>-----Original Message-----
>From: Shameer Kolothum Thodi <[email protected]>
>Subject: RE: [PATCH v4 03/15] vfio/iommufd: Create nesting parent hwpt with
>IOMMU_HWPT_ALLOC_PASID flag
>
>
>
>> -----Original Message-----
>> From: Zhenzhong Duan <[email protected]>
>> Sent: 30 April 2026 08:13
>> To: [email protected]
>> Cc: [email protected]; [email protected]; [email protected];
>> [email protected]; [email protected]; Jason Gunthorpe
>> <[email protected]>; Nicolin Chen <[email protected]>; Shameer Kolothum
>> Thodi <[email protected]>; [email protected];
>> [email protected]; [email protected]; [email protected];
>> [email protected]; Zhenzhong Duan <[email protected]>
>> Subject: [PATCH v4 03/15] vfio/iommufd: Create nesting parent hwpt with
>> IOMMU_HWPT_ALLOC_PASID flag
>>
>> External email: Use caution opening links or attachments
>>
>>
>> When both device and vIOMMU have PASID enabled, then guest may setup
>> pasid attached translation.
>>
>> VFIO needs to be aware of potential pasid usage and should attach the
>> non-pasid part of pasid-capable device to hwpt flagged with
>> IOMMU_HWPT_ALLOC_PASID.
>>
>> ARM SMMU doesn't support IOMMU_HWPT_ALLOC_PASID, only VTD need it.
>> So
>> we can't check the existing vIOMMU flag VIOMMU_FLAG_PASID_SUPPORTED
>> to
>> determine if set flag IOMMU_HWPT_ALLOC_PASID. Instead, introduce a new
>> flag VIOMMU_FLAG_WANT_PASID_ATTACH which will only be exposed by
>> VTD.
>>
>> Opportunistically add documentation for VIOMMU_FLAG_PASID_SUPPORTED
>> and explain the difference with VIOMMU_FLAG_WANT_PASID_ATTACH.
>>
>> Signed-off-by: Zhenzhong Duan <[email protected]>
>> Reviewed-by: Yi Liu <[email protected]>
>> Tested-by: Xudong Hao <[email protected]>
>> ---
>> include/hw/core/iommu.h | 11 +++++++++++
>> include/hw/vfio/vfio-device.h | 1 +
>> hw/vfio/device.c | 11 +++++++++++
>> hw/vfio/iommufd.c | 8 +++++++-
>> 4 files changed, 30 insertions(+), 1 deletion(-)
>>
>> diff --git a/include/hw/core/iommu.h b/include/hw/core/iommu.h
>> index 77739d4214..20d6d79062 100644
>> --- a/include/hw/core/iommu.h
>> +++ b/include/hw/core/iommu.h
>> @@ -20,9 +20,20 @@
>> enum viommu_flags {
>> /* vIOMMU needs nesting parent HWPT to create nested HWPT */
>> VIOMMU_FLAG_WANT_NESTING_PARENT = BIT_ULL(0),
>> + /*
>> + * vIOMMU supports PASID capability, VFIO checks this flag and
>> synthesize
>> + * a PASID capability.
>> + */
>> VIOMMU_FLAG_PASID_SUPPORTED = BIT_ULL(1),
>> /* vIOMMU needs dirty tracking on the nesting parent HWPT for nested use
>> */
>> VIOMMU_FLAG_WANT_NESTING_DIRTY_TRACKING = BIT_ULL(2),
>> + /*
>> + * vIOMMU requests other sub-system like VFIO to create a HWPT that can
>> be
>> + * used with PASID attachment. VIOMMU_FLAG_PASID_SUPPORTED can't
>> be used
>> + * for this purpose as PASID attachment is needed by VTD IOMMU but not
>> ARM
>> + * SMMU.
>> + */
>> + VIOMMU_FLAG_WANT_PASID_ATTACH = BIT_ULL(3),
>> };
>>
>> /* Host IOMMU quirks. Extracted from host IOMMU capabilities */
>> diff --git a/include/hw/vfio/vfio-device.h b/include/hw/vfio/vfio-device.h
>> index 380a55d6e5..8472420d3f 100644
>> --- a/include/hw/vfio/vfio-device.h
>> +++ b/include/hw/vfio/vfio-device.h
>> @@ -282,6 +282,7 @@ void vfio_device_unprepare(VFIODevice *vbasedev);
>>
>> bool vfio_device_get_viommu_flags_want_nesting(VFIODevice *vbasedev);
>> bool vfio_device_get_viommu_flags_want_nesting_dirty(VFIODevice
>> *vbasedev);
>> +bool vfio_device_get_viommu_flags_want_pasid_attach(VFIODevice
>> *vbasedev);
>> bool vfio_device_get_host_iommu_quirk_bypass_ro(VFIODevice *vbasedev,
>> uint32_t type, void *caps,
>> uint32_t size);
>> diff --git a/hw/vfio/device.c b/hw/vfio/device.c
>> index 3ffd69a579..b954b44d31 100644
>> --- a/hw/vfio/device.c
>> +++ b/hw/vfio/device.c
>> @@ -544,6 +544,17 @@ bool
>> vfio_device_get_viommu_flags_want_nesting(VFIODevice *vbasedev)
>> return false;
>> }
>>
>> +bool vfio_device_get_viommu_flags_want_pasid_attach(VFIODevice
>> *vbasedev)
>> +{
>> + VFIOPCIDevice *vdev = vfio_pci_from_vfio_device(vbasedev);
>> +
>> + if (vdev) {
>> + return !!(pci_device_get_viommu_flags(PCI_DEVICE(vdev)) &
>> + VIOMMU_FLAG_WANT_PASID_ATTACH);
>> + }
>> + return false;
>> +}
>> +
>> bool vfio_device_get_host_iommu_quirk_bypass_ro(VFIODevice *vbasedev,
>> uint32_t type, void *caps,
>> uint32_t size)
>> diff --git a/hw/vfio/iommufd.c b/hw/vfio/iommufd.c
>> index 78e7b6a045..0718f029ca 100644
>> --- a/hw/vfio/iommufd.c
>> +++ b/hw/vfio/iommufd.c
>> @@ -364,6 +364,7 @@ static bool
>> iommufd_cdev_autodomains_get(VFIODevice *vbasedev,
>> VendorCaps caps;
>> VFIOIOASHwpt *hwpt;
>> uint32_t hwpt_id;
>> + uint8_t max_pasid_log2 = 0;
>> int ret;
>>
>> /* Try to find a domain */
>> @@ -409,7 +410,7 @@ static bool
>> iommufd_cdev_autodomains_get(VFIODevice *vbasedev,
>> */
>> if (!iommufd_backend_get_device_info(vbasedev->iommufd, vbasedev-
>> >devid,
>> &type, &caps, sizeof(caps),
>> &hw_caps,
>> - NULL, errp)) {
>> + &max_pasid_log2, errp)) {
>> return false;
>> }
>>
>> @@ -437,6 +438,11 @@ static bool
>> iommufd_cdev_autodomains_get(VFIODevice *vbasedev,
>> }
>> }
>>
>> + if (max_pasid_log2 &&
>> + vfio_device_get_viommu_flags_want_pasid_attach(vbasedev)) {
>> + flags |= IOMMU_HWPT_ALLOC_PASID;
>> + }
>
>Here, suppose a container already has an hwpt without
>IOMMU_HWPT_ALLOC_PASID
>(from a no PASID device), and a PASID capable device then tries to attach to it
>via the FOREACH loop above, does the kernel return -EINVAL:
>
>ret = iommufd_cdev_pasid_attach_ioas_hwpt(vbasedev, IOMMU_NO_PASID,
> hwpt->hwpt_id, errp);
>
>If not, it will get attached to NO_PASID HWPT and return there, isn't it?
We had below commit:
commit 42fa8991b5e3501f07f43b8f40cd8eab306164a8
Author: Zhenzhong Duan <[email protected]>
Date: Tue Jan 6 01:12:52 2026 -0500
intel_iommu_accel: Fail passthrough device under PCI bridge if x-flts=on
which ensures each passthrough device is attached to separate AddressSpace
in nesting cases, so it never shares a container with other passthrough devices.
Thanks
Zhenzhong
