[PATCH 4/4] Fix task switching.

[Gleb Natapov <gleb@redhat.com>]

The patch fixes two problems with task switching.
1. Back link is written to a wrong TSS.
2. Instruction emulation is not needed if the reason for task switch
   is a task gate in IDT and access to it is caused by an external even.

2 is currently solved only for VMX since there is not reliable way to
skip an instruction in SVM. We should emulate it instead.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
---

 arch/x86/include/asm/svm.h |    1 +
 arch/x86/kvm/svm.c         |   25 ++++++++++++++++++-------
 arch/x86/kvm/vmx.c         |   40 +++++++++++++++++++++++++++++-----------
 arch/x86/kvm/x86.c         |   40 +++++++++++++++++++++++++++++++---------
 4 files changed, 79 insertions(+), 27 deletions(-)

diff --git a/arch/x86/include/asm/svm.h b/arch/x86/include/asm/svm.h
index 82ada75..85574b7 100644
--- a/arch/x86/include/asm/svm.h
+++ b/arch/x86/include/asm/svm.h
@@ -225,6 +225,7 @@ struct __attribute__ ((__packed__)) vmcb {
 #define SVM_EVTINJ_VALID_ERR (1 << 11)
 
 #define SVM_EXITINTINFO_VEC_MASK SVM_EVTINJ_VEC_MASK
+#define SVM_EXITINTINFO_TYPE_MASK SVM_EVTINJ_TYPE_MASK
 
 #define	SVM_EXITINTINFO_TYPE_INTR SVM_EVTINJ_TYPE_INTR
 #define	SVM_EXITINTINFO_TYPE_NMI SVM_EVTINJ_TYPE_NMI
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index 1fcbc17..3ffb695 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -1823,17 +1823,28 @@ static int task_switch_interception(struct vcpu_svm *svm,
 				    struct kvm_run *kvm_run)
 {
 	u16 tss_selector;
+	int reason;
+	int int_type = svm->vmcb->control.exit_int_info &
+		SVM_EXITINTINFO_TYPE_MASK;
 
 	tss_selector = (u16)svm->vmcb->control.exit_info_1;
+
 	if (svm->vmcb->control.exit_info_2 &
 	    (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
-		return kvm_task_switch(&svm->vcpu, tss_selector,
-				       TASK_SWITCH_IRET);
-	if (svm->vmcb->control.exit_info_2 &
-	    (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
-		return kvm_task_switch(&svm->vcpu, tss_selector,
-				       TASK_SWITCH_JMP);
-	return kvm_task_switch(&svm->vcpu, tss_selector, TASK_SWITCH_CALL);
+		reason = TASK_SWITCH_IRET;
+	else if (svm->vmcb->control.exit_info_2 &
+		 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
+		reason = TASK_SWITCH_JMP;
+	else if (svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID)
+		reason = TASK_SWITCH_GATE;
+	else
+		reason = TASK_SWITCH_CALL;
+
+
+	if (reason != TASK_SWITCH_GATE || int_type == SVM_EXITINTINFO_TYPE_SOFT)
+		skip_emulated_instruction(&svm->vcpu);
+
+	return kvm_task_switch(&svm->vcpu, tss_selector, reason);
 }
 
 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 0da7a9e..01db958 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -3025,22 +3025,40 @@ static int handle_task_switch(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 	struct vcpu_vmx *vmx = to_vmx(vcpu);
 	unsigned long exit_qualification;
 	u16 tss_selector;
-	int reason;
+	int reason, type, idt_v;
+
+	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
+	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
 
 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
 
 	reason = (u32)exit_qualification >> 30;
-	if (reason == TASK_SWITCH_GATE && vmx->vcpu.arch.nmi_injected &&
-	    (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
-	    (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK)
-	    == INTR_TYPE_NMI_INTR) {
-		vcpu->arch.nmi_injected = false;
-		if (cpu_has_virtual_nmis())
-			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
-				      GUEST_INTR_STATE_NMI);
+	if (reason == TASK_SWITCH_GATE && idt_v) {
+		switch (type) {
+		case INTR_TYPE_NMI_INTR:
+			vcpu->arch.nmi_injected = false;
+			if (cpu_has_virtual_nmis())
+				vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
+					      GUEST_INTR_STATE_NMI);
+			break;
+		case INTR_TYPE_EXT_INTR:
+			kvm_clear_interrupt_queue(vcpu);
+			break;
+		case INTR_TYPE_HARD_EXCEPTION:
+		case INTR_TYPE_SOFT_EXCEPTION:
+			kvm_clear_exception_queue(vcpu);
+			break;
+		default:
+			break;
+		}
 	}
 	tss_selector = exit_qualification;
 
+	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
+		       type != INTR_TYPE_EXT_INTR &&
+		       type != INTR_TYPE_NMI_INTR))
+		skip_emulated_instruction(vcpu);
+
 	if (!kvm_task_switch(vcpu, tss_selector, reason))
 		return 0;
 
@@ -3292,8 +3310,8 @@ static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
 
 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
-	
-	switch(type) {
+
+	switch (type) {
 	case INTR_TYPE_NMI_INTR:
 		vmx->vcpu.arch.nmi_injected = true;
 		/*
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index ae4918c..573bb3f 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -3697,7 +3697,6 @@ static void save_state_to_tss32(struct kvm_vcpu *vcpu,
 	tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
 	tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
 	tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
-	tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
 }
 
 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
@@ -3794,8 +3793,8 @@ static int load_state_from_tss16(struct kvm_vcpu *vcpu,
 }
 
 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
-		       u32 old_tss_base,
-		       struct desc_struct *nseg_desc)
+			      u16 old_tss_sel, u32 old_tss_base,
+			      struct desc_struct *nseg_desc)
 {
 	struct tss_segment_16 tss_segment_16;
 	int ret = 0;
@@ -3814,6 +3813,16 @@ static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
 			   &tss_segment_16, sizeof tss_segment_16))
 		goto out;
 
+	if (old_tss_sel != 0xffff) {
+		tss_segment_16.prev_task_link = old_tss_sel;
+
+		if (kvm_write_guest(vcpu->kvm,
+				    get_tss_base_addr(vcpu, nseg_desc),
+				    &tss_segment_16.prev_task_link,
+				    sizeof tss_segment_16.prev_task_link))
+			goto out;
+	}
+
 	if (load_state_from_tss16(vcpu, &tss_segment_16))
 		goto out;
 
@@ -3823,7 +3832,7 @@ out:
 }
 
 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
-		       u32 old_tss_base,
+		       u16 old_tss_sel, u32 old_tss_base,
 		       struct desc_struct *nseg_desc)
 {
 	struct tss_segment_32 tss_segment_32;
@@ -3843,6 +3852,16 @@ static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
 			   &tss_segment_32, sizeof tss_segment_32))
 		goto out;
 
+	if (old_tss_sel != 0xffff) {
+		tss_segment_32.prev_task_link = old_tss_sel;
+
+		if (kvm_write_guest(vcpu->kvm,
+				    get_tss_base_addr(vcpu, nseg_desc),
+				    &tss_segment_32.prev_task_link,
+				    sizeof tss_segment_32.prev_task_link))
+			goto out;
+	}
+
 	if (load_state_from_tss32(vcpu, &tss_segment_32))
 		goto out;
 
@@ -3896,14 +3915,17 @@ int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
 		kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
 	}
 
-	kvm_x86_ops->skip_emulated_instruction(vcpu);
+	/* set back link to prev task only if NT bit is set in eflags
+	   note that old_tss_sel is not used afetr this point */
+	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
+		old_tss_sel = 0xffff;
 
 	if (nseg_desc.type & 8)
-		ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
-					 &nseg_desc);
+		ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
+					 old_tss_base, &nseg_desc);
 	else
-		ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
-					 &nseg_desc);
+		ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
+					 old_tss_base, &nseg_desc);
 
 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
 		u32 eflags = kvm_x86_ops->get_rflags(vcpu);