[dpdk-dev] [PATCH] net/bnx2x: enhance old ethdev close behavior

[dpdk-dev] [PATCH] net/bnx2x: enhance old ethdev close behavior

[Rasesh Mody]

Set RTE_ETH_DEV_CLOSE_REMOVE upon probe so all the private resources
for the port can be freed by rte_eth_dev_close(). With this change the
private port resources are released in the .dev_close callback.

Signed-off-by: Rasesh Mody <rmody@marvell.com>
---
 drivers/net/bnx2x/bnx2x_ethdev.c | 20 ++++++++++++++++++--
 1 file changed, 18 insertions(+), 2 deletions(-)

diff --git a/drivers/net/bnx2x/bnx2x_ethdev.c b/drivers/net/bnx2x/bnx2x_ethdev.c
index b2ea5fafa..00d76b9f8 100644
--- a/drivers/net/bnx2x/bnx2x_ethdev.c
+++ b/drivers/net/bnx2x/bnx2x_ethdev.c
@@ -295,6 +295,9 @@ bnx2x_dev_close(struct rte_eth_dev *dev)
 
 	/* free ilt */
 	bnx2x_free_ilt_mem(sc);
+
+	/* mac_addrs must not be freed alone because part of dev_private */
+	dev->data->mac_addrs = NULL;
 }
 
 static int
@@ -726,6 +729,11 @@ bnx2x_common_dev_init(struct rte_eth_dev *eth_dev, int is_vf)
 			goto out;
 	}
 
+	/* Pass the information to the rte_eth_dev_close() that it should also
+	 * release the private port resources.
+	 */
+	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
+
 	return 0;
 
 out:
@@ -753,8 +761,16 @@ eth_bnx2xvf_dev_init(struct rte_eth_dev *eth_dev)
 
 static int eth_bnx2x_dev_uninit(struct rte_eth_dev *eth_dev)
 {
-	/* mac_addrs must not be freed alone because part of dev_private */
-	eth_dev->data->mac_addrs = NULL;
+	struct bnx2x_softc *sc = eth_dev->data->dev_private;
+
+	PMD_INIT_FUNC_TRACE(sc);
+
+	/* only uninitialize in the primary process */
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+		return 0;
+
+	bnx2x_dev_close(eth_dev);
+
 	return 0;
 }
 
-- 
2.18.1

[dpdk-dev] [PATCH] net/qede: enhance old ethdev close behavior

[Rasesh Mody]

Set RTE_ETH_DEV_CLOSE_REMOVE upon probe so all the private resources
for the port can be freed by rte_eth_dev_close(). With this change the
private port resources are released in the .dev_close callback.

Signed-off-by: Rasesh Mody <rmody@marvell.com>
---
 drivers/net/qede/qede_ethdev.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/drivers/net/qede/qede_ethdev.c b/drivers/net/qede/qede_ethdev.c
index 59f1746ee..86a0ea26f 100644
--- a/drivers/net/qede/qede_ethdev.c
+++ b/drivers/net/qede/qede_ethdev.c
@@ -1586,6 +1586,10 @@ static void qede_dev_close(struct rte_eth_dev *eth_dev)
 
 	if (ECORE_IS_CMT(edev))
 		rte_eal_alarm_cancel(qede_poll_sp_sb_cb, (void *)eth_dev);
+
+	eth_dev->dev_ops = NULL;
+	eth_dev->rx_pkt_burst = NULL;
+	eth_dev->tx_pkt_burst = NULL;
 }
 
 static int
@@ -2701,6 +2705,11 @@ static int qede_common_dev_init(struct rte_eth_dev *eth_dev, bool is_vf)
 		adapter->ipgre.enable = false;
 	}
 
+	/* Pass the information to the rte_eth_dev_close() that it should also
+	 * release the private port resources.
+	 */
+	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
+
 	DP_INFO(edev, "MAC address : %02x:%02x:%02x:%02x:%02x:%02x\n",
 		adapter->primary_mac.addr_bytes[0],
 		adapter->primary_mac.addr_bytes[1],
@@ -2745,10 +2754,6 @@ static int qede_dev_common_uninit(struct rte_eth_dev *eth_dev)
 	/* safe to close dev here */
 	qede_dev_close(eth_dev);
 
-	eth_dev->dev_ops = NULL;
-	eth_dev->rx_pkt_burst = NULL;
-	eth_dev->tx_pkt_burst = NULL;
-
 	return 0;
 }
 
-- 
2.18.1

Re: [dpdk-dev] [PATCH] net/bnx2x: enhance old ethdev close behavior

[Thomas Monjalon]

25/09/2020 06:17, Rasesh Mody:
> Set RTE_ETH_DEV_CLOSE_REMOVE upon probe so all the private resources
> for the port can be freed by rte_eth_dev_close(). With this change the
> private port resources are released in the .dev_close callback.
> 
> Signed-off-by: Rasesh Mody <rmody@marvell.com>
> ---
>  static int eth_bnx2x_dev_uninit(struct rte_eth_dev *eth_dev)
>  {
> -	/* mac_addrs must not be freed alone because part of dev_private */
> -	eth_dev->data->mac_addrs = NULL;
> +	struct bnx2x_softc *sc = eth_dev->data->dev_private;
> +
> +	PMD_INIT_FUNC_TRACE(sc);
> +
> +	/* only uninitialize in the primary process */
> +	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
> +		return 0;
> +
> +	bnx2x_dev_close(eth_dev);

The check for primary process should be in the close function.

About the title, I would suggest net/bnx2x: release port upon close

Re: [dpdk-dev] [EXT] Re: [PATCH] net/bnx2x: enhance old ethdev close behavior

[Rasesh Mody]

Hi Thomas,

>From: Thomas Monjalon <thomas@monjalon.net>
>Sent: Friday, September 25, 2020 12:11 AM
>
>25/09/2020 06:17, Rasesh Mody:
>> Set RTE_ETH_DEV_CLOSE_REMOVE upon probe so all the private resources
>> for the port can be freed by rte_eth_dev_close(). With this change the
>> private port resources are released in the .dev_close callback.
>>
>> Signed-off-by: Rasesh Mody <rmody@marvell.com>
>> ---
>>  static int eth_bnx2x_dev_uninit(struct rte_eth_dev *eth_dev)  {
>> -	/* mac_addrs must not be freed alone because part of dev_private */
>> -	eth_dev->data->mac_addrs = NULL;
>> +	struct bnx2x_softc *sc = eth_dev->data->dev_private;
>> +
>> +	PMD_INIT_FUNC_TRACE(sc);
>> +
>> +	/* only uninitialize in the primary process */
>> +	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
>> +		return 0;
>> +
>> +	bnx2x_dev_close(eth_dev);
>
>The check for primary process should be in the close function.
>
PMDs check for primary process as part of PCI probe, hence it was put in as part of PCI remove. I also observed other PMDs implementing a similar scheme.
If it needs to be part of close function, I can move it there.

>About the title, I would suggest net/bnx2x: release port upon close
>
I'll change the title and resubmit.

Thanks!
-Rasesh

Re: [dpdk-dev] [EXT] Re: [PATCH] net/bnx2x: enhance old ethdev close behavior

[Thomas Monjalon]

25/09/2020 21:24, Rasesh Mody:
> Hi Thomas,
> 
> >From: Thomas Monjalon <thomas@monjalon.net>
> >Sent: Friday, September 25, 2020 12:11 AM
> >
> >25/09/2020 06:17, Rasesh Mody:
> >> Set RTE_ETH_DEV_CLOSE_REMOVE upon probe so all the private resources
> >> for the port can be freed by rte_eth_dev_close(). With this change the
> >> private port resources are released in the .dev_close callback.
> >>
> >> Signed-off-by: Rasesh Mody <rmody@marvell.com>
> >> ---
> >>  static int eth_bnx2x_dev_uninit(struct rte_eth_dev *eth_dev)  {
> >> -	/* mac_addrs must not be freed alone because part of dev_private */
> >> -	eth_dev->data->mac_addrs = NULL;
> >> +	struct bnx2x_softc *sc = eth_dev->data->dev_private;
> >> +
> >> +	PMD_INIT_FUNC_TRACE(sc);
> >> +
> >> +	/* only uninitialize in the primary process */
> >> +	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
> >> +		return 0;
> >> +
> >> +	bnx2x_dev_close(eth_dev);
> >
> >The check for primary process should be in the close function.
> >
> PMDs check for primary process as part of PCI probe, hence it was put in as part of PCI remove. I also observed other PMDs implementing a similar scheme.

Yes I am reviewing other PMDs to fix the same.

> If it needs to be part of close function, I can move it there.

Yes please.

> >About the title, I would suggest net/bnx2x: release port upon close
> >
> I'll change the title and resubmit.

Thanks