Matias Elo

Sounds like an issue in the Mellanox drivers, so there is very little we can do. I'd verify that the latest Mellanox drivers for the correct OS are being used....

Thanks for the report and comprehensive logs. I assume the test application was working with v1.31.0.1_DPDK_19.11 tag? How about if you try to use packet pool without packet IO? For...

Hmm, I ran your test application on a Neoverse N1 SoC without problems. To verify the configuration: DPDK v20.11: - Otherwise standard build except : `meson configure -Ddefault_library=static` ODP v1.32.1.0_DPDK_19.11:...

Looking closer at the backtrace & logs, this issue could potentially be caused by DPAA2_MEMPOOL (assuming it's used) not being linked/initialized correctly. You could try checking if there is some...

> Hi, just to add tested with the following odp's v1.25 good (current using), v1.26 good v1.27 good v1.28 good v1.29 good v1.30 bad Thanks for testing this @Carisbrooke. Are...

> Thanks for looking into this. Would try to investigate that deeply. However except DIRECT mode initial app may be run in QUEUE mode with mem free over odp_event_free() which...

Thank you @rivasiv for the test code. Looking at the code, there is still one initialization bug: - `odp_init_local()` should be called once per thread (in this case right after...

> It's not a problem that in the case where rte_eth_link_get_nowait() would have to wait, we just get a link status DOWN and a success return status? Assuming a following...

Hi, The current ODP-DPDK implementation only supports software timestamping. If I understand the current DPDK API correctly, HW timestamping can only be enabled for PTP packets (IEEE1588/802.1AS). Would you need...

Thanks for the info. There really isn't much documentation to go with but it seems like these new APIs could be useful, although they are still experimental.