Fence wait in mmu_interval_notifier_ops::invalidate

[Thomas Hellström (Intel)]

On 12/9/20 5:37 PM, Jason Gunthorpe wrote:
> On Wed, Dec 09, 2020 at 05:36:16PM +0100, Thomas Hellström (Intel) wrote:
>> Jason, Christian
>>
>> In most implementations of the callback mentioned in the subject there's a
>> fence wait.
>> What exactly is it needed for?
> Invalidate must stop DMA before returning, so presumably drivers using
> a dma fence are relying on a dma fence mechanism to stop DMA.

Yes, so far I follow, but what's the reason drivers need to stop DMA?

Is it for invlidation before breaking COW after fork or something related?

Thanks,

Thomas

>
> Jason