Re: Revised memory-management stuff (was: OOM killer)

[Gregory Maxwell]

On Sat, Mar 31, 2001 at 10:03:28PM -0800, Jonathan Morton wrote:
[snip]
> Issue 3:
> 	The OOM killer was frequently killing the "wrong" process.  I have
> developed an improved badness selector, and devised a possible means of
> specifying "don't touch" PIDs at runtime.  PID 1 is never selected for
> killing.  I am debating whether to allow selection of *any* process
> labelled "init" and running as root for the chop, since one of the "unusual
> but frequently encountered" scenarios is for a second init to be running
> during an install or recovery procedure.  This might make it's way in as an
> optional feature.

IO/memory direct hardware access capability holding processes should be at
the bottom of the list.