20/04/2021 11:05, Min Hu (Connor):
> 在 2021/4/20 8:57, Thomas Monjalon 写道:
> > 08/04/2021 12:14, Min Hu (Connor):
> >> From: Chengwen Feng <fengcheng...@huawei.com>
> >>
> >> The ethtool use the socket_id which get from rte_eth_dev_socket_id API
> >> in the init stage, but use the rte_socket_id API to get socket_id when
> >> setting ringparam.
> >>
> >> This patch make sure it call rte_eth_dev_socket_id API to get
> >> socket_id when setting ringparam.
> >>
> >> Fixes: bda68ab9d1e7 ("examples/ethtool: add user-space ethtool sample
> >> application")
> >> Cc: sta...@dpdk.org
> >
> > The explanation is missing something.
> > Please tell why it is wrong.
> >
> >
> Hi, Thomas,
> It is advised that bind queue resources to the NUMA node where the
> device resides, not the NUMA node corresponding to the thread where the
> command line resides. And that will be better.
I know, but in the commit message, you give function name
without saying that rte_socket_id is the running socket
while rte_eth_dev_socket_id is the device socket.
Please make your commit messages more obvious for anyone reading.
Story telling is usually a good approach.
