On 10/14/2017 05:43 PM, Linus Walleij wrote:

So I guess the driver needs to know what pin registers it can't
access so the user does not get a gun to shoot in the foot with.

If we augment gpiolib to just handle -EACCES or something
(-EIO?) from the driver .get_direction() callback for these lines,
things should be smooth?

You mean like this:

diff --git a/drivers/pinctrl/qcom/pinctrl-msm.c 

b/drivers/pinctrl/qcom/pinctrl-msm.c
index ff491da64dab..ca4ae3d76eb4 100644

--- a/drivers/pinctrl/qcom/pinctrl-msm.c
+++ b/drivers/pinctrl/qcom/pinctrl-msm.c

@@ -443,6 +443,14 @@ static int msm_gpio_get_direction(struct gpio_chip 

*chip, unsigned int offset)

  	g = &pctrl->soc->groups[offset];

+	/*
+	 * During initialization, gpiolib may query all GPIOs for their
+	 * initial direction, regardless if they exist, so block access
+	 * to those that are unavailable.
+	 */
+	if (!g->npins)
+		return -ENODEV;
+
  	val = readl(pctrl->regs + g->ctl_reg);

  	/* 0 = output, 1 = input */


This is what I have in my patch already.  I can return any error message 
you like, but -ENODEV already works.

The problem is that it's insufficient.  I also want the non-available 
GPIOs to be as absent as possible.  I don't want them to show up in 
/sys/kernel/debug/gpio, and I don't want to be able to create them via 
/sys/class/gpio/export.

-- 
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm
Technologies, Inc.  Qualcomm Technologies, Inc. is a member of the
Code Aurora Forum, a Linux Foundation Collaborative Project.