ngene: fix dvb_pll_attach failure

Before dvb_pll_attch call, be sure that drxd demodulator was initialized, otherwise, dvb_pll_attach() will always failed.

In dvb_pll_attach(), first thing done is to enable the I2C gate control in order to probe the pll by performing a read access. As demodulator was not initialized, every i2c access failed.

Reported-by [email protected]

Signed-off-by Patrice Chotard <[email protected]>

---
  drivers/media/pci/ngene/ngene-cards.c |    2 ++
  1 file changed, 2 insertions(+)

drivers/media/pci/ngene/ngene-cards.c +0 -0

index 96a13ed..e2192db 100644

Quoted 27 lines · Patrice Chotard

Before dvb_pll_attch call, be sure that drxd demodulator was initialized, otherwise, dvb_pll_attach() will always failed.

In dvb_pll_attach(), first thing done is to enable the I2C gate control in order to probe the pll by performing a read access. As demodulator was not initialized, every i2c access failed.

Reported-by [email protected]

Signed-off-by Patrice Chotard <[email protected]>

---
  drivers/media/pci/ngene/ngene-cards.c |    2 ++
  1 file changed, 2 insertions(+)

drivers/media/pci/ngene/ngene-cards.c +2 -0

index 96a13ed..e2192db 100644
@@ -328,6 +328,8 @@ static int demod_attach_drxd(struct ngene_channel
 *chan)
          return -ENODEV;
      }
 
+    /* initialized the DRXD demodulator */
+    chan->fe->ops.init(chan->fe);
      if (!dvb_attach(dvb_pll_attach, chan->fe, feconf->pll_address,
              &chan->i2c_adapter,

Quoted 2 lines

              feconf->pll_type)) {

I don't like that as this causes again more deviation against normal procedures. If gate open is needed (for probe or id check?) then pll/tuner attach should open it. If that is not easily possible then calling gate_control() before pll attach is allowed. init() is very, very, bad as generally starts whole chip => starts eating power etc.

Even better would be to let whole gate-control to responsibility of DVB-core, but unfortunately current situation is quite mess. Gate is operated sometimes by DVB-core (like for init/sleep) and for some cases it is left for responsibility of tuner driver. So on real life there is mixed solutions and for init/sleep gate is even double controlled.

regards
Antti

Quoted 45 lines · Antti Palosaari

Quoted 27 lines · Patrice Chotard

Before dvb_pll_attch call, be sure that drxd demodulator was initialized, otherwise, dvb_pll_attach() will always failed.

In dvb_pll_attach(), first thing done is to enable the I2C gate control in order to probe the pll by performing a read access. As demodulator was not initialized, every i2c access failed.

Reported-by [email protected]

Signed-off-by Patrice Chotard <[email protected]>

---
   drivers/media/pci/ngene/ngene-cards.c |    2 ++
   1 file changed, 2 insertions(+)

drivers/media/pci/ngene/ngene-cards.c +2 -0

index 96a13ed..e2192db 100644
@@ -328,6 +328,8 @@ static int demod_attach_drxd(struct ngene_channel
 *chan)
           return -ENODEV;
       }
 
+    /* initialized the DRXD demodulator */
+    chan->fe->ops.init(chan->fe);
       if (!dvb_attach(dvb_pll_attach, chan->fe, feconf->pll_address,
               &chan->i2c_adapter,

Quoted 2 lines

               feconf->pll_type)) {

I don't like that as this causes again more deviation against normal procedures. If gate open is needed (for probe or id check?) then pll/tuner attach should open it. If that is not easily possible then calling gate_control() before pll attach is allowed. init() is very, very, bad as generally starts whole chip => starts eating power etc.

Even better would be to let whole gate-control to responsibility of DVB-core, but unfortunately current situation is quite mess. Gate is operated sometimes by DVB-core (like for init/sleep) and for some cases it is left for responsibility of tuner driver. So on real life there is mixed solutions and for init/sleep gate is even double controlled.

Quoted 3 lines

regards
Antti

Hi Antti,

Unfortunately, opening gate doesn't work with drxd demodulator before its initialization (drxd_init() call).
Also, in dvb_pll_attach(), first thing done is opening the gate control.

I was aware by calling init() directly in demod_attach_drxd() was not good at the power consumption point of view, but it was the only solution to make it work.

To stop the power consumption, it would be possible to call the frontends :

  	}

(quoted diff) +3 -0

+	/* initialized the DRXD demodulator */
+	chan->fe->ops.init(chan->fe);
  	if (!dvb_attach(dvb_pll_attach, chan->fe, feconf->pll_address,
  			&chan->i2c_adapter,
  			feconf->pll_type)) {
  		pr_err("No pll(%d) found!\n", feconf->pll_type);
  		return -ENODEV;
  	}
+	chan->fe->ops.sleep(chan->fe);
  	return 0;
  }

Thanks for your feedback.

That one is better solution...

but it is clearly DRXD driver bug - it should offer working I2C gate control after attach() :-( I looked quickly DRXD driver and there is clearly some other places to enhance too. But I suspect there is no maintainer, nor interest from anyone to start fix things properly, so only reasonable way is to add that hack to get it working...

Honestly, I hate this kind of hacks :/ That makes our live hard on long run. It goes slowly more and more hard to make any core changes as regressions will happen due to this kind of hacks.

So send new patch which put demod chip sleeping after tuner attach. Or even better, find out what is minimal set of commands needed do execute during attach in order to offer working I2C gate (I suspect firmware load is needed).

Quoted 77 lines · Patrice Chotard

Quoted 45 lines · Antti Palosaari

Quoted 27 lines · Patrice Chotard

Before dvb_pll_attch call, be sure that drxd demodulator was initialized, otherwise, dvb_pll_attach() will always failed.

In dvb_pll_attach(), first thing done is to enable the I2C gate control in order to probe the pll by performing a read access. As demodulator was not initialized, every i2c access failed.

Reported-by [email protected]

Signed-off-by Patrice Chotard <[email protected]>

---
   drivers/media/pci/ngene/ngene-cards.c |    2 ++
   1 file changed, 2 insertions(+)

drivers/media/pci/ngene/ngene-cards.c +2 -0

index 96a13ed..e2192db 100644
@@ -328,6 +328,8 @@ static int demod_attach_drxd(struct ngene_channel
 *chan)
           return -ENODEV;
       }
 
+    /* initialized the DRXD demodulator */
+    chan->fe->ops.init(chan->fe);
       if (!dvb_attach(dvb_pll_attach, chan->fe, feconf->pll_address,
               &chan->i2c_adapter,

Quoted 2 lines

               feconf->pll_type)) {

I don't like that as this causes again more deviation against normal procedures. If gate open is needed (for probe or id check?) then pll/tuner attach should open it. If that is not easily possible then calling gate_control() before pll attach is allowed. init() is very, very, bad as generally starts whole chip => starts eating power etc.

Even better would be to let whole gate-control to responsibility of DVB-core, but unfortunately current situation is quite mess. Gate is operated sometimes by DVB-core (like for init/sleep) and for some cases it is left for responsibility of tuner driver. So on real life there is mixed solutions and for init/sleep gate is even double controlled.

Quoted 3 lines

regards
Antti

Hi Antti,

Unfortunately, opening gate doesn't work with drxd demodulator before its initialization (drxd_init() call).
Also, in dvb_pll_attach(), first thing done is opening the gate control.

I was aware by calling init() directly in demod_attach_drxd() was not good at the power consumption point of view, but it was the only solution to make it work.

To stop the power consumption, it would be possible to call the frontends :

      }

(quoted diff) +3 -0

+    /* initialized the DRXD demodulator */
+    chan->fe->ops.init(chan->fe);
      if (!dvb_attach(dvb_pll_attach, chan->fe, feconf->pll_address,
              &chan->i2c_adapter,
              feconf->pll_type)) {
          pr_err("No pll(%d) found!\n", feconf->pll_type);
          return -ENODEV;
      }
+    chan->fe->ops.sleep(chan->fe);
      return 0;
  }

Quoted 1 line

Thanks for your feedback.

That one is better solution...

but it is clearly DRXD driver bug - it should offer working I2C gate control after attach() :-( I looked quickly DRXD driver and there is clearly some other places to enhance too. But I suspect there is no maintainer, nor interest from anyone to start fix things properly, so only reasonable way is to add that hack to get it working...

Honestly, I hate this kind of hacks :/ That makes our live hard on long run. It goes slowly more and more hard to make any core changes as regressions will happen due to this kind of hacks.

So send new patch which put demod chip sleeping after tuner attach. Or even better, find out what is minimal set of commands needed do execute during attach in order to offer working I2C gate (I suspect firmware load is needed).

regards
Antti