[PATCH v2 00/17] Add Nested Translation Support for SMMUv3

[Zhangfei Gao]

Hi, Nico

On Wed, 10 May 2023 at 11:34, Nicolin Chen <nicolinc at nvidia.com> wrote:
>
> [ This series is rebased on top of v6.4-rc1 merging Jason's iommu_hwpt
>   branch and Yi's vfio cdev v11 branch, then the replace v7 series and
>   the nesting v2 (candidate) series and Intel VT-d series. Note that
>   some of them are still getting finalized. So, there can be potential
>   minor API changes that would not be reflected in this series. Yet, we
>   can start the review at the SMMU driver specific things.
>
>   @robin, the hw_info patch still requires the errata patch that you
>   mentioned. Perhaps we can merge that separately or include it in v3.
>
>   Thanks! ]
>
> Changelog
> v2:
>  * Added arm_smmu_set_dev_data after the set_dev_data series.
>  * Added Jason's patch "vfio: Remove VFIO_TYPE1_NESTING_IOMMU"
>  * Replaced the iommu_get_unmanaged_domain() helper with Robin's patch.
>  * Reworked the code in arm_smmu_cmdq_build_cmd() to make NH_VA to be
>    a superset of NH_VAA.
>  * Added inline comments and a bug-report link to the patch unsetting
>    dst[2] and dst[3] of STE.
>  * Dropped the to_s2_cfg helper since only one place really needs it.
>  * Dropped the VMID (override) flag and s2vmid in iommu_hwpt_arm_smmuv3
>    structure, because it's expected for user space to use a shared S2
>    domain/hwpt for all devices, i.e. the VMID (allocated with the S2
>    domain is already unified. If there's some special case that still
>    needs a VMID unification, we should probably add it incrementally.
>  * Move the introduction of the "struct arm_smmu_domain *s2" function
>    parameter to the proper patch.
>  * Redefined "struct iommu_hwpt_arm_smmuv3" by adding ste_uptr/len and
>    out_event_uptr/len. Then added an arm_smmu_domain_finalise_nested()
>    function to read guest Stream Table Entry with a proper sanity.
>  * Reworked arm_smmu_cache_invalidate_user() by reading the guest CMDQ
>    directly, to support batching. Also, added return value feedback of
>    -ETIMEDOUT at CMD_SYNC, and reported CERROR_ILL errors via the CONS
>    in the user_data structure.
>  * Updated data/functions following the nesting infrastructure updates.
>  * Added/fixed multiple comments per v1 review inputs.
> v1:
>  https://lore.kernel.org/all/cover.1678348754.git.nicolinc@nvidia.com/
>
> --------------------------------------------------------------------------
>
> Hi all,
>
> This series of patches add nested translation support for ARM SMMUv3.
>
> Eric Auger made a huge effort previously with the VFIO uAPIs, and sent
> his v16 a year ago. Now, the nested translation should follow the new
> IOMMUFD uAPIs design. So, most of the key features are ported from the
> privous VFIO solution, and then rebuilt on top of the IOMMUFD nesting
> infrastructure.
>
> The essential parts in the driver to support a nested translation are
> ->hw_info, ->domain_alloc_user and ->cache_invalidate_user ops. So this
> series fundamentally adds these three functions in the SMMUv3 driver,
> along with several preparations and cleanups for them.
>
> One unique requirement for SMMUv3 nested translation support is the MSI
> doorbell address translation, which is a 2-stage translation too. And,
> to working with the ITS driver, an msi_cookie needs to be setup on the
> kernel-managed domain, the stage-2 domain of the nesting setup. And the
> same msi_cookie will be fetched, via iommu_dma_get_msi_mapping_domain(),
> in the iommu core to allocate and creates IOVA mappings for MSI doorbell
> page(s). However, with the nesting design, the device is attached to a
> user-managed domain, the stage-1 domain. So both the setup and fetching
> of the msi_cookie would not work at the level of stage-2 domain. Thus,
> on both sides, the msi_cookie setup and fetching require a redirection
> of the domain pointer. It's easy to do so in iommufd core, but needs a
> new op in the iommu core and driver.
>
> You can also find this series on the Github:
> https://github.com/nicolinc/iommufd/commits/iommufd_nesting-v2
>
> The kernel branch is tested with this QEMU branch:
> https://github.com/nicolinc/qemu/commits/wip/iommufd_rfcv4+nesting+smmuv3-v2
>

I rebased on these two branches and did some basic tests.

The basic functions work after backport
iommufd: Add IOMMU_PAGE_RESPONSE
iommufd: Add device fault handler support

https://github.com/Linaro/linux-kernel-warpdrive/tree/uacce-devel-6.4
https://github.com/Linaro/qemu/tree/iommufd-6.4-nesting-smmuv3-v2

However when debugging hotplug PCI device, it still does not work,
Segmentation fault same as 6.2.

guest kernel
CONFIG_HOTPLUG_PCI_PCIE=y

boot guest (this info does not appear in 6.2)
qemu-system-aarch64: -device
vfio-pci,host=0000:76:00.1,bus=pci.1,addr=0x0,id=acc1,iommufd=iommufd0:
Failed to set data -1
qemu-system-aarch64: -device
vfio-pci,host=0000:76:00.1,bus=pci.1,addr=0x0,id=acc1,iommufd=iommufd0:
failed to set device data

$ sudo nc -U /tmp/qmpm_1.socket
(qemu) info pci
(qemu) device_del acc1

guest:
qemu-system-aarch64: IOMMU_IOAS_UNMAP failed: No such file or directory
qemu-system-aarch64: vfio_container_dma_unmap(0xaaaae1fc0380,
0x8000000000, 0x10000) = -2 (No such file or directory)
qemu-system-aarch64: Failed to unset data -1
Segmentation fault (core dumped).  // also happened in 6.2

Thanks