Re: [dpdk-dev] [PATCH v3] net: fix Intel-specific Prepare the outer ipv4 hdr for checksum

[Mohsin Kazmi <mohsin.kazmi14@gmail.com>]

Are we good with this patch with the current state? @Olivier: Any comments
on the above suggestions?

On Tue, Aug 3, 2021 at 1:49 PM Mohsin Kazmi <mohsin.kazmi14@gmail.com>
wrote:

>
>
> On Sat, Jul 31, 2021 at 1:49 PM Andrew Rybchenko <
> andrew.rybchenko@oktetlabs.ru> wrote:
>
>> On 7/30/21 2:11 PM, Olivier Matz wrote:
>> > On Wed, Jul 28, 2021 at 06:46:53PM +0300, Andrew Rybchenko wrote:
>> >> On 7/7/21 12:40 PM, Mohsin Kazmi wrote:
>> >>> Preparation the headers for the hardware offload
>> >>> misses the outer ipv4 checksum offload.
>> >>> It results in bad checksum computed by hardware NIC.
>> >>>
>> >>> This patch fixes the issue by setting the outer ipv4
>> >>> checksum field to 0.
>> >>>
>> >>> Fixes: 4fb7e803eb1a ("ethdev: add Tx preparation")
>> >>> Cc: stable@dpdk.org
>> >>>
>> >>> Signed-off-by: Mohsin Kazmi <mohsin.kazmi14@gmail.com>
>> >>> Acked-by: Qi Zhang <qi.z.zhang@intel.com>
>> >>> ---
>> >>> v3:
>> >>>      * Update the conditional test with PKT_TX_OUTER_IP_CKSUM.
>> >>>      * Update the commit title with "Intel-specific".
>> >>>
>> >>> v2:
>> >>>      * Update the commit message with Fixes.
>> >>>
>> >>>    lib/net/rte_net.h | 15 +++++++++++++--
>> >>>    1 file changed, 13 insertions(+), 2 deletions(-)
>> >>>
>> >>> diff --git a/lib/net/rte_net.h b/lib/net/rte_net.h
>> >>> index 434435ffa2..3f4c8c58b9 100644
>> >>> --- a/lib/net/rte_net.h
>> >>> +++ b/lib/net/rte_net.h
>> >>> @@ -125,11 +125,22 @@ rte_net_intel_cksum_flags_prepare(struct
>> rte_mbuf *m, uint64_t ol_flags)
>> >>>      * Mainly it is required to avoid fragmented headers check if
>> >>>      * no offloads are requested.
>> >>>      */
>> >>> -   if (!(ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_L4_MASK |
>> PKT_TX_TCP_SEG)))
>> >>> +   if (!(ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_L4_MASK |
>> PKT_TX_TCP_SEG |
>> >>> +                     PKT_TX_OUTER_IP_CKSUM)))
>> >>>             return 0;
>> >>> -   if (ol_flags & (PKT_TX_OUTER_IPV4 | PKT_TX_OUTER_IPV6))
>> >>> +   if (ol_flags & (PKT_TX_OUTER_IPV4 | PKT_TX_OUTER_IPV6)) {
>> >>>             inner_l3_offset += m->outer_l2_len + m->outer_l3_len;
>> >>> +           /*
>> >>> +            * prepare outer ipv4 header checksum by setting it to 0,
>> >>> +            * in order to be computed by hardware NICs.
>> >>> +            */
>> >>> +           if (ol_flags & PKT_TX_OUTER_IP_CKSUM) {
>> >>> +                   ipv4_hdr = rte_pktmbuf_mtod_offset(m,
>> >>> +                                   struct rte_ipv4_hdr *,
>> m->outer_l2_len);
>> >>> +                   ipv4_hdr->hdr_checksum = 0;
>> >>
>> >> Here we assume that the field is located in the first segment.
>> >> Unlikely but it still could be false. We must handle it properly.
>> >
>> > This is specified in the API comment, so I think it has to be checked
>> > by the caller.
>>
>> If no, what's the point to spoil memory here if stricter check is
>> done few lines below.
>>
> We have two possibilities:
> 1) take the whole block of above code after the strict check: Then strict
> check will use m->outer_l2_len + m->outer_l3_len directly without any
> condition and we will be on the mercy of drivers to initialize these to 0
> if outer headers are not use. Drivers usually don't set the fields which
> they are not interested in because of performance reasons as
> setting these values per packet will cost them additional cycles.
> 2) Taking just PKT_TX_OUTER_IP_CKSUM conditional check after the strict
> fragmented check: In that case, each packet will hit an extra conditional
> check without getting benefit from it, again with a performance penalty.
>
> I am more inclined towards solution 1. But I also welcome other
> suggestions/comments.
>
>>
>> >>> +           }
>> >>> +   }
>> >>>     /*
>> >>>      * Check if headers are fragmented.
>> >>>
>> >>
>>
>>