rcu: Check for callback invocation from offline CPUs

Because quiescent states are now reported from offline CPUs in
CPU_DYING state, there is some possibility that such a CPU might
note the end of a grace period and attempt to start invoking
callbacks.  This would be a very bad thing, and is supposed to
be prevented by the fact that the CPU_DYING CPU gets rid of all
its callbacks before reporting the quiescent state.  However,
there is other CPU-offline code in the kernel, and it is quite
possible that someone will invoke RCU core processing from that
code.  Therefore, this commit adds a warning for this case.

Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

diff --git a/kernel/rcutree.c b/kernel/rcutree.c
index 575f91d03f06e36af9c5ecc5b438159f21d0ea0f..ac3a810d2db7bf75521d8b6312b5870846818c5f 100644 (file)
--- a/kernel/rcutree.c
+++ b/kernel/rcutree.c
@@ -1373,6 +1373,7 @@ static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
         * races with call_rcu() from interrupt handlers.
         */
        local_irq_save(flags);
+       WARN_ON_ONCE(cpu_is_offline(smp_processor_id()));
        bl = rdp->blimit;
        trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, bl);
        list = rdp->nxtlist;