Re: [RFC] iommu: Fix virtio-iommu probing

[Eric Auger <eric.auger@redhat.com>]

On 8/14/25 5:28 PM, Robin Murphy wrote:
> On 14/08/2025 4:19 pm, Eric Auger wrote:
>> Hi Robin
>>
>> On 8/14/25 4:53 PM, Robin Murphy wrote:
>>> On 14/08/2025 3:17 pm, Eric Auger wrote:
>>>> Commit bcb81ac6ae3c ("iommu: Get DT/ACPI parsing into the proper
>>>> probe path") broke virtio-iommu probing and no iommu group are
>>>> produced anymore.
>>>>
>>>> When probe_iommu_group() gets called viommu_probe_device() fails
>>>> because viommu_get_by_fwnode(fwspec->iommu_fwnode) returns NULL.
>>>
>>> ...which it's not supposed to. And *now* I remember, we never got this
>>> finished, did we?
>> Seems we did not ;-)
>>>
>>> https://lore.kernel.org/linux-iommu/9beaed48da83a0882dba153e65e6cfd0a8e21482.1742484773.git.robin.murphy@arm.com/
>>>
>>>
>>
>> Unfortunately it does not fix my issue. Still no iommu group when
>> booting with ACPI.
>
> Indeed the evidence at the time suggested the patch isn't quite right
> as-is, but that is definitely the place which needs fixing. Since
> Jean-Philippe's occupied with more exciting things at the moment, do
> you happen to have an easy recipe for testing virtio-iommu so I can
> try debugging it myself?
No easy recipe. on my end I am using qemu/kvm with a rhel guest and
custom kernel. I guess it was integrated in kvmtool, wasn't it? I am
just testing with protected virtio devices.

Thanks

Eric
>
> Cheers,
> Robin.
>
>>
>> Thanks
>>
>> Eric
>>>
>>> Thanks,
>>> Robin.
>>>
>>>> So it seems we need to restore the original iommu_probe_device
>>>> call site in acpi_iommu_configure_id() to get a chance to probe
>>>> the device again.
>>>>
>>>> Maybe this defeats the whole purpose of the original commit but
>>>> at least it fixes the virtio-iommu probing.
>>>>
>>>> Fixes: bcb81ac6ae3c ("iommu: Get DT/ACPI parsing into the proper
>>>> probe path")
>>>> Cc: stable@vger.kernel.org # v6.15+
>>>> Signed-off-by: Eric Auger <eric.auger@redhat.com>
>>>>
>>>> ---
>>>>
>>>> I also tested smmu probing and this seems to work fine.
>>>> ---
>>>>    drivers/acpi/scan.c | 7 +++++++
>>>>    1 file changed, 7 insertions(+)
>>>>
>>>> diff --git a/drivers/acpi/scan.c b/drivers/acpi/scan.c
>>>> index fb1fe9f3b1a3..9f4efa8f75a6 100644
>>>> --- a/drivers/acpi/scan.c
>>>> +++ b/drivers/acpi/scan.c
>>>> @@ -1632,6 +1632,13 @@ static int acpi_iommu_configure_id(struct
>>>> device *dev, const u32 *id_in)
>>>>            err = viot_iommu_configure(dev);
>>>>        mutex_unlock(&iommu_probe_device_lock);
>>>>    +    /*
>>>> +     * If we have reason to believe the IOMMU driver missed the
>>>> initial
>>>> +     * iommu_probe_device() call for dev, replay it to get things in
>>>> order.
>>>> +     */
>>>> +    if (!err && dev->bus)
>>>> +        err = iommu_probe_device(dev);
>>>> +
>>>>        return err;
>>>>    }
>>>>    
>>>
>>
>