1 /*
2 * Copyright (C) 2019 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 #define LOG_TAG "neuralnetworks_hidl_hal_test"
18
19 #include <android-base/logging.h>
20 #include <fcntl.h>
21 #include <ftw.h>
22 #include <gtest/gtest.h>
23 #include <hidlmemory/mapping.h>
24 #include <unistd.h>
25
26 #include <cstdio>
27 #include <cstdlib>
28 #include <random>
29 #include <thread>
30
31 #include "1.2/Callbacks.h"
32 #include "GeneratedTestHarness.h"
33 #include "MemoryUtils.h"
34 #include "TestHarness.h"
35 #include "VtsHalNeuralnetworks.h"
36
37 // Forward declaration of the mobilenet generated test models in
38 // frameworks/ml/nn/runtime/test/generated/.
39 namespace generated_tests::mobilenet_224_gender_basic_fixed {
40 const test_helper::TestModel& get_test_model();
41 } // namespace generated_tests::mobilenet_224_gender_basic_fixed
42
43 namespace generated_tests::mobilenet_quantized {
44 const test_helper::TestModel& get_test_model();
45 } // namespace generated_tests::mobilenet_quantized
46
47 namespace android::hardware::neuralnetworks::V1_2::vts::functional {
48
49 using namespace test_helper;
50 using implementation::PreparedModelCallback;
51 using V1_0::ErrorStatus;
52 using V1_1::ExecutionPreference;
53
54 namespace float32_model {
55
56 constexpr auto get_test_model = generated_tests::mobilenet_224_gender_basic_fixed::get_test_model;
57
58 } // namespace float32_model
59
60 namespace quant8_model {
61
62 constexpr auto get_test_model = generated_tests::mobilenet_quantized::get_test_model;
63
64 } // namespace quant8_model
65
66 namespace {
67
68 enum class AccessMode { READ_WRITE, READ_ONLY, WRITE_ONLY };
69
70 // Creates cache handles based on provided file groups.
71 // The outer vector corresponds to handles and the inner vector is for fds held by each handle.
createCacheHandles(const std::vector<std::vector<std::string>> & fileGroups,const std::vector<AccessMode> & mode,hidl_vec<hidl_handle> * handles)72 void createCacheHandles(const std::vector<std::vector<std::string>>& fileGroups,
73 const std::vector<AccessMode>& mode, hidl_vec<hidl_handle>* handles) {
74 handles->resize(fileGroups.size());
75 for (uint32_t i = 0; i < fileGroups.size(); i++) {
76 std::vector<int> fds;
77 for (const auto& file : fileGroups[i]) {
78 int fd;
79 if (mode[i] == AccessMode::READ_ONLY) {
80 fd = open(file.c_str(), O_RDONLY);
81 } else if (mode[i] == AccessMode::WRITE_ONLY) {
82 fd = open(file.c_str(), O_WRONLY | O_CREAT, S_IRUSR | S_IWUSR);
83 } else if (mode[i] == AccessMode::READ_WRITE) {
84 fd = open(file.c_str(), O_RDWR | O_CREAT, S_IRUSR | S_IWUSR);
85 } else {
86 FAIL();
87 }
88 ASSERT_GE(fd, 0);
89 fds.push_back(fd);
90 }
91 native_handle_t* cacheNativeHandle = native_handle_create(fds.size(), 0);
92 ASSERT_NE(cacheNativeHandle, nullptr);
93 std::copy(fds.begin(), fds.end(), &cacheNativeHandle->data[0]);
94 (*handles)[i].setTo(cacheNativeHandle, /*shouldOwn=*/true);
95 }
96 }
97
createCacheHandles(const std::vector<std::vector<std::string>> & fileGroups,AccessMode mode,hidl_vec<hidl_handle> * handles)98 void createCacheHandles(const std::vector<std::vector<std::string>>& fileGroups, AccessMode mode,
99 hidl_vec<hidl_handle>* handles) {
100 createCacheHandles(fileGroups, std::vector<AccessMode>(fileGroups.size(), mode), handles);
101 }
102
103 // Create a chain of broadcast operations. The second operand is always constant tensor [1].
104 // For simplicity, activation scalar is shared. The second operand is not shared
105 // in the model to let driver maintain a non-trivial size of constant data and the corresponding
106 // data locations in cache.
107 //
108 // --------- activation --------
109 // ↓ ↓ ↓ ↓
110 // E.g. input -> ADD -> ADD -> ADD -> ... -> ADD -> output
111 // ↑ ↑ ↑ ↑
112 // [1] [1] [1] [1]
113 //
114 // This function assumes the operation is either ADD or MUL.
115 template <typename CppType, TestOperandType operandType>
createLargeTestModelImpl(TestOperationType op,uint32_t len)116 TestModel createLargeTestModelImpl(TestOperationType op, uint32_t len) {
117 EXPECT_TRUE(op == TestOperationType::ADD || op == TestOperationType::MUL);
118
119 // Model operations and operands.
120 std::vector<TestOperation> operations(len);
121 std::vector<TestOperand> operands(len * 2 + 2);
122
123 // The activation scalar, value = 0.
124 operands[0] = {
125 .type = TestOperandType::INT32,
126 .dimensions = {},
127 .numberOfConsumers = len,
128 .scale = 0.0f,
129 .zeroPoint = 0,
130 .lifetime = TestOperandLifeTime::CONSTANT_COPY,
131 .data = TestBuffer::createFromVector<int32_t>({0}),
132 };
133
134 // The buffer value of the constant second operand. The logical value is always 1.0f.
135 CppType bufferValue;
136 // The scale of the first and second operand.
137 float scale1, scale2;
138 if (operandType == TestOperandType::TENSOR_FLOAT32) {
139 bufferValue = 1.0f;
140 scale1 = 0.0f;
141 scale2 = 0.0f;
142 } else if (op == TestOperationType::ADD) {
143 bufferValue = 1;
144 scale1 = 1.0f;
145 scale2 = 1.0f;
146 } else {
147 // To satisfy the constraint on quant8 MUL: input0.scale * input1.scale < output.scale,
148 // set input1 to have scale = 0.5f and bufferValue = 2, i.e. 1.0f in floating point.
149 bufferValue = 2;
150 scale1 = 1.0f;
151 scale2 = 0.5f;
152 }
153
154 for (uint32_t i = 0; i < len; i++) {
155 const uint32_t firstInputIndex = i * 2 + 1;
156 const uint32_t secondInputIndex = firstInputIndex + 1;
157 const uint32_t outputIndex = secondInputIndex + 1;
158
159 // The first operation input.
160 operands[firstInputIndex] = {
161 .type = operandType,
162 .dimensions = {1},
163 .numberOfConsumers = 1,
164 .scale = scale1,
165 .zeroPoint = 0,
166 .lifetime = (i == 0 ? TestOperandLifeTime::MODEL_INPUT
167 : TestOperandLifeTime::TEMPORARY_VARIABLE),
168 .data = (i == 0 ? TestBuffer::createFromVector<CppType>({1}) : TestBuffer()),
169 };
170
171 // The second operation input, value = 1.
172 operands[secondInputIndex] = {
173 .type = operandType,
174 .dimensions = {1},
175 .numberOfConsumers = 1,
176 .scale = scale2,
177 .zeroPoint = 0,
178 .lifetime = TestOperandLifeTime::CONSTANT_COPY,
179 .data = TestBuffer::createFromVector<CppType>({bufferValue}),
180 };
181
182 // The operation. All operations share the same activation scalar.
183 // The output operand is created as an input in the next iteration of the loop, in the case
184 // of all but the last member of the chain; and after the loop as a model output, in the
185 // case of the last member of the chain.
186 operations[i] = {
187 .type = op,
188 .inputs = {firstInputIndex, secondInputIndex, /*activation scalar*/ 0},
189 .outputs = {outputIndex},
190 };
191 }
192
193 // For TestOperationType::ADD, output = 1 + 1 * len = len + 1
194 // For TestOperationType::MUL, output = 1 * 1 ^ len = 1
195 CppType outputResult = static_cast<CppType>(op == TestOperationType::ADD ? len + 1u : 1u);
196
197 // The model output.
198 operands.back() = {
199 .type = operandType,
200 .dimensions = {1},
201 .numberOfConsumers = 0,
202 .scale = scale1,
203 .zeroPoint = 0,
204 .lifetime = TestOperandLifeTime::MODEL_OUTPUT,
205 .data = TestBuffer::createFromVector<CppType>({outputResult}),
206 };
207
208 return {
209 .main = {.operands = std::move(operands),
210 .operations = std::move(operations),
211 .inputIndexes = {1},
212 .outputIndexes = {len * 2 + 1}},
213 .isRelaxed = false,
214 };
215 }
216
217 } // namespace
218
219 // Tag for the compilation caching tests.
220 class CompilationCachingTestBase : public testing::Test {
221 protected:
CompilationCachingTestBase(sp<IDevice> device,OperandType type)222 CompilationCachingTestBase(sp<IDevice> device, OperandType type)
223 : kDevice(std::move(device)), kOperandType(type) {}
224
SetUp()225 void SetUp() override {
226 testing::Test::SetUp();
227 ASSERT_NE(kDevice.get(), nullptr);
228 const bool deviceIsResponsive = kDevice->ping().isOk();
229 ASSERT_TRUE(deviceIsResponsive);
230
231 // Create cache directory. The cache directory and a temporary cache file is always created
232 // to test the behavior of prepareModelFromCache, even when caching is not supported.
233 char cacheDirTemp[] = "/data/local/tmp/TestCompilationCachingXXXXXX";
234 char* cacheDir = mkdtemp(cacheDirTemp);
235 ASSERT_NE(cacheDir, nullptr);
236 mCacheDir = cacheDir;
237 mCacheDir.push_back('/');
238
239 Return<void> ret = kDevice->getNumberOfCacheFilesNeeded(
240 [this](ErrorStatus status, uint32_t numModelCache, uint32_t numDataCache) {
241 EXPECT_EQ(ErrorStatus::NONE, status);
242 mNumModelCache = numModelCache;
243 mNumDataCache = numDataCache;
244 });
245 EXPECT_TRUE(ret.isOk());
246 mIsCachingSupported = mNumModelCache > 0 || mNumDataCache > 0;
247
248 // Create empty cache files.
249 mTmpCache = mCacheDir + "tmp";
250 for (uint32_t i = 0; i < mNumModelCache; i++) {
251 mModelCache.push_back({mCacheDir + "model" + std::to_string(i)});
252 }
253 for (uint32_t i = 0; i < mNumDataCache; i++) {
254 mDataCache.push_back({mCacheDir + "data" + std::to_string(i)});
255 }
256 // Sample handles, use AccessMode::WRITE_ONLY for createCacheHandles to create files.
257 hidl_vec<hidl_handle> modelHandle, dataHandle, tmpHandle;
258 createCacheHandles(mModelCache, AccessMode::WRITE_ONLY, &modelHandle);
259 createCacheHandles(mDataCache, AccessMode::WRITE_ONLY, &dataHandle);
260 createCacheHandles({{mTmpCache}}, AccessMode::WRITE_ONLY, &tmpHandle);
261
262 if (!mIsCachingSupported) {
263 LOG(INFO) << "NN VTS: Early termination of test because vendor service does not "
264 "support compilation caching.";
265 std::cout << "[ ] Early termination of test because vendor service does not "
266 "support compilation caching."
267 << std::endl;
268 }
269 }
270
TearDown()271 void TearDown() override {
272 // If the test passes, remove the tmp directory. Otherwise, keep it for debugging purposes.
273 if (!testing::Test::HasFailure()) {
274 // Recursively remove the cache directory specified by mCacheDir.
275 auto callback = [](const char* entry, const struct stat*, int, struct FTW*) {
276 return remove(entry);
277 };
278 nftw(mCacheDir.c_str(), callback, 128, FTW_DEPTH | FTW_MOUNT | FTW_PHYS);
279 }
280 testing::Test::TearDown();
281 }
282
283 // Model and examples creators. According to kOperandType, the following methods will return
284 // either float32 model/examples or the quant8 variant.
createTestModel()285 TestModel createTestModel() {
286 if (kOperandType == OperandType::TENSOR_FLOAT32) {
287 return float32_model::get_test_model();
288 } else {
289 return quant8_model::get_test_model();
290 }
291 }
292
createLargeTestModel(OperationType op,uint32_t len)293 TestModel createLargeTestModel(OperationType op, uint32_t len) {
294 if (kOperandType == OperandType::TENSOR_FLOAT32) {
295 return createLargeTestModelImpl<float, TestOperandType::TENSOR_FLOAT32>(
296 static_cast<TestOperationType>(op), len);
297 } else {
298 return createLargeTestModelImpl<uint8_t, TestOperandType::TENSOR_QUANT8_ASYMM>(
299 static_cast<TestOperationType>(op), len);
300 }
301 }
302
303 // See if the service can handle the model.
isModelFullySupported(const Model & model)304 bool isModelFullySupported(const Model& model) {
305 bool fullySupportsModel = false;
306 Return<void> supportedCall = kDevice->getSupportedOperations_1_2(
307 model,
308 [&fullySupportsModel, &model](ErrorStatus status, const hidl_vec<bool>& supported) {
309 ASSERT_EQ(ErrorStatus::NONE, status);
310 ASSERT_EQ(supported.size(), model.operations.size());
311 fullySupportsModel = std::all_of(supported.begin(), supported.end(),
312 [](bool valid) { return valid; });
313 });
314 EXPECT_TRUE(supportedCall.isOk());
315 return fullySupportsModel;
316 }
317
saveModelToCache(const Model & model,const hidl_vec<hidl_handle> & modelCache,const hidl_vec<hidl_handle> & dataCache,sp<IPreparedModel> * preparedModel=nullptr)318 void saveModelToCache(const Model& model, const hidl_vec<hidl_handle>& modelCache,
319 const hidl_vec<hidl_handle>& dataCache,
320 sp<IPreparedModel>* preparedModel = nullptr) {
321 if (preparedModel != nullptr) *preparedModel = nullptr;
322
323 // Launch prepare model.
324 sp<PreparedModelCallback> preparedModelCallback = new PreparedModelCallback();
325 hidl_array<uint8_t, sizeof(mToken)> cacheToken(mToken);
326 Return<ErrorStatus> prepareLaunchStatus =
327 kDevice->prepareModel_1_2(model, ExecutionPreference::FAST_SINGLE_ANSWER,
328 modelCache, dataCache, cacheToken, preparedModelCallback);
329 ASSERT_TRUE(prepareLaunchStatus.isOk());
330 ASSERT_EQ(static_cast<ErrorStatus>(prepareLaunchStatus), ErrorStatus::NONE);
331
332 // Retrieve prepared model.
333 preparedModelCallback->wait();
334 ASSERT_EQ(preparedModelCallback->getStatus(), ErrorStatus::NONE);
335 if (preparedModel != nullptr) {
336 *preparedModel = IPreparedModel::castFrom(preparedModelCallback->getPreparedModel())
337 .withDefault(nullptr);
338 }
339 }
340
checkEarlyTermination(ErrorStatus status)341 bool checkEarlyTermination(ErrorStatus status) {
342 if (status == ErrorStatus::GENERAL_FAILURE) {
343 LOG(INFO) << "NN VTS: Early termination of test because vendor service cannot "
344 "save the prepared model that it does not support.";
345 std::cout << "[ ] Early termination of test because vendor service cannot "
346 "save the prepared model that it does not support."
347 << std::endl;
348 return true;
349 }
350 return false;
351 }
352
checkEarlyTermination(const Model & model)353 bool checkEarlyTermination(const Model& model) {
354 if (!isModelFullySupported(model)) {
355 LOG(INFO) << "NN VTS: Early termination of test because vendor service cannot "
356 "prepare model that it does not support.";
357 std::cout << "[ ] Early termination of test because vendor service cannot "
358 "prepare model that it does not support."
359 << std::endl;
360 return true;
361 }
362 return false;
363 }
364
prepareModelFromCache(const hidl_vec<hidl_handle> & modelCache,const hidl_vec<hidl_handle> & dataCache,sp<IPreparedModel> * preparedModel,ErrorStatus * status)365 void prepareModelFromCache(const hidl_vec<hidl_handle>& modelCache,
366 const hidl_vec<hidl_handle>& dataCache,
367 sp<IPreparedModel>* preparedModel, ErrorStatus* status) {
368 // Launch prepare model from cache.
369 sp<PreparedModelCallback> preparedModelCallback = new PreparedModelCallback();
370 hidl_array<uint8_t, sizeof(mToken)> cacheToken(mToken);
371 Return<ErrorStatus> prepareLaunchStatus = kDevice->prepareModelFromCache(
372 modelCache, dataCache, cacheToken, preparedModelCallback);
373 ASSERT_TRUE(prepareLaunchStatus.isOk());
374 if (static_cast<ErrorStatus>(prepareLaunchStatus) != ErrorStatus::NONE) {
375 *preparedModel = nullptr;
376 *status = static_cast<ErrorStatus>(prepareLaunchStatus);
377 return;
378 }
379
380 // Retrieve prepared model.
381 preparedModelCallback->wait();
382 *status = preparedModelCallback->getStatus();
383 *preparedModel = IPreparedModel::castFrom(preparedModelCallback->getPreparedModel())
384 .withDefault(nullptr);
385 }
386
387 // Absolute path to the temporary cache directory.
388 std::string mCacheDir;
389
390 // Groups of file paths for model and data cache in the tmp cache directory, initialized with
391 // outer_size = mNum{Model|Data}Cache, inner_size = 1. The outer vector corresponds to handles
392 // and the inner vector is for fds held by each handle.
393 std::vector<std::vector<std::string>> mModelCache;
394 std::vector<std::vector<std::string>> mDataCache;
395
396 // A separate temporary file path in the tmp cache directory.
397 std::string mTmpCache;
398
399 uint8_t mToken[static_cast<uint32_t>(Constant::BYTE_SIZE_OF_CACHE_TOKEN)] = {};
400 uint32_t mNumModelCache;
401 uint32_t mNumDataCache;
402 uint32_t mIsCachingSupported;
403
404 const sp<IDevice> kDevice;
405 // The primary data type of the testModel.
406 const OperandType kOperandType;
407 };
408
409 using CompilationCachingTestParam = std::tuple<NamedDevice, OperandType>;
410
411 // A parameterized fixture of CompilationCachingTestBase. Every test will run twice, with the first
412 // pass running with float32 models and the second pass running with quant8 models.
413 class CompilationCachingTest : public CompilationCachingTestBase,
414 public testing::WithParamInterface<CompilationCachingTestParam> {
415 protected:
CompilationCachingTest()416 CompilationCachingTest()
417 : CompilationCachingTestBase(getData(std::get<NamedDevice>(GetParam())),
418 std::get<OperandType>(GetParam())) {}
419 };
420
TEST_P(CompilationCachingTest,CacheSavingAndRetrieval)421 TEST_P(CompilationCachingTest, CacheSavingAndRetrieval) {
422 // Create test HIDL model and compile.
423 const TestModel& testModel = createTestModel();
424 const Model model = createModel(testModel);
425 if (checkEarlyTermination(model)) return;
426 sp<IPreparedModel> preparedModel = nullptr;
427
428 // Save the compilation to cache.
429 {
430 hidl_vec<hidl_handle> modelCache, dataCache;
431 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
432 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
433 saveModelToCache(model, modelCache, dataCache);
434 }
435
436 // Retrieve preparedModel from cache.
437 {
438 preparedModel = nullptr;
439 ErrorStatus status;
440 hidl_vec<hidl_handle> modelCache, dataCache;
441 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
442 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
443 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
444 if (!mIsCachingSupported) {
445 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
446 ASSERT_EQ(preparedModel, nullptr);
447 return;
448 } else if (checkEarlyTermination(status)) {
449 ASSERT_EQ(preparedModel, nullptr);
450 return;
451 } else {
452 ASSERT_EQ(status, ErrorStatus::NONE);
453 ASSERT_NE(preparedModel, nullptr);
454 }
455 }
456
457 // Execute and verify results.
458 EvaluatePreparedModel(preparedModel, testModel,
459 /*testDynamicOutputShape=*/false);
460 }
461
TEST_P(CompilationCachingTest,CacheSavingAndRetrievalNonZeroOffset)462 TEST_P(CompilationCachingTest, CacheSavingAndRetrievalNonZeroOffset) {
463 // Create test HIDL model and compile.
464 const TestModel& testModel = createTestModel();
465 const Model model = createModel(testModel);
466 if (checkEarlyTermination(model)) return;
467 sp<IPreparedModel> preparedModel = nullptr;
468
469 // Save the compilation to cache.
470 {
471 hidl_vec<hidl_handle> modelCache, dataCache;
472 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
473 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
474 uint8_t sampleBytes[] = {0, 0};
475 // Write a sample integer to the cache.
476 // The driver should be able to handle non-empty cache and non-zero fd offset.
477 for (uint32_t i = 0; i < modelCache.size(); i++) {
478 ASSERT_EQ(write(modelCache[i].getNativeHandle()->data[0], &sampleBytes,
479 sizeof(sampleBytes)),
480 sizeof(sampleBytes));
481 }
482 for (uint32_t i = 0; i < dataCache.size(); i++) {
483 ASSERT_EQ(
484 write(dataCache[i].getNativeHandle()->data[0], &sampleBytes, sizeof(sampleBytes)),
485 sizeof(sampleBytes));
486 }
487 saveModelToCache(model, modelCache, dataCache);
488 }
489
490 // Retrieve preparedModel from cache.
491 {
492 preparedModel = nullptr;
493 ErrorStatus status;
494 hidl_vec<hidl_handle> modelCache, dataCache;
495 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
496 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
497 uint8_t sampleByte = 0;
498 // Advance the offset of each handle by one byte.
499 // The driver should be able to handle non-zero fd offset.
500 for (uint32_t i = 0; i < modelCache.size(); i++) {
501 ASSERT_GE(read(modelCache[i].getNativeHandle()->data[0], &sampleByte, 1), 0);
502 }
503 for (uint32_t i = 0; i < dataCache.size(); i++) {
504 ASSERT_GE(read(dataCache[i].getNativeHandle()->data[0], &sampleByte, 1), 0);
505 }
506 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
507 if (!mIsCachingSupported) {
508 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
509 ASSERT_EQ(preparedModel, nullptr);
510 return;
511 } else if (checkEarlyTermination(status)) {
512 ASSERT_EQ(preparedModel, nullptr);
513 return;
514 } else {
515 ASSERT_EQ(status, ErrorStatus::NONE);
516 ASSERT_NE(preparedModel, nullptr);
517 }
518 }
519
520 // Execute and verify results.
521 EvaluatePreparedModel(preparedModel, testModel,
522 /*testDynamicOutputShape=*/false);
523 }
524
TEST_P(CompilationCachingTest,SaveToCacheInvalidNumCache)525 TEST_P(CompilationCachingTest, SaveToCacheInvalidNumCache) {
526 // Create test HIDL model and compile.
527 const TestModel& testModel = createTestModel();
528 const Model model = createModel(testModel);
529 if (checkEarlyTermination(model)) return;
530
531 // Test with number of model cache files greater than mNumModelCache.
532 {
533 hidl_vec<hidl_handle> modelCache, dataCache;
534 // Pass an additional cache file for model cache.
535 mModelCache.push_back({mTmpCache});
536 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
537 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
538 mModelCache.pop_back();
539 sp<IPreparedModel> preparedModel = nullptr;
540 saveModelToCache(model, modelCache, dataCache, &preparedModel);
541 ASSERT_NE(preparedModel, nullptr);
542 // Execute and verify results.
543 EvaluatePreparedModel(preparedModel, testModel,
544 /*testDynamicOutputShape=*/false);
545 // Check if prepareModelFromCache fails.
546 preparedModel = nullptr;
547 ErrorStatus status;
548 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
549 if (status != ErrorStatus::INVALID_ARGUMENT) {
550 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
551 }
552 ASSERT_EQ(preparedModel, nullptr);
553 }
554
555 // Test with number of model cache files smaller than mNumModelCache.
556 if (mModelCache.size() > 0) {
557 hidl_vec<hidl_handle> modelCache, dataCache;
558 // Pop out the last cache file.
559 auto tmp = mModelCache.back();
560 mModelCache.pop_back();
561 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
562 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
563 mModelCache.push_back(tmp);
564 sp<IPreparedModel> preparedModel = nullptr;
565 saveModelToCache(model, modelCache, dataCache, &preparedModel);
566 ASSERT_NE(preparedModel, nullptr);
567 // Execute and verify results.
568 EvaluatePreparedModel(preparedModel, testModel,
569 /*testDynamicOutputShape=*/false);
570 // Check if prepareModelFromCache fails.
571 preparedModel = nullptr;
572 ErrorStatus status;
573 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
574 if (status != ErrorStatus::INVALID_ARGUMENT) {
575 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
576 }
577 ASSERT_EQ(preparedModel, nullptr);
578 }
579
580 // Test with number of data cache files greater than mNumDataCache.
581 {
582 hidl_vec<hidl_handle> modelCache, dataCache;
583 // Pass an additional cache file for data cache.
584 mDataCache.push_back({mTmpCache});
585 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
586 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
587 mDataCache.pop_back();
588 sp<IPreparedModel> preparedModel = nullptr;
589 saveModelToCache(model, modelCache, dataCache, &preparedModel);
590 ASSERT_NE(preparedModel, nullptr);
591 // Execute and verify results.
592 EvaluatePreparedModel(preparedModel, testModel,
593 /*testDynamicOutputShape=*/false);
594 // Check if prepareModelFromCache fails.
595 preparedModel = nullptr;
596 ErrorStatus status;
597 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
598 if (status != ErrorStatus::INVALID_ARGUMENT) {
599 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
600 }
601 ASSERT_EQ(preparedModel, nullptr);
602 }
603
604 // Test with number of data cache files smaller than mNumDataCache.
605 if (mDataCache.size() > 0) {
606 hidl_vec<hidl_handle> modelCache, dataCache;
607 // Pop out the last cache file.
608 auto tmp = mDataCache.back();
609 mDataCache.pop_back();
610 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
611 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
612 mDataCache.push_back(tmp);
613 sp<IPreparedModel> preparedModel = nullptr;
614 saveModelToCache(model, modelCache, dataCache, &preparedModel);
615 ASSERT_NE(preparedModel, nullptr);
616 // Execute and verify results.
617 EvaluatePreparedModel(preparedModel, testModel,
618 /*testDynamicOutputShape=*/false);
619 // Check if prepareModelFromCache fails.
620 preparedModel = nullptr;
621 ErrorStatus status;
622 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
623 if (status != ErrorStatus::INVALID_ARGUMENT) {
624 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
625 }
626 ASSERT_EQ(preparedModel, nullptr);
627 }
628 }
629
TEST_P(CompilationCachingTest,PrepareModelFromCacheInvalidNumCache)630 TEST_P(CompilationCachingTest, PrepareModelFromCacheInvalidNumCache) {
631 // Create test HIDL model and compile.
632 const TestModel& testModel = createTestModel();
633 const Model model = createModel(testModel);
634 if (checkEarlyTermination(model)) return;
635
636 // Save the compilation to cache.
637 {
638 hidl_vec<hidl_handle> modelCache, dataCache;
639 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
640 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
641 saveModelToCache(model, modelCache, dataCache);
642 }
643
644 // Test with number of model cache files greater than mNumModelCache.
645 {
646 sp<IPreparedModel> preparedModel = nullptr;
647 ErrorStatus status;
648 hidl_vec<hidl_handle> modelCache, dataCache;
649 mModelCache.push_back({mTmpCache});
650 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
651 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
652 mModelCache.pop_back();
653 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
654 if (status != ErrorStatus::GENERAL_FAILURE) {
655 ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
656 }
657 ASSERT_EQ(preparedModel, nullptr);
658 }
659
660 // Test with number of model cache files smaller than mNumModelCache.
661 if (mModelCache.size() > 0) {
662 sp<IPreparedModel> preparedModel = nullptr;
663 ErrorStatus status;
664 hidl_vec<hidl_handle> modelCache, dataCache;
665 auto tmp = mModelCache.back();
666 mModelCache.pop_back();
667 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
668 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
669 mModelCache.push_back(tmp);
670 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
671 if (status != ErrorStatus::GENERAL_FAILURE) {
672 ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
673 }
674 ASSERT_EQ(preparedModel, nullptr);
675 }
676
677 // Test with number of data cache files greater than mNumDataCache.
678 {
679 sp<IPreparedModel> preparedModel = nullptr;
680 ErrorStatus status;
681 hidl_vec<hidl_handle> modelCache, dataCache;
682 mDataCache.push_back({mTmpCache});
683 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
684 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
685 mDataCache.pop_back();
686 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
687 if (status != ErrorStatus::GENERAL_FAILURE) {
688 ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
689 }
690 ASSERT_EQ(preparedModel, nullptr);
691 }
692
693 // Test with number of data cache files smaller than mNumDataCache.
694 if (mDataCache.size() > 0) {
695 sp<IPreparedModel> preparedModel = nullptr;
696 ErrorStatus status;
697 hidl_vec<hidl_handle> modelCache, dataCache;
698 auto tmp = mDataCache.back();
699 mDataCache.pop_back();
700 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
701 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
702 mDataCache.push_back(tmp);
703 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
704 if (status != ErrorStatus::GENERAL_FAILURE) {
705 ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
706 }
707 ASSERT_EQ(preparedModel, nullptr);
708 }
709 }
710
TEST_P(CompilationCachingTest,SaveToCacheInvalidNumFd)711 TEST_P(CompilationCachingTest, SaveToCacheInvalidNumFd) {
712 // Create test HIDL model and compile.
713 const TestModel& testModel = createTestModel();
714 const Model model = createModel(testModel);
715 if (checkEarlyTermination(model)) return;
716
717 // Go through each handle in model cache, test with NumFd greater than 1.
718 for (uint32_t i = 0; i < mNumModelCache; i++) {
719 hidl_vec<hidl_handle> modelCache, dataCache;
720 // Pass an invalid number of fds for handle i.
721 mModelCache[i].push_back(mTmpCache);
722 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
723 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
724 mModelCache[i].pop_back();
725 sp<IPreparedModel> preparedModel = nullptr;
726 saveModelToCache(model, modelCache, dataCache, &preparedModel);
727 ASSERT_NE(preparedModel, nullptr);
728 // Execute and verify results.
729 EvaluatePreparedModel(preparedModel, testModel,
730 /*testDynamicOutputShape=*/false);
731 // Check if prepareModelFromCache fails.
732 preparedModel = nullptr;
733 ErrorStatus status;
734 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
735 if (status != ErrorStatus::INVALID_ARGUMENT) {
736 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
737 }
738 ASSERT_EQ(preparedModel, nullptr);
739 }
740
741 // Go through each handle in model cache, test with NumFd equal to 0.
742 for (uint32_t i = 0; i < mNumModelCache; i++) {
743 hidl_vec<hidl_handle> modelCache, dataCache;
744 // Pass an invalid number of fds for handle i.
745 auto tmp = mModelCache[i].back();
746 mModelCache[i].pop_back();
747 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
748 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
749 mModelCache[i].push_back(tmp);
750 sp<IPreparedModel> preparedModel = nullptr;
751 saveModelToCache(model, modelCache, dataCache, &preparedModel);
752 ASSERT_NE(preparedModel, nullptr);
753 // Execute and verify results.
754 EvaluatePreparedModel(preparedModel, testModel,
755 /*testDynamicOutputShape=*/false);
756 // Check if prepareModelFromCache fails.
757 preparedModel = nullptr;
758 ErrorStatus status;
759 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
760 if (status != ErrorStatus::INVALID_ARGUMENT) {
761 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
762 }
763 ASSERT_EQ(preparedModel, nullptr);
764 }
765
766 // Go through each handle in data cache, test with NumFd greater than 1.
767 for (uint32_t i = 0; i < mNumDataCache; i++) {
768 hidl_vec<hidl_handle> modelCache, dataCache;
769 // Pass an invalid number of fds for handle i.
770 mDataCache[i].push_back(mTmpCache);
771 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
772 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
773 mDataCache[i].pop_back();
774 sp<IPreparedModel> preparedModel = nullptr;
775 saveModelToCache(model, modelCache, dataCache, &preparedModel);
776 ASSERT_NE(preparedModel, nullptr);
777 // Execute and verify results.
778 EvaluatePreparedModel(preparedModel, testModel,
779 /*testDynamicOutputShape=*/false);
780 // Check if prepareModelFromCache fails.
781 preparedModel = nullptr;
782 ErrorStatus status;
783 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
784 if (status != ErrorStatus::INVALID_ARGUMENT) {
785 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
786 }
787 ASSERT_EQ(preparedModel, nullptr);
788 }
789
790 // Go through each handle in data cache, test with NumFd equal to 0.
791 for (uint32_t i = 0; i < mNumDataCache; i++) {
792 hidl_vec<hidl_handle> modelCache, dataCache;
793 // Pass an invalid number of fds for handle i.
794 auto tmp = mDataCache[i].back();
795 mDataCache[i].pop_back();
796 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
797 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
798 mDataCache[i].push_back(tmp);
799 sp<IPreparedModel> preparedModel = nullptr;
800 saveModelToCache(model, modelCache, dataCache, &preparedModel);
801 ASSERT_NE(preparedModel, nullptr);
802 // Execute and verify results.
803 EvaluatePreparedModel(preparedModel, testModel,
804 /*testDynamicOutputShape=*/false);
805 // Check if prepareModelFromCache fails.
806 preparedModel = nullptr;
807 ErrorStatus status;
808 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
809 if (status != ErrorStatus::INVALID_ARGUMENT) {
810 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
811 }
812 ASSERT_EQ(preparedModel, nullptr);
813 }
814 }
815
TEST_P(CompilationCachingTest,PrepareModelFromCacheInvalidNumFd)816 TEST_P(CompilationCachingTest, PrepareModelFromCacheInvalidNumFd) {
817 // Create test HIDL model and compile.
818 const TestModel& testModel = createTestModel();
819 const Model model = createModel(testModel);
820 if (checkEarlyTermination(model)) return;
821
822 // Save the compilation to cache.
823 {
824 hidl_vec<hidl_handle> modelCache, dataCache;
825 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
826 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
827 saveModelToCache(model, modelCache, dataCache);
828 }
829
830 // Go through each handle in model cache, test with NumFd greater than 1.
831 for (uint32_t i = 0; i < mNumModelCache; i++) {
832 sp<IPreparedModel> preparedModel = nullptr;
833 ErrorStatus status;
834 hidl_vec<hidl_handle> modelCache, dataCache;
835 mModelCache[i].push_back(mTmpCache);
836 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
837 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
838 mModelCache[i].pop_back();
839 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
840 if (status != ErrorStatus::GENERAL_FAILURE) {
841 ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
842 }
843 ASSERT_EQ(preparedModel, nullptr);
844 }
845
846 // Go through each handle in model cache, test with NumFd equal to 0.
847 for (uint32_t i = 0; i < mNumModelCache; i++) {
848 sp<IPreparedModel> preparedModel = nullptr;
849 ErrorStatus status;
850 hidl_vec<hidl_handle> modelCache, dataCache;
851 auto tmp = mModelCache[i].back();
852 mModelCache[i].pop_back();
853 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
854 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
855 mModelCache[i].push_back(tmp);
856 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
857 if (status != ErrorStatus::GENERAL_FAILURE) {
858 ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
859 }
860 ASSERT_EQ(preparedModel, nullptr);
861 }
862
863 // Go through each handle in data cache, test with NumFd greater than 1.
864 for (uint32_t i = 0; i < mNumDataCache; i++) {
865 sp<IPreparedModel> preparedModel = nullptr;
866 ErrorStatus status;
867 hidl_vec<hidl_handle> modelCache, dataCache;
868 mDataCache[i].push_back(mTmpCache);
869 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
870 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
871 mDataCache[i].pop_back();
872 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
873 if (status != ErrorStatus::GENERAL_FAILURE) {
874 ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
875 }
876 ASSERT_EQ(preparedModel, nullptr);
877 }
878
879 // Go through each handle in data cache, test with NumFd equal to 0.
880 for (uint32_t i = 0; i < mNumDataCache; i++) {
881 sp<IPreparedModel> preparedModel = nullptr;
882 ErrorStatus status;
883 hidl_vec<hidl_handle> modelCache, dataCache;
884 auto tmp = mDataCache[i].back();
885 mDataCache[i].pop_back();
886 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
887 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
888 mDataCache[i].push_back(tmp);
889 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
890 if (status != ErrorStatus::GENERAL_FAILURE) {
891 ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
892 }
893 ASSERT_EQ(preparedModel, nullptr);
894 }
895 }
896
TEST_P(CompilationCachingTest,SaveToCacheInvalidAccessMode)897 TEST_P(CompilationCachingTest, SaveToCacheInvalidAccessMode) {
898 // Create test HIDL model and compile.
899 const TestModel& testModel = createTestModel();
900 const Model model = createModel(testModel);
901 if (checkEarlyTermination(model)) return;
902 std::vector<AccessMode> modelCacheMode(mNumModelCache, AccessMode::READ_WRITE);
903 std::vector<AccessMode> dataCacheMode(mNumDataCache, AccessMode::READ_WRITE);
904
905 // Go through each handle in model cache, test with invalid access mode.
906 for (uint32_t i = 0; i < mNumModelCache; i++) {
907 hidl_vec<hidl_handle> modelCache, dataCache;
908 modelCacheMode[i] = AccessMode::READ_ONLY;
909 createCacheHandles(mModelCache, modelCacheMode, &modelCache);
910 createCacheHandles(mDataCache, dataCacheMode, &dataCache);
911 modelCacheMode[i] = AccessMode::READ_WRITE;
912 sp<IPreparedModel> preparedModel = nullptr;
913 saveModelToCache(model, modelCache, dataCache, &preparedModel);
914 ASSERT_NE(preparedModel, nullptr);
915 // Execute and verify results.
916 EvaluatePreparedModel(preparedModel, testModel,
917 /*testDynamicOutputShape=*/false);
918 // Check if prepareModelFromCache fails.
919 preparedModel = nullptr;
920 ErrorStatus status;
921 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
922 if (status != ErrorStatus::INVALID_ARGUMENT) {
923 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
924 }
925 ASSERT_EQ(preparedModel, nullptr);
926 }
927
928 // Go through each handle in data cache, test with invalid access mode.
929 for (uint32_t i = 0; i < mNumDataCache; i++) {
930 hidl_vec<hidl_handle> modelCache, dataCache;
931 dataCacheMode[i] = AccessMode::READ_ONLY;
932 createCacheHandles(mModelCache, modelCacheMode, &modelCache);
933 createCacheHandles(mDataCache, dataCacheMode, &dataCache);
934 dataCacheMode[i] = AccessMode::READ_WRITE;
935 sp<IPreparedModel> preparedModel = nullptr;
936 saveModelToCache(model, modelCache, dataCache, &preparedModel);
937 ASSERT_NE(preparedModel, nullptr);
938 // Execute and verify results.
939 EvaluatePreparedModel(preparedModel, testModel,
940 /*testDynamicOutputShape=*/false);
941 // Check if prepareModelFromCache fails.
942 preparedModel = nullptr;
943 ErrorStatus status;
944 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
945 if (status != ErrorStatus::INVALID_ARGUMENT) {
946 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
947 }
948 ASSERT_EQ(preparedModel, nullptr);
949 }
950 }
951
TEST_P(CompilationCachingTest,PrepareModelFromCacheInvalidAccessMode)952 TEST_P(CompilationCachingTest, PrepareModelFromCacheInvalidAccessMode) {
953 // Create test HIDL model and compile.
954 const TestModel& testModel = createTestModel();
955 const Model model = createModel(testModel);
956 if (checkEarlyTermination(model)) return;
957 std::vector<AccessMode> modelCacheMode(mNumModelCache, AccessMode::READ_WRITE);
958 std::vector<AccessMode> dataCacheMode(mNumDataCache, AccessMode::READ_WRITE);
959
960 // Save the compilation to cache.
961 {
962 hidl_vec<hidl_handle> modelCache, dataCache;
963 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
964 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
965 saveModelToCache(model, modelCache, dataCache);
966 }
967
968 // Go through each handle in model cache, test with invalid access mode.
969 for (uint32_t i = 0; i < mNumModelCache; i++) {
970 sp<IPreparedModel> preparedModel = nullptr;
971 ErrorStatus status;
972 hidl_vec<hidl_handle> modelCache, dataCache;
973 modelCacheMode[i] = AccessMode::WRITE_ONLY;
974 createCacheHandles(mModelCache, modelCacheMode, &modelCache);
975 createCacheHandles(mDataCache, dataCacheMode, &dataCache);
976 modelCacheMode[i] = AccessMode::READ_WRITE;
977 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
978 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
979 ASSERT_EQ(preparedModel, nullptr);
980 }
981
982 // Go through each handle in data cache, test with invalid access mode.
983 for (uint32_t i = 0; i < mNumDataCache; i++) {
984 sp<IPreparedModel> preparedModel = nullptr;
985 ErrorStatus status;
986 hidl_vec<hidl_handle> modelCache, dataCache;
987 dataCacheMode[i] = AccessMode::WRITE_ONLY;
988 createCacheHandles(mModelCache, modelCacheMode, &modelCache);
989 createCacheHandles(mDataCache, dataCacheMode, &dataCache);
990 dataCacheMode[i] = AccessMode::READ_WRITE;
991 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
992 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
993 ASSERT_EQ(preparedModel, nullptr);
994 }
995 }
996
997 // Copy file contents between file groups.
998 // The outer vector corresponds to handles and the inner vector is for fds held by each handle.
999 // The outer vector sizes must match and the inner vectors must have size = 1.
copyCacheFiles(const std::vector<std::vector<std::string>> & from,const std::vector<std::vector<std::string>> & to)1000 static void copyCacheFiles(const std::vector<std::vector<std::string>>& from,
1001 const std::vector<std::vector<std::string>>& to) {
1002 constexpr size_t kBufferSize = 1000000;
1003 uint8_t buffer[kBufferSize];
1004
1005 ASSERT_EQ(from.size(), to.size());
1006 for (uint32_t i = 0; i < from.size(); i++) {
1007 ASSERT_EQ(from[i].size(), 1u);
1008 ASSERT_EQ(to[i].size(), 1u);
1009 int fromFd = open(from[i][0].c_str(), O_RDONLY);
1010 int toFd = open(to[i][0].c_str(), O_WRONLY | O_CREAT, S_IRUSR | S_IWUSR);
1011 ASSERT_GE(fromFd, 0);
1012 ASSERT_GE(toFd, 0);
1013
1014 ssize_t readBytes;
1015 while ((readBytes = read(fromFd, &buffer, kBufferSize)) > 0) {
1016 ASSERT_EQ(write(toFd, &buffer, readBytes), readBytes);
1017 }
1018 ASSERT_GE(readBytes, 0);
1019
1020 close(fromFd);
1021 close(toFd);
1022 }
1023 }
1024
1025 // Number of operations in the large test model.
1026 constexpr uint32_t kLargeModelSize = 100;
1027 constexpr uint32_t kNumIterationsTOCTOU = 100;
1028
TEST_P(CompilationCachingTest,SaveToCache_TOCTOU)1029 TEST_P(CompilationCachingTest, SaveToCache_TOCTOU) {
1030 if (!mIsCachingSupported) return;
1031
1032 // Create test models and check if fully supported by the service.
1033 const TestModel testModelMul = createLargeTestModel(OperationType::MUL, kLargeModelSize);
1034 const Model modelMul = createModel(testModelMul);
1035 if (checkEarlyTermination(modelMul)) return;
1036 const TestModel testModelAdd = createLargeTestModel(OperationType::ADD, kLargeModelSize);
1037 const Model modelAdd = createModel(testModelAdd);
1038 if (checkEarlyTermination(modelAdd)) return;
1039
1040 // Save the modelMul compilation to cache.
1041 auto modelCacheMul = mModelCache;
1042 for (auto& cache : modelCacheMul) {
1043 cache[0].append("_mul");
1044 }
1045 {
1046 hidl_vec<hidl_handle> modelCache, dataCache;
1047 createCacheHandles(modelCacheMul, AccessMode::READ_WRITE, &modelCache);
1048 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
1049 saveModelToCache(modelMul, modelCache, dataCache);
1050 }
1051
1052 // Use a different token for modelAdd.
1053 mToken[0]++;
1054
1055 // This test is probabilistic, so we run it multiple times.
1056 for (uint32_t i = 0; i < kNumIterationsTOCTOU; i++) {
1057 // Save the modelAdd compilation to cache.
1058 {
1059 hidl_vec<hidl_handle> modelCache, dataCache;
1060 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
1061 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
1062
1063 // Spawn a thread to copy the cache content concurrently while saving to cache.
1064 std::thread thread(copyCacheFiles, std::cref(modelCacheMul), std::cref(mModelCache));
1065 saveModelToCache(modelAdd, modelCache, dataCache);
1066 thread.join();
1067 }
1068
1069 // Retrieve preparedModel from cache.
1070 {
1071 sp<IPreparedModel> preparedModel = nullptr;
1072 ErrorStatus status;
1073 hidl_vec<hidl_handle> modelCache, dataCache;
1074 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
1075 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
1076 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
1077
1078 // The preparation may fail or succeed, but must not crash. If the preparation succeeds,
1079 // the prepared model must be executed with the correct result and not crash.
1080 if (status != ErrorStatus::NONE) {
1081 ASSERT_EQ(preparedModel, nullptr);
1082 } else {
1083 ASSERT_NE(preparedModel, nullptr);
1084 EvaluatePreparedModel(preparedModel, testModelAdd,
1085 /*testDynamicOutputShape=*/false);
1086 }
1087 }
1088 }
1089 }
1090
TEST_P(CompilationCachingTest,PrepareFromCache_TOCTOU)1091 TEST_P(CompilationCachingTest, PrepareFromCache_TOCTOU) {
1092 if (!mIsCachingSupported) return;
1093
1094 // Create test models and check if fully supported by the service.
1095 const TestModel testModelMul = createLargeTestModel(OperationType::MUL, kLargeModelSize);
1096 const Model modelMul = createModel(testModelMul);
1097 if (checkEarlyTermination(modelMul)) return;
1098 const TestModel testModelAdd = createLargeTestModel(OperationType::ADD, kLargeModelSize);
1099 const Model modelAdd = createModel(testModelAdd);
1100 if (checkEarlyTermination(modelAdd)) return;
1101
1102 // Save the modelMul compilation to cache.
1103 auto modelCacheMul = mModelCache;
1104 for (auto& cache : modelCacheMul) {
1105 cache[0].append("_mul");
1106 }
1107 {
1108 hidl_vec<hidl_handle> modelCache, dataCache;
1109 createCacheHandles(modelCacheMul, AccessMode::READ_WRITE, &modelCache);
1110 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
1111 saveModelToCache(modelMul, modelCache, dataCache);
1112 }
1113
1114 // Use a different token for modelAdd.
1115 mToken[0]++;
1116
1117 // This test is probabilistic, so we run it multiple times.
1118 for (uint32_t i = 0; i < kNumIterationsTOCTOU; i++) {
1119 // Save the modelAdd compilation to cache.
1120 {
1121 hidl_vec<hidl_handle> modelCache, dataCache;
1122 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
1123 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
1124 saveModelToCache(modelAdd, modelCache, dataCache);
1125 }
1126
1127 // Retrieve preparedModel from cache.
1128 {
1129 sp<IPreparedModel> preparedModel = nullptr;
1130 ErrorStatus status;
1131 hidl_vec<hidl_handle> modelCache, dataCache;
1132 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
1133 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
1134
1135 // Spawn a thread to copy the cache content concurrently while preparing from cache.
1136 std::thread thread(copyCacheFiles, std::cref(modelCacheMul), std::cref(mModelCache));
1137 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
1138 thread.join();
1139
1140 // The preparation may fail or succeed, but must not crash. If the preparation succeeds,
1141 // the prepared model must be executed with the correct result and not crash.
1142 if (status != ErrorStatus::NONE) {
1143 ASSERT_EQ(preparedModel, nullptr);
1144 } else {
1145 ASSERT_NE(preparedModel, nullptr);
1146 EvaluatePreparedModel(preparedModel, testModelAdd,
1147 /*testDynamicOutputShape=*/false);
1148 }
1149 }
1150 }
1151 }
1152
TEST_P(CompilationCachingTest,ReplaceSecuritySensitiveCache)1153 TEST_P(CompilationCachingTest, ReplaceSecuritySensitiveCache) {
1154 if (!mIsCachingSupported) return;
1155
1156 // Create test models and check if fully supported by the service.
1157 const TestModel testModelMul = createLargeTestModel(OperationType::MUL, kLargeModelSize);
1158 const Model modelMul = createModel(testModelMul);
1159 if (checkEarlyTermination(modelMul)) return;
1160 const TestModel testModelAdd = createLargeTestModel(OperationType::ADD, kLargeModelSize);
1161 const Model modelAdd = createModel(testModelAdd);
1162 if (checkEarlyTermination(modelAdd)) return;
1163
1164 // Save the modelMul compilation to cache.
1165 auto modelCacheMul = mModelCache;
1166 for (auto& cache : modelCacheMul) {
1167 cache[0].append("_mul");
1168 }
1169 {
1170 hidl_vec<hidl_handle> modelCache, dataCache;
1171 createCacheHandles(modelCacheMul, AccessMode::READ_WRITE, &modelCache);
1172 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
1173 saveModelToCache(modelMul, modelCache, dataCache);
1174 }
1175
1176 // Use a different token for modelAdd.
1177 mToken[0]++;
1178
1179 // Save the modelAdd compilation to cache.
1180 {
1181 hidl_vec<hidl_handle> modelCache, dataCache;
1182 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
1183 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
1184 saveModelToCache(modelAdd, modelCache, dataCache);
1185 }
1186
1187 // Replace the model cache of modelAdd with modelMul.
1188 copyCacheFiles(modelCacheMul, mModelCache);
1189
1190 // Retrieve the preparedModel from cache, expect failure.
1191 {
1192 sp<IPreparedModel> preparedModel = nullptr;
1193 ErrorStatus status;
1194 hidl_vec<hidl_handle> modelCache, dataCache;
1195 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
1196 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
1197 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
1198 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
1199 ASSERT_EQ(preparedModel, nullptr);
1200 }
1201 }
1202
1203 static const auto kNamedDeviceChoices = testing::ValuesIn(getNamedDevices());
1204 static const auto kOperandTypeChoices =
1205 testing::Values(OperandType::TENSOR_FLOAT32, OperandType::TENSOR_QUANT8_ASYMM);
1206
printCompilationCachingTest(const testing::TestParamInfo<CompilationCachingTestParam> & info)1207 std::string printCompilationCachingTest(
1208 const testing::TestParamInfo<CompilationCachingTestParam>& info) {
1209 const auto& [namedDevice, operandType] = info.param;
1210 const std::string type = (operandType == OperandType::TENSOR_FLOAT32 ? "float32" : "quant8");
1211 return gtestCompliantName(getName(namedDevice) + "_" + type);
1212 }
1213
1214 GTEST_ALLOW_UNINSTANTIATED_PARAMETERIZED_TEST(CompilationCachingTest);
1215 INSTANTIATE_TEST_SUITE_P(TestCompilationCaching, CompilationCachingTest,
1216 testing::Combine(kNamedDeviceChoices, kOperandTypeChoices),
1217 printCompilationCachingTest);
1218
1219 using CompilationCachingSecurityTestParam = std::tuple<NamedDevice, OperandType, uint32_t>;
1220
1221 class CompilationCachingSecurityTest
1222 : public CompilationCachingTestBase,
1223 public testing::WithParamInterface<CompilationCachingSecurityTestParam> {
1224 protected:
CompilationCachingSecurityTest()1225 CompilationCachingSecurityTest()
1226 : CompilationCachingTestBase(getData(std::get<NamedDevice>(GetParam())),
1227 std::get<OperandType>(GetParam())) {}
1228
SetUp()1229 void SetUp() {
1230 CompilationCachingTestBase::SetUp();
1231 generator.seed(kSeed);
1232 }
1233
1234 // Get a random integer within a closed range [lower, upper].
1235 template <typename T>
getRandomInt(T lower,T upper)1236 T getRandomInt(T lower, T upper) {
1237 std::uniform_int_distribution<T> dis(lower, upper);
1238 return dis(generator);
1239 }
1240
1241 // Randomly flip one single bit of the cache entry.
flipOneBitOfCache(const std::string & filename,bool * skip)1242 void flipOneBitOfCache(const std::string& filename, bool* skip) {
1243 FILE* pFile = fopen(filename.c_str(), "r+");
1244 ASSERT_EQ(fseek(pFile, 0, SEEK_END), 0);
1245 long int fileSize = ftell(pFile);
1246 if (fileSize == 0) {
1247 fclose(pFile);
1248 *skip = true;
1249 return;
1250 }
1251 ASSERT_EQ(fseek(pFile, getRandomInt(0l, fileSize - 1), SEEK_SET), 0);
1252 int readByte = fgetc(pFile);
1253 ASSERT_NE(readByte, EOF);
1254 ASSERT_EQ(fseek(pFile, -1, SEEK_CUR), 0);
1255 ASSERT_NE(fputc(static_cast<uint8_t>(readByte) ^ (1U << getRandomInt(0, 7)), pFile), EOF);
1256 fclose(pFile);
1257 *skip = false;
1258 }
1259
1260 // Randomly append bytes to the cache entry.
appendBytesToCache(const std::string & filename,bool * skip)1261 void appendBytesToCache(const std::string& filename, bool* skip) {
1262 FILE* pFile = fopen(filename.c_str(), "a");
1263 uint32_t appendLength = getRandomInt(1, 256);
1264 for (uint32_t i = 0; i < appendLength; i++) {
1265 ASSERT_NE(fputc(getRandomInt<uint16_t>(0, 255), pFile), EOF);
1266 }
1267 fclose(pFile);
1268 *skip = false;
1269 }
1270
1271 enum class ExpectedResult { GENERAL_FAILURE, NOT_CRASH };
1272
1273 // Test if the driver behaves as expected when given corrupted cache or token.
1274 // The modifier will be invoked after save to cache but before prepare from cache.
1275 // The modifier accepts one pointer argument "skip" as the returning value, indicating
1276 // whether the test should be skipped or not.
testCorruptedCache(ExpectedResult expected,std::function<void (bool *)> modifier)1277 void testCorruptedCache(ExpectedResult expected, std::function<void(bool*)> modifier) {
1278 const TestModel& testModel = createTestModel();
1279 const Model model = createModel(testModel);
1280 if (checkEarlyTermination(model)) return;
1281
1282 // Save the compilation to cache.
1283 {
1284 hidl_vec<hidl_handle> modelCache, dataCache;
1285 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
1286 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
1287 saveModelToCache(model, modelCache, dataCache);
1288 }
1289
1290 bool skip = false;
1291 modifier(&skip);
1292 if (skip) return;
1293
1294 // Retrieve preparedModel from cache.
1295 {
1296 sp<IPreparedModel> preparedModel = nullptr;
1297 ErrorStatus status;
1298 hidl_vec<hidl_handle> modelCache, dataCache;
1299 createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
1300 createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
1301 prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
1302
1303 switch (expected) {
1304 case ExpectedResult::GENERAL_FAILURE:
1305 ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
1306 ASSERT_EQ(preparedModel, nullptr);
1307 break;
1308 case ExpectedResult::NOT_CRASH:
1309 ASSERT_EQ(preparedModel == nullptr, status != ErrorStatus::NONE);
1310 break;
1311 default:
1312 FAIL();
1313 }
1314 }
1315 }
1316
1317 const uint32_t kSeed = std::get<uint32_t>(GetParam());
1318 std::mt19937 generator;
1319 };
1320
TEST_P(CompilationCachingSecurityTest,CorruptedModelCache)1321 TEST_P(CompilationCachingSecurityTest, CorruptedModelCache) {
1322 if (!mIsCachingSupported) return;
1323 for (uint32_t i = 0; i < mNumModelCache; i++) {
1324 testCorruptedCache(ExpectedResult::GENERAL_FAILURE,
1325 [this, i](bool* skip) { flipOneBitOfCache(mModelCache[i][0], skip); });
1326 }
1327 }
1328
TEST_P(CompilationCachingSecurityTest,WrongLengthModelCache)1329 TEST_P(CompilationCachingSecurityTest, WrongLengthModelCache) {
1330 if (!mIsCachingSupported) return;
1331 for (uint32_t i = 0; i < mNumModelCache; i++) {
1332 testCorruptedCache(ExpectedResult::GENERAL_FAILURE,
1333 [this, i](bool* skip) { appendBytesToCache(mModelCache[i][0], skip); });
1334 }
1335 }
1336
TEST_P(CompilationCachingSecurityTest,CorruptedDataCache)1337 TEST_P(CompilationCachingSecurityTest, CorruptedDataCache) {
1338 if (!mIsCachingSupported) return;
1339 for (uint32_t i = 0; i < mNumDataCache; i++) {
1340 testCorruptedCache(ExpectedResult::NOT_CRASH,
1341 [this, i](bool* skip) { flipOneBitOfCache(mDataCache[i][0], skip); });
1342 }
1343 }
1344
TEST_P(CompilationCachingSecurityTest,WrongLengthDataCache)1345 TEST_P(CompilationCachingSecurityTest, WrongLengthDataCache) {
1346 if (!mIsCachingSupported) return;
1347 for (uint32_t i = 0; i < mNumDataCache; i++) {
1348 testCorruptedCache(ExpectedResult::NOT_CRASH,
1349 [this, i](bool* skip) { appendBytesToCache(mDataCache[i][0], skip); });
1350 }
1351 }
1352
TEST_P(CompilationCachingSecurityTest,WrongToken)1353 TEST_P(CompilationCachingSecurityTest, WrongToken) {
1354 if (!mIsCachingSupported) return;
1355 testCorruptedCache(ExpectedResult::GENERAL_FAILURE, [this](bool* skip) {
1356 // Randomly flip one single bit in mToken.
1357 uint32_t ind =
1358 getRandomInt(0u, static_cast<uint32_t>(Constant::BYTE_SIZE_OF_CACHE_TOKEN) - 1);
1359 mToken[ind] ^= (1U << getRandomInt(0, 7));
1360 *skip = false;
1361 });
1362 }
1363
printCompilationCachingSecurityTest(const testing::TestParamInfo<CompilationCachingSecurityTestParam> & info)1364 std::string printCompilationCachingSecurityTest(
1365 const testing::TestParamInfo<CompilationCachingSecurityTestParam>& info) {
1366 const auto& [namedDevice, operandType, seed] = info.param;
1367 const std::string type = (operandType == OperandType::TENSOR_FLOAT32 ? "float32" : "quant8");
1368 return gtestCompliantName(getName(namedDevice) + "_" + type + "_" + std::to_string(seed));
1369 }
1370
1371 GTEST_ALLOW_UNINSTANTIATED_PARAMETERIZED_TEST(CompilationCachingSecurityTest);
1372 INSTANTIATE_TEST_SUITE_P(TestCompilationCaching, CompilationCachingSecurityTest,
1373 testing::Combine(kNamedDeviceChoices, kOperandTypeChoices,
1374 testing::Range(0U, 10U)),
1375 printCompilationCachingSecurityTest);
1376
1377 } // namespace android::hardware::neuralnetworks::V1_2::vts::functional
1378