Skip to contentSkip to conversation

ASoC: soc-pcm: Use conditional PCM hardware parameter initialization

RFC In review Peter Ujfalusi 1 participant 2 messages Oct 20, 2025 – Oct 21, 2025

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

RFC 1/1 ASoC: soc-pcm: Use conditional PCM hardware parameter initialization

Component drivers can prepare snd_pcm_hardware struct based on the hardware capabilities which information should not be discarded.

Only touch the rates, channels_max and formats if they were left to 0, otherwise keep the provided configuration intact for the parameter cross checking decision.

---
Hi,

this patch in essence extends the special casing of formats done by 083a25b18d6a ("ASoC: soc-pcm: fix hw->formats cleared by soc_pcm_hw_init() for dpcm")

Other parameters might have been set in the same way as the formats and preserving these are equally important.

A case for this is SOF used with HDA codec (analog or more importantly, HDMI) where the hw-> params are set based on the connected display/device and should be preserved so we can report correct rate, format and channels supported by the equipment.

If the hw-> parameters are left uninitialized then we still need to set the UINT/ULLONG_MAX for the refining code to work.

This applies only for FE setup, in other cases we shall (as before) do a full re-initialization.

I think this makes sense and I cannot think where this might flop, but sent as RFC to see what people think.

Br,
Peter

 sound/soc/soc-pcm.c | 39 ++++++++++++++++++++++++++-------------
 1 file changed, 26 insertions(+), 13 deletions(-)
sound/soc/soc-pcm.c +26 -13
index 2c21fd528afd..e4bbcf4bcc9d 100644
@@ -570,14 +570,31 @@ static void soc_pcm_apply_msb(struct snd_pcm_substream *substream)
 	soc_pcm_set_msb(substream, cpu_bits);
 }
 
-static void soc_pcm_hw_init(struct snd_pcm_hardware *hw)
+static void soc_pcm_hw_init(struct snd_pcm_hardware *hw, bool preserve_config)
 {
-	hw->rates		= UINT_MAX;
-	hw->rate_min		= 0;
-	hw->rate_max		= UINT_MAX;
-	hw->channels_min	= 0;
-	hw->channels_max	= UINT_MAX;
-	hw->formats		= ULLONG_MAX;
+	if (preserve_config) {
+		/*
+		 * preserve the configuration which might be done by components
+		 * Note: if the rates/rate_max/channels_max/formats are left to
+		 * 0 we still need to initialize them for the parameter updates
+		 * to work
+		 */
+		if (!hw->rates)
+			hw->rates = UINT_MAX;
+		if (!hw->rate_max)
+			hw->rate_max = UINT_MAX;
+		if (!hw->channels_max)
+			hw->channels_max = UINT_MAX;
+		if (!hw->formats)
+			hw->formats = ULLONG_MAX;
+	} else {
+		hw->rates = UINT_MAX;
+		hw->rate_min = 0;
+		hw->rate_max = UINT_MAX;
+		hw->channels_min = 0;
+		hw->channels_max = UINT_MAX;
+		hw->formats = ULLONG_MAX;
+	}
 }
 
 static void soc_pcm_hw_update_rate(struct snd_pcm_hardware *hw,
@@ -626,7 +643,7 @@ int snd_soc_runtime_calc_hw(struct snd_soc_pcm_runtime *rtd,
 	unsigned int cpu_chan_min = 0, cpu_chan_max = UINT_MAX;
 	int i;
 
-	soc_pcm_hw_init(hw);
+	soc_pcm_hw_init(hw, false);
 
 	/* first calculate min/max only for CPUs in the DAI link */
 	for_each_rtd_cpu_dais(rtd, i, cpu_dai) {
@@ -1738,13 +1755,9 @@ static void dpcm_runtime_setup_fe(struct snd_pcm_substream *substream)
 	struct snd_pcm_hardware *hw = &runtime->hw;
 	struct snd_soc_dai *dai;
 	int stream = substream->stream;
-	u64 formats = hw->formats;
 	int i;
 
-	soc_pcm_hw_init(hw);
-
-	if (formats)
-		hw->formats &= formats;
+	soc_pcm_hw_init(hw, true);
 
 	for_each_rtd_cpu_dais(fe, i, dai) {
 		const struct snd_soc_pcm_stream *cpu_stream;
Signature · 1 line
2.51.1.dirty
#1 lore raw
Quoted 92 lines · Peter Ujfalusi

Component drivers can prepare snd_pcm_hardware struct based on the hardware capabilities which information should not be discarded.

Only touch the rates, channels_max and formats if they were left to 0, otherwise keep the provided configuration intact for the parameter cross checking decision.

---
Hi,

this patch in essence extends the special casing of formats done by 083a25b18d6a ("ASoC: soc-pcm: fix hw->formats cleared by soc_pcm_hw_init() for dpcm")

Other parameters might have been set in the same way as the formats and preserving these are equally important.

A case for this is SOF used with HDA codec (analog or more importantly, HDMI) where the hw-> params are set based on the connected display/device and should be preserved so we can report correct rate, format and channels supported by the equipment.

If the hw-> parameters are left uninitialized then we still need to set the UINT/ULLONG_MAX for the refining code to work.

This applies only for FE setup, in other cases we shall (as before) do a full re-initialization.

I think this makes sense and I cannot think where this might flop, but sent as RFC to see what people think.

Br,
Peter

 sound/soc/soc-pcm.c | 39 ++++++++++++++++++++++++++-------------
 1 file changed, 26 insertions(+), 13 deletions(-)
sound/soc/soc-pcm.c +25 -8
index 2c21fd528afd..e4bbcf4bcc9d 100644
@@ -570,14 +570,31 @@ static void soc_pcm_apply_msb(struct snd_pcm_substream *substream)
 	soc_pcm_set_msb(substream, cpu_bits);
 }
 
-static void soc_pcm_hw_init(struct snd_pcm_hardware *hw)
+static void soc_pcm_hw_init(struct snd_pcm_hardware *hw, bool preserve_config)
 {
-	hw->rates		= UINT_MAX;
-	hw->rate_min		= 0;
-	hw->rate_max		= UINT_MAX;
-	hw->channels_min	= 0;
-	hw->channels_max	= UINT_MAX;
-	hw->formats		= ULLONG_MAX;
+	if (preserve_config) {
+		/*
+		 * preserve the configuration which might be done by components
+		 * Note: if the rates/rate_max/channels_max/formats are left to
+		 * 0 we still need to initialize them for the parameter updates
+		 * to work
+		 */
+		if (!hw->rates)
+			hw->rates = UINT_MAX;
+		if (!hw->rate_max)
+			hw->rate_max = UINT_MAX;
+		if (!hw->channels_max)
+			hw->channels_max = UINT_MAX;
+		if (!hw->formats)
+			hw->formats = ULLONG_MAX;
+	} else {
+		hw->rates = UINT_MAX;
+		hw->rate_min = 0;
+		hw->rate_max = UINT_MAX;
+		hw->channels_min = 0;
+		hw->channels_max = UINT_MAX;
+		hw->formats = ULLONG_MAX;
+	}
 }
 
 static void soc_pcm_hw_update_rate(struct snd_pcm_hardware *hw,
@@ -626,7 +643,7 @@ int snd_soc_runtime_calc_hw(struct snd_soc_pcm_runtime *rtd,
 	unsigned int cpu_chan_min = 0, cpu_chan_max = UINT_MAX;
 	int i;
 
-	soc_pcm_hw_init(hw);
+	soc_pcm_hw_init(hw, false);
 
 	/* first calculate min/max only for CPUs in the DAI link */
 	for_each_rtd_cpu_dais(rtd, i, cpu_dai) {
@@ -1738,13 +1755,9 @@ static void dpcm_runtime_setup_fe(struct snd_pcm_substream *substream)
 	struct snd_pcm_hardware *hw = &runtime->hw;

sound/soc/soc-pcm.c static void dpcm_runtime_setup_fe(struct snd_pcm_substream *substream) · line 1760

Quoted 10 lines
sound/soc/soc-pcm.c +1 -5
 	struct snd_soc_dai *dai;
 	int stream = substream->stream;
-	u64 formats = hw->formats;
 	int i;
 
-	soc_pcm_hw_init(hw);
-
-	if (formats)
-		hw->formats &= formats;
+	soc_pcm_hw_init(hw, true);

Instead of changing the soc_pcm_hw_init() it might be better to move the relevant implementation inline here. I will do this for the patch.

Quoted 3 lines
sound/soc/soc-pcm.c +0 -0
 	for_each_rtd_cpu_dais(fe, i, dai) {
 		const struct snd_soc_pcm_stream *cpu_stream;
Signature · 1 line
Péter