Disable ATS via iommu during PCI resets

Hi all,

PCIe permits a device to ignore ATS invalidation TLPs while processing a reset. This creates a problem visible to the OS where an ATS invalidation command will time out: e.g. an SVA domain will have no coordination with a reset event and can racily issue ATS invalidations to a resetting device.

The OS should do something to mitigate this as we do not want production systems to be reporting critical ATS failures, especially in a hypervisor environment. Broadly, OS could arrange to ignore the timeouts, block page table mutations to prevent invalidations, or disable and block ATS.

The PCIe spec in sec 10.3.1 IMPLEMENTATION NOTE recommends to disable and block ATS before initiating a Function Level Reset. It also mentions that other reset methods could have the same vulnerability as well.

Provide a callback from the PCI subsystem that will enclose the reset and have the iommu core temporarily change domains to group->blocking_domain, so IOMMU drivers would fence any incoming ATS queries, synchronously stop issuing new ATS invalidations, and wait for existing ATS invalidations to complete. Doing this can avoid any ATS invaliation timeouts.

When a device is resetting, any new domain attachment has to be rejected, until the reset is finished, to prevent ATS activity from being activated between the two callback functions. Introduce a new resetting_domain, and reject a concurrent __iommu_attach_device/set_group_pasid().

Finally, call these pci_dev_reset_iommu/done() functions in the PCI reset functions.

This is on Github:
https://github.com/nicolinc/iommufd/commits/iommu_dev_reset-v8

Changelog
v8
* Rebase on v6.19-rc1
* Add Tested-by from Dheeraj
* Add Reviewed-by from Jason
* [pci] Add Acked-by from Bjorn Helgaas
v7
https://lore.kernel.org/all/[email protected]/
* Rebase on Joerg's next tree
* Add Reviewed-by from Kevin
* [iommu] Fix inline functions when !CONFIG_IOMMU_API
v6
https://lore.kernel.org/all/[email protected]/
* Add Reviewed-by from Baolu and Kevin
* Revise inline comments, kdocs, commit messages, uAPI
* [iommu] s/iommu_dev_reset/pci_dev_reset_iommu/g for PCI exclusively
* [iommu] Disallow iommu group sibling devices to attach concurrently
* [pci] Drop unnecessary initializations to "ret" and "rc"
* [pci] Improve pci_err message unpon a prepare() failure
* [pci] Move pci_ats_supported() check inside the IOMMU callbacks
* [pci] Apply callbacks to pci_reset_bus_function() that was missed
v5
https://lore.kernel.org/all/[email protected]/
* Rebase on Joerg's next tree
* [iommu] Skip in shared iommu_group cases
* [iommu] Pass in default_domain to iommu_setup_dma_ops
* [iommu] Add kdocs to iommu_get_domain_for_dev_locked()
* [iommu] s/get_domain_for_dev_locked/driver_get_domain_for_dev
* [iommu] Replace per-gdev pending_reset with per-group resetting_domain
v4
https://lore.kernel.org/all/[email protected]/
* Add Reviewed-by from Baolu
* [iommu] Use guard(mutex)
* [iommu] Update kdocs for typos and revisings
* [iommu] Skip two corner cases (alias and SRIOV)
* [iommu] Rework attach_dev to pass in old domain pointer
* [iommu] Reject concurrent attach_dev/set_dev_pasid for compatibility concern
* [smmuv3] Drop the old_domain depedency in its release_dev callback
* [pci] Add pci_reset_iommu_prepare/_done() wrappers checking ATS cap
v3
https://lore.kernel.org/all/[email protected]/
* Add Reviewed-by from Jason
* [iommu] Add a fast return in iommu_deferred_attach()
* [iommu] Update kdocs, inline comments, and commit logs
* [iommu] Use group->blocking_domain v.s. ops->blocked_domain
* [iommu] Drop require_direct, iommu_group_get(), and xa_lock()
* [iommu] Set the pending_reset flag after RID/PASID domain setups
* [iommu] Do not bypass PASID domains when RID domain is already the blocking_domain
* [iommu] Add iommu_get_domain_for_dev_locked to correctly return the blocking_domain
v2
https://lore.kernel.org/all/[email protected]/
* [iommu] Update kdocs, inline comments, and commit logs
* [iommu] Replace long-holding group->mutex with a pending_reset flag
* [pci] Abort reset routines if iommu_dev_reset_prepare() fails
* [pci] Apply the same vulnerability fix to other reset functions
v1
https://lore.kernel.org/all/[email protected]/

Thanks
Nicolin

Nicolin Chen (5):
iommu: Lock group->mutex in iommu_deferred_attach()
iommu: Tidy domain for iommu_setup_dma_ops()
iommu: Add iommu_driver_get_domain_for_dev() helper
iommu: Introduce pci_dev_reset_iommu_prepare/done()
PCI: Suspend iommu function prior to resetting a device

 drivers/iommu/dma-iommu.h                   |   5 +-
 include/linux/iommu.h                       |  14 ++
 include/uapi/linux/vfio.h                   |   4 +
 drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c |   5 +-
 drivers/iommu/dma-iommu.c                   |   4 +-
 drivers/iommu/iommu.c                       | 220 +++++++++++++++++++-
 drivers/pci/pci-acpi.c                      |  13 +-
 drivers/pci/pci.c                           |  65 +++++-
 drivers/pci/quirks.c                        |  19 +-
 9 files changed, 326 insertions(+), 23 deletions(-)

2.43.0

The iommu_deferred_attach() function invokes __iommu_attach_device(), but doesn't hold the group->mutex like other __iommu_attach_device() callers.

Though there is no pratical bug being triggered so far, it would be better to apply the same locking to this __iommu_attach_device(), since the IOMMU drivers nowaday are more aware of the group->mutex -- some of them use the iommu_group_mutex_assert() function that could be potentially in the path of an attach_dev callback function invoked by the __iommu_attach_device().

Worth mentioning that the iommu_deferred_attach() will soon need to check group->resetting_domain that must be locked also.

Thus, grab the mutex to guard __iommu_attach_device() like other callers.

---
 drivers/iommu/iommu.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

index 2ca990dfbb88..170e522b5bda 100644
@@ -2185,10 +2185,17 @@ EXPORT_SYMBOL_GPL(iommu_attach_device);
 
 int iommu_deferred_attach(struct device *dev, struct iommu_domain *domain)
 {
-	if (dev->iommu && dev->iommu->attach_deferred)
-		return __iommu_attach_device(domain, dev, NULL);
+	/*
+	 * This is called on the dma mapping fast path so avoid locking. This is
+	 * racy, but we have an expectation that the driver will setup its DMAs
+	 * inside probe while being single threaded to avoid racing.
+	 */
+	if (!dev->iommu || !dev->iommu->attach_deferred)
+		return 0;
 
-	return 0;
+	guard(mutex)(&dev->iommu_group->mutex);
+
+	return __iommu_attach_device(domain, dev, NULL);
 }
 
 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)

2.43.0

This function can only be called on the default_domain. Trivally pass it in. In all three existing cases, the default domain was just attached to the device.

This avoids iommu_setup_dma_ops() calling iommu_get_domain_for_dev() that will be used by external callers.

---
 drivers/iommu/dma-iommu.h | 5 +++--
 drivers/iommu/dma-iommu.c | 4 +---
 drivers/iommu/iommu.c     | 6 +++---
 3 files changed, 7 insertions(+), 8 deletions(-)

index eca201c1f963..040d00252563 100644
@@ -9,7 +9,7 @@
 
 #ifdef CONFIG_IOMMU_DMA
 
-void iommu_setup_dma_ops(struct device *dev);
+void iommu_setup_dma_ops(struct device *dev, struct iommu_domain *domain);
 
 int iommu_get_dma_cookie(struct iommu_domain *domain);
 void iommu_put_dma_cookie(struct iommu_domain *domain);

@@ -26,7 +26,8 @@ extern bool iommu_dma_forcedac;
 
 #else /* CONFIG_IOMMU_DMA */
 
-static inline void iommu_setup_dma_ops(struct device *dev)
+static inline void iommu_setup_dma_ops(struct device *dev,
+				       struct iommu_domain *domain)
 {
 }
 

index c92088855450..aeaf8fad985c 100644
@@ -2097,10 +2097,8 @@ void dma_iova_destroy(struct device *dev, struct dma_iova_state *state,
 }
 EXPORT_SYMBOL_GPL(dma_iova_destroy);
 
-void iommu_setup_dma_ops(struct device *dev)
+void iommu_setup_dma_ops(struct device *dev, struct iommu_domain *domain)
 {
-	struct iommu_domain *domain = iommu_get_domain_for_dev(dev);
-
 	if (dev_is_pci(dev))
 		dev->iommu->pci_32bit_workaround = !iommu_dma_forcedac;
 

index 170e522b5bda..1e322f87b171 100644
@@ -661,7 +661,7 @@ static int __iommu_probe_device(struct device *dev, struct list_head *group_list
 	}
 
 	if (group->default_domain)
-		iommu_setup_dma_ops(dev);
+		iommu_setup_dma_ops(dev, group->default_domain);
 
 	mutex_unlock(&group->mutex);
 

@@ -1949,7 +1949,7 @@ static int bus_iommu_probe(const struct bus_type *bus)
 			return ret;
 		}
 		for_each_group_device(group, gdev)
-			iommu_setup_dma_ops(gdev->dev);
+			iommu_setup_dma_ops(gdev->dev, group->default_domain);
 		mutex_unlock(&group->mutex);
 
 		/*

@@ -3155,7 +3155,7 @@ static ssize_t iommu_group_store_type(struct iommu_group *group,
 
 	/* Make sure dma_ops is appropriatley set */
 	for_each_group_device(group, gdev)
-		iommu_setup_dma_ops(gdev->dev);
+		iommu_setup_dma_ops(gdev->dev, group->default_domain);
 
 out_unlock:
 	mutex_unlock(&group->mutex);

2.43.0

There is a need to stage a resetting PCI device to temporarily the blocked domain and then attach back to its previously attached domain after reset.

This can be simply done by keeping the "previously attached domain" in the iommu_group->domain pointer while adding an iommu_group->resetting_domain, which gives troubles to IOMMU drivers using the iommu_get_domain_for_dev() for a device's physical domain in order to program IOMMU hardware.

And in such for-driver use cases, the iommu_group->mutex must be held, so it doesn't fit in external callers that don't hold the iommu_group->mutex.

Introduce a new iommu_driver_get_domain_for_dev() helper, exclusively for driver use cases that hold the iommu_group->mutex, to separate from those external use cases.

Add a lockdep_assert_not_held to the existing iommu_get_domain_for_dev() and highlight that in a kdoc.

---
 include/linux/iommu.h                       |  1 +
 drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c |  5 ++--
 drivers/iommu/iommu.c                       | 28 +++++++++++++++++++++
 3 files changed, 32 insertions(+), 2 deletions(-)

index 8c66284a91a8..ff097df318b9 100644
@@ -910,6 +910,7 @@ extern int iommu_attach_device(struct iommu_domain *domain,
 extern void iommu_detach_device(struct iommu_domain *domain,
 				struct device *dev);
 extern struct iommu_domain *iommu_get_domain_for_dev(struct device *dev);
+struct iommu_domain *iommu_driver_get_domain_for_dev(struct device *dev);
 extern struct iommu_domain *iommu_get_dma_domain(struct device *dev);
 extern int iommu_map(struct iommu_domain *domain, unsigned long iova,
 		     phys_addr_t paddr, size_t size, int prot, gfp_t gfp);

index d16d35c78c06..b8d2fef3ee6b 100644
@@ -3125,7 +3125,8 @@ int arm_smmu_set_pasid(struct arm_smmu_master *master,
 		       struct arm_smmu_domain *smmu_domain, ioasid_t pasid,
 		       struct arm_smmu_cd *cd, struct iommu_domain *old)
 {
-	struct iommu_domain *sid_domain = iommu_get_domain_for_dev(master->dev);
+	struct iommu_domain *sid_domain =
+		iommu_driver_get_domain_for_dev(master->dev);
 	struct arm_smmu_attach_state state = {
 		.master = master,
 		.ssid = pasid,

@@ -3191,7 +3192,7 @@ static int arm_smmu_blocking_set_dev_pasid(struct iommu_domain *new_domain,
 	 */
 	if (!arm_smmu_ssids_in_use(&master->cd_table)) {
 		struct iommu_domain *sid_domain =
-			iommu_get_domain_for_dev(master->dev);
+			iommu_driver_get_domain_for_dev(master->dev);
 
 		if (sid_domain->type == IOMMU_DOMAIN_IDENTITY ||
 		    sid_domain->type == IOMMU_DOMAIN_BLOCKED)

index 1e322f87b171..672597100e9a 100644
@@ -2217,6 +2217,15 @@ void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
 }
 EXPORT_SYMBOL_GPL(iommu_detach_device);
 
+/**
+ * iommu_get_domain_for_dev() - Return the DMA API domain pointer
+ * @dev: Device to query
+ *
+ * This function can be called within a driver bound to dev. The returned
+ * pointer is valid for the lifetime of the bound driver.
+ *
+ * It should not be called by drivers with driver_managed_dma = true.
+ */
 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
 {
 	/* Caller must be a probed driver on dev */

@@ -2225,10 +2234,29 @@ struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
 	if (!group)
 		return NULL;
 
+	lockdep_assert_not_held(&group->mutex);
+
 	return group->domain;
 }
 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
 
+/**
+ * iommu_driver_get_domain_for_dev() - Return the driver-level domain pointer
+ * @dev: Device to query
+ *
+ * This function can be called by an iommu driver that wants to get the physical
+ * domain within an iommu callback function where group->mutex is held.
+ */
+struct iommu_domain *iommu_driver_get_domain_for_dev(struct device *dev)
+{
+	struct iommu_group *group = dev->iommu_group;
+
+	lockdep_assert_held(&group->mutex);
+
+	return group->domain;
+}
+EXPORT_SYMBOL_GPL(iommu_driver_get_domain_for_dev);
+
 /*
  * For IOMMU_DOMAIN_DMA implementations which already provide their own
  * guarantees that the group and its default domain are valid and correct.

2.43.0

PCIe permits a device to ignore ATS invalidation TLPs while processing a reset. This creates a problem visible to the OS where an ATS invalidation command will time out. E.g. an SVA domain will have no coordination with a reset event and can racily issue ATS invalidations to a resetting device.

The OS should do something to mitigate this as we do not want production systems to be reporting critical ATS failures, especially in a hypervisor environment. Broadly, OS could arrange to ignore the timeouts, block page table mutations to prevent invalidations, or disable and block ATS.

The PCIe r6.0, sec 10.3.1 IMPLEMENTATION NOTE recommends SW to disable and block ATS before initiating a Function Level Reset. It also mentions that other reset methods could have the same vulnerability as well.

Provide a callback from the PCI subsystem that will enclose the reset and have the iommu core temporarily change all the attached RID/PASID domains group->blocking_domain so that the IOMMU hardware would fence any incoming ATS queries. And IOMMU drivers should also synchronously stop issuing new ATS invalidations and wait for all ATS invalidations to complete. This can avoid any ATS invaliation timeouts.

However, if there is a domain attachment/replacement happening during an ongoing reset, ATS routines may be re-activated between the two function calls. So, introduce a new resetting_domain in the iommu_group structure to reject any concurrent attach_dev/set_dev_pasid call during a reset for a concern of compatibility failure. Since this changes the behavior of an attach operation, update the uAPI accordingly.

Note that there are two corner cases:
1. Devices in the same iommu_group
Since an attachment is always per iommu_group, this means that any sibling devices in the iommu_group cannot change domain, to prevent race conditions.
2. An SR-IOV PF that is being reset while its VF is not
In such case, the VF itself is already broken. So, there is no point in preventing PF from going through the iommu reset.

---
 include/linux/iommu.h     |  13 +++
 include/uapi/linux/vfio.h |   4 +
 drivers/iommu/iommu.c     | 173 ++++++++++++++++++++++++++++++++++++++
 3 files changed, 190 insertions(+)

index ff097df318b9..54b8b48c762e 100644
@@ -1188,6 +1188,10 @@ void iommu_detach_device_pasid(struct iommu_domain *domain,
 			       struct device *dev, ioasid_t pasid);
 ioasid_t iommu_alloc_global_pasid(struct device *dev);
 void iommu_free_global_pasid(ioasid_t pasid);
+
+/* PCI device reset functions */
+int pci_dev_reset_iommu_prepare(struct pci_dev *pdev);
+void pci_dev_reset_iommu_done(struct pci_dev *pdev);
 #else /* CONFIG_IOMMU_API */
 
 struct iommu_ops {};

@@ -1511,6 +1515,15 @@ static inline ioasid_t iommu_alloc_global_pasid(struct device *dev)
 }
 
 static inline void iommu_free_global_pasid(ioasid_t pasid) {}
+
+static inline int pci_dev_reset_iommu_prepare(struct pci_dev *pdev)
+{
+	return 0;
+}
+
+static inline void pci_dev_reset_iommu_done(struct pci_dev *pdev)
+{
+}
 #endif /* CONFIG_IOMMU_API */
 
 #ifdef CONFIG_IRQ_MSI_IOMMU

index ac2329f24141..bb7b89330d35 100644
@@ -964,6 +964,10 @@ struct vfio_device_bind_iommufd {
  * hwpt corresponding to the given pt_id.
  *
  * Return: 0 on success, -errno on failure.
+ *
+ * When a device is resetting, -EBUSY will be returned to reject any concurrent
+ * attachment to the resetting device itself or any sibling device in the IOMMU
+ * group having the resetting device.
  */
 struct vfio_device_attach_iommufd_pt {
 	__u32	argsz;

2.43.0

PCIe permits a device to ignore ATS invalidation TLPs while processing a reset. This creates a problem visible to the OS where an ATS invalidation command will time out: e.g. an SVA domain will have no coordination with a reset event and can racily issue ATS invalidations to a resetting device.

The PCIe r6.0, sec 10.3.1 IMPLEMENTATION NOTE recommends SW to disable and block ATS before initiating a Function Level Reset. It also mentions that other reset methods could have the same vulnerability as well.

The IOMMU subsystem provides pci_dev_reset_iommu_prepare/done() callback helpers for this matter. Use them in all the existing reset functions.

This will attach the device to its iommu_group->blocking_domain during the device reset, so as to allow IOMMU driver to:
- invoke pci_disable_ats() and pci_enable_ats(), if necessary
- wait for all ATS invalidations to complete
- stop issuing new ATS invalidations
- fence any incoming ATS queries

---
 drivers/pci/pci-acpi.c | 13 +++++++--
 drivers/pci/pci.c      | 65 +++++++++++++++++++++++++++++++++++++-----
 drivers/pci/quirks.c   | 19 +++++++++++-
 3 files changed, 87 insertions(+), 10 deletions(-)

index 9369377725fa..651d9b5561ff 100644
@@ -9,6 +9,7 @@
 
 #include <linux/delay.h>
 #include <linux/init.h>
+#include <linux/iommu.h>
 #include <linux/irqdomain.h>
 #include <linux/pci.h>
 #include <linux/msi.h>

@@ -971,6 +972,7 @@ void pci_set_acpi_fwnode(struct pci_dev *dev)
 int pci_dev_acpi_reset(struct pci_dev *dev, bool probe)
 {
 	acpi_handle handle = ACPI_HANDLE(&dev->dev);
+	int ret;
 
 	if (!handle || !acpi_has_method(handle, "_RST"))
 		return -ENOTTY;

@@ -978,12 +980,19 @@ int pci_dev_acpi_reset(struct pci_dev *dev, bool probe)
 	if (probe)
 		return 0;
 
+	ret = pci_dev_reset_iommu_prepare(dev);
+	if (ret) {
+		pci_err(dev, "failed to stop IOMMU for a PCI reset: %d\n", ret);
+		return ret;
+	}
+
 	if (ACPI_FAILURE(acpi_evaluate_object(handle, "_RST", NULL, NULL))) {
 		pci_warn(dev, "ACPI _RST failed\n");
-		return -ENOTTY;
+		ret = -ENOTTY;
 	}
 
-	return 0;
+	pci_dev_reset_iommu_done(dev);
+	return ret;
 }
 
 bool acpi_pci_power_manageable(struct pci_dev *dev)

index b9c252aa6fe0..c6b999045c70 100644
@@ -21,6 +21,7 @@
 #include <linux/pci.h>
 #include <linux/isa-dma.h> /* isa_dma_bridge_buggy */
 #include <linux/init.h>
+#include <linux/iommu.h>
 #include <linux/delay.h>
 #include <linux/acpi.h>
 #include <linux/dmi.h>

@@ -4228,6 +4229,22 @@ static const struct pci_dev_reset_methods pci_dev_reset_methods[] = {
 	{ 0 }
 };
 
+static int __pci_dev_specific_reset(struct pci_dev *dev, bool probe,
+				    const struct pci_dev_reset_methods *i)
+{
+	int ret;
+
+	ret = pci_dev_reset_iommu_prepare(dev);
+	if (ret) {
+		pci_err(dev, "failed to stop IOMMU for a PCI reset: %d\n", ret);
+		return ret;
+	}
+
+	ret = i->reset(dev, probe);
+	pci_dev_reset_iommu_done(dev);
+	return ret;
+}
+
 /*
  * These device-specific reset methods are here rather than in a driver
  * because when a host assigns a device to a guest VM, the host may need

@@ -4242,7 +4259,7 @@ int pci_dev_specific_reset(struct pci_dev *dev, bool probe)
 		     i->vendor == (u16)PCI_ANY_ID) &&
 		    (i->device == dev->device ||
 		     i->device == (u16)PCI_ANY_ID))
-			return i->reset(dev, probe);
+			return __pci_dev_specific_reset(dev, probe, i);
 	}
 
 	return -ENOTTY;

2.43.0

[+cc Danny; this question was prompted by sashiko review of a pci_ats_supported() patch at
https://lore.kernel.org/all/[email protected]]

I've posted a fix for a few times actually. The latest one was posted yesterday after a split:
https://lore.kernel.org/linux-iommu/c66d5221a8e4356316d4f614c8584a80dc9d71b6.1793729707.git.nicolinc@nvidia.com/

Please help review it (and the other PCI patch) as well.

Thanks
Nicolin

Well, that's embarrassing ;)

I acked the PCI patches there on the assumption that they'll go via the IOMMU tree since the interesting bits are there.

Thanks! The acks are in time.

Nicolin

Applied, thanks.