Assertion `child_sig_gt_zero' failed to hold. Replaying `SCHED': expecting tracee signal or trap, but instead at `ioctl' (ticks: 55628)

I'm trying to enable -fPIC for librrpreload, since x86-64 is unhappy about having to link relocations related to __thread in non-PIC code. As a start, I was trying to make sure that -fPIC works for 32-bit mode, but I get several test failures:

• async_signal_syscalls_100
• async_signal_syscalls_1000
• checkpoint_async_signal_syscalls_1000
• deliver_async_signal_during_syscalls
• syscallbuf_timeslice_250

In attempting to debug syscallbuf_timeslice_250, I hit the assertion which is the title of this issue. What can I do to figure out what's going wrong? The recorded trace doesn't have any ioctl syscalls whatsoever in it.

The dump has a SCHED event at ticks:55606, an intervening SYSCALLBUF_RESET event, and then a SCHED event at ticks:55928. Is it possible that the ioctl from fiddling with the syscallbuf is the system call we're seeing in the replay, and that somehow didn't happen in the record, or that it got recorded incorrectly?

Start by reproducing the listed async-signal and syscallbuf tests in 32-bit mode, especially syscallbuf_timeslice_250, and inspect the replay around the SCHED events at ticks 55606 and 55928. Compare the recorded events across the intervening SYSCALLBUF_RESET and the reported ioctl assertion; done means identifying the record/replay discrepancy and making the affected tests pass.