Skip to contentSkip to conversation

rework on the IRQ hardening of virtio

PATCH In review Jason Wang 9 participants 55 messages May 27, 2022 – Jul 5, 2022

Subsystems (5)
lore
Share & tools
lore

Jason Wang

author
#0 lore raw

v6 cover rework on the IRQ hardening of virtio

Hi All:

This is a rework on the IRQ hardening for virtio which is done previously by the following commits are reverted:

9e35276a5344 ("virtio_pci: harden MSI-X interrupts") 080cd7c3ac87 ("virtio-pci: harden INTX interrupts")

The reason is that it depends on the IRQF_NO_AUTOEN which may conflict with the assumption of the affinity managed IRQ that is used by some virtio drivers. And what's more, it is only done for virtio-pci but not other transports.

In this rework, I try to implement a general virtio solution which borrows the idea of the INTX hardening by re-using per virtqueue boolean vq->broken and toggle it in virtio_device_ready() and virtio_reset_device(). Then we can simply reuse the existing checks in the vring_interrupt() and return early if the driver is not ready.

Note that, I only did compile test on ccw and MMIO transport.

Please review.

Changes since V5:

- Various tweaks on the comments

Changes since V4:

- use spin_lock_irq()/spin_unlock_irq() to synchronize with vring_interrupt() for ccw
- use spin_lock()/spin_unlock() to protect vring_interrupt() for non airq
- add comment to explain the ordering implications of set_status() for PCI, ccw and MMIO
- various tweaks on the comments and changelogs

Changes since V3:

- Rename synchornize_vqs() to synchronize_cbs()
- tweak the comment for synchronize_cbs()
- switch to use a dedicated helper __virtio_unbreak_device() and document it should be only used for probing
- switch to use rwlock to synchornize the non airq for ccw

Changes since V2:

- add ccw and MMIO support
- rename synchronize_vqs() to synchronize_cbs()
- switch to re-use vq->broken instead of introducing new device attributes for the future virtqueue reset support
- remove unnecssary READ_ONCE()/WRITE_ONCE()
- a new patch to remove device triggerable BUG_ON()
- more tweaks on the comments

Changes since v1:

- Use transport specific irq synchronization method when possible
- Drop the module parameter and enable the hardening unconditonally
- Tweak the barrier/ordering facilities used in the code
- Reanme irq_soft_enabled to driver_ready
- Avoid unnecssary IRQ synchornization (e.g during boot)

Jason Wang (8):
virtio: use virtio_reset_device() when possible
virtio: introduce config op to synchronize vring callbacks
virtio-pci: implement synchronize_cbs()
virtio-mmio: implement synchronize_cbs()
virtio-ccw: implement synchronize_cbs()
virtio: allow to unbreak virtqueue
virtio: harden vring IRQ
virtio: use WARN_ON() to warning illegal status value

Stefano Garzarella (1):
virtio: use virtio_device_ready() in virtio_device_restore()

 drivers/s390/virtio/virtio_ccw.c       | 34 +++++++++++++++++++
 drivers/virtio/virtio.c                | 24 +++++++++----
 drivers/virtio/virtio_mmio.c           | 13 +++++++
 drivers/virtio/virtio_pci_legacy.c     |  1 +
 drivers/virtio/virtio_pci_modern.c     |  2 ++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++
 drivers/virtio/virtio_ring.c           | 33 +++++++++++++++---
 include/linux/virtio.h                 |  1 +
 include/linux/virtio_config.h          | 47 +++++++++++++++++++++++++-
 9 files changed, 148 insertions(+), 12 deletions(-)
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
author
#1 lore raw

PATCH v6 1/9 virtio: use virtio_device_ready() in virtio_device_restore()

From: Stefano Garzarella <[email protected]>

It will allow us to do extension on virtio_device_ready() without duplicating code.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
 drivers/virtio/virtio.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/virtio/virtio.c +3 -2
index ce424c16997d..938e975029d4 100644
@@ -526,8 +526,9 @@ int virtio_device_restore(struct virtio_device *dev)
 			goto err;
 	}
 
-	/* Finally, tell the device we're all set */
-	virtio_add_status(dev, VIRTIO_CONFIG_S_DRIVER_OK);
+	/* If restore didn't do it, mark device DRIVER_OK ourselves. */
+	if (!(dev->config->get_status(dev) & VIRTIO_CONFIG_S_DRIVER_OK))
+		virtio_device_ready(dev);
 
 	virtio_config_enable(dev);
 
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
reviewer Reviewed-by on v6 1/9
#2 lore raw

Re: [PATCH V6 1/9] virtio: use virtio_device_ready() in virtio_device_restore()

Quoted 5 lines · Jason Wang

From: Stefano Garzarella <[email protected]>

It will allow us to do extension on virtio_device_ready() without duplicating code.

Quoted 12 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 21 lines
---
 drivers/virtio/virtio.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/virtio/virtio.c +3 -3
index ce424c16997d..938e975029d4 100644
@@ -526,8 +526,9 @@ int virtio_device_restore(struct virtio_device *dev)
 			goto err;
 	}
 
-	/* Finally, tell the device we're all set */
-	virtio_add_status(dev, VIRTIO_CONFIG_S_DRIVER_OK);
+	/* If restore didn't do it, mark device DRIVER_OK ourselves. */
+	if (!(dev->config->get_status(dev) & VIRTIO_CONFIG_S_DRIVER_OK))
+		virtio_device_ready(dev);
 
 	virtio_config_enable(dev);
 
--
Quoted 2 lines

2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

author
#3 lore raw

PATCH v6 2/9 virtio: use virtio_reset_device() when possible

In reply to Jason Wang

This allows us to do common extension without duplicating code.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
 drivers/virtio/virtio.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/virtio/virtio.c +2 -2
index 938e975029d4..aa1eb5132767 100644
@@ -430,7 +430,7 @@ int register_virtio_device(struct virtio_device *dev)
 
 	/* We always start by resetting the device, in case a previous
 	 * driver messed it up.  This also tests that code path a little. */
-	dev->config->reset(dev);
+	virtio_reset_device(dev);
 
 	/* Acknowledge that we've seen the device. */
 	virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
@@ -496,7 +496,7 @@ int virtio_device_restore(struct virtio_device *dev)
 
 	/* We always start by resetting the device, in case a previous
 	 * driver messed it up. */
-	dev->config->reset(dev);
+	virtio_reset_device(dev);
 
 	/* Acknowledge that we've seen the device. */
 	virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
reviewer Reviewed-by on v6 2/9
#4 lore raw

Re: [PATCH V6 2/9] virtio: use virtio_reset_device() when possible

Quoted 13 lines · Jason Wang

This allows us to do common extension without duplicating code.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 27 lines
---
 drivers/virtio/virtio.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/virtio/virtio.c +2 -3
index 938e975029d4..aa1eb5132767 100644
@@ -430,7 +430,7 @@ int register_virtio_device(struct virtio_device *dev)
 
 	/* We always start by resetting the device, in case a previous
 	 * driver messed it up.  This also tests that code path a little. */
-	dev->config->reset(dev);
+	virtio_reset_device(dev);
 
 	/* Acknowledge that we've seen the device. */
 	virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
@@ -496,7 +496,7 @@ int virtio_device_restore(struct virtio_device *dev)
 
 	/* We always start by resetting the device, in case a previous
 	 * driver messed it up. */
-	dev->config->reset(dev);
+	virtio_reset_device(dev);
 
 	/* Acknowledge that we've seen the device. */
 	virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
--
Quoted 2 lines

2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

maintainer Reviewed-by on v6 2/9
#5 lore raw

Re: [PATCH V6 2/9] virtio: use virtio_reset_device() when possible

In reply to Jason Wang
Quoted 14 lines · Jason Wang

This allows us to do common extension without duplicating code.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 29 lines
---
 drivers/virtio/virtio.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/virtio/virtio.c +2 -3
index 938e975029d4..aa1eb5132767 100644
@@ -430,7 +430,7 @@ int register_virtio_device(struct virtio_device *dev)
 
        /* We always start by resetting the device, in case a previous
         * driver messed it up.  This also tests that code path a little. */
-       dev->config->reset(dev);
+       virtio_reset_device(dev);
 
        /* Acknowledge that we've seen the device. */
        virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
@@ -496,7 +496,7 @@ int virtio_device_restore(struct virtio_device *dev)
 
        /* We always start by resetting the device, in case a previous
         * driver messed it up. */
-       dev->config->reset(dev);
+       virtio_reset_device(dev);
 
        /* Acknowledge that we've seen the device. */
        virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
--

2.25.1

Reviewed-by on v6 2/9
#6 lore raw

Re: [PATCH V6 2/9] virtio: use virtio_reset_device() when possible

In reply to Jason Wang
Quoted 13 lines · Jason Wang

This allows us to do common extension without duplicating code.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 3 lines

---
drivers/virtio/virtio.c | 4 ++--

1 file changed, 2 insertions(+), 2 deletions(-)

Maybe I had already reviewed it :-), anyway:

author
#7 lore raw

PATCH v6 3/9 virtio: introduce config op to synchronize vring callbacks

In reply to Jason Wang

This patch introduces new virtio config op to vring callbacks. Transport specific method is required to make sure the write before this function is visible to the vring_interrupt() that is called after the return of this function. For the transport that doesn't provide synchronize_vqs(), use synchornize_rcu() which synchronize with IRQ implicitly as a fallback.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
 include/linux/virtio_config.h | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)
include/linux/virtio_config.h +25 -0
index b341dd62aa4d..25be018810a7 100644
@@ -57,6 +57,11 @@ struct virtio_shm_region {
  *		include a NULL entry for vqs unused by driver
  *	Returns 0 on success or error status
  * @del_vqs: free virtqueues found by find_vqs().
+ * @synchronize_cbs: synchronize with the virtqueue callbacks (optional)
+ *      The function guarantees that all memory operations on the
+ *      queue before it are visible to the vring_interrupt() that is
+ *      called after it.
+ *      vdev: the virtio_device
  * @get_features: get the array of feature bits for this device.
  *	vdev: the virtio_device
  *	Returns the first 64 feature bits (all we currently need).
@@ -89,6 +94,7 @@ struct virtio_config_ops {
 			const char * const names[], const bool *ctx,
 			struct irq_affinity *desc);
 	void (*del_vqs)(struct virtio_device *);
+	void (*synchronize_cbs)(struct virtio_device *);
 	u64 (*get_features)(struct virtio_device *vdev);
 	int (*finalize_features)(struct virtio_device *vdev);
 	const char *(*bus_name)(struct virtio_device *vdev);
@@ -217,6 +223,25 @@ int virtio_find_vqs_ctx(struct virtio_device *vdev, unsigned nvqs,
 				      desc);
 }
 
+/**
+ * virtio_synchronize_cbs - synchronize with virtqueue callbacks
+ * @vdev: the device
+ */
+static inline
+void virtio_synchronize_cbs(struct virtio_device *dev)
+{
+	if (dev->config->synchronize_cbs) {
+		dev->config->synchronize_cbs(dev);
+	} else {
+		/*
+		 * A best effort fallback to synchronize with
+		 * interrupts, preemption and softirq disabled
+		 * regions. See comment above synchronize_rcu().
+		 */
+		synchronize_rcu();
+	}
+}
+
 /**
  * virtio_device_ready - enable vq use in probe function
  * @vdev: the device
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
reviewer Reviewed-by on v6 3/9
#8 lore raw

Re: [PATCH V6 3/9] virtio: introduce config op to synchronize vring callbacks

Quoted 7 lines · Jason Wang

This patch introduces new virtio config op to vring callbacks. Transport specific method is required to make sure the write before this function is visible to the vring_interrupt() that is called after the return of this function. For the transport that doesn't provide synchronize_vqs(), use synchornize_rcu() which synchronize with IRQ implicitly as a fallback.

Quoted 11 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 55 lines
---
 include/linux/virtio_config.h | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)
include/linux/virtio_config.h +25 -1
index b341dd62aa4d..25be018810a7 100644
@@ -57,6 +57,11 @@ struct virtio_shm_region {
  *		include a NULL entry for vqs unused by driver
  *	Returns 0 on success or error status
  * @del_vqs: free virtqueues found by find_vqs().
+ * @synchronize_cbs: synchronize with the virtqueue callbacks (optional)
+ *      The function guarantees that all memory operations on the
+ *      queue before it are visible to the vring_interrupt() that is
+ *      called after it.
+ *      vdev: the virtio_device
  * @get_features: get the array of feature bits for this device.
  *	vdev: the virtio_device
  *	Returns the first 64 feature bits (all we currently need).
@@ -89,6 +94,7 @@ struct virtio_config_ops {
 			const char * const names[], const bool *ctx,
 			struct irq_affinity *desc);
 	void (*del_vqs)(struct virtio_device *);
+	void (*synchronize_cbs)(struct virtio_device *);
 	u64 (*get_features)(struct virtio_device *vdev);
 	int (*finalize_features)(struct virtio_device *vdev);
 	const char *(*bus_name)(struct virtio_device *vdev);
@@ -217,6 +223,25 @@ int virtio_find_vqs_ctx(struct virtio_device *vdev, unsigned nvqs,
 				      desc);
 }
 
+/**
+ * virtio_synchronize_cbs - synchronize with virtqueue callbacks
+ * @vdev: the device
+ */
+static inline
+void virtio_synchronize_cbs(struct virtio_device *dev)
+{
+	if (dev->config->synchronize_cbs) {
+		dev->config->synchronize_cbs(dev);
+	} else {
+		/*
+		 * A best effort fallback to synchronize with
+		 * interrupts, preemption and softirq disabled
+		 * regions. See comment above synchronize_rcu().
+		 */
+		synchronize_rcu();
+	}
+}
+
 /**
  * virtio_device_ready - enable vq use in probe function
  * @vdev: the device
--
Quoted 2 lines

2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Reviewed-by on v6 3/9
#9 lore raw

Re: [PATCH V6 3/9] virtio: introduce config op to synchronize vring callbacks

In reply to Jason Wang
Quoted 18 lines · Jason Wang

This patch introduces new virtio config op to vring callbacks. Transport specific method is required to make sure the write before this function is visible to the vring_interrupt() that is called after the return of this function. For the transport that doesn't provide synchronize_vqs(), use synchornize_rcu() which synchronize with IRQ implicitly as a fallback.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 3 lines

---
include/linux/virtio_config.h | 25 +++++++++++++++++++++++++

1 file changed, 25 insertions(+)
author
#10 lore raw

PATCH v6 4/9 virtio-pci: implement synchronize_cbs()

In reply to Jason Wang

We can simply reuse vp_synchronize_vectors() for .synchronize_cbs().

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
 drivers/virtio/virtio_pci_legacy.c | 1 +
 drivers/virtio/virtio_pci_modern.c | 2 ++
 2 files changed, 3 insertions(+)
drivers/virtio/virtio_pci_legacy.c +1 -0
index 7fe4caa4b519..a5e5721145c7 100644
@@ -192,6 +192,7 @@ static const struct virtio_config_ops virtio_pci_config_ops = {
 	.reset		= vp_reset,
 	.find_vqs	= vp_find_vqs,
 	.del_vqs	= vp_del_vqs,
+	.synchronize_cbs = vp_synchronize_vectors,
 	.get_features	= vp_get_features,
 	.finalize_features = vp_finalize_features,
 	.bus_name	= vp_bus_name,
drivers/virtio/virtio_pci_modern.c +2 -0
index 4acb34409f0b..623906b4996c 100644
@@ -394,6 +394,7 @@ static const struct virtio_config_ops virtio_pci_config_nodev_ops = {
 	.reset		= vp_reset,
 	.find_vqs	= vp_modern_find_vqs,
 	.del_vqs	= vp_del_vqs,
+	.synchronize_cbs = vp_synchronize_vectors,
 	.get_features	= vp_get_features,
 	.finalize_features = vp_finalize_features,
 	.bus_name	= vp_bus_name,
@@ -411,6 +412,7 @@ static const struct virtio_config_ops virtio_pci_config_ops = {
 	.reset		= vp_reset,
 	.find_vqs	= vp_modern_find_vqs,
 	.del_vqs	= vp_del_vqs,
+	.synchronize_cbs = vp_synchronize_vectors,
 	.get_features	= vp_get_features,
 	.finalize_features = vp_finalize_features,
 	.bus_name	= vp_bus_name,
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
reviewer Reviewed-by on v6 4/9
#11 lore raw

Re: [PATCH V6 4/9] virtio-pci: implement synchronize_cbs()

Quoted 13 lines · Jason Wang

We can simply reuse vp_synchronize_vectors() for .synchronize_cbs().

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 40 lines
---
 drivers/virtio/virtio_pci_legacy.c | 1 +
 drivers/virtio/virtio_pci_modern.c | 2 ++
 2 files changed, 3 insertions(+)
drivers/virtio/virtio_pci_legacy.c +1 -0
index 7fe4caa4b519..a5e5721145c7 100644
@@ -192,6 +192,7 @@ static const struct virtio_config_ops virtio_pci_config_ops = {
 	.reset		= vp_reset,
 	.find_vqs	= vp_find_vqs,
 	.del_vqs	= vp_del_vqs,
+	.synchronize_cbs = vp_synchronize_vectors,
 	.get_features	= vp_get_features,
 	.finalize_features = vp_finalize_features,
 	.bus_name	= vp_bus_name,
drivers/virtio/virtio_pci_modern.c +2 -1
index 4acb34409f0b..623906b4996c 100644
@@ -394,6 +394,7 @@ static const struct virtio_config_ops virtio_pci_config_nodev_ops = {
 	.reset		= vp_reset,
 	.find_vqs	= vp_modern_find_vqs,
 	.del_vqs	= vp_del_vqs,
+	.synchronize_cbs = vp_synchronize_vectors,
 	.get_features	= vp_get_features,
 	.finalize_features = vp_finalize_features,
 	.bus_name	= vp_bus_name,
@@ -411,6 +412,7 @@ static const struct virtio_config_ops virtio_pci_config_ops = {
 	.reset		= vp_reset,
 	.find_vqs	= vp_modern_find_vqs,
 	.del_vqs	= vp_del_vqs,
+	.synchronize_cbs = vp_synchronize_vectors,
 	.get_features	= vp_get_features,
 	.finalize_features = vp_finalize_features,
 	.bus_name	= vp_bus_name,
--

2.25.1

author
#12 lore raw

PATCH v6 5/9 virtio-mmio: implement synchronize_cbs()

In reply to Jason Wang

Simply synchronize the platform irq that is used by us.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
 drivers/virtio/virtio_mmio.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/virtio/virtio_mmio.c +8 -0
index 839684d672af..c9699a59f93c 100644
@@ -345,6 +345,13 @@ static void vm_del_vqs(struct virtio_device *vdev)
 	free_irq(platform_get_irq(vm_dev->pdev, 0), vm_dev);
 }
 
+static void vm_synchronize_cbs(struct virtio_device *vdev)
+{
+	struct virtio_mmio_device *vm_dev = to_virtio_mmio_device(vdev);
+
+	synchronize_irq(platform_get_irq(vm_dev->pdev, 0));
+}
+
 static struct virtqueue *vm_setup_vq(struct virtio_device *vdev, unsigned int index,
 				  void (*callback)(struct virtqueue *vq),
 				  const char *name, bool ctx)
@@ -541,6 +548,7 @@ static const struct virtio_config_ops virtio_mmio_config_ops = {
 	.finalize_features = vm_finalize_features,
 	.bus_name	= vm_bus_name,
 	.get_shm_region = vm_get_shm_region,
+	.synchronize_cbs = vm_synchronize_cbs,
 };
 
 
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
reviewer Reviewed-by on v6 5/9
#13 lore raw

Re: [PATCH V6 5/9] virtio-mmio: implement synchronize_cbs()

Quoted 13 lines · Jason Wang

Simply synchronize the platform irq that is used by us.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 33 lines
---
 drivers/virtio/virtio_mmio.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/virtio/virtio_mmio.c +8 -1
index 839684d672af..c9699a59f93c 100644
@@ -345,6 +345,13 @@ static void vm_del_vqs(struct virtio_device *vdev)
 	free_irq(platform_get_irq(vm_dev->pdev, 0), vm_dev);
 }
 
+static void vm_synchronize_cbs(struct virtio_device *vdev)
+{
+	struct virtio_mmio_device *vm_dev = to_virtio_mmio_device(vdev);
+
+	synchronize_irq(platform_get_irq(vm_dev->pdev, 0));
+}
+
 static struct virtqueue *vm_setup_vq(struct virtio_device *vdev, unsigned int index,
 				  void (*callback)(struct virtqueue *vq),
 				  const char *name, bool ctx)
@@ -541,6 +548,7 @@ static const struct virtio_config_ops virtio_mmio_config_ops = {
 	.finalize_features = vm_finalize_features,
 	.bus_name	= vm_bus_name,
 	.get_shm_region = vm_get_shm_region,
+	.synchronize_cbs = vm_synchronize_cbs,
 };
 
 
--

2.25.1

author
#14 lore raw

PATCH v6 6/9 virtio-ccw: implement synchronize_cbs()

In reply to Jason Wang

This patch tries to implement the synchronize_cbs() for ccw. For the vring_interrupt() that is called via virtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For the vring_interrupt() that is called via virtio_ccw_int_handler(), a per device rwlock is introduced and used in the synchronization method.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
 drivers/s390/virtio/virtio_ccw.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)
drivers/s390/virtio/virtio_ccw.c +30 -0
index d35e7a3f7067..c188e4f20ca3 100644
@@ -62,6 +62,7 @@ struct virtio_ccw_device {
 	unsigned int revision; /* Transport revision */
 	wait_queue_head_t wait_q;
 	spinlock_t lock;
+	rwlock_t irq_lock;
 	struct mutex io_lock; /* Serializes I/O requests */
 	struct list_head virtqueues;
 	bool is_thinint;
@@ -984,6 +985,30 @@ static const char *virtio_ccw_bus_name(struct virtio_device *vdev)
 	return dev_name(&vcdev->cdev->dev);
 }
 
+static void virtio_ccw_synchronize_cbs(struct virtio_device *vdev)
+{
+	struct virtio_ccw_device *vcdev = to_vc_device(vdev);
+	struct airq_info *info = vcdev->airq_info;
+
+	if (info) {
+		/*
+		 * This device uses adapter interrupts: synchronize with
+		 * vring_interrupt() called by virtio_airq_handler()
+		 * via the indicator area lock.
+		 */
+		write_lock_irq(&info->lock);
+		write_unlock_irq(&info->lock);
+	} else {
+		/* This device uses classic interrupts: synchronize
+		 * with vring_interrupt() called by
+		 * virtio_ccw_int_handler() via the per-device
+		 * irq_lock
+		 */
+		write_lock_irq(&vcdev->irq_lock);
+		write_unlock_irq(&vcdev->irq_lock);
+	}
+}
+
 static const struct virtio_config_ops virtio_ccw_config_ops = {
 	.get_features = virtio_ccw_get_features,
 	.finalize_features = virtio_ccw_finalize_features,
@@ -995,6 +1020,7 @@ static const struct virtio_config_ops virtio_ccw_config_ops = {
 	.find_vqs = virtio_ccw_find_vqs,
 	.del_vqs = virtio_ccw_del_vqs,
 	.bus_name = virtio_ccw_bus_name,
+	.synchronize_cbs = virtio_ccw_synchronize_cbs,
 };
 
 
@@ -1106,6 +1132,8 @@ static void virtio_ccw_int_handler(struct ccw_device *cdev,
 			vcdev->err = -EIO;
 	}
 	virtio_ccw_check_activity(vcdev, activity);
+	/* Interrupts are disabled here */
+	read_lock(&vcdev->irq_lock);
 	for_each_set_bit(i, indicators(vcdev),
 			 sizeof(*indicators(vcdev)) * BITS_PER_BYTE) {
 		/* The bit clear must happen before the vring kick. */
@@ -1114,6 +1142,7 @@ static void virtio_ccw_int_handler(struct ccw_device *cdev,
 		vq = virtio_ccw_vq_by_ind(vcdev, i);
 		vring_interrupt(0, vq);
 	}
+	read_unlock(&vcdev->irq_lock);
 	if (test_bit(0, indicators2(vcdev))) {
 		virtio_config_changed(&vcdev->vdev);
 		clear_bit(0, indicators2(vcdev));
@@ -1284,6 +1313,7 @@ static int virtio_ccw_online(struct ccw_device *cdev)
 	init_waitqueue_head(&vcdev->wait_q);
 	INIT_LIST_HEAD(&vcdev->virtqueues);
 	spin_lock_init(&vcdev->lock);
+	rwlock_init(&vcdev->irq_lock);
 	mutex_init(&vcdev->io_lock);
 
 	spin_lock_irqsave(get_ccwdev_lock(cdev), flags);
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
maintainer Reviewed-by on v6 6/9
#15 lore raw

Re: [PATCH V6 6/9] virtio-ccw: implement synchronize_cbs()

Quoted 17 lines · Jason Wang

This patch tries to implement the synchronize_cbs() for ccw. For the vring_interrupt() that is called via virtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For the vring_interrupt() that is called via virtio_ccw_int_handler(), a per device rwlock is introduced and used in the synchronization method.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 3 lines
---
 drivers/s390/virtio/virtio_ccw.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

author
#16 lore raw

PATCH v6 7/9 virtio: allow to unbreak virtqueue

In reply to Jason Wang

This patch allows the new introduced __virtio_break_device() to unbreak the virtqueue.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
 drivers/virtio/virtio_ring.c | 22 ++++++++++++++++++++++
 include/linux/virtio.h       |  1 +
 2 files changed, 23 insertions(+)
drivers/virtio/virtio_ring.c +22 -0
index 9d0bae4293be..9c231e1fded7 100644
@@ -2395,6 +2395,28 @@ void virtio_break_device(struct virtio_device *dev)
 }
 EXPORT_SYMBOL_GPL(virtio_break_device);
 
+/*
+ * This should allow the device to be used by the driver. You may
+ * need to grab appropriate locks to flush the write to
+ * vq->broken. This should only be used in some specific case e.g
+ * (probing and restoring). This function should only be called by the
+ * core, not directly by the driver.
+ */
+void __virtio_unbreak_device(struct virtio_device *dev)
+{
+	struct virtqueue *_vq;
+
+	spin_lock(&dev->vqs_list_lock);
+	list_for_each_entry(_vq, &dev->vqs, list) {
+		struct vring_virtqueue *vq = to_vvq(_vq);
+
+		/* Pairs with READ_ONCE() in virtqueue_is_broken(). */
+		WRITE_ONCE(vq->broken, false);
+	}
+	spin_unlock(&dev->vqs_list_lock);
+}
+EXPORT_SYMBOL_GPL(__virtio_unbreak_device);
+
 dma_addr_t virtqueue_get_desc_addr(struct virtqueue *_vq)
 {
 	struct vring_virtqueue *vq = to_vvq(_vq);
include/linux/virtio.h +1 -0
index 5464f398912a..d8fdf170637c 100644
@@ -131,6 +131,7 @@ void unregister_virtio_device(struct virtio_device *dev);
 bool is_virtio_device(struct device *dev);
 
 void virtio_break_device(struct virtio_device *dev);
+void __virtio_unbreak_device(struct virtio_device *dev);
 
 void virtio_config_changed(struct virtio_device *dev);
 #ifdef CONFIG_PM_SLEEP
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
reviewer Reviewed-by on v6 7/9
#17 lore raw

Re: [PATCH V6 7/9] virtio: allow to unbreak virtqueue

Quoted 13 lines · Jason Wang

This patch allows the new introduced __virtio_break_device() to unbreak the virtqueue.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 51 lines
---
 drivers/virtio/virtio_ring.c | 22 ++++++++++++++++++++++
 include/linux/virtio.h       |  1 +
 2 files changed, 23 insertions(+)
drivers/virtio/virtio_ring.c +22 -0
index 9d0bae4293be..9c231e1fded7 100644
@@ -2395,6 +2395,28 @@ void virtio_break_device(struct virtio_device *dev)
 }
 EXPORT_SYMBOL_GPL(virtio_break_device);
 
+/*
+ * This should allow the device to be used by the driver. You may
+ * need to grab appropriate locks to flush the write to
+ * vq->broken. This should only be used in some specific case e.g
+ * (probing and restoring). This function should only be called by the
+ * core, not directly by the driver.
+ */
+void __virtio_unbreak_device(struct virtio_device *dev)
+{
+	struct virtqueue *_vq;
+
+	spin_lock(&dev->vqs_list_lock);
+	list_for_each_entry(_vq, &dev->vqs, list) {
+		struct vring_virtqueue *vq = to_vvq(_vq);
+
+		/* Pairs with READ_ONCE() in virtqueue_is_broken(). */
+		WRITE_ONCE(vq->broken, false);
+	}
+	spin_unlock(&dev->vqs_list_lock);
+}
+EXPORT_SYMBOL_GPL(__virtio_unbreak_device);
+
 dma_addr_t virtqueue_get_desc_addr(struct virtqueue *_vq)
 {
 	struct vring_virtqueue *vq = to_vvq(_vq);
include/linux/virtio.h +1 -1
index 5464f398912a..d8fdf170637c 100644
@@ -131,6 +131,7 @@ void unregister_virtio_device(struct virtio_device *dev);
 bool is_virtio_device(struct device *dev);
 
 void virtio_break_device(struct virtio_device *dev);
+void __virtio_unbreak_device(struct virtio_device *dev);
 
 void virtio_config_changed(struct virtio_device *dev);
 #ifdef CONFIG_PM_SLEEP
--
Quoted 2 lines

2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

maintainer Reviewed-by on v6 7/9
#18 lore raw

Re: [PATCH V6 7/9] virtio: allow to unbreak virtqueue

In reply to Jason Wang
Quoted 23 lines · Jason Wang

This patch allows the new introduced __virtio_break_device() to unbreak the virtqueue.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
 drivers/virtio/virtio_ring.c | 22 ++++++++++++++++++++++
 include/linux/virtio.h       |  1 +
 2 files changed, 23 insertions(+)
drivers/virtio/virtio_ring.c +0 -0
index 9d0bae4293be..9c231e1fded7 100644

drivers/virtio/virtio_ring.c void virtio_break_device(struct virtio_device *dev) · line 2402

Quoted 8 lines
drivers/virtio/virtio_ring.c +5 -0
@@ -2395,6 +2395,28 @@ void virtio_break_device(struct virtio_device *dev)
 }
 EXPORT_SYMBOL_GPL(virtio_break_device);
 
+/*
+ * This should allow the device to be used by the driver. You may
+ * need to grab appropriate locks to flush the write to
+ * vq->broken. This should only be used in some specific case e.g
+ * (probing and restoring). This function should only be called by the

Minor: "...some specific cases, e.g. probing and restoring."

But no need to respin from my side.

drivers/virtio/virtio_ring.c void virtio_break_device(struct virtio_device *dev) · line 2404

Quoted 2 lines
drivers/virtio/virtio_ring.c +2 -0
+ * core, not directly by the driver.
+ */

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

author
#19 lore raw

PATCH v6 8/9 virtio: harden vring IRQ

In reply to Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
 	ccw->flags = 0;
 	ccw->count = sizeof(status);
 	ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+	/* We use ssch for setting the status which is a serializing
+	 * instruction that guarantees the memory writes have
+	 * completed before ssch.
+	 */
 	ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
 	/* Write failed? We assume status is unchanged. */
 	if (ret)
drivers/virtio/virtio.c +12 -3
index aa1eb5132767..95fac4c97c8b 100644
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+	/*
+	 * The below virtio_synchronize_cbs() guarantees that any
+	 * interrupt for this line arriving after
+	 * virtio_synchronize_vqs() has completed is guaranteed to see
+	 * vq->broken as true.
+	 */
+	virtio_break_device(dev);
+	virtio_synchronize_cbs(dev);
+
 	dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
 	dev->config_enabled = false;
 	dev->config_change_pending = false;
 
+	INIT_LIST_HEAD(&dev->vqs);
+	spin_lock_init(&dev->vqs_list_lock);
+
 	/* We always start by resetting the device, in case a previous
 	 * driver messed it up.  This also tests that code path a little. */
 	virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
 	/* Acknowledge that we've seen the device. */
 	virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-	INIT_LIST_HEAD(&dev->vqs);
-	spin_lock_init(&dev->vqs_list_lock);
-
 	/*
 	 * device_add() causes the bus infrastructure to look for a matching
 	 * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
 	/* We should never be setting status to 0. */
 	BUG_ON(status == 0);
 
+	/*
+	 * Per memory-barriers.txt, wmb() is not needed to guarantee
+	 * that the the cache coherent memory writes have completed
+	 * before writing to the MMIO region.
+	 */
 	writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
 	struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+	/*
+	 * Per memory-barriers.txt, wmb() is not needed to guarantee
+	 * that the the cache coherent memory writes have completed
+	 * before writing to the MMIO region.
+	 */
 	vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +7 -4
index 9c231e1fded7..13a7348cedff 100644
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
 	vq->we_own_ring = true;
 	vq->notify = notify;
 	vq->weak_barriers = weak_barriers;
-	vq->broken = false;
+	vq->broken = true;
 	vq->last_used_idx = 0;
 	vq->event_triggered = false;
 	vq->num_added = 0;
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
 		return IRQ_NONE;
 	}
 
-	if (unlikely(vq->broken))
-		return IRQ_HANDLED;
+	if (unlikely(vq->broken)) {
+		dev_warn_once(&vq->vq.vdev->dev,
+			      "virtio vring IRQ raised before DRIVER_OK");
+		return IRQ_NONE;
+	}
 
 	/* Just a hint for performance: so it's ok that this can be racy! */
 	if (vq->event)
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
 	vq->we_own_ring = false;
 	vq->notify = notify;
 	vq->weak_barriers = weak_barriers;
-	vq->broken = false;
+	vq->broken = true;
 	vq->last_used_idx = 0;
 	vq->event_triggered = false;
 	vq->num_added = 0;
include/linux/virtio_config.h +20 -0
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
 	unsigned status = dev->config->get_status(dev);
 
 	BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+	/*
+	 * The virtio_synchronize_cbs() makes sure vring_interrupt()
+	 * will see the driver specific setup if it sees vq->broken
+	 * as false (even if the notifications come before DRIVER_OK).
+	 */
+	virtio_synchronize_cbs(dev);
+	__virtio_unbreak_device(dev);
+	/*
+	 * The transport should ensure the visibility of vq->broken
+	 * before setting DRIVER_OK. See the comments for the transport
+	 * specific set_status() method.
+	 *
+	 * A well behaved device will only notify a virtqueue after
+	 * DRIVER_OK, this means the device should "see" the coherenct
+	 * memory write that set vq->broken as false which is done by
+	 * the driver when it sees DRIVER_OK, then the following
+	 * driver's vring_interrupt() will see vq->broken as false so
+	 * we won't lose any notification.
+	 */
 	dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
reviewer Reviewed-by on v6 8/9
#20 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 166 lines
---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
 	ccw->flags = 0;
 	ccw->count = sizeof(status);
 	ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+	/* We use ssch for setting the status which is a serializing
+	 * instruction that guarantees the memory writes have
+	 * completed before ssch.
+	 */
 	ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
 	/* Write failed? We assume status is unchanged. */
 	if (ret)
drivers/virtio/virtio.c +12 -3
index aa1eb5132767..95fac4c97c8b 100644
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+	/*
+	 * The below virtio_synchronize_cbs() guarantees that any
+	 * interrupt for this line arriving after
+	 * virtio_synchronize_vqs() has completed is guaranteed to see
+	 * vq->broken as true.
+	 */
+	virtio_break_device(dev);
+	virtio_synchronize_cbs(dev);
+
 	dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
 	dev->config_enabled = false;
 	dev->config_change_pending = false;
 
+	INIT_LIST_HEAD(&dev->vqs);
+	spin_lock_init(&dev->vqs_list_lock);
+
 	/* We always start by resetting the device, in case a previous
 	 * driver messed it up.  This also tests that code path a little. */
 	virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
 	/* Acknowledge that we've seen the device. */
 	virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-	INIT_LIST_HEAD(&dev->vqs);
-	spin_lock_init(&dev->vqs_list_lock);
-
 	/*
 	 * device_add() causes the bus infrastructure to look for a matching
 	 * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
 	/* We should never be setting status to 0. */
 	BUG_ON(status == 0);
 
+	/*
+	 * Per memory-barriers.txt, wmb() is not needed to guarantee
+	 * that the the cache coherent memory writes have completed
+	 * before writing to the MMIO region.
+	 */
 	writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
 	struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+	/*
+	 * Per memory-barriers.txt, wmb() is not needed to guarantee
+	 * that the the cache coherent memory writes have completed
+	 * before writing to the MMIO region.
+	 */
 	vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +7 -4
index 9c231e1fded7..13a7348cedff 100644
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
 	vq->we_own_ring = true;
 	vq->notify = notify;
 	vq->weak_barriers = weak_barriers;
-	vq->broken = false;
+	vq->broken = true;
 	vq->last_used_idx = 0;
 	vq->event_triggered = false;
 	vq->num_added = 0;
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
 		return IRQ_NONE;
 	}
 
-	if (unlikely(vq->broken))
-		return IRQ_HANDLED;
+	if (unlikely(vq->broken)) {
+		dev_warn_once(&vq->vq.vdev->dev,
+			      "virtio vring IRQ raised before DRIVER_OK");
+		return IRQ_NONE;
+	}
 
 	/* Just a hint for performance: so it's ok that this can be racy! */
 	if (vq->event)
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
 	vq->we_own_ring = false;
 	vq->notify = notify;
 	vq->weak_barriers = weak_barriers;
-	vq->broken = false;
+	vq->broken = true;
 	vq->last_used_idx = 0;
 	vq->event_triggered = false;
 	vq->num_added = 0;
include/linux/virtio_config.h +20 -1
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
 	unsigned status = dev->config->get_status(dev);
 
 	BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+	/*
+	 * The virtio_synchronize_cbs() makes sure vring_interrupt()
+	 * will see the driver specific setup if it sees vq->broken
+	 * as false (even if the notifications come before DRIVER_OK).
+	 */
+	virtio_synchronize_cbs(dev);
+	__virtio_unbreak_device(dev);
+	/*
+	 * The transport should ensure the visibility of vq->broken
+	 * before setting DRIVER_OK. See the comments for the transport
+	 * specific set_status() method.
+	 *
+	 * A well behaved device will only notify a virtqueue after
+	 * DRIVER_OK, this means the device should "see" the coherenct
+	 * memory write that set vq->broken as false which is done by
+	 * the driver when it sees DRIVER_OK, then the following
+	 * driver's vring_interrupt() will see vq->broken as false so
+	 * we won't lose any notification.
+	 */
 	dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
--

2.25.1

maintainer Reviewed-by on v6 8/9
#21 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

In reply to Jason Wang
Quoted 35 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 8 lines
---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#22 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

In reply to Jason Wang
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

Quoted 29 lines
---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
 	ccw->flags = 0;
 	ccw->count = sizeof(status);
 	ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+	/* We use ssch for setting the status which is a serializing
+	 * instruction that guarantees the memory writes have
+	 * completed before ssch.
+	 */
 	ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
 	/* Write failed? We assume status is unchanged. */
 	if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+	/*
+	 * The below virtio_synchronize_cbs() guarantees that any
+	 * interrupt for this line arriving after
+	 * virtio_synchronize_vqs() has completed is guaranteed to see
+	 * vq->broken as true.
+	 */
+	virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+	virtio_synchronize_cbs(dev);
+
 	dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
 	dev->config_enabled = false;
 	dev->config_change_pending = false;
 
+	INIT_LIST_HEAD(&dev->vqs);
+	spin_lock_init(&dev->vqs_list_lock);
+
 	/* We always start by resetting the device, in case a previous
 	 * driver messed it up.  This also tests that code path a little. */
 	virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
 	/* Acknowledge that we've seen the device. */
 	virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-	INIT_LIST_HEAD(&dev->vqs);
-	spin_lock_init(&dev->vqs_list_lock);
-
 	/*
 	 * device_add() causes the bus infrastructure to look for a matching
 	 * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
 	/* We should never be setting status to 0. */
 	BUG_ON(status == 0);
 
+	/*
+	 * Per memory-barriers.txt, wmb() is not needed to guarantee
+	 * that the the cache coherent memory writes have completed
+	 * before writing to the MMIO region.
+	 */
 	writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
 	struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+	/*
+	 * Per memory-barriers.txt, wmb() is not needed to guarantee
+	 * that the the cache coherent memory writes have completed
+	 * before writing to the MMIO region.
+	 */
 	vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
 	vq->we_own_ring = true;
 	vq->notify = notify;
 	vq->weak_barriers = weak_barriers;
-	vq->broken = false;
+	vq->broken = true;
 	vq->last_used_idx = 0;
 	vq->event_triggered = false;
 	vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
 		return IRQ_NONE;
 	}
 
-	if (unlikely(vq->broken))
-		return IRQ_HANDLED;
+	if (unlikely(vq->broken)) {
+		dev_warn_once(&vq->vq.vdev->dev,
+			      "virtio vring IRQ raised before DRIVER_OK");
+		return IRQ_NONE;
+	}
 
 	/* Just a hint for performance: so it's ok that this can be racy! */
 	if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
 	vq->we_own_ring = false;
 	vq->notify = notify;
 	vq->weak_barriers = weak_barriers;
-	vq->broken = false;
+	vq->broken = true;
 	vq->last_used_idx = 0;
 	vq->event_triggered = false;
 	vq->num_added = 0;

and make this conditional

Quoted 33 lines
include/linux/virtio_config.h +20 -0
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
 	unsigned status = dev->config->get_status(dev);
 
 	BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+	/*
+	 * The virtio_synchronize_cbs() makes sure vring_interrupt()
+	 * will see the driver specific setup if it sees vq->broken
+	 * as false (even if the notifications come before DRIVER_OK).
+	 */
+	virtio_synchronize_cbs(dev);
+	__virtio_unbreak_device(dev);
+	/*
+	 * The transport should ensure the visibility of vq->broken
+	 * before setting DRIVER_OK. See the comments for the transport
+	 * specific set_status() method.
+	 *
+	 * A well behaved device will only notify a virtqueue after
+	 * DRIVER_OK, this means the device should "see" the coherenct
+	 * memory write that set vq->broken as false which is done by
+	 * the driver when it sees DRIVER_OK, then the following
+	 * driver's vring_interrupt() will see vq->broken as false so
+	 * we won't lose any notification.
+	 */
 	dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
Signature · 1 line
2.25.1
author
#23 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Thanks

Quoted 177 lines
Quoted 31 lines

---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 33 lines
include/linux/virtio_config.h +20 -1
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
--

2.25.1

#24 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 47 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

Quoted 8 lines

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c
drivers/gpu/drm/virtio/virtgpu_drv.c
drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

Quoted 177 lines
Quoted 177 lines
Quoted 31 lines

---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 33 lines
include/linux/virtio_config.h +20 -1
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
--

2.25.1

author
#25 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 1 line

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

I will post a patch for mac80211_hwsim.c.

Thanks

Quoted 178 lines
Quoted 178 lines
Quoted 178 lines
Quoted 35 lines





---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 31 lines
include/linux/virtio_config.h +20 -0
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
Quoted 4 lines
Quoted 3 lines
Quoted 3 lines
Quoted 2 lines

--
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#26 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 59 lines · Jason Wang
Quoted 58 lines · Michael S. Tsirkin
Quoted 55 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Thanks

Quoted 7 lines
Quoted 5 lines

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

Quoted 4 lines
Quoted 1 line

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 6 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

Quoted 7 lines

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 182 lines
Quoted 182 lines
Quoted 181 lines
Quoted 181 lines
Quoted 35 lines





---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 33 lines
include/linux/virtio_config.h +20 -1
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
--

2.25.1

author
#27 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 65 lines · Michael S. Tsirkin
Quoted 63 lines · Jason Wang
Quoted 62 lines · Michael S. Tsirkin
Quoted 47 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

Quoted 8 lines

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

Quoted 8 lines
Quoted 3 lines
Quoted 1 line

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK. If yes, the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device. If not, we need to clarify it in the spec and call virtio_device_ready() before subsystem registration.

Quoted 20 lines
Quoted 6 lines
Quoted 3 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

For the rest we need to do virtio_device_ready() before registration.

Thanks

Quoted 179 lines
Quoted 179 lines
Quoted 179 lines
Quoted 179 lines
Quoted 179 lines
Quoted 36 lines






---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 31 lines
include/linux/virtio_config.h +20 -0
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
Quoted 3 lines
Quoted 3 lines
Quoted 2 lines

--
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#28 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 68 lines · Jason Wang
Quoted 67 lines · Michael S. Tsirkin
Quoted 65 lines · Jason Wang
Quoted 63 lines · Michael S. Tsirkin
Quoted 47 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

Quoted 8 lines

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

Quoted 8 lines
Quoted 6 lines
Quoted 1 line

This driver seems to be fine, it doesn't use the device/vq in its probe().

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Quoted 3 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

Quoted 2 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

Quoted 26 lines
Quoted 20 lines
Quoted 6 lines
Quoted 3 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well? And without kick, won't everything just be blocked?

Quoted 3 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

Quoted 184 lines
Quoted 184 lines
Quoted 183 lines
Quoted 183 lines
Quoted 182 lines
Quoted 182 lines
Quoted 36 lines






---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 33 lines
include/linux/virtio_config.h +20 -1
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
--

2.25.1

author
#29 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang
Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 5 lines
Quoted 3 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

Quoted 35 lines
Quoted 27 lines
Quoted 21 lines
Quoted 7 lines
Quoted 4 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

Quoted 8 lines

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well?

Kick itself is not blocked.

Quoted 1 line

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

        virtio_device_ready(vdev);

        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 38 lines








---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 31 lines
include/linux/virtio_config.h +20 -0
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
Quoted 6 lines
Quoted 5 lines
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
Quoted 3 lines
Quoted 3 lines
Quoted 2 lines

--
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

author
#30 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 96 lines · Jason Wang
Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang
Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

Quoted 7 lines

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

btw, hci has an open and close method and we do rx refill in hdev->open, so we're probably fine here.

Thanks

Quoted 75 lines
Quoted 6 lines
Quoted 4 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

Quoted 35 lines
Quoted 27 lines
Quoted 21 lines
Quoted 7 lines
Quoted 4 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

Quoted 8 lines

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well?

Kick itself is not blocked.

Quoted 1 line

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

Quoted 195 lines
        virtio_device_ready(vdev);

        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

Quoted 187 lines
Quoted 186 lines
Quoted 186 lines
Quoted 185 lines
Quoted 185 lines
Quoted 184 lines
Quoted 184 lines
Quoted 38 lines








---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 33 lines
include/linux/virtio_config.h +20 -1
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
--

2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#31 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 102 lines · Jason Wang
Quoted 101 lines · Jason Wang
Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang
Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

Quoted 7 lines
Quoted 2 lines

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

btw, hci has an open and close method and we do rx refill in hdev->open, so we're probably fine here.

Thanks

Sounds good. Now to audit the rest of them from this POV ;)

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c
arm_scmi
virtio_rpmsg_bus.c

Quoted 264 lines
Quoted 264 lines
Quoted 6 lines
Quoted 4 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

Quoted 35 lines
Quoted 27 lines
Quoted 21 lines
Quoted 7 lines
Quoted 4 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

Quoted 8 lines

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well?

Kick itself is not blocked.

Quoted 1 line

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

        virtio_device_ready(vdev);

        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 38 lines








---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 31 lines
include/linux/virtio_config.h +20 -0
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
Quoted 3 lines
Quoted 3 lines
Quoted 2 lines

--
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

author
#32 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 107 lines · Michael S. Tsirkin
Quoted 105 lines · Jason Wang
Quoted 103 lines · Jason Wang
Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang
Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 7 lines
Quoted 4 lines

btw, hci has an open and close method and we do rx refill in hdev->open, so we're probably fine here.

Thanks

Sounds good. Now to audit the rest of them from this POV ;)

Adding maintainers.

Quoted 2 lines

drivers/i2c/busses/i2c-virtio.c

It looks to me the device could be used immediately after i2c_add_adapter() return. So we probably need to add virtio_device_ready() before that. Fortunately, there's no rx vq in i2c and the callback looks safe if the callback is called before the i2c registration and after virtio_device_ready().

Quoted 1 line

drivers/net/caif/caif_virtio.c

A networking device, RX is backed by vringh so we don't need to refill. TX is backed by virtio and is available until ndo_open. So it's fine to let the core to set DRIVER_OK after probe().

Quoted 1 line

drivers/nvdimm/virtio_pmem.c

It doesn't use interrupt so far, so it has nothing to do with the IRQ hardening.

But the device could be used by the subsystem immediately after nvdimm_pmem_region_create(), this means the flush could be issued before DRIVER_OK. We need virtio_device_ready() before. We don't have a RX virtqueue and the callback looks safe if the callback is called after virtio_device_ready() but before the nvdimm region creating.

And it looks to me there's a race between the assignment of provider_data and virtio_pmem_flush(). If the flush was issued before the assignment we will end up with a NULL pointer dereference. This is something we need to fix.

Quoted 1 line

arm_scmi

It looks to me the singleton device could be used by SCMI immediately after

        /* Ensure initialized scmi_vdev is visible */
        smp_store_mb(scmi_vdev, vdev);

So we probably need to do virtio_device_ready() before that. It has an optional rx queue but the filling is done after the above assignment, so it's safe. And the callback looks safe is a callback is triggered after virtio_device_ready() buy before the above assignment.

Quoted 2 lines

virtio_rpmsg_bus.c

This is somehow more complicated. It has an rx queue, the rx filling is done before virtio_device_ready() but the kick is done after. And it looks to me the device could be used by subsystem immediately rpmsg_virtio_add_ctrl_dev() returns.

This means, if we do virtio_device_ready() after rpmsg_virtio_add_ctrl_dev(), we may get kick before DRIVER_OK. If we do virtio_device_ready() before rpmsg_virtio_add_ctrl_dev(), there's a race between the callbacks and rpmsg_virtio_add_ctrl_dev() that could be exploited.

It requires more thoughts.

Thanks

Quoted 266 lines
Quoted 266 lines
Quoted 266 lines
Quoted 8 lines
Quoted 6 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

Quoted 35 lines
Quoted 27 lines
Quoted 21 lines
Quoted 7 lines
Quoted 4 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

Quoted 8 lines

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well?

Kick itself is not blocked.

Quoted 1 line

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

        virtio_device_ready(vdev);

        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 181 lines
Quoted 38 lines








---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 31 lines
include/linux/virtio_config.h +20 -0
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
Quoted 7 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
Quoted 3 lines
Quoted 3 lines
Quoted 2 lines

--
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#33 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 127 lines · Jason Wang
Quoted 107 lines · Michael S. Tsirkin
Quoted 105 lines · Jason Wang
Quoted 103 lines · Jason Wang
Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang
Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 7 lines
Quoted 4 lines

btw, hci has an open and close method and we do rx refill in hdev->open, so we're probably fine here.

Thanks

Sounds good. Now to audit the rest of them from this POV ;)

Adding maintainers.

Quoted 2 lines

drivers/i2c/busses/i2c-virtio.c

It looks to me the device could be used immediately after i2c_add_adapter() return. So we probably need to add virtio_device_ready() before that. Fortunately, there's no rx vq in i2c and the callback looks safe if the callback is called before the i2c registration and after virtio_device_ready().

Quoted 5 lines
Quoted 1 line

drivers/net/caif/caif_virtio.c

A networking device, RX is backed by vringh so we don't need to refill. TX is backed by virtio and is available until ndo_open. So it's fine to let the core to set DRIVER_OK after probe().

How about we just add an explicit ready in the driver anyway? I think the implicit ready is just creating a mess as people tend to forget to think about it.

Quoted 318 lines
Quoted 1 line

drivers/nvdimm/virtio_pmem.c

It doesn't use interrupt so far, so it has nothing to do with the IRQ hardening.

But the device could be used by the subsystem immediately after nvdimm_pmem_region_create(), this means the flush could be issued before DRIVER_OK. We need virtio_device_ready() before. We don't have a RX virtqueue and the callback looks safe if the callback is called after virtio_device_ready() but before the nvdimm region creating.

And it looks to me there's a race between the assignment of provider_data and virtio_pmem_flush(). If the flush was issued before the assignment we will end up with a NULL pointer dereference. This is something we need to fix.

Quoted 1 line

arm_scmi

It looks to me the singleton device could be used by SCMI immediately after

        /* Ensure initialized scmi_vdev is visible */
        smp_store_mb(scmi_vdev, vdev);

So we probably need to do virtio_device_ready() before that. It has an optional rx queue but the filling is done after the above assignment, so it's safe. And the callback looks safe is a callback is triggered after virtio_device_ready() buy before the above assignment.

Quoted 2 lines

virtio_rpmsg_bus.c

This is somehow more complicated. It has an rx queue, the rx filling is done before virtio_device_ready() but the kick is done after. And it looks to me the device could be used by subsystem immediately rpmsg_virtio_add_ctrl_dev() returns.

This means, if we do virtio_device_ready() after rpmsg_virtio_add_ctrl_dev(), we may get kick before DRIVER_OK. If we do virtio_device_ready() before rpmsg_virtio_add_ctrl_dev(), there's a race between the callbacks and rpmsg_virtio_add_ctrl_dev() that could be exploited.

It requires more thoughts.

Thanks

Quoted 273 lines
Quoted 272 lines
Quoted 272 lines
Quoted 8 lines
Quoted 6 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

Quoted 35 lines
Quoted 27 lines
Quoted 21 lines
Quoted 7 lines
Quoted 4 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

Quoted 8 lines

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well?

Kick itself is not blocked.

Quoted 1 line

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

        virtio_device_ready(vdev);

        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

Quoted 187 lines
Quoted 186 lines
Quoted 186 lines
Quoted 185 lines
Quoted 185 lines
Quoted 184 lines
Quoted 184 lines
Quoted 38 lines








---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 33 lines
include/linux/virtio_config.h +20 -1
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
--

2.25.1

author
#34 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 131 lines · Michael S. Tsirkin
Quoted 129 lines · Jason Wang
Quoted 107 lines · Michael S. Tsirkin
Quoted 105 lines · Jason Wang
Quoted 103 lines · Jason Wang
Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang
Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 35 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for
virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ
that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ
hardening. The vq->broken is set to true during both initialization
and reset. And the vq->broken is set to false in
virtio_device_ready(). Then vring_interrupt() can check and return
when vq->broken is true. And in this case, switch to return IRQ_NONE
to let the interrupt core aware of such invalid interrupt to prevent
IRQ storm.

The reason of using a per queue variable instead of a per device one
is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the
config interrupt hardening is already done in commit 22b7050a024d7
("virtio: defer config changed notifications"). But the method that is
used by config interrupt can't be reused by the vring interrupt
handler because it uses spinlock to do the synchronization which is
expensive.

Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Cc: Vineeth Vijayan <[email protected]>
Cc: Peter Oberparleiter <[email protected]>
Cc: [email protected]
Signed-off-by: Jason Wang <[email protected]>

Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 7 lines
Quoted 4 lines

btw, hci has an open and close method and we do rx refill in hdev->open, so we're probably fine here.

Thanks

Sounds good. Now to audit the rest of them from this POV ;)

Adding maintainers.

Quoted 2 lines

drivers/i2c/busses/i2c-virtio.c

It looks to me the device could be used immediately after i2c_add_adapter() return. So we probably need to add virtio_device_ready() before that. Fortunately, there's no rx vq in i2c and the callback looks safe if the callback is called before the i2c registration and after virtio_device_ready().

Quoted 1 line

drivers/net/caif/caif_virtio.c

Quoted 7 lines
Quoted 3 lines

A networking device, RX is backed by vringh so we don't need to refill. TX is backed by virtio and is available until ndo_open. So it's fine to let the core to set DRIVER_OK after probe().

How about we just add an explicit ready in the driver anyway? I think the implicit ready is just creating a mess as people tend to forget to think about it.

This is possible, and we could fail the probe if ready is not set by a driver.

Thanks

Quoted 320 lines
Quoted 319 lines
Quoted 2 lines

drivers/nvdimm/virtio_pmem.c

It doesn't use interrupt so far, so it has nothing to do with the IRQ hardening.

But the device could be used by the subsystem immediately after nvdimm_pmem_region_create(), this means the flush could be issued before DRIVER_OK. We need virtio_device_ready() before. We don't have a RX virtqueue and the callback looks safe if the callback is called after virtio_device_ready() but before the nvdimm region creating.

And it looks to me there's a race between the assignment of provider_data and virtio_pmem_flush(). If the flush was issued before the assignment we will end up with a NULL pointer dereference. This is something we need to fix.

Quoted 1 line

arm_scmi

It looks to me the singleton device could be used by SCMI immediately after

        /* Ensure initialized scmi_vdev is visible */
        smp_store_mb(scmi_vdev, vdev);

So we probably need to do virtio_device_ready() before that. It has an optional rx queue but the filling is done after the above assignment, so it's safe. And the callback looks safe is a callback is triggered after virtio_device_ready() buy before the above assignment.

Quoted 2 lines

virtio_rpmsg_bus.c

This is somehow more complicated. It has an rx queue, the rx filling is done before virtio_device_ready() but the kick is done after. And it looks to me the device could be used by subsystem immediately rpmsg_virtio_add_ctrl_dev() returns.

This means, if we do virtio_device_ready() after rpmsg_virtio_add_ctrl_dev(), we may get kick before DRIVER_OK. If we do virtio_device_ready() before rpmsg_virtio_add_ctrl_dev(), there's a race between the callbacks and rpmsg_virtio_add_ctrl_dev() that could be exploited.

It requires more thoughts.

Thanks

Quoted 273 lines
Quoted 272 lines
Quoted 272 lines
Quoted 8 lines
Quoted 6 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

Quoted 35 lines
Quoted 27 lines
Quoted 21 lines
Quoted 7 lines
Quoted 4 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

Quoted 8 lines

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well?

Kick itself is not blocked.

Quoted 1 line

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

        virtio_device_ready(vdev);

        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

Quoted 187 lines
Quoted 186 lines
Quoted 186 lines
Quoted 185 lines
Quoted 185 lines
Quoted 184 lines
Quoted 184 lines
Quoted 48 lines

---
drivers/s390/virtio/virtio_ccw.c | 4 ++++
drivers/virtio/virtio.c | 15 ++++++++++++---
drivers/virtio/virtio_mmio.c | 5 +++++
drivers/virtio/virtio_pci_modern_dev.c | 5 +++++
drivers/virtio/virtio_ring.c | 11 +++++++----
include/linux/virtio_config.h | 20 ++++++++++++++++++++
6 files changed, 53 insertions(+), 7 deletions(-)

diff --git a/drivers/s390/virtio/virtio_ccw.c b/drivers/s390/virtio/virtio_ccw.c
index c188e4f20ca3..97e51c34e6cf 100644
--- a/drivers/s390/virtio/virtio_ccw.c
+++ b/drivers/s390/virtio/virtio_ccw.c
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
ccw->flags = 0;
ccw->count = sizeof(status);
ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+ /* We use ssch for setting the status which is a serializing
+ * instruction that guarantees the memory writes have
+ * completed before ssch.
+ */
ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
/* Write failed? We assume status is unchanged. */
if (ret)
diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c
index aa1eb5132767..95fac4c97c8b 100644
--- a/drivers/virtio/virtio.c
+++ b/drivers/virtio/virtio.c
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
* */
void virtio_reset_device(struct virtio_device *dev)
{
+ /*
+ * The below virtio_synchronize_cbs() guarantees that any
+ * interrupt for this line arriving after
+ * virtio_synchronize_vqs() has completed is guaranteed to see
+ * vq->broken as true.
+ */
+ virtio_break_device(dev);

So make this conditional

Quoted 70 lines

+ virtio_synchronize_cbs(dev);
+
dev->config->reset(dev);
}
EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
dev->config_enabled = false;
dev->config_change_pending = false;

+ INIT_LIST_HEAD(&dev->vqs);
+ spin_lock_init(&dev->vqs_list_lock);
+
/* We always start by resetting the device, in case a previous
* driver messed it up. This also tests that code path a little. */
virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
/* Acknowledge that we've seen the device. */
virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);

- INIT_LIST_HEAD(&dev->vqs);
- spin_lock_init(&dev->vqs_list_lock);
-
/*
* device_add() causes the bus infrastructure to look for a matching
* driver.
diff --git a/drivers/virtio/virtio_mmio.c b/drivers/virtio/virtio_mmio.c
index c9699a59f93c..f9a36bc7ac27 100644
--- a/drivers/virtio/virtio_mmio.c
+++ b/drivers/virtio/virtio_mmio.c
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
/* We should never be setting status to 0. */
BUG_ON(status == 0);

+ /*
+ * Per memory-barriers.txt, wmb() is not needed to guarantee
+ * that the the cache coherent memory writes have completed
+ * before writing to the MMIO region.
+ */
writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
}

diff --git a/drivers/virtio/virtio_pci_modern_dev.c b/drivers/virtio/virtio_pci_modern_dev.c
index 4093f9cca7a6..a0fa14f28a7f 100644
--- a/drivers/virtio/virtio_pci_modern_dev.c
+++ b/drivers/virtio/virtio_pci_modern_dev.c
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
{
struct virtio_pci_common_cfg __iomem *cfg = mdev->common;

+ /*
+ * Per memory-barriers.txt, wmb() is not needed to guarantee
+ * that the the cache coherent memory writes have completed
+ * before writing to the MMIO region.
+ */
vp_iowrite8(status, &cfg->device_status);
}
EXPORT_SYMBOL_GPL(vp_modern_set_status);
diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index 9c231e1fded7..13a7348cedff 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
vq->we_own_ring = true;
vq->notify = notify;
vq->weak_barriers = weak_barriers;
- vq->broken = false;
+ vq->broken = true;
vq->last_used_idx = 0;
vq->event_triggered = false;
vq->num_added = 0;

and make this conditional

Quoted 23 lines

@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
return IRQ_NONE;
}

- if (unlikely(vq->broken))
- return IRQ_HANDLED;
+ if (unlikely(vq->broken)) {
+ dev_warn_once(&vq->vq.vdev->dev,
+ "virtio vring IRQ raised before DRIVER_OK");
+ return IRQ_NONE;
+ }

/* Just a hint for performance: so it's ok that this can be racy! */
if (vq->event)
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
vq->we_own_ring = false;
vq->notify = notify;
vq->weak_barriers = weak_barriers;
- vq->broken = false;
+ vq->broken = true;
vq->last_used_idx = 0;
vq->event_triggered = false;
vq->num_added = 0;

and make this conditional

Quoted 33 lines

diff --git a/include/linux/virtio_config.h b/include/linux/virtio_config.h
index 25be018810a7..d4edfd7d91bb 100644
--- a/include/linux/virtio_config.h
+++ b/include/linux/virtio_config.h
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
unsigned status = dev->config->get_status(dev);

BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+ /*
+ * The virtio_synchronize_cbs() makes sure vring_interrupt()
+ * will see the driver specific setup if it sees vq->broken
+ * as false (even if the notifications come before DRIVER_OK).
+ */
+ virtio_synchronize_cbs(dev);
+ __virtio_unbreak_device(dev);
+ /*
+ * The transport should ensure the visibility of vq->broken
+ * before setting DRIVER_OK. See the comments for the transport
+ * specific set_status() method.
+ *
+ * A well behaved device will only notify a virtqueue after
+ * DRIVER_OK, this means the device should "see" the coherenct
+ * memory write that set vq->broken as false which is done by
+ * the driver when it sees DRIVER_OK, then the following
+ * driver's vring_interrupt() will see vq->broken as false so
+ * we won't lose any notification.
+ */
dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
}

--
2.25.1

#35 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

In reply to Jason Wang
Quoted 174 lines · Jason Wang
Quoted 107 lines · Michael S. Tsirkin
Quoted 105 lines · Jason Wang
Quoted 103 lines · Jason Wang
Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang
Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 7 lines
Quoted 4 lines

btw, hci has an open and close method and we do rx refill in hdev->open, so we're probably fine here.

Thanks

Sounds good. Now to audit the rest of them from this POV ;)

Adding maintainers.

Quoted 2 lines

drivers/i2c/busses/i2c-virtio.c

It looks to me the device could be used immediately after i2c_add_adapter() return. So we probably need to add virtio_device_ready() before that. Fortunately, there's no rx vq in i2c and the callback looks safe if the callback is called before the i2c registration and after virtio_device_ready().

Quoted 1 line

drivers/net/caif/caif_virtio.c

A networking device, RX is backed by vringh so we don't need to refill. TX is backed by virtio and is available until ndo_open. So it's fine to let the core to set DRIVER_OK after probe().

Quoted 1 line

drivers/nvdimm/virtio_pmem.c

It doesn't use interrupt so far, so it has nothing to do with the IRQ hardening.

But the device could be used by the subsystem immediately after nvdimm_pmem_region_create(), this means the flush could be issued before DRIVER_OK. We need virtio_device_ready() before. We don't have a RX virtqueue and the callback looks safe if the callback is called after virtio_device_ready() but before the nvdimm region creating.

And it looks to me there's a race between the assignment of provider_data and virtio_pmem_flush(). If the flush was issued before the assignment we will end up with a NULL pointer dereference. This is something we need to fix.

Quoted 1 line

arm_scmi

It looks to me the singleton device could be used by SCMI immediately after

        /* Ensure initialized scmi_vdev is visible */
        smp_store_mb(scmi_vdev, vdev);

So we probably need to do virtio_device_ready() before that. It has an optional rx queue but the filling is done after the above assignment, so it's safe. And the callback looks safe is a callback is triggered after virtio_device_ready() buy before the above assignment.

Quoted 2 lines

virtio_rpmsg_bus.c

This is somehow more complicated. It has an rx queue, the rx filling is done before virtio_device_ready() but the kick is done after. And it looks to me the device could be used by subsystem immediately rpmsg_virtio_add_ctrl_dev() returns.

This means, if we do virtio_device_ready() after rpmsg_virtio_add_ctrl_dev(), we may get kick before DRIVER_OK. If we do virtio_device_ready() before rpmsg_virtio_add_ctrl_dev(), there's a race between the callbacks and rpmsg_virtio_add_ctrl_dev() that could be exploited.

Quoted 3 lines

It requires more thoughts.

Thanks

I think at this point let's do it before so we at least do not get a regression with your patches, add a big comment and work on fixing properly in the next Linux version. Do you think you can commit to a full fix in the next linux version?

Quoted 273 lines
Quoted 273 lines
Quoted 272 lines
Quoted 272 lines
Quoted 8 lines
Quoted 6 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

Quoted 35 lines
Quoted 27 lines
Quoted 21 lines
Quoted 7 lines
Quoted 4 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

Quoted 8 lines

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well?

Kick itself is not blocked.

Quoted 1 line

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

        virtio_device_ready(vdev);

        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

Quoted 187 lines
Quoted 186 lines
Quoted 186 lines
Quoted 185 lines
Quoted 185 lines
Quoted 184 lines
Quoted 184 lines
Quoted 38 lines








---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 33 lines
include/linux/virtio_config.h +20 -1
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
--

2.25.1

author
#36 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 176 lines · Michael S. Tsirkin
Quoted 174 lines · Jason Wang
Quoted 107 lines · Michael S. Tsirkin
Quoted 105 lines · Jason Wang
Quoted 103 lines · Jason Wang
Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang
Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 35 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for
virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ
that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ
hardening. The vq->broken is set to true during both initialization
and reset. And the vq->broken is set to false in
virtio_device_ready(). Then vring_interrupt() can check and return
when vq->broken is true. And in this case, switch to return IRQ_NONE
to let the interrupt core aware of such invalid interrupt to prevent
IRQ storm.

The reason of using a per queue variable instead of a per device one
is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the
config interrupt hardening is already done in commit 22b7050a024d7
("virtio: defer config changed notifications"). But the method that is
used by config interrupt can't be reused by the vring interrupt
handler because it uses spinlock to do the synchronization which is
expensive.

Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Cc: Vineeth Vijayan <[email protected]>
Cc: Peter Oberparleiter <[email protected]>
Cc: [email protected]
Signed-off-by: Jason Wang <[email protected]>

Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 7 lines
Quoted 4 lines

btw, hci has an open and close method and we do rx refill in hdev->open, so we're probably fine here.

Thanks

Sounds good. Now to audit the rest of them from this POV ;)

Adding maintainers.

Quoted 2 lines

drivers/i2c/busses/i2c-virtio.c

It looks to me the device could be used immediately after i2c_add_adapter() return. So we probably need to add virtio_device_ready() before that. Fortunately, there's no rx vq in i2c and the callback looks safe if the callback is called before the i2c registration and after virtio_device_ready().

Quoted 1 line

drivers/net/caif/caif_virtio.c

A networking device, RX is backed by vringh so we don't need to refill. TX is backed by virtio and is available until ndo_open. So it's fine to let the core to set DRIVER_OK after probe().

Quoted 1 line

drivers/nvdimm/virtio_pmem.c

It doesn't use interrupt so far, so it has nothing to do with the IRQ hardening.

But the device could be used by the subsystem immediately after nvdimm_pmem_region_create(), this means the flush could be issued before DRIVER_OK. We need virtio_device_ready() before. We don't have a RX virtqueue and the callback looks safe if the callback is called after virtio_device_ready() but before the nvdimm region creating.

And it looks to me there's a race between the assignment of provider_data and virtio_pmem_flush(). If the flush was issued before the assignment we will end up with a NULL pointer dereference. This is something we need to fix.

Quoted 1 line

arm_scmi

It looks to me the singleton device could be used by SCMI immediately after

        /* Ensure initialized scmi_vdev is visible */
        smp_store_mb(scmi_vdev, vdev);

So we probably need to do virtio_device_ready() before that. It has an optional rx queue but the filling is done after the above assignment, so it's safe. And the callback looks safe is a callback is triggered after virtio_device_ready() buy before the above assignment.

Quoted 2 lines

virtio_rpmsg_bus.c

This is somehow more complicated. It has an rx queue, the rx filling is done before virtio_device_ready() but the kick is done after. And it looks to me the device could be used by subsystem immediately rpmsg_virtio_add_ctrl_dev() returns.

This means, if we do virtio_device_ready() after rpmsg_virtio_add_ctrl_dev(), we may get kick before DRIVER_OK. If we do virtio_device_ready() before rpmsg_virtio_add_ctrl_dev(), there's a race between the callbacks and rpmsg_virtio_add_ctrl_dev() that could be exploited.

Quoted 8 lines
Quoted 3 lines

It requires more thoughts.

Thanks

I think at this point let's do it before so we at least do not get a regression with your patches, add a big comment and work on fixing properly in the next Linux version. Do you think you can commit to a full fix in the next linux version?

I think it should be ok.

If I understand you correctly, you meant to disable the hardening in this release?

(Actually, my understanding is that since we are developing mainline instead of a downstream version with a hardening features, bug reports are somehow expected, especially consider most of the bugs are not related to hardening itself)

Thanks

drivers/virtio/virtio_ring.c +1 -1
index 13a7348cedff..7ef3115efbad 100644
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
        vq->we_own_ring = true;
        vq->notify = notify;
        vq->weak_barriers = weak_barriers;
-       vq->broken = true;
+       vq->broken = false;
        vq->last_used_idx = 0;
        vq->event_triggered = false;
        vq->num_added = 0;
Quoted 276 lines
Quoted 275 lines
Quoted 275 lines
Quoted 274 lines
Quoted 274 lines
Quoted 10 lines
Quoted 8 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

Quoted 35 lines
Quoted 27 lines
Quoted 21 lines
Quoted 7 lines
Quoted 4 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

Quoted 8 lines

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well?

Kick itself is not blocked.

Quoted 1 line

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

        virtio_device_ready(vdev);

        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

Quoted 187 lines
Quoted 186 lines
Quoted 186 lines
Quoted 185 lines
Quoted 185 lines
Quoted 184 lines
Quoted 184 lines
Quoted 48 lines

---
drivers/s390/virtio/virtio_ccw.c | 4 ++++
drivers/virtio/virtio.c | 15 ++++++++++++---
drivers/virtio/virtio_mmio.c | 5 +++++
drivers/virtio/virtio_pci_modern_dev.c | 5 +++++
drivers/virtio/virtio_ring.c | 11 +++++++----
include/linux/virtio_config.h | 20 ++++++++++++++++++++
6 files changed, 53 insertions(+), 7 deletions(-)

diff --git a/drivers/s390/virtio/virtio_ccw.c b/drivers/s390/virtio/virtio_ccw.c
index c188e4f20ca3..97e51c34e6cf 100644
--- a/drivers/s390/virtio/virtio_ccw.c
+++ b/drivers/s390/virtio/virtio_ccw.c
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
ccw->flags = 0;
ccw->count = sizeof(status);
ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+ /* We use ssch for setting the status which is a serializing
+ * instruction that guarantees the memory writes have
+ * completed before ssch.
+ */
ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
/* Write failed? We assume status is unchanged. */
if (ret)
diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c
index aa1eb5132767..95fac4c97c8b 100644
--- a/drivers/virtio/virtio.c
+++ b/drivers/virtio/virtio.c
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
* */
void virtio_reset_device(struct virtio_device *dev)
{
+ /*
+ * The below virtio_synchronize_cbs() guarantees that any
+ * interrupt for this line arriving after
+ * virtio_synchronize_vqs() has completed is guaranteed to see
+ * vq->broken as true.
+ */
+ virtio_break_device(dev);

So make this conditional

Quoted 70 lines

+ virtio_synchronize_cbs(dev);
+
dev->config->reset(dev);
}
EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
dev->config_enabled = false;
dev->config_change_pending = false;

+ INIT_LIST_HEAD(&dev->vqs);
+ spin_lock_init(&dev->vqs_list_lock);
+
/* We always start by resetting the device, in case a previous
* driver messed it up. This also tests that code path a little. */
virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
/* Acknowledge that we've seen the device. */
virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);

- INIT_LIST_HEAD(&dev->vqs);
- spin_lock_init(&dev->vqs_list_lock);
-
/*
* device_add() causes the bus infrastructure to look for a matching
* driver.
diff --git a/drivers/virtio/virtio_mmio.c b/drivers/virtio/virtio_mmio.c
index c9699a59f93c..f9a36bc7ac27 100644
--- a/drivers/virtio/virtio_mmio.c
+++ b/drivers/virtio/virtio_mmio.c
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
/* We should never be setting status to 0. */
BUG_ON(status == 0);

+ /*
+ * Per memory-barriers.txt, wmb() is not needed to guarantee
+ * that the the cache coherent memory writes have completed
+ * before writing to the MMIO region.
+ */
writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
}

diff --git a/drivers/virtio/virtio_pci_modern_dev.c b/drivers/virtio/virtio_pci_modern_dev.c
index 4093f9cca7a6..a0fa14f28a7f 100644
--- a/drivers/virtio/virtio_pci_modern_dev.c
+++ b/drivers/virtio/virtio_pci_modern_dev.c
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
{
struct virtio_pci_common_cfg __iomem *cfg = mdev->common;

+ /*
+ * Per memory-barriers.txt, wmb() is not needed to guarantee
+ * that the the cache coherent memory writes have completed
+ * before writing to the MMIO region.
+ */
vp_iowrite8(status, &cfg->device_status);
}
EXPORT_SYMBOL_GPL(vp_modern_set_status);
diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index 9c231e1fded7..13a7348cedff 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
vq->we_own_ring = true;
vq->notify = notify;
vq->weak_barriers = weak_barriers;
- vq->broken = false;
+ vq->broken = true;
vq->last_used_idx = 0;
vq->event_triggered = false;
vq->num_added = 0;

and make this conditional

Quoted 23 lines

@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
return IRQ_NONE;
}

- if (unlikely(vq->broken))
- return IRQ_HANDLED;
+ if (unlikely(vq->broken)) {
+ dev_warn_once(&vq->vq.vdev->dev,
+ "virtio vring IRQ raised before DRIVER_OK");
+ return IRQ_NONE;
+ }

/* Just a hint for performance: so it's ok that this can be racy! */
if (vq->event)
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
vq->we_own_ring = false;
vq->notify = notify;
vq->weak_barriers = weak_barriers;
- vq->broken = false;
+ vq->broken = true;
vq->last_used_idx = 0;
vq->event_triggered = false;
vq->num_added = 0;

and make this conditional

Quoted 33 lines

diff --git a/include/linux/virtio_config.h b/include/linux/virtio_config.h
index 25be018810a7..d4edfd7d91bb 100644
--- a/include/linux/virtio_config.h
+++ b/include/linux/virtio_config.h
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
unsigned status = dev->config->get_status(dev);

BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+ /*
+ * The virtio_synchronize_cbs() makes sure vring_interrupt()
+ * will see the driver specific setup if it sees vq->broken
+ * as false (even if the notifications come before DRIVER_OK).
+ */
+ virtio_synchronize_cbs(dev);
+ __virtio_unbreak_device(dev);
+ /*
+ * The transport should ensure the visibility of vq->broken
+ * before setting DRIVER_OK. See the comments for the transport
+ * specific set_status() method.
+ *
+ * A well behaved device will only notify a virtqueue after
+ * DRIVER_OK, this means the device should "see" the coherenct
+ * memory write that set vq->broken as false which is done by
+ * the driver when it sees DRIVER_OK, then the following
+ * driver's vring_interrupt() will see vq->broken as false so
+ * we won't lose any notification.
+ */
dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
}

--
2.25.1

#37 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 188 lines · Jason Wang
Quoted 176 lines · Michael S. Tsirkin
Quoted 174 lines · Jason Wang
Quoted 107 lines · Michael S. Tsirkin
Quoted 105 lines · Jason Wang
Quoted 103 lines · Jason Wang
Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang
Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Jason Wang
Quoted 45 lines · Michael S. Tsirkin

On Fri, May 27, 2022 at 02:01:19PM +0800, Jason Wang wrote:
This is a rework on the previous IRQ hardening that is done for
virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ
that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ
hardening. The vq->broken is set to true during both initialization
and reset. And the vq->broken is set to false in
virtio_device_ready(). Then vring_interrupt() can check and return
when vq->broken is true. And in this case, switch to return IRQ_NONE
to let the interrupt core aware of such invalid interrupt to prevent
IRQ storm.

The reason of using a per queue variable instead of a per device one
is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the
config interrupt hardening is already done in commit 22b7050a024d7
("virtio: defer config changed notifications"). But the method that is
used by config interrupt can't be reused by the vring interrupt
handler because it uses spinlock to do the synchronization which is
expensive.

Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Cc: Vineeth Vijayan <[email protected]>
Cc: Peter Oberparleiter <[email protected]>
Cc: [email protected]
Signed-off-by: Jason Wang <[email protected]>

Jason, I am really concerned by all the fallout.
I propose adding a flag to suppress the hardening -
this will be a debugging aid and a work around for
users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 7 lines
Quoted 4 lines

btw, hci has an open and close method and we do rx refill in hdev->open, so we're probably fine here.

Thanks

Sounds good. Now to audit the rest of them from this POV ;)

Adding maintainers.

Quoted 2 lines

drivers/i2c/busses/i2c-virtio.c

It looks to me the device could be used immediately after i2c_add_adapter() return. So we probably need to add virtio_device_ready() before that. Fortunately, there's no rx vq in i2c and the callback looks safe if the callback is called before the i2c registration and after virtio_device_ready().

Quoted 1 line

drivers/net/caif/caif_virtio.c

A networking device, RX is backed by vringh so we don't need to refill. TX is backed by virtio and is available until ndo_open. So it's fine to let the core to set DRIVER_OK after probe().

Quoted 1 line

drivers/nvdimm/virtio_pmem.c

It doesn't use interrupt so far, so it has nothing to do with the IRQ hardening.

But the device could be used by the subsystem immediately after nvdimm_pmem_region_create(), this means the flush could be issued before DRIVER_OK. We need virtio_device_ready() before. We don't have a RX virtqueue and the callback looks safe if the callback is called after virtio_device_ready() but before the nvdimm region creating.

And it looks to me there's a race between the assignment of provider_data and virtio_pmem_flush(). If the flush was issued before the assignment we will end up with a NULL pointer dereference. This is something we need to fix.

Quoted 1 line

arm_scmi

It looks to me the singleton device could be used by SCMI immediately after

        /* Ensure initialized scmi_vdev is visible */
        smp_store_mb(scmi_vdev, vdev);

So we probably need to do virtio_device_ready() before that. It has an optional rx queue but the filling is done after the above assignment, so it's safe. And the callback looks safe is a callback is triggered after virtio_device_ready() buy before the above assignment.

Quoted 2 lines

virtio_rpmsg_bus.c

This is somehow more complicated. It has an rx queue, the rx filling is done before virtio_device_ready() but the kick is done after. And it looks to me the device could be used by subsystem immediately rpmsg_virtio_add_ctrl_dev() returns.

This means, if we do virtio_device_ready() after rpmsg_virtio_add_ctrl_dev(), we may get kick before DRIVER_OK. If we do virtio_device_ready() before rpmsg_virtio_add_ctrl_dev(), there's a race between the callbacks and rpmsg_virtio_add_ctrl_dev() that could be exploited.

Quoted 8 lines
Quoted 3 lines

It requires more thoughts.

Thanks

I think at this point let's do it before so we at least do not get a regression with your patches, add a big comment and work on fixing properly in the next Linux version. Do you think you can commit to a full fix in the next linux version?

I think it should be ok.

Quoted 7 lines

If I understand you correctly, you meant to disable the hardening in this release?

(Actually, my understanding is that since we are developing mainline instead of a downstream version with a hardening features, bug reports are somehow expected, especially consider most of the bugs are not related to hardening itself)

Absolutely. Question is do you think we can fix everything by the release? At least for rpmsg we don't seem to have a handle on it yet.

Quoted 7 lines

Thanks

drivers/virtio/virtio_ring.c +0 -0
index 13a7348cedff..7ef3115efbad 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
        vq->we_own_ring = true;
        vq->notify = notify;
        vq->weak_barriers = weak_barriers;
-       vq->broken = true;
+       vq->broken = false;
        vq->last_used_idx = 0;
        vq->event_triggered = false;
        vq->num_added = 0;

and drop it on reset?

Quoted 276 lines
Quoted 276 lines
Quoted 275 lines
Quoted 275 lines
Quoted 274 lines
Quoted 274 lines
Quoted 10 lines
Quoted 8 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

Quoted 35 lines
Quoted 27 lines
Quoted 21 lines
Quoted 7 lines
Quoted 4 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

Quoted 8 lines

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well?

Kick itself is not blocked.

Quoted 1 line

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

        virtio_device_ready(vdev);

        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

Quoted 187 lines
Quoted 186 lines
Quoted 186 lines
Quoted 185 lines
Quoted 185 lines
Quoted 184 lines
Quoted 184 lines

---
drivers/s390/virtio/virtio_ccw.c | 4 ++++
drivers/virtio/virtio.c | 15 ++++++++++++---
drivers/virtio/virtio_mmio.c | 5 +++++
drivers/virtio/virtio_pci_modern_dev.c | 5 +++++
drivers/virtio/virtio_ring.c | 11 +++++++----
include/linux/virtio_config.h | 20 ++++++++++++++++++++
6 files changed, 53 insertions(+), 7 deletions(-)

diff --git a/drivers/s390/virtio/virtio_ccw.c b/drivers/s390/virtio/virtio_ccw.c
index c188e4f20ca3..97e51c34e6cf 100644
--- a/drivers/s390/virtio/virtio_ccw.c
+++ b/drivers/s390/virtio/virtio_ccw.c
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
ccw->flags = 0;
ccw->count = sizeof(status);
ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+ /* We use ssch for setting the status which is a serializing
+ * instruction that guarantees the memory writes have
+ * completed before ssch.
+ */
ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
/* Write failed? We assume status is unchanged. */
if (ret)
diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c
index aa1eb5132767..95fac4c97c8b 100644
--- a/drivers/virtio/virtio.c
+++ b/drivers/virtio/virtio.c
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
* */
void virtio_reset_device(struct virtio_device *dev)
{
+ /*
+ * The below virtio_synchronize_cbs() guarantees that any
+ * interrupt for this line arriving after
+ * virtio_synchronize_vqs() has completed is guaranteed to see
+ * vq->broken as true.
+ */
+ virtio_break_device(dev);

So make this conditional

+ virtio_synchronize_cbs(dev);
+
dev->config->reset(dev);
}
EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
dev->config_enabled = false;
dev->config_change_pending = false;

+ INIT_LIST_HEAD(&dev->vqs);
+ spin_lock_init(&dev->vqs_list_lock);
+
/* We always start by resetting the device, in case a previous
* driver messed it up. This also tests that code path a little. */
virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
/* Acknowledge that we've seen the device. */
virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);

- INIT_LIST_HEAD(&dev->vqs);
- spin_lock_init(&dev->vqs_list_lock);
-
/*
* device_add() causes the bus infrastructure to look for a matching
* driver.
diff --git a/drivers/virtio/virtio_mmio.c b/drivers/virtio/virtio_mmio.c
index c9699a59f93c..f9a36bc7ac27 100644
--- a/drivers/virtio/virtio_mmio.c
+++ b/drivers/virtio/virtio_mmio.c
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
/* We should never be setting status to 0. */
BUG_ON(status == 0);

+ /*
+ * Per memory-barriers.txt, wmb() is not needed to guarantee
+ * that the the cache coherent memory writes have completed
+ * before writing to the MMIO region.
+ */
writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
}

diff --git a/drivers/virtio/virtio_pci_modern_dev.c b/drivers/virtio/virtio_pci_modern_dev.c
index 4093f9cca7a6..a0fa14f28a7f 100644
--- a/drivers/virtio/virtio_pci_modern_dev.c
+++ b/drivers/virtio/virtio_pci_modern_dev.c
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
{
struct virtio_pci_common_cfg __iomem *cfg = mdev->common;

+ /*
+ * Per memory-barriers.txt, wmb() is not needed to guarantee
+ * that the the cache coherent memory writes have completed
+ * before writing to the MMIO region.
+ */
vp_iowrite8(status, &cfg->device_status);
}
EXPORT_SYMBOL_GPL(vp_modern_set_status);
diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index 9c231e1fded7..13a7348cedff 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
vq->we_own_ring = true;
vq->notify = notify;
vq->weak_barriers = weak_barriers;
- vq->broken = false;
+ vq->broken = true;
vq->last_used_idx = 0;
vq->event_triggered = false;
vq->num_added = 0;

and make this conditional

@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
return IRQ_NONE;
}

- if (unlikely(vq->broken))
- return IRQ_HANDLED;
+ if (unlikely(vq->broken)) {
+ dev_warn_once(&vq->vq.vdev->dev,
+ "virtio vring IRQ raised before DRIVER_OK");
+ return IRQ_NONE;
+ }

/* Just a hint for performance: so it's ok that this can be racy! */
if (vq->event)
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
vq->we_own_ring = false;
vq->notify = notify;
vq->weak_barriers = weak_barriers;
- vq->broken = false;
+ vq->broken = true;
vq->last_used_idx = 0;
vq->event_triggered = false;
vq->num_added = 0;

and make this conditional

diff --git a/include/linux/virtio_config.h b/include/linux/virtio_config.h
index 25be018810a7..d4edfd7d91bb 100644
--- a/include/linux/virtio_config.h
+++ b/include/linux/virtio_config.h
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
unsigned status = dev->config->get_status(dev);

BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+ /*
+ * The virtio_synchronize_cbs() makes sure vring_interrupt()
+ * will see the driver specific setup if it sees vq->broken
+ * as false (even if the notifications come before DRIVER_OK).
+ */
+ virtio_synchronize_cbs(dev);
+ __virtio_unbreak_device(dev);
+ /*
+ * The transport should ensure the visibility of vq->broken
+ * before setting DRIVER_OK. See the comments for the transport
+ * specific set_status() method.
+ *
+ * A well behaved device will only notify a virtqueue after
+ * DRIVER_OK, this means the device should "see" the coherenct
+ * memory write that set vq->broken as false which is done by
+ * the driver when it sees DRIVER_OK, then the following
+ * driver's vring_interrupt() will see vq->broken as false so
+ * we won't lose any notification.
+ */
dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
}

--
2.25.1

author
#38 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 193 lines · Michael S. Tsirkin
Quoted 191 lines · Jason Wang
Quoted 176 lines · Michael S. Tsirkin
Quoted 174 lines · Jason Wang
Quoted 107 lines · Michael S. Tsirkin
Quoted 105 lines · Jason Wang
Quoted 103 lines · Jason Wang
Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang
Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Jason Wang

On Sat, Jun 11, 2022 at 1:12 PM Michael S. Tsirkin <[email protected]> wrote:

On Fri, May 27, 2022 at 02:01:19PM +0800, Jason Wang wrote:
This is a rework on the previous IRQ hardening that is done for
virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ
that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ
hardening. The vq->broken is set to true during both initialization
and reset. And the vq->broken is set to false in
virtio_device_ready(). Then vring_interrupt() can check and return
when vq->broken is true. And in this case, switch to return IRQ_NONE
to let the interrupt core aware of such invalid interrupt to prevent
IRQ storm.

The reason of using a per queue variable instead of a per device one
is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the
config interrupt hardening is already done in commit 22b7050a024d7
("virtio: defer config changed notifications"). But the method that is
used by config interrupt can't be reused by the vring interrupt
handler because it uses spinlock to do the synchronization which is
expensive.

Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Cc: Vineeth Vijayan <[email protected]>
Cc: Peter Oberparleiter <[email protected]>
Cc: [email protected]
Signed-off-by: Jason Wang <[email protected]>

Jason, I am really concerned by all the fallout.
I propose adding a flag to suppress the hardening -
this will be a debugging aid and a work around for
users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it
won't be used by the users so there's no way for us to receive the bug
report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we
simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 7 lines
Quoted 4 lines

btw, hci has an open and close method and we do rx refill in hdev->open, so we're probably fine here.

Thanks

Sounds good. Now to audit the rest of them from this POV ;)

Adding maintainers.

Quoted 2 lines

drivers/i2c/busses/i2c-virtio.c

It looks to me the device could be used immediately after i2c_add_adapter() return. So we probably need to add virtio_device_ready() before that. Fortunately, there's no rx vq in i2c and the callback looks safe if the callback is called before the i2c registration and after virtio_device_ready().

Quoted 1 line

drivers/net/caif/caif_virtio.c

A networking device, RX is backed by vringh so we don't need to refill. TX is backed by virtio and is available until ndo_open. So it's fine to let the core to set DRIVER_OK after probe().

Quoted 1 line

drivers/nvdimm/virtio_pmem.c

It doesn't use interrupt so far, so it has nothing to do with the IRQ hardening.

But the device could be used by the subsystem immediately after nvdimm_pmem_region_create(), this means the flush could be issued before DRIVER_OK. We need virtio_device_ready() before. We don't have a RX virtqueue and the callback looks safe if the callback is called after virtio_device_ready() but before the nvdimm region creating.

And it looks to me there's a race between the assignment of provider_data and virtio_pmem_flush(). If the flush was issued before the assignment we will end up with a NULL pointer dereference. This is something we need to fix.

Quoted 1 line

arm_scmi

It looks to me the singleton device could be used by SCMI immediately after

        /* Ensure initialized scmi_vdev is visible */
        smp_store_mb(scmi_vdev, vdev);

So we probably need to do virtio_device_ready() before that. It has an optional rx queue but the filling is done after the above assignment, so it's safe. And the callback looks safe is a callback is triggered after virtio_device_ready() buy before the above assignment.

Quoted 2 lines

virtio_rpmsg_bus.c

This is somehow more complicated. It has an rx queue, the rx filling is done before virtio_device_ready() but the kick is done after. And it looks to me the device could be used by subsystem immediately rpmsg_virtio_add_ctrl_dev() returns.

This means, if we do virtio_device_ready() after rpmsg_virtio_add_ctrl_dev(), we may get kick before DRIVER_OK. If we do virtio_device_ready() before rpmsg_virtio_add_ctrl_dev(), there's a race between the callbacks and rpmsg_virtio_add_ctrl_dev() that could be exploited.

Quoted 8 lines
Quoted 3 lines

It requires more thoughts.

Thanks

I think at this point let's do it before so we at least do not get a regression with your patches, add a big comment and work on fixing properly in the next Linux version. Do you think you can commit to a full fix in the next linux version?

I think it should be ok.

If I understand you correctly, you meant to disable the hardening in this release?

Quoted 8 lines
Quoted 4 lines

(Actually, my understanding is that since we are developing mainline instead of a downstream version with a hardening features, bug reports are somehow expected, especially consider most of the bugs are not related to hardening itself)

Absolutely. Question is do you think we can fix everything by the release?

Probably not, I'm auditing all the virtio drivers and it seems we have many issues:

1) race between subsystem registration/use and virtio_device_ready()
2) race between notifications and subsystem registerstiation/use

And it looks to me even virtio-net has this race.

So I think I will post a patch to disable this like below for this release.

Quoted 1 line

At least for rpmsg we don't seem to have a handle on it yet.

Yes.

Quoted 4 lines
Quoted 4 lines

Thanks

Quoted 16 lines

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 13 lines
drivers/virtio/virtio_ring.c +1 -1
index 13a7348cedff..7ef3115efbad 100644
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
        vq->we_own_ring = true;
        vq->notify = notify;
        vq->weak_barriers = weak_barriers;
-       vq->broken = true;
+       vq->broken = false;
        vq->last_used_idx = 0;
        vq->event_triggered = false;
        vq->num_added = 0;

and drop it on reset?

Right.

Thanks

Quoted 278 lines
Quoted 277 lines
Quoted 277 lines
Quoted 276 lines
Quoted 276 lines
Quoted 275 lines
Quoted 275 lines
Quoted 11 lines
Quoted 9 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

Quoted 35 lines
Quoted 27 lines
Quoted 21 lines
Quoted 7 lines
Quoted 4 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

Quoted 8 lines

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well?

Kick itself is not blocked.

Quoted 1 line

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

        virtio_device_ready(vdev);

        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

Quoted 187 lines
Quoted 186 lines
Quoted 186 lines
Quoted 185 lines
Quoted 185 lines
Quoted 184 lines

---
drivers/s390/virtio/virtio_ccw.c | 4 ++++
drivers/virtio/virtio.c | 15 ++++++++++++---
drivers/virtio/virtio_mmio.c | 5 +++++
drivers/virtio/virtio_pci_modern_dev.c | 5 +++++
drivers/virtio/virtio_ring.c | 11 +++++++----
include/linux/virtio_config.h | 20 ++++++++++++++++++++
6 files changed, 53 insertions(+), 7 deletions(-)

diff --git a/drivers/s390/virtio/virtio_ccw.c b/drivers/s390/virtio/virtio_ccw.c
index c188e4f20ca3..97e51c34e6cf 100644
--- a/drivers/s390/virtio/virtio_ccw.c
+++ b/drivers/s390/virtio/virtio_ccw.c
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
ccw->flags = 0;
ccw->count = sizeof(status);
ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+ /* We use ssch for setting the status which is a serializing
+ * instruction that guarantees the memory writes have
+ * completed before ssch.
+ */
ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
/* Write failed? We assume status is unchanged. */
if (ret)
diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c
index aa1eb5132767..95fac4c97c8b 100644
--- a/drivers/virtio/virtio.c
+++ b/drivers/virtio/virtio.c
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
* */
void virtio_reset_device(struct virtio_device *dev)
{
+ /*
+ * The below virtio_synchronize_cbs() guarantees that any
+ * interrupt for this line arriving after
+ * virtio_synchronize_vqs() has completed is guaranteed to see
+ * vq->broken as true.
+ */
+ virtio_break_device(dev);

So make this conditional

+ virtio_synchronize_cbs(dev);
+
dev->config->reset(dev);
}
EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
dev->config_enabled = false;
dev->config_change_pending = false;

+ INIT_LIST_HEAD(&dev->vqs);
+ spin_lock_init(&dev->vqs_list_lock);
+
/* We always start by resetting the device, in case a previous
* driver messed it up. This also tests that code path a little. */
virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
/* Acknowledge that we've seen the device. */
virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);

- INIT_LIST_HEAD(&dev->vqs);
- spin_lock_init(&dev->vqs_list_lock);
-
/*
* device_add() causes the bus infrastructure to look for a matching
* driver.
diff --git a/drivers/virtio/virtio_mmio.c b/drivers/virtio/virtio_mmio.c
index c9699a59f93c..f9a36bc7ac27 100644
--- a/drivers/virtio/virtio_mmio.c
+++ b/drivers/virtio/virtio_mmio.c
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
/* We should never be setting status to 0. */
BUG_ON(status == 0);

+ /*
+ * Per memory-barriers.txt, wmb() is not needed to guarantee
+ * that the the cache coherent memory writes have completed
+ * before writing to the MMIO region.
+ */
writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
}

diff --git a/drivers/virtio/virtio_pci_modern_dev.c b/drivers/virtio/virtio_pci_modern_dev.c
index 4093f9cca7a6..a0fa14f28a7f 100644
--- a/drivers/virtio/virtio_pci_modern_dev.c
+++ b/drivers/virtio/virtio_pci_modern_dev.c
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
{
struct virtio_pci_common_cfg __iomem *cfg = mdev->common;

+ /*
+ * Per memory-barriers.txt, wmb() is not needed to guarantee
+ * that the the cache coherent memory writes have completed
+ * before writing to the MMIO region.
+ */
vp_iowrite8(status, &cfg->device_status);
}
EXPORT_SYMBOL_GPL(vp_modern_set_status);
diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index 9c231e1fded7..13a7348cedff 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
vq->we_own_ring = true;
vq->notify = notify;
vq->weak_barriers = weak_barriers;
- vq->broken = false;
+ vq->broken = true;
vq->last_used_idx = 0;
vq->event_triggered = false;
vq->num_added = 0;

and make this conditional

@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
return IRQ_NONE;
}

- if (unlikely(vq->broken))
- return IRQ_HANDLED;
+ if (unlikely(vq->broken)) {
+ dev_warn_once(&vq->vq.vdev->dev,
+ "virtio vring IRQ raised before DRIVER_OK");
+ return IRQ_NONE;
+ }

/* Just a hint for performance: so it's ok that this can be racy! */
if (vq->event)
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
vq->we_own_ring = false;
vq->notify = notify;
vq->weak_barriers = weak_barriers;
- vq->broken = false;
+ vq->broken = true;
vq->last_used_idx = 0;
vq->event_triggered = false;
vq->num_added = 0;

and make this conditional

diff --git a/include/linux/virtio_config.h b/include/linux/virtio_config.h
index 25be018810a7..d4edfd7d91bb 100644
--- a/include/linux/virtio_config.h
+++ b/include/linux/virtio_config.h
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
unsigned status = dev->config->get_status(dev);

BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+ /*
+ * The virtio_synchronize_cbs() makes sure vring_interrupt()
+ * will see the driver specific setup if it sees vq->broken
+ * as false (even if the notifications come before DRIVER_OK).
+ */
+ virtio_synchronize_cbs(dev);
+ __virtio_unbreak_device(dev);
+ /*
+ * The transport should ensure the visibility of vq->broken
+ * before setting DRIVER_OK. See the comments for the transport
+ * specific set_status() method.
+ *
+ * A well behaved device will only notify a virtqueue after
+ * DRIVER_OK, this means the device should "see" the coherenct
+ * memory write that set vq->broken as false which is done by
+ * the driver when it sees DRIVER_OK, then the following
+ * driver's vring_interrupt() will see vq->broken as false so
+ * we won't lose any notification.
+ */
dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
}

--
2.25.1

#39 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 203 lines · Jason Wang
Quoted 193 lines · Michael S. Tsirkin
Quoted 191 lines · Jason Wang
Quoted 176 lines · Michael S. Tsirkin
Quoted 174 lines · Jason Wang
Quoted 107 lines · Michael S. Tsirkin
Quoted 105 lines · Jason Wang
Quoted 103 lines · Jason Wang
Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang
Quoted 63 lines · Michael S. Tsirkin

On Mon, Jun 13, 2022 at 01:26:59PM +0800, Jason Wang wrote:
On Sat, Jun 11, 2022 at 1:12 PM Michael S. Tsirkin <[email protected]> wrote:

On Fri, May 27, 2022 at 02:01:19PM +0800, Jason Wang wrote:
This is a rework on the previous IRQ hardening that is done for
virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ
that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ
hardening. The vq->broken is set to true during both initialization
and reset. And the vq->broken is set to false in
virtio_device_ready(). Then vring_interrupt() can check and return
when vq->broken is true. And in this case, switch to return IRQ_NONE
to let the interrupt core aware of such invalid interrupt to prevent
IRQ storm.

The reason of using a per queue variable instead of a per device one
is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the
config interrupt hardening is already done in commit 22b7050a024d7
("virtio: defer config changed notifications"). But the method that is
used by config interrupt can't be reused by the vring interrupt
handler because it uses spinlock to do the synchronization which is
expensive.

Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Cc: Vineeth Vijayan <[email protected]>
Cc: Peter Oberparleiter <[email protected]>
Cc: [email protected]
Signed-off-by: Jason Wang <[email protected]>

Jason, I am really concerned by all the fallout.
I propose adding a flag to suppress the hardening -
this will be a debugging aid and a work around for
users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it
won't be used by the users so there's no way for us to receive the bug
report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we
simply warn instead of disable it by default.

Thanks

I meant more like a flag in struct virtio_driver.
For now, could you audit all drivers which don't call _ready?
I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 7 lines
Quoted 4 lines

btw, hci has an open and close method and we do rx refill in hdev->open, so we're probably fine here.

Thanks

Sounds good. Now to audit the rest of them from this POV ;)

Adding maintainers.

Quoted 2 lines

drivers/i2c/busses/i2c-virtio.c

It looks to me the device could be used immediately after i2c_add_adapter() return. So we probably need to add virtio_device_ready() before that. Fortunately, there's no rx vq in i2c and the callback looks safe if the callback is called before the i2c registration and after virtio_device_ready().

Quoted 1 line

drivers/net/caif/caif_virtio.c

A networking device, RX is backed by vringh so we don't need to refill. TX is backed by virtio and is available until ndo_open. So it's fine to let the core to set DRIVER_OK after probe().

Quoted 1 line

drivers/nvdimm/virtio_pmem.c

It doesn't use interrupt so far, so it has nothing to do with the IRQ hardening.

But the device could be used by the subsystem immediately after nvdimm_pmem_region_create(), this means the flush could be issued before DRIVER_OK. We need virtio_device_ready() before. We don't have a RX virtqueue and the callback looks safe if the callback is called after virtio_device_ready() but before the nvdimm region creating.

And it looks to me there's a race between the assignment of provider_data and virtio_pmem_flush(). If the flush was issued before the assignment we will end up with a NULL pointer dereference. This is something we need to fix.

Quoted 1 line

arm_scmi

It looks to me the singleton device could be used by SCMI immediately after

        /* Ensure initialized scmi_vdev is visible */
        smp_store_mb(scmi_vdev, vdev);

So we probably need to do virtio_device_ready() before that. It has an optional rx queue but the filling is done after the above assignment, so it's safe. And the callback looks safe is a callback is triggered after virtio_device_ready() buy before the above assignment.

Quoted 2 lines

virtio_rpmsg_bus.c

This is somehow more complicated. It has an rx queue, the rx filling is done before virtio_device_ready() but the kick is done after. And it looks to me the device could be used by subsystem immediately rpmsg_virtio_add_ctrl_dev() returns.

This means, if we do virtio_device_ready() after rpmsg_virtio_add_ctrl_dev(), we may get kick before DRIVER_OK. If we do virtio_device_ready() before rpmsg_virtio_add_ctrl_dev(), there's a race between the callbacks and rpmsg_virtio_add_ctrl_dev() that could be exploited.

Quoted 8 lines
Quoted 3 lines

It requires more thoughts.

Thanks

I think at this point let's do it before so we at least do not get a regression with your patches, add a big comment and work on fixing properly in the next Linux version. Do you think you can commit to a full fix in the next linux version?

I think it should be ok.

If I understand you correctly, you meant to disable the hardening in this release?

Quoted 8 lines
Quoted 4 lines

(Actually, my understanding is that since we are developing mainline instead of a downstream version with a hardening features, bug reports are somehow expected, especially consider most of the bugs are not related to hardening itself)

Absolutely. Question is do you think we can fix everything by the release?

Quoted 7 lines

Probably not, I'm auditing all the virtio drivers and it seems we have many issues:

1) race between subsystem registration/use and virtio_device_ready()
2) race between notifications and subsystem registerstiation/use

And it looks to me even virtio-net has this race.

Interesting. How does it look for virtio-net?

Quoted 1 line

So I think I will post a patch to disable this like below for this release.

However please do post patches that add device_ready as appropriate. This is basic spec compliance.

Also do you think we should do a full revert? Maybe a Kconfig option is ok for now.

Quoted 307 lines
Quoted 1 line

At least for rpmsg we don't seem to have a handle on it yet.

Yes.

Quoted 4 lines
Quoted 4 lines

Thanks

Quoted 16 lines

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 13 lines
drivers/virtio/virtio_ring.c +1 -1
index 13a7348cedff..7ef3115efbad 100644
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
        vq->we_own_ring = true;
        vq->notify = notify;
        vq->weak_barriers = weak_barriers;
-       vq->broken = true;
+       vq->broken = false;
        vq->last_used_idx = 0;
        vq->event_triggered = false;
        vq->num_added = 0;

and drop it on reset?

Right.

Thanks

Quoted 278 lines
Quoted 277 lines
Quoted 277 lines
Quoted 276 lines
Quoted 276 lines
Quoted 275 lines
Quoted 275 lines
Quoted 11 lines
Quoted 9 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

Quoted 35 lines
Quoted 27 lines
Quoted 21 lines
Quoted 7 lines
Quoted 4 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

Quoted 8 lines

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well?

Kick itself is not blocked.

Quoted 1 line

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

        virtio_device_ready(vdev);

        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

Quoted 187 lines
Quoted 186 lines
Quoted 186 lines
Quoted 185 lines
Quoted 185 lines

---
drivers/s390/virtio/virtio_ccw.c | 4 ++++
drivers/virtio/virtio.c | 15 ++++++++++++---
drivers/virtio/virtio_mmio.c | 5 +++++
drivers/virtio/virtio_pci_modern_dev.c | 5 +++++
drivers/virtio/virtio_ring.c | 11 +++++++----
include/linux/virtio_config.h | 20 ++++++++++++++++++++
6 files changed, 53 insertions(+), 7 deletions(-)

diff --git a/drivers/s390/virtio/virtio_ccw.c b/drivers/s390/virtio/virtio_ccw.c
index c188e4f20ca3..97e51c34e6cf 100644
--- a/drivers/s390/virtio/virtio_ccw.c
+++ b/drivers/s390/virtio/virtio_ccw.c
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
ccw->flags = 0;
ccw->count = sizeof(status);
ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+ /* We use ssch for setting the status which is a serializing
+ * instruction that guarantees the memory writes have
+ * completed before ssch.
+ */
ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
/* Write failed? We assume status is unchanged. */
if (ret)
diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c
index aa1eb5132767..95fac4c97c8b 100644
--- a/drivers/virtio/virtio.c
+++ b/drivers/virtio/virtio.c
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
* */
void virtio_reset_device(struct virtio_device *dev)
{
+ /*
+ * The below virtio_synchronize_cbs() guarantees that any
+ * interrupt for this line arriving after
+ * virtio_synchronize_vqs() has completed is guaranteed to see
+ * vq->broken as true.
+ */
+ virtio_break_device(dev);

So make this conditional

+ virtio_synchronize_cbs(dev);
+
dev->config->reset(dev);
}
EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
dev->config_enabled = false;
dev->config_change_pending = false;

+ INIT_LIST_HEAD(&dev->vqs);
+ spin_lock_init(&dev->vqs_list_lock);
+
/* We always start by resetting the device, in case a previous
* driver messed it up. This also tests that code path a little. */
virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
/* Acknowledge that we've seen the device. */
virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);

- INIT_LIST_HEAD(&dev->vqs);
- spin_lock_init(&dev->vqs_list_lock);
-
/*
* device_add() causes the bus infrastructure to look for a matching
* driver.
diff --git a/drivers/virtio/virtio_mmio.c b/drivers/virtio/virtio_mmio.c
index c9699a59f93c..f9a36bc7ac27 100644
--- a/drivers/virtio/virtio_mmio.c
+++ b/drivers/virtio/virtio_mmio.c
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
/* We should never be setting status to 0. */
BUG_ON(status == 0);

+ /*
+ * Per memory-barriers.txt, wmb() is not needed to guarantee
+ * that the the cache coherent memory writes have completed
+ * before writing to the MMIO region.
+ */
writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
}

diff --git a/drivers/virtio/virtio_pci_modern_dev.c b/drivers/virtio/virtio_pci_modern_dev.c
index 4093f9cca7a6..a0fa14f28a7f 100644
--- a/drivers/virtio/virtio_pci_modern_dev.c
+++ b/drivers/virtio/virtio_pci_modern_dev.c
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
{
struct virtio_pci_common_cfg __iomem *cfg = mdev->common;

+ /*
+ * Per memory-barriers.txt, wmb() is not needed to guarantee
+ * that the the cache coherent memory writes have completed
+ * before writing to the MMIO region.
+ */
vp_iowrite8(status, &cfg->device_status);
}
EXPORT_SYMBOL_GPL(vp_modern_set_status);
diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index 9c231e1fded7..13a7348cedff 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
vq->we_own_ring = true;
vq->notify = notify;
vq->weak_barriers = weak_barriers;
- vq->broken = false;
+ vq->broken = true;
vq->last_used_idx = 0;
vq->event_triggered = false;
vq->num_added = 0;

and make this conditional

@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
return IRQ_NONE;
}

- if (unlikely(vq->broken))
- return IRQ_HANDLED;
+ if (unlikely(vq->broken)) {
+ dev_warn_once(&vq->vq.vdev->dev,
+ "virtio vring IRQ raised before DRIVER_OK");
+ return IRQ_NONE;
+ }

/* Just a hint for performance: so it's ok that this can be racy! */
if (vq->event)
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
vq->we_own_ring = false;
vq->notify = notify;
vq->weak_barriers = weak_barriers;
- vq->broken = false;
+ vq->broken = true;
vq->last_used_idx = 0;
vq->event_triggered = false;
vq->num_added = 0;

and make this conditional

diff --git a/include/linux/virtio_config.h b/include/linux/virtio_config.h
index 25be018810a7..d4edfd7d91bb 100644
--- a/include/linux/virtio_config.h
+++ b/include/linux/virtio_config.h
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
unsigned status = dev->config->get_status(dev);

BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+ /*
+ * The virtio_synchronize_cbs() makes sure vring_interrupt()
+ * will see the driver specific setup if it sees vq->broken
+ * as false (even if the notifications come before DRIVER_OK).
+ */
+ virtio_synchronize_cbs(dev);
+ __virtio_unbreak_device(dev);
+ /*
+ * The transport should ensure the visibility of vq->broken
+ * before setting DRIVER_OK. See the comments for the transport
+ * specific set_status() method.
+ *
+ * A well behaved device will only notify a virtqueue after
+ * DRIVER_OK, this means the device should "see" the coherenct
+ * memory write that set vq->broken as false which is done by
+ * the driver when it sees DRIVER_OK, then the following
+ * driver's vring_interrupt() will see vq->broken as false so
+ * we won't lose any notification.
+ */
dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
}

--
2.25.1

author
#40 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 208 lines · Michael S. Tsirkin
Quoted 206 lines · Jason Wang
Quoted 193 lines · Michael S. Tsirkin
Quoted 191 lines · Jason Wang
Quoted 176 lines · Michael S. Tsirkin
Quoted 174 lines · Jason Wang
Quoted 107 lines · Michael S. Tsirkin
Quoted 105 lines · Jason Wang
Quoted 103 lines · Jason Wang
Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang

On Mon, Jun 13, 2022 at 3:23 PM Michael S. Tsirkin <[email protected]> wrote:

On Mon, Jun 13, 2022 at 01:26:59PM +0800, Jason Wang wrote:
On Sat, Jun 11, 2022 at 1:12 PM Michael S. Tsirkin <[email protected]> wrote:

On Fri, May 27, 2022 at 02:01:19PM +0800, Jason Wang wrote:
This is a rework on the previous IRQ hardening that is done for
virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ
that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ
hardening. The vq->broken is set to true during both initialization
and reset. And the vq->broken is set to false in
virtio_device_ready(). Then vring_interrupt() can check and return
when vq->broken is true. And in this case, switch to return IRQ_NONE
to let the interrupt core aware of such invalid interrupt to prevent
IRQ storm.

The reason of using a per queue variable instead of a per device one
is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the
config interrupt hardening is already done in commit 22b7050a024d7
("virtio: defer config changed notifications"). But the method that is
used by config interrupt can't be reused by the vring interrupt
handler because it uses spinlock to do the synchronization which is
expensive.

Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Cc: Vineeth Vijayan <[email protected]>
Cc: Peter Oberparleiter <[email protected]>
Cc: [email protected]
Signed-off-by: Jason Wang <[email protected]>

Jason, I am really concerned by all the fallout.
I propose adding a flag to suppress the hardening -
this will be a debugging aid and a work around for
users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it
won't be used by the users so there's no way for us to receive the bug
report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we
simply warn instead of disable it by default.

Thanks

I meant more like a flag in struct virtio_driver.
For now, could you audit all drivers which don't call _ready?
I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 7 lines
Quoted 4 lines

btw, hci has an open and close method and we do rx refill in hdev->open, so we're probably fine here.

Thanks

Sounds good. Now to audit the rest of them from this POV ;)

Adding maintainers.

Quoted 2 lines

drivers/i2c/busses/i2c-virtio.c

It looks to me the device could be used immediately after i2c_add_adapter() return. So we probably need to add virtio_device_ready() before that. Fortunately, there's no rx vq in i2c and the callback looks safe if the callback is called before the i2c registration and after virtio_device_ready().

Quoted 1 line

drivers/net/caif/caif_virtio.c

A networking device, RX is backed by vringh so we don't need to refill. TX is backed by virtio and is available until ndo_open. So it's fine to let the core to set DRIVER_OK after probe().

Quoted 1 line

drivers/nvdimm/virtio_pmem.c

It doesn't use interrupt so far, so it has nothing to do with the IRQ hardening.

But the device could be used by the subsystem immediately after nvdimm_pmem_region_create(), this means the flush could be issued before DRIVER_OK. We need virtio_device_ready() before. We don't have a RX virtqueue and the callback looks safe if the callback is called after virtio_device_ready() but before the nvdimm region creating.

And it looks to me there's a race between the assignment of provider_data and virtio_pmem_flush(). If the flush was issued before the assignment we will end up with a NULL pointer dereference. This is something we need to fix.

Quoted 1 line

arm_scmi

It looks to me the singleton device could be used by SCMI immediately after

        /* Ensure initialized scmi_vdev is visible */
        smp_store_mb(scmi_vdev, vdev);

So we probably need to do virtio_device_ready() before that. It has an optional rx queue but the filling is done after the above assignment, so it's safe. And the callback looks safe is a callback is triggered after virtio_device_ready() buy before the above assignment.

Quoted 2 lines

virtio_rpmsg_bus.c

This is somehow more complicated. It has an rx queue, the rx filling is done before virtio_device_ready() but the kick is done after. And it looks to me the device could be used by subsystem immediately rpmsg_virtio_add_ctrl_dev() returns.

This means, if we do virtio_device_ready() after rpmsg_virtio_add_ctrl_dev(), we may get kick before DRIVER_OK. If we do virtio_device_ready() before rpmsg_virtio_add_ctrl_dev(), there's a race between the callbacks and rpmsg_virtio_add_ctrl_dev() that could be exploited.

Quoted 8 lines
Quoted 3 lines

It requires more thoughts.

Thanks

I think at this point let's do it before so we at least do not get a regression with your patches, add a big comment and work on fixing properly in the next Linux version. Do you think you can commit to a full fix in the next linux version?

I think it should be ok.

If I understand you correctly, you meant to disable the hardening in this release?

Quoted 8 lines
Quoted 4 lines

(Actually, my understanding is that since we are developing mainline instead of a downstream version with a hardening features, bug reports are somehow expected, especially consider most of the bugs are not related to hardening itself)

Absolutely. Question is do you think we can fix everything by the release?

Probably not, I'm auditing all the virtio drivers and it seems we have many issues:

Quoted 6 lines
Quoted 4 lines

1) race between subsystem registration/use and virtio_device_ready()
2) race between notifications and subsystem registerstiation/use

And it looks to me even virtio-net has this race.

Interesting. How does it look for virtio-net?

Will post a patch soon.

Quoted 6 lines
Quoted 2 lines

So I think I will post a patch to disable this like below for this release.

However please do post patches that add device_ready as appropriate. This is basic spec compliance.

Working on this.

Quoted 3 lines

Also do you think we should do a full revert? Maybe a Kconfig option is ok for now.

Yes, Kconfig should be fine.

Patch will be posted soon.

Thanks

Quoted 310 lines
Quoted 309 lines
Quoted 3 lines

At least for rpmsg we don't seem to have a handle on it yet.

Yes.

Quoted 4 lines
Quoted 4 lines

Thanks

Quoted 16 lines

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 13 lines
drivers/virtio/virtio_ring.c +1 -1
index 13a7348cedff..7ef3115efbad 100644
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
        vq->we_own_ring = true;
        vq->notify = notify;
        vq->weak_barriers = weak_barriers;
-       vq->broken = true;
+       vq->broken = false;
        vq->last_used_idx = 0;
        vq->event_triggered = false;
        vq->num_added = 0;

and drop it on reset?

Right.

Thanks

Quoted 278 lines
Quoted 277 lines
Quoted 277 lines
Quoted 276 lines
Quoted 276 lines
Quoted 275 lines
Quoted 275 lines
Quoted 11 lines
Quoted 9 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

Quoted 35 lines
Quoted 27 lines
Quoted 21 lines
Quoted 7 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to
me the code is correct.

OK.

Quoted 11 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx,
it looks to me we need a device_ready before the kick.
virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

Quoted 8 lines

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well?

Kick itself is not blocked.

Quoted 1 line

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

        virtio_device_ready(vdev);

        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

Quoted 187 lines
Quoted 186 lines
Quoted 186 lines
Quoted 185 lines

---
drivers/s390/virtio/virtio_ccw.c | 4 ++++
drivers/virtio/virtio.c | 15 ++++++++++++---
drivers/virtio/virtio_mmio.c | 5 +++++
drivers/virtio/virtio_pci_modern_dev.c | 5 +++++
drivers/virtio/virtio_ring.c | 11 +++++++----
include/linux/virtio_config.h | 20 ++++++++++++++++++++
6 files changed, 53 insertions(+), 7 deletions(-)

diff --git a/drivers/s390/virtio/virtio_ccw.c b/drivers/s390/virtio/virtio_ccw.c
index c188e4f20ca3..97e51c34e6cf 100644
--- a/drivers/s390/virtio/virtio_ccw.c
+++ b/drivers/s390/virtio/virtio_ccw.c
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
ccw->flags = 0;
ccw->count = sizeof(status);
ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+ /* We use ssch for setting the status which is a serializing
+ * instruction that guarantees the memory writes have
+ * completed before ssch.
+ */
ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
/* Write failed? We assume status is unchanged. */
if (ret)
diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c
index aa1eb5132767..95fac4c97c8b 100644
--- a/drivers/virtio/virtio.c
+++ b/drivers/virtio/virtio.c
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
* */
void virtio_reset_device(struct virtio_device *dev)
{
+ /*
+ * The below virtio_synchronize_cbs() guarantees that any
+ * interrupt for this line arriving after
+ * virtio_synchronize_vqs() has completed is guaranteed to see
+ * vq->broken as true.
+ */
+ virtio_break_device(dev);

So make this conditional

+ virtio_synchronize_cbs(dev);
+
dev->config->reset(dev);
}
EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
dev->config_enabled = false;
dev->config_change_pending = false;

+ INIT_LIST_HEAD(&dev->vqs);
+ spin_lock_init(&dev->vqs_list_lock);
+
/* We always start by resetting the device, in case a previous
* driver messed it up. This also tests that code path a little. */
virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
/* Acknowledge that we've seen the device. */
virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);

- INIT_LIST_HEAD(&dev->vqs);
- spin_lock_init(&dev->vqs_list_lock);
-
/*
* device_add() causes the bus infrastructure to look for a matching
* driver.
diff --git a/drivers/virtio/virtio_mmio.c b/drivers/virtio/virtio_mmio.c
index c9699a59f93c..f9a36bc7ac27 100644
--- a/drivers/virtio/virtio_mmio.c
+++ b/drivers/virtio/virtio_mmio.c
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
/* We should never be setting status to 0. */
BUG_ON(status == 0);

+ /*
+ * Per memory-barriers.txt, wmb() is not needed to guarantee
+ * that the the cache coherent memory writes have completed
+ * before writing to the MMIO region.
+ */
writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
}

diff --git a/drivers/virtio/virtio_pci_modern_dev.c b/drivers/virtio/virtio_pci_modern_dev.c
index 4093f9cca7a6..a0fa14f28a7f 100644
--- a/drivers/virtio/virtio_pci_modern_dev.c
+++ b/drivers/virtio/virtio_pci_modern_dev.c
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
{
struct virtio_pci_common_cfg __iomem *cfg = mdev->common;

+ /*
+ * Per memory-barriers.txt, wmb() is not needed to guarantee
+ * that the the cache coherent memory writes have completed
+ * before writing to the MMIO region.
+ */
vp_iowrite8(status, &cfg->device_status);
}
EXPORT_SYMBOL_GPL(vp_modern_set_status);
diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index 9c231e1fded7..13a7348cedff 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
vq->we_own_ring = true;
vq->notify = notify;
vq->weak_barriers = weak_barriers;
- vq->broken = false;
+ vq->broken = true;
vq->last_used_idx = 0;
vq->event_triggered = false;
vq->num_added = 0;

and make this conditional

@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
return IRQ_NONE;
}

- if (unlikely(vq->broken))
- return IRQ_HANDLED;
+ if (unlikely(vq->broken)) {
+ dev_warn_once(&vq->vq.vdev->dev,
+ "virtio vring IRQ raised before DRIVER_OK");
+ return IRQ_NONE;
+ }

/* Just a hint for performance: so it's ok that this can be racy! */
if (vq->event)
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
vq->we_own_ring = false;
vq->notify = notify;
vq->weak_barriers = weak_barriers;
- vq->broken = false;
+ vq->broken = true;
vq->last_used_idx = 0;
vq->event_triggered = false;
vq->num_added = 0;

and make this conditional

diff --git a/include/linux/virtio_config.h b/include/linux/virtio_config.h
index 25be018810a7..d4edfd7d91bb 100644
--- a/include/linux/virtio_config.h
+++ b/include/linux/virtio_config.h
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
unsigned status = dev->config->get_status(dev);

BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+ /*
+ * The virtio_synchronize_cbs() makes sure vring_interrupt()
+ * will see the driver specific setup if it sees vq->broken
+ * as false (even if the notifications come before DRIVER_OK).
+ */
+ virtio_synchronize_cbs(dev);
+ __virtio_unbreak_device(dev);
+ /*
+ * The transport should ensure the visibility of vq->broken
+ * before setting DRIVER_OK. See the comments for the transport
+ * specific set_status() method.
+ *
+ * A well behaved device will only notify a virtqueue after
+ * DRIVER_OK, this means the device should "see" the coherenct
+ * memory write that set vq->broken as false which is done by
+ * the driver when it sees DRIVER_OK, then the following
+ * driver's vring_interrupt() will see vq->broken as false so
+ * we won't lose any notification.
+ */
dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
}

--
2.25.1

#41 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 19 lines · Jason Wang
Quoted 18 lines · Michael S. Tsirkin
Quoted 16 lines · Jason Wang
Quoted 15 lines · Michael S. Tsirkin
Quoted 13 lines · Jason Wang
Quoted 12 lines · Michael S. Tsirkin
Quoted 10 lines · Jason Wang
Quoted 9 lines · Michael S. Tsirkin
Quoted 7 lines · Jason Wang
Quoted 6 lines · Jason Wang
Quoted 4 lines · Michael S. Tsirkin
Quoted 2 lines · Jason Wang
Quoted 1 line · Michael S. Tsirkin
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines · Jason Wang

On Mon, Jun 13, 2022 at 3:23 PM Michael S. Tsirkin <[email protected]> wrote:

On Mon, Jun 13, 2022 at 01:26:59PM +0800, Jason Wang wrote:
On Sat, Jun 11, 2022 at 1:12 PM Michael S. Tsirkin <[email protected]> wrote:

On Fri, May 27, 2022 at 02:01:19PM +0800, Jason Wang wrote:
This is a rework on the previous IRQ hardening that is done for

Guys; have you heard about trimming emails on reply?
_______________________________________________
Virtualization mailing list [email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#42 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

In reply to Jason Wang
Quoted 2 lines · Jason Wang
Quoted 1 line · Michael S. Tsirkin

Hi Jason,

Quoted 150 lines
Quoted 106 lines · Jason Wang
Quoted 105 lines · Jason Wang
Quoted 103 lines · Michael S. Tsirkin
Quoted 73 lines · Jason Wang
Quoted 71 lines · Michael S. Tsirkin
Quoted 70 lines · Jason Wang
Quoted 66 lines · Michael S. Tsirkin
Quoted 56 lines · Jason Wang
Quoted 54 lines · Michael S. Tsirkin
Quoted 39 lines · Jason Wang
Quoted 25 lines

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 7 lines
Quoted 4 lines

btw, hci has an open and close method and we do rx refill in hdev->open, so we're probably fine here.

Thanks

Sounds good. Now to audit the rest of them from this POV ;)

Adding maintainers.

Quoted 2 lines

drivers/i2c/busses/i2c-virtio.c

It looks to me the device could be used immediately after i2c_add_adapter() return. So we probably need to add virtio_device_ready() before that. Fortunately, there's no rx vq in i2c and the callback looks safe if the callback is called before the i2c registration and after virtio_device_ready().

Quoted 1 line

drivers/net/caif/caif_virtio.c

A networking device, RX is backed by vringh so we don't need to refill. TX is backed by virtio and is available until ndo_open. So it's fine to let the core to set DRIVER_OK after probe().

Quoted 1 line

drivers/nvdimm/virtio_pmem.c

It doesn't use interrupt so far, so it has nothing to do with the IRQ hardening.

But the device could be used by the subsystem immediately after nvdimm_pmem_region_create(), this means the flush could be issued before DRIVER_OK. We need virtio_device_ready() before. We don't have a RX virtqueue and the callback looks safe if the callback is called after virtio_device_ready() but before the nvdimm region creating.

And it looks to me there's a race between the assignment of provider_data and virtio_pmem_flush(). If the flush was issued before the assignment we will end up with a NULL pointer dereference. This is something we need to fix.

Quoted 1 line

arm_scmi

It looks to me the singleton device could be used by SCMI immediately after

Quoted 8 lines
        /* Ensure initialized scmi_vdev is visible */
        smp_store_mb(scmi_vdev, vdev);

So we probably need to do virtio_device_ready() before that. It has an optional rx queue but the filling is done after the above assignment, so it's safe. And the callback looks safe is a callback is triggered after virtio_device_ready() buy before the above assignment.

I wanted to give it a go at this series testing it on the context of SCMI but it does not apply

- not on a v5.18:

17:33 $ git rebase -i v5.18
17:33 $ git am ./v6_20220527_jasowang_rework_on_the_irq_hardening_of_virtio.mbx
Applying: virtio: use virtio_device_ready() in virtio_device_restore()
Applying: virtio: use virtio_reset_device() when possible
Applying: virtio: introduce config op to synchronize vring callbacks
Applying: virtio-pci: implement synchronize_cbs()
Applying: virtio-mmio: implement synchronize_cbs()
error: patch failed: drivers/virtio/virtio_mmio.c:345
error: drivers/virtio/virtio_mmio.c: patch does not apply
Patch failed at 0005 virtio-mmio: implement synchronize_cbs()

- neither on a v5.19-rc2:

17:33 $ git rebase -i v5.19-rc2
17:35 $ git am ./v6_20220527_jasowang_rework_on_the_irq_hardening_of_virtio.mbx
Applying: virtio: use virtio_device_ready() in virtio_device_restore()
error: patch failed: drivers/virtio/virtio.c:526
error: drivers/virtio/virtio.c: patch does not apply
Patch failed at 0001 virtio: use virtio_device_ready() in virtio_device_restore()
hint: Use 'git am --show-current-patch=diff' to see the failed patch When you have resolved this problem, run "git am --continue".

... what I should take as base ?

Thanks,
Cristian

author
#43 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 194 lines · Cristian Marussi
Quoted 2 lines · Jason Wang
Quoted 1 line · Michael S. Tsirkin

Hi Jason,

Quoted 150 lines
Quoted 106 lines · Jason Wang
Quoted 105 lines · Jason Wang
Quoted 103 lines · Michael S. Tsirkin
Quoted 73 lines · Jason Wang
Quoted 71 lines · Michael S. Tsirkin
Quoted 70 lines · Jason Wang
Quoted 66 lines · Michael S. Tsirkin
Quoted 56 lines · Jason Wang
Quoted 54 lines · Michael S. Tsirkin
Quoted 39 lines · Jason Wang
Quoted 35 lines

This is a rework on the previous IRQ hardening that is done for
virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ
that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ
hardening. The vq->broken is set to true during both initialization
and reset. And the vq->broken is set to false in
virtio_device_ready(). Then vring_interrupt() can check and return
when vq->broken is true. And in this case, switch to return IRQ_NONE
to let the interrupt core aware of such invalid interrupt to prevent
IRQ storm.

The reason of using a per queue variable instead of a per device one
is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the
config interrupt hardening is already done in commit 22b7050a024d7
("virtio: defer config changed notifications"). But the method that is
used by config interrupt can't be reused by the vring interrupt
handler because it uses spinlock to do the synchronization which is
expensive.

Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Cc: Vineeth Vijayan <[email protected]>
Cc: Peter Oberparleiter <[email protected]>
Cc: [email protected]
Signed-off-by: Jason Wang <[email protected]>

Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 7 lines
Quoted 4 lines

btw, hci has an open and close method and we do rx refill in hdev->open, so we're probably fine here.

Thanks

Sounds good. Now to audit the rest of them from this POV ;)

Adding maintainers.

Quoted 2 lines

drivers/i2c/busses/i2c-virtio.c

It looks to me the device could be used immediately after i2c_add_adapter() return. So we probably need to add virtio_device_ready() before that. Fortunately, there's no rx vq in i2c and the callback looks safe if the callback is called before the i2c registration and after virtio_device_ready().

Quoted 1 line

drivers/net/caif/caif_virtio.c

A networking device, RX is backed by vringh so we don't need to refill. TX is backed by virtio and is available until ndo_open. So it's fine to let the core to set DRIVER_OK after probe().

Quoted 1 line

drivers/nvdimm/virtio_pmem.c

It doesn't use interrupt so far, so it has nothing to do with the IRQ hardening.

But the device could be used by the subsystem immediately after nvdimm_pmem_region_create(), this means the flush could be issued before DRIVER_OK. We need virtio_device_ready() before. We don't have a RX virtqueue and the callback looks safe if the callback is called after virtio_device_ready() but before the nvdimm region creating.

And it looks to me there's a race between the assignment of provider_data and virtio_pmem_flush(). If the flush was issued before the assignment we will end up with a NULL pointer dereference. This is something we need to fix.

Quoted 1 line

arm_scmi

It looks to me the singleton device could be used by SCMI immediately after

Quoted 8 lines
        /* Ensure initialized scmi_vdev is visible */
        smp_store_mb(scmi_vdev, vdev);

So we probably need to do virtio_device_ready() before that. It has an optional rx queue but the filling is done after the above assignment, so it's safe. And the callback looks safe is a callback is triggered after virtio_device_ready() buy before the above assignment.

I wanted to give it a go at this series testing it on the context of SCMI but it does not apply

- not on a v5.18:

17:33 $ git rebase -i v5.18
17:33 $ git am ./v6_20220527_jasowang_rework_on_the_irq_hardening_of_virtio.mbx
Applying: virtio: use virtio_device_ready() in virtio_device_restore()
Applying: virtio: use virtio_reset_device() when possible
Applying: virtio: introduce config op to synchronize vring callbacks
Applying: virtio-pci: implement synchronize_cbs()
Applying: virtio-mmio: implement synchronize_cbs()
error: patch failed: drivers/virtio/virtio_mmio.c:345
error: drivers/virtio/virtio_mmio.c: patch does not apply
Patch failed at 0005 virtio-mmio: implement synchronize_cbs()

- neither on a v5.19-rc2:

17:33 $ git rebase -i v5.19-rc2
17:35 $ git am ./v6_20220527_jasowang_rework_on_the_irq_hardening_of_virtio.mbx
Applying: virtio: use virtio_device_ready() in virtio_device_restore()
error: patch failed: drivers/virtio/virtio.c:526
error: drivers/virtio/virtio.c: patch does not apply
Patch failed at 0001 virtio: use virtio_device_ready() in virtio_device_restore()
hint: Use 'git am --show-current-patch=diff' to see the failed patch When you have resolved this problem, run "git am --continue".

Quoted 1 line

... what I should take as base ?

It should have already been included in rc2, so there's no need to apply patch manually.

Thanks

Quoted 4 lines

Thanks,
Cristian

#44 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 2 lines · Jason Wang

On Wed, Jun 15, 2022 at 12:46 AM Cristian Marussi
<[email protected]> wrote:

Hi Jason,

Quoted 5 lines
Quoted 5 lines · Jason Wang
Quoted 2 lines · Michael S. Tsirkin
Quoted 1 line

[snip]

Quoted 42 lines
Quoted 42 lines
Quoted 5 lines
Quoted 2 lines

arm_scmi

It looks to me the singleton device could be used by SCMI immediately after

Quoted 8 lines
        /* Ensure initialized scmi_vdev is visible */
        smp_store_mb(scmi_vdev, vdev);

So we probably need to do virtio_device_ready() before that. It has an optional rx queue but the filling is done after the above assignment, so it's safe. And the callback looks safe is a callback is triggered after virtio_device_ready() buy before the above assignment.

I wanted to give it a go at this series testing it on the context of SCMI but it does not apply

- not on a v5.18:

17:33 $ git rebase -i v5.18
17:33 $ git am ./v6_20220527_jasowang_rework_on_the_irq_hardening_of_virtio.mbx
Applying: virtio: use virtio_device_ready() in virtio_device_restore()
Applying: virtio: use virtio_reset_device() when possible
Applying: virtio: introduce config op to synchronize vring callbacks
Applying: virtio-pci: implement synchronize_cbs()
Applying: virtio-mmio: implement synchronize_cbs()
error: patch failed: drivers/virtio/virtio_mmio.c:345
error: drivers/virtio/virtio_mmio.c: patch does not apply
Patch failed at 0005 virtio-mmio: implement synchronize_cbs()

- neither on a v5.19-rc2:

17:33 $ git rebase -i v5.19-rc2
17:35 $ git am ./v6_20220527_jasowang_rework_on_the_irq_hardening_of_virtio.mbx
Applying: virtio: use virtio_device_ready() in virtio_device_restore()
error: patch failed: drivers/virtio/virtio.c:526
error: drivers/virtio/virtio.c: patch does not apply
Patch failed at 0001 virtio: use virtio_device_ready() in virtio_device_restore()
hint: Use 'git am --show-current-patch=diff' to see the failed patch When you have resolved this problem, run "git am --continue".

Quoted 5 lines
Quoted 1 line

... what I should take as base ?

It should have already been included in rc2, so there's no need to apply patch manually.

I tested this series as included in v5.19-rc2 (WITHOUT adding a virtio_device_ready in SCMI virtio as you mentioned above ... if I got it right) and I have NOT seen any issue around SCMI virtio using my usual test setup (using both SCMI vqueues).

No anomalies even when using SCMI virtio in atomic/polling mode.

Adding a virtio_device_ready() at the end of the SCMI virtio probe() works fine either, it does not make any difference in my setup. (both using QEMU and kvmtool with this latter NOT supporting
virtio_V1...not sure if it makes a difference but I thought was worth mentioning)

Thanks,
Cristian

author
#45 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 67 lines · Cristian Marussi
Quoted 2 lines · Jason Wang

On Wed, Jun 15, 2022 at 12:46 AM Cristian Marussi
<[email protected]> wrote:

Hi Jason,

Quoted 5 lines
Quoted 5 lines · Jason Wang
Quoted 2 lines · Michael S. Tsirkin
Quoted 1 line

[snip]

Quoted 42 lines
Quoted 42 lines
Quoted 5 lines
Quoted 2 lines

arm_scmi

It looks to me the singleton device could be used by SCMI immediately after

Quoted 8 lines
        /* Ensure initialized scmi_vdev is visible */
        smp_store_mb(scmi_vdev, vdev);

So we probably need to do virtio_device_ready() before that. It has an optional rx queue but the filling is done after the above assignment, so it's safe. And the callback looks safe is a callback is triggered after virtio_device_ready() buy before the above assignment.

I wanted to give it a go at this series testing it on the context of SCMI but it does not apply

- not on a v5.18:

17:33 $ git rebase -i v5.18
17:33 $ git am ./v6_20220527_jasowang_rework_on_the_irq_hardening_of_virtio.mbx
Applying: virtio: use virtio_device_ready() in virtio_device_restore()
Applying: virtio: use virtio_reset_device() when possible
Applying: virtio: introduce config op to synchronize vring callbacks
Applying: virtio-pci: implement synchronize_cbs()
Applying: virtio-mmio: implement synchronize_cbs()
error: patch failed: drivers/virtio/virtio_mmio.c:345
error: drivers/virtio/virtio_mmio.c: patch does not apply
Patch failed at 0005 virtio-mmio: implement synchronize_cbs()

- neither on a v5.19-rc2:

17:33 $ git rebase -i v5.19-rc2
17:35 $ git am ./v6_20220527_jasowang_rework_on_the_irq_hardening_of_virtio.mbx
Applying: virtio: use virtio_device_ready() in virtio_device_restore()
error: patch failed: drivers/virtio/virtio.c:526
error: drivers/virtio/virtio.c: patch does not apply
Patch failed at 0001 virtio: use virtio_device_ready() in virtio_device_restore()
hint: Use 'git am --show-current-patch=diff' to see the failed patch When you have resolved this problem, run "git am --continue".

Quoted 5 lines
Quoted 1 line

... what I should take as base ?

It should have already been included in rc2, so there's no need to apply patch manually.

I tested this series as included in v5.19-rc2 (WITHOUT adding a virtio_device_ready in SCMI virtio as you mentioned above ... if I got it right) and I have NOT seen any issue around SCMI virtio using my usual test setup (using both SCMI vqueues).

Quoted 7 lines

No anomalies even when using SCMI virtio in atomic/polling mode.

Adding a virtio_device_ready() at the end of the SCMI virtio probe() works fine either, it does not make any difference in my setup. (both using QEMU and kvmtool with this latter NOT supporting
virtio_V1...not sure if it makes a difference but I thought was worth mentioning)

Thanks a lot for the testing.

We want to prevent malicious hypervisors from attacking us. So more questions:

Assuming we do:

virtio_device_ready();
/* Ensure initialized scmi_vdev is visible */
smp_store_mb(scmi_vdev, vdev);

This means we allow the callbacks (scmi_vio_complete) to be called before smp_store_mb(). We need to make sure the callbacks are robust. And this looks fine since we have the check of scmi_vio_channel_acquire() and if the notification is called before smp_store_mb(), the acquire will fail.

If we put virtio_device_ready() after smp_store_mb() like:

/* Ensure initialized scmi_vdev is visible */
smp_store_mb(scmi_vdev, vdev);
virtio_device_ready();

If I understand correctly, there will be a race since the SCMI may try to use the device before virtio_device_ready(), this violates the virtio spec somehow.

Thanks

Quoted 4 lines

Thanks,
Cristian

#46 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

In reply to Jason Wang
Quoted 104 lines · Jason Wang
Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang
Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 1 line
Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

I am not sure I understand. Are you asking whether device must check vqs for buffers upon DRIVER_OK? I don't think so, if driver wants buffers processed it must kick after DRIVER_OK.

And kicking before DRIVER_OK is out of spec.

Quoted 37 lines
Quoted 37 lines
Quoted 27 lines
Quoted 21 lines
Quoted 7 lines
Quoted 4 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

exactly

Quoted 8 lines
Quoted 6 lines
Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Well first won't driver code normally kick as well?

Kick itself is not blocked.

It is out of spec though.

Quoted 18 lines
Quoted 1 line

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

Quoted 7 lines
        virtio_device_ready(vdev);

        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

yes - as long as no buffers are used, no callback is expected. However wasn't the point of your patches to handle a malicious device?

Quoted 187 lines
Quoted 187 lines
Quoted 186 lines
Quoted 186 lines
Quoted 185 lines
Quoted 185 lines
Quoted 184 lines
Quoted 184 lines
Quoted 38 lines








---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 33 lines
include/linux/virtio_config.h +20 -1
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
--

2.25.1

author
#47 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 109 lines · Michael S. Tsirkin
Quoted 107 lines · Jason Wang
Quoted 73 lines · Michael S. Tsirkin
Quoted 71 lines · Jason Wang
Quoted 70 lines · Michael S. Tsirkin
Quoted 66 lines · Jason Wang
Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Jason Wang
Quoted 39 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 6 lines

Jason, I am really concerned by all the fallout. I propose adding a flag to suppress the hardening - this will be a debugging aid and a work around for users if we find more buggy drivers.

suppress_interrupt_hardening ?

I can post a patch but I'm afraid if we disable it by default, it won't be used by the users so there's no way for us to receive the bug report. Or we need a plan to enable it by default.

It's rc2, how about waiting for 1 and 2 rc? Or it looks better if we simply warn instead of disable it by default.

Quoted 7 lines
Quoted 1 line

Thanks

I meant more like a flag in struct virtio_driver. For now, could you audit all drivers which don't call _ready? I found 5 of these:

drivers/bluetooth/virtio_bt.c

This driver seems to be fine, it doesn't use the device/vq in its probe().

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

But it calls hci_register_dev and that in turn queues all kind of work. Also, can linux start using the device immediately after it's registered?

So I think the driver is allowed to queue before DRIVER_OK.

it's not allowed to kick

Yes.

Quoted 6 lines
Quoted 4 lines

If yes,
the only side effect is the delay of the tx interrupt after DRIVER_OK for a well behaved device.

your patches drop the interrupt though, it won't be just delayed.

For a well behaved device, it can only trigger the interrupt after DRIVER_OK.

So for virtio bt, it works like:

1) driver queue buffer and kick
2) driver set DRIVER_OK
3) device start to process the buffer
4) device send an notification

The only risk is that the virtqueue could be filled before DRIVER_OK, or anything I missed?

Quoted 4 lines
Quoted 3 lines

If not, we need to clarify it in the spec
and call virtio_device_ready() before subsystem registration.

Quoted 8 lines
Quoted 5 lines
Quoted 1 line

hmm, i don't get what we need to clarify

E.g the driver is not allowed to kick or after DRIVER_OK should the device only process the buffer after a kick after DRIVER_OK (I think no)?

I am not sure I understand. Are you asking whether device must check vqs for buffers upon DRIVER_OK?

Yes.

Quoted 2 lines

I don't think so,
if driver wants buffers processed it must kick after DRIVER_OK.

Ok.

Quoted 2 lines

And kicking before DRIVER_OK is out of spec.

Right.

Quoted 44 lines
Quoted 44 lines
Quoted 44 lines
Quoted 29 lines
Quoted 23 lines
Quoted 9 lines
Quoted 6 lines

drivers/gpu/drm/virtio/virtgpu_drv.c

It calles virtio_device_ready() in virtio_gpu_init(), and it looks to me the code is correct.

OK.

Quoted 11 lines
Quoted 3 lines

drivers/i2c/busses/i2c-virtio.c
drivers/net/caif/caif_virtio.c
drivers/nvdimm/virtio_pmem.c

The above looks fine and we have three more:

arm_scmi: probe() doesn't use vq
mac80211_hwsim.c: doesn't use vq (only fill rx), but it kicks the rx, it looks to me we need a device_ready before the kick. virtio_rpmsg_bus.c: doesn't use vq

Quoted 5 lines
Quoted 2 lines

I will post a patch for mac80211_hwsim.c.
Thanks

Same comments for all of the above. Might linux not start using the device once it's registered?

Quoted 7 lines

It depends on the specific subsystem.

For the subsystem that can't use the device immediately, calling virtio_device_ready() after the subsystem's registration should be fine. E.g for the networking subsystem, the TX won't happen if ndo_open() is not called, calling virtio_device_ready() after netdev_register() seems to be fine.

exactly

Quoted 4 lines

For the subsystem that can use the device immediately, if the subsystem does not depend on the result of a request in the probe to proceed, we are still fine. Since those requests will be proceed after DRIVER_OK.

Quoted 5 lines
Quoted 3 lines
Quoted 1 line

Well first won't driver code normally kick as well?

Kick itself is not blocked.

It is out of spec though.

Yes.

Quoted 21 lines
Quoted 21 lines
Quoted 2 lines

And without kick, won't everything just be blocked?

It depends on the subsystem. E.g driver can choose to use a callback instead of polling the used buffer in the probe.

Quoted 7 lines
Quoted 5 lines

For the rest we need to do virtio_device_ready() before registration.

Thanks

Then we can get an interrupt for an unregistered device.

It depends on the device. For the device that doesn't have an rx queue (or device to driver queue), we are fine:

E.g in virtio-blk:

        virtio_device_ready(vdev);
Quoted 8 lines
Quoted 5 lines
        err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
        if (err)
                goto out_cleanup_disk;

Thanks

yes - as long as no buffers are used, no callback is expected. However wasn't the point of your patches to handle a malicious device?

Right, but for subsystems that don't have a way to open/close devices like the block layer, there's not much we can do except depending on the callback - virtblk_done itself which seems to be safe.

We guard the malicious hypervisor by checking:

1) checking the token
2) when token is NULL, no dereference to any subsystem's data structure

Thanks

Quoted 182 lines
Quoted 182 lines
Quoted 182 lines
Quoted 182 lines
Quoted 182 lines
Quoted 182 lines
Quoted 182 lines
Quoted 182 lines
Quoted 182 lines
Quoted 39 lines









---
 drivers/s390/virtio/virtio_ccw.c       |  4 ++++
 drivers/virtio/virtio.c                | 15 ++++++++++++---
 drivers/virtio/virtio_mmio.c           |  5 +++++
 drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
 drivers/virtio/virtio_ring.c           | 11 +++++++----
 include/linux/virtio_config.h          | 20 ++++++++++++++++++++
 6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +0 -0
index aa1eb5132767..95fac4c97c8b 100644

drivers/virtio/virtio.c static int virtio_features_ok(struct virtio_device *dev) · line 229

Quoted 10 lines
drivers/virtio/virtio.c +7 -0
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
  * */
 void virtio_reset_device(struct virtio_device *dev)
 {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);

So make this conditional

Quoted 62 lines
drivers/virtio/virtio.c +5 -3
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
 }
 EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
 }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
 {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +0 -0
index 9c231e1fded7..13a7348cedff 100644

drivers/virtio/virtio_ring.c static struct virtqueue *vring_create_virtqueue_packed( · line 1694

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 15 lines
drivers/virtio/virtio_ring.c +5 -2
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)

drivers/virtio/virtio_ring.c struct virtqueue *__vring_new_virtqueue(unsigned int index, · line 2186

Quoted 8 lines
drivers/virtio/virtio_ring.c +1 -1
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;

and make this conditional

Quoted 31 lines
include/linux/virtio_config.h +20 -0
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
 }
 
Quoted 7 lines
Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
Quoted 3 lines
Quoted 3 lines
Quoted 2 lines

--
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#48 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

In reply to Jason Wang

Hi,

I encounter a issue when testing virtio-balloon on my platform (ARM64) with kernel 5.19-rc4 to boot VM with "-device virtio-balloon ", and

then change the size of balloon in qemu monitor, but it isn't valid, and the log is as follows:

QEMU 6.1.50 monitor - type 'help' for more information (qemu) info balloon
info balloon
balloon: actual=4096
(qemu) balloon 3172
balloon 3172
(qemu) info balloon
info balloon
balloon: actual=4096

I git bisect the patch, and find this patch ([8b4ec69d7e098a7ddf832e1e7840de53ed474c77] virtio: harden vring IRQ) at last.

Do you have any idea about it?

Best regards,

Xiang Chen

Quoted 198 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
  drivers/s390/virtio/virtio_ccw.c       |  4 ++++
  drivers/virtio/virtio.c                | 15 ++++++++++++---
  drivers/virtio/virtio_mmio.c           |  5 +++++
  drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
  drivers/virtio/virtio_ring.c           | 11 +++++++----
  include/linux/virtio_config.h          | 20 ++++++++++++++++++++
  6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
  	ccw->flags = 0;
  	ccw->count = sizeof(status);
  	ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+	/* We use ssch for setting the status which is a serializing
+	 * instruction that guarantees the memory writes have
+	 * completed before ssch.
+	 */
  	ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
  	/* Write failed? We assume status is unchanged. */
  	if (ret)
drivers/virtio/virtio.c +12 -3
index aa1eb5132767..95fac4c97c8b 100644
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
   * */
  void virtio_reset_device(struct virtio_device *dev)
  {
+	/*
+	 * The below virtio_synchronize_cbs() guarantees that any
+	 * interrupt for this line arriving after
+	 * virtio_synchronize_vqs() has completed is guaranteed to see
+	 * vq->broken as true.
+	 */
+	virtio_break_device(dev);
+	virtio_synchronize_cbs(dev);
+
  	dev->config->reset(dev);
  }
  EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
  	dev->config_enabled = false;
  	dev->config_change_pending = false;
 
+	INIT_LIST_HEAD(&dev->vqs);
+	spin_lock_init(&dev->vqs_list_lock);
+
  	/* We always start by resetting the device, in case a previous
  	 * driver messed it up.  This also tests that code path a little. */
  	virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
  	/* Acknowledge that we've seen the device. */
  	virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-	INIT_LIST_HEAD(&dev->vqs);
-	spin_lock_init(&dev->vqs_list_lock);
-
  	/*
  	 * device_add() causes the bus infrastructure to look for a matching
  	 * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
  	/* We should never be setting status to 0. */
  	BUG_ON(status == 0);
 
+	/*
+	 * Per memory-barriers.txt, wmb() is not needed to guarantee
+	 * that the the cache coherent memory writes have completed
+	 * before writing to the MMIO region.
+	 */
  	writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
  }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
  {
  	struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+	/*
+	 * Per memory-barriers.txt, wmb() is not needed to guarantee
+	 * that the the cache coherent memory writes have completed
+	 * before writing to the MMIO region.
+	 */
  	vp_iowrite8(status, &cfg->device_status);
  }
  EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +7 -4
index 9c231e1fded7..13a7348cedff 100644
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
  	vq->we_own_ring = true;
  	vq->notify = notify;
  	vq->weak_barriers = weak_barriers;
-	vq->broken = false;
+	vq->broken = true;
  	vq->last_used_idx = 0;
  	vq->event_triggered = false;
  	vq->num_added = 0;
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
  		return IRQ_NONE;
  	}
 
-	if (unlikely(vq->broken))
-		return IRQ_HANDLED;
+	if (unlikely(vq->broken)) {
+		dev_warn_once(&vq->vq.vdev->dev,
+			      "virtio vring IRQ raised before DRIVER_OK");
+		return IRQ_NONE;
+	}
 
  	/* Just a hint for performance: so it's ok that this can be racy! */
  	if (vq->event)
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
  	vq->we_own_ring = false;
  	vq->notify = notify;
  	vq->weak_barriers = weak_barriers;
-	vq->broken = false;
+	vq->broken = true;
  	vq->last_used_idx = 0;
  	vq->event_triggered = false;
  	vq->num_added = 0;
include/linux/virtio_config.h +20 -0
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
  	unsigned status = dev->config->get_status(dev);
 
  	BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+	/*
+	 * The virtio_synchronize_cbs() makes sure vring_interrupt()
+	 * will see the driver specific setup if it sees vq->broken
+	 * as false (even if the notifications come before DRIVER_OK).
+	 */
+	virtio_synchronize_cbs(dev);
+	__virtio_unbreak_device(dev);
+	/*
+	 * The transport should ensure the visibility of vq->broken
+	 * before setting DRIVER_OK. See the comments for the transport
+	 * specific set_status() method.
+	 *
+	 * A well behaved device will only notify a virtqueue after
+	 * DRIVER_OK, this means the device should "see" the coherenct
+	 * memory write that set vq->broken as false which is done by
+	 * the driver when it sees DRIVER_OK, then the following
+	 * driver's vring_interrupt() will see vq->broken as false so
+	 * we won't lose any notification.
+	 */
  	dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
  }
 
author
#49 lore raw

Re: [PATCH V6 8/9] virtio: harden vring IRQ

Quoted 19 lines · chenxiang (M)

Hi,

I encounter a issue when testing virtio-balloon on my platform (ARM64) with kernel 5.19-rc4 to boot VM with "-device virtio-balloon ", and

then change the size of balloon in qemu monitor, but it isn't valid, and the log is as follows:

QEMU 6.1.50 monitor - type 'help' for more information (qemu) info balloon
info balloon
balloon: actual=4096
(qemu) balloon 3172
balloon 3172
(qemu) info balloon
info balloon
balloon: actual=4096

Quoted 5 lines

I git bisect the patch, and find this patch ([8b4ec69d7e098a7ddf832e1e7840de53ed474c77] virtio: harden vring IRQ) at last.

Do you have any idea about it?

Yes, we noticed this issue and have disable the hardening feature via:

c346dae4f3fbce51bbd4f2ec5e8c6f9b91e93163 ("virtio: disable notification hardening by default") 6a9720576cd00d30722c5f755bd17d4cfa9df636 ("virtio:
VIRTIO_HARDEN_NOTIFICATION is broken")

which have been merged in Michael's tree.

https://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost.git/commit/?h=linux-next&id=6a9720576cd00d30722c5f755bd17d4cfa9df636

Thanks

Quoted 197 lines

Best regards,

Xiang Chen

Quoted 190 lines · Jason Wang

This is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:

1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport

The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt() can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.

The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.

Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
  drivers/s390/virtio/virtio_ccw.c       |  4 ++++
  drivers/virtio/virtio.c                | 15 ++++++++++++---
  drivers/virtio/virtio_mmio.c           |  5 +++++
  drivers/virtio/virtio_pci_modern_dev.c |  5 +++++
  drivers/virtio/virtio_ring.c           | 11 +++++++----
  include/linux/virtio_config.h          | 20 ++++++++++++++++++++
  6 files changed, 53 insertions(+), 7 deletions(-)
drivers/s390/virtio/virtio_ccw.c +4 -0
index c188e4f20ca3..97e51c34e6cf 100644
@@ -971,6 +971,10 @@ static void virtio_ccw_set_status(struct virtio_device *vdev, u8 status)
      ccw->flags = 0;
      ccw->count = sizeof(status);
      ccw->cda = (__u32)(unsigned long)&vcdev->dma_area->status;
+     /* We use ssch for setting the status which is a serializing
+      * instruction that guarantees the memory writes have
+      * completed before ssch.
+      */
      ret = ccw_io_helper(vcdev, ccw, VIRTIO_CCW_DOING_WRITE_STATUS);
      /* Write failed? We assume status is unchanged. */
      if (ret)
drivers/virtio/virtio.c +12 -3
index aa1eb5132767..95fac4c97c8b 100644
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)
   * */
  void virtio_reset_device(struct virtio_device *dev)
  {
+     /*
+      * The below virtio_synchronize_cbs() guarantees that any
+      * interrupt for this line arriving after
+      * virtio_synchronize_vqs() has completed is guaranteed to see
+      * vq->broken as true.
+      */
+     virtio_break_device(dev);
+     virtio_synchronize_cbs(dev);
+
      dev->config->reset(dev);
  }
  EXPORT_SYMBOL_GPL(virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev)
      dev->config_enabled = false;
      dev->config_change_pending = false;
 
+     INIT_LIST_HEAD(&dev->vqs);
+     spin_lock_init(&dev->vqs_list_lock);
+
      /* We always start by resetting the device, in case a previous
       * driver messed it up.  This also tests that code path a little. */
      virtio_reset_device(dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)
      /* Acknowledge that we've seen the device. */
      virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
 
-     INIT_LIST_HEAD(&dev->vqs);
-     spin_lock_init(&dev->vqs_list_lock);
-
      /*
       * device_add() causes the bus infrastructure to look for a matching
       * driver.
drivers/virtio/virtio_mmio.c +5 -0
index c9699a59f93c..f9a36bc7ac27 100644
@@ -253,6 +253,11 @@ static void vm_set_status(struct virtio_device *vdev, u8 status)
      /* We should never be setting status to 0. */
      BUG_ON(status == 0);
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      writel(status, vm_dev->base + VIRTIO_MMIO_STATUS);
  }
 
drivers/virtio/virtio_pci_modern_dev.c +5 -0
index 4093f9cca7a6..a0fa14f28a7f 100644
@@ -467,6 +467,11 @@ void vp_modern_set_status(struct virtio_pci_modern_device *mdev,
  {
      struct virtio_pci_common_cfg __iomem *cfg = mdev->common;
 
+     /*
+      * Per memory-barriers.txt, wmb() is not needed to guarantee
+      * that the the cache coherent memory writes have completed
+      * before writing to the MMIO region.
+      */
      vp_iowrite8(status, &cfg->device_status);
  }
  EXPORT_SYMBOL_GPL(vp_modern_set_status);
drivers/virtio/virtio_ring.c +7 -4
index 9c231e1fded7..13a7348cedff 100644
@@ -1688,7 +1688,7 @@ static struct virtqueue *vring_create_virtqueue_packed(
      vq->we_own_ring = true;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;
@@ -2134,8 +2134,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)
              return IRQ_NONE;
      }
 
-     if (unlikely(vq->broken))
-             return IRQ_HANDLED;
+     if (unlikely(vq->broken)) {
+             dev_warn_once(&vq->vq.vdev->dev,
+                           "virtio vring IRQ raised before DRIVER_OK");
+             return IRQ_NONE;
+     }
 
      /* Just a hint for performance: so it's ok that this can be racy! */
      if (vq->event)
@@ -2177,7 +2180,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index,
      vq->we_own_ring = false;
      vq->notify = notify;
      vq->weak_barriers = weak_barriers;
-     vq->broken = false;
+     vq->broken = true;
      vq->last_used_idx = 0;
      vq->event_triggered = false;
      vq->num_added = 0;
include/linux/virtio_config.h +15 -0
index 25be018810a7..d4edfd7d91bb 100644
@@ -256,6 +256,26 @@ void virtio_device_ready(struct virtio_device *dev)
      unsigned status = dev->config->get_status(dev);
 
      BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+
+     /*
+      * The virtio_synchronize_cbs() makes sure vring_interrupt()
+      * will see the driver specific setup if it sees vq->broken
+      * as false (even if the notifications come before DRIVER_OK).
+      */
+     virtio_synchronize_cbs(dev);
+     __virtio_unbreak_device(dev);
+     /*
+      * The transport should ensure the visibility of vq->broken
+      * before setting DRIVER_OK. See the comments for the transport
+      * specific set_status() method.
+      *
+      * A well behaved device will only notify a virtqueue after
+      * DRIVER_OK, this means the device should "see" the coherenct
Quoted 9 lines

include/linux/virtio_config.h void virtio_device_ready(struct virtio_device *dev) · line 281

Quoted 8 lines
include/linux/virtio_config.h +5 -0
+      * memory write that set vq->broken as false which is done by
+      * the driver when it sees DRIVER_OK, then the following
+      * driver's vring_interrupt() will see vq->broken as false so
+      * we won't lose any notification.
+      */
      dev->config->set_status(dev, status | VIRTIO_CONFIG_S_DRIVER_OK);
  }
 

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

author
#50 lore raw

PATCH v6 9/9 virtio: use WARN_ON() to warning illegal status value

In reply to Jason Wang

We used to use BUG_ON() in virtio_device_ready() to detect illegal status value, this seems sub-optimal since the value is under the control of the device. Switch to use WARN_ON() instead.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
 include/linux/virtio_config.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/virtio_config.h +1 -1
index d4edfd7d91bb..9a36051ceb76 100644
@@ -255,7 +255,7 @@ void virtio_device_ready(struct virtio_device *dev)
 {
 	unsigned status = dev->config->get_status(dev);
 
-	BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+	WARN_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
 
 	/*
 	 * The virtio_synchronize_cbs() makes sure vring_interrupt()
Signature · 6 lines
2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
reviewer Reviewed-by on v6 9/9
#51 lore raw

Re: [PATCH V6 9/9] virtio: use WARN_ON() to warning illegal status value

Quoted 4 lines · Jason Wang

We used to use BUG_ON() in virtio_device_ready() to detect illegal status value, this seems sub-optimal since the value is under the control of the device. Switch to use WARN_ON() instead.

Quoted 10 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 18 lines
---
 include/linux/virtio_config.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/virtio_config.h +1 -2
index d4edfd7d91bb..9a36051ceb76 100644
@@ -255,7 +255,7 @@ void virtio_device_ready(struct virtio_device *dev)
 {
 	unsigned status = dev->config->get_status(dev);
 
-	BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+	WARN_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
 
 	/*
 	 * The virtio_synchronize_cbs() makes sure vring_interrupt()
--
Quoted 2 lines

2.25.1

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Reviewed-by on v6 9/9
#52 lore raw

Re: [PATCH V6 9/9] virtio: use WARN_ON() to warning illegal status value

In reply to Jason Wang
Quoted 14 lines · Jason Wang

We used to use BUG_ON() in virtio_device_ready() to detect illegal status value, this seems sub-optimal since the value is under the control of the device. Switch to use WARN_ON() instead.

Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
Quoted 3 lines

---
include/linux/virtio_config.h | 2 +-

1 file changed, 1 insertion(+), 1 deletion(-)
#53 lore raw

Re: [PATCH V6 9/9] virtio: use WARN_ON() to warning illegal status value

In reply to Jason Wang

At a minimum, I don't see why it's part of the series. Host can always crash the guest if it wants to ...
The point of BUG_ON is device or driver is already corrupted so we should not try to drive it. If you still want this in pls come up with a better commit log explaining the why.

Quoted 1 line · Jason Wang

We used to use BUG_ON() in virtio_device_ready() to detect illegal

not really, BUG_ON just crashes the kernel. we detect by checking status.

Quoted 2 lines

status value, this seems sub-optimal since the value is under the control of the device. Switch to use WARN_ON() instead.

some people use crash on warn so ...

Quoted 21 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
 include/linux/virtio_config.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/virtio_config.h +0 -0
index d4edfd7d91bb..9a36051ceb76 100644

include/linux/virtio_config.h void virtio_device_ready(struct virtio_device *dev) · line 259

Quoted 6 lines
include/linux/virtio_config.h +1 -1
@@ -255,7 +255,7 @@ void virtio_device_ready(struct virtio_device *dev)
 {
 	unsigned status = dev->config->get_status(dev);
 
-	BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+	WARN_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
 

we lose debuggability as guest will try to continue.
if we are doing this let us print a helpful message and dump a lot of state right here.

Quoted 4 lines
include/linux/virtio_config.h +0 -0
 	/*
 	 * The virtio_synchronize_cbs() makes sure vring_interrupt()
Signature · 1 line
2.25.1
author
#54 lore raw

Re: [PATCH V6 9/9] virtio: use WARN_ON() to warning illegal status value

Quoted 3 lines · Michael S. Tsirkin

At a minimum, I don't see why it's part of the series. Host can always crash the guest if it wants to ...

Probably not with some recent technology. In those cases, a fault will be generated if the hypervisor tries to access the memory that is private to the guest.

Quoted 3 lines

The point of BUG_ON is device or driver is already corrupted so we should not try to drive it. If you still want this in pls come up with a better commit log explaining the why.

A question here, should we always use BUG_ON for the buggy/malicious hypervisor?

The interrupt hardening logic in this series tries to make guest survive, so did this patch.

Quoted 6 lines · Jason Wang
Quoted 1 line

We used to use BUG_ON() in virtio_device_ready() to detect illegal

not really, BUG_ON just crashes the kernel. we detect by checking status.

We need a kind of notification otherwise there's no way for the user to know about this expected value.

Quoted 5 lines
Quoted 3 lines

status value, this seems sub-optimal since the value is under the control of the device. Switch to use WARN_ON() instead.

some people use crash on warn so ...

Yes, but the policy is under the control of the user.

Quoted 25 lines
Quoted 25 lines
Cc Thomas Gleixner <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Paul E. McKenney" <[email protected]>
Cc Marc Zyngier <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Vineeth Vijayan <[email protected]>
Cc Peter Oberparleiter <[email protected]>
---
 include/linux/virtio_config.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/virtio_config.h +0 -0
index d4edfd7d91bb..9a36051ceb76 100644
@@ -255,7 +255,7 @@ void virtio_device_ready(struct virtio_device *dev)
 {
      unsigned status = dev->config->get_status(dev);
 
Quoted 7 lines

include/linux/virtio_config.h void virtio_device_ready(struct virtio_device *dev) · line 259

Quoted 3 lines
include/linux/virtio_config.h +1 -1
-     BUG_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
+     WARN_ON(status & VIRTIO_CONFIG_S_DRIVER_OK);
 

we lose debuggability as guest will try to continue.
if we are doing this let us print a helpful message and dump a lot of state right here.

I'm ok with dropping this patch from the series. And revisit it in the future.

Thanks

Quoted 6 lines
Quoted 5 lines
include/linux/virtio_config.h +0 -1
      /*
       * The virtio_synchronize_cbs() makes sure vring_interrupt()
--

2.25.1