On Thu, Jan 28, 2021 at 06:12:12PM +0100, Frederic Weisbecker wrote:
> rdp->nocb_cb_sleep is first set to true by default after processing
> the callbacks then set back to false if we still find ready callbacks
> to invoke.
>
> This is confusing and even unsafe if it ever happens to be read
> locklessly at some point. So make sure we write it only once per
> nocb_cb_wait() loop.
>
> Reported-by: Paul E. McKenney <[email protected]>
> Cc: Josh Triplett <[email protected]>
> Cc: Lai Jiangshan <[email protected]>
> Cc: Joel Fernandes <[email protected]>
> Cc: Neeraj Upadhyay <[email protected]>
> Cc: Boqun Feng <[email protected]>
> Signed-off-by: Frederic Weisbecker <[email protected]>
Nice, queued, thank you! The usual wordsmithing &c...
Thanx, Paul
------------------------------------------------------------------------
commit cbc3fbfe8424edc90668d5878eb493ae2ff1b888
Author: Frederic Weisbecker <[email protected]>
Date: Thu Jan 28 18:12:12 2021 +0100
rcu/nocb: Avoid confusing double write of rdp->nocb_cb_sleep
The nocb_cb_wait() function first sets the rdp->nocb_cb_sleep flag to
true by after invoking the callbacks, and then sets it back to false if
it finds more callbacks that are ready to invoke.
This is confusing and will become unsafe if this flag is ever read
locklessly. This commit therefore writes it only once, based on the
state after both callback invocation and checking.
Reported-by: Paul E. McKenney <[email protected]>
Cc: Josh Triplett <[email protected]>
Cc: Lai Jiangshan <[email protected]>
Cc: Joel Fernandes <[email protected]>
Cc: Neeraj Upadhyay <[email protected]>
Cc: Boqun Feng <[email protected]>
Signed-off-by: Frederic Weisbecker <[email protected]>
Signed-off-by: Paul E. McKenney <[email protected]>
diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
index c61613a..a3db700 100644
--- a/kernel/rcu/tree_plugin.h
+++ b/kernel/rcu/tree_plugin.h
@@ -2229,6 +2229,7 @@ static void nocb_cb_wait(struct rcu_data *rdp)
unsigned long flags;
bool needwake_state = false;
bool needwake_gp = false;
+ bool can_sleep = true;
struct rcu_node *rnp = rdp->mynode;
local_irq_save(flags);
@@ -2252,8 +2253,6 @@ static void nocb_cb_wait(struct rcu_data *rdp)
raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */
}
- WRITE_ONCE(rdp->nocb_cb_sleep, true);
-
if (rcu_segcblist_test_flags(cblist, SEGCBLIST_OFFLOADED)) {
if (!rcu_segcblist_test_flags(cblist, SEGCBLIST_KTHREAD_CB)) {
rcu_segcblist_set_flags(cblist, SEGCBLIST_KTHREAD_CB);
@@ -2261,7 +2260,7 @@ static void nocb_cb_wait(struct rcu_data *rdp)
needwake_state = true;
}
if (rcu_segcblist_ready_cbs(cblist))
- WRITE_ONCE(rdp->nocb_cb_sleep, false);
+ can_sleep = false;
} else {
/*
* De-offloading. Clear our flag and notify the de-offload
worker.
@@ -2274,6 +2273,8 @@ static void nocb_cb_wait(struct rcu_data *rdp)
needwake_state = true;
}
+ WRITE_ONCE(rdp->nocb_cb_sleep, can_sleep);
+
if (rdp->nocb_cb_sleep)
trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("CBSleep"));
