[ofa-general] Re: [PATCH] opensm: Bug in coding trying to set vl_arb_high_limit

[Sasha Khapyorsky]

Hi Eitan,

On 19:31 Wed 18 Jul     , Eitan Zahavi wrote:
> Hi Sasha
> 
> When QoS setup is done the code was trying to send updates of
> vl_arb_high_limit by req_set of PORT_INFO with the new data.
> However, at that stage the SM still did not assign LIDs to the ports.
> So the sent PortInfo.base_lid was still zero. The specification does not
> allow for such LIDs (they are considered ilegal). 
> 
> the patch below fixes this by storing the calculated value and later 
> using it in link and lid managers.

Good, Thanks (and this also saves one PortInfo update MAD). One question below:


> 
> Eitan
> 
> Signed-off-by:  Eitan Zahavi <eitan at mellanox.co.il>
> 

[snip...]

> diff --git a/opensm/opensm/osm_lid_mgr.c b/opensm/opensm/osm_lid_mgr.c
> index bc3f8b3..ed76382 100644
> --- a/opensm/opensm/osm_lid_mgr.c
> +++ b/opensm/opensm/osm_lid_mgr.c
> @@ -1182,6 +1182,14 @@ __osm_lid_mgr_set_physp_pi(
>             ib_port_info_get_port_state(p_old_pi) )
>          send_set = TRUE;
>      }
> +
> +	 /* provide the vl_high_limit from the qos mgr */
> +	 if (p_mgr->p_subn->opt.no_qos == FALSE) 
> +		 if (p_physp->vl_high_limit != p_old_pi->vl_high_limit)
> +		 {
> +			 send_set = TRUE;
> +			 p_pi->vl_high_limit = p_physp->vl_high_limit;
> +		 }

This part of code is for port_num != 0, so VLHighLimit setup will be
skipped for switch enhanced port 0. Is it something expected? If so why?

Sasha