On Thu, Mar 23, 2023 at 10:30:12AM -0700, Douglas Anderson wrote:
> 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.
>
Seemed like a good idea at the time...
> 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.
>
It made total sense to to spell input-disable as output-{high,low} back
then, but we're wiser now. Thanks for fixing it.
> 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 <dianders@xxxxxxxxxxxx>
Reviewed-by: Bjorn Andersson <andersson@xxxxxxxxxx>
Regards,
Bjorn
> ---
>
> 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);
> --
> 2.40.0.348.gf938b09366-goog
>
