Hi Eric,
>-----Original Message-----
>From: Eric Auger <eric.au...@redhat.com>
>Sent: Tuesday, May 6, 2025 12:15 AM
>To: Cédric Le Goater <c...@redhat.com>; Duan, Zhenzhong
><zhenzhong.d...@intel.com>; qemu-devel@nongnu.org
>Cc: alex.william...@redhat.com; nicol...@nvidia.com;
>joao.m.mart...@oracle.com; Peng, Chao P <chao.p.p...@intel.com>; Liu, Yi L
><yi.l....@intel.com>
>Subject: Re: [PATCH 1/5] vfio/iommufd: Save host iommu capabilities in
>VFIODevice.caps
>
>Hi Zhenzhong,
>
>On 4/11/25 1:28 PM, Cédric Le Goater wrote:
>> On 4/11/25 12:17, Zhenzhong Duan wrote:
>>> The saved caps copy can be used to check dirty tracking capability.
>>>
>>> The capabilities is gotten through IOMMUFD interface, so define a
>>> new structure HostIOMMUDeviceIOMMUFDCaps which contains vendor
>>> caps raw data in "include/system/iommufd.h".
>>>
>>> This is a prepare work for moving .realize() after .attach_device().
>>>
>>> Suggested-by: Cédric Le Goater <c...@redhat.com>
>>> Suggested-by: Eric Auger <eric.au...@redhat.com>
>>> Suggested-by: Nicolin Chen <nicol...@nvidia.com>
>>> Signed-off-by: Zhenzhong Duan <zhenzhong.d...@intel.com>
>>> ---
>>> include/hw/vfio/vfio-device.h | 1 +
>>> include/system/iommufd.h | 22 ++++++++++++++++++++++
>>> hw/vfio/iommufd.c | 10 +++++++++-
>>> 3 files changed, 32 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/include/hw/vfio/vfio-device.h
>>> b/include/hw/vfio/vfio-device.h
>>> index 66797b4c92..09a7af891a 100644
>>> --- a/include/hw/vfio/vfio-device.h
>>> +++ b/include/hw/vfio/vfio-device.h
>>> @@ -77,6 +77,7 @@ typedef struct VFIODevice {
>>> bool dirty_tracking; /* Protected by BQL */
>>> bool iommu_dirty_tracking;
>>> HostIOMMUDevice *hiod;
>>> + HostIOMMUDeviceIOMMUFDCaps caps;
>>
>> IMO, these capabilities belong to HostIOMMUDevice and not VFIODevice.
>I do agree with Cédric that it looks a wrong place to put this caps. I
>feel this somehow breaks the abstraction layering.
This change was dropped in "[PATCH v2 0/5] vfio: Move realize after attach_dev".
>
>Now "[PATCH v2 0/5] vfio: Move realize after attach_dev" has landed, I
>think it would help if you could respin with a clear functional goal
>such as the one targeted in[PATCH v2 0/5] Check host IOMMU compatilibity
>with vIOMMU
><https://lore.kernel.org/all/20240408084404.1111628-1-
>zhenzhong.d...@intel.com/>
See a rfcv3 candidate at link
https://github.com/yiliu1765/qemu/commits/zhenzhong/iommufd_nesting_rfcv3.wip/
for example implementation.
In this example, I used .get_cap() interface for vIOMMU to get cap.
vIOMMU could also access HostIOMMUDevice::HostIOMMUDeviceCaps
directly as it's passed from VFIO to vIOMMU along with HostIOMMUDevice.
Thanks
Zhenzhong
