[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH v2 00/11] vfio/migration: Implement VFIO migration protocol v
|
From: |
Jason Gunthorpe |
|
Subject: |
Re: [PATCH v2 00/11] vfio/migration: Implement VFIO migration protocol v2 |
|
Date: |
Thu, 23 Jun 2022 11:56:47 -0300 |
On Fri, Jun 17, 2022 at 03:51:29PM -0600, Alex Williamson wrote:
> It's ok by me if QEMU vfio is the one that marks all mapped pages dirty
> if the host interface provides no way to do so. Would we toggle that
> based on whether the device has bus-master enabled?
I don't think so, that is a very niche optimization, it would only
happen if a device is plugged in but never used.
If a device truely doesn't have bus master capability at all then it's
VFIO migration driver should implement report dirties and report no
dirties.
> Regarding SPAPR, I'd tend to think that if we're dirtying in QEMU then
> nothing prevents us from implementing the same there, but also I'm not
> going to stand in the way of simply disabling migration for that IOMMU
> backend unless someone speaks up that they think it deserves parity.
If the VFIO device internal tracker is being used it should work with
SPAPR too.
The full algorithm should be to try to find a dirty tracker for each
VFIO migration device and if none is found then always dirty
everything at STOP_COPY.
iommufd will provide the only global dirty tracker, so if SPAPR or
legacy VFIO type1 is used without a device internal tracker then it
should do the all-dirties.
Jason