[dpdk-dev] [PATCH] ethdev: stop the device before close

[Andrew Rybchenko]

From: Ivan Ilchenko <ivan.ilchenko at oktetlabs.ru>

In drivers important cleanup could happen on the device stop.
Do stop in the rte_eth_dev_close() function for robustness and
to simplify drivers code.

Signed-off-by: Ivan Ilchenko <ivan.ilchenko at oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <andrew.rybchenko at oktetlabs.ru>
---
In fact the patch is required to fix segfault in the case of
net/virtio on close without stop after Rx interrupts enabled.

I believe that the right way to address the problem is automated
stop from close, but I guess it cannot not be backported and
may be fix in a different way required in stable branches.

 lib/ethdev/rte_ethdev.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/lib/ethdev/rte_ethdev.c b/lib/ethdev/rte_ethdev.c
index a151c05849..b9f0938f20 100644
--- a/lib/ethdev/rte_ethdev.c
+++ b/lib/ethdev/rte_ethdev.c
@@ -1894,6 +1894,17 @@ rte_eth_dev_close(uint16_t port_id)
 	dev = &rte_eth_devices[port_id];
 
 	RTE_FUNC_PTR_OR_ERR_RET(*dev->dev_ops->dev_close, -ENOTSUP);
+
+	if (dev->data->dev_started) {
+		*lasterr = rte_eth_dev_stop(port_id);
+		if (*lasterr != 0) {
+			RTE_ETHDEV_LOG(ERR,
+				"Failed to stop device (port %u) before close: %s - ignore\n",
+				port_id, rte_strerror(-*lasterr));
+			lasterr = &binerr;
+		}
+	}
+
 	*lasterr = (*dev->dev_ops->dev_close)(dev);
 	if (*lasterr != 0)
 		lasterr = &binerr;
-- 
2.30.2