fix(sysman): Fix the Warm Device Reset

The Workaround added earlier to resize the VF Bar has been removed. Related-To: NEO-13775 Signed-off-by: Pratik Bari <pratik.bari@intel.com>
2025-12-20 00:24:58 +08:00 · 2025-02-18 06:54:58 +00:00
parent b33b4233ad
commit 568d90902d
4 changed files with 39 additions and 170 deletions
--- a/level_zero/sysman/source/shared/linux/zes_os_sysman_imp.cpp
+++ b/level_zero/sysman/source/shared/linux/zes_os_sysman_imp.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2023-2024 Intel Corporation
+ * Copyright (C) 2023-2025 Intel Corporation
 *
 * SPDX-License-Identifier: MIT
 *
@@ -341,50 +341,6 @@ void LinuxSysmanImp::clearHPIE(int fd) {
    NEO::sleep(std::chrono::seconds(10)); // Sleep for 10seconds just to make sure the change is propagated.
 }

-// Function to adjust VF BAR size i.e Modify VF BAR Control register.
-// size param is an encoded value described as follows:
-// 0  - 1 MB (2^20 bytes)
-// 1  - 2 MB (2^21 bytes)
-// 2  - 4 MB (2^22 bytes)
-// 3  - 8 MB (2^23 bytes)
-// .
-// .
-// .
-// b  - 2 GB (2^31 bytes)
-// 43 - 8 EB (2^63 bytes)
-ze_result_t LinuxSysmanImp::resizeVfBar(uint8_t size) {
-    std::string pciConfigNode;
-    pciConfigNode = gtDevicePath + "/config";
-
-    auto fdConfig = NEO::FileDescriptor(pciConfigNode.c_str(), O_RDWR);
-    if (fdConfig < 0) {
-        NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stdout,
-                              "Config node open failed\n");
-        return ZE_RESULT_ERROR_UNKNOWN;
-    }
-    std::unique_ptr<uint8_t[]> configMemory = std::make_unique<uint8_t[]>(PCI_CFG_SPACE_EXP_SIZE);
-    memset(configMemory.get(), 0, PCI_CFG_SPACE_EXP_SIZE);
-    if (this->preadFunction(fdConfig, configMemory.get(), PCI_CFG_SPACE_EXP_SIZE, 0) < 0) {
-        NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stdout,
-                              "Read to get config space failed\n");
-        return ZE_RESULT_ERROR_UNKNOWN;
-    }
-    auto reBarCapPos = L0::Sysman::LinuxPciImp::getRebarCapabilityPos(configMemory.get(), true);
-    if (!reBarCapPos) {
-        NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stdout,
-                              "VF BAR capability not found\n");
-        return ZE_RESULT_ERROR_UNKNOWN;
-    }
-
-    auto barSizePos = reBarCapPos + PCI_REBAR_CTRL + 1; // position of VF(0) BAR SIZE.
-    if (this->pwriteFunction(fdConfig, &size, 0x01, barSizePos) < 0) {
-        NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stdout,
-                              "Write to change VF bar size failed\n");
-        return ZE_RESULT_ERROR_UNKNOWN;
-    }
-    return ZE_RESULT_SUCCESS;
-}
-
 // A 'warm reset' is a conventional reset that is triggered across a PCI express link.
 // A warm reset is triggered either when a link is forced into electrical idle or
 // by sending TS1 and TS2 ordered sets with the hot reset bit set.
@@ -443,33 +399,6 @@ ze_result_t LinuxSysmanImp::osWarmReset() {
    }
    NEO::sleep(std::chrono::seconds(10)); // Sleep for 10seconds, allows the rescan to complete on all devices attached to the root port.

-    // PCIe port driver uses the BIOS allocated VF bars on bootup. A known bug exists in pcie port driver
-    // and is causing VF bar allocation failure in PCIe port driver after an SBR - https://bugzilla.kernel.org/show_bug.cgi?id=216795
-
-    // WA to adjust VF bar size to 2GB. The default VF bar size is 8GB and for 63VFs, 504GB need to be allocated which is failing on SBR.
-    // When configured VF bar size to 2GB, an allocation of 126GB is successful. This WA resizes VF0 bar to 2GB. Once pcie port driver
-    // issue is resolved, this WA may not be necessary. Description for 0xb is explained at function definition - resizeVfVar.
-    if (NEO::debugManager.flags.VfBarResourceAllocationWa.get()) {
-        if (ZE_RESULT_SUCCESS != (result = resizeVfBar(0xb))) {
-            return result;
-        }
-
-        result = pFsAccess->write(devicePath + '/' + "remove", "1");
-        if (ZE_RESULT_SUCCESS != result) {
-            NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stdout,
-                                  "Card Bus remove after resizing VF bar failed\n");
-            return result;
-        }
-        NEO::sleep(std::chrono::seconds(10)); // Sleep for 10seconds to make sure that the config spaces of all devices are saved correctly.
-
-        result = pFsAccess->write(rootPortPath + '/' + "rescan", "1");
-        if (ZE_RESULT_SUCCESS != result) {
-            NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stdout,
-                                  "Rescanning root port failed after resizing VF bar failed\n");
-            return result;
-        }
-        NEO::sleep(std::chrono::seconds(10)); // Sleep for 10seconds, allows the rescan to complete on all devices attached to the root port.
-    }
    return result;
 }

--- a/level_zero/sysman/test/unit_tests/sources/diagnostics/linux/test_zes_sysman_diagnostics.cpp
+++ b/level_zero/sysman/test/unit_tests/sources/diagnostics/linux/test_zes_sysman_diagnostics.cpp
@@ -25,22 +25,11 @@ inline static int openMockDiag(const char *pathname, int flags) {
    }
    return -1;
 }
-void mockSleepFunctionSecs(int64_t secs) {
-    return;
-}

 inline static int openMockDiagFail(const char *pathname, int flags) {
    return -1;
 }

-inline static int gtPciConfigOpenFail(const char *pathname, int flags) {
-    if (strcmp(pathname, mockRealPathConfig.c_str()) == 0) {
-        return mockFileDescriptor;
-    } else {
-        return -1;
-    }
-}
-
 ssize_t preadMockDiag(int fd, void *buf, size_t count, off_t offset) {
    uint8_t *mockBuf = static_cast<uint8_t *>(buf);
    if (fd == mockGtPciConfigFd) {
@@ -78,24 +67,6 @@ ssize_t preadMockDiag(int fd, void *buf, size_t count, off_t offset) {
    return count;
 }

-ssize_t mockGtConfigPreadInvalid(int fd, void *buf, size_t count, off_t offset) {
-    return count;
-}
-
-ssize_t mockGtConfigPreadFail(int fd, void *buf, size_t count, off_t offset) {
-    if (fd == mockGtPciConfigFd) {
-        return -1;
-    }
-    return count;
-}
-
-ssize_t mockGtConfigPwriteFail(int fd, const void *buf, size_t count, off_t offset) {
-    if (fd == mockGtPciConfigFd) {
-        return -1;
-    }
-    return count;
-}
-
 ssize_t pwriteMockDiag(int fd, const void *buf, size_t count, off_t offset) {
    return count;
 }
@@ -529,66 +500,6 @@ TEST_F(ZesDiagnosticsFixture, GivenValidSysmanImpPointerWhenCallingWarmResetThen
    EXPECT_EQ(ZE_RESULT_SUCCESS, pLinuxSysmanImp->osWarmReset());
 }

-TEST_F(ZesDiagnosticsFixture, GivenValidSysmanImpPointerAndVfBarIsResizedWhenCallingWarmResetAndGtPciConfigOpenFailsThenCallReturnsFailure) {
-    pLinuxSysmanImp->gtDevicePath = "/sys/devices/pci0000:89/0000:89:02.0/0000:8a:00.0/0000:8b:01.0/0000:8c:00.0";
-    VariableBackup<decltype(NEO::SysCalls::sysCallsOpen)> openBackup(&NEO::SysCalls::sysCallsOpen, gtPciConfigOpenFail);
-    pLinuxSysmanImp->preadFunction = preadMockDiag;
-    pLinuxSysmanImp->pwriteFunction = pwriteMockDiag;
-
-    EXPECT_EQ(ZE_RESULT_ERROR_UNKNOWN, pLinuxSysmanImp->osWarmReset());
-}
-
-TEST_F(ZesDiagnosticsFixture, GivenValidSysmanImpPointerAndVfBarIsResizedWhenCallingWarmResetAndConfigHeaderIsInvalidThenCallReturnsFailure) {
-    pLinuxSysmanImp->gtDevicePath = "/sys/devices/pci0000:89/0000:89:02.0/0000:8a:00.0/0000:8b:01.0/0000:8c:00.0";
-    VariableBackup<decltype(NEO::SysCalls::sysCallsOpen)> openBackup(&NEO::SysCalls::sysCallsOpen, openMockDiag);
-    pLinuxSysmanImp->preadFunction = mockGtConfigPreadInvalid;
-    pLinuxSysmanImp->pwriteFunction = pwriteMockDiag;
-
-    EXPECT_EQ(ZE_RESULT_ERROR_UNKNOWN, pLinuxSysmanImp->osWarmReset());
-}
-
-TEST_F(ZesDiagnosticsFixture, GivenValidSysmanImpPointerAndVfBarIsResizedWhenCallingWarmResetAndGtConfigPreadFailsThenCallReturnsFailure) {
-    pLinuxSysmanImp->gtDevicePath = "/sys/devices/pci0000:89/0000:89:02.0/0000:8a:00.0/0000:8b:01.0/0000:8c:00.0";
-    VariableBackup<decltype(NEO::SysCalls::sysCallsOpen)> openBackup(&NEO::SysCalls::sysCallsOpen, openMockDiag);
-    pLinuxSysmanImp->preadFunction = mockGtConfigPreadFail;
-    pLinuxSysmanImp->pwriteFunction = pwriteMockDiag;
-
-    EXPECT_EQ(ZE_RESULT_ERROR_UNKNOWN, pLinuxSysmanImp->osWarmReset());
-}
-
-TEST_F(ZesDiagnosticsFixture, GivenValidSysmanImpPointerAndVfBarIsResizedWhenCallingWarmResetAndGtConfigPwriteFailsThenCallReturnsFailure) {
-    pLinuxSysmanImp->gtDevicePath = "/sys/devices/pci0000:89/0000:89:02.0/0000:8a:00.0/0000:8b:01.0/0000:8c:00.0";
-    VariableBackup<decltype(NEO::SysCalls::sysCallsOpen)> openBackup(&NEO::SysCalls::sysCallsOpen, openMockDiag);
-    pLinuxSysmanImp->preadFunction = preadMockDiag;
-    pLinuxSysmanImp->pwriteFunction = mockGtConfigPwriteFail;
-
-    EXPECT_EQ(ZE_RESULT_ERROR_UNKNOWN, pLinuxSysmanImp->osWarmReset());
-}
-
-TEST_F(ZesDiagnosticsFixture, GivenValidSysmanImpPointerAndVfBarIsResizedWhenCallingWarmResetAndCardBusRemoveFailsThenCallReturnsFailure) {
-    pLinuxSysmanImp->gtDevicePath = "/sys/devices/pci0000:89/0000:89:02.0/0000:8a:00.0/0000:8b:01.0/0000:8c:00.0";
-    VariableBackup<decltype(NEO::SysCalls::sysCallsOpen)> openBackup(&NEO::SysCalls::sysCallsOpen, openMockDiag);
-    pLinuxSysmanImp->preadFunction = preadMockDiag;
-    pLinuxSysmanImp->pwriteFunction = pwriteMockDiag;
-
-    pMockFsAccess->checkErrorAfterCount = 2;
-    pMockFsAccess->mockWriteError = ZE_RESULT_ERROR_NOT_AVAILABLE;
-
-    EXPECT_EQ(ZE_RESULT_ERROR_NOT_AVAILABLE, pLinuxSysmanImp->osWarmReset());
-}
-
-TEST_F(ZesDiagnosticsFixture, GivenValidSysmanImpPointerAndVfBarIsResizedWhenCallingWarmResetAndRootPortRescanFailsThenCallReturnsFailure) {
-    pLinuxSysmanImp->gtDevicePath = "/sys/devices/pci0000:89/0000:89:02.0/0000:8a:00.0/0000:8b:01.0/0000:8c:00.0";
-    VariableBackup<decltype(NEO::SysCalls::sysCallsOpen)> openBackup(&NEO::SysCalls::sysCallsOpen, openMockDiag);
-    pLinuxSysmanImp->preadFunction = preadMockDiag;
-    pLinuxSysmanImp->pwriteFunction = pwriteMockDiag;
-
-    pMockFsAccess->checkErrorAfterCount = 3;
-    pMockFsAccess->mockWriteError = ZE_RESULT_ERROR_NOT_AVAILABLE;
-
-    EXPECT_EQ(ZE_RESULT_ERROR_NOT_AVAILABLE, pLinuxSysmanImp->osWarmReset());
-}
-
 TEST_F(ZesDiagnosticsFixture, GivenValidSysmanImpPointerAndVfBarIsResizedWhenCallingWarmResetThenCallSucceeds) {
    pLinuxSysmanImp->gtDevicePath = "/sys/devices/pci0000:89/0000:89:02.0/0000:8a:00.0/0000:8b:01.0/0000:8c:00.0";
    VariableBackup<decltype(NEO::SysCalls::sysCallsOpen)> openBackup(&NEO::SysCalls::sysCallsOpen, openMockDiag);
--- a/level_zero/sysman/test/unit_tests/sources/global_operations/linux/test_zes_global_operations.cpp
+++ b/level_zero/sysman/test/unit_tests/sources/global_operations/linux/test_zes_global_operations.cpp
@@ -845,8 +845,6 @@ TEST_F(SysmanGlobalOperationsFixture, GivenDeviceInUseWhenCallingzesDeviceResetE
    std::unique_ptr<SysmanProductHelper> pSysmanProductHelper = std::make_unique<MockSysmanProductHelper>();
    std::swap(pLinuxSysmanImp->pSysmanProductHelper, pSysmanProductHelper);

-    DebugManagerStateRestore dbgRestore;
-    debugManager.flags.VfBarResourceAllocationWa.set(false);
    zes_reset_properties_t pProperties = {.stype = ZES_STRUCTURE_TYPE_RESET_PROPERTIES, .pNext = nullptr, .force = true, .resetType = ZES_RESET_TYPE_WARM};
    ze_result_t result = zesDeviceResetExt(pSysmanDevice->toHandle(), &pProperties);
    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
@@ -894,8 +892,6 @@ TEST_F(SysmanGlobalOperationsFixture, GivenDeviceInUseWhenCallingZesDeviceResetE

 TEST_F(SysmanGlobalOperationsFixture, GivenDeviceInUseWhenCallingResetExtWithInvalidTypeThenFailureIsReturned) {
    init(true);
-    DebugManagerStateRestore dbgRestore;
-    debugManager.flags.VfBarResourceAllocationWa.set(false);
    zes_reset_properties_t pProperties = {.stype = ZES_STRUCTURE_TYPE_RESET_PROPERTIES, .pNext = nullptr, .force = true, .resetType = ZES_RESET_TYPE_FORCE_UINT32};
    ze_result_t result = zesDeviceResetExt(pSysmanDevice->toHandle(), &pProperties);
    EXPECT_EQ(ZE_RESULT_ERROR_INVALID_ARGUMENT, result);
@@ -903,8 +899,6 @@ TEST_F(SysmanGlobalOperationsFixture, GivenDeviceInUseWhenCallingResetExtWithInv

 TEST_F(SysmanGlobalOperationsFixture, GivenGettingSysfsPathFailsWhenCallingResetExtThenFailureIsReturned) {
    init(true);
-    DebugManagerStateRestore dbgRestore;
-    debugManager.flags.VfBarResourceAllocationWa.set(false);
    pSysfsAccess->mockDeviceUnbound = true;
    zes_reset_properties_t pProperties = {.stype = ZES_STRUCTURE_TYPE_RESET_PROPERTIES, .pNext = nullptr, .force = true, .resetType = ZES_RESET_TYPE_FORCE_UINT32};
    ze_result_t result = zesDeviceResetExt(pSysmanDevice->toHandle(), &pProperties);
@@ -912,8 +906,6 @@ TEST_F(SysmanGlobalOperationsFixture, GivenGettingSysfsPathFailsWhenCallingReset
 }

 TEST_F(SysmanGlobalOperationsFixture, GivenForceTrueWhenCallingResetThenSuccessIsReturned) {
-    DebugManagerStateRestore dbgRestore;
-    debugManager.flags.VfBarResourceAllocationWa.set(false);
    ze_result_t result = zesDeviceReset(device, true);
    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
 }
--- a/level_zero/sysman/test/unit_tests/sources/pci/linux/test_zes_pci.cpp
+++ b/level_zero/sysman/test/unit_tests/sources/pci/linux/test_zes_pci.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2023-2024 Intel Corporation
+ * Copyright (C) 2023-2025 Intel Corporation
 *
 * SPDX-License-Identifier: MIT
 *
@@ -668,6 +668,43 @@ TEST_F(ZesPciFixture, WhenConvertingLinkSpeedFromGigaTransfersPerSecondToBytesPe
    EXPECT_EQ(0, L0::Sysman::convertPcieSpeedFromGTsToBs(0.0));
 }

+TEST_F(ZesPciFixture, GivenValidConfigMemoryDataWhenCallingGetRebarCapabilityPosThenTrueValueIsReturned) {
+    std::unique_ptr<PublicLinuxPciImp> pLinuxPciImp = std::make_unique<PublicLinuxPciImp>(pOsSysman);
+    std::vector<uint8_t> configMemory(PCI_CFG_SPACE_EXP_SIZE);
+    uint8_t *mockBuf = configMemory.data();
+    mockBuf[0x006] = 0x10;
+    mockBuf[0x034] = 0x40;
+    mockBuf[0x040] = 0x0d;
+    mockBuf[0x041] = 0x50;
+    mockBuf[0x050] = 0x10;
+    mockBuf[0x051] = 0x70;
+    mockBuf[0x052] = 0x90;
+    mockBuf[0x070] = 0x10;
+    mockBuf[0x071] = 0xac;
+    mockBuf[0x072] = 0xa0;
+    mockBuf[0x0ac] = 0x10;
+    mockBuf[0x0b8] = 0x11;
+    mockBuf[0x100] = 0x0e;
+    mockBuf[0x102] = 0x01;
+    mockBuf[0x103] = 0x42;
+    mockBuf[0x420] = 0x15;
+    mockBuf[0x422] = 0x01;
+    mockBuf[0x423] = 0x22;
+    mockBuf[0x425] = 0xf0;
+    mockBuf[0x426] = 0x3f;
+    mockBuf[0x428] = 0x22;
+    mockBuf[0x429] = 0x11;
+    mockBuf[0x220] = 0x24;
+    mockBuf[0x222] = 0x01;
+    mockBuf[0x223] = 0x32;
+    mockBuf[0x320] = 0x10;
+    mockBuf[0x322] = 0x01;
+    mockBuf[0x323] = 0x40;
+    mockBuf[0x400] = 0x18;
+    mockBuf[0x402] = 0x01;
+    EXPECT_TRUE(pLinuxPciImp->getRebarCapabilityPos(mockBuf, true));
+}
+
 } // namespace ult
 } // namespace Sysman
 } // namespace L0