lkml.org 
[lkml]   [2010]   [Dec]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
Subject[PATCH tip/core/rcu 0/20] second preview of RCU patches for 2.6.38
Hello!

This patchset contains RCU priority boosting for the tiny RCU
implementations and a number of fixes and cleanups. The patches
are as follows, with the first 12 being updates from the earlier
posting at https://lkml.org/lkml/2010/11/6/187:

1. Add rcutorture tests to check for correct priority boosting.
2-6. RCU priority boosting for the tiny RCU implementations.
RCU priority boosting for the tree RCU implementations is
in progress, but not yet ready for prime time.
7. Fix a few naming holdouts from old Classic RCU.
8. Move synchronize_sched_expedited() out of sched.c
(from Lai Jiangshan).
9. Simplify the RCU callback orphan/adopt code for CPU hotplug
(from Lai Jiangshan).
10. Update RCU tracing documentation to reflect patch #9.
11. Fix a race condition in synchronize_sched_expedited(),
but Tejun Heo has suggested another approach that might
be better longer term.
12. Make synchronize_srcu_expedited() spin for a bit to avoid
blocking -- again, there may be a better long-term fix.
That said, this adaptive approach seems to work very well
in practice.
13. Improve synchronize_sched_expedited()'s ability to batch
concurrent requests, as foreshadowed above.
14-15. Make RCU avoid looking for quiescent states in cases when
they are not required, thereby improving RCU's energy
efficiency (from Frederic Weisbecker).
16. Enlist the rnp->qsmask data to simplify #14 and #15 and
also handle another case where RCU was unnecessarily asking
CPUs to go through quiescent states.
17. Limit tree RCU's leaf-level fanout to reduce lock contention.
Lock contention is higher at the leaf level that elsewhere
in the tree of rcu_node structures due to the fact that
CPUs synchronize at the leaf level to detect grace-period
events.
18. Reduce tree RCU's leaf-level lock contention still further
by making __call_rcu() less aggressive about starting new
grace periods. The aggression is a holdover from ancient
times predating synchronize_rcu_expedited().
19. Fix a mismatched-parentheses error in __list_for_each_rcu()
(from Mariusz Kozlowski).
20. Given that there are no in-tree users, remove the aforementioned
__list_for_each_rcu(). This is maintained as a separate
commit in case there is a use out there somewhere making its
way into mainline. (Though I would rather keep this API
out -- list_for_each_entry_rcu() is much better in most cases.)

For a testing-only version of this patchset from git, please see:

git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-2.6-rcu.git rcu/testing

Thanx, Paul

Documentation/RCU/trace.txt | 12 -
b/Documentation/RCU/trace.txt | 132 +++++++++++-
b/include/linux/init_task.h | 9
b/include/linux/rculist.h | 2
b/include/linux/rcupdate.h | 1
b/include/linux/rcutiny.h | 8
b/include/linux/rcutree.h | 1
b/include/linux/sched.h | 11 -
b/init/Kconfig | 39 +++
b/kernel/rcutiny.c | 71 +++++-
b/kernel/rcutiny_plugin.h | 15 -
b/kernel/rcutorture.c | 270 ++++++++++++++++++++++++-
b/kernel/rcutree.c | 81 +------
b/kernel/rcutree.h | 16 -
b/kernel/rcutree_plugin.h | 71 ++++++
b/kernel/rcutree_trace.c | 8
b/kernel/sched.c | 69 ------
b/kernel/srcu.c | 8
include/linux/rculist.h | 5
include/linux/rcupdate.h | 3
include/linux/rcutiny.h | 5
include/linux/rcutree.h | 1
init/Kconfig | 16 +
kernel/rcutiny.c | 70 ++----
kernel/rcutiny_plugin.h | 444 ++++++++++++++++++++++++++++++++++++++++--
kernel/rcutree.c | 95 ++++++--
kernel/rcutree.h | 45 ++--
kernel/rcutree_plugin.h | 110 +++++++---
kernel/rcutree_trace.c | 4
29 files changed, 1272 insertions(+), 350 deletions(-)

\
 
 \ /
  Last update: 2010-12-17 21:57    [from the cache]
©2003-2011 Jasper Spaans