Re: iavf: rte_eth_stats_get() fails upon upgrade to 23.07.0

[Isaac Boukris <iboukris@gmail.com>]

On Tue, Feb 27, 2024 at 9:24 PM Isaac Boukris <iboukris@gmail.com> wrote:
>
> On Tue, Feb 27, 2024 at 4:28 PM Isaac Boukris <iboukris@gmail.com> wrote:
> >
> > On Tue, Feb 27, 2024 at 1:27 PM Isaac Boukris <iboukris@gmail.com> wrote:
> > >
> > > Hello,
> > >
> > > This works fine with 23.03.0 but starting from 23.07.0 up to master it
> > > fails as follows.
> > >
> > > Initialization is ok:
> > > EAL: Probe PCI driver: net_iavf (8086:154c) device: 0000:09:00.0 (socket -1)
> > > iavf_set_rx_function(): request RXDID[1] in Queue[0] is legacy, set
> > > rx_pkt_burst as legacy for all queues
> > >
> > > telemetry.py /ethdev/link_status,0 | jq
> > >   {
> > >     "/ethdev/link_status": {
> > >       "status": "UP",
> > >       "speed": 40000,
> > >       "duplex": "full-duplex"
> > >     }
> > >   }
> > >
> > > But the following command hang and fail:
> > > telemetry.py /ethdev/stats,0 | jq
> > > telemetry.py /ethdev/xstats,0 | jq
> > >
> > > On the logs I see:
> > > iavf_execute_vf_cmd(): No response for cmd 15
> > > iavf_query_stats(): fail to execute command OP_GET_STATS
> > > iavf_dev_stats_get(): Get statistics failed
> >
> > I did some git bisect and found the commit below, help appreciated!
> >
> > 5712bf9d6e14d0eaa7a07814d59351ffeb8e02ed is the first bad commit
> > commit 5712bf9d6e14d0eaa7a07814d59351ffeb8e02ed
> > Author: Wenzhuo Lu <wenzhuo.lu@intel.com>
> > Date:   Tue Apr 18 13:11:05 2023 +0800
> >
> >     net/iavf: add Tx AVX2 offload path
> >
> >     Add a specific path for TX AVX2.
> >     In this path, support the HW offload features, like,
> >     checksum insertion, VLAN insertion.
> >     This path is chosen automatically according to the
> >     configuration.
> >
> >     'inline' is used, then the duplicate code is generated
> >     by the compiler.
> >
> >     Signed-off-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
> >     Acked-by: Qi Zhang <qi.z.zhang@intel.com>
> >
> >  drivers/net/iavf/iavf_rxtx.c          | 33 +++++++++++++--------
> >  drivers/net/iavf/iavf_rxtx.h          |  2 ++
> >  drivers/net/iavf/iavf_rxtx_vec_avx2.c | 54 +++++++++++++++++++++++++----------
> >  3 files changed, 62 insertions(+), 27 deletions(-)
>
> Notte that I was able to workaround it by forcing the old logic with:
>
> diff --git a/drivers/net/iavf/iavf_rxtx.c b/drivers/net/iavf/iavf_rxtx.c
> index f19aa14646..1537a5f19c 100644
> --- a/drivers/net/iavf/iavf_rxtx.c
> +++ b/drivers/net/iavf/iavf_rxtx.c
> @@ -4003,6 +4003,8 @@ iavf_set_tx_function(struct rte_eth_dev *dev)
>                         use_avx512 = true;
>  #endif
>
> +               if (!use_sse) use_avx2 = false;
> +
>                 if (!use_sse && !use_avx2 && !use_avx512)
>                         goto normal;

FYI, I was able to fix this and drop my patch by upgrading the
firmware on the host machine.

From:
kernel: i40e 0000:37:00.0: fw 8.5.67516 api 1.15 nvm 8.50 0x8000b6d7
1.3082.0 [8086:1583] [8086:0002]

To:
kernel: i40e 0000:37:00.0: fw 9.150.77492 api 1.15 nvm 9.50 0x8000f146
1.3597.0 [8086:1583] [8086:0002]

Perhaps we should have a compatibility matrix on the iavf doc page,
like the one we have on the i40e PF one.

HTH