Re: [PATCH] ethdev: prefer offload names in logs

[David Marchand <david.marchand@redhat.com>]

On Wed, May 17, 2023 at 4:53 PM Ferruh Yigit <ferruh.yigit@amd.com> wrote:
>
> On 3/9/2023 4:21 PM, Stephen Hemminger wrote:
> > On Thu,  9 Mar 2023 09:16:33 +0100
> > David Marchand <david.marchand@redhat.com> wrote:
> >
> >> +            RTE_ETHDEV_LOG(ERR, "Ethdev port_id=%u requested Rx offloads '%s' in %s(). "
> >> +                    "Device supports '%s' Rx offloads but does not support '%s'.\n",
> >> +                    port_id, requested != NULL ? requested : "N/A", __func__,
> >> +                    available != NULL ? available : "N/A",
> >> +                    unavailable != NULL ? unavailable : "N/A");
> >> +            free(requested);
> >
> > Please shorten message and make sure it does not cross line boundaries.
> > Best to allow users to do simple search for message.
>
> Agree that using offload names are more user friendly.
>
> To keep the log more reasonable length, what would you think to split
> into two, one in ERR level other is in info/debug:
> ERR, "Ethdev port_id=%u does not support '%s'.\n", unavailable
> DEBUG, "Ethdev port_id=%u requested Rx offloads '%s', device supports
> '%s'.\n", requested, available
>
> And I think we can drop __func__, we don't use in many other logs anyway.

Splitting seems the simpler and won't require an application involvement.
I would even split the debug message in two (after all, if we have two
logs, why not three :-)).

I'll also revisit the patch wrt allocations.

>
>
>
> Other option can be to provide APIs to print all offloads (similar to
> 'rte_eth_dev_tx_offload_name()'), so application does its own logging,
> and ethdev just prints 'unavailable' part of the log.
>


-- 
David Marchand