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 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 <[email protected]>
---
Notes:
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 | 15 +++++++++++++++
1 file changed, 15 insertions(+)
diff --git a/drivers/i2c/busses/i2c-mv64xxx.c b/drivers/i2c/busses/i2c-mv64xxx.c
index efd28bbecf61..28f11d2e800b 100644
--- a/drivers/i2c/busses/i2c-mv64xxx.c
+++ b/drivers/i2c/busses/i2c-mv64xxx.c
@@ -160,6 +160,7 @@ 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 +1037,7 @@ 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 +1085,14 @@ 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, "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, "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 +1131,11 @@ mv64xxx_i2c_probe(struct platform_device *pd)
goto exit_disable_pm;
}
+ if (drv_data->reset_gpio) {
+ usleep_range(reset_duration, reset_duration + 10);
+ 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) {
--
2.42.0
On 19/10/23 10:08, 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 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 <[email protected]>
> ---
>
> Notes:
> 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 | 15 +++++++++++++++
> 1 file changed, 15 insertions(+)
>
> diff --git a/drivers/i2c/busses/i2c-mv64xxx.c b/drivers/i2c/busses/i2c-mv64xxx.c
> index efd28bbecf61..28f11d2e800b 100644
> --- a/drivers/i2c/busses/i2c-mv64xxx.c
> +++ b/drivers/i2c/busses/i2c-mv64xxx.c
kernel test robot points out I'm missing an include of gpio/consumer.h
I'll fix that with a v4. I'll give it a couple of days before sending it
out just in case there are any more comments.
> @@ -160,6 +160,7 @@ 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 +1037,7 @@ 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 +1085,14 @@ 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, "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, "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 +1131,11 @@ mv64xxx_i2c_probe(struct platform_device *pd)
> goto exit_disable_pm;
> }
>
> + if (drv_data->reset_gpio) {
> + usleep_range(reset_duration, reset_duration + 10);
> + 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) {
Hi Chris,
as you are working on the v4...
...
> + if (drv_data->reset_gpio) {
> + usleep_range(reset_duration, reset_duration + 10);
I'm not against this, but it's not optimal unless we know more or
less what to expect from reset_duration.
Do we have a rough idea of what reset_duration is? If we don't
then you could consider using a generic "fsleep(reset_duration);"
Would it work?
Rest looks good.
Andi
On 25/10/23 08:18, Andi Shyti wrote:
> Hi Chris,
>
> as you are working on the v4...
>
> ...
>
>> + if (drv_data->reset_gpio) {
>> + usleep_range(reset_duration, reset_duration + 10);
> I'm not against this, but it's not optimal unless we know more or
> less what to expect from reset_duration.
>
> Do we have a rough idea of what reset_duration is? If we don't
> then you could consider using a generic "fsleep(reset_duration);"
> Would it work?
flseep() would work for me. All of the devices I'm testing with seem to
be fine with a very short reset pulse, they'd probably be fine with no
delay at all.
>
> Rest looks good.
>
> Andi
Hi Chris,
> > as you are working on the v4...
> >
> > ...
> >
> >> + if (drv_data->reset_gpio) {
> >> + usleep_range(reset_duration, reset_duration + 10);
> > I'm not against this, but it's not optimal unless we know more or
> > less what to expect from reset_duration.
> >
> > Do we have a rough idea of what reset_duration is? If we don't
> > then you could consider using a generic "fsleep(reset_duration);"
> > Would it work?
> flseep() would work for me. All of the devices I'm testing with seem to
> be fine with a very short reset pulse, they'd probably be fine with no
> delay at all.
you know this better than me :-)
If you say that a delay is not necessary, then I'm also fine.
In any case, we are in probe and I don't think it's time
critical, so that a little delay wouldn't hurt and make everyone
happy.
Either way I'm fine as long as you use the correct sleeping
function.
Andi
On 25/10/23 09:37, Andi Shyti wrote:
> Hi Chris,
>
>>> as you are working on the v4...
>>>
>>> ...
>>>
>>>> + if (drv_data->reset_gpio) {
>>>> + usleep_range(reset_duration, reset_duration + 10);
>>> I'm not against this, but it's not optimal unless we know more or
>>> less what to expect from reset_duration.
>>>
>>> Do we have a rough idea of what reset_duration is? If we don't
>>> then you could consider using a generic "fsleep(reset_duration);"
>>> Would it work?
>> flseep() would work for me. All of the devices I'm testing with seem to
>> be fine with a very short reset pulse, they'd probably be fine with no
>> delay at all.
> you know this better than me :-)
> If you say that a delay is not necessary, then I'm also fine.
>
> In any case, we are in probe and I don't think it's time
> critical, so that a little delay wouldn't hurt and make everyone
> happy.
>
> Either way I'm fine as long as you use the correct sleeping
> function.
My particular hardware doesn't need it but for this to be generally
usable I think it is necessary to provide the capability for some kind
of hardware specific reset-duration. I'll look at fsleep() for v4 (or
say why I've stuck with usleep_range() in the changelog).
> Andi