On Thu, 2013-03-14 at 22:07 +0200, Tanu Kaskinen wrote: > I investigated a bug in module-loopback reported[1] by Fr?d?ric > Dalleau last year. The instructions for reproduction were as follows: > > 1. Have one sound card (index 0) loaded with one sink (index 0) and > one source (index 1), plus the monitor source (index 0) > > 2. pactl load-module module-loopback sink=0 source=0 > > 3. pactl set-card-profile 0 off > > These steps caused reliable crashing, and after fixing the first > issue, two more cropped up. The first three patches fix these three > issues. When I investigated in more detail the exact conditions that > caused the first crash, I found out that module-alsa-card was doing > something that it shouldn't do: it was moving streams. The last patch > fixes that. > > Removing the stream moving from module-alsa-card causes a change in > behavior that perhaps isn't acceptable: if a sink is removed due to > a profile change and a new sink is created, the streams connected to > the old sink aren't necessarily moved to the new sink. If this is > unacceptable, the old policy should be reimplemented somewhere else, > perhaps in module-rescue-streams. > > [1] http://thread.gmane.org/gmane.comp.audio.pulseaudio.general/13408 > > Tanu Kaskinen (4): > loopback: Fix segfault in may_move_to() callbacks > filter-apply: Fix segfault with moving streams > loopback: Flush asyncmsgq from the right context > alsa: Don't move streams when changing profiles > > src/modules/alsa/module-alsa-card.c | 19 ------------ > src/modules/module-filter-apply.c | 10 +++++-- > src/modules/module-loopback.c | 55 ++++++++++++++++++++++++----------- > 3 files changed, 45 insertions(+), 39 deletions(-) No feedback received, other than that the fourth patch should not go in before the removed policy code is reimplemented elsewhere. I've pushed the first three patches. -- Tanu
