[ofa-general] Re: [PATCH] opensm: update LFTs when entering master state

[Yevgeny Kliteynik]

Hi Sasha,

Sasha Khapyorsky wrote:
> Hi Yevgeny,
> 
> On 23:42 Sun 11 Jan     , Yevgeny Kliteynik wrote:
>> Hi Sasha,
>>
>> Sasha Khapyorsky wrote:
>>> When we are going to setup LFTs we need to ignore its previous images if
>>> OpenSM enters master after standby, so need to check for subnet
>>> need_update flag too.
>> Nice catch. I think there will be a similar problem with
>> cached routing too - need to invalidate the cache when SM
>> enters master state.
> 
> Yes, obviously. Maybe something like this will do:
> 
> 
> diff --git a/opensm/opensm/osm_state_mgr.c b/opensm/opensm/osm_state_mgr.c
> index 625e026..fc7ceb9 100644
> --- a/opensm/opensm/osm_state_mgr.c
> +++ b/opensm/opensm/osm_state_mgr.c
> @@ -1089,7 +1089,7 @@ static void do_sweep(osm_sm_t * sm)
>  	 */
>  	if (sm->p_subn->opt.use_ucast_cache &&
>  	    (sm->p_subn->subnet_initialization_error ||
> -	     sm->p_subn->force_reroute))
> +	     sm->p_subn->force_reroute || sm->p_subn->coming_out_of_standby))

Sure, looks like it would do the job.

>  		osm_ucast_cache_invalidate(&sm->ucast_mgr);
>  
>  	/*
> 
> 
> (and basically I think tah all those "flags flow" mess requires cleanup
> already :)).

Oh yes...

-- Yevgeny

> Sasha
>