iommufd: Add vIOMMU infrastructure (Part-1)

From: Nicolin Chen <[email protected]>
Sent: Tuesday, October 22, 2024 8:19 AM

Prepare for an embedded structure design for driver-level iommufd_viommu objects:

    // include/linux/iommufd.h
    struct iommufd_viommu {
        struct iommufd_object obj;
        ....
    };

    // Some IOMMU driver
    struct iommu_driver_viommu {
        struct iommufd_viommu core;
        ....
    };

It has to expose struct iommufd_object and enum iommufd_object_type from
the core-level private header to the public iommufd header.

Reviewed-by Jason Gunthorpe <[email protected]>

Signed-off-by Nicolin Chen <[email protected]>

+ * @viommu_alloc: Allocate an iommufd_viommu on a physical IOMMU instance behind

+ *                the @dev, as the set of virtualization resources shared/passed
+ *                to user space IOMMU instance. And associate it with a nesting
+ *                @parent_domain. The @viommu_type must be defined in the header
+ *                include/uapi/linux/iommufd.h
+ *                It is suggested to call iommufd_viommu_alloc() helper for
+ *                a bundled allocation of the core and the driver structures,
+ *                using the given @ictx pointer.

* @pgsize_bitmap: bitmap of all possible supported page sizes
* @owner: Driver module providing these ops
* @identity_domain: An always available, always attachable identity

Quoted 15 lines · Nicolin Chen

+ * @viommu_alloc: Allocate an iommufd_viommu on a physical IOMMU instance behind

+ *                the @dev, as the set of virtualization resources shared/passed
+ *                to user space IOMMU instance. And associate it with a nesting
+ *                @parent_domain. The @viommu_type must be defined in the header
+ *                include/uapi/linux/iommufd.h
+ *                It is suggested to call iommufd_viommu_alloc() helper for
+ *                a bundled allocation of the core and the driver structures,
+ *                using the given @ictx pointer.

* @pgsize_bitmap: bitmap of all possible supported page sizes
* @owner: Driver module providing these ops
* @identity_domain: An always available, always attachable identity

include/linux/iommu.h +0 -0

@@ -591,6 +601,10 @@ struct iommu_ops {
      void (*remove_dev_pasid)(struct device *dev, ioasid_t pasid,
                               struct iommu_domain *domain);
 

include/linux/iommu.h struct iommu_ops { · line 606

Quoted 3 lines

include/linux/iommu.h +3 -0

+     struct iommufd_viommu *(*viommu_alloc)(
+             struct device *dev, struct iommu_domain *parent_domain,
+             struct iommufd_ctx *ictx, unsigned int viommu_type);

Is the vIOMMU object limited to a parent domain?

Quoted 16 lines · Baolu Lu

Quoted 15 lines · Nicolin Chen

+ * @viommu_alloc: Allocate an iommufd_viommu on a physical IOMMU instance behind

+ *                the @dev, as the set of virtualization resources shared/passed
+ *                to user space IOMMU instance. And associate it with a nesting
+ *                @parent_domain. The @viommu_type must be defined in the header
+ *                include/uapi/linux/iommufd.h
+ *                It is suggested to call iommufd_viommu_alloc() helper for
+ *                a bundled allocation of the core and the driver structures,
+ *                using the given @ictx pointer.

* @pgsize_bitmap: bitmap of all possible supported page sizes
* @owner: Driver module providing these ops
* @identity_domain: An always available, always attachable identity

include/linux/iommu.h +0 -0

@@ -591,6 +601,10 @@ struct iommu_ops {
       void (*remove_dev_pasid)(struct device *dev, ioasid_t pasid,
                                struct iommu_domain *domain);
 

Quoted 4 lines

include/linux/iommu.h struct iommu_ops { · line 606

Quoted 3 lines

include/linux/iommu.h +3 -0

+     struct iommufd_viommu *(*viommu_alloc)(
+             struct device *dev, struct iommu_domain *parent_domain,
+             struct iommufd_ctx *ictx, unsigned int viommu_type);

Is the vIOMMU object limited to a parent domain?

Yes, for each vIOMMU (a slice of physical IOMMU per VM), one S2 parent domain is enough. Typically, it has the mappings between gPAs and hPAs. If its format/compatibility allows, this single parent domain can be shared with other vIOMMUs.

Is it feasible to make vIOMMU object more generic, rather than strictly tying it to nested translation? For example, a normal paging domain that translates gPAs to hPAs could also have a vIOMMU object associated with it.

While we can only support vIOMMU object allocation uAPI for S2 paging domains in the context of this series, we could consider leaving the option open to associate a vIOMMU object with other normal paging domains that are not a nested parent?

Quoted 5 lines · Baolu Lu

Is it feasible to make vIOMMU object more generic, rather than strictly tying it to nested translation? For example, a normal paging domain that translates gPAs to hPAs could also have a vIOMMU object associated with it.

Quoted 4 lines

While we can only support vIOMMU object allocation uAPI for S2 paging domains in the context of this series, we could consider leaving the option open to associate a vIOMMU object with other normal paging domains that are not a nested parent?

Why? The nested parent flavour of the domain is basically free to create, what reason would be to not do that?

If the HW doesn't support it, then does the HW really need/support a VIOMMU?

I suppose it could be made up to the driver, but for now I think we should leave it as is in the core code requiring it until we have a reason to relax it.

From: Jason Gunthorpe <[email protected]>
Sent: Tuesday, October 22, 2024 9:16 PM

Quoted 9 lines · Baolu Lu

Is it feasible to make vIOMMU object more generic, rather than strictly tying it to nested translation? For example, a normal paging domain that translates gPAs to hPAs could also have a vIOMMU object associated with it.

While we can only support vIOMMU object allocation uAPI for S2 paging domains in the context of this series, we could consider leaving the option open to associate a vIOMMU object with other normal paging domains that are not a nested parent?

Why? The nested parent flavour of the domain is basically free to create, what reason would be to not do that?

If the HW doesn't support it, then does the HW really need/support a VIOMMU?

Quoted 15 lines

From: Jason Gunthorpe <[email protected]>
Sent: Tuesday, October 22, 2024 9:16 PM

Quoted 9 lines · Baolu Lu

Is it feasible to make vIOMMU object more generic, rather than strictly tying it to nested translation? For example, a normal paging domain that translates gPAs to hPAs could also have a vIOMMU object associated with it.

While we can only support vIOMMU object allocation uAPI for S2 paging domains in the context of this series, we could consider leaving the option open to associate a vIOMMU object with other normal paging domains that are not a nested parent?

Quoted 5 lines

Why? The nested parent flavour of the domain is basically free to create, what reason would be to not do that?

If the HW doesn't support it, then does the HW really need/support a VIOMMU?

Now it's agreed to build trusted I/O on top of this new vIOMMU object. format-wise probably it's free to assume that nested parent is supported on any new platform which will support trusted I/O. But I'm not sure all the conditions around allowing nested are same as for trusted I/O, e.g. for ARM nesting is allowed only for CANWBS/S2FWB. Are they always guaranteed in trusted I/O configuration?

Baolu did raise a good open to confirm given it will be used beyond nesting. 😊

From: Jason Gunthorpe <[email protected]>
Sent: Friday, October 25, 2024 11:24 PM

Quoted 7 lines · Kevin

Quoted 7 lines

From: Jason Gunthorpe <[email protected]>
Sent: Tuesday, October 22, 2024 9:16 PM

Quoted 2 lines · Baolu Lu

Is it feasible to make vIOMMU object more generic, rather than strictly tying it to nested translation? For example, a normal paging domain

that

Quoted 1 line

Quoted 1 line

Quoted 1 line

translates gPAs to hPAs could also have a vIOMMU object associated

with

Quoted 13 lines

Quoted 7 lines

Quoted 6 lines

it.

While we can only support vIOMMU object allocation uAPI for S2 paging domains in the context of this series, we could consider leaving the option open to associate a vIOMMU object with other normal paging domains that are not a nested parent?

Quoted 5 lines

Why? The nested parent flavour of the domain is basically free to create, what reason would be to not do that?

If the HW doesn't support it, then does the HW really need/support a VIOMMU?

Quoted 6 lines

Now it's agreed to build trusted I/O on top of this new vIOMMU object. format-wise probably it's free to assume that nested parent is supported on any new platform which will support trusted I/O. But I'm not sure all the conditions around allowing nested are same as for trusted I/O, e.g. for ARM nesting is allowed only for CANWBS/S2FWB. Are they always guaranteed in trusted I/O configuration?

ARM is a big ? what exactly will come, but I'm expecting that to be resolved either with continued HW support or Linux will add the cache flushing and relax the test.

Quoted 2 lines

Baolu did raise a good open to confirm given it will be used beyond nesting. 😊

Even CC is "nesting", it is just nested with a fixed Identity S1 in the baseline case. The S2 translation still exists and still has to be consistent with whatever the secure world is doing.

So, my feeling is that the S2 nested domain is mandatory for the viommu, especially for CC, it must exists. In the end there may be more options than just a nested parent.

For instance if the CC design relies on the secure world sharing the CPU and IOMMU page table we might need a new HWPT type to represent that configuration.

From a uapi perspective we seem OK here as the hwpt input could be anything. We might have to adjust some checks in the kernel someday.

From: Nicolin Chen <[email protected]>
Sent: Tuesday, October 22, 2024 8:19 AM

To support driver-allocated vIOMMU objects, it's suggested to call the allocator helper in IOMMU dirvers. However, there is no guarantee that drivers will all use it and allocate objects properly.

Add a helper for iommufd core to verify if an unfinalized object is at least reserved in the ictx.

Signed-off-by Nicolin Chen <[email protected]>

From: Nicolin Chen <[email protected]>
Sent: Tuesday, October 22, 2024 8:19 AM

Add a new ioctl for user space to do a vIOMMU allocation. It must be based on a nesting parent HWPT, so take its refcount.

If an IOMMU driver supports a driver-managed vIOMMU object, it must define

+/**
+ * struct iommu_viommu_alloc - ioctl(IOMMU_VIOMMU_ALLOC)
+ * @size: sizeof(struct iommu_viommu_alloc)
+ * @flags: Must be 0
+ * @type: Type of the virtual IOMMU. Must be defined in enum iommu_viommu_type

+ * @dev_id: The device's physical IOMMU will be used to back the virtual IOMMU
+ * @hwpt_id: ID of a nesting parent HWPT to associate to
+ * @out_viommu_id: Output virtual IOMMU ID for the allocated object
+ *
+ * Allocate a virtual IOMMU object that represents the underlying physical
+ * IOMMU's virtualization support. The vIOMMU object is a security-isolated
+ * slice of the physical IOMMU HW that is unique to a specific VM.

Quoted 8 lines

From: Nicolin Chen <[email protected]>
Sent: Tuesday, October 22, 2024 8:19 AM

Add a new ioctl for user space to do a vIOMMU allocation. It must be based on a nesting parent HWPT, so take its refcount.

If an IOMMU driver supports a driver-managed vIOMMU object, it must define

why highlight 'driver-managed', implying a core-managed vIOMMU object some day?

Quoted 6 lines

+/**
+ * struct iommu_viommu_alloc - ioctl(IOMMU_VIOMMU_ALLOC)
+ * @size: sizeof(struct iommu_viommu_alloc)
+ * @flags: Must be 0
+ * @type: Type of the virtual IOMMU. Must be defined in enum iommu_viommu_type

Quoted 8 lines

+ * @dev_id: The device's physical IOMMU will be used to back the virtual IOMMU
+ * @hwpt_id: ID of a nesting parent HWPT to associate to
+ * @out_viommu_id: Output virtual IOMMU ID for the allocated object
+ *
+ * Allocate a virtual IOMMU object that represents the underlying physical
+ * IOMMU's virtualization support. The vIOMMU object is a security-isolated
+ * slice of the physical IOMMU HW that is unique to a specific VM.

the object itself is a software abstraction, while a 'slice' is a set of real hw resources.

Quoted 4 lines

From: Nicolin Chen <[email protected]>
Sent: Tuesday, October 22, 2024 8:19 AM

(quoted diff) +2 -0

+
+     viommu->type = cmd->type;

Review

Quoted 6 lines

drivers/iommu/iommufd/viommu.c new file +3 -0

+     viommu->ictx = ucmd->ictx;
+     viommu->hwpt = hwpt_paging;
+     /* Assume physical IOMMUs are unpluggable (the most likely case)

*/

(quoted diff) +2 -0

+     viommu->iommu_dev = __iommu_get_iommu_dev(idev->dev);
+

so what would happen if this assumption breaks?

From: Nicolin Chen <[email protected]>
Sent: Tuesday, October 22, 2024 8:19 AM

Allow IOMMU driver to use a vIOMMU object that holds a nesting parent hwpt/domain to allocate a nested domain.

Suggested-by Jason Gunthorpe <[email protected]>

Signed-off-by Nicolin Chen <[email protected]>

drivers/iommu/iommufd/hw_pagetable.c +10 -0

+{
+	struct iommufd_hwpt_nested *hwpt_nested;
+	struct iommufd_hw_pagetable *hwpt;
+	int rc;
+
+	if (!viommu->ops || !viommu->ops->domain_alloc_nested)
+		return ERR_PTR(-EOPNOTSUPP);
+
+	hwpt_nested = __iommufd_object_alloc(
+		viommu->ictx, hwpt_nested, IOMMUFD_OBJ_HWPT_NESTED,

common.obj);

drivers/iommu/iommufd/hw_pagetable.c +9 -0

+	if (IS_ERR(hwpt_nested))
+		return ERR_CAST(hwpt_nested);
+	hwpt = &hwpt_nested->common;
+
+	hwpt_nested->viommu = viommu;
+	hwpt_nested->parent = viommu->hwpt;
+	refcount_inc(&viommu->obj.users);
+
+	hwpt->domain = viommu->ops->domain_alloc_nested(viommu,

user_data);

drivers/iommu/iommufd/hw_pagetable.c +8 -0

+	if (IS_ERR(hwpt->domain)) {
+		rc = PTR_ERR(hwpt->domain);
+		hwpt->domain = NULL;
+		goto out_abort;
+	}
+	hwpt->domain->owner = viommu->iommu_dev->ops;
+
+	if (WARN_ON_ONCE(hwpt->domain->type !=

IOMMU_DOMAIN_NESTED)) {

Quoted 7 lines

From: Nicolin Chen <[email protected]>
Sent: Tuesday, October 22, 2024 8:19 AM

(quoted diff) +2 -0

+static struct iommufd_hwpt_nested *
+iommufd_hwpt_nested_alloc_for_viommu(struct iommufd_viommu

*viommu,
+ const struct iommu_user_data *user_data)

probably "_for" can be skipped to reduce the name length

looks there missed a check on flags in this path.

From: Nicolin Chen <[email protected]>
Sent: Tuesday, October 22, 2024 8:20 AM

With the introduction of the new object and its infrastructure, update the doc to reflect that and add a new graph.

Reviewed-by Jason Gunthorpe <[email protected]>

Signed-off-by Nicolin Chen <[email protected]>

From: Nicolin Chen <[email protected]>
Sent: Tuesday, October 22, 2024 8:20 AM

Add a new driver-type for ARM SMMUv3 to enum iommu_viommu_type. Implement
an arm_vsmmu_alloc() with its viommu op arm_vsmmu_domain_alloc_nested(),
to replace arm_smmu_domain_alloc_nesting(). As an initial step, copy the VMID from s2_parent. A later cleanup series is required to move the VMID allocation out of the stage-2 domain allocation routine to this.

After that, replace nested_domain->s2_parent with nested_domain->vsmmu.

Note that the validatting conditions for a nested_domain allocation are moved from arm_vsmmu_domain_alloc_nested to arm_vsmmu_alloc, since there
is no point in creating a vIOMMU (vsmmu) from the beginning if it would not support a nested_domain.

Signed-off-by Nicolin Chen <[email protected]>

Quoted 19 lines

From: Nicolin Chen <[email protected]>
Sent: Tuesday, October 22, 2024 8:20 AM

Add a new driver-type for ARM SMMUv3 to enum iommu_viommu_type. Implement
an arm_vsmmu_alloc() with its viommu op arm_vsmmu_domain_alloc_nested(),
to replace arm_smmu_domain_alloc_nesting(). As an initial step, copy the VMID from s2_parent. A later cleanup series is required to move the VMID allocation out of the stage-2 domain allocation routine to this.

After that, replace nested_domain->s2_parent with nested_domain->vsmmu.

Note that the validatting conditions for a nested_domain allocation are moved from arm_vsmmu_domain_alloc_nested to arm_vsmmu_alloc, since there
is no point in creating a vIOMMU (vsmmu) from the beginning if it would not support a nested_domain.

Quoted 1 line

Signed-off-by Nicolin Chen <[email protected]>

hmm I wonder whether this series should be merged with Jason's nesting series together and directly use vIOMMU to create nesting. Otherwise it looks a bit weird for one series to first enable a uAPI which is immediately replaced by another uAPI from the following series.

From: Nicolin Chen <[email protected]>
Sent: Tuesday, October 22, 2024 8:19 AM

This series introduces a new vIOMMU infrastructure and related ioctls.

IOMMUFD has been using the HWPT infrastructure for all cases, including a nested IO page table support. Yet, there're limitations for an HWPT-based structure to support some advanced HW-accelerated features, such as CMDQV
on NVIDIA Grace, and HW-accelerated vIOMMU on AMD. Even for a multi- IOMMU
environment, it is not straightforward for nested HWPTs to share the same parent HWPT (stage-2 IO pagetable), with the HWPT infrastructure alone: a parent HWPT typically hold one stage-2 IO pagetable and tag it with only one ID in the cache entries. When sharing one large stage-2 IO pagetable across physical IOMMU instances, that one ID may not always be available across all the IOMMU instances. In other word, it's ideal for SW to have a different container for the stage-2 IO pagetable so it can hold another ID that's available.

For this "different container", add vIOMMU, an additional layer to hold extra virtualization information:

________________________________________________________________
_______

 |                      iommufd (with vIOMMU)                            |
 |                                                                       |
 |                             [5]                                       |
 |                        _____________                                  |
 |                       |             |                                 |
 |      |----------------|    vIOMMU   |                                 |
 |      |                |             |                                 |
 |      |                |             |                                 |
 |      |      [1]       |             |          [4]             [2]    |
 |      |     ______     |             |     _____________     ________  |
 |      |    |      |    |     [3]     |    |             |   |        | |
 |      |    | IOAS |<---|(HWPT_PAGING)|<---| HWPT_NESTED |<--| DEVICE | |
 |      |    |______|    |_____________|    |_____________|   |________| |
 |      |        |              |                  |               |     |

|______|________|______________|__________________|_____________
__|_____|
        |        |              |                  |               |
  ______v_____   |        ______v_____       ______v_____       ___v__
 |   struct   |  |  PFN  |  (paging)  |     |  (nested)  |     |struct|
 |iommu_device|  |------>|iommu_domain|<----|iommu_domain|<----
|device|
 |____________|   storage|____________|     |____________|     |______|

The vIOMMU object should be seen as a slice of a physical IOMMU instance that is passed to or shared with a VM. That can be some HW/SW resources:
- Security namespace for guest owned ID, e.g. guest-controlled cache tags
- Access to a sharable nesting parent pagetable across physical IOMMUs
- Virtualization of various platforms IDs, e.g. RIDs and others
- Delivery of paravirtualized invalidation
- Direct assigned invalidation queues
- Direct assigned interrupts
- Non-affiliated event reporting

On a multi-IOMMU system, the vIOMMU object must be instanced to the number
of the physical IOMMUs that are passed to (via devices) a guest VM, while

being able to hold the shareable parent HWPT. Each vIOMMU then just needs
to allocate its own individual ID to tag its own cache:

                     ----------------------------
 ----------------    |         |  paging_hwpt0  |
 | hwpt_nested0 |--->| viommu0 ------------------
 ----------------    |         |      IDx       |
                     ----------------------------
                     ----------------------------
 ----------------    |         |  paging_hwpt0  |
 | hwpt_nested1 |--->| viommu1 ------------------
 ----------------    |         |      IDy       |
                     ----------------------------

As an initial part-1, add IOMMUFD_CMD_VIOMMU_ALLOC ioctl for an allocation
only. And implement it in arm-smmu-v3 driver as a real world use case.

More vIOMMU-based structs and ioctls will be introduced in the follow-up series to support vDEVICE, vIRQ (vEVENT) and vQUEUE objects. Although we repurposed the vIOMMU object from an earlier RFC, just for a referece:
https://lore.kernel.org/all/[email protected]/

This series is on Github:
https://github.com/nicolinc/iommufd/commits/iommufd_viommu_p1-v4
(paring QEMU branch for testing will be provided with the part2 series)

Changelog
v4
* Added "Reviewed-by" from Jason
* Dropped IOMMU_VIOMMU_TYPE_DEFAULT support
* Dropped iommufd_object_alloc_elm renamings
* Renamed iommufd's viommu_api.c to driver.c
* Reworked iommufd_viommu_alloc helper
* Added a separate iommufd_hwpt_nested_alloc_for_viommu function for hwpt_nested allocations on a vIOMMU, and added comparison between viommu->iommu_dev->ops and dev_iommu_ops(idev->dev)
* Replaced s2_parent with vsmmu in arm_smmu_nested_domain
* Replaced domain_alloc_user in iommu_ops with domain_alloc_nested in viommu_ops
* Replaced wait_queue_head_t with a completion, to delay the unplug of mock_iommu_dev
* Corrected documentation graph that was missing struct iommu_device
* Added an iommufd_verify_unfinalized_object helper to verify driver- allocated vIOMMU/vDEVICE objects
* Added missing test cases for TEST_LENGTH and fail_nth
v3
https://lore.kernel.org/all/[email protected]/
* Rebased on top of Jason's nesting v3 series
https://lore.kernel.org/all/0-v3-e2e16cd7467f+2a6a1-
[email protected]/
* Split the series into smaller parts
* Added Jason's Reviewed-by
* Added back viommu->iommu_dev
* Added support for driver-allocated vIOMMU v.s. core-allocated
* Dropped arm_smmu_cache_invalidate_user
* Added an iommufd_test_wait_for_users() in selftest
* Reworked test code to make viommu an individual FIXTURE
* Added missing TEST_LENGTH case for the new ioctl command
v2
https://lore.kernel.org/all/[email protected]/
* Limited vdev_id to one per idev
* Added a rw_sem to protect the vdev_id list
* Reworked driver-level APIs with proper lockings
* Added a new viommu_api file for IOMMUFD_DRIVER config
* Dropped useless iommu_dev point from the viommu structure
* Added missing index numnbers to new types in the uAPI header
* Dropped IOMMU_VIOMMU_INVALIDATE uAPI; Instead, reuse the HWPT
one
* Reworked mock_viommu_cache_invalidate() using the new iommu helper
* Reordered details of set/unset_vdev_id handlers for proper lockings
v1
https://lore.kernel.org/all/[email protected]/

Thanks!
Nicolin

Nicolin Chen (11):
iommufd: Move struct iommufd_object to public iommufd header
iommufd: Introduce IOMMUFD_OBJ_VIOMMU and its related struct
iommufd: Add iommufd_verify_unfinalized_object iommufd/viommu: Add IOMMU_VIOMMU_ALLOC ioctl
iommufd: Add domain_alloc_nested op to iommufd_viommu_ops
iommufd: Allow pt_id to carry viommu_id for IOMMU_HWPT_ALLOC iommufd/selftest: Add refcount to mock_iommu_device iommufd/selftest: Add IOMMU_VIOMMU_TYPE_SELFTEST iommufd/selftest: Add IOMMU_VIOMMU_ALLOC test coverage
Documentation: userspace-api: iommufd: Update vIOMMU iommu/arm-smmu-v3: Add IOMMU_VIOMMU_TYPE_ARM_SMMUV3
support

 drivers/iommu/iommufd/Makefile                |  5 +-
 drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h   | 26 +++---
 drivers/iommu/iommufd/iommufd_private.h       | 36 ++------
 drivers/iommu/iommufd/iommufd_test.h          |  2 +
 include/linux/iommu.h                         | 14 +++
 include/linux/iommufd.h                       | 89 +++++++++++++++++++
 include/uapi/linux/iommufd.h                  | 56 ++++++++++--
 tools/testing/selftests/iommu/iommufd_utils.h | 28 ++++++
 .../arm/arm-smmu-v3/arm-smmu-v3-iommufd.c     | 79 ++++++++++------
 drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c   |  9 +-
 drivers/iommu/iommufd/driver.c                | 38 ++++++++
 drivers/iommu/iommufd/hw_pagetable.c          | 69 +++++++++++++-
 drivers/iommu/iommufd/main.c                  | 58 ++++++------
 drivers/iommu/iommufd/selftest.c              | 73 +++++++++++++--
 drivers/iommu/iommufd/viommu.c                | 85 ++++++++++++++++++
 tools/testing/selftests/iommu/iommufd.c       | 78 ++++++++++++++++
 .../selftests/iommu/iommufd_fail_nth.c        | 11 +++
 Documentation/userspace-api/iommufd.rst       | 69 +++++++++++++-
 18 files changed, 701 insertions(+), 124 deletions(-)
 create mode 100644 drivers/iommu/iommufd/driver.c
 create mode 100644 drivers/iommu/iommufd/viommu.c

--
2.43.0

Quoted 9 lines

The vIOMMU object should be seen as a slice of a physical IOMMU instance that is passed to or shared with a VM. That can be some HW/SW resources:
- Security namespace for guest owned ID, e.g. guest-controlled cache tags
- Access to a sharable nesting parent pagetable across physical IOMMUs
- Virtualization of various platforms IDs, e.g. RIDs and others
- Delivery of paravirtualized invalidation
- Direct assigned invalidation queues
- Direct assigned interrupts
- Non-affiliated event reporting

sorry no idea about 'non-affiliated event'. Can you elaborate?

From: Jason Gunthorpe <[email protected]>
Sent: Friday, October 25, 2024 11:43 PM

Quoted 1 line · Kevin

Quoted 1 line

The vIOMMU object should be seen as a slice of a physical IOMMU

instance

Quoted 1 line

Quoted 1 line

that is passed to or shared with a VM. That can be some HW/SW

resources:

Quoted 1 line

Quoted 1 line

- Security namespace for guest owned ID, e.g. guest-controlled cache

tags

Quoted 7 lines

Quoted 6 lines

- Access to a sharable nesting parent pagetable across physical IOMMUs
- Virtualization of various platforms IDs, e.g. RIDs and others
- Delivery of paravirtualized invalidation
- Direct assigned invalidation queues
- Direct assigned interrupts
- Non-affiliated event reporting

Quoted 1 line

sorry no idea about 'non-affiliated event'. Can you elaborate?

This would be an even that is not a connected to a device

For instance a CMDQ experienced a problem.

Quoted 5 lines

From: Nicolin Chen <[email protected]>
Sent: Tuesday, October 22, 2024 8:19 AM

This series introduces a new vIOMMU infrastructure and related ioctls.

Quoted 14 lines

IOMMUFD has been using the HWPT infrastructure for all cases, including a nested IO page table support. Yet, there're limitations for an HWPT-based structure to support some advanced HW-accelerated features, such as CMDQV
on NVIDIA Grace, and HW-accelerated vIOMMU on AMD. Even for a multi- IOMMU
environment, it is not straightforward for nested HWPTs to share the same parent HWPT (stage-2 IO pagetable), with the HWPT infrastructure alone: a parent HWPT typically hold one stage-2 IO pagetable and tag it with only one ID in the cache entries. When sharing one large stage-2 IO pagetable across physical IOMMU instances, that one ID may not always be available across all the IOMMU instances. In other word, it's ideal for SW to have a different container for the stage-2 IO pagetable so it can hold another ID that's available.

Just holding multiple IDs doesn't require a different container. This is just a side effect when vIOMMU will be required for other said reasons.

If we have to put more words here I'd prefer to adding a bit more for CMDQV which is more compelling. not a big deal though. 😊

Quoted 20 lines

For this "different container", add vIOMMU, an additional layer to hold extra virtualization information:

________________________________________________________________
_______

 |                      iommufd (with vIOMMU)                            |
 |                                                                       |
 |                             [5]                                       |
 |                        _____________                                  |
 |                       |             |                                 |
 |      |----------------|    vIOMMU   |                                 |
 |      |                |             |                                 |
 |      |                |             |                                 |
 |      |      [1]       |             |          [4]             [2]    |
 |      |     ______     |             |     _____________     ________  |
 |      |    |      |    |     [3]     |    |             |   |        | |
 |      |    | IOAS |<---|(HWPT_PAGING)|<---| HWPT_NESTED |<--| DEVICE | |
 |      |    |______|    |_____________|    |_____________|   |________| |
 |      |        |              |                  |               |     |

Quoted 9 lines

|______|________|______________|__________________|_____________
__|_____|
        |        |              |                  |               |
  ______v_____   |        ______v_____       ______v_____       ___v__
 |   struct   |  |  PFN  |  (paging)  |     |  (nested)  |     |struct|
 |iommu_device|  |------>|iommu_domain|<----|iommu_domain|<----
|device|
 |____________|   storage|____________|     |____________|     |______|

nit - [1] ... [5] can be removed.

Quoted 9 lines

The vIOMMU object should be seen as a slice of a physical IOMMU instance that is passed to or shared with a VM. That can be some HW/SW resources:
- Security namespace for guest owned ID, e.g. guest-controlled cache tags
- Access to a sharable nesting parent pagetable across physical IOMMUs
- Virtualization of various platforms IDs, e.g. RIDs and others
- Delivery of paravirtualized invalidation
- Direct assigned invalidation queues
- Direct assigned interrupts
- Non-affiliated event reporting

sorry no idea about 'non-affiliated event'. Can you elaborate?

Quoted 3 lines

On a multi-IOMMU system, the vIOMMU object must be instanced to the number
of the physical IOMMUs that are passed to (via devices) a guest VM, while

'to the number of the physical IOMMUs that have a slice passed to ..."