[bardliao]

The regcache_sync() function actually do nothing since regcache_mark_dirty() is not called in the existing code. We call regcache_mark_dirty() in rt711_sdca_io_init(), but rt711_sdca_io_init() will not be invoked since rt711->hw_init is true.

Fixes: #2908

[RanderWang]

LGTM

[shumingfan]

@bardliao LGTM. Thanks.

[bardliao]

Update commit message.

[plbossart]

Good finding @bardliao but see alternative suggestion below.

[plbossart]

that's not the right fix, we need to extend the existing one with this

	if (rt711->first_hw_init) {
		regcache_cache_bypass(rt711->regmap, false);
		regcache_mark_dirty(rt711->regmap);
>>		regcache_cache_bypass(rt711->regmap_mbq, false);
>>		regcache_mark_dirty(rt711->regmap_mbq);
	} else
		rt711->first_hw_init = true;

[bardliao]

that's not the right fix, we need to extend the existing one with this

	if (rt711->first_hw_init) {
		regcache_cache_bypass(rt711->regmap, false);
		regcache_mark_dirty(rt711->regmap);
>>		regcache_cache_bypass(rt711->regmap_mbq, false);
>>		regcache_mark_dirty(rt711->regmap_mbq);
	} else
		rt711->first_hw_init = true;

I did it in the next patch. I think we need both because there are two scenarios.

1. Codec is unattached. In this case we need to do rt711_sdca_io_init
2. Common suspend/resume case. In this case we don't need to do rt711_sdca_io_init, but need to sync some registers since they will be restore to default when we do bus reset.

[plbossart]

@bardliao I think we are not aligned on the io_init case. This function will be used when we reattach, and there's no difference between the codec resetting on its own and the bus reset. The codec becoming un-attached can be for different reasons with different side effects, but there is nothing that tells us what the source of the reset was. It's a miss in the SoundWire spec and as a result software has to assume that all the state was lost every time the device re-attaches.

[bardliao]

@plbossart So the problem is actually sdw_update_slave_status(SDW_SLAVE_UNATTACHED) is never triggered in the reattach case. See dmesg.txt with bard.diff.txt. Therefore, rt711->hw_init = false; and rt711_sdca_io_init will not be invoked
I will look into it.

[bardliao]

@plbossart I got why sdw_update_slave_status(SDW_SLAVE_UNATTACHED) is never triggered now. We call sdw_modify_slave_status without sdw_update_slave_status in sdw_clear_slave_status. As a result, sdw_update_slave_status(SDW_SLAVE_UNATTACHED) will never be invoked elsewhere due to the
status is not changed. Please see #2928.
However, with #2928 and this PR updated, I see

[   76.370376] rt711-sdca sdw:0:025d:0711:01: DPN_PortCtrl register write failed for port 3
[   76.370380] soundwire sdw-master-0: Program transport params failed: -61
[   76.370382] soundwire sdw-master-0: Program params failed: -61

dmesg.txt
I think it could be something wrong with rt711_sdca_io_init @shumingfan FYI

[plbossart]

the patches look good @bardliao but they both need a Fixes tag (from Mark Brown's tree) so that they are applied to the stable kernel and propagated to distros. Thanks!

[bardliao]

the patches look good @bardliao but they both need a Fixes tag (from Mark Brown's tree) so that they are applied to the stable kernel and propagated to distros. Thanks!

Done

[plbossart]

thanks @bardliao