08/11/2019 13:09, Andrew Rybchenko: > On 11/8/19 2:48 PM, Matan Azrad wrote: > > From: Andrew Rybchenko > >> It does 3 things for all port at the end of rte_eth_dev_configure() based > >> on > >> dev->data->dev_conf (if PMD updates it in the case of violations). > >> > >> 1. If requested offload is not enabled, log error message. > >> > >> 2. If requested offload is not enabled, fail dev_configure by > >> returning error and deconfiguring all queues. > >> > >> 3. If not requested offload is enabled, log info message. > >> > >> Above is done for Rx and Tx offloads. > > > > Agree with your summary. > > > > My problem in this patch is part 3: > > This part implies by a formal way(ethdev) that there are 2 configurations: > > 1. The application configuration. > > 2. The actual device configuration. > > > > I think that formally if there is a difference between both, an error must > > be returned. > > Informally, PMD can do whatever it want with good documentation in its doc > > file. > > OK I see. The problem that you'd like make (3) stronger > check to log and return error. > > Above discussion cover it. We can discuss it later, but > right now it cannot be done (at least from my point of > view) as I said above.
Yes, making checks stronger can always be done later. It looks a step is done in this patch. In general, the gap between app needs and PMD support needs to be better handled in ethdev. Matan, I suggest you submit a patch for next release, so it will be a base for continuing this discussion.
