intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 06:49:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: update new RAS errors for sysman

Browse Source 
Related-To: NEO-12603

Signed-off-by: Shreyas Kunder <shreyas.kunder@intel.com>
This commit is contained in:
Shreyas Kunder
2025-01-06 10:49:36 +00:00
committed by Compute-Runtime-Automation
parent bb1a125f0c
commit ee3ef684ad
8 changed files with 72 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
level_zero/tools/test/unit_tests/sources/sysman/ras/linux/mock_sysman_ras.h
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2022-2024 Intel Corporation
* Copyright (C) 2022-2025 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -29,6 +29,7 @@ constexpr uint64_t correctableEuErrorCount = 75u;
constexpr uint64_t fatalEuErrorCount = 50u;
constexpr uint64_t fatalTlb = 3u;
constexpr uint64_t socFatalPsfCsc0Count = 5u;
constexpr uint64_t socFatalIosfPciaer = 3u;
constexpr uint64_t fatalEngineResetCount = 45u;
constexpr uint64_t correctableGrfErrorCountTile0 = 90u;
constexpr uint64_t correctableEuErrorCountTile0 = 70u;
@@ -60,7 +61,7 @@ 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 = 4u;
constexpr uint64_t initialUncorrectableFabricErrors = 8u;
constexpr uint64_t initialUncorrectableComputeErrors = 7u;
constexpr uint64_t initialCorrectableComputeErrors = 6u;
@@ -70,7 +71,7 @@ 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 = 11u;
constexpr uint64_t initialCorrectableNonComputeErrorsTile0 = 2u;
constexpr uint64_t initialUncorrectableComputeErrorsTile0 = 8u;
constexpr uint64_t initialUncorrectableFabricErrorsTile0 = 8u;
@@ -139,9 +140,10 @@ struct MockRasPmuInterfaceImp : public PmuInterfaceImp {
data[7] = 0;
data[8] = fatalEuErrorCount;
data[9] = socFatalPsfCsc0Count;
data[10] = fatalTlb;
data[11] = 0;
data[12] = socFatalMdfiEastCount;
data[10] = socFatalIosfPciaer;
data[11] = fatalTlb;
data[12] = 0;
data[13] = socFatalMdfiEastCount;
return 0;
}
@@ -167,11 +169,12 @@ struct MockRasPmuInterfaceImp : public PmuInterfaceImp {
data[8] = fatalSubslice;
data[9] = fatalEuErrorCount;
data[10] = socFatalPsfCsc0Count;
data[11] = nonFatalGscAonParity;
data[12] = nonFataGscSelfmBist;
data[13] = fatalTlb;
data[14] = 0;
data[15] = socFatalMdfiEastCount;
data[11] = socFatalIosfPciaer;
data[12] = nonFatalGscAonParity;
data[13] = nonFataGscSelfmBist;
data[14] = fatalTlb;
data[15] = 0;
data[16] = socFatalMdfiEastCount;
return 0;
}
@@ -382,6 +385,9 @@ struct MockRasSysfsAccess : public SysfsAccess {
} else if (file.compare("gt/gt1/error_counter/correctable_subslice") == 0) {
val = 2u;
return ZE_RESULT_SUCCESS;
} else if (file.compare("gt/gt0/error_counter/soc_fatal_iosf_pciaer") == 0) {
val = 1u;
return ZE_RESULT_SUCCESS;
}
return ZE_RESULT_ERROR_NOT_AVAILABLE;
}
@@ -453,6 +459,9 @@ struct MockRasSysfsAccess : public SysfsAccess {
} else if (file.compare("gt/gt1/error_counter/driver_engine_other") == 0) {
val = 3u;
return ZE_RESULT_SUCCESS;
} else if (file.compare("gt/gt0/error_counter/soc_fatal_iosf_pciaer") == 0) {
val = 1u;
return ZE_RESULT_SUCCESS;
}
return ZE_RESULT_ERROR_NOT_AVAILABLE;
}
@@ -513,6 +522,7 @@ struct MockRasFsAccess : public FsAccess {
events.push_back("error--fatal-fpu");
events.push_back("error--fatal-l3-fabric");
events.push_back("ccs0-busy");
events.push_back("error--soc-fatal-iosf-pciaer");
return ZE_RESULT_SUCCESS;
}
return ZE_RESULT_ERROR_NOT_AVAILABLE;
@@ -560,6 +570,7 @@ struct MockRasFsAccess : public FsAccess {
events.push_back("error-gt0--fatal-subslice");
events.push_back("error-gt1--fatal-l3bank");
events.push_back("error-gt1--correctable-subslice");
events.push_back("error-gt0--soc-fatal-iosf-pciaer");
return ZE_RESULT_SUCCESS;
}
return ZE_RESULT_ERROR_NOT_AVAILABLE;

8
level_zero/tools/test/unit_tests/sources/sysman/ras/linux/test_zes_ras.cpp
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2022-2024 Intel Corporation
* Copyright (C) 2022-2025 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -244,7 +244,7 @@ TEST_F(SysmanRasFixture, GivenValidRasHandleWhenCallingzesRasGetStateForGtThenSu
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 + socFatalIosfPciaer + 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);
}
@@ -281,7 +281,7 @@ 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], socFatalPsfCsc0Count + socFatalIosfPciaer + 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);
}
@@ -720,7 +720,7 @@ 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], socFatalPsfCsc0Count + socFatalIosfPciaer + 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) {

12
level_zero/tools/test/unit_tests/sources/sysman/ras/linux/test_zes_ras_exp.cpp
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2023-2024 Intel Corporation
* Copyright (C) 2023-2025 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -125,7 +125,7 @@ TEST_F(SysmanRasExpFixture, GivenValidRasHandleWhenCallingZesRasGetStateExpForGt
expectedErrCount = fatalEuErrorCount + initialUncorrectableComputeErrors;
EXPECT_EQ(rasStates[i].errorCounter, expectedErrCount);
} else if (rasStates[i].category == ZES_RAS_ERROR_CATEGORY_EXP_NON_COMPUTE_ERRORS) {
expectedErrCount = socFatalPsfCsc0Count + initialUncorrectableNonComputeErrors;
expectedErrCount = socFatalPsfCsc0Count + socFatalIosfPciaer + initialUncorrectableNonComputeErrors;
EXPECT_EQ(rasStates[i].errorCounter, expectedErrCount);
} else if (rasStates[i].category == ZES_RAS_ERROR_CATEGORY_EXP_PROGRAMMING_ERRORS) {
expectedErrCount = euAttention + initialProgrammingErrors;
@@ -197,7 +197,7 @@ TEST_F(SysmanRasExpFixture, GivenValidRasHandleWhenCallingZesRasGetStateExpForGt
expectedErrCount = fatalEuErrorCount + initialUncorrectableComputeErrors;
EXPECT_EQ(rasStates[i].errorCounter, expectedErrCount);
} else if (rasStates[i].category == ZES_RAS_ERROR_CATEGORY_EXP_NON_COMPUTE_ERRORS) {
expectedErrCount = socFatalPsfCsc0Count + initialUncorrectableNonComputeErrors;
expectedErrCount = socFatalPsfCsc0Count + socFatalIosfPciaer + initialUncorrectableNonComputeErrors;
EXPECT_EQ(rasStates[i].errorCounter, expectedErrCount);
} else if (rasStates[i].category == ZES_RAS_ERROR_CATEGORY_EXP_PROGRAMMING_ERRORS) {
expectedErrCount = euAttention + initialProgrammingErrors;
@@ -507,7 +507,7 @@ TEST_F(SysmanRasExpFixture, GivenValidRasHandleWhenCallingzesRasClearStateExpAnd
expectedErrCount = fatalEuErrorCount + initialUncorrectableComputeErrors;
EXPECT_EQ(rasStates[i].errorCounter, expectedErrCount);
} else if (rasStates[i].category == ZES_RAS_ERROR_CATEGORY_EXP_NON_COMPUTE_ERRORS) {
expectedErrCount = socFatalPsfCsc0Count + initialUncorrectableNonComputeErrors;
expectedErrCount = socFatalPsfCsc0Count + socFatalIosfPciaer + initialUncorrectableNonComputeErrors;
EXPECT_EQ(rasStates[i].errorCounter, expectedErrCount);
} else if (rasStates[i].category == ZES_RAS_ERROR_CATEGORY_EXP_PROGRAMMING_ERRORS) {
expectedErrCount = euAttention + initialProgrammingErrors;
@@ -806,7 +806,7 @@ TEST_F(SysmanRasExpMultiDeviceFixture, GivenValidRasHandleWhenCallingZesRasGetSt
expectedErrCount = fatalSubslice + fatalEuErrorCount + initialUncorrectableComputeErrorsTile0;
EXPECT_EQ(rasStates[i].errorCounter, expectedErrCount);
} else if (rasStates[i].category == ZES_RAS_ERROR_CATEGORY_EXP_NON_COMPUTE_ERRORS) {
expectedErrCount = socFatalPsfCsc0Count + nonFatalGscAonParity + nonFataGscSelfmBist + initialUncorrectableNonComputeErrorsTile0;
expectedErrCount = socFatalPsfCsc0Count + socFatalIosfPciaer + nonFatalGscAonParity + nonFataGscSelfmBist + initialUncorrectableNonComputeErrorsTile0;
EXPECT_EQ(rasStates[i].errorCounter, expectedErrCount);
} else if (rasStates[i].category == ZES_RAS_ERROR_CATEGORY_EXP_PROGRAMMING_ERRORS) {
expectedErrCount = euAttention + initialProgrammingErrors;
@@ -975,7 +975,7 @@ TEST_F(SysmanRasExpMultiDeviceFixture, GivenValidRasHandleWhenCallingzesRasClear
expectedErrCount = fatalSubslice + fatalEuErrorCount + initialUncorrectableComputeErrorsTile0;
EXPECT_EQ(rasStates[i].errorCounter, expectedErrCount);
} else if (rasStates[i].category == ZES_RAS_ERROR_CATEGORY_EXP_NON_COMPUTE_ERRORS) {
expectedErrCount = socFatalPsfCsc0Count + nonFatalGscAonParity + nonFataGscSelfmBist + initialUncorrectableNonComputeErrorsTile0;
expectedErrCount = socFatalPsfCsc0Count + socFatalIosfPciaer + nonFatalGscAonParity + nonFataGscSelfmBist + initialUncorrectableNonComputeErrorsTile0;
EXPECT_EQ(rasStates[i].errorCounter, expectedErrCount);
} else if (rasStates[i].category == ZES_RAS_ERROR_CATEGORY_EXP_PROGRAMMING_ERRORS) {
expectedErrCount = euAttention + initialProgrammingErrors;