DVBSky V3 PCIe card: add some changes to M88DS3103 for supporting the demod of M88RS6000

M88RS6000 is the integrated chip, which includes tuner and demod.
Its internal demod is similar with M88DS3103 except some registers definition.
The main different part of this internal demod from others is its clock/pll generation IP block sitting inside the tuner die. So clock/pll functions should be configed through its tuner i2c bus, NOT its demod i2c bus.
The demod of M88RS6000 need the firmware: dvb-demod-m88rs6000.fw
firmware download link: http://www.dvbsky.net/download/linux/dvbsky-firmware.tar.gz

drivers/media/dvb-frontends/m88ds3103.c +10 -0

@@ -250,6 +251,7 @@ static int m88ds3103_set_frontend(struct dvb_frontend *fe)
  	u16 u16tmp, divide_ratio;
  	u32 tuner_frequency, target_mclk;
  	s32 s32tmp;
+	struct m88rs6000_mclk_config mclk_cfg;
 
  	dev_dbg(&priv->i2c->dev,
  			"%s: delivery_system=%d modulation=%d frequency=%d symbol_rate=%d inversion=%d pilot=%d rolloff=%d\n",

@@ -291,6 +293,26 @@ static int m88ds3103_set_frontend(struct dvb_frontend *fe)
  	if (ret)
  		goto err;
 
+	if (priv->chip_id == M88RS6000_CHIP_ID) {
+		ret = m88ds3103_wr_reg(priv, 0x06, 0xe0);
+		if (ret)
+			goto err;
+		if (fe->ops.tuner_ops.set_config) {
+			/* select main mclk */
+			mclk_cfg.config_op = 0;
+			mclk_cfg.TunerfreqMHz = c->frequency / 1000;
+			mclk_cfg.SymRateKSs = c->symbol_rate / 1000;

Quoted 28 lines

M88RS6000 is the integrated chip, which includes tuner and demod.
Its internal demod is similar with M88DS3103 except some registers definition.
The main different part of this internal demod from others is its clock/pll generation IP block sitting inside the tuner die. So clock/pll functions should be configed through its tuner i2c bus, NOT its demod i2c bus.
The demod of M88RS6000 need the firmware: dvb-demod-m88rs6000.fw
firmware download link: http://www.dvbsky.net/download/linux/dvbsky-firmware.tar.gz

drivers/media/dvb-frontends/m88ds3103.c +10 -0

@@ -250,6 +251,7 @@ static int m88ds3103_set_frontend(struct dvb_frontend *fe)
  	u16 u16tmp, divide_ratio;
  	u32 tuner_frequency, target_mclk;
  	s32 s32tmp;
+	struct m88rs6000_mclk_config mclk_cfg;
 
  	dev_dbg(&priv->i2c->dev,
  			"%s: delivery_system=%d modulation=%d frequency=%d symbol_rate=%d inversion=%d pilot=%d rolloff=%d\n",

@@ -291,6 +293,26 @@ static int m88ds3103_set_frontend(struct dvb_frontend *fe)
  	if (ret)
  		goto err;
 
+	if (priv->chip_id == M88RS6000_CHIP_ID) {
+		ret = m88ds3103_wr_reg(priv, 0x06, 0xe0);
+		if (ret)
+			goto err;
+		if (fe->ops.tuner_ops.set_config) {
+			/* select main mclk */
+			mclk_cfg.config_op = 0;
+			mclk_cfg.TunerfreqMHz = c->frequency / 1000;
+			mclk_cfg.SymRateKSs = c->symbol_rate / 1000;

drivers/media/dvb-frontends/m88ds3103.c static int m88ds3103_set_frontend(struct dvb_frontend *fe) · line 314

Quoted 10 lines

drivers/media/dvb-frontends/m88ds3103.c +10 -0

+			ret = fe->ops.tuner_ops.set_config(fe, &mclk_cfg);
+			if (ret)
+				goto err;
+			priv->mclk_khz = mclk_cfg.MclkKHz;
+		}
+		ret = m88ds3103_wr_reg(priv, 0x06, 0x00);
+		if (ret)
+			goto err;
+		usleep_range(10000, 20000);
+	}

That looks odd and also ugly. You pass some values from demod to tuner using set_config callback. Tuner driver can get symbol_rate and frequency just similarly from property cache than demod. Why you do it like that?

Clock is provided by tuner as you mention. I see you use that to pass used clock frequency from tuner to demod. This does not look nice and I would like to see clock framework instead. Or calculate clock on both drivers. Does the demod clock even needs to be changed? I think it is only TS stream size which defines used clock frequency - smaller the TS bitstream, the smaller the clock frequency needed => optimizes power consumption a little. But TS clock is calculated on tuner driver in any case?

regards
Antti

Signature · 1 line

http://palosaari.fi/

Quoted 28 lines · Nibble Max

M88RS6000 is the integrated chip, which includes tuner and demod.
Its internal demod is similar with M88DS3103 except some registers definition.
The main different part of this internal demod from others is its clock/pll generation IP block sitting inside the tuner die. So clock/pll functions should be configed through its tuner i2c bus, NOT its demod i2c bus.
The demod of M88RS6000 need the firmware: dvb-demod-m88rs6000.fw
firmware download link: http://www.dvbsky.net/download/linux/dvbsky-firmware.tar.gz

drivers/media/dvb-frontends/m88ds3103.c +10 -0

@@ -250,6 +251,7 @@ static int m88ds3103_set_frontend(struct dvb_frontend *fe)
  	u16 u16tmp, divide_ratio;
  	u32 tuner_frequency, target_mclk;
  	s32 s32tmp;
+	struct m88rs6000_mclk_config mclk_cfg;
 
  	dev_dbg(&priv->i2c->dev,
  			"%s: delivery_system=%d modulation=%d frequency=%d symbol_rate=%d inversion=%d pilot=%d rolloff=%d\n",

@@ -291,6 +293,26 @@ static int m88ds3103_set_frontend(struct dvb_frontend *fe)
  	if (ret)
  		goto err;
 
+	if (priv->chip_id == M88RS6000_CHIP_ID) {
+		ret = m88ds3103_wr_reg(priv, 0x06, 0xe0);
+		if (ret)
+			goto err;
+		if (fe->ops.tuner_ops.set_config) {
+			/* select main mclk */
+			mclk_cfg.config_op = 0;
+			mclk_cfg.TunerfreqMHz = c->frequency / 1000;
+			mclk_cfg.SymRateKSs = c->symbol_rate / 1000;

drivers/media/dvb-frontends/m88ds3103.c static int m88ds3103_set_frontend(struct dvb_frontend *fe) · line 314

Quoted 10 lines

drivers/media/dvb-frontends/m88ds3103.c +10 -0

+			ret = fe->ops.tuner_ops.set_config(fe, &mclk_cfg);
+			if (ret)
+				goto err;
+			priv->mclk_khz = mclk_cfg.MclkKHz;
+		}
+		ret = m88ds3103_wr_reg(priv, 0x06, 0x00);
+		if (ret)
+			goto err;
+		usleep_range(10000, 20000);
+	}

That looks odd and also ugly. You pass some values from demod to tuner using set_config callback. Tuner driver can get symbol_rate and frequency just similarly from property cache than demod. Why you do it like that?

Clock is provided by tuner as you mention. I see you use that to pass used clock frequency from tuner to demod. This does not look nice and I would like to see clock framework instead. Or calculate clock on both drivers. Does the demod clock even needs to be changed? I think it is only TS stream size which defines used clock frequency - smaller the TS bitstream, the smaller the clock frequency needed => optimizes power consumption a little. But TS clock is calculated on tuner driver in any case?

regards
Antti

Signature · 1 line

http://palosaari.fi/

Quoted 47 lines · Antti Palosaari

Quoted 28 lines · Nibble Max

M88RS6000 is the integrated chip, which includes tuner and demod.
Its internal demod is similar with M88DS3103 except some registers definition.
The main different part of this internal demod from others is its clock/pll generation IP block sitting inside the tuner die. So clock/pll functions should be configed through its tuner i2c bus, NOT its demod i2c bus.
The demod of M88RS6000 need the firmware: dvb-demod-m88rs6000.fw
firmware download link: http://www.dvbsky.net/download/linux/dvbsky-firmware.tar.gz

drivers/media/dvb-frontends/m88ds3103.c +10 -0

@@ -250,6 +251,7 @@ static int m88ds3103_set_frontend(struct dvb_frontend *fe)
   	u16 u16tmp, divide_ratio;
   	u32 tuner_frequency, target_mclk;
   	s32 s32tmp;
+	struct m88rs6000_mclk_config mclk_cfg;
 
   	dev_dbg(&priv->i2c->dev,
   			"%s: delivery_system=%d modulation=%d frequency=%d symbol_rate=%d inversion=%d pilot=%d rolloff=%d\n",

@@ -291,6 +293,26 @@ static int m88ds3103_set_frontend(struct dvb_frontend *fe)
   	if (ret)
   		goto err;
 
+	if (priv->chip_id == M88RS6000_CHIP_ID) {
+		ret = m88ds3103_wr_reg(priv, 0x06, 0xe0);
+		if (ret)
+			goto err;
+		if (fe->ops.tuner_ops.set_config) {
+			/* select main mclk */
+			mclk_cfg.config_op = 0;
+			mclk_cfg.TunerfreqMHz = c->frequency / 1000;
+			mclk_cfg.SymRateKSs = c->symbol_rate / 1000;

drivers/media/dvb-frontends/m88ds3103.c static int m88ds3103_set_frontend(struct dvb_frontend *fe) · line 314

Quoted 10 lines

drivers/media/dvb-frontends/m88ds3103.c +10 -0

+			ret = fe->ops.tuner_ops.set_config(fe, &mclk_cfg);
+			if (ret)
+				goto err;
+			priv->mclk_khz = mclk_cfg.MclkKHz;
+		}
+		ret = m88ds3103_wr_reg(priv, 0x06, 0x00);
+		if (ret)
+			goto err;
+		usleep_range(10000, 20000);
+	}

That looks odd and also ugly. You pass some values from demod to tuner using set_config callback. Tuner driver can get symbol_rate and frequency just similarly from property cache than demod. Why you do it like that?

Quoted 9 lines

Clock is provided by tuner as you mention. I see you use that to pass used clock frequency from tuner to demod. This does not look nice and I would like to see clock framework instead. Or calculate clock on both drivers. Does the demod clock even needs to be changed? I think it is only TS stream size which defines used clock frequency - smaller the TS bitstream, the smaller the clock frequency needed => optimizes power consumption a little. But TS clock is calculated on tuner driver in any case?

Yes, M88RS6000 looks odd. This integrated chip has two part die, tuner die and demod die.
Its demod's clock(PLL) block is sitting insided the tuner die. The demod has no PLL ip block that makes demod die smaller. The demod's clock can be adjusted according to the transponder's frequency and symbol rate.
So that the demod's clock and its harmonic frequency will not overlap with the transponder's frequency range. It will improve its tuner's sensitivity.

However the tuner driver can get the values from property cache. Tuner driver does not know when need adjust this demod pll and return the current demod pll value to the demod driver. in "struct dvb_tuner_ops", there is no call back to do this directly. So I select the general "set_config" call back.
TS main clock of demdod also need to be controlled in the tuner die.

These demod's PLL registers have no relationship with tuner at all.
Logically, These demod's PLL registers should go with demod die as usual.
But in this case they goes with the tuner die physically and controlled through tuner i2c bus.

The current dvb-frontend driver requires the tuner and demod to split into the seperate drivers. The demod driver will not access tuner i2c bus directly.
But this integrate chip has more tighter relationship of its tuner and demod die. That is reason why these odd call backs happen.

Moikka!

Quoted 70 lines · Nibble Max

Quoted 47 lines · Antti Palosaari

Quoted 28 lines · Nibble Max

M88RS6000 is the integrated chip, which includes tuner and demod.
Its internal demod is similar with M88DS3103 except some registers definition.
The main different part of this internal demod from others is its clock/pll generation IP block sitting inside the tuner die. So clock/pll functions should be configed through its tuner i2c bus, NOT its demod i2c bus.
The demod of M88RS6000 need the firmware: dvb-demod-m88rs6000.fw
firmware download link: http://www.dvbsky.net/download/linux/dvbsky-firmware.tar.gz

drivers/media/dvb-frontends/m88ds3103.c +10 -0

@@ -250,6 +251,7 @@ static int m88ds3103_set_frontend(struct dvb_frontend *fe)
   	u16 u16tmp, divide_ratio;
   	u32 tuner_frequency, target_mclk;
   	s32 s32tmp;
+	struct m88rs6000_mclk_config mclk_cfg;
 
   	dev_dbg(&priv->i2c->dev,
   			"%s: delivery_system=%d modulation=%d frequency=%d symbol_rate=%d inversion=%d pilot=%d rolloff=%d\n",

@@ -291,6 +293,26 @@ static int m88ds3103_set_frontend(struct dvb_frontend *fe)
   	if (ret)
   		goto err;
 
+	if (priv->chip_id == M88RS6000_CHIP_ID) {
+		ret = m88ds3103_wr_reg(priv, 0x06, 0xe0);
+		if (ret)
+			goto err;
+		if (fe->ops.tuner_ops.set_config) {
+			/* select main mclk */
+			mclk_cfg.config_op = 0;
+			mclk_cfg.TunerfreqMHz = c->frequency / 1000;
+			mclk_cfg.SymRateKSs = c->symbol_rate / 1000;

drivers/media/dvb-frontends/m88ds3103.c static int m88ds3103_set_frontend(struct dvb_frontend *fe) · line 314

Quoted 10 lines

drivers/media/dvb-frontends/m88ds3103.c +10 -0

+			ret = fe->ops.tuner_ops.set_config(fe, &mclk_cfg);
+			if (ret)
+				goto err;
+			priv->mclk_khz = mclk_cfg.MclkKHz;
+		}
+		ret = m88ds3103_wr_reg(priv, 0x06, 0x00);
+		if (ret)
+			goto err;
+		usleep_range(10000, 20000);
+	}

That looks odd and also ugly. You pass some values from demod to tuner using set_config callback. Tuner driver can get symbol_rate and frequency just similarly from property cache than demod. Why you do it like that?

Quoted 9 lines

Clock is provided by tuner as you mention. I see you use that to pass used clock frequency from tuner to demod. This does not look nice and I would like to see clock framework instead. Or calculate clock on both drivers. Does the demod clock even needs to be changed? I think it is only TS stream size which defines used clock frequency - smaller the TS bitstream, the smaller the clock frequency needed => optimizes power consumption a little. But TS clock is calculated on tuner driver in any case?

Yes, M88RS6000 looks odd. This integrated chip has two part die, tuner die and demod die.
Its demod's clock(PLL) block is sitting insided the tuner die. The demod has no PLL ip block that makes demod die smaller. The demod's clock can be adjusted according to the transponder's frequency and symbol rate.
So that the demod's clock and its harmonic frequency will not overlap with the transponder's frequency range. It will improve its tuner's sensitivity.

However the tuner driver can get the values from property cache. Tuner driver does not know when need adjust this demod pll and return the current demod pll value to the demod driver. in "struct dvb_tuner_ops", there is no call back to do this directly. So I select the general "set_config" call back.
TS main clock of demdod also need to be controlled in the tuner die.

Quoted 8 lines

These demod's PLL registers have no relationship with tuner at all.
Logically, These demod's PLL registers should go with demod die as usual.
But in this case they goes with the tuner die physically and controlled through tuner i2c bus.

The current dvb-frontend driver requires the tuner and demod to split into the seperate drivers. The demod driver will not access tuner i2c bus directly.
But this integrate chip has more tighter relationship of its tuner and demod die. That is reason why these odd call backs happen.

I understand situation pretty well.
Let me say it shortly still to make it clear for the possible others; M88RS6000 is integrated DVB-S2 receiver containing demod IP block that is very near to M88DS3103 demod and tuner which is not so near any existing tuner chip. Both IP blocks, demod and tuner, are connected to I2C bus. However, all demodulator clocks (master clock and TS master clock) are coming from tuner die. Tuner gets reference clock from 27 MHz Xtal and there is PLL or multiple PLLs to derive/generate all the other needed clocks.

So now we need to get demod MCLK and TS MCLK from tuner to demod. I looked the tuner driver and saw following possible M88RS6000 MCLKs 93.000, 96.000, 99.000, 110.250 MHz. Old M88DS3103 uses only 96.000 MHz. There was rather simple logic to select suitable (smallest possible) demod MCLK according to symbol rate. TS master clock calculation looked more harder.

1) demod MCLK calculation
Could you write down smallest possible logic to calculate needed demod MCLK frequency? I looked m88rs6000t_select_mclk(), but I am not very happy as that function contains both dead-code and variables that are initialized multiple times. So take a pen and paper, or text editor, and write down code snippet for it.

2) TS MCLK calculation
Initial TS master clock is calculated by demod driver, which then pass that clock to tuner driver, which then programs it to PLL, and finally PLL multipliers/dividers are read back and actual frequency is calculated. Demod driver took that MCLK and uses integer divider in order to get final TS clock. As there is rounding errors in any case and TS clock is not critical at all, I wonder if it is wise to read it back from PLL registers at all (on m88rs6000t driver). Do you have idea how much there is rounding error in real life when TS MCLK is programmed? I mean if you ask 96MHz TS MCLK, what the result is after you read registers back?

I mean these functions:
m88rs6000t_set_ts_mclk()
m88rs6000_get_ts_mclk()

regards
Antti

Signature · 1 line

http://palosaari.fi/