2020. november 25., szerda 11:57 keltezéssel, Coiby Xu írta:
> On Mon, Nov 23, 2020 at 04:32:40PM +0000, Barnabás Pőcze wrote:
> >> [...]
> >> >> >> +static int get_gpio_pin_state(struct irq_desc *irq_desc)
> >> >> >> +{
> >> >> >> + struct gpio_chip *gc = irq_data_get_irq_chip_data(&irq_desc->irq_data);
> >> >> >> +
> >> >> >> + return gc->get(gc, irq_desc->irq_data.hwirq);
> >> >> >> +}
> >> >> [...]
> >> >> >> + ssize_t status = get_gpio_pin_state(irq_desc);
> >> >> >
> >> >> >`get_gpio_pin_state()` returns an `int`, so I am not sure why `ssize_t` is used here.
> >> >> >
> >> >>
> >> >> I used `ssize_t` because I found gpiolib-sysfs.c uses `ssize_t`
> >> >>
> >> >> // drivers/gpio/gpiolib-sysfs.c
> >> >> static ssize_t value_show(struct device *dev,
> >> >> struct device_attribute *attr, char *buf)
> >> >> {
> >> >> struct gpiod_data *data = dev_get_drvdata(dev);
> >> >> struct gpio_desc *desc = data->desc;
> >> >> ssize_t status;
> >> >>
> >> >> mutex_lock(&data->mutex);
> >> >>
> >> >> status = gpiod_get_value_cansleep(desc);
> >> >> ...
> >> >> return status;
> >> >> }
> >> >>
> >> >> According to the book Advanced Programming in the UNIX Environment by
> >> >> W. Richard Stevens,
> >> >> With the 1990 POSIX.1 standard, the primitive system data type
> >> >> ssize_t was introduced to provide the signed return value...
> >> >>
> >> >> So ssize_t is fairly common, for example, the read and write syscall
> >> >> return a value of type ssize_t. But I haven't found out why ssize_t is
> >> >> better int.
> >> >> >
> >> >
> >> >Sorry if I wasn't clear, what prompted me to ask that question is the following:
> >> >`gc->get()` returns `int`, `get_gpio_pin_state()` returns `int`, yet you still
> >> >save the return value of `get_gpio_pin_state()` into a variable with type
> >> >`ssize_t` for no apparent reason. In the example you cited, `ssize_t` is used
> >> >because the show() callback of a sysfs attribute must return `ssize_t`, but here,
> >> >`interrupt_line_active()` returns `bool`, so I don't see any advantage over a
> >> >plain `int`. Anyways, I believe either one is fine, I just found it odd.
> >> >
> >> I don't understand why "the show() callback of a sysfs attribute
> >> must return `ssize_t`" instead of int. Do you think the rationale
> >> behind it is the same for this case? If yes, using "ssize_t" for
> >> status could be justified.
> >> [...]
> >
> >Because it was decided that way, `ssize_t` is a better choice for that purpose
> >than plain `int`. You can see it in include/linux/device.h, that both the
> >show() and store() methods must return `ssize_t`.
> >
>
> Could you explain why `ssize_t` is a better choice? AFAIU, ssize_t
> is used because we can return negative value to indicate an error.
ssize_t: "Signed integer type used for a count of bytes or an error indication."[1]
And POSIX mandates that the return type of read() and write() be `ssize_t`,
so it makes sense to keep a similar interface in the kernel since show() and store()
are called as a direct result of the user using the read() and write() system
calls, respectively.
> If
> we use ssize_t here, it's a reminder that reading a GPIO pin's status
> could fail. And ssize_t reminds us it's a operation similar to read
> or write. So ssize_t is better than int here. And maybe it's the same
> reason why "it was decided that way".
> [...]
I believe it's more appropriate to use ssize_t when it's about a "count of elements",
but the GPIO pin state is a single boolean value (or an error indication), which
is returned as an `int`. Since it's returned as an `int` - I'm arguing that -
there is no reason to use `ssize_t` here. Anyways, both `ssize_t` and `int` work fine
in this case.
[1]: https://pubs.opengroup.org/onlinepubs/9699919799/functions/V2_chap02.html#tag_15_12
Regards,
Barnabás Pőcze
