Do not initialize Sysman if OsSysman Init failed

Related-To: LOCI-2552

Signed-off-by: Jitendra Sharma <jitendra.sharma@intel.com>
This commit is contained in:
Jitendra Sharma
2021-09-15 17:20:49 +05:30
committed by Compute-Runtime-Automation
parent aba7d74bcd
commit c46f591a99
12 changed files with 436 additions and 65 deletions

View File

@@ -14,13 +14,15 @@ namespace L0 {
ze_result_t LinuxSysmanImp::init() {
pFwUtilInterface = FirmwareUtil::create();
pFsAccess = FsAccess::create();
UNRECOVERABLE_IF(nullptr == pFsAccess);
DEBUG_BREAK_IF(nullptr == pFsAccess);
pProcfsAccess = ProcfsAccess::create();
UNRECOVERABLE_IF(nullptr == pProcfsAccess);
if (pProcfsAccess == nullptr) {
pProcfsAccess = ProcfsAccess::create();
}
DEBUG_BREAK_IF(nullptr == pProcfsAccess);
pDevice = Device::fromHandle(pParentSysmanDeviceImp->hCoreDevice);
UNRECOVERABLE_IF(nullptr == pDevice);
DEBUG_BREAK_IF(nullptr == pDevice);
NEO::OSInterface &OsInterface = pDevice->getOsInterface();
pDrm = OsInterface.getDriverModel()->as<NEO::Drm>();
int myDeviceFd = pDrm->getFileDescriptor();
@@ -30,8 +32,10 @@ ze_result_t LinuxSysmanImp::init() {
return result;
}
pSysfsAccess = SysfsAccess::create(myDeviceName);
UNRECOVERABLE_IF(nullptr == pSysfsAccess);
if (pSysfsAccess == nullptr) {
pSysfsAccess = SysfsAccess::create(myDeviceName);
}
DEBUG_BREAK_IF(nullptr == pSysfsAccess);
std::string realRootPath;
result = pSysfsAccess->getRealPath("device", realRootPath);
@@ -42,7 +46,7 @@ ze_result_t LinuxSysmanImp::init() {
PlatformMonitoringTech::create(pParentSysmanDeviceImp->deviceHandles, pFsAccess, rootPciPathOfGpuDevice, mapOfSubDeviceIdToPmtObject);
pPmuInterface = PmuInterface::create(this);
UNRECOVERABLE_IF(nullptr == pPmuInterface);
DEBUG_BREAK_IF(nullptr == pPmuInterface);
return ZE_RESULT_SUCCESS;
}

View File

@@ -18,8 +18,11 @@ namespace L0 {
SysmanDevice *SysmanDeviceHandleContext::init(ze_device_handle_t coreDevice) {
SysmanDeviceImp *sysmanDevice = new SysmanDeviceImp(coreDevice);
UNRECOVERABLE_IF(!sysmanDevice);
sysmanDevice->init();
DEBUG_BREAK_IF(!sysmanDevice);
if (ZE_RESULT_SUCCESS != sysmanDevice->init()) {
delete sysmanDevice;
sysmanDevice = nullptr;
}
L0::DeviceImp *device = static_cast<DeviceImp *>(Device::fromHandle(coreDevice));
for (auto &subDevice : device->subDevices) {
static_cast<DeviceImp *>(subDevice)->setSysmanHandle(sysmanDevice);
@@ -87,4 +90,179 @@ ze_result_t DriverHandleImp::sysmanEventsListenEx(
return ZE_RESULT_SUCCESS;
}
ze_result_t SysmanDevice::performanceGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_perf_handle_t *phPerformance) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->performanceGet(pCount, phPerformance);
}
ze_result_t SysmanDevice::powerGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_pwr_handle_t *phPower) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->powerGet(pCount, phPower);
}
ze_result_t SysmanDevice::frequencyGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_freq_handle_t *phFrequency) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->frequencyGet(pCount, phFrequency);
}
ze_result_t SysmanDevice::fabricPortGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_fabric_port_handle_t *phPort) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->fabricPortGet(pCount, phPort);
}
ze_result_t SysmanDevice::temperatureGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_temp_handle_t *phTemperature) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->temperatureGet(pCount, phTemperature);
}
ze_result_t SysmanDevice::standbyGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_standby_handle_t *phStandby) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->standbyGet(pCount, phStandby);
}
ze_result_t SysmanDevice::deviceGetProperties(zes_device_handle_t hDevice, zes_device_properties_t *pProperties) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->deviceGetProperties(pProperties);
}
ze_result_t SysmanDevice::processesGetState(zes_device_handle_t hDevice, uint32_t *pCount, zes_process_state_t *pProcesses) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->processesGetState(pCount, pProcesses);
}
ze_result_t SysmanDevice::deviceReset(zes_device_handle_t hDevice, ze_bool_t force) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->deviceReset(force);
}
ze_result_t SysmanDevice::deviceGetState(zes_device_handle_t hDevice, zes_device_state_t *pState) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->deviceGetState(pState);
}
ze_result_t SysmanDevice::engineGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_engine_handle_t *phEngine) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->engineGet(pCount, phEngine);
}
ze_result_t SysmanDevice::pciGetProperties(zes_device_handle_t hDevice, zes_pci_properties_t *pProperties) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->pciGetProperties(pProperties);
}
ze_result_t SysmanDevice::pciGetState(zes_device_handle_t hDevice, zes_pci_state_t *pState) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->pciGetState(pState);
}
ze_result_t SysmanDevice::pciGetBars(zes_device_handle_t hDevice, uint32_t *pCount, zes_pci_bar_properties_t *pProperties) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->pciGetBars(pCount, pProperties);
}
ze_result_t SysmanDevice::pciGetStats(zes_device_handle_t hDevice, zes_pci_stats_t *pStats) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->pciGetStats(pStats);
}
ze_result_t SysmanDevice::schedulerGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_sched_handle_t *phScheduler) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->schedulerGet(pCount, phScheduler);
}
ze_result_t SysmanDevice::rasGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_ras_handle_t *phRas) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->rasGet(pCount, phRas);
}
ze_result_t SysmanDevice::memoryGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_mem_handle_t *phMemory) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->memoryGet(pCount, phMemory);
}
ze_result_t SysmanDevice::fanGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_fan_handle_t *phFan) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->fanGet(pCount, phFan);
}
ze_result_t SysmanDevice::diagnosticsGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_diag_handle_t *phDiagnostics) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->diagnosticsGet(pCount, phDiagnostics);
}
ze_result_t SysmanDevice::firmwareGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_firmware_handle_t *phFirmware) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->firmwareGet(pCount, phFirmware);
}
ze_result_t SysmanDevice::deviceEventRegister(zes_device_handle_t hDevice, zes_event_type_flags_t events) {
auto pSysmanDevice = L0::SysmanDevice::fromHandle(hDevice);
if (pSysmanDevice == nullptr) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
return pSysmanDevice->deviceEventRegister(events);
}
} // namespace L0

View File

@@ -31,6 +31,29 @@ struct Device;
struct SysmanDevice : _ze_device_handle_t {
static SysmanDevice *fromHandle(zes_device_handle_t handle) { return Device::fromHandle(handle)->getSysmanHandle(); }
static ze_result_t performanceGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_perf_handle_t *phPerformance);
static ze_result_t powerGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_pwr_handle_t *phPower);
static ze_result_t frequencyGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_freq_handle_t *phFrequency);
static ze_result_t fabricPortGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_fabric_port_handle_t *phPort);
static ze_result_t temperatureGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_temp_handle_t *phTemperature);
static ze_result_t standbyGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_standby_handle_t *phStandby);
static ze_result_t deviceGetProperties(zes_device_handle_t hDevice, zes_device_properties_t *pProperties);
static ze_result_t processesGetState(zes_device_handle_t hDevice, uint32_t *pCount, zes_process_state_t *pProcesses);
static ze_result_t deviceReset(zes_device_handle_t hDevice, ze_bool_t force);
static ze_result_t deviceGetState(zes_device_handle_t hDevice, zes_device_state_t *pState);
static ze_result_t engineGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_engine_handle_t *phEngine);
static ze_result_t pciGetProperties(zes_device_handle_t hDevice, zes_pci_properties_t *pProperties);
static ze_result_t pciGetState(zes_device_handle_t hDevice, zes_pci_state_t *pState);
static ze_result_t pciGetBars(zes_device_handle_t hDevice, uint32_t *pCount, zes_pci_bar_properties_t *pProperties);
static ze_result_t pciGetStats(zes_device_handle_t hDevice, zes_pci_stats_t *pStats);
static ze_result_t schedulerGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_sched_handle_t *phScheduler);
static ze_result_t rasGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_ras_handle_t *phRas);
static ze_result_t memoryGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_mem_handle_t *phMemory);
static ze_result_t fanGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_fan_handle_t *phFan);
static ze_result_t diagnosticsGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_diag_handle_t *phDiagnostics);
static ze_result_t firmwareGet(zes_device_handle_t hDevice, uint32_t *pCount, zes_firmware_handle_t *phFirmware);
static ze_result_t deviceEventRegister(zes_device_handle_t hDevice, zes_event_type_flags_t events);
static bool deviceEventListen(zes_device_handle_t hDevice, zes_event_type_flags_t &pEvent, uint64_t timeout);
virtual ze_result_t performanceGet(uint32_t *pCount, zes_perf_handle_t *phPerformance) = 0;
virtual ze_result_t powerGet(uint32_t *pCount, zes_pwr_handle_t *phPower) = 0;

View File

@@ -60,7 +60,7 @@ SysmanDeviceImp::~SysmanDeviceImp() {
freeResource(pOsSysman);
}
void SysmanDeviceImp::init() {
ze_result_t SysmanDeviceImp::init() {
uint32_t subDeviceCount = 0;
// We received a device handle. Check for subdevices in this device
Device::fromHandle(hCoreDevice)->getSubDevices(&subDeviceCount, nullptr);
@@ -71,7 +71,10 @@ void SysmanDeviceImp::init() {
Device::fromHandle(hCoreDevice)->getSubDevices(&subDeviceCount, deviceHandles.data());
}
pOsSysman->init();
auto result = pOsSysman->init();
if (ZE_RESULT_SUCCESS != result) {
return result;
}
if (pPowerHandleContext) {
pPowerHandleContext->init();
}
@@ -120,6 +123,7 @@ void SysmanDeviceImp::init() {
if (pPerformanceHandleContext) {
pPerformanceHandleContext->init(deviceHandles);
}
return result;
}
ze_result_t SysmanDeviceImp::frequencyGet(uint32_t *pCount, zes_freq_handle_t *phFrequency) {

View File

@@ -22,7 +22,7 @@ struct SysmanDeviceImp : SysmanDevice, NEO::NonCopyableOrMovableClass {
~SysmanDeviceImp() override;
SysmanDeviceImp() = delete;
void init();
ze_result_t init();
ze_device_handle_t hCoreDevice = nullptr;
std::vector<ze_device_handle_t> deviceHandles; // handles of all subdevices