KVM: Do not communicate to userspace through cpu registers during PIO

Currently when passing the a PIO emulation request to userspace, we
rely on userspace updating %rax (on 'in' instructions) and %rsi/%rdi/%rcx
(on string instructions).  This (a) requires two extra ioctls for getting
and setting the registers and (b) is unfriendly to non-x86 archs, when
they get kvm ports.

So fix by doing the register fixups in the kernel and passing to userspace
only an abstract description of the PIO to be done.

Signed-off-by: Avi Kivity <avi@qumranet.com>

4 files changed, 50 insertions, 3 deletions

diff --git a/drivers/kvm/kvm.h b/drivers/kvm/kvm.h
index 901b8d9..59cbc5b 100644
--- a/drivers/kvm/kvm.h
+++ b/drivers/kvm/kvm.h
@@ -274,6 +274,7 @@ struct kvm_vcpu {
 	int mmio_size;
 	unsigned char mmio_data[8];
 	gpa_t mmio_phys_addr;
+	int pio_pending;
 
 	struct {
 		int active;
diff --git a/drivers/kvm/kvm_main.c b/drivers/kvm/kvm_main.c
index 42be8a8..ff8bcfe 100644
--- a/drivers/kvm/kvm_main.c
+++ b/drivers/kvm/kvm_main.c
@@ -1504,6 +1504,44 @@ void save_msrs(struct vmx_msr_entry *e, int n)
 }
 EXPORT_SYMBOL_GPL(save_msrs);
 
+static void complete_pio(struct kvm_vcpu *vcpu)
+{
+	struct kvm_io *io = &vcpu->run->io;
+	long delta;
+
+	kvm_arch_ops->cache_regs(vcpu);
+
+	if (!io->string) {
+		if (io->direction == KVM_EXIT_IO_IN)
+			memcpy(&vcpu->regs[VCPU_REGS_RAX], &io->value,
+			       io->size);
+	} else {
+		delta = 1;
+		if (io->rep) {
+			delta *= io->count;
+			/*
+			 * The size of the register should really depend on
+			 * current address size.
+			 */
+			vcpu->regs[VCPU_REGS_RCX] -= delta;
+		}
+		if (io->string_down)
+			delta = -delta;
+		delta *= io->size;
+		if (io->direction == KVM_EXIT_IO_IN)
+			vcpu->regs[VCPU_REGS_RDI] += delta;
+		else
+			vcpu->regs[VCPU_REGS_RSI] += delta;
+	}
+
+	vcpu->pio_pending = 0;
+	vcpu->run->io_completed = 0;
+
+	kvm_arch_ops->decache_regs(vcpu);
+
+	kvm_arch_ops->skip_emulated_instruction(vcpu);
+}
+
 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 {
 	int r;
@@ -1518,9 +1556,13 @@ static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 		kvm_run->emulated = 0;
 	}
 
-	if (kvm_run->mmio_completed) {
-		memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
-		vcpu->mmio_read_completed = 1;
+	if (kvm_run->io_completed) {
+		if (vcpu->pio_pending)
+			complete_pio(vcpu);
+		else {
+			memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
+			vcpu->mmio_read_completed = 1;
+		}
 	}
 
 	vcpu->mmio_needed = 0;
diff --git a/drivers/kvm/svm.c b/drivers/kvm/svm.c
index 6787f11..c35b8c8 100644
--- a/drivers/kvm/svm.c
+++ b/drivers/kvm/svm.c
@@ -1037,6 +1037,7 @@ static int io_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 	kvm_run->io.size = ((io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT);
 	kvm_run->io.string = (io_info & SVM_IOIO_STR_MASK) != 0;
 	kvm_run->io.rep = (io_info & SVM_IOIO_REP_MASK) != 0;
+	kvm_run->io.count = 1;
 
 	if (kvm_run->io.string) {
 		unsigned addr_mask;
@@ -1056,6 +1057,7 @@ static int io_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 		}
 	} else
 		kvm_run->io.value = vcpu->svm->vmcb->save.rax;
+	vcpu->pio_pending = 1;
 	return 0;
 }
 
diff --git a/drivers/kvm/vmx.c b/drivers/kvm/vmx.c
index a721b60..4d5f40f 100644
--- a/drivers/kvm/vmx.c
+++ b/drivers/kvm/vmx.c
@@ -1459,12 +1459,14 @@ static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 		= (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
 	kvm_run->io.rep = (exit_qualification & 32) != 0;
 	kvm_run->io.port = exit_qualification >> 16;
+	kvm_run->io.count = 1;
 	if (kvm_run->io.string) {
 		if (!get_io_count(vcpu, &kvm_run->io.count))
 			return 1;
 		kvm_run->io.address = vmcs_readl(GUEST_LINEAR_ADDRESS);
 	} else
 		kvm_run->io.value = vcpu->regs[VCPU_REGS_RAX]; /* rax */
+	vcpu->pio_pending = 1;
 	return 0;
 }