Several Cores Crash When input_max_users Set < 5

[Awakened0]

I'm getting a crash with this core when loading content on today's nightly (10/22). It doesn't happen in the previous one available on the buildbot (10/20). A bt in gdb gives me:

Program received signal SIGSEGV, Segmentation fault.
0x00000000004f7789 in dinput_is_pressed (di=0x4f6f110, binds=0x0, port=2, id=7)
at input/drivers/dinput.c:282
282 input/drivers/dinput.c: No such file or directory.
(gdb) bt

0 0x00000000004f7789 in dinput_is_pressed (di=0x4f6f110, binds=0x0, port=2, id=7)

at input/drivers/dinput.c:282

1 0x00000000004f7e4e in dinput_input_state (data=0x4f6f110,

binds=0x96de80 <libretro_input_binds>, port=2, device=1, idx=0, id=7)
at input/drivers/dinput.c:489

2 0x000000000042a74e in input_state (port=2, device=1, idx=0, id=7) at input/input_driver.c:464

3 0x0000000000402443 in core_input_state_poll (port=2, device=1, idx=0, id=7) at core_impl.c:73

4 0x000000001224472a in inputGetKeys_default ()

from F:\Blue\Emulation\Emulators\RetroArch\cores\glupen64_libretro.dll

5 0x0000000012244dc4 in egcvip_get_input ()

from F:\Blue\Emulation\Emulators\RetroArch\cores\glupen64_libretro.dll

6 0x0000000012205315 in read_controller ()

from F:\Blue\Emulation\Emulators\RetroArch\cores\glupen64_libretro.dll

7 0x0000000012205b8d in update_pif_read ()

from F:\Blue\Emulation\Emulators\RetroArch\cores\glupen64_libretro.dll

8 0x0000000012205f25 in write_si_regs ()

from F:\Blue\Emulation\Emulators\RetroArch\cores\glupen64_libretro.dll

9 0x0000000010678b7a in ?? ()

Backtrace stopped: previous frame inner to this frame (corrupt stack?)

[Awakened0]

Found out this happens with the Mupen core too. I've found Banjo Kazooie works fine in both cores for some reason. Here's a backtrace:

Program received signal SIGSEGV, Segmentation fault.
0x00000000004f7ec9 in dinput_is_pressed (di=0x5ef8100, binds=0x0, port=2, id=7)
at input/drivers/dinput.c:282
282 input/drivers/dinput.c: No such file or directory.
(gdb) bt
#0 0x00000000004f7ec9 in dinput_is_pressed (di=0x5ef8100, binds=0x0, port=2, id=7)
at input/drivers/dinput.c:282
#1 0x00000000004f85a6 in dinput_input_state (data=0x5ef8100,
binds=0x96ef00 <libretro_input_binds>, port=2, device=1, idx=0, id=7)
at input/drivers/dinput.c:489
#2 0x000000000042a503 in input_state (port=2, device=1, idx=0, id=7) at input/input_driver.c:457
#3 0x0000000000402443 in core_input_state_poll (port=2, device=1, idx=0, id=7) at core_impl.c:73
#4 0x00000000179f596a in inputGetKeys_default (Control=2, Keys=0x14b16050)
at mupen64plus-core/src/plugin/emulate_game_controller_via_libretro.c:710
#5 0x00000000179f7814 in egcvip_get_input (opaque=)
at mupen64plus-core/src/plugin/emulate_game_controller_via_libretro.c:1040
#6 0x000000001799f53e in game_controller_get_input (cont=0x183e2c00 <g_si+384>)
at mupen64plus-core/src/si/game_controller.c:185
#7 read_controller_read_buttons (
cmd=0x183e2ab1 <g_si+49> "\001\004\001\377\377\377\377\377\001\004\001\377\377\377\377_",
cont=0x183e2c00 <g_si+384>) at mupen64plus-core/src/si/game_controller.c:58
#8 read_controller (cont=cont@entry=0x183e2c00 <g_si+384>,
cmd=cmd@entry=0x183e2ab1 <g_si+49> "\001\004\001\377\377\377\377\377\001\004\001\377\377\377\377_") at mupen64plus-core/src/si/game_controller.c:214
#9 0x000000001799ecca in update_pif_read (si=0x183e2a80 <g_si>)
at mupen64plus-core/src/si/pif.c:225
#10 0x00000000179b6675 in dma_si_read (si=0x183e2a80 <g_si>)
at mupen64plus-core/src/si/si_controller.c:68
#11 write_si_regs (opaque=0x183e2a80 <g_si>, address=, value=,
mask=) at mupen64plus-core/src/si/si_controller.c:122
#12 0x000000002354dccf in ?? ()
Backtrace stopped: previous frame inner to this frame (corrupt stack?)

[Awakened0]

Ok, I should have tested multiple cores first. I just tested glupen not working and desmume working and thought it was something with glupen, but several cores are crashing.

Working cores: mGBA, mednafen-wonderswan, Desmume, gambatte, Picodrive, Genesis Plus GX, mednafen-saturn, mednafen-psx, mednafen-ngp, PPSSPP, BSNES-Mercury Balanced.
Non-working cores: Glupen64, Mupen64, MAME, mednafen-supergrafx, Nestopia.

Edit: Here are backtraces for the last three crashing cores:
http://hastebin.com/qevacuduso.swift
http://pastebin.com/ZY2zp496

[andres-asm]

Have you tried to bisect RetroArch? Or at least tested with 1.3.6?

[andres-asm]

I think it's happening in reicast too

[Awakened0]

I don't have the stuff to bisect, but at least it seems to have started somewhere after a build from the 20th.

[andres-asm]

I just tested up-to-date mupen and RetroArch and I couldn't reproduce
Edit: ditto for Glupen64

Are you running insider builds?
I had some gamepad related breakage recently (steam crashing, xbone pad not connecting, etc)

[Awakened0]

No, I'm on stable anniversary with all the latest cumulative updates. No problems with my controller or Steam.

[Awakened0]

Ok, seems setting input_max_users to anything less than 5 results in a crash for certain cores.

[hizzlekizzle]

Someone reported this same issue on Vita a few days ago. I figured it was Vita-specific, but evidently not.

[inactive123]

I was able to reproduce this -

8ce882c

From this point on, it no longer crashes for me on Linux with Mupen64plus when I set max users to 1.

Try to see if it fixes it on all other platforms as well.

[lasers]

Confirmed! I apparently have this problem. I can't play a NES game (Nestopia) with anything below 4 on input-max_users... on Archlinux. I assume my config was awful. I had it at 1.

Not related. If you set input_max_users to 20~ or above, it'll crash.

Necessary 4 for Nestopia.
Necessary 5 for Mednafen SuperGrafx.
Necessary 2 for BSNES.

[inactive123]

Instead of saying 'confirmed', at least test if it happens with the latest sources.

[lasers]

Noted. This happened with the latest sources.

[inactive123]

Which input driver/joypad driver are you using, and give me an exact scenario to reproduce this.

[lasers]

lasers@z420 ~ $ retroarch --version
RetroArch: Frontend for libretro -- v1.3.6 -- 4ec356d --
Compiler: GCC (6.2.1) 64-bitBuilt: Oct 25 2016

I start with input_max_users = "1" before starting in a custom.cfg

I start with no retroarch.cfg.
retroarch --menu --verbose --appendconfig ~/.config/retroarch/config/retroarch.cfg.custom

I didn't explicitly set any input/joypad driver. It is... udev and udev

Navigate to start a random game or pick something from history. I picked something from history. If it didn't start, then that's it.

Repeat with a new higher number until the game eventually starts.

[inactive123]

Which core do you get this crash with? Be specific. I can't reproduce these issues anymore on my end with udev/udev drivers.

[lasers]

All cores came from Online Updaters.
Necessary max_input 4 for Nestopia.
Necessary max_input 5 for Mednafen SuperGrafx.
Necessary max_input 2 for BSNES.

I will try the cores that I recently got from ./libretro-install.sh right now.

[inactive123]

Cannot reproduce with Nestopia anymore.

I think you don't have the latest nightly built from source, it simply cannot happen anymore after my recent commits.

[lasers]

I install retroarch-git again from time to time. Compiling latest cores right now with libretro-super. Is this an issue with RetroArch or the cores? I can try and install RetroArch w/ libretro-super too if you prefer that instead.

[inactive123]

It has nothing to do with the cores.

Look, you are not being specific here again. Are you compiling it directly from source or not? It's a very simple question really. You are only adding to the confusion here by not being directly specific here.

[lasers]

I thought using retroarch-git was the latest source. It looks like it might not be. I tested it with aur/retroarch-git. I'm diving into libretro-super for first time... and am currently compiling the cores. I will compile RetroArch next... then I can tell you if it works fine now.

[inactive123]

Your log above shows you were on an old build.

4ec356d

[lasers]

YES.

chris@z420 ~ $ pa -Ss retroarch-git
aur/retroarch-git 1.3.6.r1588.2c0d8e8-1 (73, 1.41) [installed: 1.3.6.r2186.4ec356d-1]
    Reference frontend for the libretro API (Git-latest)

NOW.

chris@z420 ~ $ pa -Ss retroarch-git
aur/retroarch-git 1.3.6.r1588.2c0d8e8-1 (73, 1.41) [installed: 1.3.6.r2234.07421c3-1]
    Reference frontend for the libretro API (Git-latest)

Necessary max_input 4 for Nestopia. // Works with 1 👍
Necessary max_input 5 for Mednafen SuperGrafx. // Works with 1 👍
Necessary max_input 2 for BSNES. // Works with 1 👍

PROBLEM SOLVED? 👍

I notice this different issue just now... Try typing 'x' in search prompt. It comes up x1.

[inactive123]

The x/z issue is known for now. It will require several input code changes in order to fix that.

Anyway, that is a different issue, open a different issue for that if one doesn't already exist.

[lasers]

Sure! I will file it under the "z" issue I made. Sorry I never knew about x.

[Awakened0]

On windows with the latest buildbot build (b1d1875) SuperGrafx and Nestopia now work with input_max_users = "2", but Glupen64 and MAME 2015 are still crashing.

MAME 2015 backtrace:

Program received signal SIGSEGV, Segmentation fault.
0x00000000004f5c57 in dinput_pressed_analog (di=0xd110680, binds=0x0, idx=0, id=0)
at input/drivers/dinput.c:303
303 input/drivers/dinput.c: No such file or directory.
(gdb) bt
#0 0x00000000004f5c57 in dinput_pressed_analog (di=0xd110680, binds=0x0, idx=0, id=0)
at input/drivers/dinput.c:303
#1 0x00000000004f62c1 in dinput_input_state (data=0xd110680,
binds=0x968f00 <libretro_input_binds>, port=2, device=5, idx=0, id=0)
at input/drivers/dinput.c:496
#2 0x000000000042afa3 in input_state (port=2, device=5, idx=0, id=0) at input/input_driver.c:457
#3 0x0000000000402443 in core_input_state_poll (port=2, device=5, idx=0, id=0) at core_impl.c:73
#4 0x0000000013162914 in retro_set_controller_port_device ()
from F:\Blue\Emulation\Emulators\RetroArch\cores\mame_libretro.dll
#5 0x0000000013162ac3 in retro_run ()
from F:\Blue\Emulation\Emulators\RetroArch\cores\mame_libretro.dll
#6 0x0000000000402c1f in core_run () at core_impl.c:406
#7 0x0000000000415891 in runloop_iterate (sleep_ms=0x2dafd90) at runloop.c:1234
#8 0x00000000004016e7 in rarch_main (argc=3, argv=0x3b51370, data=0x0) at frontend/frontend.c:127
#9 0x0000000000401755 in SDL_main (argc=3, argv=0x3b51370) at frontend/frontend.c:145
#10 0x00000000006492da in console_main ()
#11 0x0000000000649395 in WinMain ()
#12 0x00000000004013e8 in __tmainCRTStartup ()
at C:/repo/mingw-w64-crt-git/src/mingw-w64/mingw-w64-crt/crt/crtexe.c:332
#13 0x00000000004014eb in WinMainCRTStartup ()
at C:/repo/mingw-w64-crt-git/src/mingw-w64/mingw-w64-crt/crt/crtexe.c:184

Glupen64 backtrace:

Program received signal SIGSEGV, Segmentation fault.
0x00000000004f5c57 in dinput_pressed_analog (di=0x54b0570, binds=0x0, idx=1, id=0)
at input/drivers/dinput.c:303
303 input/drivers/dinput.c: No such file or directory.
(gdb) bt
#0 0x00000000004f5c57 in dinput_pressed_analog (di=0x54b0570, binds=0x0, idx=1, id=0)
at input/drivers/dinput.c:303
#1 0x00000000004f62c1 in dinput_input_state (data=0x54b0570,
binds=0x968f00 <libretro_input_binds>, port=2, device=5, idx=1, id=0)
at input/drivers/dinput.c:496
#2 0x000000000042afa3 in input_state (port=2, device=5, idx=1, id=0) at input/input_driver.c:457
#3 0x0000000000402443 in core_input_state_poll (port=2, device=5, idx=1, id=0) at core_impl.c:73
#4 0x000000001292494c in inputGetKeys_default ()
from F:\Blue\Emulation\Emulators\RetroArch\cores\glupen64_libretro.dll
#5 0x0000000012924e14 in egcvip_get_input ()
from F:\Blue\Emulation\Emulators\RetroArch\cores\glupen64_libretro.dll
#6 0x00000000128e5365 in read_controller ()
from F:\Blue\Emulation\Emulators\RetroArch\cores\glupen64_libretro.dll
#7 0x00000000128e5bdd in update_pif_read ()
from F:\Blue\Emulation\Emulators\RetroArch\cores\glupen64_libretro.dll
#8 0x00000000128e5f75 in write_si_regs ()
from F:\Blue\Emulation\Emulators\RetroArch\cores\glupen64_libretro.dll
#9 0x000000001bb8dccf in ?? ()
Backtrace stopped: previous frame inner to this frame (corrupt stack?)

They will both work if I set max_users to at least 4 now.

[inactive123]

Test if this fixes it -

26547be

[Awakened0]

Everything works fine with it set to 2 or 1 now. Thanks!