[ltt-dev] Trace "destroy" hangs, lttd keeps running

[Mathieu Desnoyers]

* Ashwin Tanugula (ashwin.tanugula at broadcom.com) wrote:
>  
> > OK. Can you please try with this patch applied on top of 0.176 ? This 
> > should be very close to mips64 which, AFAIK, should work and does not 
> > need this workaround to deal with 64-bit values. Sorry it's taking so 
> > long. It would help if I could get my hands on a MIPS32 board. More 
> > comments after the patch.
> 
> > [snip snip]
> 
> > I compared the mips code to the approach I took for intel (they were quite similar) and found the discrepancy. Please ignore the previous patch I just sent, and try this one instead:
> 
> 
> Mathieu,
> 
> Even this patch does not fix the issue.
> 

Well, it does seem to fix the trace clock hang issue on mips32, right ?
So we're getting there...

> 
> [Merging emails]
> 
> 
> > > 
> > > With "return 1;" I can sometimes generate the trace and sometime I cannot as lttd is still running even after destroying the trace.
> > > 
> > > Please look at the attached log file.
> > >
> 
> > running lttd in verbose mode could eventually help here. (see lttd --help). You would have to start it separately from lttctl though (see lttctl --help to see how to create a trace, then start lttd, and then start tracing with lttctl).
> 
> 
> Running lttd in verbose mode didn't help either.
> 
> Here'swhat I did
> 
>   lttctl -c trace2
>   lttd/lttd -t /tmp/trace2 -c /proc/sys/debug/ltt/trace2 -N 1 -d -n -v
>   lttctl -s trace2
> 
> To stop the trace 
> 
>   lttctl -D trace2
> 
> 
> lttd still is running at this point for traces longer than 5 minutes. 
> 

It would be useful to have the output of what lttd is doing after you
issue "lttctl -D trace2".

Hrm, and you might want to remove the "-d" parameter for lttd. I'm not
sure it keeps the verbose output on the console.

Is the system very busy while you take your trace ? I suppose the mips32
board you have is single-cpu, am I correct ?

Also please make sure to double-check the lttd version you are using
(e.g. not a binary left over from an old build). This is especially
important because the newer lttng-lttv-lttctl-lttd versions break
compatibility with older versions.

> I can generate and view the trace data for smaller traces (1 min) without any problems.
> 

Great, at least this tells us that the trace clock now works. I'm having
trouble reproducing your issue on my x86-64 machine though.

Thanks,

Mathieu


> 
> > > i) ERROR:tracefile.c:1676:ltt_seek_next_event: assertion failed: (tf->event.offset < tf->buffer.data_size)
> > > ./lttv-gui: line 10: 12602 Aborted                 (core dumped) $LTTV_CMD.real -m lttvwindow -m guievents -m guifilter -m guicontrolflow -m resourceview -m guistatistics -m guitracecontrol $*
> > >
> > > 
> 
> > This is a bit weird. Make sure you do a make clean of your lttv tree and rebuild. Also, compiling lttv with:
> 
> > ./configure CFLAGS=-g
> > make
> > make install
> 
> > and running:
> 
> > gdb lttv.real
> > set args -v -d -m textDump -t /tmp/trace run
> 
> > and showing the stack trace with:
> 
> > bt full
> 
> > would help here. (btw, lttv -m textDump --help   shows you the textDump help information)
> 
> > You could also put the trace somewhere I can access it, so I can look at it with lttv on my local machine.
> 
> 
> I was not able to reproduce this error.
> 
> Thanks,
> Ashwin

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68