Re: [PATCH v11 2/4] leds: add basic support for TI/National Semiconductor LP5812 LED Driver
From: Nam Tran <trannamatk@gmail.com>
Date: 2025-07-23 15:32:29
Also in:
linux-doc, linux-leds, lkml
On Fri, 18 Jul 2025, Krzysztof Kozlowski wrote:
On 14/07/2025 19:23, Nam Tran wrote:quoted
quoted
+static int lp5812_parse_led(struct device_node *np,+ struct lp5812_led_config *cfg, + int led_index) +{ + int num_colors = 0, ret; + + of_property_read_string(np, "label", &cfg[led_index].name); + + ret = of_property_read_u32(np, "reg", &cfg[led_index].chan_nr);You mix code for probe with code for regular operation. This is not expected and confusing. All functions related to probe must be before the probe function is defined.
I will restructured the code to move all probe-related helpers above the lp5812_probe() function.
quoted
+ +static struct lp5812_led *lp5812_dev_to_led(struct device *dev) +{ + struct led_classdev *cdev = dev_get_drvdata(dev); + const char *name = dev->platform_data; + + if (strcmp(name, LP5812_SC_LED) == 0) + return container_of(cdev, struct lp5812_led, cdev); + + return container_of((struct led_classdev_mc *)cdev, struct lp5812_led, mc_cdev);No, just pass correct pointer to platform data, no need with strcmp and then different container of...
I will remove the string-based `platform_data` handling and now store a direct pointer to `struct lp5812_led` in `dev->platform_data`. This allows simplifying `lp5812_dev_to_led()` without the need for `strcmp()` or multiple `container_of()` usages.
quoted
+static int lp5812_init_led(struct lp5812_led *led, struct lp5812_chip *chip, int chan) +{ + struct device *dev = &chip->client->dev; + struct mc_subled *mc_led_info; + struct led_classdev *led_cdev; + int i, ret = 0; + + if (chip->led_config[chan].name) { + led->cdev.name = chip->led_config[chan].name; + } else { + led->cdev.name = devm_kasprintf(dev, GFP_KERNEL, "%s:channel%d", + chip->label ? : chip->client->name, chan); + if (!led->cdev.name) + return -ENOMEM; + } + + if (!chip->led_config[chan].is_sc_led) { + mc_led_info = devm_kcalloc(dev, + chip->led_config[chan].num_colors, + sizeof(*mc_led_info), GFP_KERNEL); + if (!mc_led_info) + return -ENOMEM; + + led_cdev = &led->mc_cdev.led_cdev; + led_cdev->name = led->cdev.name; + led_cdev->brightness_set_blocking = lp5812_set_mc_brightness; + led->mc_cdev.num_colors = chip->led_config[chan].num_colors; + for (i = 0; i < led->mc_cdev.num_colors; i++) { + mc_led_info[i].color_index = + chip->led_config[chan].color_id[i]; + mc_led_info[i].channel = + chip->led_config[chan].led_id[i]; + } + + led->mc_cdev.subled_info = mc_led_info; + } else { + led->cdev.brightness_set_blocking = lp5812_set_brightness; + } + + led->cdev.groups = lp5812_led_groups; + led->chan_nr = chan; + + if (chip->led_config[chan].is_sc_led) { + ret = devm_led_classdev_register(dev, &led->cdev); + if (ret == 0) { + led->cdev.dev->platform_data = devm_kstrdup(dev, LP5812_SC_LED, GFP_KERNEL); + if (!led->cdev.dev->platform_data) + return -ENOMEM; + } + } else { + ret = devm_led_classdev_multicolor_register(dev, &led->mc_cdev); + if (ret == 0) { + led->mc_cdev.led_cdev.dev->platform_data = + devm_kstrdup(dev, LP5812_MC_LED, GFP_KERNEL); + if (!led->mc_cdev.led_cdev.dev->platform_data) + return -ENOMEM; + + ret = sysfs_create_groups(&led->mc_cdev.led_cdev.dev->kobj, + lp5812_led_groups); + if (ret) + dev_err(dev, "sysfs_create_groups failed\n"); + } + } + + if (ret) { + dev_err(dev, "led register err: %d\n", ret);Why are you printing same error multiple times?
I will remove the redundant error message at the end of `lp5812_init_led()` to avoid double-printing. Now only the specific failure points log detailed errors.
quoted
+static int lp5812_probe(struct i2c_client *client) +{ + struct lp5812_chip *chip; + const struct i2c_device_id *id = i2c_client_get_device_id(client); + struct device_node *np = dev_of_node(&client->dev); + struct lp5812_led *led; + int ret; + + chip = devm_kzalloc(&client->dev, sizeof(*chip), GFP_KERNEL); + if (!chip) + return -ENOMEM; + + chip->cfg = i2c_get_match_data(client); + + if (np) { + ret = lp5812_of_populate_pdata(&client->dev, np, chip); + if (ret) + goto err_init; + } else { + return dev_err_probe(&client->dev, -EINVAL, "No platform data\n");This is confusing syntax. Expected is: if (!missing something) return -EINVAL
I will update it.
The other problem is that you claim this can match and bind without OF, but here you say it is a requirement. So either this code is wrong or your I2C ID table should be removed.
I will update the probe logic for clarity and removed the i2c_device_id table and .id_table field.
quoted
+ led = devm_kcalloc(&client->dev, chip->num_channels, sizeof(*led), GFP_KERNEL); + if (!led) + return -ENOMEM; + + chip->client = client; + + mutex_init(&chip->lock); + + i2c_set_clientdata(client, led); + + ret = lp5812_init_device(chip); + if (ret) + goto err_init; + + dev_info(&client->dev, "%s Programmable led chip found\n", id->name);Drop. You have only one device type (look at your binding), so you cannot "find" devices.
I will drop it.
quoted
+static void lp5812_remove(struct i2c_client *client) +{ + struct lp5812_led *led = i2c_get_clientdata(client); + + lp5812_unregister_sysfs(led, led->chip); + lp5812_deinit_device(led->chip); + + dev_info(&client->dev, "Removed driver\n");No, drop, useless.
I will remove it. Thank you for your thorough review and helpful suggestions. Best regards, Nam Tran