On 10/19/21 12:04 PM, Andrew Rybchenko wrote: > On 10/19/21 11:49 AM, David Marchand wrote: >> On Mon, Oct 18, 2021 at 4:50 PM Andrew Rybchenko >> <andrew.rybche...@oktetlabs.ru> wrote: >>> >>> Add RTE_ prefix to macro used to register mempool driver. >>> The old one is still available but deprecated. >> >> ODP seems to use its own mempools. >> >> $ git grep-all -w MEMPOOL_REGISTER_OPS >> OpenDataplane/platform/linux-generic/pktio/dpdk.c:MEMPOOL_REGISTER_OPS(odp_pool_ops); >> >> I'd say it counts as a driver macro. >> If so, we could hide it in a driver-only header, along with >> rte_mempool_register_ops getting marked as internal. >> >> $ git grep-all -w rte_mempool_register_ops >> FD.io-VPP/src/plugins/dpdk/buffer.c: rte_mempool_register_ops (&ops); >> FD.io-VPP/src/plugins/dpdk/buffer.c: rte_mempool_register_ops (&ops); > > Do I understand correctly that it is required to remove it from > stable ABI/API, but still allow external SW to use it? > > Should I add one more patch to the series? >
I'm afraid not now. It is too invasive or too illogical. Basically it should more rte_mempool_ops to the header as well, but it is heavily used by inline functions in rte_mempool.h. Of course, it is possible to move just register API to the mempool_driver.h header, but value of such changes is not really big.
