Hendrik Greving
Hendrik Greving
Looks like the cdash failure is 32-bit linux.
xref #3366
I have not been able to find out what exactly is going wrong - yet. But I believe it is 32-bit specific and will lower priority to go after it...
Investigated further: * I am printing xax at the beginning of dispatch in `if (get_at_syscall(dcontext) && was_thread_create_syscall(dcontext))` which should print the tid of a child when the parent is post-syscall....
Investigated further: * When inspecting sc->xsp of the thread that runs into SIGSEGV, it looks like the value is always not at a map's bound. I am getting a delta...
Investigated further: * Moving faulting instruction pre-syscall has no effect (does not fault). * Enabling the checks in get_thread_private_dcontext has no effect either (doesn't fire). * As suggested by Derek,...
Conclusion (no fix yet): a native thread that hadn't been taken over yet follows the vsyscall hook in the vdso page, and our post-syscall code assuming valid TLS is causing...
This issue also covers / will cover 32-bit flakiness of the new test api.detach_spawn_stress.