thanks to Carsten Weinhold for finding and fixing this:

pthread locks are _not_ fair (shocking news for me), so audio out has a
problem: the audio out loop holds the driver_lock almost always (since the
write() to the driver takes most of the time); other threads who want that
lock (like the frontend trying to change the volume) can starve;
the problem shows on kernel 2.6 with NPTL, but the potential is there in
all pthread implementations

CVS patchset: 5938
CVS date: 2003/12/24 13:06:10

0 files changed, 0 insertions, 0 deletions