[ofa-general] Re: [PATCH v2] opensm: free lft_buf if it matches switch's lft

[Yevgeny Kliteynik]

Sasha Khapyorsky wrote:
> Hi Yevgeny,
> 
> On 18:03 Thu 30 Oct     , Yevgeny Kliteynik wrote:
>> Sasha,
>>
>> This patch frees the switch's lft_buf if it matches the
>> LFT that is currently configured on switch.
>>
>> Signed-off-by: Yevgeny Kliteynik <kliteyn at dev.mellanox.co.il>
> 
> I applied this. Thanks. However one question below.
> 
> [snip...]
> 
>> diff --git a/opensm/opensm/osm_ucast_mgr.c b/opensm/opensm/osm_ucast_mgr.c
>> index adb6688..175817c 100644
>> --- a/opensm/opensm/osm_ucast_mgr.c
>> +++ b/opensm/opensm/osm_ucast_mgr.c
>> @@ -392,6 +392,20 @@ int osm_ucast_mgr_set_fwd_table(IN osm_ucast_mgr_t * const p_mgr,
>>  	context.lft_context.node_guid = osm_node_get_node_guid(p_node);
>>  	context.lft_context.set_method = TRUE;
>>
>> +	if (!p_sw->lft_buf) {
>> +		/* any routing should provide the lft_buf */
>> +		CL_ASSERT(p_mgr->p_subn->opt.use_ucast_cache &&
>> +			  p_mgr->cache_valid && !p_sw->need_update);
>> +		goto Exit;
>> +	}
>> +
>> +	if (!p_sw->need_update &&
>> +	    !memcmp(p_sw->lft, p_sw->lft_buf, IB_LID_UCAST_END_HO + 1)) {
>> +		free(p_sw->lft_buf);
>> +		p_sw->lft_buf = NULL;
>> +		goto Exit;
>> +	}
>> +
> 
> So buffers are freed only on next routing iteration (heavy sweep). Isn't
> it better to drop it when LFT images from switches are received in
> osm_lin_fwd_rcv.c?

Sure, why not. That way the memory would be freed faster.

-- Yevgeny

> Sasha
>