[dpdk-dev] [PATCH] net/i40e: stop lldp before setting local lldp MIB

[Zhang, Qi Z]

From: Chas Williams [mailto:3chas3 at gmail.com]
Sent: Wednesday, August 1, 2018 11:31 PM
To: Zhang, Qi Z <qi.z.zhang at intel.com>
Cc: dev at dpdk.org; Xing, Beilei <beilei.xing at intel.com>; Chas Williams <chas3 at att.com>
Subject: Re: [PATCH] net/i40e: stop lldp before setting local lldp MIB


On Wed, Aug 1, 2018 at 10:00 AM Zhang, Qi Z <qi.z.zhang at intel.com<mailto:qi.z.zhang at intel.com>> wrote:
Hi Williams:

> -----Original Message-----
> From: Chas Williams [mailto:3chas3 at gmail.com<mailto:3chas3 at gmail.com>]
> Sent: Wednesday, August 1, 2018 12:07 PM
> To: dev at dpdk.org<mailto:dev at dpdk.org>
> Cc: Xing, Beilei <beilei.xing at intel.com<mailto:beilei.xing at intel.com>>; Zhang, Qi Z <qi.z.zhang at intel.com<mailto:qi.z.zhang at intel.com>>;
> Charles (Chas) Williams <chas3 at att.com<mailto:chas3 at att.com>>
> Subject: [PATCH] net/i40e: stop lldp before setting local lldp MIB
>
> From: "Charles (Chas) Williams" <chas3 at att.com<mailto:chas3 at att.com>>
>
> From the Intel Ethernet Controller X710/XXV710/XL710 Specifiction
> Update:
>
>     Starting from NVM 5.02, if the Set Local LLDP MIB command is
>     received while the DCBx specific agent is stopped, the command
>     returns an EPERM error. If the command is received while the
>     LLDP agent is stopped, it sets the local MIB without exchanging
>     LLDP with peer, and returns SUCCESS.
>
> This results in the harmless, but annoying, diagnostic:
>
>     default dcb config fails. err = -53, aq_err = 1.
>
> So, always stop the lldp daemon when we are in software mode before we
> attempt to call i40e_set_dcb_config.
>
> Signed-off-by: Chas Williams <chas3 at att.com<mailto:chas3 at att.com>>
> ---
>  drivers/net/i40e/i40e_ethdev.c | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/drivers/net/i40e/i40e_ethdev.c b/drivers/net/i40e/i40e_ethdev.c
> index a340540ef..03bedf5c1 100644
> --- a/drivers/net/i40e/i40e_ethdev.c
> +++ b/drivers/net/i40e/i40e_ethdev.c
> @@ -11237,6 +11237,7 @@ i40e_dcb_init_configure(struct rte_eth_dev
> *dev, bool sw_dcb)
>                * i40e_init_dcb we expect is failure with I40E_AQ_RC_EPERM
>                * adminq status. Otherwise, it should return success.
>                */
> +             i40e_aq_stop_lldp(hw, TRUE, NULL);

I40e_aq_stop_lldp is intended to be removed with below two patches.

commit c6431c891d9e9691e3205fe5c5350071cbaeb852
commit fcbd40d4327b36725c4de9f33f57809edc359f4a

Yeah about that.  The i40e driver seems to go out of its way to program the flow
control outside of LLDP see i40e_update_flow_control() and i40e_flow_ctrl_set().
So it's not clear to me what is going on here with DPDK and LLDP.  Do you really
want LLDP running in this mode?  The other branch in i40e_dcb_init_configure()
explicitly starts LLDP, so I assume at some point, LLDP wasn't/isn't running.  How did
it get this way?  With respect to item #70, shouldn't you therefore always start
LLDP to workaround the bug?


LLDP is enabled by firmware as default,  from my view it is not necessary to start LLDP, if  We never stop it, and maybe we should not allow the case sw_dcb = TRUE since we can’t stop LLDP.

Item#70 is fixed in later NVM's (6.01 or later).  Perhaps a check to see if the NVM
is new enough to safely allow LLDP stopping?

Yes, I agree, would you mind to add a firmware check for lldp_start and lldp_stop in v2? So that fix your concern in the case with a new firmware
For old firmware case, let’s just keep exist way , there should be some re-work in future, and more test and review is required.

Thanks
Qi



Regards.
Qi


>               if ((ret == I40E_SUCCESS) || (ret != I40E_SUCCESS &&
>                   hw->aq.asq_last_status == I40E_AQ_RC_EPERM)) {
>                       memset(&hw->local_dcbx_config, 0,
> --
> 2.14.4