Re: [net] 9317bb6982: INFO: task cat-kmsg:893 blocked for more than 300 seconds.

[Huang, Ying <ying.huang@intel.com>]

[-- Attachment #1: Type: text/plain, Size: 3660 bytes --]

"Huang, Ying" <ying.huang@intel.com> writes:

> Eric Dumazet <edumazet@google.com> writes:
>> On Mon, May 9, 2016 at 6:26 PM, Huang, Ying <ying.huang@linux.intel.com> wrote:
>>> Hi, Eric,
>>>
>>> kernel test robot <ying.huang@linux.intel.com> writes:
>>>> FYI, we noticed the following commit:
>>>>
>>>> git://internal_merge_and_test_tree devel-catchup-201604281529
>>>> commit 9317bb69824ec8d078b0b786b6971aedb0af3d4f ("net: SOCKWQ_ASYNC_NOSPACE optimizations")
>>>>
>>>> on test machine: vm-kbuild-2G: 2 threads qemu-system-x86_64 -enable-kvm -cpu Haswell,+smep,+smap with 2G memory
>>>>
>>>> caused below changes:
>>>>
>>>>
>>>> +--------------------------------------------------+------------+------------+
>>>> |                                                  | 210732d16d | 9317bb6982 |
>>>> +--------------------------------------------------+------------+------------+
>>>> | boot_successes                                   | 40         | 13         |
>>>> | boot_failures                                    | 0          | 27         |
>>>> | INFO:task_blocked_for_more_than#seconds          | 0          | 27         |
>>>> | RIP:native_safe_halt                             | 0          | 20         |
>>>> | RIP:native_write_msr_safe                        | 0          | 27         |
>>>> | Kernel_panic-not_syncing:hung_task:blocked_tasks | 0          | 27         |
>>>> | backtrace:__close_fd                             | 0          | 27         |
>>>> | backtrace:SyS_close                              | 0          | 27         |
>>>> | backtrace:cpu_startup_entry                      | 0          | 19         |
>>>> | backtrace:watchdog                               | 0          | 27         |
>>>> | RIP:__lock_acquire                               | 0          | 2          |
>>>> | backtrace:rpc_async_schedule                     | 0          | 2          |
>>>> | backtrace:lock_acquire                           | 0          | 1          |
>>>> | RIP:delay_tsc                                    | 0          | 1          |
>>>> | backtrace:SYSC_epoll_wait                        | 0          | 1          |
>>>> | backtrace:SyS_epoll_wait                         | 0          | 1          |
>>>> | RIP:pvclock_clocksource_read                     | 0          | 1          |
>>>> | RIP:xs_reclassify_socket                         | 0          | 1          |
>>>> | backtrace:xs_tcp_setup_socket                    | 0          | 2          |
>>>> | RIP:insert_work                                  | 0          | 1          |
>>>> +--------------------------------------------------+------------+------------+
>>>
>>> We recently found this patch cause NFS hang in 0day/LKP test system.
>>> The NFS export can be mounted, but after a while all read/write to NFS
>>> mount blocked.  This influenced the 0day/LKP testing.  Could you help us
>>> to fix this?
>>>
>>> Best Regards,
>>> Huang, Ying
>>
>>
>> I need to officially submit this patch :
>> http://www.spinics.net/lists/netdev/msg375777.html
>
> Thanks a lot!  The patch fixed our NFS hang.
>
> Tested-by: Huang, Ying <ying.huang@intel.com>
>
> Is it possible to put the fixing patch near (preferably next) the patch
> trigger regression?  Otherwise a bigger range of patches will not be
> bisectable.

Hi, Eric,

Is it possible for this to be fixed in linux-next ASAP?  This has
blocked us (0day functionality and performance test) from testing
linux-next and many other git trees for near 1 week.  Because NFS is
used in our test infrastructure to save test result data.

Best Regards,
Huang, Ying