dma
commented
7 years ago Just writing down some notes on this from testing some time ago:
-
Using zero-copy for samples in PA didn't work in Oz, and some light (i.e. strace) analysis revealed that the PA client code was trying to open a path in /dev/shm that wasn't in the sandbox (/dev/shm is created per-sandbox), and I attributed the failure to that..
-
If fds for open shared memory objects are passed through PA socket, why is there a need to open objects on the filesystem at specific paths?
-
If zero copy is used for audio chunks, then it seems quite feasible that we can filter the PA control packets
This is an exposure we've been aware of for a while. Creating an issue to track it for work in coming weeks.
Some useful info:
https://gavv.github.io/blog/pulseaudio-under-the-hood/