| 1 |
/* |
| 2 |
* Copyright (C) 2021-2026 Intel Corporation |
| 3 |
* |
| 4 |
* SPDX-License-Identifier: MIT |
| 5 |
* |
| 6 |
*/ |
| 7 |
|
| 8 |
#pragma once |
| 9 |
#include "shared/source/helpers/aligned_memory.h" |
| 10 |
#include "shared/source/helpers/hw_info.h" |
| 11 |
#include "shared/source/os_interface/os_interface.h" |
| 12 |
#include "shared/test/common/mocks/mock_sip.h" |
| 13 |
|
| 14 |
#include "level_zero/core/source/device/device.h" |
| 15 |
#include "level_zero/tools/source/debug/debug_session.h" |
| 16 |
#include "level_zero/tools/source/debug/debug_session_imp.h" |
| 17 |
|
| 18 |
namespace NEO { |
| 19 |
class EuDebugInterface; |
| 20 |
struct EuDebugConnect; |
| 21 |
} // namespace NEO |
| 22 |
|
| 23 |
namespace L0 { |
| 24 |
DebugSession *createDebugSessionHelper(const zet_debug_config_t &config, Device *device, int debugFd, void *params); |
| 25 |
ze_result_t openConnectionUpstreamHelper(int pid, Device *device, NEO::EuDebugInterface &debugInterface, NEO::EuDebugConnect *open, int &debugFd); |
| 26 |
|
| 27 |
namespace ult { |
| 28 |
|
| 29 |
using CreateDebugSessionHelperFunc = decltype(&L0::createDebugSessionHelper); |
| 30 |
using OpenConnectionUpstreamHelperFunc = std::function<ze_result_t(int, L0::Device *, NEO::EuDebugInterface &, NEO::EuDebugConnect *, int &)>; |
| 31 |
|
| 32 |
class OsInterfaceWithDebugAttach : public NEO::OSInterface { |
| 33 |
public: |
| 34 |
OsInterfaceWithDebugAttach() : OSInterface() {} |
| 35 |
bool isDebugAttachAvailable() const override { |
| 36 |
return debugAttachAvailable; |
| 37 |
} |
| 38 |
|
| 39 |
bool debugAttachAvailable = true; |
| 40 |
}; |
| 41 |
|
| 42 |
struct DebugSessionMock : public L0::DebugSession { |
| 43 |
using L0::DebugSession::allThreads; |
| 44 |
using L0::DebugSession::debugArea; |
| 45 |
using L0::DebugSession::fillDevicesFromThread; |
| 46 |
using L0::DebugSession::getPerThreadScratchOffset; |
| 47 |
using L0::DebugSession::getSingleThreadsForDevice; |
| 48 |
using L0::DebugSession::isBindlessSystemRoutine; |
| 49 |
|
| 50 |
DebugSessionMock(const zet_debug_config_t &config, L0::Device *device) : DebugSession(config, device), config(config) { |
| 51 |
topologyMap = buildMockTopology(device); |
| 52 |
} |
| 53 |
|
| 54 |
bool closeConnection() override { |
| 55 |
closeConnectionCalled = true; |
| 56 |
return true; |
| 57 |
} |
| 58 |
|
| 59 |
static NEO::TopologyMap buildMockTopology(L0::Device *device) { |
| 60 |
NEO::TopologyMap topologyMap; |
| 61 |
auto &hwInfo = device->getHwInfo(); |
| 62 |
uint32_t tileCount; |
| 63 |
if (hwInfo.gtSystemInfo.MultiTileArchInfo.IsValid) { |
| 64 |
tileCount = hwInfo.gtSystemInfo.MultiTileArchInfo.TileCount; |
| 65 |
} else { |
| 66 |
tileCount = 1; |
| 67 |
} |
| 68 |
|
| 69 |
for (uint32_t deviceIdx = 0; deviceIdx < tileCount; deviceIdx++) { |
| 70 |
NEO::TopologyMapping mapping; |
| 71 |
for (uint32_t i = 0; i < hwInfo.gtSystemInfo.SliceCount; i++) { |
| 72 |
mapping.sliceIndices.push_back(static_cast<int>(i)); |
| 73 |
} |
| 74 |
if (hwInfo.gtSystemInfo.SliceCount == 1) { |
| 75 |
for (uint32_t i = 0; i < hwInfo.gtSystemInfo.SubSliceCount; i++) { |
| 76 |
mapping.subsliceIndices.push_back(static_cast<int>(i)); |
| 77 |
} |
| 78 |
} |
| 79 |
topologyMap[deviceIdx] = mapping; |
| 80 |
} |
| 81 |
return topologyMap; |
| 82 |
} |
| 83 |
|
| 84 |
ze_result_t initialize() override { |
| 85 |
if (config.pid == 0) { |
| 86 |
return ZE_RESULT_ERROR_UNKNOWN; |
| 87 |
} |
| 88 |
createEuThreads(); |
| 89 |
return ZE_RESULT_SUCCESS; |
| 90 |
} |
| 91 |
ze_result_t readEvent(uint64_t timeout, zet_debug_event_t *event) override { |
| 92 |
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE; |
| 93 |
} |
| 94 |
ze_result_t interrupt(ze_device_thread_t thread) override { |
| 95 |
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE; |
| 96 |
} |
| 97 |
ze_result_t resume(ze_device_thread_t thread) override { |
| 98 |
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE; |
| 99 |
} |
| 100 |
ze_result_t readMemory(ze_device_thread_t thread, const zet_debug_memory_space_desc_t *desc, size_t size, void *buffer) override { |
| 101 |
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE; |
| 102 |
} |
| 103 |
ze_result_t writeMemory(ze_device_thread_t thread, const zet_debug_memory_space_desc_t *desc, size_t size, const void *buffer) override { |
| 104 |
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE; |
| 105 |
} |
| 106 |
ze_result_t acknowledgeEvent(const zet_debug_event_t *event) override { |
| 107 |
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE; |
| 108 |
} |
| 109 |
ze_result_t readRegisters(ze_device_thread_t thread, uint32_t type, uint32_t start, uint32_t count, void *pRegisterValues) override { |
| 110 |
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE; |
| 111 |
} |
| 112 |
ze_result_t writeRegisters(ze_device_thread_t thread, uint32_t type, uint32_t start, uint32_t count, void *pRegisterValues) override { |
| 113 |
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE; |
| 114 |
} |
| 115 |
ze_result_t readSbaBuffer(EuThread::ThreadId threadId, NEO::SbaTrackedAddresses &sbaBuffer) override { |
| 116 |
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE; |
| 117 |
} |
| 118 |
void startAsyncThread() override { |
| 119 |
asyncThreadStarted = true; |
| 120 |
} |
| 121 |
|
| 122 |
bool readModuleDebugArea() override { |
| 123 |
return true; |
| 124 |
} |
| 125 |
|
| 126 |
ze_result_t readRegistersImp(EuThread::ThreadId thread, uint32_t type, uint32_t start, uint32_t count, void *pRegisterValues) override { |
| 127 |
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE; |
| 128 |
} |
| 129 |
|
| 130 |
uint32_t getRegisterSize(uint32_t type) override { |
| 131 |
return 0; |
| 132 |
} |
| 133 |
|
| 134 |
const SIP::regset_desc *typeToRegsetDesc(const NEO::StateSaveAreaHeader *pStateSaveAreaHeader, uint32_t type, L0::Device *device) { |
| 135 |
return DebugSessionImp::typeToRegsetDesc(pStateSaveAreaHeader, type, device); |
| 136 |
} |
| 137 |
|
| 138 |
bool getRegisterAccessProperties(EuThread::ThreadId *threadId, uint32_t *pCount, zet_debug_regset_properties_t *pRegisterSetProperties) override { |
| 139 |
registerAccessPropertiesCalled = true; |
| 140 |
return true; // ignore parameters for mock |
| 141 |
} |
| 142 |
|
| 143 |
bool registerAccessPropertiesCalled = false; |
| 144 |
|
| 145 |
void detachTileDebugSession(DebugSession *tileSession) override {} |
| 146 |
bool areAllTileDebugSessionDetached() override { return true; } |
| 147 |
|
| 148 |
L0::DebugSession *attachTileDebugSession(L0::Device *device) override { |
| 149 |
return nullptr; |
| 150 |
} |
| 151 |
|
| 152 |
void setAttachMode(bool isRootAttach) override {} |
| 153 |
|
| 154 |
const NEO::TopologyMap &getTopologyMap() override { |
| 155 |
return this->topologyMap; |
| 156 |
} |
| 157 |
|
| 158 |
bool openSipWrapper(NEO::Device *neoDevice, uint64_t contextHandle, uint64_t gpuVa) override { |
| 159 |
return true; |
| 160 |
} |
| 161 |
|
| 162 |
bool closeSipWrapper(NEO::Device *neoDevice, uint64_t contextHandle) override { |
| 163 |
return true; |
| 164 |
} |
| 165 |
|
| 166 |
void closeExternalSipHandles() override { |
| 167 |
// Mock implementation - no-op |
| 168 |
} |
| 169 |
|
| 170 |
NEO::TopologyMap topologyMap; |
| 171 |
zet_debug_config_t config; |
| 172 |
bool asyncThreadStarted = false; |
| 173 |
bool closeConnectionCalled = false; |
| 174 |
}; |
| 175 |
|
| 176 |
struct MockDebugSession : public L0::DebugSessionImp { |
| 177 |
|
| 178 |
using L0::DebugSession::allThreads; |
| 179 |
using L0::DebugSession::debugArea; |
| 180 |
using L0::DebugSessionImp::addThreadToNewlyStoppedFromRaisedAttention; |
| 181 |
using L0::DebugSessionImp::allocateStateSaveAreaMemory; |
| 182 |
using L0::DebugSessionImp::apiEvents; |
| 183 |
using L0::DebugSessionImp::applyResumeWa; |
| 184 |
using L0::DebugSessionImp::attentionEventContext; |
| 185 |
using L0::DebugSessionImp::calculateSrMagicOffset; |
| 186 |
using L0::DebugSessionImp::calculateThreadSlotOffset; |
| 187 |
using L0::DebugSessionImp::checkTriggerEventsForAttention; |
| 188 |
using L0::DebugSessionImp::cmdRegisterAccessHelper; |
| 189 |
using L0::DebugSessionImp::dumpDebugSurfaceToFile; |
| 190 |
using L0::DebugSessionImp::fillResumeAndStoppedThreadsFromNewlyStopped; |
| 191 |
using L0::DebugSessionImp::generateEventsAndResumeStoppedThreads; |
| 192 |
using L0::DebugSessionImp::generateEventsForPendingInterrupts; |
| 193 |
using L0::DebugSessionImp::generateEventsForStoppedThreads; |
| 194 |
using L0::DebugSessionImp::getRegisterSize; |
| 195 |
using L0::DebugSessionImp::getSipCommandRegisterValues; |
| 196 |
using L0::DebugSessionImp::getSlmAccessProtocol; |
| 197 |
using L0::DebugSessionImp::getStateSaveAreaHeader; |
| 198 |
using L0::DebugSessionImp::interruptTimeout; |
| 199 |
using L0::DebugSessionImp::isValidNode; |
| 200 |
using L0::DebugSessionImp::newAttentionRaised; |
| 201 |
using L0::DebugSessionImp::pollFifo; |
| 202 |
using L0::DebugSessionImp::readDebugScratchRegisters; |
| 203 |
using L0::DebugSessionImp::readFifo; |
| 204 |
using L0::DebugSessionImp::readModeFlags; |
| 205 |
using L0::DebugSessionImp::readSbaRegisters; |
| 206 |
using L0::DebugSessionImp::readThreadScratchRegisters; |
| 207 |
using L0::DebugSessionImp::registersAccessHelper; |
| 208 |
using L0::DebugSessionImp::resumeAccidentallyStoppedThreads; |
| 209 |
using L0::DebugSessionImp::sendInterrupts; |
| 210 |
using L0::DebugSessionImp::stateSaveAreaMemory; |
| 211 |
using L0::DebugSessionImp::typeToRegsetDesc; |
| 212 |
using L0::DebugSessionImp::updateStoppedThreadsAndCheckTriggerEvents; |
| 213 |
using L0::DebugSessionImp::validateAndSetStateSaveAreaHeader; |
| 214 |
|
| 215 |
using L0::DebugSessionImp::interruptSent; |
| 216 |
using L0::DebugSessionImp::stateSaveAreaHeader; |
| 217 |
using L0::DebugSessionImp::triggerEvents; |
| 218 |
|
| 219 |
using L0::DebugSessionImp::expectedAttentionEvents; |
| 220 |
using L0::DebugSessionImp::fifoPollInterval; |
| 221 |
using L0::DebugSessionImp::interruptMutex; |
| 222 |
using L0::DebugSessionImp::interruptRequests; |
| 223 |
using L0::DebugSessionImp::isValidGpuAddress; |
| 224 |
using L0::DebugSessionImp::lastFifoReadTime; |
| 225 |
using L0::DebugSessionImp::minSlmSipVersion; |
| 226 |
using L0::DebugSessionImp::newlyStoppedThreads; |
| 227 |
using L0::DebugSessionImp::pendingInterrupts; |
| 228 |
using L0::DebugSessionImp::readBarrierMemory; |
| 229 |
using L0::DebugSessionImp::readStateSaveAreaHeader; |
| 230 |
using L0::DebugSessionImp::sipSupportsSlm; |
| 231 |
using L0::DebugSessionImp::SlmAccessProtocol; |
| 232 |
using L0::DebugSessionImp::slmMemoryAccess; |
| 233 |
using L0::DebugSessionImp::slmMemoryReadV2; |
| 234 |
using L0::DebugSessionImp::slmMemoryWriteV2; |
| 235 |
using L0::DebugSessionImp::slmSipVersionCheck; |
| 236 |
using L0::DebugSessionImp::tileAttachEnabled; |
| 237 |
using L0::DebugSessionImp::tileSessions; |
| 238 |
|
| 239 |
MockDebugSession(const zet_debug_config_t &config, L0::Device *device) : MockDebugSession(config, device, true) {} |
| 240 |
|
| 241 |
MockDebugSession(const zet_debug_config_t &config, L0::Device *device, bool rootAttach) : MockDebugSession(config, device, rootAttach, 2) { |
| 242 |
} |
| 243 |
|
| 244 |
MockDebugSession(const zet_debug_config_t &config, L0::Device *device, bool rootAttach, uint32_t stateSaveHeaderVersion) : DebugSessionImp(config, device) { |
| 245 |
if (device) { |
| 246 |
topologyMap = DebugSessionMock::buildMockTopology(device); |
| 247 |
} |
| 248 |
setAttachMode(rootAttach); |
| 249 |
if (rootAttach) { |
| 250 |
createEuThreads(); |
| 251 |
} |
| 252 |
if (stateSaveHeaderVersion == 3) { |
| 253 |
stateSaveAreaHeader = NEO::MockSipData::createStateSaveAreaHeader(3); |
| 254 |
} else { |
| 255 |
stateSaveAreaHeader = NEO::MockSipData::createStateSaveAreaHeader(2); |
| 256 |
} |
| 257 |
} |
| 258 |
~MockDebugSession() override { |
| 259 |
for (auto session : tileSessions) { |
| 260 |
delete session.first; |
| 261 |
} |
| 262 |
} |
| 263 |
|
| 264 |
bool closeConnection() override { return true; } |
| 265 |
ze_result_t initialize() override { |
| 266 |
|
| 267 |
auto numTiles = connectedDevice->getNEODevice()->getNumSubDevices(); |
| 268 |
if (numTiles > 0 && tileAttachEnabled) { |
| 269 |
tileSessions.resize(numTiles); |
| 270 |
zet_debug_config_t config = {}; |
| 271 |
|
| 272 |
for (uint32_t i = 0; i < numTiles; i++) { |
| 273 |
auto subDevice = connectedDevice->getNEODevice()->getSubDevice(i)->getSpecializedDevice<Device>(); |
| 274 |
tileSessions[i] = std::pair<DebugSessionImp *, bool>{new MockDebugSession(config, subDevice), false}; |
| 275 |
} |
| 276 |
tileSessionsEnabled = true; |
| 277 |
} |
| 278 |
|
| 279 |
return ZE_RESULT_SUCCESS; |
| 280 |
} |
| 281 |
|
| 282 |
void attachTile() override { attachTileCalled = true; }; |
| 283 |
void detachTile() override { detachTileCalled = true; }; |
| 284 |
void cleanRootSessionAfterDetach(uint32_t deviceIndex) override { cleanRootSessionDeviceIndices.push_back(deviceIndex); }; |
| 285 |
|
| 286 |
ze_result_t readMemory(ze_device_thread_t thread, const zet_debug_memory_space_desc_t *desc, size_t size, void *buffer) override { |
| 287 |
return readMemoryResult; |
| 288 |
} |
| 289 |
|
| 290 |
ze_result_t writeMemory(ze_device_thread_t thread, const zet_debug_memory_space_desc_t *desc, size_t size, const void *buffer) override { |
| 291 |
return ZE_RESULT_SUCCESS; |
| 292 |
} |
| 293 |
ze_result_t acknowledgeEvent(const zet_debug_event_t *event) override { |
| 294 |
return ZE_RESULT_SUCCESS; |
| 295 |
} |
| 296 |
|
| 297 |
ze_result_t isValidNode(uint64_t vmHandle, uint64_t gpuVa, SIP::fifo_node &node) override { |
| 298 |
if (isValidNodeResult != ZE_RESULT_SUCCESS) { |
| 299 |
return isValidNodeResult; |
| 300 |
} else { |
| 301 |
return DebugSessionImp::isValidNode(vmHandle, gpuVa, node); |
| 302 |
} |
| 303 |
} |
| 304 |
|
| 305 |
ze_result_t readRegistersImp(EuThread::ThreadId thread, uint32_t type, uint32_t start, uint32_t count, void *pRegisterValues) override { |
| 306 |
readRegistersCallCount++; |
| 307 |
readRegistersReg = type; |
| 308 |
|
| 309 |
if (readRegistersSizeToFill > 0 && readRegistersSizeToFill <= sizeof(regs)) { |
| 310 |
memcpy_s(pRegisterValues, readRegistersSizeToFill, regs, readRegistersSizeToFill); |
| 311 |
} |
| 312 |
if (readRegistersResult != ZE_RESULT_FORCE_UINT32) { |
| 313 |
return readRegistersResult; |
| 314 |
} |
| 315 |
return DebugSessionImp::readRegistersImp(thread, type, start, count, pRegisterValues); |
| 316 |
} |
| 317 |
|
| 318 |
ze_result_t writeRegistersImp(EuThread::ThreadId thread, uint32_t type, uint32_t start, uint32_t count, void *pRegisterValues) override { |
| 319 |
writeRegistersCallCount++; |
| 320 |
writeRegistersReg = type; |
| 321 |
|
| 322 |
if (writeRegistersResult != ZE_RESULT_FORCE_UINT32) { |
| 323 |
return writeRegistersResult; |
| 324 |
} |
| 325 |
|
| 326 |
return DebugSessionImp::writeRegistersImp(thread, type, start, count, pRegisterValues); |
| 327 |
} |
| 328 |
|
| 329 |
ze_result_t readSbaBuffer(EuThread::ThreadId threadId, NEO::SbaTrackedAddresses &sbaBuffer) override { |
| 330 |
sbaBuffer = sba; |
| 331 |
return readSbaBufferResult; |
| 332 |
} |
| 333 |
|
| 334 |
void readStateSaveAreaHeader() override { |
| 335 |
readStateSaveAreaHeaderCalled++; |
| 336 |
DebugSessionImp::readStateSaveAreaHeader(); |
| 337 |
} |
| 338 |
ze_result_t resumeImp(const std::vector<EuThread::ThreadId> &threads, uint32_t deviceIndex) override { |
| 339 |
resumeImpCalled++; |
| 340 |
resumeThreadCount = threads.size(); |
| 341 |
resumedThreads.push_back(std::move(threads)); |
| 342 |
resumedDevices.push_back(deviceIndex); |
| 343 |
return resumeImpResult; |
| 344 |
} |
| 345 |
|
| 346 |
ze_result_t interruptImp(uint32_t deviceIndex) override { |
| 347 |
interruptImpCalled++; |
| 348 |
interruptedDevices.push_back(deviceIndex); |
| 349 |
return interruptImpResult; |
| 350 |
} |
| 351 |
|
| 352 |
ze_result_t readGpuMemory(uint64_t memoryHandle, char *output, size_t size, uint64_t gpuVa) override { |
| 353 |
readGpuMemoryCallCount++; |
| 354 |
if (forcereadGpuMemoryFailOnCount == readGpuMemoryCallCount) { |
| 355 |
return ZE_RESULT_ERROR_UNKNOWN; |
| 356 |
} |
| 357 |
if (gpuVa != 0 && gpuVa >= reinterpret_cast<uint64_t>(stateSaveAreaHeader.data()) && |
| 358 |
((gpuVa + size) <= reinterpret_cast<uint64_t>(stateSaveAreaHeader.data() + stateSaveAreaHeader.size()))) { |
| 359 |
[[maybe_unused]] auto offset = ptrDiff(gpuVa, reinterpret_cast<uint64_t>(stateSaveAreaHeader.data())); |
| 360 |
memcpy_s(output, size, reinterpret_cast<void *>(gpuVa), size); |
| 361 |
} |
| 362 |
|
| 363 |
else if (!readMemoryBuffer.empty() && |
| 364 |
(gpuVa != 0 && gpuVa >= reinterpret_cast<uint64_t>(readMemoryBuffer.data()) && |
| 365 |
gpuVa <= reinterpret_cast<uint64_t>(ptrOffset(readMemoryBuffer.data(), readMemoryBuffer.size())))) { |
| 366 |
memcpy_s(output, size, reinterpret_cast<void *>(gpuVa), std::min(size, readMemoryBuffer.size())); |
| 367 |
} |
| 368 |
return readMemoryResult; |
| 369 |
} |
| 370 |
ze_result_t writeGpuMemory(uint64_t memoryHandle, const char *input, size_t size, uint64_t gpuVa) override { |
| 371 |
writeGpuMemoryCallCount++; |
| 372 |
if (forceWriteGpuMemoryFailOnCount == writeGpuMemoryCallCount) { |
| 373 |
return ZE_RESULT_ERROR_UNKNOWN; |
| 374 |
} |
| 375 |
|
| 376 |
if (gpuVa != 0 && gpuVa >= reinterpret_cast<uint64_t>(stateSaveAreaHeader.data()) && |
| 377 |
((gpuVa + size) <= reinterpret_cast<uint64_t>(stateSaveAreaHeader.data() + stateSaveAreaHeader.size()))) { |
| 378 |
[[maybe_unused]] auto offset = ptrDiff(gpuVa, reinterpret_cast<uint64_t>(stateSaveAreaHeader.data())); |
| 379 |
memcpy_s(reinterpret_cast<void *>(gpuVa), size, input, size); |
| 380 |
} |
| 381 |
return writeMemoryResult; |
| 382 |
} |
| 383 |
|
| 384 |
ze_result_t readThreadScratchRegisters(EuThread::ThreadId thread, uint32_t start, uint32_t count, void *pRegisterValues) override { |
| 385 |
|
| 386 |
if (readThreadScratchRegistersResult != ZE_RESULT_FORCE_UINT32) { |
| 387 |
return readThreadScratchRegistersResult; |
| 388 |
} |
| 389 |
return DebugSessionImp::readThreadScratchRegisters(thread, start, count, pRegisterValues); |
| 390 |
} |
| 391 |
|
| 392 |
ze_result_t updateStoppedThreadsAndCheckTriggerEvents(const AttentionEventFields &attention, uint32_t tileIndex, std::vector<EuThread::ThreadId> &threadsWithAttention) override { return ZE_RESULT_SUCCESS; } |
| 393 |
void resumeAccidentallyStoppedThreads(const std::vector<EuThread::ThreadId> &threadIds) override { |
| 394 |
resumeAccidentallyStoppedCalled++; |
| 395 |
return DebugSessionImp::resumeAccidentallyStoppedThreads(threadIds); |
| 396 |
} |
| 397 |
|
| 398 |
void startAsyncThread() override {} |
| 399 |
bool readModuleDebugArea() override { return true; } |
| 400 |
|
| 401 |
bool getRegisterAccessProperties(EuThread::ThreadId *threadId, uint32_t *pCount, zet_debug_regset_properties_t *pRegisterSetProperties) override { |
| 402 |
registerAccessPropertiesCalled = true; |
| 403 |
return true; // simulate successful retrieval for tests that validate SIP path usage |
| 404 |
} |
| 405 |
|
| 406 |
void enqueueApiEvent(zet_debug_event_t &debugEvent) override { |
| 407 |
apiEvents.push(debugEvent); |
| 408 |
apiEventCondition.notify_all(); |
| 409 |
} |
| 410 |
|
| 411 |
bool isAIPequalToThreadStartIP(uint32_t *cr0, uint32_t *dbg0) override { |
| 412 |
if (callBaseisAIPequalToThreadStartIP) { |
| 413 |
return DebugSessionImp::isAIPequalToThreadStartIP(cr0, dbg0); |
| 414 |
} |
| 415 |
return forceAIPEqualStartIP; |
| 416 |
} |
| 417 |
|
| 418 |
bool isForceExceptionOrForceExternalHaltOnlyExceptionReason(uint32_t *cr0) override { |
| 419 |
if (callBaseIsForceExceptionOrForceExternalHaltOnlyExceptionReason) { |
| 420 |
return isForceExceptionOrForceExternalHaltOnlyExceptionReasonBase(cr0); |
| 421 |
} |
| 422 |
return onlyForceException; |
| 423 |
} |
| 424 |
|
| 425 |
bool isForceExceptionOrForceExternalHaltOnlyExceptionReasonBase(uint32_t *cr0) { |
| 426 |
return DebugSessionImp::isForceExceptionOrForceExternalHaltOnlyExceptionReason(cr0); |
| 427 |
} |
| 428 |
|
| 429 |
bool readSystemRoutineIdent(EuThread *thread, uint64_t vmHandle, SIP::sr_ident &srMagic) override { |
| 430 |
if (skipReadSystemRoutineIdent) { |
| 431 |
srMagic.count = threadStopped ? 1 : 0; |
| 432 |
return readSystemRoutineIdentRetVal; |
| 433 |
} |
| 434 |
return DebugSessionImp::readSystemRoutineIdent(thread, vmHandle, srMagic); |
| 435 |
} |
| 436 |
|
| 437 |
bool readSystemRoutineIdentFromMemory(EuThread *thread, const void *stateSaveArea, SIP::sr_ident &srMagic) override { |
| 438 |
if (skipReadSystemRoutineIdent) { |
| 439 |
srMagic.count = threadStopped ? 1 : 0; |
| 440 |
return readSystemRoutineIdentRetVal; |
| 441 |
} |
| 442 |
return DebugSessionImp::readSystemRoutineIdentFromMemory(thread, stateSaveArea, srMagic); |
| 443 |
} |
| 444 |
|
| 445 |
bool areRequestedThreadsStopped(ze_device_thread_t thread) override { |
| 446 |
if (areRequestedThreadsStoppedReturnValue != -1) { |
| 447 |
return areRequestedThreadsStoppedReturnValue != 0; |
| 448 |
} |
| 449 |
return DebugSessionImp::areRequestedThreadsStopped(thread); |
| 450 |
} |
| 451 |
|
| 452 |
bool writeResumeCommand(const std::vector<EuThread::ThreadId> &threadIds) override { |
| 453 |
writeResumeCommandCalled++; |
| 454 |
if (skipWriteResumeCommand) { |
| 455 |
return true; |
| 456 |
} |
| 457 |
return L0::DebugSessionImp::writeResumeCommand(threadIds); |
| 458 |
} |
| 459 |
|
| 460 |
ze_result_t waitForCmdReady(EuThread::ThreadId threadId, uint16_t retryCount) override { |
| 461 |
return DebugSessionImp::waitForCmdReady(threadId, 1); |
| 462 |
} |
| 463 |
|
| 464 |
ze_result_t cmdRegisterAccessHelper(const EuThread::ThreadId &threadId, NEO::SipCommandRegisterValues &command, bool write) override { |
| 465 |
|
| 466 |
ze_result_t status = ZE_RESULT_SUCCESS; |
| 467 |
|
| 468 |
if (slmTesting) { |
| 469 |
uint32_t size = command.sip_commandValues.size * slmSendBytesSize; |
| 470 |
|
| 471 |
// initial wait for ready |
| 472 |
if (!write && slmCmdRegisterCmdvalue == static_cast<uint32_t>(NEO::SipKernel::Command::resume)) { |
| 473 |
if (!memcmp(slmMemory, "FailWaiting", strlen("FailWaiting"))) { |
| 474 |
return ZE_RESULT_FORCE_UINT32; |
| 475 |
} |
| 476 |
command.sip_commandValues.command = static_cast<uint32_t>(NEO::SipKernel::Command::ready); |
| 477 |
status = ZE_RESULT_SUCCESS; |
| 478 |
} else { |
| 479 |
if (write) { // writing to CMD register |
| 480 |
if (forceCmdAccessFail) { |
| 481 |
return ZE_RESULT_FORCE_UINT32; |
| 482 |
} else if (command.sip_commandValues.command == static_cast<uint32_t>(NEO::SipKernel::Command::slmWrite)) { |
| 483 |
memcpy_s(slmMemory + command.sip_commandValues.offset, size, command.sip_commandValues.buffer, size); |
| 484 |
} |
| 485 |
slmCmdRegisterCmdvalue = command.sip_commandValues.command; |
| 486 |
} |
| 487 |
|
| 488 |
if (slmCmdRegisterCmdvalue != static_cast<uint32_t>(NEO::SipKernel::Command::resume)) { |
| 489 |
slmCmdRegisterAccessCount++; |
| 490 |
} |
| 491 |
|
| 492 |
if (slmCmdRegisterAccessCount == slmCmdRegisterAccessReadyCount) { // SIP restores cmd to READY |
| 493 |
|
| 494 |
command.sip_commandValues.command = static_cast<uint32_t>(NEO::SipKernel::Command::ready); |
| 495 |
|
| 496 |
if (slmCmdRegisterCmdvalue == static_cast<uint32_t>(NEO::SipKernel::Command::slmWrite)) { |
| 497 |
slmCmdRegisterAccessCount = 0; |
| 498 |
slmCmdRegisterCmdvalue = static_cast<uint32_t>(NEO::SipKernel::Command::resume); |
| 499 |
} |
| 500 |
} else if (slmCmdRegisterAccessCount == slmCmdRegisterAccessReadyCount + 1) { // Reading the data does an extra call to retrieve the data |
| 501 |
if (!memcmp(slmMemory, "FailReadingData", strlen("FailReadingData"))) { |
| 502 |
status = ZE_RESULT_FORCE_UINT32; |
| 503 |
} else if (slmCmdRegisterCmdvalue == static_cast<uint32_t>(NEO::SipKernel::Command::slmRead)) { |
| 504 |
memcpy_s(command.sip_commandValues.buffer, size, slmMemory + command.sip_commandValues.offset, size); |
| 505 |
command.sip_commandValues.offset = 0; |
| 506 |
command.sip_commandValues.size = 0; |
| 507 |
} |
| 508 |
|
| 509 |
slmCmdRegisterAccessCount = 0; |
| 510 |
slmCmdRegisterCmdvalue = static_cast<uint32_t>(NEO::SipKernel::Command::resume); |
| 511 |
} |
| 512 |
} |
| 513 |
return status; |
| 514 |
} |
| 515 |
|
| 516 |
return L0::DebugSessionImp::cmdRegisterAccessHelper(threadId, command, write); |
| 517 |
} |
| 518 |
|
| 519 |
bool checkThreadIsResumed(const EuThread::ThreadId &threadID) override { |
| 520 |
checkThreadIsResumedCalled++; |
| 521 |
if (skipCheckThreadIsResumed) { |
| 522 |
return true; |
| 523 |
} |
| 524 |
return L0::DebugSessionImp::checkThreadIsResumed(threadID); |
| 525 |
} |
| 526 |
bool checkThreadIsResumed(const EuThread::ThreadId &threadID, const void *stateSaveArea) override { |
| 527 |
checkThreadIsResumedFromPassedSaveAreaCalled++; |
| 528 |
if (skipCheckThreadIsResumed) { |
| 529 |
return true; |
| 530 |
} |
| 531 |
return L0::DebugSessionImp::checkThreadIsResumed(threadID, stateSaveArea); |
| 532 |
} |
| 533 |
|
| 534 |
bool getThreadSipCounter(const void *stateSaveArea, L0::EuThread *thread, const NEO::StateSaveAreaHeader *stateSaveAreaHeader, uint64_t *sipThreadCounter) override { |
| 535 |
if (skipGetThreadSipCounter) { |
| 536 |
if (sipThreadCounter) { |
| 537 |
*sipThreadCounter = mockSipCounter; |
| 538 |
} |
| 539 |
return getThreadSipCounterRetVal; |
| 540 |
} |
| 541 |
return L0::DebugSessionImp::getThreadSipCounter(stateSaveArea, thread, stateSaveAreaHeader, sipThreadCounter); |
| 542 |
} |
| 543 |
|
| 544 |
uint64_t getContextStateSaveAreaGpuVa(uint64_t memoryHandle) override { |
| 545 |
if (returnStateSaveAreaGpuVa && !this->stateSaveAreaHeader.empty()) { |
| 546 |
return reinterpret_cast<uint64_t>(this->stateSaveAreaHeader.data()); |
| 547 |
} |
| 548 |
return 0; |
| 549 |
}; |
| 550 |
|
| 551 |
size_t getContextStateSaveAreaSize(uint64_t memoryHandle) override { |
| 552 |
if (forceStateSaveAreaSize.has_value()) { |
| 553 |
return forceStateSaveAreaSize.value(); |
| 554 |
} |
| 555 |
if (stateSaveAreaHeader.size()) { |
| 556 |
auto header = getStateSaveAreaHeader(); |
| 557 |
auto stateSaveAreaSize = header->regHeader.num_slices * |
| 558 |
header->regHeader.num_subslices_per_slice * |
| 559 |
header->regHeader.num_eus_per_subslice * |
| 560 |
header->regHeader.num_threads_per_eu * |
| 561 |
header->regHeader.state_save_size + |
| 562 |
header->versionHeader.size * 8 + header->regHeader.state_area_offset; |
| 563 |
|
| 564 |
return alignUp(stateSaveAreaSize, MemoryConstants::pageSize); |
| 565 |
} |
| 566 |
|
| 567 |
return 0; |
| 568 |
} |
| 569 |
|
| 570 |
int64_t getTimeDifferenceMilliseconds(std::chrono::high_resolution_clock::time_point time) override { |
| 571 |
if (returnTimeDiff != -1) { |
| 572 |
return returnTimeDiff; |
| 573 |
} |
| 574 |
return L0::DebugSessionImp::getTimeDifferenceMilliseconds(time); |
| 575 |
} |
| 576 |
|
| 577 |
const NEO::TopologyMap &getTopologyMap() override { |
| 578 |
return this->topologyMap; |
| 579 |
} |
| 580 |
|
| 581 |
const SIP::regset_desc *typeToRegsetDesc(const NEO::StateSaveAreaHeader *pStateSaveAreaHeader, uint32_t type, L0::Device *device) { |
| 582 |
return DebugSessionImp::typeToRegsetDesc(pStateSaveAreaHeader, type, device); |
| 583 |
} |
| 584 |
|
| 585 |
void checkStoppedThreadsAndGenerateEvents(const std::vector<EuThread::ThreadId> &threads, uint64_t memoryHandle, uint32_t deviceIndex) override { |
| 586 |
checkStoppedThreadsAndGenerateEventsCallCount++; |
| 587 |
return L0::DebugSessionImp::checkStoppedThreadsAndGenerateEvents(threads, memoryHandle, deviceIndex); |
| 588 |
} |
| 589 |
|
| 590 |
ze_result_t readFifo(uint64_t vmHandle, std::vector<EuThread::ThreadId> &threadsWithAttention) override { |
| 591 |
readFifoCallCount++; |
| 592 |
return L0::DebugSessionImp::readFifo(vmHandle, threadsWithAttention); |
| 593 |
} |
| 594 |
|
| 595 |
NEO::TopologyMap topologyMap; |
| 596 |
|
| 597 |
uint32_t readFifoCallCount = 0; |
| 598 |
uint32_t interruptImpCalled = 0; |
| 599 |
uint32_t resumeImpCalled = 0; |
| 600 |
uint32_t resumeAccidentallyStoppedCalled = 0; |
| 601 |
size_t resumeThreadCount = 0; |
| 602 |
ze_result_t interruptImpResult = ZE_RESULT_SUCCESS; |
| 603 |
ze_result_t resumeImpResult = ZE_RESULT_SUCCESS; |
| 604 |
bool onlyForceException = true; |
| 605 |
bool forceAIPEqualStartIP = false; |
| 606 |
bool callBaseIsForceExceptionOrForceExternalHaltOnlyExceptionReason = false; |
| 607 |
bool callBaseisAIPequalToThreadStartIP = false; |
| 608 |
bool threadStopped = true; |
| 609 |
int areRequestedThreadsStoppedReturnValue = -1; |
| 610 |
bool readSystemRoutineIdentRetVal = true; |
| 611 |
size_t readRegistersSizeToFill = 0; |
| 612 |
ze_result_t readSbaBufferResult = ZE_RESULT_SUCCESS; |
| 613 |
ze_result_t readRegistersResult = ZE_RESULT_FORCE_UINT32; |
| 614 |
ze_result_t readMemoryResult = ZE_RESULT_SUCCESS; |
| 615 |
ze_result_t writeMemoryResult = ZE_RESULT_SUCCESS; |
| 616 |
ze_result_t writeRegistersResult = ZE_RESULT_FORCE_UINT32; |
| 617 |
ze_result_t readThreadScratchRegistersResult = ZE_RESULT_FORCE_UINT32; |
| 618 |
ze_result_t isValidNodeResult = ZE_RESULT_SUCCESS; |
| 619 |
|
| 620 |
uint32_t readStateSaveAreaHeaderCalled = 0; |
| 621 |
uint32_t readRegistersCallCount = 0; |
| 622 |
uint32_t readRegistersReg = 0; |
| 623 |
uint32_t writeRegistersCallCount = 0; |
| 624 |
uint32_t writeRegistersReg = 0; |
| 625 |
|
| 626 |
uint32_t readGpuMemoryCallCount = 0; |
| 627 |
uint32_t forcereadGpuMemoryFailOnCount = 0; |
| 628 |
|
| 629 |
uint32_t writeGpuMemoryCallCount = 0; |
| 630 |
uint32_t forceWriteGpuMemoryFailOnCount = 0; |
| 631 |
|
| 632 |
// Instrumentation flag: set when getRegisterAccessProperties is invoked through SIP external lib path |
| 633 |
bool registerAccessPropertiesCalled = false; |
| 634 |
|
| 635 |
bool skipWriteResumeCommand = true; |
| 636 |
uint32_t writeResumeCommandCalled = 0; |
| 637 |
|
| 638 |
bool skipCheckThreadIsResumed = true; |
| 639 |
uint32_t checkThreadIsResumedCalled = 0; |
| 640 |
uint32_t checkThreadIsResumedFromPassedSaveAreaCalled = 0; |
| 641 |
uint32_t checkStoppedThreadsAndGenerateEventsCallCount = 0; |
| 642 |
|
| 643 |
bool skipReadSystemRoutineIdent = true; |
| 644 |
bool skipGetThreadSipCounter = false; |
| 645 |
bool getThreadSipCounterRetVal = true; |
| 646 |
uint64_t mockSipCounter = 0; |
| 647 |
std::vector<uint32_t> interruptedDevices; |
| 648 |
std::vector<uint32_t> resumedDevices; |
| 649 |
std::vector<std::vector<EuThread::ThreadId>> resumedThreads; |
| 650 |
|
| 651 |
NEO::SbaTrackedAddresses sba; |
| 652 |
std::vector<uint8_t> readMemoryBuffer; |
| 653 |
uint64_t regs[16]; |
| 654 |
|
| 655 |
int64_t returnTimeDiff = -1; |
| 656 |
bool returnStateSaveAreaGpuVa = true; |
| 657 |
std::optional<size_t> forceStateSaveAreaSize = std::nullopt; |
| 658 |
|
| 659 |
bool attachTileCalled = false; |
| 660 |
bool detachTileCalled = false; |
| 661 |
std::vector<uint32_t> cleanRootSessionDeviceIndices; |
| 662 |
|
| 663 |
int slmCmdRegisterAccessCount = 0; |
| 664 |
uint32_t slmCmdRegisterCmdvalue = static_cast<uint32_t>(NEO::SipKernel::Command::resume); |
| 665 |
bool forceCmdAccessFail = false; |
| 666 |
int slmCmdRegisterAccessReadyCount = 2; |
| 667 |
bool slmTesting = false; |
| 668 |
static constexpr int slmSize = 256; |
| 669 |
char slmMemory[slmSize]; |
| 670 |
char originalSlmMemory[slmSize]; |
| 671 |
}; |
| 672 |
|
| 673 |
} // namespace ult |
| 674 |
} // namespace L0 |