Message-ID: <c2dce65a-6e7a-0c5b-49e9-b50b4f4e91ea@roeck-us.net>
Date:   Sun, 29 Jan 2017 11:50:55 -0800
From:   Guenter Roeck <linux@...ck-us.net>
To:     eajames.ibm@...il.com
Cc:     devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
        linux-hwmon@...r.kernel.org, linux-doc@...r.kernel.org,
        jdelvare@...e.com, corbet@....net, mark.rutland@....com,
        robh+dt@...nel.org, =wsa@...-dreams.de, andrew@...id.au,
        joel@....id.au, benh@...nel.crashing.org,
        "Edward A. James" <eajames@...ibm.com>
Subject: Re: [PATCH linux v4 2/6] hwmon: occ: Add sysfs interface

On 01/26/2017 03:19 PM, eajames.ibm@...il.com wrote:
> From: "Edward A. James" <eajames@...ibm.com>
>
> Add a generic mechanism to expose the sensors provided by the OCC in
> sysfs.
>
> Signed-off-by: Edward A. James <eajames@...ibm.com>
> Signed-off-by: Andrew Jeffery <andrew@...id.au>
> ---
>  Documentation/hwmon/occ       |  62 ++++++++++
>  drivers/hwmon/occ/Makefile    |   2 +-
>  drivers/hwmon/occ/occ_sysfs.c | 259 ++++++++++++++++++++++++++++++++++++++++++
>  drivers/hwmon/occ/occ_sysfs.h |  30 +++++
>  4 files changed, 352 insertions(+), 1 deletion(-)
>  create mode 100644 drivers/hwmon/occ/occ_sysfs.c
>  create mode 100644 drivers/hwmon/occ/occ_sysfs.h
>
> diff --git a/Documentation/hwmon/occ b/Documentation/hwmon/occ
> index 79d1642..d0bdf06 100644
> --- a/Documentation/hwmon/occ
> +++ b/Documentation/hwmon/occ
> @@ -25,6 +25,68 @@ Currently, all versions of the OCC support four types of sensor data: power,
>  temperature, frequency, and "caps," which indicate limits and thresholds used
>  internally on the OCC.
>
> +sysfs Entries
> +-------------
> +
> +The OCC driver uses the hwmon sysfs framework to provide data to userspace.
> +
> +The driver exports a number of sysfs files for each type of sensor. The
> +sensor-specific files vary depending on the processor type, though many of the
> +attributes are common for both the POWER8 and POWER9.
> +
> +The hwmon interface cannot define every type of sensor that may be used.
> +Therefore, the frequency sensor on the OCC uses the "input" type sensor defined
> +by the hwmon interface, rather than defining a new type of custom sensor.
> +
> +Below are detailed the names and meaning of each sensor file for both types of
> +processors. All sensors are read-only unless otherwise specified. <x> indicates
> +the hwmon index. sensor id indicates the unique internal OCC identifer. Please
> +see the POWER OCC specification for details on all these sensor values.
> +
> +frequency:
> +	all processors:
> +		in<x>_input - frequency value
> +		in<x>_label - sensor id
> +temperature:
> +	POWER8:
> +		temp<x>_input - temperature value
> +		temp<x>_label - sensor id
> +	POWER9 (in addition to above):
> +		temp<x>_type - FRU type
> +
> +power:
> +	POWER8:
> +		power<x>_input - power value
> +		power<x>_label - sensor id
> +		power<x>_average - accumulator
> +		power<x>_average_interval - update tag (number of samples in
> +			accumulator)
> +	POWER9:
> +		power<x>_input - power value
> +		power<x>_label - sensor id
> +		power<x>_average_min - accumulator[0]
> +		power<x>_average_max - accumulator[1] (64 bits total)
> +		power<x>_average_interval - update tag
> +		power<x>_reset_history - (function_id | (apss_channel << 8)
> +
> +caps:
> +	POWER8:
> +		power<x>_cap - current powercap
> +		power<x>_cap_max - max powercap
> +		power<x>_cap_min - min powercap
> +		power<x>_max - normal powercap
> +		power<x>_alarm - user powercap, r/w
> +	POWER9:
> +		power<x>_cap_alarm - user powercap source
> +
> +The driver also provides two sysfs entries through hwmon to better
> +control the driver and monitor the master OCC. Though there may be multiple
> +OCCs present on the system, these two files are only present for the "master"
> +OCC.
> +	name - read the name of the driver
> +	update_interval - read or write the minimum interval for polling the
> +		OCC.
> +
>  BMC - Host Communications
>  -------------------------
>
> diff --git a/drivers/hwmon/occ/Makefile b/drivers/hwmon/occ/Makefile
> index 93cb52f..a6881f9 100644
> --- a/drivers/hwmon/occ/Makefile
> +++ b/drivers/hwmon/occ/Makefile
> @@ -1 +1 @@
> -obj-$(CONFIG_SENSORS_PPC_OCC) += occ.o
> +obj-$(CONFIG_SENSORS_PPC_OCC) += occ.o occ_sysfs.o
> diff --git a/drivers/hwmon/occ/occ_sysfs.c b/drivers/hwmon/occ/occ_sysfs.c
> new file mode 100644
> index 0000000..ea2aa04
> --- /dev/null
> +++ b/drivers/hwmon/occ/occ_sysfs.c
> @@ -0,0 +1,259 @@
> +/*
> + * occ_sysfs.c - OCC sysfs interface
> + *
> + * This file contains the methods and data structures for implementing the OCC
> + * hwmon sysfs entries.
> + *
> + * Copyright 2016 IBM Corp.
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License as published by
> + * the Free Software Foundation; either version 2 of the License, or
> + * (at your option) any later version.
> + *
> + * This program is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + * GNU General Public License for more details.
> + */
> +
> +#include <linux/delay.h>
> +#include <linux/device.h>
> +#include <linux/err.h>
> +#include <linux/hwmon.h>
> +#include <linux/hwmon-sysfs.h>
> +#include <linux/init.h>
> +#include <linux/jiffies.h>
> +#include <linux/kernel.h>
> +#include <linux/module.h>
> +#include <linux/mutex.h>
> +#include <linux/slab.h>
> +#include "occ.h"
> +#include "occ_sysfs.h"
> +
> +#define RESP_RETURN_CMD_INVAL	0x13
> +#define OCC_HWMON_NAME_LENGTH	32
> +
> +struct occ_sysfs {
> +	struct device *dev;
> +	struct occ *occ;
> +
> +	char hwmon_name[OCC_HWMON_NAME_LENGTH + 1];
> +	const u32 *sensor_hwmon_configs;
> +	struct hwmon_channel_info **occ_sensors;
> +	struct hwmon_chip_info occ_info;
> +	u16 user_powercap;
> +};
> +
> +static int occ_hwmon_read(struct device *dev, enum hwmon_sensor_types type,
> +			  u32 attr, int channel, long *val)
> +{
> +	int rc;
> +	struct occ_sysfs *driver = dev_get_drvdata(dev);
> +	struct occ *occ = driver->occ;
> +
> +	switch (type) {
> +	case hwmon_in:
> +		rc = occ_get_sensor_field(occ, FREQ, channel, attr, val);
> +		break;
> +	case hwmon_temp:
> +		rc = occ_get_sensor_field(occ, TEMP, channel, attr, val);
> +		break;
> +	case hwmon_power:
> +		rc = occ_get_sensor_field(occ, POWER, channel, attr, val);
> +		break;
> +	default:
> +		rc = -EOPNOTSUPP;
> +	}
> +
> +	return rc;
> +}
> +
> +static int occ_hwmon_read_string(struct device *dev,
> +				 enum hwmon_sensor_types type, u32 attr,
> +				 int channel, char **str)
> +{
> +	int rc;
> +	unsigned long val = 0;
> +
> +	if (!((type == hwmon_in && attr == hwmon_in_label) ||
> +	    (type == hwmon_temp && attr == hwmon_temp_label) ||
> +	    (type == hwmon_power && attr == hwmon_power_label)))
> +		return -EOPNOTSUPP;
> +
> +	/* will fetch the "label", the sensor_id */
> +	rc = occ_hwmon_read(dev, type, attr, channel, &val);
> +	if (rc < 0)
> +		return rc;
> +
> +	rc = snprintf(*str, PAGE_SIZE - 1, "%lu", val);
> +	if (rc > 0)
> +		rc = 0;
> +
> +	return rc;
> +}
> +
> +static int occ_hwmon_write(struct device *dev, enum hwmon_sensor_types type,
> +			   u32 attr, int channel, long val)
> +{
> +	int rc = 0;

Unnecessary initialization.

> +	struct occ_sysfs *driver = dev_get_drvdata(dev);
> +
> +	if (type == hwmon_power && attr == hwmon_power_alarm) {
> +		rc = occ_set_user_powercap(driver->occ, val);
> +		if (rc) {
> +			if (rc == RESP_RETURN_CMD_INVAL) {

Is the mix of private and public error codes necessary ? Any chance
the called code can just return -EINVAL ?

> +				dev_err(dev,
> +					"set invalid powercap value: %ld\n",
> +					val);
> +				return -EINVAL;
> +			}
> +
> +			dev_err(dev, "set user powercap failed: 0x:%x\n", rc);
> +			return rc;
> +		}
> +
> +		driver->user_powercap = val;
> +
> +		return rc;
> +	}
> +
> +	return -EOPNOTSUPP;
> +}
> +
> +static umode_t occ_is_visible(const void *data, enum hwmon_sensor_types type,
> +			      u32 attr, int channel)
> +{
> +	const struct occ_sysfs *driver = data;
> +
> +	switch (type) {
> +	case hwmon_chip:
> +		if (attr == hwmon_chip_update_interval)
> +			return S_IRUGO | S_IWUSR;

It might be better to just use octals directly to avoid the annoying checkpatch
warnings.

> +		break;
> +	case hwmon_in:
> +		if (BIT(attr) & driver->sensor_hwmon_configs[0])
> +			return S_IRUGO;
> +		break;
> +	case hwmon_temp:
> +		if (BIT(attr) & driver->sensor_hwmon_configs[1])
> +			return S_IRUGO;
> +		break;
> +	case hwmon_power:
> +		/* user power limit */
> +		if (attr == hwmon_power_alarm)
> +			return S_IRUGO | S_IWUSR;
> +		else if ((BIT(attr) & driver->sensor_hwmon_configs[2]) ||
> +			 (BIT(attr) & driver->sensor_hwmon_configs[3]))
> +			return S_IRUGO;
> +		break;
> +	default:
> +		return 0;
> +	}
> +
> +	return 0;
> +}
> +
> +static const struct hwmon_ops occ_hwmon_ops = {
> +	.is_visible = occ_is_visible,
> +	.read = occ_hwmon_read,
> +	.read_string = occ_hwmon_read_string,
> +	.write = occ_hwmon_write,
> +};
> +
> +static const enum hwmon_sensor_types occ_sensor_types[MAX_OCC_SENSOR_TYPE] = {
> +	hwmon_in,
> +	hwmon_temp,
> +	hwmon_power,
> +	hwmon_power

So you have two sets of power sensors, meaning you'll end up with two sets
of hwmon_channel_info describing hwmon_power ? This kind of misuses the API.
I hope that doesn't create duplicate attributes; I think it might if
the actually supported attributes overlap. Also, this means that the attribute
index association gets lost. For example, if the alarm attributes
are in the second group, power1_alarm would not necessarily align with,
say, power1_max.

Is there any chance to avoid this ?

> +};
> +
> +struct occ_sysfs *occ_sysfs_start(struct device *dev, struct occ *occ,
> +				  const u32 *sensor_hwmon_configs,
> +				  const char *name)
> +{
> +	int rc, i, j, num_sensors, index = 0, id;
> +	char *brk;
> +	struct occ_blocks *resp = NULL;
> +	u32 *sensor_config;
> +	struct occ_sysfs *hwmon = devm_kzalloc(dev, sizeof(struct occ_sysfs),
> +					       GFP_KERNEL);
> +	if (!hwmon)
> +		return ERR_PTR(-ENOMEM);
> +
> +	/* need space for null-termination */
> +	hwmon->occ_sensors =
> +		devm_kzalloc(dev, sizeof(struct hwmon_channel_info *) *
> +			     (MAX_OCC_SENSOR_TYPE + 1), GFP_KERNEL);
> +	if (!hwmon->occ_sensors)
> +		return ERR_PTR(-ENOMEM);
> +
> +	hwmon->occ = occ;
> +	hwmon->sensor_hwmon_configs = sensor_hwmon_configs;
> +	hwmon->occ_info.ops = &occ_hwmon_ops;
> +	hwmon->occ_info.info =
> +		(const struct hwmon_channel_info **)hwmon->occ_sensors;
> +
> +	occ_get_response_blocks(occ, &resp);
> +
> +	for (i = 0; i < MAX_OCC_SENSOR_TYPE; ++i)
> +		resp->sensor_block_id[i] = -1;
> +
> +	/* read sensor data from occ */
> +	rc = occ_update_device(occ);
> +	if (rc)
> +		return ERR_PTR(rc);
> +
> +	for (i = 0; i < MAX_OCC_SENSOR_TYPE; i++) {
> +		id = resp->sensor_block_id[i];
> +		if (id < 0)
> +			continue;
> +
> +		num_sensors = resp->blocks[id].header.num_sensors;
> +		/* need null-termination */
> +		sensor_config = devm_kzalloc(dev,
> +					     sizeof(u32) * (num_sensors + 1),
> +					     GFP_KERNEL);
> +		if (!sensor_config)
> +			return ERR_PTR(-ENOMEM);
> +
> +		for (j = 0; j < num_sensors; j++)
> +			sensor_config[j] = sensor_hwmon_configs[i];
> +
> +		hwmon->occ_sensors[index] =
> +			devm_kzalloc(dev, sizeof(struct hwmon_channel_info),
> +				     GFP_KERNEL);
> +		if (!hwmon->occ_sensors[index])
> +			return ERR_PTR(-ENOMEM);
> +
> +		hwmon->occ_sensors[index]->type = occ_sensor_types[i];
> +		hwmon->occ_sensors[index]->config = sensor_config;
> +		index++;
> +	}
> +
> +	/* search for bad chars */
> +	strncpy(hwmon->hwmon_name, name, OCC_HWMON_NAME_LENGTH);
> +	brk = strpbrk(hwmon->hwmon_name, "-* \t\n");
> +	while (brk) {
> +		*brk = '_';
> +		brk = strpbrk(brk,  "-* \t\n");
> +	}
> +
> +	hwmon->dev = devm_hwmon_device_register_with_info(dev,
> +							  hwmon->hwmon_name,
> +							  hwmon,
> +							  &hwmon->occ_info,
> +							  NULL);
> +	if (IS_ERR(hwmon->dev)) {
> +		dev_err(dev, "cannot register hwmon device %s: %ld\n",
> +			hwmon->hwmon_name, PTR_ERR(hwmon->dev));
> +		return ERR_CAST(hwmon->dev);
> +	}
> +
> +	return hwmon;
> +}
> +EXPORT_SYMBOL(occ_sysfs_start);
> +
> +MODULE_AUTHOR("Eddie James <eajames@...ibm.com>");
> +MODULE_DESCRIPTION("OCC sysfs driver");
> +MODULE_LICENSE("GPL");
> diff --git a/drivers/hwmon/occ/occ_sysfs.h b/drivers/hwmon/occ/occ_sysfs.h
> new file mode 100644
> index 0000000..462f9d2
> --- /dev/null
> +++ b/drivers/hwmon/occ/occ_sysfs.h
> @@ -0,0 +1,30 @@
> +/*
> + * occ_sysfs.h - OCC sysfs interface
> + *
> + * This file contains the data structures and function prototypes for the OCC
> + * hwmon sysfs entries.
> + *
> + * Copyright 2016 IBM Corp.
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License as published by
> + * the Free Software Foundation; either version 2 of the License, or
> + * (at your option) any later version.
> + *
> + * This program is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + * GNU General Public License for more details.
> + */
> +
> +#ifndef __OCC_SYSFS_H__
> +#define __OCC_SYSFS_H__
> +
> +struct occ;
> +struct device;
> +struct occ_sysfs;
> +
> +struct occ_sysfs *occ_sysfs_start(struct device *dev, struct occ *occ,
> +				  const u32 *sensor_hwmon_configs,
> +				  const char *name);
> +#endif /* __OCC_SYSFS_H__ */
>

Powered by blists - more mailing lists