Re: [PATCH] net/iavf: fix Tx offloading flags check

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

On Tue, Oct 24, 2023 at 11:04 AM Radu Nicolau <radu.nicolau@intel.com> wrote:
> On 24-Oct-23 9:44 AM, David Marchand wrote:
> > On Tue, Oct 24, 2023 at 10:40 AM Radu Nicolau <radu.nicolau@intel.com> wrote:
> >>
> >> On 24-Oct-23 6:42 AM, Zhang, Qi Z wrote:
> >>>> -----Original Message-----
> >>>> From: Radu Nicolau <radu.nicolau@intel.com>
> >>>> Sent: Tuesday, October 24, 2023 12:38 AM
> >>>> To: Wu, Jingjing <jingjing.wu@intel.com>; Xing, Beilei <beilei.xing@intel.com>
> >>>> Cc: dev@dpdk.org; Nicolau, Radu <radu.nicolau@intel.com>;
> >>>> stable@dpdk.org; Marchand, David <david.marchand@redhat.com>
> >>>> Subject: [PATCH] net/iavf: fix Tx offloading flags check
> >>>>
> >>>> Use IAVF_TX_OFFLOAD_MASK flags instead of
> >>>> IAVF_TX_CKSUM_OFFLOAD_MASK.
> >>>>
> >>>> Fixes: 3c715591ece0 ("net/iavf: fix checksum offloading")
> >>>> Cc: stable@dpdk.org
> >>>> Cc: david.marchand@redhat.com
> >>>>
> >>>> Signed-off-by: Radu Nicolau <radu.nicolau@intel.com>
> >>>> ---
> >>>>    drivers/net/iavf/iavf_rxtx.c | 2 +-
> >>>>    1 file changed, 1 insertion(+), 1 deletion(-)
> >>>>
> >>>> diff --git a/drivers/net/iavf/iavf_rxtx.c b/drivers/net/iavf/iavf_rxtx.c index
> >>>> c6ef6af1d8..85f8c141ce 100644
> >>>> --- a/drivers/net/iavf/iavf_rxtx.c
> >>>> +++ b/drivers/net/iavf/iavf_rxtx.c
> >>>> @@ -2664,7 +2664,7 @@ iavf_build_data_desc_cmd_offset_fields(volatile
> >>>> uint64_t *qw1,
> >>>>               l2tag1 |= m->vlan_tci;
> >>>>       }
> >>>>
> >>>> -    if ((m->ol_flags & IAVF_TX_CKSUM_OFFLOAD_MASK) == 0)
> >>>> +    if ((m->ol_flags & IAVF_TX_OFFLOAD_MASK) == 0)
> >>> Not sure if this will break previous fix.
> >>> Could you please provide some clarification regarding the specific offload flags that not in IAVF_TX_CKSUM_OFFLOAD_MASK, but you still don't want to skip?
> >> A specific flag is RTE_ETH_TX_OFFLOAD_SECURITY, and because this is not
> >> contained in IAVF_TX_OFFLOAD_MASK the previous fix broke the inline
> >> crypto feature.
> > RTE_ETH_TX_OFFLOAD_SECURITY is a ethdev level flag.
> > This is not supposed to be in a mbuf ol_flags, is it?
>
> No, it's not, you are right. Actually it's RTE_MBUF_F_TX_SEC_OFFLOAD,
> and that also means the IAVF_TX_OFFLOAD_MASK definition is not correct.
> I will send another patch to fix the definition of IAVF_TX_OFFLOAD_MASK.

Thanks.

>
> As for this fix, if you prefer a safer approach I can add another check
> only for RTE_MBUF_F_TX_SEC_OFFLOAD, but one way or the other we need to
> have this fix.

There are multiple helpers touching offloads in this code and I don't
have a clear view of what the best fix for this driver is.

My concern is that IAVF_TX_OFFLOAD_MASK mixes both packet types
(RTE_MBUF_F_TX_IPV[46], RTE_MBUF_F_TX_OUTER_IPV6[46]...) and actual
offloading requests (RTE_MBUF_F_TX_VLAN, RTE_MBUF_F_TX_IP_CKSUM
etc...).

In the mbuf API, the presence of "packet types" tx flags is not
described as something that requires filling l2_len / l3_len etc...
For a similar reason, the presence of RTE_MBUF_F_TX_VLAN tells nothing
about l2_len / l3_len.
So switching to IAVF_TX_OFFLOAD_MASK seems dangerous.

As for your suggestion, it seems safer only checking
RTE_MBUF_F_TX_SEC_OFFLOAD, I agree.


This code is still really hard to follow, so I'll let Intel
maintainers advise on the best fix.


-- 
David Marchand