Re: [PATCH] ethdev: introduce generic dummy packet burst function

[Thomas Monjalon <thomas@monjalon.net>]

10/02/2022 17:30, Stephen Hemminger:
> On Thu, 10 Feb 2022 13:58:43 +0000
> Ferruh Yigit <ferruh.yigit@intel.com> wrote:
> > On 2/8/2022 7:44 PM, Ferruh Yigit wrote:
> > > --- a/drivers/net/mlx5/mlx5.c
> > > +++ b/drivers/net/mlx5/mlx5.c
> > > @@ -1559,8 +1559,8 @@ mlx5_dev_close(struct rte_eth_dev *dev)
> > >   	mlx5_action_handle_flush(dev);
> > >   	mlx5_flow_meter_flush(dev, NULL);
> > >   	/* Prevent crashes when queues are still in use. */
> > > -	dev->rx_pkt_burst = removed_rx_burst;
> > > -	dev->tx_pkt_burst = removed_tx_burst;
> > > +	dev->rx_pkt_burst = rte_eth_pkt_burst_dummy;
> > > +	dev->tx_pkt_burst = rte_eth_pkt_burst_dummy;
> > >   	rte_wmb();
> > >   	/* Disable datapath on secondary process. */
> > >   	mlx5_mp_os_req_stop_rxtx(dev);
> > > diff --git a/drivers/net/mlx5/mlx5_rx.c b/drivers/net/mlx5/mlx5_rx.c
> > > index f388fcc31395..11ea935d72f0 100644
> > > --- a/drivers/net/mlx5/mlx5_rx.c
> > > +++ b/drivers/net/mlx5/mlx5_rx.c
> > > @@ -252,7 +252,7 @@ mlx5_rx_queue_count(void *rx_queue)
> > >   	dev = &rte_eth_devices[rxq->port_id];
> > >   
> > >   	if (dev->rx_pkt_burst == NULL ||
> > > -	    dev->rx_pkt_burst == removed_rx_burst) {
> > > +	    dev->rx_pkt_burst == rte_eth_pkt_burst_dummy) {
> > >   		rte_errno = ENOTSUP;
> > >   		return -rte_errno;
> > >   	}  
> > 
> > Thinking twice I am not sure if above change works.
> > 
> > Since function is in the header file, and the .c file that assign
> > the 'dev->rx_pkt_burst' and the .c file that check function pointer
> > are different, these two different instance of same function may
> > have different addresses and above check may fail when it should match.
> > 
> > I guess solution is move the function to a .c file and export it
> > internally.
> > I was thinking to add ethdev_driver.c file, perhaps this can be
> > a motivation to start that file.
> > Thomas, Andrew, what do you think about ethdev_driver.c file?
> 
> Right putting it the header file ends up with multiple copies of same
> code compiled into each driver.

I'm OK with introducing such .c file.