Some games load, some don't (NES/FCEUmm Core)

bmn001

Thanks for that. The cache directory is definitely set in the cfg. I even tried changing it temporarily to a persistent dir, and no change.

So next I tried manually entering the same command that runcommand is running on the command line, based on the error log:

/opt/retropie/emulators/retroarch/bin/retroarch -L /opt/retropie/libretrocores/lr-fceumm/fceumm_libretro.so --config /opt/retropie/configs/nes/retroarch.cfg "/home/pi/RetroPie/roms/nes/Ice Climber (USA, Europe).zip" --verbose --appendconfig /dev/shm/retroarch.cfg

This is the output:
http://pastebin.com/C36xN96W

Compare that with the output of a zipped rom that does work properly:

/opt/retropie/emulators/retroarch/bin/retroarch -L /opt/retropie/libretrocores/lr-fceumm/fceumm_libretro.so --config /opt/retropie/configs/nes/retroarch.cfg "/home/pi/RetroPie/roms/nes/Tetris (USA).zip" --verbose --appendconfig /dev/shm/retroarch.cfg

http://pastebin.com/7hwpgJxC

The error I'm seeing is "*** Error in `/opt/retropie/emulators/retroarch/bin/retroarch': free(): invalid pointer: 0x01ff8979 ***"

Is this definitive proof that it's a Retroarch bug? If so, I can file a report with them.

hansolo77

When you compare the two.. the one that works continues right after checking for Environ GET_VARIABLE fceumm_overclocking:. The next step that is failing on the one that doesn't work is Environ GET_VARIABLE fceumm_overscan:. So maybe it's an overscan setting?

bmn001

Good eyes, thank you.

Ok, I just enabled overscan in /boot/config.txt and rebooted.

Didn't do much. Same error, slightly different pointer address:
*** Error in `/opt/retropie/emulators/retroarch/bin/retroarch': free(): invalid pointer: 0x01cb89e9 ***

I'm not sure if this is a retroarch bug or a fceumm core bug.

hansolo77

At least the invalid pointer error is now pointing to a different address..

0x01cb89e9
vs
0x01ff8979

The 2nd to last digit changed :)

bmn001

I submitted this as a Retroarch bug. Hopefully I don't get laughed out of their bug tracker for not knowing what I'm doing.

https://github.com/libretro/RetroArch/issues/4063

synack

I experienced the same bug with lr-fceumm using a no-intro set. I can confirm it's specific to lr-fceumm (lr-nestopia works fine).

One foolproof workaround is use an unzipped rom, or to store (zero compression) the rom in the zip. The commands below will re-zip the files at zero compression (so no extra scraping needs to be done, gamelist tweaking, etc).

As always, make a backup first.

sudo apt-get install zip
cd ~/RetroPie/roms/nes
mkdir old
for f in *.zip; do unzip "$f"; mv "$f" old; zip -0 "${f%.*}".zip "${f%.*}".nes; mv "${f%.*}".nes old; done

The old zip files will be placed in the "old" folder. They can be removed when you're satisfied with the new set.

chromium

Is there a fix for this yet? I just spent ages setting up retropie only to find that most of my nes roms are failing to load with a "failed to extract content" error followed by an "invalid pointer" error. Nothing wrong with the zips as they unzip fine and they worked in my other retropie setups

I'm sure starting again from scratch will fix, but I really don't want to do that. I've spent days on this setup :(

BuZz

@chromium there is a retroarch ticket link above - doesn't look resolved yet.

Unpack your ROMs or use another emulator?

chromium

@BuZz said in Some games load, some don't (NES/FCEUmm Core):

@chromium there is a retroarch ticket link above - doesn't look resolved yet.

Unpack your ROMs or use another emulator?

I'm stubborn and am trying to fix it properly :)

I had this issue with a new Retropie install a couple of days ago and then I updated everything and the problem went away. Updating isn't helping in this one though. I tried updating from binaries and sources for both retroarch and lr-fceumm with no joy.

I actually only get the invalid pointer error from command line. If I run from emulationstation I don't get that, I just get "Aborted."

Eg.

RetroArch [libretro INFO] :: Loading /tmp/retroarch/1942 (Japan, USA).nes...

RetroArch [libretro INFO] :: PRG ROM: 2 x 16KiB
RetroArch [libretro INFO] :: CHR ROM: 1 x 8KiB
RetroArch [libretro INFO] :: ROM CRC32: 0x171251e3
RetroArch [libretro INFO] :: ROM MD5: 0x0b66fdf88964235c434daff62837af60
RetroArch [libretro INFO] :: Mapper #: 0
RetroArch [libretro INFO] :: Mapper name: NROM
RetroArch [libretro INFO] :: Mirroring: Horizontal
RetroArch [libretro INFO] :: Battery-backed: No
RetroArch [libretro INFO] :: Trained: No
RetroArch [libretro INFO] ::
RetroArch [INFO] :: Environ GET_VARIABLE fceumm_palette:
RetroArch [INFO] :: asqrealc
RetroArch [INFO] :: Environ GET_VARIABLE fceumm_nospritelimit:
RetroArch [INFO] :: disabled
RetroArch [INFO] :: Environ GET_VARIABLE fceumm_overclocking:
RetroArch [INFO] :: disabled
/opt/retropie/supplementary/runcommand/runcommand.sh: line 947: 8304 Aborted /opt/retropie/emulators/retroarch/bin/retroarch --verbose -L /opt/retropie/libretrocores/lr-fceumm/fceumm_libretro.so --config /opt/retropie/configs/nes/retroarch.cfg "/home/pi/RetroPie/roms/nes/1942 (Japan, USA).zip" --appendconfig /dev/shm/retroarch.cfg

So it looks like it's unzipping the rom ok because it's reading information from the rom. Pretty weird. Not sure why zipping with no compression works. Everything works ok in nestopia.

BuZz

@chromium also make sure /tmp/retroarch exists (runcommand makes it usually), if starting from command line without using runcommand. That's the default path for unpacking.

PS. Put logs in a code block when posting to forum.

BuZz

I may look into this issue at some point, but it might help to add more info to ticket as they still have it marked down as unconfirmed.

chromium

@BuZz said in Some games load, some don't (NES/FCEUmm Core):

@chromium also make sure /tmp/retroarch exists (runcommand makes it usually), if starting from command line without using runcommand. That's the default path for unpacking.

PS. Put logs in a code block when posting to forum.

I checked and it does get created because I can see the unzipped rom there for working games while they are running and then it disappears once I quit. I'm assuming for games that don't work it gets cleaned up as soon as it fails because if you look at the log I posted before you'll see it actually reads information from the unzipped from in /tmp/retroarch

chromium

This is insane. I tried my other SD card where nes games work in lr-fceumm and I compared the retroarch binaries, the fceumm_libretro.so binaries, the runcommand.sh and all the config files. EVERYTHING was identical. So I have no idea what the problem is. It doesn't seem to be retroarch, lr-fceumm or runcommand, what on earth else can it be?

synack

One other interesting tidbit is that regardless of the zip file used (normally compressed file which fails to load or one using store [0% compression]), they both write the .nes file correctly to /tmp/retroarch -- same md5. It appears that something is continuing to read the original zip file.

chromium

@synack said in Some games load, some don't (NES/FCEUmm Core):

One other interesting tidbit is that regardless of the zip file used (normally compressed file which fails to load or one using store [0% compression]), they both write the .nes file correctly to /tmp/retroarch -- same md5. It appears that something is continuing to read the original zip file.

Yep, this is the bit that I can't understand. The zip file is being unzipped correctly in both cases. I added a comment saying that yesterday in the ticket that was raised. It makes no sense. Also my other fully update retropie setup is running all these roms fine. Same binaries, same kernel. It's madness :) I'll probably use that one, but I have put a lot more work into the one with the lr-fceumm issue. I might just rezip all my roms with 0 compression, but I'd really love to know what this issue is :)

synack

just fyi, the issue has been identified and fixed

https://github.com/libretro/libretro-fceumm/commit/821ca02

BuZz

@synack That didn't fix the free crash on RetroPie (there was a separate issue though also) but I have debugged it and worked out the problem. Updating from source should sort it now. (binaries will be updated in a moment).

synack

@BuZz Thank you. Appreciate the time and effort you put in to solve these type of issues.

chromium

Great work @BuZz!

Out of interest why do some zips work and others don't and why did zipping with 0 compression make the issue go away? Also, why could I not reproduce the issue on my other retropie setup that had the same retroarch and fceumm_libretro.so binaries?

BuZz

@chromium Luck. Well, it was due to some code that after allocating some memory, tried to align a pointer. If the pointer got changed it would fail (since the wrong pointer would be used to free the memory). The code was wrong as the free should have been done on the original pointer value.

That's why when bisecting it, a completely unrelated piece of code in retroarch seemed to be the problem, as it just affected where the memory was allocated, but the real problem was the code that tried to align the memory block.