Segfault in tb_target_set_jmp_target_arm

[futhewo]

In a very long emulation, I have a segfault (dereferencing null pointer) in qemu/tcg/aarch64/tcg-target.inc.c:tb_target_set_jmp_target.

In qemu/accel/tcg/cpu-exec.c, the function tb_set_jmp_target calls the previous function tb_target_set_jmp_target with parameters tc_ptr = 0 and tc_ptr + offset = 0.

Then tb_target_set_jmp_target calls atomic_set((uint64_t*)jmp_addr_, pair) with jmp_addr = tc_ptr + offset = 0, which segfaults.

I patched it (dirty) by adding the following code in tb_set_jmp_target:
if (TCG_TARGET_HAS_direct_jump && tb->tc.ptr) {
[…]
}

I do not know what this patch may break, but it solves the crash.
I am on commit 6ae0c97.

Feel free to ask me anything.

[wtdcode]

Thanks for the report and investigation. Your version seems pretty old, how about current dev branch?

[1144822034]

I have also encountered the same problem, and this issue has been very helpful to me

[futhewo]

Unfortunately, I cannot try on a newer unicorn version, sorry.
I am using unicorn as part of unicornafl++ and it uses the version I pointed out.

[wtdcode]

unicornafl++

What is that? Are you mentioning unicornafl?

[futhewo]

Yes, the one that is part of afl++, that you may find here:
https://github.com/AFLplusplus/unicornafl

[wtdcode]

Yes, the one that is part of afl++, that you may find here: https://github.com/AFLplusplus/unicornafl

Okay, unicornafl will bump to 2.1.2 once I fixed it.

[futhewo]

That would be great! Thanks.