[RFC PATCH 27/73] KVM: x86/PVM: Implement event injection related callbacks

[Lai Jiangshan <jiangshanlai@gmail.com>]

From: Lai Jiangshan <jiangshan.ljs@antgroup.com>

In PVM, events are injected and consumed directly. The PVM hypervisor
does not follow the IDT-based event delivery mechanism but instead
utilizes a new PVM-specific event delivery ABI, which is similar to FRED
event delivery.

Signed-off-by: Lai Jiangshan <jiangshan.ljs@antgroup.com>
Signed-off-by: Hou Wenlong <houwenlong.hwl@antgroup.com>
---
 arch/x86/kvm/pvm/pvm.c | 193 +++++++++++++++++++++++++++++++++++++++++
 arch/x86/kvm/pvm/pvm.h |   1 +
 2 files changed, 194 insertions(+)

diff --git a/arch/x86/kvm/pvm/pvm.c b/arch/x86/kvm/pvm/pvm.c
index 3d2a3c472664..57d987903791 100644
--- a/arch/x86/kvm/pvm/pvm.c
+++ b/arch/x86/kvm/pvm/pvm.c
@@ -648,6 +648,150 @@ static void pvm_event_flags_update(struct kvm_vcpu *vcpu, unsigned long set,
 	pvm_put_vcpu_struct(pvm, new_flags != old_flags);
 }
 
+static void pvm_standard_event_entry(struct kvm_vcpu *vcpu, unsigned long entry)
+{
+	// Change rip, rflags, rcx and r11 per PVM event delivery specification,
+	// this allows to use sysret in VM enter.
+	kvm_rip_write(vcpu, entry);
+	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
+	kvm_rcx_write(vcpu, entry);
+	kvm_r11_write(vcpu, X86_EFLAGS_IF | X86_EFLAGS_FIXED);
+}
+
+/* handle pvm user event per PVM Spec. */
+static int do_pvm_user_event(struct kvm_vcpu *vcpu, int vector,
+			     bool has_err_code, u64 err_code)
+{
+	struct vcpu_pvm *pvm = to_pvm(vcpu);
+	unsigned long entry = vector == PVM_SYSCALL_VECTOR ?
+			      pvm->msr_lstar : pvm->msr_event_entry;
+	struct pvm_vcpu_struct *pvcs;
+
+	pvcs = pvm_get_vcpu_struct(pvm);
+	if (!pvcs) {
+		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
+		return 1;
+	}
+
+	pvcs->user_cs = pvm->hw_cs;
+	pvcs->user_ss = pvm->hw_ss;
+	pvcs->eflags = kvm_get_rflags(vcpu);
+	pvcs->pkru = 0;
+	pvcs->user_gsbase = pvm_read_guest_gs_base(pvm);
+	pvcs->rip = kvm_rip_read(vcpu);
+	pvcs->rsp = kvm_rsp_read(vcpu);
+	pvcs->rcx = kvm_rcx_read(vcpu);
+	pvcs->r11 = kvm_r11_read(vcpu);
+
+	if (has_err_code)
+		pvcs->event_errcode = err_code;
+	if (vector != PVM_SYSCALL_VECTOR)
+		pvcs->event_vector = vector;
+
+	if (vector == PF_VECTOR)
+		pvcs->cr2 = vcpu->arch.cr2;
+
+	pvm_put_vcpu_struct(pvm, true);
+
+	switch_to_smod(vcpu);
+
+	pvm_standard_event_entry(vcpu, entry);
+
+	return 1;
+}
+
+static int do_pvm_supervisor_exception(struct kvm_vcpu *vcpu, int vector,
+				       bool has_error_code, u64 error_code)
+{
+	struct vcpu_pvm *pvm = to_pvm(vcpu);
+	unsigned long stack;
+	struct pvm_supervisor_event frame;
+	struct x86_exception e;
+	int ret;
+
+	memset(&frame, 0, sizeof(frame));
+	frame.cs = kernel_cs_by_msr(pvm->msr_star);
+	frame.ss = kernel_ds_by_msr(pvm->msr_star);
+	frame.rip = kvm_rip_read(vcpu);
+	frame.rflags = kvm_get_rflags(vcpu);
+	frame.rsp = kvm_rsp_read(vcpu);
+	frame.errcode = ((unsigned long)vector << 32) | error_code;
+	frame.r11 = kvm_r11_read(vcpu);
+	frame.rcx = kvm_rcx_read(vcpu);
+
+	stack = ((frame.rsp - pvm->msr_supervisor_redzone) & ~15UL) - sizeof(frame);
+
+	ret = kvm_write_guest_virt_system(vcpu, stack, &frame, sizeof(frame), &e);
+	if (ret) {
+		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
+		return 1;
+	}
+
+	if (vector == PF_VECTOR) {
+		struct pvm_vcpu_struct *pvcs;
+
+		pvcs = pvm_get_vcpu_struct(pvm);
+		if (!pvcs) {
+			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
+			return 1;
+		}
+
+		pvcs->cr2 = vcpu->arch.cr2;
+		pvm_put_vcpu_struct(pvm, true);
+	}
+
+	kvm_rsp_write(vcpu, stack);
+
+	pvm_standard_event_entry(vcpu, pvm->msr_event_entry + 256);
+
+	return 1;
+}
+
+static int do_pvm_supervisor_interrupt(struct kvm_vcpu *vcpu, int vector,
+				       bool has_error_code, u64 error_code)
+{
+	struct vcpu_pvm *pvm = to_pvm(vcpu);
+	unsigned long stack = kvm_rsp_read(vcpu);
+	struct pvm_vcpu_struct *pvcs;
+
+	pvcs = pvm_get_vcpu_struct(pvm);
+	if (!pvcs) {
+		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
+		return 1;
+	}
+	pvcs->eflags = kvm_get_rflags(vcpu);
+	pvcs->rip = kvm_rip_read(vcpu);
+	pvcs->rsp = stack;
+	pvcs->rcx = kvm_rcx_read(vcpu);
+	pvcs->r11 = kvm_r11_read(vcpu);
+
+	pvcs->event_vector = vector;
+	if (has_error_code)
+		pvcs->event_errcode = error_code;
+
+	pvm_put_vcpu_struct(pvm, true);
+
+	stack = (stack - pvm->msr_supervisor_redzone) & ~15UL;
+	kvm_rsp_write(vcpu, stack);
+
+	pvm_standard_event_entry(vcpu, pvm->msr_event_entry + 512);
+
+	return 1;
+}
+
+static int do_pvm_event(struct kvm_vcpu *vcpu, int vector,
+			bool has_error_code, u64 error_code)
+{
+	if (!is_smod(to_pvm(vcpu)))
+		return do_pvm_user_event(vcpu, vector, has_error_code, error_code);
+
+	if (vector < 32)
+		return do_pvm_supervisor_exception(vcpu, vector,
+						   has_error_code, error_code);
+
+	return do_pvm_supervisor_interrupt(vcpu, vector, has_error_code, error_code);
+}
+
 static unsigned long pvm_get_rflags(struct kvm_vcpu *vcpu)
 {
 	return to_pvm(vcpu)->rflags;
@@ -722,6 +866,51 @@ static int pvm_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
 	return !pvm->nmi_mask && !pvm->int_shadow;
 }
 
+/* Always inject the exception directly and consume the event. */
+static void pvm_inject_exception(struct kvm_vcpu *vcpu)
+{
+	unsigned int vector = vcpu->arch.exception.vector;
+	bool has_error_code = vcpu->arch.exception.has_error_code;
+	u32 error_code = vcpu->arch.exception.error_code;
+
+	kvm_deliver_exception_payload(vcpu, &vcpu->arch.exception);
+
+	if (do_pvm_event(vcpu, vector, has_error_code, error_code))
+		kvm_clear_exception_queue(vcpu);
+}
+
+/* Always inject the interrupt directly and consume the event. */
+static void pvm_inject_irq(struct kvm_vcpu *vcpu, bool reinjected)
+{
+	int irq = vcpu->arch.interrupt.nr;
+
+	trace_kvm_inj_virq(irq, vcpu->arch.interrupt.soft, false);
+
+	if (do_pvm_event(vcpu, irq, false, 0))
+		kvm_clear_interrupt_queue(vcpu);
+
+	++vcpu->stat.irq_injections;
+}
+
+/* Always inject the NMI directly and consume the event. */
+static void pvm_inject_nmi(struct kvm_vcpu *vcpu)
+{
+	if (do_pvm_event(vcpu, NMI_VECTOR, false, 0)) {
+		vcpu->arch.nmi_injected = false;
+		pvm_set_nmi_mask(vcpu, true);
+	}
+
+	++vcpu->stat.nmi_injections;
+}
+
+static void pvm_cancel_injection(struct kvm_vcpu *vcpu)
+{
+	/*
+	 * Nothing to do. Since exceptions/interrupts are delivered immediately
+	 * during event injection, so they cannot be cancelled and reinjected.
+	 */
+}
+
 static void pvm_setup_mce(struct kvm_vcpu *vcpu)
 {
 }
@@ -1282,6 +1471,10 @@ static struct kvm_x86_ops pvm_x86_ops __initdata = {
 	.handle_exit = pvm_handle_exit,
 	.set_interrupt_shadow = pvm_set_interrupt_shadow,
 	.get_interrupt_shadow = pvm_get_interrupt_shadow,
+	.inject_irq = pvm_inject_irq,
+	.inject_nmi = pvm_inject_nmi,
+	.inject_exception = pvm_inject_exception,
+	.cancel_injection = pvm_cancel_injection,
 	.interrupt_allowed = pvm_interrupt_allowed,
 	.nmi_allowed = pvm_nmi_allowed,
 	.get_nmi_mask = pvm_get_nmi_mask,
diff --git a/arch/x86/kvm/pvm/pvm.h b/arch/x86/kvm/pvm/pvm.h
index b0c633ce2987..39506ddbe5c5 100644
--- a/arch/x86/kvm/pvm/pvm.h
+++ b/arch/x86/kvm/pvm/pvm.h
@@ -7,6 +7,7 @@
 
 #define SWITCH_FLAGS_INIT	(SWITCH_FLAGS_SMOD)
 
+#define PVM_SYSCALL_VECTOR		SWITCH_EXIT_REASONS_SYSCALL
 #define PVM_FAILED_VMENTRY_VECTOR	SWITCH_EXIT_REASONS_FAILED_VMETNRY
 
 #define PT_L4_SHIFT		39
-- 
2.19.1.6.gb485710b