On 2/9/2015 11:20 AM, Dmitry Torokhov wrote:
> Hi Ray,
>
> On Wed, Feb 04, 2015 at 09:21:01AM -0800, Ray Jui wrote:
>> +static int cygnus_gpio_pinmux_add_range(struct cygnus_gpio *chip)
>> +{
>> + struct device_node *node = chip->dev->of_node;
>> + struct device_node *pinmux_node;
>> + struct platform_device *pinmux_pdev;
>> + struct gpio_chip *gc = &chip->gc;
>> + int i, ret;
>> +
>> + /* parse DT to find the phandle to the pinmux controller */
>> + pinmux_node = of_parse_phandle(node, "pinmux", 0);
>> + if (!pinmux_node)
>> + return -ENODEV;
>> +
>> + pinmux_pdev = of_find_device_by_node(pinmux_node);
>> + if (!pinmux_pdev) {
>> + dev_err(chip->dev, "failed to get pinmux device\n");
>> + return -EINVAL;
>> + }
>> +
>> + /* now need to create the mapping between local GPIO and PINMUX pins */
>> + for (i = 0; i < ARRAY_SIZE(cygnus_gpio_pintable); i++) {
>> + ret = gpiochip_add_pin_range(gc, dev_name(&pinmux_pdev->dev),
>> + cygnus_gpio_pintable[i].offset,
>> + cygnus_gpio_pintable[i].pin_base,
>> + cygnus_gpio_pintable[i].num_pins);
>> + if (ret) {
>> + dev_err(chip->dev, "unable to add GPIO pin range\n");
>> + goto err_put_device;
>> + }
>> + }
>> +
>> + chip->pinmux_is_supported = true;
>> +
>> + /* no need for pinmux_pdev device reference anymore */
>> + put_device(&pinmux_pdev->dev);
>
> Sorry I did not notice it before, but of_parse_phandle() takes reference
> to the returned device node, so you need to "put" it here and in error
> path as well. Actually you can do:
>
> int ret = 0;
>
> pinmux_node = of_parse_phandle(node, "pinmux", 0);
> if (!pinmux_node)
> return -ENODEV;
>
> pinmux_pdev = of_find_device_by_node(pinmux_node);
> /* We do not longer need pinmux node */
> of_node_put(pinmux_node);
>
> if (!pinmux_dev)
> ....
>
> for (..) {
> ...
> if (ret) {
> dev_err(...);
> break;
> }
> }
>
> chip->pinmux_is_supported = (ret == 0);
> put_device(..);
> return ret;
> }
>
> This way you free resources in the same path.
>
Thanks. I'll make the change.
> ...
>
>> +
>> +static struct platform_driver cygnus_gpio_driver = {
>> + .driver = {
>> + .name = "cygnus-gpio",
>> + .of_match_table = cygnus_gpio_of_match,
>> + .suppress_bind_attrs = true,
>> + },
>> + .probe = cygnus_gpio_probe,
>> +};
>> +
>> +static int __init cygnus_gpio_init(void)
>> +{
>> + return platform_driver_probe(&cygnus_gpio_driver, cygnus_gpio_probe);
>
> When I asked you to add ".suppress_bind_attrs = true" I missed the fact
> that you were using platform_driver_probe() which already does this
> internally. However platform_driver_probe() can't handle deferred
> probing, which may or may not be OK. Is there a chance that any of the
> resources needed by the driver return -EPROBE_DEFER? If not then it is
> safe to continue using platform_driver_probe() and you can drop
> suppress_bind_attrs assignment, otherwise it may be better to switch to
> platform_driver_register().
>
> Thanks.
>
No I do not expect any resource that this driver depends on to return
-EPROBE_DEFER. The IOMUX driver that this driver depends on should be
initialized before this driver.
I'll drop .suppress_bind_attrs then. Thanks.
Ray
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
