Add lifetime handle validation to zeInitDrivers (#470)

* Previous mako generation had missed epilogue hooking
  for zeInitDrivers, added it into the Mako
* Added Unit tests to verify

Signed-off-by: Russell McGuire <russell.w.mcguire@intel.com>

Author: rwmcguir

scripts/templates/validation/valddi.cpp.mako

@@ -197,7 +197,7 @@ return ${failure_return};
 
         <% 
         func_name = th.make_func_name(n, tags, obj)
-        generate_post_call = re.match(r"\w+Create\w*$|\w+Get$|\w+Get\w*Exp$|\w+GetIpcHandle$|\w+GetSubDevices$", func_name)
+        generate_post_call = re.match(r"\w+Create\w*$|\w+Get$|\w+Get\w*Exp$|\w+GetIpcHandle$|\w+GetSubDevices$|\w+InitDrivers$", func_name)
         %>
         if(context.enableHandleLifetime ){
             auto result = context.handleLifetime->${n}HandleLifetime.${th.make_func_name(n, tags, obj)}Prologue( \

source/layers/validation/ze_valddi.cpp

@@ -7748,6 +7748,15 @@ namespace validation_layer
             if(result!=ZE_RESULT_SUCCESS) return logAndPropagateResult_zeInitDrivers(result, pCount, phDrivers, desc);
         }
 
+
+        if( driver_result == ZE_RESULT_SUCCESS && context.enableHandleLifetime ){
+            for (size_t i = 0; ( nullptr != phDrivers) && (i < *pCount); ++i){
+                if (phDrivers[i]){
+                    context.handleLifetime->addHandle( phDrivers[i] );
+                    context.handleLifetime->addDependent( pCount, phDrivers[i] );
+                }
+            }
+        }
         return logAndPropagateResult_zeInitDrivers(driver_result, pCount, phDrivers, desc);
     }
 

test/CMakeLists.txt

@@ -1130,6 +1130,17 @@ set_property(TEST test_zer_validation_layer_positive_case PROPERTY ENVIRONMENT "
 add_test(NAME test_zer_parameter_validation_layer_negative_case COMMAND tests --gtest_filter=*GivenLevelZeroLoaderPresentWhenCallingZerApiWithParameterValidationEnabledThenExpectValidationsAreTriggered)
 set_property(TEST test_zer_parameter_validation_layer_negative_case PROPERTY ENVIRONMENT "ZE_ENABLE_LOADER_DEBUG_TRACE=1;ZE_ENABLE_NULL_DRIVER=1;ZE_ENABLE_VALIDATION_LAYER=1;ZE_ENABLE_PARAMETER_VALIDATION=1;ZE_ENABLE_HANDLE_LIFETIME=1;ZEL_ENABLE_EVENTS_CHECKER=1;ZEL_ENABLE_BASIC_LEAK_CHECKER=1;ZEL_ENABLE_CERTIFICATION_CHECKER=1")
 
+# Handle Lifetime Validation Tests
+# Regression: zeInitDrivers must register driver handles so downstream calls succeed.
+add_test(NAME test_handle_lifetime_zeInitDrivers_path COMMAND tests --gtest_filter=*HandleLifetimeValidation*GivenHandleLifetimeEnabledWhenCallingzeDeviceGetWithDriverHandleFromzeInitDriversThenSucceeds*)
+set_property(TEST test_handle_lifetime_zeInitDrivers_path PROPERTY ENVIRONMENT "ZE_ENABLE_LOADER_DEBUG_TRACE=1;ZE_ENABLE_NULL_DRIVER=1;ZEL_TEST_NULL_DRIVER_TYPE=GPU;ZE_ENABLE_VALIDATION_LAYER=1;ZE_ENABLE_HANDLE_LIFETIME=1")
+
+add_test(NAME test_handle_lifetime_zeDriverGet_path COMMAND tests --gtest_filter=*HandleLifetimeValidation*GivenHandleLifetimeEnabledWhenCallingzeDeviceGetWithDriverHandleFromzeDriverGetThenSucceeds*)
+set_property(TEST test_handle_lifetime_zeDriverGet_path PROPERTY ENVIRONMENT "ZE_ENABLE_LOADER_DEBUG_TRACE=1;ZE_ENABLE_NULL_DRIVER=1;ZEL_TEST_NULL_DRIVER_TYPE=GPU;ZE_ENABLE_VALIDATION_LAYER=1;ZE_ENABLE_HANDLE_LIFETIME=1")
+
+add_test(NAME test_handle_lifetime_null_handle_rejected COMMAND tests --gtest_filter=*HandleLifetimeValidation*GivenHandleLifetimeEnabledWhenCallingzeDeviceGetWithNullDriverHandleThenReturnsInvalidHandle*)
+set_property(TEST test_handle_lifetime_null_handle_rejected PROPERTY ENVIRONMENT "ZE_ENABLE_LOADER_DEBUG_TRACE=1;ZE_ENABLE_NULL_DRIVER=1;ZEL_TEST_NULL_DRIVER_TYPE=GPU;ZE_ENABLE_VALIDATION_LAYER=1;ZE_ENABLE_HANDLE_LIFETIME=1")
+
 # API Call Tracing Tests - Validation Layer with Trace Logging
 add_test(NAME test_validation_layer_api_tracing_basic COMMAND tests --gtest_filter=ValidationLayerApiTracing.GivenValidationLayerEnabledWithTraceLevelLoggingWhenCallingBasicApisThenTracingDoesNotCrash)
 set_property(TEST test_validation_layer_api_tracing_basic PROPERTY ENVIRONMENT "ZE_ENABLE_VALIDATION_LAYER=1;ZEL_ENABLE_LOADER_LOGGING=1;ZEL_LOADER_LOGGING_LEVEL=trace;ZE_ENABLE_NULL_DRIVER=1")

test/loader_validation_layer.cpp

@@ -970,3 +970,89 @@ TEST(
     EXPECT_EQ(ZE_RESULT_SUCCESS, zeCommandQueueDestroy(commandQueue));
     EXPECT_EQ(ZE_RESULT_SUCCESS, zeContextDestroy(context));
 }
+
+///////////////////////////////////////////////////////////////////////////////
+// Handle Lifetime Tests
+//
+// These tests validate the ZE_ENABLE_HANDLE_LIFETIME feature of the validation
+// layer. Handle lifetime tracking registers handles when they are created and
+// validates them on every subsequent API call, returning
+// ZE_RESULT_ERROR_INVALID_NULL_HANDLE for any unregistered handle.
+//
+// The tests are run as separate CTest invocations with the environment:
+//   ZE_ENABLE_VALIDATION_LAYER=1
+//   ZE_ENABLE_HANDLE_LIFETIME=1
+//   ZE_ENABLE_NULL_DRIVER=1
+
+// Regression test: driver handles obtained via zeInitDrivers must be registered
+// in the handle lifetime tracker so that subsequent API calls succeed.
+// Before the fix, zeInitDrivers had no handle-registration epilogue (unlike
+// zeDriverGet), causing zeDeviceGet to return ZE_RESULT_ERROR_INVALID_NULL_HANDLE.
+TEST(
+    HandleLifetimeValidation,
+    GivenHandleLifetimeEnabledWhenCallingzeDeviceGetWithDriverHandleFromzeInitDriversThenSucceeds) {
+
+    ze_init_driver_type_desc_t driverTypeDesc = {};
+    driverTypeDesc.stype = ZE_STRUCTURE_TYPE_INIT_DRIVER_TYPE_DESC;
+    driverTypeDesc.pNext = nullptr;
+    driverTypeDesc.flags = UINT32_MAX;
+
+    uint32_t count = 0;
+    EXPECT_EQ(ZE_RESULT_SUCCESS, zeInitDrivers(&count, nullptr, &driverTypeDesc));
+    EXPECT_GT(count, 0);
+
+    std::vector<ze_driver_handle_t> drivers(count);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, zeInitDrivers(&count, drivers.data(), &driverTypeDesc));
+
+    // Regression: driver handles from zeInitDrivers must be registered with the
+    // handle lifetime tracker so that zeDeviceGet does not fail with
+    // ZE_RESULT_ERROR_INVALID_NULL_HANDLE.
+    uint32_t deviceCount = 0;
+    EXPECT_EQ(ZE_RESULT_SUCCESS, zeDeviceGet(drivers[0], &deviceCount, nullptr));
+    EXPECT_GT(deviceCount, 0);
+
+    std::vector<ze_device_handle_t> devices(deviceCount);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, zeDeviceGet(drivers[0], &deviceCount, devices.data()));
+}
+
+// Verify that driver handles obtained via zeDriverGet are registered correctly
+// and can be used with zeDeviceGet when handle lifetime tracking is enabled.
+TEST(
+    HandleLifetimeValidation,
+    GivenHandleLifetimeEnabledWhenCallingzeDeviceGetWithDriverHandleFromzeDriverGetThenSucceeds) {
+
+    EXPECT_EQ(ZE_RESULT_SUCCESS, zeInit(ZE_INIT_FLAG_GPU_ONLY));
+
+    uint32_t driverCount = 0;
+    EXPECT_EQ(ZE_RESULT_SUCCESS, zeDriverGet(&driverCount, nullptr));
+    EXPECT_GT(driverCount, 0);
+
+    std::vector<ze_driver_handle_t> drivers(driverCount);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, zeDriverGet(&driverCount, drivers.data()));
+
+    uint32_t deviceCount = 0;
+    EXPECT_EQ(ZE_RESULT_SUCCESS, zeDeviceGet(drivers[0], &deviceCount, nullptr));
+    EXPECT_GT(deviceCount, 0);
+
+    std::vector<ze_device_handle_t> devices(deviceCount);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, zeDeviceGet(drivers[0], &deviceCount, devices.data()));
+}
+
+// Verify that calling an API with a null (unregistered) handle returns
+// ZE_RESULT_ERROR_INVALID_NULL_HANDLE when handle lifetime tracking is enabled.
+TEST(
+    HandleLifetimeValidation,
+    GivenHandleLifetimeEnabledWhenCallingzeDeviceGetWithNullDriverHandleThenReturnsInvalidHandle) {
+
+    ze_init_driver_type_desc_t driverTypeDesc = {};
+    driverTypeDesc.stype = ZE_STRUCTURE_TYPE_INIT_DRIVER_TYPE_DESC;
+    driverTypeDesc.pNext = nullptr;
+    driverTypeDesc.flags = UINT32_MAX;
+
+    uint32_t count = 0;
+    EXPECT_EQ(ZE_RESULT_SUCCESS, zeInitDrivers(&count, nullptr, &driverTypeDesc));
+
+    // A null handle is never registered; the handle lifetime checker must reject it.
+    uint32_t deviceCount = 0;
+    EXPECT_EQ(ZE_RESULT_ERROR_INVALID_NULL_HANDLE, zeDeviceGet(nullptr, &deviceCount, nullptr));
+}