Re: [PATCH v2 08/10] gpio: aggregator: expoose aggregator created via legacy sysfs to configfs

[Bartosz Golaszewski]

On Mon, Feb 3, 2025 at 4:12 AM Koichiro Den <koichiro.den@xxxxxxxxxxxxx> wrote:
>
> Expose settings for aggregators created using the sysfs 'new_device'
> interface to configfs. Once written to 'new_device', an "_auto.<N>" path

I would prefer this to be called "_sysfs.<N>" as it's not really
"automatic" - the user did create this, just with a different
interface.

> appears in the configfs regardless of whether the probe succeeds.
> Consequently, users can no longer use that prefix for custom GPIO
> aggregator names. The 'live' attribute changes to 1 when the probe
> succeeds and the GPIO forwarder is instantiated.
>
> Note that the aggregator device created via sysfs is asynchrnous, i.e.
> writing into 'new_device' returns without waiting for probe completion,
> and the probe may succeed, fail, or eventually succeed via deferred
> probe. Thus, the 'live' attribute may change from 0 to 1 asynchronously
> without notice. So, editting key/offset/name while it's waiting for
> deferred probe is prohibited.
>
> The configfs auto-generation relies on create_default_group(), which
> inherently prohibits rmdir(2). To align with the limitation, this commit
> also prohibits mkdir(2) for them. When users want to change the number
> of lines for an aggregator initialized via 'new_device', they need to
> tear down the device using 'delete_device' and reconfigure it from
> scratch. This does not break previous behaviour; users of legacy sysfs
> interface simply gain additional almost read-only configfs exposure.
>
> Still, users can write into 'live' attribute to toggle the device unless
> it's waiting for deferred probe. So once probe succeeds, they can
> deactivate it in the same manner as the devices initialized via
> configfs.
>
> Signed-off-by: Koichiro Den <koichiro.den@xxxxxxxxxxxxx>
> ---

Bart