fix: serialize printf kernel accesses using device-wise locks

Related-To: LOCI-4114

Signed-off-by: Lu, Wenbin <wenbin.lu@intel.com>

level_zero/core/source/cmdlist/cmdlist.cpp

@@ -22,6 +22,7 @@
 #include "level_zero/core/source/driver/driver_handle_imp.h"
 #include "level_zero/core/source/event/event.h"
 #include "level_zero/core/source/kernel/kernel.h"
+#include "level_zero/core/source/kernel/kernel_imp.h"
 
 namespace L0 {
 
@@ -38,11 +39,11 @@ CommandList::~CommandList() {
 }
 
 void CommandList::storePrintfKernel(Kernel *kernel) {
-    auto it = std::find(this->printfKernelContainer.begin(), this->printfKernelContainer.end(),
+    auto it = std::find_if(this->printfKernelContainer.begin(), this->printfKernelContainer.end(), [&kernel](const auto &kernelWeakPtr) { return kernelWeakPtr.lock().get() == kernel; });
-                        kernel);
 
     if (it == this->printfKernelContainer.end()) {
-        this->printfKernelContainer.push_back(kernel);
+        auto module = static_cast<const ModuleImp *>(&static_cast<KernelImp *>(kernel)->getParentModule());
+        this->printfKernelContainer.push_back(module->getPrintfKernelWeakPtr(kernel->toHandle()));
     }
 }
 

level_zero/core/source/cmdlist/cmdlist.h

@@ -231,7 +231,7 @@ struct CommandList : _ze_command_list_handle_t {
         return indirectAllocationsAllowed;
     }
 
-    std::vector<Kernel *> &getPrintfKernelContainer() {
+    std::vector<std::weak_ptr<Kernel>> &getPrintfKernelContainer() {
         return this->printfKernelContainer;
     }
 
@@ -360,7 +360,7 @@ struct CommandList : _ze_command_list_handle_t {
     std::map<const void *, NEO::GraphicsAllocation *> hostPtrMap;
     NEO::PrivateAllocsToReuseContainer ownedPrivateAllocations;
     std::vector<NEO::GraphicsAllocation *> patternAllocations;
-    std::vector<Kernel *> printfKernelContainer;
+    std::vector<std::weak_ptr<Kernel>> printfKernelContainer;
 
     NEO::CommandContainer commandContainer;
 

level_zero/core/source/cmdlist/cmdlist_hw.inl

@@ -434,7 +434,10 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernelIndirect(ze_
     if (hEvent) {
         event = Event::fromHandle(hEvent);
         if (Kernel::fromHandle(kernelHandle)->getPrintfBufferAllocation() != nullptr) {
-            event->setKernelForPrintf(Kernel::fromHandle(kernelHandle));
+            Kernel *kernel = Kernel::fromHandle(kernelHandle);
+            auto module = static_cast<const ModuleImp *>(&static_cast<KernelImp *>(kernel)->getParentModule());
+            event->setKernelForPrintf(module->getPrintfKernelWeakPtr(kernelHandle));
+            event->setKernelWithPrintfDeviceMutex(kernel->getDevicePrintfKernelMutex());
         }
         launchParams.isHostSignalScopeEvent = event->isSignalScope(ZE_EVENT_SCOPE_FLAG_HOST);
     }

level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl

@@ -387,7 +387,7 @@ inline ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::executeCommand
 
     this->cmdQImmediate->setTaskCount(completionStamp.taskCount);
 
-    if (this->isSyncModeQueue || this->printfKernelContainer.size() > 0u) {
+    if (this->isSyncModeQueue) {
         status = hostSynchronize(std::numeric_limits<uint64_t>::max(), completionStamp.taskCount, true);
     }
 
@@ -1238,9 +1238,11 @@ bool CommandListCoreFamilyImmediate<gfxCoreFamily>::isBarrierRequired() {
 
 template <GFXCORE_FAMILY gfxCoreFamily>
 void CommandListCoreFamilyImmediate<gfxCoreFamily>::printKernelsPrintfOutput(bool hangDetected) {
-    size_t size = this->printfKernelContainer.size();
+    for (auto &kernelWeakPtr : this->printfKernelContainer) {
-    for (size_t i = 0; i < size; i++) {
+        std::lock_guard<std::mutex> lock(static_cast<DeviceImp *>(this->device)->printfKernelMutex);
-        this->printfKernelContainer[i]->printPrintfOutput(hangDetected);
+        if (!kernelWeakPtr.expired()) {
+            kernelWeakPtr.lock()->printPrintfOutput(hangDetected);
+        }
     }
     this->printfKernelContainer.clear();
 }

level_zero/core/source/cmdlist/cmdlist_hw_skl_to_tgllp.inl

@@ -271,7 +271,9 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernelWithParams(K
     }
 
     if (event != nullptr && kernel->getPrintfBufferAllocation() != nullptr) {
-        event->setKernelForPrintf(kernel);
+        auto module = static_cast<const ModuleImp *>(&static_cast<KernelImp *>(kernel)->getParentModule());
+        event->setKernelForPrintf(module->getPrintfKernelWeakPtr(kernel->toHandle()));
+        event->setKernelWithPrintfDeviceMutex(kernel->getDevicePrintfKernelMutex());
     }
 
     if (this->isInOrderExecutionEnabled() && !launchParams.isKernelSplitOperation) {

level_zero/core/source/cmdlist/cmdlist_hw_xehp_and_later.inl

@@ -23,7 +23,9 @@
 #include "shared/source/utilities/software_tags_manager.h"
 
 #include "level_zero/core/source/cmdlist/cmdlist_hw.h"
+#include "level_zero/core/source/device/device.h"
 #include "level_zero/core/source/driver/driver_handle_imp.h"
+#include "level_zero/core/source/event/event.h"
 #include "level_zero/core/source/gfx_core_helpers/l0_gfx_core_helper.h"
 #include "level_zero/core/source/kernel/kernel_imp.h"
 #include "level_zero/core/source/module/module.h"
@@ -180,7 +182,9 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernelWithParams(K
     Event *eventForInOrderExec = event;
     if (event) {
         if (kernel->getPrintfBufferAllocation() != nullptr) {
-            event->setKernelForPrintf(kernel);
+            auto module = static_cast<const ModuleImp *>(&static_cast<KernelImp *>(kernel)->getParentModule());
+            event->setKernelForPrintf(module->getPrintfKernelWeakPtr(kernel->toHandle()));
+            event->setKernelWithPrintfDeviceMutex(kernel->getDevicePrintfKernelMutex());
         }
         isHostSignalScopeEvent = event->isSignalScope(ZE_EVENT_SCOPE_FLAG_HOST);
         if (compactL3FlushEvent(getDcFlushRequired(event->isSignalScope()))) {

level_zero/core/source/cmdqueue/cmdqueue.cpp

@@ -190,9 +190,11 @@ ze_result_t CommandQueueImp::synchronizeByPollingForTaskCount(uint64_t timeout)
 }
 
 void CommandQueueImp::printKernelsPrintfOutput(bool hangDetected) {
-    size_t size = this->printfKernelContainer.size();
+    for (auto &kernelWeakPtr : this->printfKernelContainer) {
-    for (size_t i = 0; i < size; i++) {
+        std::lock_guard<std::mutex> lock(static_cast<DeviceImp *>(this->getDevice())->printfKernelMutex);
-        this->printfKernelContainer[i]->printPrintfOutput(hangDetected);
+        if (!kernelWeakPtr.expired()) {
+            kernelWeakPtr.lock()->printPrintfOutput(hangDetected);
+        }
     }
     this->printfKernelContainer.clear();
 }

level_zero/core/source/cmdqueue/cmdqueue_imp.h

@@ -146,7 +146,7 @@ struct CommandQueueImp : public CommandQueue {
     NEO::LinearStream firstCmdListStream{};
     NEO::HeapContainer heapContainer;
     ze_command_queue_desc_t desc;
-    std::vector<Kernel *> printfKernelContainer;
+    std::vector<std::weak_ptr<Kernel>> printfKernelContainer;
 
     Device *device = nullptr;
     NEO::CommandStreamReceiver *csr = nullptr;

level_zero/core/source/device/device_imp.h

@@ -132,6 +132,8 @@ struct DeviceImp : public Device, NEO::NonCopyableOrMovableClass {
     ze_pci_speed_ext_t pciMaxSpeed = {-1, -1, -1};
     Device *rootDevice = nullptr;
 
+    std::mutex printfKernelMutex;
+
     BcsSplit bcsSplit;
 
     bool resourcesReleased = false;

level_zero/core/source/event/event.h

@@ -18,6 +18,7 @@
 #include <chrono>
 #include <limits>
 #include <memory>
+#include <mutex>
 #include <vector>
 
 struct _ze_event_handle_t {};
@@ -206,12 +207,24 @@ struct Event : _ze_event_handle_t {
     uint32_t getMaxKernelCount() const {
         return maxKernelCount;
     }
-    void setKernelForPrintf(Kernel *inputKernelPtr) {
+    void setKernelForPrintf(std::weak_ptr<Kernel> inputKernelWeakPtr) {
-        kernelWithPrintf = inputKernelPtr;
+        kernelWithPrintf = inputKernelWeakPtr;
     }
-    Kernel *getKernelForPrintf() {
+    std::weak_ptr<Kernel> getKernelForPrintf() {
         return kernelWithPrintf;
     }
+    void resetKernelForPrintf() {
+        kernelWithPrintf.reset();
+    }
+    void setKernelWithPrintfDeviceMutex(std::mutex *mutexPtr) {
+        kernelWithPrintfDeviceMutex = mutexPtr;
+    }
+    std::mutex *getKernelWithPrintfDeviceMutex() {
+        return kernelWithPrintfDeviceMutex;
+    }
+    void resetKernelWithPrintfDeviceMutex() {
+        kernelWithPrintfDeviceMutex = nullptr;
+    }
 
     bool isSignalScope() const {
         return !!signalScope;
@@ -280,7 +293,8 @@ struct Event : _ze_event_handle_t {
     void *hostAddress = nullptr;
     Device *device = nullptr;
     EventPool *eventPool = nullptr;
-    Kernel *kernelWithPrintf = nullptr;
+    std::weak_ptr<Kernel> kernelWithPrintf = std::weak_ptr<Kernel>{};
+    std::mutex *kernelWithPrintfDeviceMutex = nullptr;
     std::shared_ptr<InOrderExecInfo> inOrderExecInfo;
     CommandQueue *latestUsedCmdQueue = nullptr;
 

level_zero/core/source/event/event_impl.inl

@@ -15,6 +15,7 @@
 #include "shared/source/os_interface/os_context.h"
 #include "shared/source/os_interface/os_time.h"
 
+#include "level_zero/core/source/device/device.h"
 #include "level_zero/core/source/event/event_imp.h"
 #include "level_zero/core/source/gfx_core_helpers/l0_gfx_core_helper.h"
 #include "level_zero/core/source/kernel/kernel.h"
@@ -467,9 +468,13 @@ ze_result_t EventImp<TagSizeT>::hostSynchronize(uint64_t timeout) {
             ret = queryStatus();
         }
         if (ret == ZE_RESULT_SUCCESS) {
-            if (this->getKernelForPrintf() != nullptr) {
+            if (this->getKernelWithPrintfDeviceMutex() != nullptr) {
-                static_cast<Kernel *>(this->getKernelForPrintf())->printPrintfOutput(true);
+                std::lock_guard<std::mutex> lock(*this->getKernelWithPrintfDeviceMutex());
-                this->setKernelForPrintf(nullptr);
+                if (!this->getKernelForPrintf().expired()) {
+                    this->getKernelForPrintf().lock()->printPrintfOutput(true);
+                }
+                this->resetKernelForPrintf();
+                this->resetKernelWithPrintfDeviceMutex();
             }
             if (device->getNEODevice()->getRootDeviceEnvironment().assertHandler.get()) {
                 device->getNEODevice()->getRootDeviceEnvironment().assertHandler->printAssertAndAbort();

level_zero/core/source/kernel/kernel.h

@@ -151,6 +151,7 @@ struct Kernel : _ze_kernel_handle_t, virtual NEO::DispatchKernelEncoderI {
     virtual UnifiedMemoryControls getUnifiedMemoryControls() const = 0;
     virtual bool hasIndirectAllocationsAllowed() const = 0;
 
+    virtual std::mutex *getDevicePrintfKernelMutex() = 0;
     virtual NEO::GraphicsAllocation *getPrintfBufferAllocation() = 0;
     virtual void printPrintfOutput(bool hangDetected) = 0;
 

level_zero/core/source/kernel/kernel_imp.cpp

@@ -1114,11 +1114,11 @@ void KernelImp::createPrintfBuffer() {
         if (pImplicitArgs) {
             pImplicitArgs->printfBufferPtr = printfBuffer->getGpuAddress();
         }
+        this->devicePrintfKernelMutex = &(static_cast<DeviceImp *>(this->module->getDevice())->printfKernelMutex);
     }
 }
 
 void KernelImp::printPrintfOutput(bool hangDetected) {
-    std::lock_guard<std::mutex> lock(this->printfLock);
     PrintfHandler::printOutput(kernelImmData, this->printfBuffer, module->getDevice(), hangDetected);
 }
 

level_zero/core/source/kernel/kernel_imp.h

@@ -14,6 +14,8 @@
 #include "shared/source/unified_memory/unified_memory.h"
 
 #include "level_zero/core/source/kernel/kernel.h"
+#include "level_zero/core/source/module/module.h"
+#include "level_zero/core/source/module/module_imp.h"
 
 #include <memory>
 #include <mutex>
@@ -38,8 +40,12 @@ struct KernelImp : Kernel {
     ~KernelImp() override;
 
     ze_result_t destroy() override {
-        delete this;
+        if (this->devicePrintfKernelMutex == nullptr) {
-        return ZE_RESULT_SUCCESS;
+            delete this;
+            return ZE_RESULT_SUCCESS;
+        } else {
+            return static_cast<ModuleImp *>(this->module)->destroyPrintfKernel(this);
+        }
     }
 
     ze_result_t getBaseAddress(uint64_t *baseAddress) override;
@@ -99,6 +105,7 @@ struct KernelImp : Kernel {
     uint32_t getNumThreadsPerThreadGroup() const override { return numThreadsPerThreadGroup; }
     uint32_t getThreadExecutionMask() const override { return threadExecutionMask; }
 
+    std::mutex *getDevicePrintfKernelMutex() override { return this->devicePrintfKernelMutex; }
     NEO::GraphicsAllocation *getPrintfBufferAllocation() override { return this->printfBuffer; }
     void printPrintfOutput(bool hangDetected) override;
 
@@ -207,6 +214,7 @@ struct KernelImp : Kernel {
     std::vector<KernelImp::KernelArgHandler> kernelArgHandlers;
     std::vector<NEO::GraphicsAllocation *> residencyContainer;
 
+    std::mutex *devicePrintfKernelMutex = nullptr;
     NEO::GraphicsAllocation *printfBuffer = nullptr;
 
     uint32_t groupSize[3] = {0u, 0u, 0u};
@@ -248,7 +256,6 @@ struct KernelImp : Kernel {
     std::unique_ptr<NEO::ImplicitArgs> pImplicitArgs;
 
     std::unique_ptr<KernelExt> pExtension;
-    std::mutex printfLock;
 
     struct SuggestGroupSizeCacheEntry {
         Vec3<size_t> groupSize;

level_zero/core/source/module/module_imp.cpp

@@ -41,6 +41,7 @@
 #include "shared/source/program/program_initialization.h"
 
 #include "level_zero/core/source/device/device.h"
+#include "level_zero/core/source/device/device_imp.h"
 #include "level_zero/core/source/driver/driver_handle.h"
 #include "level_zero/core/source/driver/driver_handle_imp.h"
 #include "level_zero/core/source/gfx_core_helpers/l0_gfx_core_helper.h"
@@ -507,6 +508,11 @@ ModuleImp::ModuleImp(Device *device, ModuleBuildLog *moduleBuildLog, ModuleType
 }
 
 ModuleImp::~ModuleImp() {
+    for (auto &kernel : this->printfKernelContainer) {
+        if (kernel.get() != nullptr) {
+            destroyPrintfKernel(kernel->toHandle());
+        }
+    }
     this->kernelImmDatas.clear();
     if (this->kernelsIsaParentRegion) {
         DEBUG_BREAK_IF(this->device->getNEODevice()->getMemoryManager() == nullptr);
@@ -955,6 +961,9 @@ ze_result_t ModuleImp::createKernel(const ze_kernel_desc_t *desc,
 
     if (res == ZE_RESULT_SUCCESS) {
         *kernelHandle = kernel->toHandle();
+        if (kernel->getPrintfBufferAllocation() != nullptr) {
+            this->printfKernelContainer.push_back(std::shared_ptr<Kernel>(kernel));
+        }
     } else {
         driverHandle->clearErrorDescription();
     }
@@ -973,6 +982,28 @@ ze_result_t ModuleImp::createKernel(const ze_kernel_desc_t *desc,
     return res;
 }
 
+std::weak_ptr<Kernel> ModuleImp::getPrintfKernelWeakPtr(ze_kernel_handle_t kernelHandle) const {
+    std::lock_guard<std::mutex> lock(static_cast<DeviceImp *>(device)->printfKernelMutex);
+    Kernel *kernel = Kernel::fromHandle(kernelHandle);
+    auto it = std::find_if(this->printfKernelContainer.begin(), this->printfKernelContainer.end(), [&kernel](const auto &kernelSharedPtr) { return kernelSharedPtr.get() == kernel; });
+    if (it == this->printfKernelContainer.end()) {
+        return std::weak_ptr<Kernel>{};
+    } else {
+        return std::weak_ptr<Kernel>{*it};
+    }
+}
+
+ze_result_t ModuleImp::destroyPrintfKernel(ze_kernel_handle_t kernelHandle) {
+    std::lock_guard<std::mutex> lock(static_cast<DeviceImp *>(device)->printfKernelMutex);
+    Kernel *kernel = Kernel::fromHandle(kernelHandle);
+    auto it = std::find_if(this->printfKernelContainer.begin(), this->printfKernelContainer.end(), [&kernel](const auto &kernelSharedPtr) { return kernelSharedPtr.get() == kernel; });
+    if (it == this->printfKernelContainer.end()) {
+        return ZE_RESULT_ERROR_INVALID_ARGUMENT;
+    }
+    it->reset();
+    return ZE_RESULT_SUCCESS;
+}
+
 ze_result_t ModuleImp::getNativeBinary(size_t *pSize, uint8_t *pModuleNativeBinary) {
     auto genBinary = this->translationUnit->packedDeviceBinary.get();
 

level_zero/core/source/module/module_imp.h

@@ -11,6 +11,7 @@
 #include "shared/source/compiler_interface/linker.h"
 #include "shared/source/program/program_info.h"
 
+#include "level_zero/core/source/kernel/kernel.h"
 #include "level_zero/core/source/module/module.h"
 
 #include "igfxfmid.h"
@@ -155,6 +156,10 @@ struct ModuleImp : public Module {
         return this->translationUnit.get();
     }
 
+    std::vector<std::shared_ptr<Kernel>> &getPrintfKernelContainer() { return this->printfKernelContainer; }
+    std::weak_ptr<Kernel> getPrintfKernelWeakPtr(ze_kernel_handle_t kernelHandle) const;
+    ze_result_t destroyPrintfKernel(ze_kernel_handle_t kernelHandle);
+
   protected:
     MOCKABLE_VIRTUAL ze_result_t initializeTranslationUnit(const ze_module_desc_t *desc, NEO::Device *neoDevice);
     ze_result_t checkIfBuildShouldBeFailed(NEO::Device *neoDevice);
@@ -182,6 +187,7 @@ struct ModuleImp : public Module {
     ModuleBuildLog *moduleBuildLog = nullptr;
     NEO::GraphicsAllocation *exportedFunctionsSurface = nullptr;
     std::unique_ptr<NEO::GraphicsAllocation> kernelsIsaParentRegion;
+    std::vector<std::shared_ptr<Kernel>> printfKernelContainer;
     std::vector<std::unique_ptr<KernelImmutableData>> kernelImmDatas;
     NEO::Linker::RelocatedSymbolsMap symbols;
 

level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_launch_kernel_1.cpp

@@ -226,36 +226,41 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenNotEnoughSpaceInCommandStreamWhenA
 }
 
 HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfUsedWhenAppendedToCommandListThenKernelIsStored) {
-    createKernel();
     ze_result_t returnValue;
     std::unique_ptr<L0::CommandList> commandList(CommandList::create(productFamily, device, NEO::EngineGroupType::RenderCompute, 0u, returnValue));
     ze_group_count_t groupCount{1, 1, 1};
 
-    EXPECT_TRUE(kernel->kernelImmData->getDescriptor().kernelAttributes.flags.usesPrintf);
+    auto kernel = new Mock<KernelImp>{};
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
+
     CmdListKernelLaunchParams launchParams = {};
-    auto result = commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false));
-    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
 
     EXPECT_EQ(1u, commandList->getPrintfKernelContainer().size());
-    EXPECT_EQ(kernel.get(), commandList->getPrintfKernelContainer()[0]);
+    EXPECT_EQ(kernel, commandList->getPrintfKernelContainer()[0].lock().get());
 }
 
 HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfUsedWhenAppendedToCommandListMultipleTimesThenKernelIsStoredOnce) {
-    createKernel();
     ze_result_t returnValue;
     std::unique_ptr<L0::CommandList> commandList(CommandList::create(productFamily, device, NEO::EngineGroupType::RenderCompute, 0u, returnValue));
     ze_group_count_t groupCount{1, 1, 1};
 
-    EXPECT_TRUE(kernel->kernelImmData->getDescriptor().kernelAttributes.flags.usesPrintf);
+    auto kernel = new Mock<KernelImp>{};
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
+
     CmdListKernelLaunchParams launchParams = {};
-    auto result = commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false));
-    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
 
     EXPECT_EQ(1u, commandList->getPrintfKernelContainer().size());
-    EXPECT_EQ(kernel.get(), commandList->getPrintfKernelContainer()[0]);
+    EXPECT_EQ(kernel, commandList->getPrintfKernelContainer()[0].lock().get());
 
-    result = commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false));
-    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
     EXPECT_EQ(1u, commandList->getPrintfKernelContainer().size());
 }
 
@@ -269,25 +274,27 @@ HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfWhenAppendedToSynch
 
     std::unique_ptr<L0::CommandList> commandList(CommandList::createImmediate(productFamily, device, &queueDesc, false, NEO::EngineGroupType::RenderCompute, returnValue));
 
-    Mock<KernelImp> kernel;
+    auto kernel = new Mock<KernelImp>{};
-    kernel.descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
 
+    EXPECT_EQ(0u, kernel->printPrintfOutputCalledTimes);
     ze_group_count_t groupCount{1, 1, 1};
     CmdListKernelLaunchParams launchParams = {};
-    auto result = commandList->appendLaunchKernel(kernel.toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false));
-    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
-    EXPECT_EQ(1u, kernel.printPrintfOutputCalledTimes);
+    EXPECT_FALSE(kernel->hangDetectedPassedToPrintfOutput);
-    EXPECT_FALSE(kernel.hangDetectedPassedToPrintfOutput);
     EXPECT_EQ(0u, commandList->getPrintfKernelContainer().size());
 
-    result = commandList->appendLaunchKernel(kernel.toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false));
-    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+    EXPECT_EQ(2u, kernel->printPrintfOutputCalledTimes);
-    EXPECT_EQ(2u, kernel.printPrintfOutputCalledTimes);
+    EXPECT_FALSE(kernel->hangDetectedPassedToPrintfOutput);
-    EXPECT_FALSE(kernel.hangDetectedPassedToPrintfOutput);
     EXPECT_EQ(0u, commandList->getPrintfKernelContainer().size());
 }
 
-HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfWhenAppendedToAsynchronousImmCommandListThenPrintfBufferIsPrinted) {
+HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfWhenAppendedToAsynchronousImmCommandListThenPrintfBufferIsNotPrintedUntilHostSync) {
     DebugManagerStateRestore dbgRestorer;
     DebugManager.flags.EnableFlushTaskSubmission.set(1);
 
@@ -297,21 +304,26 @@ HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfWhenAppendedToAsync
 
     std::unique_ptr<L0::CommandList> commandList(CommandList::createImmediate(productFamily, device, &queueDesc, false, NEO::EngineGroupType::RenderCompute, returnValue));
 
-    Mock<KernelImp> kernel;
+    auto kernel = new Mock<KernelImp>{};
-    kernel.descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
 
     ze_group_count_t groupCount{1, 1, 1};
     CmdListKernelLaunchParams launchParams = {};
-    auto result = commandList->appendLaunchKernel(kernel.toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false));
-    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+    EXPECT_EQ(0u, kernel->printPrintfOutputCalledTimes);
-    EXPECT_EQ(1u, kernel.printPrintfOutputCalledTimes);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->hostSynchronize(std::numeric_limits<uint64_t>::max()));
-    EXPECT_FALSE(kernel.hangDetectedPassedToPrintfOutput);
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_FALSE(kernel->hangDetectedPassedToPrintfOutput);
     EXPECT_EQ(0u, commandList->getPrintfKernelContainer().size());
 
-    result = commandList->appendLaunchKernel(kernel.toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false));
-    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
-    EXPECT_EQ(2u, kernel.printPrintfOutputCalledTimes);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->hostSynchronize(std::numeric_limits<uint64_t>::max()));
-    EXPECT_FALSE(kernel.hangDetectedPassedToPrintfOutput);
+    EXPECT_EQ(2u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_FALSE(kernel->hangDetectedPassedToPrintfOutput);
     EXPECT_EQ(0u, commandList->getPrintfKernelContainer().size());
 }
 
@@ -330,22 +342,476 @@ HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfWhenAppendToSynchro
 
     std::unique_ptr<L0::CommandList> commandList(CommandList::createImmediate(productFamily, device, &queueDesc, false, NEO::EngineGroupType::RenderCompute, returnValue));
 
-    Mock<KernelImp> kernel;
+    auto kernel = new Mock<KernelImp>{};
-    kernel.descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
+
+    EXPECT_EQ(0u, kernel->printPrintfOutputCalledTimes);
+    ze_group_count_t groupCount{1, 1, 1};
+    CmdListKernelLaunchParams launchParams = {};
+    EXPECT_EQ(ZE_RESULT_ERROR_DEVICE_LOST, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false));
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_TRUE(kernel->hangDetectedPassedToPrintfOutput);
+    EXPECT_EQ(0u, commandList->getPrintfKernelContainer().size());
+
+    EXPECT_EQ(ZE_RESULT_ERROR_DEVICE_LOST, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false));
+    EXPECT_EQ(2u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_TRUE(kernel->hangDetectedPassedToPrintfOutput);
+    EXPECT_EQ(0u, commandList->getPrintfKernelContainer().size());
+}
+
+HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfAppendedToCommandListAndDestroyKernelAfterQueueSyncThenSuccessIsReturned) {
+    ze_result_t result;
+
+    ze_command_queue_handle_t commandQueueHandle;
+    ze_command_queue_desc_t queueDesc{};
+    queueDesc.stype = ZE_STRUCTURE_TYPE_COMMAND_QUEUE_DESC;
+    queueDesc.pNext = nullptr;
+    queueDesc.mode = ZE_COMMAND_QUEUE_MODE_ASYNCHRONOUS;
+    queueDesc.ordinal = 0u;
+    queueDesc.index = 0u;
+    EXPECT_EQ(ZE_RESULT_SUCCESS, device->createCommandQueue(&queueDesc, &commandQueueHandle));
+    auto commandQueue = static_cast<L0::ult::CommandQueue *>(L0::CommandQueue::fromHandle(commandQueueHandle));
+    EXPECT_NE(nullptr, commandQueue);
+
+    std::unique_ptr<L0::CommandList> commandList(CommandList::create(productFamily, device, NEO::EngineGroupType::RenderCompute, 0u, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    auto kernel = new Mock<KernelImp>{};
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
 
     ze_group_count_t groupCount{1, 1, 1};
     CmdListKernelLaunchParams launchParams = {};
-    auto result = commandList->appendLaunchKernel(kernel.toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false));
-    EXPECT_EQ(ZE_RESULT_ERROR_DEVICE_LOST, result);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->close());
-    EXPECT_EQ(1u, kernel.printPrintfOutputCalledTimes);
+    EXPECT_EQ(1u, commandList->getPrintfKernelContainer().size());
-    EXPECT_TRUE(kernel.hangDetectedPassedToPrintfOutput);
+    EXPECT_EQ(kernel, commandList->getPrintfKernelContainer()[0].lock().get());
-    EXPECT_EQ(0u, commandList->getPrintfKernelContainer().size());
 
-    result = commandList->appendLaunchKernel(kernel.toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false);
+    auto commandListHandle = commandList->toHandle();
-    EXPECT_EQ(ZE_RESULT_ERROR_DEVICE_LOST, result);
+    EXPECT_EQ(0u, kernel->printPrintfOutputCalledTimes);
-    EXPECT_EQ(2u, kernel.printPrintfOutputCalledTimes);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandQueue->executeCommandLists(1, &commandListHandle, nullptr, false));
-    EXPECT_TRUE(kernel.hangDetectedPassedToPrintfOutput);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandQueue->synchronize(std::numeric_limits<uint64_t>::max()));
-    EXPECT_EQ(0u, commandList->getPrintfKernelContainer().size());
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, kernel->destroy());
+    EXPECT_EQ(nullptr, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer()[0].get());
+    EXPECT_EQ(1u, commandList->getPrintfKernelContainer().size());
+    EXPECT_EQ(nullptr, commandList->getPrintfKernelContainer()[0].lock().get());
+
+    commandQueue->destroy();
+}
+
+HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfAndEventAppendedToCommandListAndDestroyKernelAfterQueueSyncThenSuccessIsReturnedAndEventSyncDoesNotAccessKernel) {
+    ze_result_t result;
+
+    ze_command_queue_handle_t commandQueueHandle;
+    ze_command_queue_desc_t queueDesc{};
+    queueDesc.stype = ZE_STRUCTURE_TYPE_COMMAND_QUEUE_DESC;
+    queueDesc.pNext = nullptr;
+    queueDesc.mode = ZE_COMMAND_QUEUE_MODE_ASYNCHRONOUS;
+    queueDesc.ordinal = 0u;
+    queueDesc.index = 0u;
+    EXPECT_EQ(ZE_RESULT_SUCCESS, device->createCommandQueue(&queueDesc, &commandQueueHandle));
+    auto commandQueue = static_cast<L0::ult::CommandQueue *>(L0::CommandQueue::fromHandle(commandQueueHandle));
+    EXPECT_NE(nullptr, commandQueue);
+
+    std::unique_ptr<L0::CommandList> commandList(CommandList::create(productFamily, device, NEO::EngineGroupType::RenderCompute, 0u, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    auto kernel = new Mock<KernelImp>{};
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
+
+    ze_event_pool_desc_t eventPoolDesc{};
+    eventPoolDesc.stype = ZE_STRUCTURE_TYPE_EVENT_POOL_DESC;
+    eventPoolDesc.pNext = nullptr;
+    eventPoolDesc.count = 1;
+    eventPoolDesc.flags = ZE_EVENT_POOL_FLAG_HOST_VISIBLE;
+    auto eventPool = std::unique_ptr<::L0::EventPool>(::L0::EventPool::create(driverHandle.get(), context, 0, nullptr, &eventPoolDesc, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    ze_event_desc_t eventDesc{};
+    eventDesc.stype = ZE_STRUCTURE_TYPE_EVENT_DESC;
+    eventDesc.pNext = nullptr;
+    eventDesc.index = 0;
+    eventDesc.signal = ZE_EVENT_SCOPE_FLAG_HOST;
+    eventDesc.wait = ZE_EVENT_SCOPE_FLAG_HOST;
+    auto event = std::unique_ptr<::L0::Event>(::L0::Event::create<typename FamilyType::TimestampPacketType>(eventPool.get(), &eventDesc, device));
+    auto hEventHandle = event->toHandle();
+
+    ze_group_count_t groupCount{1, 1, 1};
+    CmdListKernelLaunchParams launchParams = {};
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, hEventHandle, 0, nullptr, launchParams, false));
+    EXPECT_EQ(kernel, event->getKernelForPrintf().lock().get());
+    EXPECT_NE(nullptr, event->getKernelWithPrintfDeviceMutex());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->close());
+    EXPECT_EQ(1u, commandList->getPrintfKernelContainer().size());
+    EXPECT_EQ(kernel, commandList->getPrintfKernelContainer()[0].lock().get());
+
+    auto commandListHandle = commandList->toHandle();
+    EXPECT_EQ(0u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandQueue->executeCommandLists(1, &commandListHandle, nullptr, false));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandQueue->synchronize(std::numeric_limits<uint64_t>::max()));
+    *reinterpret_cast<uint32_t *>(event->getHostAddress()) = Event::STATE_SIGNALED;
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, kernel->destroy());
+    EXPECT_EQ(nullptr, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer()[0].get());
+    EXPECT_EQ(1u, commandList->getPrintfKernelContainer().size());
+    EXPECT_EQ(nullptr, commandList->getPrintfKernelContainer()[0].lock().get());
+    EXPECT_EQ(nullptr, event->getKernelForPrintf().lock().get());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, event->queryStatus());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, event->hostSynchronize(std::numeric_limits<uint64_t>::max()));
+    EXPECT_EQ(nullptr, event->getKernelWithPrintfDeviceMutex());
+
+    commandQueue->destroy();
+}
+
+HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfAndEventAppendedToCommandListAndDestroyKernelAfterEventSyncThenSuccessIsReturnedAndQueueSyncDoesNotAccessKernel) {
+    ze_result_t result;
+
+    ze_command_queue_handle_t commandQueueHandle;
+    ze_command_queue_desc_t queueDesc{};
+    queueDesc.stype = ZE_STRUCTURE_TYPE_COMMAND_QUEUE_DESC;
+    queueDesc.pNext = nullptr;
+    queueDesc.mode = ZE_COMMAND_QUEUE_MODE_ASYNCHRONOUS;
+    queueDesc.ordinal = 0u;
+    queueDesc.index = 0u;
+    EXPECT_EQ(ZE_RESULT_SUCCESS, device->createCommandQueue(&queueDesc, &commandQueueHandle));
+    auto commandQueue = static_cast<L0::ult::CommandQueue *>(L0::CommandQueue::fromHandle(commandQueueHandle));
+    EXPECT_NE(nullptr, commandQueue);
+
+    std::unique_ptr<L0::CommandList> commandList(CommandList::create(productFamily, device, NEO::EngineGroupType::RenderCompute, 0u, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    auto kernel = new Mock<KernelImp>{};
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
+
+    ze_event_pool_desc_t eventPoolDesc{};
+    eventPoolDesc.stype = ZE_STRUCTURE_TYPE_EVENT_POOL_DESC;
+    eventPoolDesc.pNext = nullptr;
+    eventPoolDesc.count = 1;
+    eventPoolDesc.flags = ZE_EVENT_POOL_FLAG_HOST_VISIBLE;
+    auto eventPool = std::unique_ptr<::L0::EventPool>(::L0::EventPool::create(driverHandle.get(), context, 0, nullptr, &eventPoolDesc, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    ze_event_desc_t eventDesc{};
+    eventDesc.stype = ZE_STRUCTURE_TYPE_EVENT_DESC;
+    eventDesc.pNext = nullptr;
+    eventDesc.index = 0;
+    eventDesc.signal = ZE_EVENT_SCOPE_FLAG_HOST;
+    eventDesc.wait = ZE_EVENT_SCOPE_FLAG_HOST;
+    auto event = std::unique_ptr<::L0::Event>(::L0::Event::create<typename FamilyType::TimestampPacketType>(eventPool.get(), &eventDesc, device));
+    auto hEventHandle = event->toHandle();
+
+    ze_group_count_t groupCount{1, 1, 1};
+    CmdListKernelLaunchParams launchParams = {};
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, hEventHandle, 0, nullptr, launchParams, false));
+    EXPECT_EQ(kernel, event->getKernelForPrintf().lock().get());
+    EXPECT_NE(nullptr, event->getKernelWithPrintfDeviceMutex());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->close());
+    EXPECT_EQ(1u, commandList->getPrintfKernelContainer().size());
+    EXPECT_EQ(kernel, commandList->getPrintfKernelContainer()[0].lock().get());
+
+    auto commandListHandle = commandList->toHandle();
+    EXPECT_EQ(0u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandQueue->executeCommandLists(1, &commandListHandle, nullptr, false));
+    *reinterpret_cast<uint32_t *>(event->getHostAddress()) = Event::STATE_SIGNALED;
+    EXPECT_EQ(ZE_RESULT_SUCCESS, event->hostSynchronize(std::numeric_limits<uint64_t>::max()));
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_EQ(nullptr, event->getKernelForPrintf().lock().get());
+    EXPECT_EQ(nullptr, event->getKernelWithPrintfDeviceMutex());
+    EXPECT_EQ(1u, commandList->getPrintfKernelContainer().size());
+    EXPECT_EQ(kernel, commandList->getPrintfKernelContainer()[0].lock().get());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, kernel->destroy());
+    EXPECT_EQ(nullptr, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer()[0].get());
+    EXPECT_EQ(1u, commandList->getPrintfKernelContainer().size());
+    EXPECT_EQ(nullptr, commandList->getPrintfKernelContainer()[0].lock().get());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandQueue->synchronize(std::numeric_limits<uint64_t>::max()));
+
+    commandQueue->destroy();
+}
+
+HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfAppendedToImmCommandListWithoutFlushTaskSubmissionAndDestroyKernelAfterListSyncThenSuccessIsReturned) {
+    DebugManagerStateRestore restorer;
+    NEO::DebugManager.flags.EnableFlushTaskSubmission.set(0);
+
+    ze_result_t result;
+    ze_command_queue_desc_t queueDesc{};
+    queueDesc.stype = ZE_STRUCTURE_TYPE_COMMAND_QUEUE_DESC;
+    queueDesc.pNext = nullptr;
+    queueDesc.mode = ZE_COMMAND_QUEUE_MODE_ASYNCHRONOUS;
+    queueDesc.ordinal = 0u;
+    queueDesc.index = 0u;
+    std::unique_ptr<L0::CommandList> commandList(CommandList::createImmediate(productFamily, device, &queueDesc, false, NEO::EngineGroupType::RenderCompute, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    auto kernel = new Mock<KernelImp>{};
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
+
+    EXPECT_EQ(0u, kernel->printPrintfOutputCalledTimes);
+    ze_group_count_t groupCount{1, 1, 1};
+    CmdListKernelLaunchParams launchParams = {};
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->hostSynchronize(std::numeric_limits<uint64_t>::max()));
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, kernel->destroy());
+    EXPECT_EQ(nullptr, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer()[0].get());
+}
+
+HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfAndEventAppendedToImmCommandListWithoutFlushTaskSubmissionAndDestroyKernelAfterListSyncThenSuccessIsReturnedAndEventSyncDoesNotAccessKernel) {
+    DebugManagerStateRestore restorer;
+    NEO::DebugManager.flags.EnableFlushTaskSubmission.set(0);
+
+    ze_result_t result;
+    ze_command_queue_desc_t queueDesc{};
+    queueDesc.stype = ZE_STRUCTURE_TYPE_COMMAND_QUEUE_DESC;
+    queueDesc.pNext = nullptr;
+    queueDesc.mode = ZE_COMMAND_QUEUE_MODE_ASYNCHRONOUS;
+    queueDesc.ordinal = 0u;
+    queueDesc.index = 0u;
+    std::unique_ptr<L0::CommandList> commandList(CommandList::createImmediate(productFamily, device, &queueDesc, false, NEO::EngineGroupType::RenderCompute, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    auto kernel = new Mock<KernelImp>{};
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
+
+    ze_event_pool_desc_t eventPoolDesc{};
+    eventPoolDesc.stype = ZE_STRUCTURE_TYPE_EVENT_POOL_DESC;
+    eventPoolDesc.pNext = nullptr;
+    eventPoolDesc.count = 1;
+    eventPoolDesc.flags = ZE_EVENT_POOL_FLAG_HOST_VISIBLE;
+    auto eventPool = std::unique_ptr<::L0::EventPool>(::L0::EventPool::create(driverHandle.get(), context, 0, nullptr, &eventPoolDesc, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    ze_event_desc_t eventDesc{};
+    eventDesc.stype = ZE_STRUCTURE_TYPE_EVENT_DESC;
+    eventDesc.pNext = nullptr;
+    eventDesc.index = 0;
+    eventDesc.signal = ZE_EVENT_SCOPE_FLAG_HOST;
+    eventDesc.wait = ZE_EVENT_SCOPE_FLAG_HOST;
+    auto event = std::unique_ptr<::L0::Event>(::L0::Event::create<typename FamilyType::TimestampPacketType>(eventPool.get(), &eventDesc, device));
+    auto hEventHandle = event->toHandle();
+
+    EXPECT_EQ(0u, kernel->printPrintfOutputCalledTimes);
+    ze_group_count_t groupCount{1, 1, 1};
+    CmdListKernelLaunchParams launchParams = {};
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, hEventHandle, 0, nullptr, launchParams, false));
+    EXPECT_EQ(kernel, event->getKernelForPrintf().lock().get());
+    EXPECT_NE(nullptr, event->getKernelWithPrintfDeviceMutex());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->hostSynchronize(std::numeric_limits<uint64_t>::max()));
+    *reinterpret_cast<uint32_t *>(event->getHostAddress()) = Event::STATE_SIGNALED;
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, kernel->destroy());
+    EXPECT_EQ(nullptr, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer()[0].get());
+    EXPECT_EQ(nullptr, event->getKernelForPrintf().lock().get());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, event->queryStatus());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, event->hostSynchronize(std::numeric_limits<uint64_t>::max()));
+    EXPECT_EQ(nullptr, event->getKernelWithPrintfDeviceMutex());
+}
+
+HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfAndEventAppendedToImmCommandListWithoutFlushTaskSubmissionAndDestroyKernelAfterEventSyncThenSuccessIsReturnedAndListSyncDoesNotAccessKernel) {
+    DebugManagerStateRestore restorer;
+    NEO::DebugManager.flags.EnableFlushTaskSubmission.set(0);
+
+    ze_result_t result;
+    ze_command_queue_desc_t queueDesc{};
+    queueDesc.stype = ZE_STRUCTURE_TYPE_COMMAND_QUEUE_DESC;
+    queueDesc.pNext = nullptr;
+    queueDesc.mode = ZE_COMMAND_QUEUE_MODE_ASYNCHRONOUS;
+    queueDesc.ordinal = 0u;
+    queueDesc.index = 0u;
+    std::unique_ptr<L0::CommandList> commandList(CommandList::createImmediate(productFamily, device, &queueDesc, false, NEO::EngineGroupType::RenderCompute, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    auto kernel = new Mock<KernelImp>{};
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
+
+    ze_event_pool_desc_t eventPoolDesc{};
+    eventPoolDesc.stype = ZE_STRUCTURE_TYPE_EVENT_POOL_DESC;
+    eventPoolDesc.pNext = nullptr;
+    eventPoolDesc.count = 1;
+    eventPoolDesc.flags = ZE_EVENT_POOL_FLAG_HOST_VISIBLE;
+    auto eventPool = std::unique_ptr<::L0::EventPool>(::L0::EventPool::create(driverHandle.get(), context, 0, nullptr, &eventPoolDesc, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    ze_event_desc_t eventDesc{};
+    eventDesc.stype = ZE_STRUCTURE_TYPE_EVENT_DESC;
+    eventDesc.pNext = nullptr;
+    eventDesc.index = 0;
+    eventDesc.signal = ZE_EVENT_SCOPE_FLAG_HOST;
+    eventDesc.wait = ZE_EVENT_SCOPE_FLAG_HOST;
+    auto event = std::unique_ptr<::L0::Event>(::L0::Event::create<typename FamilyType::TimestampPacketType>(eventPool.get(), &eventDesc, device));
+    auto hEventHandle = event->toHandle();
+
+    EXPECT_EQ(0u, kernel->printPrintfOutputCalledTimes);
+    ze_group_count_t groupCount{1, 1, 1};
+    CmdListKernelLaunchParams launchParams = {};
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, hEventHandle, 0, nullptr, launchParams, false));
+    EXPECT_EQ(kernel, event->getKernelForPrintf().lock().get());
+    EXPECT_NE(nullptr, event->getKernelWithPrintfDeviceMutex());
+    *reinterpret_cast<uint32_t *>(event->getHostAddress()) = Event::STATE_SIGNALED;
+    EXPECT_EQ(ZE_RESULT_SUCCESS, event->hostSynchronize(std::numeric_limits<uint64_t>::max()));
+    EXPECT_EQ(2u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_EQ(nullptr, event->getKernelForPrintf().lock().get());
+    EXPECT_EQ(nullptr, event->getKernelWithPrintfDeviceMutex());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, kernel->destroy());
+    EXPECT_EQ(nullptr, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer()[0].get());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->hostSynchronize(std::numeric_limits<uint64_t>::max()));
+}
+
+HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfAppendedToImmCommandListWithFlushTaskSubmissionAndDestroyKernelAfterListSyncThenSuccessIsReturned) {
+    DebugManagerStateRestore restorer;
+    NEO::DebugManager.flags.EnableFlushTaskSubmission.set(1);
+
+    ze_result_t result;
+    ze_command_queue_desc_t queueDesc{};
+    queueDesc.stype = ZE_STRUCTURE_TYPE_COMMAND_QUEUE_DESC;
+    queueDesc.pNext = nullptr;
+    queueDesc.mode = ZE_COMMAND_QUEUE_MODE_ASYNCHRONOUS;
+    queueDesc.ordinal = 0u;
+    queueDesc.index = 0u;
+    std::unique_ptr<L0::CommandList> commandList(CommandList::createImmediate(productFamily, device, &queueDesc, false, NEO::EngineGroupType::RenderCompute, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    auto kernel = new Mock<KernelImp>{};
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
+
+    EXPECT_EQ(0u, kernel->printPrintfOutputCalledTimes);
+    ze_group_count_t groupCount{1, 1, 1};
+    CmdListKernelLaunchParams launchParams = {};
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->hostSynchronize(std::numeric_limits<uint64_t>::max()));
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, kernel->destroy());
+    EXPECT_EQ(nullptr, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer()[0].get());
+}
+
+HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfAndEventAppendedToImmCommandListWithFlushTaskSubmissionAndDestroyKernelAfterListSyncThenSuccessIsReturnedAndEventSyncDoesNotAccessKernel) {
+    DebugManagerStateRestore restorer;
+    NEO::DebugManager.flags.EnableFlushTaskSubmission.set(1);
+
+    ze_result_t result;
+    ze_command_queue_desc_t queueDesc{};
+    queueDesc.stype = ZE_STRUCTURE_TYPE_COMMAND_QUEUE_DESC;
+    queueDesc.pNext = nullptr;
+    queueDesc.mode = ZE_COMMAND_QUEUE_MODE_ASYNCHRONOUS;
+    queueDesc.ordinal = 0u;
+    queueDesc.index = 0u;
+    std::unique_ptr<L0::CommandList> commandList(CommandList::createImmediate(productFamily, device, &queueDesc, false, NEO::EngineGroupType::RenderCompute, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    auto kernel = new Mock<KernelImp>{};
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
+
+    ze_event_pool_desc_t eventPoolDesc{};
+    eventPoolDesc.stype = ZE_STRUCTURE_TYPE_EVENT_POOL_DESC;
+    eventPoolDesc.pNext = nullptr;
+    eventPoolDesc.count = 1;
+    eventPoolDesc.flags = ZE_EVENT_POOL_FLAG_HOST_VISIBLE;
+    auto eventPool = std::unique_ptr<::L0::EventPool>(::L0::EventPool::create(driverHandle.get(), context, 0, nullptr, &eventPoolDesc, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    ze_event_desc_t eventDesc{};
+    eventDesc.stype = ZE_STRUCTURE_TYPE_EVENT_DESC;
+    eventDesc.pNext = nullptr;
+    eventDesc.index = 0;
+    eventDesc.signal = ZE_EVENT_SCOPE_FLAG_HOST;
+    eventDesc.wait = ZE_EVENT_SCOPE_FLAG_HOST;
+    auto event = std::unique_ptr<::L0::Event>(::L0::Event::create<typename FamilyType::TimestampPacketType>(eventPool.get(), &eventDesc, device));
+    auto hEventHandle = event->toHandle();
+
+    EXPECT_EQ(0u, kernel->printPrintfOutputCalledTimes);
+    ze_group_count_t groupCount{1, 1, 1};
+    CmdListKernelLaunchParams launchParams = {};
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, hEventHandle, 0, nullptr, launchParams, false));
+    EXPECT_EQ(kernel, event->getKernelForPrintf().lock().get());
+    EXPECT_NE(nullptr, event->getKernelWithPrintfDeviceMutex());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->hostSynchronize(std::numeric_limits<uint64_t>::max()));
+    *reinterpret_cast<uint32_t *>(event->getHostAddress()) = Event::STATE_SIGNALED;
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, kernel->destroy());
+    EXPECT_EQ(nullptr, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer()[0].get());
+    EXPECT_EQ(nullptr, event->getKernelForPrintf().lock().get());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, event->queryStatus());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, event->hostSynchronize(std::numeric_limits<uint64_t>::max()));
+    EXPECT_EQ(nullptr, event->getKernelWithPrintfDeviceMutex());
+}
+
+HWTEST_F(CommandListAppendLaunchKernel, givenKernelWithPrintfAndEventAppendedToImmCommandListWithFlushTaskSubmissionAndDestroyKernelAfterEventSyncThenSuccessIsReturnedAndListSyncDoesNotAccessKernel) {
+    DebugManagerStateRestore restorer;
+    NEO::DebugManager.flags.EnableFlushTaskSubmission.set(1);
+
+    ze_result_t result;
+    ze_command_queue_desc_t queueDesc{};
+    queueDesc.stype = ZE_STRUCTURE_TYPE_COMMAND_QUEUE_DESC;
+    queueDesc.pNext = nullptr;
+    queueDesc.mode = ZE_COMMAND_QUEUE_MODE_ASYNCHRONOUS;
+    queueDesc.ordinal = 0u;
+    queueDesc.index = 0u;
+    std::unique_ptr<L0::CommandList> commandList(CommandList::createImmediate(productFamily, device, &queueDesc, false, NEO::EngineGroupType::RenderCompute, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    auto kernel = new Mock<KernelImp>{};
+    kernel->module = module.get();
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->createPrintfBuffer();
+    static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().push_back(std::shared_ptr<Kernel>{kernel});
+
+    ze_event_pool_desc_t eventPoolDesc{};
+    eventPoolDesc.stype = ZE_STRUCTURE_TYPE_EVENT_POOL_DESC;
+    eventPoolDesc.pNext = nullptr;
+    eventPoolDesc.count = 1;
+    eventPoolDesc.flags = ZE_EVENT_POOL_FLAG_HOST_VISIBLE;
+    auto eventPool = std::unique_ptr<::L0::EventPool>(::L0::EventPool::create(driverHandle.get(), context, 0, nullptr, &eventPoolDesc, result));
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    ze_event_desc_t eventDesc{};
+    eventDesc.stype = ZE_STRUCTURE_TYPE_EVENT_DESC;
+    eventDesc.pNext = nullptr;
+    eventDesc.index = 0;
+    eventDesc.signal = ZE_EVENT_SCOPE_FLAG_HOST;
+    eventDesc.wait = ZE_EVENT_SCOPE_FLAG_HOST;
+    auto event = std::unique_ptr<::L0::Event>(::L0::Event::create<typename FamilyType::TimestampPacketType>(eventPool.get(), &eventDesc, device));
+    auto hEventHandle = event->toHandle();
+
+    EXPECT_EQ(0u, kernel->printPrintfOutputCalledTimes);
+    ze_group_count_t groupCount{1, 1, 1};
+    CmdListKernelLaunchParams launchParams = {};
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->appendLaunchKernel(kernel->toHandle(), groupCount, hEventHandle, 0, nullptr, launchParams, false));
+    EXPECT_EQ(kernel, event->getKernelForPrintf().lock().get());
+    EXPECT_NE(nullptr, event->getKernelWithPrintfDeviceMutex());
+    *reinterpret_cast<uint32_t *>(event->getHostAddress()) = Event::STATE_SIGNALED;
+    EXPECT_EQ(ZE_RESULT_SUCCESS, event->hostSynchronize(std::numeric_limits<uint64_t>::max()));
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
+    EXPECT_EQ(nullptr, event->getKernelForPrintf().lock().get());
+    EXPECT_EQ(nullptr, event->getKernelWithPrintfDeviceMutex());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, kernel->destroy());
+    EXPECT_EQ(nullptr, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer()[0].get());
+    EXPECT_EQ(ZE_RESULT_SUCCESS, commandList->hostSynchronize(std::numeric_limits<uint64_t>::max()));
 }
 
 HWTEST_F(CommandListAppendLaunchKernel, WhenAppendingMultipleTimesThenSshIsNotDepletedButReallocated) {

level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_launch_kernel_3.cpp

@@ -274,7 +274,8 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenImmediateCommandListWhenAppendingL
 
 HWTEST2_F(CommandListAppendLaunchKernel, givenNonemptyAllocPrintfBufferKernelWhenAppendingLaunchKernelIndirectThenKernelIsStoredOnEvent, IsAtLeastSkl) {
     Mock<Module> module(this->device, nullptr);
-    Mock<::L0::KernelImp> kernel;
+    auto kernel = new Mock<::L0::KernelImp>{};
+    static_cast<ModuleImp *>(&module)->getPrintfKernelContainer().push_back(std::shared_ptr<Mock<::L0::KernelImp>>{kernel});
 
     ze_result_t returnValue;
     std::unique_ptr<L0::CommandList> commandList(L0::CommandList::create(productFamily, device, NEO::EngineGroupType::RenderCompute, 0u, returnValue));
@@ -282,9 +283,9 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenNonemptyAllocPrintfBufferKernelWhe
     eventPoolDesc.flags = ZE_EVENT_POOL_FLAG_HOST_VISIBLE;
     eventPoolDesc.count = 1;
 
-    kernel.module = &module;
+    kernel->module = &module;
-    kernel.descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
-    kernel.createPrintfBuffer();
+    kernel->createPrintfBuffer();
 
     ze_event_desc_t eventDesc = {};
     eventDesc.index = 0;
@@ -294,15 +295,16 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenNonemptyAllocPrintfBufferKernelWhe
     auto event = std::unique_ptr<L0::Event>(L0::Event::create<typename FamilyType::TimestampPacketType>(eventPool.get(), &eventDesc, device));
 
     ze_group_count_t groupCount{1, 1, 1};
-    auto result = commandList->appendLaunchKernelIndirect(kernel.toHandle(), groupCount, event->toHandle(), 0, nullptr, false);
+    auto result = commandList->appendLaunchKernelIndirect(kernel->toHandle(), groupCount, event->toHandle(), 0, nullptr, false);
     EXPECT_EQ(ZE_RESULT_SUCCESS, result);
 
-    ASSERT_NE(nullptr, event->getKernelForPrintf());
+    ASSERT_FALSE(event->getKernelForPrintf().expired());
 }
 
 HWTEST2_F(CommandListAppendLaunchKernel, givenEmptyAllocPrintfBufferKernelWhenAppendingLaunchKernelIndirectThenKernelIsNotStoredOnEvent, IsAtLeastSkl) {
     Mock<Module> module(this->device, nullptr);
-    Mock<::L0::KernelImp> kernel;
+    auto kernel = new Mock<::L0::KernelImp>{};
+    static_cast<ModuleImp *>(&module)->getPrintfKernelContainer().push_back(std::shared_ptr<Mock<::L0::KernelImp>>{kernel});
 
     ze_result_t returnValue;
     std::unique_ptr<L0::CommandList> commandList(L0::CommandList::create(productFamily, device, NEO::EngineGroupType::RenderCompute, 0u, returnValue));
@@ -310,8 +312,8 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenEmptyAllocPrintfBufferKernelWhenAp
     eventPoolDesc.flags = ZE_EVENT_POOL_FLAG_HOST_VISIBLE;
     eventPoolDesc.count = 1;
 
-    kernel.module = &module;
+    kernel->module = &module;
-    kernel.descriptor.kernelAttributes.flags.usesPrintf = false;
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = false;
 
     ze_event_desc_t eventDesc = {};
     eventDesc.index = 0;
@@ -321,24 +323,25 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenEmptyAllocPrintfBufferKernelWhenAp
     auto event = std::unique_ptr<L0::Event>(Event::create<typename FamilyType::TimestampPacketType>(eventPool.get(), &eventDesc, device));
 
     ze_group_count_t groupCount{1, 1, 1};
-    auto result = commandList->appendLaunchKernelIndirect(kernel.toHandle(), groupCount, event->toHandle(), 0, nullptr, false);
+    auto result = commandList->appendLaunchKernelIndirect(kernel->toHandle(), groupCount, event->toHandle(), 0, nullptr, false);
     EXPECT_EQ(ZE_RESULT_SUCCESS, result);
 
-    ASSERT_EQ(nullptr, event->getKernelForPrintf());
+    ASSERT_EQ(nullptr, event->getKernelForPrintf().lock().get());
 }
 
 HWTEST2_F(CommandListAppendLaunchKernel, givenNonemptyAllocPrintfBufferKernelWhenAppendingLaunchKernelWithParamThenKernelIsStoredOnEvent, IsAtLeastSkl) {
     Mock<Module> module(this->device, nullptr);
-    Mock<::L0::KernelImp> kernel;
+    auto kernel = new Mock<::L0::KernelImp>{};
+    static_cast<ModuleImp *>(&module)->getPrintfKernelContainer().push_back(std::shared_ptr<Mock<::L0::KernelImp>>{kernel});
 
     ze_result_t returnValue;
     ze_event_pool_desc_t eventPoolDesc = {};
     eventPoolDesc.flags = ZE_EVENT_POOL_FLAG_HOST_VISIBLE;
     eventPoolDesc.count = 1;
 
-    kernel.module = &module;
+    kernel->module = &module;
-    kernel.descriptor.kernelAttributes.flags.usesPrintf = true;
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = true;
-    kernel.createPrintfBuffer();
+    kernel->createPrintfBuffer();
 
     ze_event_desc_t eventDesc = {};
     eventDesc.index = 0;
@@ -354,23 +357,24 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenNonemptyAllocPrintfBufferKernelWhe
     auto pCommandList = std::make_unique<WhiteBox<::L0::CommandListCoreFamily<gfxCoreFamily>>>();
     pCommandList->initialize(device, NEO::EngineGroupType::Compute, 0u);
 
-    auto result = pCommandList->appendLaunchKernelWithParams(&kernel, groupCount, event.get(), launchParams);
+    auto result = pCommandList->appendLaunchKernelWithParams(kernel, groupCount, event.get(), launchParams);
     EXPECT_EQ(ZE_RESULT_SUCCESS, result);
 
-    ASSERT_NE(nullptr, event->getKernelForPrintf());
+    ASSERT_FALSE(event->getKernelForPrintf().expired());
 }
 
 HWTEST2_F(CommandListAppendLaunchKernel, givenEmptyAllocPrintfBufferKernelWhenAppendingLaunchKernelWithParamThenKernelIsNotStoredOnEvent, IsAtLeastSkl) {
     Mock<Module> module(this->device, nullptr);
-    Mock<::L0::KernelImp> kernel;
+    auto kernel = new Mock<::L0::KernelImp>{};
+    static_cast<ModuleImp *>(&module)->getPrintfKernelContainer().push_back(std::shared_ptr<Mock<::L0::KernelImp>>{kernel});
 
     ze_result_t returnValue;
     ze_event_pool_desc_t eventPoolDesc = {};
     eventPoolDesc.flags = ZE_EVENT_POOL_FLAG_HOST_VISIBLE;
     eventPoolDesc.count = 1;
 
-    kernel.module = &module;
+    kernel->module = &module;
-    kernel.descriptor.kernelAttributes.flags.usesPrintf = false;
+    kernel->descriptor.kernelAttributes.flags.usesPrintf = false;
 
     ze_event_desc_t eventDesc = {};
     eventDesc.index = 0;
@@ -386,10 +390,10 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenEmptyAllocPrintfBufferKernelWhenAp
     auto pCommandList = std::make_unique<WhiteBox<::L0::CommandListCoreFamily<gfxCoreFamily>>>();
     pCommandList->initialize(device, NEO::EngineGroupType::Compute, 0u);
 
-    auto result = pCommandList->appendLaunchKernelWithParams(&kernel, groupCount, event.get(), launchParams);
+    auto result = pCommandList->appendLaunchKernelWithParams(kernel, groupCount, event.get(), launchParams);
     EXPECT_EQ(ZE_RESULT_SUCCESS, result);
 
-    ASSERT_EQ(nullptr, event->getKernelForPrintf());
+    ASSERT_EQ(nullptr, event->getKernelForPrintf().lock().get());
 }
 
 HWTEST2_F(CommandListAppendLaunchKernel, givenImmediateCommandListWhenAppendingLaunchKernelIndirectThenKernelIsExecutedOnImmediateCmdQ, IsAtLeastSkl) {

level_zero/core/test/unit_tests/sources/cmdqueue/test_cmdqueue_1.cpp

@@ -82,16 +82,17 @@ TEST_F(CommandQueueCreate, whenSynchronizeByPollingTaskCountThenCallsPrintOutput
                                                           false,
                                                           returnValue));
 
-    Mock<KernelImp> kernel1, kernel2;
+    std::shared_ptr<Mock<KernelImp>> kernel1{new Mock<KernelImp>{}};
+    std::shared_ptr<Mock<KernelImp>> kernel2{new Mock<KernelImp>{}};
 
-    commandQueue->printfKernelContainer.push_back(&kernel1);
+    commandQueue->printfKernelContainer.push_back(std::weak_ptr<Kernel>{kernel1});
-    commandQueue->printfKernelContainer.push_back(&kernel2);
+    commandQueue->printfKernelContainer.push_back(std::weak_ptr<Kernel>{kernel2});
 
     commandQueue->synchronizeByPollingForTaskCount(0u);
 
     EXPECT_EQ(0u, commandQueue->printfKernelContainer.size());
-    EXPECT_EQ(1u, kernel1.printPrintfOutputCalledTimes);
+    EXPECT_EQ(1u, kernel1->printPrintfOutputCalledTimes);
-    EXPECT_EQ(1u, kernel2.printPrintfOutputCalledTimes);
+    EXPECT_EQ(1u, kernel2->printPrintfOutputCalledTimes);
 
     commandQueue->destroy();
 }
@@ -108,20 +109,20 @@ HWTEST_F(CommandQueueCreate, givenPrintfKernelAndDetectedHangWhenSynchronizingBy
                                                           false,
                                                           returnValue));
 
-    Mock<KernelImp> kernel1;
+    std::shared_ptr<Mock<KernelImp>> kernel{new Mock<KernelImp>{}};
     TaskCountType currentTaskCount = 33u;
     auto &csr = neoDevice->getUltCommandStreamReceiver<FamilyType>();
     csr.callBaseWaitForCompletionWithTimeout = false;
     csr.latestWaitForCompletionWithTimeoutTaskCount = currentTaskCount;
     csr.returnWaitForCompletionWithTimeout = WaitStatus::GpuHang;
 
-    commandQueue->printfKernelContainer.push_back(&kernel1);
+    commandQueue->printfKernelContainer.push_back(std::weak_ptr<Kernel>{kernel});
 
     commandQueue->synchronizeByPollingForTaskCount(0u);
 
     EXPECT_EQ(0u, commandQueue->printfKernelContainer.size());
-    EXPECT_EQ(1u, kernel1.printPrintfOutputCalledTimes);
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
-    EXPECT_TRUE(kernel1.hangDetectedPassedToPrintfOutput);
+    EXPECT_TRUE(kernel->hangDetectedPassedToPrintfOutput);
 
     commandQueue->destroy();
 }
@@ -141,18 +142,18 @@ HWTEST_F(CommandQueueCreate, givenPrintfKernelAndDetectedHangWhenSynchronizingTh
                                                           false,
                                                           returnValue));
 
-    Mock<KernelImp> kernel1;
+    std::shared_ptr<Mock<KernelImp>> kernel{new Mock<KernelImp>{}};
     TaskCountType currentTaskCount = 33u;
     auto &csr = neoDevice->getUltCommandStreamReceiver<FamilyType>();
     csr.latestWaitForCompletionWithTimeoutTaskCount = currentTaskCount;
     csr.waitForTaskCountWithKmdNotifyFallbackReturnValue = WaitStatus::GpuHang;
 
-    commandQueue->printfKernelContainer.push_back(&kernel1);
+    commandQueue->printfKernelContainer.push_back(std::weak_ptr<Kernel>{kernel});
     commandQueue->synchronize(std::numeric_limits<uint64_t>::max());
 
     EXPECT_EQ(0u, commandQueue->printfKernelContainer.size());
-    EXPECT_EQ(1u, kernel1.printPrintfOutputCalledTimes);
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
-    EXPECT_TRUE(kernel1.hangDetectedPassedToPrintfOutput);
+    EXPECT_TRUE(kernel->hangDetectedPassedToPrintfOutput);
 
     commandQueue->destroy();
 }

level_zero/core/test/unit_tests/sources/event/test_event.cpp

@@ -3702,31 +3702,34 @@ TEST_F(EventTests, GivenResetAllPacketsFalseWhenResetPacketsThenKernelCountAndPa
 TEST_F(EventTests, givenCallToEventQueryStatusWithKernelPointerReturnsCounter) {
     auto event = std::make_unique<MockEventCompletion<uint32_t>>(eventPool.get(), 1u, device);
     Mock<Module> mockModule(this->device, nullptr);
-    Mock<KernelImp> mockKernel;
+    std::shared_ptr<Mock<KernelImp>> mockKernel{new Mock<KernelImp>{}};
-    mockKernel.descriptor.kernelAttributes.flags.usesPrintf = true;
+    mockKernel->descriptor.kernelAttributes.flags.usesPrintf = true;
-    mockKernel.module = &mockModule;
+    mockKernel->module = &mockModule;
 
-    event->setKernelForPrintf(&mockKernel);
+    event->setKernelForPrintf(std::weak_ptr<Kernel>{mockKernel});
-    EXPECT_NE(nullptr, event->getKernelForPrintf());
+    event->setKernelWithPrintfDeviceMutex(&static_cast<DeviceImp *>(this->device)->printfKernelMutex);
+    EXPECT_FALSE(event->getKernelForPrintf().expired());
+    EXPECT_NE(nullptr, event->getKernelWithPrintfDeviceMutex());
 
     constexpr uint64_t timeout = std::numeric_limits<std::uint64_t>::max();
     event->hostSynchronize(timeout);
-    EXPECT_EQ(1u, mockKernel.printPrintfOutputCalledTimes);
+    EXPECT_EQ(1u, mockKernel->printPrintfOutputCalledTimes);
 }
 
 TEST_F(EventTests, givenCallToEventQueryStatusWithNullKernelPointerReturnsCounter) {
     auto event = std::make_unique<MockEventCompletion<uint32_t>>(eventPool.get(), 1u, device);
     Mock<Module> mockModule(this->device, nullptr);
-    Mock<KernelImp> mockKernel;
+    std::shared_ptr<Mock<KernelImp>> mockKernel{new Mock<KernelImp>{}};
-    mockKernel.descriptor.kernelAttributes.flags.usesPrintf = true;
+    mockKernel->descriptor.kernelAttributes.flags.usesPrintf = true;
-    mockKernel.module = &mockModule;
+    mockKernel->module = &mockModule;
 
-    event->setKernelForPrintf(nullptr);
+    event->resetKernelForPrintf();
-    EXPECT_EQ(nullptr, event->getKernelForPrintf());
+    EXPECT_TRUE(event->getKernelForPrintf().expired());
 
     constexpr uint64_t timeout = std::numeric_limits<std::uint64_t>::max();
     event->hostSynchronize(timeout);
-    EXPECT_EQ(0u, mockKernel.printPrintfOutputCalledTimes);
+    EXPECT_EQ(0u, mockKernel->printPrintfOutputCalledTimes);
+    EXPECT_EQ(nullptr, event->getKernelWithPrintfDeviceMutex());
 }
 
 TEST_F(EventSynchronizeTest, whenEventSetCsrThenCorrectCsrSet) {

level_zero/core/test/unit_tests/sources/fence/test_fence.cpp

@@ -365,13 +365,13 @@ HWTEST_F(FenceTest, givenPrintfKernelWhenSynchronizingFenceThenPrintPrintfOutput
                                                           false,
                                                           false,
                                                           returnValue));
-    Mock<KernelImp> kernel;
+    std::shared_ptr<Mock<KernelImp>> kernel{new Mock<KernelImp>{}};
     TaskCountType currentTaskCount = 33u;
     auto &csr = neoDevice->getUltCommandStreamReceiver<FamilyType>();
     csr.returnWaitForCompletionWithTimeout = WaitStatus::Ready;
     csr.latestWaitForCompletionWithTimeoutTaskCount = currentTaskCount;
     *csr.tagAddress = currentTaskCount;
-    commandQueue->printfKernelContainer.push_back(&kernel);
+    commandQueue->printfKernelContainer.push_back(std::weak_ptr<Kernel>{kernel});
 
     ze_fence_desc_t fenceDesc = {ZE_STRUCTURE_TYPE_FENCE_DESC,
                                  nullptr,
@@ -383,8 +383,8 @@ HWTEST_F(FenceTest, givenPrintfKernelWhenSynchronizingFenceThenPrintPrintfOutput
     EXPECT_EQ(ZE_RESULT_SUCCESS, result);
 
     EXPECT_EQ(0u, commandQueue->printfKernelContainer.size());
-    EXPECT_EQ(1u, kernel.printPrintfOutputCalledTimes);
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
-    EXPECT_FALSE(kernel.hangDetectedPassedToPrintfOutput);
+    EXPECT_FALSE(kernel->hangDetectedPassedToPrintfOutput);
 
     delete fence;
 
@@ -408,14 +408,14 @@ HWTEST_F(FenceTest, givenPrintfKernelAndDetectedHangWhenSynchronizingFenceThenPr
                                                           false,
                                                           returnValue));
 
-    Mock<KernelImp> kernel;
+    std::shared_ptr<Mock<KernelImp>> kernel{new Mock<KernelImp>{}};
     TaskCountType currentTaskCount = 33u;
     auto &csr = neoDevice->getUltCommandStreamReceiver<FamilyType>();
     csr.latestWaitForCompletionWithTimeoutTaskCount = currentTaskCount;
     csr.returnWaitForCompletionWithTimeout = WaitStatus::GpuHang;
     *csr.tagAddress = 0;
     csr.gpuHangCheckPeriod = 0us;
-    commandQueue->printfKernelContainer.push_back(&kernel);
+    commandQueue->printfKernelContainer.push_back(std::weak_ptr<Kernel>{kernel});
 
     ze_fence_desc_t fenceDesc = {ZE_STRUCTURE_TYPE_FENCE_DESC,
                                  nullptr,
@@ -427,8 +427,8 @@ HWTEST_F(FenceTest, givenPrintfKernelAndDetectedHangWhenSynchronizingFenceThenPr
     EXPECT_EQ(ZE_RESULT_ERROR_DEVICE_LOST, result);
 
     EXPECT_EQ(0u, commandQueue->printfKernelContainer.size());
-    EXPECT_EQ(1u, kernel.printPrintfOutputCalledTimes);
+    EXPECT_EQ(1u, kernel->printPrintfOutputCalledTimes);
-    EXPECT_TRUE(kernel.hangDetectedPassedToPrintfOutput);
+    EXPECT_TRUE(kernel->hangDetectedPassedToPrintfOutput);
 
     delete fence;
     commandQueue->destroy();
@@ -445,14 +445,14 @@ HWTEST_F(FenceTest, givenPrintfKernelNotCompletedWhenSynchronizingFenceWithZeroT
                                                           false,
                                                           false,
                                                           returnValue));
-    Mock<KernelImp> kernel;
+    std::shared_ptr<Mock<KernelImp>> kernel{new Mock<KernelImp>{}};
     TaskCountType currentTaskCount = 33u;
     auto &csr = neoDevice->getUltCommandStreamReceiver<FamilyType>();
     csr.returnWaitForCompletionWithTimeout = WaitStatus::Ready;
     csr.latestWaitForCompletionWithTimeoutTaskCount = currentTaskCount;
     *csr.tagAddress = currentTaskCount - 1;
 
-    commandQueue->printfKernelContainer.push_back(&kernel);
+    commandQueue->printfKernelContainer.push_back(std::weak_ptr<Kernel>{kernel});
 
     ze_fence_desc_t fenceDesc = {ZE_STRUCTURE_TYPE_FENCE_DESC,
                                  nullptr,
@@ -464,7 +464,7 @@ HWTEST_F(FenceTest, givenPrintfKernelNotCompletedWhenSynchronizingFenceWithZeroT
     EXPECT_EQ(ZE_RESULT_NOT_READY, result);
 
     EXPECT_EQ(1u, commandQueue->printfKernelContainer.size());
-    EXPECT_EQ(0u, kernel.printPrintfOutputCalledTimes);
+    EXPECT_EQ(0u, kernel->printPrintfOutputCalledTimes);
 
     delete fence;
     commandQueue->destroy();

level_zero/core/test/unit_tests/sources/kernel/test_kernel.cpp

@@ -3127,6 +3127,40 @@ TEST_F(KernelPrintHandlerTest, whenPrintPrintfOutputIsCalledThenPrintfBufferIsUs
     EXPECT_EQ(buffer, MyPrintfHandler::getPrintfSurfaceInitialDataSize());
 }
 
+using PrintfKernelOwnershipTest = Test<ModuleFixture>;
+
+TEST_F(PrintfKernelOwnershipTest, givenKernelWithPrintfThenModuleStoresItCorrectly) {
+    ze_kernel_handle_t kernelHandle;
+    ze_kernel_desc_t kernelDesc{};
+    kernelDesc.pKernelName = kernelName.c_str();
+    EXPECT_EQ(ZE_RESULT_SUCCESS, module->createKernel(&kernelDesc, &kernelHandle));
+
+    auto kernel = static_cast<Mock<KernelImp> *>(Kernel::fromHandle(kernelHandle));
+    EXPECT_NE(nullptr, kernel->getPrintfBufferAllocation());
+    EXPECT_NE(nullptr, kernel->getDevicePrintfKernelMutex());
+    EXPECT_EQ(1u, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().size());
+    EXPECT_EQ(kernel, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer()[0].get());
+}
+
+TEST_F(PrintfKernelOwnershipTest, givenKernelWithPrintfDestroyedFirstThenModuleDestructionReturnsTheCorrectStatus) {
+    ze_kernel_handle_t kernelHandle;
+    ze_kernel_desc_t kernelDesc{};
+    kernelDesc.pKernelName = kernelName.c_str();
+    EXPECT_EQ(ZE_RESULT_SUCCESS, module->createKernel(&kernelDesc, &kernelHandle));
+
+    auto kernel = static_cast<Mock<KernelImp> *>(Kernel::fromHandle(kernelHandle));
+    EXPECT_NE(nullptr, kernel->getPrintfBufferAllocation());
+    EXPECT_NE(nullptr, kernel->getDevicePrintfKernelMutex());
+    EXPECT_EQ(1u, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().size());
+    EXPECT_EQ(kernel, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer()[0].get());
+
+    EXPECT_EQ(ZE_RESULT_SUCCESS, kernel->destroy());
+
+    EXPECT_EQ(1u, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer().size());
+    EXPECT_EQ(nullptr, static_cast<ModuleImp *>(module.get())->getPrintfKernelContainer()[0].get());
+    EXPECT_EQ(ZE_RESULT_ERROR_INVALID_ARGUMENT, static_cast<ModuleImp *>(module.get())->destroyPrintfKernel(kernelHandle));
+}
+
 using PrintfTest = Test<DeviceFixture>;
 
 TEST_F(PrintfTest, givenKernelWithPrintfThenPrintfBufferIsCreated) {