L0 Windows Debugger - Implement GPU mem read/write

Signed-off-by: Yates, Brandon <brandon.yates@intel.com>
This commit is contained in:
Yates, Brandon
2022-06-22 23:25:15 +00:00
committed by Compute-Runtime-Automation
parent 153811998f
commit 60b88806d5
4 changed files with 534 additions and 10 deletions

View File

@@ -70,7 +70,7 @@ ze_result_t DebugSessionWindows::initialize() {
}
bool DebugSessionWindows::closeConnection() {
if (debugHandle == 0) {
if (debugHandle == invalidHandle) {
return false;
}
@@ -303,12 +303,94 @@ ze_result_t DebugSessionWindows::readEvent(uint64_t timeout, zet_debug_event_t *
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE;
}
ze_result_t DebugSessionWindows::readMemory(ze_device_thread_t thread, const zet_debug_memory_space_desc_t *desc, size_t size, void *buffer) {
ze_result_t DebugSessionWindows::readElfSpace(const zet_debug_memory_space_desc_t *desc, size_t size, void *buffer) {
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE;
}
bool DebugSessionWindows::isVAElf(const zet_debug_memory_space_desc_t *desc, size_t size) {
std::unique_lock<std::mutex> lock(asyncThreadMutex);
for (auto elf : allElfs) {
if (desc->address >= elf.startVA && desc->address <= elf.endVA) {
if (desc->address + size > elf.endVA) {
return false;
}
return true;
}
}
return false;
}
ze_result_t DebugSessionWindows::readMemory(ze_device_thread_t thread, const zet_debug_memory_space_desc_t *desc, size_t size, void *buffer) {
if (debugHandle == invalidHandle) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
if (!isValidGpuAddress(desc->address)) {
return ZE_RESULT_ERROR_INVALID_ARGUMENT;
}
ze_result_t status = ZE_RESULT_ERROR_UNINITIALIZED;
status = sanityMemAccessThreadCheck(thread, desc);
if (status != ZE_RESULT_SUCCESS) {
return status;
}
if (isVAElf(desc, size)) {
return readElfSpace(desc, size, buffer);
}
uint64_t memoryHandle = DebugSessionWindows::invalidHandle;
if (DebugSession::isThreadAll(thread)) {
std::unique_lock<std::mutex> lock(asyncThreadMutex);
if (allContexts.empty()) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
memoryHandle = *allContexts.begin();
} else {
auto threadId = convertToThreadId(thread);
memoryHandle = allThreads[threadId]->getMemoryHandle();
if (memoryHandle == EuThread::invalidHandle) {
return ZE_RESULT_ERROR_NOT_AVAILABLE;
}
}
return readGpuMemory(memoryHandle, static_cast<char *>(buffer), size, desc->address);
}
ze_result_t DebugSessionWindows::writeMemory(ze_device_thread_t thread, const zet_debug_memory_space_desc_t *desc, size_t size, const void *buffer) {
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE;
if (debugHandle == invalidHandle) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
if (!isValidGpuAddress(desc->address)) {
return ZE_RESULT_ERROR_INVALID_ARGUMENT;
}
ze_result_t status = ZE_RESULT_ERROR_UNINITIALIZED;
status = sanityMemAccessThreadCheck(thread, desc);
if (status != ZE_RESULT_SUCCESS) {
return status;
}
uint64_t memoryHandle = DebugSessionWindows::invalidHandle;
if (DebugSession::isThreadAll(thread)) {
std::unique_lock<std::mutex> lock(asyncThreadMutex);
if (allContexts.empty()) {
return ZE_RESULT_ERROR_UNINITIALIZED;
}
memoryHandle = *allContexts.begin();
} else {
auto threadId = convertToThreadId(thread);
memoryHandle = allThreads[threadId]->getMemoryHandle();
if (memoryHandle == EuThread::invalidHandle) {
return ZE_RESULT_ERROR_NOT_AVAILABLE;
}
}
return writeGpuMemory(memoryHandle, static_cast<const char *>(buffer), size, desc->address);
}
ze_result_t DebugSessionWindows::acknowledgeEvent(const zet_debug_event_t *event) {
@@ -324,11 +406,43 @@ ze_result_t DebugSessionWindows::interruptImp(uint32_t deviceIndex) {
}
ze_result_t DebugSessionWindows::readGpuMemory(uint64_t memoryHandle, char *output, size_t size, uint64_t gpuVa) {
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE;
auto gmmHelper = connectedDevice->getNEODevice()->getGmmHelper();
gpuVa = gmmHelper->decanonize(gpuVa);
KM_ESCAPE_INFO escapeInfo = {0};
escapeInfo.KmEuDbgL0EscapeInfo.EscapeActionType = DBGUMD_ACTION_READ_GFX_MEMORY;
escapeInfo.KmEuDbgL0EscapeInfo.ReadGfxMemoryParams.hContextHandle = memoryHandle;
escapeInfo.KmEuDbgL0EscapeInfo.ReadGfxMemoryParams.GpuVirtualAddr = gpuVa;
escapeInfo.KmEuDbgL0EscapeInfo.ReadGfxMemoryParams.MemoryBufferSize = static_cast<uint32_t>(size);
escapeInfo.KmEuDbgL0EscapeInfo.ReadGfxMemoryParams.MemoryBufferPtr = reinterpret_cast<uint64_t>(output);
auto status = runEscape(escapeInfo);
if (STATUS_SUCCESS != status || DBGUMD_RETURN_ESCAPE_SUCCESS != escapeInfo.KmEuDbgL0EscapeInfo.EscapeReturnStatus) {
PRINT_DEBUGGER_ERROR_LOG("DBGUMD_ACTION_READ_GFX_MEMORY: Failed - ProcessId: %d Status: %d EscapeReturnStatus: %d\n", processId, status, escapeInfo.KmEuDbgL0EscapeInfo.EscapeReturnStatus);
return DebugSessionWindows::translateEscapeReturnStatusToZeResult(escapeInfo.KmEuDbgL0EscapeInfo.EscapeReturnStatus);
}
return ZE_RESULT_SUCCESS;
}
ze_result_t DebugSessionWindows::writeGpuMemory(uint64_t memoryHandle, const char *input, size_t size, uint64_t gpuVa) {
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE;
auto gmmHelper = connectedDevice->getNEODevice()->getGmmHelper();
gpuVa = gmmHelper->decanonize(gpuVa);
KM_ESCAPE_INFO escapeInfo = {0};
escapeInfo.KmEuDbgL0EscapeInfo.EscapeActionType = DBGUMD_ACTION_WRITE_GFX_MEMORY;
escapeInfo.KmEuDbgL0EscapeInfo.ReadGfxMemoryParams.hContextHandle = memoryHandle;
escapeInfo.KmEuDbgL0EscapeInfo.ReadGfxMemoryParams.GpuVirtualAddr = gpuVa;
escapeInfo.KmEuDbgL0EscapeInfo.ReadGfxMemoryParams.MemoryBufferSize = static_cast<uint32_t>(size);
escapeInfo.KmEuDbgL0EscapeInfo.ReadGfxMemoryParams.MemoryBufferPtr = reinterpret_cast<uint64_t>(input);
auto status = runEscape(escapeInfo);
if (STATUS_SUCCESS != status || DBGUMD_RETURN_ESCAPE_SUCCESS != escapeInfo.KmEuDbgL0EscapeInfo.EscapeReturnStatus) {
PRINT_DEBUGGER_ERROR_LOG("DBGUMD_ACTION_READ_GFX_MEMORY: Failed - ProcessId: %d Status: %d EscapeReturnStatus: %d\n", processId, status, escapeInfo.KmEuDbgL0EscapeInfo.EscapeReturnStatus);
return DebugSessionWindows::translateEscapeReturnStatusToZeResult(escapeInfo.KmEuDbgL0EscapeInfo.EscapeReturnStatus);
}
return ZE_RESULT_SUCCESS;
}
void DebugSessionWindows::enqueueApiEvent(zet_debug_event_t &debugEvent) {