Occasional segfaults

[kitchen-ace]

Using Arch Linux 64-bit, dsda-doom compiled locally, either stable or git.

Occasionally dsda-doom will segfault for no apparent reason. I use it most to play/record demos, and it can crash when playing a demo but then trying to play that demo again will work fine. I don't have a good sense of how frequently it happens but when playing multi-hour demos a crash is pretty much guaranteed.

I made a debug build and ran it with gdb, I don't know if there's a way to get more info or if this is helpful.

Thread 50 "dsda-doo:gdrv0" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fff977fc6c0 (LWP 1916)]
0x00007fffa7235b8d in ?? () from /usr/lib/libgallium-25.1.7-arch1.1.so

(gdb) bt
#0  0x00007fffa7235b8d in ?? () from /usr/lib/libgallium-25.1.7-arch1.1.so
#1  0x00007fffa6750799 in ?? () from /usr/lib/libgallium-25.1.7-arch1.1.so
#2  0x00007fffa651a795 in ?? () from /usr/lib/libgallium-25.1.7-arch1.1.so
#3  0x00007fffa656000d in ?? () from /usr/lib/libgallium-25.1.7-arch1.1.so
#4  0x00007ffff74969cb in ?? () from /usr/lib/libc.so.6
#5  0x00007ffff751aa0c in ?? () from /usr/lib/libc.so.6

(gdb) p $_siginfo._sifields._sigfault.si_addr
$1 = (void *) 0x0

[Pedro-Beirao]

That log does not provide any meaningful info.
Could you make a build of master with -DDSDA_ENABLE_ASAN=true and try to get a crash?
This will output a readable trace.

There was an issue where demo playback could crash, but that only seemed to affect macOS and is already fixed on master.

[rfomin]

I have a theory that misalignment in the saving code can cause random crashes. Please test this PR: #733, which might fix the issue.

[kitchen-ace]

The problem is related to GL rendering. I ran Coincident's 6-hour Okuplok demo overnight on software rendering without issue. I am wondering if it's some kind of system instability on my end :/

With GL rendering, using -DDSDA_ENABLE_ASAN=true did not give me any more information with a crash. These lines came up while running CMAKE but I assume they're about something else.

-- Performing Test HAVE_ASAN
-- Performing Test HAVE_ASAN - Failed
-- Performing Test HAVE_UBSAN
-- Performing Test HAVE_UBSAN - Failed
-- Performing Test HAVE_TSAN
-- Performing Test HAVE_TSAN - Failed

@rfomin 's linked PR also did not fix the issue.

[rfomin]

These lines came up while running CMAKE but I assume they're about something else.

These lines mean that ASAN was not found. What compiler are you using? ASAN should print a report on any crash.

[kitchen-ace]

Oh my mistake, I ran the ASAN build with gdb still which I think I wasn't supposed to do? Will give it another run later, debugging something that takes several hours to crash is not the most convenient :p

What compiler are you using?

Just regular gcc 15.2.1

[rfomin]

If these lines are printed, then something is wrong.

-- Performing Test HAVE_ASAN
-- Performing Test HAVE_ASAN - Failed

GCC 15 should support ASAN.

I ran the ASAN build with gdb

I think it should work, but I've never tried it.

Try this miminal test:

int main(void)
{
    char str[10];
    char a = str[11];
}

gcc test.c -fsanitize=address
./a

You will get ASAN report (it's on my Windows machine):

=================================================================
==6628==ERROR: AddressSanitizer: stack-buffer-overflow on address 0x00043ffcf84b at pc 0x7ff7a17514e7 bp 0x00043ffcf7a0 sp 0x00043ffcf7e8
READ of size 1 at 0x00043ffcf84b thread T0
    #0 0x7ff7a17514e6 in main (C:\develop\woof\src\a.exe+0x1400014e6)
    #1 0x7ff7a17510fd in __tmainCRTStartup D:/W/B/src/mingw-w64/mingw-w64-crt/crt/crtexe.c:236:11
    #2 0x7ff7a17513b5 in .l_start D:/W/B/src/mingw-w64/mingw-w64-crt/crt/crtexe.c:122:9
    #3 0x7ff89c4f7373  (C:\Windows\System32\KERNEL32.DLL+0x180017373)
    #4 0x7ff89d43cc90  (C:\Windows\SYSTEM32\ntdll.dll+0x18004cc90)

Address 0x00043ffcf84b is located in stack of thread T0 at offset 43 in frame
    #0 0x7ff7a17513ef in main (C:\develop\woof\src\a.exe+0x1400013ef)

  This frame has 1 object(s):
    [32, 42) 'str' <== Memory access at offset 43 overflows this variable
HINT: this may be a false positive if your program uses some custom stack unwind mechanism, swapcontext or vfork
      (longjmp, SEH and C++ exceptions *are* supported)
SUMMARY: AddressSanitizer: stack-buffer-overflow (C:\develop\woof\src\a.exe+0x1400014e6) in main
...

[kitchen-ace]

Try this miminal test:

Yeah that works fine and gives me the large ASAN dump similar to what you posted.

[kitchen-ace]

If I set my compiler to Clang with export CC=/usr/bin/clang; export CXX=/usr/bin/clang++ then ASAN is detected properly.

[rfomin]

If I set my compiler to Clang with export CC=/usr/bin/clang; export CXX=/usr/bin/clang++ then ASAN is detected properly.

Is there any news on the ASAN report yet?

[kitchen-ace]

Here's the stderr from running that build. You can ignore the alsa stuff at the start.

dsdaasan-stderr.txt

[rfomin]

Here's the stderr from running that build. You can ignore the alsa stuff at the start.

dsdaasan-stderr.txt

Unfortunately, this report only covers "leaks" at program exit. We don't free memory upon exiting because the OS does it anyway. It's common behavior; on Windows, ASAN doesn't report these "leaks".
Perhaps the signal handler is preventing ASAN from working properly. I will try removing it.

[kitchen-ace]

OK with the latest commit @fabiangreffrath made, ASAN seems to be working. Here's what is hopefully a more useful dump.

dsdaasan-stderr2.txt

I'm getting a lot of crashes playing one specific map so this might not actually be the original problem though. I'll run through okuplok later today, or tonight.

[rfomin]

I'm getting a lot of crashes playing one specific map so this might not actually be the original problem though. I'll run through okuplok later today, or tonight.

Are you using a sound pack with WAV/Ogg? It seems the crash is related to the code we recently added to support various SFX formats.

[kitchen-ace]

Are you using a sound pack with WAV/Ogg? It seems the crash is related to the code we recently added to support various SFX formats.

That map used MP3 music. MAP05 from https://www.doomworld.com/forum/topic/154798-the-oven-an-8-level-wad/

Normally I don't use any extra sound wads etc.

[rfomin]

That map used MP3 music. MAP05 from https://www.doomworld.com/forum/topic/154798-the-oven-an-8-level-wad/

It also has DSPLASMA sound in WAV format. I will look into it.