[ltt-dev] Problem with tracing network events in 0.128 to 0.132

[Gregory Haskins]

Mathieu Desnoyers wrote:
> * Gregory Haskins (ghaskins at novell.com) wrote:
>   
>> Mathieu Desnoyers wrote:
>>     
>>> * Gregory Haskins (ghaskins at novell.com) wrote:
>>>   
>>>       
>>>> I am willing to dig in and see if I can fix it, but I figured I would
>>>> post first to see if this is a known issue, has a workaround, or perhaps
>>>> some guiding wisdom on what to look for first.
>>>>
>>>>     
>>>>         
>>> Hi Greg,
>>>
>>> It should be fixed in LTTng 0.133. Can you confirm ?
>>>   
>>>       
>> Hi Mathieu,
>>   I confirmed that the ethernet IRQ is now showing up.  Thanks!  Note,
>> however, that the softirq-net-rx (softirq-3) is still notably missing. 
>> Is there a similar issue on the softirq side?  Note that I do see some
>> of the softirq vectors (1, 6, and 8 specifically), so its not a general
>> problem.
>>
>>     
>
> Hi Greg,
>
> Hrm, which events are missing for softirq-net-rx exactly ? only softirq
> raise or also softirq entry/exit ?
>
> I know I need to update my softirq raise instrumentation. Should be done
> soon, but I wonder about entry/exit.
>   

Hmm...not sure, but I will check later.

>   
>> As an aside, note that ive been seeing a problem in lttv-gui for a while
>> now where the interrupts/softirqs can sometimes get "stuck".  (e.g. the
>> color bar goes to infinity even if there are no events).  I suspect that
>> part of the problem is perhaps overlapping events when you allow things
>> like interrupts to preempt each other.  As an additional datapoint, I do
>> not see the problem when I turn on lockdep, which disables overlapping
>> interrupts.
>>     
>
> Hrm, in which view do you have this problem ? The per-thread control
> flow view or the ressource view ?
>   

The RV view.  I noticed that if you get something like irq_entry(x)
irq_entry(y), irq_exit(y), irq_exit(x), its like the last event gets
lost and (I believe) x will stay "lit" forever in the RV window (tho it
could be y, will have to confirm).  Turning on lockdep makes that
nesting impossible and Ive noted that the issue seems to goes away in
this config.  Thus my theory about the nesting.  I can post a screen
shot if it will help.


-Greg

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 266 bytes
Desc: OpenPGP digital signature
URL: <http://lists.casi.polymtl.ca/pipermail/lttng-dev/attachments/20090505/fda227ec/attachment-0003.pgp>