On Thu, May 16, 2019 at 11:06:19AM +0800, Wanpeng Li wrote:
> From: Wanpeng Li <wanpengli@xxxxxxxxxxx>
>
> wait_lapic_expire() call was moved above guest_enter_irqoff() because of
> its tracepoint, which violated the RCU extended quiescent state invoked
> by guest_enter_irqoff()[1][2]. This patch simply moves the tracepoint
> below guest_exit_irqoff() in vcpu_enter_guest(). Snapshot the delta before
> VM-Enter, but trace it after VM-Exit. This can help us to move
> wait_lapic_expire() just before vmentry in the later patch.
>
> [1] Commit 8b89fe1f6c43 ("kvm: x86: move tracepoints outside extended quiescent state")
> [2] https://patchwork.kernel.org/patch/7821111/
>
> Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx>
> Cc: Radim Krčmář <rkrcmar@xxxxxxxxxx>
> Cc: Sean Christopherson <sean.j.christopherson@xxxxxxxxx>
> Cc: Liran Alon <liran.alon@xxxxxxxxxx>
> Signed-off-by: Wanpeng Li <wanpengli@xxxxxxxxxxx>
> ---
> arch/x86/kvm/lapic.c | 16 ++++++++--------
> arch/x86/kvm/lapic.h | 1 +
> arch/x86/kvm/x86.c | 2 ++
> 3 files changed, 11 insertions(+), 8 deletions(-)
>
> diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
> index 2f364fe..af38ece 100644
> --- a/arch/x86/kvm/lapic.c
> +++ b/arch/x86/kvm/lapic.c
> @@ -1502,27 +1502,27 @@ static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
> }
>
> static inline void adaptive_tune_timer_advancement(struct kvm_vcpu *vcpu,
> - u64 guest_tsc, u64 tsc_deadline)
> + s64 advance_expire_delta)
> {
> struct kvm_lapic *apic = vcpu->arch.apic;
> u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
> u64 ns;
>
> /* too early */
> - if (guest_tsc < tsc_deadline) {
> - ns = (tsc_deadline - guest_tsc) * 1000000ULL;
> + if (advance_expire_delta < 0) {
> + ns = -advance_expire_delta * 1000000ULL;
> do_div(ns, vcpu->arch.virtual_tsc_khz);
> timer_advance_ns -= min((u32)ns,
> timer_advance_ns / LAPIC_TIMER_ADVANCE_ADJUST_STEP);
> } else {
> /* too late */
> - ns = (guest_tsc - tsc_deadline) * 1000000ULL;
> + ns = advance_expire_delta * 1000000ULL;
> do_div(ns, vcpu->arch.virtual_tsc_khz);
> timer_advance_ns += min((u32)ns,
> timer_advance_ns / LAPIC_TIMER_ADVANCE_ADJUST_STEP);
> }
>
> - if (abs(guest_tsc - tsc_deadline) < LAPIC_TIMER_ADVANCE_ADJUST_DONE)
> + if (abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_DONE)
> apic->lapic_timer.timer_advance_adjust_done = true;
> if (unlikely(timer_advance_ns > 5000)) {
> timer_advance_ns = 0;
> @@ -1545,13 +1545,13 @@ void wait_lapic_expire(struct kvm_vcpu *vcpu)
> tsc_deadline = apic->lapic_timer.expired_tscdeadline;
> apic->lapic_timer.expired_tscdeadline = 0;
> guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
> - trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
> + apic->lapic_timer.advance_expire_delta = guest_tsc - tsc_deadline;
>
> - if (guest_tsc < tsc_deadline)
> + if (apic->lapic_timer.advance_expire_delta < 0)
I'd prefer to keep "guest_tsc < tsc_deadline" here, just so that it's
obvious that the call to __wait_lapic_expire() is safe. My eyes did a
few double takes reading this code :-)
> __wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
>
> if (unlikely(!apic->lapic_timer.timer_advance_adjust_done))
> - adaptive_tune_timer_advancement(vcpu, guest_tsc, tsc_deadline);
> + adaptive_tune_timer_advancement(vcpu, apic->lapic_timer.advance_expire_delta);
> }
>
> static void start_sw_tscdeadline(struct kvm_lapic *apic)
> diff --git a/arch/x86/kvm/lapic.h b/arch/x86/kvm/lapic.h
> index d6d049b..3e72a25 100644
> --- a/arch/x86/kvm/lapic.h
> +++ b/arch/x86/kvm/lapic.h
> @@ -32,6 +32,7 @@ struct kvm_timer {
> u64 tscdeadline;
> u64 expired_tscdeadline;
> u32 timer_advance_ns;
> + s64 advance_expire_delta;
> atomic_t pending; /* accumulated triggered timers */
> bool hv_timer_in_use;
> bool timer_advance_adjust_done;
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index f2e3847..4a7b00c 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -7961,6 +7961,8 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
> ++vcpu->stat.exits;
>
> guest_exit_irqoff();
> + trace_kvm_wait_lapic_expire(vcpu->vcpu_id,
> + vcpu->arch.apic->lapic_timer.advance_expire_delta);
This needs to be guarded with lapic_in_kernel(vcpu). But, since this is
all in the same flow, a better approach would be to return the delta from
wait_lapic_expire(). That saves 8 bytes in struct kvm_timer and avoids
additional checks for tracing the delta.
E.g.:
s64 lapic_expire_delta;
...
if (lapic_in_kernel(vcpu) &&
vcpu->arch.apic->lapic_timer.timer_advance_ns)
lapic_expire_delta = wait_lapic_expire(vcpu);
else
lapic_expire_delta = 0;
...
trace_kvm_wait_lapic_expire(vcpu->vcpu_id, lapic_expire_delta);
>
> local_irq_enable();
> preempt_enable();
> --
> 2.7.4
>
