[PATCH v5 2/2] hwmon: (coretemp) Report power and energy consumption

From: Guenter Roeck
Date: Mon Sep 23 2013 - 16:35:34 EST

Supported by SandyBridge and newer CPUs.

Cc: Fenghua Yu <fenghua.yu@xxxxxxxxx>
Signed-off-by: Guenter Roeck <linux@xxxxxxxxxxxx>
---
v5: Merged to 3.12-rc1
v4: Replaced INIT_DELAYED_WORK with INIT_DEFERRABLE_WORK
v3: Added Cc:
v2: Fix missing symbol error on i386 builds
Dropped leftover pr_info from debugging

I am inclined to drop this patch set, as I received neither test nor code
review feedback even though the patch set has been out there for a long time.
Also, I am concerned that it overlaps or even conflicts with the proposed
Power Cap/RAPL driver (https://lkml.org/lkml/2013/9/19/260).

Consider this to be a Last Call.

Documentation/hwmon/coretemp | 16 ++
drivers/hwmon/coretemp.c | 386 +++++++++++++++++++++++++++++++++++++++---
2 files changed, 376 insertions(+), 26 deletions(-)

diff --git a/Documentation/hwmon/coretemp b/Documentation/hwmon/coretemp
index fec5a9b..cdb533d 100644
--- a/Documentation/hwmon/coretemp
+++ b/Documentation/hwmon/coretemp
@@ -43,6 +43,22 @@ tempX_crit_alarm - Set when Out-of-spec bit is set, never clears.
tempX_label - Contains string "Core X", where X is processor
number. For Package temp, this will be "Physical id Y",
where Y is the package number.
+energy1_input - Package energy consumption since driver was loaded
+power1_cap - Package power limit
+power1_input - Package power consumption
+power1_max - Maximum package power
+energy2_input - Cores energy consumption since driver was loaded
+power2_cap - Cores power limit
+power2_input - Cores power consumption
+energy3_input - Graphics domain energy consumption since driver was loaded
+power3_cap - Graphics domain power limit
+power3_input - Graphics domain power consumption
+energy4_input - DRAM domain energy consumption since driver was loaded
+power4_cap - DRAM domain power limit
+power4_input - DRAM domain power consumption
+
+Graphics and DRAM power domains are not supported on all chip variants.
+powerX_cap is only reported if enabled.

On CPU models which support it, TjMax is read from a model-specific register.
On other models, it is set to an arbitrary value based on weak heuristics.
diff --git a/drivers/hwmon/coretemp.c b/drivers/hwmon/coretemp.c
index 38d458b..d79fc8c 100644
--- a/drivers/hwmon/coretemp.c
+++ b/drivers/hwmon/coretemp.c
@@ -54,6 +54,7 @@ MODULE_PARM_DESC(tjmax, "TjMax value in degrees Celsius");
#define NUM_REAL_CORES 32 /* Number of Real cores per cpu */
#define CORETEMP_NAME_LENGTH 17 /* String Length of attrs */
#define MAX_CORE_ATTRS 4 /* Maximum no of basic attrs */
+#define MAX_POWER_ATTRS 6 /* Maximum no of power attrs */
#define TOTAL_ATTRS (MAX_CORE_ATTRS + 1)
#define MAX_CORE_DATA (NUM_REAL_CORES + BASE_SYSFS_ATTR_NO)

@@ -67,6 +68,36 @@ MODULE_PARM_DESC(tjmax, "TjMax value in degrees Celsius");
#define for_each_sibling(i, cpu) for (i = 0; false; )
#endif

+#define RAPL_PKG_INDEX 0
+#define RAPL_CORE_INDEX 1
+#define RAPL_GFX_INDEX 2
+#define RAPL_DRAM_INDEX 3
+#define RAPL_DOMAINS (RAPL_DRAM_INDEX + 1)
+
+#define HAS_RAPL_PKG (1 << RAPL_PKG_INDEX)
+#define HAS_RAPL_CORE (1 << RAPL_CORE_INDEX)
+
+static const u32 power_limit_msr[] = {
+ MSR_PKG_POWER_LIMIT,
+ MSR_PP0_POWER_LIMIT,
+ MSR_PP1_POWER_LIMIT,
+ MSR_DRAM_POWER_LIMIT,
+};
+
+static const u32 energy_status_msr[] = {
+ MSR_PKG_ENERGY_STATUS,
+ MSR_PP0_ENERGY_STATUS,
+ MSR_PP1_ENERGY_STATUS,
+ MSR_DRAM_ENERGY_STATUS,
+};
+
+static const u32 power_info_msr[] = {
+ MSR_PKG_POWER_INFO,
+ 0,
+ 0,
+ MSR_DRAM_POWER_INFO,
+};
+
/*
* Per-Core Temperature Data
* @last_updated: The time when the current temperature value was updated
@@ -75,10 +106,20 @@ MODULE_PARM_DESC(tjmax, "TjMax value in degrees Celsius");
* This value is passed as "id" field to rdmsr/wrmsr functions.
* @status_reg: One of IA32_THERM_STATUS or IA32_PACKAGE_THERM_STATUS,
* from where the temperature values should be read.
- * @attr_size: Total number of pre-core attrs displayed in the sysfs.
+ * @attr_size: Total number of per-core attrs displayed in the sysfs.
* @is_pkg_data: If this is true, the core_data holds pkgtemp data.
* Otherwise, core_data holds coretemp data.
* @valid: If this is true, the current temperature is valid.
+ * @rapl: Bit mask or supported RAPL (energy measurement) domains
+ * @rapl_attr_mask: Masks for enabled power attributes per domain
+ * @rapl_timestamp: Last time energy/power was retrieved
+ * @rapl_power_units: Units of power as reported by the chip
+ * @rapl_energy_units: Units of energy as reported by the chip
+ * @rapl_energy_raw: Most recent energy measurement (raw) per domain
+ * @rapl_energy: cumulative energy (mJ) per domain
+ * @rapl_power: current power usage (mW) per domain
+ * @rapl_power_max: maximum power (TDP, mW) per domain as reported
+ * by the chip
*/
struct core_data {
int temp;
@@ -91,9 +132,24 @@ struct core_data {
int attr_size;
bool is_pkg_data;
bool valid;
- struct sensor_device_attribute sd_attrs[TOTAL_ATTRS];
+ struct sensor_device_attribute_2 sd_attrs[TOTAL_ATTRS];
char attr_name[TOTAL_ATTRS][CORETEMP_NAME_LENGTH];
struct mutex update_lock;
+ /* power/energy */
+ struct sensor_device_attribute_2 sd_power_attrs[MAX_POWER_ATTRS
+ * RAPL_DOMAINS];
+ char power_attr_name[MAX_POWER_ATTRS * RAPL_DOMAINS]
+ [CORETEMP_NAME_LENGTH];
+ u32 rapl;
+ u32 rapl_attr_mask[RAPL_DOMAINS];
+ unsigned long rapl_timestamp;
+ u32 rapl_power_units;
+ u32 rapl_energy_units;
+ u32 rapl_energy_raw[RAPL_DOMAINS];
+ u64 rapl_energy[RAPL_DOMAINS];
+ u32 rapl_power[RAPL_DOMAINS];
+ u32 rapl_power_max[RAPL_DOMAINS];
+ struct delayed_work rapl_wq;
};

/* Platform Data per Physical CPU */
@@ -122,7 +178,7 @@ static ssize_t show_name(struct device *dev,
static ssize_t show_label(struct device *dev,
struct device_attribute *devattr, char *buf)
{
- struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
+ struct sensor_device_attribute_2 *attr = to_sensor_dev_attr_2(devattr);
struct platform_data *pdata = dev_get_drvdata(dev);
struct core_data *tdata = pdata->core_data[attr->index];

@@ -132,11 +188,34 @@ static ssize_t show_label(struct device *dev,
return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
}

+static const char * const power_domains[] = {
+ "Pkg",
+ "Cores",
+ "Graphics",
+ "DRAM",
+};
+
+static ssize_t show_power_label(struct device *dev,
+ struct device_attribute *devattr, char *buf)
+{
+ struct sensor_device_attribute_2 *attr = to_sensor_dev_attr_2(devattr);
+
+ return sprintf(buf, "%s power\n", power_domains[attr->nr]);
+}
+
+static ssize_t show_energy_label(struct device *dev,
+ struct device_attribute *devattr, char *buf)
+{
+ struct sensor_device_attribute_2 *attr = to_sensor_dev_attr_2(devattr);
+
+ return sprintf(buf, "%s energy\n", power_domains[attr->nr]);
+}
+
static ssize_t show_crit_alarm(struct device *dev,
struct device_attribute *devattr, char *buf)
{
u32 eax, edx;
- struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
+ struct sensor_device_attribute_2 *attr = to_sensor_dev_attr_2(devattr);
struct platform_data *pdata = dev_get_drvdata(dev);
struct core_data *tdata = pdata->core_data[attr->index];

@@ -148,7 +227,7 @@ static ssize_t show_crit_alarm(struct device *dev,
static ssize_t show_tjmax(struct device *dev,
struct device_attribute *devattr, char *buf)
{
- struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
+ struct sensor_device_attribute_2 *attr = to_sensor_dev_attr_2(devattr);
struct platform_data *pdata = dev_get_drvdata(dev);

return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
@@ -157,7 +236,7 @@ static ssize_t show_tjmax(struct device *dev,
static ssize_t show_ttarget(struct device *dev,
struct device_attribute *devattr, char *buf)
{
- struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
+ struct sensor_device_attribute_2 *attr = to_sensor_dev_attr_2(devattr);
struct platform_data *pdata = dev_get_drvdata(dev);

return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
@@ -167,7 +246,7 @@ static ssize_t show_temp(struct device *dev,
struct device_attribute *devattr, char *buf)
{
u32 eax, edx;
- struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
+ struct sensor_device_attribute_2 *attr = to_sensor_dev_attr_2(devattr);
struct platform_data *pdata = dev_get_drvdata(dev);
struct core_data *tdata = pdata->core_data[attr->index];

@@ -190,6 +269,58 @@ static ssize_t show_temp(struct device *dev,
return tdata->valid ? sprintf(buf, "%d\n", tdata->temp) : -EAGAIN;
}

+static ssize_t show_power(struct device *dev,
+ struct device_attribute *devattr, char *buf)
+{
+ struct sensor_device_attribute_2 *attr = to_sensor_dev_attr_2(devattr);
+ struct platform_data *pdata = dev_get_drvdata(dev);
+ struct core_data *tdata = pdata->core_data[attr->index];
+
+ return sprintf(buf, "%u\n", tdata->rapl_power[attr->nr] * 1000);
+}
+
+static ssize_t show_power_max(struct device *dev,
+ struct device_attribute *devattr, char *buf)
+{
+ struct sensor_device_attribute_2 *attr = to_sensor_dev_attr_2(devattr);
+ struct platform_data *pdata = dev_get_drvdata(dev);
+ struct core_data *tdata = pdata->core_data[attr->index];
+
+ return sprintf(buf, "%u\n", tdata->rapl_power_max[attr->nr] * 1000);
+}
+
+static ssize_t show_power_cap(struct device *dev,
+ struct device_attribute *devattr, char *buf)
+{
+ struct sensor_device_attribute_2 *attr = to_sensor_dev_attr_2(devattr);
+ struct platform_data *pdata = dev_get_drvdata(dev);
+ struct core_data *tdata = pdata->core_data[attr->index];
+ u32 eax, edx;
+ u64 cap;
+
+ rdmsr_on_cpu(tdata->cpu, power_limit_msr[attr->nr], &eax, &edx);
+
+ /* Report lowest configured cap limit */
+ cap = (eax & 0x8000) ? (eax & 0x7fff) : 0;
+ if (attr->nr && (edx & 0x8000) && (!(eax & 0x8000) ||
+ (edx & 0x7fff) < cap))
+ cap = edx & 0x7fff;
+
+ cap = (cap * 1000000LL) >> tdata->rapl_power_units;
+
+ return sprintf(buf, "%llu\n", cap);
+}
+
+static ssize_t show_energy(struct device *dev,
+ struct device_attribute *devattr, char *buf)
+{
+ struct sensor_device_attribute_2 *attr = to_sensor_dev_attr_2(devattr);
+ struct platform_data *pdata = dev_get_drvdata(dev);
+ struct core_data *tdata = pdata->core_data[attr->index];
+
+ return sprintf(buf, "%llu\n", tdata->rapl_energy[attr->nr] * 1000ULL);
+}
+
struct tjmax {
char const *id;
int tjmax;
@@ -377,10 +508,55 @@ static int create_name_attr(struct platform_data *pdata,
return device_create_file(dev, &pdata->name_attr);
}

-static int create_core_attrs(struct core_data *tdata, struct device *dev,
- int attr_no)
+static void remove_core_files(struct device *dev, struct core_data *tdata)
{
- int err, i;
+ int d, i;
+
+ for (i = 0; i < tdata->attr_size; i++)
+ device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
+
+ if (!tdata->is_pkg_data)
+ return;
+
+ for (d = 0; d < RAPL_DOMAINS; d++) {
+ if (!(tdata->rapl & (1 << d)))
+ continue;
+ for (i = 0; i < MAX_POWER_ATTRS; i++) {
+ int index = d * MAX_POWER_ATTRS + i;
+
+ if (!(tdata->rapl_attr_mask[d] & (1 << i)))
+ continue;
+ device_remove_file(dev,
+ &tdata->sd_power_attrs[index].dev_attr);
+ }
+ }
+}
+
+static int create_core_attr(struct device *dev,
+ struct sensor_device_attribute_2 *attr,
+ char *attr_name,
+ int index, int nr,
+ ssize_t (*const rd_ptr)(struct device *,
+ struct device_attribute *, char *),
+ const char *const template)
+{
+ int err;
+
+ snprintf(attr_name, CORETEMP_NAME_LENGTH, template, nr);
+ sysfs_attr_init(&attr->dev_attr.attr);
+ attr->dev_attr.attr.name = attr_name;
+ attr->dev_attr.attr.mode = S_IRUGO;
+ attr->dev_attr.show = rd_ptr;
+ attr->index = index;
+ attr->nr = nr - 1;
+ err = device_create_file(dev, &attr->dev_attr);
+ return err;
+}
+
+static int create_core_attrs(struct core_data *tdata,
+ struct device *dev, int attr_no)
+{
+ int err, d, i;
static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
struct device_attribute *devattr, char *buf) = {
show_label, show_crit_alarm, show_temp, show_tjmax,
@@ -389,24 +565,51 @@ static int create_core_attrs(struct core_data *tdata, struct device *dev,
"temp%d_label", "temp%d_crit_alarm",
"temp%d_input", "temp%d_crit",
"temp%d_max" };
+ static ssize_t (*const p_rd_ptr[MAX_POWER_ATTRS]) (struct device *dev,
+ struct device_attribute *devattr, char *buf) = {
+ show_energy_label, show_energy,
+ show_power_label, show_power, show_power_max,
+ show_power_cap };
+ static const char *const power_names[MAX_POWER_ATTRS] = {
+ "energy%d_label", "energy%d_input",
+ "power%d_label", "power%d_input",
+ "power%d_max", "power%d_cap" };

for (i = 0; i < tdata->attr_size; i++) {
- snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH, names[i],
- attr_no);
- sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
- tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
- tdata->sd_attrs[i].dev_attr.attr.mode = S_IRUGO;
- tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
- tdata->sd_attrs[i].index = attr_no;
- err = device_create_file(dev, &tdata->sd_attrs[i].dev_attr);
+ err = create_core_attr(dev, &tdata->sd_attrs[i],
+ tdata->attr_name[i],
+ attr_no, attr_no,
+ rd_ptr[i],
+ names[i]);
if (err)
goto exit_free;
}
+ if (tdata->is_pkg_data) {
+ for (d = 0; d < RAPL_DOMAINS; d++) {
+ if (!(tdata->rapl & (1 << d)))
+ continue;
+ for (i = 0; i < MAX_POWER_ATTRS; i++) {
+ int index = d * MAX_POWER_ATTRS + i;
+
+ if (!(tdata->rapl_attr_mask[d] & (1 << i)))
+ continue;
+
+ err = create_core_attr(dev,
+ &tdata->sd_power_attrs[index],
+ tdata->power_attr_name[index],
+ attr_no, d + 1,
+ p_rd_ptr[i],
+ power_names[i]);
+ if (err)
+ goto exit_free;
+ }
+ }
+ }
+
return 0;

exit_free:
- while (--i >= 0)
- device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
+ remove_core_files(dev, tdata);
return err;
}

@@ -462,8 +665,83 @@ static struct core_data *init_core_data(unsigned int cpu, bool pkg_flag)
return tdata;
}

-static int create_core_data(struct platform_device *pdev, unsigned int cpu,
- bool pkg_flag)
+static void coretemp_rapl_work(struct work_struct *work)
+{
+ struct core_data *tdata = container_of(work, struct core_data,
+ rapl_wq.work);
+ u32 eax, edx;
+ u32 delta;
+ u32 power;
+ int d;
+ unsigned long elapsed = jiffies - tdata->rapl_timestamp;
+
+ for (d = 0; d < RAPL_DOMAINS; d++) {
+ if (!(tdata->rapl & (1 << d)))
+ continue;
+
+ rdmsr_on_cpu(tdata->cpu, energy_status_msr[d], &eax, &edx);
+ delta = eax - tdata->rapl_energy_raw[d];
+ power = (delta * 1000LL) >> tdata->rapl_energy_units;
+ tdata->rapl_energy_raw[d] = eax;
+ tdata->rapl_energy[d] += power;
+ if (elapsed)
+ power = DIV_ROUND_CLOSEST(power * HZ, elapsed);
+ tdata->rapl_power[d] = power;
+ }
+
+ tdata->rapl_timestamp = jiffies;
+ schedule_delayed_work(&tdata->rapl_wq, HZ);
+}
+
+static void coretemp_init_rapl(struct platform_device *pdev,
+ int cpu, struct core_data *tdata)
+{
+ u32 eax, edx;
+ int d, err;
+
+ /* Test if we can access rapl registers */
+ err = rdmsr_safe_on_cpu(cpu, MSR_RAPL_POWER_UNIT, &eax, &edx);
+ if (err)
+ return;
+
+ tdata->rapl_power_units = eax & 0x000f;
+ tdata->rapl_energy_units = (eax >> 8) & 0x001f;
+
+ INIT_DEFERRABLE_WORK(&tdata->rapl_wq, coretemp_rapl_work);
+
+ tdata->rapl_timestamp = jiffies;
+
+ for (d = 0; d < RAPL_DOMAINS; d++) {
+ err = rdmsr_safe_on_cpu(cpu, energy_status_msr[d], &eax, &edx);
+ if (err)
+ continue;
+ tdata->rapl |= 1 << d;
+ tdata->rapl_energy_raw[d] = eax;
+ tdata->rapl_attr_mask[d] = BIT(0) | BIT(1) | BIT(2) | BIT(3);
+
+ /*
+ * Only report power cap if supported for domain and enabled.
+ * Note: package domain (index 0) has two cap limits.
+ */
+ err = rdmsr_safe_on_cpu(tdata->cpu, power_limit_msr[d],
+ &eax, &edx);
+ if (!err && ((eax & 0x8000) || (d && (edx & 0x8000))))
+ tdata->rapl_attr_mask[d] |= BIT(5);
+
+ /* Only report max power if it exists for the domain */
+ if (!power_info_msr[d])
+ continue;
+ err = rdmsr_safe_on_cpu(cpu, power_info_msr[d], &eax, &edx);
+ if (err)
+ continue;
+ tdata->rapl_power_max[d] =
+ ((eax & 0x7fff) * 1000) >> tdata->rapl_power_units;
+ tdata->rapl_attr_mask[d] |= BIT(4);
+ }
+}
+
+static int create_core_data(struct platform_device *pdev,
+ unsigned int cpu, bool pkg_flag)
{
struct core_data *tdata;
struct platform_data *pdata = platform_get_drvdata(pdev);
@@ -519,6 +797,9 @@ static int create_core_data(struct platform_device *pdev, unsigned int cpu,
}
}

+ if (tdata->is_pkg_data)
+ coretemp_init_rapl(pdev, cpu, tdata);
+
pdata->core_data[attr_no] = tdata;

/* Create sysfs interfaces */
@@ -526,6 +807,9 @@ static int create_core_data(struct platform_device *pdev, unsigned int cpu,
if (err)
goto exit_free;

+ if (tdata->rapl)
+ schedule_delayed_work(&tdata->rapl_wq, HZ);
+
return 0;
exit_free:
pdata->core_data[attr_no] = NULL;
@@ -549,12 +833,12 @@ static void coretemp_add_core(unsigned int cpu, bool pkg_flag)
static void coretemp_remove_core(struct platform_data *pdata,
struct device *dev, int indx)
{
- int i;
struct core_data *tdata = pdata->core_data[indx];

- /* Remove the sysfs attributes */
- for (i = 0; i < tdata->attr_size; i++)
- device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
+ if (tdata->rapl)
+ cancel_delayed_work_sync(&tdata->rapl_wq);
+
+ remove_core_files(dev, tdata);

kfree(pdata->core_data[indx]);
pdata->core_data[indx] = NULL;
@@ -607,10 +891,60 @@ static int coretemp_remove(struct platform_device *pdev)
return 0;
}

+#ifdef CONFIG_PM_SLEEP
+static int coretemp_suspend(struct device *dev)
+{
+ struct platform_data *pdata = dev_get_drvdata(dev);
+ struct core_data *tdata;
+ int i;
+
+ for (i = 0; i < MAX_CORE_DATA; i++) {
+ tdata = pdata->core_data[i];
+ if (tdata && tdata->rapl)
+ cancel_delayed_work_sync(&tdata->rapl_wq);
+ }
+ return 0;
+}
+
+static int coretemp_resume(struct device *dev)
+{
+ struct platform_data *pdata = dev_get_drvdata(dev);
+ struct core_data *tdata;
+ int d, i;
+ u32 eax, edx;
+
+ for (i = 0; i < MAX_CORE_DATA; i++) {
+ tdata = pdata->core_data[i];
+ if (tdata && tdata->rapl) {
+ tdata->rapl_timestamp = jiffies;
+ for (d = 0; d < RAPL_DOMAINS; d++) {
+ if (!(tdata->rapl & (1 << d)))
+ continue;
+
+ rdmsr_on_cpu(tdata->cpu,
+ energy_status_msr[d],
+ &eax, &edx);
+ tdata->rapl_energy_raw[d] = eax;
+ }
+ schedule_delayed_work(&tdata->rapl_wq, HZ);
+ }
+ }
+ return 0;
+}
+
+static SIMPLE_DEV_PM_OPS(coretemp_dev_pm_ops, coretemp_suspend,
+ coretemp_resume);
+
+#define CORETEMP_DEV_PM_OPS (&coretemp_dev_pm_ops)
+#else
+#define CORETEMP_DEV_PM_OPS NULL
+#endif /* CONFIG_PM_SLEEP */
+
static struct platform_driver coretemp_driver = {
.driver = {
.owner = THIS_MODULE,
.name = DRVNAME,
+ .pm = CORETEMP_DEV_PM_OPS,
},
.probe = coretemp_probe,
.remove = coretemp_remove,
--
1.7.9.7

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/