Re: pull request: bluetooth-next 2014-11-07

From: Kirill A. Shutemov
Date: Wed Nov 19 2014 - 06:07:19 EST


On Tue, Nov 18, 2014 at 08:09:19PM +0200, Johan Hedberg wrote:
> Hi Kirill,
>
> On Tue, Nov 18, 2014, Kirill A. Shutemov wrote:
> > On Fri, Nov 07, 2014 at 11:27:54AM +0200, Johan Hedberg wrote:
> > > Chan-yeol Park (1):
> > > Bluetooth: Fix hci_sync missing wakeup interrupt
> >
> > Look like this commit causes problem for me:
> >
> > [ 2.018671] ------------[ cut here ]------------
> > [ 2.022836] WARNING: CPU: 2 PID: 109 at /home/kas/git/public/linux/kernel/sched/core.c:7323 __might_sleep+0xbd/0xd0()
> > [ 2.023166] Freeing unused kernel memory: 944K (ffff880001b14000 - ffff880001c00000)
> > [ 2.030362] do not call blocking ops when !TASK_RUNNING; state=1 set at [<ffffffff819ab67b>] __hci_req_sync+0x7b/0x2a0
> > [ 2.034193] Modules linked in:
> > [ 2.036133] CPU: 2 PID: 109 Comm: kworker/u17:0 Not tainted 3.18.0-rc4-next-20141117-07404-g9dad2ab6df8b #66
> > [ 2.036383] Freeing unused kernel memory: 1404K (ffff8800020a1000 - ffff880002200000)
> > [ 2.038940] Hardware name: LENOVO 3460CC6/3460CC6, BIOS G6ET93WW (2.53 ) 02/04/2013
> > [ 2.040233] Workqueue: hci0 hci_power_on
> > [ 2.041517] ffffffff81f51d50 ffff8800d3ecfb28 ffffffff81afc316 0000000000000000
> > [ 2.042811] ffff8800d3ecfb78 ffff8800d3ecfb68 ffffffff810fc11a ffff880000000000
> > [ 2.042812] ffffffff81f52a28 000000000000026d 0000000000000000 ffff8800d3ec9200
> > [ 2.042813] Call Trace:
> > [ 2.042818] [<ffffffff81afc316>] dump_stack+0x4c/0x6e
> > [ 2.042821] [<ffffffff810fc11a>] warn_slowpath_common+0x8a/0xc0
> > [ 2.042823] [<ffffffff810fc196>] warn_slowpath_fmt+0x46/0x50
> > [ 2.042825] [<ffffffff819ab67b>] ? __hci_req_sync+0x7b/0x2a0
> > [ 2.042827] [<ffffffff819ab67b>] ? __hci_req_sync+0x7b/0x2a0
> > [ 2.042830] [<ffffffff811295ed>] __might_sleep+0xbd/0xd0
> > [ 2.042832] [<ffffffff81b04b8f>] mutex_lock_nested+0x2f/0x450
> > [ 2.042835] [<ffffffff81b0812b>] ? _raw_spin_unlock+0x2b/0x50
> > [ 2.042837] [<ffffffff81131db0>] ? wake_up_process+0x50/0x50
> > [ 2.042840] [<ffffffff813b9ea1>] __create_file+0x71/0x2c0
> > [ 2.042842] [<ffffffff813ba10f>] debugfs_create_file+0x1f/0x30
> > [ 2.042844] [<ffffffff819ac461>] hci_dev_do_open+0x431/0xa70
> > [ 2.042846] [<ffffffff8111a72c>] ? process_one_work+0x13c/0x810
> > [ 2.042847] [<ffffffff819ad2b1>] hci_power_on+0x31/0x1e0
> > [ 2.042849] [<ffffffff8111a7c8>] process_one_work+0x1d8/0x810
> > [ 2.042850] [<ffffffff8111a72c>] ? process_one_work+0x13c/0x810
> > [ 2.042851] [<ffffffff8111b2db>] worker_thread+0x6b/0x4b0
> > [ 2.042852] [<ffffffff8111b270>] ? init_pwq+0xf0/0xf0
> > [ 2.042854] [<ffffffff81120ef9>] kthread+0x119/0x130
> > [ 2.042855] [<ffffffff8112923d>] ? finish_task_switch+0x4d/0x140
> > [ 2.042857] [<ffffffff81120de0>] ? kthread_create_on_node+0x240/0x240
> > [ 2.042859] [<ffffffff81b08cfc>] ret_from_fork+0x7c/0xb0
> > [ 2.042861] [<ffffffff81120de0>] ? kthread_create_on_node+0x240/0x240
> > [ 2.042863] ---[ end trace 3a40e80ec4ca7159 ]---
>
> At least one obvious bug that the patch in question seems to have is
> that it fails to set the state back to TASK_RUNNING in case
> hci_req_run() fails and we never call schedule_timeout(). This would
> also seem to match the error you're getting.
>
> The attached patch fixes the missing set_current_state() calls - could
> you please check if it fixes the issue for you? Out of curiosity, which
> HW are you reproducing this with and are there any special steps
> involved? Looking at the code it seems that at least some older
> Bluetooth adapters could cause some hci_req_run() calls to return
> ENODATA, however I wasn't able to get the warning with any of my own
> adapters (I have one for pretty much every Bluetooth version).
>
> Johan

> From a75be9ae3c163db6a812330b5b50079891e1f7bd Mon Sep 17 00:00:00 2001
> From: Johan Hedberg <johan.hedberg@xxxxxxxxx>
> Date: Tue, 18 Nov 2014 20:00:15 +0200
> Subject: [PATCH] Bluetooth: Fix setting state back to TASK_RUNNING
>
> In __hci_cmd_sync_ev() and __hci_req_sync() if the hci_req_run() call
> fails and we return from the functions we should ensure that the state
> doesn't remain in TASK_INTERRUPTIBLE that we just set it to. This patch
> fixes missing calls to set_current_state(TASK_RUNNING) in both places.
>
> Signed-off-by: Johan Hedberg <johan.hedberg@xxxxxxxxx>

This patch helps -- warning is gone.

Trace from btmon on 'hciconfig hci0 up' is attached.

--
Kirill A. Shutemov

Attachment: trace
Description: Binary data