Re: [patch] Real-Time Preemption, -RT-2.6.12-rc1-V0.7.41-07

["Paul E. McKenney" <paulmck@us.ibm.com>]

On Wed, Mar 23, 2005 at 10:46:45PM +0100, Ingo Molnar wrote:
> 
> * Ingo Molnar <mingo@elte.hu> wrote:
> 
> > i think the 'migrate read-count' method is not adequate either, 
> > because all callbacks queued within an RCU read section must be called 
> > after the lock has been dropped - while with the migration method 
> > CPU#1 would be free to process callbacks queued in the RCU read 
> > section still active on CPU#2.
> > 
> > i'm wondering how much of a problem this is though. Can there be stale 
> > pointers at that point? Yes in theory, because code like:
> > 
> > 	rcu_read_lock();
> > 	call_rcu(&dentry->d_rcu, d_callback);
> > 	func(dentry->whatever);
> > 	rcu_read_unlock();
> 
> but, this cannot happen, because call_rcu() is used by RCU-write code.

The code would not look exactly like this, but acquiring the update-side
lock inside an RCU read-side critical section can be thought of as
a reader-to-writer lock upgrade.  RCU can do this unconditionally,
which was one of the walls I was banging my head against when trying
to think up a realtime-safe RCU implementation.

So something like the following would be legitimate RCU code:

	rcu_read_lock();
	spin_lock(&dcache_lock);
	call_rcu(&dentry->d_rcu, d_callback);
	spin_unlock(&dcache_lock);
	rcu_read_unlock();

The spin_lock() call upgrades from a read-side to a write-side critical
section.

> so the important property seems to be that any active RCU-read section 
> should keep at least one CPU's active_readers count elevated 
> permanently, for the duration of the RCU-read section.

Yep!

>                                                         It doesnt matter 
> that the reader migrates between CPUs - because the RCU code itself 
> guarantees that no callbacks will be executed until _all_ CPUs have been 
> in quiescent state. I.e. all CPUs have to go through a zero 
> active_readers value before the callback is done.

Yep again!

						Thanx, Paul