Re: [RFC PATCH v4 01/13] iommu: Introduce dirty log tracking framework

[Keqian Zhu]

Hi Baolu,

On 2021/5/11 11:12, Lu Baolu wrote:
> Hi Keqian,
> 
> On 5/10/21 7:07 PM, Keqian Zhu wrote:
>>>>> I suppose this interface is to ask the vendor IOMMU driver to check
>>>>> whether each device/iommu in the domain supports dirty bit tracking.
>>>>> But what will happen if new devices with different tracking capability
>>>>> are added afterward?
>>>> Yep, this is considered in the vfio part. We will query again after attaching or
>>>> detaching devices from the domain.  When the domain becomes capable, we enable
>>>> dirty log for it. When it becomes not capable, we disable dirty log for it.
>>> If that's the case, why not putting this logic in the iommu subsystem so
>>> that it doesn't need to be duplicate in different upper layers?
>>>
>>> For example, add something like dirty_page_trackable in the struct of
>>> iommu_domain and ask the vendor iommu driver to update it once any
>>> device is added/removed to/from the domain. It's also better to disallow
>> If we do it, the upper layer still needs to query the capability from domain and switch
>> dirty log tracking for it. Or do you mean the domain can switch dirty log tracking automatically
>> when its capability change? If so, I think we're lack of some flexibility. The upper layer
>> may have it's own policy, such as only enable dirty log tracking when all domains are capable,
>> and disable dirty log tracking when just one domain is not capable.
> 
> I may not get you.
> 
> Assume that dirty_page_trackable is an attribution of an iommu_domain.
> This attribution might be changed once a new device (with different
> capability) added or removed. So it should be updated every time a new
> device is attached or detached. This work could be done by the vendor
> iommu driver on the path of dev_attach/dev_detach callback.
Yes, this is what I understand you.

> 
> For upper layers, before starting page tracking, they check the
> dirty_page_trackable attribution of the domain and start it only it's
> capable. Once the page tracking is switched on the vendor iommu driver
> (or iommu core) should block further device attach/detach operations
> until page tracking is stopped.
But when a domain becomes capable after detaching a device, the upper layer
still needs to query it and enable dirty log for it...

To make things coordinated, maybe the upper layer can register a notifier,
when the domain's capability change, the upper layer do not need to query, instead
they just need to realize a callback, and do their specific policy in the callback.
What do you think?

> 
>>
>>> any domain attach/detach once the dirty page tracking is on.
>> Yep, this can greatly simplify our code logic, but I don't know whether our maintainers
>> agree that, as they may think that IOMMU dirty logging should not change original domain
>> behaviors.
> 
> The maintainer owns the last word, but we need to work out a generic and
> self-contained API set.
OK, I see.

Thanks,
Keqian