Re: [PATCH v2 1/2] gpio: tps65219: add GPIO support for TPS65219 PMIC

From: jerome Neanne
Date: Mon May 22 2023 - 08:30:34 EST




On 17/05/2023 08:33, Tony Lindgren wrote:
* jerome Neanne <jneanne@xxxxxxxxxxxx> [230512 07:13]:


On 11/05/2023 22:57, Linus Walleij wrote:
+ /* Documentation is stating that GPIO0 direction must not be changed in Linux:
+ * Table 8-34. MFP_1_CONFIG(3): MULTI_DEVICE_ENABLE,
+ * Should only be changed in INITIALIZE state (prior to ON Request).
+ * Set statically by NVM, changing direction in application can cause a hang.
+ * Below can be used for test purpose only:
+ */
+
+#if 0
+ int ret = regmap_update_bits(gpio->tps->regmap, TPS65219_REG_MFP_1_CONFIG,
+ TPS65219_GPIO0_DIR_MASK, direction);
+ if (ret)
+ return ret;
+#endif
+ dev_err(gpio->tps->dev,
+ "GPIO%d direction set by NVM, change to %u failed, not allowed by specification\n",
+ offset, direction);
+ return -EOPNOTSUPP;
+}

Normally people would complain about #if 0 code.

But this is a special case!

I definitely want the code to be in there somehow.

What about:

if (IS_ENABLED(DEBUG))?

If someone enables debug with an explicit -DDEBUG to the compiler
this could be allowed.
I'm fine with your proposal. Will wait few days just in case anyone wants to
add any comment then go for this.

Just wondering.. Would it help for the driver probe to set gpio0 as a gpio
hog after reading the configured value?

Maybe the multi device enable just means the pin may be shared with no
specific hardware reason to not change it during the runtime?

Your point looks valid. But I think we can't simply add a regular "gpio-hog" as a property in the device tree because we don't have a one to one mapping for GPIO consumer (theoretically we can have more than 2 PMICs).

I think your suggestion is to read the multi-device value through regmap then "kind of" hog gpio with devm_gpio_request_one
So that gpio0 is preserved from being requested by other user.
Is this correct understanding?

Practically the board I'm using for test currently only have one PMIC.
I'm reluctant to implement additional logic that does not look so "conventional" that I can't test.

If maintainers agree, I'll postpone the implementation of your proposal until a platform compatible with this implementation is available for testing. So that we can check what is most accurate in real life.

Side Note: wo this implementation use of the driver is less optimal in multi-device configuration but still usable.

Regards,
Jerome.