sysfs: Disable lockdep for driver bind/unbind files
authorDaniel Vetter <daniel.vetter@ffwll.ch>
Wed, 19 Dec 2018 12:39:09 +0000 (13:39 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 19 Dec 2018 14:58:40 +0000 (15:58 +0100)
commit4f4b374332ec0ae9c738ff8ec9bed5cd97ff9adc
tree0073b302437eb4efd66a33f764209e6d229021b9
parente121a833745b4708b660e3fe6776129c2956b041
sysfs: Disable lockdep for driver bind/unbind files

This is the much more correct fix for my earlier attempt at:

https://lkml.org/lkml/2018/12/10/118

Short recap:

- There's not actually a locking issue, it's just lockdep being a bit
  too eager to complain about a possible deadlock.

- Contrary to what I claimed the real problem is recursion on
  kn->count. Greg pointed me at sysfs_break_active_protection(), used
  by the scsi subsystem to allow a sysfs file to unbind itself. That
  would be a real deadlock, which isn't what's happening here. Also,
  breaking the active protection means we'd need to manually handle
  all the lifetime fun.

- With Rafael we discussed the task_work approach, which kinda works,
  but has two downsides: It's a functional change for a lockdep
  annotation issue, and it won't work for the bind file (which needs
  to get the errno from the driver load function back to userspace).

- Greg also asked why this never showed up: To hit this you need to
  unregister a 2nd driver from the unload code of your first driver. I
  guess only gpus do that. The bug has always been there, but only
  with a recent patch series did we add more locks so that lockdep
  built a chain from unbinding the snd-hda driver to the
  acpi_video_unregister call.

Full lockdep splat:

[12301.898799] ============================================
[12301.898805] WARNING: possible recursive locking detected
[12301.898811] 4.20.0-rc7+ #84 Not tainted
[12301.898815] --------------------------------------------
[12301.898821] bash/5297 is trying to acquire lock:
[12301.898826] 00000000f61c6093 (kn->count#39){++++}, at: kernfs_remove_by_name_ns+0x3b/0x80
[12301.898841] but task is already holding lock:
[12301.898847] 000000005f634021 (kn->count#39){++++}, at: kernfs_fop_write+0xdc/0x190
[12301.898856] other info that might help us debug this:
[12301.898862]  Possible unsafe locking scenario:
[12301.898867]        CPU0
[12301.898870]        ----
[12301.898874]   lock(kn->count#39);
[12301.898879]   lock(kn->count#39);
[12301.898883] *** DEADLOCK ***
[12301.898891]  May be due to missing lock nesting notation
[12301.898899] 5 locks held by bash/5297:
[12301.898903]  #0: 00000000cd800e54 (sb_writers#4){.+.+}, at: vfs_write+0x17f/0x1b0
[12301.898915]  #1: 000000000465e7c2 (&of->mutex){+.+.}, at: kernfs_fop_write+0xd3/0x190
[12301.898925]  #2: 000000005f634021 (kn->count#39){++++}, at: kernfs_fop_write+0xdc/0x190
[12301.898936]  #3: 00000000414ef7ac (&dev->mutex){....}, at: device_release_driver_internal+0x34/0x240
[12301.898950]  #4: 000000003218fbdf (register_count_mutex){+.+.}, at: acpi_video_unregister+0xe/0x40
[12301.898960] stack backtrace:
[12301.898968] CPU: 1 PID: 5297 Comm: bash Not tainted 4.20.0-rc7+ #84
[12301.898974] Hardware name: Hewlett-Packard HP EliteBook 8460p/161C, BIOS 68SCF Ver. F.01 03/11/2011
[12301.898982] Call Trace:
[12301.898989]  dump_stack+0x67/0x9b
[12301.898997]  __lock_acquire+0x6ad/0x1410
[12301.899003]  ? kernfs_remove_by_name_ns+0x3b/0x80
[12301.899010]  ? find_held_lock+0x2d/0x90
[12301.899017]  ? mutex_spin_on_owner+0xe4/0x150
[12301.899023]  ? find_held_lock+0x2d/0x90
[12301.899030]  ? lock_acquire+0x90/0x180
[12301.899036]  lock_acquire+0x90/0x180
[12301.899042]  ? kernfs_remove_by_name_ns+0x3b/0x80
[12301.899049]  __kernfs_remove+0x296/0x310
[12301.899055]  ? kernfs_remove_by_name_ns+0x3b/0x80
[12301.899060]  ? kernfs_name_hash+0xd/0x80
[12301.899066]  ? kernfs_find_ns+0x6c/0x100
[12301.899073]  kernfs_remove_by_name_ns+0x3b/0x80
[12301.899080]  bus_remove_driver+0x92/0xa0
[12301.899085]  acpi_video_unregister+0x24/0x40
[12301.899127]  i915_driver_unload+0x42/0x130 [i915]
[12301.899160]  i915_pci_remove+0x19/0x30 [i915]
[12301.899169]  pci_device_remove+0x36/0xb0
[12301.899176]  device_release_driver_internal+0x185/0x240
[12301.899183]  unbind_store+0xaf/0x180
[12301.899189]  kernfs_fop_write+0x104/0x190
[12301.899195]  __vfs_write+0x31/0x180
[12301.899203]  ? rcu_read_lock_sched_held+0x6f/0x80
[12301.899209]  ? rcu_sync_lockdep_assert+0x29/0x50
[12301.899216]  ? __sb_start_write+0x13c/0x1a0
[12301.899221]  ? vfs_write+0x17f/0x1b0
[12301.899227]  vfs_write+0xb9/0x1b0
[12301.899233]  ksys_write+0x50/0xc0
[12301.899239]  do_syscall_64+0x4b/0x180
[12301.899247]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
[12301.899253] RIP: 0033:0x7f452ac7f7a4
[12301.899259] Code: 00 f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b7 0f 1f 80 00 00 00 00 8b 05 aa f0 2c 00 48 63 ff 85 c0 75 13 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 54 f3 c3 66 90 55 53 48 89 d5 48 89 f3 48 83
[12301.899273] RSP: 002b:00007ffceafa6918 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
[12301.899282] RAX: ffffffffffffffda RBX: 000000000000000d RCX: 00007f452ac7f7a4
[12301.899288] RDX: 000000000000000d RSI: 00005612a1abf7c0 RDI: 0000000000000001
[12301.899295] RBP: 00005612a1abf7c0 R08: 000000000000000a R09: 00005612a1c46730
[12301.899301] R10: 000000000000000a R11: 0000000000000246 R12: 000000000000000d
[12301.899308] R13: 0000000000000001 R14: 00007f452af4a740 R15: 000000000000000d

Looking around I've noticed that usb and i2c already handle similar
recursion problems, where a sysfs file can unbind the same type of
sysfs somewhere else in the hierarchy. Relevant commits are:

commit 356c05d58af05d582e634b54b40050c73609617b
Author: Alan Stern <stern@rowland.harvard.edu>
Date:   Mon May 14 13:30:03 2012 -0400

    sysfs: get rid of some lockdep false positives

commit e9b526fe704812364bca07edd15eadeba163ebfb
Author: Alexander Sverdlin <alexander.sverdlin@nsn.com>
Date:   Fri May 17 14:56:35 2013 +0200

    i2c: suppress lockdep warning on delete_device

Implement the same trick for driver bind/unbind.

v2: Put the macro into bus.c (Greg).

Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: Ramalingam C <ramalingam.c@intel.com>
Cc: Arend van Spriel <aspriel@gmail.com>
Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Geert Uytterhoeven <geert+renesas@glider.be>
Cc: Bartosz Golaszewski <brgl@bgdev.pl>
Cc: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Cc: Vivek Gautam <vivek.gautam@codeaurora.org>
Cc: Joe Perches <joe@perches.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/base/bus.c