dma-trace: send ipc msg after update host pointer

[kuanhsuncheng]

When I use logger to trace the sof message, there some left data in dma buffer.

I find trace_work updates host pointer after dma_copy_to_host_nowait.

but dma_copy_to_host_nowait will send ipc msg before updating the new host offset.

I think the ipc_msg_send (send the host_offset information) should be after update host pointer.

If there is anything I don't well considered, please let me know.

Thanks.

[lgirdwood]

@marc-hb would this explain the missing data we see from time to time ?

[marc-hb]

The description sounds indeed similar to "stuck traces" issue #4333.

@kuanhsuncheng can you please try the check-sof-logger.sh test after locally disabling the dma_nudge() workaround?
thesofproject/sof-test@b0b7c2d5e2fba8

Everyone please correct me but dma_copy_to_host_nowait() is not just for the trace, so moving the IPC like this would break non-trace DMA wouldn't it?

[keyonjie]

Good finding to me actually.

w.r.t the usage in the probe feature, it looks odd to me that it sends the dma_trace position to host side, haven't tracked it so might need Cezary's comment here.

[lgirdwood]

This is correct. This wont affect probes() or trace on HDA since the drivers use the LIPB to get position updates instead of an IPC.
Looks like a network error for BYT timeout in CI. https://sof-ci.01.org/sofpr/PR4763/build10333/devicetest/?model=BYT_MB_NOCODEC&testcase=verify-firmware-presence

[plbossart]

This wont affect probes() or trace on HDA since the drivers use the LIPB to get position updates instead of an IPC.

Did you mean DPIB? I've never seen LPIB being used for data transfers.

[lgirdwood]

This wont affect probes() or trace on HDA since the drivers use the LIPB to get position updates instead of an IPC.

Did you mean DPIB? I've never seen LPIB being used for data transfers.

Yes.