refactor: Merge Ras prelim files with non-prelim files

Related-To: NEO-9469

Signed-off-by: Bellekallu Rajkiran <bellekallu.rajkiran@intel.com>
This commit is contained in:
Bellekallu Rajkiran
2023-11-09 13:20:25 +00:00
committed by Compute-Runtime-Automation
parent 616ef4c9c7
commit 16725e2438
12 changed files with 695 additions and 1022 deletions

View File

@@ -8,21 +8,9 @@ if(UNIX)
target_sources(${L0_STATIC_LIB_NAME}
PRIVATE
${CMAKE_CURRENT_SOURCE_DIR}/CMakeLists.txt
${CMAKE_CURRENT_SOURCE_DIR}/sysman_os_ras_imp.cpp
${CMAKE_CURRENT_SOURCE_DIR}/sysman_os_ras_imp.h
${CMAKE_CURRENT_SOURCE_DIR}/sysman_os_ras_imp_gt.cpp
${CMAKE_CURRENT_SOURCE_DIR}/sysman_os_ras_imp_hbm.cpp
)
if(NEO_ENABLE_i915_PRELIM_DETECTION)
target_sources(${L0_STATIC_LIB_NAME}
PRIVATE
${CMAKE_CURRENT_SOURCE_DIR}/sysman_os_ras_imp_prelim.cpp
${CMAKE_CURRENT_SOURCE_DIR}/sysman_os_ras_imp_prelim.h
${CMAKE_CURRENT_SOURCE_DIR}/sysman_os_ras_imp_gt.cpp
${CMAKE_CURRENT_SOURCE_DIR}/sysman_os_ras_imp_hbm.cpp
)
else()
target_sources(${L0_STATIC_LIB_NAME}
PRIVATE
${CMAKE_CURRENT_SOURCE_DIR}/sysman_os_ras_imp.cpp
${CMAKE_CURRENT_SOURCE_DIR}/sysman_os_ras_imp.h
)
endif()
endif()

View File

@@ -8,35 +8,46 @@
#include "level_zero/sysman/source/api/ras/linux/sysman_os_ras_imp.h"
#include "shared/source/debug_settings/debug_settings_manager.h"
#include "shared/source/helpers/string.h"
#include "shared/source/os_interface/linux/system_info.h"
#include "level_zero/sysman/source/shared/linux/zes_os_sysman_imp.h"
#include <cstring>
#include "drm/intel_hwconfig_types.h"
namespace L0 {
namespace Sysman {
LinuxRasImp::LinuxRasImp(OsSysman *pOsSysman, zes_ras_error_type_t type, ze_bool_t onSubdevice, uint32_t subdeviceId) : osRasErrorType(type), isSubdevice(onSubdevice), subdeviceId(subdeviceId) {
pLinuxSysmanImp = static_cast<LinuxSysmanImp *>(pOsSysman);
pFsAccess = &pLinuxSysmanImp->getFsAccess();
static bool isMemoryTypeHbm(LinuxSysmanImp *pLinuxSysmanImp) {
uint32_t memType = pLinuxSysmanImp->getMemoryType();
if (memType == INTEL_HWCONFIG_MEMORY_TYPE_HBM2e || memType == INTEL_HWCONFIG_MEMORY_TYPE_HBM2) {
return true;
}
return false;
}
void OsRas::getSupportedRasErrorTypes(std::set<zes_ras_error_type_t> &errorType, OsSysman *pOsSysman, ze_bool_t isSubDevice, uint32_t subDeviceId) {}
void OsRas::getSupportedRasErrorTypes(std::set<zes_ras_error_type_t> &errorType, OsSysman *pOsSysman, ze_bool_t isSubDevice, uint32_t subDeviceId) {
ze_result_t LinuxRasImp::osRasGetState(zes_ras_state_t &state, ze_bool_t clear) {
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE;
constexpr auto maxErrorTypes = 2;
LinuxRasSourceGt::getSupportedRasErrorTypes(errorType, pOsSysman, isSubDevice, subDeviceId);
if (errorType.size() < maxErrorTypes) {
auto pLinuxSysmanImp = static_cast<LinuxSysmanImp *>(pOsSysman);
if (isMemoryTypeHbm(pLinuxSysmanImp) == true) {
LinuxRasSourceHbm::getSupportedRasErrorTypes(errorType, pOsSysman, isSubDevice, subDeviceId);
}
}
}
ze_result_t LinuxRasImp::osRasGetConfig(zes_ras_config_t *config) {
config->totalThreshold = totalThreshold;
memcpy(config->detailedThresholds.category, categoryThreshold, maxRasErrorCategoryCount * sizeof(uint64_t));
memcpy_s(config->detailedThresholds.category, maxRasErrorCategoryCount * sizeof(uint64_t), categoryThreshold, maxRasErrorCategoryCount * sizeof(uint64_t));
return ZE_RESULT_SUCCESS;
}
ze_result_t LinuxRasImp::osRasSetConfig(const zes_ras_config_t *config) {
if (pFsAccess->isRootUser() == true) {
totalThreshold = config->totalThreshold;
memcpy(categoryThreshold, config->detailedThresholds.category, maxRasErrorCategoryCount * sizeof(uint64_t));
memcpy_s(categoryThreshold, maxRasErrorCategoryCount * sizeof(uint64_t), config->detailedThresholds.category, maxRasErrorCategoryCount * sizeof(uint64_t));
return ZE_RESULT_SUCCESS;
}
NEO::printDebugString(NEO::DebugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s(): Insufficient permissions and returning error:0x%x \n", __FUNCTION__, ZE_RESULT_ERROR_INSUFFICIENT_PERMISSIONS);
@@ -51,6 +62,42 @@ ze_result_t LinuxRasImp::osRasGetProperties(zes_ras_properties_t &properties) {
return ZE_RESULT_SUCCESS;
}
ze_result_t LinuxRasImp::osRasGetState(zes_ras_state_t &state, ze_bool_t clear) {
if (clear == true) {
if (pFsAccess->isRootUser() == false) {
NEO::printDebugString(NEO::DebugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s(): Insufficient permissions and returning error:0x%x \n", __FUNCTION__, ZE_RESULT_ERROR_INSUFFICIENT_PERMISSIONS);
return ZE_RESULT_ERROR_INSUFFICIENT_PERMISSIONS;
}
}
ze_result_t result = ZE_RESULT_ERROR_UNSUPPORTED_FEATURE;
for (auto &rasSource : rasSources) {
zes_ras_state_t localState = {};
ze_result_t localResult = rasSource->osRasGetState(localState, clear);
if (localResult != ZE_RESULT_SUCCESS) {
continue;
}
for (uint32_t i = 0; i < maxRasErrorCategoryCount; i++) {
state.category[i] += localState.category[i];
}
result = ZE_RESULT_SUCCESS;
}
return result;
}
void LinuxRasImp::initSources() {
rasSources.push_back(std::make_unique<L0::Sysman::LinuxRasSourceGt>(pLinuxSysmanImp, osRasErrorType, isSubdevice, subdeviceId));
if (isMemoryTypeHbm(pLinuxSysmanImp) == true) {
rasSources.push_back(std::make_unique<L0::Sysman::LinuxRasSourceHbm>(pLinuxSysmanImp, osRasErrorType, subdeviceId));
}
}
LinuxRasImp::LinuxRasImp(OsSysman *pOsSysman, zes_ras_error_type_t type, ze_bool_t onSubdevice, uint32_t subdeviceId) : osRasErrorType(type), isSubdevice(onSubdevice), subdeviceId(subdeviceId) {
pLinuxSysmanImp = static_cast<LinuxSysmanImp *>(pOsSysman);
pFsAccess = &pLinuxSysmanImp->getFsAccess();
initSources();
}
OsRas *OsRas::create(OsSysman *pOsSysman, zes_ras_error_type_t type, ze_bool_t onSubdevice, uint32_t subdeviceId) {
LinuxRasImp *pLinuxRasImp = new LinuxRasImp(pOsSysman, type, onSubdevice, subdeviceId);
return static_cast<OsRas *>(pLinuxRasImp);

View File

@@ -9,12 +9,28 @@
#include "shared/source/helpers/non_copyable_or_moveable.h"
#include "level_zero/sysman/source/api/ras/sysman_os_ras.h"
#include "level_zero/sysman/source/device/sysman_device_imp.h"
#include "level_zero/sysman/source/shared/linux/pmu/sysman_pmu_imp.h"
#include "level_zero/sysman/source/shared/linux/sysman_fs_access.h"
#include "level_zero/sysman/source/sysman_const.h"
#include <map>
#include <memory>
#include <string>
#include <vector>
namespace L0 {
namespace Sysman {
class LinuxSysmanImp;
class FirmwareUtil;
class LinuxRasSources : NEO::NonCopyableOrMovableClass {
public:
virtual ze_result_t osRasGetState(zes_ras_state_t &state, ze_bool_t clear) = 0;
virtual ~LinuxRasSources() = default;
};
class LinuxRasImp : public OsRas, NEO::NonCopyableOrMovableClass {
public:
ze_result_t osRasGetProperties(zes_ras_properties_t &properties) override;
@@ -29,13 +45,70 @@ class LinuxRasImp : public OsRas, NEO::NonCopyableOrMovableClass {
zes_ras_error_type_t osRasErrorType = {};
FsAccess *pFsAccess = nullptr;
LinuxSysmanImp *pLinuxSysmanImp = nullptr;
std::vector<std::unique_ptr<L0::Sysman::LinuxRasSources>> rasSources = {};
private:
void initSources();
bool isSubdevice = false;
uint32_t subdeviceId = 0;
uint64_t totalThreshold = 0;
uint64_t categoryThreshold[maxRasErrorCategoryCount] = {0};
};
class LinuxRasSourceGt : public LinuxRasSources {
public:
ze_result_t osRasGetState(zes_ras_state_t &state, ze_bool_t clear) override;
static void getSupportedRasErrorTypes(std::set<zes_ras_error_type_t> &errorType, OsSysman *pOsSysman, ze_bool_t isSubDevice, uint32_t subDeviceId);
LinuxRasSourceGt(LinuxSysmanImp *pLinuxSysmanImp, zes_ras_error_type_t type, ze_bool_t onSubdevice, uint32_t subdeviceId);
LinuxRasSourceGt() = default;
~LinuxRasSourceGt() override;
protected:
LinuxSysmanImp *pLinuxSysmanImp = nullptr;
zes_ras_error_type_t osRasErrorType = {};
PmuInterface *pPmuInterface = nullptr;
FsAccess *pFsAccess = nullptr;
SysfsAccess *pSysfsAccess = nullptr;
private:
void initRasErrors(ze_bool_t clear);
ze_result_t getPmuConfig(
const std::string &eventDirectory,
const std::vector<std::string> &listOfEvents,
const std::string &errorFileToGetConfig,
std::string &pmuConfig);
ze_result_t getBootUpErrorCountFromSysfs(
std::string nameOfError,
const std::string &errorCounterDir,
uint64_t &errorVal);
void closeFds();
int64_t groupFd = -1;
std::vector<int64_t> memberFds = {};
uint64_t initialErrorCount[maxRasErrorCategoryCount] = {0};
std::map<zes_ras_error_cat_t, uint64_t> errorCategoryToEventCount;
uint64_t totalEventCount = 0;
bool isSubdevice = false;
uint32_t subdeviceId = 0;
};
class LinuxRasSourceHbm : public LinuxRasSources {
public:
ze_result_t osRasGetState(zes_ras_state_t &state, ze_bool_t clear) override;
static void getSupportedRasErrorTypes(std::set<zes_ras_error_type_t> &errorType, OsSysman *pOsSysman, ze_bool_t isSubDevice, uint32_t subDeviceId);
LinuxRasSourceHbm(LinuxSysmanImp *pLinuxSysmanImp, zes_ras_error_type_t type, uint32_t subdeviceId);
LinuxRasSourceHbm() = default;
~LinuxRasSourceHbm() override{};
protected:
LinuxSysmanImp *pLinuxSysmanImp = nullptr;
zes_ras_error_type_t osRasErrorType = {};
FirmwareUtil *pFwInterface = nullptr;
SysmanDeviceImp *pDevice = nullptr;
private:
uint64_t errorBaseline = 0;
uint32_t subdeviceId = 0;
};
} // namespace Sysman
} // namespace L0

View File

@@ -7,7 +7,7 @@
#include "shared/source/debug_settings/debug_settings_manager.h"
#include "level_zero/sysman/source/api/ras/linux/sysman_os_ras_imp_prelim.h"
#include "level_zero/sysman/source/api/ras/linux/sysman_os_ras_imp.h"
#include "level_zero/sysman/source/shared/linux/zes_os_sysman_imp.h"
#include <cstring>

View File

@@ -8,7 +8,7 @@
#include "shared/source/debug_settings/debug_settings_manager.h"
#include "shared/source/helpers/gfx_core_helper.h"
#include "level_zero/sysman/source/api/ras/linux/sysman_os_ras_imp_prelim.h"
#include "level_zero/sysman/source/api/ras/linux/sysman_os_ras_imp.h"
#include "level_zero/sysman/source/shared/firmware_util/sysman_firmware_util.h"
#include "level_zero/sysman/source/shared/linux/zes_os_sysman_imp.h"

View File

@@ -1,107 +0,0 @@
/*
* Copyright (C) 2023 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "level_zero/sysman/source/api/ras/linux/sysman_os_ras_imp_prelim.h"
#include "shared/source/debug_settings/debug_settings_manager.h"
#include "shared/source/helpers/string.h"
#include "shared/source/os_interface/linux/system_info.h"
#include "level_zero/sysman/source/shared/linux/zes_os_sysman_imp.h"
#include "drm/intel_hwconfig_types.h"
namespace L0 {
namespace Sysman {
static bool isMemoryTypeHbm(LinuxSysmanImp *pLinuxSysmanImp) {
uint32_t memType = pLinuxSysmanImp->getMemoryType();
if (memType == INTEL_HWCONFIG_MEMORY_TYPE_HBM2e || memType == INTEL_HWCONFIG_MEMORY_TYPE_HBM2) {
return true;
}
return false;
}
void OsRas::getSupportedRasErrorTypes(std::set<zes_ras_error_type_t> &errorType, OsSysman *pOsSysman, ze_bool_t isSubDevice, uint32_t subDeviceId) {
constexpr auto maxErrorTypes = 2;
LinuxRasSourceGt::getSupportedRasErrorTypes(errorType, pOsSysman, isSubDevice, subDeviceId);
if (errorType.size() < maxErrorTypes) {
auto pLinuxSysmanImp = static_cast<LinuxSysmanImp *>(pOsSysman);
if (isMemoryTypeHbm(pLinuxSysmanImp) == true) {
LinuxRasSourceHbm::getSupportedRasErrorTypes(errorType, pOsSysman, isSubDevice, subDeviceId);
}
}
}
ze_result_t LinuxRasImp::osRasGetConfig(zes_ras_config_t *config) {
config->totalThreshold = totalThreshold;
memcpy_s(config->detailedThresholds.category, maxRasErrorCategoryCount * sizeof(uint64_t), categoryThreshold, maxRasErrorCategoryCount * sizeof(uint64_t));
return ZE_RESULT_SUCCESS;
}
ze_result_t LinuxRasImp::osRasSetConfig(const zes_ras_config_t *config) {
if (pFsAccess->isRootUser() == true) {
totalThreshold = config->totalThreshold;
memcpy_s(categoryThreshold, maxRasErrorCategoryCount * sizeof(uint64_t), config->detailedThresholds.category, maxRasErrorCategoryCount * sizeof(uint64_t));
return ZE_RESULT_SUCCESS;
}
NEO::printDebugString(NEO::DebugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s(): Insufficient permissions and returning error:0x%x \n", __FUNCTION__, ZE_RESULT_ERROR_INSUFFICIENT_PERMISSIONS);
return ZE_RESULT_ERROR_INSUFFICIENT_PERMISSIONS;
}
ze_result_t LinuxRasImp::osRasGetProperties(zes_ras_properties_t &properties) {
properties.pNext = nullptr;
properties.type = osRasErrorType;
properties.onSubdevice = isSubdevice;
properties.subdeviceId = subdeviceId;
return ZE_RESULT_SUCCESS;
}
ze_result_t LinuxRasImp::osRasGetState(zes_ras_state_t &state, ze_bool_t clear) {
if (clear == true) {
if (pFsAccess->isRootUser() == false) {
NEO::printDebugString(NEO::DebugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s(): Insufficient permissions and returning error:0x%x \n", __FUNCTION__, ZE_RESULT_ERROR_INSUFFICIENT_PERMISSIONS);
return ZE_RESULT_ERROR_INSUFFICIENT_PERMISSIONS;
}
}
ze_result_t result = ZE_RESULT_ERROR_UNSUPPORTED_FEATURE;
for (auto &rasSource : rasSources) {
zes_ras_state_t localState = {};
ze_result_t localResult = rasSource->osRasGetState(localState, clear);
if (localResult != ZE_RESULT_SUCCESS) {
continue;
}
for (uint32_t i = 0; i < maxRasErrorCategoryCount; i++) {
state.category[i] += localState.category[i];
}
result = ZE_RESULT_SUCCESS;
}
return result;
}
void LinuxRasImp::initSources() {
rasSources.push_back(std::make_unique<L0::Sysman::LinuxRasSourceGt>(pLinuxSysmanImp, osRasErrorType, isSubdevice, subdeviceId));
if (isMemoryTypeHbm(pLinuxSysmanImp) == true) {
rasSources.push_back(std::make_unique<L0::Sysman::LinuxRasSourceHbm>(pLinuxSysmanImp, osRasErrorType, subdeviceId));
}
}
LinuxRasImp::LinuxRasImp(OsSysman *pOsSysman, zes_ras_error_type_t type, ze_bool_t onSubdevice, uint32_t subdeviceId) : osRasErrorType(type), isSubdevice(onSubdevice), subdeviceId(subdeviceId) {
pLinuxSysmanImp = static_cast<LinuxSysmanImp *>(pOsSysman);
pFsAccess = &pLinuxSysmanImp->getFsAccess();
initSources();
}
OsRas *OsRas::create(OsSysman *pOsSysman, zes_ras_error_type_t type, ze_bool_t onSubdevice, uint32_t subdeviceId) {
LinuxRasImp *pLinuxRasImp = new LinuxRasImp(pOsSysman, type, onSubdevice, subdeviceId);
return static_cast<OsRas *>(pLinuxRasImp);
}
} // namespace Sysman
} // namespace L0

View File

@@ -1,114 +0,0 @@
/*
* Copyright (C) 2023 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#pragma once
#include "shared/source/helpers/non_copyable_or_moveable.h"
#include "level_zero/sysman/source/api/ras/sysman_os_ras.h"
#include "level_zero/sysman/source/device/sysman_device_imp.h"
#include "level_zero/sysman/source/shared/linux/pmu/sysman_pmu_imp.h"
#include "level_zero/sysman/source/shared/linux/sysman_fs_access.h"
#include "level_zero/sysman/source/sysman_const.h"
#include <map>
#include <memory>
#include <string>
#include <vector>
namespace L0 {
namespace Sysman {
class LinuxSysmanImp;
class FirmwareUtil;
class LinuxRasSources : NEO::NonCopyableOrMovableClass {
public:
virtual ze_result_t osRasGetState(zes_ras_state_t &state, ze_bool_t clear) = 0;
virtual ~LinuxRasSources() = default;
};
class LinuxRasImp : public OsRas, NEO::NonCopyableOrMovableClass {
public:
ze_result_t osRasGetProperties(zes_ras_properties_t &properties) override;
ze_result_t osRasGetState(zes_ras_state_t &state, ze_bool_t clear) override;
ze_result_t osRasGetConfig(zes_ras_config_t *config) override;
ze_result_t osRasSetConfig(const zes_ras_config_t *config) override;
LinuxRasImp(OsSysman *pOsSysman, zes_ras_error_type_t type, ze_bool_t onSubdevice, uint32_t subdeviceId);
LinuxRasImp() = default;
~LinuxRasImp() override = default;
protected:
zes_ras_error_type_t osRasErrorType = {};
FsAccess *pFsAccess = nullptr;
LinuxSysmanImp *pLinuxSysmanImp = nullptr;
std::vector<std::unique_ptr<L0::Sysman::LinuxRasSources>> rasSources = {};
private:
void initSources();
bool isSubdevice = false;
uint32_t subdeviceId = 0;
uint64_t totalThreshold = 0;
uint64_t categoryThreshold[maxRasErrorCategoryCount] = {0};
};
class LinuxRasSourceGt : public LinuxRasSources {
public:
ze_result_t osRasGetState(zes_ras_state_t &state, ze_bool_t clear) override;
static void getSupportedRasErrorTypes(std::set<zes_ras_error_type_t> &errorType, OsSysman *pOsSysman, ze_bool_t isSubDevice, uint32_t subDeviceId);
LinuxRasSourceGt(LinuxSysmanImp *pLinuxSysmanImp, zes_ras_error_type_t type, ze_bool_t onSubdevice, uint32_t subdeviceId);
LinuxRasSourceGt() = default;
~LinuxRasSourceGt() override;
protected:
LinuxSysmanImp *pLinuxSysmanImp = nullptr;
zes_ras_error_type_t osRasErrorType = {};
PmuInterface *pPmuInterface = nullptr;
FsAccess *pFsAccess = nullptr;
SysfsAccess *pSysfsAccess = nullptr;
private:
void initRasErrors(ze_bool_t clear);
ze_result_t getPmuConfig(
const std::string &eventDirectory,
const std::vector<std::string> &listOfEvents,
const std::string &errorFileToGetConfig,
std::string &pmuConfig);
ze_result_t getBootUpErrorCountFromSysfs(
std::string nameOfError,
const std::string &errorCounterDir,
uint64_t &errorVal);
void closeFds();
int64_t groupFd = -1;
std::vector<int64_t> memberFds = {};
uint64_t initialErrorCount[maxRasErrorCategoryCount] = {0};
std::map<zes_ras_error_cat_t, uint64_t> errorCategoryToEventCount;
uint64_t totalEventCount = 0;
bool isSubdevice = false;
uint32_t subdeviceId = 0;
};
class LinuxRasSourceHbm : public LinuxRasSources {
public:
ze_result_t osRasGetState(zes_ras_state_t &state, ze_bool_t clear) override;
static void getSupportedRasErrorTypes(std::set<zes_ras_error_type_t> &errorType, OsSysman *pOsSysman, ze_bool_t isSubDevice, uint32_t subDeviceId);
LinuxRasSourceHbm(LinuxSysmanImp *pLinuxSysmanImp, zes_ras_error_type_t type, uint32_t subdeviceId);
LinuxRasSourceHbm() = default;
~LinuxRasSourceHbm() override{};
protected:
LinuxSysmanImp *pLinuxSysmanImp = nullptr;
zes_ras_error_type_t osRasErrorType = {};
FirmwareUtil *pFwInterface = nullptr;
SysmanDeviceImp *pDevice = nullptr;
private:
uint64_t errorBaseline = 0;
uint32_t subdeviceId = 0;
};
} // namespace Sysman
} // namespace L0