Hi,
Peter Chen <peter.chen@xxxxxxxxxxxxx> writes:
>> seriously ? Is this really all ? What about that reset line below ?
>
> The clock is PHY input clock on the HUB, this clock may from SoC's
> PLL.
oh, you might have misunderstood my comment. I'm saying that there is
more than one thing you could cache in your private structure. That's
all.
>> > +static int usb_hub_generic_probe(struct platform_device *pdev)
>> > +{
>> > + struct device *dev = &pdev->dev;
>> > + struct usb_hub_generic_platform_data *pdata = dev->platform_data;
>> > + struct usb_hub_generic_data *hub_data;
>> > + int reset_pol = 0, duration_us = 50, ret = 0;
>> > + struct gpio_desc *gpiod_reset = NULL;
>> > +
>> > + hub_data = devm_kzalloc(dev, sizeof(*hub_data), GFP_KERNEL);
>> > + if (!hub_data)
>> > + return -ENOMEM;
>> > +
>> > + if (dev->of_node) {
>> > + struct device_node *node = dev->of_node;
>> > +
>> > + hub_data->clk = devm_clk_get(dev, "external_clk");
>> > + if (IS_ERR(hub_data->clk)) {
>> > + dev_dbg(dev, "Can't get external clock: %ld\n",
>> > + PTR_ERR(hub_data->clk));
>>
>> how about setting clock to NULL to here ? then you don't need IS_ERR()
>> checks anywhere else.
>>
>> > + }
>>
>> braces shouldn't be used here, if you add NULL trick above,
>> however... then you can keep them.
>>
>
> Braces aren't needed, it may not too much useful to using NULL
> as a indicator for error pointer.
heh, it's not about using it as an error pointer. I'm merely trying to
make clk optional. NULL is a valid clk, meaning you won't get NULL
pointer dereferences when passing it along clk_*() calls (if you find
any, it's likely a bug in CCF), so NULL can be used to cope with
optional clocks:
clk = clk_get(dev, "foo");
if (IS_ERR(clk)) {
if (PTR_ERR(clk) == -EPROBE_DEFER)
return -EPROBE_DEFER;
else
clk = NULL;
}
>> > + /*
>> > + * Try to get the information for HUB reset, the
>> > + * default setting like below:
>> > + *
>> > + * - Reset state is low
>> > + * - The duration is 50us
>> > + */
>> > + if (of_find_property(node, "hub-reset-active-high", NULL))
>> > + reset_pol = 1;
>>
>> you see, this is definitely *not* generic. You should write a generic
>> reset-gpio.c reset controller and describe the polarity on the gpio
>> binding. This driver *always* uses reset_assert(); reset_deassert(); and
>> reset-gpio implements though by gpiod_set_value() correctly.
>>
>> Polarity _must_ be described elsewhere in DT.
>>
>> > + of_property_read_u32(node, "hub-reset-duration-us",
>> > + &duration_us);
>>
>> likewise, this should be described as a debounce time for the GPIO.
>>
>
> Yes, if we are a reset gpio driver.
even if you use a raw GPIO, polarity and duration must come through DT.
>> > + usleep_range(duration_us, duration_us + 100);
>> > + gpiod_set_value(gpiod_reset, reset_pol ? 0 : 1);
>>
>> wrong. You should _not_ have polarity checks here. You should have
>> already initialized the gpio as ACTIVE_HIGH or ACTIVE_LOW and gpiolib
>> will handle the polarity for you.
>
> Yes, you are right. I did not understand ACTIVE_LOW for gpio flag
> before.
with open source code, that's a rather poor excuse, Peter.
>> > +static int __init usb_hub_generic_init(void)
>> > +{
>> > + return platform_driver_register(&usb_hub_generic_driver);
>> > +}
>> > +subsys_initcall(usb_hub_generic_init);
>> > +
>> > +static void __exit usb_hub_generic_exit(void)
>> > +{
>> > + platform_driver_unregister(&usb_hub_generic_driver);
>> > +}
>> > +module_exit(usb_hub_generic_exit);
>>
>> module_platform_driver();
>
> I want this driver to be called before module init's.
why ?
--
balbi
Attachment:
signature.asc
Description: PGP signature
