Re: [ANNOUNCE] 3.0-rt4

[Peter Zijlstra <peterz@infradead.org>]

On Thu, 2011-07-28 at 12:16 -0700, Fernando Lopez-Lezcano wrote:
> [    0.000000] =============================================
> [    0.000000] [ INFO: possible recursive locking detected ]
> [    0.000000] 3.0.0-1.rt5.1.fc15.ccrma.i686.rtPAE #1
> [    0.000000] ---------------------------------------------
> [    0.000000] swapper/0 is trying to acquire lock:
> [    0.000000]  (&parent->list_lock){+.+...}, at: [<c04fb406>] __cache_free+0x43/0xc3
> [    0.000000]
> [    0.000000] but task is already holding lock:
> [    0.000000]  (&parent->list_lock){+.+...}, at: [<c04fc538>] do_tune_cpucache+0xf2/0x2bb
> [    0.000000]
> [    0.000000] other info that might help us debug this:
> [    0.000000]  Possible unsafe locking scenario:
> [    0.000000]
> [    0.000000]        CPU0
> [    0.000000]        ----
> [    0.000000]   lock(&parent->list_lock);
> [    0.000000]   lock(&parent->list_lock);
> [    0.000000]
> [    0.000000]  *** DEADLOCK ***
> [    0.000000]
> [    0.000000]  May be due to missing lock nesting notation
> [    0.000000]
> [    0.000000] 3 locks held by swapper/0:
> [    0.000000]  #0:  (cache_chain_mutex){+.+...}, at: [<c0bd9d2b>] kmem_cache_init_late+0xe/0x61
> [    0.000000]  #1:  (&per_cpu(slab_lock, __cpu).lock){+.+...}, at: [<c04faa65>] __local_lock_irq+0x1e/0x5b
> [    0.000000]  #2:  (&parent->list_lock){+.+...}, at: [<c04fc538>] do_tune_cpucache+0xf2/0x2bb
> [    0.000000]
> [    0.000000] stack backtrace:
> [    0.000000] Pid: 0, comm: swapper Not tainted 3.0.0-1.rt5.1.fc15.ccrma.i686.rtPAE #1
> [    0.000000] Call Trace:
> [    0.000000]  [<c0856355>] ? printk+0x2d/0x2f
> [    0.000000]  [<c0474a4b>] __lock_acquire+0x805/0xb57
> [    0.000000]  [<c0472604>] ? lock_release_holdtime.part.10+0x4b/0x51
> [    0.000000]  [<c085ecb4>] ? _raw_spin_unlock+0x31/0x3d
> [    0.000000]  [<c085dbc5>] ? rt_spin_lock_slowlock+0x75/0x190
> [    0.000000]  [<c04720c3>] ? trace_hardirqs_off+0xb/0xd
> [    0.000000]  [<c04fb406>] ? __cache_free+0x43/0xc3
> [    0.000000]  [<c0475215>] lock_acquire+0xde/0x11d
> [    0.000000]  [<c04fb406>] ? __cache_free+0x43/0xc3
> [    0.000000]  [<c085e24f>] rt_spin_lock+0x50/0x56
> [    0.000000]  [<c04fb406>] ? __cache_free+0x43/0xc3
> [    0.000000]  [<c04fb406>] __cache_free+0x43/0xc3
> [    0.000000]  [<c043646d>] ? test_ti_thread_flag+0x8/0x10
> [    0.000000]  [<c04fb23f>] kmem_cache_free+0x6c/0xdc
> [    0.000000]  [<c04fb2fe>] slab_destroy+0x4f/0x53
> [    0.000000]  [<c04fb396>] free_block+0x94/0xc1
> [    0.000000]  [<c04fc551>] do_tune_cpucache+0x10b/0x2bb
> [    0.000000]  [<c04fc8dc>] enable_cpucache+0x7b/0xa7
> [    0.000000]  [<c0bd9d3c>] kmem_cache_init_late+0x1f/0x61
> [    0.000000]  [<c0bba687>] start_kernel+0x24c/0x363
> [    0.000000]  [<c0bba1c4>] ? loglevel+0x18/0x18
> [    0.000000]  [<c0bba0ba>] i386_start_kernel+0xa9/0xaf 

Ooh, fun.. one does wonder why mainline doesn't show this..

This is the normal OFF_SLAB recursion, and the reason this shows up is
because this is ran before we do the lockdep fixup.

Fernando, does something like the below (not actually against -rt, but
it shouldn't matter much) fix things?

---
Subject: slab, lockdep: Annotate the locks before using

Fernando found we hit the regular OFF_SLAB 'recursion' before we
annotate the locks, cure this.

Reported-by: Fernando Lopez-Lezcano <nando@ccrma.Stanford.EDU>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
Index: linux-2.6/mm/slab.c
===================================================================
--- linux-2.6.orig/mm/slab.c
+++ linux-2.6/mm/slab.c
@@ -1665,6 +1665,9 @@ void __init kmem_cache_init_late(void)
 {
 	struct kmem_cache *cachep;
 
+	/* Annotate slab for lockdep -- annotate the malloc caches */
+	init_lock_keys();
+
 	/* 6) resize the head arrays to their final sizes */
 	mutex_lock(&cache_chain_mutex);
 	list_for_each_entry(cachep, &cache_chain, next)
@@ -1675,9 +1678,6 @@ void __init kmem_cache_init_late(void)
 	/* Done! */
 	g_cpucache_up = FULL;
 
-	/* Annotate slab for lockdep -- annotate the malloc caches */
-	init_lock_keys();
-
 	/*
 	 * Register a cpu startup notifier callback that initializes
 	 * cpu_cache_get for all new cpus