Message ID | 20240909-tps25990-v1-0-39b37e43e795@baylibre.com |
---|---|
Headers | show |
Series | hwmon: pmbus: add tps25990 efuse support | expand |
On 9/9/24 08:39, Jerome Brunet wrote: > Add support for POWER_GOOD_ON and POWER_GOOD_OFF standard PMBus commands. > > For PMBus devices that offer a POWER_GOOD signal, these commands are used > for setting the output voltage at which a power good signal should be > asserted and negated. > > Power Good signals are device and manufacturer specific. Many factors other > than output voltage may be used to determine whether or not the POWER_GOOD > signal is to be asserted. PMBus device users are instructed to consult the > device manufacturer’s product literature for the specifics of the device > they are using. > > Note that depending on the choice of the device manufacturer that a device > may drive a POWER_GOOD signal high or low to indicate that the signal is > asserted. > > Signed-off-by: Jerome Brunet <jbrunet@baylibre.com> > --- > drivers/hwmon/pmbus/pmbus.h | 3 +++ > drivers/hwmon/pmbus/pmbus_core.c | 6 ++++++ > 2 files changed, 9 insertions(+) > > diff --git a/drivers/hwmon/pmbus/pmbus.h b/drivers/hwmon/pmbus/pmbus.h > index 5d5dc774187b..e322d2dd9fb7 100644 > --- a/drivers/hwmon/pmbus/pmbus.h > +++ b/drivers/hwmon/pmbus/pmbus.h > @@ -78,6 +78,9 @@ enum pmbus_regs { > PMBUS_IIN_OC_FAULT_LIMIT = 0x5B, > PMBUS_IIN_OC_WARN_LIMIT = 0x5D, > > + PMBUS_POWER_GOOD_ON = 0x5E, > + PMBUS_POWER_GOOD_OFF = 0x5F, > + > PMBUS_POUT_OP_FAULT_LIMIT = 0x68, > PMBUS_POUT_OP_WARN_LIMIT = 0x6A, > PMBUS_PIN_OP_WARN_LIMIT = 0x6B, > diff --git a/drivers/hwmon/pmbus/pmbus_core.c b/drivers/hwmon/pmbus/pmbus_core.c > index 0ea6fe7eb17c..94ddf0166770 100644 > --- a/drivers/hwmon/pmbus/pmbus_core.c > +++ b/drivers/hwmon/pmbus/pmbus_core.c > @@ -1768,6 +1768,12 @@ static const struct pmbus_limit_attr vout_limit_attrs[] = { > .attr = "crit", > .alarm = "crit_alarm", > .sbit = PB_VOLTAGE_OV_FAULT, > + }, { > + .reg = PMBUS_POWER_GOOD_ON, > + .attr = "good_on", > + }, { > + .reg = PMBUS_POWER_GOOD_OFF, > + .attr = "good_off", > }, { > .reg = PMBUS_VIRT_READ_VOUT_AVG, > .update = true, > Those attributes are not hardware monitoring attributes and therefore not acceptable. In general I am not sure if they should be configurable in the first place, but definitely not from the hardware monitoring subsystem. Maybe the regulator subsystem callbacks set_over_voltage_protection and set_under_voltage_protection would be appropriate (with severity REGULATOR_SEVERITY_PROT), but that should be discussed with regulator subsystem maintainers. Thanks, Guenter
On 9/9/24 08:39, Jerome Brunet wrote: > Add initial support for the Texas Instruments TPS25990 eFuse. > This adds the basic PMBUS telemetry support for the device. > > Signed-off-by: Jerome Brunet <jbrunet@baylibre.com> > --- > Documentation/hwmon/tps25990.rst | 141 ++++++++++++ > drivers/hwmon/pmbus/Kconfig | 17 ++ > drivers/hwmon/pmbus/Makefile | 1 + > drivers/hwmon/pmbus/tps25990.c | 474 +++++++++++++++++++++++++++++++++++++++ > 4 files changed, 633 insertions(+) > > diff --git a/Documentation/hwmon/tps25990.rst b/Documentation/hwmon/tps25990.rst > new file mode 100644 > index 000000000000..7b3ef724008a > --- /dev/null > +++ b/Documentation/hwmon/tps25990.rst > @@ -0,0 +1,141 @@ > +.. SPDX-License-Identifier: GPL-2.0 > + > +Kernel driver tps25990 > +====================== > + > +Supported chips: > + > + * TI TPS25990 > + > + Prefix: 'tps25990' > + > + * Datasheet > + > + Publicly available at Texas Instruments website: https://www.ti.com/lit/gpn/tps25990 > + > +Author: > + > + Jerome Brunet <jbrunet@baylibre.com> > + > +Description > +----------- > + > +This driver implements support for TI TPS25990 eFuse. > +This is an integrated, high-current circuit protection and power > +management device with PMBUS interface > + > +Device compliant with: > + > +- PMBus rev 1.3 interface. > + > +Device supports direct format for reading input voltages, > +output voltage, input current, input power and temperature. > + > +The driver exports the following attributes via the 'sysfs' files > +for input current: > + > +**curr1_average** > + > +**curr1_crit** > + > +**curr1_crit_alarm** > + > +**curr1_highest** > + > +**curr1_input** > + > +**curr1_label** > + > +**curr1_max** > + > +**curr1_max_alarm** > + > +The driver provides the following attributes for main input voltage: > + > +**in1_average** > + > +**in1_crit** > + > +**in1_crit_alarm** > + > +**in1_highest** > + > +**in1_input** > + > +**in1_label** > + > +**in1_lcrit** > + > +**in1_lcrit_alarm** > + > +**in1_lowest** > + > +**in1_max** > + > +**in1_max_alarm** > + > +**in1_min** > + > +**in1_min_alarm** > + > +The driver provides the following attributes for auxiliary input voltage: > + > +**in2_input** > + > +**in2_label** > + > +The driver provides the following attributes for output voltage: > + > +**in3_average** > + > +**in3_good_off** As mentioned in the other patch, those mnon-standard attributes are not acceptable. > + > +**in3_input** > + > +**in3_label** > + > +**in3_lowest** > + > +**in3_min** > + > +**in3_min_alarm** > + > +The driver provides the following attributes for input power: > + > +**power1_alarm** > + > +**power1_average** > + > +**power1_input** > + > +**power1_input_highest** > + > +**power1_label** > + > +**power1_max** > + > +The driver provides the following attributes for temperature: > + > +**temp1_average** > + > +**temp1_crit** > + > +**temp1_crit_alarm** > + > +**temp1_highest** > + > +**temp1_input** > + > +**temp1_max** > + > +**temp1_max_alarm** > + > +The driver provides the following attributes for history: > + > +**samples** > + > +**average_history_reset** > + > +**highest_history_reset** > + > +**lowest_history_reset** > diff --git a/drivers/hwmon/pmbus/Kconfig b/drivers/hwmon/pmbus/Kconfig > index a4f02cad92fd..3559864e232d 100644 > --- a/drivers/hwmon/pmbus/Kconfig > +++ b/drivers/hwmon/pmbus/Kconfig > @@ -510,6 +510,23 @@ config SENSORS_TDA38640_REGULATOR > If you say yes here you get regulator support for Infineon > TDA38640 as regulator. > > +config SENSORS_TPS25990 > + tristate "TI TPS25990" > + help > + If you say yes here you get hardware monitoring support for TI > + TPS25990. > + > + This driver can also be built as a module. If so, the module will > + be called tps25990. > + > +config SENSORS_TPS25990_REGULATOR > + bool "Regulator support for TPS25990 and compatibles" > + depends on SENSORS_TPS25990 && REGULATOR > + default SENSORS_TPS2599 > + help > + If you say yes here you get regulator support for Texas Instruments > + TPS25990. > + > config SENSORS_TPS40422 > tristate "TI TPS40422" > help > diff --git a/drivers/hwmon/pmbus/Makefile b/drivers/hwmon/pmbus/Makefile > index d00bcc758b97..3d3183f8d2a7 100644 > --- a/drivers/hwmon/pmbus/Makefile > +++ b/drivers/hwmon/pmbus/Makefile > @@ -51,6 +51,7 @@ obj-$(CONFIG_SENSORS_PXE1610) += pxe1610.o > obj-$(CONFIG_SENSORS_Q54SJ108A2) += q54sj108a2.o > obj-$(CONFIG_SENSORS_STPDDC60) += stpddc60.o > obj-$(CONFIG_SENSORS_TDA38640) += tda38640.o > +obj-$(CONFIG_SENSORS_TPS25990) += tps25990.o > obj-$(CONFIG_SENSORS_TPS40422) += tps40422.o > obj-$(CONFIG_SENSORS_TPS53679) += tps53679.o > obj-$(CONFIG_SENSORS_TPS546D24) += tps546d24.o > diff --git a/drivers/hwmon/pmbus/tps25990.c b/drivers/hwmon/pmbus/tps25990.c > new file mode 100644 > index 000000000000..14290c4c71dd > --- /dev/null > +++ b/drivers/hwmon/pmbus/tps25990.c > @@ -0,0 +1,474 @@ > +// SPDX-License-Identifier: GPL-2.0 > +// > +// Copyright (c) 2024 BayLibre, SAS. > +// Author: Jerome Brunet <jbrunet@baylibre.com> > + > +#include <linux/debugfs.h> > +#include <linux/err.h> > +#include <linux/hwmon-sysfs.h> > +#include <linux/i2c.h> > +#include <linux/init.h> > +#include <linux/kernel.h> > +#include <linux/module.h> > + > +#include "pmbus.h" > + > +#define TPS25990_READ_VAUX 0xd0 > +#define TPS25990_READ_VIN_MIN 0xd1 > +#define TPS25990_READ_VIN_PEAK 0xd2 > +#define TPS25990_READ_IIN_PEAK 0xd4 > +#define TPS25990_READ_PIN_PEAK 0xd5 > +#define TPS25990_READ_TEMP_AVG 0xd6 > +#define TPS25990_READ_TEMP_PEAK 0xd7 > +#define TPS25990_READ_VOUT_MIN 0xda > +#define TPS25990_READ_VIN_AVG 0xdc > +#define TPS25990_READ_VOUT_AVG 0xdd > +#define TPS25990_READ_IIN_AVG 0xde > +#define TPS25990_READ_PIN_AVG 0xdf > +#define TPS25990_VIREF 0xe0 > +#define TPS25990_PK_MIN_AVG 0xea > +#define PK_MIN_AVG_RST_PEAK BIT(7) > +#define PK_MIN_AVG_RST_AVG BIT(6) > +#define PK_MIN_AVG_RST_MIN BIT(5) > +#define PK_MIN_AVG_AVG_CNT GENMASK(2, 0) > +#define TPS25990_MFR_WRITE_PROTECT 0xf8 > +#define TPS25990_UNLOCKED BIT(7) > + > +#define TPS25990_8B_SHIFT 2 > +#define TPS25990_VIN_OVF_NUM 525100 > +#define TPS25990_VIN_OVF_DIV 10163 > +#define TPS25990_VIN_OVF_OFF 155 > +#define TPS25990_IIN_OCF_NUM 953800 > +#define TPS25990_IIN_OCF_DIV 129278 > +#define TPS25990_IIN_OCF_OFF 157 > + > +#define TPS25990_DEFAULT_RIMON 910000 If this is the default, why would it make sense to even specify the property in milli-ohm ? I mean, sure, the devices can be stacked, but it seems unrealistic to assume that there are hundreds or even dozens of devices in parallel. > + > +static int tps25990_mfr_write_protect(struct i2c_client *client, bool protect) > +{ > + return pmbus_write_byte_data(client, -1, TPS25990_MFR_WRITE_PROTECT, > + protect ? 0x0 : 0xa2); > +} > + > +static int tps25990_mfr_write_protect_active(struct i2c_client *client) > +{ > + int ret = pmbus_read_byte_data(client, -1, TPS25990_MFR_WRITE_PROTECT); > + > + if (ret < 0) > + return ret; > + > + return !(ret & TPS25990_UNLOCKED); > +} > + > +#if IS_ENABLED(CONFIG_DEBUG_FS) > +static int tps25990_write_protect_get(void *data, u64 *val) > +{ > + struct i2c_client *client = data; > + > + return tps25990_mfr_write_protect_active(client); > +} > + > +static int tps25990_write_protect_set(void *data, u64 val) > +{ > + struct i2c_client *client = data; > + > + if (val > 1) > + return -EINVAL; > + > + return tps25990_mfr_write_protect(client, val); > +} > + > +DEFINE_DEBUGFS_ATTRIBUTE(tps25990_write_protect_fops, > + tps25990_write_protect_get, > + tps25990_write_protect_set, > + "%llu\n"); > + > +static int tps25990_init_debugfs(struct i2c_client *client) > +{ > + struct dentry *dir; > + > + dir = pmbus_get_debugfs_dir(client); > + if (!dir) > + return -ENOENT; > + > + debugfs_create_file("write_protect", 0644, dir, > + client, &tps25990_write_protect_fops); > + > + return 0; > +} > + > +#else > +static inline int tps25990_init_debugfs(struct i2c_client *client) > +{ > + return 0; > +} > +#endif > + In general it is extremely undesirable to overwrite write protection. Many chips support such attributes. If write protection is enabled, it means that the board vendor does not want to have them changed. Granted, that can be overwritten with direct i2c commands, but that is what it should be. Anyone who really wants to disable write protection should have to dig deeper than just writing into a debugfs or sysfs attribute. Otherwise the protection becomes worthless. If this is, for example, needed for production to write initial settings, the production scripts should disable (or enable) write protection by writing directly into command registers. > +/* > + * TPS25990 has history reset based on MIN/AVG/PEAK instead of per sensor type > + * Emulate the behaviour a pmbus limit_attr would have for consistency > + * - Read: Do nothing and emit 0 > + * - Write: Check the input is a number and reset > + */ > +static ssize_t tps25990_history_reset_show(struct device *dev, > + struct device_attribute *devattr, > + char *buf) > +{ > + return sysfs_emit(buf, "0\n"); > +} > + > +static ssize_t tps25990_history_reset_store(struct device *dev, > + struct device_attribute *devattr, > + const char *buf, size_t count) > +{ > + struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr); > + struct i2c_client *client = to_i2c_client(dev->parent); > + s64 val; > + int ret; > + > + if (kstrtos64(buf, 10, &val) < 0) > + return -EINVAL; > + > + ret = pmbus_update_byte_data(client, 0, TPS25990_PK_MIN_AVG, > + BIT(attr->index), BIT(attr->index)); > + if (ret < 0) > + return ret; > + > + return count; > +} > + > +static SENSOR_DEVICE_ATTR_RW(highest_history_reset, tps25990_history_reset, 7); > +static SENSOR_DEVICE_ATTR_RW(average_history_reset, tps25990_history_reset, 6); > +static SENSOR_DEVICE_ATTR_RW(lowest_history_reset, tps25990_history_reset, 5); That is not a unique problem, and not a reason to introduce non-standard attributes. Just attach the attribute to the first channel and document that it resets all channels. > + > +static struct attribute *tps25990_attrs[] = { > + &sensor_dev_attr_highest_history_reset.dev_attr.attr, > + &sensor_dev_attr_average_history_reset.dev_attr.attr, > + &sensor_dev_attr_lowest_history_reset.dev_attr.attr, > + NULL, > +}; > + > +ATTRIBUTE_GROUPS(tps25990); > + > +static int tps25990_get_addr(int reg) > +{ > + switch (reg) { > + case PMBUS_SMBALERT_MASK: > + /* > + * Note: PMBUS_SMBALERT_MASK is not implemented on this chip > + * Writing to this address raises CML errors. > + * Instead it provides ALERT_MASK which allows to set the mask > + * for each of the status registers, but not the specific bits > + * in them. > + * The default setup assert SMBA# if any bit is set in any of the > + * status registers the chip has. This is as close as we can get > + * to what pmbus_irq_setup() would set, sooo ... do nothing. > + */ > + return -ENXIO; Many chips have that problem. The core code ignores errors, and attempts to write the command are limited to initialization. This is not a reason to overwrite the command like this. If this does cause a real a problem wit hthe chip (other than setting CML errors, which many chips not supporting the command do), we should define a flag in include/linux/pmbus.h and explain its need. > + case PMBUS_IIN_OC_FAULT_LIMIT: > + /* > + * VIREF directly sets the over-current limit at which the eFuse > + * will turn the FET off and trigger a fault. Expose it through > + * this generic property instead of a manufacturer specific one. > + */ > + return TPS25990_VIREF; I don't see the value in this override. See below. > + case PMBUS_VIRT_READ_VIN_MAX: > + return TPS25990_READ_VIN_PEAK; > + case PMBUS_VIRT_READ_VIN_MIN: > + return TPS25990_READ_VIN_MIN; > + case PMBUS_VIRT_READ_VIN_AVG: > + return TPS25990_READ_VIN_AVG; > + case PMBUS_VIRT_READ_VOUT_MIN: > + return TPS25990_READ_VOUT_MIN; > + case PMBUS_VIRT_READ_VOUT_AVG: > + return TPS25990_READ_VOUT_AVG; > + case PMBUS_VIRT_READ_IIN_AVG: > + return TPS25990_READ_IIN_AVG; > + case PMBUS_VIRT_READ_IIN_MAX: > + return TPS25990_READ_IIN_PEAK; > + case PMBUS_VIRT_READ_TEMP_AVG: > + return TPS25990_READ_TEMP_AVG; > + case PMBUS_VIRT_READ_TEMP_MAX: > + return TPS25990_READ_TEMP_PEAK; > + case PMBUS_VIRT_READ_PIN_AVG: > + return TPS25990_READ_PIN_AVG; > + case PMBUS_VIRT_READ_PIN_MAX: > + return TPS25990_READ_PIN_PEAK; > + case PMBUS_VIRT_READ_VMON: > + return TPS25990_READ_VAUX; > + case PMBUS_VIRT_SAMPLES: > + return TPS25990_PK_MIN_AVG; default: missing. > + } > + > + /* Let the register check do its job */ > + if (reg < PMBUS_VIRT_BASE) > + return reg; > + > + return -ENXIO; This needs to be more specific: The code should only return -ENXIO if auto-detection can not handle the case. "Return -ENXIO for all unsupported virtual registers" is unexpected. That situation should be handled by the PMBus core. > +} > + > +/* > + * Some registers use a different scale than the one registered with > + * pmbus_driver_info. An extra conversion step is necessary to adapt > + * the register value to the conversion on the sensor type > + */ > +static int tps25990_read_adapt_value(int reg, int val) > +{ > + switch (reg) { > + case PMBUS_VIN_UV_WARN_LIMIT: > + case PMBUS_VIN_UV_FAULT_LIMIT: > + case PMBUS_VIN_OV_WARN_LIMIT: > + case PMBUS_VOUT_UV_WARN_LIMIT: > + case PMBUS_IIN_OC_WARN_LIMIT: > + case PMBUS_OT_WARN_LIMIT: > + case PMBUS_OT_FAULT_LIMIT: > + case PMBUS_PIN_OP_WARN_LIMIT: > + case PMBUS_POWER_GOOD_OFF: > + /* > + * These registers provide an 8 bits value instead of a > + * 10bits one. Just shifting twice the register value is > + * enough to make the sensor type conversion work, even > + * if the datasheet provides different m, b and R for > + * those. > + */ > + val <<= TPS25990_8B_SHIFT; > + break; > + > + case PMBUS_VIN_OV_FAULT_LIMIT: > + val = DIV_ROUND_CLOSEST(val * TPS25990_VIN_OVF_NUM, TPS25990_VIN_OVF_DIV); > + val += TPS25990_VIN_OVF_OFF; > + break; > + > + case PMBUS_IIN_OC_FAULT_LIMIT: > + val = DIV_ROUND_CLOSEST(val * TPS25990_IIN_OCF_NUM, TPS25990_IIN_OCF_DIV); > + val += TPS25990_IIN_OCF_OFF; > + break; > + > + case PMBUS_VIRT_SAMPLES: > + val = 1 << val; > + break; default case missing. > + } > + > + return val; > +} > + > +static int tps25990_read_word(struct i2c_client *client, > + int page, int phase, int reg) > +{ > + int ret, addr; > + > + addr = tps25990_get_addr(reg); > + if (addr < 0) > + return addr; > + > + switch (reg) { > + case PMBUS_VIRT_SAMPLES: > + ret = pmbus_read_byte_data(client, page, addr); Mapping the register name in tps25990_get_addr() is unnecessary and misleading. It is well known that TPS25990_PK_MIN_AVG is to be used. Do it here. > + ret = FIELD_GET(PK_MIN_AVG_AVG_CNT, ret); > + break; > + > + case PMBUS_IIN_OC_FAULT_LIMIT: > + ret = pmbus_read_byte_data(client, page, addr); > + break; > + Same here. > + default: > + ret = pmbus_read_word_data(client, page, -1, addr); This is unexpected for registers not handled locally. Expectation is that -ENODATA is returned for those, to be handled in the core. > + break; > + } > + > + if (ret >= 0) > + ret = tps25990_read_adapt_value(reg, ret); > + > + return ret; > +} > + > +static int tps25990_write_adapt_value(int reg, int val) > +{ > + switch (reg) { > + case PMBUS_VIN_UV_WARN_LIMIT: > + case PMBUS_VIN_UV_FAULT_LIMIT: > + case PMBUS_VIN_OV_WARN_LIMIT: > + case PMBUS_VOUT_UV_WARN_LIMIT: > + case PMBUS_IIN_OC_WARN_LIMIT: > + case PMBUS_OT_WARN_LIMIT: > + case PMBUS_OT_FAULT_LIMIT: > + case PMBUS_PIN_OP_WARN_LIMIT: > + case PMBUS_POWER_GOOD_OFF: > + val >>= TPS25990_8B_SHIFT; > + val = clamp(val, 0, 0xff); Why clamp() here but clamp_val() elsewhere ? > + break; > + > + case PMBUS_VIN_OV_FAULT_LIMIT: > + val -= TPS25990_VIN_OVF_OFF; > + val = DIV_ROUND_CLOSEST(val * TPS25990_VIN_OVF_DIV, TPS25990_VIN_OVF_NUM); > + val = clamp_val(val, 0, 0xf); > + break; > + > + case PMBUS_IIN_OC_FAULT_LIMIT: > + val -= TPS25990_IIN_OCF_OFF; > + val = DIV_ROUND_CLOSEST(val * TPS25990_IIN_OCF_DIV, TPS25990_IIN_OCF_NUM); > + val = clamp_val(val, 0, 0x3f); > + break; > + > + case PMBUS_VIRT_SAMPLES: > + val = clamp_val(val, 1, 1 << PK_MIN_AVG_AVG_CNT); > + val = ilog2(val); > + break; default: missing. > + } > + > + return val; > +} > + > +static int tps25990_write_word(struct i2c_client *client, > + int page, int reg, u16 value) > +{ > + int addr, ret; > + > + addr = tps25990_get_addr(reg); > + if (addr < 0) > + return addr; > + > + value = tps25990_write_adapt_value(reg, value); > + > + switch (reg) { > + case PMBUS_VIRT_SAMPLES: > + ret = pmbus_update_byte_data(client, page, addr, > + PK_MIN_AVG_AVG_CNT, > + FIELD_PREP(PK_MIN_AVG_AVG_CNT, value)); > + break; > + > + case PMBUS_IIN_OC_FAULT_LIMIT: > + ret = pmbus_write_byte_data(client, page, addr, > + value); > + break; > + > + default: > + ret = pmbus_write_word_data(client, page, addr, value); > + break; Same comments as for read functions. > + } > + > + return ret; > +} > + > +#if IS_ENABLED(CONFIG_SENSORS_TPS25990_REGULATOR) > +static const struct regulator_desc tps25990_reg_desc[] = { > + PMBUS_REGULATOR_ONE("vout"), > +}; > +#endif > + > +static const struct pmbus_driver_info tps25990_base_info = { > + .pages = 1, > + .format[PSC_VOLTAGE_IN] = direct, > + .m[PSC_VOLTAGE_IN] = 5251, > + .b[PSC_VOLTAGE_IN] = 0, > + .R[PSC_VOLTAGE_IN] = -2, > + .format[PSC_VOLTAGE_OUT] = direct, > + .m[PSC_VOLTAGE_OUT] = 5251, > + .b[PSC_VOLTAGE_OUT] = 0, > + .R[PSC_VOLTAGE_OUT] = -2, > + .format[PSC_TEMPERATURE] = direct, > + .m[PSC_TEMPERATURE] = 140, > + .b[PSC_TEMPERATURE] = 32100, > + .R[PSC_TEMPERATURE] = -2, > + /* > + * Current and Power measurement depends on the ohm value > + * of Rimon. m is multiplied by 1000 below to have an integer > + * and -3 is added to R to compensate. > + */ > + .format[PSC_CURRENT_IN] = direct, > + .m[PSC_CURRENT_IN] = 9538, > + .b[PSC_CURRENT_IN] = 0, > + .R[PSC_CURRENT_IN] = -6, > + .format[PSC_POWER] = direct, > + .m[PSC_POWER] = 4901, > + .b[PSC_POWER] = 0, > + .R[PSC_POWER] = -7, > + .func[0] = (PMBUS_HAVE_VIN | > + PMBUS_HAVE_VOUT | > + PMBUS_HAVE_VMON | > + PMBUS_HAVE_IIN | > + PMBUS_HAVE_PIN | > + PMBUS_HAVE_TEMP | > + PMBUS_HAVE_STATUS_VOUT | > + PMBUS_HAVE_STATUS_IOUT | > + PMBUS_HAVE_STATUS_INPUT | > + PMBUS_HAVE_STATUS_TEMP | > + PMBUS_HAVE_SAMPLES), > + .read_word_data = tps25990_read_word, > + .write_word_data = tps25990_write_word, > + .groups = tps25990_groups, > + > +#if IS_ENABLED(CONFIG_SENSORS_TPS25990_REGULATOR) > + .reg_desc = tps25990_reg_desc, > + .num_regulators = ARRAY_SIZE(tps25990_reg_desc), > +#endif > +}; > + > +static const struct i2c_device_id tps25990_i2c_id[] = { > + { "tps25990" }, > + {} > +}; > +MODULE_DEVICE_TABLE(i2c, tps25990_i2c_id); > + > +static const struct of_device_id tps25990_of_match[] = { > + { .compatible = "ti,tps25990" }, > + {} > +}; > +MODULE_DEVICE_TABLE(of, tps25990_of_match); > + > +static int tps25990_probe(struct i2c_client *client) > +{ > + struct device *dev = &client->dev; > + struct pmbus_driver_info *info; > + u32 rimon = TPS25990_DEFAULT_RIMON; > + int ret; > + > + ret = device_property_read_u32(dev, "ti,rimon-milli-ohms", &rimon); > + if (ret == -EINVAL) { > + dev_warn(dev, > + "using default rimon: current and power scale possibly wrong\n"); This is not an appropriate warning. It is perfectly fine to load the driver if there is no ti,rimon-milli-ohms property. > + } else if (ret < 0) { > + return dev_err_probe(dev, ret, "failed get rimon\n"); > + } > + > + /* > + * TPS25990 may be stacked with several TPS25895, allowing a higher > + * current. The higher the allowed current is, the lower rimon > + * will be. How low it can realistically get is unknown. > + * To avoid problems with precision later on, rimon is provided in > + * milli Ohms. This is a precaution to keep a stable ABI. > + * At the moment, doing the calculation with rimon in milli Ohms > + * would overflow the s32 'm' in the direct conversion. Convert it > + * back to Ohms until greater precision is actually needed. > + */ > + rimon /= 1000; > + Seems to me it would make more sense to limit the valid range of ti,rimon-milli-ohms to avoid the overflow. But then I really don't understand the reasoning to provide the property in milli-ohm, given the default value of 910 Ohm. What is a realistic lowest value that would make sense ? But even if it is less than 1 Ohm I don't understand why it would make sense to completely ignore it. > + info = devm_kmemdup(dev, &tps25990_base_info, sizeof(*info), GFP_KERNEL); > + if (!info) > + return -ENOMEM; > + > + /* Adapt the current and power scale for each instance */ > + info->m[PSC_CURRENT_IN] *= rimon; > + info->m[PSC_POWER] *= rimon; Any rimon value < 1000 mOhm will result in m values of 0. > + > + ret = pmbus_do_probe(client, info); > + if (ret < 0) > + return ret; > + > + return tps25990_init_debugfs(client); debugfs initialization is not expected to fail. > +} > + > +static struct i2c_driver tps25990_driver = { > + .driver = { > + .name = "tps25990", > + .of_match_table = tps25990_of_match, > + }, > + .probe = tps25990_probe, > + .id_table = tps25990_i2c_id, > +}; > +module_i2c_driver(tps25990_driver); > + > +MODULE_AUTHOR("Jerome Brunet <jbrunet@baylibre.com>"); > +MODULE_DESCRIPTION("PMBUS driver for TPS25990 eFuse"); > +MODULE_LICENSE("GPL"); > +MODULE_IMPORT_NS(PMBUS); >
On Mon 09 Sep 2024 at 11:16, Guenter Roeck <linux@roeck-us.net> wrote: > On 9/9/24 08:39, Jerome Brunet wrote: >> Add support for POWER_GOOD_ON and POWER_GOOD_OFF standard PMBus commands. >> For PMBus devices that offer a POWER_GOOD signal, these commands are used >> for setting the output voltage at which a power good signal should be >> asserted and negated. >> Power Good signals are device and manufacturer specific. Many factors >> other >> than output voltage may be used to determine whether or not the POWER_GOOD >> signal is to be asserted. PMBus device users are instructed to consult the >> device manufacturer’s product literature for the specifics of the device >> they are using. >> Note that depending on the choice of the device manufacturer that a >> device >> may drive a POWER_GOOD signal high or low to indicate that the signal is >> asserted. >> Signed-off-by: Jerome Brunet <jbrunet@baylibre.com> >> --- >> drivers/hwmon/pmbus/pmbus.h | 3 +++ >> drivers/hwmon/pmbus/pmbus_core.c | 6 ++++++ >> 2 files changed, 9 insertions(+) >> diff --git a/drivers/hwmon/pmbus/pmbus.h b/drivers/hwmon/pmbus/pmbus.h >> index 5d5dc774187b..e322d2dd9fb7 100644 >> --- a/drivers/hwmon/pmbus/pmbus.h >> +++ b/drivers/hwmon/pmbus/pmbus.h >> @@ -78,6 +78,9 @@ enum pmbus_regs { >> PMBUS_IIN_OC_FAULT_LIMIT = 0x5B, >> PMBUS_IIN_OC_WARN_LIMIT = 0x5D, >> + PMBUS_POWER_GOOD_ON = 0x5E, >> + PMBUS_POWER_GOOD_OFF = 0x5F, >> + >> PMBUS_POUT_OP_FAULT_LIMIT = 0x68, >> PMBUS_POUT_OP_WARN_LIMIT = 0x6A, >> PMBUS_PIN_OP_WARN_LIMIT = 0x6B, >> diff --git a/drivers/hwmon/pmbus/pmbus_core.c b/drivers/hwmon/pmbus/pmbus_core.c >> index 0ea6fe7eb17c..94ddf0166770 100644 >> --- a/drivers/hwmon/pmbus/pmbus_core.c >> +++ b/drivers/hwmon/pmbus/pmbus_core.c >> @@ -1768,6 +1768,12 @@ static const struct pmbus_limit_attr vout_limit_attrs[] = { >> .attr = "crit", >> .alarm = "crit_alarm", >> .sbit = PB_VOLTAGE_OV_FAULT, >> + }, { >> + .reg = PMBUS_POWER_GOOD_ON, >> + .attr = "good_on", >> + }, { >> + .reg = PMBUS_POWER_GOOD_OFF, >> + .attr = "good_off", >> }, { >> .reg = PMBUS_VIRT_READ_VOUT_AVG, >> .update = true, >> > > Those attributes are not hardware monitoring attributes and therefore not > acceptable. In general I am not sure if they should be configurable in the > first place, but definitely not from the hardware monitoring subsystem. > Maybe the regulator subsystem callbacks set_over_voltage_protection and > set_under_voltage_protection would be appropriate (with severity > REGULATOR_SEVERITY_PROT), but that should be discussed with regulator > subsystem maintainers. According to PMBUS spec, there is no protection associated with that command. It just tells when the output voltage is considered good, when it is not. What it does after that really depends the device, it may drive a pin for example (or an LED indicator in my case). It is very similar to 'crit' or other limits in that sense, I think. I don't really get why such property is not OK in hwmon then and why it should not be configurable, if the other limits are ? I don't mind dropping that completly, that change is not critical to me. The intent was to contribute something to overall pmbus support. > > Thanks, > Guenter
On Mon 09 Sep 2024 at 15:52, Guenter Roeck <linux@roeck-us.net> wrote: [...] >> + >> +#define TPS25990_DEFAULT_RIMON 910000 > > If this is the default, why would it make sense to even specify the > property in milli-ohm ? I mean, sure, the devices can be stacked, > but it seems unrealistic to assume that there are hundreds or even > dozens of devices in parallel. Indeed not hundreds, I suppose :) 8 would already land you in a position where milli-Ohms could be useful. Is this realistic, I honestly don't know. More on this below. >> + >> +static int tps25990_mfr_write_protect(struct i2c_client *client, bool protect) >> +{ >> + return pmbus_write_byte_data(client, -1, TPS25990_MFR_WRITE_PROTECT, >> + protect ? 0x0 : 0xa2); >> +} >> + >> +static int tps25990_mfr_write_protect_active(struct i2c_client *client) >> +{ >> + int ret = pmbus_read_byte_data(client, -1, TPS25990_MFR_WRITE_PROTECT); >> + >> + if (ret < 0) >> + return ret; >> + >> + return !(ret & TPS25990_UNLOCKED); >> +} >> + >> +#if IS_ENABLED(CONFIG_DEBUG_FS) >> +static int tps25990_write_protect_get(void *data, u64 *val) >> +{ >> + struct i2c_client *client = data; >> + >> + return tps25990_mfr_write_protect_active(client); >> +} >> + >> +static int tps25990_write_protect_set(void *data, u64 val) >> +{ >> + struct i2c_client *client = data; >> + >> + if (val > 1) >> + return -EINVAL; >> + >> + return tps25990_mfr_write_protect(client, val); >> +} >> + >> +DEFINE_DEBUGFS_ATTRIBUTE(tps25990_write_protect_fops, >> + tps25990_write_protect_get, >> + tps25990_write_protect_set, >> + "%llu\n"); >> + >> +static int tps25990_init_debugfs(struct i2c_client *client) >> +{ >> + struct dentry *dir; >> + >> + dir = pmbus_get_debugfs_dir(client); >> + if (!dir) >> + return -ENOENT; >> + >> + debugfs_create_file("write_protect", 0644, dir, >> + client, &tps25990_write_protect_fops); >> + >> + return 0; >> +} >> + >> +#else >> +static inline int tps25990_init_debugfs(struct i2c_client *client) >> +{ >> + return 0; >> +} >> +#endif >> + > > In general it is extremely undesirable to overwrite write protection. > Many chips support such attributes. If write protection is enabled, > it means that the board vendor does not want to have them changed. According to documentation, it protects against "unintented" writes, not 'wrong' or 'malicious'. If one goes in debugfs and write just '0' to a file, there is an intent at least. > Granted, that can be overwritten with direct i2c commands, but that > is what it should be. Anyone who really wants to disable write protection > should have to dig deeper than just writing into a debugfs or sysfs attribute. > Otherwise the protection becomes worthless. > If this is, for example, needed > for production to write initial settings, the production scripts should > disable (or enable) write protection by writing directly into command > registers. As I wrote in the cover letter, the write protection is always active on chip startup and it locks down almost everything, including things you may need to write past production, in the field. The history reset below is an example of such thing. To 'safely' remove the protection by writing i2c commands from userspace: * the device will need be unbinded first, * call i2cset * bind the device again That seems really cumbersome to do something like an history reset. Is this what you are suggesting ? bind/unbind could be skipped by forcing i2cset but that would add danger where we certainly don't want it. > >> +/* >> + * TPS25990 has history reset based on MIN/AVG/PEAK instead of per sensor type >> + * Emulate the behaviour a pmbus limit_attr would have for consistency >> + * - Read: Do nothing and emit 0 >> + * - Write: Check the input is a number and reset >> + */ >> +static ssize_t tps25990_history_reset_show(struct device *dev, >> + struct device_attribute *devattr, >> + char *buf) >> +{ >> + return sysfs_emit(buf, "0\n"); >> +} >> + >> +static ssize_t tps25990_history_reset_store(struct device *dev, >> + struct device_attribute *devattr, >> + const char *buf, size_t count) >> +{ >> + struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr); >> + struct i2c_client *client = to_i2c_client(dev->parent); >> + s64 val; >> + int ret; >> + >> + if (kstrtos64(buf, 10, &val) < 0) >> + return -EINVAL; >> + >> + ret = pmbus_update_byte_data(client, 0, TPS25990_PK_MIN_AVG, >> + BIT(attr->index), BIT(attr->index)); >> + if (ret < 0) >> + return ret; >> + >> + return count; >> +} >> + >> +static SENSOR_DEVICE_ATTR_RW(highest_history_reset, tps25990_history_reset, 7); >> +static SENSOR_DEVICE_ATTR_RW(average_history_reset, tps25990_history_reset, 6); >> +static SENSOR_DEVICE_ATTR_RW(lowest_history_reset, tps25990_history_reset, 5); > > That is not a unique problem, and not a reason to introduce non-standard attributes. > Just attach the attribute to the first channel and document that it resets all > channels. Not sure I got this right so I'll rephrase. I should: * Pick a channel, say vin * Map the virtual reset register to hit the 3 resets above * Put in the documentation that it resets the other channels as well * Not allow independent resets of min/max/avg, just all 3 together ? > >> + >> +static struct attribute *tps25990_attrs[] = { >> + &sensor_dev_attr_highest_history_reset.dev_attr.attr, >> + &sensor_dev_attr_average_history_reset.dev_attr.attr, >> + &sensor_dev_attr_lowest_history_reset.dev_attr.attr, >> + NULL, >> +}; >> + >> +ATTRIBUTE_GROUPS(tps25990); >> + >> +static int tps25990_get_addr(int reg) >> +{ >> + switch (reg) { >> + case PMBUS_SMBALERT_MASK: >> + /* >> + * Note: PMBUS_SMBALERT_MASK is not implemented on this chip >> + * Writing to this address raises CML errors. >> + * Instead it provides ALERT_MASK which allows to set the mask >> + * for each of the status registers, but not the specific bits >> + * in them. >> + * The default setup assert SMBA# if any bit is set in any of the >> + * status registers the chip has. This is as close as we can get >> + * to what pmbus_irq_setup() would set, sooo ... do nothing. >> + */ >> + return -ENXIO; > > Many chips have that problem. The core code ignores errors, and attempts to write > the command are limited to initialization. This is not a reason to overwrite > the command like this. If this does cause a real a problem wit hthe chip (other > than setting CML errors, which many chips not supporting the command do), > we should define a flag in include/linux/pmbus.h and explain its need. CML is error is the problem. Following pmbus_irq_setup() there is an uncleared fault because there is no register check on PMBUS_SMBALERT_MASK. When pmbus_core then gets here: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/hwmon/pmbus/pmbus_core.c?h=v6.11-rc7#n3386 pmbus_check_block_register() fails because of the uncleared fault and the 'mfr_id' property is silently not registered, eventhough the register is supported by the chip. This is how I noticed the problem. So, should I add flag in include/linux/pmbus.h to skip PMBUS_SMBALERT_MASK setup ? Another possibility is to call register_check() on it before using PMBUS_SMBALERT_MASK in pmbus_core. > >> + case PMBUS_IIN_OC_FAULT_LIMIT: >> + /* >> + * VIREF directly sets the over-current limit at which the eFuse >> + * will turn the FET off and trigger a fault. Expose it through >> + * this generic property instead of a manufacturer specific one. >> + */ >> + return TPS25990_VIREF; > > I don't see the value in this override. See below. > >> + case PMBUS_VIRT_READ_VIN_MAX: >> + return TPS25990_READ_VIN_PEAK; >> + case PMBUS_VIRT_READ_VIN_MIN: >> + return TPS25990_READ_VIN_MIN; >> + case PMBUS_VIRT_READ_VIN_AVG: >> + return TPS25990_READ_VIN_AVG; >> + case PMBUS_VIRT_READ_VOUT_MIN: >> + return TPS25990_READ_VOUT_MIN; >> + case PMBUS_VIRT_READ_VOUT_AVG: >> + return TPS25990_READ_VOUT_AVG; >> + case PMBUS_VIRT_READ_IIN_AVG: >> + return TPS25990_READ_IIN_AVG; >> + case PMBUS_VIRT_READ_IIN_MAX: >> + return TPS25990_READ_IIN_PEAK; >> + case PMBUS_VIRT_READ_TEMP_AVG: >> + return TPS25990_READ_TEMP_AVG; >> + case PMBUS_VIRT_READ_TEMP_MAX: >> + return TPS25990_READ_TEMP_PEAK; >> + case PMBUS_VIRT_READ_PIN_AVG: >> + return TPS25990_READ_PIN_AVG; >> + case PMBUS_VIRT_READ_PIN_MAX: >> + return TPS25990_READ_PIN_PEAK; >> + case PMBUS_VIRT_READ_VMON: >> + return TPS25990_READ_VAUX; >> + case PMBUS_VIRT_SAMPLES: >> + return TPS25990_PK_MIN_AVG; > > default: missing. > >> + } >> + >> + /* Let the register check do its job */ >> + if (reg < PMBUS_VIRT_BASE) >> + return reg; >> + >> + return -ENXIO; > > This needs to be more specific: The code should only return -ENXIO > if auto-detection can not handle the case. "Return -ENXIO for all > unsupported virtual registers" is unexpected. That situation should > be handled by the PMBus core. > >> +} >> + >> +/* >> + * Some registers use a different scale than the one registered with >> + * pmbus_driver_info. An extra conversion step is necessary to adapt >> + * the register value to the conversion on the sensor type >> + */ >> +static int tps25990_read_adapt_value(int reg, int val) >> +{ >> + switch (reg) { >> + case PMBUS_VIN_UV_WARN_LIMIT: >> + case PMBUS_VIN_UV_FAULT_LIMIT: >> + case PMBUS_VIN_OV_WARN_LIMIT: >> + case PMBUS_VOUT_UV_WARN_LIMIT: >> + case PMBUS_IIN_OC_WARN_LIMIT: >> + case PMBUS_OT_WARN_LIMIT: >> + case PMBUS_OT_FAULT_LIMIT: >> + case PMBUS_PIN_OP_WARN_LIMIT: >> + case PMBUS_POWER_GOOD_OFF: >> + /* >> + * These registers provide an 8 bits value instead of a >> + * 10bits one. Just shifting twice the register value is >> + * enough to make the sensor type conversion work, even >> + * if the datasheet provides different m, b and R for >> + * those. >> + */ >> + val <<= TPS25990_8B_SHIFT; >> + break; >> + >> + case PMBUS_VIN_OV_FAULT_LIMIT: >> + val = DIV_ROUND_CLOSEST(val * TPS25990_VIN_OVF_NUM, TPS25990_VIN_OVF_DIV); >> + val += TPS25990_VIN_OVF_OFF; >> + break; >> + >> + case PMBUS_IIN_OC_FAULT_LIMIT: >> + val = DIV_ROUND_CLOSEST(val * TPS25990_IIN_OCF_NUM, TPS25990_IIN_OCF_DIV); >> + val += TPS25990_IIN_OCF_OFF; >> + break; >> + >> + case PMBUS_VIRT_SAMPLES: >> + val = 1 << val; >> + break; > > default case missing. > >> + } >> + >> + return val; >> +} >> + >> +static int tps25990_read_word(struct i2c_client *client, >> + int page, int phase, int reg) >> +{ >> + int ret, addr; >> + >> + addr = tps25990_get_addr(reg); >> + if (addr < 0) >> + return addr; >> + >> + switch (reg) { >> + case PMBUS_VIRT_SAMPLES: >> + ret = pmbus_read_byte_data(client, page, addr); > > Mapping the register name in tps25990_get_addr() is unnecessary > and misleading. It is well known that TPS25990_PK_MIN_AVG is to be > used. Do it here. My intent was to do the mapping in one place instead of repeating for both read and write, not be misleading. I'll change it. > >> + ret = FIELD_GET(PK_MIN_AVG_AVG_CNT, ret); >> + break; >> + >> + case PMBUS_IIN_OC_FAULT_LIMIT: >> + ret = pmbus_read_byte_data(client, page, addr); >> + break; >> + > Same here. > >> + default: >> + ret = pmbus_read_word_data(client, page, -1, addr); > > This is unexpected for registers not handled locally. Expectation is > that -ENODATA is returned for those, to be handled in the core. Got it. Thanks. > >> + break; >> + } >> + >> + if (ret >= 0) >> + ret = tps25990_read_adapt_value(reg, ret); >> + >> + return ret; >> +} >> + >> +static int tps25990_write_adapt_value(int reg, int val) >> +{ >> + switch (reg) { >> + case PMBUS_VIN_UV_WARN_LIMIT: >> + case PMBUS_VIN_UV_FAULT_LIMIT: >> + case PMBUS_VIN_OV_WARN_LIMIT: >> + case PMBUS_VOUT_UV_WARN_LIMIT: >> + case PMBUS_IIN_OC_WARN_LIMIT: >> + case PMBUS_OT_WARN_LIMIT: >> + case PMBUS_OT_FAULT_LIMIT: >> + case PMBUS_PIN_OP_WARN_LIMIT: >> + case PMBUS_POWER_GOOD_OFF: >> + val >>= TPS25990_8B_SHIFT; >> + val = clamp(val, 0, 0xff); > > Why clamp() here but clamp_val() elsewhere ? > >> + break; >> + >> + case PMBUS_VIN_OV_FAULT_LIMIT: >> + val -= TPS25990_VIN_OVF_OFF; >> + val = DIV_ROUND_CLOSEST(val * TPS25990_VIN_OVF_DIV, TPS25990_VIN_OVF_NUM); >> + val = clamp_val(val, 0, 0xf); >> + break; >> + >> + case PMBUS_IIN_OC_FAULT_LIMIT: >> + val -= TPS25990_IIN_OCF_OFF; >> + val = DIV_ROUND_CLOSEST(val * TPS25990_IIN_OCF_DIV, TPS25990_IIN_OCF_NUM); >> + val = clamp_val(val, 0, 0x3f); >> + break; >> + >> + case PMBUS_VIRT_SAMPLES: >> + val = clamp_val(val, 1, 1 << PK_MIN_AVG_AVG_CNT); >> + val = ilog2(val); >> + break; > > default: missing. > >> + } >> + >> + return val; >> +} >> + >> +static int tps25990_write_word(struct i2c_client *client, >> + int page, int reg, u16 value) >> +{ >> + int addr, ret; >> + >> + addr = tps25990_get_addr(reg); >> + if (addr < 0) >> + return addr; >> + >> + value = tps25990_write_adapt_value(reg, value); >> + >> + switch (reg) { >> + case PMBUS_VIRT_SAMPLES: >> + ret = pmbus_update_byte_data(client, page, addr, >> + PK_MIN_AVG_AVG_CNT, >> + FIELD_PREP(PK_MIN_AVG_AVG_CNT, value)); >> + break; >> + >> + case PMBUS_IIN_OC_FAULT_LIMIT: >> + ret = pmbus_write_byte_data(client, page, addr, >> + value); >> + break; >> + >> + default: >> + ret = pmbus_write_word_data(client, page, addr, value); >> + break; > > Same comments as for read functions. > >> + } >> + >> + return ret; >> +} >> + >> +#if IS_ENABLED(CONFIG_SENSORS_TPS25990_REGULATOR) >> +static const struct regulator_desc tps25990_reg_desc[] = { >> + PMBUS_REGULATOR_ONE("vout"), >> +}; >> +#endif >> + >> +static const struct pmbus_driver_info tps25990_base_info = { >> + .pages = 1, >> + .format[PSC_VOLTAGE_IN] = direct, >> + .m[PSC_VOLTAGE_IN] = 5251, >> + .b[PSC_VOLTAGE_IN] = 0, >> + .R[PSC_VOLTAGE_IN] = -2, >> + .format[PSC_VOLTAGE_OUT] = direct, >> + .m[PSC_VOLTAGE_OUT] = 5251, >> + .b[PSC_VOLTAGE_OUT] = 0, >> + .R[PSC_VOLTAGE_OUT] = -2, >> + .format[PSC_TEMPERATURE] = direct, >> + .m[PSC_TEMPERATURE] = 140, >> + .b[PSC_TEMPERATURE] = 32100, >> + .R[PSC_TEMPERATURE] = -2, >> + /* >> + * Current and Power measurement depends on the ohm value >> + * of Rimon. m is multiplied by 1000 below to have an integer >> + * and -3 is added to R to compensate. >> + */ >> + .format[PSC_CURRENT_IN] = direct, >> + .m[PSC_CURRENT_IN] = 9538, >> + .b[PSC_CURRENT_IN] = 0, >> + .R[PSC_CURRENT_IN] = -6, >> + .format[PSC_POWER] = direct, >> + .m[PSC_POWER] = 4901, >> + .b[PSC_POWER] = 0, >> + .R[PSC_POWER] = -7, >> + .func[0] = (PMBUS_HAVE_VIN | >> + PMBUS_HAVE_VOUT | >> + PMBUS_HAVE_VMON | >> + PMBUS_HAVE_IIN | >> + PMBUS_HAVE_PIN | >> + PMBUS_HAVE_TEMP | >> + PMBUS_HAVE_STATUS_VOUT | >> + PMBUS_HAVE_STATUS_IOUT | >> + PMBUS_HAVE_STATUS_INPUT | >> + PMBUS_HAVE_STATUS_TEMP | >> + PMBUS_HAVE_SAMPLES), >> + .read_word_data = tps25990_read_word, >> + .write_word_data = tps25990_write_word, >> + .groups = tps25990_groups, >> + >> +#if IS_ENABLED(CONFIG_SENSORS_TPS25990_REGULATOR) >> + .reg_desc = tps25990_reg_desc, >> + .num_regulators = ARRAY_SIZE(tps25990_reg_desc), >> +#endif >> +}; >> + >> +static const struct i2c_device_id tps25990_i2c_id[] = { >> + { "tps25990" }, >> + {} >> +}; >> +MODULE_DEVICE_TABLE(i2c, tps25990_i2c_id); >> + >> +static const struct of_device_id tps25990_of_match[] = { >> + { .compatible = "ti,tps25990" }, >> + {} >> +}; >> +MODULE_DEVICE_TABLE(of, tps25990_of_match); >> + >> +static int tps25990_probe(struct i2c_client *client) >> +{ >> + struct device *dev = &client->dev; >> + struct pmbus_driver_info *info; >> + u32 rimon = TPS25990_DEFAULT_RIMON; >> + int ret; >> + >> + ret = device_property_read_u32(dev, "ti,rimon-milli-ohms", &rimon); >> + if (ret == -EINVAL) { >> + dev_warn(dev, >> + "using default rimon: current and power scale possibly wrong\n"); > > This is not an appropriate warning. It is perfectly fine to load the driver > if there is no ti,rimon-milli-ohms property. I should have commented more on the default value. It is meant for the case where the device is instanciated through i2c sys 'new_device', which is meant for debugging purpose. In that particular case, it does not really matter if the current and power scale are wrong. There is no way to pass device properties when instanciating device through that interface, as far as I know. In every other cases, a correct Rimon value is expected. I could turn the above to an error. It means loading through i2c sys would not possible for this driver. Would it be better ? > >> + } else if (ret < 0) { >> + return dev_err_probe(dev, ret, "failed get rimon\n"); >> + } >> + >> + /* >> + * TPS25990 may be stacked with several TPS25895, allowing a higher >> + * current. The higher the allowed current is, the lower rimon >> + * will be. How low it can realistically get is unknown. >> + * To avoid problems with precision later on, rimon is provided in >> + * milli Ohms. This is a precaution to keep a stable ABI. >> + * At the moment, doing the calculation with rimon in milli Ohms >> + * would overflow the s32 'm' in the direct conversion. Convert it >> + * back to Ohms until greater precision is actually needed. >> + */ >> + rimon /= 1000; >> + > > Seems to me it would make more sense to limit the valid range of ti,rimon-milli-ohms > to avoid the overflow. But then I really don't understand the reasoning to provide > the property in milli-ohm, given the default value of 910 Ohm. What is a realistic > lowest value that would make sense ? The highest value I've seen, when the tps25990 is alone, is 1370 Ohms. That means a 30A overcurrent fault limit. With one TPS25895, I've seen 608 Ohms (110A limit) I have no idea what the realistic low limit is. To get to ~100 Ohms, you'd need 8 devices (not hundreds ;) ) If one gets there, it might be desirable to have 3 digits to play with, and not be limited by the unit. The DT folks really don't like when a property changes. Going with milli-Ohms is way to anticipate the problem. The other way could be to use Ohms now, and if we ever get to point where milli-Ohms precision is needed, add it then. The downside is that the driver will need to support both properties. Would you prefer this ? > But even if it is less than 1 Ohm I don't > understand why it would make sense to completely ignore it. It would not make sense to ignore it. > >> + info = devm_kmemdup(dev, &tps25990_base_info, sizeof(*info), GFP_KERNEL); >> + if (!info) >> + return -ENOMEM; >> + >> + /* Adapt the current and power scale for each instance */ >> + info->m[PSC_CURRENT_IN] *= rimon; >> + info->m[PSC_POWER] *= rimon; > > Any rimon value < 1000 mOhm will result in m values of 0. Indeed. Such Rimon value would mean an over current limit > 50kA. I admit I did really think much about such value. The idea was more keep some precision if we get somewhere near a 100 Ohms. > >> + >> + ret = pmbus_do_probe(client, info); >> + if (ret < 0) >> + return ret; >> + >> + return tps25990_init_debugfs(client); > > debugfs initialization is not expected to fail. > >> +} >> + >> +static struct i2c_driver tps25990_driver = { >> + .driver = { >> + .name = "tps25990", >> + .of_match_table = tps25990_of_match, >> + }, >> + .probe = tps25990_probe, >> + .id_table = tps25990_i2c_id, >> +}; >> +module_i2c_driver(tps25990_driver); >> + >> +MODULE_AUTHOR("Jerome Brunet <jbrunet@baylibre.com>"); >> +MODULE_DESCRIPTION("PMBUS driver for TPS25990 eFuse"); >> +MODULE_LICENSE("GPL"); >> +MODULE_IMPORT_NS(PMBUS); >>
On 9/9/24 23:43, Jerome Brunet wrote: > On Mon 09 Sep 2024 at 11:16, Guenter Roeck <linux@roeck-us.net> wrote: > >> On 9/9/24 08:39, Jerome Brunet wrote: >>> Add support for POWER_GOOD_ON and POWER_GOOD_OFF standard PMBus commands. >>> For PMBus devices that offer a POWER_GOOD signal, these commands are used >>> for setting the output voltage at which a power good signal should be >>> asserted and negated. >>> Power Good signals are device and manufacturer specific. Many factors >>> other >>> than output voltage may be used to determine whether or not the POWER_GOOD >>> signal is to be asserted. PMBus device users are instructed to consult the >>> device manufacturer’s product literature for the specifics of the device >>> they are using. >>> Note that depending on the choice of the device manufacturer that a >>> device >>> may drive a POWER_GOOD signal high or low to indicate that the signal is >>> asserted. >>> Signed-off-by: Jerome Brunet <jbrunet@baylibre.com> >>> --- >>> drivers/hwmon/pmbus/pmbus.h | 3 +++ >>> drivers/hwmon/pmbus/pmbus_core.c | 6 ++++++ >>> 2 files changed, 9 insertions(+) >>> diff --git a/drivers/hwmon/pmbus/pmbus.h b/drivers/hwmon/pmbus/pmbus.h >>> index 5d5dc774187b..e322d2dd9fb7 100644 >>> --- a/drivers/hwmon/pmbus/pmbus.h >>> +++ b/drivers/hwmon/pmbus/pmbus.h >>> @@ -78,6 +78,9 @@ enum pmbus_regs { >>> PMBUS_IIN_OC_FAULT_LIMIT = 0x5B, >>> PMBUS_IIN_OC_WARN_LIMIT = 0x5D, >>> + PMBUS_POWER_GOOD_ON = 0x5E, >>> + PMBUS_POWER_GOOD_OFF = 0x5F, >>> + >>> PMBUS_POUT_OP_FAULT_LIMIT = 0x68, >>> PMBUS_POUT_OP_WARN_LIMIT = 0x6A, >>> PMBUS_PIN_OP_WARN_LIMIT = 0x6B, >>> diff --git a/drivers/hwmon/pmbus/pmbus_core.c b/drivers/hwmon/pmbus/pmbus_core.c >>> index 0ea6fe7eb17c..94ddf0166770 100644 >>> --- a/drivers/hwmon/pmbus/pmbus_core.c >>> +++ b/drivers/hwmon/pmbus/pmbus_core.c >>> @@ -1768,6 +1768,12 @@ static const struct pmbus_limit_attr vout_limit_attrs[] = { >>> .attr = "crit", >>> .alarm = "crit_alarm", >>> .sbit = PB_VOLTAGE_OV_FAULT, >>> + }, { >>> + .reg = PMBUS_POWER_GOOD_ON, >>> + .attr = "good_on", >>> + }, { >>> + .reg = PMBUS_POWER_GOOD_OFF, >>> + .attr = "good_off", >>> }, { >>> .reg = PMBUS_VIRT_READ_VOUT_AVG, >>> .update = true, >>> >> >> Those attributes are not hardware monitoring attributes and therefore not >> acceptable. In general I am not sure if they should be configurable in the >> first place, but definitely not from the hardware monitoring subsystem. >> Maybe the regulator subsystem callbacks set_over_voltage_protection and >> set_under_voltage_protection would be appropriate (with severity >> REGULATOR_SEVERITY_PROT), but that should be discussed with regulator >> subsystem maintainers. > > According to PMBUS spec, there is no protection associated with that > command. It just tells when the output voltage is considered good, when > it is not. What it does after that really depends the device, it may > drive a pin for example (or an LED indicator in my case). > It is much more likely that it connects to the reset signal on the board, or it enables/disables power to parts of the board. > It is very similar to 'crit' or other limits in that sense, > I think. I don't really get why such property is not OK in hwmon then > and why it should not be configurable, if the other limits are ? > Its use is for hardware control, not monitoring, even if it may be connected to a status LED. MAX15301, for example, groups the command under "Voltage Sequencing Commands". On top of that, the voltages are value/hysteresis values. The "off" voltage is lower than the "on" voltage. TPS25990 doesn't even support the command according to its datasheet, so I am at loss about your use case in the context of this patch series (the PGOOD pin on this chip signals to the downstream load that it is ok to draw power). Guenter
On Tue 10 Sep 2024 at 07:37, Guenter Roeck <linux@roeck-us.net> wrote: > On 9/9/24 23:43, Jerome Brunet wrote: >> On Mon 09 Sep 2024 at 11:16, Guenter Roeck <linux@roeck-us.net> wrote: >> >>> On 9/9/24 08:39, Jerome Brunet wrote: >>>> Add support for POWER_GOOD_ON and POWER_GOOD_OFF standard PMBus commands. >>>> For PMBus devices that offer a POWER_GOOD signal, these commands are used >>>> for setting the output voltage at which a power good signal should be >>>> asserted and negated. >>>> Power Good signals are device and manufacturer specific. Many factors >>>> other >>>> than output voltage may be used to determine whether or not the POWER_GOOD >>>> signal is to be asserted. PMBus device users are instructed to consult the >>>> device manufacturer’s product literature for the specifics of the device >>>> they are using. >>>> Note that depending on the choice of the device manufacturer that a >>>> device >>>> may drive a POWER_GOOD signal high or low to indicate that the signal is >>>> asserted. >>>> Signed-off-by: Jerome Brunet <jbrunet@baylibre.com> >>>> --- >>>> drivers/hwmon/pmbus/pmbus.h | 3 +++ >>>> drivers/hwmon/pmbus/pmbus_core.c | 6 ++++++ >>>> 2 files changed, 9 insertions(+) >>>> diff --git a/drivers/hwmon/pmbus/pmbus.h b/drivers/hwmon/pmbus/pmbus.h >>>> index 5d5dc774187b..e322d2dd9fb7 100644 >>>> --- a/drivers/hwmon/pmbus/pmbus.h >>>> +++ b/drivers/hwmon/pmbus/pmbus.h >>>> @@ -78,6 +78,9 @@ enum pmbus_regs { >>>> PMBUS_IIN_OC_FAULT_LIMIT = 0x5B, >>>> PMBUS_IIN_OC_WARN_LIMIT = 0x5D, >>>> + PMBUS_POWER_GOOD_ON = 0x5E, >>>> + PMBUS_POWER_GOOD_OFF = 0x5F, >>>> + >>>> PMBUS_POUT_OP_FAULT_LIMIT = 0x68, >>>> PMBUS_POUT_OP_WARN_LIMIT = 0x6A, >>>> PMBUS_PIN_OP_WARN_LIMIT = 0x6B, >>>> diff --git a/drivers/hwmon/pmbus/pmbus_core.c b/drivers/hwmon/pmbus/pmbus_core.c >>>> index 0ea6fe7eb17c..94ddf0166770 100644 >>>> --- a/drivers/hwmon/pmbus/pmbus_core.c >>>> +++ b/drivers/hwmon/pmbus/pmbus_core.c >>>> @@ -1768,6 +1768,12 @@ static const struct pmbus_limit_attr vout_limit_attrs[] = { >>>> .attr = "crit", >>>> .alarm = "crit_alarm", >>>> .sbit = PB_VOLTAGE_OV_FAULT, >>>> + }, { >>>> + .reg = PMBUS_POWER_GOOD_ON, >>>> + .attr = "good_on", >>>> + }, { >>>> + .reg = PMBUS_POWER_GOOD_OFF, >>>> + .attr = "good_off", >>>> }, { >>>> .reg = PMBUS_VIRT_READ_VOUT_AVG, >>>> .update = true, >>>> >>> >>> Those attributes are not hardware monitoring attributes and therefore not >>> acceptable. In general I am not sure if they should be configurable in the >>> first place, but definitely not from the hardware monitoring subsystem. >>> Maybe the regulator subsystem callbacks set_over_voltage_protection and >>> set_under_voltage_protection would be appropriate (with severity >>> REGULATOR_SEVERITY_PROT), but that should be discussed with regulator >>> subsystem maintainers. >> According to PMBUS spec, there is no protection associated with that >> command. It just tells when the output voltage is considered good, when >> it is not. What it does after that really depends the device, it may >> drive a pin for example (or an LED indicator in my case). >> > > It is much more likely that it connects to the reset signal on the board, > or it enables/disables power to parts of the board. That's not what PMBus spec says about it: """ 15.32. POWER_GOOD Signal Limits For PMBus devices that offer a POWER_GOOD signal, these commands are used for setting the output voltage at which a power good signal should be asserted and negated. Power Good signals will be device and manufacturer specific. Many factors other than output voltage may be used to determine whether or not the POWER_GOOD signal is to be asserted. PMBus device users are instructed to consult the device manufacturer’s product literature for the specifics of the device they are using. """ It's only supposed to have an effect on the power_good signal, not the reset. I guess someone could wire that signal to a reset. Same could be done with the alert or the fault one, I suppose > >> It is very similar to 'crit' or other limits in that sense, >> I think. I don't really get why such property is not OK in hwmon then >> and why it should not be configurable, if the other limits are ? >> > > Its use is for hardware control, not monitoring, even if it may be connected > to a status LED. MAX15301, for example, groups the command under "Voltage > Sequencing Commands". > > On top of that, the voltages are value/hysteresis values. The "off" voltage > is lower than the "on" voltage. > > TPS25990 doesn't even support the command according to its datasheet, so I am > at loss about your use case in the context of this patch series (the PGOOD pin > on this chip signals to the downstream load that it is ok to draw > power). It does support GOOD_OFF, althought TI renamed the register to VOUT_PGTH (Section 8.3.14.7.1.52, p87): """ VOUT_PGTH is a standard PMBus® command for setting or reading an 8-bit output voltage threshold at which Power Good (PGOOD) is be de-asserted. """ Same as the PMBus spec. Changing the value through this command does affect the signal as intented. How the signal is depends on the implementation. It just drives an LED on the EVM. Anyway, I don't want to hold things on this. I'll drop it from the next version. > > Guenter
On 9/10/24 08:00, Jerome Brunet wrote: > On Tue 10 Sep 2024 at 07:37, Guenter Roeck <linux@roeck-us.net> wrote: > >> On 9/9/24 23:43, Jerome Brunet wrote: >>> On Mon 09 Sep 2024 at 11:16, Guenter Roeck <linux@roeck-us.net> wrote: >>> >>>> On 9/9/24 08:39, Jerome Brunet wrote: >>>>> Add support for POWER_GOOD_ON and POWER_GOOD_OFF standard PMBus commands. >>>>> For PMBus devices that offer a POWER_GOOD signal, these commands are used >>>>> for setting the output voltage at which a power good signal should be >>>>> asserted and negated. >>>>> Power Good signals are device and manufacturer specific. Many factors >>>>> other >>>>> than output voltage may be used to determine whether or not the POWER_GOOD >>>>> signal is to be asserted. PMBus device users are instructed to consult the >>>>> device manufacturer’s product literature for the specifics of the device >>>>> they are using. >>>>> Note that depending on the choice of the device manufacturer that a >>>>> device >>>>> may drive a POWER_GOOD signal high or low to indicate that the signal is >>>>> asserted. >>>>> Signed-off-by: Jerome Brunet <jbrunet@baylibre.com> >>>>> --- >>>>> drivers/hwmon/pmbus/pmbus.h | 3 +++ >>>>> drivers/hwmon/pmbus/pmbus_core.c | 6 ++++++ >>>>> 2 files changed, 9 insertions(+) >>>>> diff --git a/drivers/hwmon/pmbus/pmbus.h b/drivers/hwmon/pmbus/pmbus.h >>>>> index 5d5dc774187b..e322d2dd9fb7 100644 >>>>> --- a/drivers/hwmon/pmbus/pmbus.h >>>>> +++ b/drivers/hwmon/pmbus/pmbus.h >>>>> @@ -78,6 +78,9 @@ enum pmbus_regs { >>>>> PMBUS_IIN_OC_FAULT_LIMIT = 0x5B, >>>>> PMBUS_IIN_OC_WARN_LIMIT = 0x5D, >>>>> + PMBUS_POWER_GOOD_ON = 0x5E, >>>>> + PMBUS_POWER_GOOD_OFF = 0x5F, >>>>> + >>>>> PMBUS_POUT_OP_FAULT_LIMIT = 0x68, >>>>> PMBUS_POUT_OP_WARN_LIMIT = 0x6A, >>>>> PMBUS_PIN_OP_WARN_LIMIT = 0x6B, >>>>> diff --git a/drivers/hwmon/pmbus/pmbus_core.c b/drivers/hwmon/pmbus/pmbus_core.c >>>>> index 0ea6fe7eb17c..94ddf0166770 100644 >>>>> --- a/drivers/hwmon/pmbus/pmbus_core.c >>>>> +++ b/drivers/hwmon/pmbus/pmbus_core.c >>>>> @@ -1768,6 +1768,12 @@ static const struct pmbus_limit_attr vout_limit_attrs[] = { >>>>> .attr = "crit", >>>>> .alarm = "crit_alarm", >>>>> .sbit = PB_VOLTAGE_OV_FAULT, >>>>> + }, { >>>>> + .reg = PMBUS_POWER_GOOD_ON, >>>>> + .attr = "good_on", >>>>> + }, { >>>>> + .reg = PMBUS_POWER_GOOD_OFF, >>>>> + .attr = "good_off", >>>>> }, { >>>>> .reg = PMBUS_VIRT_READ_VOUT_AVG, >>>>> .update = true, >>>>> >>>> >>>> Those attributes are not hardware monitoring attributes and therefore not >>>> acceptable. In general I am not sure if they should be configurable in the >>>> first place, but definitely not from the hardware monitoring subsystem. >>>> Maybe the regulator subsystem callbacks set_over_voltage_protection and >>>> set_under_voltage_protection would be appropriate (with severity >>>> REGULATOR_SEVERITY_PROT), but that should be discussed with regulator >>>> subsystem maintainers. >>> According to PMBUS spec, there is no protection associated with that >>> command. It just tells when the output voltage is considered good, when >>> it is not. What it does after that really depends the device, it may >>> drive a pin for example (or an LED indicator in my case). >>> >> >> It is much more likely that it connects to the reset signal on the board, >> or it enables/disables power to parts of the board. > > That's not what PMBus spec says about it: > > """ > 15.32. POWER_GOOD Signal Limits > For PMBus devices that offer a POWER_GOOD signal, these commands are used for > setting the output voltage at which a power good signal should be asserted and negated. > Power Good signals will be device and manufacturer specific. Many factors other than > output voltage may be used to determine whether or not the POWER_GOOD signal is to > be asserted. PMBus device users are instructed to consult the device manufacturer’s > product literature for the specifics of the device they are using. > """ > > It's only supposed to have an effect on the power_good signal, not the > reset. I guess someone could wire that signal to a reset. Same could be > done with the alert or the fault one, I suppose > It doesn't say anything about the _use_ of that signal. The PMBus specification says "Power Good signals will be device and manufacturer specific", and that is exactly what it is. TPS25990 specifically states that the signal indicates that it is ok for downstream chips to draw power, which is a very typical use. The ability to connect it it to an LED does not reflect its core use. >> >>> It is very similar to 'crit' or other limits in that sense, >>> I think. I don't really get why such property is not OK in hwmon then >>> and why it should not be configurable, if the other limits are ? >>> >> >> Its use is for hardware control, not monitoring, even if it may be connected >> to a status LED. MAX15301, for example, groups the command under "Voltage >> Sequencing Commands". >> >> On top of that, the voltages are value/hysteresis values. The "off" voltage >> is lower than the "on" voltage. >> >> TPS25990 doesn't even support the command according to its datasheet, so I am >> at loss about your use case in the context of this patch series (the PGOOD pin >> on this chip signals to the downstream load that it is ok to draw >> power). > > It does support GOOD_OFF, althought TI renamed the register to > VOUT_PGTH (Section 8.3.14.7.1.52, p87): > > """ > VOUT_PGTH is a standard PMBus® command for setting or reading an 8-bit > output voltage threshold at which Power Good (PGOOD) is be de-asserted. > """ > Ah yes, typical for PMBus chips :-(. Why use standard register/command names if one can rename them. It actually also states that pgood is asserted first when the voltage reaches VOUT_PGTH + 250mV, so even with this chip it is really a hysteresis. > Same as the PMBus spec. Changing the value through this command does > affect the signal as intented. How the signal is depends on the > implementation. It just drives an LED on the EVM. > Yes, but that doesn't make it a hardware _monitoring_ attribute. Guenter
On Tue, Sep 10, 2024 at 11:07:57AM +0200, Jerome Brunet wrote: > On Mon 09 Sep 2024 at 15:52, Guenter Roeck <linux@roeck-us.net> wrote: > > [...] > Unrelated to the other comments: Documentation/hwmon/tps25990.rst | 141 ++++++++++++ Needs to be added to Documentation/hwmon/index.rst. +config SENSORS_TPS25990_REGULATOR + bool "Regulator support for TPS25990 and compatibles" + depends on SENSORS_TPS25990 && REGULATOR + default SENSORS_TPS2599 ^^^^^^^^^^^^^^^ TPS2599 ??? > >> + > >> +#define TPS25990_DEFAULT_RIMON 910000 Where does the default come from anyway ? I don't immediately see the number in the datasheet. > >> +static int tps25990_write_protect_get(void *data, u64 *val) > >> +{ > >> + struct i2c_client *client = data; > >> + > >> + return tps25990_mfr_write_protect_active(client); > >> +} > >> + > >> +static int tps25990_write_protect_set(void *data, u64 val) > >> +{ > >> + struct i2c_client *client = data; > >> + > >> + if (val > 1) > >> + return -EINVAL; > >> + > >> + return tps25990_mfr_write_protect(client, val); > >> +} > >> + > >> +DEFINE_DEBUGFS_ATTRIBUTE(tps25990_write_protect_fops, > >> + tps25990_write_protect_get, > >> + tps25990_write_protect_set, > >> + "%llu\n"); > >> + > >> +static int tps25990_init_debugfs(struct i2c_client *client) > >> +{ > >> + struct dentry *dir; > >> + > >> + dir = pmbus_get_debugfs_dir(client); > >> + if (!dir) > >> + return -ENOENT; > >> + > >> + debugfs_create_file("write_protect", 0644, dir, > >> + client, &tps25990_write_protect_fops); > >> + > >> + return 0; > >> +} > >> + > >> +#else > >> +static inline int tps25990_init_debugfs(struct i2c_client *client) > >> +{ > >> + return 0; > >> +} > >> +#endif > >> + > > > > In general it is extremely undesirable to overwrite write protection. > > Many chips support such attributes. If write protection is enabled, > > it means that the board vendor does not want to have them changed. > > According to documentation, it protects against "unintented" writes, > not 'wrong' or 'malicious'. If one goes in debugfs and write just '0' to > a file, there is an intent at least. > > > Granted, that can be overwritten with direct i2c commands, but that > > is what it should be. Anyone who really wants to disable write protection > > should have to dig deeper than just writing into a debugfs or sysfs attribute. > > Otherwise the protection becomes worthless. > > If this is, for example, needed > > for production to write initial settings, the production scripts should > > disable (or enable) write protection by writing directly into command > > registers. > > As I wrote in the cover letter, the write protection is always active on > chip startup and it locks down almost everything, including things you may > need to write past production, in the field. The history reset below is > an example of such thing. > > To 'safely' remove the protection by writing i2c commands from > userspace: > * the device will need be unbinded first, > * call i2cset > * bind the device again > > That seems really cumbersome to do something like an history > reset. Is this what you are suggesting ? > > bind/unbind could be skipped by forcing i2cset but that would add danger > where we certainly don't want it. > Not sure I understand the "danger" part. Either case, the problem is deeper. The driver enables regulator support, which includes enabling and disabling the output voltage. But that doesn't work unles write protect is disabled. debugfs doesn't help there; that is way too late. > > > >> +/* > >> + * TPS25990 has history reset based on MIN/AVG/PEAK instead of per sensor type > >> + * Emulate the behaviour a pmbus limit_attr would have for consistency > >> + * - Read: Do nothing and emit 0 > >> + * - Write: Check the input is a number and reset > >> + */ > >> +static ssize_t tps25990_history_reset_show(struct device *dev, > >> + struct device_attribute *devattr, > >> + char *buf) > >> +{ > >> + return sysfs_emit(buf, "0\n"); > >> +} > >> + > >> +static ssize_t tps25990_history_reset_store(struct device *dev, > >> + struct device_attribute *devattr, > >> + const char *buf, size_t count) > >> +{ > >> + struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr); > >> + struct i2c_client *client = to_i2c_client(dev->parent); > >> + s64 val; > >> + int ret; > >> + > >> + if (kstrtos64(buf, 10, &val) < 0) > >> + return -EINVAL; > >> + > >> + ret = pmbus_update_byte_data(client, 0, TPS25990_PK_MIN_AVG, > >> + BIT(attr->index), BIT(attr->index)); > >> + if (ret < 0) > >> + return ret; > >> + > >> + return count; > >> +} > >> + > >> +static SENSOR_DEVICE_ATTR_RW(highest_history_reset, tps25990_history_reset, 7); > >> +static SENSOR_DEVICE_ATTR_RW(average_history_reset, tps25990_history_reset, 6); > >> +static SENSOR_DEVICE_ATTR_RW(lowest_history_reset, tps25990_history_reset, 5); > > > > That is not a unique problem, and not a reason to introduce non-standard attributes. > > Just attach the attribute to the first channel and document that it resets all > > channels. > > Not sure I got this right so I'll rephrase. I should: > * Pick a channel, say vin > * Map the virtual reset register to hit the 3 resets above > * Put in the documentation that it resets the other channels as well > * Not allow independent resets of min/max/avg, just all 3 together ? > Correct. It is amazing what hardware designers come up with (here: resetting history based on min/max/average instead of the sensor type is novel), but I really don't want to introduce new attributes to accommodate each variant. I'd be open to introducing a global PMBUS_VIRT_RESET_HISTORY virtual register and reset_history attribute if you want to go there, but that would have to be in the PMBus core. > > > >> + > >> +static struct attribute *tps25990_attrs[] = { > >> + &sensor_dev_attr_highest_history_reset.dev_attr.attr, > >> + &sensor_dev_attr_average_history_reset.dev_attr.attr, > >> + &sensor_dev_attr_lowest_history_reset.dev_attr.attr, > >> + NULL, > >> +}; > >> + > >> +ATTRIBUTE_GROUPS(tps25990); > >> + > >> +static int tps25990_get_addr(int reg) > >> +{ > >> + switch (reg) { > >> + case PMBUS_SMBALERT_MASK: > >> + /* > >> + * Note: PMBUS_SMBALERT_MASK is not implemented on this chip > >> + * Writing to this address raises CML errors. > >> + * Instead it provides ALERT_MASK which allows to set the mask > >> + * for each of the status registers, but not the specific bits > >> + * in them. > >> + * The default setup assert SMBA# if any bit is set in any of the > >> + * status registers the chip has. This is as close as we can get > >> + * to what pmbus_irq_setup() would set, sooo ... do nothing. > >> + */ > >> + return -ENXIO; > > > > Many chips have that problem. The core code ignores errors, and attempts to write > > the command are limited to initialization. This is not a reason to overwrite > > the command like this. If this does cause a real a problem wit hthe chip (other > > than setting CML errors, which many chips not supporting the command do), > > we should define a flag in include/linux/pmbus.h and explain its need. > > CML is error is the problem. Following pmbus_irq_setup() there is an > uncleared fault because there is no register check on PMBUS_SMBALERT_MASK. > > When pmbus_core then gets here: > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/hwmon/pmbus/pmbus_core.c?h=v6.11-rc7#n3386 > > pmbus_check_block_register() fails because of the uncleared fault and > the 'mfr_id' property is silently not registered, eventhough the > register is supported by the chip. This is how I noticed the problem. > > So, should I add flag in include/linux/pmbus.h to skip > PMBUS_SMBALERT_MASK setup ? > > Another possibility is to call register_check() > on it before using PMBUS_SMBALERT_MASK in pmbus_core. > The problem, as you point out, is in pmbus_irq_setup(). Since the function explicitly ignores errors from accessing PMBUS_SMBALERT_MASK, it should either clear faults after it is done. I don't think we can rely on register_check() because the register might exist but be read-only. > > > >> + case PMBUS_IIN_OC_FAULT_LIMIT: > >> + /* > >> + * VIREF directly sets the over-current limit at which the eFuse > >> + * will turn the FET off and trigger a fault. Expose it through > >> + * this generic property instead of a manufacturer specific one. > >> + */ > >> + return TPS25990_VIREF; > > > > I don't see the value in this override. See below. > > > >> + case PMBUS_VIRT_READ_VIN_MAX: > >> + return TPS25990_READ_VIN_PEAK; > >> + case PMBUS_VIRT_READ_VIN_MIN: > >> + return TPS25990_READ_VIN_MIN; > >> + case PMBUS_VIRT_READ_VIN_AVG: > >> + return TPS25990_READ_VIN_AVG; > >> + case PMBUS_VIRT_READ_VOUT_MIN: > >> + return TPS25990_READ_VOUT_MIN; > >> + case PMBUS_VIRT_READ_VOUT_AVG: > >> + return TPS25990_READ_VOUT_AVG; > >> + case PMBUS_VIRT_READ_IIN_AVG: > >> + return TPS25990_READ_IIN_AVG; > >> + case PMBUS_VIRT_READ_IIN_MAX: > >> + return TPS25990_READ_IIN_PEAK; > >> + case PMBUS_VIRT_READ_TEMP_AVG: > >> + return TPS25990_READ_TEMP_AVG; > >> + case PMBUS_VIRT_READ_TEMP_MAX: > >> + return TPS25990_READ_TEMP_PEAK; > >> + case PMBUS_VIRT_READ_PIN_AVG: > >> + return TPS25990_READ_PIN_AVG; > >> + case PMBUS_VIRT_READ_PIN_MAX: > >> + return TPS25990_READ_PIN_PEAK; > >> + case PMBUS_VIRT_READ_VMON: > >> + return TPS25990_READ_VAUX; > >> + case PMBUS_VIRT_SAMPLES: > >> + return TPS25990_PK_MIN_AVG; > > > > default: missing. > > > >> + } > >> + > >> + /* Let the register check do its job */ > >> + if (reg < PMBUS_VIRT_BASE) > >> + return reg; > >> + > >> + return -ENXIO; > > > > This needs to be more specific: The code should only return -ENXIO > > if auto-detection can not handle the case. "Return -ENXIO for all > > unsupported virtual registers" is unexpected. That situation should > > be handled by the PMBus core. > > > >> +} > >> + > >> +/* > >> + * Some registers use a different scale than the one registered with > >> + * pmbus_driver_info. An extra conversion step is necessary to adapt > >> + * the register value to the conversion on the sensor type > >> + */ > >> +static int tps25990_read_adapt_value(int reg, int val) > >> +{ > >> + switch (reg) { > >> + case PMBUS_VIN_UV_WARN_LIMIT: > >> + case PMBUS_VIN_UV_FAULT_LIMIT: > >> + case PMBUS_VIN_OV_WARN_LIMIT: > >> + case PMBUS_VOUT_UV_WARN_LIMIT: > >> + case PMBUS_IIN_OC_WARN_LIMIT: > >> + case PMBUS_OT_WARN_LIMIT: > >> + case PMBUS_OT_FAULT_LIMIT: > >> + case PMBUS_PIN_OP_WARN_LIMIT: > >> + case PMBUS_POWER_GOOD_OFF: > >> + /* > >> + * These registers provide an 8 bits value instead of a > >> + * 10bits one. Just shifting twice the register value is > >> + * enough to make the sensor type conversion work, even > >> + * if the datasheet provides different m, b and R for > >> + * those. > >> + */ > >> + val <<= TPS25990_8B_SHIFT; > >> + break; > >> + > >> + case PMBUS_VIN_OV_FAULT_LIMIT: > >> + val = DIV_ROUND_CLOSEST(val * TPS25990_VIN_OVF_NUM, TPS25990_VIN_OVF_DIV); > >> + val += TPS25990_VIN_OVF_OFF; > >> + break; > >> + > >> + case PMBUS_IIN_OC_FAULT_LIMIT: > >> + val = DIV_ROUND_CLOSEST(val * TPS25990_IIN_OCF_NUM, TPS25990_IIN_OCF_DIV); > >> + val += TPS25990_IIN_OCF_OFF; > >> + break; > >> + > >> + case PMBUS_VIRT_SAMPLES: > >> + val = 1 << val; > >> + break; > > > > default case missing. > > > >> + } > >> + > >> + return val; > >> +} > >> + > >> +static int tps25990_read_word(struct i2c_client *client, > >> + int page, int phase, int reg) > >> +{ > >> + int ret, addr; > >> + > >> + addr = tps25990_get_addr(reg); > >> + if (addr < 0) > >> + return addr; > >> + > >> + switch (reg) { > >> + case PMBUS_VIRT_SAMPLES: > >> + ret = pmbus_read_byte_data(client, page, addr); > > > > Mapping the register name in tps25990_get_addr() is unnecessary > > and misleading. It is well known that TPS25990_PK_MIN_AVG is to be > > used. Do it here. > > My intent was to do the mapping in one place instead of repeating for > both read and write, not be misleading. I'll change it. > Mapping is only needed in the default: case below, and should be limited to that. > > > >> + ret = FIELD_GET(PK_MIN_AVG_AVG_CNT, ret); > >> + break; > >> + > >> + case PMBUS_IIN_OC_FAULT_LIMIT: > >> + ret = pmbus_read_byte_data(client, page, addr); > >> + break; > >> + > > Same here. > > > >> + default: > >> + ret = pmbus_read_word_data(client, page, -1, addr); > > > > This is unexpected for registers not handled locally. Expectation is > > that -ENODATA is returned for those, to be handled in the core. > > Got it. Thanks. > > > > >> + break; > >> + } > >> + > >> + if (ret >= 0) > >> + ret = tps25990_read_adapt_value(reg, ret); > >> + > >> + return ret; > >> +} > >> + > >> +static int tps25990_write_adapt_value(int reg, int val) > >> +{ > >> + switch (reg) { > >> + case PMBUS_VIN_UV_WARN_LIMIT: > >> + case PMBUS_VIN_UV_FAULT_LIMIT: > >> + case PMBUS_VIN_OV_WARN_LIMIT: > >> + case PMBUS_VOUT_UV_WARN_LIMIT: > >> + case PMBUS_IIN_OC_WARN_LIMIT: > >> + case PMBUS_OT_WARN_LIMIT: > >> + case PMBUS_OT_FAULT_LIMIT: > >> + case PMBUS_PIN_OP_WARN_LIMIT: > >> + case PMBUS_POWER_GOOD_OFF: > >> + val >>= TPS25990_8B_SHIFT; > >> + val = clamp(val, 0, 0xff); > > > > Why clamp() here but clamp_val() elsewhere ? > > > >> + break; > >> + > >> + case PMBUS_VIN_OV_FAULT_LIMIT: > >> + val -= TPS25990_VIN_OVF_OFF; > >> + val = DIV_ROUND_CLOSEST(val * TPS25990_VIN_OVF_DIV, TPS25990_VIN_OVF_NUM); > >> + val = clamp_val(val, 0, 0xf); > >> + break; > >> + > >> + case PMBUS_IIN_OC_FAULT_LIMIT: > >> + val -= TPS25990_IIN_OCF_OFF; > >> + val = DIV_ROUND_CLOSEST(val * TPS25990_IIN_OCF_DIV, TPS25990_IIN_OCF_NUM); > >> + val = clamp_val(val, 0, 0x3f); > >> + break; > >> + > >> + case PMBUS_VIRT_SAMPLES: > >> + val = clamp_val(val, 1, 1 << PK_MIN_AVG_AVG_CNT); > >> + val = ilog2(val); > >> + break; > > > > default: missing. > > > >> + } > >> + > >> + return val; > >> +} > >> + > >> +static int tps25990_write_word(struct i2c_client *client, > >> + int page, int reg, u16 value) > >> +{ > >> + int addr, ret; > >> + > >> + addr = tps25990_get_addr(reg); > >> + if (addr < 0) > >> + return addr; > >> + > >> + value = tps25990_write_adapt_value(reg, value); > >> + > >> + switch (reg) { > >> + case PMBUS_VIRT_SAMPLES: > >> + ret = pmbus_update_byte_data(client, page, addr, > >> + PK_MIN_AVG_AVG_CNT, > >> + FIELD_PREP(PK_MIN_AVG_AVG_CNT, value)); > >> + break; > >> + > >> + case PMBUS_IIN_OC_FAULT_LIMIT: > >> + ret = pmbus_write_byte_data(client, page, addr, > >> + value); > >> + break; > >> + > >> + default: > >> + ret = pmbus_write_word_data(client, page, addr, value); > >> + break; > > > > Same comments as for read functions. > > > >> + } > >> + > >> + return ret; > >> +} > >> + > >> +#if IS_ENABLED(CONFIG_SENSORS_TPS25990_REGULATOR) > >> +static const struct regulator_desc tps25990_reg_desc[] = { > >> + PMBUS_REGULATOR_ONE("vout"), > >> +}; > >> +#endif > >> + > >> +static const struct pmbus_driver_info tps25990_base_info = { > >> + .pages = 1, > >> + .format[PSC_VOLTAGE_IN] = direct, > >> + .m[PSC_VOLTAGE_IN] = 5251, > >> + .b[PSC_VOLTAGE_IN] = 0, > >> + .R[PSC_VOLTAGE_IN] = -2, > >> + .format[PSC_VOLTAGE_OUT] = direct, > >> + .m[PSC_VOLTAGE_OUT] = 5251, > >> + .b[PSC_VOLTAGE_OUT] = 0, > >> + .R[PSC_VOLTAGE_OUT] = -2, > >> + .format[PSC_TEMPERATURE] = direct, > >> + .m[PSC_TEMPERATURE] = 140, > >> + .b[PSC_TEMPERATURE] = 32100, > >> + .R[PSC_TEMPERATURE] = -2, > >> + /* > >> + * Current and Power measurement depends on the ohm value > >> + * of Rimon. m is multiplied by 1000 below to have an integer > >> + * and -3 is added to R to compensate. > >> + */ > >> + .format[PSC_CURRENT_IN] = direct, > >> + .m[PSC_CURRENT_IN] = 9538, > >> + .b[PSC_CURRENT_IN] = 0, > >> + .R[PSC_CURRENT_IN] = -6, > >> + .format[PSC_POWER] = direct, > >> + .m[PSC_POWER] = 4901, > >> + .b[PSC_POWER] = 0, > >> + .R[PSC_POWER] = -7, > >> + .func[0] = (PMBUS_HAVE_VIN | > >> + PMBUS_HAVE_VOUT | > >> + PMBUS_HAVE_VMON | > >> + PMBUS_HAVE_IIN | > >> + PMBUS_HAVE_PIN | > >> + PMBUS_HAVE_TEMP | > >> + PMBUS_HAVE_STATUS_VOUT | > >> + PMBUS_HAVE_STATUS_IOUT | > >> + PMBUS_HAVE_STATUS_INPUT | > >> + PMBUS_HAVE_STATUS_TEMP | > >> + PMBUS_HAVE_SAMPLES), > >> + .read_word_data = tps25990_read_word, > >> + .write_word_data = tps25990_write_word, > >> + .groups = tps25990_groups, > >> + > >> +#if IS_ENABLED(CONFIG_SENSORS_TPS25990_REGULATOR) > >> + .reg_desc = tps25990_reg_desc, > >> + .num_regulators = ARRAY_SIZE(tps25990_reg_desc), > >> +#endif > >> +}; > >> + > >> +static const struct i2c_device_id tps25990_i2c_id[] = { > >> + { "tps25990" }, > >> + {} > >> +}; > >> +MODULE_DEVICE_TABLE(i2c, tps25990_i2c_id); > >> + > >> +static const struct of_device_id tps25990_of_match[] = { > >> + { .compatible = "ti,tps25990" }, > >> + {} > >> +}; > >> +MODULE_DEVICE_TABLE(of, tps25990_of_match); > >> + > >> +static int tps25990_probe(struct i2c_client *client) > >> +{ > >> + struct device *dev = &client->dev; > >> + struct pmbus_driver_info *info; > >> + u32 rimon = TPS25990_DEFAULT_RIMON; > >> + int ret; > >> + > >> + ret = device_property_read_u32(dev, "ti,rimon-milli-ohms", &rimon); > >> + if (ret == -EINVAL) { > >> + dev_warn(dev, > >> + "using default rimon: current and power scale possibly wrong\n"); > > > > This is not an appropriate warning. It is perfectly fine to load the driver > > if there is no ti,rimon-milli-ohms property. > > I should have commented more on the default value. It is meant for the > case where the device is instanciated through i2c sys 'new_device', > which is meant for debugging purpose. In that particular case, it does > not really matter if the current and power scale are wrong. > > There is no way to pass device properties when instanciating device > through that interface, as far as I know. > > In every other cases, a correct Rimon value is expected. > I could turn the above to an error. It means loading through i2c sys > would not possible for this driver. > > Would it be better ? > We use default values for pretty much all drivers, so I don't see why this one should be different. The driver should still be usable on a system without devicetree support. There is a reason for the sensors configuration file. > > > >> + } else if (ret < 0) { > >> + return dev_err_probe(dev, ret, "failed get rimon\n"); > >> + } > >> + > >> + /* > >> + * TPS25990 may be stacked with several TPS25895, allowing a higher > >> + * current. The higher the allowed current is, the lower rimon > >> + * will be. How low it can realistically get is unknown. > >> + * To avoid problems with precision later on, rimon is provided in > >> + * milli Ohms. This is a precaution to keep a stable ABI. > >> + * At the moment, doing the calculation with rimon in milli Ohms > >> + * would overflow the s32 'm' in the direct conversion. Convert it > >> + * back to Ohms until greater precision is actually needed. > >> + */ > >> + rimon /= 1000; > >> + > > > > Seems to me it would make more sense to limit the valid range of ti,rimon-milli-ohms > > to avoid the overflow. But then I really don't understand the reasoning to provide > > the property in milli-ohm, given the default value of 910 Ohm. What is a realistic > > lowest value that would make sense ? > > The highest value I've seen, when the tps25990 is alone, is 1370 > Ohms. That means a 30A overcurrent fault limit. > > With one TPS25895, I've seen 608 Ohms (110A limit) > > I have no idea what the realistic low limit is. To get to ~100 Ohms, you'd > need 8 devices (not hundreds ;) ) If one gets there, it might be > desirable to have 3 digits to play with, and not be limited by the unit. > > The DT folks really don't like when a property changes. Going with > milli-Ohms is way to anticipate the problem. > > The other way could be to use Ohms now, and if we ever get to point > where milli-Ohms precision is needed, add it then. The downside is that > the driver will need to support both properties. > > Would you prefer this ? > In practice the driver, as submitted, does _not_ support milli-Ohms to start with. It only supports Ohms. Worse, it doesn't range check the value, causing bad behavior (everything will be reported 0) if a value below 1,000 is provided, and still overflows if the value gets close to UINT_MAX. > > > But even if it is less than 1 Ohm I don't > > understand why it would make sense to completely ignore it. > > It would not make sense to ignore it. > But you do ... by setting m to 0 in that case. > > > >> + info = devm_kmemdup(dev, &tps25990_base_info, sizeof(*info), GFP_KERNEL); > >> + if (!info) > >> + return -ENOMEM; > >> + > >> + /* Adapt the current and power scale for each instance */ > >> + info->m[PSC_CURRENT_IN] *= rimon; > >> + info->m[PSC_POWER] *= rimon; > > > > Any rimon value < 1000 mOhm will result in m values of 0. > > Indeed. Such Rimon value would mean an over current limit > 50kA. I admit > I did really think much about such value. > > The idea was more keep some precision if we get somewhere near a 100 Ohms. > It doesn't, though, since the provided milli-Ohm value is divided by 1,000 (and the division doesn't even use DIV_ROUND_CLOSEST). Even though certain values don't make sense, there still needs to be a range check. And that missing range check triggers the next question: Why not just limit the upper range instead of ignoring the milli-part of the value ? Note that you might have used micro-Ohm (which is a standard devicetree resolution) and divide it by 1,000. That would have been perfectly fine. It would result in an upper resistor limit of 4,294 Ohm, which I'd assume should be acceptable. Overflows due to large values of m could have been avoided by adjusting .R if .m gets otherwise too large. Guenter
On Tue 10 Sep 2024 at 10:07, Guenter Roeck <linux@roeck-us.net> wrote: > On Tue, Sep 10, 2024 at 11:07:57AM +0200, Jerome Brunet wrote: >> On Mon 09 Sep 2024 at 15:52, Guenter Roeck <linux@roeck-us.net> wrote: >> >> [...] >> > Unrelated to the other comments: > > Documentation/hwmon/tps25990.rst | 141 ++++++++++++ > > Needs to be added to Documentation/hwmon/index.rst. > > +config SENSORS_TPS25990_REGULATOR > + bool "Regulator support for TPS25990 and compatibles" > + depends on SENSORS_TPS25990 && REGULATOR > + default SENSORS_TPS2599 > ^^^^^^^^^^^^^^^ TPS2599 ??? > >> >> + >> >> +#define TPS25990_DEFAULT_RIMON 910000 > > Where does the default come from anyway ? I don't immediately see the number > in the datasheet. It is Rimon value for the maximum current supported when the TPS25990 is alone (60A) with Viref on its default value: 1V - Section 8.3.4.2. There is no reason for it beside that. > >> >> +static int tps25990_write_protect_get(void *data, u64 *val) >> >> +{ >> >> + struct i2c_client *client = data; >> >> + >> >> + return tps25990_mfr_write_protect_active(client); >> >> +} >> >> + >> >> +static int tps25990_write_protect_set(void *data, u64 val) >> >> +{ >> >> + struct i2c_client *client = data; >> >> + >> >> + if (val > 1) >> >> + return -EINVAL; >> >> + >> >> + return tps25990_mfr_write_protect(client, val); >> >> +} >> >> + >> >> +DEFINE_DEBUGFS_ATTRIBUTE(tps25990_write_protect_fops, >> >> + tps25990_write_protect_get, >> >> + tps25990_write_protect_set, >> >> + "%llu\n"); >> >> + >> >> +static int tps25990_init_debugfs(struct i2c_client *client) >> >> +{ >> >> + struct dentry *dir; >> >> + >> >> + dir = pmbus_get_debugfs_dir(client); >> >> + if (!dir) >> >> + return -ENOENT; >> >> + >> >> + debugfs_create_file("write_protect", 0644, dir, >> >> + client, &tps25990_write_protect_fops); >> >> + >> >> + return 0; >> >> +} >> >> + >> >> +#else >> >> +static inline int tps25990_init_debugfs(struct i2c_client *client) >> >> +{ >> >> + return 0; >> >> +} >> >> +#endif >> >> + >> > >> > In general it is extremely undesirable to overwrite write protection. >> > Many chips support such attributes. If write protection is enabled, >> > it means that the board vendor does not want to have them changed. >> >> According to documentation, it protects against "unintented" writes, >> not 'wrong' or 'malicious'. If one goes in debugfs and write just '0' to >> a file, there is an intent at least. >> >> > Granted, that can be overwritten with direct i2c commands, but that >> > is what it should be. Anyone who really wants to disable write protection >> > should have to dig deeper than just writing into a debugfs or sysfs attribute. >> > Otherwise the protection becomes worthless. >> > If this is, for example, needed >> > for production to write initial settings, the production scripts should >> > disable (or enable) write protection by writing directly into command >> > registers. >> >> As I wrote in the cover letter, the write protection is always active on >> chip startup and it locks down almost everything, including things you may >> need to write past production, in the field. The history reset below is >> an example of such thing. >> >> To 'safely' remove the protection by writing i2c commands from >> userspace: >> * the device will need be unbinded first, >> * call i2cset >> * bind the device again >> >> That seems really cumbersome to do something like an history >> reset. Is this what you are suggesting ? >> >> bind/unbind could be skipped by forcing i2cset but that would add danger >> where we certainly don't want it. >> > > Not sure I understand the "danger" part. Either case, the problem is > deeper. If the driver is bound, i2cset will require the '-f' flag. Man page says it is dangerous do so, if 2 i2c commands happens at the same time I suppose. > The driver enables regulator support, which includes enabling and disabling > the output voltage. But that doesn't work unles write protect is disabled. > debugfs doesn't help there; that is way too late. Indeed OPERATION command is locked as well, I missed that. I'll drop that from the initial submission. The fact that is comes too late is also why I did not add extra features yet, things like GPIO support, GPDAC regulators, NVMEM blackbox, etc... I know we are not supposed to (and never will) support all the shiny features HW designers can think of, but it would be nice to unlock some of its potential. Do you have an idea ? (for later I mean) A module parm to do the unlock might work but seems a bit extreme. > >> > >> >> +/* >> >> + * TPS25990 has history reset based on MIN/AVG/PEAK instead of per sensor type >> >> + * Emulate the behaviour a pmbus limit_attr would have for consistency >> >> + * - Read: Do nothing and emit 0 >> >> + * - Write: Check the input is a number and reset >> >> + */ >> >> +static ssize_t tps25990_history_reset_show(struct device *dev, >> >> + struct device_attribute *devattr, >> >> + char *buf) >> >> +{ >> >> + return sysfs_emit(buf, "0\n"); >> >> +} >> >> + >> >> +static ssize_t tps25990_history_reset_store(struct device *dev, >> >> + struct device_attribute *devattr, >> >> + const char *buf, size_t count) >> >> +{ >> >> + struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr); >> >> + struct i2c_client *client = to_i2c_client(dev->parent); >> >> + s64 val; >> >> + int ret; >> >> + >> >> + if (kstrtos64(buf, 10, &val) < 0) >> >> + return -EINVAL; >> >> + >> >> + ret = pmbus_update_byte_data(client, 0, TPS25990_PK_MIN_AVG, >> >> + BIT(attr->index), BIT(attr->index)); >> >> + if (ret < 0) >> >> + return ret; >> >> + >> >> + return count; >> >> +} >> >> + >> >> +static SENSOR_DEVICE_ATTR_RW(highest_history_reset, tps25990_history_reset, 7); >> >> +static SENSOR_DEVICE_ATTR_RW(average_history_reset, tps25990_history_reset, 6); >> >> +static SENSOR_DEVICE_ATTR_RW(lowest_history_reset, tps25990_history_reset, 5); >> > >> > That is not a unique problem, and not a reason to introduce non-standard attributes. >> > Just attach the attribute to the first channel and document that it resets all >> > channels. >> >> Not sure I got this right so I'll rephrase. I should: >> * Pick a channel, say vin >> * Map the virtual reset register to hit the 3 resets above >> * Put in the documentation that it resets the other channels as well >> * Not allow independent resets of min/max/avg, just all 3 together ? >> > Correct. It is amazing what hardware designers come up with (here: > resetting history based on min/max/average instead of the sensor type > is novel), but I really don't want to introduce new attributes to > accommodate each variant. Sure. Make sense > I'd be open to introducing a global > PMBUS_VIRT_RESET_HISTORY virtual register and reset_history attribute > if you want to go there, but that would have to be in the PMBus core. Both solutions are fine by me. Do you have a preference ? > >> > >> >> + >> >> +static struct attribute *tps25990_attrs[] = { >> >> + &sensor_dev_attr_highest_history_reset.dev_attr.attr, >> >> + &sensor_dev_attr_average_history_reset.dev_attr.attr, >> >> + &sensor_dev_attr_lowest_history_reset.dev_attr.attr, >> >> + NULL, >> >> +}; >> >> + >> >> +ATTRIBUTE_GROUPS(tps25990); >> >> + >> >> +static int tps25990_get_addr(int reg) >> >> +{ >> >> + switch (reg) { >> >> + case PMBUS_SMBALERT_MASK: >> >> + /* >> >> + * Note: PMBUS_SMBALERT_MASK is not implemented on this chip >> >> + * Writing to this address raises CML errors. >> >> + * Instead it provides ALERT_MASK which allows to set the mask >> >> + * for each of the status registers, but not the specific bits >> >> + * in them. >> >> + * The default setup assert SMBA# if any bit is set in any of the >> >> + * status registers the chip has. This is as close as we can get >> >> + * to what pmbus_irq_setup() would set, sooo ... do nothing. >> >> + */ >> >> + return -ENXIO; >> > >> > Many chips have that problem. The core code ignores errors, and attempts to write >> > the command are limited to initialization. This is not a reason to overwrite >> > the command like this. If this does cause a real a problem wit hthe chip (other >> > than setting CML errors, which many chips not supporting the command do), >> > we should define a flag in include/linux/pmbus.h and explain its need. >> >> CML is error is the problem. Following pmbus_irq_setup() there is an >> uncleared fault because there is no register check on PMBUS_SMBALERT_MASK. >> >> When pmbus_core then gets here: >> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/hwmon/pmbus/pmbus_core.c?h=v6.11-rc7#n3386 >> >> pmbus_check_block_register() fails because of the uncleared fault and >> the 'mfr_id' property is silently not registered, eventhough the >> register is supported by the chip. This is how I noticed the problem. >> >> So, should I add flag in include/linux/pmbus.h to skip >> PMBUS_SMBALERT_MASK setup ? >> >> Another possibility is to call register_check() >> on it before using PMBUS_SMBALERT_MASK in pmbus_core. >> > > The problem, as you point out, is in pmbus_irq_setup(). Since the function > explicitly ignores errors from accessing PMBUS_SMBALERT_MASK, it should > either clear faults after it is done. I don't think we can rely on > register_check() because the register might exist but be read-only. > Noted. I'll add the fault clearing. >> > >> >> + case PMBUS_IIN_OC_FAULT_LIMIT: [...] >> >> +static int tps25990_probe(struct i2c_client *client) >> >> +{ >> >> + struct device *dev = &client->dev; >> >> + struct pmbus_driver_info *info; >> >> + u32 rimon = TPS25990_DEFAULT_RIMON; >> >> + int ret; >> >> + >> >> + ret = device_property_read_u32(dev, "ti,rimon-milli-ohms", &rimon); >> >> + if (ret == -EINVAL) { >> >> + dev_warn(dev, >> >> + "using default rimon: current and power scale possibly wrong\n"); >> > >> > This is not an appropriate warning. It is perfectly fine to load the driver >> > if there is no ti,rimon-milli-ohms property. >> >> I should have commented more on the default value. It is meant for the >> case where the device is instanciated through i2c sys 'new_device', >> which is meant for debugging purpose. In that particular case, it does >> not really matter if the current and power scale are wrong. >> >> There is no way to pass device properties when instanciating device >> through that interface, as far as I know. >> >> In every other cases, a correct Rimon value is expected. >> I could turn the above to an error. It means loading through i2c sys >> would not possible for this driver. >> >> Would it be better ? >> > > We use default values for pretty much all drivers, so I don't see why > this one should be different. The driver should still be usable on a > system without devicetree support. There is a reason for the sensors > configuration file. > Supporting more than DT is a concern. That is why I did not use the DT specific API. In theory, the one used should support other sources, such as ACPI, I think. Thanks for pointing out the sensor configuration file. I did not know calculation were possible, and acceptable, at that stage. So, IIUC, I could just drop the device property, then the device would be used in the same way on all the platform, DT or not ? I like that a lot. All I would have to do is add something in the documentation about it, I guess. With default value of 1000, instead 910, the range would still be good in sysfs and calculation simple in userspace. That would solve the unit discussion as well, a nice bonus ;) >> > >> >> + } else if (ret < 0) { >> >> + return dev_err_probe(dev, ret, "failed get rimon\n"); >> >> + } >> >> + >> >> + /* >> >> + * TPS25990 may be stacked with several TPS25895, allowing a higher >> >> + * current. The higher the allowed current is, the lower rimon >> >> + * will be. How low it can realistically get is unknown. >> >> + * To avoid problems with precision later on, rimon is provided in >> >> + * milli Ohms. This is a precaution to keep a stable ABI. >> >> + * At the moment, doing the calculation with rimon in milli Ohms >> >> + * would overflow the s32 'm' in the direct conversion. Convert it >> >> + * back to Ohms until greater precision is actually needed. >> >> + */ >> >> + rimon /= 1000; >> >> + >> > >> > Seems to me it would make more sense to limit the valid range of ti,rimon-milli-ohms >> > to avoid the overflow. But then I really don't understand the reasoning to provide >> > the property in milli-ohm, given the default value of 910 Ohm. What is a realistic >> > lowest value that would make sense ? >> >> The highest value I've seen, when the tps25990 is alone, is 1370 >> Ohms. That means a 30A overcurrent fault limit. >> >> With one TPS25895, I've seen 608 Ohms (110A limit) >> >> I have no idea what the realistic low limit is. To get to ~100 Ohms, you'd >> need 8 devices (not hundreds ;) ) If one gets there, it might be >> desirable to have 3 digits to play with, and not be limited by the unit. >> >> The DT folks really don't like when a property changes. Going with >> milli-Ohms is way to anticipate the problem. >> >> The other way could be to use Ohms now, and if we ever get to point >> where milli-Ohms precision is needed, add it then. The downside is that >> the driver will need to support both properties. >> >> Would you prefer this ? >> > > In practice the driver, as submitted, does _not_ support milli-Ohms > to start with. It only supports Ohms. Worse, it doesn't range check the > value, causing bad behavior (everything will be reported 0) if a value > below 1,000 is provided, and still overflows if the value gets close to > UINT_MAX. Yes I relied on the value being sane-ish. 4 MOhms (or even 2) is not. > >> >> > But even if it is less than 1 Ohm I don't >> > understand why it would make sense to completely ignore it. >> >> It would not make sense to ignore it. >> > But you do ... by setting m to 0 in that case. I did not intentionally set 0. Let's just call it for what it is: a bug that needs fixing, if the property stays. > >> > >> >> + info = devm_kmemdup(dev, &tps25990_base_info, sizeof(*info), GFP_KERNEL); >> >> + if (!info) >> >> + return -ENOMEM; >> >> + >> >> + /* Adapt the current and power scale for each instance */ >> >> + info->m[PSC_CURRENT_IN] *= rimon; >> >> + info->m[PSC_POWER] *= rimon; >> > >> > Any rimon value < 1000 mOhm will result in m values of 0. >> >> Indeed. Such Rimon value would mean an over current limit > 50kA. I admit >> I did really think much about such value. >> >> The idea was more keep some precision if we get somewhere near a 100 Ohms. >> > > It doesn't, though, since the provided milli-Ohm value is divided by 1,000 > (and the division doesn't even use DIV_ROUND_CLOSEST). Even though certain > values don't make sense, there still needs to be a range check. And that > missing range check triggers the next question: Why not just limit the upper > range instead of ignoring the milli-part of the value ? If there was an actual range in the documentation, I'd be happy to check it, but there is not one. I don't think I should make a range out of thin air. That being said, we are not going get into Mega-Ohms or Micro-Ohms territory with this chip, that's for sure. I was trying to strike a balance in between. The upper limit of Rimon would be set by what you consider the lower acceptable limit for over current protection (Iocp). Isn't 10A ? or 5A or even 1A ? I don't know. I'm not sure the driver should disallow any sane value if the HW can do it. What is sane then ... > > Note that you might have used micro-Ohm (which is a standard devicetree > resolution) and divide it by 1,000. Take Rimon = 1370 Ohms, which provide an Iocp of 30A. Micro-Ohms divided by 1000: 13700000000 / 1000 = 1370000. For current: m = 9538 * 1370000 = 13067060000 This will overflow m on 32bits systems for struct pmbus_driver_info. In pmbus_reg2data_direct, m is an s32 so it would overflow there as well regardless of the arch. Micro or milli-Ohms, the matter is the same. Solution might simply be to do m calculation over 64bits then divide result so it fits the 32bits type used in pmbus_core. > That would have been perfectly fine. > It would result in an upper resistor limit of 4,294 Ohm, which I'd assume > should be acceptable. Means the driver will not allow a Iocp lower that 234mA. It is probably realistic. > Overflows due to large values of m could have been > avoided by adjusting .R if .m gets otherwise too large. I think we are down 2 solutions then: 1) Drop the device property completely, use a sane default and rely on libsensor for the final calculation. 2) Property in micro-ohms, with some calculation to fit m in 32bits. Probably need to expose Rimon in debugfs too, so the users may check the value used and revert back to lib-sensor calc if necessary. I tend to prefer 1) for its simplicity and lack of added constraints. Is it Ok with you ? > > Guenter