[linux-audio-user] about 2.6.x woes [nptl?]

[Jack O'Quin]

Florian Schmidt <mista.tapas at gmx.net> writes:

> well, i stumbled across this thread on the jackit-devel mailing list
> about trying to set LD_ASSUME_KERNEL=2.4.22 and run all jack stuff
> [server and clients with this env var set] and it worked
> wonders.. 2*512 works very nice again..

Yes, this works fine for me.  I can even run 2.6.7 with realtime-lsm
at 2*32 (using a delta-66).  It gets quite a few xruns at that
setting, but will run all night without crashing.  This is as good I
get with 2.4-lowlat.

Fernando has observed additional problems on Fedora Core 2, which are
not fixed by setting this environment variable.  He developed a JACK
patch to hack around some additional bugs in pthread_create().  These
have not yet been integrated into JACK CVS, but should be soon.
Because of these problems, CCRMA currently does not recommend using
FC2.

> i suspect this is nptl related (see the jackit-devel thread for
> details:
> http://sourceforge.net/mailarchive/message.php?msg_id=8629219)..  i
> think the LD_ASSUME_KERNEL switch somehow makes ld.so link to another
> libc or something w/o nptl? anyone knows details?

Similar problems have been observed with 2.6 kernels even without
NPTL.  The environment variable seems to force use of the 2.4-style
kernel interfaces, instead.

> so i just wanted to urge all the users of 2.6.x and who experience
> strange weird non rt operation with jack to try it. maybe it
> helps...
> 
> i wrote myself a bash script called "realrealtime" and put that in my
> path:
> 
> LD_ASSUME_KERNEL=2.4.22 $@
> 
> and use that to start al my jack stuff :) 

That works.  

Or, you can just set it globally in your ~/.profile (or wherever you
choose to put things like that).

> Any ideas about why jackd behaves so weird in the first place
> appreciated..

I don't think anyone fully understands the problem, yet.  It *seems*
to be somewhere in the interaction between the kernel and the pthreads
library.
-- 
  joq