On Mon, Jun 01, 2020 at 02:41:20PM +0100, Jonathan Cameron wrote: > ... > > > > > > > + > > > > > > + return ret; > > > > > > +} > > > > > > + > > > > > > +int scd30_probe(struct device *dev, int irq, const char *name, void *priv, > > > > > > + scd30_command_t command) > > > > > > +{ > > > > > > + static const unsigned long scd30_scan_masks[] = { 0x07, 0x00 }; > > > > > > + struct scd30_state *state; > > > > > > + struct iio_dev *indio_dev; > > > > > > + int ret; > > > > > > + u16 val; > > > > > > + > > > > > > + indio_dev = devm_iio_device_alloc(dev, sizeof(*state)); > > > > > > + if (!indio_dev) > > > > > > + return -ENOMEM; > > > > > > + > > > > > > + state = iio_priv(indio_dev); > > > > > > + state->dev = dev; > > > > > > > > > > Doesn't seem to be used. > > > > > > > > > > > + state->priv = priv; > > > > > > > > > > What's this for? At least at first glance I can't find it being used > > > > > anywhere. > > > > > > > > > > > + state->irq = irq; > > > > > > + state->pressure_comp = SCD30_PRESSURE_COMP_DEFAULT; > > > > > > + state->meas_interval = SCD30_MEAS_INTERVAL_DEFAULT; > > > > > > + state->command = command; > > > > > > + mutex_init(&state->lock); > > > > > > + init_completion(&state->meas_ready); > > > > > > + > > > > > > + dev_set_drvdata(dev, indio_dev); > > > > > > + > > > > > > + indio_dev->dev.parent = dev; > > > > > > > > > > Side note that there is a series moving this into the core under revision at > > > > > the moment. Hopefully I'll remember to fix this up when applying your patch > > > > > if that one has gone in ahead of it. > > > > > > > > > > > + indio_dev->info = &scd30_info; > > > > > > + indio_dev->name = name; > > > > > > + indio_dev->channels = scd30_channels; > > > > > > + indio_dev->num_channels = ARRAY_SIZE(scd30_channels); > > > > > > + indio_dev->modes = INDIO_DIRECT_MODE; > > > > > > + indio_dev->available_scan_masks = scd30_scan_masks; > > > > > > + > > > > > > + state->vdd = devm_regulator_get(dev, "vdd"); > > > > > > + if (IS_ERR(state->vdd)) { > > > > > > + if (PTR_ERR(state->vdd) == -EPROBE_DEFER) > > > > > > + return -EPROBE_DEFER; > > > > > > + > > > > > > + dev_err(dev, "failed to get regulator\n"); > > > > > > + return PTR_ERR(state->vdd); > > > > > > + } > > > > > > + > > > > > > + ret = regulator_enable(state->vdd); > > > > > > + if (ret) > > > > > > + return ret; > > > > > > + > > > > > > + ret = devm_add_action_or_reset(dev, scd30_disable_regulator, state); > > > > > > + if (ret) > > > > > > + return ret; > > > > > > + > > > > > > > > > > A comment here on why it makes sense to register this here. What > > > > > started mesurement? It seems that happens well below here so > > > > > we should really call this after that start all. > > > > > > > > > > > > > Sensor after being powered up starts in mode it was left in. > > > > Chances are it was continuous mode and we want to shut it down. > > > > > > That's fine. The question is why 'here' as opposed to after the below where you > > > put it into continuous mode. > > > > > > > Let's suppose sensor got energized and started measuring. Then without > > registering action which stops measurement we jump to device reset etc. > > > > Now if reset failed for whatever reason (same applies to everything > > below reset) devm will gracefully unwind previous actions but sensor > > will continue doing his job. But there's no point. Better to save some > > milliaps for later. > > I'm not convinced. Note that in your example, the sensor was already on. > If we never loaded the driver it would stay on. We should > deal with only problems we have potentially created. > Fair enough. > If reset fails, it shouldn't 'enable' the sensor hence we are just in > the same state as if the driver hadn't loaded at all. Also if reset > fails the chance of a disable succeeding is very low. > > The code is much more logical with this being done immediately after > the action to enable thing we are disabling. So please put it down > there. > > > > > > > In case we have real regulator then there's no real issue because > > power gets cut off during cleanup. > > > > Quite often though there's only a dummy one which does nothing useful > > except making regulator framework happy. > > > > So my thinking here is that we're slightly better off registering > > scd30_stop_meas() action earlier to prevent such scenario from happening. > > > > > > > > > > > > + ret = devm_add_action_or_reset(dev, scd30_stop_meas, state); > > > > > > + if (ret) > > > > > > + return ret; > > > > > > + > > > > > > + ret = scd30_reset(state); > > > > > > + if (ret) { > > > > > > + dev_err(dev, "failed to reset device: %d\n", ret); > > > > > > + return ret; > > > > > > + } > > > > > > + > > > > > > + if (state->irq > 0) { > > > > > > + ret = scd30_setup_trigger(indio_dev); > > > > > > + if (ret) { > > > > > > + dev_err(dev, "failed to setup trigger: %d\n", ret); > > > > > > + return ret; > > > > > > + } > > > > > > + } > > > > > > + > > > > > > + ret = devm_iio_triggered_buffer_setup(dev, indio_dev, NULL, > > > > > > + scd30_trigger_handler, NULL); > > > > > > + if (ret) > > > > > > + return ret; > > > > > > + > > > > > > + ret = scd30_command_read(state, CMD_FW_VERSION, &val); > > > > > > + if (ret) { > > > > > > + dev_err(dev, "failed to read firmware version: %d\n", ret); > > > > > > + return ret; > > > > > > + } > > > > > > + dev_info(dev, "firmware version: %d.%d\n", val >> 8, (char)val); > > > > > > + > > > > > > + ret = scd30_command_write(state, CMD_MEAS_INTERVAL, > > > > > > + state->meas_interval); > > > > > > + if (ret) { > > > > > > + dev_err(dev, "failed to set measurement interval: %d\n", ret); > > > > > > + return ret; > > > > > > + } > > > > > > + > > > > > > + ret = scd30_command_write(state, CMD_START_MEAS, state->pressure_comp); > > > > > > + if (ret) { > > > > > > + dev_err(dev, "failed to start measurement: %d\n", ret); > > > > > > + return ret; > > > > > > + } > Here is where we should register that cleanup handler. > > Jonathan > > > > > > > + > > > > > > + return devm_iio_device_register(dev, indio_dev); > > > > > > +} > > > > > > +EXPORT_SYMBOL(scd30_probe); > > > > > > + > > > > > > +MODULE_AUTHOR("Tomasz Duszynski <tomasz.duszynski@xxxxxxxxxxx>"); > > > > > > +MODULE_DESCRIPTION("Sensirion SCD30 carbon dioxide sensor core driver"); > > > > > > +MODULE_LICENSE("GPL v2"); > > > > > > -- > > > > > > 2.26.2 > > > > > > > > > > > > > > > > > > >