ARM: 7457/1: smp: Fix suspicious RCU originating from cpu_die()
authorStephen Boyd <sboyd@codeaurora.org>
Fri, 6 Jul 2012 21:03:42 +0000 (22:03 +0100)
committerRussell King <rmk+kernel@arm.linux.org.uk>
Mon, 9 Jul 2012 16:35:32 +0000 (17:35 +0100)
While running hotplug tests I ran into this RCU splat

===============================
[ INFO: suspicious RCU usage. ]
3.4.0 #3275 Tainted: G        W
-------------------------------
include/linux/rcupdate.h:729 rcu_read_lock() used illegally while idle!

other info that might help us debug this:

RCU used illegally from idle CPU!
rcu_scheduler_active = 1, debug_locks = 0
RCU used illegally from extended quiescent state!
4 locks held by swapper/2/0:
 #0:  ((cpu_died).wait.lock){......}, at: [<c00ab128>] complete+0x1c/0x5c
 #1:  (&p->pi_lock){-.-.-.}, at: [<c00b275c>] try_to_wake_up+0x2c/0x388
 #2:  (&rq->lock){-.-.-.}, at: [<c00b2860>] try_to_wake_up+0x130/0x388
 #3:  (rcu_read_lock){.+.+..}, at: [<c00abe5c>] cpuacct_charge+0x28/0x1f4

stack backtrace:
[<c001521c>] (unwind_backtrace+0x0/0x12c) from [<c00abec8>] (cpuacct_charge+0x94/0x1f4)
[<c00abec8>] (cpuacct_charge+0x94/0x1f4) from [<c00b395c>] (update_curr+0x24c/0x2c8)
[<c00b395c>] (update_curr+0x24c/0x2c8) from [<c00b59c4>] (enqueue_task_fair+0x50/0x194)
[<c00b59c4>] (enqueue_task_fair+0x50/0x194) from [<c00afea4>] (enqueue_task+0x30/0x34)
[<c00afea4>] (enqueue_task+0x30/0x34) from [<c00b0908>] (ttwu_activate+0x14/0x38)
[<c00b0908>] (ttwu_activate+0x14/0x38) from [<c00b28a8>] (try_to_wake_up+0x178/0x388)
[<c00b28a8>] (try_to_wake_up+0x178/0x388) from [<c00a82a0>] (__wake_up_common+0x34/0x78)
[<c00a82a0>] (__wake_up_common+0x34/0x78) from [<c00ab154>] (complete+0x48/0x5c)
[<c00ab154>] (complete+0x48/0x5c) from [<c07db7cc>] (cpu_die+0x2c/0x58)
[<c07db7cc>] (cpu_die+0x2c/0x58) from [<c000f954>] (cpu_idle+0x64/0xfc)
[<c000f954>] (cpu_idle+0x64/0xfc) from [<80208160>] (0x80208160)

When a cpu is marked offline during its idle thread it calls
cpu_die() during an RCU idle period. cpu_die() calls complete()
to notify the killing process that the cpu has died. complete()
calls into the scheduler code and eventually grabs an RCU read
lock in cpuacct_charge().

Mark complete() as RCU_NONIDLE so that RCU pays attention to this
CPU for the duration of the complete() function even though it's
in idle.

Suggested-by: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
arch/arm/kernel/smp.c

index 2c7217d..aea74f5 100644 (file)
@@ -179,7 +179,7 @@ void __ref cpu_die(void)
        mb();
 
        /* Tell __cpu_die() that this CPU is now safe to dispose of */
-       complete(&cpu_died);
+       RCU_NONIDLE(complete(&cpu_died));
 
        /*
         * actual CPU shutdown procedure is at least platform (if not