diff --git a/level_zero/tools/source/sysman/ras/linux/os_ras_imp_gt.cpp b/level_zero/tools/source/sysman/ras/linux/os_ras_imp_gt.cpp index 0eb99b13d1..f2c1097574 100644 --- a/level_zero/tools/source/sysman/ras/linux/os_ras_imp_gt.cpp +++ b/level_zero/tools/source/sysman/ras/linux/os_ras_imp_gt.cpp @@ -27,10 +27,12 @@ static const std::map> categoryToL "sgunit-fatal", "soc-nonfatal-punit", "sgunit-fatal", "sgunit-nonfatal", "gsc-nonfatal-mia-shutdown", "gsc-nonfatal-aon-parity", "gsc-nonfatal-rom-parity", "gsc-nonfatal-fuse-crc-check", "gsc-nonfatal-selfmbist", "gsc-nonfatal-fuse-pull", "gsc-nonfatal-sram-ecc", "gsc-nonfatal-glitch-det", - "gsc-nonfatal-ucode-parity", "gsc-nonfatal-mia-int", "gsc-nonfatal-wdg-timeout"}}, + "gsc-nonfatal-ucode-parity", "gsc-nonfatal-mia-int", "gsc-nonfatal-wdg-timeout", "soc-fatal-mdfi-east", + "soc-fatal-mdfi-south", "soc-nonfatal-mdfi-east", "soc-nonfatal-mdfi-south", "soc-fatal-mdfi-west", + "soc-fatal-cd0-mdfi", "soc-nonfatal-cd0-mdfi"}}, {ZES_RAS_ERROR_CAT_COMPUTE_ERRORS, {"fatal-fpu", "fatal-eu-grf", "fatal-sampler", "fatal-slm", - "fatal-guc", "fatal-eu-ic", "fatal-subslice"}}, + "fatal-guc", "fatal-eu-ic", "fatal-subslice", "fatal-l3-fabric"}}, {ZES_RAS_ERROR_CAT_DRIVER_ERRORS, {"driver-object-migration", "driver-engine-other", "driver-ggtt", "driver-gt-interrupt", "driver-gt-other", "driver-guc-communication", diff --git a/level_zero/tools/test/unit_tests/sources/sysman/ras/linux/mock_fs_ras_prelim.h b/level_zero/tools/test/unit_tests/sources/sysman/ras/linux/mock_fs_ras_prelim.h index cf006bad41..218ef64d9f 100644 --- a/level_zero/tools/test/unit_tests/sources/sysman/ras/linux/mock_fs_ras_prelim.h +++ b/level_zero/tools/test/unit_tests/sources/sysman/ras/linux/mock_fs_ras_prelim.h @@ -63,9 +63,9 @@ constexpr uint64_t driverEngineOther = 3u; constexpr uint64_t initialUncorrectableCacheErrors = 2u; constexpr uint64_t initialEngineReset = 2u; constexpr uint64_t initialProgrammingErrors = 7u; -constexpr uint64_t initialUncorrectableNonComputeErrors = 3u; +constexpr uint64_t initialUncorrectableNonComputeErrors = 8u; constexpr uint64_t initialUncorrectableFabricErrors = 8u; -constexpr uint64_t initialUncorrectableComputeErrors = 7u; +constexpr uint64_t initialUncorrectableComputeErrors = 10u; constexpr uint64_t initialCorrectableComputeErrors = 6u; constexpr uint64_t initialUncorrectableDriverErrors = 5u; @@ -73,9 +73,9 @@ constexpr uint64_t initialUncorrectableCacheErrorsTile0 = 2u; constexpr uint64_t initialCorrectableCacheErrorTile0 = 2u; constexpr uint64_t initialEngineResetTile0 = 2u; constexpr uint64_t initialProgrammingErrorsTile0 = 7u; -constexpr uint64_t initialUncorrectableNonComputeErrorsTile0 = 10u; +constexpr uint64_t initialUncorrectableNonComputeErrorsTile0 = 15u; constexpr uint64_t initialCorrectableNonComputeErrorsTile0 = 2u; -constexpr uint64_t initialUncorrectableComputeErrorsTile0 = 8u; +constexpr uint64_t initialUncorrectableComputeErrorsTile0 = 11u; constexpr uint64_t initialUncorrectableFabricErrorsTile0 = 8u; constexpr uint64_t initialCorrectableComputeErrorsTile0 = 6u; constexpr uint64_t initialUncorrectableDriverErrorsTile0 = 5u; @@ -83,7 +83,7 @@ constexpr uint64_t initialUncorrectableCacheErrorsTile1 = 1u; constexpr uint64_t initialEngineResetTile1 = 4u; constexpr uint64_t initialProgrammingErrorsTile1 = 5u; constexpr uint64_t initialCorrectableComputeErrorsTile1 = 7u; -constexpr uint64_t initialUncorrectableNonComputeErrorsTile1 = 3u; +constexpr uint64_t initialUncorrectableNonComputeErrorsTile1 = 5u; constexpr uint64_t initialUncorrectableFabricErrorsTile1 = 2u; constexpr uint64_t initialUncorrectableComputeErrorsTile1 = 6u; constexpr uint64_t initialUncorrectableDriverErrorsTile1 = 4u; @@ -140,9 +140,11 @@ struct MockRasPmuInterfaceImp : public PmuInterfaceImp { data[5] = driverGgtt; data[6] = driverRps; data[7] = 0; - data[8] = fatalEuErrorCount; - data[9] = socFatalPsfCsc0Count; - data[10] = fatalTlb; + data[8] = 0; + data[9] = fatalEuErrorCount; + data[10] = socFatalPsfCsc0Count; + data[11] = socFatalMdfiEastCount; + data[12] = fatalTlb; return 0; } @@ -165,12 +167,14 @@ struct MockRasPmuInterfaceImp : public PmuInterfaceImp { data[5] = driverGgtt; data[6] = driverRps; data[7] = 0; - data[8] = fatalSubslice; - data[9] = fatalEuErrorCount; - data[10] = socFatalPsfCsc0Count; - data[11] = nonFatalGscAonParity; - data[12] = nonFataGscSelfmBist; - data[13] = fatalTlb; + data[8] = 0; + data[9] = fatalSubslice; + data[10] = fatalEuErrorCount; + data[11] = socFatalPsfCsc0Count; + data[12] = socFatalMdfiEastCount; + data[13] = nonFatalGscAonParity; + data[14] = nonFataGscSelfmBist; + data[15] = fatalTlb; return 0; } @@ -191,9 +195,10 @@ struct MockRasPmuInterfaceImp : public PmuInterfaceImp { data[4] = driverMigration; data[5] = driverEngineOther; data[6] = fatalGucErrorCountTile1; - data[7] = socFatalPunitTile1; - data[8] = fatalIdiParityErrorCountTile1; - data[9] = fatalL3BankTile1; + data[7] = socFatalMdfiWestCountTile1; + data[8] = socFatalPunitTile1; + data[9] = fatalIdiParityErrorCountTile1; + data[10] = fatalL3BankTile1; return 0; } diff --git a/level_zero/tools/test/unit_tests/sources/sysman/ras/linux/test_zes_ras_prelim.cpp b/level_zero/tools/test/unit_tests/sources/sysman/ras/linux/test_zes_ras_prelim.cpp index 634ea070bc..b336a331b5 100644 --- a/level_zero/tools/test/unit_tests/sources/sysman/ras/linux/test_zes_ras_prelim.cpp +++ b/level_zero/tools/test/unit_tests/sources/sysman/ras/linux/test_zes_ras_prelim.cpp @@ -242,7 +242,7 @@ TEST_F(SysmanRasFixture, GivenValidRasHandleWhenCallingzesRasGeStateForGtThenSuc EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_RESET], fatalEngineResetCount + initialEngineReset); EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_PROGRAMMING_ERRORS], euAttention + initialProgrammingErrors); EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_COMPUTE_ERRORS], fatalEuErrorCount + initialUncorrectableComputeErrors); - EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_NON_COMPUTE_ERRORS], socFatalPsfCsc0Count + initialUncorrectableNonComputeErrors); + EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_NON_COMPUTE_ERRORS], socFatalPsfCsc0Count + socFatalMdfiEastCount + initialUncorrectableNonComputeErrors); EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_DISPLAY_ERRORS], 0u); EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_DRIVER_ERRORS], driverMigration + driverGgtt + driverRps + initialUncorrectableDriverErrors); } @@ -278,7 +278,7 @@ TEST_F(SysmanRasFixture, GivenValidRasHandleWhenCallingzesRasGeStateForGtAfterCl EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_RESET], fatalEngineResetCount + initialEngineReset); EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_PROGRAMMING_ERRORS], euAttention + initialProgrammingErrors); EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_COMPUTE_ERRORS], fatalEuErrorCount + initialUncorrectableComputeErrors); - EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_NON_COMPUTE_ERRORS], socFatalPsfCsc0Count + initialUncorrectableNonComputeErrors); + EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_NON_COMPUTE_ERRORS], socFatalMdfiEastCount + socFatalPsfCsc0Count + initialUncorrectableNonComputeErrors); EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_DISPLAY_ERRORS], 0u); EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_DRIVER_ERRORS], driverMigration + driverGgtt + driverRps + initialUncorrectableDriverErrors); } @@ -702,7 +702,7 @@ TEST_F(SysmanRasMultiDeviceFixture, GivenValidRasHandleWhenCallingzesRasGeStateF EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_RESET], fatalEngineResetCount + initialEngineResetTile0); EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_PROGRAMMING_ERRORS], euAttention + initialProgrammingErrorsTile0); EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_COMPUTE_ERRORS], fatalSubslice + fatalEuErrorCount + initialUncorrectableComputeErrorsTile0); - EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_NON_COMPUTE_ERRORS], socFatalPsfCsc0Count + nonFatalGscAonParity + nonFataGscSelfmBist + initialUncorrectableNonComputeErrorsTile0); + EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_NON_COMPUTE_ERRORS], socFatalMdfiEastCount + socFatalPsfCsc0Count + nonFatalGscAonParity + nonFataGscSelfmBist + initialUncorrectableNonComputeErrorsTile0); EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_DISPLAY_ERRORS], 0u); EXPECT_EQ(state.category[ZES_RAS_ERROR_CAT_DRIVER_ERRORS], driverMigration + driverGgtt + driverRps + initialUncorrectableDriverErrorsTile0); } else if (handleIndex == 2u) {