Link Copied
Hello! Thank you for your repro attempt. I have received the traceback. In terms of my code, it is similar to the implementation of glfw and glad. For the initialization of the render context, I used the documentation from Windows and Khronos. You can find my source code and a README.txt file that explains the notable parts. GLFW simply calls SwapBuffers and doesn't do anything else (similar to my implementation).
// glfw/src/wgl_context.c
static void swapBuffersWGL(_GLFWwindow* window)
{
if (!window->monitor)
{
// HACK: Use DwmFlush when desktop composition is enabled on Windows Vista and 7
if (!IsWindows8OrGreater() && IsWindowsVistaOrGreater())
{
BOOL enabled = FALSE;
if (SUCCEEDED(DwmIsCompositionEnabled(&enabled)) && enabled)
{
int count = abs(window->context.wgl.interval);
while (count--)
DwmFlush();
}
}
}
SwapBuffers(window->context.wgl.dc);
}
By the way, at that time, my thread was stopped while calling `NtSetEvent`.
My thread:
00007FFE8CF9F8C8 nop dword ptr [rax+rax*1], eax
00007FFE8CF9F8D0 mov r10, rcx
00007FFE8CF9F8D3 mov eax, 0xe // NOTE: 0x000e = NtSetEvent: https://j00ru.vexillium.org/syscalls/nt/64/
00007FFE8CF9F8D8 test byte ptr [0x7ffe0308], 0x1
00007FFE8CF9F8E0 jnz 0x7ffe8cf9f8e5
00007FFE8CF9F8E2 syscall
00007FFE8CF9F8E4 ret
Here is a stacktrace:
And it keeps crashing at the same point and at the same time. I tried four out of five times, and the counter reached 98230 before the crash happened. On the third attempt, it was 76733 (perhaps the fact that I was not focused on the window had something to do with the result).
Here is code with counter:
int x = 0;
while (!glfwWindowShouldClose(window))
{
if (glfwGetKey(window, GLFW_KEY_ESCAPE) == GLFW_PRESS)
glfwSetWindowShouldClose(window, true);
glClear(GL_COLOR_BUFFER_BIT);
glClearColor(0, 0, 0, 1);
x += 1;
printf("x = (%d) win32_error=(%d) gl_error=(%d)\n", x, GetLastError(), glGetError());
// glfwSwapBuffers(window);
glfwPollEvents();
}
And here is where driver's thread segfaults:
00007FFDEB182A1E add rax, 0x8
00007FFDEB182A22 jmp 0x7ffdeb182a0a
00007FFDEB182A24 mov eax, dword ptr [rcx+0x18]
00007FFDEB182A27 mov r9, rcx
00007FFDEB182A2A mov rcx, qword ptr [rcx+0x8]
00007FFDEB182A2E test rcx, rcx
00007FFDEB182A31 jz 0x7ffdeb182a4f
00007FFDEB182A33 lea r10, ptr [rcx+rax*1]
00007FFDEB182A37 cmp r10, 0x1000
00007FFDEB182A3E jbe 0x7ffdeb182a4a
00007FFDEB182A40 test rdx, rdx
00007FFDEB182A43 jz 0x7ffdeb182a4a
00007FFDEB182A45 mov ecx, dword ptr [rdx]
00007FFDEB182A47 mov dword ptr [r10], ecx // NOTE: Here is segfault
00007FFDEB182A4A add dword ptr [r9+0x18], 0x4
00007FFDEB182A4F ret
Registers:
Stacktrace:
I still believe that this issue is not only a problem with the game engine, and it should be addressed by the driver development team. Perhaps my code is causing this issue, but it's still strange that it's happening without any error codes.
It would be great if you could try to reproduce using my sources. I am still suspect race-condition of synchronization. After what I saw was called in glGetError implementation of `igxelpicd64.dll`.
---
Best regard,
Ilya
---
P.S. Sorry for the off-topic, but in terms of the reply's user interface, the previous forum board was much better. I was unable to format the code snippets properly. Come on, Intel is a multi-billion corporation. Why is the web UI such a mess? At this point, I am ready to throw the markdown files with the replies.
