On Thu, Jun 26, 2025 at 11:16:42PM +0000, Yogev Chaimovich wrote: > OK. > I wished there was a DPDK function to verify if transceiver is plugged in or > not, without log errors. Such as rte_eth_dev_is_module_exist() > So, 1'st I would have checked if it's plugged in. Then I would have used > rte_eth_dev_get_module_info().
I think such an API is a good idea, but in mlx5 PMD case, this would possibly also require additional changes on kernel driver side. Alternatively, if ETHTOOL_GMODULEINFO returned EIO, then in this new ethdev API, driver could return that transceiver does not exist and nothing would be logged. But still, the same kind of uncertainty as I mentioned in the previous mail would be reached - EIO does not necessarily mean that transceiver is not plugged in. Because of that uncertainty, I think it's invalid to reduce the log level or introduce the special handling for EIO as proposed in existing implementation of rte_eth_dev_get_module_info(). > > The workaround is setting the module's log level so the error won't be seen > before calling rte_eth_dev_get_module_info(), and set it back to default > afterwards. > > Thanks, > Yogev > ________________________________ > From: Dariusz Sosnowski <[email protected]> > Sent: Thursday, June 26, 2025 14:36 > To: Yogev Chaimovich <[email protected]> > Cc: Maayan Kashani <[email protected]>; Stephen Hemminger > <[email protected]>; Slava Ovsiienko <[email protected]>; Bing > Zhao <[email protected]>; Ori Kam <[email protected]>; Suanming Mou > <[email protected]>; Matan Azrad <[email protected]>; [email protected] > <[email protected]> > Subject: Re: [PATCH] net/mlx5: fix transceiver warning when not exist > > Hi Yogev, > > On Mon, Jun 09, 2025 at 01:11:27PM +0000, Yogev Chaimovich wrote: > > Hi, > > Actually looks like we're going with another workaround, so I might not be > > needing my original patch. > > Could you tell us what kind of workaround you applied? > > Also, I'm not sure if this warning should be suppressed on EIO. > mlx5 PMD relies on the kernel driver to get the module info, > and from what I've seen for ETHTOOL_GMODULEINFO kernel can return EIO > also for cases other than transceiver not existing > (for example error on NIC register access). > So mlx5 PMD cannot discern between "no transceiver" and > "other I/O error". > For this reason I would keep the log and keep it at warn level. > > > Also, I do not understand how my patch (not printing a log) makes a test > > fail - I couldn't find that test in DPDK to verify. I'm not that familiar > > though.. > > > > Yogev > > *snip* > > Best regards, > Dariusz Sosnowski
