Re: [PATCH v4 09/10] iommu: Make iommu_queue_iopf() more generic

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

On 2023/8/30 15:55, Tian, Kevin wrote:
>> From: Baolu Lu <baolu.lu@linux.intel.com>
>> Sent: Saturday, August 26, 2023 4:04 PM
>>
>> On 8/25/23 4:17 PM, Tian, Kevin wrote:
>>>> +static void assert_no_pending_iopf(struct device *dev, ioasid_t pasid)
>>>> +{
>>>> +	struct iommu_fault_param *iopf_param = dev->iommu-
>>>>> fault_param;
>>>> +	struct iopf_fault *iopf;
>>>> +
>>>> +	if (!iopf_param)
>>>> +		return;
>>>> +
>>>> +	mutex_lock(&iopf_param->lock);
>>>> +	list_for_each_entry(iopf, &iopf_param->partial, list) {
>>>> +		if (WARN_ON(iopf->fault.prm.pasid == pasid))
>>>> +			break;
>>>> +	}
>>> partial list is protected by dev_iommu lock.
>>>
>>
>> Ah, do you mind elaborating a bit more? In my mind, partial list is
>> protected by dev_iommu->fault_param->lock.
>>
> 
> well, it's not how the code is currently written. iommu_queue_iopf()
> doesn't hold dev_iommu->fault_param->lock to update the partial
> list.
> 
> while at it looks there is also a mislocking in iopf_queue_discard_partial()
> which only acquires queue->lock.
> 
> So we have three places touching the partial list all with different locks:
> 
> - iommu_queue_iopf() relies on dev_iommu->lock
> - iopf_queue_discard_partial() relies on queue->lock
> - this new assert function uses dev_iommu->fault_param->lock

Yeah, I see your point now. Thanks for the explanation.

So, my understanding is that dev_iommu->lock protects the whole
pointer of dev_iommu->fault_param, while dev_iommu->fault_param->lock
protects the lists inside it.

Is this locking mechanism different from what you think?

Best regards,
baolu