Re: 2.6.35-rc2-git1 - include/linux/cgroup.h:534 invoked rcu_dereference_check() without protection!
From: Peter Zijlstra <peterz@infradead.org>
Date: 2010-06-09 17:57:37
Also in:
lkml
Subsystem:
the rest · Maintainer:
Linus Torvalds
On Wed, 2010-06-09 at 13:00 -0400, Miles Lane wrote:
ACPI: Core revision 20100428 [ 0.061088] [ 0.061090] =================================================== [ 0.062009] [ INFO: suspicious rcu_dereference_check() usage. ] [ 0.062138] --------------------------------------------------- [ 0.062268] kernel/sched.c:616 invoked rcu_dereference_check() without protection! [ 0.062470] [ 0.062471] other info that might help us debug this: [ 0.062472] [ 0.062835] [ 0.062836] rcu_scheduler_active = 1, debug_locks = 1 [ 0.063009] no locks held by swapper/0. [ 0.063134] [ 0.063135] stack backtrace: [ 0.063378] Pid: 0, comm: swapper Not tainted 2.6.35-rc2-git3 #3 [ 0.063507] Call Trace: [ 0.063638] [<ffffffff81072205>] lockdep_rcu_dereference+0x9d/0xa5 [ 0.063773] [<ffffffff810379f9>] task_group+0x7b/0x8a [ 0.064012] [<ffffffff81037a1d>] set_task_rq+0x15/0x6e [ 0.064143] [<ffffffff8103e50f>] set_task_cpu+0xa9/0xba [ 0.064274] [<ffffffff81042dbb>] sched_fork+0x10a/0x1b3 [ 0.064405] [<ffffffff810446f9>] copy_process+0x617/0x10e6 [ 0.064537] [<ffffffff8104533d>] do_fork+0x175/0x39b [ 0.064670] [<ffffffff8106589b>] ? up+0xf/0x39 [ 0.064800] [<ffffffff8106589b>] ? up+0xf/0x39 [ 0.065013] [<ffffffff811dbf73>] ? do_raw_spin_lock+0x79/0x13e [ 0.065148] [<ffffffff81011526>] kernel_thread+0x70/0x72 [ 0.065279] [<ffffffff816cc5e4>] ? kernel_init+0x0/0x1ce [ 0.065411] [<ffffffff8100aba0>] ? kernel_thread_helper+0x0/0x10 [ 0.065545] [<ffffffff81096bea>] ? rcu_scheduler_starting+0x2a/0x4c [ 0.065679] [<ffffffff813a8a4d>] rest_init+0x21/0xde [ 0.065810] [<ffffffff816cce28>] start_kernel+0x448/0x453 [ 0.066013] [<ffffffff816cc2c8>] x86_64_start_reservations+0xb3/0xb7 [ 0.066148] [<ffffffff816cc418>] x86_64_start_kernel+0x14c/0x15b [ 0.066499] Setting APIC routing to flat
Argh, moar funkeh stuff.. Either we do something like the below, or add something like (p->flags & PF_STARTING) to the task_subsys_state_check(), opinions? --- kernel/sched.c | 8 ++++++++ 1 files changed, 8 insertions(+), 0 deletions(-)
diff --git a/kernel/sched.c b/kernel/sched.c
index 19b3c5d..bfd3128 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c@@ -2564,7 +2564,15 @@ void sched_fork(struct task_struct *p, int clone_flags) if (p->sched_class->task_fork) p->sched_class->task_fork(p); + /* + * We're not in the pid-hash yet so no cgroup attach races, and the + * cgroup is pinned by the parent running this. + * + * Silence PROVE_RCU. + */ + rcu_read_lock(); set_task_cpu(p, cpu); + rcu_read_unlock(); #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) if (likely(sched_info_on()))