Threads::print_on_error calls Universe::heap()->gc_threads_do(...) without checking the return value of Universe::heap(). Since this function can be called before the heap has been set up, I think we need to NULL check here.

hs_err output from early failure without the fix:
Other Threads:

[error occurred during error reporting (printing all threads), id 0xb, SIGSEGV (0xb) at pc=0x0000000104cbb609]

and with a null check:
Other Threads:

=>0x00007ff10980e220 (exited) JavaThread "Unknown thread" [_thread_in_vm, id=7171, stack(0x00007000083dc000,0x00007000084dc000)]