Hello Martin,
On 8/15/24 04:20, Martin KaFai Lau wrote:
> On 8/13/24 5:45 AM, Alexis Lothoré (eBPF Foundation) wrote:
>
>> +#define DST_ADDR "ff02::1"
>
> [ ... ]
>
>> +static int wait_local_ip(void)
>> +{
>> + char *ping_cmd = ping_command(AF_INET6);
>> + int i, err;
>> +
>> + for (i = 0; i < WAIT_AUTO_IP_MAX_ATTEMPT; i++) {
>> + err = SYS_NOFAIL("%s -c 1 -W 1 %s%%%s", ping_cmd, DST_ADDR,
>> + VETH_1);
>
> I tried in my qemu. This loop takes at least 3-4 iteration to get the ping
> through. This test could become flaky if the CI is busy.
>
> I have been always wondering why some of the (non) test_progs has this practice.
> I traced a little. I think it has something to do with the "ff02::1" used in the
> test and/or the local link address is not ready. I have not further nailed it
> down but I think it is close enough.
>
> It will be easier to use a nodad configured v6 addr.
>
> I take this chance to use an easier "::1" address for the test here instead of
> ff02::1. This also removed the need to add veth pair and no need to ping first.
>
> Applied with the "::1" changes mentioned above.
Thanks for the improvement. So far I tried to preserve as much as possible the
original test behavior to avoid accidentally removing some important features,
but I guess the more tests I work on, the more confident I will be to
"challenge" some parts ;)
> Thanks for migrating the tests to test_progs. This is long overdue.
For the record, the next test I am targeting is test_xdp_features.sh. I have
undertaken the conversion and have a working base but it still needs some work.
Thanks,
Alexis
--
Alexis Lothoré, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
