Re: [PATCH net-next] net/mlnx5: Use generic code for page_pool statistics.

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

On 2025-03-06 12:22:42 [+0200], Tariq Toukan wrote:
> 
> 
> On 06/03/2025 11:56, Sebastian Andrzej Siewior wrote:
> > On 2025-03-06 11:50:27 [+0200], Tariq Toukan wrote:
> > > On 06/03/2025 10:32, Sebastian Andrzej Siewior wrote:
> > > > Could I keep it as-is for now with the removal of the counter from the
> > > > RQ since we don't have the per-queue/ ring API for it now?
> > > 
> > > I'm fine with transition to generic APIs, as long as we get no regression.
> > > We must keep the per-ring counters exposed.
> > 
> > I don't see a regression.
> > Could you please show me how per-ring counters for page_pool_stats are
> > exposed at the moment? Maybe I am missing something important.
> > 
> 
> After a ring is created for the first time, we start exposing its stats
> forever.
> 
> So after your interface is up for the first time, you should be able to see
> the following per-ring stats with respective indices in ethtool -S.
> 
> pp_alloc_fast
> pp_alloc_slow
> pp_alloc_slow_high_order
> pp_alloc_empty
> pp_alloc_refill
> pp_alloc_waive
> pp_recycle_cached
> pp_recycle_cache_full
> pp_recycle_ring
> pp_recycle_ring_full
> pp_recycle_released_ref
> 
> Obviously, depending on CONFIG_PAGE_POOL_STATS.

That is correct. But if you have 4 rings, you only see one pp_alloc_fast
which combines the stats of all 4 rings. Or do I miss something?

Sebastian