Re: [LSF/MM TOPIC] Un-addressable device memory and block/fs implications

On Wed, Dec 14, 2016 at 5:15 AM, Jerome Glisse [off-list ref] wrote:

I would like to discuss un-addressable device memory in the context of
filesystem and block device. Specificaly how to handle write-back, read,
... when a filesystem page is migrated to device memory that CPU can not
access.

I intend to post a patchset leveraging the same idea as the existing
block bounce helper (block/bounce.c) to handle this. I believe this is
worth discussing during summit see how people feels about such plan and
if they have better ideas.

Yes, that would be interesting. I presume all of this is for
ZONE_DEVICE and HMM.
I think designing such an interface requires careful thought on tracking pages
to ensure we don't lose writes and also the impact on things like the
writeback subsytem.

From a HMM perspective and an overall MM perspective, I worry that our

accounting
system is broken with the proposed mirroring and unaddressable memory that
needs to be addressed as well.

It would also be nice to have a discussion on migration patches currently on the
list

1. THP migration
2. HMM migration
3. Async migration

I also like to join discussions on:
  - Peer-to-Peer DMAs between PCIe devices
  - CDM coherent device memory

Yes, this needs discussion. Specifically from is all of CDM memory NORMAL or not
and the special requirements we have today for CDM.

  - PMEM
  - overall mm discussions

Balbir Singh.