On Wed, 27 Feb 2013, Arve Hjønnevåg wrote:

> > Explain how this can possibly be consistent if your system includes two
> > processes, one writing oom_adj values and one writing oom_score_adj
> > values, and the kernel has this new option enabled?
> 
> These values are all written by the same process. If you want to
> improve this, the lowmemorykiller will need a separate score_adj
> parameter so that it knows for sure which type of value it receives.
> You changed the meaning of /sys/module/lowmemorykiller/parameters/adj.
> This patch restores the old behavior as long as the values you write
> are within the range oom_adj, while still supporting the new behavior
> you added in the likely case that one of the values are out of range
> for oom_adj.
> 

Then remove the config option entirely and add a new score_adj file that 
gets mapped from writes to adj.  No reason to tie the kernel config to a 
particular userspace implementation (even though it's been like this for 
over a year).  Nack to this patch.