Re: Real-Time Preemption and RCU

[Ingo Molnar <mingo@elte.hu>]

* Ingo Molnar <mingo@elte.hu> wrote:

> 
> * Paul E. McKenney <paulmck@us.ibm.com> wrote:
> 
> > I have tested this approach, but in user-level scaffolding.  All of
> > these implementations should therefore be regarded with great
> > suspicion: untested, probably don't even compile.  Besides which, I
> > certainly can't claim to fully understand the real-time preempt patch,
> > so I am bound to have gotten something wrong somewhere. [...]
> 
> you dont even have to consider the -RT patchset: if the scheme allows
> forced preemption of read-side RCU sections on current upstream
> CONFIG_PREEMPT, then it's perfect for PREEMPT_RT too.

there's one detail on PREEMPT_RT though (which i think you noticed too). 

Priority inheritance handling can be done in a pretty straightforward
way as long as no true read-side nesting is allowed for rwsems and
rwlocks - i.e. there's only one owner of a lock at a time. So PREEMPT_RT
restricts rwsem and rwlock concurrency: readers are writers, with the
only exception that they are allowed to 'self-nest'. I.e. things like:

	read_lock(&rwlock);
	...
		read_lock(&rwlock);

are still legal. (it's also done quite often.)

(it is virtually impossible to implement priority inheritance for true
multi-reader locks in any sane way: i've done it initially and it sucks
very much. It also fundamentally increases the 'lock-dependent'
worst-case latencies - imagine 4 readers having to finish first if a
higher-prio writer comes along. It's insane.)

	Ingo