| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Fri, 30 Apr 2021 20:45:11 +0000
From: Sean Christopherson <seanjc@...gle.com>
To: Jon Kohler <jon@...anix.com>
Cc: Bijan Mottahedeh <bijan.mottahedeh@...anix.com>,
Raphael Norwitz <raphael.norwitz@...anix.com>,
Junaid Shahid <junaids@...gle.com>,
Paolo Bonzini <pbonzini@...hat.com>,
Vitaly Kuznetsov <vkuznets@...hat.com>,
Wanpeng Li <wanpengli@...cent.com>,
Jim Mattson <jmattson@...gle.com>,
Joerg Roedel <joro@...tes.org>,
Thomas Gleixner <tglx@...utronix.de>,
Ingo Molnar <mingo@...hat.com>, Borislav Petkov <bp@...en8.de>,
x86@...nel.org, "H. Peter Anvin" <hpa@...or.com>,
kvm@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] kvm: x86: move srcu lock out of kvm_vcpu_check_block
On Wed, Apr 28, 2021, Jon Kohler wrote:
> To improve performance, this moves kvm->srcu lock logic from
> kvm_vcpu_check_block to kvm_vcpu_running and wraps directly around
> check_events. Also adds a hint for callers to tell
> kvm_vcpu_running whether or not to acquire srcu, which is useful in
> situations where the lock may already be held. With this in place, we
> see roughly 5% improvement in an internal benchmark [3] and no more
> impact from this lock on non-nested workloads.
...
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index efc7a82ab140..354f690cc982 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -9273,10 +9273,24 @@ static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
> return 1;
> }
>
> -static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
> +static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu, bool acquire_srcu)
> {
> - if (is_guest_mode(vcpu))
> - kvm_x86_ops.nested_ops->check_events(vcpu);
> + if (is_guest_mode(vcpu)) {
> + if (acquire_srcu) {
> + /*
> + * We need to lock because check_events could call
> + * nested_vmx_vmexit() which might need to resolve a
> + * valid memslot. We will have this lock only when
> + * called from vcpu_run but not when called from
> + * kvm_vcpu_check_block > kvm_arch_vcpu_runnable.
> + */
> + int idx = srcu_read_lock(&vcpu->kvm->srcu);
> + kvm_x86_ops.nested_ops->check_events(vcpu);
> + srcu_read_unlock(&vcpu->kvm->srcu, idx);
> + } else {
> + kvm_x86_ops.nested_ops->check_events(vcpu);
> + }
> + }
Obviously not your fault, but I absolutely detest calling check_events() from
kvm_vcpu_running. I would much prefer to make baby steps toward cleaning up the
existing mess instead of piling more weirdness on top.
Ideally, APICv support would be fixed to not require a deep probe into nested
events just to see if a vCPU can run. But, that's probably more than we want to
bite off at this time.
What if we add another nested_ops API to check if the vCPU has an event, but not
actually process the event? I think that would allow eliminating the SRCU lock,
and would get rid of the most egregious behavior of triggering a nested VM-Exit
in a seemingly innocuous helper.
If this works, we could even explore moving the call to nested_ops->has_events()
out of kvm_vcpu_running() and into kvm_vcpu_has_events(); I can't tell if the
side effects in vcpu_block() would get messed up with that change :-/
Incomplete patch...
diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
index 00339d624c92..15f514891326 100644
--- a/arch/x86/kvm/vmx/nested.c
+++ b/arch/x86/kvm/vmx/nested.c
@@ -3771,15 +3771,17 @@ static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
to_vmx(vcpu)->nested.preemption_timer_expired;
}
-static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
+static int __vmx_check_nested_events(struct kvm_vcpu *vcpu, bool only_check)
{
struct vcpu_vmx *vmx = to_vmx(vcpu);
unsigned long exit_qual;
- bool block_nested_events =
- vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
bool mtf_pending = vmx->nested.mtf_pending;
struct kvm_lapic *apic = vcpu->arch.apic;
+ bool block_nested_events = only_check ||
+ vmx->nested.nested_run_pending ||
+ kvm_event_needs_reinjection(vcpu);
+
/*
* Clear the MTF state. If a higher priority VM-exit is delivered first,
* this state is discarded.
@@ -3837,7 +3839,7 @@ static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
}
if (vcpu->arch.exception.pending) {
- if (vmx->nested.nested_run_pending)
+ if (vmx->nested.nested_run_pending || only_check)
return -EBUSY;
if (!nested_vmx_check_exception(vcpu, &exit_qual))
goto no_vmexit;
@@ -3886,10 +3888,23 @@ static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
}
no_vmexit:
- vmx_complete_nested_posted_interrupt(vcpu);
+ if (!check_only)
+ vmx_complete_nested_posted_interrupt(vcpu);
+ else if (vmx->nested.pi_desc && vmx->nested.pi_pending)
+ return -EBUSY;
return 0;
}
+static bool vmx_has_nested_event(struct kvm_vcpu *vcpu)
+{
+ return !!__vmx_check_nested_events(vcpu, true);
+}
+
+static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
+{
+ return __vmx_check_nested_events(vcpu, false);
+}
+
static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
{
ktime_t remaining =
@@ -6627,6 +6642,7 @@ __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
}
struct kvm_x86_nested_ops vmx_nested_ops = {
+ .has_event = vmx_has_nested_event,
.check_events = vmx_check_nested_events,
.hv_timer_pending = nested_vmx_preemption_timer_pending,
.triple_fault = nested_vmx_triple_fault,
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index a829f1ab60c3..5df01012cb1f 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -9310,6 +9310,10 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
update_cr8_intercept(vcpu);
kvm_lapic_sync_to_vapic(vcpu);
}
+ } else if (is_guest_mode(vcpu)) {
+ r = kvm_check_nested_events(vcpu);
+ if (r < 0)
+ req_immediate_exit = true;
}
r = kvm_mmu_reload(vcpu);
@@ -9516,8 +9520,10 @@ static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
{
- if (is_guest_mode(vcpu))
- kvm_check_nested_events(vcpu);
+ if (is_guest_mode(vcpu) &&
+ (kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu) ||
+ kvm_x86_ops.nested_ops->has_event(vcpu)))
+ return true;
return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
!vcpu->arch.apf.halted);
Powered by blists - more mailing lists