Re: WTH is going on with memory hotplug sysf interface (was: Re: [RFC PATCH] mm, hotplug: get rid of auto_online_blocks)

From: Michal Hocko
Date: Mon Mar 13 2017 - 05:22:48 EST

On Fri 10-03-17 13:00:37, Reza Arbab wrote:
> On Fri, Mar 10, 2017 at 04:53:33PM +0100, Michal Hocko wrote:
> >OK, so while I was playing with this setup some more I probably got why
> >this is done this way. All new memblocks are added to the zone Normal
> >where they are accounted as spanned but not present.
> It's not always zone Normal. See zone_for_memory(). This leads to a
> workaround for having to do online_movable in descending block order.
> Instead of this:
> 1. probe block 34, probe block 33, probe block 32, ...
> 2. online_movable 34, online_movable 33, online_movable 32, ...
> you can online_movable the first block before adding the rest:

I do I enforce that behavior when the probe happens automagically?

> 1. probe block 32, online_movable 32
> 2. probe block 33, probe block 34, ...
> - zone_for_memory() will cause these to start Movable
> 3. online 33, online 34, ...
> - they're already in Movable, so online_movable is equivalentr
> I agree with your general sentiment that this stuff is very nonintuitive.

My criterion for nonintuitive is probably different because I would call
this _completely_unusable_. Sorry for being so loud about this but the
more I look into this area the more WTF code I see. This has seen close
to zero review and seems to be building up more single usecase code on
top of previous. We need to change this, seriously!
Michal Hocko