Messages in this thread | | | Date | Sat, 26 Sep 2009 23:47:53 -0700 | From | "Paul E. McKenney" <> | Subject | [PATCH tip/core/rcu 0/2] rcu: more review feedback |
| |
This patchset applies the last of the review comments:
o Whitespace fixes, updated comments, and code movement.
o Improve an existing abstraction and introduce two additional abstractions for traversing the rcu_node tree.
Issues remaining:
o I can force hangs given sufficient rcutorture testing in conjunction with random CPU-hotplug operations. I am working on a simpler interface between RCU and CPU hotplug that I believe will clear this up.
o TREE_PREEMPT_RCU currently uses the TREE_RCU variant of synchronize_rcu_expedited, which is wrong. I am working on an expedited variant for TREE_PREEMPT_RCU, but will fall back on synchronize_rcu() for the short term if it gives me much trouble.
o The IPI hyperactivity from call_rcu() noted by Nick Piggin. I believe I have identified a clean design for a fix for this.
o Make RCU less dependent on IPIs for forcing grace periods (but this might be best deferred to 2.6.33).
o RCU priority boosting (2.6.33!).
o TINY_PREEMPT_RCU (2.6.33!).
o Squeeze another 400 bytes or so out of TINY_RCU (2.6.33!).
| |