Much like arm-smmu in commit 7d835134d4e1 ("iommu/arm-smmu: Make instance lookup robust"), virtio-iommu appears to have the same issue where iommu_device_register() makes the IOMMU instance visible to other API callers (including itself) straight away, but internally the instance isn't ready to recognise itself for viommu_probe_device() to work correctly until after viommu_probe() has returned. This matters a lot more now that bus_iommu_probe() has the DT/VIOT knowledge to probe client devices the way that was always intended. Tweak the lookup and initialisation in much the same way as for arm-smmu, to ensure that what we register is functional and ready to go.
Cc: stable@vger.kernel.org Fixes: bcb81ac6ae3c ("iommu: Get DT/ACPI parsing into the proper probe path") Signed-off-by: Robin Murphy robin.murphy@arm.com ---
v2: Of course generic bus_find_device_by_fwnode() didn't work, since it's dev->parent we need to check rather than dev itself, sigh... --- drivers/iommu/virtio-iommu.c | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 deletions(-)
diff --git a/drivers/iommu/virtio-iommu.c b/drivers/iommu/virtio-iommu.c index 532db1de201b..b39d6f134ab2 100644 --- a/drivers/iommu/virtio-iommu.c +++ b/drivers/iommu/virtio-iommu.c @@ -998,8 +998,7 @@ static void viommu_get_resv_regions(struct device *dev, struct list_head *head) iommu_dma_get_resv_regions(dev, head); }
-static const struct iommu_ops viommu_ops; -static struct virtio_driver virtio_iommu_drv; +static const struct bus_type *virtio_bus_type;
static int viommu_match_node(struct device *dev, const void *data) { @@ -1008,8 +1007,9 @@ static int viommu_match_node(struct device *dev, const void *data)
static struct viommu_dev *viommu_get_by_fwnode(struct fwnode_handle *fwnode) { - struct device *dev = driver_find_device(&virtio_iommu_drv.driver, NULL, - fwnode, viommu_match_node); + struct device *dev = bus_find_device(virtio_bus_type, NULL, fwnode, + viommu_match_node); + put_device(dev);
return dev ? dev_to_virtio(dev)->priv : NULL; @@ -1160,6 +1160,9 @@ static int viommu_probe(struct virtio_device *vdev) if (!viommu) return -ENOMEM;
+ /* Borrow this for easy lookups later */ + virtio_bus_type = dev->bus; + spin_lock_init(&viommu->request_lock); ida_init(&viommu->domain_ids); viommu->dev = dev; @@ -1229,10 +1232,10 @@ static int viommu_probe(struct virtio_device *vdev) if (ret) goto err_free_vqs;
- iommu_device_register(&viommu->iommu, &viommu_ops, parent_dev); - vdev->priv = viommu;
+ iommu_device_register(&viommu->iommu, &viommu_ops, parent_dev); + dev_info(dev, "input address: %u bits\n", order_base_2(viommu->geometry.aperture_end)); dev_info(dev, "page mask: %#llx\n", viommu->pgsize_bitmap);
Hi Robin,
On 8/14/25 6:47 PM, Robin Murphy wrote:
Much like arm-smmu in commit 7d835134d4e1 ("iommu/arm-smmu: Make instance lookup robust"), virtio-iommu appears to have the same issue where iommu_device_register() makes the IOMMU instance visible to other API callers (including itself) straight away, but internally the instance isn't ready to recognise itself for viommu_probe_device() to work correctly until after viommu_probe() has returned. This matters a lot more now that bus_iommu_probe() has the DT/VIOT knowledge to probe client devices the way that was always intended. Tweak the lookup and initialisation in much the same way as for arm-smmu, to ensure that what we register is functional and ready to go.
Cc: stable@vger.kernel.org Fixes: bcb81ac6ae3c ("iommu: Get DT/ACPI parsing into the proper probe path") Signed-off-by: Robin Murphy robin.murphy@arm.com
This version works for me. Thanks for the diligent respin.
Tested-by: Eric Auger eric.auger@redhat.com
Thanks
Eric
v2: Of course generic bus_find_device_by_fwnode() didn't work, since it's dev->parent we need to check rather than dev itself, sigh...
drivers/iommu/virtio-iommu.c | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 deletions(-)
diff --git a/drivers/iommu/virtio-iommu.c b/drivers/iommu/virtio-iommu.c index 532db1de201b..b39d6f134ab2 100644 --- a/drivers/iommu/virtio-iommu.c +++ b/drivers/iommu/virtio-iommu.c @@ -998,8 +998,7 @@ static void viommu_get_resv_regions(struct device *dev, struct list_head *head) iommu_dma_get_resv_regions(dev, head); } -static const struct iommu_ops viommu_ops; -static struct virtio_driver virtio_iommu_drv; +static const struct bus_type *virtio_bus_type; static int viommu_match_node(struct device *dev, const void *data) { @@ -1008,8 +1007,9 @@ static int viommu_match_node(struct device *dev, const void *data) static struct viommu_dev *viommu_get_by_fwnode(struct fwnode_handle *fwnode) {
- struct device *dev = driver_find_device(&virtio_iommu_drv.driver, NULL,
fwnode, viommu_match_node);
- struct device *dev = bus_find_device(virtio_bus_type, NULL, fwnode,
viommu_match_node);- put_device(dev);
return dev ? dev_to_virtio(dev)->priv : NULL; @@ -1160,6 +1160,9 @@ static int viommu_probe(struct virtio_device *vdev) if (!viommu) return -ENOMEM;
- /* Borrow this for easy lookups later */
- virtio_bus_type = dev->bus;
- spin_lock_init(&viommu->request_lock); ida_init(&viommu->domain_ids); viommu->dev = dev;
@@ -1229,10 +1232,10 @@ static int viommu_probe(struct virtio_device *vdev) if (ret) goto err_free_vqs;
- iommu_device_register(&viommu->iommu, &viommu_ops, parent_dev);
- vdev->priv = viommu;
- iommu_device_register(&viommu->iommu, &viommu_ops, parent_dev);
- dev_info(dev, "input address: %u bits\n", order_base_2(viommu->geometry.aperture_end)); dev_info(dev, "page mask: %#llx\n", viommu->pgsize_bitmap);
Hi Robin, Joerg,
On 8/14/25 6:47 PM, Robin Murphy wrote:
Much like arm-smmu in commit 7d835134d4e1 ("iommu/arm-smmu: Make instance lookup robust"), virtio-iommu appears to have the same issue where iommu_device_register() makes the IOMMU instance visible to other API callers (including itself) straight away, but internally the instance isn't ready to recognise itself for viommu_probe_device() to work correctly until after viommu_probe() has returned. This matters a lot more now that bus_iommu_probe() has the DT/VIOT knowledge to probe client devices the way that was always intended. Tweak the lookup and initialisation in much the same way as for arm-smmu, to ensure that what we register is functional and ready to go.
Cc: stable@vger.kernel.org Fixes: bcb81ac6ae3c ("iommu: Get DT/ACPI parsing into the proper probe path") Signed-off-by: Robin Murphy robin.murphy@arm.com
could we get that fix merged as soon as possible? the virtio-iommu is currently totally broken.
Thanks
Eric
v2: Of course generic bus_find_device_by_fwnode() didn't work, since it's dev->parent we need to check rather than dev itself, sigh...
drivers/iommu/virtio-iommu.c | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 deletions(-)
diff --git a/drivers/iommu/virtio-iommu.c b/drivers/iommu/virtio-iommu.c index 532db1de201b..b39d6f134ab2 100644 --- a/drivers/iommu/virtio-iommu.c +++ b/drivers/iommu/virtio-iommu.c @@ -998,8 +998,7 @@ static void viommu_get_resv_regions(struct device *dev, struct list_head *head) iommu_dma_get_resv_regions(dev, head); } -static const struct iommu_ops viommu_ops; -static struct virtio_driver virtio_iommu_drv; +static const struct bus_type *virtio_bus_type; static int viommu_match_node(struct device *dev, const void *data) { @@ -1008,8 +1007,9 @@ static int viommu_match_node(struct device *dev, const void *data) static struct viommu_dev *viommu_get_by_fwnode(struct fwnode_handle *fwnode) {
- struct device *dev = driver_find_device(&virtio_iommu_drv.driver, NULL,
fwnode, viommu_match_node);
- struct device *dev = bus_find_device(virtio_bus_type, NULL, fwnode,
viommu_match_node);- put_device(dev);
return dev ? dev_to_virtio(dev)->priv : NULL; @@ -1160,6 +1160,9 @@ static int viommu_probe(struct virtio_device *vdev) if (!viommu) return -ENOMEM;
- /* Borrow this for easy lookups later */
- virtio_bus_type = dev->bus;
- spin_lock_init(&viommu->request_lock); ida_init(&viommu->domain_ids); viommu->dev = dev;
@@ -1229,10 +1232,10 @@ static int viommu_probe(struct virtio_device *vdev) if (ret) goto err_free_vqs;
- iommu_device_register(&viommu->iommu, &viommu_ops, parent_dev);
- vdev->priv = viommu;
- iommu_device_register(&viommu->iommu, &viommu_ops, parent_dev);
- dev_info(dev, "input address: %u bits\n", order_base_2(viommu->geometry.aperture_end)); dev_info(dev, "page mask: %#llx\n", viommu->pgsize_bitmap);
linux-stable-mirror@lists.linaro.org
-
Eric Auger -
Joerg Roedel -
Robin Murphy