On 5/22/2023 4:21 PM, Stephen Hemminger wrote:
On Mon, 22 May 2023 11:19:24 +0100
"Burakov, Anatoly" <anatoly.bura...@intel.com> wrote:
This case validates mbuf. IMO there is no need to do validation in a secondary
process.
Unit test for rte_panic() also uses fork() and could have the same issue.
In that case, rte_panic() test should be fixed as well.
My concern is that ideally, we shouldn't intentionally crash the test
app if something goes wrong, and calling rte_panic() accomplishes just
that - which is why I suggested running them in secondary processes
instead, so that any call into rte_panic happens inside a secondary
process, and the main test process doesn't crash even if the test has
failed.
All forks outside of EAL are bad. The test should be removed, it was buggy
when first written.
I agree that none of the tests (or anything else in DPDK) should fork,
but some things still crash process (as in, cause rte_panic), and the
purpose of using fork() was for that. We can rewrite the tests to not
fork and instead use other methods (such as spinning up secondary
processes), we don't have to remove them.
--
Thanks,
Anatoly
