Date:	Sun, 31 Aug 2014 11:40:43 -0400
From:	Douglas Gilbert <dgilbert@...erlog.com>
To:	Milan Broz <gmazyland@...il.com>
CC:	linux-scsi@...r.kernel.org,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: scsi_debug module deadlock on 3.17-rc2

On 14-08-31 05:49 AM, Milan Broz wrote:
> On 08/31/2014 01:14 AM, Douglas Gilbert wrote:
>> On 14-08-30 04:56 PM, Milan Broz wrote:
>>> Hi,
>>>
>>> I am using scsi_debug in cryptsetup testsuite and with recent 3.17-rc kernel
>>> it deadlocks on rmmod of scsi_debug module.
>>>
>>> For me even this simple reproducer causes deadlock:
>>>     modprobe scsi_debug dev_size_mb=16 sector_size=512 num_tgts=1
>>>     DEV="/dev/"$(grep -l -e scsi_debug /sys/block/*/device/model | cut -f4 -d /)
>>>     mkfs -t ext4 $DEV
>>>     rmmod scsi_debug
>>>
>>> (adding small delay before rmmod obviously helps here)
>>
>> So I used this slight variation for testing:
>>
>> modprobe scsi_debug dev_size_mb=16 sector_size=512 num_tgts=1 num_parts=1
>> DEV="/dev/"$(grep -l -e scsi_debug /sys/block/*/device/model | cut -f4 -d /)"1"
>> echo "mkfs -t ext4 ${DEV}"
>> mkfs -t ext4 ${DEV}
>> sleep 0.1
>> rmmod scsi_debug
>>
>>> Bisect tracked it to commit
>>>     commit cbf67842c3d9e7af8ccc031332b79e88d9cca592
>>>     Author: Douglas Gilbert <dgilbert@...erlog.com>
>>>     Date:   Sat Jul 26 11:55:35 2014 -0400
>>>     scsi_debug: support scsi-mq, queues and locks
>>>
>>> I guess that with introducing mq the del_timer_sync() must not be called
>>> with acquired queued_arr_lock.
>>> (to me it looks like situation described in comment before
>>> del_timer_sync() in kernel/time/timer.c...)
>>
>> Looks like something a lawyer would write.
>
> :-)
>
> ...
>
>> Anyway the attached patch removes the lock(queued_arr_lock)
>> from around the del_timer calls. Could you try it and report
>> back.
>
> Yes, the patch fixes the deadlock.
> Thanks!
>
> Tested-and-reported-by: Milan Broz <gmazyland@...il.com>
>
> However, removing the device (rmmod scsi_debug) now take up to 20 or more seconds.
> (Nothing like flush buffers or udev settle helps seems...)

I did not see that in my tests with your script, perhaps
I need to increase the 'delay=<jiffies>' argument to
scsi_debug and run rmmod in a loop until it succeeds.

And are you saying that in lk 3.16 and earlier you did
not observe a 20 second delay?


Now, what your test is doing, is telling a low level
driver at the bottom of the scsi stack to shut down,
before all the WRITE commands associated with the
mkfs may have completed and perhaps the upper layers have
assumed they are completed. The default timeout on those
WRITE commands (set in the upper layers) may well be
the 20 or more seconds you are observing.

As far as I can see the scsi_debug driver has two options
when rmmod is issued:
   1) shut everything; done in as orderly fashion as
      possible (with no deadlocks). This may mean deleting
      the timers associated with "in play" commands and
      forgetting about them. This will cause pain for the
      upper levels but they should be able to cope.
   2) if there are no scsi commands "in play" then
      shut down as in 1). Otherwise reject any new
      commands and create a work queue to handle the
      remaining "in play" commands. When they are
      completed, do 1)

The scsi_debug driver takes the simpler approach and
does 1) ***. This also has the advantage of exercising
the error handling (EH) logic in the scsi mid-layer.
And it may not be that unrealistic, perhaps mimicking
a real HBA driver whose firmware has crashed, as seen
from the upper layers.

Also, for drivers that do 2), how do they cope with
'rmmod scsi_debug; modprobe scsi_debug ...'. Perhaps
an EBUSY on that modprobe if the rmmod hasn't really
finished?

> It is somehow related to udev events processing (killing udev process
> will remove this delay).
>
> I have full process trace (let me know if you want to send it),

Yes, please send me the full trace. I will probably be
talking to the scsi EH maintainer about it. Perhaps
the scsi_debug driver needs to make more calls back to
the scsi mid-layer (its parent) in this situation.

Doug Gilbert


*** there is no change in lk 3.17 for scsi_debug's shutdown
     policy to 3.16 and earlier

> this is just where rmmod is while waiting (note the delay in the end):
>
> [  132.918141] sd 3:0:0:0: [sdh] Attached SCSI disk
> [  137.476448] SysRq : Show State
> ...
> [  137.485711] kworker/0:2     D c16ce000     0   274      2 0x00000000
> [  137.485724] Workqueue: events_freezable_power_ disk_events_workfn
> [  137.485728]  dc915c2c 00000046 dc915bcc c16ce000 debad340 0000001f de15e250 c00530f0
> [  137.485741]  00000002 c00530f0 00000001 c00530f0 c0053698 dc915c0c c106e934 c1465cdd
> [  137.485753]  c00530f0 dc915c44 dc915c0c c128925f dc915c1c c106ea49 00000282 c1ff7d80
> [  137.485765] Call Trace:
> [  137.485772]  [<c106e934>] ? mark_held_locks+0x64/0x90
> [  137.485779]  [<c1465cdd>] ? _raw_spin_unlock_irqrestore+0x4d/0x70
> [  137.485785]  [<c128925f>] ? __this_cpu_preempt_check+0xf/0x20
> [  137.485792]  [<c106ea49>] ? trace_hardirqs_on_caller+0xe9/0x220
> [  137.485798]  [<c146194e>] schedule+0x1e/0x60
> [  137.485804]  [<c146520f>] schedule_timeout+0x11f/0x1b0
> [  137.485811]  [<c1084270>] ? internal_add_timer+0x60/0x60
> [  137.485817]  [<c106ea49>] ? trace_hardirqs_on_caller+0xe9/0x220
> [  137.485824]  [<c1461db6>] io_schedule_timeout+0x46/0x60
> [  137.485831]  [<c146230b>] wait_for_common_io.constprop.1+0x7b/0xd0
> [  137.485836]  [<c1059c60>] ? wake_up_process+0x40/0x40
> [  137.485843]  [<c1462368>] wait_for_completion_io_timeout+0x8/0x10
> [  137.485849]  [<c12614a1>] blk_execute_rq+0x81/0x100
> [  137.485856]  [<c14622a7>] ? wait_for_common_io.constprop.1+0x17/0xd0
> [  137.485863]  [<c12730f0>] ? cfq_exit_icq+0x40/0x40
> [  137.485870]  [<c125a2d5>] ? elv_set_request+0x15/0x30
> [  137.485877]  [<c125bd4a>] ? get_request+0x2ea/0x490
> [  137.485884]  [<c1066fd0>] ? __wake_up_sync+0x10/0x10
> [  137.485891]  [<c125bf59>] ? blk_get_request+0x69/0xd0
> [  137.485897]  [<c13312bc>] scsi_execute+0xfc/0x190
> [  137.485904]  [<c1331423>] scsi_execute_req_flags+0x53/0xb0
> [  137.485912]  [<c13319be>] scsi_test_unit_ready+0x5e/0x100
> [  137.485918]  [<c1349df3>] sd_check_events+0xe3/0x120
> [  137.485925]  [<c12697f7>] disk_check_events+0x37/0x100
> [  137.485932]  [<c12698d3>] disk_events_workfn+0x13/0x20
> [  137.485939]  [<c104d096>] process_one_work+0x156/0x390
> [  137.485945]  [<c104d043>] ? process_one_work+0x103/0x390
> [  137.485951]  [<c104d8f9>] worker_thread+0x39/0x440
> [  137.485958]  [<c104d8c0>] ? execute_in_process_context+0x70/0x70
> [  137.485965]  [<c10513f3>] kthread+0xa3/0xc0
> [  137.485970]  [<c106cc9d>] ? put_lock_stats.isra.21+0xd/0x30
> [  137.485976]  [<c106eb8b>] ? trace_hardirqs_on+0xb/0x10
> [  137.485982]  [<c14667c1>] ret_from_kernel_thread+0x21/0x30
> [  137.485989]  [<c1051350>] ? kthread_create_on_node+0x160/0x160
> ...
> [  137.493862] rmmod           D c16ce000     0  1434   1422 0x00000000
> [  137.493869]  da1bbca4 00000046 deb5e050 c16ce000 199fe46e 0000001f df11c190 deb5e050
> [  137.493881]  b02da1a3 da1bbcc0 c10708cb c1c0f9c8 00000004 00000000 c1c75a90 00000000
> [  137.493893]  0ac0a782 b02da1a3 00000003 deb5e670 00000001 deb5e050 deb5e050 deb5e670
> [  137.493905] Call Trace:
> [  137.493911]  [<c10708cb>] ? __lock_acquire+0x48b/0x1c30
> [  137.493917]  [<c146194e>] schedule+0x1e/0x60
> [  137.493924]  [<c146523d>] schedule_timeout+0x14d/0x1b0
> [  137.493930]  [<c106e934>] ? mark_held_locks+0x64/0x90
> [  137.493936]  [<c1465d22>] ? _raw_spin_unlock_irq+0x22/0x50
> [  137.493942]  [<c128925f>] ? __this_cpu_preempt_check+0xf/0x20
> [  137.493948]  [<c106ea49>] ? trace_hardirqs_on_caller+0xe9/0x220
> [  137.493954]  [<c106eb8b>] ? trace_hardirqs_on+0xb/0x10
> [  137.493960]  [<c1462156>] wait_for_common+0x86/0x110
> [  137.493965]  [<c1059c60>] ? wake_up_process+0x40/0x40
> [  137.493971]  [<c14621f2>] wait_for_completion+0x12/0x20
> [  137.493977]  [<c104c9bf>] flush_work+0x13f/0x1e0
> [  137.493982]  [<c104abe0>] ? work_busy+0x80/0x80
> [  137.493988]  [<c14620ea>] ? wait_for_common+0x1a/0x110
> [  137.493994]  [<c104d7a7>] __cancel_work_timer+0x67/0xf0
> [  137.494000]  [<c104d84d>] cancel_delayed_work_sync+0xd/0x10
> [  137.494006]  [<c126ad93>] disk_block_events+0x73/0x80
> [  137.494011]  [<c126ae32>] del_gendisk+0x22/0x1d0
> [  137.494017]  [<c1318de1>] ? device_del+0x131/0x180
> [  137.494022]  [<c134a929>] sd_remove+0x49/0xa0
> [  137.494029]  [<c131bc59>] __device_release_driver+0x49/0xb0
> [  137.494034]  [<c131bcda>] device_release_driver+0x1a/0x30
> [  137.494040]  [<c131b7cb>] bus_remove_device+0xbb/0xf0
> [  137.494045]  [<c1318d9c>] device_del+0xec/0x180
> [  137.494051]  [<c1337981>] __scsi_remove_device+0xb1/0xc0
> [  137.494058]  [<c13364f7>] scsi_forget_host+0x57/0x60
> [  137.494064]  [<c132c75e>] scsi_remove_host+0x5e/0xf0
> [  137.494074]  [<e10e4fdb>] sdebug_driver_remove+0x1b/0x80 [scsi_debug]
> [  137.494081]  [<c131bc59>] __device_release_driver+0x49/0xb0
> [  137.494086]  [<c131bcda>] device_release_driver+0x1a/0x30
> [  137.494092]  [<c131b7cb>] bus_remove_device+0xbb/0xf0
> [  137.494097]  [<c1318d9c>] device_del+0xec/0x180
> [  137.494103]  [<c1318e3b>] device_unregister+0xb/0x20
> [  137.494110]  [<e10e48bd>] sdebug_remove_adapter+0x4d/0x70 [scsi_debug]
> [  137.494118]  [<e10e8ffd>] scsi_debug_exit+0x6e/0xac [scsi_debug]
> [  137.494125]  [<c109a74d>] SyS_delete_module+0xfd/0x180
> [  137.494131]  [<c14669e1>] ? restore_all+0xf/0xf
> [  137.494137]  [<c128925f>] ? __this_cpu_preempt_check+0xf/0x20
> [  137.494143]  [<c106ea49>] ? trace_hardirqs_on_caller+0xe9/0x220
> [  137.494148]  [<c14669ae>] syscall_call+0x7/0x7
> ...
> [  137.496322] Showing all locks held in the system:
> [  137.496332] 2 locks held by kworker/0:2/274:
> [  137.496334]  #0:  ("events_freezable_power_efficient"){.+.+.+}, at: [<c104d043>] process_one_work+0x103/0x390
> [  137.496347]  #1:  ((&(&ev->dwork)->work)){+.+.+.}, at: [<c104d043>] process_one_work+0x103/0x390
> [  137.496360] 2 locks held by agetty/1395:
> [  137.496363]  #0:  (&tty->ldisc_sem){++++++}, at: [<c12d8c83>] tty_ldisc_ref_wait+0x13/0x40
> [  137.496374]  #1:  (&ldata->atomic_read_lock){+.+...}, at: [<c12d6c04>] n_tty_read+0x84/0x920
> [  137.496386] 3 locks held by bash/1416:
> [  137.496388]  #0:  (sb_writers#5){.+.+.+}, at: [<c10f4282>] vfs_write+0x152/0x1a0
> [  137.496399]  #1:  (rcu_read_lock){......}, at: [<c12dc9d0>] __handle_sysrq+0x0/0x150
> [  137.496409]  #2:  (tasklist_lock){.+.+..}, at: [<c106d636>] debug_show_all_locks+0x36/0x1b0
> [  137.496420] 4 locks held by rmmod/1434:
> [  137.496423]  #0:  (&dev->mutex){......}, at: [<c131bcd3>] device_release_driver+0x13/0x30
> [  137.496433]  #1:  (&shost->scan_mutex){+.+.+.}, at: [<c132c714>] scsi_remove_host+0x14/0xf0
> [  137.496443]  #2:  (&dev->mutex){......}, at: [<c131bcd3>] device_release_driver+0x13/0x30
> [  137.496453]  #3:  (&ev->block_mutex){+.+...}, at: [<c126ad42>] disk_block_events+0x22/0x80
> [  137.496464]
> [  137.496467] =============================================
> [  137.496467]
> [  164.035571] sd 3:0:0:0: [sdh] Synchronizing SCSI cache

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists