Re: [PATCH v4 26/41] backends/iommufd: Introduce the iommufd object

[Markus Armbruster <armbru@redhat.com>]

Cédric Le Goater <clg@redhat.com> writes:

> On 11/8/23 11:30, Markus Armbruster wrote:
>> Cédric Le Goater <clg@redhat.com> writes:
>> 
>>> Hello Markus,
>>>
>>> On 11/8/23 06:50, Markus Armbruster wrote:
>>>> Cédric Le Goater <clg@redhat.com> writes:
>>>>
>>>>> On 11/2/23 08:12, Zhenzhong Duan wrote:
>>>>>> From: Eric Auger <eric.auger@redhat.com>
>>>>>> Introduce an iommufd object which allows the interaction
>>>>>> with the host /dev/iommu device.
>>>>>> The /dev/iommu can have been already pre-opened outside of qemu,
>>>>>> in which case the fd can be passed directly along with the
>>>>>> iommufd object:
>>>>>> This allows the iommufd object to be shared accross several
>>>>>> subsystems (VFIO, VDPA, ...). For example, libvirt would open
>>>>>> the /dev/iommu once.
>>>>>> If no fd is passed along with the iommufd object, the /dev/iommu
>>>>>> is opened by the qemu code.
>>>>>> The CONFIG_IOMMUFD option must be set to compile this new object.
>>>>>> Suggested-by: Alex Williamson <alex.williamson@redhat.com>
>>>>>> Signed-off-by: Eric Auger <eric.auger@redhat.com>
>>>>>> Signed-off-by: Yi Liu <yi.l.liu@intel.com>
>>>>>> Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
>>>>>> ---
>>>>>> v4: add CONFIG_IOMMUFD check, document default case
>>>>>>     MAINTAINERS              |   7 ++
>>>>>>     qapi/qom.json            |  22 ++++
>>>>>>     include/sysemu/iommufd.h |  46 +++++++
>>>>>>     backends/iommufd-stub.c  |  59 +++++++++
>>>>>>     backends/iommufd.c       | 257 +++++++++++++++++++++++++++++++++++++++
>>>>>>     backends/Kconfig         |   4 +
>>>>>>     backends/meson.build     |   5 +
>>>>>>     backends/trace-events    |  12 ++
>>>>>>     qemu-options.hx          |  13 ++
>>>>>>     9 files changed, 425 insertions(+)
>>>>>>     create mode 100644 include/sysemu/iommufd.h
>>>>>>     create mode 100644 backends/iommufd-stub.c
>>>>>>     create mode 100644 backends/iommufd.c
>>>>>> diff --git a/MAINTAINERS b/MAINTAINERS
>>>>>> index cd8d6b140f..6f35159255 100644
>>>>>> --- a/MAINTAINERS
>>>>>> +++ b/MAINTAINERS
>>>>>> @@ -2135,6 +2135,13 @@ F: hw/vfio/ap.c
>>>>>>     F: docs/system/s390x/vfio-ap.rst
>>>>>>     L: qemu-s390x@nongnu.org
>>>>>>     +iommufd
>>>>>> +M: Yi Liu <yi.l.liu@intel.com>
>>>>>> +M: Eric Auger <eric.auger@redhat.com>
>>>>>> +S: Supported
>>>>>> +F: backends/iommufd.c
>>>>>> +F: include/sysemu/iommufd.h
>>>>>> +
>>>>>>     vhost
>>>>>>     M: Michael S. Tsirkin <mst@redhat.com>
>>>>>>     S: Supported
>>>>>> diff --git a/qapi/qom.json b/qapi/qom.json
>>>>>> index c53ef978ff..27300add48 100644
>>>>>> --- a/qapi/qom.json
>>>>>> +++ b/qapi/qom.json
>>>>>> @@ -794,6 +794,24 @@
>>>>>>     { 'struct': 'VfioUserServerProperties',
>>>>>>       'data': { 'socket': 'SocketAddress', 'device': 'str' } }
>>>>>> +##
>>>>>> +# @IOMMUFDProperties:
>>>>>> +#
>>>>>> +# Properties for iommufd objects.
>>>>>> +#
>>>>>> +# @fd: file descriptor name previously passed via 'getfd' command,
>>>>>> +#     which represents a pre-opened /dev/iommu.  This allows the
>>>>>> +#     iommufd object to be shared accross several subsystems
>>>>>> +#     (VFIO, VDPA, ...), and the file descriptor to be shared
>>>>>> +#     with other process, e.g. DPDK.  (default: QEMU opens
>>>>>> +#     /dev/iommu by itself)
>>>>>> +#
>>>>>> +# Since: 8.2
>>>>>> +##
>>>>>> +{ 'struct': 'IOMMUFDProperties',
>>>>>> +  'data': { '*fd': 'str' },
>>>>>> +  'if': 'CONFIG_IOMMUFD' }
>>>>>
>>>>>
>>>>> Activating or not IOMMUFD on a platform is a configuration choice
>>>>> and it is not a dependency on an external resource. I would make
>>>>> things simpler and drop all the #ifdef in the documentation files.
>>>>
>>>> What exactly are you proposing?
>>>
>>> I would like to simplify the configuration part of this new IOMMUFD
>>> feature and avoid a ./configure option to enable/disable the feature
>>> since it has no external dependencies and can be compiled on all
>>> platforms.
>>>
>>> However, we know that it only makes sense to have the IOMMUFD backend
>>> on platforms s390x, aarch64, x86_64. So I am proposing as an improvement
>>> to enable IOMMUFD only on these platforms with this addition :
>>>
>>>    imply IOMMUFD
>>>
>>> to hw/{i386,s390x,arm}/Kconfig files.
>>>
>>> This gives us the possibility to compile out the feature downstream
>>> if something goes wrong, using the files under : configs/devices/.
>> 
>> Shouldn't we then compile out the relevant parts of the QAPI schema,
>> too?
>
> Is it possible with Kconfig options ?

See below.

>>> Given that the IOMMUFD feature doesn't have any external dependencies
>>> and that the IOMMUFD backend object is common to all platforms, I am
>>> also proposing to remove all the CONFIG_IOMMUFD define usage in the
>>> documentation file "qemu-options.hx" and the schema file "qapi/qom.json".
>> 
>> Any CONFIG_IOMMUFD left elsewhere?
>
> There are. To expose or not vfio properties. Stuff like :
>
> ifdef CONFIG_IOMMUFD
>      DEFINE_PROP_LINK("iommufd", VFIOPCIDevice, vbasedev.iommufd,
>                       TYPE_IOMMUFD_BACKEND, IOMMUFDBackend *),
> #endif
>      DEFINE_PROP_END_OF_LIST(),
>
> and
>
> #ifdef CONFIG_IOMMUFD
>      object_class_property_add_str(klass, "fd", NULL, vfio_pci_set_fd);
> #endif
>
>
>>>> The use of 'if': 'CONFIG_IOMMUFD' in the QAPI schema enables
>>>> introspection with query-qmp-schema: when ObjectType @iommufd exists,
>>>> QEMU supports creating the object.  Or am I confused?
>>>
>>> Object iommufd should always exist since it is common to all.
>>>
>>> Is that acceptable ?
>> 
>> Perhaps the question to ask is whether a management application needs to
>> know whether this version of QEMU supports iommufd objects.  If yes,
>> then query-qmp-schema is an obvious way to find out.  
>
> Thanks for reminding me of this possibility. In that case, we should
> indeed avoid returning iommufd support when !CONFIG_IOMMUFD.
>
> Can it be implemented in qapi/qom.json with Kconfig options ?

The only tool we have for configuring the schema is the 'if'
conditional.  'if': 'CONFIG_IOMMUFD' compiles to #if
defined(CONFIG_IOMMUFD) ... #endif.  Your use of #ifdef CONFIG_IOMMUFD
above suggests this is fine here.

Symbols that are only defined in target-dependent compiles (see
exec/poison.h) can only be used in target-dependent schema modules,
i.e. the *-target.json.

>> What could go
>> wrong when this returns "supported" when it actually isn't?
>   
> The management layer would build an invalid QEMU command line and
> QEMU would return "invalid object type: iommufd"
>
> Thanks,
>
> C.