Backend SEGV under TVRec::TuningSignalCheck

[martin@…]

Backend is normally completely stable, but r9247 has crashed 6 times in the last 24 hours. Backend is idle at the time (apart from processing EIT from UK DVB).

Backtrace attached, and last few minutes of log from one crash (log was record,channel,siparser).

[martin@…]

mythbackend backtrace

[martin@…]

Last few minutes of log

[martin@…]

This is another of the backtraces. Looks quite different though. End of log was full of lines with "SIParser: NIT Present on this transport on PID 0x0"

[martin@…]

Added trace from another of the crashes (gdb-ex2.txt), though looks different. This died with:

Program terminated with signal 11, Segmentation fault.
#0  0x00002aaaab327c74 in ProgramAssociationTable::ProgramNumber (this=0x4aa0e8f0, i=4756403) at mpeg/mpegtables.h:421
421             { return (psipdata()[(i<<2)] << 8) | psipdata()[(i<<2) + 1]; }

[danielk]

[invalid]

The first backtrace is for a memory error, I need you do run valgrind to find where the error actually occurs.

The second backtrace is for a completely different problem and should have it's own ticket. Before you create another ticket though, I will need a matching "mythbackend -v siparser,channel,record" log with the backtrace. This looks like it is having problems with the reported length of the pes packet being inaccurate.

[Martin Ebourne <martin@…>]

first run of valgrind

[Martin Ebourne <martin@…>]

Added a run with valgrind. It stopped after 7 mins due to a pure virtual method call. I'll run it again, see if it has anything else.

[Martin Ebourne <martin@…>]

record,channel,siparser from first run of valgrind

[Martin Ebourne <martin@…>]

Valgrind output from second run

[Martin Ebourne <martin@…>]

Backend log from second valgrind run

[Martin Ebourne <martin@…>]

I've added a second valgrind run. Doesn't look too helpful though, and eventually valgrind fails on an assert. Also it's not clear if the pure virtual call from the first run was in myth or valgrind itself. I'm on x86_64 which is relatively new to valgrind, so it may have bugs.

[Janne <janne-mythtv@…>]

please test current svn with fixes from #1459 and #1493 if the tickets are open or the pathes aren't applied to svn.

[danielk]

no reply from reporter about whether the fixes in SVN worked, assuming they did.