Hi Guenter thanks for the review, answers bellow. Marc. Le 23/10/2015 18:52, Guenter Roeck a écrit :
The current code checks for the elapsed time since the last read. If this time is less than the default Conversion Time (CT = 2.2ms) times the averaging setting, then the driver will not actually access the registers. Hence the user can at best read fresh values every 2.2 x 1 ms, roughly 500Hz.On 10/23/2015 09:13 AM, Marc Titinger wrote:With the current implementation, the driver will prevent a readout at a pace faster than the default conversion time (2ms) times the averaging setting, min AVG being 1:1.Any sysfs "show" read access from the client app faster than 500 Hz will be500 Hz ?
Note that since the driver does bulk-read of all 8 registers (on ina226), it takes about 0.8 x 8 = 6.4 ms to update all/any values on BeagleBoneBlack, hence a client app cannot read fresh values faster than 160Hz per seconds.
This is the problem that motivates this rework.
'cached' by the driver, but actually since do_update reads all 8 registers,the best achievable measurement rate is roughly 8*800 us (for the time spent in i2c-core) i.e. <= 156Hz with Beagle Bone Black. Registers are now accessed individually through the regmap facility. For four measurements the readout rate is now around (1/(4*800us) = 312 Hz.I'll need some time to go through the patch in detail Some quick comments below.
ok
Signed-off-by: Marc Titinger <mtitinger@xxxxxxxxxxxx> ---drivers/hwmon/ina2xx.c | 359 ++++++++++++++++++++++---------------------------1 file changed, 159 insertions(+), 200 deletions(-) diff --git a/drivers/hwmon/ina2xx.c b/drivers/hwmon/ina2xx.c index 4d28150..e7c1aaa 100644 --- a/drivers/hwmon/ina2xx.c +++ b/drivers/hwmon/ina2xx.c @@ -37,6 +37,7 @@ #include <linux/of.h> #include <linux/delay.h> #include <linux/util_macros.h> +#include <linux/regmap.h> #include <linux/platform_data/ina2xx.h> @@ -48,32 +49,25 @@ #define INA2XX_CURRENT 0x04 /* readonly */ #define INA2XX_CALIBRATION 0x05 -/* INA226 register definitions */ -#define INA226_MASK_ENABLE 0x06 -#define INA226_ALERT_LIMIT 0x07 -#define INA226_DIE_ID 0xFF - -/* register count */ -#define INA219_REGISTERS 6 -#define INA226_REGISTERS 8 - -#define INA2XX_MAX_REGISTERS 8 +/* CONFIG register fields */ +#define INA2XX_AVG_MASK 0x0E00 +#define INA2XX_AVG_SHFT 9 /* settings - depend on use case */ #define INA219_CONFIG_DEFAULT 0x399F /* PGA=8 */ #define INA226_CONFIG_DEFAULT 0x4527 /* averages=16 */ /* worst case is 68.10 ms (~14.6Hz, ina219) */ -#define INA2XX_CONVERSION_RATE 15 #define INA2XX_MAX_DELAY 69 /* worst case delay in ms */ #define INA2XX_RSHUNT_DEFAULT 10000-/* bit mask for reading the averaging setting in the configuration register */-#define INA226_AVG_RD_MASK 0x0E00 - -#define INA226_READ_AVG(reg) (((reg) & INA226_AVG_RD_MASK) >> 9) -#define INA226_SHIFT_AVG(val) ((val) << 9) +/* Currently only handling common register set */ +static const struct regmap_config INA2XX_regmap_config = { + .reg_bits = 8, + .val_bits = 16, + .max_register = INA2XX_CALIBRATION, +}; /* common attrs, ina226 attrs and NULL */ #define INA2XX_MAX_ATTRIBUTE_GROUPS 3 @@ -89,7 +83,6 @@ enum ina2xx_ids { ina219, ina226 }; struct ina2xx_config { u16 config_default; int calibration_factor; - int registers; int shunt_div; int bus_voltage_shift; int bus_voltage_lsb; /* uV */ @@ -99,25 +92,16 @@ struct ina2xx_config { struct ina2xx_data { struct i2c_client *client; const struct ina2xx_config *config; - + struct regmap *regmap; long rshunt; - u16 curr_config; - - struct mutex update_lock; - bool valid; - unsigned long last_updated; - int update_interval; /* in jiffies */ - - int kind; + int valid;bool, please. But then valid was supposed to be used to indicate if the cache is valid, and should no longer be needed since there is no more cache. I'll have to spend more time to understand if and why it is still needed and how it is used.
Ah yes thanks, re-reading the initial code, I figure I 'invented' another need for this by mistake...
I'm happy to remove it. Good catch!
Agreed, as I said, it "morphed" the use of valid because of my earlier patch...const struct attribute_group *groups[INA2XX_MAX_ATTRIBUTE_GROUPS]; - u16 regs[INA2XX_MAX_REGISTERS]; }; static const struct ina2xx_config ina2xx_config[] = { [ina219] = { .config_default = INA219_CONFIG_DEFAULT, .calibration_factor = 40960000, - .registers = INA219_REGISTERS, .shunt_div = 100, .bus_voltage_shift = 3, .bus_voltage_lsb = 4000,@@ -126,7 +110,6 @@ static const struct ina2xx_config ina2xx_config[] = {[ina226] = { .config_default = INA226_CONFIG_DEFAULT, .calibration_factor = 5120000, - .registers = INA226_REGISTERS, .shunt_div = 400, .bus_voltage_shift = 0, .bus_voltage_lsb = 1250,@@ -142,9 +125,9 @@ static const struct ina2xx_config ina2xx_config[] = {*/static const int ina226_avg_tab[] = { 1, 4, 16, 64, 128, 256, 512, 1024 };-static int ina226_reg_to_interval(u16 config) +static int ina226_field_to_interval(int field) { - int avg = ina226_avg_tab[INA226_READ_AVG(config)]; + int avg = ina226_avg_tab[field]; /* * Multiply the total conversion time by the number of averages. @@ -153,24 +136,11 @@ static int ina226_reg_to_interval(u16 config)return DIV_ROUND_CLOSEST(avg * INA226_TOTAL_CONV_TIME_DEFAULT, 1000);} -static u16 ina226_interval_to_reg(int interval, u16 config) -{ - int avg, avg_bits; - - avg = DIV_ROUND_CLOSEST(interval * 1000, - INA226_TOTAL_CONV_TIME_DEFAULT); - avg_bits = find_closest(avg, ina226_avg_tab, - ARRAY_SIZE(ina226_avg_tab)); - - return (config & ~INA226_AVG_RD_MASK) | INA226_SHIFT_AVG(avg_bits); -} - -static void ina226_set_update_interval(struct ina2xx_data *data) +static int ina226_interval_to_field(int interval) { - int ms; - - ms = ina226_reg_to_interval(data->curr_config); - data->update_interval = msecs_to_jiffies(ms); + int avg = DIV_ROUND_CLOSEST(interval * 1000, + INA226_TOTAL_CONV_TIME_DEFAULT);+ return find_closest(avg, ina226_avg_tab, ARRAY_SIZE(ina226_avg_tab));} static int ina2xx_calibrate(struct ina2xx_data *data)@@ -178,8 +148,7 @@ static int ina2xx_calibrate(struct ina2xx_data *data)u16 val = DIV_ROUND_CLOSEST(data->config->calibration_factor, data->rshunt); - return i2c_smbus_write_word_swapped(data->client, - INA2XX_CALIBRATION, val); + return regmap_write(data->regmap, INA2XX_CALIBRATION, val); } /*@@ -187,15 +156,10 @@ static int ina2xx_calibrate(struct ina2xx_data *data)*/ static int ina2xx_init(struct ina2xx_data *data) { - struct i2c_client *client = data->client; - int ret; - - /* device configuration */ - ret = i2c_smbus_write_word_swapped(client, INA2XX_CONFIG, - data->curr_config); - if (ret < 0) + int ret = regmap_write(data->regmap, INA2XX_CONFIG, + data->config->config_default); + if (ret) return ret; - /* * Set current LSB to 1mA, shunt is in uOhms * (equation 13 in datasheet). @@ -203,22 +167,22 @@ static int ina2xx_init(struct ina2xx_data *data) return ina2xx_calibrate(data); } -static int ina2xx_do_update(struct device *dev) +static int ina2xx_show_common(struct device *dev, int index, int *val) { struct ina2xx_data *data = dev_get_drvdata(dev); - struct i2c_client *client = data->client; - int i, rv, retry; + int err, retry; - dev_dbg(&client->dev, "Starting ina2xx update\n"); + if (unlikely(!val)) + return -EINVAL;Please no unnecessary parameter checks. If val is NULL here, something is wrong with the implementation and needs to get fixed. Besides, the return value is only negative on errors. There is no need to pass an extra pointer to the value. Furthermore, regmap_read expects a pointer to an unsigned int, not a pointer to an int. Just return an error or the value returned by regmap. The calling codecan determine if there is an error by checking if the returned value is < 0.for (retry = 5; retry; retry--) { - /* Read all registers */ - for (i = 0; i < data->config->registers; i++) { - rv = i2c_smbus_read_word_swapped(client, i); - if (rv < 0) - return rv; - data->regs[i] = rv; - } + + /* Check for remaining registers in mask. */ + err = regmap_read(data->regmap, index, val); + if (err) + return err; + + dev_dbg(dev, "read %d, val = 0x%04x\n", index, *val); /* * If the current value in the calibration register is 0, the @@ -226,24 +190,24 @@ static int ina2xx_do_update(struct device *dev) * the chip has been reset let's check the calibration * register and reinitialize if needed. */ - if (data->regs[INA2XX_CALIBRATION] == 0) { - dev_warn(dev, "chip not calibrated, reinitializing\n"); - - rv = ina2xx_init(data); - if (rv < 0) - return rv; + if (!data->valid) { + dev_warn(dev, + "chip needs calibration, reinitializing\n"); + err = ina2xx_calibrate(data); + if (err) + return err;Ah, that is the context of "valid". I don't immediately see why moving its scope out of this function, and setting it to false in the calling code, is beneficial here.This also changes the dynamic of the driver. Previously it was specifically able to detect if the chip was power cycled (for whatever reason) and it would re-initialize itself. Now you use 'valid' and you defer (re-)initialization until chip registersare read. I don't immediately see if and how this is beneficial.
/* * Let's make sure the power and current registers * have been updated before trying again. */ msleep(INA2XX_MAX_DELAY); + + /* data valid once we have a cal value. */ + regmap_read(data->regmap, INA2XX_CALIBRATION, + &data->valid); continue; } - - data->last_updated = jiffies; - data->valid = 1; - return 0; } @@ -256,86 +220,89 @@ static int ina2xx_do_update(struct device *dev) return -ENODEV; } -static struct ina2xx_data *ina2xx_update_device(struct device *dev) +static ssize_t ina2xx_show_shunt(struct device *dev, + struct device_attribute *da, char *buf) { struct ina2xx_data *data = dev_get_drvdata(dev); - struct ina2xx_data *ret = data; - unsigned long after; - int rv; + int val, err; - mutex_lock(&data->update_lock); + err = ina2xx_show_common(dev, INA2XX_SHUNT_VOLTAGE, &val); + if (err) + return err; - after = data->last_updated + data->update_interval; - if (time_after(jiffies, after) || !data->valid) { - rv = ina2xx_do_update(dev); - if (rv < 0) - ret = ERR_PTR(rv); - } + val = DIV_ROUND_CLOSEST((s16) val, data->config->shunt_div); + return snprintf(buf, PAGE_SIZE, "%d\n", val); +} + +static ssize_t ina2xx_show_bus(struct device *dev, + struct device_attribute *da, char *buf) +{ + struct ina2xx_data *data = dev_get_drvdata(dev); + int val, err; + + err = ina2xx_show_common(dev, INA2XX_BUS_VOLTAGE, &val); + if (err) + return err; - mutex_unlock(&data->update_lock); - return ret; + val = (val >> data->config->bus_voltage_shift) + * data->config->bus_voltage_lsb; + val = DIV_ROUND_CLOSEST(val, 1000); + + return snprintf(buf, PAGE_SIZE, "%d\n", val); } -static int ina2xx_get_value(struct ina2xx_data *data, u8 reg) +static ssize_t ina2xx_show_pow(struct device *dev, + struct device_attribute *da, char *buf) { - int val; - - switch (reg) { - case INA2XX_SHUNT_VOLTAGE: - /* signed register */ - val = DIV_ROUND_CLOSEST((s16)data->regs[reg], - data->config->shunt_div); - break; - case INA2XX_BUS_VOLTAGE: - val = (data->regs[reg] >> data->config->bus_voltage_shift) - * data->config->bus_voltage_lsb; - val = DIV_ROUND_CLOSEST(val, 1000); - break; - case INA2XX_POWER: - val = data->regs[reg] * data->config->power_lsb; - break; - case INA2XX_CURRENT: - /* signed register, LSB=1mA (selected), in mA */ - val = (s16)data->regs[reg]; - break; - case INA2XX_CALIBRATION: - val = DIV_ROUND_CLOSEST(data->config->calibration_factor, - data->regs[reg]); - break; - default: - /* programmer goofed */ - WARN_ON_ONCE(1); - val = 0; - break; - } + struct ina2xx_data *data = dev_get_drvdata(dev); + int val, err; + + err = ina2xx_show_common(dev, INA2XX_POWER, &val); + if (err) + return err; + + val *= data->config->power_lsb; - return val; + return snprintf(buf, PAGE_SIZE, "%d\n", val); } -static ssize_t ina2xx_show_value(struct device *dev, - struct device_attribute *da, char *buf) +static ssize_t ina2xx_show_curr(struct device *dev, + struct device_attribute *da, char *buf) { - struct sensor_device_attribute *attr = to_sensor_dev_attr(da); - struct ina2xx_data *data = ina2xx_update_device(dev); + int val, err; - if (IS_ERR(data)) - return PTR_ERR(data); + err = ina2xx_show_common(dev, INA2XX_CURRENT, &val); + if (err) + return err; - return snprintf(buf, PAGE_SIZE, "%d\n", - ina2xx_get_value(data, attr->index)); + /* signed register, LSB=1mA (selected), in mA */ + return snprintf(buf, PAGE_SIZE, "%d\n", val); +} + +static ssize_t ina2xx_show_cal(struct device *dev, + struct device_attribute *da, char *buf) +{ + struct ina2xx_data *data = dev_get_drvdata(dev); + int val, err; + + err = ina2xx_show_common(dev, INA2XX_CALIBRATION, &val); + if (err) + return err; + + val = DIV_ROUND_CLOSEST(data->config->calibration_factor, val); + + return snprintf(buf, PAGE_SIZE, "%d\n", val); } static ssize_t ina2xx_set_shunt(struct device *dev, struct device_attribute *da, const char *buf, size_t count) { - struct ina2xx_data *data = ina2xx_update_device(dev); + struct ina2xx_data *data = dev_get_drvdata(dev); + unsigned long val; int status; - if (IS_ERR(data)) - return PTR_ERR(data); - status = kstrtoul(buf, 10, &val); if (status < 0) return status; @@ -345,12 +312,8 @@ static ssize_t ina2xx_set_shunt(struct device *dev, val > data->config->calibration_factor) return -EINVAL; - mutex_lock(&data->update_lock); data->rshunt = val; - status = ina2xx_calibrate(data); - mutex_unlock(&data->update_lock); - if (status < 0) - return status; + data->valid = 0; return count; }@@ -370,65 +333,58 @@ static ssize_t ina226_set_interval(struct device *dev,if (val > INT_MAX || val == 0) return -EINVAL; - mutex_lock(&data->update_lock); - data->curr_config = ina226_interval_to_reg(val, - data->regs[INA2XX_CONFIG]); - status = i2c_smbus_write_word_swapped(data->client, - INA2XX_CONFIG, - data->curr_config); + val = ina226_interval_to_field(val); - ina226_set_update_interval(data); - /* Make sure the next access re-reads all registers. */ - data->valid = 0; - mutex_unlock(&data->update_lock); + status = regmap_update_bits(data->regmap, INA2XX_CONFIG, + INA2XX_AVG_MASK, (val << INA2XX_AVG_SHFT)); if (status < 0) return status; + data->valid = 0; +This now causes re-calibration. Why ?
Mistake.
The interval value is no longer needed to compute a read "gard" delay, but the client Application may still need to set a different averaging value through this interval setting, based on the DUT characteristics. As of using separate functions, it fits better in the new logic of "per-register" accesses. It does not bloat the code either.return count; } static ssize_t ina226_show_interval(struct device *dev, struct device_attribute *da, char *buf) { - struct ina2xx_data *data = ina2xx_update_device(dev); + struct ina2xx_data *data = dev_get_drvdata(dev); + int status, val; - if (IS_ERR(data)) - return PTR_ERR(data); + status = regmap_read(data->regmap, INA2XX_CONFIG, &val); + if (status) + return status; + + val = (val & INA2XX_AVG_MASK) >> INA2XX_AVG_SHFT; + val = ina226_field_to_interval(val); /* - * We don't use data->update_interval here as we want to display - * the actual interval used by the chip and jiffies_to_msecs() - * doesn't seem to be accurate enough. + * We want to display the actual interval used by the chip. */ - return snprintf(buf, PAGE_SIZE, "%d\n", - ina226_reg_to_interval(data->regs[INA2XX_CONFIG])); + return snprintf(buf, PAGE_SIZE, "%d\n", val); } /* shunt voltage */ -static SENSOR_DEVICE_ATTR(in0_input, S_IRUGO, ina2xx_show_value, NULL, +static SENSOR_DEVICE_ATTR(in0_input, S_IRUGO, ina2xx_show_shunt, NULL, INA2XX_SHUNT_VOLTAGE);If you no longer use the register parameter, there is no need to provide it. But then I'll have to spend some time trying to understand _why_ you don'tuse it anymore and why you introduced separate show functions. Some explanation might help.
/* bus voltage */ -static SENSOR_DEVICE_ATTR(in1_input, S_IRUGO, ina2xx_show_value, NULL, +static SENSOR_DEVICE_ATTR(in1_input, S_IRUGO, ina2xx_show_bus, NULL, INA2XX_BUS_VOLTAGE); /* calculated current */-static SENSOR_DEVICE_ATTR(curr1_input, S_IRUGO, ina2xx_show_value, NULL,+static SENSOR_DEVICE_ATTR(curr1_input, S_IRUGO, ina2xx_show_curr, NULL, INA2XX_CURRENT); /* calculated power */-static SENSOR_DEVICE_ATTR(power1_input, S_IRUGO, ina2xx_show_value, NULL,+static SENSOR_DEVICE_ATTR(power1_input, S_IRUGO, ina2xx_show_pow, NULL, INA2XX_POWER); /* shunt resistance */ static SENSOR_DEVICE_ATTR(shunt_resistor, S_IRUGO | S_IWUSR, - ina2xx_show_value, ina2xx_set_shunt, + ina2xx_show_cal, ina2xx_set_shunt, INA2XX_CALIBRATION); -/* update interval (ina226 only) */ -static SENSOR_DEVICE_ATTR(update_interval, S_IRUGO | S_IWUSR, - ina226_show_interval, ina226_set_interval, 0); - /* pointers to created device attributes */ static struct attribute *ina2xx_attrs[] = { &sensor_dev_attr_in0_input.dev_attr.attr,@@ -443,6 +399,10 @@ static const struct attribute_group ina2xx_group = {.attrs = ina2xx_attrs, }; +/* update interval (ina226 only) */ +static SENSOR_DEVICE_ATTR(update_interval, S_IRUGO | S_IWUSR, + ina226_show_interval, ina226_set_interval, 0); +I don't really see the value of moving this code around. Please no unnecessary changes.static struct attribute *ina226_attrs[] = { &sensor_dev_attr_update_interval.dev_attr.attr, NULL,@@ -452,65 +412,65 @@ static const struct attribute_group ina226_group = {.attrs = ina226_attrs, }; + static int ina2xx_probe(struct i2c_client *client, const struct i2c_device_id *id) { - struct i2c_adapter *adapter = client->adapter; struct ina2xx_platform_data *pdata; struct device *dev = &client->dev; struct ina2xx_data *data; struct device *hwmon_dev; + struct regmap *regmap; u32 val; int ret, group = 0; - if (!i2c_check_functionality(adapter, I2C_FUNC_SMBUS_WORD_DATA)) - return -ENODEV; + /* Register regmap */ + regmap = devm_regmap_init_i2c(client, &INA2XX_regmap_config); + if (IS_ERR(regmap)) { + dev_err(dev, "failed to allocate register map\n"); + return PTR_ERR(regmap); + } data = devm_kzalloc(dev, sizeof(*data), GFP_KERNEL); if (!data) return -ENOMEM; - if (dev_get_platdata(dev)) { - pdata = dev_get_platdata(dev); - data->rshunt = pdata->shunt_uohms; - } else if (!of_property_read_u32(dev->of_node, - "shunt-resistor", &val)) { - data->rshunt = val; - } else { - data->rshunt = INA2XX_RSHUNT_DEFAULT; - } + data->regmap = regmap; - /* set the device type */ - data->kind = id->driver_data; - data->config = &ina2xx_config[data->kind]; - data->curr_config = data->config->config_default; + /* Set config according to device type. */ + data->config = &ina2xx_config[id->driver_data]; data->client = client; - /* - * Ina226 has a variable update_interval. For ina219 we - * use a constant value. + /* Check for shunt resistor value. + * Give precedence to device tree over must-recompile. */ - if (data->kind == ina226) - ina226_set_update_interval(data); - else - data->update_interval = HZ / INA2XX_CONVERSION_RATE;+ if (of_property_read_u32(dev->of_node, "shunt-resistor", &val) < 0) {+ pdata = dev_get_platdata(dev); + if (pdata) + val = pdata->shunt_uohms; + else + val = INA2XX_RSHUNT_DEFAULT; + }This changes priority from platform data first to devicetree configuration first. As such, it is an unrelated change. If needed, split into a separate patch, and
Yes I would do a separate patch normaly, agreed.
Changing the platform data requires changes in the kernel code, and hence recompilation. It seems a bit unexpected that setting a new value in the dtb will be ignored because there is a compiled-in platform data. Should'nt the dtb allow to override platform data ?explain the reasoning, please.
- if (data->rshunt <= 0 || - data->rshunt > data->config->calibration_factor) + if (val <= 0 || val > data->config->calibration_factor) { + dev_err(dev, "Invalid shunt resistor value %li", val); return -ENODEV; + } + data->rshunt = val; + /* Write config to chip, and calibrate */That comment, if needed, would be more appropriate at the top of the called function.ret = ina2xx_init(data); - if (ret < 0) { - dev_err(dev, "error configuring the device: %d\n", ret); - return -ENODEV; + if (ret) { + dev_err(dev, "error configuring the device.\n"); + return ret;Please explain whuy this code is better than before. To me it looks like personal preference, and I don't see the improvement.
It's not, it came from successive changes, will revert...
Appropriate it is, In some instances. Usually I dislike it too, question of style and opinion really I'd say.} - mutex_init(&data->update_lock); - + /* Set sensor group according to device type. */Seems to be obvious to me.data->groups[group++] = &ina2xx_group; - if (data->kind == ina226) + if (ina226 == id->driver_data)Not likely to accept Yoda programming I am.
data->groups[group++] = &ina226_group; + /* register to hwmon */Isn't that obvious ? I don't see the value of this comment.hwmon_dev = devm_hwmon_device_register_with_groups(dev, client->name,data, data->groups); if (IS_ERR(hwmon_dev)) @@ -541,7 +501,6 @@ static struct i2c_driver ina2xx_driver = { }; module_i2c_driver(ina2xx_driver); -Does this change fix a checkpatch problem ? If so, please submit as separate patch.If not, please no unnecessary whitespace changes.
Ok.
MODULE_AUTHOR("Lothar Felten <l-felten@xxxxxx>"); MODULE_DESCRIPTION("ina2xx driver"); MODULE_LICENSE("GPL");
_______________________________________________ lm-sensors mailing list lm-sensors@xxxxxxxxxxxxxx http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
