Re: [PATCH] kthread: Don't depend on work queues

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Wednesday, 11 April 2007 09:03, Andrew Morton wrote:
> On Tue, 10 Apr 2007 23:44:36 -0600 ebiederm@xmission.com (Eric W. Biederman) wrote:
> 
> > Currently there is a circular reference between work queue initialization
> > and kthread initialization.   This prevents the kernel thread
> > infrastructure from initializing until after work queues have been
> > initialized.
> > 
> > For kernel threads we want something that is as close as possible to the
> > init_task and is not contaminated by user processes.  The later we start
> > our kthreadd that forks the rest of the kernel threads the harder this is
> > to do and the more of a mess we have to clean up because the defaults have
> > changed on us.
> > 
> > So this patch modifies the kthread support to not use work queues but to
> > instead use a simple list of structures, and to have kthreadd start from
> > init_task immediately after our kernel thread that execs /sbin/init.
> > 
> > By being a true child of init_task we only have to change those process
> > settings that we want to have different from init_task, such as our
> > process name, blocking all signals and setting SIGCHLD to SIG_IGN
> > so that all of our children are reaped automatically.
> > 
> > By being a tre child of init_task we also naturally get our ppid set to 0
> > and do not wind up as a child of PID == 1.  Ensuring that kernel threads
> > will not slow down the functioning of the wait family of functions.
> 
> argh.  Your description freely confuddles the terms "kernel thread" and
> "kthread".  Can we not do that?  Henceforth the term "kernel thread" refers
> to something which was started with kernel_thread() and "kthread" refers to
> something which was created by kthread_create(), OK?
> 
> Your patch gets midly tangled up with Oleg's recent
> 
> reduce-reparent_to_init.patch
> make-kernel-threads-invisible-to-sbin-init.patch
> reparent-kernel-threads-to-swapper.patch
> 
> but they seemed fairly unpopular anyway so I'll drop 'em.
> 
> Your wait_event() will contribute to load average, I expect.  We get mail. 
> I converted it to wait_event_interruptible().
> 
> I guess using PF_NOFREEZE rather than try_to_freeze() is OK, but one
> wonders what thinking led to that?

It should be calling try_to_freeze() somewhere anyway.  We may need to freeze
all tasks in some cases.

Greetings,
Rafael