PREEMPT_RT has made rcu_core() preemptible, making it unsafe against
concurrent NOCB (de-)offloading.
Thomas suggested to drop the local_lock() based solution and simply
check the offloaded state while context looks safe but that's not
enough. Here is a bit of rework.
git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks.git
rcu/rt
HEAD: aac1c58961446c731f2e989bd822ca1fd2659bad
Thanks,
Frederic
---
Frederic Weisbecker (10):
rcu/nocb: Make local rcu_nocb_lock_irqsave() safe against concurrent deoffloading
rcu/nocb: Prepare state machine for a new step
rcu/nocb: Invoke rcu_core() at the start of deoffloading
rcu/nocb: Make rcu_core() callbacks acceleration (de-)offloading safe
rcu/nocb: Check a stable offloaded state to manipulate qlen_last_fqs_check
rcu/nocb: Use appropriate rcu_nocb_lock_irqsave()
rcu/nocb: Limit number of softirq callbacks only on softirq
rcu: Fix callbacks processing time limit retaining cond_resched()
rcu: Apply callbacks processing time limit only on softirq
rcu/nocb: Don't invoke local rcu core on callback overload from nocb kthread
Thomas Gleixner (1):
rcu/nocb: Make rcu_core() callbacks acceleration preempt-safe
include/linux/rcu_segcblist.h | 51 +++++++++++++++++++-------
kernel/rcu/rcu_segcblist.c | 10 ++---
kernel/rcu/rcu_segcblist.h | 7 ++--
kernel/rcu/tree.c | 85 ++++++++++++++++++++++++++++++-------------
kernel/rcu/tree.h | 16 +++++---
kernel/rcu/tree_nocb.h | 29 ++++++++++++---
6 files changed, 136 insertions(+), 62 deletions(-)
