>>> In fact it's a solution for such firmware hasing ?low hw_ptr updating rate. >>> Pulseaudio use hw_unused field to track the unused buffer, the real overrun >>> means the "avail" space is geting close to hwbuf_size. I think this patch has adverse effects for regular hardware. In the case where the hardware only updates the hw_ptr at every period and the specified latency is lower than a period, indeed the driver will either see zero or more bytes than requested. This is interpreted wrongly as an overflow. The additional test only reports an overflow if the number of bytes is smaller than the ring buffer and thus avoids any watermark increase. Let me point out this is a workaround, the system still does not meet the latency specified as it can only capture data with a granularity equal to a period. In addition for regular hardware, the fact that more data is available means the system does not meet the latency requirements. By adding this test, the watermark will never be increased to the point where the latency matches ALSA actual buffering. There is no real solution other than fixing the firmware if indeed low-latency is a requirement. Hope this helps, -Pierre
