Why did you change this?
compiledMethod() is called at a known location where all pointers should be valid. If it's not the case, there must be a bug somewhere.

Check this segfault stacktrace: https://github.com/fandreuz/async-profiler/actions/runs/15069387768/job/42361695062#step:16:164

In the same run, I also added some debug print statements:

    VMMethod* compiledMethod() {
        const char* env = *(const char**) at(_comp_env_offset);
        if (env != NULL) {
            const char* task = *(const char**) (env + _comp_task_offset);
            if (task != NULL) {
+               fprintf(stderr, "env: %p, taks_ptr: %p, task_offset: %d, method_ptr: %p, method_offset: %d\n", 
+                  env, task, _comp_task_offset, task + _comp_method_offset, _comp_method_offset);
                return *(VMMethod**) (task + _comp_method_offset);
            }
        }
        return NULL;
    }

and these are the last few lines before the segfault:

env: 0x7f6014cf9a10, taks_ptr: 0x7f60310208d0, task_offset: 112, method_ptr: 0x7f60310208e0, method_offset: 16
env: 0x7f6014cf9a10, taks_ptr: 0x7f60310208d0, task_offset: 112, method_ptr: 0x7f60310208e0, method_offset: 16
env: 0x7f6014cf9a10, taks_ptr: 0x7f60310208d0, task_offset: 112, method_ptr: 0x7f60310208e0, method_offset: 16
env: 0x7f6014cf9a10, taks_ptr: 0x1, task_offset: 112, method_ptr: 0x11, method_offset: 16

It seems that the dereferenced pointer to CompileTask points to some random location. In a different run, I also got a valid task_ptr and an invalid method_ptr.

Do we have any strong guarantee about the memory we're reading? I guess we don't, the JVM seems to be free to reuse it however it likes. My theory here is that the CompileTask completed successfully, and we're just getting access to some memory which has been recycled.

OK, so your test found a real bug, and that's actually a good news.
The reason is, CompilerThread::_env is initialized inside invoke_compiler_on_method. This means, it's not always safe to access _env inside invoke_compiler_on_method, we should check the presence of another frame like compile_method.

See https://github.com/openjdk/jdk/blob/265d630125db448ba0cdc3ab7e938beb50e93ed0/src/hotspot/share/compiler/compileBroker.cpp#L2266

Fixed