2013-10-10 08:39:56

by Ionut Nicu

[permalink] [raw]
Subject: [PATCH 2/2] i2c-mux-gpio: eliminate i2c channel order assumptions

The i2c-mux driver uses the chan_id parameter provided
in i2c_add_mux_adapter as a parameter to the select
and deselect callbacks while the i2c-mux-gpio driver
uses the chan_id as an index in the mux->data.values
array.

A simple example of where this doesn't work is when we
have a device tree like this:

i2cmux {
i2c@1 {
reg = <1>;
...
};

i2c@0 {
reg = <0>;
...
};
};

The mux->data.values array will be { 1, 0 }, but when
the i2-mux driver will try to select channel 0, the
i2c-mux-gpio driver will actually use values[0], hence 1
as the gpio selection value.

Signed-off-by: Ionut Nicu <[email protected]>
---
drivers/i2c/muxes/i2c-mux-gpio.c | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/i2c/muxes/i2c-mux-gpio.c b/drivers/i2c/muxes/i2c-mux-gpio.c
index b5f17ef..3505d0e 100644
--- a/drivers/i2c/muxes/i2c-mux-gpio.c
+++ b/drivers/i2c/muxes/i2c-mux-gpio.c
@@ -43,7 +43,7 @@ static int i2c_mux_gpio_select(struct i2c_adapter *adap, void *data, u32 chan)
{
struct gpiomux *mux = data;

- i2c_mux_gpio_set(mux, mux->data.values[chan]);
+ i2c_mux_gpio_set(mux, chan);

return 0;
}
@@ -233,7 +233,7 @@ static int i2c_mux_gpio_probe(struct platform_device *pdev)
unsigned int class = mux->data.classes ? mux->data.classes[i] : 0;

mux->adap[i] = i2c_add_mux_adapter(parent, &pdev->dev, mux, nr,
- i, class,
+ mux->data.values[i], class,
i2c_mux_gpio_select, deselect);
if (!mux->adap[i]) {
ret = -ENODEV;
--
1.7.1


2013-10-10 10:34:39

by Alexander Sverdlin

[permalink] [raw]
Subject: Re: [PATCH 2/2] i2c-mux-gpio: eliminate i2c channel order assumptions

Hi!

On 10/10/2013 10:39 AM, Ionut Nicu wrote:
> The i2c-mux driver uses the chan_id parameter provided
> in i2c_add_mux_adapter as a parameter to the select
> and deselect callbacks while the i2c-mux-gpio driver
> uses the chan_id as an index in the mux->data.values
> array.
>
> A simple example of where this doesn't work is when we
> have a device tree like this:
>
> i2cmux {
> i2c@1 {
> reg = <1>;
> ...
> };
>
> i2c@0 {
> reg = <0>;
> ...
> };
> };
>
> The mux->data.values array will be { 1, 0 }, but when
> the i2-mux driver will try to select channel 0, the
> i2c-mux-gpio driver will actually use values[0], hence 1
> as the gpio selection value.

The patch itself is correct, but the description is not precise,
I suppose... i2c-mux-gpio is consistent inside itself, it will
receive for every child adapter the value it has configured.
The problem happens inside i2c-mux.c, i2c_add_mux_adapter():

for_each_child_of_node(mux_dev->of_node, child) {
ret = of_property_read_u32(child, "reg", &reg);
if (ret)
continue;
if (chan_id == reg) {
priv->adap.dev.of_node = child;

Which means, i2c-mux-gpio MUST pass reg, not its logical index inside
array. Otherwise node will not be correctly assigned and i2c-mux will
have problems selecting right adapter for the multiplexed devices.

> Signed-off-by: Ionut Nicu <[email protected]>

So, for the code itself

Acked-by: Alexander Sverdlin <[email protected]>

> ---
> drivers/i2c/muxes/i2c-mux-gpio.c | 4 ++--
> 1 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/i2c/muxes/i2c-mux-gpio.c b/drivers/i2c/muxes/i2c-mux-gpio.c
> index b5f17ef..3505d0e 100644
> --- a/drivers/i2c/muxes/i2c-mux-gpio.c
> +++ b/drivers/i2c/muxes/i2c-mux-gpio.c
> @@ -43,7 +43,7 @@ static int i2c_mux_gpio_select(struct i2c_adapter *adap, void *data, u32 chan)
> {
> struct gpiomux *mux = data;
>
> - i2c_mux_gpio_set(mux, mux->data.values[chan]);
> + i2c_mux_gpio_set(mux, chan);
>
> return 0;
> }
> @@ -233,7 +233,7 @@ static int i2c_mux_gpio_probe(struct platform_device *pdev)
> unsigned int class = mux->data.classes ? mux->data.classes[i] : 0;
>
> mux->adap[i] = i2c_add_mux_adapter(parent, &pdev->dev, mux, nr,
> - i, class,
> + mux->data.values[i], class,
> i2c_mux_gpio_select, deselect);
> if (!mux->adap[i]) {
> ret = -ENODEV;
>

--
Best regards,
Alexander Sverdlin.

2013-10-11 08:46:12

by Ionut Nicu

[permalink] [raw]
Subject: Re: [PATCH 2/2] i2c-mux-gpio: eliminate i2c channel order assumptions

Hi,

On 10.10.2013 12:34, Alexander Sverdlin wrote:
> Hi!
>
> On 10/10/2013 10:39 AM, Ionut Nicu wrote:
>> The i2c-mux driver uses the chan_id parameter provided
>> in i2c_add_mux_adapter as a parameter to the select
>> and deselect callbacks while the i2c-mux-gpio driver
>> uses the chan_id as an index in the mux->data.values
>> array.
>>
>> A simple example of where this doesn't work is when we
>> have a device tree like this:
>>
>> i2cmux {
>> i2c@1 {
>> reg = <1>;
>> ...
>> };
>>
>> i2c@0 {
>> reg = <0>;
>> ...
>> };
>> };
>>
>> The mux->data.values array will be { 1, 0 }, but when
>> the i2-mux driver will try to select channel 0, the
>> i2c-mux-gpio driver will actually use values[0], hence 1
>> as the gpio selection value.
>
> The patch itself is correct, but the description is not precise,
> I suppose... i2c-mux-gpio is consistent inside itself, it will
> receive for every child adapter the value it has configured.
> The problem happens inside i2c-mux.c, i2c_add_mux_adapter():
>
> for_each_child_of_node(mux_dev->of_node, child) {
> ret = of_property_read_u32(child, "reg", &reg);
> if (ret)
> continue;
> if (chan_id == reg) {
> priv->adap.dev.of_node = child;
>
> Which means, i2c-mux-gpio MUST pass reg, not its logical index inside
> array. Otherwise node will not be correctly assigned and i2c-mux will
> have problems selecting right adapter for the multiplexed devices.
>
>> Signed-off-by: Ionut Nicu <[email protected]>
>
> So, for the code itself
>
> Acked-by: Alexander Sverdlin <[email protected]>
>

You are right, the patch description is not so good. I will try to change
it so it's clearer for everyone what I'm trying to fix here and after that
I will re-submit the series.

>> ---
>> drivers/i2c/muxes/i2c-mux-gpio.c | 4 ++--
>> 1 files changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/i2c/muxes/i2c-mux-gpio.c b/drivers/i2c/muxes/i2c-mux-gpio.c
>> index b5f17ef..3505d0e 100644
>> --- a/drivers/i2c/muxes/i2c-mux-gpio.c
>> +++ b/drivers/i2c/muxes/i2c-mux-gpio.c
>> @@ -43,7 +43,7 @@ static int i2c_mux_gpio_select(struct i2c_adapter *adap, void *data, u32 chan)
>> {
>> struct gpiomux *mux = data;
>>
>> - i2c_mux_gpio_set(mux, mux->data.values[chan]);
>> + i2c_mux_gpio_set(mux, chan);
>>
>> return 0;
>> }
>> @@ -233,7 +233,7 @@ static int i2c_mux_gpio_probe(struct platform_device *pdev)
>> unsigned int class = mux->data.classes ? mux->data.classes[i] : 0;
>>
>> mux->adap[i] = i2c_add_mux_adapter(parent, &pdev->dev, mux, nr,
>> - i, class,
>> + mux->data.values[i], class,
>> i2c_mux_gpio_select, deselect);
>> if (!mux->adap[i]) {
>> ret = -ENODEV;
>>
>