Bard Liao

> Humm very interesting. > > the first two commits of PR #5074 are ace-2.x specific so that would point to the last patch with the deferred workqueue. > >...

@plbossart Apply the "soundwire: cadence: re-check Peripheral status with delayed_work" commit can reproduce the issue.

@plbossart Yes, `[ 33.214013] soundwire_cadence:cdns_check_attached_status_dwork: soundwire_intel soundwire_intel.link.0: Peripheral 6 status: 2` And `sdw_handle_slave_alerts` is called twice. ``` [ 33.214031] rt711-sdca sdw:0:0:025d:0711:01: bard: sdw_handle_slave_alerts [ 33.294601] rt711-sdca sdw:0:0:025d:0711:01: bard: sdw_handle_slave_alerts ```...

@plbossart Potential fix: https://github.com/thesofproject/linux/pull/5095

> the scenario that worries me is > > a) an alert occurs b) sdw_cdns_irq() detects the alert and schedules the workqueue c) the dwork is scheduled and it cancels...

> These parity errors have been noising test results for 2 weeks now. As usual, this will cause other, unrelated regressions to go unnoticed. Could there be some temporary workaround/revert...

@plbossart Device number has been assigned. And there is no trace in rt715_sdca_io_init(). I think it is probably why we don't see the next action in the log. ``` [...

> I saw that confusion @bardliao but if the enumeration didn't complete by the deadline, then the initialization would not complete either? Fair point. Indeed, I can't find `soundwire sdw-master-0-1:...

> The mechanism for interrupts is that we have ONE interrupt, and then we deal with all the links with this loop > > ``` > list_for_each_entry(link, &ctx->link_list, list) >...

> @bardliao I am not sure this theory is correct, if it is that would mean the solution we implemented years ago was never robust... For LNL we start the...