[media] rc-core: don't lock device at rc_register_device()
authorMauro Carvalho Chehab <mchehab@osg.samsung.com>
Thu, 11 Feb 2016 12:33:31 +0000 (10:33 -0200)
committerMauro Carvalho Chehab <mchehab@osg.samsung.com>
Tue, 16 Feb 2016 10:40:41 +0000 (08:40 -0200)
commitc73bbaa4ec3eb225ffe468f80d45724d0496bf03
tree99d5c2efef9c6f452cc8750c8873bbd2514bf275
parenteee7d353a19032b48c0f71504081de84a0ee79d8
[media] rc-core: don't lock device at rc_register_device()

The mutex lock at rc_register_device() was added by commit 08aeb7c9a42a
("[media] rc: add locking to fix register/show race").

It is meant to avoid race issues when trying to open a sysfs file while
the RC register didn't complete.

Adding a lock there causes troubles, as detected by the Kernel lock
debug instrumentation at the Kernel:

    ======================================================
    [ INFO: possible circular locking dependency detected ]
    4.5.0-rc3+ #46 Not tainted
    -------------------------------------------------------
    systemd-udevd/2681 is trying to acquire lock:
     (s_active#171){++++.+}, at: [<ffffffff8171a115>] kernfs_remove_by_name_ns+0x45/0xa0

    but task is already holding lock:
     (&dev->lock){+.+.+.}, at: [<ffffffffa0724def>] rc_register_device+0xb2f/0x1450 [rc_core]

    which lock already depends on the new lock.

    the existing dependency chain (in reverse order) is:

    -> #1 (&dev->lock){+.+.+.}:
           [<ffffffff8124817d>] lock_acquire+0x13d/0x320
           [<ffffffff822de966>] mutex_lock_nested+0xb6/0x860
           [<ffffffffa0721f2b>] show_protocols+0x3b/0x3f0 [rc_core]
           [<ffffffff81cdaba5>] dev_attr_show+0x45/0xc0
           [<ffffffff8171f1b3>] sysfs_kf_seq_show+0x203/0x3c0
           [<ffffffff8171a6a1>] kernfs_seq_show+0x121/0x1b0
           [<ffffffff81617c71>] seq_read+0x2f1/0x1160
           [<ffffffff8171c911>] kernfs_fop_read+0x321/0x460
           [<ffffffff815abc20>] __vfs_read+0xe0/0x3d0
           [<ffffffff815ae90e>] vfs_read+0xde/0x2d0
           [<ffffffff815b1d01>] SyS_read+0x111/0x230
           [<ffffffff822e8636>] entry_SYSCALL_64_fastpath+0x16/0x76

    -> #0 (s_active#171){++++.+}:
           [<ffffffff81244f24>] __lock_acquire+0x4304/0x5990
           [<ffffffff8124817d>] lock_acquire+0x13d/0x320
           [<ffffffff81717d3a>] __kernfs_remove+0x58a/0x810
           [<ffffffff8171a115>] kernfs_remove_by_name_ns+0x45/0xa0
           [<ffffffff81721592>] remove_files.isra.0+0x72/0x190
           [<ffffffff8172174b>] sysfs_remove_group+0x9b/0x150
           [<ffffffff81721854>] sysfs_remove_groups+0x54/0xa0
           [<ffffffff81cd97d0>] device_remove_attrs+0xb0/0x140
           [<ffffffff81cdb27c>] device_del+0x38c/0x6b0
           [<ffffffffa0724b8b>] rc_register_device+0x8cb/0x1450 [rc_core]
           [<ffffffffa1326a7b>] dvb_usb_remote_init+0x66b/0x14d0 [dvb_usb]
           [<ffffffffa1321c81>] dvb_usb_device_init+0xf21/0x1860 [dvb_usb]
           [<ffffffffa13517dc>] dib0700_probe+0x14c/0x410 [dvb_usb_dib0700]
           [<ffffffff81dbb1dd>] usb_probe_interface+0x45d/0x940
           [<ffffffff81ce7e7a>] driver_probe_device+0x21a/0xc30
           [<ffffffff81ce89b1>] __driver_attach+0x121/0x160
           [<ffffffff81ce21bf>] bus_for_each_dev+0x11f/0x1a0
           [<ffffffff81ce6cdd>] driver_attach+0x3d/0x50
           [<ffffffff81ce5df9>] bus_add_driver+0x4c9/0x770
           [<ffffffff81cea39c>] driver_register+0x18c/0x3b0
           [<ffffffff81db6e98>] usb_register_driver+0x1f8/0x440
           [<ffffffffa074001e>] dib0700_driver_init+0x1e/0x1000 [dvb_usb_dib0700]
           [<ffffffff810021b1>] do_one_initcall+0x141/0x300
           [<ffffffff8144d8eb>] do_init_module+0x1d0/0x5ad
           [<ffffffff812f27b6>] load_module+0x6666/0x9ba0
           [<ffffffff812f5fe8>] SyS_finit_module+0x108/0x130
           [<ffffffff822e8636>] entry_SYSCALL_64_fastpath+0x16/0x76

    other info that might help us debug this:

     Possible unsafe locking scenario:

           CPU0                    CPU1
           ----                    ----
      lock(&dev->lock);
                                   lock(s_active#171);
                                   lock(&dev->lock);
      lock(s_active#171);

     *** DEADLOCK ***

    3 locks held by systemd-udevd/2681:
     #0:  (&dev->mutex){......}, at: [<ffffffff81ce8933>] __driver_attach+0xa3/0x160
     #1:  (&dev->mutex){......}, at: [<ffffffff81ce8941>] __driver_attach+0xb1/0x160
     #2:  (&dev->lock){+.+.+.}, at: [<ffffffffa0724def>] rc_register_device+0xb2f/0x1450 [rc_core]

In this specific case, some error happened during device init,
causing IR to be disabled.

Let's fix it by adding a var that will tell when the device is
initialized. Any calls before that will return a -EINVAL.

That should prevent the race issues.

Signed-off-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
drivers/media/rc/rc-main.c
include/media/rc-core.h