media: s5p-mfc: Fix lock contention - request_firmware() once
authorShuah Khan <shuahkh@osg.samsung.com>
Sat, 4 Nov 2017 02:01:58 +0000 (22:01 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sat, 24 Mar 2018 10:01:24 +0000 (11:01 +0100)
commitbadf37254bc7906899915b464d0c203e645e48dc
treedac08ef5933bf3d740eef278e4b4cc76b7e9f83c
parent639dab36edbdb186dad9dad40750723a4f820873
media: s5p-mfc: Fix lock contention - request_firmware() once

[ Upstream commit f45ce9877561044090010e0eb0fad644232ded04 ]

Driver calls request_firmware() whenever the device is opened for the
first time. As the device gets opened and closed, dev->num_inst == 1
is true several times. This is not necessary since the firmware is saved
in the fw_buf. s5p_mfc_load_firmware() copies the buffer returned by
the request_firmware() to dev->fw_buf.

fw_buf sticks around until it gets released from s5p_mfc_remove(), hence
there is no need to keep requesting firmware and copying it to fw_buf.

This might have been overlooked when changes are made to free fw_buf from
the device release interface s5p_mfc_release().

Fix s5p_mfc_load_firmware() to call request_firmware() once and keep state.
Change _probe() to load firmware once fw_buf has been allocated.

s5p_mfc_open() and it continues to call s5p_mfc_load_firmware() and init
hardware which is the step where firmware is written to the device.

This addresses the mfc_mutex contention due to repeated request_firmware()
calls from open() in the following circular locking warning:

[  552.194115] qtdemux0:sink/2710 is trying to acquire lock:
[  552.199488]  (&dev->mfc_mutex){+.+.}, at: [<bf145544>] s5p_mfc_mmap+0x28/0xd4 [s5p_mfc]
[  552.207459]
               but task is already holding lock:
[  552.213264]  (&mm->mmap_sem){++++}, at: [<c01df2e4>] vm_mmap_pgoff+0x44/0xb8
[  552.220284]
               which lock already depends on the new lock.

[  552.228429]
               the existing dependency chain (in reverse order) is:
[  552.235881]
               -> #2 (&mm->mmap_sem){++++}:
[  552.241259]        __might_fault+0x80/0xb0
[  552.245331]        filldir64+0xc0/0x2f8
[  552.249144]        call_filldir+0xb0/0x14c
[  552.253214]        ext4_readdir+0x768/0x90c
[  552.257374]        iterate_dir+0x74/0x168
[  552.261360]        SyS_getdents64+0x7c/0x1a0
[  552.265608]        ret_fast_syscall+0x0/0x28
[  552.269850]
               -> #1 (&type->i_mutex_dir_key#2){++++}:
[  552.276180]        down_read+0x48/0x90
[  552.279904]        lookup_slow+0x74/0x178
[  552.283889]        walk_component+0x1a4/0x2e4
[  552.288222]        link_path_walk+0x174/0x4a0
[  552.292555]        path_openat+0x68/0x944
[  552.296541]        do_filp_open+0x60/0xc4
[  552.300528]        file_open_name+0xe4/0x114
[  552.304772]        filp_open+0x28/0x48
[  552.308499]        kernel_read_file_from_path+0x30/0x78
[  552.313700]        _request_firmware+0x3ec/0x78c
[  552.318291]        request_firmware+0x3c/0x54
[  552.322642]        s5p_mfc_load_firmware+0x54/0x150 [s5p_mfc]
[  552.328358]        s5p_mfc_open+0x4e4/0x550 [s5p_mfc]
[  552.333394]        v4l2_open+0xa0/0x104 [videodev]
[  552.338137]        chrdev_open+0xa4/0x18c
[  552.342121]        do_dentry_open+0x208/0x310
[  552.346454]        path_openat+0x28c/0x944
[  552.350526]        do_filp_open+0x60/0xc4
[  552.354512]        do_sys_open+0x118/0x1c8
[  552.358586]        ret_fast_syscall+0x0/0x28
[  552.362830]
               -> #0 (&dev->mfc_mutex){+.+.}:
               -> #0 (&dev->mfc_mutex){+.+.}:
[  552.368379]        lock_acquire+0x6c/0x88
[  552.372364]        __mutex_lock+0x68/0xa34
[  552.376437]        mutex_lock_interruptible_nested+0x1c/0x24
[  552.382086]        s5p_mfc_mmap+0x28/0xd4 [s5p_mfc]
[  552.386939]        v4l2_mmap+0x54/0x88 [videodev]
[  552.391601]        mmap_region+0x3a8/0x638
[  552.395673]        do_mmap+0x330/0x3a4
[  552.399400]        vm_mmap_pgoff+0x90/0xb8
[  552.403472]        SyS_mmap_pgoff+0x90/0xc0
[  552.407632]        ret_fast_syscall+0x0/0x28
[  552.411876]
               other info that might help us debug this:

[  552.419848] Chain exists of:
                 &dev->mfc_mutex --> &type->i_mutex_dir_key#2 --> &mm->mmap_sem

[  552.431200]  Possible unsafe locking scenario:

[  552.437092]        CPU0                    CPU1
[  552.441598]        ----                    ----
[  552.446104]   lock(&mm->mmap_sem);
[  552.449484]                                lock(&type->i_mutex_dir_key#2);
[  552.456329]                                lock(&mm->mmap_sem);
[  552.462222]   lock(&dev->mfc_mutex);
[  552.465775]
                *** DEADLOCK ***

Signed-off-by: Shuah Khan <shuahkh@osg.samsung.com>
Signed-off-by: Sylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@s-opensource.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/media/platform/s5p-mfc/s5p_mfc.c
drivers/media/platform/s5p-mfc/s5p_mfc_common.h
drivers/media/platform/s5p-mfc/s5p_mfc_ctrl.c