Linux Virtualization - Feb 2015 - [RFC 0/2] virtio_pci: patches never to apply.

This should allow testing when QEMU gets VIRTIO_PCI_CAP_PCI_CFG support,
but I'm pretty sure we should never allow these patches upstream.

Tested with lguest (in virtio-next), which supports VIRTIO_PCI_CAP_PCI_CFG.

Rusty Russell (2):
  virtio_pci: abstract all MMIO accesses.
  virtio: Introducing virtio_pci.no_mmio, the worst boot option in
    history.

 drivers/virtio/virtio_pci_common.c |  70 +++++++-
 drivers/virtio/virtio_pci_common.h |  62 ++++++++
 drivers/virtio/virtio_pci_legacy.c |   1 +
 drivers/virtio/virtio_pci_modern.c | 317 ++++++++++++++++++++++++++++++-------
 4 files changed, 391 insertions(+), 59 deletions(-)

-- 
2.1.0

This is in preparation for testing the virtio pci config mmio backdoor.

Signed-off-by: Rusty Russell <rusty at rustcorp.com.au>
---
 drivers/virtio/virtio_pci_common.c |  47 ++++++++++++-
 drivers/virtio/virtio_pci_common.h |  41 +++++++++++
 drivers/virtio/virtio_pci_modern.c | 140 ++++++++++++++++++++++++-------------
 3 files changed, 177 insertions(+), 51 deletions(-)

diff --git a/drivers/virtio/virtio_pci_common.c
b/drivers/virtio/virtio_pci_common.c
index e894eb278d83..4e6132dd0ca3 100644
--- a/drivers/virtio/virtio_pci_common.c
+++ b/drivers/virtio/virtio_pci_common.c
@@ -27,6 +27,49 @@ MODULE_PARM_DESC(force_legacy,
 		 "Force legacy mode for transitional virtio 1 devices");
 #endif
 
+u8 vp_read_isr(const struct virtio_pci_device *dev)
+{
+	return ioread8(dev->isr);
+}
+
+void vp_write_notify(const struct virtqueue *vq, u16 vqindex)
+{
+	iowrite16(vqindex, (void __iomem *)vq->priv);
+}
+
+u32 __vp_read_common32(const struct virtio_pci_device *dev, size_t offset)
+{
+	return ioread32((void __iomem *)dev->common + offset);
+}
+
+u16 __vp_read_common16(const struct virtio_pci_device *dev, size_t offset)
+{
+	return ioread16((void __iomem *)dev->common + offset);
+}
+
+u8 __vp_read_common8(const struct virtio_pci_device *dev, size_t offset)
+{
+	return ioread8((void __iomem *)dev->common + offset);
+}
+
+void __vp_write_common32(const struct virtio_pci_device *dev,
+			 size_t offset, u32 val)
+{
+	iowrite32(val, (void __iomem *)dev->common + offset);
+}
+
+void __vp_write_common16(const struct virtio_pci_device *dev,
+			 size_t offset, u16 val)
+{
+	iowrite16(val, (void __iomem *)dev->common + offset);
+}
+
+void __vp_write_common8(const struct virtio_pci_device *dev,
+			size_t offset, u8 val)
+{
+	iowrite8(val, (void __iomem *)dev->common + offset);
+}
+
 /* wait for pending irq handlers */
 void vp_synchronize_vectors(struct virtio_device *vdev)
 {
@@ -45,7 +88,7 @@ bool vp_notify(struct virtqueue *vq)
 {
 	/* we write the queue's selector into the notification register to
 	 * signal the other end */
-	iowrite16(vq->index, (void __iomem *)vq->priv);
+	vp_write_notify(vq, vq->index);
 	return true;
 }
 
@@ -89,7 +132,7 @@ static irqreturn_t vp_interrupt(int irq, void *opaque)
 
 	/* reading the ISR has the effect of also clearing it so it's very
 	 * important to save off the value. */
-	isr = ioread8(vp_dev->isr);
+	isr = vp_read_isr(vp_dev);
 
 	/* It's definitely not us if the ISR was not high */
 	if (!isr)
diff --git a/drivers/virtio/virtio_pci_common.h
b/drivers/virtio/virtio_pci_common.h
index 28ee4e56badf..15a20c968ae7 100644
--- a/drivers/virtio/virtio_pci_common.h
+++ b/drivers/virtio/virtio_pci_common.h
@@ -113,6 +113,47 @@ struct virtio_pci_device {
 	u16 (*config_vector)(struct virtio_pci_device *vp_dev, u16 vector);
 };
 
+/* Accessor functions. */
+u8 vp_read_isr(const struct virtio_pci_device *dev);
+void vp_write_notify(const struct virtqueue *vq, u16 vqindex);
+u32 __vp_read_common32(const struct virtio_pci_device *dev, size_t offset);
+u16 __vp_read_common16(const struct virtio_pci_device *dev, size_t offset);
+u8 __vp_read_common8(const struct virtio_pci_device *dev, size_t offset);
+void __vp_write_common32(const struct virtio_pci_device *dev,
+			 size_t offset, u32 val);
+void __vp_write_common16(const struct virtio_pci_device *dev,
+			 size_t offset, u16 val);
+void __vp_write_common8(const struct virtio_pci_device *dev,
+			size_t offset, u8 val);
+
+#define vp_read_common32(dev, fieldname)				\
+	__vp_read_common32((dev),					\
+		   offsetof(struct virtio_pci_common_cfg, fieldname) + \
+		   BUILD_BUG_ON_ZERO(sizeof((dev)->common->fieldname) != 4))
+#define vp_read_common16(dev, fieldname)				\
+	__vp_read_common16((dev),					\
+		offsetof(struct virtio_pci_common_cfg, fieldname) +	\
+		BUILD_BUG_ON_ZERO(sizeof((dev)->common->fieldname) != 2))
+#define vp_read_common8(dev, fieldname)				\
+	__vp_read_common8((dev),					\
+		offsetof(struct virtio_pci_common_cfg, fieldname) +	\
+		BUILD_BUG_ON_ZERO(sizeof((dev)->common->fieldname) != 1))
+#define vp_write_common32(dev, fieldname, val)				\
+	__vp_write_common32((dev),					\
+		   offsetof(struct virtio_pci_common_cfg, fieldname) + \
+		   BUILD_BUG_ON_ZERO(sizeof((dev)->common->fieldname) != 4), \
+		   (val))
+#define vp_write_common16(dev, fieldname, val)				\
+	__vp_write_common16((dev),					\
+		   offsetof(struct virtio_pci_common_cfg, fieldname) + \
+		   BUILD_BUG_ON_ZERO(sizeof((dev)->common->fieldname) != 2), \
+		   (val))
+#define vp_write_common8(dev, fieldname, val)				\
+	__vp_write_common8((dev),					\
+		   offsetof(struct virtio_pci_common_cfg, fieldname) + \
+		   BUILD_BUG_ON_ZERO(sizeof((dev)->common->fieldname) != 1), \
+		   (val))
+
 /* Constants for MSI-X */
 /* Use first vector for configuration changes, second and the rest for
  * virtqueues Thus, we need at least 2 vectors for MSI. */
diff --git a/drivers/virtio/virtio_pci_modern.c
b/drivers/virtio/virtio_pci_modern.c
index 2aa38e59db2e..daa990ef3df0 100644
--- a/drivers/virtio/virtio_pci_modern.c
+++ b/drivers/virtio/virtio_pci_modern.c
@@ -94,22 +94,29 @@ static void __iomem *map_capability(struct pci_dev *dev, int
off,
 	return p;
 }
 
-static void iowrite64_twopart(u64 val, __le32 __iomem *lo, __le32 __iomem *hi)
+static void __vp_write_common64(const struct virtio_pci_device *dev,
+				size_t off_lo, size_t off_hi, u64 val)
 {
-	iowrite32((u32)val, lo);
-	iowrite32(val >> 32, hi);
+	__vp_write_common32(dev, off_lo, val);
+	__vp_write_common32(dev, off_hi, val >> 32);
 }
 
+#define vp_write_common64(dev, name_lo, name_hi, val)			\
+	__vp_write_common64((dev),					\
+			    offsetof(struct virtio_pci_common_cfg, name_lo), \
+			    offsetof(struct virtio_pci_common_cfg, name_hi), \
+			    (val))
+
 /* virtio config->get_features() implementation */
 static u64 vp_get_features(struct virtio_device *vdev)
 {
 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 	u64 features;
 
-	iowrite32(0, &vp_dev->common->device_feature_select);
-	features = ioread32(&vp_dev->common->device_feature);
-	iowrite32(1, &vp_dev->common->device_feature_select);
-	features |= ((u64)ioread32(&vp_dev->common->device_feature) <<
32);
+	vp_write_common32(vp_dev, device_feature_select, 0);
+	features = vp_read_common32(vp_dev, device_feature);
+	vp_write_common32(vp_dev, device_feature_select, 1);
+	features |= ((u64)vp_read_common32(vp_dev, device_feature) << 32);
 
 	return features;
 }
@@ -128,13 +135,49 @@ static int vp_finalize_features(struct virtio_device
*vdev)
 		return -EINVAL;
 	}
 
-	iowrite32(0, &vp_dev->common->guest_feature_select);
-	iowrite32((u32)vdev->features, &vp_dev->common->guest_feature);
-	iowrite32(1, &vp_dev->common->guest_feature_select);
-	iowrite32(vdev->features >> 32,
&vp_dev->common->guest_feature);
+	vp_write_common32(vp_dev, guest_feature_select, 0);
+	vp_write_common32(vp_dev, guest_feature, (u32)vdev->features);
+	vp_write_common32(vp_dev, guest_feature_select, 1);
+	vp_write_common32(vp_dev, guest_feature, vdev->features >> 32);
 
 	return 0;
 }
+			   
+static u32 vp_read_device32(const struct virtio_pci_device *dev,
+			    size_t offset)
+{
+	return ioread32((void __iomem *)dev->device + offset);
+}
+
+static u16 vp_read_device16(const struct virtio_pci_device *dev,
+			    size_t offset)
+{
+	return ioread16((void __iomem *)dev->device + offset);
+}
+
+static u8 vp_read_device8(const struct virtio_pci_device *dev,
+			  size_t offset)
+{
+	return ioread8((void __iomem *)dev->device + offset);
+}
+
+static void vp_write_device32(const struct virtio_pci_device *dev,
+			      size_t offset, u32 val)
+{
+	iowrite32(val, (void __iomem *)dev->device + offset);
+}
+
+static void vp_write_device16(const struct virtio_pci_device *dev,
+			      size_t offset, u16 val)
+{
+	iowrite16(val, (void __iomem *)dev->device + offset);
+}
+
+static void vp_write_device8(const struct virtio_pci_device *dev,
+			      size_t offset, u8 val)
+{
+	iowrite8(val, (void __iomem *)dev->device + offset);
+}
 
 /* virtio config->get() implementation */
 static void vp_get(struct virtio_device *vdev, unsigned offset,
@@ -149,21 +192,21 @@ static void vp_get(struct virtio_device *vdev, unsigned
offset,
 
 	switch (len) {
 	case 1:
-		b = ioread8(vp_dev->device + offset);
+		b = vp_read_device8(vp_dev, offset);
 		memcpy(buf, &b, sizeof b);
 		break;
 	case 2:
-		w = cpu_to_le16(ioread16(vp_dev->device + offset));
+		w = cpu_to_le16(vp_read_device16(vp_dev, offset));
 		memcpy(buf, &w, sizeof w);
 		break;
 	case 4:
-		l = cpu_to_le32(ioread32(vp_dev->device + offset));
+		l = cpu_to_le32(vp_read_device32(vp_dev, offset));
 		memcpy(buf, &l, sizeof l);
 		break;
 	case 8:
-		l = cpu_to_le32(ioread32(vp_dev->device + offset));
+		l = cpu_to_le32(vp_read_device32(vp_dev, offset));
 		memcpy(buf, &l, sizeof l);
-		l = cpu_to_le32(ioread32(vp_dev->device + offset + sizeof l));
+		l = cpu_to_le32(vp_read_device32(vp_dev, offset + sizeof l));
 		memcpy(buf + sizeof l, &l, sizeof l);
 		break;
 	default:
@@ -186,21 +229,21 @@ static void vp_set(struct virtio_device *vdev, unsigned
offset,
 	switch (len) {
 	case 1:
 		memcpy(&b, buf, sizeof b);
-		iowrite8(b, vp_dev->device + offset);
+		vp_write_device8(vp_dev, offset, b);
 		break;
 	case 2:
 		memcpy(&w, buf, sizeof w);
-		iowrite16(le16_to_cpu(w), vp_dev->device + offset);
+		vp_write_device16(vp_dev, offset, le16_to_cpu(w));
 		break;
 	case 4:
 		memcpy(&l, buf, sizeof l);
-		iowrite32(le32_to_cpu(l), vp_dev->device + offset);
+		vp_write_device32(vp_dev, offset, le32_to_cpu(l));
 		break;
 	case 8:
 		memcpy(&l, buf, sizeof l);
-		iowrite32(le32_to_cpu(l), vp_dev->device + offset);
+		vp_write_device32(vp_dev, offset, le32_to_cpu(l));
 		memcpy(&l, buf + sizeof l, sizeof l);
-		iowrite32(le32_to_cpu(l), vp_dev->device + offset + sizeof l);
+		vp_write_device32(vp_dev, offset + sizeof l, le32_to_cpu(l));
 		break;
 	default:
 		BUG();
@@ -210,14 +253,14 @@ static void vp_set(struct virtio_device *vdev, unsigned
offset,
 static u32 vp_generation(struct virtio_device *vdev)
 {
 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
-	return ioread8(&vp_dev->common->config_generation);
+	return vp_read_common8(vp_dev, config_generation);
 }
 
 /* config->{get,set}_status() implementations */
 static u8 vp_get_status(struct virtio_device *vdev)
 {
 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
-	return ioread8(&vp_dev->common->device_status);
+	return vp_read_common8(vp_dev, device_status);
 }
 
 static void vp_set_status(struct virtio_device *vdev, u8 status)
@@ -225,17 +268,17 @@ static void vp_set_status(struct virtio_device *vdev, u8
status)
 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 	/* We should never be setting status to 0. */
 	BUG_ON(status == 0);
-	iowrite8(status, &vp_dev->common->device_status);
+	vp_write_common8(vp_dev, device_status, status);
 }
 
 static void vp_reset(struct virtio_device *vdev)
 {
 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 	/* 0 status means a reset. */
-	iowrite8(0, &vp_dev->common->device_status);
+	vp_write_common8(vp_dev, device_status, 0);
 	/* Flush out the status write, and flush in device writes,
 	 * including MSI-X interrupts, if any. */
-	ioread8(&vp_dev->common->device_status);
+	vp_read_common8(vp_dev, device_status);
 	/* Flush pending VQ/configuration callbacks. */
 	vp_synchronize_vectors(vdev);
 }
@@ -243,10 +286,10 @@ static void vp_reset(struct virtio_device *vdev)
 static u16 vp_config_vector(struct virtio_pci_device *vp_dev, u16 vector)
 {
 	/* Setup the vector used for configuration events */
-	iowrite16(vector, &vp_dev->common->msix_config);
+	vp_write_common16(vp_dev, msix_config, vector);
 	/* Verify we had enough resources to assign the vector */
 	/* Will also flush the write out to device */
-	return ioread16(&vp_dev->common->msix_config);
+	return vp_read_common16(vp_dev, msix_config);
 }
 
 static size_t vring_pci_size(u16 num)
@@ -281,20 +324,19 @@ static struct virtqueue *setup_vq(struct virtio_pci_device
*vp_dev,
 				  const char *name,
 				  u16 msix_vec)
 {
-	struct virtio_pci_common_cfg __iomem *cfg = vp_dev->common;
 	struct virtqueue *vq;
 	u16 num, off;
 	int err;
 
-	if (index >= ioread16(&cfg->num_queues))
+	if (index >= vp_read_common16(vp_dev, num_queues))
 		return ERR_PTR(-ENOENT);
 
 	/* Select the queue we're interested in */
-	iowrite16(index, &cfg->queue_select);
+	vp_write_common16(vp_dev, queue_select, index);
 
 	/* Check if queue is either not available or already active. */
-	num = ioread16(&cfg->queue_size);
-	if (!num || ioread16(&cfg->queue_enable))
+	num = vp_read_common16(vp_dev, queue_size);
+	if (!num || vp_read_common16(vp_dev, queue_enable))
 		return ERR_PTR(-ENOENT);
 
 	if (num & (num - 1)) {
@@ -303,7 +345,7 @@ static struct virtqueue *setup_vq(struct virtio_pci_device
*vp_dev,
 	}
 
 	/* get offset of notification word for this vq */
-	off = ioread16(&cfg->queue_notify_off);
+	off = vp_read_common16(vp_dev, queue_notify_off);
 
 	info->num = num;
 	info->msix_vector = msix_vec;
@@ -322,13 +364,13 @@ static struct virtqueue *setup_vq(struct virtio_pci_device
*vp_dev,
 	}
 
 	/* activate the queue */
-	iowrite16(num, &cfg->queue_size);
-	iowrite64_twopart(virt_to_phys(info->queue),
-			  &cfg->queue_desc_lo, &cfg->queue_desc_hi);
-	iowrite64_twopart(virt_to_phys(virtqueue_get_avail(vq)),
-			  &cfg->queue_avail_lo, &cfg->queue_avail_hi);
-	iowrite64_twopart(virt_to_phys(virtqueue_get_used(vq)),
-			  &cfg->queue_used_lo, &cfg->queue_used_hi);
+	vp_write_common16(vp_dev, queue_size, num);
+	vp_write_common64(vp_dev, queue_desc_lo, queue_desc_hi,
+			  virt_to_phys(info->queue));
+	vp_write_common64(vp_dev, queue_avail_lo, queue_avail_hi,
+			  virt_to_phys(virtqueue_get_avail(vq)));
+	vp_write_common64(vp_dev, queue_used_lo, queue_used_hi,
+			  virt_to_phys(virtqueue_get_used(vq)));
 
 	if (vp_dev->notify_base) {
 		/* offset should not wrap */
@@ -357,8 +399,8 @@ static struct virtqueue *setup_vq(struct virtio_pci_device
*vp_dev,
 	}
 
 	if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
-		iowrite16(msix_vec, &cfg->queue_msix_vector);
-		msix_vec = ioread16(&cfg->queue_msix_vector);
+		vp_write_common16(vp_dev, queue_msix_vector, msix_vec);
+		msix_vec = vp_read_common16(vp_dev, queue_msix_vector);
 		if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
 			err = -EBUSY;
 			goto err_assign_vector;
@@ -393,8 +435,8 @@ static int vp_modern_find_vqs(struct virtio_device *vdev,
unsigned nvqs,
 	 * this, there's no way to go back except reset.
 	 */
 	list_for_each_entry(vq, &vdev->vqs, list) {
-		iowrite16(vq->index, &vp_dev->common->queue_select);
-		iowrite16(1, &vp_dev->common->queue_enable);
+		vp_write_common16(vp_dev, queue_select, vq->index);
+		vp_write_common16(vp_dev, queue_enable, 1);
 	}
 
 	return 0;
@@ -405,13 +447,13 @@ static void del_vq(struct virtio_pci_vq_info *info)
 	struct virtqueue *vq = info->vq;
 	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 
-	iowrite16(vq->index, &vp_dev->common->queue_select);
+	vp_write_common16(vp_dev, queue_select, vq->index);
 
 	if (vp_dev->msix_enabled) {
-		iowrite16(VIRTIO_MSI_NO_VECTOR,
-			  &vp_dev->common->queue_msix_vector);
+		vp_write_common16(vp_dev, queue_msix_vector,
+				  VIRTIO_MSI_NO_VECTOR);
 		/* Flush the write out to device */
-		ioread16(&vp_dev->common->queue_msix_vector);
+		vp_read_common16(vp_dev, queue_msix_vector);
 	}
 
 	if (!vp_dev->notify_base)
-- 
2.1.0

Sometimes, devices are just too damn fast.  Wouldn't it be nice if we
could (1) have an option to access them in the most baroque way
possible, and (2) ensure that even the normal case caused extra bloat?

Well, wish no longer: This patch does exactly that!  Since every
complaint virtio 1.0 device has to have a capability to allow backdoor
access into the BARs, we can use that:

	4.1.4.7.1 Device Requirements: PCI configuration access capability

	The device MUST present at least one VIRTIO_PCI_CAP_PCI_CFG
	capability.

	Upon detecting driver write access to pci_cfg_data, the device
	MUST execute a write access at offset cap.offset at BAR
	selected by cap.bar using the first cap.length bytes from
	pci_cfg_data.

	Upon detecting driver read access to pci_cfg_data, the device
	MUST execute a read access of length cap.length at offset
	cap.offset at BAR selected by cap.bar and store the first
	cap.length bytes in pci_cfg_data.

Signed-off-by: Rusty Russell <rusty at rustcorp.com.au>
---
 drivers/virtio/virtio_pci_common.c |  31 ++++++-
 drivers/virtio/virtio_pci_common.h |  21 +++++
 drivers/virtio/virtio_pci_legacy.c |   1 +
 drivers/virtio/virtio_pci_modern.c | 183 ++++++++++++++++++++++++++++++++++---
 4 files changed, 221 insertions(+), 15 deletions(-)

diff --git a/drivers/virtio/virtio_pci_common.c
b/drivers/virtio/virtio_pci_common.c
index 4e6132dd0ca3..0030180411cc 100644
--- a/drivers/virtio/virtio_pci_common.c
+++ b/drivers/virtio/virtio_pci_common.c
@@ -29,45 +29,68 @@ MODULE_PARM_DESC(force_legacy,
 
 u8 vp_read_isr(const struct virtio_pci_device *dev)
 {
+	if (dev->no_mmio)
+		return dev->no_mmio->read8(dev, dev->isr, 0);
 	return ioread8(dev->isr);
 }
 
 void vp_write_notify(const struct virtqueue *vq, u16 vqindex)
 {
-	iowrite16(vqindex, (void __iomem *)vq->priv);
+	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
+
+	if (vp_dev->no_mmio)
+		vp_dev->no_mmio->write16(vp_dev, (void __iomem *)vq->priv,
+					 0, vqindex);
+	else
+		iowrite16(vqindex, (void __iomem *)vq->priv);
 }
 
 u32 __vp_read_common32(const struct virtio_pci_device *dev, size_t offset)
 {
+	if (dev->no_mmio)
+		return dev->no_mmio->read32(dev, dev->common, offset);
 	return ioread32((void __iomem *)dev->common + offset);
 }
 
 u16 __vp_read_common16(const struct virtio_pci_device *dev, size_t offset)
 {
+	if (dev->no_mmio)
+		return dev->no_mmio->read16(dev, dev->common, offset);
 	return ioread16((void __iomem *)dev->common + offset);
 }
 
 u8 __vp_read_common8(const struct virtio_pci_device *dev, size_t offset)
 {
+	if (dev->no_mmio)
+		return dev->no_mmio->read8(dev, dev->common, offset);
 	return ioread8((void __iomem *)dev->common + offset);
 }
 
 void __vp_write_common32(const struct virtio_pci_device *dev,
 			 size_t offset, u32 val)
 {
-	iowrite32(val, (void __iomem *)dev->common + offset);
+	if (dev->no_mmio)
+		dev->no_mmio->write32(dev, dev->common, offset, val);
+	else
+		iowrite32(val, (void __iomem *)dev->common + offset);
 }
 
 void __vp_write_common16(const struct virtio_pci_device *dev,
 			 size_t offset, u16 val)
 {
-	iowrite16(val, (void __iomem *)dev->common + offset);
+	if (dev->no_mmio)
+		dev->no_mmio->write16(dev, dev->common, offset, val);
+	else
+		iowrite16(val, (void __iomem *)dev->common + offset);
 }
 
 void __vp_write_common8(const struct virtio_pci_device *dev,
 			size_t offset, u8 val)
 {
-	iowrite8(val, (void __iomem *)dev->common + offset);
+	if (dev->no_mmio)
+		dev->no_mmio->write8(dev, dev->common, offset, val);
+	else
+		iowrite8(val, (void __iomem *)dev->common + offset);
 }
 
 /* wait for pending irq handlers */
diff --git a/drivers/virtio/virtio_pci_common.h
b/drivers/virtio/virtio_pci_common.h
index 15a20c968ae7..d891e3123cdd 100644
--- a/drivers/virtio/virtio_pci_common.h
+++ b/drivers/virtio/virtio_pci_common.h
@@ -53,6 +53,9 @@ struct virtio_pci_device {
 	struct virtio_device vdev;
 	struct pci_dev *pci_dev;
 
+	/* This is only valid for modern devices. */
+	const struct virtio_pci_no_mmio_ops *no_mmio;
+
 	/* In legacy mode, these two point to within ->legacy. */
 	/* Where to read and clear interrupt */
 	u8 __iomem *isr;
@@ -75,6 +78,9 @@ struct virtio_pci_device {
 	/* Multiply queue_notify_off by this value. (non-legacy mode). */
 	u32 notify_offset_multiplier;
 
+	/* PCI config window for BAR access (non-legacy mode). */
+	int window;
+
 	/* Legacy only field */
 	/* the IO mapping for the PCI config space */
 	void __iomem *ioaddr;
@@ -113,6 +119,21 @@ struct virtio_pci_device {
 	u16 (*config_vector)(struct virtio_pci_device *vp_dev, u16 vector);
 };
 
+struct virtio_pci_no_mmio_ops {
+	u8 (*read8)(const struct virtio_pci_device *,
+		    const void __iomem *, size_t);
+	u16 (*read16)(const struct virtio_pci_device *,
+		      const void __iomem *, size_t);
+	u32 (*read32)(const struct virtio_pci_device *,
+		      const void __iomem *, size_t);
+	void (*write8)(const struct virtio_pci_device *,
+		       const void __iomem *, size_t, u8);
+	void (*write16)(const struct virtio_pci_device *,
+			const void __iomem *, size_t, u16);
+	void (*write32)(const struct virtio_pci_device *,
+			const void __iomem *, size_t, u32);
+};
+
 /* Accessor functions. */
 u8 vp_read_isr(const struct virtio_pci_device *dev);
 void vp_write_notify(const struct virtqueue *vq, u16 vqindex);
diff --git a/drivers/virtio/virtio_pci_legacy.c
b/drivers/virtio/virtio_pci_legacy.c
index 256a5278a515..26aa036fc494 100644
--- a/drivers/virtio/virtio_pci_legacy.c
+++ b/drivers/virtio/virtio_pci_legacy.c
@@ -226,6 +226,7 @@ int virtio_pci_legacy_probe(struct virtio_pci_device
*vp_dev)
 		return -ENODEV;
 	}
 
+	vp_dev->no_mmio = NULL;
 	vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
 	if (!vp_dev->ioaddr)
 		return -ENOMEM;
diff --git a/drivers/virtio/virtio_pci_modern.c
b/drivers/virtio/virtio_pci_modern.c
index daa990ef3df0..76882b467519 100644
--- a/drivers/virtio/virtio_pci_modern.c
+++ b/drivers/virtio/virtio_pci_modern.c
@@ -20,6 +20,125 @@
 #define VIRTIO_PCI_NO_LEGACY
 #include "virtio_pci_common.h"
 
+static bool no_mmio = false; /* Very, very false.  Never set this! */
+module_param(no_mmio, bool, 0644);
+
+/* We jam bar and offset into the pointer for non-MMIO */
+static u8 bar_off_from_pointer(const void *ptr, u32 *offset)
+{
+	*offset = (unsigned long)ptr >> 3;
+	return ((unsigned long)ptr & 0x7) - 1;
+}
+
+static void *bar_off_to_pointer(u8 bar, u32 offset)
+{
+	void *p;
+	u32 check_offset;
+
+	BUG_ON(bar >= 7);
+	p = (void *)((bar + 1) | (offset << 3));
+	if (bar_off_from_pointer(p, &check_offset) != bar
+	    || check_offset != offset) {
+		printk(KERN_ERR "Cannot pack bar %u offset %u into pointer!\n",
+		       bar, offset);
+		return NULL;
+	}
+	return p;
+}
+
+/* This sets up the PCI config space window ready for an access. */
+static void set_cfg_window(const struct virtio_pci_device *dev,
+			   const void __iomem *ptr, size_t offset, size_t len)
+{
+	u32 bar_offset;
+	u8 bar;
+
+	BUG_ON(!dev->no_mmio);
+
+	bar = bar_off_from_pointer(ptr, &bar_offset);
+
+	pci_write_config_byte(dev->pci_dev,
+			      dev->window + offsetof(struct virtio_pci_cap, bar),
+			      bar);
+	pci_write_config_dword(dev->pci_dev,
+			       dev->window
+			       + offsetof(struct virtio_pci_cap, length),
+			       len);
+	pci_write_config_dword(dev->pci_dev,
+			       dev->window
+			       + offsetof(struct virtio_pci_cap, offset),
+			       bar_offset + offset);
+}
+
+static u32 vp_window_read32(const struct virtio_pci_device *dev,
+			    const void __iomem *ptr,
+			    size_t offset)
+{
+	u32 ret = -1;
+	set_cfg_window(dev, ptr, offset, sizeof(ret));
+	pci_read_config_dword(dev->pci_dev,
+			      dev->window + sizeof(struct virtio_pci_cap), &ret);
+	return ret;
+}
+
+static u16 vp_window_read16(const struct virtio_pci_device *dev,
+			    const void __iomem *ptr,
+			    size_t offset)
+{
+	u16 ret = -1;
+	set_cfg_window(dev, ptr, offset, sizeof(ret));
+	pci_read_config_word(dev->pci_dev,
+			     dev->window + sizeof(struct virtio_pci_cap), &ret);
+	return ret;
+}
+
+static u8 vp_window_read8(const struct virtio_pci_device *dev,
+			  const void __iomem *ptr,
+			  size_t offset)
+{
+	u8 ret = -1;
+	set_cfg_window(dev, ptr, offset, sizeof(ret));
+	pci_read_config_byte(dev->pci_dev,
+			     dev->window + sizeof(struct virtio_pci_cap), &ret);
+	return ret;
+}
+
+static void vp_window_write32(const struct virtio_pci_device *dev,
+			      const void __iomem *ptr, size_t offset, u32 val)
+{
+	set_cfg_window(dev, ptr, offset, sizeof(val));
+	pci_write_config_dword(dev->pci_dev,
+			       dev->window + sizeof(struct virtio_pci_cap),
+			       val);
+}
+
+static void vp_window_write16(const struct virtio_pci_device *dev,
+			      const void __iomem *ptr, size_t offset, u16 val)
+{
+	set_cfg_window(dev, ptr, offset, sizeof(val));
+	pci_write_config_word(dev->pci_dev,
+			      dev->window + sizeof(struct virtio_pci_cap),
+			      val);
+}
+
+static void vp_window_write8(const struct virtio_pci_device *dev,
+			      const void __iomem *ptr, size_t offset, u8 val)
+{
+	set_cfg_window(dev, ptr, offset, sizeof(val));
+	pci_write_config_byte(dev->pci_dev,
+			      dev->window + sizeof(struct virtio_pci_cap),
+			      val);
+}
+
+static const struct virtio_pci_no_mmio_ops vp_window_ops = {
+	.read8 = vp_window_read8,
+	.read16 = vp_window_read16,
+	.read32 = vp_window_read32,
+	.write8 = vp_window_write8,
+	.write16 = vp_window_write16,
+	.write32 = vp_window_write32,
+};
+
 static void __iomem *map_capability(struct pci_dev *dev, int off,
 				    size_t minlen,
 				    u32 align,
@@ -86,6 +205,10 @@ static void __iomem *map_capability(struct pci_dev *dev, int
off,
 		return NULL;
 	}
 
+	/* We encode bar and offset into the pointer for the no-mmio case. */
+	if (no_mmio)
+		return bar_off_to_pointer(bar, offset);
+
 	p = pci_iomap_range(dev, bar, offset, length);
 	if (!p)
 		dev_err(&dev->dev,
@@ -146,37 +269,52 @@ static int vp_finalize_features(struct virtio_device
*vdev)
 static u32 vp_read_device32(const struct virtio_pci_device *dev,
 			    size_t offset)
 {
+	if (dev->no_mmio)
+		return vp_window_read32(dev, dev->device, offset);
 	return ioread32((void __iomem *)dev->device + offset);
 }
 
 static u16 vp_read_device16(const struct virtio_pci_device *dev,
 			    size_t offset)
 {
+	if (dev->no_mmio)
+		return vp_window_read16(dev, dev->device, offset);
 	return ioread16((void __iomem *)dev->device + offset);
 }
 
 static u8 vp_read_device8(const struct virtio_pci_device *dev,
 			  size_t offset)
 {
+	if (dev->no_mmio)
+		return vp_window_read8(dev, dev->device, offset);
 	return ioread8((void __iomem *)dev->device + offset);
 }
 
 static void vp_write_device32(const struct virtio_pci_device *dev,
 			      size_t offset, u32 val)
 {
-	iowrite32(val, (void __iomem *)dev->device + offset);
+	if (dev->no_mmio)
+		vp_window_write32(dev, dev->device, offset, val);
+	else
+		iowrite32(val, (void __iomem *)dev->device + offset);
 }
 
 static void vp_write_device16(const struct virtio_pci_device *dev,
 			      size_t offset, u16 val)
 {
-	iowrite16(val, (void __iomem *)dev->device + offset);
+	if (dev->no_mmio)
+		vp_window_write16(dev, dev->device, offset, val);
+	else
+		iowrite16(val, (void __iomem *)dev->device + offset);
 }
 
 static void vp_write_device8(const struct virtio_pci_device *dev,
 			      size_t offset, u8 val)
 {
-	iowrite8(val, (void __iomem *)dev->device + offset);
+	if (dev->no_mmio)
+		vp_window_write8(dev, dev->device, offset, val);
+	else
+		iowrite8(val, (void __iomem *)dev->device + offset);
 }
 
 /* virtio config->get() implementation */
@@ -494,6 +632,12 @@ static const struct virtio_config_ops virtio_pci_config_ops
= {
 	.set_vq_affinity = vp_set_vq_affinity,
 };
 
+static void vp_iounmap(struct virtio_pci_device *dev, void __iomem *p)
+{
+	if (!dev->no_mmio)
+		pci_iounmap(dev->pci_dev, p);
+}
+
 /**
  * virtio_pci_find_capability - walk capabilities to find device info.
  * @dev: the pci device
@@ -598,7 +742,7 @@ static inline void check_offsets(void)
 int virtio_pci_modern_probe(struct virtio_pci_device *vp_dev)
 {
 	struct pci_dev *pci_dev = vp_dev->pci_dev;
-	int err, common, isr, notify, device;
+	int err, common, isr, notify, device, window;
 	u32 notify_length;
 	u32 notify_offset;
 
@@ -631,6 +775,13 @@ int virtio_pci_modern_probe(struct virtio_pci_device
*vp_dev)
 		return -ENODEV;
 	}
 
+	if (no_mmio) {
+		vp_dev->no_mmio = &vp_window_ops;
+		dev_warn(&pci_dev->dev,
+			 "virtio_pci: using SLOW non-MMIO access\n");
+ 	} else
+		vp_dev->no_mmio = NULL;
+
 	/* If common is there, these should be too... */
 	isr = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_ISR_CFG,
 					 IORESOURCE_IO | IORESOURCE_MEM);
@@ -643,6 +794,16 @@ int virtio_pci_modern_probe(struct virtio_pci_device
*vp_dev)
 		return -EINVAL;
 	}
 
+	vp_dev->window = virtio_pci_find_capability(pci_dev,
+						    VIRTIO_PCI_CAP_PCI_CFG,
+						    IORESOURCE_IO |
+						    IORESOURCE_MEM);
+	if (!vp_dev->window) {
+		dev_err(&pci_dev->dev,
+			"virtio_pci: missing pci window capability\n");
+		return -EINVAL;
+	}
+
 	/* Device capability is only mandatory for devices that have
 	 * device-specific configuration.
 	 */
@@ -715,11 +876,11 @@ int virtio_pci_modern_probe(struct virtio_pci_device
*vp_dev)
 
 err_map_device:
 	if (vp_dev->notify_base)
-		pci_iounmap(pci_dev, vp_dev->notify_base);
+		vp_iounmap(vp_dev, vp_dev->notify_base);
 err_map_notify:
-	pci_iounmap(pci_dev, vp_dev->isr);
+	vp_iounmap(vp_dev, vp_dev->isr);
 err_map_isr:
-	pci_iounmap(pci_dev, vp_dev->common);
+	vp_iounmap(vp_dev, vp_dev->common);
 err_map_common:
 	return err;
 }
@@ -729,9 +890,9 @@ void virtio_pci_modern_remove(struct virtio_pci_device
*vp_dev)
 	struct pci_dev *pci_dev = vp_dev->pci_dev;
 
 	if (vp_dev->device)
-		pci_iounmap(pci_dev, vp_dev->device);
+		vp_iounmap(vp_dev, vp_dev->device);
 	if (vp_dev->notify_base)
-		pci_iounmap(pci_dev, vp_dev->notify_base);
-	pci_iounmap(pci_dev, vp_dev->isr);
-	pci_iounmap(pci_dev, vp_dev->common);
+		vp_iounmap(vp_dev, vp_dev->notify_base);
+	vp_iounmap(vp_dev, vp_dev->isr);
+	vp_iounmap(vp_dev, vp_dev->common);
 }
-- 
2.1.0