[v7,2/4] i2c-acpi : exclude ARA address for smbus device

Message ID 20180113133705.25044-2-m.capdeville@no-log.org
State New
Headers show
Series
  • [v7,1/4] i2c-smbus : Add client discovered ARA support
Related show

Commit Message

CAPDEVILLE Marc Jan. 13, 2018, 1:37 p.m.
Somme ACPI enumerated devices are known to support smbus alert protocol.
Theses devices may be miss-enumerated with the reserved smbus ARA address.

This is the case on Asus T100 tablet where cm3218 ambiant light sensor
expose two i2c serial bus connections, with the first one being the alert
response address.

This patch make a match on known ACPI ids for which devices are smbus ARA
capable, then skip the connection if it has the reserved 0x0c address and
mark it with I2C_CLIENT_ALERT flag. So device is enumerated with the
correct address.

Signed-off-by: Marc CAPDEVILLE <m.capdeville@no-log.org>
---
 drivers/i2c/i2c-core-acpi.c | 23 +++++++++++++++++++++--
 include/linux/i2c.h         |  1 +
 2 files changed, 22 insertions(+), 2 deletions(-)

Comments

Jonathan Cameron Jan. 14, 2018, 11:28 a.m. | #1
On Sat, 13 Jan 2018 14:37:03 +0100
Marc CAPDEVILLE <m.capdeville@no-log.org> wrote:

> Somme ACPI enumerated devices are known to support smbus alert protocol.
> Theses devices may be miss-enumerated with the reserved smbus ARA address.
> 
> This is the case on Asus T100 tablet where cm3218 ambiant light sensor
> expose two i2c serial bus connections, with the first one being the alert
> response address.
> 
> This patch make a match on known ACPI ids for which devices are smbus ARA
> capable, then skip the connection if it has the reserved 0x0c address and
> mark it with I2C_CLIENT_ALERT flag. So device is enumerated with the
> correct address.

I wonder if we are safe to always skip 0x0c address whether or not
we know the device supports ARA.  The exception may be for devices
that do support ARA but are rolling their own support the hard way...

I suppose it's possible there are i2c devices (not smbus where the
spec says you must not use 0x0c for normal address IIRC) that use
this address.  Does anyone know of any?

Jonathan

> 
> Signed-off-by: Marc CAPDEVILLE <m.capdeville@no-log.org>
> ---
>  drivers/i2c/i2c-core-acpi.c | 23 +++++++++++++++++++++--
>  include/linux/i2c.h         |  1 +
>  2 files changed, 22 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/i2c/i2c-core-acpi.c b/drivers/i2c/i2c-core-acpi.c
> index a9126b3cda61..5a8886f14329 100644
> --- a/drivers/i2c/i2c-core-acpi.c
> +++ b/drivers/i2c/i2c-core-acpi.c
> @@ -59,8 +59,14 @@ static int i2c_acpi_fill_info(struct acpi_resource *ares, void *data)
>  	if (sb->type != ACPI_RESOURCE_SERIAL_TYPE_I2C)
>  		return 1;
>  
> -	if (lookup->index != -1 && lookup->n++ != lookup->index)
> -		return 1;
> +	if (lookup->index != -1) {
> +		if (lookup->n++ != lookup->index)
> +			return 1;
> +	} else {
> +		if (lookup->info->flags & I2C_CLIENT_ALERT &&
> +		    sb->slave_address == 0x0c)
> +			return 1;
> +	}
>  
>  	status = acpi_get_handle(lookup->device_handle,
>  				 sb->resource_source.string_ptr,
> @@ -85,6 +91,15 @@ static const struct acpi_device_id i2c_acpi_ignored_device_ids[] = {
>  	{}
>  };
>  
> +static const struct acpi_device_id i2c_acpi_alert_device_ids[] = {
> +	/*
> +	 * Smbus alert capable device which may have the reserved ARA address
> +	 * in their serial bus resources list.
> +	 */
> +	{ "CPLM3218", 0 },
> +	{}
> +};
> +
>  static int i2c_acpi_do_lookup(struct acpi_device *adev,
>  			      struct i2c_acpi_lookup *lookup)
>  {
> @@ -100,6 +115,10 @@ static int i2c_acpi_do_lookup(struct acpi_device *adev,
>  		return -ENODEV;
>  
>  	memset(info, 0, sizeof(*info));
> +
> +	if (acpi_match_device_ids(adev, i2c_acpi_alert_device_ids) == 0)
> +		info->flags |= I2C_CLIENT_ALERT;
> +
>  	lookup->device_handle = acpi_device_handle(adev);
>  
>  	/* Look up for I2cSerialBus resource */
> diff --git a/include/linux/i2c.h b/include/linux/i2c.h
> index 7592dce12923..b0d6f1333442 100644
> --- a/include/linux/i2c.h
> +++ b/include/linux/i2c.h
> @@ -743,6 +743,7 @@ i2c_unlock_adapter(struct i2c_adapter *adapter)
>  #define I2C_CLIENT_SLAVE	0x20	/* we are the slave */
>  #define I2C_CLIENT_HOST_NOTIFY	0x40	/* We want to use I2C host notify */
>  #define I2C_CLIENT_WAKE		0x80	/* for board_info; true iff can wake */
> +#define I2C_CLIENT_ALERT	0x100	/* Client use SMBUS alert protocol */
>  #define I2C_CLIENT_SCCB		0x9000	/* Use Omnivision SCCB protocol */
>  					/* Must match I2C_M_STOP|IGNORE_NAK */
>
CAPDEVILLE Marc Jan. 17, 2018, 10:28 a.m. | #2
Le dimanche 14 janvier 2018 à 11:28 +0000, Jonathan Cameron a écrit :
> On Sat, 13 Jan 2018 14:37:03 +0100
> Marc CAPDEVILLE <m.capdeville@no-log.org> wrote:
> 
> > Somme ACPI enumerated devices are known to support smbus alert protocol.
> > Theses devices may be miss-enumerated with the reserved smbus ARA address.
> > 
> > This is the case on Asus T100 tablet where cm3218 ambiant light sensor
> > expose two i2c serial bus connections, with the first one being the alert
> > response address.
> > 
> > This patch make a match on known ACPI ids for which devices are smbus ARA
> > capable, then skip the connection if it has the reserved 0x0c address and
> > mark it with I2C_CLIENT_ALERT flag. So device is enumerated with the
> > correct address.
> 
> I wonder if we are safe to always skip 0x0c address whether or not
> we know the device supports ARA.  The exception may be for devices
> that do support ARA but are rolling their own support the hard way...

The device is only flagged with I2C_CLIENT_ALERT if its ID is in the
i2c_acpi_alert_device_ids. IF a driver want to be enumerated with the
reserved Ox0c address, It just have to not put its Id in this table. But
this may break with registering the smbus_alert device as the address will
be marked as busy for this adapter.

> I suppose it's possible there are i2c devices (not smbus where the
> spec says you must not use 0x0c for normal address IIRC) that use
> this address.  Does anyone know of any?

I think this possible, but a such device can't share the bus with a smbus
alert capable device.
> 
> 
> Jonathan
> 
> > 
> > Signed-off-by: Marc CAPDEVILLE <m.capdeville@no-log.org>
> > ---
> >  drivers/i2c/i2c-core-acpi.c | 23 +++++++++++++++++++++--
> >  include/linux/i2c.h         |  1 +
> >  2 files changed, 22 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/i2c/i2c-core-acpi.c b/drivers/i2c/i2c-core-acpi.c
> > index a9126b3cda61..5a8886f14329 100644
> > --- a/drivers/i2c/i2c-core-acpi.c
> > +++ b/drivers/i2c/i2c-core-acpi.c
> > @@ -59,8 +59,14 @@ static int i2c_acpi_fill_info(struct acpi_resource *ares,
> > void *data)
> >  	if (sb->type != ACPI_RESOURCE_SERIAL_TYPE_I2C)
> >  		return 1;
> >  
> > -	if (lookup->index != -1 && lookup->n++ != lookup->index)
> > -		return 1;
> > +	if (lookup->index != -1) {
> > +		if (lookup->n++ != lookup->index)
> > +			return 1;
> > +	} else {
> > +		if (lookup->info->flags & I2C_CLIENT_ALERT &&
> > +		    sb->slave_address == 0x0c)
> > +			return 1;
> > +	}
> >  
> >  	status = acpi_get_handle(lookup->device_handle,
> >  				 sb->resource_source.string_ptr,
> > @@ -85,6 +91,15 @@ static const struct acpi_device_id
> > i2c_acpi_ignored_device_ids[] = {
> >  	{}
> >  };
> >  
> > +static const struct acpi_device_id i2c_acpi_alert_device_ids[] = {
> > +	/*
> > +	 * Smbus alert capable device which may have the reserved ARA
> > address
> > +	 * in their serial bus resources list.
> > +	 */
> > +	{ "CPLM3218", 0 },
> > +	{}
> > +};
> > +
> >  static int i2c_acpi_do_lookup(struct acpi_device *adev,
> >  			      struct i2c_acpi_lookup *lookup)
> >  {
> > @@ -100,6 +115,10 @@ static int i2c_acpi_do_lookup(struct acpi_device *adev,
> >  		return -ENODEV;
> >  
> >  	memset(info, 0, sizeof(*info));
> > +
> > +	if (acpi_match_device_ids(adev, i2c_acpi_alert_device_ids) == 0)
> > +		info->flags |= I2C_CLIENT_ALERT;
> > +
> >  	lookup->device_handle = acpi_device_handle(adev);
> >  
> >  	/* Look up for I2cSerialBus resource */
> > diff --git a/include/linux/i2c.h b/include/linux/i2c.h
> > index 7592dce12923..b0d6f1333442 100644
> > --- a/include/linux/i2c.h
> > +++ b/include/linux/i2c.h
> > @@ -743,6 +743,7 @@ i2c_unlock_adapter(struct i2c_adapter *adapter)
> >  #define I2C_CLIENT_SLAVE	0x20	/* we are the slave */
> >  #define I2C_CLIENT_HOST_NOTIFY	0x40	/* We want to use I2C
> > host notify */
> >  #define I2C_CLIENT_WAKE		0x80	/* for board_info; true
> > iff can wake */
> > +#define I2C_CLIENT_ALERT	0x100	/* Client use SMBUS alert
> > protocol */
> >  #define I2C_CLIENT_SCCB		0x9000	/* Use Omnivision SCCB
> > protocol */
> >  					/* Must match I2C_M_STOP|IGNORE_NAK
> > */
> >  
> 
>
Jonathan Cameron Jan. 20, 2018, 4:30 p.m. | #3
On Wed, 17 Jan 2018 11:28:01 +0100
CAPDEVILLE Marc <m.capdeville@no-log.org> wrote:

> Le dimanche 14 janvier 2018 à 11:28 +0000, Jonathan Cameron a écrit :
> > On Sat, 13 Jan 2018 14:37:03 +0100
> > Marc CAPDEVILLE <m.capdeville@no-log.org> wrote:
> >   
> > > Somme ACPI enumerated devices are known to support smbus alert protocol.
> > > Theses devices may be miss-enumerated with the reserved smbus ARA address.
> > > 
> > > This is the case on Asus T100 tablet where cm3218 ambiant light sensor
> > > expose two i2c serial bus connections, with the first one being the alert
> > > response address.
> > > 
> > > This patch make a match on known ACPI ids for which devices are smbus ARA
> > > capable, then skip the connection if it has the reserved 0x0c address and
> > > mark it with I2C_CLIENT_ALERT flag. So device is enumerated with the
> > > correct address.  
> > 
> > I wonder if we are safe to always skip 0x0c address whether or not
> > we know the device supports ARA.  The exception may be for devices
> > that do support ARA but are rolling their own support the hard way...  
> 
> The device is only flagged with I2C_CLIENT_ALERT if its ID is in the
> i2c_acpi_alert_device_ids. IF a driver want to be enumerated with the
> reserved Ox0c address, It just have to not put its Id in this table. But
> this may break with registering the smbus_alert device as the address will
> be marked as busy for this adapter.
> 
> > I suppose it's possible there are i2c devices (not smbus where the
> > spec says you must not use 0x0c for normal address IIRC) that use
> > this address.  Does anyone know of any?  
> 
> I think this possible, but a such device can't share the bus with a smbus
> alert capable device.

Hmm. It seems it would also be broken under the i2c specifications:
https://www.totalphase.com/support/articles/200349176-7-bit-8-bit-and-10-bit-I2C-Slave-Addressing

There are a load of reserved addresses that a device is not allowed
to use and this is one of them.

Not that it really restricts things given the number of implementations
that definitely aren't i2c but just look like it ;)

> > 
> > 
> > Jonathan
> >   
> > > 
> > > Signed-off-by: Marc CAPDEVILLE <m.capdeville@no-log.org>
> > > ---
> > >  drivers/i2c/i2c-core-acpi.c | 23 +++++++++++++++++++++--
> > >  include/linux/i2c.h         |  1 +
> > >  2 files changed, 22 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/i2c/i2c-core-acpi.c b/drivers/i2c/i2c-core-acpi.c
> > > index a9126b3cda61..5a8886f14329 100644
> > > --- a/drivers/i2c/i2c-core-acpi.c
> > > +++ b/drivers/i2c/i2c-core-acpi.c
> > > @@ -59,8 +59,14 @@ static int i2c_acpi_fill_info(struct acpi_resource *ares,
> > > void *data)
> > >  	if (sb->type != ACPI_RESOURCE_SERIAL_TYPE_I2C)
> > >  		return 1;
> > >  
> > > -	if (lookup->index != -1 && lookup->n++ != lookup->index)
> > > -		return 1;
> > > +	if (lookup->index != -1) {
> > > +		if (lookup->n++ != lookup->index)
> > > +			return 1;
> > > +	} else {
> > > +		if (lookup->info->flags & I2C_CLIENT_ALERT &&
> > > +		    sb->slave_address == 0x0c)
> > > +			return 1;
> > > +	}
> > >  
> > >  	status = acpi_get_handle(lookup->device_handle,
> > >  				 sb->resource_source.string_ptr,
> > > @@ -85,6 +91,15 @@ static const struct acpi_device_id
> > > i2c_acpi_ignored_device_ids[] = {
> > >  	{}
> > >  };
> > >  
> > > +static const struct acpi_device_id i2c_acpi_alert_device_ids[] = {
> > > +	/*
> > > +	 * Smbus alert capable device which may have the reserved ARA
> > > address
> > > +	 * in their serial bus resources list.
> > > +	 */
> > > +	{ "CPLM3218", 0 },
> > > +	{}
> > > +};
> > > +
> > >  static int i2c_acpi_do_lookup(struct acpi_device *adev,
> > >  			      struct i2c_acpi_lookup *lookup)
> > >  {
> > > @@ -100,6 +115,10 @@ static int i2c_acpi_do_lookup(struct acpi_device *adev,
> > >  		return -ENODEV;
> > >  
> > >  	memset(info, 0, sizeof(*info));
> > > +
> > > +	if (acpi_match_device_ids(adev, i2c_acpi_alert_device_ids) == 0)
> > > +		info->flags |= I2C_CLIENT_ALERT;
> > > +
> > >  	lookup->device_handle = acpi_device_handle(adev);
> > >  
> > >  	/* Look up for I2cSerialBus resource */
> > > diff --git a/include/linux/i2c.h b/include/linux/i2c.h
> > > index 7592dce12923..b0d6f1333442 100644
> > > --- a/include/linux/i2c.h
> > > +++ b/include/linux/i2c.h
> > > @@ -743,6 +743,7 @@ i2c_unlock_adapter(struct i2c_adapter *adapter)
> > >  #define I2C_CLIENT_SLAVE	0x20	/* we are the slave */
> > >  #define I2C_CLIENT_HOST_NOTIFY	0x40	/* We want to use I2C
> > > host notify */
> > >  #define I2C_CLIENT_WAKE		0x80	/* for board_info; true
> > > iff can wake */
> > > +#define I2C_CLIENT_ALERT	0x100	/* Client use SMBUS alert
> > > protocol */
> > >  #define I2C_CLIENT_SCCB		0x9000	/* Use Omnivision SCCB
> > > protocol */
> > >  					/* Must match I2C_M_STOP|IGNORE_NAK
> > > */
> > >    
> > 
> >

Patch

diff --git a/drivers/i2c/i2c-core-acpi.c b/drivers/i2c/i2c-core-acpi.c
index a9126b3cda61..5a8886f14329 100644
--- a/drivers/i2c/i2c-core-acpi.c
+++ b/drivers/i2c/i2c-core-acpi.c
@@ -59,8 +59,14 @@  static int i2c_acpi_fill_info(struct acpi_resource *ares, void *data)
 	if (sb->type != ACPI_RESOURCE_SERIAL_TYPE_I2C)
 		return 1;
 
-	if (lookup->index != -1 && lookup->n++ != lookup->index)
-		return 1;
+	if (lookup->index != -1) {
+		if (lookup->n++ != lookup->index)
+			return 1;
+	} else {
+		if (lookup->info->flags & I2C_CLIENT_ALERT &&
+		    sb->slave_address == 0x0c)
+			return 1;
+	}
 
 	status = acpi_get_handle(lookup->device_handle,
 				 sb->resource_source.string_ptr,
@@ -85,6 +91,15 @@  static const struct acpi_device_id i2c_acpi_ignored_device_ids[] = {
 	{}
 };
 
+static const struct acpi_device_id i2c_acpi_alert_device_ids[] = {
+	/*
+	 * Smbus alert capable device which may have the reserved ARA address
+	 * in their serial bus resources list.
+	 */
+	{ "CPLM3218", 0 },
+	{}
+};
+
 static int i2c_acpi_do_lookup(struct acpi_device *adev,
 			      struct i2c_acpi_lookup *lookup)
 {
@@ -100,6 +115,10 @@  static int i2c_acpi_do_lookup(struct acpi_device *adev,
 		return -ENODEV;
 
 	memset(info, 0, sizeof(*info));
+
+	if (acpi_match_device_ids(adev, i2c_acpi_alert_device_ids) == 0)
+		info->flags |= I2C_CLIENT_ALERT;
+
 	lookup->device_handle = acpi_device_handle(adev);
 
 	/* Look up for I2cSerialBus resource */
diff --git a/include/linux/i2c.h b/include/linux/i2c.h
index 7592dce12923..b0d6f1333442 100644
--- a/include/linux/i2c.h
+++ b/include/linux/i2c.h
@@ -743,6 +743,7 @@  i2c_unlock_adapter(struct i2c_adapter *adapter)
 #define I2C_CLIENT_SLAVE	0x20	/* we are the slave */
 #define I2C_CLIENT_HOST_NOTIFY	0x40	/* We want to use I2C host notify */
 #define I2C_CLIENT_WAKE		0x80	/* for board_info; true iff can wake */
+#define I2C_CLIENT_ALERT	0x100	/* Client use SMBUS alert protocol */
 #define I2C_CLIENT_SCCB		0x9000	/* Use Omnivision SCCB protocol */
 					/* Must match I2C_M_STOP|IGNORE_NAK */