On Sun, 18.10.09 20:30, Ilyes Gouta (ilyes.gouta at gmail.com) wrote: > Lennart, Heya, > Does PA have the pull mode (as they call it) where the audio card > calls back the user-space layer to tell the software mixer how much > samples (I know, this is not a new science) we need to feed it in > order to avoid an underrun, accompanying the time-based, not interrupt > based, buffer submission? Yes, of course. That's how we always did it. Just use pa_stream_set_write_callback() and pass a function to it you defined that will then be called each time your app needs to write new data. It's the default mode of operation. Always has been. And timer-based scheduling we have been doing since 0.9.11. > What about the high priority thread (belonging to a special > scheduling class) that manages all of this? PA was able to make use of RT scheduling for quite some time already. Howver, handing out RT scheduling priviliges to normal users is a security issue, since that effectively allows them to freeze the machine. The rtkit daemon most distributions have now integrated works as a protection against that: it supervises user processes that got RT and makes sure they don't misuse that. The whole idea is very similar to what MS decided to do, except that rtkit is a bit less crackish I guess. ;-) Lennart -- Lennart Poettering Red Hat, Inc. lennart [at] poettering [dot] net http://0pointer.net/lennart/ GnuPG 0x1A015CC4
