Bug 373307 – media-sound/rosegarden-11.06 using media-sound/jack-audio-connection-kit-0.120.2 likely to crash with std::bad_alloc due to mlockall call

• Product: Gentoo Linux
• Status: RESOLVED INVALID

[Description Martin von Gagern 2011-06-28 09:40:01 UTC]

When starting rosegarden with a midi file passed on the command line, it's likely to die again soon with the following message:

terminate called after throwing an instance of 'std::bad_alloc'
  what():  std::bad_alloc

As far as I understand things, this is because
1. I have a memlock limit of 500.000 KiB, as suggested by
   http://en.gentoo-wiki.com/wiki/JACK#Kernel_Configuration
2. jack_start_thread in the libjack client library will do a call to mlockall,
   thereby locking ALL pages which rosegarden has mapped into memory.
3. Rosegarden has a LOT of libs mapped, like qt4, kde stuff, and so on.
   So it reaches the limit rather easily. I guess most functions in those libs
   are never used at all, so having them memlocked is a waste of memory.

The problem occurred only some of the time, not always. I guess when things worked, that's because the vsize had already exceeded 500MB by the time the jack thread was started. In that case, jack does fall back to a non-memlocked operation.

I guess there are several options how to deal with this:
1. Issue an ewarn, instructing users that rosegarden might require a larger
   memlock limit than the Gentoo wiki suggests.
2. Adjust rosegarden to prevent all memlocking. Perhaps based on a USE flag.
3. Adjust rosegarden in such a way that only the relevant areas are actually
   memlocked. This might be a bit hard, but would be the best solution in the
   long run. Probably a task for upstream.

[Comment 1 Martin von Gagern 2011-06-28 20:45:55 UTC]

OK, I found out that the -m argument to jackd does solve the issue as well, although the -u argument does not. I had looked for such options in the rosegarden settings, but hadn't expected the jack daemon to control this aspect of client behaviour. If you add an elog (ewarn was a mistake) message, you might suggest the -m flag instead of or as an alternative to the memlock limit.

It looks like -u should come close to solution 3, but otoh it looks as though it doesn't work as it should. Haven't looked at the code accompanying -u yet.

If you consider all of this as a mere configuration issue that rosegarden users should know how to deal with, then close this as WONTFIX. In that case I'd write a forum article about it instead.