'slot get frame fail'/'dropped pkts as no slot' errors

[ZeroDawn]

Hi,

During our overnight tests we noticed some dropped frames were occurring sporadically. Checking the Rx logs, we saw some errors which correspond to the dropped frames:

The tests in question were recording 5 or 6 1080p50 streams simultaneously, so using a fair amount of cpu generally (60-80%). Those errors occurred across all streams at the same time. Other tests with similar metrics did pass.

Just wondering if you have any thoughts on what might be causing these? Let me know if I can provide anymore information.

Thanks,
Matt

[Sakoram]

This appears to be application-side back-pressure: MTL is producing frames faster than your app is releasing them. Two things to try:

Increase framebuff_cnt in the session ops to give the producer more headroom.
Ensure your app has enough CPU budget to handle every frame on time — keep the notify_frame_ready callback non-blocking and offload heavy work to a worker thread; make sure st20_rx_put_framebuff() is called promptly.
I've also improved the diagnostics in #1589. Once that lands, MTL will emit a unified warn line whenever it detects this condition, e.g.:

Warn: RX_VIDEO_SESSION(0,0): back-pressure: framebuff pool empty (0/3 free),
dropped 5 frames / 12 pkts in interval; raise framebuff_cnt or drain via st20_rx_put_framebuff

A (sustained Nx) suffix is appended once the condition persists for 3+ stat intervals, so transient spikes are easy to distinguish from a stuck consumer. Equivalent log lines are added for ST30 audio and ST40 ancillary sessions.