diff mbox series

[02/11] KVM: VMX: Handle APIC-write offset wrangling in VMX code

Message ID 20220204214205.3306634-3-seanjc@google.com (mailing list archive)
State New, archived
Headers show
Series KVM: x86: Prep work for VMX IPI virtualization | expand

Commit Message

Sean Christopherson Feb. 4, 2022, 9:41 p.m. UTC
Move the vAPIC offset adjustments done in the APIC-write trap path from
common x86 to VMX in anticipation of using the nodecode path for SVM's
AVIC.  The adjustment reflects hardware behavior, i.e. it's technically a
property of VMX, no common x86.  SVM's AVIC behavior is identical, so
it's a bit of a moot point, the goal is purely to make it easier to
understand why the adjustment is ok.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/lapic.c   |  3 ---
 arch/x86/kvm/vmx/vmx.c | 11 +++++++++--
 2 files changed, 9 insertions(+), 5 deletions(-)

Comments

Chao Gao Feb. 15, 2022, 2:22 a.m. UTC | #1
>--- a/arch/x86/kvm/vmx/vmx.c
>+++ b/arch/x86/kvm/vmx/vmx.c
>@@ -5302,9 +5302,16 @@ static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
> static int handle_apic_write(struct kvm_vcpu *vcpu)
> {
> 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
>-	u32 offset = exit_qualification & 0xfff;
> 
>-	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
>+	/*
>+	 * APIC-write VM-Exit is trap-like, KVM doesn't need to advance RIP and
>+	 * hardware has done any necessary aliasing, offset adjustments, etc...
>+	 * for the access.  I.e. the correct value has already been  written to
>+	 * the vAPIC page for the correct 16-byte chunk.  KVM needs only to
>+	 * retrieve the register value and emulate the access.
>+	 */
>+	u32 offset = exit_qualification & 0xff0;

Can we take this opportunity to remove offset/exit_qualification?
They are used just once.

>+
> 	kvm_apic_write_nodecode(vcpu, offset);
> 	return 1;
> }
>-- 
>2.35.0.263.gb82422642f-goog
>
Sean Christopherson Feb. 15, 2022, 4:30 p.m. UTC | #2
On Tue, Feb 15, 2022, Chao Gao wrote:
> >--- a/arch/x86/kvm/vmx/vmx.c
> >+++ b/arch/x86/kvm/vmx/vmx.c
> >@@ -5302,9 +5302,16 @@ static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
> > static int handle_apic_write(struct kvm_vcpu *vcpu)
> > {
> > 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
> >-	u32 offset = exit_qualification & 0xfff;
> > 
> >-	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
> >+	/*
> >+	 * APIC-write VM-Exit is trap-like, KVM doesn't need to advance RIP and
> >+	 * hardware has done any necessary aliasing, offset adjustments, etc...
> >+	 * for the access.  I.e. the correct value has already been  written to
> >+	 * the vAPIC page for the correct 16-byte chunk.  KVM needs only to
> >+	 * retrieve the register value and emulate the access.
> >+	 */
> >+	u32 offset = exit_qualification & 0xff0;
> 
> Can we take this opportunity to remove offset/exit_qualification?
> They are used just once.

Definitely should have dropped exit_qualification, not sure why I didn't.

I'd prefer to keep offset to document what is held in vmcs.EXIT_QUALIFICATION
without having to add an explicit comment.
diff mbox series

Patch

diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index 4662469240bc..fbce455a9d17 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -2188,9 +2188,6 @@  void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
 {
 	u32 val = 0;
 
-	/* hw has done the conditional check and inst decode */
-	offset &= 0xff0;
-
 	kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
 
 	/* TODO: optimize to just emulate side effect w/o one more write */
diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index b1165bb13a5a..1b135473677b 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -5302,9 +5302,16 @@  static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
 static int handle_apic_write(struct kvm_vcpu *vcpu)
 {
 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
-	u32 offset = exit_qualification & 0xfff;
 
-	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
+	/*
+	 * APIC-write VM-Exit is trap-like, KVM doesn't need to advance RIP and
+	 * hardware has done any necessary aliasing, offset adjustments, etc...
+	 * for the access.  I.e. the correct value has already been  written to
+	 * the vAPIC page for the correct 16-byte chunk.  KVM needs only to
+	 * retrieve the register value and emulate the access.
+	 */
+	u32 offset = exit_qualification & 0xff0;
+
 	kvm_apic_write_nodecode(vcpu, offset);
 	return 1;
 }