[v5,2/2] i2c: mv64xxx: add an optional bus-reset-gpios property

Message ID 20231027033104.1348921-3-chris.packham@alliedtelesis.co.nz
State New
Headers
Series i2c: mv64xxx: bus-reset-gpios |

Commit Message

Chris Packham Oct. 27, 2023, 3:31 a.m. UTC
  Some hardware designs have a GPIO used to control the reset of all the
devices on and I2C bus. It's not possible for every child node to
declare a reset-gpios property as only the first device probed would be
able to successfully request it (the others will get -EBUSY). Represent
this kind of hardware design by associating the bus-reset-gpios with the
parent I2C bus. The reset line will be released prior to the child I2C
devices being probed.

Signed-off-by: Chris Packham <chris.packham@alliedtelesis.co.nz>
---

Notes:
    Changes in v5:
    - Rename reset-gpios and reset-duration-us to bus-reset-gpios and
      bus-reset-duration-us as requested by Wolfram
    Changes in v4:
    - Add missing gpio/consumer.h
    - use fsleep() for enforcing reset-duration
    Changes in v3:
    - Rename reset-delay to reset-duration
    - Use reset-duration-us property to control the reset pulse rather than
      delaying after the reset
    Changes in v2:
    - Add a property to cover the length of delay after releasing the reset
      GPIO
    - Use dev_err_probe() when requesing the GPIO fails

 drivers/i2c/busses/i2c-mv64xxx.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
  

Comments

Andi Shyti Oct. 27, 2023, 8:48 a.m. UTC | #1
Hi Chris,

On Fri, Oct 27, 2023 at 04:31:04PM +1300, Chris Packham wrote:
> Some hardware designs have a GPIO used to control the reset of all the
> devices on and I2C bus. It's not possible for every child node to
> declare a reset-gpios property as only the first device probed would be
> able to successfully request it (the others will get -EBUSY). Represent
> this kind of hardware design by associating the bus-reset-gpios with the
> parent I2C bus. The reset line will be released prior to the child I2C
> devices being probed.
> 
> Signed-off-by: Chris Packham <chris.packham@alliedtelesis.co.nz>
> ---
> 
> Notes:
>     Changes in v5:
>     - Rename reset-gpios and reset-duration-us to bus-reset-gpios and
>       bus-reset-duration-us as requested by Wolfram

for such change you could have kept my r-b:

Reviewed-by: Andi Shyti <andi.shyti@kernel.org>

Andi
  
Krzysztof Kozlowski Oct. 27, 2023, 11:27 a.m. UTC | #2
On 27/10/2023 05:31, Chris Packham wrote:
> Some hardware designs have a GPIO used to control the reset of all the
> devices on and I2C bus. It's not possible for every child node to
> declare a reset-gpios property as only the first device probed would be
> able to successfully request it (the others will get -EBUSY). Represent
> this kind of hardware design by associating the bus-reset-gpios with the
> parent I2C bus. The reset line will be released prior to the child I2C
> devices being probed.
> 
> Signed-off-by: Chris Packham <chris.packham@alliedtelesis.co.nz>
> ---
> 
> Notes:
>     Changes in v5:
>     - Rename reset-gpios and reset-duration-us to bus-reset-gpios and
>       bus-reset-duration-us as requested by Wolfram
>     Changes in v4:
>     - Add missing gpio/consumer.h
>     - use fsleep() for enforcing reset-duration
>     Changes in v3:
>     - Rename reset-delay to reset-duration
>     - Use reset-duration-us property to control the reset pulse rather than
>       delaying after the reset
>     Changes in v2:
>     - Add a property to cover the length of delay after releasing the reset
>       GPIO
>     - Use dev_err_probe() when requesing the GPIO fails
> 
>  drivers/i2c/busses/i2c-mv64xxx.c | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
> 
> diff --git a/drivers/i2c/busses/i2c-mv64xxx.c b/drivers/i2c/busses/i2c-mv64xxx.c
> index efd28bbecf61..6e2762d22e5a 100644
> --- a/drivers/i2c/busses/i2c-mv64xxx.c
> +++ b/drivers/i2c/busses/i2c-mv64xxx.c
> @@ -13,6 +13,7 @@
>  #include <linux/slab.h>
>  #include <linux/module.h>
>  #include <linux/spinlock.h>
> +#include <linux/gpio/consumer.h>
>  #include <linux/i2c.h>
>  #include <linux/interrupt.h>
>  #include <linux/mv643xx_i2c.h>
> @@ -160,6 +161,7 @@ struct mv64xxx_i2c_data {
>  	bool			clk_n_base_0;
>  	struct i2c_bus_recovery_info	rinfo;
>  	bool			atomic;
> +	struct gpio_desc	*reset_gpio;
>  };
>  
>  static struct mv64xxx_i2c_regs mv64xxx_i2c_regs_mv64xxx = {
> @@ -1036,6 +1038,7 @@ mv64xxx_i2c_probe(struct platform_device *pd)
>  	struct mv64xxx_i2c_data		*drv_data;
>  	struct mv64xxx_i2c_pdata	*pdata = dev_get_platdata(&pd->dev);
>  	struct resource *res;
> +	u32	reset_duration;
>  	int	rc;
>  
>  	if ((!pdata && !pd->dev.of_node))
> @@ -1083,6 +1086,14 @@ mv64xxx_i2c_probe(struct platform_device *pd)
>  	if (drv_data->irq < 0)
>  		return drv_data->irq;
>  
> +	drv_data->reset_gpio = devm_gpiod_get_optional(&pd->dev, "bus-reset", GPIOD_OUT_HIGH);
> +	if (IS_ERR(drv_data->reset_gpio))
> +		return dev_err_probe(&pd->dev, PTR_ERR(drv_data->reset_gpio),
> +				     "Cannot get reset gpio\n");
> +	rc = device_property_read_u32(&pd->dev, "bus-reset-duration-us", &reset_duration);
> +	if (rc)
> +		reset_duration = 1;

No, this should be solved by core - for entire I2C at minimum. This is
not specific to this device.

Best regards,
Krzysztof
  
Krzysztof Kozlowski Oct. 27, 2023, 11:37 a.m. UTC | #3
On 27/10/2023 13:27, Krzysztof Kozlowski wrote:
> On 27/10/2023 05:31, Chris Packham wrote:
>> Some hardware designs have a GPIO used to control the reset of all the
>> devices on and I2C bus. It's not possible for every child node to
>> declare a reset-gpios property as only the first device probed would be
>> able to successfully request it (the others will get -EBUSY). Represent

Cc: Mark,

Also this part is not true. If the bus is non-discoverable, then it is
possible to have reset-gpios in each probed device. You can share GPIOs,
so no problem with -EBUSY at all.

The problem is doing reset:
1. in proper moment for all devices
2. without affecting other devices when one unbinds/remove()

The (2) above is not solveable easy in kernel and we already had nice
talks about it just few days ago:
1. Apple case:
https://social.treehouse.systems/@marcan/111268780311634160

2. my WSA884x:
https://lore.kernel.org/alsa-devel/84f9f1c4-0627-4986-8160-b4ab99469b81@linaro.org/

Last,
I would like to apologize to you Chris. I understand that bringing such
feedback at v5 is not that good. I had plenty of time to say something
earlier, so this is not really professional from my side. I am sorry,
just my brain did not connect all these topics together.

I apologize.

Best regards,
Krzysztof

>> this kind of hardware design by associating the bus-reset-gpios with the
>> parent I2C bus. The reset line will be released prior to the child I2C
>> devices being probed.
>>
>> Signed-off-by: Chris Packham <chris.packham@alliedtelesis.co.nz>
>> ---
>>
>> Notes:
>>     Changes in v5:
>>     - Rename reset-gpios and reset-duration-us to bus-reset-gpios and
>>       bus-reset-duration-us as requested by Wolfram
>>     Changes in v4:
>>     - Add missing gpio/consumer.h
>>     - use fsleep() for enforcing reset-duration
>>     Changes in v3:
>>     - Rename reset-delay to reset-duration
>>     - Use reset-duration-us property to control the reset pulse rather than
>>       delaying after the reset
>>     Changes in v2:
>>     - Add a property to cover the length of delay after releasing the reset
>>       GPIO
>>     - Use dev_err_probe() when requesing the GPIO fails
>>
>>  drivers/i2c/busses/i2c-mv64xxx.c | 16 ++++++++++++++++
>>  1 file changed, 16 insertions(+)
>>
>> diff --git a/drivers/i2c/busses/i2c-mv64xxx.c b/drivers/i2c/busses/i2c-mv64xxx.c
>> index efd28bbecf61..6e2762d22e5a 100644
>> --- a/drivers/i2c/busses/i2c-mv64xxx.c
>> +++ b/drivers/i2c/busses/i2c-mv64xxx.c
>> @@ -13,6 +13,7 @@
>>  #include <linux/slab.h>
>>  #include <linux/module.h>
>>  #include <linux/spinlock.h>
>> +#include <linux/gpio/consumer.h>
>>  #include <linux/i2c.h>
>>  #include <linux/interrupt.h>
>>  #include <linux/mv643xx_i2c.h>
>> @@ -160,6 +161,7 @@ struct mv64xxx_i2c_data {
>>  	bool			clk_n_base_0;
>>  	struct i2c_bus_recovery_info	rinfo;
>>  	bool			atomic;
>> +	struct gpio_desc	*reset_gpio;
>>  };
>>  
>>  static struct mv64xxx_i2c_regs mv64xxx_i2c_regs_mv64xxx = {
>> @@ -1036,6 +1038,7 @@ mv64xxx_i2c_probe(struct platform_device *pd)
>>  	struct mv64xxx_i2c_data		*drv_data;
>>  	struct mv64xxx_i2c_pdata	*pdata = dev_get_platdata(&pd->dev);
>>  	struct resource *res;
>> +	u32	reset_duration;
>>  	int	rc;
>>  
>>  	if ((!pdata && !pd->dev.of_node))
>> @@ -1083,6 +1086,14 @@ mv64xxx_i2c_probe(struct platform_device *pd)
>>  	if (drv_data->irq < 0)
>>  		return drv_data->irq;
>>  
>> +	drv_data->reset_gpio = devm_gpiod_get_optional(&pd->dev, "bus-reset", GPIOD_OUT_HIGH);
>> +	if (IS_ERR(drv_data->reset_gpio))
>> +		return dev_err_probe(&pd->dev, PTR_ERR(drv_data->reset_gpio),
>> +				     "Cannot get reset gpio\n");
>> +	rc = device_property_read_u32(&pd->dev, "bus-reset-duration-us", &reset_duration);
>> +	if (rc)
>> +		reset_duration = 1;
> 
> No, this should be solved by core - for entire I2C at minimum. This is
> not specific to this device.
  
Andi Shyti Oct. 27, 2023, 12:55 p.m. UTC | #4
Hi Krzysztof,

On Fri, Oct 27, 2023 at 01:37:05PM +0200, Krzysztof Kozlowski wrote:
> On 27/10/2023 13:27, Krzysztof Kozlowski wrote:
> > On 27/10/2023 05:31, Chris Packham wrote:
> >> Some hardware designs have a GPIO used to control the reset of all the
> >> devices on and I2C bus. It's not possible for every child node to
> >> declare a reset-gpios property as only the first device probed would be
> >> able to successfully request it (the others will get -EBUSY). Represent
> 
> Cc: Mark,
> 
> Also this part is not true. If the bus is non-discoverable, then it is
> possible to have reset-gpios in each probed device. You can share GPIOs,
> so no problem with -EBUSY at all.
> 
> The problem is doing reset:
> 1. in proper moment for all devices
> 2. without affecting other devices when one unbinds/remove()

yes, I thought that we could get to this point, but I did not
object the patch as I didn't see an immediate better solution. I
would still be OK to merge it until we develop something better.

Let me mull this over and will be back to the topic.

Thanks, Krzysztof!
Andi

> The (2) above is not solveable easy in kernel and we already had nice
> talks about it just few days ago:
> 1. Apple case:
> https://social.treehouse.systems/@marcan/111268780311634160
> 
> 2. my WSA884x:
> https://lore.kernel.org/alsa-devel/84f9f1c4-0627-4986-8160-b4ab99469b81@linaro.org/
> 
> Last,
> I would like to apologize to you Chris. I understand that bringing such
> feedback at v5 is not that good. I had plenty of time to say something
> earlier, so this is not really professional from my side. I am sorry,
> just my brain did not connect all these topics together.
> 
> I apologize.
> 
> Best regards,
> Krzysztof
  
Mark Brown Oct. 27, 2023, 3:10 p.m. UTC | #5
On Fri, Oct 27, 2023 at 01:37:05PM +0200, Krzysztof Kozlowski wrote:

> The (2) above is not solveable easy in kernel and we already had nice
> talks about it just few days ago:
> 1. Apple case:
> https://social.treehouse.systems/@marcan/111268780311634160

Note that the Apple case is not a reset, it's a low power mode for the
device, though at the GPIO level the beheviour with tying together
multiple devices that need to coordinate their usage looks very similar.
  
Chris Packham Oct. 29, 2023, 8:48 p.m. UTC | #6
On 28/10/23 00:37, Krzysztof Kozlowski wrote:
> On 27/10/2023 13:27, Krzysztof Kozlowski wrote:
>> On 27/10/2023 05:31, Chris Packham wrote:
>>> Some hardware designs have a GPIO used to control the reset of all the
>>> devices on and I2C bus. It's not possible for every child node to
>>> declare a reset-gpios property as only the first device probed would be
>>> able to successfully request it (the others will get -EBUSY). Represent
> Cc: Mark,
>
> Also this part is not true. If the bus is non-discoverable, then it is
> possible to have reset-gpios in each probed device. You can share GPIOs,
> so no problem with -EBUSY at all.

Last time I checked you couldn't share GPIOs. If that's no longer the 
case then I can probably make what I need to happen work. It still 
creates an issue that I have multiple PCA954x muxes connected to a 
common reset GPIO so as each mux is probed the PCA954x driver will 
toggle the reset. That's probably OK as the PCA954x is sufficiently 
stateless that the extra resets won't do any harm but if it were a more 
complicated device then there would be issues.

Having some kind of ref-counted reset controller that is implemented 
with GPIOs is probably the better solution. I was kind of surprised that 
nothing existed like that in drivers/reset.

> The problem is doing reset:
> 1. in proper moment for all devices
> 2. without affecting other devices when one unbinds/remove()
>
> The (2) above is not solveable easy in kernel and we already had nice
> talks about it just few days ago:
> 1. Apple case:
> https://scanmail.trustwave.com/?c=20988&d=6qC75SLs-9PNM1ZHpLa6reGv82R6opEUmyI62vCytQ&u=https%3a%2f%2fsocial%2etreehouse%2esystems%2f%40marcan%2f111268780311634160
>
> 2. my WSA884x:
> https://scanmail.trustwave.com/?c=20988&d=6qC75SLs-9PNM1ZHpLa6reGv82R6opEUmyJk3q3j7g&u=https%3a%2f%2flore%2ekernel%2eorg%2falsa-devel%2f84f9f1c4-0627-4986-8160-b4ab99469b81%40linaro%2eorg%2f
Apologies for the mangled links (they're more secure now at least that's 
what our IS team have been sold).
> Last,
> I would like to apologize to you Chris. I understand that bringing such
> feedback at v5 is not that good. I had plenty of time to say something
> earlier, so this is not really professional from my side. I am sorry,
> just my brain did not connect all these topics together.
>
> I apologize.

Actually I kind of expected this feedback. I figured I could start with 
the driver that is currently causing me issues and once the dt-binding 
was considered good enough it might migrate to the i2c core.

>
> Best regards,
> Krzysztof
>
>>> this kind of hardware design by associating the bus-reset-gpios with the
>>> parent I2C bus. The reset line will be released prior to the child I2C
>>> devices being probed.
>>>
>>> Signed-off-by: Chris Packham <chris.packham@alliedtelesis.co.nz>
>>> ---
>>>
>>> Notes:
>>>      Changes in v5:
>>>      - Rename reset-gpios and reset-duration-us to bus-reset-gpios and
>>>        bus-reset-duration-us as requested by Wolfram
>>>      Changes in v4:
>>>      - Add missing gpio/consumer.h
>>>      - use fsleep() for enforcing reset-duration
>>>      Changes in v3:
>>>      - Rename reset-delay to reset-duration
>>>      - Use reset-duration-us property to control the reset pulse rather than
>>>        delaying after the reset
>>>      Changes in v2:
>>>      - Add a property to cover the length of delay after releasing the reset
>>>        GPIO
>>>      - Use dev_err_probe() when requesing the GPIO fails
>>>
>>>   drivers/i2c/busses/i2c-mv64xxx.c | 16 ++++++++++++++++
>>>   1 file changed, 16 insertions(+)
>>>
>>> diff --git a/drivers/i2c/busses/i2c-mv64xxx.c b/drivers/i2c/busses/i2c-mv64xxx.c
>>> index efd28bbecf61..6e2762d22e5a 100644
>>> --- a/drivers/i2c/busses/i2c-mv64xxx.c
>>> +++ b/drivers/i2c/busses/i2c-mv64xxx.c
>>> @@ -13,6 +13,7 @@
>>>   #include <linux/slab.h>
>>>   #include <linux/module.h>
>>>   #include <linux/spinlock.h>
>>> +#include <linux/gpio/consumer.h>
>>>   #include <linux/i2c.h>
>>>   #include <linux/interrupt.h>
>>>   #include <linux/mv643xx_i2c.h>
>>> @@ -160,6 +161,7 @@ struct mv64xxx_i2c_data {
>>>   	bool			clk_n_base_0;
>>>   	struct i2c_bus_recovery_info	rinfo;
>>>   	bool			atomic;
>>> +	struct gpio_desc	*reset_gpio;
>>>   };
>>>   
>>>   static struct mv64xxx_i2c_regs mv64xxx_i2c_regs_mv64xxx = {
>>> @@ -1036,6 +1038,7 @@ mv64xxx_i2c_probe(struct platform_device *pd)
>>>   	struct mv64xxx_i2c_data		*drv_data;
>>>   	struct mv64xxx_i2c_pdata	*pdata = dev_get_platdata(&pd->dev);
>>>   	struct resource *res;
>>> +	u32	reset_duration;
>>>   	int	rc;
>>>   
>>>   	if ((!pdata && !pd->dev.of_node))
>>> @@ -1083,6 +1086,14 @@ mv64xxx_i2c_probe(struct platform_device *pd)
>>>   	if (drv_data->irq < 0)
>>>   		return drv_data->irq;
>>>   
>>> +	drv_data->reset_gpio = devm_gpiod_get_optional(&pd->dev, "bus-reset", GPIOD_OUT_HIGH);
>>> +	if (IS_ERR(drv_data->reset_gpio))
>>> +		return dev_err_probe(&pd->dev, PTR_ERR(drv_data->reset_gpio),
>>> +				     "Cannot get reset gpio\n");
>>> +	rc = device_property_read_u32(&pd->dev, "bus-reset-duration-us", &reset_duration);
>>> +	if (rc)
>>> +		reset_duration = 1;
>> No, this should be solved by core - for entire I2C at minimum. This is
>> not specific to this device.
  
Chris Packham Oct. 29, 2023, 9:02 p.m. UTC | #7
On 28/10/23 01:55, Andi Shyti wrote:
> Hi Krzysztof,
>
> On Fri, Oct 27, 2023 at 01:37:05PM +0200, Krzysztof Kozlowski wrote:
>> On 27/10/2023 13:27, Krzysztof Kozlowski wrote:
>>> On 27/10/2023 05:31, Chris Packham wrote:
>>>> Some hardware designs have a GPIO used to control the reset of all the
>>>> devices on and I2C bus. It's not possible for every child node to
>>>> declare a reset-gpios property as only the first device probed would be
>>>> able to successfully request it (the others will get -EBUSY). Represent
>> Cc: Mark,
>>
>> Also this part is not true. If the bus is non-discoverable, then it is
>> possible to have reset-gpios in each probed device. You can share GPIOs,
>> so no problem with -EBUSY at all.
>>
>> The problem is doing reset:
>> 1. in proper moment for all devices
>> 2. without affecting other devices when one unbinds/remove()
> yes, I thought that we could get to this point, but I did not
> object the patch as I didn't see an immediate better solution. I
> would still be OK to merge it until we develop something better.
>
> Let me mull this over and will be back to the topic.

If we're happy with plain GPIOs I can move what I've done so far to 
somewhere in the I2C core. I know we've got other hardware designs with 
different controllers that also have muxes connected to a common reset 
GPIO so I would have ended up moving this code to I2C core eventually.

If we're talking a proper reset driver implemented using GPIOs then that 
might be a bit of bigger task.

> Thanks, Krzysztof!
> Andi
>
>> The (2) above is not solveable easy in kernel and we already had nice
>> talks about it just few days ago:
>> 1. Apple case:
>> https://scanmail.trustwave.com/?c=20988&d=1LO75R2nre1LP3TyEWMYg1Is4Mz-YROPQ8JxsJqwkg&u=https%3a%2f%2fsocial%2etreehouse%2esystems%2f%40marcan%2f111268780311634160
>>
>> 2. my WSA884x:
>> https://scanmail.trustwave.com/?c=20988&d=1LO75R2nre1LP3TyEWMYg1Is4Mz-YROPQ8IvtMfhyQ&u=https%3a%2f%2flore%2ekernel%2eorg%2falsa-devel%2f84f9f1c4-0627-4986-8160-b4ab99469b81%40linaro%2eorg%2f
>>
>> Last,
>> I would like to apologize to you Chris. I understand that bringing such
>> feedback at v5 is not that good. I had plenty of time to say something
>> earlier, so this is not really professional from my side. I am sorry,
>> just my brain did not connect all these topics together.
>>
>> I apologize.
>>
>> Best regards,
>> Krzysztof
  
Krzysztof Kozlowski Oct. 31, 2023, 6:01 a.m. UTC | #8
On 29/10/2023 21:48, Chris Packham wrote:
> 
> On 28/10/23 00:37, Krzysztof Kozlowski wrote:
>> On 27/10/2023 13:27, Krzysztof Kozlowski wrote:
>>> On 27/10/2023 05:31, Chris Packham wrote:
>>>> Some hardware designs have a GPIO used to control the reset of all the
>>>> devices on and I2C bus. It's not possible for every child node to
>>>> declare a reset-gpios property as only the first device probed would be
>>>> able to successfully request it (the others will get -EBUSY). Represent
>> Cc: Mark,
>>
>> Also this part is not true. If the bus is non-discoverable, then it is
>> possible to have reset-gpios in each probed device. You can share GPIOs,
>> so no problem with -EBUSY at all.
> 
> Last time I checked you couldn't share GPIOs. If that's no longer the 
> case then I can probably make what I need to happen work. It still 
> creates an issue that I have multiple PCA954x muxes connected to a 
> common reset GPIO so as each mux is probed the PCA954x driver will 
> toggle the reset. That's probably OK as the PCA954x is sufficiently 
> stateless that the extra resets won't do any harm but if it were a more 
> complicated device then there would be issues.

I know, but this is a broader problem, not really specific to this one
device. I also argue that your I2C controller does not actually have
this reset line.

> 
> Having some kind of ref-counted reset controller that is implemented 
> with GPIOs is probably the better solution. I was kind of surprised that 
> nothing existed like that in drivers/reset.

reset controller framework already supports this. The point is that GPIO
reset is not a reset controller, so in terms of bindings "resets"
property does not fit it.

> 
>> The problem is doing reset:
>> 1. in proper moment for all devices
>> 2. without affecting other devices when one unbinds/remove()
>>
>> The (2) above is not solveable easy in kernel and we already had nice
>> talks about it just few days ago:
>> 1. Apple case:
>> https://scanmail.trustwave.com/?c=20988&d=6qC75SLs-9PNM1ZHpLa6reGv82R6opEUmyI62vCytQ&u=https%3a%2f%2fsocial%2etreehouse%2esystems%2f%40marcan%2f111268780311634160
>>
>> 2. my WSA884x:
>> https://scanmail.trustwave.com/?c=20988&d=6qC75SLs-9PNM1ZHpLa6reGv82R6opEUmyJk3q3j7g&u=https%3a%2f%2flore%2ekernel%2eorg%2falsa-devel%2f84f9f1c4-0627-4986-8160-b4ab99469b81%40linaro%2eorg%2f
> Apologies for the mangled links (they're more secure now at least that's 
> what our IS team have been sold).
>> Last,
>> I would like to apologize to you Chris. I understand that bringing such
>> feedback at v5 is not that good. I had plenty of time to say something
>> earlier, so this is not really professional from my side. I am sorry,
>> just my brain did not connect all these topics together.
>>
>> I apologize.
> 
> Actually I kind of expected this feedback. I figured I could start with 
> the driver that is currently causing me issues and once the dt-binding 
> was considered good enough it might migrate to the i2c core.
> 
>>


Best regards,
Krzysztof
  
Chris Packham Oct. 31, 2023, 7:59 p.m. UTC | #9
On 31/10/23 19:01, Krzysztof Kozlowski wrote:
> On 29/10/2023 21:48, Chris Packham wrote:
>> On 28/10/23 00:37, Krzysztof Kozlowski wrote:
>>> On 27/10/2023 13:27, Krzysztof Kozlowski wrote:
>>>> On 27/10/2023 05:31, Chris Packham wrote:
>>>>> Some hardware designs have a GPIO used to control the reset of all the
>>>>> devices on and I2C bus. It's not possible for every child node to
>>>>> declare a reset-gpios property as only the first device probed would be
>>>>> able to successfully request it (the others will get -EBUSY). Represent
>>> Cc: Mark,
>>>
>>> Also this part is not true. If the bus is non-discoverable, then it is
>>> possible to have reset-gpios in each probed device. You can share GPIOs,
>>> so no problem with -EBUSY at all.
>> Last time I checked you couldn't share GPIOs. If that's no longer the
>> case then I can probably make what I need to happen work. It still
>> creates an issue that I have multiple PCA954x muxes connected to a
>> common reset GPIO so as each mux is probed the PCA954x driver will
>> toggle the reset. That's probably OK as the PCA954x is sufficiently
>> stateless that the extra resets won't do any harm but if it were a more
>> complicated device then there would be issues.
> I know, but this is a broader problem, not really specific to this one
> device. I also argue that your I2C controller does not actually have
> this reset line.

Yes absolutely. Because the reset line is common to multiple pca954x 
muxes the only option I have (I think) is to associate the reset line 
with the controller. It happens to be true for my case that everything 
connected to that bus is affected by the reset line but I can completely 
see that there may be other designs where there are a mix of muxes and 
other devices on the root bus.

So associating the reset line with the I2C controller is a pragmatic 
solution (or an egregious hack depending on your point of view) that 
works with this kind of hardware design.

Another complete hack I've experimented with is adding the muxes defined 
with `status = "disabled";` in the dts and having a custom driver that 
requests the GPIO and manipulates the live device tree. It works but is 
quite a lot more code and will invariably break if I need to tweak the 
device tree.

>> Having some kind of ref-counted reset controller that is implemented
>> with GPIOs is probably the better solution. I was kind of surprised that
>> nothing existed like that in drivers/reset.
> reset controller framework already supports this. The point is that GPIO
> reset is not a reset controller, so in terms of bindings "resets"
> property does not fit it.

So I need some way of representing a GPIO line associated with multiple 
devices that must be requested and driven appropriately before the 
devices are probed. In lieu of anything else a "bus-reset-gpios" 
property recognized by the generic I2C framework kind of sounds like the 
best solution so far. Unless maybe there's some kind of pinctrl type 
thing that would already work.

>>> The problem is doing reset:
>>> 1. in proper moment for all devices
>>> 2. without affecting other devices when one unbinds/remove()
>>>
>>> The (2) above is not solveable easy in kernel and we already had nice
>>> talks about it just few days ago:
>>> 1. Apple case:
>>> https://scanmail.trustwave.com/?c=20988&d=tZjA5R77ysliRyWfDvgX9JnmLZr-TqhRWpYjsNO-5A&u=https%3a%2f%2fsocial%2etreehouse%2esystems%2f%40marcan%2f111268780311634160
>>>
>>> 2. my WSA884x:
>>> https://scanmail.trustwave.com/?c=20988&d=tZjA5R77ysliRyWfDvgX9JnmLZr-TqhRWpZ9tI7vvw&u=https%3a%2f%2flore%2ekernel%2eorg%2falsa-devel%2f84f9f1c4-0627-4986-8160-b4ab99469b81%40linaro%2eorg%2f
>> Apologies for the mangled links (they're more secure now at least that's
>> what our IS team have been sold).
>>> Last,
>>> I would like to apologize to you Chris. I understand that bringing such
>>> feedback at v5 is not that good. I had plenty of time to say something
>>> earlier, so this is not really professional from my side. I am sorry,
>>> just my brain did not connect all these topics together.
>>>
>>> I apologize.
>> Actually I kind of expected this feedback. I figured I could start with
>> the driver that is currently causing me issues and once the dt-binding
>> was considered good enough it might migrate to the i2c core.
>>
>
> Best regards,
> Krzysztof
>
  

Patch

diff --git a/drivers/i2c/busses/i2c-mv64xxx.c b/drivers/i2c/busses/i2c-mv64xxx.c
index efd28bbecf61..6e2762d22e5a 100644
--- a/drivers/i2c/busses/i2c-mv64xxx.c
+++ b/drivers/i2c/busses/i2c-mv64xxx.c
@@ -13,6 +13,7 @@ 
 #include <linux/slab.h>
 #include <linux/module.h>
 #include <linux/spinlock.h>
+#include <linux/gpio/consumer.h>
 #include <linux/i2c.h>
 #include <linux/interrupt.h>
 #include <linux/mv643xx_i2c.h>
@@ -160,6 +161,7 @@  struct mv64xxx_i2c_data {
 	bool			clk_n_base_0;
 	struct i2c_bus_recovery_info	rinfo;
 	bool			atomic;
+	struct gpio_desc	*reset_gpio;
 };
 
 static struct mv64xxx_i2c_regs mv64xxx_i2c_regs_mv64xxx = {
@@ -1036,6 +1038,7 @@  mv64xxx_i2c_probe(struct platform_device *pd)
 	struct mv64xxx_i2c_data		*drv_data;
 	struct mv64xxx_i2c_pdata	*pdata = dev_get_platdata(&pd->dev);
 	struct resource *res;
+	u32	reset_duration;
 	int	rc;
 
 	if ((!pdata && !pd->dev.of_node))
@@ -1083,6 +1086,14 @@  mv64xxx_i2c_probe(struct platform_device *pd)
 	if (drv_data->irq < 0)
 		return drv_data->irq;
 
+	drv_data->reset_gpio = devm_gpiod_get_optional(&pd->dev, "bus-reset", GPIOD_OUT_HIGH);
+	if (IS_ERR(drv_data->reset_gpio))
+		return dev_err_probe(&pd->dev, PTR_ERR(drv_data->reset_gpio),
+				     "Cannot get reset gpio\n");
+	rc = device_property_read_u32(&pd->dev, "bus-reset-duration-us", &reset_duration);
+	if (rc)
+		reset_duration = 1;
+
 	if (pdata) {
 		drv_data->freq_m = pdata->freq_m;
 		drv_data->freq_n = pdata->freq_n;
@@ -1121,6 +1132,11 @@  mv64xxx_i2c_probe(struct platform_device *pd)
 			goto exit_disable_pm;
 	}
 
+	if (drv_data->reset_gpio) {
+		fsleep(reset_duration);
+		gpiod_set_value_cansleep(drv_data->reset_gpio, 0);
+	}
+
 	rc = request_irq(drv_data->irq, mv64xxx_i2c_intr, 0,
 			 MV64XXX_I2C_CTLR_NAME, drv_data);
 	if (rc) {