mm, slub: move slub_debug static key enabling outside slab_mutex
authorVlastimil Babka <vbabka@suse.cz>
Sat, 15 May 2021 00:27:10 +0000 (17:27 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 15 May 2021 02:41:32 +0000 (19:41 -0700)
commitafe0c26d1968fe3bbef6a45df945bfeff774ca75
tree4b63f454766984d3aba48d946f8da59b34a5fac9
parent84894e1c42e9f25c17f2888e0c0e1505cb727538
mm, slub: move slub_debug static key enabling outside slab_mutex

Paul E.  McKenney reported [1] that commit 1f0723a4c0df ("mm, slub: enable
slub_debug static key when creating cache with explicit debug flags")
results in the lockdep complaint:

 ======================================================
 WARNING: possible circular locking dependency detected
 5.12.0+ #15 Not tainted
 ------------------------------------------------------
 rcu_torture_sta/109 is trying to acquire lock:
 ffffffff96063cd0 (cpu_hotplug_lock){++++}-{0:0}, at: static_key_enable+0x9/0x20

 but task is already holding lock:
 ffffffff96173c28 (slab_mutex){+.+.}-{3:3}, at: kmem_cache_create_usercopy+0x2d/0x250

 which lock already depends on the new lock.

 the existing dependency chain (in reverse order) is:

 -> #1 (slab_mutex){+.+.}-{3:3}:
        lock_acquire+0xb9/0x3a0
        __mutex_lock+0x8d/0x920
        slub_cpu_dead+0x15/0xf0
        cpuhp_invoke_callback+0x17a/0x7c0
        cpuhp_invoke_callback_range+0x3b/0x80
        _cpu_down+0xdf/0x2a0
        cpu_down+0x2c/0x50
        device_offline+0x82/0xb0
        remove_cpu+0x1a/0x30
        torture_offline+0x80/0x140
        torture_onoff+0x147/0x260
        kthread+0x10a/0x140
        ret_from_fork+0x22/0x30

 -> #0 (cpu_hotplug_lock){++++}-{0:0}:
        check_prev_add+0x8f/0xbf0
        __lock_acquire+0x13f0/0x1d80
        lock_acquire+0xb9/0x3a0
        cpus_read_lock+0x21/0xa0
        static_key_enable+0x9/0x20
        __kmem_cache_create+0x38d/0x430
        kmem_cache_create_usercopy+0x146/0x250
        kmem_cache_create+0xd/0x10
        rcu_torture_stats+0x79/0x280
        kthread+0x10a/0x140
        ret_from_fork+0x22/0x30

 other info that might help us debug this:

  Possible unsafe locking scenario:

        CPU0                    CPU1
        ----                    ----
   lock(slab_mutex);
                                lock(cpu_hotplug_lock);
                                lock(slab_mutex);
   lock(cpu_hotplug_lock);

  *** DEADLOCK ***

 1 lock held by rcu_torture_sta/109:
  #0: ffffffff96173c28 (slab_mutex){+.+.}-{3:3}, at: kmem_cache_create_usercopy+0x2d/0x250

 stack backtrace:
 CPU: 3 PID: 109 Comm: rcu_torture_sta Not tainted 5.12.0+ #15
 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.13.0-1ubuntu1.1 04/01/2014
 Call Trace:
  dump_stack+0x6d/0x89
  check_noncircular+0xfe/0x110
  ? lock_is_held_type+0x98/0x110
  check_prev_add+0x8f/0xbf0
  __lock_acquire+0x13f0/0x1d80
  lock_acquire+0xb9/0x3a0
  ? static_key_enable+0x9/0x20
  ? mark_held_locks+0x49/0x70
  cpus_read_lock+0x21/0xa0
  ? static_key_enable+0x9/0x20
  static_key_enable+0x9/0x20
  __kmem_cache_create+0x38d/0x430
  kmem_cache_create_usercopy+0x146/0x250
  ? rcu_torture_stats_print+0xd0/0xd0
  kmem_cache_create+0xd/0x10
  rcu_torture_stats+0x79/0x280
  ? rcu_torture_stats_print+0xd0/0xd0
  kthread+0x10a/0x140
  ? kthread_park+0x80/0x80
  ret_from_fork+0x22/0x30

This is because there's one order of locking from the hotplug callbacks:

lock(cpu_hotplug_lock); // from hotplug machinery itself
lock(slab_mutex); // in e.g. slab_mem_going_offline_callback()

And commit 1f0723a4c0df made the reverse sequence possible:
lock(slab_mutex); // in kmem_cache_create_usercopy()
lock(cpu_hotplug_lock); // kmem_cache_open() -> static_key_enable()

The simplest fix is to move static_key_enable() to a place before slab_mutex is
taken. That means kmem_cache_create_usercopy() in mm/slab_common.c which is not
ideal for SLUB-specific code, but the #ifdef CONFIG_SLUB_DEBUG makes it
at least self-contained and obvious.

[1] https://lore.kernel.org/lkml/20210502171827.GA3670492@paulmck-ThinkPad-P17-Gen-1/

Link: https://lkml.kernel.org/r/20210504120019.26791-1-vbabka@suse.cz
Fixes: 1f0723a4c0df ("mm, slub: enable slub_debug static key when creating cache with explicit debug flags")
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Reported-by: Paul E. McKenney <paulmck@kernel.org>
Tested-by: Paul E. McKenney <paulmck@kernel.org>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/slab_common.c
mm/slub.c