Re: [PATCH 5/5] platform/x86: dell-ddv: Add hwmon support

[Armin Wolf <W_Armin@gmx.de>]

Am 27.01.23 um 14:08 schrieb Guenter Roeck:

> On Thu, Jan 26, 2023 at 08:40:21PM +0100, Armin Wolf wrote:
>> Thanks to bugreport 216655 on bugzilla triggered by the
>> dell-smm-hwmon driver, the contents of the sensor buffers
>> could be almost completely decoded.
>> Add an hwmon interface for exposing the fan and thermal
>> sensor values. The debugfs interface remains in place to
>> aid in reverse-engineering of unknown sensor types
>> and the thermal buffer.
>>
>> Tested-by: Antonín Skala <skala.antonin@gmail.com>
>> Tested-by: Gustavo Walbon <gustavowalbon@gmail.com>
>> Signed-off-by: Armin Wolf <W_Armin@gmx.de>
>> ---
>>   drivers/platform/x86/dell/Kconfig        |   1 +
>>   drivers/platform/x86/dell/dell-wmi-ddv.c | 435 ++++++++++++++++++++++-
>>   2 files changed, 435 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/platform/x86/dell/Kconfig b/drivers/platform/x86/dell/Kconfig
>> index d319de8f2132..21a74b63d9b1 100644
>> --- a/drivers/platform/x86/dell/Kconfig
>> +++ b/drivers/platform/x86/dell/Kconfig
>> @@ -194,6 +194,7 @@ config DELL_WMI_DDV
>>   	default m
>>   	depends on ACPI_BATTERY
>>   	depends on ACPI_WMI
>> +	depends on HWMON
> Not sure if adding such a dependency is a good idea.
> Up to the maintainer to decide. Personally I would prefer
> something like
> 	depends on HWMON || HWMON=n
> and some conditionals in the code, as it is done with other drivers
> outside the hwmon directory.
>
Good point, i will include this in the next patch revision.

>>   	help
>>   	  This option adds support for WMI-based sensors like
>>   	  battery temperature sensors found on some Dell notebooks.
>> diff --git a/drivers/platform/x86/dell/dell-wmi-ddv.c b/drivers/platform/x86/dell/dell-wmi-ddv.c
>> index 9695bf493ea6..5b30bb85199e 100644
>> --- a/drivers/platform/x86/dell/dell-wmi-ddv.c
>> +++ b/drivers/platform/x86/dell/dell-wmi-ddv.c
>> @@ -13,6 +13,7 @@
>>   #include <linux/dev_printk.h>
>>   #include <linux/errno.h>
>>   #include <linux/kernel.h>
>> +#include <linux/hwmon.h>
>>   #include <linux/kstrtox.h>
>>   #include <linux/math.h>
>>   #include <linux/module.h>
>> @@ -21,10 +22,13 @@
>>   #include <linux/printk.h>
>>   #include <linux/seq_file.h>
>>   #include <linux/sysfs.h>
>> +#include <linux/types.h>
>>   #include <linux/wmi.h>
>>
>>   #include <acpi/battery.h>
>>
>> +#include <asm/unaligned.h>
>> +
>>   #define DRIVER_NAME	"dell-wmi-ddv"
>>
>>   #define DELL_DDV_SUPPORTED_VERSION_MIN	2
>> @@ -63,6 +67,29 @@ enum dell_ddv_method {
>>   	DELL_DDV_THERMAL_SENSOR_INFORMATION	= 0x22,
>>   };
>>
>> +struct fan_sensor_entry {
>> +	u8 type;
>> +	__le16 rpm;
>> +} __packed;
>> +
>> +struct thermal_sensor_entry {
>> +	u8 type;
>> +	s8 now;
>> +	s8 min;
>> +	s8 max;
>> +	u8 unknown;
>> +} __packed;
>> +
>> +struct combined_channel_info {
>> +	struct hwmon_channel_info info;
>> +	u32 config[];
>> +};
>> +
>> +struct combined_chip_info {
>> +	struct hwmon_chip_info chip;
>> +	const struct hwmon_channel_info *info[];
>> +};
>> +
>>   struct dell_wmi_ddv_data {
>>   	struct acpi_battery_hook hook;
>>   	struct device_attribute temp_attr;
>> @@ -70,6 +97,24 @@ struct dell_wmi_ddv_data {
>>   	struct wmi_device *wdev;
>>   };
>>
>> +static const char * const fan_labels[] = {
>> +	"CPU Fan",
>> +	"Chassis Motherboard Fan",
>> +	"Video Fan",
>> +	"Power Supply Fan",
>> +	"Chipset Fan",
>> +	"Memory Fan",
>> +	"PCI Fan",
>> +	"HDD Fan",
>> +};
>> +
>> +static const char * const fan_dock_labels[] = {
>> +	"Docking Chassis/Motherboard Fan",
>> +	"Docking Video Fan",
>> +	"Docking Power Supply Fan",
>> +	"Docking Chipset Fan",
>> +};
>> +
>>   static int dell_wmi_ddv_query_type(struct wmi_device *wdev, enum dell_ddv_method method, u32 arg,
>>   				   union acpi_object **result, acpi_object_type type)
>>   {
>> @@ -171,6 +216,386 @@ static int dell_wmi_ddv_query_string(struct wmi_device *wdev, enum dell_ddv_meth
>>   	return dell_wmi_ddv_query_type(wdev, method, arg, result, ACPI_TYPE_STRING);
>>   }
>>
>> +static int dell_wmi_ddv_query_sensors(struct wmi_device *wdev, enum dell_ddv_method method,
>> +				      size_t entry_size, union acpi_object **result, u64 *count)
>> +{
>> +	union acpi_object *obj;
>> +	u64 buffer_size;
>> +	u8 *buffer;
>> +	int ret;
>> +
>> +	ret = dell_wmi_ddv_query_buffer(wdev, method, 0, &obj);
>> +	if (ret < 0)
>> +		return ret;
>> +
>> +	buffer_size = obj->package.elements[0].integer.value;
>> +	buffer = obj->package.elements[1].buffer.pointer;
>> +	if (buffer_size % entry_size != 1 || buffer[buffer_size - 1] != 0xff) {
>> +		kfree(obj);
>> +
> Stray empty line
>
>> +		return -ENOMSG;
>> +	}
>> +
>> +	*count = (buffer_size - 1) / entry_size;
>> +	*result = obj;
>> +
>> +	return 0;
>> +}
>> +
>> +static umode_t dell_wmi_ddv_is_visible(const void *drvdata, enum hwmon_sensor_types type, u32 attr,
>> +				       int channel)
>> +{
>> +	return 0444;
>> +}
>> +
>> +static int dell_wmi_ddv_fan_read_channel(struct dell_wmi_ddv_data *data, u32 attr, int channel,
>> +					 long *val)
>> +{
>> +	struct fan_sensor_entry *entry;
>> +	union acpi_object *obj;
>> +	u64 count;
>> +	int ret;
>> +
>> +	ret = dell_wmi_ddv_query_sensors(data->wdev, DELL_DDV_FAN_SENSOR_INFORMATION,
>> +					 sizeof(*entry), &obj, &count);
>> +	if (ret < 0)
>> +		return ret;
>> +
>> +	entry = (struct fan_sensor_entry *)obj->package.elements[1].buffer.pointer;
>> +	if (count > channel) {
>> +		switch (attr) {
>> +		case hwmon_fan_input:
>> +			*val = get_unaligned_le16(&entry[channel].rpm);
>> +
> Another stray empty line. I see that "empty line before return or break"
> is common. Looks odd to me, and I don't see the point (it confuses
> the code flow for me and lets my brain focus on the empty line instead
> of the code in question), but I guess that is PoV. I won't comment on
> it further and let the maintainer decide.
>
>> +			break;
>> +		default:
>> +			ret = -EOPNOTSUPP;
>> +		}
>> +	} else {
>> +		ret = -ENXIO;
>> +	}
> Error handling should come first.
>> +
>> +	kfree(obj);
>> +
>> +	return ret;
>> +}
>> +
>> +static int dell_wmi_ddv_temp_read_channel(struct dell_wmi_ddv_data *data, u32 attr, int channel,
>> +					  long *val)
>> +{
>> +	struct thermal_sensor_entry *entry;
>> +	union acpi_object *obj;
>> +	u64 count;
>> +	int ret;
>> +
>> +	ret = dell_wmi_ddv_query_sensors(data->wdev, DELL_DDV_THERMAL_SENSOR_INFORMATION,
>> +					 sizeof(*entry), &obj, &count);
>> +	if (ret < 0)
>> +		return ret;
>> +
>> +	entry = (struct thermal_sensor_entry *)obj->package.elements[1].buffer.pointer;
>> +	if (count > channel) {
> This is a bit of Joda programming. It is really "channel < count",
> ie the requested channel number is in the range of channels reported
> by the WMI code. PoV, of course, but I find that the above makes the
> code more difficult to read.
>> +		switch (attr) {
>> +		case hwmon_temp_input:
>> +			*val = entry[channel].now * 1000;
>> +
>> +			break;
>> +		case hwmon_temp_min:
>> +			*val = entry[channel].min * 1000;
>> +
>> +			break;
>> +		case hwmon_temp_max:
>> +			*val = entry[channel].max * 1000;
>> +
>> +			break;
>> +		default:
>> +			ret = -EOPNOTSUPP;
> break; missing
>
>> +		}
>> +	} else {
>> +		ret = -ENXIO;
>> +	}
> Same as above - error handling should come first.
>
>> +
>> +	kfree(obj);
>> +
>> +	return ret;
>> +}
>> +
>> +static int dell_wmi_ddv_read(struct device *dev, enum hwmon_sensor_types type, u32 attr,
>> +			     int channel, long *val)
>> +{
>> +	struct dell_wmi_ddv_data *data = dev_get_drvdata(dev);
>> +
>> +	switch (type) {
>> +	case hwmon_fan:
>> +		return dell_wmi_ddv_fan_read_channel(data, attr, channel, val);
>> +	case hwmon_temp:
>> +		return dell_wmi_ddv_temp_read_channel(data, attr, channel, val);
>> +	default:
>> +		break;
>> +	}
>> +
>> +	return -EOPNOTSUPP;
>> +}
>> +
>> +static int dell_wmi_ddv_fan_read_string(struct dell_wmi_ddv_data *data, int channel,
>> +					const char **str)
>> +{
>> +	struct fan_sensor_entry *entry;
>> +	union acpi_object *obj;
>> +	u64 count;
>> +	u8 type;
>> +	int ret;
>> +
>> +	ret = dell_wmi_ddv_query_sensors(data->wdev, DELL_DDV_FAN_SENSOR_INFORMATION,
>> +					 sizeof(*entry), &obj, &count);
>> +	if (ret < 0)
>> +		return ret;
>> +
>> +	entry = (struct fan_sensor_entry *)obj->package.elements[1].buffer.pointer;
>> +	if (count > channel) {
>> +		type = entry[channel].type;
>> +
>> +		switch (type) {
>> +		case 0x00 ... 0x07:
>> +			*str = fan_labels[type];
>> +
>> +			break;
>> +		case 0x11 ... 0x14:
>> +			*str = fan_dock_labels[type - 0x11];
>> +
>> +			break;
>> +		default:
>> +			*str = "Unknown Fan";
> break; missing.
>
>> +		}
>> +	} else {
>> +		ret = -ENXIO;
>> +	}
> And again.
>
>> +
>> +	kfree(obj);
>> +
>> +	return ret;
>> +}
>> +
>> +static int dell_wmi_ddv_temp_read_string(struct dell_wmi_ddv_data *data, int channel,
>> +					 const char **str)
>> +{
>> +	struct thermal_sensor_entry *entry;
>> +	union acpi_object *obj;
>> +	u64 count;
>> +	int ret;
>> +
>> +	ret = dell_wmi_ddv_query_sensors(data->wdev, DELL_DDV_THERMAL_SENSOR_INFORMATION,
>> +					 sizeof(*entry), &obj, &count);
>> +	if (ret < 0)
>> +		return ret;
>> +
>> +	entry = (struct thermal_sensor_entry *)obj->package.elements[1].buffer.pointer;
>> +	if (count > channel) {
>> +		switch (entry[channel].type) {
>> +		case 0x00:
>> +			*str = "CPU";
>> +
>> +			break;
>> +		case 0x11:
>> +			*str = "Video";
>> +
>> +			break;
>> +		case 0x22:
>> +			*str = "Memory"; // sometimes called DIMM
> Personally I don't permit C++ style comments in a hwmon driver
> unless _all_ comments are C++ style. Just a remark here.
>
>> +
>> +			break;
>> +		case 0x33:
>> +			*str = "Other";
>> +
>> +			break;
>> +		case 0x44:
>> +			*str = "Ambient"; // sometimes called SKIN
>> +
>> +			break;
>> +		case 0x52:
>> +			*str = "SODIMM";
>> +
>> +			break;
>> +		case 0x55:
>> +			*str = "HDD";
>> +
>> +			break;
>> +		case 0x62:
>> +			*str = "SODIMM 2";
>> +
>> +			break;
>> +		case 0x73:
>> +			*str = "NB";
>> +
>> +			break;
>> +		case 0x83:
>> +			*str = "Charger";
>> +
>> +			break;
>> +		case 0xbb:
>> +			*str = "Memory 3";
>> +
>> +			break;
>> +		default:
>> +			*str = "Unknown";
> break; missing
> Ok, I guess this is on purpose. I personally don't permit
> that since it always leaves the question if it was on purpose or not.
>
>> +		}
>> +	} else {
>> +		ret = -ENXIO;
>> +	}
>> +
>> +	kfree(obj);
>> +
>> +	return ret;
>> +}
>> +
>> +static int dell_wmi_ddv_read_string(struct device *dev, enum hwmon_sensor_types type, u32 attr,
>> +				    int channel, const char **str)
>> +{
>> +	struct dell_wmi_ddv_data *data = dev_get_drvdata(dev);
>> +
>> +	switch (type) {
>> +	case hwmon_fan:
>> +		switch (attr) {
>> +		case hwmon_fan_label:
>> +			return dell_wmi_ddv_fan_read_string(data, channel, str);
>> +		default:
>> +			break;
>> +		}
>> +		break;
>> +	case hwmon_temp:
>> +		switch (attr) {
>> +		case hwmon_temp_label:
>> +			return dell_wmi_ddv_temp_read_string(data, channel, str);
>> +		default:
>> +			break;
>> +		}
>> +		break;
>> +	default:
>> +		break;
>> +	}
>> +
>> +	return -EOPNOTSUPP;
>> +}
>> +
>> +static const struct hwmon_ops dell_wmi_ddv_ops = {
>> +	.is_visible = dell_wmi_ddv_is_visible,
>> +	.read = dell_wmi_ddv_read,
>> +	.read_string = dell_wmi_ddv_read_string,
>> +};
>> +
>> +static struct hwmon_channel_info *dell_wmi_ddv_channel_create(struct device *dev, u64 count,
>> +							      enum hwmon_sensor_types type,
>> +							      u32 config)
>> +{
>> +	struct combined_channel_info *cinfo;
>> +	int i;
>> +
>> +	cinfo = devm_kzalloc(dev, struct_size(cinfo, config, count + 1), GFP_KERNEL);
>> +	if (!cinfo)
>> +		return ERR_PTR(-ENOMEM);
>> +
>> +	cinfo->info.type = type;
>> +	cinfo->info.config = cinfo->config;
>> +
>> +	for (i = 0; i < count; i++)
>> +		cinfo->config[i] = config;
>> +
>> +	return &cinfo->info;
>> +}
>> +
>> +static struct hwmon_channel_info *dell_wmi_ddv_channel_init(struct wmi_device *wdev,
>> +							    enum dell_ddv_method method,
>> +							    size_t entry_size,
>> +							    enum hwmon_sensor_types type,
>> +							    u32 config)
>> +{
>> +	union acpi_object *obj;
>> +	u64 count;
>> +	int ret;
>> +
>> +	ret = dell_wmi_ddv_query_sensors(wdev, method, entry_size, &obj, &count);
>> +	if (ret < 0)
>> +		return ERR_PTR(ret);
>> +
>> +	kfree(obj);
>> +
>> +	if (!count)
>> +		return ERR_PTR(-ENODEV);
>> +
>> +	return dell_wmi_ddv_channel_create(&wdev->dev, count, type, config);
>> +}
>> +
>> +static int dell_wmi_ddv_hwmon_add(struct dell_wmi_ddv_data *data)
>> +{
>> +	struct wmi_device *wdev = data->wdev;
>> +	struct combined_chip_info *cinfo;
>> +	struct device *hdev;
>> +	int index = 0;
>> +	int ret;
>> +
>> +	if (!devres_open_group(&wdev->dev, dell_wmi_ddv_hwmon_add, GFP_KERNEL))
>> +		return -ENOMEM;
>> +
>> +	cinfo = devm_kzalloc(&wdev->dev, struct_size(cinfo, info, 4), GFP_KERNEL);
>> +	if (!cinfo) {
>> +		ret = -ENOMEM;
>> +
>> +		goto err_release;
>> +	}
>> +
>> +	cinfo->chip.ops = &dell_wmi_ddv_ops;
>> +	cinfo->chip.info = cinfo->info;
>> +
>> +	cinfo->info[index] = dell_wmi_ddv_channel_create(&wdev->dev, 1, hwmon_chip,
>> +							 HWMON_C_REGISTER_TZ);
>> +
>> +	if (IS_ERR(cinfo->info[index])) {
>> +		ret = PTR_ERR(cinfo->info[index]);
>> +
>> +		goto err_release;
>> +	}
>> +
>> +	index++;
>> +
>> +	cinfo->info[index] = dell_wmi_ddv_channel_init(wdev, DELL_DDV_FAN_SENSOR_INFORMATION,
>> +						       sizeof(struct fan_sensor_entry), hwmon_fan,
>> +						       (HWMON_F_INPUT | HWMON_F_LABEL));
>> +	if (!IS_ERR(cinfo->info[index]))
>> +		index++;
>> +
>> +	cinfo->info[index] = dell_wmi_ddv_channel_init(wdev, DELL_DDV_THERMAL_SENSOR_INFORMATION,
>> +						       sizeof(struct thermal_sensor_entry),
>> +						       hwmon_temp, (HWMON_T_INPUT | HWMON_T_MIN |
>> +						       HWMON_T_MAX | HWMON_T_LABEL));
>> +	if (!IS_ERR(cinfo->info[index]))
>> +		index++;
>> +
>> +	if (!index) {
>> +		ret = -ENODEV;
>> +
>> +		goto err_release;
>> +	}
>> +
>> +	cinfo->info[index] = NULL;
>> +
>> +	hdev = devm_hwmon_device_register_with_info(&wdev->dev, "dell_ddv", data, &cinfo->chip,
>> +						    NULL);
>> +	if (IS_ERR(hdev)) {
>> +		ret = PTR_ERR(hdev);
>> +
>> +		goto err_release;
>> +	}
>> +
>> +	devres_close_group(&wdev->dev, dell_wmi_ddv_hwmon_add);
>> +
>> +	return 0;
>> +
>> +err_release:
>> +	devres_release_group(&wdev->dev, dell_wmi_ddv_hwmon_add);
>> +
>> +	return ret;
>> +}
>> +
>>   static int dell_wmi_ddv_battery_index(struct acpi_device *acpi_dev, u32 *index)
>>   {
>>   	const char *uid_str;
>> @@ -370,7 +795,15 @@ static int dell_wmi_ddv_probe(struct wmi_device *wdev, const void *context)
>>
>>   	dell_wmi_ddv_debugfs_init(wdev);
>>
>> -	return dell_wmi_ddv_battery_add(data);
>> +	ret = dell_wmi_ddv_hwmon_add(data);
>> +	if (ret < 0)
>> +		dev_dbg(&wdev->dev, "Unable to register hwmon interface: %d\n", ret);
>> +
>> +	ret = dell_wmi_ddv_battery_add(data);
>> +	if (ret < 0)
>> +		dev_dbg(&wdev->dev, "Unable to register acpi battery hook: %d\n", ret);
>> +
> This used to be an error, but no longer is. Not my call to make
> if this is acceptable, just pointing it out.

I decided to not treat the battery hook as essential function anymore because
the battery hook and hwmon functionality are more or less disconnected from
each other, so having the driver abort probing just because one functionality
could not be initialized seemed unreasonable to me.

I already thought about putting the battery hook and hwmon functionality into
separate drivers, with the main driver registering a MFD device or something similar.
Because apart from some generic routines, both functions are not connected in any way.

Is it acceptable to split the driver for such a thing?

Armin Wolf

>> +	return 0;
>>   }
>>
>>   static const struct wmi_device_id dell_wmi_ddv_id_table[] = {
>> --
>> 2.30.2
>>