[PATCH] iommu/arm-smmu-v3: Enable PCI ATS in passthrough mode as well

[Robin Murphy]

On 2023-02-03 10:44, Ganapatrao Kulkarni wrote:
> 
> 
> On 02-02-2023 06:52 pm, Robin Murphy wrote:
>> On 2023-02-02 12:40, Ganapatrao Kulkarni wrote:
>>> The current smmu-v3 driver does not enable PCI ATS for physical 
>>> functions
>>> of ATS capable End Points when booted in smmu bypass mode
>>> (iommu.passthrough=1). This will not allow virtual functions to enable
>>> ATS(even though EP supports it) while they are attached to a VM using
>>> VFIO driver.
>>>
>>> This patch adds changes to enable ATS support for physical functions
>>> in passthrough/bypass mode as well.
>>>
>>> Also, adding check to avoid disabling of ATS if it is not enabled,
>>> to avoid unnecessary call-traces.
>>>
>>> Signed-off-by: Ganapatrao Kulkarni <gankulkarni at os.amperecomputing.com>
>>> ---
>>>   drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c | 10 +++++++---
>>>   1 file changed, 7 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c 
>>> b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
>>> index 6d5df91c5c46..5a605cb5ccef 100644
>>> --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
>>> +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
>>> @@ -2313,11 +2313,16 @@ static void arm_smmu_enable_ats(struct 
>>> arm_smmu_master *master)
>>>   static void arm_smmu_disable_ats(struct arm_smmu_master *master)
>>>   {
>>>       struct arm_smmu_domain *smmu_domain = master->domain;
>>> +    struct pci_dev *pdev;
>>>       if (!master->ats_enabled)
>>>           return;
>>> -    pci_disable_ats(to_pci_dev(master->dev));
>>> +    pdev = to_pci_dev(master->dev);
>>> +
>>> +    if (pdev->ats_enabled)
>>
>> If the master->ats_enabled check above passes when ATS isn't actually 
>> enabled, surely that's a bug?
> 
> IIUC, It means ATS feature is supported (just check for existence of ATS 
> extended capability and smmu capability) and not necessarily enabled.
> Function pci_enable_ats(called by arm_smmu_enable_ats) enables the ATS 
> by setting bit 15 of ATS Control Register (Offset 06h).
> If pci_enable_ats is not successful, it will not set dev->ats_enabled 
> flag. So calling pci_disable_ats later results in call-trace, if 
> dev->ats_enabled is not set.

And like I say, that appears to be a bug, or at least something 
deserving of improvement. If we *know* that enabling ATS failed, leaving 
it hanging in some half-enabled state seems wrong. I know the PCI spec 
says that functions must accept ATS invalidate requests even when ATS is 
disabled, but that still doesn't make it a great idea for the driver to 
spend time and effort sending them when it should know they are 
unnecessarily (especially given the infamous 90-second maximum timeout).

> Function arm_smmu_enable_ats already prints error message if ATS enable 
> is failed.

Printing a message hardly constitutes robust error handling...

Thanks,
Robin.