iommu: Add iommu_group_get/set_domain()

The hardware assistant vfio mediated device is a use case of iommu aux-domain. The interactions between vfio/mdev and iommu during mdev creation and passthr are:

- Create a group for mdev with iommu_group_alloc();
- Add the device to the group with

         group = iommu_group_alloc();
         if (IS_ERR(group))
                 return PTR_ERR(group);

         ret = iommu_group_add_device(group, &mdev->dev);
         if (!ret)
                 dev_info(&mdev->dev, "MDEV: group_id = %d\n",
                          iommu_group_id(group));

- Allocate an aux-domain
iommu_domain_alloc()
- Attach the aux-domain to the physical device from which the mdev is created.
iommu_aux_attach_device()

In the whole process, an iommu group was allocated for the mdev and an iommu domain was attached to the group, but the group->domain leaves NULL. As the result, iommu_get_domain_for_dev() doesn't work anymore.

This adds iommu_group_get/set_domain() so that group->domain could be managed whenever a domain is attached or detached through the aux-domain api's.

Fixes 7bd50f0cd2fd5 ("vfio/type1: Add domain at(de)taching group helpers")

Signed-off-by Lu Baolu <[email protected]>

---
  drivers/iommu/iommu.c | 28 ++++++++++++++++++++++++++++
  include/linux/iommu.h | 14 ++++++++++++++
  2 files changed, 42 insertions(+)

drivers/iommu/iommu.c +28 -0

index d43120eb1dc5..e2b665303d70 100644
@@ -715,6 +715,34 @@ int iommu_group_set_name(struct iommu_group *group, const char *name)
  }
  EXPORT_SYMBOL_GPL(iommu_group_set_name);
 
+/**
+ * iommu_group_get_domain - get domain of a group
+ * @group: the group
+ *
+ * This is called to get the domain of a group.
+ */
+struct iommu_domain *iommu_group_get_domain(struct iommu_group *group)
+{
+	return group->domain;
+}
+EXPORT_SYMBOL_GPL(iommu_group_get_domain);
+
+/**
+ * iommu_group_set_domain - set domain for a group
+ * @group: the group
+ * @domain: iommu domain
+ *
+ * This is called to set the domain for a group. In aux-domain case, a domain
+ * might attach or detach to an iommu group through the aux-domain apis, but
+ * the group->domain doesn't get a chance to be updated there.
+ */
+void iommu_group_set_domain(struct iommu_group *group,
+			    struct iommu_domain *domain)
+{
+	group->domain = domain;
+}
+EXPORT_SYMBOL_GPL(iommu_group_set_domain);
+
  static int iommu_create_device_direct_mappings(struct iommu_group *group,
  					       struct device *dev)
  {

include/linux/iommu.h +7 -0

index 5f0b7859d2eb..ff88d548a870 100644
@@ -496,6 +496,9 @@ extern void iommu_group_set_iommudata(struct iommu_group *group,
  				      void *iommu_data,
  				      void (*release)(void *iommu_data));
  extern int iommu_group_set_name(struct iommu_group *group, const char *name);
+extern struct iommu_domain *iommu_group_get_domain(struct iommu_group *group);
+extern void iommu_group_set_domain(struct iommu_group *group,
+				   struct iommu_domain *domain);
  extern int iommu_group_add_device(struct iommu_group *group,
  				  struct device *dev);
  extern void iommu_group_remove_device(struct device *dev);

@@ -840,6 +843,17 @@ static inline int iommu_group_set_name(struct iommu_group *group,
  	return -ENODEV;
  }
 
+static inline
+struct iommu_domain *iommu_group_get_domain(struct iommu_group *group)
+{
+	return NULL;

Quoted 24 lines · Lu Baolu

The hardware assistant vfio mediated device is a use case of iommu aux-domain. The interactions between vfio/mdev and iommu during mdev creation and passthr are:

- Create a group for mdev with iommu_group_alloc();
- Add the device to the group with
         group = iommu_group_alloc();
         if (IS_ERR(group))
                 return PTR_ERR(group);

         ret = iommu_group_add_device(group, &mdev->dev); if (!ret)
                 dev_info(&mdev->dev, "MDEV: group_id = %d\n", iommu_group_id(group));
- Allocate an aux-domain
    iommu_domain_alloc()
- Attach the aux-domain to the physical device from which the mdev is created.
    iommu_aux_attach_device()

In the whole process, an iommu group was allocated for the mdev and an iommu domain was attached to the group, but the group->domain leaves NULL. As the result, iommu_get_domain_for_dev() doesn't work anymore.

Quoted 3 lines

This adds iommu_group_get/set_domain() so that group->domain could be managed whenever a domain is attached or detached through the aux-domain api's.

Letting external callers poke around directly in the internals of iommu_group doesn't look right to me.

If a regular device is attached to one or more aux domains for PASID use, iommu_get_domain_for_dev() is still going to return the primary domain, so why should it be expected to behave differently for mediated

devices? AFAICS it's perfectly legitimate to have no primary domain if traffic-without-PASID is invalid.

Robin.

Hi Robin,

Quoted 29 lines · Robin Murphy

Quoted 27 lines · Lu Baolu

The hardware assistant vfio mediated device is a use case of iommu aux-domain. The interactions between vfio/mdev and iommu during mdev creation and passthr are:

- Create a group for mdev with iommu_group_alloc();
- Add the device to the group with
         group = iommu_group_alloc();
         if (IS_ERR(group))
                 return PTR_ERR(group);

         ret = iommu_group_add_device(group, &mdev->dev); if (!ret)
                 dev_info(&mdev->dev, "MDEV: group_id = %d\n", iommu_group_id(group));
- Allocate an aux-domain
    iommu_domain_alloc()
- Attach the aux-domain to the physical device from which the mdev is created.
    iommu_aux_attach_device()

In the whole process, an iommu group was allocated for the mdev and an iommu domain was attached to the group, but the group->domain leaves NULL. As the result, iommu_get_domain_for_dev() doesn't work anymore.

This adds iommu_group_get/set_domain() so that group->domain could be managed whenever a domain is attached or detached through the aux-domain api's.

Quoted 1 line

Letting external callers poke around directly in the internals of iommu_group doesn't look right to me.

Unfortunately, it seems that the vifo iommu abstraction is deeply bound to the IOMMU subsystem. We can easily find other examples:

iommu_group_get/set_iommudata()
iommu_group_get/set_name()
...

Quoted 1 line

If a regular device is attached to one or more aux domains for PASID use, iommu_get_domain_for_dev() is still going to return the primary domain, so why should it be expected to behave differently for mediated

Unlike the normal device attach, we will encounter two devices when it comes to aux-domain.

- Parent physical device - this might be, for example, a PCIe device with PASID feature support, hence it is able to tag an unique PASID for DMA transfers originated from its subset. The device driver hence is able to wrapper this subset into an isolated:

- Mediated device - a fake device created by the device driver mentioned above.

Yes. All you mentioned are right for the parent device. But for mediated device, iommu_get_domain_for_dev() doesn't work even it has an valid iommu_group and iommu_domain.

iommu_get_domain_for_dev() is a necessary interface for device drivers which want to support aux-domain. For example,

          struct iommu_domain *domain;
          struct device *dev = mdev_dev(mdev);
      unsigned long pasid;

          domain = iommu_get_domain_for_dev(dev); if (!domain)
                  return -ENODEV;

          pasid = iommu_aux_get_pasid(domain, dev->parent); if (pasid == IOASID_INVALID)
          return -EINVAL;

      /* Program the device context with the PASID value */
      ....

Without this fix, iommu_get_domain_for_dev() always returns NULL and the device driver has no means to support aux-domain.

Quoted 32 lines · Lu Baolu

Hi Robin,

Quoted 29 lines · Robin Murphy

Quoted 27 lines · Lu Baolu

The hardware assistant vfio mediated device is a use case of iommu aux-domain. The interactions between vfio/mdev and iommu during mdev creation and passthr are:

- Create a group for mdev with iommu_group_alloc();
- Add the device to the group with
         group = iommu_group_alloc();
         if (IS_ERR(group))
                 return PTR_ERR(group);

         ret = iommu_group_add_device(group, &mdev->dev); if (!ret)
                 dev_info(&mdev->dev, "MDEV: group_id = %d\n", iommu_group_id(group));
- Allocate an aux-domain
    iommu_domain_alloc()
- Attach the aux-domain to the physical device from which the mdev is created.
    iommu_aux_attach_device()

In the whole process, an iommu group was allocated for the mdev and an iommu domain was attached to the group, but the group->domain leaves NULL. As the result, iommu_get_domain_for_dev() doesn't work anymore.

This adds iommu_group_get/set_domain() so that group->domain could be managed whenever a domain is attached or detached through the aux-domain api's.

Quoted 8 lines

Quoted 1 line

Letting external callers poke around directly in the internals of iommu_group doesn't look right to me.

Unfortunately, it seems that the vifo iommu abstraction is deeply bound to the IOMMU subsystem. We can easily find other examples:

iommu_group_get/set_iommudata()
iommu_group_get/set_name()
...

Sure, but those are ways for users of a group to attach useful information of their own to it, that doesn't matter to the IOMMU subsystem itself. The interface you've proposed gives callers rich new opportunities to fundamentally break correct operation of the API:

    dom = iommu_domain_alloc(); iommu_attach_group(dom, grp);
    ...
    iommu_group_set_domain(grp, NULL);
    // oops, leaked and can't ever detach properly now

or perhaps:

    grp = iommu_group_alloc();
    iommu_group_add_device(grp, dev);
    iommu_group_set_domain(grp, dom);
    ...
    iommu_detach_group(dom, grp);
    // oops, IOMMU driver might not handle this

Quoted 13 lines

Quoted 1 line

If a regular device is attached to one or more aux domains for PASID use, iommu_get_domain_for_dev() is still going to return the primary domain, so why should it be expected to behave differently for mediated

Unlike the normal device attach, we will encounter two devices when it comes to aux-domain.

- Parent physical device - this might be, for example, a PCIe device with PASID feature support, hence it is able to tag an unique PASID for DMA transfers originated from its subset. The device driver hence is able to wrapper this subset into an isolated:

- Mediated device - a fake device created by the device driver mentioned above.

Quoted 6 lines

Yes. All you mentioned are right for the parent device. But for mediated device, iommu_get_domain_for_dev() doesn't work even it has an valid iommu_group and iommu_domain.

iommu_get_domain_for_dev() is a necessary interface for device drivers which want to support aux-domain. For example,

Only if they want to follow this very specific notion of using made-up devices and groups to represent aux attachments. Even if a driver managing its own aux domains entirely privately does create child devices for them, it's not like it can't keep its domain pointers in drvdata if it wants to ;)

Let's not conflate the current implementation of vfio_mdev with the general concepts involved here.

Quoted 12 lines

           struct iommu_domain *domain;
           struct device *dev = mdev_dev(mdev);
       unsigned long pasid;

           domain = iommu_get_domain_for_dev(dev); if (!domain)
                   return -ENODEV;

           pasid = iommu_aux_get_pasid(domain, dev->parent); if (pasid == IOASID_INVALID)
           return -EINVAL;

Quoted 5 lines

       /* Program the device context with the PASID value */
       ....

Without this fix, iommu_get_domain_for_dev() always returns NULL and the device driver has no means to support aux-domain.

So either the IOMMU API itself is missing the ability to do the right thing internally, or the mdev layer isn't using it appropriately. Either way, simply punching holes in the API for mdev to hack around its own mess doesn't seem like the best thing to do.

The initial impression I got was that it's implicitly assumed here that the mdev itself is attached to exactly one aux domain and nothing else, at which point I would wonder why it's using aux at all, but are you saying that in fact no attach happens with the mdev group either way, only to the parent device?

I'll admit I'm not hugely familiar with any of this, but it seems to me that the logical flow should be:

    - allocate domain
    - attach as aux to parent
    - retrieve aux domain PASID
    - create mdev child based on PASID
    - attach mdev to domain (normally)

Of course that might require giving the IOMMU API a proper first-class notion of mediated devices, such that it knows the mdev represents the PASID, and can recognise the mdev attach is equivalent to the earlier parent aux attach so not just blindly hand it down to an IOMMU driver that's never heard of this new device before. Or perhaps the IOMMU drivers do their own bookkeeping for the mdev bus, such that they do handle the attach call, and just validate it internally based on the associated parent device and PASID. Either way, the inside maintains self-consistency and from the outside it looks like standard API usage without nasty hacks.

I'm pretty sure I've heard suggestions of using mediated devices beyond VFIO (e.g. within the kernel itself), so chances are this is a direction that we'll have to take at some point anyway.

And, that said, even if people do want an immediate quick fix regardless of technical debt, I'd still be a lot happier to see iommu_group_set_domain() lightly respun as iommu_attach_mdev() ;)

Hi Robin,

Quoted 113 lines · Robin Murphy

Quoted 32 lines · Lu Baolu

Hi Robin,

Quoted 29 lines · Robin Murphy

Quoted 27 lines · Lu Baolu

The hardware assistant vfio mediated device is a use case of iommu aux-domain. The interactions between vfio/mdev and iommu during mdev creation and passthr are:

- Create a group for mdev with iommu_group_alloc();
- Add the device to the group with
         group = iommu_group_alloc();
         if (IS_ERR(group))
                 return PTR_ERR(group);

         ret = iommu_group_add_device(group, &mdev->dev); if (!ret)
                 dev_info(&mdev->dev, "MDEV: group_id = %d\n", iommu_group_id(group));
- Allocate an aux-domain
    iommu_domain_alloc()
- Attach the aux-domain to the physical device from which the mdev is created.
    iommu_aux_attach_device()

In the whole process, an iommu group was allocated for the mdev and an iommu domain was attached to the group, but the group->domain leaves NULL. As the result, iommu_get_domain_for_dev() doesn't work anymore.

This adds iommu_group_get/set_domain() so that group->domain could be managed whenever a domain is attached or detached through the aux-domain api's.

Quoted 8 lines

Quoted 1 line

Letting external callers poke around directly in the internals of iommu_group doesn't look right to me.

Unfortunately, it seems that the vifo iommu abstraction is deeply bound to the IOMMU subsystem. We can easily find other examples:

iommu_group_get/set_iommudata()
iommu_group_get/set_name()
...

Sure, but those are ways for users of a group to attach useful information of their own to it, that doesn't matter to the IOMMU subsystem itself. The interface you've proposed gives callers rich new opportunities to fundamentally break correct operation of the API:

     dom = iommu_domain_alloc(); iommu_attach_group(dom, grp);
     ...
     iommu_group_set_domain(grp, NULL);
     // oops, leaked and can't ever detach properly now

or perhaps:

     grp = iommu_group_alloc();
     iommu_group_add_device(grp, dev);
     iommu_group_set_domain(grp, dom);
     ...
     iommu_detach_group(dom, grp);
     // oops, IOMMU driver might not handle this

Quoted 13 lines

Quoted 1 line

If a regular device is attached to one or more aux domains for PASID use, iommu_get_domain_for_dev() is still going to return the primary domain, so why should it be expected to behave differently for mediated

Unlike the normal device attach, we will encounter two devices when it comes to aux-domain.

- Parent physical device - this might be, for example, a PCIe device with PASID feature support, hence it is able to tag an unique PASID for DMA transfers originated from its subset. The device driver hence is able to wrapper this subset into an isolated:

- Mediated device - a fake device created by the device driver mentioned above.

Quoted 6 lines

Yes. All you mentioned are right for the parent device. But for mediated device, iommu_get_domain_for_dev() doesn't work even it has an valid iommu_group and iommu_domain.

iommu_get_domain_for_dev() is a necessary interface for device drivers which want to support aux-domain. For example,

Only if they want to follow this very specific notion of using made-up devices and groups to represent aux attachments. Even if a driver managing its own aux domains entirely privately does create child devices for them, it's not like it can't keep its domain pointers in drvdata if it wants to ;)

Let's not conflate the current implementation of vfio_mdev with the general concepts involved here.

Quoted 12 lines

           struct iommu_domain *domain;
           struct device *dev = mdev_dev(mdev);
       unsigned long pasid;

           domain = iommu_get_domain_for_dev(dev); if (!domain)
                   return -ENODEV;

           pasid = iommu_aux_get_pasid(domain, dev->parent); if (pasid == IOASID_INVALID)
           return -EINVAL;

Quoted 5 lines

       /* Program the device context with the PASID value */
       ....

Without this fix, iommu_get_domain_for_dev() always returns NULL and the device driver has no means to support aux-domain.

So either the IOMMU API itself is missing the ability to do the right thing internally, or the mdev layer isn't using it appropriately. Either way, simply punching holes in the API for mdev to hack around its own mess doesn't seem like the best thing to do.

The initial impression I got was that it's implicitly assumed here that the mdev itself is attached to exactly one aux domain and nothing else, at which point I would wonder why it's using aux at all, but are you saying that in fact no attach happens with the mdev group either way, only to the parent device?

I'll admit I'm not hugely familiar with any of this, but it seems to me that the logical flow should be:

     - allocate domain
     - attach as aux to parent
     - retrieve aux domain PASID
     - create mdev child based on PASID
     - attach mdev to domain (normally)

Quoted 5 lines

Of course that might require giving the IOMMU API a proper first-class notion of mediated devices, such that it knows the mdev represents the PASID, and can recognise the mdev attach is equivalent to the earlier parent aux attach so not just blindly hand it down to an IOMMU driver that's never heard of this new device before. Or perhaps the IOMMU drivers do their own bookkeeping for the mdev bus, such that they do handle the attach call, and just validate it internally based on the associated parent device and PASID. Either way, the inside maintains self-consistency and from the outside it looks like standard API usage without nasty hacks.

I'm pretty sure I've heard suggestions of using mediated devices beyond VFIO (e.g. within the kernel itself), so chances are this is a direction that we'll have to take at some point anyway.

And, that said, even if people do want an immediate quick fix regardless of technical debt, I'd still be a lot happier to see iommu_group_set_domain() lightly respun as iommu_attach_mdev() ;)

Get your point and I agree with your concerns.

To maintain the relationship between mdev's iommu_group and iommu_domain, how about extending below existing aux_attach api

int iommu_aux_attach_device(struct iommu_domain *domain,
                struct device *dev)

by adding the mdev's iommu_group?

int iommu_aux_attach_device(struct iommu_domain *domain,
                struct device *dev,
                struct iommu_group *group)

And, in iommu_aux_attach_device(), we require,
 - @group only has a single device;
 - @group hasn't been attached by any devices;
 - Set the @domain to @group

Just like what we've done in iommu_attach_device().

Any thoughts?

Quoted 142 lines · Lu Baolu

Hi Robin,

Quoted 114 lines · Robin Murphy

Quoted 32 lines · Lu Baolu

Hi Robin,

Quoted 29 lines · Robin Murphy

Quoted 27 lines · Lu Baolu

The hardware assistant vfio mediated device is a use case of iommu aux-domain. The interactions between vfio/mdev and iommu during mdev creation and passthr are:

- Create a group for mdev with iommu_group_alloc();
- Add the device to the group with
         group = iommu_group_alloc();
         if (IS_ERR(group))
                 return PTR_ERR(group);

         ret = iommu_group_add_device(group, &mdev->dev); if (!ret)
                 dev_info(&mdev->dev, "MDEV: group_id = %d\n", iommu_group_id(group));
- Allocate an aux-domain
    iommu_domain_alloc()
- Attach the aux-domain to the physical device from which the mdev is created.
    iommu_aux_attach_device()

In the whole process, an iommu group was allocated for the mdev and an iommu domain was attached to the group, but the group->domain leaves NULL. As the result, iommu_get_domain_for_dev() doesn't work anymore.

This adds iommu_group_get/set_domain() so that group->domain could be managed whenever a domain is attached or detached through the aux-domain api's.

Quoted 8 lines

Quoted 1 line

Letting external callers poke around directly in the internals of iommu_group doesn't look right to me.

Unfortunately, it seems that the vifo iommu abstraction is deeply bound to the IOMMU subsystem. We can easily find other examples:

iommu_group_get/set_iommudata()
iommu_group_get/set_name()
...

Sure, but those are ways for users of a group to attach useful information of their own to it, that doesn't matter to the IOMMU subsystem itself. The interface you've proposed gives callers rich new opportunities to fundamentally break correct operation of the API:

     dom = iommu_domain_alloc(); iommu_attach_group(dom, grp);
     ...
     iommu_group_set_domain(grp, NULL);
     // oops, leaked and can't ever detach properly now

or perhaps:

     grp = iommu_group_alloc();
     iommu_group_add_device(grp, dev);
     iommu_group_set_domain(grp, dom);
     ...
     iommu_detach_group(dom, grp);
     // oops, IOMMU driver might not handle this

Quoted 14 lines

Quoted 1 line

If a regular device is attached to one or more aux domains for PASID use, iommu_get_domain_for_dev() is still going to return the primary domain, so why should it be expected to behave differently for mediated

Unlike the normal device attach, we will encounter two devices when it comes to aux-domain.

- Parent physical device - this might be, for example, a PCIe device with PASID feature support, hence it is able to tag an unique PASID for DMA transfers originated from its subset. The device driver hence is able to wrapper this subset into an isolated:

- Mediated device - a fake device created by the device driver mentioned
above.

Quoted 6 lines

Yes. All you mentioned are right for the parent device. But for mediated device, iommu_get_domain_for_dev() doesn't work even it has an valid iommu_group and iommu_domain.

iommu_get_domain_for_dev() is a necessary interface for device drivers which want to support aux-domain. For example,

Only if they want to follow this very specific notion of using made-up devices and groups to represent aux attachments. Even if a driver managing its own aux domains entirely privately does create child devices for them, it's not like it can't keep its domain pointers in drvdata if it wants to ;)

Let's not conflate the current implementation of vfio_mdev with the general concepts involved here.

Quoted 12 lines

           struct iommu_domain *domain;
           struct device *dev = mdev_dev(mdev);
       unsigned long pasid;

           domain = iommu_get_domain_for_dev(dev); if (!domain)
                   return -ENODEV;

           pasid = iommu_aux_get_pasid(domain, dev->parent); if (pasid == IOASID_INVALID)
           return -EINVAL;

Quoted 5 lines

       /* Program the device context with the PASID value */
       ....

Without this fix, iommu_get_domain_for_dev() always returns NULL and the device driver has no means to support aux-domain.

So either the IOMMU API itself is missing the ability to do the right thing internally, or the mdev layer isn't using it appropriately. Either way, simply punching holes in the API for mdev to hack around its own mess doesn't seem like the best thing to do.

The initial impression I got was that it's implicitly assumed here that the mdev itself is attached to exactly one aux domain and nothing else, at which point I would wonder why it's using aux at all, but are you saying that in fact no attach happens with the mdev group either way, only to the parent device?

I'll admit I'm not hugely familiar with any of this, but it seems to me that the logical flow should be:

     - allocate domain
     - attach as aux to parent
     - retrieve aux domain PASID
     - create mdev child based on PASID
     - attach mdev to domain (normally)

Quoted 5 lines

Of course that might require giving the IOMMU API a proper first-class notion of mediated devices, such that it knows the mdev represents the PASID, and can recognise the mdev attach is equivalent to the earlier parent aux attach so not just blindly hand it down to an IOMMU driver that's never heard of this new device before. Or perhaps the IOMMU drivers do their own bookkeeping for the mdev bus, such that they do handle the attach call, and just validate it internally based on the associated parent device and PASID. Either way, the inside maintains self-consistency and from the outside it looks like standard API usage without nasty hacks.

I'm pretty sure I've heard suggestions of using mediated devices beyond VFIO (e.g. within the kernel itself), so chances are this is a direction that we'll have to take at some point anyway.

And, that said, even if people do want an immediate quick fix regardless of technical debt, I'd still be a lot happier to see iommu_group_set_domain() lightly respun as iommu_attach_mdev() ;)

Get your point and I agree with your concerns.

To maintain the relationship between mdev's iommu_group and iommu_domain, how about extending below existing aux_attach api

int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev)

by adding the mdev's iommu_group?

int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev,
                 struct iommu_group *group)

And, in iommu_aux_attach_device(), we require,
  - @group only has a single device;
  - @group hasn't been attached by any devices;
  - Set the @domain to @group

Quoted 3 lines

Just like what we've done in iommu_attach_device().

Any thoughts?

Rather than pass a bare iommu_group with implicit restrictions, it might be neater to just pass an mdev_device, so that the IOMMU core can also take care of allocating and setting up the group. Then we flag the group internally as a special "mdev group" such that we can prevent callers from subsequently trying to add/remove devices or attach/detach its domain directly. That seems like it would make a pretty straightforward and robust API extension, as long as the mdev argument here is optional so that SVA and other aux users don't have to care. Other than the slightly different ordering where caller would have to allocate the mdev first, then finish it's PASID-based configuration afterwards, I guess it's not far off what I was thinking yesterday :)

Quoted 142 lines · Lu Baolu

Hi Robin,

Quoted 114 lines · Robin Murphy

Quoted 32 lines · Lu Baolu

Hi Robin,

Quoted 29 lines · Robin Murphy

Quoted 27 lines · Lu Baolu

The hardware assistant vfio mediated device is a use case of iommu aux-domain. The interactions between vfio/mdev and iommu during mdev creation and passthr are:

- Create a group for mdev with iommu_group_alloc();
- Add the device to the group with
         group = iommu_group_alloc();
         if (IS_ERR(group))
                 return PTR_ERR(group);

         ret = iommu_group_add_device(group, &mdev->dev); if (!ret)
                 dev_info(&mdev->dev, "MDEV: group_id = %d\n", iommu_group_id(group));
- Allocate an aux-domain
    iommu_domain_alloc()
- Attach the aux-domain to the physical device from which the mdev is created.
    iommu_aux_attach_device()

In the whole process, an iommu group was allocated for the mdev and an iommu domain was attached to the group, but the group->domain leaves NULL. As the result, iommu_get_domain_for_dev() doesn't work anymore.

This adds iommu_group_get/set_domain() so that group->domain could be managed whenever a domain is attached or detached through the aux-domain api's.

Quoted 8 lines

Quoted 1 line

Letting external callers poke around directly in the internals of iommu_group doesn't look right to me.

Unfortunately, it seems that the vifo iommu abstraction is deeply bound to the IOMMU subsystem. We can easily find other examples:

iommu_group_get/set_iommudata()
iommu_group_get/set_name()
...

Sure, but those are ways for users of a group to attach useful information of their own to it, that doesn't matter to the IOMMU subsystem itself. The interface you've proposed gives callers rich new opportunities to fundamentally break correct operation of the API:

     dom = iommu_domain_alloc(); iommu_attach_group(dom, grp);
     ...
     iommu_group_set_domain(grp, NULL);
     // oops, leaked and can't ever detach properly now

or perhaps:

     grp = iommu_group_alloc();
     iommu_group_add_device(grp, dev);
     iommu_group_set_domain(grp, dom);
     ...
     iommu_detach_group(dom, grp);
     // oops, IOMMU driver might not handle this

Quoted 14 lines

Quoted 1 line

If a regular device is attached to one or more aux domains for PASID use, iommu_get_domain_for_dev() is still going to return the primary domain, so why should it be expected to behave differently for mediated

Unlike the normal device attach, we will encounter two devices when it comes to aux-domain.

- Parent physical device - this might be, for example, a PCIe device with PASID feature support, hence it is able to tag an unique PASID for DMA transfers originated from its subset. The device driver hence is able to wrapper this subset into an isolated:

- Mediated device - a fake device created by the device driver mentioned
above.

Quoted 6 lines

Yes. All you mentioned are right for the parent device. But for mediated device, iommu_get_domain_for_dev() doesn't work even it has an valid iommu_group and iommu_domain.

iommu_get_domain_for_dev() is a necessary interface for device drivers which want to support aux-domain. For example,

Only if they want to follow this very specific notion of using made-up devices and groups to represent aux attachments. Even if a driver managing its own aux domains entirely privately does create child devices for them, it's not like it can't keep its domain pointers in drvdata if it wants to ;)

Let's not conflate the current implementation of vfio_mdev with the general concepts involved here.

Quoted 12 lines

           struct iommu_domain *domain;
           struct device *dev = mdev_dev(mdev);
       unsigned long pasid;

           domain = iommu_get_domain_for_dev(dev); if (!domain)
                   return -ENODEV;

           pasid = iommu_aux_get_pasid(domain, dev->parent); if (pasid == IOASID_INVALID)
           return -EINVAL;

Quoted 5 lines

       /* Program the device context with the PASID value */
       ....

Without this fix, iommu_get_domain_for_dev() always returns NULL and the device driver has no means to support aux-domain.

So either the IOMMU API itself is missing the ability to do the right thing internally, or the mdev layer isn't using it appropriately. Either way, simply punching holes in the API for mdev to hack around its own mess doesn't seem like the best thing to do.

The initial impression I got was that it's implicitly assumed here that the mdev itself is attached to exactly one aux domain and nothing else, at which point I would wonder why it's using aux at all, but are you saying that in fact no attach happens with the mdev group either way, only to the parent device?

I'll admit I'm not hugely familiar with any of this, but it seems to me that the logical flow should be:

     - allocate domain
     - attach as aux to parent
     - retrieve aux domain PASID
     - create mdev child based on PASID
     - attach mdev to domain (normally)

Quoted 5 lines

Of course that might require giving the IOMMU API a proper first-class notion of mediated devices, such that it knows the mdev represents the PASID, and can recognise the mdev attach is equivalent to the earlier parent aux attach so not just blindly hand it down to an IOMMU driver that's never heard of this new device before. Or perhaps the IOMMU drivers do their own bookkeeping for the mdev bus, such that they do handle the attach call, and just validate it internally based on the associated parent device and PASID. Either way, the inside maintains self-consistency and from the outside it looks like standard API usage without nasty hacks.

I'm pretty sure I've heard suggestions of using mediated devices beyond VFIO (e.g. within the kernel itself), so chances are this is a direction that we'll have to take at some point anyway.

And, that said, even if people do want an immediate quick fix regardless of technical debt, I'd still be a lot happier to see iommu_group_set_domain() lightly respun as iommu_attach_mdev() ;)

Get your point and I agree with your concerns.

To maintain the relationship between mdev's iommu_group and iommu_domain, how about extending below existing aux_attach api

int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev)

by adding the mdev's iommu_group?

int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev,
                 struct iommu_group *group)

And, in iommu_aux_attach_device(), we require,
  - @group only has a single device;
  - @group hasn't been attached by any devices;
  - Set the @domain to @group

Quoted 3 lines

Just like what we've done in iommu_attach_device().

Any thoughts?

Rather than pass a bare iommu_group with implicit restrictions, it might be neater to just pass an mdev_device, so that the IOMMU core can also take care of allocating and setting up the group. Then we flag the group internally as a special "mdev group" such that we can prevent callers from subsequently trying to add/remove devices or attach/detach its domain directly. That seems like it would make a pretty straightforward and robust API extension, as long as the mdev argument here is optional so that SVA and other aux users don't have to care. Other than the slightly different ordering where caller would have to allocate the mdev first, then finish it's PASID-based configuration afterwards, I guess it's not far off what I was thinking yesterday :)

Hi Robin,

Quoted 144 lines · Robin Murphy

Quoted 143 lines · Lu Baolu

Hi Robin,

Quoted 115 lines · Robin Murphy

Quoted 33 lines · Lu Baolu

Hi Robin,

Quoted 30 lines · Robin Murphy

Quoted 28 lines · Lu Baolu

The hardware assistant vfio mediated device is a use case of iommu aux-domain. The interactions between vfio/mdev and iommu during mdev creation and passthr are:

- Create a group for mdev with iommu_group_alloc();
- Add the device to the group with
         group = iommu_group_alloc();
         if (IS_ERR(group))
                 return PTR_ERR(group);

         ret = iommu_group_add_device(group, &mdev->dev); if (!ret)
                 dev_info(&mdev->dev, "MDEV: group_id = %d\n", iommu_group_id(group));
- Allocate an aux-domain
    iommu_domain_alloc()
- Attach the aux-domain to the physical device from which the mdev is
   created.
    iommu_aux_attach_device()

In the whole process, an iommu group was allocated for the mdev and an iommu domain was attached to the group, but the group->domain leaves NULL. As the result, iommu_get_domain_for_dev() doesn't work anymore.

This adds iommu_group_get/set_domain() so that group->domain could be managed whenever a domain is attached or detached through the aux-domain api's.

Quoted 8 lines

Quoted 1 line

Letting external callers poke around directly in the internals of iommu_group doesn't look right to me.

Unfortunately, it seems that the vifo iommu abstraction is deeply bound to the IOMMU subsystem. We can easily find other examples:

iommu_group_get/set_iommudata()
iommu_group_get/set_name()
...

Sure, but those are ways for users of a group to attach useful information of their own to it, that doesn't matter to the IOMMU subsystem itself. The interface you've proposed gives callers rich new opportunities to fundamentally break correct operation of the API:

     dom = iommu_domain_alloc(); iommu_attach_group(dom, grp);
     ...
     iommu_group_set_domain(grp, NULL);
     // oops, leaked and can't ever detach properly now

or perhaps:

     grp = iommu_group_alloc();
     iommu_group_add_device(grp, dev);
     iommu_group_set_domain(grp, dom);
     ...
     iommu_detach_group(dom, grp);
     // oops, IOMMU driver might not handle this

Quoted 14 lines

Quoted 1 line

If a regular device is attached to one or more aux domains for PASID use, iommu_get_domain_for_dev() is still going to return the primary domain, so why should it be expected to behave differently for mediated

Unlike the normal device attach, we will encounter two devices when it comes to aux-domain.

- Parent physical device - this might be, for example, a PCIe device with PASID feature support, hence it is able to tag an unique PASID for DMA transfers originated from its subset. The device driver hence is able to wrapper this subset into an isolated:

- Mediated device - a fake device created by the device driver mentioned
above.

Quoted 6 lines

Yes. All you mentioned are right for the parent device. But for mediated device, iommu_get_domain_for_dev() doesn't work even it has an valid iommu_group and iommu_domain.

iommu_get_domain_for_dev() is a necessary interface for device drivers which want to support aux-domain. For example,

Only if they want to follow this very specific notion of using made-up devices and groups to represent aux attachments. Even if a driver managing its own aux domains entirely privately does create child devices for them, it's not like it can't keep its domain pointers in drvdata if it wants to ;)

Let's not conflate the current implementation of vfio_mdev with the general concepts involved here.

Quoted 12 lines

           struct iommu_domain *domain;
           struct device *dev = mdev_dev(mdev);
       unsigned long pasid;

           domain = iommu_get_domain_for_dev(dev); if (!domain)
                   return -ENODEV;

           pasid = iommu_aux_get_pasid(domain, dev->parent); if (pasid == IOASID_INVALID)
           return -EINVAL;

Quoted 5 lines

       /* Program the device context with the PASID value */
       ....

Without this fix, iommu_get_domain_for_dev() always returns NULL and the device driver has no means to support aux-domain.

So either the IOMMU API itself is missing the ability to do the right thing internally, or the mdev layer isn't using it appropriately. Either way, simply punching holes in the API for mdev to hack around its own mess doesn't seem like the best thing to do.

The initial impression I got was that it's implicitly assumed here that the mdev itself is attached to exactly one aux domain and nothing else, at which point I would wonder why it's using aux at all, but are you saying that in fact no attach happens with the mdev group either way, only to the parent device?

I'll admit I'm not hugely familiar with any of this, but it seems to me that the logical flow should be:

     - allocate domain
     - attach as aux to parent
     - retrieve aux domain PASID
     - create mdev child based on PASID
     - attach mdev to domain (normally)

Quoted 5 lines

Of course that might require giving the IOMMU API a proper first-class notion of mediated devices, such that it knows the mdev represents the PASID, and can recognise the mdev attach is equivalent to the earlier parent aux attach so not just blindly hand it down to an IOMMU driver that's never heard of this new device before. Or perhaps the IOMMU drivers do their own bookkeeping for the mdev bus, such that they do handle the attach call, and just validate it internally based on the associated parent device and PASID. Either way, the inside maintains self-consistency and from the outside it looks like standard API usage without nasty hacks.

I'm pretty sure I've heard suggestions of using mediated devices beyond VFIO (e.g. within the kernel itself), so chances are this is a direction that we'll have to take at some point anyway.

And, that said, even if people do want an immediate quick fix regardless of technical debt, I'd still be a lot happier to see iommu_group_set_domain() lightly respun as iommu_attach_mdev() ;)

Get your point and I agree with your concerns.

To maintain the relationship between mdev's iommu_group and iommu_domain, how about extending below existing aux_attach api

int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev)

by adding the mdev's iommu_group?

int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev,
                 struct iommu_group *group)

And, in iommu_aux_attach_device(), we require,
  - @group only has a single device;
  - @group hasn't been attached by any devices;
  - Set the @domain to @group

Quoted 5 lines

Quoted 3 lines

Just like what we've done in iommu_attach_device().

Any thoughts?

Rather than pass a bare iommu_group with implicit restrictions, it might be neater to just pass an mdev_device, so that the IOMMU core can also take care of allocating and setting up the group. Then we flag the group internally as a special "mdev group" such that we can prevent callers from subsequently trying to add/remove devices or attach/detach its domain directly. That seems like it would make a pretty straightforward and robust API extension, as long as the mdev argument here is optional so that SVA and other aux users don't have to care. Other than the slightly different ordering where caller would have to allocate the mdev first, then finish it's PASID-based configuration afterwards, I guess it's not far off what I was thinking yesterday :)

It looks good to me if we pass an *optional* made-up device instead of iommu_group. But it seems that vfio/mdev assumes an iommu_group first and then attaches domains to the groups. Hence, it's hard to move the group allocation and setting up into the attach interface.

As proposed, the new iommu_aux_attach_device() might look like this:

int iommu_aux_attach_device(struct iommu_domain *domain,
                            struct device *phys_dev, struct device *dev)

where,

@phys_dev: The physical device which supports IOMMU_DEV_FEAT_AUX; @dev: a made-up device which presents the subset resources binding to
      the aux-domain. An example use case is vfio/mdev. For cases where no made-up devices are used, pass NULL instead.

With @dev passed, we can require

- single device in group;
- no previous attaching;
- set up internal logistics between group and domain;

The iommu_aux_detach_device() needs the equivalent extensions.