From c4a48b0df8bfcf45256e8690f959a5af39c20ce7 Mon Sep 17 00:00:00 2001 From: Douglas Anderson Date: Thu, 23 Mar 2023 10:30:10 -0700 Subject: [PATCH 1/3] dt-bindings: pinctrl: qcom: tlmm should use output-disable, not input-enable As evidenced by the Qualcomm TLMM Linux driver, the TLMM IP block in Qualcomm SoCs has a bit to enable/disable the output for a pin that's configured as a GPIO but _not_ a bit to enable/disable an input buffer. Current device trees that are specifying "input-enable" for pins managed by TLMM are either doing so needlessly or are using it to mean "output-disable". Presumably the current convention of using "input-enable" to mean "output-disable" stems from the fact that "output-disable" is a "new" property from 2017. It was introduced in commit 425562429d4f ("pinctrl: generic: Add output-enable property"). The "input-enable" handling in Qualcomm drivers is from 2015 introduced in commit 407f5e392f9c ("pinctrl: qcom: handle input-enable pinconf property"). Given that there's no other use for "input-enable" for TLMM, we can still handle old device trees in code, but let's encourage people to move to the proper / documented property by updating the bindings. Signed-off-by: Douglas Anderson Reviewed-by: Krzysztof Kozlowski Acked-by: Linus Walleij Link: https://lore.kernel.org/r/20230323102605.6.I291ce0ba2c6ea80b341659c4f75a567a76dd7ca6@changeid Signed-off-by: Linus Walleij --- .../devicetree/bindings/pinctrl/qcom,tlmm-common.yaml | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/Documentation/devicetree/bindings/pinctrl/qcom,tlmm-common.yaml b/Documentation/devicetree/bindings/pinctrl/qcom,tlmm-common.yaml index cb5ba1bd6f8d..5a815c199642 100644 --- a/Documentation/devicetree/bindings/pinctrl/qcom,tlmm-common.yaml +++ b/Documentation/devicetree/bindings/pinctrl/qcom,tlmm-common.yaml @@ -75,7 +75,8 @@ $defs: bias-pull-down: true bias-pull-up: true bias-disable: true - input-enable: true + input-enable: false + output-disable: true output-high: true output-low: true From 988a9eab4d9a0eb07339b9d3f2bc9bee66af47b1 Mon Sep 17 00:00:00 2001 From: Douglas Anderson Date: Thu, 23 Mar 2023 10:30:11 -0700 Subject: [PATCH 2/3] dt-bindings: pinctrl: qcom: Add output-enable In the patch ("dt-bindings: pinctrl: qcom: tlmm should use output-disable, not input-enable") we allowed setting "output-disable" for TLMM pinctrl states. Let's also add "output-enable". At first blush this seems a needless thing to do. Specifically: - In Linux (and presumably any other OSes using the same device trees) the GPIO/pinctrl driver knows to automatically enable the output when a GPIO is changed to an output. Thus in most cases specifying "output-enable" is superfluous and should be avoided. - If we need to set a pin's default state we already have "output-high" and "output-low" and these properties already imply "output-enabled" (at least on the Linux Qualcomm TLMM driver). However, there is one instance where "output-enable" seems like it could be useful: sleep states. It's not uncommon to want to configure pins as inputs (with appropriate pulls) when the driver controlling them is in a low power state. Then we want the pins back to outputs when the driver wants things running normally. To accomplish this we'd want to be able to use "output-enable". Then the "default" state could have "output-enable" and the "sleep" state could have "output-disable". NOTE: in all instances I'm aware of, we'd only want to use "output-enable" on pins that are configured as "gpio". The Qualcomm documentation that I have access to says that "output-enable" only does something useful when in GPIO mode. Signed-off-by: Douglas Anderson Reviewed-by: Krzysztof Kozlowski Acked-by: Linus Walleij Link: https://lore.kernel.org/r/20230323102605.7.I7874c00092115c45377c2a06f7f133356956686e@changeid Signed-off-by: Linus Walleij --- Documentation/devicetree/bindings/pinctrl/qcom,tlmm-common.yaml | 1 + 1 file changed, 1 insertion(+) diff --git a/Documentation/devicetree/bindings/pinctrl/qcom,tlmm-common.yaml b/Documentation/devicetree/bindings/pinctrl/qcom,tlmm-common.yaml index 5a815c199642..90b7d75840c1 100644 --- a/Documentation/devicetree/bindings/pinctrl/qcom,tlmm-common.yaml +++ b/Documentation/devicetree/bindings/pinctrl/qcom,tlmm-common.yaml @@ -77,6 +77,7 @@ $defs: bias-disable: true input-enable: false output-disable: true + output-enable: true output-high: true output-low: true From e49eabe3e13f295e4f46daa897a0575654789a8c Mon Sep 17 00:00:00 2001 From: Douglas Anderson Date: Thu, 23 Mar 2023 10:30:12 -0700 Subject: [PATCH 3/3] pinctrl: qcom: Support OUTPUT_ENABLE; deprecate INPUT_ENABLE The Qualcomm pinctrl driver has been violating the documented meaning of PIN_CONFIG_INPUT_ENABLE. That documentation says: Note that this does not affect the pin's ability to drive output. ...yet the Qualcomm driver's sole action when asked to "enable input" on a pin is to disable its output. The Qualcomm driver's implementation stems from the fact that "output-disable" is a "new" property from 2017. It was introduced in commit 425562429d4f ("pinctrl: generic: Add output-enable property"). The "input-enable" handling in Qualcomm drivers is from 2015 introduced in commit 407f5e392f9c ("pinctrl: qcom: handle input-enable pinconf property"). Let's change the Qualcomm driver to move us in the right direction. As part of this: 1. We'll now support PIN_CONFIG_OUTPUT_ENABLE 2. We'll still support using PIN_CONFIG_INPUT_ENABLE to disable a pin's output (in violation of the docs) with a big comment in the code. This is needed because old device trees have "input-enable" in them and, in some cases, people might need the old behavior. While we could programmatically change all old device trees, it doesn't really hurt to keep supporting the old behavior and we're _supposed_ to try to be compatible with old device trees anyway. It can also be noted that the PIN_CONFIG_INPUT_ENABLE handling code seems to have purposefully ignored its argument. That means that old boards that had _either_ "input-disable" or "input-enable" in them would have had the effect of disabling a pin's output. While we could change this behavior, since we're only leaving the PIN_CONFIG_INPUT_ENABLE there for backward compatibility we might as well be fully backward compatible. NOTE: despite the fact that we'll still support PIN_CONFIG_INPUT_ENABLE for _setting_ config, we take it away from msm_config_group_get(). This appears to be only used for populating debugfs and fixing debugfs to "output enabled" where relevant instead of "input enabled" makes more sense and has more truthiness. Signed-off-by: Douglas Anderson Reviewed-by: Bjorn Andersson Acked-by: Linus Walleij Link: https://lore.kernel.org/r/20230323102605.8.Id740ae6a993f9313b58add6b10f6a92795d510d4@changeid Signed-off-by: Linus Walleij --- drivers/pinctrl/qcom/pinctrl-msm.c | 36 +++++++++++++++++++++++++----- 1 file changed, 31 insertions(+), 5 deletions(-) diff --git a/drivers/pinctrl/qcom/pinctrl-msm.c b/drivers/pinctrl/qcom/pinctrl-msm.c index daeb79a9a602..4515f375c5e8 100644 --- a/drivers/pinctrl/qcom/pinctrl-msm.c +++ b/drivers/pinctrl/qcom/pinctrl-msm.c @@ -323,6 +323,7 @@ static int msm_config_reg(struct msm_pinctrl *pctrl, break; case PIN_CONFIG_OUTPUT: case PIN_CONFIG_INPUT_ENABLE: + case PIN_CONFIG_OUTPUT_ENABLE: *bit = g->oe_bit; *mask = 1; break; @@ -414,11 +415,9 @@ static int msm_config_group_get(struct pinctrl_dev *pctldev, val = msm_readl_io(pctrl, g); arg = !!(val & BIT(g->in_bit)); break; - case PIN_CONFIG_INPUT_ENABLE: - /* Pin is output */ - if (arg) + case PIN_CONFIG_OUTPUT_ENABLE: + if (!arg) return -EINVAL; - arg = 1; break; default: return -ENOTSUPP; @@ -502,9 +501,36 @@ static int msm_config_group_set(struct pinctrl_dev *pctldev, arg = 1; break; case PIN_CONFIG_INPUT_ENABLE: - /* disable output */ + /* + * According to pinctrl documentation this should + * actually be a no-op. + * + * The docs are explicit that "this does not affect + * the pin's ability to drive output" but what we do + * here is to modify the output enable bit. Thus, to + * follow the docs we should remove that. + * + * The docs say that we should enable any relevant + * input buffer, but TLMM there is no input buffer that + * can be enabled/disabled. It's always on. + * + * The points above, explain why this _should_ be a + * no-op. However, for historical reasons and to + * support old device trees, we'll violate the docs + * still affect the output. + * + * It should further be noted that this old historical + * behavior actually overrides arg to 0. That means + * that "input-enable" and "input-disable" in a device + * tree would _both_ disable the output. We'll + * continue to preserve this behavior as well since + * we have no other use for this attribute. + */ arg = 0; break; + case PIN_CONFIG_OUTPUT_ENABLE: + arg = !!arg; + break; default: dev_err(pctrl->dev, "Unsupported config parameter: %x\n", param);