[SWDEV-520665] Add support for board voltage (#303)

* Add the API and CLI to show the board voltage. --------- Change-Id: Icb25bd653bb1d004704b5a21b378ca31b2b242c7 Signed-off-by: Arif, Maisam <Maisam.Arif@amd.com> Signed-off-by: AL Musaffar, Yazen <Yazen.ALMusaffar@amd.com> [ROCm/amdsmi commit: 970560fc7c]
2025-05-29 19:55:08 -04:00
@@ -229,6 +229,14 @@ Full documentation for amd_smi_lib is available at [https://rocm.docs.amd.com/pr
  - Increasing available JPEG engines to 40.  
  Current ASICs may not support all 40. These will be indicated as UINT16_MAX or N/A in CLI.

+- **Added support to get GPU voltage**.
+    ```shell
+        $ amd-smi metric --voltage
+            GPU: 0
+                VOLTAGE:
+                    VDDBOARD: 52536 mV
+                    ...
+    ```
 - **Added bad page threshold count**.  
  - Added `amdsmi_get_gpu_bad_page_threshold` to Python API and CLI; root/sudo permissions required to display the count.

@@ -1383,7 +1383,7 @@ class AMDSMICommands():
                usage=None, watch=None, watch_time=None, iterations=None, power=None,
                clock=None, temperature=None, ecc=None, ecc_blocks=None, pcie=None,
                fan=None, voltage_curve=None, overdrive=None, perf_level=None,
-                xgmi_err=None, energy=None, mem_usage=None, schedule=None,
+                xgmi_err=None, energy=None, mem_usage=None, voltage=None, schedule=None,
                guard=None, guest_data=None, fb_usage=None, xgmi=None, throttle=None,
                ):
        """Get Metric information for target gpu
@@ -1410,6 +1410,7 @@ class AMDSMICommands():
            xgmi_err (bool, optional): Value override for args.xgmi_err. Defaults to None.
            energy (bool, optional): Value override for args.energy. Defaults to None.
            mem_usage (bool, optional): Value override for args.mem_usage. Defaults to None.
+            voltage (bool, optional): Value override for args.voltage. Defaults to None.
            schedule (bool, optional): Value override for args.schedule. Defaults to None.
            guard (bool, optional): Value override for args.guard. Defaults to None.
            guest_data (bool, optional): Value override for args.guest_data. Defaults to None.
@@ -1458,9 +1459,9 @@ class AMDSMICommands():
                args.ecc = ecc
            if ecc_blocks:
                args.ecc_blocks = ecc_blocks
-            current_platform_args += ["usage", "power", "clock", "temperature", "pcie", "ecc", "ecc_blocks"]
+            current_platform_args += ["usage", "power", "clock", "temperature", "voltage", "pcie", "ecc", "ecc_blocks"]
            current_platform_values += [args.usage, args.power, args.clock,
-                                        args.temperature, args.pcie]
+                                        args.temperature, args.voltage, args.pcie]
            current_platform_values += [args.ecc, args.ecc_blocks]

        if self.helpers.is_baremetal() and self.helpers.is_linux():
@@ -2385,6 +2386,22 @@ class AMDSMICommands():
                except amdsmi_exception.AmdSmiLibraryException as e:
                    values_dict['xgmi_err'] = "N/A"
                    logging.debug("Failed to get xgmi error status for gpu %s | %s", gpu_id, e.get_error_info())
+        if "voltage" in current_platform_args:
+            if args.voltage:
+                voltage_dict = {}
+                all_voltage = {
+                    "vddboard": amdsmi_interface.AmdSmiVoltageType.VDDBOARD
+                }
+                for volt_type, volt_metric in all_voltage.items():
+                    try:
+                        voltage = amdsmi_interface.amdsmi_get_gpu_volt_metric(args.gpu, volt_metric, amdsmi_interface.AmdSmiVoltageMetric.CURRENT)
+                        if voltage == 0:
+                            voltage = "N/A"
+                        voltage_dict[volt_type] = self.helpers.unit_format(self.logger, voltage, "mV")
+                    except amdsmi_exception.AmdSmiLibraryException as e:
+                        voltage_dict[volt_type] = "N/A"
+                        logging.debug("Failed to get voltage for gpu %s | %s", gpu_id, e.get_error_info())
+                values_dict['voltage'] = voltage_dict
        if "energy" in current_platform_args:
            if args.energy:
                try:
@@ -2991,7 +3008,7 @@ class AMDSMICommands():
                usage=None, watch=None, watch_time=None, iterations=None, power=None,
                clock=None, temperature=None, ecc=None, ecc_blocks=None, pcie=None,
                fan=None, voltage_curve=None, overdrive=None, perf_level=None,
-                xgmi_err=None, energy=None, mem_usage=None, schedule=None,
+                xgmi_err=None, energy=None, mem_usage=None, voltage=None, schedule=None,
                guard=None, guest_data=None, fb_usage=None, xgmi=None,
                cpu=None, cpu_power_metrics=None, cpu_prochot=None, cpu_freq_metrics=None,
                cpu_c0_res=None, cpu_lclk_dpm_level=None, cpu_pwr_svi_telemetry_rails=None,
@@ -3025,6 +3042,7 @@ class AMDSMICommands():
            xgmi_err (bool, optional): Value override for args.xgmi_err. Defaults to None.
            energy (bool, optional): Value override for args.energy. Defaults to None.
            mem_usage (bool, optional): Value override for args.mem_usage. Defaults to None.
+            voltage (bool, optional): Value override for args.voltage. Defaults to None.
            schedule (bool, optional): Value override for args.schedule. Defaults to None.
            guard (bool, optional): Value override for args.guard. Defaults to None.
            guest_data (bool, optional): Value override for args.guest_data. Defaults to None.
@@ -3073,7 +3091,7 @@ class AMDSMICommands():
        gpu_args_enabled = False
        gpu_attributes = ["usage", "watch", "watch_time", "iterations", "power", "clock",
                          "temperature", "ecc", "ecc_blocks", "pcie", "fan", "voltage_curve",
-                          "overdrive", "perf_level", "xgmi_err", "energy", "mem_usage", "schedule",
+                          "overdrive", "perf_level", "xgmi_err", "energy", "mem_usage", "voltage", "schedule",
                          "guard", "guest_data", "fb_usage", "xgmi", "throttle"]
        for attr in gpu_attributes:
            if hasattr(args, attr):
@@ -3146,7 +3164,7 @@ class AMDSMICommands():
                                usage, watch, watch_time, iterations, power,
                                clock, temperature, ecc, ecc_blocks, pcie,
                                fan, voltage_curve, overdrive, perf_level,
-                                xgmi_err, energy, mem_usage, schedule,
+                                xgmi_err, energy, mem_usage, voltage, schedule,
                                guard, guest_data, fb_usage, xgmi, throttle,
                                )
        elif self.helpers.is_amd_hsmp_initialized(): # Only CPU is initialized
@@ -3182,7 +3200,7 @@ class AMDSMICommands():
                                usage, watch, watch_time, iterations, power,
                                clock, temperature, ecc, ecc_blocks, pcie,
                                fan, voltage_curve, overdrive, perf_level,
-                                xgmi_err, energy, mem_usage, schedule, throttle,
+                                xgmi_err, energy, mem_usage, voltage, schedule, throttle,
                                )
        if self.logger.is_json_format():
            self.logger.combine_arrays_to_json()
@@ -874,6 +874,7 @@ class AMDSMIParser(argparse.ArgumentParser):
        ecc_help = "Total number of ECC errors"
        ecc_blocks_help = "Number of ECC errors per block"
        pcie_help = "Current PCIe speed, width, and replay count"
+        voltage_help = "GPU voltage"

        # Help text for Arguments only on Linux Baremetal platforms
        fan_help = "Current fan speed"
@@ -939,6 +940,7 @@ class AMDSMIParser(argparse.ArgumentParser):
                metric_parser.add_argument('-P', '--pcie', action='store_true', required=False, help=pcie_help)
                metric_parser.add_argument('-e', '--ecc', action='store_true', required=False, help=ecc_help)
                metric_parser.add_argument('-k', '--ecc-blocks', action='store_true', required=False, help=ecc_blocks_help)
+                metric_parser.add_argument('-V', '--voltage', action='store_true', required=False, help=voltage_help)

            # Options that only apply to Hypervisors and Baremetal Linux
            if self.helpers.is_hypervisor() or (self.helpers.is_baremetal() and self.helpers.is_linux()):
@@ -313,6 +313,7 @@ Metric arguments:
  -P, --pcie                                Current PCIe speed, width, and replay count
  -e, --ecc                                 Total number of ECC errors
  -k, --ecc-blocks                          Number of ECC errors per block
+  -V, --voltage                             GPU voltage
  -f, --fan                                 Current fan speed
  -C, --voltage-curve                       Display voltage curve
  -o, --overdrive                           Current GPU clock overdrive and GPU memory clock overdrive level
@@ -2112,16 +2112,17 @@ machine guest

 Input parameters:

-* `processor_handle` handle for the given device
-* `sensor_type` part of device from which voltage should be obtained
-* `metric` enum indicated which voltage value should be retrieved
+Parameters | Description
+---|---
+`processor_handle` |  Handle for the given device
+`sensor_type` | <table><thead><tr><th> Possible Values </th><th> Description </th></tr></thead><tbody><tr><td>`AmdSmiVoltageType.VDDGFX`</td><td>Represents the voltage supplied to the GPU's graphics core.</td></tr><tr><td>`AmdSmiVoltageType.VDDBOARD`</td><td>Represents the voltage supplied to the entire GPU board, including auxiliary components. Intended for Mi300+</td></tr></tbody></table>
+`metric` | <table><thead><tr><th> Possible Values </th><th> Description </th></tr></thead><tbody><tr><td>`AmdSmiVoltageMetric.CURRENT`</td><td>Represents the current voltage value measured at the specified sensor.</td></tr><tr><td>`AmdSmiVoltageMetric.MAX`</td><td>Represents the maximum voltage value recorded at the specified sensor.</td></tr><tr><td>`AmdSmiVoltageMetric.MIN`</td><td>Represents the minimum voltage value recorded at the specified sensor.</td></tr><tr><td>`AmdSmiVoltageMetric.AVERAGE`</td><td>Represents the average voltage value calculated over a period of time at the specified sensor.</td></tr><tr><td>`AmdSmiVoltageMetric.MAX_CRIT`</td><td>Represents the critical maximum voltage value that should not be exceeded.</td></tr><tr><td>`AmdSmiVoltageMetric.MIN_CRIT`</td><td>Represents the critical minimum voltage value that should not be dropped below.</td></tr><tr><td>`AmdSmiVoltageMetric.LOWEST`</td><td>Represents the lowest voltage value recorded during the monitoring period.</td></tr><tr><td>`AmdSmiVoltageMetric.HIGHEST`</td><td>Represents the highest voltage value recorded during the monitoring period.</td></tr></tbody></table>

 Output: Voltage as integer in millivolts

 Exceptions that can be thrown by `amdsmi_get_gpu_volt_metric` function:

 * `AmdSmiLibraryException`
-* `AmdSmiRetryException`
 * `AmdSmiParameterException`

 Example:
@@ -2133,8 +2134,11 @@ try:
        print("No GPUs on machine")
    else:
        for device in devices:
-            voltage =  amdsmi_get_gpu_volt_metric(device, AmdSmiVoltageType.VDDGFX,
-                        AmdSmiVoltageMetric.AVERAGE)
+            voltage = amdsmi_get_gpu_volt_metric(
+                device,
+                AmdSmiVoltageType.VDDBOARD,
+                AmdSmiVoltageMetric.AVERAGE
+            )
            print(voltage)
 except AmdSmiException as e:
    print(e)
@@ -2692,7 +2696,7 @@ except AmdSmiException as e:

 ### amdsmi_get_gpu_power_profile_presets

-Description:  Get the list of available preset power profiles and an indication of
+Description: Get the list of available preset power profiles and an indication of
 which profile is currently active. It is not supported on virtual machine guest

 Input parameters:
@@ -1360,7 +1360,8 @@ typedef enum {
    AMDSMI_VOLT_TYPE_FIRST = 0,

    AMDSMI_VOLT_TYPE_VDDGFX = AMDSMI_VOLT_TYPE_FIRST,  //!< Vddgfx GPU voltage
-    AMDSMI_VOLT_TYPE_LAST = AMDSMI_VOLT_TYPE_VDDGFX,
+    AMDSMI_VOLT_TYPE_VDDBOARD,                        //!< Voltage for VDDBOARD
+    AMDSMI_VOLT_TYPE_LAST = AMDSMI_VOLT_TYPE_VDDBOARD,
    AMDSMI_VOLT_TYPE_INVALID = 0xFFFFFFFF              //!< Invalid type
 } amdsmi_voltage_type_t;

@@ -309,6 +309,7 @@ class AmdSmiVoltageMetric(IntEnum):

 class AmdSmiVoltageType(IntEnum):
    VDDGFX = amdsmi_wrapper.AMDSMI_VOLT_TYPE_VDDGFX
+    VDDBOARD = amdsmi_wrapper.AMDSMI_VOLT_TYPE_VDDBOARD
    INVALID = amdsmi_wrapper.AMDSMI_VOLT_TYPE_INVALID

 class AmdSmiAcceleratorPartitionResourceType(IntEnum):
@@ -1530,12 +1530,14 @@ amdsmi_voltage_metric_t = ctypes.c_uint32 # enum
 amdsmi_voltage_type_t__enumvalues = {
    0: 'AMDSMI_VOLT_TYPE_FIRST',
    0: 'AMDSMI_VOLT_TYPE_VDDGFX',
-    0: 'AMDSMI_VOLT_TYPE_LAST',
+    1: 'AMDSMI_VOLT_TYPE_VDDBOARD',
+    1: 'AMDSMI_VOLT_TYPE_LAST',
    4294967295: 'AMDSMI_VOLT_TYPE_INVALID',
 }
 AMDSMI_VOLT_TYPE_FIRST = 0
 AMDSMI_VOLT_TYPE_VDDGFX = 0
-AMDSMI_VOLT_TYPE_LAST = 0
+AMDSMI_VOLT_TYPE_VDDBOARD = 1
+AMDSMI_VOLT_TYPE_LAST = 1
 AMDSMI_VOLT_TYPE_INVALID = 4294967295
 amdsmi_voltage_type_t = ctypes.c_uint32 # enum

@@ -3161,6 +3163,7 @@ __all__ = \
    'AMDSMI_VOLT_MAX_CRIT', 'AMDSMI_VOLT_MIN', 'AMDSMI_VOLT_MIN_CRIT',
    'AMDSMI_VOLT_TYPE_FIRST', 'AMDSMI_VOLT_TYPE_INVALID',
    'AMDSMI_VOLT_TYPE_LAST', 'AMDSMI_VOLT_TYPE_VDDGFX',
+    'AMDSMI_VOLT_TYPE_VDDBOARD',
    'AMDSMI_VRAM_TYPE_DDR2', 'AMDSMI_VRAM_TYPE_DDR3',
    'AMDSMI_VRAM_TYPE_DDR4', 'AMDSMI_VRAM_TYPE_GDDR1',
    'AMDSMI_VRAM_TYPE_GDDR2', 'AMDSMI_VRAM_TYPE_GDDR3',
@@ -574,9 +574,10 @@ typedef enum {
 typedef enum {
  RSMI_VOLT_TYPE_FIRST = 0,

-  RSMI_VOLT_TYPE_VDDGFX = RSMI_VOLT_TYPE_FIRST,  //!< Vddgfx GPU
-                                                 //!< voltage
-  RSMI_VOLT_TYPE_LAST = RSMI_VOLT_TYPE_VDDGFX,
+  RSMI_VOLT_TYPE_VDDGFX = RSMI_VOLT_TYPE_FIRST,  //!< Vddgfx GPU voltage
+  RSMI_VOLT_TYPE_VDDBOARD,                       //!< Voltage for VDDBOARD
+
+  RSMI_VOLT_TYPE_LAST = RSMI_VOLT_TYPE_VDDBOARD,
  RSMI_VOLT_TYPE_INVALID = 0xFFFFFFFF            //!< Invalid type
 } rsmi_voltage_type_t;

@@ -88,7 +88,7 @@ static const char *kTempSensorTypeJunctionName = "junction";
 static const char *kTempSensorTypeEdgeName = "edge";

 static const char *kTempSensorTypeVddgfxName = "vddgfx";
-
+static const char *kTempSensorTypeVddboardName = "vddboard";

 static const std::map<std::string, rsmi_temperature_type_t>
                                                        kTempSensorNameMap = {
@@ -100,6 +100,7 @@ static const std::map<std::string, rsmi_temperature_type_t>
 static const std::map<std::string, rsmi_voltage_type_t>
                                                        kVoltSensorNameMap = {
    {kTempSensorTypeVddgfxName, RSMI_VOLT_TYPE_VDDGFX},
+    {kTempSensorTypeVddboardName, RSMI_VOLT_TYPE_VDDBOARD},
 };

 static const std::map<MonitorTypes, const char *> kMonitorNameMap = {
@@ -378,7 +379,13 @@ Monitor::setVoltSensorLabelMap(void) {
  };

  for (uint32_t i = 0; i < RSMI_VOLT_TYPE_LAST + 1; ++i) {
-    ret = add_volt_sensor_entry(i);
+    // VDDGFX -> 0, VDDNB -> 1, VDDBOARD -> 2
+    // Here the VDDNB will be skipped as it is not defined in the enum and not supported by AMD.
+    auto file_index = i;
+    if (i >= RSMI_VOLT_TYPE_VDDBOARD) {
+      file_index = i + 1;
+    }
+    ret = add_volt_sensor_entry(file_index);
    if (ret) {
      return ret;
    }
@@ -123,6 +123,7 @@ static const char* short_options = "i:v:m:fr";

 static const std::map<uint32_t, std::string> kVoltSensorNameMap = {
    {AMDSMI_VOLT_TYPE_VDDGFX, "Vddgfx"},
+    {AMDSMI_VOLT_TYPE_VDDBOARD, "Vddboard"},
 };

 static void PrintHelp(void) {