This patch eliminates the counter-moving during CPU-offline
notifiers, eliminating potential confusion if counters are
scanned during counter-movement process.
This confusion could result in premature ending of an RCU grace
period. For example, if there are two tasks in RCU read-side
critical sections (so that the sum of the counters is two), and
the counter for the CPU going offline is -2, then moving the
count to another CPU can result in the sum momentarily
appearing to be zero. Since there are no memory barriers in
either case, many more such scenarios are possible.
So just don't move the counts!!!
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: laijs@cn.fujitsu.com
Cc: dipankar@in.ibm.com
Cc: josht@linux.vnet.ibm.com
Cc: akpm@linux-foundation.org
Cc: mathieu.desnoyers@polymtl.ca
Cc: dvhltc@us.ibm.com
Cc: niv@us.ibm.com
Cc: peterz@infradead.org
Cc: rostedt@goodmis.org
Cc: hugh.dickins@tiscali.co.uk
Cc: benh@kernel.crashing.org
LKML-Reference: <
12503552312863-git-send-email->
Signed-off-by: Ingo Molnar <mingo@elte.hu>
/* Check to see if the sum of the "last" counters is zero. */
RCU_TRACE_ME(rcupreempt_trace_try_flip_z1);
- for_each_cpu(cpu, to_cpumask(rcu_cpu_online_map))
+ for_each_possible_cpu(cpu)
sum += RCU_DATA_CPU(cpu)->rcu_flipctr[lastidx];
if (sum != 0) {
RCU_TRACE_ME(rcupreempt_trace_try_flip_ze1);
/* seen -after- acknowledgement. */
}
- RCU_DATA_ME()->rcu_flipctr[0] += RCU_DATA_CPU(cpu)->rcu_flipctr[0];
- RCU_DATA_ME()->rcu_flipctr[1] += RCU_DATA_CPU(cpu)->rcu_flipctr[1];
-
- RCU_DATA_CPU(cpu)->rcu_flipctr[0] = 0;
- RCU_DATA_CPU(cpu)->rcu_flipctr[1] = 0;
-
cpumask_clear_cpu(cpu, to_cpumask(rcu_cpu_online_map));
spin_unlock_irqrestore(&rcu_ctrlblk.fliplock, flags);