[PATCH 3/6] drivers: Introduce AIODEV subsystem

[Andrey Smirnov]

On Mon, May 2, 2016 at 11:13 PM, Sascha Hauer <s.hauer at pengutronix.de> wrote:
> On Fri, Apr 29, 2016 at 10:24:03AM -0700, Andrey Smirnov wrote:
>> From: Sascha Hauer <s.hauer at pengutronix.de>
>>
>> AIODEV/Aiodevice is a analog I/O framework that can be thought of as a
>> simplified hybrid between 'hwmon' and 'IIO' subsystems of Linux kernel
>>
>> This commit is very heavily based on 'iodevice' framework proposal
>> written by Sascha Hauer.
>>
>> Signed-off-by: Andrey Smirnov <andrew.smrinov at gmail.com>
>> Signed-off-by: Sascha Hauer <s.hauer at pengutronix.de>
>> ---
>>  drivers/Kconfig         |   1 +
>>  drivers/Makefile        |   1 +
>>  drivers/aiodev/Kconfig  |   8 +++
>>  drivers/aiodev/Makefile |   2 +
>>  drivers/aiodev/core.c   | 135 ++++++++++++++++++++++++++++++++++++++++++++++++
>>  include/aiodev.h        |  39 ++++++++++++++
>>  6 files changed, 186 insertions(+)
>>  create mode 100644 drivers/aiodev/Kconfig
>>  create mode 100644 drivers/aiodev/Makefile
>>  create mode 100644 drivers/aiodev/core.c
>>  create mode 100644 include/aiodev.h
>>
>> diff --git a/drivers/aiodev/Makefile b/drivers/aiodev/Makefile
>> new file mode 100644
>> index 0000000..806464e
>> --- /dev/null
>> +++ b/drivers/aiodev/Makefile
>> @@ -0,0 +1,2 @@
>> +
>> +obj-$(CONFIG_AIODEV) += core.o
>> diff --git a/drivers/aiodev/core.c b/drivers/aiodev/core.c
>> new file mode 100644
>> index 0000000..6dcb917
>> --- /dev/null
>> +++ b/drivers/aiodev/core.c
>> @@ -0,0 +1,135 @@
>> +#include <common.h>
>> +#include <aiodev.h>
>> +#include <linux/list.h>
>> +#include <malloc.h>
>
> GPL Header missing.

OK, will fix in v2.


>
>> +
>> +LIST_HEAD(aiodevices);
>> +EXPORT_SYMBOL(aiodevices);
>> +
>> +struct aiochannel *aiochannel_get_by_name(const char *name)
>> +{
>> +     struct aiodevice *aiodev;
>> +     int i;
>> +
>> +     list_for_each_entry(aiodev, &aiodevices, list) {
>> +             for (i = 0; i < aiodev->num_channels; i++)
>> +                     if (!strcmp(name, aiodev->channels[i]->name))
>> +                             return aiodev->channels[i];
>> +     }
>> +
>> +     return ERR_PTR(-ENOENT);
>> +}
>> +EXPORT_SYMBOL(aiochannel_get_by_name);
>> +
>> +struct aiochannel *aiochannel_get(struct device_d *dev, int index)
>> +{
>> +     struct of_phandle_args spec;
>> +     struct aiodevice *aiodev;
>> +     int ret, chnum = 0;
>> +
>> +     if (!dev->device_node)
>> +             return ERR_PTR(-EINVAL);
>> +
>> +     ret = of_parse_phandle_with_args(dev->device_node,
>> +                                      "aio-channels",
>> +                                      "#aio-channel-cells",
>> +                                      index, &spec);
>
> #io-channel-cells is part of the official binding in
> /dts/Bindings/iio/iio-bindings.txt. We should work with this existing
> binding.

Oh, I didn't realize that it was original IIO DT binding. Will fix in v2.

>
>> +        if (ret)
>> +                return ERR_PTR(ret);
>> +
>> +     list_for_each_entry(aiodev, &aiodevices, list) {
>> +             if (aiodev->hwdev->device_node == spec.np)
>> +                     goto found;
>> +     }
>> +
>> +     return ERR_PTR(-EPROBE_DEFER);
>> +
>> +found:
>> +     if (spec.args_count)
>> +             chnum = spec.args[0];
>> +
>> +     if (chnum >= aiodev->num_channels)
>> +             return ERR_PTR(-EINVAL);
>> +
>> +     return aiodev->channels[chnum];
>> +}
>> +EXPORT_SYMBOL(aiochannel_get);
>> +
>> +int aiochannel_get_value(struct aiochannel *aiochan, int *value)
>> +{
>> +     struct aiodevice *aiodev = aiochan->aiodev;
>> +
>> +     return aiodev->read(aiochan, value);
>> +}
>> +EXPORT_SYMBOL(aiochannel_get_value);
>> +
>> +int aiochannel_get_index(struct aiochannel *aiochan)
>> +{
>> +     int i;
>> +     struct aiodevice *aiodev = aiochan->aiodev;
>> +
>> +     for (i = 0; i < aiodev->num_channels; i++)
>> +             if (aiodev->channels[i] == aiochan)
>> +                     return i;
>
> This function is unused in your patches. If the information this
> function provides is needed, maybe better add a index member to struct
> aiochannel to get rid of this loop?
>

It's used in one of the custom drivers I have in my tree. And yeah, I
think it's a good idea to store index and get rid of the loop. Will do
that in v2.

>> +
>> +     return -ENOENT;
>> +}
>> +EXPORT_SYMBOL(aiochannel_get_index);
>> +
>> +static int aiochannel_param_get_value(struct param_d *p, void *priv)
>> +{
>> +     struct aiochannel *aiochan = priv;
>> +
>> +     return aiochannel_get_value(aiochan, &aiochan->value);
>> +}
>> +
>> +int aiodevice_register(struct aiodevice *aiodev)
>> +{
>> +     int i, ret;
>> +
>> +     if (!aiodev->name) {
>> +             if (aiodev->hwdev &&
>> +                 aiodev->hwdev->device_node) {
>
>         if (!aiodev->name && aiodev->hwdev &&
>             aiodev->hwdev->device_node)
>
> ?

Agreed, there's no need to have a standalone if. Will fix in v2.

Thank you for reviewing my pathes!

Andrey