Re: [PATCH] selftests: ublk: propagate recovery worker failures

[shao.mingyin]

> Please run the real tests next time, it just needs one kernel with
> ublk enabled, and stubs can't cover what the tests are supposed to
> verify.

Agreed, and sorry for sending it that way.  The stubs only exercised the
ERR_CODE aggregation in the shell: they can show that the parent now sees a
worker's status, not that a real failed recovery reaches the final result.
The note should not have claimed more than that.  The follow-up for
test_stress_*.sh will be run on a kernel built with CONFIG_BLK_DEV_UBLK,
and I'll do the same for any future revision of this patch.

> Nit: the above can be simplified as:
>
> 		wait "$pid" && continue
> 		ret=$?
>
> Not a big deal.

I'd rather not make you re-review a v2 just for this, so I'll leave it for
now; if something else forces a revision I'll fold it in.  Say the word if
you'd prefer a v2 with it either way.

> BTW, test_stress_*.sh have the same issue (ERR_CODE set in background
> worker + bare 'wait'), which may be handled in a follow-up.

Yes, they set ERR_CODE in a worker and call a bare 'wait' the same way.  I
will send that as a separate patch on top of this one, and make the helper
in test_common.sh generic (_wait_test_pids) there, so this patch stays as
reviewed.

Thanks for the review and the R-b.