kvm: Virtio block device emulation

This patch implements a rough first version of virtio block device emulation.
You can enable it with the '--enable-virtio' command line option. We're still
missing virtio queue handling so enabling virtio will cause a guest kernel to
hang.

Signed-off-by: Pekka Enberg <penberg@kernel.org>

blk-virtio.c

@@ -2,9 +2,10 @@
 
 #include "kvm/virtio_pci.h"
 #include "kvm/ioport.h"
+#include "kvm/kvm.h"
 #include "kvm/pci.h"
 
-#define VIRTIO_PCI_IOPORT_SIZE		24
+#define VIRTIO_BLK_IRQ		14
 
 struct device {
 	uint32_t		guest_features;
@@ -19,13 +20,22 @@ static bool blk_virtio_in(struct kvm *self, uint16_t port, void *data, int size,
 
 	offset		= port - IOPORT_VIRTIO;
 
+	/* XXX: Let virtio block device handle this */
+	if (offset >= VIRTIO_PCI_CONFIG_NOMSI)
+		return true;
+
 	switch (offset) {
 	case VIRTIO_PCI_HOST_FEATURES:
 		ioport__write32(data, 0x00);
 		break;
 	case VIRTIO_PCI_GUEST_FEATURES:
+		return false;
 	case VIRTIO_PCI_QUEUE_PFN:
+		ioport__write32(data, 0x00);
+		break;
 	case VIRTIO_PCI_QUEUE_NUM:
+		ioport__write16(data, 0x10);
+		break;
 	case VIRTIO_PCI_QUEUE_SEL:
 	case VIRTIO_PCI_QUEUE_NOTIFY:
 		return false;
@@ -33,6 +43,9 @@ static bool blk_virtio_in(struct kvm *self, uint16_t port, void *data, int size,
 		ioport__write8(data, device.status);
 		break;
 	case VIRTIO_PCI_ISR:
+		ioport__write8(data, 0x1);
+		kvm__irq_line(self, VIRTIO_BLK_IRQ, 0);
+		return true;
 	case VIRTIO_MSI_CONFIG_VECTOR:
 	default:
 		return false;
@@ -41,22 +54,28 @@ static bool blk_virtio_in(struct kvm *self, uint16_t port, void *data, int size,
 	return true;
 }
 
-#include <stdio.h>
-
 static bool blk_virtio_out(struct kvm *self, uint16_t port, void *data, int size, uint32_t count)
 {
 	unsigned long offset;
 
 	offset		= port - IOPORT_VIRTIO;
 
+	/* XXX: Let virtio block device handle this */
+	if (offset >= VIRTIO_PCI_CONFIG_NOMSI)
+		return true;
+
 	switch (offset) {
 	case VIRTIO_PCI_GUEST_FEATURES:
 		device.guest_features	= ioport__read32(data);
 		break;
 	case VIRTIO_PCI_QUEUE_PFN:
+		return true;
 	case VIRTIO_PCI_QUEUE_SEL:
-	case VIRTIO_PCI_QUEUE_NOTIFY:
-		return false;
+		return true;
+	case VIRTIO_PCI_QUEUE_NOTIFY: {
+		kvm__irq_line(self, VIRTIO_BLK_IRQ, 1);
+		return true;
+	}
 	case VIRTIO_PCI_STATUS:
 		device.status		= ioport__read8(data);
 		break;
@@ -88,11 +107,14 @@ static struct pci_device_header blk_virtio_pci_device = {
 	.subsys_vendor_id	= PCI_SUBSYSTEM_VENDOR_ID_REDHAT_QUMRANET,
 	.subsys_id		= PCI_SUBSYSTEM_ID_VIRTIO_BLK,
 	.bar[0]			= IOPORT_VIRTIO | PCI_BASE_ADDRESS_SPACE_IO,
+	/* XXX: Is this IRQ setup OK? */
+	.irq_pin		= 1,
+	.irq_line		= VIRTIO_BLK_IRQ,
 };
 
 void blk_virtio__init(void)
 {
 	pci__register(&blk_virtio_pci_device, 1);
 
-	ioport__register(IOPORT_VIRTIO, &blk_virtio_io_ops, VIRTIO_PCI_IOPORT_SIZE);
+	ioport__register(IOPORT_VIRTIO, &blk_virtio_io_ops, 256);
 }

include/kvm/ioport.h

@@ -22,6 +22,11 @@ static inline uint8_t ioport__read8(uint8_t *data)
 	return *data;
 }
 
+static inline uint16_t ioport__read16(uint16_t *data)
+{
+	return *data;
+}
+
 static inline uint32_t ioport__read32(uint32_t *data)
 {
 	return *data;
@@ -32,6 +37,11 @@ static inline void ioport__write8(uint8_t *data, uint8_t value)
 	*data		 = value;
 }
 
+static inline void ioport__write16(uint16_t *data, uint16_t value)
+{
+	*data		 = value;
+}
+
 static inline void ioport__write32(uint32_t *data, uint32_t value)
 {
 	*data		 = value;

include/kvm/kvm.h

@@ -40,6 +40,7 @@ bool kvm__load_kernel(struct kvm *kvm, const char *kernel_filename,
 void kvm__reset_vcpu(struct kvm *self);
 void kvm__setup_mem(struct kvm *self);
 void kvm__run(struct kvm *self);
+void kvm__irq_line(struct kvm *self, int irq, int level);
 bool kvm__emulate_io(struct kvm *self, uint16_t port, void *data, int direction, int size, uint32_t count);
 bool kvm__emulate_mmio(struct kvm *self, uint64_t phys_addr, uint8_t *data, uint32_t len, uint8_t is_write);
 

include/kvm/virtio_pci.h

@@ -49,4 +49,8 @@
 /* A 16-bit vector for selected queue notifications. */
 #define VIRTIO_MSI_QUEUE_VECTOR         22
 
+/* Config space size */
+#define VIRTIO_PCI_CONFIG_NOMSI         20
+#define VIRTIO_PCI_CONFIG_MSI           24
+
 #endif /* _LINUX_VIRTIO_PCI_H */

kvm.c

@@ -611,6 +611,21 @@ void kvm__run(struct kvm *self)
 		die_perror("KVM_RUN failed");
 }
 
+void kvm__irq_line(struct kvm *self, int irq, int level)
+{
+	struct kvm_irq_level irq_level;
+
+	irq_level	= (struct kvm_irq_level) {
+		{
+			.irq		= irq,
+		},
+		.level		= level,
+	};
+
+	if (ioctl(self->vm_fd, KVM_IRQ_LINE, &irq_level) < 0)
+		die_perror("KVM_IRQ_LINE failed");
+}
+
 static void print_dtable(const char *name, struct kvm_dtable *dtable)
 {
 	printf(" %s                 %016" PRIx64 "  %08" PRIx16 "\n",

main.c

@@ -51,6 +51,7 @@ int main(int argc, char *argv[])
 	const char *kernel_cmdline = NULL;
 	const char *kvm_dev = "/dev/kvm";
 	unsigned long ram_size = 64UL << 20;
+	bool enable_virtio = false;
 	bool single_step = false;
 	int i;
 
@@ -72,6 +73,9 @@ int main(int argc, char *argv[])
 		} else if (option_matches(argv[i], "--single-step")) {
 			single_step	= true;
 			continue;
+		} else if (option_matches(argv[i], "--enable-virtio")) {
+			enable_virtio	= true;
+			continue;
 		} else if (option_matches(argv[i], "--mem=")) {
 			unsigned long val = atol(&argv[i][6]) << 20;
 			if (val < ram_size)
@@ -117,7 +121,9 @@ int main(int argc, char *argv[])
 
 	early_printk__init();
 	pci__init();
-	blk_virtio__init();
+
+	if (enable_virtio)
+		blk_virtio__init();
 
 	for (;;) {
 		kvm__run(kvm);