drm/xe/hwmon: Expose hwmon energy attribute
authorBadal Nilawar <badal.nilawar@intel.com>
Mon, 25 Sep 2023 08:18:41 +0000 (13:48 +0530)
committerRodrigo Vivi <rodrigo.vivi@intel.com>
Thu, 21 Dec 2023 16:42:08 +0000 (11:42 -0500)
Expose hwmon energy attribute to show device level energy usage

v2:
  - %s/hwm_/hwmon_/
  - Convert enums to upper case
v3:
  - %s/hwmon_/xe_hwmon
  - Remove gt specific hwmon attributes
v4:
 - %s/REG_PKG_ENERGY_STATUS/REG_ENERGY_STATUS_ALL (Riana)
 - %s/hwmon_energy_info/xe_hwmon_energy_info (Riana)

Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Riana Tauro <riana.tauro@intel.com>
Signed-off-by: Badal Nilawar <badal.nilawar@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://lore.kernel.org/r/20230925081842.3566834-5-badal.nilawar@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Documentation/ABI/testing/sysfs-driver-intel-xe-hwmon
drivers/gpu/drm/xe/regs/xe_gt_regs.h
drivers/gpu/drm/xe/regs/xe_mchbar_regs.h
drivers/gpu/drm/xe/xe_hwmon.c

index 7f9407c20864612ba7c3493420dddea2266ad3a2..1a7a6c23e14170c681449e192dd5e87660a65d38 100644 (file)
@@ -52,3 +52,10 @@ Description: RO. Current Voltage in millivolt.
 
                Only supported for particular Intel xe graphics platforms.
 
+What:          /sys/devices/.../hwmon/hwmon<i>/energy1_input
+Date:          September 2023
+KernelVersion: 6.5
+Contact:       intel-xe@lists.freedesktop.org
+Description:   RO. Energy input of device in microjoules.
+
+               Only supported for particular Intel xe graphics platforms.
index a9a91195c6d5655d420472bc2490679ce7abc6c4..83519a424aab1c6200a3497a06c7121d2d3846b0 100644 (file)
 #define XEHPC_BCS5_BCS6_INTR_MASK              XE_REG(0x190118)
 #define XEHPC_BCS7_BCS8_INTR_MASK              XE_REG(0x19011c)
 
+#define PVC_GT0_PACKAGE_ENERGY_STATUS          XE_REG(0x281004)
 #define PVC_GT0_PACKAGE_RAPL_LIMIT             XE_REG(0x281008)
 #define PVC_GT0_PACKAGE_POWER_SKU_UNIT         XE_REG(0x281068)
+#define PVC_GT0_PLATFORM_ENERGY_STATUS         XE_REG(0x28106c)
 #define PVC_GT0_PACKAGE_POWER_SKU              XE_REG(0x281080)
 
 #endif
index 27f1d42baf6d5a561ba7ebcda1bb292d7f1a6fb1..d8ecbe1858d122f4bcf44d66e72060de14f0f4c1 100644 (file)
@@ -25,6 +25,9 @@
 
 #define PCU_CR_PACKAGE_POWER_SKU_UNIT          XE_REG(MCHBAR_MIRROR_BASE_SNB + 0x5938)
 #define   PKG_PWR_UNIT                         REG_GENMASK(3, 0)
+#define   PKG_ENERGY_UNIT                      REG_GENMASK(12, 8)
+
+#define PCU_CR_PACKAGE_ENERGY_STATUS           XE_REG(MCHBAR_MIRROR_BASE_SNB + 0x593c)
 
 #define PCU_CR_PACKAGE_RAPL_LIMIT              XE_REG(MCHBAR_MIRROR_BASE_SNB + 0x59a0)
 #define   PKG_PWR_LIM_1                                REG_GENMASK(14, 0)
index d89345b85e76b6c8ab07d7df417c0860b8a765b4..734fcca9f71fab4a07e64f3834c26d27943eacd7 100644 (file)
@@ -22,6 +22,7 @@ enum xe_hwmon_reg {
        REG_PKG_POWER_SKU,
        REG_PKG_POWER_SKU_UNIT,
        REG_GT_PERF_STATUS,
+       REG_PKG_ENERGY_STATUS,
 };
 
 enum xe_hwmon_reg_operation {
@@ -36,12 +37,20 @@ enum xe_hwmon_reg_operation {
 #define SF_POWER       1000000         /* microwatts */
 #define SF_CURR                1000            /* milliamperes */
 #define SF_VOLTAGE     1000            /* millivolts */
+#define SF_ENERGY      1000000         /* microjoules */
+
+struct xe_hwmon_energy_info {
+       u32 reg_val_prev;
+       long accum_energy;              /* Accumulated energy for energy1_input */
+};
 
 struct xe_hwmon {
        struct device *hwmon_dev;
        struct xe_gt *gt;
        struct mutex hwmon_lock; /* rmw operations*/
        int scl_shift_power;
+       int scl_shift_energy;
+       struct xe_hwmon_energy_info ei; /*  Energy info for energy1_input */
 };
 
 static u32 xe_hwmon_get_reg(struct xe_hwmon *hwmon, enum xe_hwmon_reg hwmon_reg)
@@ -72,6 +81,12 @@ static u32 xe_hwmon_get_reg(struct xe_hwmon *hwmon, enum xe_hwmon_reg hwmon_reg)
                if (xe->info.platform == XE_DG2)
                        reg = GT_PERF_STATUS;
                break;
+       case REG_PKG_ENERGY_STATUS:
+               if (xe->info.platform == XE_DG2)
+                       reg = PCU_CR_PACKAGE_ENERGY_STATUS;
+               else if (xe->info.platform == XE_PVC)
+                       reg = PVC_GT0_PLATFORM_ENERGY_STATUS;
+               break;
        default:
                drm_warn(&xe->drm, "Unknown xe hwmon reg id: %d\n", hwmon_reg);
                break;
@@ -194,10 +209,59 @@ static int xe_hwmon_power_rated_max_read(struct xe_hwmon *hwmon, long *value)
        return 0;
 }
 
+/*
+ * xe_hwmon_energy_get - Obtain energy value
+ *
+ * The underlying energy hardware register is 32-bits and is subject to
+ * overflow. How long before overflow? For example, with an example
+ * scaling bit shift of 14 bits (see register *PACKAGE_POWER_SKU_UNIT) and
+ * a power draw of 1000 watts, the 32-bit counter will overflow in
+ * approximately 4.36 minutes.
+ *
+ * Examples:
+ *    1 watt:  (2^32 >> 14) /    1 W / (60 * 60 * 24) secs/day -> 3 days
+ * 1000 watts: (2^32 >> 14) / 1000 W / 60             secs/min -> 4.36 minutes
+ *
+ * The function significantly increases overflow duration (from 4.36
+ * minutes) by accumulating the energy register into a 'long' as allowed by
+ * the hwmon API. Using x86_64 128 bit arithmetic (see mul_u64_u32_shr()),
+ * a 'long' of 63 bits, SF_ENERGY of 1e6 (~20 bits) and
+ * hwmon->scl_shift_energy of 14 bits we have 57 (63 - 20 + 14) bits before
+ * energy1_input overflows. This at 1000 W is an overflow duration of 278 years.
+ */
+static void
+xe_hwmon_energy_get(struct xe_hwmon *hwmon, long *energy)
+{
+       struct xe_hwmon_energy_info *ei = &hwmon->ei;
+       u32 reg_val;
+
+       xe_device_mem_access_get(gt_to_xe(hwmon->gt));
+
+       mutex_lock(&hwmon->hwmon_lock);
+
+       xe_hwmon_process_reg(hwmon, REG_PKG_ENERGY_STATUS, REG_READ,
+                            &reg_val, 0, 0);
+
+       if (reg_val >= ei->reg_val_prev)
+               ei->accum_energy += reg_val - ei->reg_val_prev;
+       else
+               ei->accum_energy += UINT_MAX - ei->reg_val_prev + reg_val;
+
+       ei->reg_val_prev = reg_val;
+
+       *energy = mul_u64_u32_shr(ei->accum_energy, SF_ENERGY,
+                                 hwmon->scl_shift_energy);
+
+       mutex_unlock(&hwmon->hwmon_lock);
+
+       xe_device_mem_access_put(gt_to_xe(hwmon->gt));
+}
+
 static const struct hwmon_channel_info *hwmon_info[] = {
        HWMON_CHANNEL_INFO(power, HWMON_P_MAX | HWMON_P_RATED_MAX | HWMON_P_CRIT),
        HWMON_CHANNEL_INFO(curr, HWMON_C_CRIT),
        HWMON_CHANNEL_INFO(in, HWMON_I_INPUT),
+       HWMON_CHANNEL_INFO(energy, HWMON_E_INPUT),
        NULL
 };
 
@@ -371,6 +435,29 @@ xe_hwmon_in_read(struct xe_hwmon *hwmon, u32 attr, long *val)
        return ret;
 }
 
+static umode_t
+xe_hwmon_energy_is_visible(struct xe_hwmon *hwmon, u32 attr)
+{
+       switch (attr) {
+       case hwmon_energy_input:
+               return xe_hwmon_get_reg(hwmon, REG_PKG_ENERGY_STATUS) ? 0444 : 0;
+       default:
+               return 0;
+       }
+}
+
+static int
+xe_hwmon_energy_read(struct xe_hwmon *hwmon, u32 attr, long *val)
+{
+       switch (attr) {
+       case hwmon_energy_input:
+               xe_hwmon_energy_get(hwmon, val);
+               return 0;
+       default:
+               return -EOPNOTSUPP;
+       }
+}
+
 static umode_t
 xe_hwmon_is_visible(const void *drvdata, enum hwmon_sensor_types type,
                    u32 attr, int channel)
@@ -390,6 +477,9 @@ xe_hwmon_is_visible(const void *drvdata, enum hwmon_sensor_types type,
        case hwmon_in:
                ret = xe_hwmon_in_is_visible(hwmon, attr);
                break;
+       case hwmon_energy:
+               ret = xe_hwmon_energy_is_visible(hwmon, attr);
+               break;
        default:
                ret = 0;
                break;
@@ -419,6 +509,9 @@ xe_hwmon_read(struct device *dev, enum hwmon_sensor_types type, u32 attr,
        case hwmon_in:
                ret = xe_hwmon_in_read(hwmon, attr, val);
                break;
+       case hwmon_energy:
+               ret = xe_hwmon_energy_read(hwmon, attr, val);
+               break;
        default:
                ret = -EOPNOTSUPP;
                break;
@@ -470,6 +563,7 @@ static void
 xe_hwmon_get_preregistration_info(struct xe_device *xe)
 {
        struct xe_hwmon *hwmon = xe->hwmon;
+       long energy;
        u32 val_sku_unit = 0;
        int ret;
 
@@ -478,8 +572,17 @@ xe_hwmon_get_preregistration_info(struct xe_device *xe)
         * The contents of register PKG_POWER_SKU_UNIT do not change,
         * so read it once and store the shift values.
         */
-       if (!ret)
+       if (!ret) {
                hwmon->scl_shift_power = REG_FIELD_GET(PKG_PWR_UNIT, val_sku_unit);
+               hwmon->scl_shift_energy = REG_FIELD_GET(PKG_ENERGY_UNIT, val_sku_unit);
+       }
+
+       /*
+        * Initialize 'struct xe_hwmon_energy_info', i.e. set fields to the
+        * first value of the energy register read
+        */
+       if (xe_hwmon_is_visible(hwmon, hwmon_energy, hwmon_energy_input, 0))
+               xe_hwmon_energy_get(hwmon, &energy);
 }
 
 void xe_hwmon_register(struct xe_device *xe)