On Mon, Jan 25, 2021 at 10:52 AM Stephen Boyd <swboyd@xxxxxxxxxxxx> wrote:
>
> Quoting Gwendal Grignou (2021-01-24 13:41:44)
> > On Sun, Jan 24, 2021 at 9:38 AM Jonathan Cameron <jic23@xxxxxxxxxx> wrote:
> > >
> > > On Fri, 22 Jan 2021 14:54:43 -0800
> > > Stephen Boyd <swboyd@xxxxxxxxxxxx> wrote:
> > >
> > > > ---
> > > > drivers/iio/proximity/Kconfig | 11 +
> > > > drivers/iio/proximity/Makefile | 1 +
> > > > drivers/iio/proximity/cros_ec_proximity.c | 252 ++++++++++++++++++++++
>
> I suppose I'll change this to cros_ec_mkbp_proximity as well.
>
> > > > diff --git a/drivers/iio/proximity/cros_ec_proximity.c b/drivers/iio/proximity/cros_ec_proximity.c
> > > > new file mode 100644
> > > > index 000000000000..a3aef911e3cc
> > > > --- /dev/null
> > > > +++ b/drivers/iio/proximity/cros_ec_proximity.c
> > > > @@ -0,0 +1,252 @@
> [...]
> > > > +
> > > > +static int cros_ec_proximity_query(struct cros_ec_device *ec_dev, int *state)
> > > > +{
> > > > + struct ec_params_mkbp_info *params;
> > > > + struct cros_ec_command *msg;
> > > > + int ret;
> > > > +
> > > > + msg = kzalloc(sizeof(*msg) + max(sizeof(u32), sizeof(*params)),
> > > > + GFP_KERNEL);
> > >
> > > Given this is known at build time, perhaps better to add it to the
> > > iio_priv() accessed structure and avoid having to handle allocations
> > > separately.
> > As Jonathan said, it can be preallocated in iio private structure. We
> > can also use the stack, given the response size is known beforehand.
> > See cros_ec_cec_set_log_addr() or cros_ec_pwm_get_duty() for example.
>
> I suppose stack is even simpler. I'll try that.
>
> > > > +
> > > > +static int cros_ec_proximity_notify(struct notifier_block *nb,
> > > > + unsigned long queued_during_suspend, void *_ec)
> > > > +{
> > > > + struct cros_ec_proximity_data *data;
> > > > + struct cros_ec_device *ec = _ec;
> > > > + u8 event_type = ec->event_data.event_type & EC_MKBP_EVENT_TYPE_MASK;
> > > > + void *switches = &ec->event_data.data.switches;
> > > > + struct iio_dev *indio_dev;
> > > > + s64 timestamp;
> > > > + int state, dir;
> > > > + u64 ev;
> > > > +
> > > > + if (event_type == EC_MKBP_EVENT_SWITCH) {
> > > > + data = container_of(nb, struct cros_ec_proximity_data, notifier);
> > > > + indio_dev = data->indio_dev;
> > > > +
> > > > + mutex_lock(&data->lock);
> > > > + if (data->enabled) {
> > > > + timestamp = iio_get_time_ns(indio_dev);
> > For Android, given the timestamp must be time it happens, not reported
> > [https://source.android.com/devices/sensors/sensors-hal2] """The
> > timestamp must be accurate and correspond to the time at which the
> > event physically happened, not the time it was reported.""", consider
> > using ec_dev->last_event_time and apply a delta if the iio clock base
> > is different from CLOCK_BOOTTIME.
>
> Ah alright. Is there a reason why cros_ec_get_time_ns() is using
> boottime instead of plain ktime_get(), i.e. CLOCK_MONOTONIC? Otherwise I
> suppose some sort of cros_ec API should be exposed to convert the
> last_event_time to whatever clock base is desired. Does that exist?
CLOCK_BOOTTIME was chosen to be Android compliant, as it includes
suspend time and match elapsedRealtime() [see
https://developer.android.com/reference/android/os/SystemClock#elapsedRealtime()]
Chromebook set iio clock reference for all sensor to CLOCK_BOOTTIME
(see https://chromium.googlesource.com/chromiumos/platform2/+/HEAD/mems_setup/configuration.cc#127).
In case the iio device clock_id is not CLOCK_BOOTTIME/"bootime", we
need to add a delta, like in cros_ec_sensors_push_data()
[https://elixir.bootlin.com/linux/latest/source/drivers/iio/common/cros_ec_sensors/cros_ec_sensors_core.c#L210]
>
> > > > +static int cros_ec_proximity_probe(struct platform_device *pdev)
> > > > +{
> > > > + struct device *dev = &pdev->dev;
> > > > + struct cros_ec_device *ec = dev_get_drvdata(dev->parent);
> > > > + struct iio_dev *indio_dev;
> > > > + struct cros_ec_proximity_data *data;
> > > > + int ret;
> > > > +
> > > > + indio_dev = devm_iio_device_alloc(dev, sizeof(*data));
> > > > + if (!indio_dev)
> > > > + return -ENOMEM;
> > > > +
> > > > + data = iio_priv(indio_dev);
> > > > + data->ec = ec;
> > > > + data->indio_dev = indio_dev;
> > > > + mutex_init(&data->lock);
> > > > + platform_set_drvdata(pdev, data);
> > > > +
> > > > + indio_dev->name = "cros_ec_proximity";
> > Define a constant CROS_EC_[MKBP_]PROXIMITY_DRIVER_NAME and use it here
> > and in struct platform_driver cros_ec_proximity_driver.
>
> I used dev->driver->name instead. Yay for no define!
>
> > > > + indio_dev->dev.parent = dev;
> > Not needed, done by iio_device_alloc(), called by devm_iio_device_alloc().
>
> Ok.
>
> > > > +static const struct of_device_id cros_ec_proximity_of_match[] = {
> > > > + { .compatible = "google,cros-ec-proximity" },
> > > > + {}
> > > > +};
> > > > +MODULE_DEVICE_TABLE(of, cros_ec_proximity_of_match);
> > > > +#endif
> > > > +
> > > > +static struct platform_driver cros_ec_proximity_driver = {
> > > > + .driver = {
> > > > + .name = "cros-ec-proximity",
> > > > + .of_match_table = of_match_ptr(cros_ec_proximity_of_match),
> > Add a ACPI match table to match.
>
> I don't have an ACPI system in hand. What should the ACPI table look
> like? Can ACPI use the of_match_table logic?
AFAIK, ACPI uses .acpi_match_table, see
drivers/iio/magnetometer/ak8975.c for a simple example.
>
> > > > + },
> > > > + .probe = cros_ec_proximity_probe,
> > > > + .remove = cros_ec_proximity_remove,
> > > > +};
