[SWDEV-488303] Updated CU occupancy for per-process retrieval (#243)
Change-Id: I2990597c6dd4b2e8cf3e11ce60f72049ebdd9a8c Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
This commit is contained in:
Arif, Maisam
GitHub
@@ -216,6 +216,9 @@ int32_t AMDSmiGPUDevice::get_compute_process_list_impl(GPUComputeProcessList_t&
|
||||
|
||||
/**
|
||||
* Complete the process information
|
||||
* This is where we copy rsmi_process_info_t into the larger amdsmi_proc_info_t
|
||||
* Then populate the remaining fields with the gpuvsmi_get_pid_info()
|
||||
* TODO FIX HERE TO GRAB KFD VRAM if /proc is inconsistent
|
||||
*/
|
||||
auto get_process_info = [&](const rsmi_process_info_t& rsmi_proc_info, amdsmi_proc_info_t& asmi_proc_info) {
|
||||
auto status_code = gpuvsmi_get_pid_info(get_bdf(), rsmi_proc_info.process_id, asmi_proc_info);
|
||||
@@ -225,6 +228,9 @@ int32_t AMDSmiGPUDevice::get_compute_process_list_impl(GPUComputeProcessList_t&
|
||||
asmi_proc_info.memory_usage.vram_mem = rsmi_proc_info.vram_usage;
|
||||
}
|
||||
|
||||
// Copy the cu occupancy from rsmi_process_info_t to amdsmi_proc_info_t
|
||||
asmi_proc_info.cu_occupancy = rsmi_proc_info.cu_occupancy;
|
||||
|
||||
return status_code;
|
||||
};
|
||||
|
||||
|
||||
+157
-190
@@ -20,16 +20,17 @@
|
||||
* THE SOFTWARE.
|
||||
*/
|
||||
|
||||
#include <sys/types.h>
|
||||
#include <dirent.h>
|
||||
#include <inttypes.h>
|
||||
#include <string.h>
|
||||
#include <sys/types.h>
|
||||
#include <unistd.h>
|
||||
#include <memory>
|
||||
#include <vector>
|
||||
|
||||
#include <algorithm>
|
||||
#include <cstdlib>
|
||||
#include <fstream>
|
||||
#include <algorithm>
|
||||
#include <string.h>
|
||||
#include <memory>
|
||||
#include <vector>
|
||||
|
||||
#include "amd_smi/amdsmi.h"
|
||||
#include "amd_smi/impl/amd_smi_utils.h"
|
||||
@@ -37,230 +38,196 @@
|
||||
extern "C" {
|
||||
|
||||
static const char *container_type_name[AMDSMI_MAX_CONTAINER_TYPE] = {
|
||||
[AMDSMI_CONTAINER_LXC] = "lxc",
|
||||
[AMDSMI_CONTAINER_DOCKER] = "docker",
|
||||
[AMDSMI_CONTAINER_LXC] = "lxc",
|
||||
[AMDSMI_CONTAINER_DOCKER] = "docker",
|
||||
};
|
||||
|
||||
amdsmi_status_t gpuvsmi_pid_is_gpu(const std::string &path, const char *bdf)
|
||||
{
|
||||
DIR *d;
|
||||
struct dirent *dir;
|
||||
amdsmi_status_t gpuvsmi_pid_is_gpu(const std::string &path, const char *bdf) {
|
||||
DIR *d;
|
||||
struct dirent *dir;
|
||||
|
||||
d = opendir(path.c_str());
|
||||
if (!d)
|
||||
return AMDSMI_STATUS_NO_PERM;
|
||||
d = opendir(path.c_str());
|
||||
if (!d) return AMDSMI_STATUS_NO_PERM;
|
||||
|
||||
/* iterate through all the fds, try to find
|
||||
* a match for the GPU bdf
|
||||
*/
|
||||
while ((dir = readdir(d)) != NULL) {
|
||||
std::string file = path + dir->d_name;
|
||||
std::ifstream fdinfo(file.c_str());
|
||||
for (std::string line; std::getline(fdinfo, line);) {
|
||||
if (line.find(bdf) != std::string::npos) {
|
||||
closedir(d);
|
||||
return AMDSMI_STATUS_SUCCESS;
|
||||
}
|
||||
}
|
||||
}
|
||||
/* iterate through all the fds, try to find
|
||||
* a match for the GPU bdf
|
||||
*/
|
||||
while ((dir = readdir(d)) != NULL) {
|
||||
std::string file = path + dir->d_name;
|
||||
std::ifstream fdinfo(file.c_str());
|
||||
for (std::string line; std::getline(fdinfo, line);) {
|
||||
if (line.find(bdf) != std::string::npos) {
|
||||
closedir(d);
|
||||
return AMDSMI_STATUS_SUCCESS;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
closedir(d);
|
||||
closedir(d);
|
||||
|
||||
return AMDSMI_STATUS_NOT_FOUND;
|
||||
return AMDSMI_STATUS_NOT_FOUND;
|
||||
}
|
||||
|
||||
amdsmi_status_t gpuvsmi_get_pids(const amdsmi_bdf_t &bdf, std::vector<long int> &pids, uint64_t *size)
|
||||
{
|
||||
char bdf_str[13];
|
||||
DIR *d;
|
||||
struct dirent *dir;
|
||||
amdsmi_status_t gpuvsmi_get_pids(const amdsmi_bdf_t &bdf, std::vector<long int> &pids,
|
||||
uint64_t *size) {
|
||||
char bdf_str[13];
|
||||
DIR *d;
|
||||
struct dirent *dir;
|
||||
|
||||
/* 0000:00:00.0 */
|
||||
snprintf(bdf_str, 13, "%04" PRIx32 ":%02" PRIx32 ":%02" PRIx32 ".%" PRIu32,
|
||||
static_cast<uint32_t>(bdf.domain_number & 0xffff),
|
||||
static_cast<uint32_t>(bdf.bus_number & 0xff),
|
||||
static_cast<uint32_t>(bdf.device_number & 0x1f),
|
||||
static_cast<uint32_t>(bdf.function_number & 0x7));
|
||||
/* 0000:00:00.0 */
|
||||
snprintf(bdf_str, 13, "%04" PRIx32 ":%02" PRIx32 ":%02" PRIx32 ".%" PRIu32,
|
||||
static_cast<uint32_t>(bdf.domain_number & 0xffff),
|
||||
static_cast<uint32_t>(bdf.bus_number & 0xff),
|
||||
static_cast<uint32_t>(bdf.device_number & 0x1f),
|
||||
static_cast<uint32_t>(bdf.function_number & 0x7));
|
||||
|
||||
d = opendir("/proc");
|
||||
if (!d)
|
||||
return AMDSMI_STATUS_NO_PERM;
|
||||
d = opendir("/proc");
|
||||
if (!d) return AMDSMI_STATUS_NO_PERM;
|
||||
|
||||
pids.clear();
|
||||
/* Find the pid folders in /proc/ that we have access to */
|
||||
while ((dir = readdir(d)) != NULL) {
|
||||
if (dir->d_type == DT_DIR) {
|
||||
/* Try to cast the name of the folder to a
|
||||
* number, if it fails, it is not */
|
||||
char *p;
|
||||
long int pid;
|
||||
pids.clear();
|
||||
/* Find the pid folders in /proc/ that we have access to */
|
||||
while ((dir = readdir(d)) != NULL) {
|
||||
if (dir->d_type == DT_DIR) {
|
||||
/* Try to cast the name of the folder to a
|
||||
* number, if it fails, it is not */
|
||||
char *p;
|
||||
long int pid;
|
||||
|
||||
pid = strtol(dir->d_name, &p, 10);
|
||||
if (*p != 0)
|
||||
continue;
|
||||
pid = strtol(dir->d_name, &p, 10);
|
||||
if (*p != 0) continue;
|
||||
|
||||
/* Check if fdinfo is accesible */
|
||||
std::string path = "/proc/" + std::string(dir->d_name) + "/fdinfo/";
|
||||
/* Check if fdinfo is accesible */
|
||||
std::string path = "/proc/" + std::string(dir->d_name) + "/fdinfo/";
|
||||
|
||||
if (access(path.c_str(), R_OK))
|
||||
continue;
|
||||
if (access(path.c_str(), R_OK)) continue;
|
||||
|
||||
/* check if GPU is present */
|
||||
if (gpuvsmi_pid_is_gpu(path, bdf_str))
|
||||
continue;
|
||||
pids.push_back(pid);
|
||||
}
|
||||
}
|
||||
closedir(d);
|
||||
/* check if GPU is present */
|
||||
if (gpuvsmi_pid_is_gpu(path, bdf_str)) continue;
|
||||
pids.push_back(pid);
|
||||
}
|
||||
}
|
||||
closedir(d);
|
||||
|
||||
*size = pids.size();
|
||||
return AMDSMI_STATUS_SUCCESS;
|
||||
*size = pids.size();
|
||||
return AMDSMI_STATUS_SUCCESS;
|
||||
}
|
||||
|
||||
amdsmi_status_t gpuvsmi_get_pid_info(const amdsmi_bdf_t &bdf, long int pid,
|
||||
amdsmi_proc_info_t &info)
|
||||
{
|
||||
char bdf_str[13];
|
||||
DIR *d;
|
||||
struct dirent *dir;
|
||||
amdsmi_proc_info_t &info) {
|
||||
char bdf_str[13];
|
||||
DIR *d;
|
||||
struct dirent *dir;
|
||||
|
||||
/* 0000:00:00.0 */
|
||||
snprintf(bdf_str, 13, "%04" PRIx32 ":%02" PRIx32 ":%02" PRIx32 ".%" PRIu32,
|
||||
static_cast<uint32_t>(bdf.domain_number & 0xffff),
|
||||
static_cast<uint32_t>(bdf.bus_number & 0xff),
|
||||
static_cast<uint32_t>(bdf.device_number & 0x1f),
|
||||
static_cast<uint32_t>(bdf.function_number & 0x7));
|
||||
/* 0000:00:00.0 */
|
||||
snprintf(bdf_str, 13, "%04" PRIx32 ":%02" PRIx32 ":%02" PRIx32 ".%" PRIu32,
|
||||
static_cast<uint32_t>(bdf.domain_number & 0xffff),
|
||||
static_cast<uint32_t>(bdf.bus_number & 0xff),
|
||||
static_cast<uint32_t>(bdf.device_number & 0x1f),
|
||||
static_cast<uint32_t>(bdf.function_number & 0x7));
|
||||
|
||||
std::string path = "/proc/" + std::to_string(pid) + "/fdinfo/";
|
||||
std::string name_path = "/proc/" + std::to_string(pid) + "/comm";
|
||||
std::string cgroup_path = "/proc/" + std::to_string(pid) + "/cgroup";
|
||||
std::string path = "/proc/" + std::to_string(pid) + "/fdinfo/";
|
||||
std::string name_path = "/proc/" + std::to_string(pid) + "/comm";
|
||||
std::string cgroup_path = "/proc/" + std::to_string(pid) + "/cgroup";
|
||||
|
||||
if (gpuvsmi_pid_is_gpu(path.c_str(), bdf_str)) {
|
||||
return AMDSMI_STATUS_INVAL;
|
||||
}
|
||||
if (gpuvsmi_pid_is_gpu(path.c_str(), bdf_str)) {
|
||||
return AMDSMI_STATUS_INVAL;
|
||||
}
|
||||
|
||||
d = opendir(path.c_str());
|
||||
if (!d)
|
||||
return AMDSMI_STATUS_NO_PERM;
|
||||
d = opendir(path.c_str());
|
||||
if (!d) return AMDSMI_STATUS_NO_PERM;
|
||||
|
||||
/* Vectors to check if repated fd pasid */
|
||||
// TODO remove pasid Not working in ROCm 6.4+, deprecating in 7.0
|
||||
std::vector<int> pasids;
|
||||
/* Vectors to check if repated fd pasid */
|
||||
// TODO remove pasid Not working in ROCm 6.4+, deprecating in 7.0
|
||||
std::vector<int> pasids;
|
||||
|
||||
memset(&info, 0, sizeof(info));
|
||||
/* Iterate through all fdinfos */
|
||||
while ((dir = readdir(d)) != NULL) {
|
||||
memset(&info, 0, sizeof(info));
|
||||
/* Iterate through all fdinfos */
|
||||
while ((dir = readdir(d)) != NULL) {
|
||||
std::string file = path + dir->d_name;
|
||||
std::ifstream fdinfo(file.c_str());
|
||||
|
||||
std::string file = path + dir->d_name;
|
||||
std::ifstream fdinfo(file.c_str());
|
||||
for (std::string bdfline; getline(fdinfo, bdfline);) {
|
||||
if (bdfline.find("drm-pdev:") != std::string::npos) {
|
||||
char fd_bdf_str[13];
|
||||
|
||||
for (std::string bdfline; getline(fdinfo, bdfline);) {
|
||||
if (bdfline.find("drm-pdev:") != std::string::npos) {
|
||||
char fd_bdf_str[13];
|
||||
/* Only check against fdinfo files that contain a bdf */
|
||||
if (sscanf(bdfline.c_str(), "drm-pdev: %s", &fd_bdf_str[0]) != 1) continue;
|
||||
|
||||
/* Only check against fdinfo files that contain a bdf */
|
||||
if (sscanf(bdfline.c_str(), "drm-pdev: %s", &fd_bdf_str[0]) != 1)
|
||||
continue;
|
||||
/* Populate amdsmi_proc_info_t struct only if the bdf in
|
||||
* the fdinfo file matches the passed bdf */
|
||||
if (strncmp(bdf_str, fd_bdf_str, 13) == 0) {
|
||||
std::ifstream fdinfo(file.c_str());
|
||||
|
||||
/* Populate amdsmi_proc_info_t struct only if the bdf in
|
||||
* the fdinfo file matches the passed bdf */
|
||||
if (strncmp(bdf_str, fd_bdf_str, 13) == 0){
|
||||
std::ifstream fdinfo(file.c_str());
|
||||
for (std::string line; getline(fdinfo, line);) {
|
||||
if (line.find("pasid:") != std::string::npos) {
|
||||
int pasid;
|
||||
if (sscanf(line.c_str(), "pasid: %d", &pasid) != 1) continue;
|
||||
auto it = std::find(pasids.begin(), pasids.end(), pasid);
|
||||
if (it == pasids.end()) pasids.push_back(pasid);
|
||||
} else if (line.find("drm-memory-gtt:") != std::string::npos) {
|
||||
unsigned long mem;
|
||||
if (sscanf(line.c_str(), "drm-memory-gtt: %lu", &mem) != 1) continue;
|
||||
info.mem += mem * 1000;
|
||||
info.memory_usage.gtt_mem += mem * 1000;
|
||||
} else if (line.find("drm-memory-cpu:") != std::string::npos) {
|
||||
unsigned long mem;
|
||||
if (sscanf(line.c_str(), "drm-memory-cpu: %lu", &mem) != 1) continue;
|
||||
info.mem += mem * 1000;
|
||||
info.memory_usage.cpu_mem += mem * 1000;
|
||||
} else if (line.find("drm-memory-vram:") != std::string::npos) {
|
||||
unsigned long mem;
|
||||
if (sscanf(line.c_str(), "drm-memory-vram: %lu", &mem) != 1) continue;\
|
||||
info.mem += mem * 1000;
|
||||
info.memory_usage.vram_mem += mem * 1000;
|
||||
} else if (line.find("drm-engine-gfx") != std::string::npos) {
|
||||
uint64_t engine_gfx;
|
||||
if (sscanf(line.c_str(), "drm-engine-gfx: %lu", &engine_gfx) != 1) continue;
|
||||
info.engine_usage.gfx = engine_gfx;
|
||||
} else if (line.find("drm-engine-enc") != std::string::npos) {
|
||||
uint64_t engine_enc;
|
||||
if (sscanf(line.c_str(), "drm-engine-enc: %lu", &engine_enc) != 1) continue;
|
||||
info.engine_usage.enc = engine_enc;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
for (std::string line; getline(fdinfo, line);) {
|
||||
if (line.find("pasid:") != std::string::npos) {
|
||||
int pasid;
|
||||
|
||||
if (sscanf(line.c_str(), "pasid: %d", &pasid) != 1)
|
||||
continue;
|
||||
|
||||
auto it = std::find(pasids.begin(), pasids.end(), pasid);
|
||||
|
||||
if (it == pasids.end())
|
||||
pasids.push_back(pasid);
|
||||
} else if (line.find("drm-memory-gtt:") != std::string::npos) {
|
||||
unsigned long mem;
|
||||
|
||||
if (sscanf(line.c_str(), "drm-memory-gtt: %lu", &mem) != 1)
|
||||
continue;
|
||||
|
||||
info.mem += mem * 1024;
|
||||
info.memory_usage.gtt_mem += mem * 1024;
|
||||
} else if (line.find("drm-memory-cpu:") != std::string::npos) {
|
||||
unsigned long mem;
|
||||
|
||||
if (sscanf(line.c_str(), "drm-memory-cpu: %lu", &mem) != 1)
|
||||
continue;
|
||||
|
||||
info.mem += mem * 1024;
|
||||
info.memory_usage.cpu_mem += mem * 1024;
|
||||
} else if (line.find("drm-memory-vram:") != std::string::npos) {
|
||||
unsigned long mem;
|
||||
|
||||
if (sscanf(line.c_str(), "drm-memory-vram: %lu", &mem) != 1)
|
||||
continue;
|
||||
|
||||
info.mem += mem * 1024;
|
||||
info.memory_usage.vram_mem += mem * 1024;
|
||||
} else if (line.find("drm-engine-gfx") != std::string::npos) {
|
||||
uint64_t engine_gfx;
|
||||
|
||||
if (sscanf(line.c_str(), "drm-engine-gfx: %lu", &engine_gfx) != 1)
|
||||
continue;
|
||||
|
||||
info.engine_usage.gfx = engine_gfx;
|
||||
} else if (line.find("drm-engine-enc") != std::string::npos) {
|
||||
uint64_t engine_enc;
|
||||
|
||||
if (sscanf(line.c_str(), "drm-engine-enc: %lu", &engine_enc) != 1)
|
||||
continue;
|
||||
|
||||
info.engine_usage.enc = engine_enc;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
closedir(d);
|
||||
closedir(d);
|
||||
|
||||
// Note: If possible at all, try to get the name of the process/container.
|
||||
// In case the other info fail, get at least something.
|
||||
std::ifstream filename(name_path.c_str());
|
||||
std::string name;
|
||||
std::ifstream filename(name_path.c_str());
|
||||
std::string name;
|
||||
|
||||
getline(filename, name);
|
||||
getline(filename, name);
|
||||
|
||||
if (name.empty())
|
||||
return AMDSMI_STATUS_API_FAILED;
|
||||
if (name.empty()) return AMDSMI_STATUS_API_FAILED;
|
||||
|
||||
strncpy(info.name, name.c_str(), std::min(
|
||||
(unsigned long) AMDSMI_MAX_STRING_LENGTH,
|
||||
name.length()));
|
||||
strncpy(info.name, name.c_str(),
|
||||
std::min((unsigned long)AMDSMI_MAX_STRING_LENGTH, name.length()));
|
||||
|
||||
for (int i = 0; i < AMDSMI_MAX_CONTAINER_TYPE; i++) {
|
||||
std::ifstream cgroup_info(cgroup_path.c_str());
|
||||
std::string container_id;
|
||||
for (std::string line; getline(cgroup_info, line);) {
|
||||
if (line.find(container_type_name[i]) != std::string::npos) {
|
||||
container_id = line.substr(line.find(container_type_name[i]) +
|
||||
strlen(container_type_name[i]) + 1, 16);
|
||||
strcpy(info.container_name, container_id.c_str());
|
||||
break;
|
||||
}
|
||||
}
|
||||
if (strlen(info.container_name) > 0)
|
||||
break;
|
||||
}
|
||||
info.pid = (uint32_t)pid;
|
||||
for (int i = 0; i < AMDSMI_MAX_CONTAINER_TYPE; i++) {
|
||||
std::ifstream cgroup_info(cgroup_path.c_str());
|
||||
std::string container_id;
|
||||
for (std::string line; getline(cgroup_info, line);) {
|
||||
if (line.find(container_type_name[i]) != std::string::npos) {
|
||||
container_id = line.substr(line.find(container_type_name[i]) +
|
||||
strlen(container_type_name[i]) + 1, 16);
|
||||
strcpy(info.container_name, container_id.c_str());
|
||||
break;
|
||||
}
|
||||
}
|
||||
if (strlen(info.container_name) > 0) break;
|
||||
}
|
||||
info.pid = (uint32_t)pid;
|
||||
|
||||
if (!pasids.size()) {
|
||||
return AMDSMI_STATUS_NOT_FOUND;
|
||||
if (!pasids.size()) {
|
||||
return AMDSMI_STATUS_NOT_FOUND;
|
||||
}
|
||||
|
||||
return AMDSMI_STATUS_SUCCESS;
|
||||
return AMDSMI_STATUS_SUCCESS;
|
||||
}
|
||||
|
||||
|
||||
} // extern "C"
|
||||
} // extern "C"
|
||||
|
||||
Reference in New Issue
Block a user