Messages in this thread | | | Date | Mon, 8 Nov 2010 16:15:38 +0000 (UTC) | From | houston.jim@comcast ... | Subject | Re: [PATCH] a local-timer-free version of RCU |
| |
Hi Everyone,
I'm sorry started this thread and have not been able to keep up with the discussion. I agree that the problems described are real.
> > UAS> PEM> o CPU 1 continues in rcu_grace_period_complete(), > > UAS> PEM> incorrectly ending the new grace period. > > UAS> PEM> > > UAS> PEM> Or am I missing something here? > > UAS> > > UAS> The scenario you describe seems possible. However, it should be easily > > UAS> fixed by passing the perceived batch number as another parameter to > > UAS> rcu_set_state() and making it part of the cmpxchg. So if the caller > > UAS> tries to set state bits on a stale batch number (e.g., batch != > > UAS> rcu_batch), it can be detected.
My thought on how to fix this case is to only hand off the DO_RCU_COMPLETION to a single cpu. The rcu_unlock which receives this hand off would clear its own bit and then call rcu_poll_other_cpus to complete the process.
> What is scary with this is that it also changes rcu sched semantics, and users > of call_rcu_sched() and synchronize_sched(), who rely on that to do more > tricky things than just waiting for rcu_derefence_sched() pointer grace periods, > like really wanting for preempt_disable and local_irq_save/disable, those > users will be screwed... :-( ...unless we also add relevant rcu_read_lock_sched() > for them...
I need to stare at the code and get back up to speed. I expect that the synchronize_sched path in my patch is just plain broken.
Jim Houston
| |