Consider "s390/mm: avoid empty zero pages for KVM guests to avoid postcopy hangs" for Artful

[Christian Ehrhardt ]

Hi,
there is this patch in discussion [1].

If I read correctly between the lines that is supposed to go to -stable trees as well as to linxu-s390x next (4.14) submissions.
But the change is rather big and we have to consider taking that early or as SRU or not at all.

I wanted to ask the kernel Team to keep an eye on that discussion and take the patch into consideration - IMHO it might change KVM behavior on s390 quite a bit.
So I'm not sure how SRUable it is while on Artful we migth add it rather soon before release.

To some extend we should wait for the discussion on the upstreaming, but then I wanted to make you aware asap.

[1]: https://lists.gnu.org/archive/html/qemu-devel/2017-08/msg05286.html

[Christian Ehrhardt  (paelzer)]

Also asking Lou to reverse mirror and let Christian Borntraeger (Author) be added on the mirror of this.

[Ubuntu Kernel Bot (ubuntu-kernel-bot)]

This bug is missing log files that will aid in diagnosing the problem. While running an Ubuntu kernel (not a mainline or third-party kernel) please enter the following command in a terminal window:

apport-collect 1713765

and then change the status of the bug to 'Confirmed'.

If, due to the nature of the issue you have encountered, you are unable to run this command, please add a comment stating that fact and change the bug status to 'Confirmed'.

This change has been made by an automated script, maintained by the Ubuntu Kernel Team.

[Seth Forshee (sforshee)]

Looks like the patch just was only just sent yesterday, will give it a couple days to see if it collects any feedback or acks.

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2017-08-30 04:09 EDT-------
FWIW, this fix is only necessary if you have guest that use storage keys - e.g. an 16.04 guest should not trigger this problem.

[Christian Ehrhardt  (paelzer)]

Continued the discussion on IRC:

[10:19] <hws> cpaelzer: just to shortcut..... FWIW, this fix is only necessary if you have guest that use storage keys - e.g. an 16.04 guest should not trigger this problem.
[10:21] <cpaelzer> hws: do we have any linux guests that use storgae keys in the meantime?
[10:21] <cpaelzer> I might have lost track
[10:21] <hws> cpaelzer: question for christian...
[10:21] <cpaelzer> and as a HV host it is a bit weird, as you can only make "so much" assumption on your guests
[10:22] <cpaelzer> who says no one uses a Ubuntu Xenial as host in 2020 (still supported then) with a new Guest that suddenly uses stroage keys
[10:22] <cpaelzer> ...
[10:22] <cpaelzer> anyway this is just the sort of discussion that I wanted on that bugzilla
[10:22] <cpaelzer> for a more permanent documentation of the case
[10:23] <cpaelzer> and since it is mirrored now this will happen
[10:23] <cpaelzer> I don't mind loosing a few hours for the mirroring
[10:32] <borntraeger> cpaelzer, hws no current Linux guest uses storage keys
[10:32] <borntraeger> sles 11 and rhel6 do
[10:32] <borntraeger> sles12, rhel7 ubuntu 16.04 do not
[10:32] <cpaelzer> well the old one before we removed it
[10:33] <cpaelzer> so since there is no intention to add storage keys back in Linux or host non Linux (not intention I'd know of) - the TL;DR is "not needed"?
[10:40] <borntraeger> cpaelzer, unless you want to support some old debians as guest or so
[10:45] <cpaelzer> borntraeger: did anybody check if there really is no perf impact?
[10:45] <cpaelzer> I know that over time all pages will be used
[10:45] <cpaelzer> but I thought the zero page would still be overloaded
[10:45] <cpaelzer> or is even the fix
[10:45] <cpaelzer> only active on storage key using guests
[10:46] <cpaelzer> borntraeger: and so it is not only "not needed" but even when applied "not actively changing" the zero page handling as long as you have only "new" guests?
[10:48] <borntraeger> cpaelzer, if a guest has a pattern of only reading a page, it will be a zero page inside the guest
[10:49] <borntraeger> cpaelzer, the fix is to not use zero pages in the host to back guest pages

So leaving it to the kernel Team, but I'd think won't fix is correct for now.

[Seth Forshee (sforshee)]

Closing won't fix based on comment #5, please re-open if something changes.

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2017-09-06 03:23 EDT-------
IBM bugzilla status-> closed.. See last comment from Canonical

[Christian Ehrhardt  (paelzer)]

Seth missed to actually close it, doing so.