[dpdk-dev] [PATCH v2 07/30] net/mlx5: fix reta update can segfault

[Nelio Laranjeiro]

Reta update needs to stop/start the port but stopping the port does not
disable the polling functions which may end in a segfault if a core is
polling the queue while the control thread is modifying it.

This patch changes the sequences to an order where such situation cannot
happen.

Fixes: aa13338faf5e ("net/mlx5: rebuild flows on updating RETA")
Cc: ys...@mellanox.com
Cc: sta...@dpdk.org

Signed-off-by: Nelio Laranjeiro <nelio.laranje...@6wind.com>
---
 drivers/net/mlx5/mlx5_rss.c     | 9 +++++----
 drivers/net/mlx5/mlx5_trigger.c | 7 +++++++
 2 files changed, 12 insertions(+), 4 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_rss.c b/drivers/net/mlx5/mlx5_rss.c
index d3d2603..8942879 100644
--- a/drivers/net/mlx5/mlx5_rss.c
+++ b/drivers/net/mlx5/mlx5_rss.c
@@ -351,11 +351,12 @@ mlx5_dev_rss_reta_update(struct rte_eth_dev *dev,
        struct priv *priv = dev->data->dev_private;
 
        assert(!mlx5_is_secondary());
-       mlx5_dev_stop(dev);
        priv_lock(priv);
        ret = priv_dev_rss_reta_update(priv, reta_conf, reta_size);
        priv_unlock(priv);
-       if (ret)
-               return -ret;
-       return mlx5_dev_start(dev);
+       if (dev->data->dev_started) {
+               mlx5_dev_stop(dev);
+               mlx5_dev_start(dev);
+       }
+       return -ret;
 }
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index 212b4df..eeb9585 100644
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -30,6 +30,7 @@
  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
+#include <unistd.h>
 
 #include <rte_ether.h>
 #include <rte_ethdev.h>
@@ -118,6 +119,12 @@ mlx5_dev_stop(struct rte_eth_dev *dev)
                return;
 
        priv_lock(priv);
+       dev->data->dev_started = 0;
+       /* Prevent crashes when queues are still in use. */
+       dev->rx_pkt_burst = removed_rx_burst;
+       dev->tx_pkt_burst = removed_tx_burst;
+       rte_wmb();
+       usleep(1000 * priv->rxqs_n);
        DEBUG("%p: cleaning up and destroying hash RX queues", (void *)dev);
        priv_special_flow_disable_all(priv);
        priv_mac_addrs_disable(priv);
-- 
2.1.4