You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Start seeing Unhandled signal SIGSEGV from our rust app when upgrading to version 0.1.48 from 0.1.47 #153
Since we upgrade mimalloc crate from 0.1.47 to 0.1.48, we started seeing Unhandled signal SIGSEGV error from our error logs. From our stack trace related to these errors, we found the following:
#24 0x00007e3c6cbd7d13 in malloc_consolidate (av=0x7e3a6c000030) at ./malloc/malloc.c:4876
#25 0x00007e3c6cbd8da5 in _int_free_maybe_consolidate (av=0x7e3a6c000030, size=<optimized out>) at ./malloc/malloc.c:4779
#26 0x00007e3c6cbd943a in _int_free (av=0x7e3a6c000030, p=<optimized out>, have_lock=<optimized out>) at ./malloc/malloc.c:4646
#27 0x00007e3c6cbdc001 in __GI___libc_free (mem=<optimized out>) at ./malloc/malloc.c:3398
#28 tcache_thread_shutdown () at ./malloc/malloc.c:3234
#29 __malloc_arena_thread_freeres () at ./malloc/arena.c:897
#30 0x00007e3c6cbde502 in __libc_thread_freeres () at ./malloc/thread-freeres.c:41
#31 0x00007e3c6cbca895 in start_thread (arg=<optimized out>) at ./nptl/pthread_create.c:458
#32 0x00007e3c6cc57c6c in clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:78
For now, we have reverted back to use version 0.1.47. Hopefully, the future release will handle this error. Thanks
The transition from mimalloc_rust 0.1.47 to 0.1.48 seems significant, because mimalloc_rust commit 747b5b bumps the version of the underlying C library (Microsoft's mimalloc) from v2 to v3, if I'm reading the diffs correctly.
It looks like the usage of mimalloc v3 is behind a cargo feature in libmimalloc-sys. I'm not sure if end-users are able to force that flag off for that crate, but that would be a way to test whether switching back to mimalloc v2 fixes the problem.
Activity
tomkel commentedon Feb 10, 2026
Yes i'm also seeing this issue, getting segfault
vladh commentedon Mar 1, 2026
The transition from mimalloc_rust 0.1.47 to 0.1.48 seems significant, because mimalloc_rust commit 747b5b bumps the version of the underlying C library (Microsoft's mimalloc) from v2 to v3, if I'm reading the diffs correctly.
It looks like the usage of mimalloc v3 is behind a cargo feature in libmimalloc-sys. I'm not sure if end-users are able to force that flag off for that crate, but that would be a way to test whether switching back to mimalloc v2 fixes the problem.