Re: [dpdk-dev] [PATCH v4] mlx5: Support for rte_eth_rx_queue_count

[Olivier Matz <olivier.matz@6wind.com>]

Hi Tom,

On Mon, Nov 05, 2018 at 09:01:03AM +0000, Tom Barbette wrote:
> 
> > It will tell you if in a given offset on the rxq you have a packet ready. I think it will fit your needs, see below.
> So we just loose in precision here. We're looking at ML techniques that will play better with a numerical value and benefit from more entropy than arbitrarily poking of some thresholds.
> If one had to remain, I'd say rx_descriptor_* should be removed as it can be deduced from queue_count. Moreover, in the MLX5 case, calling the function 3 times to get "half busy", "partially busy" or "idle" will lead to scanning the queue 3 times...

The rationale of the descriptor_status() API was to avoid to scan the hw
ring in a linear way, like it was done in *_queue_count() functions (on
Intel NICs). This costs a lot, especially for large rings. The
alternative was to do a pci access to read the hw head/tail, but
benchmarks showed that it was also expensive.

With descriptor_status API, it is fast if you only need a threshold info.
If you need a more precise count, it is possible to do a binary search
algorithm, which is faster than the linear traversal.

That said, these assumptions may not be true for all NICs types, since
there are different ways of implementing the queue.

Regards,
Olivier