Nested VMX: Update APIC-v(RVI/SVI) when vmexit to L1
authorYang Zhang <yang.z.zhang@Intel.com>
Thu, 22 Aug 2013 08:59:01 +0000 (10:59 +0200)
committerJan Beulich <jbeulich@suse.com>
Thu, 22 Aug 2013 08:59:01 +0000 (10:59 +0200)
If enabling APIC-v, all interrupts to L1 are delivered through APIC-v.
But when L2 is running, external interrupt will casue L1 vmexit with
reason external interrupt. Then L1 will pick up the interrupt through
vmcs12. when L1 ack the interrupt, since the APIC-v is enabled when
L1 is running, so APIC-v hardware still will do vEOI updating. The problem
is that the interrupt is delivered not through APIC-v hardware, this means
SVI/RVI/vPPR are not setting, but hardware required them when doing vEOI
updating. The solution is that, when L1 tried to pick up the interrupt
from vmcs12, then hypervisor will help to update the SVI/RVI/vPPR to make
sure the following vEOI updating and vPPR updating corrently.

Also, since interrupt is delivered through vmcs12, so APIC-v hardware will
not cleare vIRR and hypervisor need to clear it before L1 running.

Signed-off-by: Yang Zhang <yang.z.zhang@Intel.com>
Acked-by: "Dong, Eddie" <eddie.dong@intel.com>
xen/arch/x86/hvm/irq.c
xen/arch/x86/hvm/vlapic.c
xen/arch/x86/hvm/vmx/intr.c
xen/arch/x86/hvm/vmx/vmx.c
xen/arch/x86/hvm/vmx/vvmx.c
xen/include/asm-x86/hvm/vlapic.h
xen/include/asm-x86/hvm/vmx/vmx.h
xen/include/asm-x86/hvm/vmx/vvmx.h

index 9eae5de21ad8c8678ce5d17cf64e5da6ac7e1fb3..6a6fb68eab0ba0b73062eced0494d56719a043ce 100644 (file)
@@ -437,7 +437,7 @@ struct hvm_intack hvm_vcpu_ack_pending_irq(
             intack.vector = (uint8_t)vector;
         break;
     case hvm_intsrc_lapic:
-        if ( !vlapic_ack_pending_irq(v, intack.vector) )
+        if ( !vlapic_ack_pending_irq(v, intack.vector, 0) )
             intack = hvm_intack_none;
         break;
     case hvm_intsrc_vector:
index f1530fdedbc377b320f4d61e5c36c7e36ae8246a..7b10ab627469126c634af7e62790383ee3fb4960 100644 (file)
@@ -168,6 +168,14 @@ static uint32_t vlapic_get_ppr(struct vlapic *vlapic)
     return ppr;
 }
 
+uint32_t vlapic_set_ppr(struct vlapic *vlapic)
+{
+   uint32_t ppr = vlapic_get_ppr(vlapic);
+
+   vlapic_set_reg(vlapic, APIC_PROCPRI, ppr);
+   return ppr;
+}
+
 static int vlapic_match_logical_addr(struct vlapic *vlapic, uint8_t mda)
 {
     int result = 0;
@@ -1050,15 +1058,15 @@ int vlapic_has_pending_irq(struct vcpu *v)
     return irr;
 }
 
-int vlapic_ack_pending_irq(struct vcpu *v, int vector)
+int vlapic_ack_pending_irq(struct vcpu *v, int vector, bool_t force_ack)
 {
     struct vlapic *vlapic = vcpu_vlapic(v);
 
-    if ( vlapic_virtual_intr_delivery_enabled() )
-        return 1;
-
-    vlapic_set_vector(vector, &vlapic->regs->data[APIC_ISR]);
-    vlapic_clear_irr(vector, vlapic);
+    if ( force_ack || !vlapic_virtual_intr_delivery_enabled() )
+    {
+        vlapic_set_vector(vector, &vlapic->regs->data[APIC_ISR]);
+        vlapic_clear_irr(vector, vlapic);
+    }
 
     return 1;
 }
index cab91094f4e633825b0c91c6b9689cd45de9ea02..45942ab43a7ab9e4b9404141237438ac99ac1f67 100644 (file)
@@ -185,7 +185,7 @@ static int nvmx_intr_intercept(struct vcpu *v, struct hvm_intack intack)
             if ( !(ctrl & PIN_BASED_EXT_INTR_MASK) )
                 return 0;
 
-            vmx_inject_extint(intack.vector);
+            vmx_inject_extint(intack.vector, intack.source);
 
             ctrl = __get_vvmcs(vcpu_nestedhvm(v).nv_vvmcx, VM_EXIT_CONTROLS);
             if ( ctrl & VM_EXIT_ACK_INTR_ON_EXIT )
@@ -314,7 +314,7 @@ void vmx_intr_assist(void)
     else
     {
         HVMTRACE_2D(INJ_VIRQ, intack.vector, /*fake=*/ 0);
-        vmx_inject_extint(intack.vector);
+        vmx_inject_extint(intack.vector, intack.source);
         pt_intr_post(v, intack);
     }
 
index 8ed70264c70cc85051d03d37fc39126f4b0150f1..011a8179d90cad0524fd6c1c6b1487af093acfcd 100644 (file)
@@ -1208,7 +1208,7 @@ static void vmx_update_guest_efer(struct vcpu *v)
 }
 
 void nvmx_enqueue_n2_exceptions(struct vcpu *v, 
-            unsigned long intr_fields, int error_code)
+            unsigned long intr_fields, int error_code, uint8_t source)
 {
     struct nestedvmx *nvmx = &vcpu_2_nvmx(v);
 
@@ -1216,6 +1216,7 @@ void nvmx_enqueue_n2_exceptions(struct vcpu *v,
         /* enqueue the exception till the VMCS switch back to L1 */
         nvmx->intr.intr_info = intr_fields;
         nvmx->intr.error_code = error_code;
+        nvmx->intr.source = source;
         vcpu_nestedhvm(v).nv_vmexit_pending = 1;
         return;
     }
@@ -1227,7 +1228,8 @@ void nvmx_enqueue_n2_exceptions(struct vcpu *v,
 
 static int nvmx_vmexit_trap(struct vcpu *v, struct hvm_trap *trap)
 {
-    nvmx_enqueue_n2_exceptions(v, trap->vector, trap->error_code);
+    nvmx_enqueue_n2_exceptions(v, trap->vector, trap->error_code,
+                               hvm_intsrc_none);
     return NESTEDHVM_VMEXIT_DONE;
 }
 
@@ -1258,7 +1260,7 @@ static void __vmx_inject_exception(int trap, int type, int error_code)
         curr->arch.hvm_vmx.vmx_emulate = 1;
 }
 
-void vmx_inject_extint(int trap)
+void vmx_inject_extint(int trap, uint8_t source)
 {
     struct vcpu *v = current;
     u32    pin_based_cntrl;
@@ -1269,7 +1271,7 @@ void vmx_inject_extint(int trap)
         if ( pin_based_cntrl & PIN_BASED_EXT_INTR_MASK ) {
             nvmx_enqueue_n2_exceptions (v, 
                INTR_INFO_VALID_MASK | (X86_EVENTTYPE_EXT_INTR<<8) | trap,
-               HVM_DELIVER_NO_ERROR_CODE);
+               HVM_DELIVER_NO_ERROR_CODE, source);
             return;
         }
     }
@@ -1288,7 +1290,7 @@ void vmx_inject_nmi(void)
         if ( pin_based_cntrl & PIN_BASED_NMI_EXITING ) {
             nvmx_enqueue_n2_exceptions (v, 
                INTR_INFO_VALID_MASK | (X86_EVENTTYPE_NMI<<8) | TRAP_nmi,
-               HVM_DELIVER_NO_ERROR_CODE);
+               HVM_DELIVER_NO_ERROR_CODE, hvm_intsrc_nmi);
             return;
         }
     }
@@ -1356,7 +1358,7 @@ static void vmx_inject_trap(struct hvm_trap *trap)
     {
         nvmx_enqueue_n2_exceptions (curr, 
             INTR_INFO_VALID_MASK | (_trap.type<<8) | _trap.vector,
-            _trap.error_code); 
+            _trap.error_code, hvm_intsrc_none);
         return;
     }
     else
index 0dc567a97278035b9e5b76cec021ef6fafc1cbb2..cecc72f05359c7dfc59eb0fb47d6ee71143036c0 100644 (file)
@@ -1295,6 +1295,36 @@ static void sync_exception_state(struct vcpu *v)
     }
 }
 
+static void nvmx_update_apicv(struct vcpu *v)
+{
+    struct nestedvmx *nvmx = &vcpu_2_nvmx(v);
+    struct nestedvcpu *nvcpu = &vcpu_nestedhvm(v);
+    unsigned long reason = __get_vvmcs(nvcpu->nv_vvmcx, VM_EXIT_REASON);
+    uint32_t intr_info = __get_vvmcs(nvcpu->nv_vvmcx, VM_EXIT_INTR_INFO);
+
+    if ( reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
+         nvmx->intr.source == hvm_intsrc_lapic &&
+         (intr_info & INTR_INFO_VALID_MASK) )
+    {
+        uint16_t status;
+        uint32_t rvi, ppr;
+        uint32_t vector = intr_info & 0xff;
+        struct vlapic *vlapic = vcpu_vlapic(v);
+
+        vlapic_ack_pending_irq(v, vector, 1);
+
+        ppr = vlapic_set_ppr(vlapic);
+        WARN_ON((ppr & 0xf0) != (vector & 0xf0));
+
+        status = vector << 8;
+        rvi = vlapic_has_pending_irq(v);
+        if ( rvi != -1 )
+            status |= rvi & 0xff;
+
+        __vmwrite(GUEST_INTR_STATUS, status);
+    }
+}
+
 static void virtual_vmexit(struct cpu_user_regs *regs)
 {
     struct vcpu *v = current;
@@ -1340,6 +1370,9 @@ static void virtual_vmexit(struct cpu_user_regs *regs)
     /* updating host cr0 to sync TS bit */
     __vmwrite(HOST_CR0, v->arch.hvm_vmx.host_cr0);
 
+    if ( cpu_has_vmx_virtual_intr_delivery )
+        nvmx_update_apicv(v);
+
     vmreturn(regs, VMSUCCEED);
 }
 
index 021a5f278fa4302bf5d65923b454caac29c136ed..eb6dec96cba9c9960327c10a0cf2b2c15d5a6951 100644 (file)
@@ -96,7 +96,7 @@ bool_t is_vlapic_lvtpc_enabled(struct vlapic *vlapic);
 void vlapic_set_irq(struct vlapic *vlapic, uint8_t vec, uint8_t trig);
 
 int vlapic_has_pending_irq(struct vcpu *v);
-int vlapic_ack_pending_irq(struct vcpu *v, int vector);
+int vlapic_ack_pending_irq(struct vcpu *v, int vector, bool_t force_ack);
 
 int  vlapic_init(struct vcpu *v);
 void vlapic_destroy(struct vcpu *v);
@@ -108,6 +108,7 @@ void vlapic_tdt_msr_set(struct vlapic *vlapic, uint64_t value);
 uint64_t vlapic_tdt_msr_get(struct vlapic *vlapic);
 
 int vlapic_accept_pic_intr(struct vcpu *v);
+uint32_t vlapic_set_ppr(struct vlapic *vlapic);
 
 void vlapic_adjust_i8259_target(struct domain *d);
 
index c33b9f9a7e8c9391a88113dbfe9b750165993103..f4d759bd5a89f5e49e1f7a2107e62abcf78081c6 100644 (file)
@@ -448,7 +448,7 @@ static inline int __vmxon(u64 addr)
 
 void vmx_get_segment_register(struct vcpu *, enum x86_segment,
                               struct segment_register *);
-void vmx_inject_extint(int trap);
+void vmx_inject_extint(int trap, uint8_t source);
 void vmx_inject_nmi(void);
 
 int ept_p2m_init(struct p2m_domain *p2m);
index 3874525c1ac6334e7ff881bae7e1b697de2a8455..848be75796360008a8a6fa65000e81ecae08400d 100644 (file)
@@ -36,6 +36,7 @@ struct nestedvmx {
     struct {
         unsigned long intr_info;
         u32           error_code;
+        u8            source;
     } intr;
     struct {
         bool_t   enabled;