| Message ID | 1462388992-25242-9-git-send-email-Suravee.Suthikulpanit@amd.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, May 04, 2016 at 02:09:47PM -0500, Suravee Suthikulpanit wrote: > This patch introduces a new mechanism to inject interrupt using AVIC. > Since VINTR is not supported when enable AVIC, we need to inject "... is not supported when AVIC is enabled ..." VINTR? Please write those things out in the commit message for maximum information transfer to the reader. :) > interrupt via APIC backing page instead. > > This patch also adds support for AVIC doorbell, which is used by > KVM to signal a running vcpu to check IRR for injected interrupts. > > Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com> > --- > arch/x86/kvm/svm.c | 39 +++++++++++++++++++++++++++++++++++---- > 1 file changed, 35 insertions(+), 4 deletions(-) > > diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c > index 4b00dc6..3a97874 100644 > --- a/arch/x86/kvm/svm.c > +++ b/arch/x86/kvm/svm.c > @@ -71,6 +71,8 @@ MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id); > #define SVM_FEATURE_DECODE_ASSIST (1 << 7) > #define SVM_FEATURE_PAUSE_FILTER (1 << 10) > > +#define SVM_AVIC_DOORBELL 0xc001011b MSR_SVM_AVIC_DOORBELL > + > #define NESTED_EXIT_HOST 0 /* Exit handled on host level */ > #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */ > #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
On 10/05/2016 11:19, Borislav Petkov wrote: >> > This patch introduces a new mechanism to inject interrupt using AVIC. >> > Since VINTR is not supported when enable AVIC, we need to inject > "... is not supported when AVIC is enabled ..." > > VINTR? The ability to request a vmexit as soon as an interrupt can be injected (IF=GIF=1, no interrupt window, etc.). It's called the "VINTR intercept". > Please write those things out in the commit message for maximum > information transfer to the reader. :) More important, where does the APM document that VINTR is not supported when AVIC is enabled? It is certainly pointless and inefficient, but I'm not sure where it says that it doesn't work. Paolo -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi, Sorry for late response on this. On 5/10/16 09:50, Paolo Bonzini wrote: > > > On 10/05/2016 11:19, Borislav Petkov wrote: >>>> This patch introduces a new mechanism to inject interrupt using AVIC. >>>> Since VINTR is not supported when enable AVIC, we need to inject >> "... is not supported when AVIC is enabled ..." >> >> VINTR? > > The ability to request a vmexit as soon as an interrupt can be injected > (IF=GIF=1, no interrupt window, etc.). It's called the "VINTR intercept". > >> Please write those things out in the commit message for maximum >> information transfer to the reader. :) > > More important, where does the APM document that VINTR is not supported > when AVIC is enabled? It is certainly pointless and inefficient, but > I'm not sure where it says that it doesn't work. > > Paolo > Basically, from the APM vol2 here: http://developer.amd.com/wordpress/media/2012/10/24593_APM_v21.pdf On page 115, section AVIC Enable—Virtual Interrupt Control, Bit 31: "... Enabling AVIC implicitly disables the V_IRQ, V_INTR_PRIO, V_IGN_TPR, and V_INTR_VECTOR fields in the VMCB Control Word." Thanks, Suravee -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c index 4b00dc6..3a97874 100644 --- a/arch/x86/kvm/svm.c +++ b/arch/x86/kvm/svm.c @@ -71,6 +71,8 @@ MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id); #define SVM_FEATURE_DECODE_ASSIST (1 << 7) #define SVM_FEATURE_PAUSE_FILTER (1 << 10) +#define SVM_AVIC_DOORBELL 0xc001011b + #define NESTED_EXIT_HOST 0 /* Exit handled on host level */ #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */ #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */ @@ -293,6 +295,17 @@ static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data) mark_dirty(svm->vmcb, VMCB_AVIC); } +static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu) +{ + struct vcpu_svm *svm = to_svm(vcpu); + u64 *entry = svm->avic_physical_id_cache; + + if (!entry) + return false; + + return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK); +} + static void recalc_intercepts(struct vcpu_svm *svm) { struct vmcb_control_area *c, *h; @@ -2865,10 +2878,11 @@ static int clgi_interception(struct vcpu_svm *svm) disable_gif(svm); /* After a CLGI no interrupts should come */ - svm_clear_vintr(svm); - svm->vmcb->control.int_ctl &= ~V_IRQ_MASK; - - mark_dirty(svm->vmcb, VMCB_INTR); + if (!kvm_vcpu_apicv_active(&svm->vcpu)) { + svm_clear_vintr(svm); + svm->vmcb->control.int_ctl &= ~V_IRQ_MASK; + mark_dirty(svm->vmcb, VMCB_INTR); + } return 1; } @@ -3762,6 +3776,7 @@ static inline void svm_inject_irq(struct vcpu_svm *svm, int irq) { struct vmcb_control_area *control; + /* The following fields are ignored when AVIC is enabled */ control = &svm->vmcb->control; control->int_vector = irq; control->int_ctl &= ~V_INTR_PRIO_MASK; @@ -3840,6 +3855,18 @@ static void svm_sync_pir_to_irr(struct kvm_vcpu *vcpu) return; } +static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec) +{ + kvm_lapic_set_irr(vec, vcpu->arch.apic); + smp_mb__after_atomic(); + + if (avic_vcpu_is_running(vcpu)) + wrmsrl(SVM_AVIC_DOORBELL, + __default_cpu_present_to_apicid(vcpu->cpu)); + else + kvm_vcpu_wake_up(vcpu); +} + static int svm_nmi_allowed(struct kvm_vcpu *vcpu) { struct vcpu_svm *svm = to_svm(vcpu); @@ -3894,6 +3921,9 @@ static void enable_irq_window(struct kvm_vcpu *vcpu) { struct vcpu_svm *svm = to_svm(vcpu); + if (kvm_vcpu_apicv_active(vcpu)) + return; + /* * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes * 1, because that's a separate STGI/VMRUN intercept. The next time we @@ -4637,6 +4667,7 @@ static struct kvm_x86_ops svm_x86_ops = { .sched_in = svm_sched_in, .pmu_ops = &amd_pmu_ops, + .deliver_posted_interrupt = svm_deliver_avic_intr, }; static int __init svm_init(void)
This patch introduces a new mechanism to inject interrupt using AVIC. Since VINTR is not supported when enable AVIC, we need to inject interrupt via APIC backing page instead. This patch also adds support for AVIC doorbell, which is used by KVM to signal a running vcpu to check IRR for injected interrupts. Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com> --- arch/x86/kvm/svm.c | 39 +++++++++++++++++++++++++++++++++++---- 1 file changed, 35 insertions(+), 4 deletions(-)