ASoC: soc-dai: pull out be_hw_params_fixup from snd_soc_dai_hw_params
A message that is not in the archive (sent off-list or not yet received): CGME20200511033107epcas2p2465bc53295b38d2723c4bc351da77c5e@epcas2p2.samsung.com
PATCH 1/1 ASoC: soc-dai: pull out be_hw_params_fixup from snd_soc_dai_hw_params
When dpcm_be_dai_hw_params() called, be_hw_params_fixup() callback is called with same arguments 3times.
It is called in be_hw_params_fixup() itself and in soc_pcm_hw_params() for cpu dai and codec dai.
Tested in 5.4.
--- sound/soc/soc-dai.c | 12 ------------ sound/soc/soc-dapm.c | 11 +++++++++++ 2 files changed, 11 insertions(+), 12 deletions(-)
sound/soc/soc-dai.c +0 -12
@@ -257,20 +257,8 @@ int snd_soc_dai_hw_params(struct snd_soc_dai *dai,struct snd_pcm_substream *substream, struct snd_pcm_hw_params *params) { - struct snd_soc_pcm_runtime *rtd = substream->private_data; int ret; -/* perform any topology hw_params fixups before DAI */ -if (rtd->dai_link->be_hw_params_fixup) { - ret = rtd->dai_link->be_hw_params_fixup (rtd, params); -if (ret <0 ) { -dev_err (rtd->dev, -"ASoC: hw_params topology fixup failed %d\n" , - ret); -return ret; - } - } -if (dai->driver->ops->hw_params) { ret = dai->driver->ops->hw_params (substream, params, dai);if (ret <0 ) {
sound/soc/soc-dapm.c +11 -0
@@ -3886,6 +3886,17 @@ snd_soc_dai_link_event_pre_pmu(struct snd_soc_dapm_widget *w,hw_param_interval (params,SNDRV_PCM_HW_PARAM_CHANNELS )->max = config->channels_max; +/* perform any topology hw_params fixups before DAI */ +if (rtd->dai_link->be_hw_params_fixup) { + ret = rtd->dai_link->be_hw_params_fixup (rtd, params); +if (ret <0 ) { +dev_err (rtd->dev, +"ASoC: hw_params topology fixup failed %d\n" , + ret); +return ret; + } + } + substream->stream =SNDRV_PCM_STREAM_CAPTURE ;snd_soc_dapm_widget_for_each_source_path (w, path) { source = path->source->priv;
base-commit: f3643491bd079c973ac6c693da7966cd17506ca3
Signature · 1 line
2.21.0
Quoted 25 lines · Gyeongtaek Lee
When
dpcm_be_dai_hw_params()called,be_hw_params_fixup()callback is called with same arguments 3times.
It is called inbe_hw_params_fixup()itself and insoc_pcm_hw_params()for cpu dai and codec dai.
Tested in 5.4.Signed-off-by Gyeongtaek Lee <[email protected]>--- sound/soc/soc-dai.c | 12 ------------ sound/soc/soc-dapm.c | 11 +++++++++++ 2 files changed, 11 insertions(+), 12 deletions(-)sound/soc/soc-dai.c +0
-2@@ -257,20 +257,8 @@ int snd_soc_dai_hw_params(struct snd_soc_dai *dai,struct snd_pcm_substream *substream, struct snd_pcm_hw_params *params) { - struct snd_soc_pcm_runtime *rtd = substream->private_data; int ret; -/* perform any topology hw_params fixups before DAI */
sound/soc/soc-dai.c int snd_soc_dai_hw_params(struct snd_soc_dai *dai, · line 273
Quoted 10 lines
sound/soc/soc-dai.c +0
-10-if (rtd->dai_link->be_hw_params_fixup) { - ret = rtd->dai_link->be_hw_params_fixup (rtd, params); -if (ret <0 ) { -dev_err (rtd->dev, -"ASoC: hw_params topology fixup failed %d\n" , - ret); -return ret; - } - } -
Sorry I don't get this change.
If the be_hw_params_fixup() callback is called three times, it's because the soc_soc_dai_hw_params() routine is called three times, so what is the problem here?
Also the comment is explicit about doing fixups before calling the dai hw_params() callback, so that is not longer the case with this change? Even if the change was legit, the comment is no longer relevant and should be updated.
Quoted 28 lines
sound/soc/soc-dai.c +0
-0if (dai->driver->ops->hw_params) { ret = dai->driver->ops->hw_params (substream, params, dai);if (ret <0 ) {sound/soc/soc-dapm.c +11
-0@@ -3886,6 +3886,17 @@ snd_soc_dai_link_event_pre_pmu(struct snd_soc_dapm_widget *w,hw_param_interval (params,SNDRV_PCM_HW_PARAM_CHANNELS )->max = config->channels_max; +/* perform any topology hw_params fixups before DAI */ +if (rtd->dai_link->be_hw_params_fixup) { + ret = rtd->dai_link->be_hw_params_fixup (rtd, params); +if (ret <0 ) { +dev_err (rtd->dev, +"ASoC: hw_params topology fixup failed %d\n" , + ret); +return ret; + } + } + substream->stream =SNDRV_PCM_STREAM_CAPTURE ;snd_soc_dapm_widget_for_each_source_path (w, path) { source = path->source->priv;base-commit:
f3643491bd079c973ac6c693da7966cd17506ca3
Quoted 39 lines · Pierre-Louis Bossart
Quoted 25 lines · Gyeongtaek Lee
When
dpcm_be_dai_hw_params()called,be_hw_params_fixup()callback is called with same arguments 3times.
It is called inbe_hw_params_fixup()itself and insoc_pcm_hw_params()for cpu dai and codec dai.
Tested in 5.4.Signed-off-by Gyeongtaek Lee <[email protected]>--- sound/soc/soc-dai.c | 12 ------------ sound/soc/soc-dapm.c | 11 +++++++++++ 2 files changed, 11 insertions(+), 12 deletions(-)sound/soc/soc-dai.c +0
-2@@ -257,20 +257,8 @@ int snd_soc_dai_hw_params(struct snd_soc_dai *dai,struct snd_pcm_substream *substream, struct snd_pcm_hw_params *params) { - struct snd_soc_pcm_runtime *rtd = substream->private_data; int ret; -/* perform any topology hw_params fixups before DAI */ sound/soc/soc-dai.c int snd_soc_dai_hw_params(struct snd_soc_dai *dai, · line 273
Quoted 10 lines
sound/soc/soc-dai.c +0
-10-if (rtd->dai_link->be_hw_params_fixup) { - ret = rtd->dai_link->be_hw_params_fixup (rtd, params); -if (ret <0 ) { -dev_err (rtd->dev, -"ASoC: hw_params topology fixup failed %d\n" , - ret); -return ret; - } - } -Sorry I don't get this change.
Quoted 9 lines
If the
be_hw_params_fixup()callback is called three times, it's because thesoc_soc_dai_hw_params()routine is called three times, so what is the problem here?Also the comment is explicit about doing fixups before calling the dai
hw_params()callback, so that is not longer the case with this change? Even if the change was legit, the comment is no longer relevant and should be updated.
Sorry, the comment was too short and inexact to explain the intention of the patch. When dpcm_be_dai_hw_params() called, be_hw_params_fixup() is called three times with same substream and params in dpcm_be_dai_hw_params() and snd_soc_dai_hw_params() in soc_pcm_hw_params() for cpu dai and codec dai. Calling same code three times may not be a problem in most systems, but in some system which has limited computing power and changes audio routing frequently, couple of milliseconds are consumed and make the system a little bit slower to audio response.
If the be_hw_params_fixup() could be pull out from soc_soc_dai_hw_params(), and make it call once at pcm start or routing change, response time can be increased.
In my search, the only point that calls snd_soc_dai_hw_params() without calling be_hw_params_fixup() callback directly is snd_soc_dai_link_event_pre_pmu(). So, I proposed pulling out be_hw_params_fixup() from snd_soc_dai_hw_params() and then add direct call to be_hw_params_fixup() callback in snd_soc_dai_link_event_pre_pmu() not to harm current working process.
Quoted 29 lines
Quoted 28 lines
sound/soc/soc-dai.c +0
-0if (dai->driver->ops->hw_params) { ret = dai->driver->ops->hw_params (substream, params, dai);if (ret <0 ) {sound/soc/soc-dapm.c +11
-0@@ -3886,6 +3886,17 @@ snd_soc_dai_link_event_pre_pmu(struct snd_soc_dapm_widget *w,hw_param_interval (params,SNDRV_PCM_HW_PARAM_CHANNELS )->max = config->channels_max; +/* perform any topology hw_params fixups before DAI */ +if (rtd->dai_link->be_hw_params_fixup) { + ret = rtd->dai_link->be_hw_params_fixup (rtd, params); +if (ret <0 ) { +dev_err (rtd->dev, +"ASoC: hw_params topology fixup failed %d\n" , + ret); +return ret; + } + } + substream->stream =SNDRV_PCM_STREAM_CAPTURE ;snd_soc_dapm_widget_for_each_source_path (w, path) { source = path->source->priv;base-commit:
f3643491bd079c973ac6c693da7966cd17506ca3
Quoted 49 lines · Gyeongtaek Lee
Quoted 39 lines · Pierre-Louis Bossart
Quoted 25 lines · Gyeongtaek Lee
When
dpcm_be_dai_hw_params()called,be_hw_params_fixup()callback is called with same arguments 3times.
It is called inbe_hw_params_fixup()itself and insoc_pcm_hw_params()for cpu dai and codec dai.
Tested in 5.4.Signed-off-by Gyeongtaek Lee <[email protected]>--- sound/soc/soc-dai.c | 12 ------------ sound/soc/soc-dapm.c | 11 +++++++++++ 2 files changed, 11 insertions(+), 12 deletions(-)sound/soc/soc-dai.c +0
-2@@ -257,20 +257,8 @@ int snd_soc_dai_hw_params(struct snd_soc_dai *dai,struct snd_pcm_substream *substream, struct snd_pcm_hw_params *params) { - struct snd_soc_pcm_runtime *rtd = substream->private_data; int ret; -/* perform any topology hw_params fixups before DAI */ sound/soc/soc-dai.c int snd_soc_dai_hw_params(struct snd_soc_dai *dai, · line 273
Quoted 10 lines
sound/soc/soc-dai.c +0
-10-if (rtd->dai_link->be_hw_params_fixup) { - ret = rtd->dai_link->be_hw_params_fixup (rtd, params); -if (ret <0 ) { -dev_err (rtd->dev, -"ASoC: hw_params topology fixup failed %d\n" , - ret); -return ret; - } - } -Sorry I don't get this change.
Quoted 9 lines
If the
be_hw_params_fixup()callback is called three times, it's because thesoc_soc_dai_hw_params()routine is called three times, so what is the problem here?Also the comment is explicit about doing fixups before calling the dai
hw_params()callback, so that is not longer the case with this change? Even if the change was legit, the comment is no longer relevant and should be updated.
Quoted 10 lines
Sorry, the comment was too short and inexact to explain the intention of the patch. When
dpcm_be_dai_hw_params()called,be_hw_params_fixup()is called three times with same substream and params indpcm_be_dai_hw_params()andsnd_soc_dai_hw_params()insoc_pcm_hw_params()for cpu dai and codec dai. Calling same code three times may not be a problem in most systems, but in some system which has limited computing power and changes audio routing frequently, couple of milliseconds are consumed and make the system a little bit slower to audio response.
If thebe_hw_params_fixup()could be pull out fromsoc_soc_dai_hw_params(), and make it call once at pcm start or routing change, response time can be increased.
I accept the argument that be_hw_params_fixup() is called from two locations, and in the second one there is a iteration to deal with capture and playback. We should indeed check if calling the same function from two places is legit or not, you pointed to a legitimate issue. We currently use this fixup in SOF to make sure the BE configuration matches what the topology provides, and doing this match 3 times is not very useful indeed.
That said, this is supposed to be a fixup, meaning just a change to the hw_params structure with no access to hardware. I believe that even if we reduce the number of calls you are not going to see any improvement in responses time, the actual time is still going to be spend in the hw_params() itself.
Quoted 5 lines
In my search, the only point that calls
snd_soc_dai_hw_params()without callingbe_hw_params_fixup()callback directly issnd_soc_dai_link_event_pre_pmu(). So, I proposed pulling outbe_hw_params_fixup()fromsnd_soc_dai_hw_params()and then add direct call tobe_hw_params_fixup()callback insnd_soc_dai_link_event_pre_pmu()not to harm current working process.
This change leaves the calls in soc-pcm, so you still have 3 calls to the same callback - and the same odd pattern where in one case the fixup is called in a direction-agnostic manner while in the two others it is called in a loop that's direction-based.
In other words my take on this is: valid problem, but let's try to see if we can remove redundant calls rather than move one around.
Quoted 30 lines
Quoted 29 lines
Quoted 28 lines
sound/soc/soc-dai.c +0
-0if (dai->driver->ops->hw_params) { ret = dai->driver->ops->hw_params (substream, params, dai);if (ret <0 ) {sound/soc/soc-dapm.c +11
-0@@ -3886,6 +3886,17 @@ snd_soc_dai_link_event_pre_pmu(struct snd_soc_dapm_widget *w,hw_param_interval (params,SNDRV_PCM_HW_PARAM_CHANNELS )->max = config->channels_max; +/* perform any topology hw_params fixups before DAI */ +if (rtd->dai_link->be_hw_params_fixup) { + ret = rtd->dai_link->be_hw_params_fixup (rtd, params); +if (ret <0 ) { +dev_err (rtd->dev, +"ASoC: hw_params topology fixup failed %d\n" , + ret); +return ret; + } + } + substream->stream =SNDRV_PCM_STREAM_CAPTURE ;snd_soc_dapm_widget_for_each_source_path (w, path) { source = path->source->priv;base-commit:
f3643491bd079c973ac6c693da7966cd17506ca3