[dpdk-dev] [PATCH 01/11] avp: implement dynamic logging

[Ferruh Yigit]

On 12/20/2017 10:58 AM, Stephen Hemminger wrote:
> On Tue, 19 Dec 2017 17:53:50 -0800
> Ferruh Yigit <ferruh.yigit at intel.com> wrote:
> 
>> On 12/18/2017 10:38 PM, Stephen Hemminger wrote:
>>> All PMD should be using dynamic log levels.
>>>
>>> Signed-off-by: Stephen Hemminger <stephen at networkplumber.org>

<...>

>>> @@ -2312,3 +2313,12 @@ avp_dev_stats_reset(struct rte_eth_dev *eth_dev)
>>>  
>>>  RTE_PMD_REGISTER_PCI(net_avp, rte_avp_pmd);
>>>  RTE_PMD_REGISTER_PCI_TABLE(net_avp, pci_id_avp_map);
>>> +
>>> +RTE_INIT(avp_init_log);
>>> +static void
>>> +avp_init_log(void)
>>> +{
>>> +	avp_logtype_driver = rte_log_register("pmd.avp.driver");
>>> +	if (avp_logtype_driver >= 0)
>>> +		rte_log_set_level(avp_logtype_driver, RTE_LOG_NOTICE);
>>> +}  
>>
>> This can be done later as well but what do you think creating a macro [1], so
>> that driver part can become just:
>>
>> PMD_INIT_LOG(avp, init, NOTICE);
>>
>>
>>
>> [1] something like:
>>  #define INIT_LOG_VAR_NAME(pmd, type)   logtype_ ## pmd ## _ ## type
>>  #define INIT_LOG_FUNC_NAME(pmd, type)  log_ ## pmd ## _ ## type
>>
>>  #define PMD_INIT_LOG(pmd, type, level)                                \
>>         int INIT_LOG_VAR_NAME(pmd, type);                       \
>>         RTE_INIT(INIT_LOG_FUNC_NAME(pmd, type));                \
>>         static void INIT_LOG_FUNC_NAME(pmd, type)(void)                 \
>>         {                                                       \
>>                 INIT_LOG_VAR_NAME(pmd, type) = rte_log_register("pmd."
>> RTE_STR(pmd) "." RTE_STR(type)); \
>>                 if (INIT_LOG_VAR_NAME(pmd, type) > 0)           \
>>                         rte_log_set_level(INIT_LOG_VAR_NAME(pmd, type),
>> RTE_LOG_##level); \
>>         }
> 
> That macro is a little complex.  Also, for better or worse, the current
> logging is done on a per driver basis. If we want to do something fancier
> it should be in common EAL core.

Of course, my intention was putting it into rte_log.h so updates in each driver
will be minimal. But this can be done better to cover library updates as well.