On 14/02/2025 13:05, Cédric Le Goater wrote:
> +Kirti
> +Joao
>
> On 2/13/25 22:45, Alex Williamson wrote:
>>> +
>>> + /*
>>> + * Migration support
>>> + */
>>> + object_class_property_set_description(klass, /* 5.2 */
>>> + "x-pre-copy-dirty-page-tracking",
>>> + "Disable dirty pages tracking
>>> during iterative phase");
>>> + object_class_property_set_description(klass, /* 9.1 */
>>> + "x-device-dirty-page-tracking",
>>> + "Disable device dirty page
>>> tracking and use container-based dirty page tracking");
>> These are really debug as well, right? They just happen to be
>> migration related debug.
>
> I suppose so. I would rather keep them under the migration topic
> and add 'debug' in the comment.
>
> Changes :
>
> commit bb0990d1740f ("vfio: Change default dirty pages tracking behavior
> during migration")
> commit 30b916778517 ("vfio/common: Allow disabling device dirty page
> tracking")
>
> do not explicitly explain why these properties are useful in any way.
>
> Kirti, Joao, could you ?
Sorry for the extreme delay but I was out travelling for the past 3 weeks for
vacation and work.
The property is marked as x- but my intent was both debug/testing and to have
something that allows me to select the dirty tracker between device vs platform
dirty tracker if something is wrong with say VF dirty tracker or if there's some
limitations around it. These days we actually use it a lot as my default to have
IOMMU dirty tracker in use. I wouldn't label it exclusively as '(debug)' just
yet.
The precopy option might be to avoid dirty tracking at all and it's generally
useful when you don't have dirty tracking at all or say snapshots or 'offline
migration'. But I have never seen real use of it, or need for that matter,
except the early days of VFIO migration during testing.
