af9015: limit I2C access to keep FW happy

AF9015 firmware does not like if it gets interrupted by I2C adapter request on some critical phases. During normal operation I2C adapter is used only 2nd demodulator and tuner on dual tuner devices.

Override demodulator callbacks and use mutex for limit access to those "critical" paths to keep AF9015 happy.

Signed-off-by Antti Palosaari<[email protected]>

---
  drivers/media/dvb/dvb-usb/af9015.c |   97 ++++++++++++++++++++++++++++++++++++
  drivers/media/dvb/dvb-usb/af9015.h |    7 +++
  2 files changed, 104 insertions(+), 0 deletions(-)

drivers/media/dvb/dvb-usb/af9015.c +97 -0

index c6c275b..033aa8a 100644
@@ -1093,9 +1093,80 @@ error:
  	return ret;
  }
 
+/* override demod callbacks for resource locking */
+static int af9015_af9013_set_frontend(struct dvb_frontend *fe,
+	struct dvb_frontend_parameters *params)
+{
+	int ret;
+	struct dvb_usb_adapter *adap = fe->dvb->priv;
+	struct af9015_state *priv = adap->dev->priv;
+
+	if (mutex_lock_interruptible(&adap->dev->usb_mutex))
+		return -EAGAIN;
+
+	ret = priv->set_frontend[adap->id](fe, params);
+
+	mutex_unlock(&adap->dev->usb_mutex);
+
+	return ret;
+}
+
+/* override demod callbacks for resource locking */
+static int af9015_af9013_read_status(struct dvb_frontend *fe,
+	fe_status_t *status)
+{
+	int ret;
+	struct dvb_usb_adapter *adap = fe->dvb->priv;
+	struct af9015_state *priv = adap->dev->priv;
+
+	if (mutex_lock_interruptible(&adap->dev->usb_mutex))
+		return -EAGAIN;
+
+	ret = priv->read_status[adap->id](fe, status);
+
+	mutex_unlock(&adap->dev->usb_mutex);
+
+	return ret;
+}
+
+/* override demod callbacks for resource locking */
+static int af9015_af9013_init(struct dvb_frontend *fe)
+{
+	int ret;
+	struct dvb_usb_adapter *adap = fe->dvb->priv;
+	struct af9015_state *priv = adap->dev->priv;
+
+	if (mutex_lock_interruptible(&adap->dev->usb_mutex))
+		return -EAGAIN;
+
+	ret = priv->init[adap->id](fe);
+
+	mutex_unlock(&adap->dev->usb_mutex);
+
+	return ret;
+}
+
+/* override demod callbacks for resource locking */
+static int af9015_af9013_sleep(struct dvb_frontend *fe)
+{
+	int ret;
+	struct dvb_usb_adapter *adap = fe->dvb->priv;
+	struct af9015_state *priv = adap->dev->priv;
+
+	if (mutex_lock_interruptible(&adap->dev->usb_mutex))
+		return -EAGAIN;
+
+	ret = priv->init[adap->id](fe);
+
+	mutex_unlock(&adap->dev->usb_mutex);
+
+	return ret;
+}
+
  static int af9015_af9013_frontend_attach(struct dvb_usb_adapter *adap)
  {
  	int ret;
+	struct af9015_state *state = adap->dev->priv;
 
  	if (adap->id == 1) {
  		/* copy firmware to 2nd demodulator */

@@ -1116,6 +1187,32 @@ static int af9015_af9013_frontend_attach(struct dvb_usb_adapter *adap)
  	adap->fe_adap[0].fe = dvb_attach(af9013_attach,&af9015_af9013_config[adap->id],
  		&adap->dev->i2c_adap);
 
+	/*
+	 * AF9015 firmware does not like if it gets interrupted by I2C adapter
+	 * request on some critical phases. During normal operation I2C adapter
+	 * is used only 2nd demodulator and tuner on dual tuner devices.
+	 * Override demodulator callbacks and use mutex for limit access to
+	 * those "critical" paths to keep AF9015 happy.
+	 * Note: we abuse unused usb_mutex here.
+	 */
+	if (adap->fe_adap[0].fe) {
+		state->set_frontend[adap->id] =
+			adap->fe_adap[0].fe->ops.set_frontend;
+		adap->fe_adap[0].fe->ops.set_frontend =
+			af9015_af9013_set_frontend;
+
+		state->read_status[adap->id] =
+			adap->fe_adap[0].fe->ops.read_status;
+		adap->fe_adap[0].fe->ops.read_status =
+			af9015_af9013_read_status;
+
+		state->init[adap->id] = adap->fe_adap[0].fe->ops.init;
+		adap->fe_adap[0].fe->ops.init = af9015_af9013_init;
+
+		state->sleep[adap->id] = adap->fe_adap[0].fe->ops.sleep;
+		adap->fe_adap[0].fe->ops.sleep = af9015_af9013_sleep;
+	}
+
  	return adap->fe_adap[0].fe == NULL ? -ENODEV : 0;
  }
 

drivers/media/dvb/dvb-usb/af9015.h +7 -0

index 6252ea6..4a12617 100644
@@ -102,6 +102,13 @@ struct af9015_state {
  	u8 rc_repeat;
  	u32 rc_keycode;
  	u8 rc_last[4];
+
+	/* for demod callback override */
+	int (*set_frontend[2]) (struct dvb_frontend *fe,
+		struct dvb_frontend_parameters *params);
+	int (*read_status[2]) (struct dvb_frontend *fe, fe_status_t *status);
+	int (*init[2]) (struct dvb_frontend *fe);
+	int (*sleep[2]) (struct dvb_frontend *fe);
  };
 
  struct af9015_config {

Mauro,
Don't still but that to the 3.2 as fix. It is not much tested and not critical so better put it master for some more testing.

regards
Antti

Quoted 123 lines · Antti Palosaari

AF9015 firmware does not like if it gets interrupted by I2C adapter request on some critical phases. During normal operation I2C adapter is used only 2nd demodulator and tuner on dual tuner devices.

Override demodulator callbacks and use mutex for limit access to those "critical" paths to keep AF9015 happy.

Signed-off-by Antti Palosaari<[email protected]>

---
  drivers/media/dvb/dvb-usb/af9015.c |   97 ++++++++++++++++++++++++++++++++++++
  drivers/media/dvb/dvb-usb/af9015.h |    7 +++
  2 files changed, 104 insertions(+), 0 deletions(-)

drivers/media/dvb/dvb-usb/af9015.c +92 -0

index c6c275b..033aa8a 100644
@@ -1093,9 +1093,80 @@ error:
  	return ret;
  }
 
+/* override demod callbacks for resource locking */
+static int af9015_af9013_set_frontend(struct dvb_frontend *fe,
+	struct dvb_frontend_parameters *params)
+{
+	int ret;
+	struct dvb_usb_adapter *adap = fe->dvb->priv;
+	struct af9015_state *priv = adap->dev->priv;
+
+	if (mutex_lock_interruptible(&adap->dev->usb_mutex))
+		return -EAGAIN;
+
+	ret = priv->set_frontend[adap->id](fe, params);
+
+	mutex_unlock(&adap->dev->usb_mutex);
+
+	return ret;
+}
+
+/* override demod callbacks for resource locking */
+static int af9015_af9013_read_status(struct dvb_frontend *fe,
+	fe_status_t *status)
+{
+	int ret;
+	struct dvb_usb_adapter *adap = fe->dvb->priv;
+	struct af9015_state *priv = adap->dev->priv;
+
+	if (mutex_lock_interruptible(&adap->dev->usb_mutex))
+		return -EAGAIN;
+
+	ret = priv->read_status[adap->id](fe, status);
+
+	mutex_unlock(&adap->dev->usb_mutex);
+
+	return ret;
+}
+
+/* override demod callbacks for resource locking */
+static int af9015_af9013_init(struct dvb_frontend *fe)
+{
+	int ret;
+	struct dvb_usb_adapter *adap = fe->dvb->priv;
+	struct af9015_state *priv = adap->dev->priv;
+
+	if (mutex_lock_interruptible(&adap->dev->usb_mutex))
+		return -EAGAIN;
+
+	ret = priv->init[adap->id](fe);
+
+	mutex_unlock(&adap->dev->usb_mutex);
+
+	return ret;
+}
+
+/* override demod callbacks for resource locking */
+static int af9015_af9013_sleep(struct dvb_frontend *fe)
+{
+	int ret;
+	struct dvb_usb_adapter *adap = fe->dvb->priv;
+	struct af9015_state *priv = adap->dev->priv;
+
+	if (mutex_lock_interruptible(&adap->dev->usb_mutex))
+		return -EAGAIN;
+
+	ret = priv->init[adap->id](fe);
+
+	mutex_unlock(&adap->dev->usb_mutex);
+
+	return ret;
+}
+
  static int af9015_af9013_frontend_attach(struct dvb_usb_adapter *adap)
  {
  	int ret;
+	struct af9015_state *state = adap->dev->priv;
 
  	if (adap->id == 1) {
  		/* copy firmware to 2nd demodulator */

@@ -1116,6 +1187,32 @@ static int af9015_af9013_frontend_attach(struct dvb_usb_adapter *adap)
  	adap->fe_adap[0].fe = dvb_attach(af9013_attach,&af9015_af9013_config[adap->id],
  		&adap->dev->i2c_adap);
 
+	/*
+	 * AF9015 firmware does not like if it gets interrupted by I2C adapter
+	 * request on some critical phases. During normal operation I2C adapter
+	 * is used only 2nd demodulator and tuner on dual tuner devices.
+	 * Override demodulator callbacks and use mutex for limit access to
+	 * those "critical" paths to keep AF9015 happy.
+	 * Note: we abuse unused usb_mutex here.
+	 */
+	if (adap->fe_adap[0].fe) {
+		state->set_frontend[adap->id] =
+			adap->fe_adap[0].fe->ops.set_frontend;
+		adap->fe_adap[0].fe->ops.set_frontend =
+			af9015_af9013_set_frontend;
+
+		state->read_status[adap->id] =
+			adap->fe_adap[0].fe->ops.read_status;
+		adap->fe_adap[0].fe->ops.read_status =
+			af9015_af9013_read_status;
+
+		state->init[adap->id] = adap->fe_adap[0].fe->ops.init;
+		adap->fe_adap[0].fe->ops.init = af9015_af9013_init;

drivers/media/dvb/dvb-usb/af9015.c static int af9015_af9013_frontend_attach(struct dvb_usb_adapter *adap) · line 1218

Quoted 8 lines

drivers/media/dvb/dvb-usb/af9015.c +5 -0

+
+		state->sleep[adap->id] = adap->fe_adap[0].fe->ops.sleep;
+		adap->fe_adap[0].fe->ops.sleep = af9015_af9013_sleep;
+	}
+
  	return adap->fe_adap[0].fe == NULL ? -ENODEV : 0;
  }
 

Quoted 7 lines · Antti Palosaari

Quoted 6 lines

AF9015 firmware does not like if it gets interrupted by I2C adapter request on some critical phases. During normal operation I2C adapter is used only 2nd demodulator and tuner on dual tuner devices.

Override demodulator callbacks and use mutex for limit access to those "critical" paths to keep AF9015 happy.

I have tried this patch, while it initially got MythTV working, there is too many call backs and some failed to acquire the lock. The device became unstable on both single and dual devices.

The callbacks

af9015_af9013_read_status,
af9015_af9013_init
af9015_af9013_sleep

had to be removed.

I take your point, a call back can be an alternative.

The patch didn't stop the firmware fails either.

The af9015 usb bridge on the whole is so unstable in its early stages, especially on a cold boot and when the USB controller has another device on it, such as card reader or wifi device.

I am, at the moment looking to see if the fails are due to interface 1 being claimed by HID.

I have tried this patch, while it initially got MythTV working, there is too many call backs and some failed to acquire the lock. The device became unstable on both single and dual devices.

The callbacks

af9015_af9013_read_status,
af9015_af9013_init
af9015_af9013_sleep

had to be removed.

I take your point, a call back can be an alternative.

The patch didn't stop the firmware fails either.

The af9015 usb bridge on the whole is so unstable in its early stages, especially on a cold boot and when the USB controller has another device on it, such as card reader or wifi device.

I am, at the moment looking to see if the fails are due to interface 1 being claimed by HID.

Quoted 14 lines · Malcolm Priestley

I have tried this patch, while it initially got MythTV working, there is too many call backs and some failed to acquire the lock. The device became unstable on both single and dual devices.

The callbacks

af9015_af9013_read_status,
af9015_af9013_init
af9015_af9013_sleep

had to be removed.

I take your point, a call back can be an alternative.

Quoted 8 lines

The patch didn't stop the firmware fails either.

The af9015 usb bridge on the whole is so unstable in its early stages, especially on a cold boot and when the USB controller has another device on it, such as card reader or wifi device.

I am, at the moment looking to see if the fails are due to interface 1 being claimed by HID.

I just got af9013 rewrite ready. Feel free to test.
http://git.linuxtv.org/anttip/media_tree.git/shortlog/refs/heads/misc

It reduces typical statistics polling load maybe 1/4 from the original.

I see still small glitch when switching to channel. That seems to come from tuner driver I2C load. There is 3 tuners used for dual devices; MXL5005S, MXL5007T and TDA18271. I have only MXL5005S and MXL5007T dual devices here. MXL5005S is worse than MXL5007T but both makes still rather much I/O.