On 02.08.2020 19:54, Wolfram Sang wrote:
> On Fri, Jun 19, 2020 at 05:19:02PM +0300, Codrin Ciubotariu wrote:
>> Multiple I2C bus drivers use similar bindings to obtain information needed
>> for I2C recovery. For example, for platforms using device-tree, the
>> properties look something like this:
>>
>> &i2c {
>> ...
>> pinctrl-names = "default", "gpio";
>> // or pinctrl-names = "default", "recovery";
>> pinctrl-0 = <&pinctrl_i2c_default>;
>> pinctrl-1 = <&pinctrl_i2c_gpio>;
>> sda-gpios = <&pio 0 GPIO_ACTIVE_HIGH>;
>> scl-gpios = <&pio 1 (GPIO_ACTIVE_HIGH | GPIO_OPEN_DRAIN)>;
>> ...
>> }
>>
>> For this reason, we can add this common initialization in the core. This
>> way, other I2C bus drivers will be able to support GPIO recovery just by
>> providing a pointer to platform's pinctrl and calling i2c_recover_bus()
>> when SDA is stuck low.
>>
>> Signed-off-by: Codrin Ciubotariu <codrin.ciubotariu@xxxxxxxxxxxxx>
>
> Thanks, it looks a lot like what I had in mind!
>
>> ---
>> drivers/i2c/i2c-core-base.c | 119 ++++++++++++++++++++++++++++++++++++
>> include/linux/i2c.h | 11 ++++
>> 2 files changed, 130 insertions(+)
>>
>> diff --git a/drivers/i2c/i2c-core-base.c b/drivers/i2c/i2c-core-base.c
>> index d1f278f73011..4ee29fec4e93 100644
>> --- a/drivers/i2c/i2c-core-base.c
>> +++ b/drivers/i2c/i2c-core-base.c
>> @@ -32,6 +32,7 @@
>> #include <linux/of_device.h>
>> #include <linux/of.h>
>> #include <linux/of_irq.h>
>> +#include <linux/pinctrl/consumer.h>
>> #include <linux/pm_domain.h>
>> #include <linux/pm_runtime.h>
>> #include <linux/pm_wakeirq.h>
>> @@ -179,6 +180,8 @@ int i2c_generic_scl_recovery(struct i2c_adapter *adap)
>> struct i2c_bus_recovery_info *bri = adap->bus_recovery_info;
>> int i = 0, scl = 1, ret = 0;
>>
>> + if (bri->pinctrl)
>> + pinctrl_select_state(bri->pinctrl, bri->pins_gpio);
>
> I think this should come after 'prepare_recovery'. It may be that
> 'prepare_recovery' already needs to select the pinctrl state to avoid a
> glitch. In this version, there would be a glitch then. If we move it
> down, the doubled 'pinctrl_select_state' would be a noop then.
Agree
>
>> if (bri->prepare_recovery)
>> bri->prepare_recovery(adap);
>>
>> @@ -236,6 +239,8 @@ int i2c_generic_scl_recovery(struct i2c_adapter *adap)
>>
>> if (bri->unprepare_recovery)
>> bri->unprepare_recovery(adap);
>> + if (bri->pinctrl)
>> + pinctrl_select_state(bri->pinctrl, bri->pins_default);
>
> Here it is OK and will still work with the PXA version which needs to
> select the state on its own.
>
>>
>> return ret;
>> }
>> @@ -251,6 +256,118 @@ int i2c_recover_bus(struct i2c_adapter *adap)
>> }
>> EXPORT_SYMBOL_GPL(i2c_recover_bus);
>>
>> +static void i2c_gpio_init_pinctrl_recovery(struct i2c_adapter *adap)
>> +{
>> + struct i2c_bus_recovery_info *bri = adap->bus_recovery_info;
>> + struct device *dev = &adap->dev;
>> + struct pinctrl *p = bri->pinctrl;
>> +
>> + /*
>> + * we can't change states without pinctrl, so remove the states if
>> + * available
>
> s/available/populated/ ?
OK
>
>> + */
>> + if (!p) {
>> + bri->pins_default = NULL;
>> + bri->pins_gpio = NULL;
>> + return;
>> + }
>> +
>> + if (!bri->pins_default) {
>> + bri->pins_default = pinctrl_lookup_state(p,
>> + PINCTRL_STATE_DEFAULT);
>> + if (IS_ERR(bri->pins_default)) {
>> + dev_dbg(dev, PINCTRL_STATE_DEFAULT " state not found for GPIO recovery\n");
>> + bri->pins_default = NULL;
>> +
>> + goto cleanup_pinctrl;
>
> I'd leave out the goto here. It is OK to check both parameters, I think.
since default state is missing, the next check can be skipped, but I
agree that removing the label makes things easier to read.
>
>> + }
>> + }
>> + if (!bri->pins_gpio) {
>> + bri->pins_gpio = pinctrl_lookup_state(p, "gpio");
>> + if (IS_ERR(bri->pins_gpio))
>> + bri->pins_gpio = pinctrl_lookup_state(p, "recovery");
>> +
>> + if (IS_ERR(bri->pins_gpio)) {
>> + dev_dbg(dev, "no gpio or recovery state found for GPIO recovery\n");
>> + bri->pins_gpio = NULL;
>> +
>> + goto cleanup_pinctrl;
>
> This goto is not needed...
right
>
>> + }
>> + }
>> +
>> +cleanup_pinctrl:
>
> ... and this label can go then. Also nicer to read, I'd say.
I will remove it.
>
>> + /* for pinctrl state changes, we need all the information */
>> + if (!bri->pins_default || !bri->pins_gpio) {
>> + bri->pinctrl = NULL;
>> + bri->pins_default = NULL;
>> + bri->pins_gpio = NULL;
>> + } else {
>> + dev_info(dev, "using pinctrl states for GPIO recovery");
>> + }
>> +}
>> +
>> +static int i2c_gpio_init_generic_recovery(struct i2c_adapter *adap)
>> +{
>> + struct i2c_bus_recovery_info *bri = adap->bus_recovery_info;
>> + struct device *dev = &adap->dev;
>> + struct gpio_desc *gpiod;
>> + int ret = 0;
>> +
>> + /* don't touch the recovery information if the driver is not using
>> + * generic SCL recovery
>> + */
>
> Not kernel comment style.
Right, sorry about this. Will fix.
>
>> + if (bri->recover_bus && bri->recover_bus != i2c_generic_scl_recovery)
>> + return 0;
>
> No need for the first condition. 'i2c_generic_scl_recovery' is
> definately not NULL :)
It is not the same thing. Without the first check, it will return 0 if
bri->recover_bus is NULL, which is not what we want. If bri->recover_bus
is NULL (and the pintrl states and gpios are in place) we can set
recover_bus to i2c_generic_scl_recovery and use the generic recovery.
>
>> +
>> + /*
>> + * pins might be taken as GPIO, so we might as well inform pinctrl about
>
> s/might as well/should/
OK
>
>> + * this and move the state to GPIO
>> + */
>> + if (bri->pinctrl)
>> + pinctrl_select_state(bri->pinctrl, bri->pins_gpio);
>> +
>> + /*
>> + * if there is incomplete or no recovery information, see if generic
>> + * GPIO recovery is available
>> + */
>> + if (!bri->scl_gpiod) {
>> + gpiod = devm_gpiod_get(dev, "scl", GPIOD_OUT_HIGH_OPEN_DRAIN);
>> + if (PTR_ERR(gpiod) == -EPROBE_DEFER) {
>> + ret = -EPROBE_DEFER;
>> + goto cleanup_pinctrl_state;
>> + }
>> + if (!IS_ERR(gpiod)) {
>> + bri->scl_gpiod = gpiod;
>> + bri->recover_bus = i2c_generic_scl_recovery;
>> + dev_info(dev, "using generic GPIOs for recovery\n");
>> + }
>> + }
>
> I think this extra code from the PXA driver makes sense in case SDA was
> released while we were executing this code:
>
> 1383 /*
> 1384 * We have SCL. Pull SCL low and wait a bit so that SDA glitches
> 1385 * have no effect.
> 1386 */
> 1387 gpiod_direction_output(bri->scl_gpiod, 0);
> 1388 udelay(10);
> 1389 bri->sda_gpiod = devm_gpiod_get(dev, "sda", GPIOD_OUT_HIGH_OPEN_DRAIN);
> 1390
> 1391 /* Wait a bit in case of a SDA glitch, and then release SCL. */
> 1392 udelay(10);
> 1393 gpiod_direction_output(bri->scl_gpiod, 1);
I agree. I will add it.
>
>> +
>> + /* SDA GPIOD line is optional, so we care about DEFER only */
>> + if (!bri->sda_gpiod) {
>> + gpiod = devm_gpiod_get(dev, "sda", GPIOD_IN);
>> + if (PTR_ERR(gpiod) == -EPROBE_DEFER) {
>> + ret = -EPROBE_DEFER;
>> + goto cleanup_pinctrl_state;
>> + }
>> + if (!IS_ERR(gpiod))
>> + bri->sda_gpiod = gpiod;
>> + }
>> +
>> +cleanup_pinctrl_state:
>> + /* change the state of the pins back to their default state */
>> + if (bri->pinctrl)
>> + pinctrl_select_state(bri->pinctrl, bri->pins_default);
>> +
>> + return ret;
>> +}
>> +
>
> Rest looks good! If you have some time for this now, I will make sure to
> get it into 5.9. With these minor things fixed, this is good to go, me
> thinks.
>
I agree will all your suggestions, except with the removal of if
(bri->recover_bus) . If you agree with this, I can send the next version
and remove the RFC.
Thanks and best regards,
Codrin
