advmame 3.9 crash on exit

gomisensei

@jmbooth2000 said in advmame 3.9 crash on exit:

Just checking to see if there is any update? The problem and work around is still there.

Any idea what the effect on performance will be of setting misc_smp to no? A quick check with one game showed no real difference.

I'm very glad I found as manually editing the config file was a huge pain when trying to setup a new controller.

Thanks
Jonathan

Turning off SMP does cause a bit of a performance hit.

you can edit your ~/.advance.rc and temporarily change it to"misc_smp no", add "config save _at_exit" , then run and do your config. After you confirm your changes have been written to config file, turn smp back on, and remove save_at_exit.

joyrider3774

i just ran into this same problem but left misc_smp off as i'm not sure if the error occurs if nvram and highscores are saved as well. did anyone verify that ?

andrewsi

I have certainly seen NVRAM and high scores be updated with SMP on, so it seems like it's the config file writing that somehow seems to be implicated in the crash. On the Pi4 turning SMP off seems like a non-starter, performance is severely impacted, evidently due to the emulated frame buffer. The official Advmame page recommends turning off the DRM/KMS driver to get a real frame buffer, but then EmulationStation doesn't run, so it's a little bit of a no-win situation at the moment.

grant2258

@andrewsi said in advmame 3.9 crash on exit:

rter, performance is severely impacted, evidently due to the emulated frame buffer. The official Advmame page recommends turning off the DRM/KMS driver to get a

compile this with debugging info and asan it will give you fault location

andrewsi

@grant2258 Well, I used ./configure --include-debug, rebuilt, and repro'd it, but not sure the info is all that useful:

Executing: /opt/retropie/emulators/advmame/bin/advmame "arkanoid"
Signal SIGSEGV[fffffffah], fault at 0x324, from code at (nil)
Compiled Jun 14 2020, 18:48:58

grant2258

well all i can suggest is add -fsanitize=address,undefined CFLAGS and LDFLAGS here https://github.com/amadvance/advancemame/blob/95a36b507aaddbc7f71691a292919a1edc1af495/configure.ac#L168 that should give you and indication of whats going wrong

andrewsi

I spent a bunch of time trying to get it to build correctly with --enable-debug AND with the configure.ac edited with -fsanitize=address,undefined in the CFLAGS and LFLAGS, but I have not been successful. I wind up with a huge pile of BFD assertion failures during link, and one unresolved reference that keep the build from completing:

/usr/bin/ld: obj/cfg/linux/debug/blit/blit.o: in function interp_16_dist3': /home/pi/RetroPie-Setup/tmp/build/advmame/advance/blit/interp.h:494: undefined reference to interp_16_dist'
/usr/bin/ld: /home/pi/RetroPie-Setup/tmp/build/advmame/advance/blit/interp.h:494: undefined reference to `interp_16_dist'
collect2: error: ld returned 1 exit status
make: *** [advance/cfg.mak:269: obj/cfg/linux/debug/advcfg] Error 1
make: *** Waiting for unfinished jobs....
/usr/bin/ld: BFD (GNU Binutils for Raspbian) 2.31.1 assertion fail ../../bfd/elf32-arm.c:9853
/usr/bin/ld: BFD (GNU Binutils for Raspbian) 2.31.1 assertion fail ../../bfd/elf32-arm.c:9853
/usr/bin/ld: BFD (GNU Binutils for Raspbian) 2.31.1 assertion fail ../../bfd/elf32-arm.c:9853
...repeat a few hundred times...

Neither of these occurs with a standard build.

andrewsi

Actually, I retract that. I did get it compiled with these options, but the data returned after the crash was equally unhelpful. The address changed but still no symbols, source location, or anything useful.

joyrider3774

Somthing corrupting the stack then or error happening in a lib?

andrewsi

@joyrider3774 Well, must be pretty bad since I can’t get any kind of location or call stack. Unfortunately I’m much better at debugging in Windows than on GDB. In did try to get it to break into GDB but between the video not being in a text mode when the error occurs and the lack of usable info on the segfault I’m out of my depth. Since it’s easy to repro, I’m hoping somebody more skilled will take up the charge.

joyrider3774

@andrewsi starting it from gdb inside a ssh session could work to have text mode, unless you tried that. But could be it wont break on correct location either then. Not used gdb that much for such problems