On Thu, Mar 5, 2020 at 10:17 PM ZY Qiu <quze...@gmail.com> wrote:
>
> When compiling with -O0,
> the compiler does not optimize two memory accesses into one.
> Leads to accessing a null pointer when queue post Rx burst callback
> removal while traffic is running.
>
> Signed-off-by: ZY Qiu <tgw_t...@tencent.com>
> ---
> lib/librte_ethdev/rte_ethdev.h | 9 +++++----
> 1 file changed, 5 insertions(+), 4 deletions(-)
>
> diff --git a/lib/librte_ethdev/rte_ethdev.h b/lib/librte_ethdev/rte_ethdev.h
> index d1a593ad1..c46612e3e 100644
> --- a/lib/librte_ethdev/rte_ethdev.h
> +++ b/lib/librte_ethdev/rte_ethdev.h
> @@ -4388,10 +4388,10 @@ rte_eth_rx_burst(uint16_t port_id, uint16_t queue_id,
> rx_pkts, nb_pkts);
>
> #ifdef RTE_ETHDEV_RXTX_CALLBACKS
> - if (unlikely(dev->post_rx_burst_cbs[queue_id] != NULL)) {
> - struct rte_eth_rxtx_callback *cb =
> - dev->post_rx_burst_cbs[queue_id];
> + struct rte_eth_rxtx_callback *volatile cb =
> + dev->post_rx_burst_cbs[queue_id];
Is adding rte_compiler_barrier() here without changing to volatile
solving the problem in your case?
If so, I prefer to change to compiler_barrier() as volatile may have a
performance impact on fast-path code.
>
> + if (unlikely(cb != NULL)) {
> do {
> nb_rx = cb->fn.rx(port_id, queue_id, rx_pkts, nb_rx,
> nb_pkts, cb->param);
> @@ -4652,7 +4652,8 @@ rte_eth_tx_burst(uint16_t port_id, uint16_t queue_id,
> #endif
>
> #ifdef RTE_ETHDEV_RXTX_CALLBACKS
> - struct rte_eth_rxtx_callback *cb = dev->pre_tx_burst_cbs[queue_id];
> + struct rte_eth_rxtx_callback *volatile cb =
> + dev->pre_tx_burst_cbs[queue_id];
>
> if (unlikely(cb != NULL)) {
> do {
> --
> 2.17.1
>
