feature(sysman): Add support for Memory properties & state for iGPU's in Linux

Related-To: NEO-14198

Signed-off-by: Shreyas Kunder <shreyas.kunder@intel.com>
This commit is contained in:
Shreyas Kunder
2025-07-04 09:47:42 +00:00
committed by Compute-Runtime-Automation
parent ff00896801
commit e1eae6a3e5
6 changed files with 178 additions and 20 deletions

View File

@@ -17,6 +17,7 @@
#include "level_zero/sysman/source/shared/firmware_util/sysman_firmware_util.h"
#include "level_zero/sysman/source/shared/linux/kmd_interface/sysman_kmd_interface.h"
#include "level_zero/sysman/source/shared/linux/product_helper/sysman_product_helper.h"
#include "level_zero/sysman/source/shared/linux/sysman_fs_access_interface.h"
#include "level_zero/sysman/source/shared/linux/zes_os_sysman_imp.h"
namespace L0 {
@@ -35,6 +36,22 @@ ze_result_t LinuxMemoryImp::getBandwidth(zes_mem_bandwidth_t *pBandwidth) {
ze_result_t LinuxMemoryImp::getState(zes_mem_state_t *pState) {
ze_result_t status = ZE_RESULT_SUCCESS;
pState->health = ZES_MEM_HEALTH_UNKNOWN;
if (pLinuxSysmanImp->getHardwareInfo().capabilityTable.isIntegratedDevice) {
const std::string memFreeKey = "MemFree";
const std::string memAvailableKey = "MemAvailable";
auto memInfoValues = readMemInfoValues(&pLinuxSysmanImp->getFsAccess(), {memFreeKey, memAvailableKey});
if (memInfoValues.find(memFreeKey) != memInfoValues.end() && memInfoValues.find(memAvailableKey) != memInfoValues.end()) {
pState->free = memInfoValues[memFreeKey] * 1024;
pState->size = memInfoValues[memAvailableKey] * 1024;
} else {
pState->free = 0;
pState->size = 0;
status = ZE_RESULT_ERROR_UNKNOWN;
}
return status;
}
FirmwareUtil *pFwInterface = pLinuxSysmanImp->getFwUtilInterface();
auto pSysmanProductHelper = pLinuxSysmanImp->getSysmanProductHelper();
pSysmanProductHelper->getMemoryHealthIndicator(pFwInterface, &pState->health);
@@ -60,6 +77,35 @@ ze_result_t LinuxMemoryImp::getState(zes_mem_state_t *pState) {
return status;
}
std::map<std::string, uint64_t> LinuxMemoryImp::readMemInfoValues(FsAccessInterface *pFsAccess, const std::vector<std::string> &keys) {
std::map<std::string, uint64_t> result = {};
const std::string memInfoFile = "/proc/meminfo";
std::vector<std::string> memInfo{};
if (pFsAccess->read(memInfoFile, memInfo) == ZE_RESULT_SUCCESS) {
for (const auto &line : memInfo) {
std::istringstream lineStream(line);
std::string label = "";
std::string unit = "";
uint64_t value = 0;
lineStream >> label >> value >> unit;
if (!label.empty() && label.back() == ':') {
label.pop_back();
}
if (std::find(keys.begin(), keys.end(), label) != keys.end()) {
result[label] = value;
if (result.size() == keys.size()) {
break;
}
}
}
}
return result;
}
LinuxMemoryImp::LinuxMemoryImp(OsSysman *pOsSysman, ze_bool_t onSubdevice, uint32_t subdeviceId) : isSubdevice(onSubdevice), subdeviceId(subdeviceId) {
pLinuxSysmanImp = static_cast<LinuxSysmanImp *>(pOsSysman);
pDrm = pLinuxSysmanImp->getDrm();
@@ -69,6 +115,9 @@ LinuxMemoryImp::LinuxMemoryImp(OsSysman *pOsSysman, ze_bool_t onSubdevice, uint3
bool LinuxMemoryImp::isMemoryModuleSupported() {
auto &gfxCoreHelper = pDevice->getRootDeviceEnvironment().getHelper<NEO::GfxCoreHelper>();
if (pLinuxSysmanImp->getHardwareInfo().capabilityTable.isIntegratedDevice) {
return true;
}
return gfxCoreHelper.getEnableLocalMemory(pDevice->getHardwareInfo());
}

View File

@@ -10,7 +10,9 @@
#include "level_zero/sysman/source/api/memory/sysman_os_memory.h"
#include <map>
#include <string>
#include <vector>
namespace NEO {
class Drm;
@@ -22,12 +24,14 @@ namespace Sysman {
class LinuxSysmanImp;
class SysmanKmdInterface;
struct SysmanDeviceImp;
class FsAccessInterface;
class LinuxMemoryImp : public OsMemory, NEO::NonCopyableAndNonMovableClass {
public:
ze_result_t getProperties(zes_mem_properties_t *pProperties) override;
ze_result_t getBandwidth(zes_mem_bandwidth_t *pBandwidth) override;
ze_result_t getState(zes_mem_state_t *pState) override;
static std::map<std::string, uint64_t> readMemInfoValues(FsAccessInterface *pFsAccess, const std::vector<std::string> &keys);
bool isMemoryModuleSupported() override;
LinuxMemoryImp(OsSysman *pOsSysman, ze_bool_t onSubdevice, uint32_t subdeviceId);
LinuxMemoryImp() = default;

View File

@@ -9,6 +9,7 @@
#include "shared/source/os_interface/linux/memory_info.h"
#include "shared/source/os_interface/linux/system_info.h"
#include "level_zero/sysman/source/api/memory/linux/sysman_os_memory_imp.h"
#include "level_zero/sysman/source/api/ras/linux/ras_util/sysman_ras_util.h"
#include "level_zero/sysman/source/shared/firmware_util/sysman_firmware_util.h"
#include "level_zero/sysman/source/shared/linux/kmd_interface/sysman_kmd_interface.h"
@@ -37,8 +38,13 @@ void SysmanProductHelperHw<gfxProduct>::getFrequencyStepSize(double *pStepSize)
template <PRODUCT_FAMILY gfxProduct>
ze_result_t SysmanProductHelperHw<gfxProduct>::getMemoryProperties(zes_mem_properties_t *pProperties, LinuxSysmanImp *pLinuxSysmanImp, NEO::Drm *pDrm, SysmanKmdInterface *pSysmanKmdInterface, uint32_t subDeviceId, bool isSubdevice) {
auto pSysFsAccess = pSysmanKmdInterface->getSysFsAccess();
bool isIntegratedDevice = pLinuxSysmanImp->getHardwareInfo().capabilityTable.isIntegratedDevice;
pProperties->location = ZES_MEM_LOC_DEVICE;
if (isIntegratedDevice) {
pProperties->location = ZES_MEM_LOC_SYSTEM;
} else {
pProperties->location = ZES_MEM_LOC_DEVICE;
}
pProperties->type = ZES_MEM_TYPE_DDR;
pProperties->onSubdevice = isSubdevice;
pProperties->subdeviceId = subDeviceId;
@@ -84,7 +90,17 @@ ze_result_t SysmanProductHelperHw<gfxProduct>::getMemoryProperties(zes_mem_prope
pProperties->busWidth = memoryBusWidth;
pProperties->physicalSize = 0;
if (pSysmanKmdInterface->isPhysicalMemorySizeSupported() == true) {
if (isIntegratedDevice) {
pProperties->busWidth = -1;
pProperties->numChannels = -1;
pProperties->type = ZES_MEM_TYPE_FORCE_UINT32;
const std::string memTotalKey = "MemTotal";
auto memInfoValues = LinuxMemoryImp::readMemInfoValues(&pLinuxSysmanImp->getFsAccess(), {memTotalKey});
if (memInfoValues.find(memTotalKey) != memInfoValues.end()) {
pProperties->physicalSize = memInfoValues[memTotalKey] * 1024;
}
} else if (pSysmanKmdInterface->isPhysicalMemorySizeSupported() == true) {
if (isSubdevice) {
std::string memval;
std::string physicalSizeFile = pSysmanKmdInterface->getSysfsFilePathForPhysicalMemorySize(subDeviceId);

View File

@@ -68,6 +68,10 @@ constexpr uint64_t mockIdiWriteVal = 9u;
constexpr uint64_t mockDisplayVc1ReadVal = 10u;
constexpr uint64_t numberMcChannels = 16;
constexpr uint64_t mockIntegratedDeviceAvailableMemory = 8192 * 1024;
constexpr uint64_t mockIntegratedDeviceFreeMemory = 4096 * 1024;
constexpr uint64_t mockIntegratedDevicePhysicalSize = 16384 * 1024;
namespace L0 {
namespace Sysman {
namespace ult {
@@ -167,6 +171,26 @@ struct MockMemorySysFsAccessInterface : public L0::Sysman::SysFsAccessInterface
}
};
struct MockMemoryFsAccessInterface : public L0::Sysman::FsAccessInterface {
bool mockMemInfoIncorrectValue = false;
ze_result_t read(std::string file, std::vector<std::string> &val) override {
if (file == "/proc/meminfo") {
if (mockMemInfoIncorrectValue) {
val.push_back("Buffers: 158772 kB");
val.push_back("Cached: 11744244 kB");
val.push_back("SwapCached: 1376 kB");
val.push_back("Active: 6777644 kB");
} else {
val.push_back("MemTotal: 16384 kB");
val.push_back("MemFree: 4096 kB");
val.push_back("MemAvailable: 8192 kB");
}
}
return ZE_RESULT_SUCCESS;
}
MockMemoryFsAccessInterface() = default;
};
class MockProcFsAccessInterface : public L0::Sysman::ProcFsAccessInterface {
public:
MockProcFsAccessInterface() = default;

View File

@@ -202,22 +202,38 @@ TEST_F(SysmanDeviceMemoryFixtureI915, GivenI915DriverVersionWhenValidCallingSysf
EXPECT_STREQ("gt/gt0/mem_RPn_freq_mhz", pSysmanKmdInterface->getSysfsFilePath(SysfsName::sysfsNameMinMemoryFrequency, 0, true).c_str());
}
TEST_F(SysmanDeviceMemoryFixtureI915, GivenComponentCountZeroWhenEnumeratingMemoryModulesWithNoLocalMemorySupportThenZeroCountIsReturned) {
setLocalSupportedAndReinit(false);
TEST_F(SysmanDeviceMemoryFixtureI915, GivenComponentCountZeroWhenEnumeratingMemoryModulesWithNoLocalMemorySupportThenZeroCountIsReturnedForDiscretePlatforms) {
if (!defaultHwInfo->capabilityTable.isIntegratedDevice) {
setLocalSupportedAndReinit(false);
}
uint32_t count = 0;
EXPECT_EQ(zesDeviceEnumMemoryModules(device->toHandle(), &count, nullptr), ZE_RESULT_SUCCESS);
EXPECT_EQ(count, 0u);
if (defaultHwInfo->capabilityTable.isIntegratedDevice) {
EXPECT_EQ(count, 1u);
} else {
EXPECT_EQ(count, 0u);
}
}
TEST_F(SysmanDeviceMemoryFixtureI915, GivenInvalidComponentCountWhenEnumeratingMemoryModulesWithNoLocalMemorySupportThenZeroCountIsReturned) {
setLocalSupportedAndReinit(false);
TEST_F(SysmanDeviceMemoryFixtureI915, GivenInvalidComponentCountWhenEnumeratingMemoryModulesWithNoLocalMemorySupportThenZeroCountIsReturnedForDiscretePlatforms) {
if (!defaultHwInfo->capabilityTable.isIntegratedDevice) {
setLocalSupportedAndReinit(false);
}
uint32_t count = 0;
EXPECT_EQ(zesDeviceEnumMemoryModules(device->toHandle(), &count, nullptr), ZE_RESULT_SUCCESS);
EXPECT_EQ(count, 0u);
if (defaultHwInfo->capabilityTable.isIntegratedDevice) {
EXPECT_EQ(count, 1u);
} else {
EXPECT_EQ(count, 0u);
}
count = count + 1;
EXPECT_EQ(zesDeviceEnumMemoryModules(device->toHandle(), &count, nullptr), ZE_RESULT_SUCCESS);
EXPECT_EQ(count, 0u);
if (defaultHwInfo->capabilityTable.isIntegratedDevice) {
EXPECT_EQ(count, 1u);
} else {
EXPECT_EQ(count, 0u);
}
}
HWTEST2_F(SysmanDeviceMemoryFixtureI915, GivenComponentCountZeroWhenEnumeratingMemoryModulesThenValidCountIsReturned, IsPVC) {
@@ -386,7 +402,11 @@ TEST_F(SysmanDeviceMemoryFixtureI915, GivenValidMemoryHandleWhenCallingZesMemory
ze_result_t result = zesMemoryGetProperties(handle, &properties);
EXPECT_EQ(result, ZE_RESULT_SUCCESS);
EXPECT_EQ(properties.type, ZES_MEM_TYPE_DDR);
EXPECT_EQ(properties.location, ZES_MEM_LOC_DEVICE);
if (defaultHwInfo->capabilityTable.isIntegratedDevice) {
EXPECT_EQ(properties.location, ZES_MEM_LOC_SYSTEM);
} else {
EXPECT_EQ(properties.location, ZES_MEM_LOC_DEVICE);
}
EXPECT_EQ(properties.numChannels, -1);
EXPECT_FALSE(properties.onSubdevice);
EXPECT_EQ(properties.subdeviceId, 0u);
@@ -682,6 +702,9 @@ HWTEST2_F(SysmanDeviceMemoryFixtureI915, GivenValidMemoryHandleWhenCallingZesSys
}
TEST_F(SysmanDeviceMemoryFixtureI915, GivenValidMemoryHandleWhenCallingZesSysmanMemoryGetStateAndIoctlReturnedErrorThenApiReturnsError) {
if (defaultHwInfo->capabilityTable.isIntegratedDevice) {
GTEST_SKIP() << "This test is not applicable for integrated devices";
}
auto ioctlHelper = static_cast<SysmanMemoryMockIoctlHelper *>(pDrm->ioctlHelper.get());
ioctlHelper->returnEmptyMemoryInfo = true;
ioctlHelper->mockErrorNumber = EBUSY;
@@ -699,6 +722,9 @@ TEST_F(SysmanDeviceMemoryFixtureI915, GivenValidMemoryHandleWhenCallingZesSysman
}
TEST_F(SysmanDeviceMemoryFixtureI915, GivenValidMemoryHandleWhenCallingZesSysmanMemoryGetStateAndDeviceIsNotAvailableThenDeviceLostErrorIsReturned) {
if (defaultHwInfo->capabilityTable.isIntegratedDevice) {
GTEST_SKIP() << "This test is not applicable for integrated devices";
}
auto ioctlHelper = static_cast<SysmanMemoryMockIoctlHelper *>(pDrm->ioctlHelper.get());
ioctlHelper->returnEmptyMemoryInfo = true;
ioctlHelper->mockErrorNumber = ENODEV;
@@ -795,6 +821,7 @@ class SysmanMultiDeviceMemoryFixture : public SysmanMultiDeviceFixture {
L0::Sysman::SysmanDevice *device = nullptr;
MockMemorySysFsAccessInterface *pSysfsAccess = nullptr;
MockSysmanKmdInterfacePrelim *pSysmanKmdInterface = nullptr;
MockMemoryFsAccessInterface *pFsAccess = nullptr;
void SetUp() override {
debugManager.flags.EnableLocalMemory.set(1);
@@ -802,9 +829,12 @@ class SysmanMultiDeviceMemoryFixture : public SysmanMultiDeviceFixture {
pSysmanKmdInterface = new MockSysmanKmdInterfacePrelim(pLinuxSysmanImp->getSysmanProductHelper());
pSysfsAccess = new MockMemorySysFsAccessInterface();
pFsAccess = new MockMemoryFsAccessInterface();
pSysmanKmdInterface->pSysfsAccess.reset(pSysfsAccess);
pSysmanKmdInterface->pFsAccess.reset(pFsAccess);
pLinuxSysmanImp->pSysmanKmdInterface.reset(pSysmanKmdInterface);
pLinuxSysmanImp->pSysfsAccess = pLinuxSysmanImp->pSysmanKmdInterface->getSysFsAccess();
pLinuxSysmanImp->pFsAccess = pLinuxSysmanImp->pSysmanKmdInterface->getFsAccess();
pDrm = new MockMemoryNeoDrm(const_cast<NEO::RootDeviceEnvironment &>(pSysmanDeviceImp->getRootDeviceEnvironment()));
pDrm->ioctlHelper = static_cast<std::unique_ptr<NEO::IoctlHelper>>(std::make_unique<NEO::MockIoctlHelper>(*pDrm));
auto &osInterface = pSysmanDeviceImp->getRootDeviceEnvironment().osInterface;
@@ -866,15 +896,39 @@ HWTEST2_F(SysmanMultiDeviceMemoryFixture, GivenValidMemoryHandleWhenCallingZesMe
ze_result_t result = zesMemoryGetState(handles[0], &state1);
EXPECT_EQ(result, ZE_RESULT_SUCCESS);
EXPECT_EQ(state1.health, ZES_MEM_HEALTH_UNKNOWN);
EXPECT_EQ(state1.size, NEO::probedSizeRegionOne);
EXPECT_EQ(state1.free, NEO::unallocatedSizeRegionOne);
if (defaultHwInfo->capabilityTable.isIntegratedDevice) {
EXPECT_EQ(state1.size, mockIntegratedDeviceAvailableMemory);
EXPECT_EQ(state1.free, mockIntegratedDeviceFreeMemory);
} else {
EXPECT_EQ(state1.size, NEO::probedSizeRegionOne);
EXPECT_EQ(state1.free, NEO::unallocatedSizeRegionOne);
}
zes_mem_state_t state2;
result = zesMemoryGetState(handles[1], &state2);
EXPECT_EQ(result, ZE_RESULT_SUCCESS);
EXPECT_EQ(state2.health, ZES_MEM_HEALTH_UNKNOWN);
EXPECT_EQ(state2.size, NEO::probedSizeRegionFour);
EXPECT_EQ(state2.free, NEO::unallocatedSizeRegionFour);
if (defaultHwInfo->capabilityTable.isIntegratedDevice) {
EXPECT_EQ(state2.size, mockIntegratedDeviceAvailableMemory);
EXPECT_EQ(state2.free, mockIntegratedDeviceFreeMemory);
} else {
EXPECT_EQ(state2.size, NEO::probedSizeRegionFour);
EXPECT_EQ(state2.free, NEO::unallocatedSizeRegionFour);
}
}
TEST_F(SysmanMultiDeviceMemoryFixture, GivenMemFreeAndMemAvailableMissingInMemInfoWhenCallingGetStateThenFreeAndSizeValuesAreZero) {
if (!defaultHwInfo->capabilityTable.isIntegratedDevice) {
GTEST_SKIP() << "This test is only applicable for integrated devices.";
}
pFsAccess->mockMemInfoIncorrectValue = true;
auto handles = getMemoryHandles(pOsSysman->getSubDeviceCount());
zes_mem_state_t state = {};
ze_result_t result = zesMemoryGetState(handles[0], &state);
EXPECT_EQ(result, ZE_RESULT_ERROR_UNKNOWN);
EXPECT_EQ(state.free, 0u);
EXPECT_EQ(state.size, 0u);
}
} // namespace ult

View File

@@ -921,20 +921,31 @@ HWTEST2_F(SysmanProductHelperMemoryTest, GivenSysmanProductHelperInstanceWhenCal
auto pSysmanKmdInterface = new MockSysmanKmdInterfacePrelim(pLinuxSysmanImp->getSysmanProductHelper());
MockMemorySysFsAccessInterface *pSysfsAccess = new MockMemorySysFsAccessInterface();
MockMemoryFsAccessInterface *pFsAccess = new MockMemoryFsAccessInterface();
pLinuxSysmanImp->pSysmanKmdInterface.reset(pSysmanKmdInterface);
pSysmanKmdInterface->pSysfsAccess.reset(pSysfsAccess);
pSysmanKmdInterface->pFsAccess.reset(pFsAccess);
pLinuxSysmanImp->pFsAccess = pLinuxSysmanImp->pSysmanKmdInterface->getFsAccess();
pSysfsAccess->mockReadStringValue.push_back(mockPhysicalSize);
pSysfsAccess->mockReadReturnStatus.push_back(ZE_RESULT_SUCCESS);
pSysfsAccess->isRepeated = true;
ze_result_t result = pSysmanProductHelper->getMemoryProperties(&properties, pLinuxSysmanImp, pDrm.get(), pLinuxSysmanImp->getSysmanKmdInterface(), subDeviceId, isSubdevice);
EXPECT_EQ(result, ZE_RESULT_SUCCESS);
EXPECT_EQ(properties.type, ZES_MEM_TYPE_HBM);
EXPECT_EQ(properties.location, ZES_MEM_LOC_DEVICE);
EXPECT_EQ(properties.subdeviceId, 0u);
EXPECT_EQ(properties.physicalSize, strtoull(mockPhysicalSize.c_str(), nullptr, 16));
EXPECT_EQ(properties.numChannels, numMemoryChannels);
EXPECT_EQ(properties.busWidth, memoryBusWidth);
if (defaultHwInfo->capabilityTable.isIntegratedDevice) {
EXPECT_EQ(properties.location, ZES_MEM_LOC_SYSTEM);
EXPECT_EQ(properties.numChannels, -1);
EXPECT_EQ(properties.busWidth, -1);
EXPECT_EQ(properties.type, ZES_MEM_TYPE_FORCE_UINT32);
EXPECT_EQ(properties.physicalSize, mockIntegratedDevicePhysicalSize);
} else {
EXPECT_EQ(properties.location, ZES_MEM_LOC_DEVICE);
EXPECT_EQ(properties.numChannels, numMemoryChannels);
EXPECT_EQ(properties.busWidth, memoryBusWidth);
EXPECT_EQ(properties.type, ZES_MEM_TYPE_HBM);
EXPECT_EQ(properties.subdeviceId, 0u);
EXPECT_EQ(properties.physicalSize, strtoull(mockPhysicalSize.c_str(), nullptr, 16));
}
}
HWTEST2_F(SysmanProductHelperMemoryTest, GivenSysmanProductHelperInstanceWhenCallingGetMemoryBandwidthAndNoTelemNodesAvailableThenFailureIsReturned, IsBMG) {