[DPDK/ethdev Bug 1441] RSS Hash returned by the gvnic pmd dpdk driver not matching Toeplitz

[bugzilla at dpdk.org]

https://bugs.dpdk.org/show_bug.cgi?id=1441

            Bug ID: 1441
           Summary: RSS Hash returned by the gvnic pmd dpdk driver not
                    matching Toeplitz
           Product: DPDK
           Version: 24.03
          Hardware: Other
                OS: Linux
            Status: UNCONFIRMED
          Severity: critical
          Priority: Normal
         Component: ethdev
          Assignee: dev at dpdk.org
          Reporter: mallappanagesh at gmail.com
  Target Milestone: ---

We are using a C3 GCP instance having GVNIC with DPDK enabled.


We are using latest version of DPDK ( dpdk-24.03 ) to configure RSS key, we
understand C3 GCP instance with gvnic supports Toeplitz rss algorithm (
Reference -
https://github.com/google/compute-virtual-ethernet-dpdk/blob/main/gve_ethdev.c
), but we are hitting below issue.


Problem Statement - The RSS Hash value returned by the gvnic driver is
different and is not matching the one computed by Toeplitz.


Using below script to compute rss hash value from Toeplitz.
(https://gist.github.com/joongh/16867705b03b49e393cbf91da3cb42a7)


Consider below data to compute RSS Hash value -


Source IP - 79.124.62.122
Destination IP - 192.168.7.28
Source Port - 49442
Destination Port - 55327
RSS Key - {0x94e3e97e, 0x4abac401, 0x1e0edbf, 0xe059e1f4, 0xb6e27dff,
0xdf55c916, 0xe9e6e4db, 0x1f9fb82c, 0x1e0edbf, 0xe059e1f4}


Hash Value returned by Toeplitz - 0x8bbafd0f


Gvnic dpdk pmd driver which uses Toeplitz RSS hash algorithm too is expected to
compute above RSS Hash value but we see the Hash value stored in dpdk mbuf to
be 0xc6a15941 which is not right.


>From the DPDK side we have ensured RSS offloading configuration is done
appropriately, but still we are seeing above mismatch.


We request you to please help us understand why this mismatch is happening and
is this expected behavior of gvnic on C3 instance.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mails.dpdk.org/archives/dev/attachments/20240515/7a68a9b5/attachment-0001.htm>