On 09.02.21 14:25, Michal Hocko wrote:
On Tue 09-02-21 11:23:35, David Hildenbrand wrote:
[...]
I am constantly trying to fight for making more stuff MOVABLE instead of
going into the other direction (e.g., because it's easier to implement,
which feels like the wrong direction).
Maybe I am the only person that really cares about ZONE_MOVABLE these days
:) I can't stop such new stuff from popping up, so at least I want it to be
documented.
MOVABLE zone is certainly an important thing to keep working. And there
is still quite a lot of work on the way. But as I've said this is more
of a outlier than a norm. On the other hand movable zone is kinda hard
requirement for a lot of application and it is to be expected that
many features will be less than 100% compatible. Some usecases even
impossible. That's why I am arguing that we should have a central
document where the movable zone is documented with all the potential
problems we have encountered over time and explicitly state which
features are fully/partially incompatible.
I'll send a mail during the next weeks to gather current restrictions to
document them (and include my brain dump). We might see more excessive
use of ZONE_MOVABLE in the future and as history told us, of CMA as
well. We really should start documenting/caring.
@Mike, it would be sufficient for me if one of your patches at least
mention the situation in the description like
"Please note that secretmem currently behaves much more like long-term
GUP instead of mlocked memory; secretmem is unmovable memory directly
consumed/controlled by user space. secretmem cannot be placed onto
ZONE_MOVABLE/CMA.
As long as there is no excessive use of secretmem (e.g., maximum of 16
MiB for selected processes) in combination with ZONE_MOVABLE/CMA, this
is barely a real issue. However, it is something to keep in mind when a
significant amount of system RAM might be used for secretmem. In the
future, we might support migration of secretmem and make it look much
more like mlocked memory instead."
Just a suggestion.
--
Thanks,
David / dhildenb
