SDL crashes on LInux ARCH w/Nvidia

Micropython
1

I just built from github master, using SDL2 2.0.12 and get:

> gdb ./micropython
Reading symbols from ./micropython...
(gdb) run
Starting program: /home/src/esp32/lvmp/ports/unix/micropython
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/usr/lib/libthread_db.so.1".
MicroPython v1.9.4-2165-ge795d3765 on 2020-06-20; linux version
Use Ctrl-D to exit, Ctrl-E for paste mode
>>> import SDL
>>> SDL.init()
[New Thread 0x7ffff45e1700 (LWP 574492)]
>>>
Thread 1 "micropython" received signal SIGSEGV, Segmentation fault.
0x00007ffff5cc9e96 in ?? () from /usr/lib/libnvidia-glcore.so.430.64
(gdb) where
#0  0x00007ffff5cc9e96 in ?? () from /usr/lib/libnvidia-glcore.so.430.64
#1  0x00007ffff5cca4d0 in ?? () from /usr/lib/libnvidia-glcore.so.430.64
#2  0x00007ffff5ca2c3e in ?? () from /usr/lib/libnvidia-glcore.so.430.64
#3  0x00007ffff5cb0867 in ?? () from /usr/lib/libnvidia-glcore.so.430.64
#4  0x00007ffff5cb9c7c in ?? () from /usr/lib/libnvidia-glcore.so.430.64
#5  0x00007ffff7cdb73f in ?? () from /usr/lib/libSDL2-2.0.so.0
#6  0x00007ffff7ccf6e1 in ?? () from /usr/lib/libSDL2-2.0.so.0
#7  0x00007ffff7cd498d in ?? () from /usr/lib/libSDL2-2.0.so.0
#8  0x00005555556c09d4 in monitor_sdl_refr_core ()
    at ../../lib/lv_bindings/driver/SDL/SDL_monitor.c:181
#9  0x00005555556c0cb1 in mp_lv_task_handler (arg=0x555555728fb0 <mp_const_none_obj>)
    at ../../lib/lv_bindings/driver/SDL/modSDL.c:17
#10 0x00005555555da246 in fun_builtin_1_call (self_in=0x555555754a20 <mp_lv_task_handler_obj>,
    n_args=1, n_kw=0, args=0x7fffffffd470) at ../../py/objfun.c:70
#11 0x00005555555ceaf4 in mp_call_function_n_kw (fun_in=0x555555754a20 <mp_lv_task_handler_obj>,
    n_args=1, n_kw=0, args=0x7fffffffd470) at ../../py/runtime.c:642
#12 0x00005555555cea39 in mp_call_function_1 (fun=0x555555754a20 <mp_lv_task_handler_obj>,
    arg=0x555555728fb0 <mp_const_none_obj>) at ../../py/runtime.c:620
#13 0x00005555555d0dc7 in mp_call_function_1_protected (
    fun=0x555555754a20 <mp_lv_task_handler_obj>, arg=0x555555728fb0 <mp_const_none_obj>)
    at ../../py/runtime_utils.c:33
#14 0x00005555555d0fb4 in mp_handle_pending_tail (atomic_state=0) at ../../py/scheduler.c:72
#15 0x00005555555d0f28 in mp_handle_pending () at ../../py/scheduler.c:59
#16 0x000055555560389f in mp_hal_stdin_rx_chr () at unix_mphal.c:177
#17 0x00005555556c1a0a in readline (line=0x7fffffffd5d0, prompt=0x5555556d3feb ">>> ")
    at ../../lib/mp-readline/readline.c:425
#18 0x00005555556022cf in do_repl () at main.c:198
#19 0x000055555560333b in main_ (argc=1, argv=0x7fffffffd878) at main.c:651
#20 0x00005555556028d2 in main (argc=1, argv=0x7fffffffd878) at main.c:425

I see the TFT Simulator window being created and after a half-second pause the crash occurs. Any suggestions?

2

It’s a known bug with Nvidia cards: https://github.com/lvgl/lv_binding_micropython/issues/46

The crash happens inside an Nvidia driver somewhere, meaning that we don’t really know why this happens. The current workaround is to use Intel graphics instead (if you have them).

3

Are you using Ubuntu 18.04 by chance?

4

Thanks for the link to the bug. I’m using an Arch linux derivative (Manjaro). Can’t switch to another graphics card 'cause there’s no other.
Is there a way to disable hardware acceleration or make it fall back to some generic driver?

5

Not sure. I’ll see if I can make it work, as I can now reproduce the same issue (I switched to an Nvidia card a few months after that issue was reported).

1 Like
6

It insists on using libnvidia no matter how many software-only APIs I try! The weirdest part is that the standard C-based PC simulator works without problems. I’ll have to compare the SDL-related code and see what’s been changed.

1 Like
7

If you don’t mind, since this is the same issue as https://github.com/lvgl/lv_binding_micropython/issues/46, let’s continue the discussion over there so we have it in one place.

8

Looks like on Ubuntu 22.04 with nvidia-driver-510 on my GeForce GTX 950M this problem has disappeared