Re: [PATCH v4 1/4] wifi: ath11k: use queue mapping for WCN6750 ring selection

[Jose Ignacio Tornos Martinez <jtornosm@redhat.com>]

Hello Tamizh,

> Sure thanks for the information. Do we have throughput values with
> this change?

I don't have WCN6750 hardware, but since WCN7850 (ath12k) already
uses skb_get_queue_mapping() for ring selection, I tested on
WCN7850 comparing queue mapping vs hash based ring selection with
the following change:

  static u8 ath12k_wifi7_hw_get_ring_selector_wcn7850(struct sk_buff *skb)
  {
  -       return skb_get_queue_mapping(skb);
  +       return skb_get_hash(skb);
  }

The hash based test was done on top of the v3 tx flow control
series, which is needed to get stable throughput results.

Results (100s iperf3 TCP TX, open radio environment with VMs):

  skb_get_queue_mapping: 764 Mbits/sec (8.89 GB, 19 retries)
  skb_get_hash:          744 Mbits/sec (8.67 GB, 25 retries)

Throughput is comparable with no measurable regression from using
queue mapping based ring selection.

Note that VM-based testing may not reach peak hardware throughput,
but I think it is sufficient to compare the two ring selection
methods under identical conditions.

Thanks

Best regards
José Ignacio