[PATCH 1/2] mm: handle multiple owners of device private pages in migrate_vma

[Christoph Hellwig]

On Mon, Mar 16, 2020 at 03:17:07PM -0300, Jason Gunthorpe wrote:
> On Mon, Mar 16, 2020 at 06:52:58PM +0100, Christoph Hellwig wrote:
> > Add a new opaque owner field to struct dev_pagemap, which will allow
> > the hmm and migrate_vma code to identify who owns ZONE_DEVICE memory,
> > and refuse to work on mappings not owned by the calling entity.
> 
> Using a pointer seems like a good solution to me.
> 
> Is this a bug fix? What is the downside for migrate on pages it
> doesn't work? I'm not up to speed on migrate..

migrating private pages not owned by driver simply won't work, as
the device can't access it.  Even inside the same driver say
GPU A can't just migrate CPU B's memory.  In that sense it is
a bug fix for the rather unlikely case of using the experimental
nouveau with multiple GPUs, or in a power secure VM (if that is
even possible).