timer_advance_ns automatically tune logic as hypervisor
> overhead, move it before wait_lapic_expire instead of between wait_lapic_expire 
> and the world switch, the wait duration should be calculated by the 
> up-to-date guest_tsc after the overhead of automatically tune logic. This 
> patch reduces ~30+ cycles for kvm-unit-tests/tscdeadline-latency when testing 
> busy waits.
> 
> Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
> ---
>  arch/x86/kvm/lapic.c | 7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
> index c0ebef560bd1..552d2acf89ab 100644
> --- a/arch/x86/kvm/lapic.c
> +++ b/arch/x86/kvm/lapic.c
> @@ -1598,11 +1598,12 @@ static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
>  	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
>  	apic->lapic_timer.advance_expire_delta = guest_tsc - tsc_deadline;
>  
> -	if (guest_tsc < tsc_deadline)
> -		__wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
> -
>  	if (lapic_timer_advance_dynamic)
>  		adjust_lapic_timer_advance(vcpu, apic->lapic_timer.advance_expire_delta);
> +
> +	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());

This is redundant and unnecessary if automatic tuning is disabled, or if the
timer did not arrive early.  A comment would also be helpful.  E.g. I think this
would micro-optimize all paths:

diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index c0ebef560bd1..5d91f2367c31 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -1598,11 +1598,19 @@ static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
        guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
        apic->lapic_timer.advance_expire_delta = guest_tsc - tsc_deadline;

+       if (lapic_timer_advance_dynamic) {
+               adjust_lapic_timer_advance(vcpu, apic->lapic_timer.advance_expire_delta);
+               /*
+                * If the timer fired early, reread the TSC to account for the
+                * overhead of the above adjustment to avoid waiting longer
+                * than is necessary.
+                */
+               if (guest_tsc < tsc_deadline)
+                       guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
+       }
+
        if (guest_tsc < tsc_deadline)
                __wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
-
-       if (lapic_timer_advance_dynamic)
-               adjust_lapic_timer_advance(vcpu, apic->lapic_timer.advance_expire_delta);
 }

 void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)

> +	if (guest_tsc < tsc_deadline)
> +		__wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
>  }
>  
>  void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
> -- 
> 2.25.1
>