The idle_bias_off flag is not configured for DAIs not mapped with a codec. This causes the pm counter to be increased at probe time for the CPU dai which unbalances the pm counter handling.

Signed-off-by: Peter Ujfalusi peter.ujfalusi@ti.com ---

Hello Mark,

I'm not sure if this is the correct way to fix the problem but this patch works for me.

The issue in short: Since commit: '18d7564 ASoC: when removing a CPU DAI, clean up its DAPM context' After booting up the kernel on BeagleBoard for example I see that the clocks are enabled for the McBSP CPU dai. This alone causing the first stream start to fail due to unbalanced pm_runtime counter. Later it stabilizes, but I see failures with full duplex operation (arecord | aplay). The root cause of this is not commit 18d7564, but the fact that we did not configured the dapm->idle_bias_off for the CPU dai to be 1 (we do this for the platform driver). The pm_runtime_get_sync is called from dapm_pre_sequence_async() - since we add the cpu_dai dapm context to the list.

With this patch we set the dapm->idle_bias_off for DAIs not linked with a codec which means cpu_dais. If this is not acceptable we need to set this flag in the cpu_dai driver itself.

Regards, Peter

sound/soc/soc-core.c | 6 ++++++ 1 files changed, 6 insertions(+), 0 deletions(-)

diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c index f81c597..f10f00b 100644 --- a/sound/soc/soc-core.c +++ b/sound/soc/soc-core.c @@ -3715,6 +3715,9 @@ int snd_soc_register_dai(struct device *dev, } }

+ if (!dai->codec) + dai->dapm.idle_bias_off = 1; + list_add(&dai->list, &dai_list);

mutex_unlock(&client_mutex); @@ -3803,6 +3806,9 @@ int snd_soc_register_dais(struct device *dev, } }

+ if (!dai->codec) + dai->dapm.idle_bias_off = 1; + list_add(&dai->list, &dai_list);

mutex_unlock(&client_mutex);