net/octeontx: fix port close

Segmentation fault has been observed while closing the ethernet
port. Reason for the segfault is, eth port close also shuts down
event device while other ethernet port is still using the event
device.

Fixes: da6c687471a3 ("net/octeontx: add start and stop support")
Cc: stable@dpdk.org
Signed-off-by: Harman Kalra <hkalra@marvell.com>

diff --git a/drivers/net/octeontx/octeontx_ethdev.c b/drivers/net/octeontx/octeontx_ethdev.c
index f5ea9de8efe571ebad69b3476dd96b8ac07f8a53..6469fd0a9627821f2e94e14f92c8f3c1edf7eb38 100644 (file)
--- a/drivers/net/octeontx/octeontx_ethdev.c
+++ b/drivers/net/octeontx/octeontx_ethdev.c
@@ -26,6 +26,11 @@
 #include "octeontx_rxtx.h"
 #include "octeontx_logs.h"
 
+/* Useful in stopping/closing event device if no of
+ * eth ports are using it.
+ */
+uint16_t evdev_refcnt;
+
 struct evdev_priv_data {
        OFFLOAD_FLAGS; /*Sequence should not be changed */
 } __rte_cache_aligned;
@@ -491,7 +496,11 @@ octeontx_dev_close(struct rte_eth_dev *dev)
        if (rte_eal_process_type() != RTE_PROC_PRIMARY)
                return 0;
 
-       rte_event_dev_close(nic->evdev);
+       /* Stopping/closing event device once all eth ports are closed. */
+       if (__atomic_sub_fetch(&evdev_refcnt, 1, __ATOMIC_ACQUIRE) == 0) {
+               rte_event_dev_stop(nic->evdev);
+               rte_event_dev_close(nic->evdev);
+       }
 
        octeontx_dev_flow_ctrl_fini(dev);
 
@@ -671,8 +680,6 @@ octeontx_dev_stop(struct rte_eth_dev *dev)
 
        PMD_INIT_FUNC_TRACE();
 
-       rte_event_dev_stop(nic->evdev);
-
        ret = octeontx_port_stop(nic);
        if (ret < 0) {
                octeontx_log_err("failed to req stop port %d res=%d",
@@ -1333,6 +1340,7 @@ octeontx_create(struct rte_vdev_device *dev, int port, uint8_t evdev,
        nic->pko_vfid = pko_vfid;
        nic->port_id = port;
        nic->evdev = evdev;
+       __atomic_add_fetch(&evdev_refcnt, 1, __ATOMIC_ACQUIRE);
 
        res = octeontx_port_open(nic);
        if (res < 0)
@@ -1582,6 +1590,7 @@ octeontx_probe(struct rte_vdev_device *dev)
                }
        }
 
+       __atomic_store_n(&evdev_refcnt, 0, __ATOMIC_RELEASE);
        /*
         * Do 1:1 links for ports & queues. All queues would be mapped to
         * one port. If there are more ports than queues, then some ports