[dpdk-dev] [PATCH v2] app/testpmd: optimize membuf pool allocation

Burakov, Anatoly anatoly.burakov at intel.com
Mon Oct 8 13:35:30 CEST 2018


On 08-Oct-18 12:33 PM, Ferruh Yigit wrote:
> On 9/12/2018 2:54 AM, dev-bounces at dpdk.org wrote:
>> By default, testpmd will create membuf pool for all NUMA nodes and
>> ignore EAL configuration.
>>
>> Count the number of available NUMA according to EAL core mask or core
>> list configuration. Optimized by only creating membuf pool for those
>> nodes.
>>
>> Fixes: c9cafcc ("app/testpmd: fix mempool creation by socket id")
>>
>> Signed-off-by: Phil Yang <phil.yang at arm.com>
>> Acked-by: Gavin Hu <gavin.hu at arm.com>
>> ---
>>   app/test-pmd/testpmd.c | 4 ++--
>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
>> index ee48db2..a56af2b 100644
>> --- a/app/test-pmd/testpmd.c
>> +++ b/app/test-pmd/testpmd.c
>> @@ -476,6 +476,8 @@ set_default_fwd_lcores_config(void)
>>   
>>   	nb_lc = 0;
>>   	for (i = 0; i < RTE_MAX_LCORE; i++) {
>> +		if (!rte_lcore_is_enabled(i))
>> +			continue;
>>   		sock_num = rte_lcore_to_socket_id(i);
>>   		if (new_socket_id(sock_num)) {
>>   			if (num_sockets >= RTE_MAX_NUMA_NODES) {
>> @@ -485,8 +487,6 @@ set_default_fwd_lcores_config(void)
>>   			}
>>   			socket_ids[num_sockets++] = sock_num;
>>   		}
>> -		if (!rte_lcore_is_enabled(i))
>> -			continue;
>>   		if (i == rte_get_master_lcore())
>>   			continue;
>>   		fwd_lcores_cpuids[nb_lc++] = i;
>>
> 
> 
> This is causing testpmd fail for the case all cores from socket 1 and added a
> virtual device which will try to allocate memory from socket 0.
> 
> 
>   $ testpmd -l<cores from socket 1> --vdev net_pcap0,iface=lo -- -i
>   ...
>   Failed to setup RX queue:No mempool allocation on the socket 0
>   EAL: Error - exiting with code: 1
>     Cause: Start ports failed
> 
> 

It's an open question as to why pcap driver tries to allocate on socket 
0 when everything is on socket 1, but perhaps a better improvement would 
be to take into account not only socket ID's of lcores, but ethdev 
devices as well?

-- 
Thanks,
Anatoly


More information about the dev mailing list