[dpdk-dev] [Bug 664] ixgbe: Calling rte_eth_dev_rss_reta_update before rte_eth_dev_start sets device in wrong state

[bugzilla at dpdk.org]

https://bugs.dpdk.org/show_bug.cgi?id=664

            Bug ID: 664
           Summary: ixgbe: Calling rte_eth_dev_rss_reta_update before
                    rte_eth_dev_start sets device in wrong state
           Product: DPDK
           Version: 20.11
          Hardware: All
                OS: All
            Status: UNCONFIRMED
          Severity: minor
          Priority: Normal
         Component: ethdev
          Assignee: dev at dpdk.org
          Reporter: barbette at kth.se
  Target Milestone: ---

If one calls rte_eth_dev_rss_reta_update with ixgbe before starting the device
(but after setting everything else), then the table will be full of zeroes
after starting the device.

I would expect one of the following behavior:

 - Either the documentation should say RETA should be set after the device
starts, and the RSS configuration should therefore be ignored and
rte_eth_dev_rss_reta_update return an error
- Ot can be done and the configuration is kept.

In this case, it's half between as if it's called without errors, then the RSS
table will be full zeroes (instead of my table content or the default
sequential list).

Of course I prefer the second option. And some devices like Mellanox ones
actually need to set the reta table before the device starts, else it will
cause a device restart... But I understand ixgbe is old so maybe returning an
error in rte_eth_dev_rss_reta_update  if the device is not started yet is an
acceptable fix.

-- 
You are receiving this mail because:
You are the assignee for the bug.