[cip-dev] [PATCH] gpiolib: Support 'gpio-reserved-ranges' property

[JohnsonCH.Chen@moxa.com (Johnson CH Chen (陳昭勳))]

Hi Fabrizio Castro,

I find a kernel panic issue by gpio when I upgrade kernel to 4.4.182-cip34 with Freescale LS1021A (gpiochip is mpc8xxx series):

Kernel log shows ?Unable to handle kernel NULL pointer dereference at virtual address 000000b4.?

Subsequently, I find the commit baff4777cdb80256cd24dede2a3d0af761356307 (gpiolib: Support 'gpio-reserved-ranges' property) could result in kernel panic because the code ?*np= chip->dev->of_node? is executed when chip->dev is NULL:

diff --git a/drivers/gpio/gpiolib-of.c b/drivers/gpio/gpiolib-of.c
index 5fe34a9df3e6..ec642bf1d976 100644
--- a/drivers/gpio/gpiolib-of.c
+++ b/drivers/gpio/gpiolib-of.c

+static void of_gpiochip_init_valid_mask(struct gpio_chip *chip)
+{
+      int len, i;
+      u32 start, count;
+      struct device_node *np = chip->dev->of_node;
+
+      len = of_property_count_u32_elems(np,  "gpio-reserved-ranges");
+      if (len < 0 || len % 2 != 0)
+              return;
+
+      for (i = 0; i < len; i += 2) {
+              of_property_read_u32_index(np, "gpio-reserved-ranges",
+                                         i, &start);
+              of_property_read_u32_index(np, "gpio-reserved-ranges",
+                                         i + 1, &count);
+              if (start >= chip->ngpio || start + count >= chip->ngpio)
+                      continue;
+
+              bitmap_clear(chip->valid_mask, start, count);
+      }
+};
+

diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
index adb474089733..487a86495163 100644
--- a/drivers/gpio/gpiolib.c
+++ b/drivers/gpio/gpiolib.c
@@ -292,6 +292,43 @@ static unsigned long *gpiochip_allocate_mask(struct gpio_chip *chip)
      return p;
}

+static int gpiochip_init_valid_mask(struct gpio_chip *gpiochip)
+{
+#ifdef CONFIG_OF_GPIO
+      int size;
+      struct device_node *np = gpiochip->dev->of_node;
+
+      size = of_property_count_u32_elems(np,  "gpio-reserved-ranges");
+      if (size > 0 && size % 2 == 0)
+              gpiochip->need_valid_mask = true;
+#endif
+
+      if (!gpiochip->need_valid_mask)
+              return 0;
+
+      gpiochip->valid_mask = gpiochip_allocate_mask(gpiochip);
+      if (!gpiochip->valid_mask)
+              return -ENOMEM;
+
+      return 0;
+}
+

Could you help me understand why we need to replaced chip->of_node with chip->dev->of_node?

The following two items I try have no kernel panic:

1. For gpio drivers:

It?s fine and no kernel panic issue if probe(struct platform_device *pdev) in gpio drivers executes the following codes:

struct cpio_chip *gc;
gc->dev = &pdev->dev; //Put platform?s dev to gpiochip?s dev, and of_node info in pdev->dev->of_node also put in gpiochip structure.

But some gpio drivers doesn?t do this, such as gpio-mpc8xxx.c (mpc8xxx series), and result in kernel panic when use this commit.

2. For gpiolib:

//set the following code in gpiolib.c and gpiolib-of.c
struct device_node *np = gpiochip->of_node;

Device node of platform will put in gpio chip when of_mm_gpiochip_add() is executed:

int of_mm_gpiochip_add(struct device_node *np,
                       struct of_mm_gpio_chip *mm_gc)
{
        int ret = -ENOMEM;
        struct gpio_chip *gc = &mm_gc->gc;
?..
        mm_gc->gc.of_node = np;
}

It?s seems chip->of_node is good.

Thanks,
Johnson
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cip-project.org/pipermail/cip-dev/attachments/20191108/02a5be16/attachment-0001.html>