Skip to contentSkip to conversation

ASoC: wm8985: rework and fix the clock calculation

PATCH In review Petr Kulhavy 2 participants 5 messages May 12, 2016

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ASoC: wm8985: rework and fix the clock calculation

The clock calculation has several issues:
1) if PLL is used in master mode the BCLK output runs at double the speed
2) de-facto only 44.1kHz and 48kHz sampling rates are supported, other rates like 8kHz, 12kHz, 24kHz fail to find the proper BCLK divider
3) the wm8985->sysclk variable has a misleading name and is used wrongly in the clock calculation in wm8985_hw_params() which is the root cause for (1)
4) wm8985->bclk is used only in wm8985_hw_params() and therefore no need to store it in the wm8985_priv structure

Therefore the clock calculation is rewritten in more clean and proper way:
- move wm8985_priv->bclk as a local variable into mw8985_hw_params()
- new variable wm8985_priv->pllout holds the actual frequency that is input to the MCLKDIV post-divider
- move wm8985_priv->sysclk as a local variable into mw8985_hw_params()
- sysclk is now always calculated as 256 * fs
- the MCLKDIV is looked up as pllout/sysclk
- fs_ratios[] is replaced by simpler mclk_divs[] lookup table

With this patch all rates: 8, 11.025, 12, 16, 22.05, 24, 32, 44.1 and 48kHz work properly and generate the correct BCLK.

---
 sound/soc/codecs/wm8985.c | 59 ++++++++++++++++++++++++-----------------------
 1 file changed, 30 insertions(+), 29 deletions(-)
sound/soc/codecs/wm8985.c +30 -29
index 18f2babe1090..628127aa3c96 100644
@@ -181,22 +181,11 @@ static const int volume_update_regs[] = {
 struct wm8985_priv {
 	struct regmap *regmap;
 	struct regulator_bulk_data supplies[WM8985_NUM_SUPPLIES];
-	unsigned int sysclk;
-	unsigned int bclk;
+	unsigned int pllout;	/* input rate to the MCLKDIV divider */
 };
 
-static const struct {
-	int div;
-	int ratio;
-} fs_ratios[] = {
-	{ 10, 128 },
-	{ 15, 192 },
-	{ 20, 256 },
-	{ 30, 384 },
-	{ 40, 512 },
-	{ 60, 768 },
-	{ 80, 1024 },
-	{ 120, 1536 }
+static const int mclk_divs[] = {
+	10, 15, 20, 30, 40, 60, 80, 120
 };
 
 static const int srates[] = { 48000, 32000, 24000, 16000, 12000, 8000 };
@@ -693,15 +682,18 @@ static int wm8985_hw_params(struct snd_pcm_substream *substream,
 	struct snd_soc_codec *codec;
 	struct wm8985_priv *wm8985;
 	u16 blen, srate_idx;
-	unsigned int tmp;
 	int srate_best;
+	unsigned int bclk;	/* bit clock matching the current params */
+	unsigned int sysclk;	/* the actual sysclk after post division */
 
 	codec = dai->codec;
 	wm8985 = snd_soc_codec_get_drvdata(codec);
 
-	wm8985->bclk = snd_soc_params_to_bclk(params);
-	if ((int)wm8985->bclk < 0)
-		return wm8985->bclk;
+	/* always use 256 * fs in order to get best filter quality */
+	sysclk = 256 * params_rate(params);
+	bclk = snd_soc_params_to_bclk(params);
+	if ((int)bclk < 0)
+		return bclk;
 
 	switch (params_format(params)) {
 	case SNDRV_PCM_FORMAT_S16_LE:
@@ -742,29 +734,28 @@ static int wm8985_hw_params(struct snd_pcm_substream *substream,
 	snd_soc_update_bits(codec, WM8985_ADDITIONAL_CONTROL,
 			    WM8985_SR_MASK, srate_idx << WM8985_SR_SHIFT);
 
-	dev_dbg(dai->dev, "Target BCLK = %uHz\n", wm8985->bclk);
-	dev_dbg(dai->dev, "SYSCLK = %uHz\n", wm8985->sysclk);
+	dev_dbg(dai->dev, "Target BCLK = %uHz\n", bclk);
+	dev_dbg(dai->dev, "SYSCLK = %uHz\n", sysclk);
 
-	for (i = 0; i < ARRAY_SIZE(fs_ratios); ++i) {
-		if (wm8985->sysclk / params_rate(params)
-				== fs_ratios[i].ratio)
+	/* select the appropriate mclk divider */
+	for (i = 0; i < ARRAY_SIZE(mclk_divs); ++i) {
+		if (wm8985->pllout / mclk_divs[i] * 10
+				== sysclk)
 			break;
 	}
 
-	if (i == ARRAY_SIZE(fs_ratios)) {
+	if (i == ARRAY_SIZE(mclk_divs)) {
 		dev_err(dai->dev, "Unable to configure MCLK ratio %u/%u\n",
-			wm8985->sysclk, params_rate(params));
+			wm8985->pllout, sysclk);
 		return -EINVAL;
 	}
 
-	dev_dbg(dai->dev, "MCLK ratio = %dfs\n", fs_ratios[i].ratio);
 	snd_soc_update_bits(codec, WM8985_CLOCK_GEN_CONTROL,
 			    WM8985_MCLKDIV_MASK, i << WM8985_MCLKDIV_SHIFT);
 
 	/* select the appropriate bclk divider */
-	tmp = (wm8985->sysclk / fs_ratios[i].div) * 10;
 	for (i = 0; i < ARRAY_SIZE(bclk_divs); ++i) {
-		if (wm8985->bclk == tmp / bclk_divs[i])
+		if (bclk == sysclk / bclk_divs[i])
 			break;
 	}
 
@@ -786,6 +777,10 @@ struct pll_div {
 };
 
 #define FIXED_PLL_SIZE ((1ULL << 24) * 10)
+/*
+ * source = MCLK input of the chip
+ * target = the f2 coming out of the PLL before /4 divider
+ */
 static int pll_factors(struct pll_div *pll_div, unsigned int target,
 		       unsigned int source)
 {
@@ -872,17 +867,23 @@ static int wm8985_set_sysclk(struct snd_soc_dai *dai,
 				    WM8985_CLKSEL_MASK, 0);
 		snd_soc_update_bits(codec, WM8985_POWER_MANAGEMENT_1,
 				    WM8985_PLLEN_MASK, 0);
+		wm8985->pllout = freq;
 		break;
 	case WM8985_CLKSRC_PLL:
 		snd_soc_update_bits(codec, WM8985_CLOCK_GEN_CONTROL,
 				    WM8985_CLKSEL_MASK, WM8985_CLKSEL);
+		/*
+		 * in order to run the PLL within the recommended 90MHz
+		 * operating range the wm8985_set_pll() configures the PLL
+		 * to output double the required frequency
+		 */
+		wm8985->pllout = 2 * freq;
 		break;
 	default:
 		dev_err(dai->dev, "Unknown clock source %d\n", clk_id);
 		return -EINVAL;
 	}
 
-	wm8985->sysclk = freq;
 	return 0;
 }
 
Signature · 1 line
1.9.1
#1 lore raw
Quoted 4 lines · Petr Kulhavy

The clock calculation has several issues:
1) if PLL is used in master mode the BCLK output runs at double the speed
2) de-facto only 44.1kHz and 48kHz sampling rates are supported, other rates like 8kHz, 12kHz, 24kHz fail to find the proper BCLK divider

These certainly seem like things that need fixed :-)

Quoted 60 lines

3) the wm8985->sysclk variable has a misleading name and is used wrongly in the clock calculation in wm8985_hw_params() which is the root cause for (1)
4) wm8985->bclk is used only in wm8985_hw_params() and therefore no need to store it in the wm8985_priv structure

Therefore the clock calculation is rewritten in more clean and proper way:
- move wm8985_priv->bclk as a local variable into mw8985_hw_params()
- new variable wm8985_priv->pllout holds the actual frequency that is input to the MCLKDIV post-divider
- move wm8985_priv->sysclk as a local variable into mw8985_hw_params()
- sysclk is now always calculated as 256 * fs
- the MCLKDIV is looked up as pllout/sysclk
- fs_ratios[] is replaced by simpler mclk_divs[] lookup table

With this patch all rates: 8, 11.025, 12, 16, 22.05, 24, 32, 44.1 and 48kHz work properly and generate the correct BCLK.

---
 sound/soc/codecs/wm8985.c | 59 ++++++++++++++++++++++++-----------------------
 1 file changed, 30 insertions(+), 29 deletions(-)
sound/soc/codecs/wm8985.c +4 -15
index 18f2babe1090..628127aa3c96 100644
@@ -181,22 +181,11 @@ static const int volume_update_regs[] = {
 struct wm8985_priv {
 	struct regmap *regmap;
 	struct regulator_bulk_data supplies[WM8985_NUM_SUPPLIES];
-	unsigned int sysclk;
-	unsigned int bclk;
+	unsigned int pllout;	/* input rate to the MCLKDIV divider */
 };
 
-static const struct {
-	int div;
-	int ratio;
-} fs_ratios[] = {
-	{ 10, 128 },
-	{ 15, 192 },
-	{ 20, 256 },
-	{ 30, 384 },
-	{ 40, 512 },
-	{ 60, 768 },
-	{ 80, 1024 },
-	{ 120, 1536 }
+static const int mclk_divs[] = {
+	10, 15, 20, 30, 40, 60, 80, 120
 };
 
 static const int srates[] = { 48000, 32000, 24000, 16000, 12000, 8000 };
@@ -693,15 +682,18 @@ static int wm8985_hw_params(struct snd_pcm_substream *substream,
 	struct snd_soc_codec *codec;
 	struct wm8985_priv *wm8985;
 	u16 blen, srate_idx;
-	unsigned int tmp;
 	int srate_best;
+	unsigned int bclk;	/* bit clock matching the current params */

sound/soc/codecs/wm8985.c static int wm8985_hw_params(struct snd_pcm_substream *substream, · line 693

Quoted 10 lines
sound/soc/codecs/wm8985.c +3 -3
+	unsigned int sysclk;	/* the actual sysclk after post division */
 
 	codec = dai->codec;
 	wm8985 = snd_soc_codec_get_drvdata(codec);
 
-	wm8985->bclk = snd_soc_params_to_bclk(params);
-	if ((int)wm8985->bclk < 0)
-		return wm8985->bclk;
+	/* always use 256 * fs in order to get best filter quality */
+	sysclk = 256 * params_rate(params);

Seems quite bold to assume we always want to use 256*fs, I am not super familiar with the part itself but are there good reasons to do that? What if someone wanted to use say a direct MCLK and a lower multiple?

The intent presumably here is that set_sysclk is where we set the target SYSCLK and set_pll should set the PLL output then here we probably should work out the correct MCLKDIV to get those.

Quoted 61 lines
sound/soc/codecs/wm8985.c +17 -10
+	bclk = snd_soc_params_to_bclk(params);
+	if ((int)bclk < 0)
+		return bclk;
 
 	switch (params_format(params)) {
 	case SNDRV_PCM_FORMAT_S16_LE:
@@ -742,29 +734,28 @@ static int wm8985_hw_params(struct snd_pcm_substream *substream,
 	snd_soc_update_bits(codec, WM8985_ADDITIONAL_CONTROL,
 			    WM8985_SR_MASK, srate_idx << WM8985_SR_SHIFT);
 
-	dev_dbg(dai->dev, "Target BCLK = %uHz\n", wm8985->bclk);
-	dev_dbg(dai->dev, "SYSCLK = %uHz\n", wm8985->sysclk);
+	dev_dbg(dai->dev, "Target BCLK = %uHz\n", bclk);
+	dev_dbg(dai->dev, "SYSCLK = %uHz\n", sysclk);
 
-	for (i = 0; i < ARRAY_SIZE(fs_ratios); ++i) {
-		if (wm8985->sysclk / params_rate(params)
-				== fs_ratios[i].ratio)
+	/* select the appropriate mclk divider */
+	for (i = 0; i < ARRAY_SIZE(mclk_divs); ++i) {
+		if (wm8985->pllout / mclk_divs[i] * 10
+				== sysclk)
 			break;
 	}
 
-	if (i == ARRAY_SIZE(fs_ratios)) {
+	if (i == ARRAY_SIZE(mclk_divs)) {
 		dev_err(dai->dev, "Unable to configure MCLK ratio %u/%u\n",
-			wm8985->sysclk, params_rate(params));
+			wm8985->pllout, sysclk);
 		return -EINVAL;
 	}
 
-	dev_dbg(dai->dev, "MCLK ratio = %dfs\n", fs_ratios[i].ratio);
 	snd_soc_update_bits(codec, WM8985_CLOCK_GEN_CONTROL,
 			    WM8985_MCLKDIV_MASK, i << WM8985_MCLKDIV_SHIFT);
 
 	/* select the appropriate bclk divider */
-	tmp = (wm8985->sysclk / fs_ratios[i].div) * 10;
 	for (i = 0; i < ARRAY_SIZE(bclk_divs); ++i) {
-		if (wm8985->bclk == tmp / bclk_divs[i])
+		if (bclk == sysclk / bclk_divs[i])
 			break;
 	}
 
@@ -786,6 +777,10 @@ struct pll_div {
 };
 
 #define FIXED_PLL_SIZE ((1ULL << 24) * 10)
+/*
+ * source = MCLK input of the chip
+ * target = the f2 coming out of the PLL before /4 divider
+ */
 static int pll_factors(struct pll_div *pll_div, unsigned int target,
 		       unsigned int source)
 {
@@ -872,17 +867,23 @@ static int wm8985_set_sysclk(struct snd_soc_dai *dai,
 				    WM8985_CLKSEL_MASK, 0);
 		snd_soc_update_bits(codec, WM8985_POWER_MANAGEMENT_1,
 				    WM8985_PLLEN_MASK, 0);
+		wm8985->pllout = freq;

sound/soc/codecs/wm8985.c static int wm8985_set_sysclk(struct snd_soc_dai *dai, · line 880

Quoted 10 lines
sound/soc/codecs/wm8985.c +6 -0
 		break;
 	case WM8985_CLKSRC_PLL:
 		snd_soc_update_bits(codec, WM8985_CLOCK_GEN_CONTROL,
 				    WM8985_CLKSEL_MASK, WM8985_CLKSEL);
+		/*
+		 * in order to run the PLL within the recommended 90MHz
+		 * operating range the wm8985_set_pll() configures the PLL
+		 * to output double the required frequency
+		 */
+		wm8985->pllout = 2 * freq;

Were does this *2 come from? Is this the PLLPRESCALE?

Quoted 12 lines
sound/soc/codecs/wm8985.c +0 -1
 		break;
 	default:
 		dev_err(dai->dev, "Unknown clock source %d\n", clk_id);
 		return -EINVAL;
 	}
 
-	wm8985->sysclk = freq;
 	return 0;
 }
 
Signature · 1 line
1.9.1

Thanks,
Charles

#2 lore raw
Quoted 23 lines · Charles Keepax

On Thu, May 12, 2016 at 08:48:54AM +0200, Petr Kulhavy wrote:

sound/soc/codecs/wm8985.c +4 -4
@@ -693,15 +682,18 @@ static int wm8985_hw_params(struct snd_pcm_substream *substream,
  	struct snd_soc_codec *codec;
  	struct wm8985_priv *wm8985;
  	u16 blen, srate_idx;
-	unsigned int tmp;
  	int srate_best;
+	unsigned int bclk;	/* bit clock matching the current params */
+	unsigned int sysclk;	/* the actual sysclk after post division */
 
  	codec = dai->codec;
  	wm8985 = snd_soc_codec_get_drvdata(codec);
 
-	wm8985->bclk = snd_soc_params_to_bclk(params);
-	if ((int)wm8985->bclk < 0)
-		return wm8985->bclk;
+	/* always use 256 * fs in order to get best filter quality */
+	sysclk = 256 * params_rate(params);

Seems quite bold to assume we always want to use 256*fs, I am not super familiar with the part itself but are there good reasons to do that? What if someone wanted to use say a direct MCLK and a lower multiple?

Quoted 3 lines

The intent presumably here is that set_sysclk is where we set the target SYSCLK and set_pll should set the PLL output then here we probably should work out the correct MCLKDIV to get those.

As I understand the datasheet the codec is basically designed for 256*fs clock:
DAC and ADC expect it, so do the cutoffs for the digital filters, ALC attack and delay times. Also all the timing diagrams are specified at 256fs clock.

Quoted 11 lines
Quoted 11 lines

static int pll_factors(struct pll_div *pll_div, unsigned int target,
unsigned int source)
{

sound/soc/codecs/wm8985.c +1 -0
@@ -872,17 +867,23 @@ static int wm8985_set_sysclk(struct snd_soc_dai *dai,
  				    WM8985_CLKSEL_MASK, 0);
  		snd_soc_update_bits(codec, WM8985_POWER_MANAGEMENT_1,
  				    WM8985_PLLEN_MASK, 0);
+		wm8985->pllout = freq;
  		break;
  	case WM8985_CLKSRC_PLL:
  		snd_soc_update_bits(codec, WM8985_CLOCK_GEN_CONTROL,
Quoted 8 lines

sound/soc/codecs/wm8985.c static int wm8985_set_sysclk(struct snd_soc_dai *dai, · line 880

Quoted 7 lines
sound/soc/codecs/wm8985.c +6 -0
  				    WM8985_CLKSEL_MASK, WM8985_CLKSEL);
+		/*
+		 * in order to run the PLL within the recommended 90MHz
+		 * operating range the wm8985_set_pll() configures the PLL
+		 * to output double the required frequency
+		 */
+		wm8985->pllout = 2 * freq;

Were does this *2 come from? Is this the PLLPRESCALE?

It comes from the following line in wm8985_set_pll() :

                 ret = pll_factors(&pll_div, freq_out * 4 * 2, freq_in);

This formula is taken over from the datasheet (the section Master Clock and PLL, calculation of the f2) and is correct.
It attempts to get the PLL to run at approx. 90MHz where it performs the best.
But since there is only f/4 fixed post-divider after the PLL (see the Figure 40 in the wm8758 or Figure 38 in the wm8985 datasheet) the F_pllout is effectively freq*2.

Petr

#3 lore raw
Quoted 24 lines · Petr Kulhavy
Quoted 21 lines · Charles Keepax

On Thu, May 12, 2016 at 08:48:54AM +0200, Petr Kulhavy wrote:

sound/soc/codecs/wm8985.c +4 -4
@@ -693,15 +682,18 @@ static int wm8985_hw_params(struct snd_pcm_substream *substream,
 	struct snd_soc_codec *codec;
 	struct wm8985_priv *wm8985;
 	u16 blen, srate_idx;
-	unsigned int tmp;
 	int srate_best;
+	unsigned int bclk;	/* bit clock matching the current params */
+	unsigned int sysclk;	/* the actual sysclk after post division */
 	codec = dai->codec;
 	wm8985 = snd_soc_codec_get_drvdata(codec);
-	wm8985->bclk = snd_soc_params_to_bclk(params);
-	if ((int)wm8985->bclk < 0)
-		return wm8985->bclk;
+	/* always use 256 * fs in order to get best filter quality */
+	sysclk = 256 * params_rate(params);

Seems quite bold to assume we always want to use 256*fs, I am not super familiar with the part itself but are there good reasons to do that? What if someone wanted to use say a direct MCLK and a lower multiple?

Quoted 8 lines
Quoted 3 lines

The intent presumably here is that set_sysclk is where we set the target SYSCLK and set_pll should set the PLL output then here we probably should work out the correct MCLKDIV to get those.

As I understand the datasheet the codec is basically designed for 256*fs clock:
DAC and ADC expect it, so do the cutoffs for the digital filters, ALC attack and delay times.
Also all the timing diagrams are specified at 256fs clock.

Ok that seems pretty conclusive that we should always have this at 256fs.

Quoted 22 lines
Quoted 11 lines
Quoted 11 lines

static int pll_factors(struct pll_div *pll_div, unsigned int target,

sound/soc/codecs/wm8985.c +1 -0
 		       unsigned int source)
 {
@@ -872,17 +867,23 @@ static int wm8985_set_sysclk(struct snd_soc_dai *dai,
 				    WM8985_CLKSEL_MASK, 0);
 		snd_soc_update_bits(codec, WM8985_POWER_MANAGEMENT_1,
 				    WM8985_PLLEN_MASK, 0);
+		wm8985->pllout = freq;
 		break;
 	case WM8985_CLKSRC_PLL:
 		snd_soc_update_bits(codec, WM8985_CLOCK_GEN_CONTROL,
Quoted 8 lines

sound/soc/codecs/wm8985.c static int wm8985_set_sysclk(struct snd_soc_dai *dai, · line 880

Quoted 7 lines
sound/soc/codecs/wm8985.c +6 -0
 				    WM8985_CLKSEL_MASK, WM8985_CLKSEL);
+		/*
+		 * in order to run the PLL within the recommended 90MHz
+		 * operating range the wm8985_set_pll() configures the PLL
+		 * to output double the required frequency
+		 */
+		wm8985->pllout = 2 * freq;

Were does this *2 come from? Is this the PLLPRESCALE?

It comes from the following line in wm8985_set_pll() :

                 ret = pll_factors(&pll_div, freq_out * 4 * 2, freq_in);
Quoted 7 lines

This formula is taken over from the datasheet (the section Master Clock and PLL, calculation of the f2) and is correct.
It attempts to get the PLL to run at approx. 90MHz where it performs the best.
But since there is only f/4 fixed post-divider after the PLL (see the Figure 40 in the wm8758 or Figure 38 in the wm8985 datasheet) the F_pllout is effectively freq*2.

Ok that also makes sense.

I will have another quick look over the patch.

Thanks,
Charles

Reviewed-by on v1 1/1
#4 lore raw
In reply to Petr Kulhavy
Quoted 19 lines · Petr Kulhavy

The clock calculation has several issues:
1) if PLL is used in master mode the BCLK output runs at double the speed
2) de-facto only 44.1kHz and 48kHz sampling rates are supported, other rates like 8kHz, 12kHz, 24kHz fail to find the proper BCLK divider
3) the wm8985->sysclk variable has a misleading name and is used wrongly in the clock calculation in wm8985_hw_params() which is the root cause for (1)
4) wm8985->bclk is used only in wm8985_hw_params() and therefore no need to store it in the wm8985_priv structure

Therefore the clock calculation is rewritten in more clean and proper way:
- move wm8985_priv->bclk as a local variable into mw8985_hw_params()
- new variable wm8985_priv->pllout holds the actual frequency that is input to the MCLKDIV post-divider
- move wm8985_priv->sysclk as a local variable into mw8985_hw_params()
- sysclk is now always calculated as 256 * fs
- the MCLKDIV is looked up as pllout/sysclk
- fs_ratios[] is replaced by simpler mclk_divs[] lookup table

Quoted 5 lines

With this patch all rates: 8, 11.025, 12, 16, 22.05, 24, 32, 44.1 and 48kHz work properly and generate the correct BCLK.

---

Thanks,
Charles