crashdump caused by unexpected halt(undefined, undefined)

[zzydxm]

Describe the bug
Sometimes during node shutdown we see such Slogan:

⁠ Slogan: System process <0.0.0> terminated: {badarg,[{erlang,halt,[undefined,undefined],[{error_info,#{cause=>badopt,module=>erl_erts_errors}}]},{erlang,halt,0,[]}]}

It seems that when init.erl calls halt() as the last step of shutdown, somehow it results in a halt(undefined, undefined)

We think it is caused by this line https://github.com/erlang/otp/blob/OTP-26.2.3/erts/emulator/beam/bif.c?fbclid=IwZXh0bgNhZW0CMTEAAR3G8wFBshuPgBzFHroABMARgLPYBKZZpU42C_f6CAwxYnTxsRIAT_a__l4_aem_sExFv1tZJSqb_KTxH1r0LQ#L4307

but not sure in which situation the line will (should) be called

To Reproduce
We haven't a way to reproduce it yet, just wonder what can be the cause so we can dive deeper, thanks!

Expected behavior
The node should be shut down gracefully

Affected versions
26.2.1

[sverker]

There seem to be a possibility for race in BIF halt_2. If two or more calls are done "at the same time" only one will succeed the cmpxchg operation in erts_halt() at

otp/erts/emulator/beam/erl_process.c

Line 14968 in 928d03e

if (-1 == erts_atomic32_cmpxchg_acqb(&erts_halt_progress,

the other(s) will return from erts_halt and may actually do the scheduled halt(undefined,undefined) call, which is not supposed to happen, before the scheduler gets halted.

#8640 is a proposed fix that suspends the calling process.

[zzydxm]

Ah I see, thanks a lot for the fix!