[SWDEV-512393] Added amdsmi_get_cpu_affinity_with_scope (#198)
Signed-off-by: Maisam Arif <Maisam.Arif@amd.com> Signed-off-by: Deepak Mewar <deepak.mewar@amd.com>
This commit is contained in:
Mewar, Deepak
GitHub
+110
-2
@@ -3941,8 +3941,8 @@ amdsmi_get_gpu_cper_entries(
|
||||
std::string path = std::string("/sys/kernel/debug/dri/") +
|
||||
std::to_string(gpu_device->get_card_id()) +
|
||||
"/amdgpu_ring_cper";
|
||||
|
||||
|
||||
|
||||
|
||||
return amdsmi_get_gpu_cper_entries_by_path(
|
||||
path.c_str(),
|
||||
severity_mask,
|
||||
@@ -4795,6 +4795,85 @@ amdsmi_get_gpu_virtualization_mode(amdsmi_processor_handle processor_handle,
|
||||
return status;
|
||||
}
|
||||
|
||||
amdsmi_status_t amdsmi_get_cpu_affinity_with_scope(amdsmi_processor_handle processor_handle,
|
||||
uint32_t cpu_set_size, uint64_t *cpu_set, amdsmi_affinity_scope_t scope)
|
||||
{
|
||||
AMDSMI_CHECK_INIT();
|
||||
|
||||
if (processor_handle == nullptr || cpu_set == nullptr || cpu_set_size == 0) {
|
||||
return AMDSMI_STATUS_INVAL;
|
||||
}
|
||||
|
||||
// Retrieve GPU device from the processor handle
|
||||
amd::smi::AMDSmiGPUDevice* gpu_device = nullptr;
|
||||
amdsmi_status_t status = get_gpu_device_from_handle(processor_handle, &gpu_device);
|
||||
if (status != AMDSMI_STATUS_SUCCESS) {
|
||||
return status;
|
||||
}
|
||||
|
||||
uint32_t numa_node;
|
||||
status = amdsmi_topo_get_numa_node_number(processor_handle, &numa_node);
|
||||
if (status != AMDSMI_STATUS_SUCCESS) {
|
||||
return status;
|
||||
}
|
||||
|
||||
int32_t node_id = static_cast<int32_t>(numa_node);
|
||||
|
||||
status = amdsmi_get_gpu_topo_numa_affinity(processor_handle, &node_id);
|
||||
if (status != AMDSMI_STATUS_SUCCESS) {
|
||||
return status;
|
||||
}
|
||||
|
||||
if(node_id < 0) {
|
||||
return AMDSMI_STATUS_NOT_FOUND;
|
||||
}
|
||||
|
||||
std::memset(cpu_set, 0, cpu_set_size * sizeof(uint64_t));
|
||||
switch(scope) {
|
||||
case AMDSMI_AFFINITY_SCOPE_NODE:
|
||||
{
|
||||
std::vector<uint64_t> bitmask = gpu_device->get_bitmask_from_numa_node(node_id, cpu_set_size);
|
||||
if(bitmask[0] == std::numeric_limits<int32_t>::max()){
|
||||
return AMDSMI_STATUS_REFCOUNT_OVERFLOW;
|
||||
} else {
|
||||
std::memcpy(cpu_set, bitmask.data(), cpu_set_size * sizeof(uint64_t));
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
||||
case AMDSMI_AFFINITY_SCOPE_SOCKET:
|
||||
{
|
||||
std::vector<uint32_t> sockets = amd::smi::AMDSmiSystem::getInstance().get_cpu_sockets_from_numa_node(node_id);
|
||||
|
||||
if(sockets[0] == std::numeric_limits<int32_t>::max()){
|
||||
return AMDSMI_STATUS_REFCOUNT_OVERFLOW;
|
||||
} else {
|
||||
for (uint32_t idx : sockets) {
|
||||
cpu_set[idx] = idx;
|
||||
}
|
||||
|
||||
std::sort(cpu_set, cpu_set + cpu_set_size);
|
||||
|
||||
// Discard duplicates
|
||||
uint32_t temp_size = 0;
|
||||
for (uint32_t i = 0; i < cpu_set_size; ++i) {
|
||||
if (i == 0 || cpu_set[i] != cpu_set[i - 1]) {
|
||||
cpu_set[temp_size++] = cpu_set[i];
|
||||
}
|
||||
}
|
||||
|
||||
// Update the size to the temp size after discarding duplicates
|
||||
cpu_set_size = temp_size;
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
||||
default:
|
||||
return AMDSMI_STATUS_INPUT_OUT_OF_BOUNDS;
|
||||
}
|
||||
|
||||
return AMDSMI_STATUS_SUCCESS;
|
||||
}
|
||||
|
||||
#ifdef ENABLE_ESMI_LIB
|
||||
static amdsmi_status_t amdsmi_errno_to_esmi_status(amdsmi_status_t status)
|
||||
@@ -5905,6 +5984,35 @@ amdsmi_status_t amdsmi_get_cpu_model_name(amdsmi_processor_handle processor_hand
|
||||
return AMDSMI_STATUS_SUCCESS;
|
||||
}
|
||||
|
||||
amdsmi_status_t amdsmi_get_cpu_cores_per_socket(uint32_t sock_count, amdsmi_sock_info_t *sock_info)
|
||||
{
|
||||
std::map<uint32_t, uint32_t> socket_core_count = amd::smi::AMDSmiSystem::getInstance().get_sys_cpu_cores_per_socket();
|
||||
|
||||
for (uint32_t i = 0; i < sock_count; ++i) {
|
||||
auto it = socket_core_count.find(sock_info[i].socket_id);
|
||||
if (it != socket_core_count.end()) {
|
||||
sock_info[i].cores_per_socket = it->second;
|
||||
} else {
|
||||
sock_info[i].cores_per_socket = 0;
|
||||
}
|
||||
}
|
||||
|
||||
return AMDSMI_STATUS_SUCCESS;
|
||||
}
|
||||
|
||||
amdsmi_status_t amdsmi_get_cpu_socket_count(uint32_t *sock_count)
|
||||
{
|
||||
amdsmi_status_t status;
|
||||
uint32_t sock_num;
|
||||
status = amd::smi::AMDSmiSystem::getInstance().get_sys_num_of_cpu_sockets(&sock_num);
|
||||
if (status != AMDSMI_STATUS_SUCCESS)
|
||||
return status;
|
||||
|
||||
*sock_count = sock_num;
|
||||
|
||||
return AMDSMI_STATUS_SUCCESS;
|
||||
}
|
||||
|
||||
amdsmi_status_t amdsmi_get_cpu_handles(uint32_t *cpu_count,
|
||||
amdsmi_processor_handle *processor_handles)
|
||||
{
|
||||
|
||||
@@ -323,7 +323,39 @@ std::string AMDSmiGPUDevice::bdf_to_string() const {
|
||||
return oss.str();
|
||||
}
|
||||
|
||||
std::vector<uint64_t> AMDSmiGPUDevice::get_bitmask_from_numa_node(int32_t node_id, uint32_t size) const {
|
||||
std::vector<uint64_t> bitmask(size, 0);
|
||||
|
||||
if (node_id < 0) {
|
||||
bitmask[0] = std::numeric_limits<int32_t>::max();
|
||||
return bitmask;
|
||||
}
|
||||
|
||||
std::string path = "/sys/devices/system/node/node" + std::to_string(node_id) + "/cpulist";
|
||||
std::ifstream file(path);
|
||||
|
||||
if (file.is_open()) {
|
||||
std::string info;
|
||||
while (std::getline(file, info)) {
|
||||
std::istringstream sstr(info);
|
||||
std::string node_cpus;
|
||||
while (std::getline(sstr, node_cpus, ',')) {
|
||||
size_t hyphen = node_cpus.find('-');
|
||||
if (hyphen != std::string::npos) {
|
||||
int start = std::stoi(node_cpus.substr(0, hyphen));
|
||||
int end = std::stoi(node_cpus.substr(hyphen + 1));
|
||||
for (int i = start; i <= end; ++i) {
|
||||
bitmask[i / 64] |= (1ULL << (i % 64));
|
||||
}
|
||||
} else {
|
||||
int core = std::stoi(node_cpus);
|
||||
bitmask[core / 64] |= (1ULL << (core % 64));
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
return bitmask;
|
||||
}
|
||||
|
||||
} // namespace smi
|
||||
} // namespace amd
|
||||
|
||||
|
||||
@@ -31,6 +31,7 @@
|
||||
#include "amd_smi/impl/amd_smi_utils.h"
|
||||
#include "rocm_smi/rocm_smi.h"
|
||||
#include "rocm_smi/rocm_smi_main.h"
|
||||
#include <map>
|
||||
|
||||
namespace amd {
|
||||
namespace smi {
|
||||
@@ -103,7 +104,7 @@ amdsmi_status_t AMDSmiSystem::get_cpu_model_name(uint32_t socket_id, std::string
|
||||
if (!cpu_info.is_open()) {
|
||||
std::cerr << "Failed to open /proc/cpuinfo:" << strerror(errno) << std::endl;
|
||||
return AMDSMI_STATUS_FILE_ERROR;
|
||||
} else {
|
||||
} else {
|
||||
uint32_t current_socket_id = -1;
|
||||
while (std::getline(cpu_info, info)) {
|
||||
if (info.find("processor") != std::string::npos) {
|
||||
@@ -126,7 +127,85 @@ amdsmi_status_t AMDSmiSystem::get_cpu_model_name(uint32_t socket_id, std::string
|
||||
}
|
||||
return AMDSMI_STATUS_SUCCESS;
|
||||
}
|
||||
|
||||
#endif
|
||||
std::map<uint32_t, uint32_t> AMDSmiSystem::get_sys_cpu_cores_per_socket() {
|
||||
std::map<uint32_t, uint32_t> socket_core_count;
|
||||
std::string base_path = "/sys/devices/system/cpu/";
|
||||
|
||||
iterate_directory(base_path, [&socket_core_count](const std::string &path) {
|
||||
std::string filename(basename(path.c_str()));
|
||||
if (filename.find("cpu") != std::string::npos) {
|
||||
std::string cpuPath = path;
|
||||
std::ifstream package_id_file(cpuPath + "/topology/physical_package_id");
|
||||
std::ifstream core_id_file(cpuPath + "/topology/core_id");
|
||||
|
||||
if (package_id_file.is_open() && core_id_file.is_open()) {
|
||||
uint32_t physical_id, core_id;
|
||||
package_id_file >> physical_id;
|
||||
core_id_file >> core_id;
|
||||
|
||||
socket_core_count[physical_id]++;
|
||||
}
|
||||
}
|
||||
});
|
||||
|
||||
return socket_core_count;
|
||||
}
|
||||
|
||||
amdsmi_status_t AMDSmiSystem::get_sys_num_of_cpu_sockets(uint32_t *sock_num) {
|
||||
std::map<uint32_t, uint32_t> socket_count_map;
|
||||
std::string base_path = "/sys/devices/system/cpu/";
|
||||
|
||||
iterate_directory(base_path, [&socket_count_map](std::string path) {
|
||||
std::string filename(basename(path.c_str()));
|
||||
if (filename.find("cpu") != std::string::npos) {
|
||||
std::string cpu_path = path;
|
||||
std::ifstream package_id_file(cpu_path + "/topology/physical_package_id");
|
||||
|
||||
if (package_id_file.is_open()) {
|
||||
uint32_t physical_id;
|
||||
package_id_file >> physical_id;
|
||||
|
||||
socket_count_map[physical_id]++;
|
||||
}
|
||||
}
|
||||
});
|
||||
|
||||
*sock_num = static_cast<uint32_t>(socket_count_map.size());
|
||||
|
||||
return AMDSMI_STATUS_SUCCESS;
|
||||
}
|
||||
|
||||
std::vector<uint32_t> AMDSmiSystem::get_cpu_sockets_from_numa_node(int32_t numa_node) {
|
||||
std::vector<uint32_t> sockets;
|
||||
if (numa_node < 0) {
|
||||
sockets[0] = std::numeric_limits<int32_t>::max();
|
||||
return sockets;
|
||||
}
|
||||
std::ifstream node_info("/sys/devices/system/node/node" + std::to_string(numa_node) + "/cpulist");
|
||||
std::string info;
|
||||
|
||||
if (node_info.is_open()) {
|
||||
std::getline(node_info, info);
|
||||
std::istringstream iss(info);
|
||||
uint32_t index;
|
||||
while (iss >> index) {
|
||||
std::ifstream cpu_info("/sys/devices/system/cpu/cpu" + std::to_string(index) + "/topology/physical_package_id");
|
||||
if (cpu_info.is_open()) {
|
||||
uint32_t socket;
|
||||
cpu_info >> socket;
|
||||
sockets.push_back(socket);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Discarding duplicate socket entries
|
||||
std::sort(sockets.begin(), sockets.end());
|
||||
sockets.erase(std::unique(sockets.begin(), sockets.end()), sockets.end());
|
||||
|
||||
return sockets;
|
||||
}
|
||||
|
||||
amdsmi_status_t AMDSmiSystem::init(uint64_t flags) {
|
||||
init_flag_ = flags;
|
||||
|
||||
@@ -36,15 +36,15 @@
|
||||
#include <dirent.h>
|
||||
#include <sys/types.h>
|
||||
|
||||
#include <memory>
|
||||
#include <random>
|
||||
#include <algorithm>
|
||||
#include <cstdio>
|
||||
#include <fstream>
|
||||
#include <iostream>
|
||||
#include <regex>
|
||||
#include <cstdio>
|
||||
#include <sstream>
|
||||
#include <iterator>
|
||||
#include <algorithm>
|
||||
#include <memory>
|
||||
#include <random>
|
||||
#include <regex>
|
||||
#include <sstream>
|
||||
|
||||
#include "amd_smi/impl/amd_smi_utils.h"
|
||||
#include "amd_smi/impl/amd_smi_system.h"
|
||||
@@ -1046,3 +1046,25 @@ void amdsmi_wait_for_user_input(void) {
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
bool iterate_directory(const std::string &base_path,
|
||||
std::function<void(const std::string &)> entry_callback) {
|
||||
|
||||
DIR *dir = opendir(base_path.c_str());
|
||||
if (!dir) {
|
||||
return false;
|
||||
}
|
||||
|
||||
struct dirent *entry = nullptr;
|
||||
while ((entry = readdir(dir)) != NULL) {
|
||||
entry_callback(entry->d_name);
|
||||
}
|
||||
|
||||
if (errno != 0) {
|
||||
closedir(dir);
|
||||
return false;
|
||||
}
|
||||
|
||||
closedir(dir);
|
||||
return true;
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user