On Fri, Apr 26, 2013 at 8:22 PM, Tejun Heo <tj@kernel.org> wrote:
> On Fri, Apr 26, 2013 at 07:40:20PM +0200, Sedat Dilek wrote:
>> Oops, NULL-pointer-deref [ __queue_work() ]
>>
>> [   25.974932] BUG: unable to handle kernel NULL pointer dereference
>> at 0000000000000100
>> [   25.974944] IP: [<ffffffff81077502>] __queue_work+0x32/0x3d0
>
> So, 0x100 deref near the top of the function.
>
> ...
>> [   25.975037] RIP: 0010:[<ffffffff81077502>]  [<ffffffff81077502>]
>> __queue_work+0x32/0x3d0
>> [   25.975047] RSP: 0018:ffff88008fed5c48  EFLAGS: 00010046
>> [   25.975052] RAX: 0000000000000096 RBX: 0000000000000292 RCX: 0000000000000000
>> [   25.975058] RDX: ffff880095281850 RSI: 0000000000000000 RDI: 0000000000000100
>> [   25.975063] RBP: ffff88008fed5c88 R08: 0000000000000000 R09: 0000000000000300
>> [   25.975069] R10: ffff880094981a00 R11: 0000000000000000 R12: ffff880095281850
>> [   25.975074] R13: 0000000000000000 R14: 0000000000000100 R15: 00000000000009c4
>> [   25.975081] FS:  00007f2f61707740(0000) GS:ffff88011fac0000(0000)
>> knlGS:0000000000000000
>> [   25.975088] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>> [   25.975093] CR2: 0000000000000100 CR3: 000000009101f000 CR4: 00000000000407e0
>> [   25.975099] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>> [   25.975104] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
> ...
>> [   25.975143] Call Trace:
>> [   25.975151]  [<ffffffff81077be5>] queue_work_on+0x45/0x50
>> [   25.975165]  [<ffffffffa016e8ff>] hci_req_run+0xbf/0xf0 [bluetooth]
>> [   25.975188]  [<ffffffffa016ea06>] __hci_req_sync+0xd6/0x1c0 [bluetooth]
>> [   25.975217]  [<ffffffffa016fad5>] hci_dev_open+0x275/0x2e0 [bluetooth]
>> [   25.975230]  [<ffffffffa0182752>] hci_sock_ioctl+0x1f2/0x3f0 [bluetooth]
>> [   25.975238]  [<ffffffff815c6050>] sock_do_ioctl+0x30/0x70
>> [   25.975245]  [<ffffffff815c75f9>] sock_ioctl+0x79/0x2f0
>> [   25.975254]  [<ffffffff811a8046>] do_vfs_ioctl+0x96/0x560
>> [   25.975262]  [<ffffffff811a85a1>] SyS_ioctl+0x91/0xb0
>> [   25.975271]  [<ffffffff816d989d>] system_call_fastpath+0x1a/0x1f
>> [   25.975276] Code: 89 e5 41 57 41 56 41 89 fe 41 55 49 89 f5 41 54
>> 49 89 d4 53 48 83 ec 18 89 7d c8 9c 58 66 66 90 66 90 f6 c4 02 0f 85
>> 56 02 00 00 <41> 8b 85 00 01 00 00 a9 00 00 01 00 0f 85 b0 02 00 00 48
>> c7 c2
>
> All code
> ========
>    0:   89 e5                   mov    %esp,%ebp
>    2:   41 57                   push   %r15
>    4:   41 56                   push   %r14
>    6:   41 89 fe                mov    %edi,%r14d
>    9:   41 55                   push   %r13
>    b:   49 89 f5                mov    %rsi,%r13
>    e:   41 54                   push   %r12
>   10:   49 89 d4                mov    %rdx,%r12
>   13:   53                      push   %rbx
>   14:   48 83 ec 18             sub    $0x18,%rsp
>   18:   89 7d c8                mov    %edi,-0x38(%rbp)
>   1b:   9c                      pushfq
>   1c:   58                      pop    %rax
>   1d:   66 66 90                data32 xchg %ax,%ax
>   20:   66 90                   xchg   %ax,%ax
>   22:   f6 c4 02                test   $0x2,%ah
>   25:   0f 85 56 02 00 00       jne    0x281
>   2b:*  41 8b 85 00 01 00 00    mov    0x100(%r13),%eax     <-- trapping instruction
>   32:   a9 00 00 01 00          test   $0x10000,%eax
>   37:   0f 85 b0 02 00 00       jne    0x2ed
>   3d:   48                      rex.W
>   3e:   c7                      .byte 0xc7
>   3f:   c2                      .byte 0xc2
>
> The second argument %rsi is zero, which got transferred to %r13 and
> then offset deref on it trapped.
>
> The second argument is @wq and the oopsing code is the wq->flags deref
> in the following if condition.
>
>         /* if dying, only works from the same workqueue are allowed */
>         if (unlikely(wq->flags & __WQ_DRAINING) &&
>                 WARN_ON_ONCE(!is_chained_work(wq)))
>                 return;
>
> So, umm, don't pass in NULL as @wq. :)
>

[ CC Frederic (linux-dynticks) ]

Great, Tejun!
Anyway a bug...

Just wanted to mention I switched to a full-cpu-dynticks config-setup:

1. TICK_CPU_ACCOUNTING -> VIRT_CPU_ACCOUNTING_GEN

2. [X] NO_HZ_FULL

>From [2]:

 config VIRT_CPU_ACCOUNTING_GEN
bool "Full dynticks CPU time accounting"
- depends on HAVE_CONTEXT_TRACKING && 64BIT
+ depends on HAVE_CONTEXT_TRACKING && 64BIT && NO_HZ_FULL

Choosing NO_HZ_FULL depends leads to a different kernel-config which
seems not to show the trace.

- Sedat -

[1] http://git.kernel.org/cgit/linux/kernel/git/frederic/linux-dynticks.git/log/?h=timers/nohz
[2] http://git.kernel.org/cgit/linux/kernel/git/frederic/linux-dynticks.git/commit/?h=timers/nohz&id=7f40072a53838380e3902d94fae49efed506b34e

> --
> tejun