On Wed, Jun 5, 2024 at 9:55 AM Patrick Robb <pr...@iol.unh.edu> wrote:
>
>
>
> On Fri, May 31, 2024 at 12:44 PM Luca Vizzarro <luca.vizza...@arm.com> wrote:
>>
>>
>> In my testing of Jeremy's patches which depend on this one ("Add second
>> scatter test case"), I've discovered that the Intel E810-C NIC I am
>> using to test does not automatically show "RX scattered packets: on".
>> But I've noticed it does if the MTU is set to something big like 9000.
>>
>> I've tested a change of this by adding:
>>
>> self.set_port_mtu(0, 9000)
>> > + rxq_info = self.send_command(command)
>> self.set_port_mtu(1, 9000)
>>
>> And it seems to work alright. I've also tested this specific change with
>> Mellanox NICs and it didn't seem to affect them at all. No errors or
>> problems and they still showed "RX scattered packets: off" as expected.
>>
>> The `set_port_mtu` method comes from Jeremy's patch...
>>
>>
>
> Hi Jeremy,
>
> Sounds like Luca's way ahead of me here, but I wanted to note that I did run
> from the capabilities patch + Jeremy's new Scatter patch, across these NICs:
>
> Mellanox CX5
> Broadcom 57414 25G
> Broadcom P2100G
> Intel XL710 40G
>
> And in call cases scatter_mbuf_2048 skips, and scatter_mbuf_2048_with_offload
> runs.
>
> The 2nd case passed in all cases, excluding the XL710 where it errors with
> "Test pmd failed to set fwd mode to mac." I can double check that to ensure
> there was no setup error on my part, but I think the more interesting part is
> the skip on the non-offload testcase, as I recall Jeremy saying that the
> XL710 was expected to natively support scatter and run the first testcase.
The "failing to set forwarding mode" is strange, I assume this is
likely because of another issue that Luca noted which was some Intel
NICs taking longer to start testpmd and the timeout not being long
enough to allow for proper startup. If this is the case it would have
essentially "poisoned" the testpmd output buffer and caused all of the
verification steps of subsequent commands to fail. This should be
fixed in the new series of my patch that increases this timeout to 5
seconds.
>
> I can do a rerun, adding in the MTU modifier, and see if the same adjustment
> happens as with the E810 as Luca describes.
