Jakub Kicinski <[email protected]> writes:
> On Wed, 19 Nov 2025 14:40:41 +0100 Petr Machata wrote: >> > @@ -163,7 +162,7 @@ KSFT_DISRUPTIVE = True >> > entry = global_defer_queue.pop() >> > try: >> > entry.exec_only() >> > - except: >> > + except Exception: >> >> This used to catch KsftTerminate, which we use for SIGTERM handling, now >> it doesn't anymore. I think it could legitimately appear in that context >> if SIGTERM si delivered while exec_only() is running. >> >> IMHO it should catch BaseException, like ksft_run() already does. > > TBH I haven't thought of this. Are you thinking that we shouldn't > interrupt the execution of deferred cleanups when SIGTERM arrives? > Fair point, but I think we'd need more code to handle that properly 🤔️ > Right now we ignore SIGTERM which isn't great. After this patch we'll > no longer ignore it and have the whole test exit. Neither actually > catches the exception and sets stop=True in ksft_run().. Well, previously at least the rest of the defer queue would be run, now it's skipped. Which -- OK, likely if you SIGTERM in the middle of a cleanup, chances are the cleanup is stuck and the sigterm then skips it, which is what you want. It's a pick your poison. Ignore C-c or ignore deferred cleanups :) Given this is supposed to be a cleanup patch, it would make sense to just s/bare except/except BaseException/ and leave the behavior broken. The commit message indicates it's a NOP make-linter-happy patch, but then it's not really. > WDYT about leaving this patch as is and doing this on top: It's a good stop-gap. > diff --git a/tools/testing/selftests/net/lib/py/ksft.py > b/tools/testing/selftests/net/lib/py/ksft.py > index 83b1574f7719..5a667ad22ef4 100644 > --- a/tools/testing/selftests/net/lib/py/ksft.py > +++ b/tools/testing/selftests/net/lib/py/ksft.py > @@ -268,7 +268,12 @@ KSFT_DISRUPTIVE = True > KSFT_RESULT = False > cnt_key = 'fail' > > - ksft_flush_defer() > + try: > + ksft_flush_defer() > + except BaseException as e: > + stop |= isinstance(e, KeyboardInterrupt) > + # Flush was interrupted, try to finish the job best we can > + ksft_flush_defer() > > if not cnt_key: > cnt_key = 'pass' if KSFT_RESULT else 'fail'
