AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

OAM: Add get dev and pci properties and sensor count

Browse Source 
Also, add amdoam_get_error_description.

On behalf of
Amber Lin <Amber.Lin@amd.com> and
Divya Shikre <DivyaUday.Shikre@amd.com>

Change-Id: I1f5ac0c5948adb2c30008e95c501e8b69b8183b6


[ROCm/amdsmi commit: 27deaea6e8]
This commit is contained in:
Amber Lin
2020-05-25 17:20:11 -04:00
committed by Chris Freehill
parent 9e57932639
commit 328aa330b2
5 changed files with 509 additions and 299 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/amdsmi/oam/CMakeLists.txt
Executable → Regular
View File
projects/amdsmi/oam/example/oam_example.c
Executable → Regular
+50 -1
View File
@@ -7,14 +7,23 @@ const oam_ops_t amd_oam_ops = {
.free = amdoam_free,
// .get_mapi_version = amdoam_get_mapi_version,
.discover_devices = amdoam_discover_devices,
.get_dev_properties = amdoam_get_dev_properties,
.get_pci_properties = amdoam_get_pci_properties,
.get_sensors_count = amdoam_get_sensors_count,
.get_error_description = amdoam_get_error_description,
};
int main()
{
uint32_t dev_cnt = 0;
oam_mapi_version_t version;
oam_dev_properties_t *devs_prop;
int i;
oam_pci_info_t pci_info;
oam_sensor_count_t sensor_count;
const char *string;
if (amd_oam_ops.init(version)) {
if (amd_oam_ops.init()) {
printf("init failed\n");
return -1;
}
@@ -22,7 +31,47 @@ int main()
// amd_oam_ops.get_mapi_version(&version);
if (!amd_oam_ops.discover_devices(&dev_cnt))
printf("%d AMD devices are discovered\n", dev_cnt);
if (!dev_cnt) {
printf("No devices are found.\n");
return -1;
}
devs_prop = calloc(dev_cnt, sizeof(oam_dev_properties_t));
if (!devs_prop) {
printf("Allocating dev_prop failed\n");
return -1;
}
amd_oam_ops.get_dev_properties(dev_cnt, devs_prop);
for (i = 0; i < dev_cnt; i++) {
printf("Device %d:\n", i);
printf(" device id %d\n", devs_prop[i].device_id);
printf(" device_vendor %s\n", devs_prop[i].device_vendor);
printf(" device_name %s\n", devs_prop[i].device_name);
printf(" sku_name %s\n", devs_prop[i].sku_name);
printf(" board_name %s\n", devs_prop[i].board_name);
printf(" board_revision %s\n", devs_prop[i].board_revision);
printf(" board_serial_number %s\n",
devs_prop[i].board_serial_number);
if (!amd_oam_ops.get_pci_properties(devs_prop[i].device_id, &pci_info)){
printf(" PCI Domain : 0x%d \n", pci_info.domain);
printf(" PCI bus : 0x%d \n", pci_info.bus);
printf(" PCI device : 0x%d \n", pci_info.device);
printf(" PCI function : 0x%d \n", pci_info.function);
}
if(!amd_oam_ops.get_sensors_count(devs_prop[i].device_id, &sensor_count)) {
printf(" Temperature Sensors : %d \n", sensor_count.num_temperature_sensors);
printf(" Power Sensors : %d \n", sensor_count.num_power_sensors);
printf(" Voltage Sensors : %d \n", sensor_count.num_voltage_sensors);
printf(" Current Sensors : %d \n", sensor_count.num_current_sensors);
printf(" Fan Sensors : %d \n", sensor_count.num_fans);
}
}
amd_oam_ops.get_error_description(1, &string);
printf("error code 1: %s\n", string);
free(devs_prop);
amd_oam_ops.free();
return 0;
projects/amdsmi/oam/include/oam/amd_oam.h
+26 -2
View File
@@ -32,10 +32,34 @@ extern "C" {
#include <stdint.h>
#endif // __cplusplus
int amdoam_init(oam_mapi_version_t version);
#include "oam/oam_mapi.h"
typedef enum {
AMDOAM_STATUS_SUCCESS = 0x0,
/* copy RSMI errors */
AMDOAM_STATUS_INVALID_ARGS,
AMDOAM_STATUS_NOT_SUPPORTED,
AMDOAM_STATUS_FILE_ERROR,
AMDOAM_STATUS_PERMISSION,
AMDOAM_STATUS_OUT_OF_RESOURCES,
AMDOAM_STATUS_INTERNAL_EXCEPTION,
AMDOAM_STATUS_INPUT_OUT_OF_BOUNDS,
AMDOAM_STATUS_INIT_ERROR,
/* end of RSMI error code */
AMDOAM_STATUS_ERROR, // Generic error return if not otherwise specified
AMDOAM_STATUS_NOT_FOUND,
} amdoam_status_t;
int amdoam_init(void);
int amdoam_free(void);
// int amdoam_get_mapi_version(oam_mapi_version_t *version);
int amdoam_discover_devices(int *device_count);
int amdoam_discover_devices(uint32_t *device_count);
int amdoam_get_dev_properties(uint32_t dev_inx,
oam_dev_properties_t *prop);
int amdoam_get_pci_properties(uint32_t device_id, oam_pci_info_t *pci_info);
int amdoam_get_sensors_count(uint32_t device_id,
oam_sensor_count_t *sensor_count);
int amdoam_get_error_description(int code, const char **description);
#ifdef __cplusplus
}
projects/amdsmi/oam/include/oam/oam_mapi.h
+261 -271
View File
@@ -42,46 +42,35 @@ extern "C" {
* \struct oam_mapi_version_t
* \brief OAM library API version
* \details TBD
* All the libraries versions are expected to be backward compatible.
* The major version increment indicates a new API has been added.
* Minor version increment indicates an interface change.
* All the libraries versions are expected to be backward compatible.
* The major version increment indicates a new API has been added.
* Minor version increment indicates an interface change.
*/
typedef struct oam_mapi_version {
uint32_t major;
uint32_t minor;
uint32_t major;
uint32_t minor;
} oam_mapi_version_t;
/**
* \struct oam_dev_properties_t
* \brief Local identifier for the device
* \details Immutable device identifier
* This is unique within the chassis.
*/
typedef struct oam_dev_id {
/*!< local identifier for the device */
int device_id;
} oam_dev_id_t;
/**
* \struct oam_dev_properties_t
* \brief Network identifier for the device
* \struct oam_dev_properties_t
* \brief Network identifier for the device
* \details Immutable network identifier for the device.
* This is unique across the entire network.
* This is unique across the entire network.
*/
typedef struct oam_net_dev_id {
/*!< unique network identifier for the device */
int network_id;
/*!< unique network identifier for the device */
int network_id;
} oam_net_dev_id_t;
/*
* various lengths for device properties
*/
#define DEVICE_VENDOR_LEN 128
#define DEVICE_NAME_LEN 128
#define DEVICE_SKU_LEN 128
#define BOARD_NAME_LEN 128
#define BOARD_REVISION_LEN 128
#define BOARD_SERIAL_NUM_LEN 128
#define DEVICE_VENDOR_LEN 128
#define DEVICE_NAME_LEN 128
#define DEVICE_SKU_LEN 128
#define BOARD_NAME_LEN 128
#define BOARD_REVISION_LEN 128
#define BOARD_SERIAL_NUM_LEN 128
/**
* \struct oam_dev_properties_t
@@ -89,23 +78,23 @@ typedef struct oam_net_dev_id {
* \details TBD
*/
typedef struct oam_dev_properties {
/*!< unique network identifier for the device */
oam_dev_id_t device_id;
/*!< vendor name */
char device_vendor[DEVICE_VENDOR_LEN];
/*!< Device name */
char device_name[DEVICE_NAME_LEN];
/*!< SKU name */
char sku_name[DEVICE_SKU_LEN];
/*!< Board name */
char board_name[BOARD_NAME_LEN];
/*!< Board revision */
char board_revision[BOARD_REVISION_LEN];
/*!<
* Board Serial Number or UUID any other identifier, which can be used
* to identify devices uniquely and physically.
*/
char board_serial_number[BOARD_SERIAL_NUM_LEN];
/*!< Immutable local identifier for the device */
uint32_t device_id;
/*!< vendor name */
char device_vendor[DEVICE_VENDOR_LEN];
/*!< Device name */
char device_name[DEVICE_NAME_LEN];
/*!< SKU name */
char sku_name[DEVICE_SKU_LEN];
/*!< Board name */
char board_name[BOARD_NAME_LEN];
/*!< Board revision */
char board_revision[BOARD_REVISION_LEN];
/*!<
* Board Serial Number or UUID any other identifier, which can be used
* to identify devices uniquely and physically.
*/
char board_serial_number[BOARD_SERIAL_NUM_LEN];
} oam_dev_properties_t;
/**
@@ -115,11 +104,11 @@ typedef struct oam_dev_properties {
* Various sensor related information
*/
typedef struct oam_sensor_count {
uint32_t num_temperature_sensors;
uint32_t num_power_sensors;
uint32_t num_voltage_sensors;
uint32_t num_current_sensors;
uint32_t num_fans;
uint32_t num_temperature_sensors;
uint32_t num_power_sensors;
uint32_t num_voltage_sensors;
uint32_t num_current_sensors;
uint32_t num_fans;
} oam_sensor_count_t;
/**
@@ -128,12 +117,12 @@ typedef struct oam_sensor_count {
* \details This enumerated type defines available sensors types.
*/
typedef enum oam_sensor_type {
OAM_SENSOR_TYPE_POWER = 0,
OAM_SENSOR_TYPE_VOLTAGE,
OAM_SENSOR_TYPE_CURRENT,
OAM_SENSOR_TYPE_TEMP,
OAM_SENSOR_TYPE_FAN_SPEED,
OAM_SENSOR_TYPE_UNKNOWN = 0xFF
OAM_SENSOR_TYPE_POWER = 0,
OAM_SENSOR_TYPE_VOLTAGE,
OAM_SENSOR_TYPE_CURRENT,
OAM_SENSOR_TYPE_TEMP,
OAM_SENSOR_TYPE_FAN_SPEED,
OAM_SENSOR_TYPE_UNKNOWN = 0xFF
} oam_sensor_type_t;
/**
@@ -142,9 +131,9 @@ typedef enum oam_sensor_type {
* \details This enumerated type defines available scales for power measurements
*/
typedef enum oam_power_sensor_scale {
OAM_POWER_SCALE_uW = 0,
OAM_POWER_SCALE_mW,
OAM_POWER_SCALE_W,
OAM_POWER_SCALE_uW = 0,
OAM_POWER_SCALE_mW,
OAM_POWER_SCALE_W,
} oam_power_sensor_scale_t;
/**
@@ -153,9 +142,9 @@ typedef enum oam_power_sensor_scale {
* \details This enumerated type defines available scales for voltage measurements
*/
typedef enum oam_voltage_sensor_scale {
OAM_VOLTAGE_SCALE_uV = 0,
OAM_VOLTAGE_SCALE_mV,
OAM_VOLTAGE_SCALE_V,
OAM_VOLTAGE_SCALE_uV = 0,
OAM_VOLTAGE_SCALE_mV,
OAM_VOLTAGE_SCALE_V,
} oam_voltage_sensor_scale_t;
/**
@@ -164,9 +153,9 @@ typedef enum oam_voltage_sensor_scale {
* \details This enumerated type defines available scales for current measurements
*/
typedef enum oam_current_sensor_scale {
OAM_CURRENT_SCALE_uA = 0,
OAM_CURRENT_SCALE_mA,
OAM_CURRENT_SCALE_A,
OAM_CURRENT_SCALE_uA = 0,
OAM_CURRENT_SCALE_mA,
OAM_CURRENT_SCALE_A,
} oam_current_sensor_scale_t;
/**
@@ -175,8 +164,8 @@ typedef enum oam_current_sensor_scale {
* \details This enumerated type defines available scales for temp measurements
*/
typedef enum oam_temp_sensor_scale {
OAM_TEMP_SCALE_C = 0,
OAM_TEMP_SCALE_F
OAM_TEMP_SCALE_C = 0,
OAM_TEMP_SCALE_F
} oam_temp_sensor_scale_t;
/**
@@ -185,17 +174,17 @@ typedef enum oam_temp_sensor_scale {
* \details This enumerated type defines available scales for power measurements
*/
typedef enum oam_fan_sensor_scale {
OAM_FAN_SPEED_Hz = 0,
OAM_FAN_SPEED_KHz,
OAM_FAN_SPEED_MHz
OAM_FAN_SPEED_Hz = 0,
OAM_FAN_SPEED_KHz,
OAM_FAN_SPEED_MHz
} oam_fan_sensor_scale_t;
typedef union oam_sensor_scale {
oam_power_sensor_scale_t power_scale;
oam_voltage_sensor_scale_t volate_scale;
oam_current_sensor_scale_t current_scale;
oam_temp_sensor_scale_t temp_scale;
oam_fan_sensor_scale_t fan_scale;
oam_power_sensor_scale_t power_scale;
oam_voltage_sensor_scale_t volate_scale;
oam_current_sensor_scale_t current_scale;
oam_temp_sensor_scale_t temp_scale;
oam_fan_sensor_scale_t fan_scale;
} oam_sensor_scale_t;
/**
@@ -206,7 +195,7 @@ typedef union oam_sensor_scale {
* specific operation on that device.
*/
typedef struct oam_dev_handle {
void *handle;
void *handle;
} oam_dev_handle_t;
/**
@@ -218,9 +207,9 @@ typedef struct oam_dev_handle {
* device there are no side effects.
*/
typedef enum oam_dev_mode {
OAM_DEV_MODE_EXCLUSIVE = 0,
OAM_DEV_MODE_NONEXLUSIVE = 1,
OAM_DEV_MODE_UNKNOWN = 0xFF
OAM_DEV_MODE_EXCLUSIVE = 0,
OAM_DEV_MODE_NONEXLUSIVE = 1,
OAM_DEV_MODE_UNKNOWN = 0xFF
} oam_dev_mode_t;
/**
@@ -237,10 +226,10 @@ typedef enum oam_dev_mode {
* specific operation on that device.
*/
typedef struct oam_sensor_info {
char sensor_name[OAM_SENSOR_NAME_MAX];
oam_sensor_type_t sensor_type;
oam_sensor_scale_t scale;
int32_t value;
char sensor_name[OAM_SENSOR_NAME_MAX];
oam_sensor_type_t sensor_type;
oam_sensor_scale_t scale;
int32_t value;
} oam_sensor_info_t;
/**
@@ -249,10 +238,10 @@ typedef struct oam_sensor_info {
* \details Various types of errors reported by device.
*/
typedef struct oam_dev_error_count {
uint32_t total_error_count;
uint32_t fatal_error_count;
uint32_t unknown_error_count;
uint32_t ecc_error_count;
uint32_t total_error_count;
uint32_t fatal_error_count;
uint32_t unknown_error_count;
uint32_t ecc_error_count;
} oam_dev_error_count_t;
/**
@@ -261,10 +250,10 @@ typedef struct oam_dev_error_count {
* \details Structure to store various firmware versions of OAM module
*/
typedef struct oam_firmware_version {
oam_mapi_version_t device_boot_fw_version;
oam_mapi_version_t device_fw_version;
oam_mapi_version_t board_boot_fw_version;
oam_mapi_version_t board_fw_version;
oam_mapi_version_t device_boot_fw_version;
oam_mapi_version_t device_fw_version;
oam_mapi_version_t board_boot_fw_version;
oam_mapi_version_t board_fw_version;
} oam_firmware_version_t;
/**
@@ -273,10 +262,10 @@ typedef struct oam_firmware_version {
* \details Structure to store PCI (Domain, BDF) information of the device
*/
typedef struct oam_pci_info {
uint16_t domain;
uint8_t bus;
uint8_t device;
uint8_t function;
uint16_t domain;
uint8_t bus;
uint8_t device;
uint8_t function;
} oam_pci_info_t;
/**
@@ -285,8 +274,8 @@ typedef struct oam_pci_info {
* \details This enumerated type defines various states of the network port
*/
typedef enum oam_net_port_state {
OAM_NET_PORT_DISABLED = 0,
OAM_NET_PORT_ENABLED = 1
OAM_NET_PORT_DISABLED = 0,
OAM_NET_PORT_ENABLED = 1
} oam_net_port_state_t;
/**
@@ -295,8 +284,8 @@ typedef enum oam_net_port_state {
* \details This enumerated type defines various status of the network port
*/
typedef enum oam_net_port_status {
OAM_NET_PORT_UP = 0,
OAM_NET_PORT_DOWN = 1,
OAM_NET_PORT_UP = 0,
OAM_NET_PORT_DOWN = 1,
} oam_net_port_status_t;
/**
@@ -305,10 +294,10 @@ typedef enum oam_net_port_status {
* \details This enumerated type defines various identifiers for network ports
*/
typedef enum oam_net_port_id {
OAM_NET_PORT0 = 0,
OAM_NET_PORT1 = 1,
OAM_NET_PORT2 = 2,
OAM_NET_PORT_MAX = 0xFFFF
OAM_NET_PORT0 = 0,
OAM_NET_PORT1 = 1,
OAM_NET_PORT2 = 2,
OAM_NET_PORT_MAX = 0xFFFF
} oam_net_port_id_t;
/**
@@ -318,8 +307,8 @@ typedef enum oam_net_port_id {
* the device to update firmware.
*/
typedef enum oam_firmware_modes {
OAM_DOWNLOAD_ONLY = 0,
OAM_DOWNLOAD_ACTIVATE = 1
OAM_DOWNLOAD_ONLY = 0,
OAM_DOWNLOAD_ACTIVATE = 1
} oam_firmware_modes_t;
/**
@@ -334,7 +323,7 @@ typedef enum oam_firmware_modes {
* \details Structure to store additional details about the network port
*/
typedef struct oam_net_port_desc {
char name[OAM_NET_PORT_NAME];
char name[OAM_NET_PORT_NAME];
} oam_net_port_desc_t;
/**
@@ -350,9 +339,9 @@ typedef struct oam_net_port_desc {
* on a particular network.
*/
typedef struct oam_net_dev_info {
oam_net_dev_id_t net_dev_id;
char host_name[OAM_DEV_HOST_NAME];
oam_pci_info_t pci_info;
oam_net_dev_id_t net_dev_id;
char host_name[OAM_DEV_HOST_NAME];
oam_pci_info_t pci_info;
} oam_net_dev_info_t;
/**
@@ -362,8 +351,8 @@ typedef struct oam_net_dev_info {
* network
*/
typedef struct oam_neighbour_info {
oam_net_port_id_t device_port;
oam_net_dev_info_t device_info;
oam_net_port_id_t device_port;
oam_net_dev_info_t device_info;
} oam_neighbour_info_t;
/**
@@ -372,10 +361,10 @@ typedef struct oam_neighbour_info {
* \details This enumerated type defines various identifiers for TPCs
*/
typedef enum oam_dev_tpc_id {
OAM_DEV_TPC0,
OAM_DEV_TPC1,
OAM_DEV_TPC2,
OAM_DEV_TPC_MAX
OAM_DEV_TPC0,
OAM_DEV_TPC1,
OAM_DEV_TPC2,
OAM_DEV_TPC_MAX
} oam_dev_tpc_id_t;
/**
@@ -391,7 +380,7 @@ typedef enum oam_dev_tpc_id {
* to the id etc.
*/
typedef struct oam_tpc_desc {
char name[256];
char name[256];
} oam_tpc_desc_t;
/**
@@ -401,7 +390,7 @@ typedef struct oam_tpc_desc {
* e.g. TPC utilization
*/
typedef struct oam_dev_tpc_stats {
double util;
double util;
} oam_dev_tpc_stats_t;
/**
@@ -410,10 +399,10 @@ typedef struct oam_dev_tpc_stats {
* \details This enumerated type defines various identifiers for device memories
*/
typedef enum oam_dev_mem_id {
OAM_DEV_MEM0,
OAM_DEV_MEM1,
OAM_DEV_MEM2,
OAM_DEV_MEM_MAX
OAM_DEV_MEM0,
OAM_DEV_MEM1,
OAM_DEV_MEM2,
OAM_DEV_MEM_MAX
} oam_dev_mem_id_t;
/**
@@ -422,7 +411,7 @@ typedef enum oam_dev_mem_id {
* \details Structure to store additional details about device memories port
*/
typedef struct oam_mem_desc {
char name[256];
char name[256];
} oam_mem_desc_t;
/**
@@ -432,9 +421,9 @@ typedef struct oam_mem_desc {
* memory.
*/
typedef struct oam_dev_mem_stats {
uint32_t total_mem;
uint32_t allocated_mem;
uint32_t free_mem;
uint32_t total_mem;
uint32_t allocated_mem;
uint32_t free_mem;
} oam_dev_mem_stats_t;
/**
@@ -444,10 +433,10 @@ typedef struct oam_dev_mem_stats {
* packets on a given port.
*/
typedef struct oam_net_port_pkt_stats {
uint64_t rx_count;
uint64_t tx_count;
uint64_t rx_errors;
uint64_t tx_errors;
uint64_t rx_count;
uint64_t tx_count;
uint64_t rx_errors;
uint64_t tx_errors;
} oam_net_port_pkt_stats_t;
/**
@@ -457,186 +446,187 @@ typedef struct oam_net_port_pkt_stats {
* supported by the OAM library.
*/
typedef struct oam_ops {
/*!<
* to initialise library instance and perform version compatibility
* check
*/
int (*init)(oam_mapi_version_t version);
int (*free)(void);
/*!<
* to initialise library instance and perform version compatibility
* check
*/
int (*init)(void);
int (*free)(void);
/*!<
* To get error description from the error code
*/
int (*get_error_description)(int error_code, const char **error_description);
/*!<
* To get error description from the error code
*/
int (*get_error_description)(int error_code, const char **error_description);
/*!<
* To retrieve the OAM Management interface version
*/
int (*get_mapi_version)(oam_mapi_version_t *version);
/*!<
* To retrieve the OAM Management interface version
*/
int (*get_mapi_version)(oam_mapi_version_t *version);
/*!<
* To retrieve the number of devices present/discovered by the library
*/
int (*discover_devices)(int *device_count);
/*!<
* To retrieve the number of devices present/discovered by the library
*/
int (*discover_devices)(uint32_t *device_count);
/*!<
* To retrieve device properties for each discovered devices
*/
int (*get_dev_properties)(oam_dev_properties_t *devices);
/*!<
* To retrieve device properties for each discovered devices
*/
int (*get_dev_properties)(uint32_t device_count,
oam_dev_properties_t *devices);
/*!<
* To retrieve PCI properties of the device
*/
int (*get_pci_properties)(oam_dev_id_t *device_id, oam_pci_info_t *pci_info);
/*!<
* To retrieve PCI properties of the device
*/
int (*get_pci_properties)(uint32_t device_id, oam_pci_info_t *pci_info);
/*!<
* To query the number of various sensors present
*/
int (*get_sensors_count)(oam_dev_id_t *device_id,
oam_sensor_count_t *sensor_count);
/*!<
* To query the number of various sensors present
*/
int (*get_sensors_count)(uint32_t device_id,
oam_sensor_count_t *sensor_count);
/*!<
* Open the device and obtain handle
*/
int (*open_device)(oam_dev_id_t *dev_id, oam_dev_mode_t mode,
oam_dev_handle_t *handle);
int (*close_device)(oam_dev_handle_t *handle);
/*!<
* Open the device and obtain handle
*/
int (*open_device)(uint32_t *dev_id, oam_dev_mode_t mode,
oam_dev_handle_t *handle);
int (*close_device)(oam_dev_handle_t *handle);
/*!<
* To read various sensor values for a given sensor type
*/
int (*get_sensors_info)(oam_dev_handle_t *handle,
oam_sensor_type_t type,
uint32_t num_sensors,
oam_sensor_info_t sensor_info[]);
/*!<
* To read current error count of the device
*/
int (*get_device_error_count)(oam_dev_handle_t *handle,
oam_dev_error_count_t *count);
/*!<
* To read various sensor values for a given sensor type
*/
int (*get_sensors_info)(oam_dev_handle_t *handle,
oam_sensor_type_t type,
uint32_t num_sensors,
oam_sensor_info_t sensor_info[]);
/*!<
* To read current error count of the device
*/
int (*get_device_error_count)(oam_dev_handle_t *handle,
oam_dev_error_count_t *count);
/*!<
* To update firmware on the device
* fw_image contains a null terminated string which specifies complete
* path where the firmware image is located
*/
int (*download_firmware)(oam_dev_id_t *device_id, char *fw_image,
oam_firmware_modes_t mode);
/*!<
* To update firmware on the device
* fw_image contains a null terminated string which specifies complete
* path where the firmware image is located
*/
int (*download_firmware)(uint32_t *device_id, char *fw_image,
oam_firmware_modes_t mode);
/*!<
* To query firmware versions
*/
int (*get_firmware_version)(oam_dev_id_t *device_id,
oam_firmware_version_t *version);
/*!<
* To query firmware versions
*/
int (*get_firmware_version)(uint32_t *device_id,
oam_firmware_version_t *version);
/*!<
* to get network id from device id
*/
int (*get_net_dev_id)(oam_dev_id_t *device_id, oam_net_dev_id_t *net_device);
/*!<
* to get network id from device id
*/
int (*get_net_dev_id)(uint32_t *device_id, oam_net_dev_id_t *net_device);
/*!<
* Network management APIs.
*/
/*!<
* Network management APIs.
*/
/*!<
* discover network.
*/
int (*discover_network)(int *net_dev_count);
int (*get_dev_net_properties)(oam_net_dev_info_t *net_dev_info);
/*!<
* discover network.
*/
int (*discover_network)(int *net_dev_count);
int (*get_dev_net_properties)(oam_net_dev_info_t *net_dev_info);
int (*get_neighbour_count)(oam_dev_id_t *device,
oam_net_port_id_t local_port_id,
uint32_t *neighbor_count);
int (*get_neighbour_count)(uint32_t *device,
oam_net_port_id_t local_port_id,
uint32_t *neighbor_count);
int (*get_neighbours_info)(oam_dev_id_t *device,
oam_net_port_id_t local_port_id,
uint32_t *neighbors_count,
oam_neighbour_info_t *neighbours_info);
int (*get_neighbours_info)(uint32_t *device,
oam_net_port_id_t local_port_id,
uint32_t *neighbors_count,
oam_neighbour_info_t *neighbours_info);
int (*configure_network)(oam_net_dev_id_t *net_devices,
uint32_t *net_device_count,
char *network_name);
int (*configure_network)(oam_net_dev_id_t *net_devices,
uint32_t *net_device_count,
char *network_name);
int (*destroy_network)(char *network_name);
int (*destroy_network)(char *network_name);
int (*query_network)(char *network_name, oam_net_dev_info_t *devices,
uint32_t *device_count);
int (*query_network)(char *network_name, oam_net_dev_info_t *devices,
uint32_t *device_count);
int (*get_network_count)(uint32_t *network_count);
int (*list_networks)(char *network_names[]);
int (*get_network_count)(uint32_t *network_count);
int (*list_networks)(char *network_names[]);
/*!<
* Various statistics related to blocks
*/
/*!<
* Various statistics related to blocks
*/
/*!<
* To query number of ports
*/
int (*get_net_port_count)(oam_dev_handle_t *handle, uint32_t *count,
oam_net_port_id_t *port_ids);
/*!<
* To query number of ports
*/
int (*get_net_port_count)(oam_dev_handle_t *handle, uint32_t *count,
oam_net_port_id_t *port_ids);
int (*get_net_port_desc)(oam_dev_handle_t *handle, oam_net_port_id_t *port,
oam_net_port_desc_t *desc);
int (*get_net_port_desc)(oam_dev_handle_t *handle, oam_net_port_id_t *port,
oam_net_port_desc_t *desc);
int (*get_net_port_state)(oam_dev_handle_t *handle, oam_net_port_id_t *port,
oam_net_port_state_t *state);
int (*get_net_port_state)(oam_dev_handle_t *handle, oam_net_port_id_t *port,
oam_net_port_state_t *state);
int (*check_net_port_status)(oam_dev_handle_t *handle,
oam_net_port_id_t *port,
oam_net_port_status_t *status);
int (*get_net_port_pkt_stats)(oam_dev_handle_t *handle,
oam_net_port_id_t *port,
uint32_t duration_sec,
oam_net_port_pkt_stats_t *stats);
int (*check_net_port_status)(oam_dev_handle_t *handle,
oam_net_port_id_t *port,
oam_net_port_status_t *status);
int (*get_net_port_pkt_stats)(oam_dev_handle_t *handle,
oam_net_port_id_t *port,
uint32_t duration_sec,
oam_net_port_pkt_stats_t *stats);
int (*query_net_port_bandwidth)(oam_dev_handle_t *handle,
oam_net_port_id_t *port,
uint32_t duration_sec,
double *bandwidth);
int (*query_net_port_bandwidth)(oam_dev_handle_t *handle,
oam_net_port_id_t *port,
uint32_t duration_sec,
double *bandwidth);
int (*get_tpc_count)(oam_dev_handle_t *handle, uint32_t *count,
oam_dev_tpc_id_t *tpc_ids);
int (*get_tpc_count)(oam_dev_handle_t *handle, uint32_t *count,
oam_dev_tpc_id_t *tpc_ids);
int (*get_tpc_desc)(oam_dev_handle_t *handle, oam_dev_tpc_id_t *tpc_id,
oam_tpc_desc_t *desc);
int (*get_tpc_desc)(oam_dev_handle_t *handle, oam_dev_tpc_id_t *tpc_id,
oam_tpc_desc_t *desc);
int (*get_tpc_stats)(oam_dev_handle_t *handle,
oam_dev_tpc_id_t *port,
oam_dev_tpc_stats_t *stats,
uint32_t duration_sec);
int (*get_tpc_stats)(oam_dev_handle_t *handle,
oam_dev_tpc_id_t *port,
oam_dev_tpc_stats_t *stats,
uint32_t duration_sec);
int (*get_mem_count)(oam_dev_handle_t *handle, uint32_t *count,
oam_dev_mem_id_t *mem_ids);
int (*get_mem_count)(oam_dev_handle_t *handle, uint32_t *count,
oam_dev_mem_id_t *mem_ids);
int (*get_mem_desc)(oam_dev_handle_t *handle, oam_dev_mem_id_t *tpc_id,
oam_mem_desc_t *desc);
int (*get_mem_desc)(oam_dev_handle_t *handle, oam_dev_mem_id_t *tpc_id,
oam_mem_desc_t *desc);
int (*get_mem_stats)(oam_dev_handle_t *handle, oam_dev_mem_id_t *mem_id,
oam_dev_mem_stats_t *stats);
int (*get_mem_stats)(oam_dev_handle_t *handle, oam_dev_mem_id_t *mem_id,
oam_dev_mem_stats_t *stats);
/*!<
* To check the health of the individual components, libraries
* generates test workload to check if the block is functioning properly
* or not. So no other workload should be running while calling these
* APIs
*/
int (*check_tpc_health)(oam_dev_id_t *device_id, oam_dev_tpc_id_t *tpc_id);
int (*check_net_port_health)(oam_dev_id_t *device_id,
oam_net_port_id_t *port);
int (*check_mem_health)(oam_dev_id_t *device_id, oam_dev_mem_id_t *port);
/*!<
* To check the health of the individual components, libraries
* generates test workload to check if the block is functioning properly
* or not. So no other workload should be running while calling these
* APIs
*/
int (*check_tpc_health)(uint32_t *device_id, oam_dev_tpc_id_t *tpc_id);
int (*check_net_port_health)(uint32_t *device_id, oam_net_port_id_t *port);
int (*check_mem_health)(uint32_t *device_id, oam_dev_mem_id_t *port);
/*
* Following needs more attention, will work on in next
int (*get_fan_speed)(oam_dev_t *oam);
int (*set_fan_speed)(oam_dev_t *oam, int speed);
/*
* Following needs more attention, will work on in next
int (*get_fan_speed)(oam_dev_t *oam);
int (*set_fan_speed)(oam_dev_t *oam, int speed);
int (*get_power_cap)(oam_dev_t *oam);
int (*set_power_cap)(oam_dev_t *oam, int power);
int (*get_power_cap)(oam_dev_t *oam);
int (*set_power_cap)(oam_dev_t *oam, int power);
int (*get_telemetry)(oam_dev_t *oam);
*/
int (*get_telemetry)(oam_dev_t *oam);
*/
} oam_ops_t;
projects/amdsmi/oam/src/amd_oam.cc
+172 -25
View File
@@ -23,8 +23,12 @@
*/
#include <assert.h>
#include <dirent.h>
#include <sstream>
#include <cstring>
#include <iostream>
#include <regex> // NOLINT
#include <map>
#include "rocm_smi/rocm_smi_common.h"
#include "rocm_smi/rocm_smi_main.h"
@@ -38,52 +42,196 @@
#include "oam/oam_mapi.h"
#include "oam/amd_oam.h"
static const std::map<int, const char *> err_map = {
{ AMDOAM_STATUS_INVALID_ARGS, "Invalid arguments" },
{ AMDOAM_STATUS_NOT_SUPPORTED, "Feature not supported" },
{ AMDOAM_STATUS_FILE_ERROR, "Problem accessing a file" },
{ AMDOAM_STATUS_PERMISSION, "Permission denied" },
{ AMDOAM_STATUS_OUT_OF_RESOURCES, "Not enough memory or other resource" },
{ AMDOAM_STATUS_INTERNAL_EXCEPTION, "An internal exception was caught" },
{ AMDOAM_STATUS_INPUT_OUT_OF_BOUNDS,
"The provided input is out of allowable or safe range" },
{ AMDOAM_STATUS_INIT_ERROR, "AMDOAM is not initialized or init failed" },
{ AMDOAM_STATUS_ERROR, "Generic error" },
{ AMDOAM_STATUS_NOT_FOUND, "An item was searched for but not found" }
};
#define TRY try {
#define CATCH } catch (...) {return handleRSMIException();}
static bool rsmi_initialized;
static int handleRSMIException() {
rsmi_status_t ret;
ret = amd::smi::handleException();
// TODO(x): convert RSMI return to OAM return
// For now, just return int equiv.
return static_cast<int>(ret);
static int rsmi_status_to_amdoam_errorcode(rsmi_status_t status) {
if (status > RSMI_STATUS_INIT_ERROR)
return -AMDOAM_STATUS_ERROR;
else
return -status;
}
int amdoam_init(oam_mapi_version_t version) {
static int handleRSMIException() {
rsmi_status_t ret = amd::smi::handleException();
return rsmi_status_to_amdoam_errorcode(ret);
}
int amdoam_get_error_description(int code, const char **description) {
if (description == nullptr)
return -AMDOAM_STATUS_INVALID_ARGS;
auto search = err_map.find(code);
if (search == err_map.end())
return -AMDOAM_STATUS_NOT_FOUND;
*description = search->second;
return AMDOAM_STATUS_SUCCESS;
}
int amdoam_init(void) {
TRY
// TODO(x): handle version argument
(void)version;
rsmi_status_t status = rsmi_init(0);
rsmi_status_t ret = rsmi_init(0);
if (status != RSMI_STATUS_SUCCESS)
return rsmi_status_to_amdoam_errorcode(status);
rsmi_initialized = true;
return AMDOAM_STATUS_SUCCESS;
return 0;
CATCH
}
int amdoam_free(void) {
rsmi_status_t ret = rsmi_shut_down();
rsmi_status_t status = rsmi_shut_down();
// TODO(x) convert rsmi return to oam return val
return static_cast<int>(ret);
if (status != RSMI_STATUS_SUCCESS)
return rsmi_status_to_amdoam_errorcode(status);
return AMDOAM_STATUS_SUCCESS;
}
int amdoam_discover_devices(int *device_count) {
uint32_t dv_cnt;
int amdoam_discover_devices(uint32_t *device_count) {
rsmi_status_t status;
if (device_count == nullptr) {
return -1; // TODO(x): return appropriate OAM code
return -AMDOAM_STATUS_INVALID_ARGS;
}
rsmi_status_t ret = rsmi_num_monitor_devices(&dv_cnt);
status = rsmi_num_monitor_devices(device_count);
if (status != RSMI_STATUS_SUCCESS)
return rsmi_status_to_amdoam_errorcode(status);
*device_count = static_cast<int>(dv_cnt);
return AMDOAM_STATUS_SUCCESS;
}
// TODO(x) convert rsmi return to oam return val
return static_cast<int>(ret);
int amdoam_get_pci_properties(uint32_t device_id, oam_pci_info_t *pci_info) {
uint64_t bdfid;
TRY
if (pci_info == nullptr) {
return -AMDOAM_STATUS_INVALID_ARGS;
}
rsmi_status_t status = rsmi_dev_pci_id_get(device_id, &bdfid);
if (status != RSMI_STATUS_SUCCESS)
return rsmi_status_to_amdoam_errorcode(status);
pci_info->domain = (uint16_t)(bdfid >> 32) & 0xffff;
pci_info->bus = (bdfid >> 8) & 0xff;
pci_info->device = (bdfid >> 3) & 0x1f;
pci_info->function = bdfid & 0x7;
CATCH
return AMDOAM_STATUS_SUCCESS;
}
int amdoam_get_dev_properties(uint32_t num_devices,
oam_dev_properties_t *devices) {
const size_t buf_size = 32;
char buf[buf_size] = "";
uint32_t dev_inx;
oam_dev_properties_t *dev = devices;
TRY
if (devices == nullptr)
return -AMDOAM_STATUS_INVALID_ARGS;
if (!rsmi_initialized)
return -AMDOAM_STATUS_INIT_ERROR;
for (dev_inx = 0; dev_inx < num_devices; dev_inx++) {
dev->device_id = dev_inx;
/* If fails to get any following properties, it's not treated as a deal
* breaker. Variable not filled means that property is not available on
* this device or AMD doesn't support that property.
*/
rsmi_dev_vendor_name_get(dev_inx, dev->device_vendor, DEVICE_VENDOR_LEN);
rsmi_dev_name_get(dev_inx, dev->device_name, DEVICE_NAME_LEN);
rsmi_dev_vbios_version_get(dev_inx, buf, buf_size);
if (std::strlen(buf) > 0) {
std::strncpy(dev->sku_name, &buf[4], 6);
std::strncpy(dev->board_name, buf, 12);
}
rsmi_dev_serial_number_get(dev_inx, dev->board_serial_number,
BOARD_SERIAL_NUM_LEN);
++dev;
}
CATCH
return AMDOAM_STATUS_SUCCESS;
}
static int get_num_sensors(std::string hwmon_path, std::string fn_reg_ex,
uint32_t *sensor_max) {
*sensor_max = 0;
fn_reg_ex = "\\b" + fn_reg_ex + "([0-9]+)([^ ]*)";
std::string fn;
std::smatch m;
uint32_t temp = 0;
std::regex re(fn_reg_ex);
auto hwmon_dir = opendir(hwmon_path.c_str());
if (hwmon_dir == nullptr)
return 0;
auto dentry = readdir(hwmon_dir);
while (dentry != nullptr) {
fn = dentry->d_name;
if (std::regex_search(fn, m, re)) {
std::string output = std::regex_replace(
fn,
std::regex("[^0-9]*([0-9]+).*"),
std::string("$1"));
temp = stoi(output);
if (temp > *sensor_max) {
*sensor_max = temp;
}
}
dentry = readdir(hwmon_dir);
}
if (closedir(hwmon_dir)) {
return -errno;
}
return 0;
}
int amdoam_get_sensors_count(uint32_t device_id,
oam_sensor_count_t *sensor_count) {
uint32_t dv_ind = device_id;
uint32_t num_sensors = 0;
TRY
GET_DEV_FROM_INDX
std::string hwmon_path = dev->monitor()->path();
get_num_sensors(hwmon_path, "temp", &num_sensors);
sensor_count->num_temperature_sensors = num_sensors;
get_num_sensors(hwmon_path, "fan", &num_sensors);
sensor_count->num_fans = num_sensors;
get_num_sensors(hwmon_path, "in", &num_sensors);
sensor_count->num_voltage_sensors = (num_sensors+1);
get_num_sensors(hwmon_path, "power", &num_sensors);
sensor_count->num_power_sensors = num_sensors;
get_num_sensors(hwmon_path, "current", &num_sensors);
sensor_count->num_current_sensors = num_sensors;
CATCH
return AMDOAM_STATUS_SUCCESS;
}
// TODO(x): This function doesn't work for OAM. It's just a version
@@ -158,4 +306,3 @@ get_device_error_count(oam_dev_handle_t *handle,
return ret;
CATCH
}