Re: [PATCH] move put_task_struct() reaping into a thread [Re: 2.6.18-rt1]

[ebiederm@xmission.com (Eric W. Biederman)]

Ingo Molnar <mingo@elte.hu> writes:

> * Eric W. Biederman <ebiederm@xmission.com> wrote:
>
>> Yes I am.  The motivator would be the RT work but I don't see a reason 
>> why the it couldn't be put in the mainline kernel.  If not at least we 
>> need the big fat comment in the mainline kernel that says 
>> put_task_struct must be safe to call with interrupts disabled.
>> 
>> The way the code is structured now it deviates from the mainline 
>> kernel in more than just changing locking behavior.  Which is what 
>> brought me into this conversation in the first place.  So removing 
>> that point of discord would be good.
>
> well, this is one of those few cases (out of ~50,000 lock uses in the 
> kernel) where such a change was unavoidable: put_task_struct() is used 
> in the scheduler context-switch path. (see sched.c:finish_task_switch())

I had missed that was in a preempt disable path when I skimmed through
the users.

> So that's why i first turned it into a separate, extra delayed-free via 
> the "desched thread", and later on picked up the RCUification from Paul 
> McKenney. The RCUification was the simpler (and hence easier to 
> maintain) change. There is no problem with putting this into the RCU 
> path on PREEMPT_RT, as this is a resource-freeing act. I.e. whatever 
> 'delay' there might be in RCU processing, it does not impact program 
> logic. I agree with you that on !PREEMPT_RT there's no reason to 
> complicate things with an extra layer of indirection.

I'm still wondering if we can move put_task_struct a little lower in
the logic in the places where it is called, so it isn't called under a
lock, or with preemption disabled.  The only downside I see is that it
might convolute the logic into unreadability.

In general I get nervous about calling big functions while holding locks.

Eric