Fix lockdep false positive in add_full()
Subsystems (2)
PATCH 1/1 Fix lockdep false positive in add_full()
Hello!
The add_full() function currently has a lockdep_assert_held() requiring that the kmem_cache_node structure's ->list_lock be held. However, this lock is not acquired by add_full()'s caller deactivate_slab() in the full-node case unless debugging is enabled. Because full nodes are accessed only by debugging code, this state of affairs results in lockdep false-positive splats like the following:
[ 43.942868] WARNING: CPU: 0 PID: 698 at /home/paulmck/public_git/linux-rcu/mm/slub.c:1007 deactivate_slab+0x509/0x720() [ 43.943016] Modules linked in: [ 43.943016] CPU: 0 PID: 698 Comm: torture_onoff Not tainted 3.14.0-rc1+ #1 [ 43.943016] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2007 [ 43.943016] 00000000000003ef ffff88001e3f5ba8 ffffffff818952ec 0000000000000046 [ 43.943016] 0000000000000000 ffff88001e3f5be8 ffffffff81049517 ffffea0000784e00 [ 43.943016] 0000000000000000 ffffea00007a9000 0000000000000002 0000000000000000 [ 43.943016] Call Trace: [ 43.943016] [<ffffffff818952ec>] dump_stack+0x46/0x58 [ 43.943016] [<ffffffff81049517>] warn_slowpath_common+0x87/0xb0 [ 43.943016] [<ffffffff81049555>] warn_slowpath_null+0x15/0x20 [ 43.943016] [<ffffffff8116e679>] deactivate_slab+0x509/0x720 [ 43.943016] [<ffffffff8116eebb>] ? slab_cpuup_callback+0x3b/0x100 [ 43.943016] [<ffffffff8116ef52>] ? slab_cpuup_callback+0xd2/0x100 [ 43.943016] [<ffffffff8116ef24>] slab_cpuup_callback+0xa4/0x100 [ 43.943016] [<ffffffff818a4c14>] notifier_call_chain+0x54/0x110 [ 43.943016] [<ffffffff81075b79>] __raw_notifier_call_chain+0x9/0x10 [ 43.943016] [<ffffffff8104963b>] __cpu_notify+0x1b/0x30 [ 43.943016] [<ffffffff81049720>] cpu_notify_nofail+0x10/0x20 [ 43.943016] [<ffffffff8188cc5d>] _cpu_down+0x10d/0x2e0 [ 43.943016] [<ffffffff8188ce60>] cpu_down+0x30/0x50 [ 43.943016] [<ffffffff811205f3>] torture_onoff+0xd3/0x3c0 [ 43.943016] [<ffffffff81120520>] ? torture_onoff_stats+0x90/0x90 [ 43.943016] [<ffffffff810710df>] kthread+0xdf/0x100 [ 43.943016] [<ffffffff818a09cb>] ? _raw_spin_unlock_irq+0x2b/0x40 [ 43.943016] [<ffffffff81071000>] ? flush_kthread_worker+0x130/0x130 [ 43.943016] [<ffffffff818a983c>] ret_from_fork+0x7c/0xb0 [ 43.943016] [<ffffffff81071000>] ? flush_kthread_worker+0x130/0x130
This commit therefore does the lockdep check only if debuggging is enabled, thus avoiding the false positives.
mm/slub.c +2 -1
@@ -1004,7 +1004,8 @@ static inline void slab_free_hook(struct kmem_cache *s, void *x)static void add_full (struct kmem_cache *s,struct kmem_cache_node *n,struct page *page) { -lockdep_assert_held (&n->list_lock); +if (kmem_cache_debug (s)) +lockdep_assert_held (&n->list_lock);if (!(s->flags &SLAB_STORE_USER ))return ;
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
David Rientjes
reviewerQuoted 38 lines · Paul E. McKenney
Hello!
The
add_full()function currently has alockdep_assert_held()requiring that thekmem_cache_nodestructure's ->list_lock be held. However, this lock is not acquired by add_full()'s callerdeactivate_slab()in the full-node case unless debugging is enabled. Because full nodes are accessed only by debugging code, this state of affairs results in lockdep false-positive splats like the following:[ 43.942868] WARNING: CPU: 0 PID: 698 at /home/paulmck/public_git/linux-rcu/mm/slub.c:1007 deactivate_slab+0x509/0x720() [ 43.943016] Modules linked in: [ 43.943016] CPU: 0 PID: 698 Comm: torture_onoff Not tainted 3.14.0-rc1+ #1 [ 43.943016] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2007 [ 43.943016] 00000000000003ef ffff88001e3f5ba8 ffffffff818952ec 0000000000000046 [ 43.943016] 0000000000000000 ffff88001e3f5be8 ffffffff81049517 ffffea0000784e00 [ 43.943016] 0000000000000000 ffffea00007a9000 0000000000000002 0000000000000000 [ 43.943016] Call Trace: [ 43.943016] [<ffffffff818952ec>] dump_stack+0x46/0x58 [ 43.943016] [<ffffffff81049517>] warn_slowpath_common+0x87/0xb0 [ 43.943016] [<ffffffff81049555>] warn_slowpath_null+0x15/0x20 [ 43.943016] [<ffffffff8116e679>] deactivate_slab+0x509/0x720 [ 43.943016] [<ffffffff8116eebb>] ? slab_cpuup_callback+0x3b/0x100 [ 43.943016] [<ffffffff8116ef52>] ? slab_cpuup_callback+0xd2/0x100 [ 43.943016] [<ffffffff8116ef24>] slab_cpuup_callback+0xa4/0x100 [ 43.943016] [<ffffffff818a4c14>] notifier_call_chain+0x54/0x110 [ 43.943016] [<ffffffff81075b79>] __raw_notifier_call_chain+0x9/0x10 [ 43.943016] [<ffffffff8104963b>] __cpu_notify+0x1b/0x30 [ 43.943016] [<ffffffff81049720>] cpu_notify_nofail+0x10/0x20 [ 43.943016] [<ffffffff8188cc5d>] _cpu_down+0x10d/0x2e0 [ 43.943016] [<ffffffff8188ce60>] cpu_down+0x30/0x50 [ 43.943016] [<ffffffff811205f3>] torture_onoff+0xd3/0x3c0 [ 43.943016] [<ffffffff81120520>] ? torture_onoff_stats+0x90/0x90 [ 43.943016] [<ffffffff810710df>] kthread+0xdf/0x100 [ 43.943016] [<ffffffff818a09cb>] ? _raw_spin_unlock_irq+0x2b/0x40 [ 43.943016] [<ffffffff81071000>] ? flush_kthread_worker+0x130/0x130 [ 43.943016] [<ffffffff818a983c>] ret_from_fork+0x7c/0xb0 [ 43.943016] [<ffffffff81071000>] ? flush_kthread_worker+0x130/0x130
Quoted 4 lines
This commit therefore does the lockdep check only if debuggging is enabled, thus avoiding the false positives.
Signed-off-by Paul E. McKenney <[email protected]>
This was discussed in http://marc.info/?t=139145791300002, what do you think about the patch in that thread instead?
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Quoted 40 lines · David Rientjes
Quoted 38 lines · Paul E. McKenney
Hello!
The
add_full()function currently has alockdep_assert_held()requiring that thekmem_cache_nodestructure's ->list_lock be held. However, this lock is not acquired by add_full()'s callerdeactivate_slab()in the full-node case unless debugging is enabled. Because full nodes are accessed only by debugging code, this state of affairs results in lockdep false-positive splats like the following:[ 43.942868] WARNING: CPU: 0 PID: 698 at /home/paulmck/public_git/linux-rcu/mm/slub.c:1007 deactivate_slab+0x509/0x720() [ 43.943016] Modules linked in: [ 43.943016] CPU: 0 PID: 698 Comm: torture_onoff Not tainted 3.14.0-rc1+ #1 [ 43.943016] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2007 [ 43.943016] 00000000000003ef ffff88001e3f5ba8 ffffffff818952ec 0000000000000046 [ 43.943016] 0000000000000000 ffff88001e3f5be8 ffffffff81049517 ffffea0000784e00 [ 43.943016] 0000000000000000 ffffea00007a9000 0000000000000002 0000000000000000 [ 43.943016] Call Trace: [ 43.943016] [<ffffffff818952ec>] dump_stack+0x46/0x58 [ 43.943016] [<ffffffff81049517>] warn_slowpath_common+0x87/0xb0 [ 43.943016] [<ffffffff81049555>] warn_slowpath_null+0x15/0x20 [ 43.943016] [<ffffffff8116e679>] deactivate_slab+0x509/0x720 [ 43.943016] [<ffffffff8116eebb>] ? slab_cpuup_callback+0x3b/0x100 [ 43.943016] [<ffffffff8116ef52>] ? slab_cpuup_callback+0xd2/0x100 [ 43.943016] [<ffffffff8116ef24>] slab_cpuup_callback+0xa4/0x100 [ 43.943016] [<ffffffff818a4c14>] notifier_call_chain+0x54/0x110 [ 43.943016] [<ffffffff81075b79>] __raw_notifier_call_chain+0x9/0x10 [ 43.943016] [<ffffffff8104963b>] __cpu_notify+0x1b/0x30 [ 43.943016] [<ffffffff81049720>] cpu_notify_nofail+0x10/0x20 [ 43.943016] [<ffffffff8188cc5d>] _cpu_down+0x10d/0x2e0 [ 43.943016] [<ffffffff8188ce60>] cpu_down+0x30/0x50 [ 43.943016] [<ffffffff811205f3>] torture_onoff+0xd3/0x3c0 [ 43.943016] [<ffffffff81120520>] ? torture_onoff_stats+0x90/0x90 [ 43.943016] [<ffffffff810710df>] kthread+0xdf/0x100 [ 43.943016] [<ffffffff818a09cb>] ? _raw_spin_unlock_irq+0x2b/0x40 [ 43.943016] [<ffffffff81071000>] ? flush_kthread_worker+0x130/0x130 [ 43.943016] [<ffffffff818a983c>] ret_from_fork+0x7c/0xb0 [ 43.943016] [<ffffffff81071000>] ? flush_kthread_worker+0x130/0x130
Quoted 7 lines
Quoted 4 lines
This commit therefore does the lockdep check only if debuggging is enabled, thus avoiding the false positives.
Signed-off-by Paul E. McKenney <[email protected]>This was discussed in http://marc.info/?t=139145791300002, what do you think about the patch in that thread instead?
Looks fine to me! I also tried it out and it avoided the splats, as noted in my mail in the other thread, so please feel free to add my Tested-by.
Thanx, Paul
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>