examples/vhost: fix port init failure in mergeable mode

  When the example starts in mergeable mode with an i40e port,
it fails to launch because the examples use default mtu MAX_MTU
to configure ethdev. The root cause is some devices have Ethernet
frame overhead and then MAX_MTU will be larger than device's max
mtu, so the ethdev configure will fail.

This patch checks the device's max mtu before setting the ethdev
configuration. If the device has a max mtu, use that value to
configure.

Fixes: 1bb4a528c41f ("ethdev: fix max Rx packet length")

Signed-off-by: Chenbo Xia <chenbo.xia@intel.com>
Reported-by: Xingguang He <xingguang.he@intel.com>
---
 examples/vhost/main.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)
  

Hi Maxime,

> -----Original Message-----
> From: dev <dev-bounces@dpdk.org> On Behalf Of Chenbo Xia
> Sent: Thursday, November 4, 2021 1:11 PM
> To: dev@dpdk.org
> Cc: maxime.coquelin@redhat.com
> Subject: [dpdk-dev] [PATCH] examples/vhost: fix port init failure in mergeable
> mode
> 
> When the example starts in mergeable mode with an i40e port,
> it fails to launch because the examples use default mtu MAX_MTU
> to configure ethdev. The root cause is some devices have Ethernet
> frame overhead and then MAX_MTU will be larger than device's max
> mtu, so the ethdev configure will fail.
> 
> This patch checks the device's max mtu before setting the ethdev
> configuration. If the device has a max mtu, use that value to
> configure.
> 
> Fixes: 1bb4a528c41f ("ethdev: fix max Rx packet length")
> 
> Signed-off-by: Chenbo Xia <chenbo.xia@intel.com>
> Reported-by: Xingguang He <xingguang.he@intel.com>
> ---

I see a unit test failure on this patch, but I think this patch should not cause
issues in unit test, could you help me confirm it?

Thanks,
Chenbo
  

On 11/5/21 08:32, Xia, Chenbo wrote:
> Hi Maxime,
> 
>> -----Original Message-----
>> From: dev <dev-bounces@dpdk.org> On Behalf Of Chenbo Xia
>> Sent: Thursday, November 4, 2021 1:11 PM
>> To: dev@dpdk.org
>> Cc: maxime.coquelin@redhat.com
>> Subject: [dpdk-dev] [PATCH] examples/vhost: fix port init failure in mergeable
>> mode
>>
>> When the example starts in mergeable mode with an i40e port,
>> it fails to launch because the examples use default mtu MAX_MTU
>> to configure ethdev. The root cause is some devices have Ethernet
>> frame overhead and then MAX_MTU will be larger than device's max
>> mtu, so the ethdev configure will fail.
>>
>> This patch checks the device's max mtu before setting the ethdev
>> configuration. If the device has a max mtu, use that value to
>> configure.
>>
>> Fixes: 1bb4a528c41f ("ethdev: fix max Rx packet length")
>>
>> Signed-off-by: Chenbo Xia <chenbo.xia@intel.com>
>> Reported-by: Xingguang He <xingguang.he@intel.com>
>> ---
> 
> I see a unit test failure on this patch, but I think this patch should not cause
> issues in unit test, could you help me confirm it?

I can confirm, mcslock_autotest is not related to your change, and it 
seems other patches are facing the same issue, eg:
https://lab.dpdk.org/results/dashboard/patchsets/19997/

Maxime

> Thanks,
> Chenbo
>
  

On 11/4/21 06:11, Chenbo Xia wrote:
> When the example starts in mergeable mode with an i40e port,
> it fails to launch because the examples use default mtu MAX_MTU
> to configure ethdev. The root cause is some devices have Ethernet
> frame overhead and then MAX_MTU will be larger than device's max
> mtu, so the ethdev configure will fail.
> 
> This patch checks the device's max mtu before setting the ethdev
> configuration. If the device has a max mtu, use that value to
> configure.
> 
> Fixes: 1bb4a528c41f ("ethdev: fix max Rx packet length")
> 
> Signed-off-by: Chenbo Xia <chenbo.xia@intel.com>
> Reported-by: Xingguang He <xingguang.he@intel.com>
> ---
>   examples/vhost/main.c | 9 +++++++--
>   1 file changed, 7 insertions(+), 2 deletions(-)
> 
> diff --git a/examples/vhost/main.c b/examples/vhost/main.c
> index 58e12aa710..09fb2382bf 100644
> --- a/examples/vhost/main.c
> +++ b/examples/vhost/main.c
> @@ -273,6 +273,13 @@ port_init(uint16_t port)
>   
>   	tx_rings = (uint16_t)rte_lcore_count();
>   
> +	if (mergeable) {
> +		if (dev_info.max_mtu != UINT16_MAX && dev_info.max_rx_pktlen > dev_info.max_mtu)
> +			vmdq_conf_default.rxmode.mtu = dev_info.max_mtu;
> +		else
> +			vmdq_conf_default.rxmode.mtu = MAX_MTU;
> +	}
> +
>   	/* Get port configuration. */
>   	retval = get_eth_conf(&port_conf, num_devices);
>   	if (retval < 0)
> @@ -631,8 +638,6 @@ us_vhost_parse_args(int argc, char **argv)
>   				return -1;
>   			}
>   			mergeable = !!ret;
> -			if (ret)
> -				vmdq_conf_default.rxmode.mtu = MAX_MTU;
>   			break;
>   
>   		case OPT_STATS_NUM:
> 

Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>

Thanks,
Maxime
  

On 11/4/21 06:11, Chenbo Xia wrote:
> When the example starts in mergeable mode with an i40e port,
> it fails to launch because the examples use default mtu MAX_MTU
> to configure ethdev. The root cause is some devices have Ethernet
> frame overhead and then MAX_MTU will be larger than device's max
> mtu, so the ethdev configure will fail.
> 
> This patch checks the device's max mtu before setting the ethdev
> configuration. If the device has a max mtu, use that value to
> configure.
> 
> Fixes: 1bb4a528c41f ("ethdev: fix max Rx packet length")
> 
> Signed-off-by: Chenbo Xia <chenbo.xia@intel.com>
> Reported-by: Xingguang He <xingguang.he@intel.com>
> ---
>   examples/vhost/main.c | 9 +++++++--
>   1 file changed, 7 insertions(+), 2 deletions(-)
> 

Applied to dpdk-next-virtio/main.

Thanks,
Maxime