[RFC 0/3] generic sw counters

[Stephen Hemminger]

On Fri, 10 May 2024 19:29:48 +0200
Morten Brørup <mb at smartsharesystems.com> wrote:

> ad, common functions are good.
> 
> I'm somewhat skeptical about putting the stats structure first in the rte_eth_dev_data's tx_queues and rx_queues.
> These are void* because their types are private to the PMD. Putting the stats structure first is somewhat a hack, partially removing that privacy.
> Perhaps we can make it look less like a hack. After all, it is still a private structure type, only the first part is public and must be the same across drivers using the SW stats counters. Overlapping certain parts of a private structure with a public structure is a common design pattern; I've seen it used elsewhere in DPDK too.
> If we get used to it for SW ethdev drivers, we might not consider it a hack anymore. ;-)

The tradeoff here, is putting them first allows for writing generic version of eth_stats_get and eth_stats_reset.
Other option is having generic code take offset for tx/rx to find the stats.

C doesn't have RTI like C++ so this was best I could think of.