Hello Ralf,

Booting 3.0.1 kernel with ROSE patches in SMP mode gives systematically
the following Inconsistent Lock State.
See attached file.

Bernard


Le 08/08/2011 16:06, Ralf Baechle a écrit :
> On Mon, Aug 08, 2011 at 03:40:48PM +0200, f6bvp wrote:
>
>> Hello Ralf,
>>
>> Since my last post my two ROSE/FPAC nodes have been running
>> flawlessly with last ROSE patches for nearly 17 days.
>> Situation remains the same, i.e. from time to time use can still be
>> negative.
>>
>> Here is a /proc/net/rose/neigh dump from the machine running dual
>> core CPU with SMP kernel :
>>
>> /proc/net/rose_neigh
>> addr  callsign  dev  count use mode restart  t0  tf digipeaters
>> 00005 F6BVP-5   ax2      1   0  DTE      no   0   0
>> 00004 F8COJ-11  ax2      2  -1  DTE      yes   0   0
>> 00003 F6BVP-11  ax1     11   0  DTE      no   0   0
>> 00002 K4GBB-12  ax2      9   0  DCE     yes   0   0
>> 00001 RSLOOP-0  ???      1   2  DCE     yes   0   0
>>
>> For the present time "use" parameter has returned to 0 for F8COJ-11
>> node neighbor on the other system without
>> manual change.
>> Thus I guess some situations can reinitialize "use" counter.
>> I will then investigate if use can also become negative on this
>> second system if I do not switch to UMP
>> mode at boot (removing maxcpus=1 boot option).
>>
>> /proc/net/rose_neigh
>> addr  callsign  dev  count use mode restart  t0  tf digipeaters
>> 00020 F6BVP-5   ax0      1   0  DTE      no   0   0
>> 00019 F6BVP-7   ax0      2   0  DTE      no   0   0
>> 00018 F6BVP-9   ax2      2   0  DTE      no   0   0
>> 00017 F3KT-11   ax0      3   0  DCE     yes   0   0
>> 00016 F8COJ-11  ax0      2   0  DCE     yes   0   0
>> 00015 F6GGY-9   ax0      3   0  DTE      no   0   0
>> ....
>> 00002 TI2HAS-9  ax0      5   0  DTE      no   0   0
>> 00001 RSLOOP-0  ???      1   0  DCE     yes   0   0
> This at least means there should be no regressions by my patch series and
> it's getting time to send it upstream in a few days when I sorted out my
> b0rked DSL line and new workstation.  With these holes plugged finding
> what's actually causing the use counters to go negative should also have
> gotten a little easier.
>
> Thanks,
>
>    Ralf