[Devel] [PATCH hci-8.1 v8 2/2] selftests: ve: introduce vtty pair teardown stress test

[Eva Kurchatova (Virtuozzo)]

For future reference:

On Thu, 8 Oct 2026 04:03:29 +0000, no-reply at sashiko.virtuozzo.com wrote:
 > - [Medium] selftests/ve: kmsg_check() fails on any unrelated host log 
record or ring overrun

Deliberate, and settled in v7: a kernel that warns during the run is a
kernel to look at, whatever warned, and the test prints the record so
the reader sees what it was. ACKed by Konstantin.

 > - [Low] selftests/ve: master loop ignores early exit of slave child 
in main()

It changes when the verdict is printed, not the verdict. A slave that
stops on its first open is reported as a skip or a failure at the end
of the run.

 > - [Low] selftests/ve: wait_child() leaves status 0 when waitpid() 
fails non-EINTR

waitpid() has no way to fail there: the child is this process's own
fork, nothing else reaps it, and SIGCHLD is not ignored, so ECHILD
cannot come back, and the arguments are fixed. The only error it can
return is EINTR, which the loop handles.

 > - [Low] selftests/ve: stop flag check races with blocking 
waitpid()/read() in main()

A child that is blocked in a syscall is interrupted by the signal,
since SA_RESTART is disabled, and a child that landed on the racy
window will still eventually stop at its own deadline.
I am not aware of an atomic way to read()/waitpid() and check variable,
like a futex, that would be useful here for sure.

 > - [Low] selftests/ve: deadline uses 1s-granularity time(NULL) in 
main() and hammer_slave()

That one is right: a run asked for one second could end immediately,
and the timeout can jitter in 1-second units, however that is not
a useful run time for this test, and adding monotonic clocksource
would in my opinion be redundant here.

Eva