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

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:

quoted

On Wed, Jul 28, 2021 at 06:46:53PM +0300, Andrew Rybchenko wrote:

quoted

On 7/7/21 12:40 PM, Mohsin Kazmi wrote:

quoted

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 <redacted>
Acked-by: Qi Zhang <redacted>
---
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)

quoted

quoted

quoted

     * 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)))

quoted

quoted

quoted

+   if (!(ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_L4_MASK |

PKT_TX_TCP_SEG |

quoted

quoted

quoted

+                     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);

quoted

quoted

quoted

+                   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.

quoted

quoted

quoted

+           }
+   }
    /*
     * Check if headers are fragmented.