Re: [PATCH] iommu/intel-iommu: set as DUMMY_DEVICE_DOMAIN_INFO if no IOMMU

[Lu Baolu <baolu.lu@linux.intel.com>]

Hi,

On 2020/2/4 17:25, Jian-Hong Pan wrote:
> Lu Baolu <baolu.lu@linux.intel.com> 於 2020年2月4日 週二 下午2:11寫道：
>>
>> Hi,
>>
>> On 2020/2/3 17:10, Jian-Hong Pan wrote:
>>> If the device has no IOMMU, it still invokes iommu_need_mapping during
>>> intel_alloc_coherent. However, iommu_need_mapping can only check the
>>> device is DUMMY_DEVICE_DOMAIN_INFO or not. This patch marks the device
>>> is a DUMMY_DEVICE_DOMAIN_INFO if the device has no IOMMU.
>>>
>>> Signed-off-by: Jian-Hong Pan <jian-hong@endlessm.com>
>>> ---
>>>    drivers/iommu/intel-iommu.c | 4 +++-
>>>    1 file changed, 3 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c
>>> index 35a4a3abedc6..878bc986a015 100644
>>> --- a/drivers/iommu/intel-iommu.c
>>> +++ b/drivers/iommu/intel-iommu.c
>>> @@ -5612,8 +5612,10 @@ static int intel_iommu_add_device(struct device *dev)
>>>        int ret;
>>>
>>>        iommu = device_to_iommu(dev, &bus, &devfn);
>>> -     if (!iommu)
>>> +     if (!iommu) {
>>> +             dev->archdata.iommu = DUMMY_DEVICE_DOMAIN_INFO;
>>
>> Is this a DMA capable device?
> 
> Do you mean is the device in DMA Remapping table?
> Dump DMAR from ACPI table.  The device is not in the table.
> So, it does not support DMAR, Intel IOMMU.
> 
> Or, should device_to_iommu be invoked in iommu_need_mapping to check
> IOMMU feature again?
> 

Normally intel_iommu_add_device() should only be called for PCI devices
and APCI name space devices (reported in ACPI/DMAR table). In both
cases, device_to_iommu() should always return a corresponding iommu. I
just tried to understand why it failed for your case.

Best regards,
baolu