> v2:
> - Cap same-pool runs at 64 to retain mempool-cache behavior.
NAK!
Why would you do that?
> - Move the sanity check after the chained-packet fallback decision.
> - Add mixed flat/chained and indirect-mbuf regression cases.
> - Remove the earlier timing claim; the capped version has not yet
> been benchmarked on x86 or with a cached mempool.
>
> + run_count++;
> + if (run_count == RTE_PKTMBUF_FREE_PENDING_SZ) {
> + rte_mbuf_raw_free_bulk(run_pool,
> + &mbufs[run_start], run_count);
> + run_count = 0;
> + }
NAK to this part.
Flushing a large bulk in smaller bulks of 64 mbufs is not necessary, and may
reduce performance.
