[dpdk-dev] [PATCH] ethdev: add queue state when retrieve queue information

[Andrew Rybchenko]

On 3/22/21 7:53 PM, Ananyev, Konstantin wrote:
> 
> 
>> -----Original Message-----
>> From: Andrew Rybchenko <andrew.rybchenko at oktetlabs.ru>
>> Sent: Monday, March 22, 2021 4:02 PM
>> To: Ananyev, Konstantin <konstantin.ananyev at intel.com>; Yigit, Ferruh <ferruh.yigit at intel.com>; Lijun Ou <oulijun at huawei.com>;
>> thomas at monjalon.net
>> Cc: dev at dpdk.org; linuxarm at openeuler.org; Andrew Rybchenko <arybchenko at solarflare.com>; David Marchand
>> <david.marchand at redhat.com>; Ray Kinsella <mdr at ashroe.eu>; Luca Boccassi <bluca at debian.org>
>> Subject: Re: [dpdk-dev] [PATCH] ethdev: add queue state when retrieve queue information
>>
>> On 3/22/21 6:45 PM, Ananyev, Konstantin wrote:
>>>
>>>
>>>> -----Original Message-----
>>>> From: dev <dev-bounces at dpdk.org> On Behalf Of Andrew Rybchenko
>>>> Sent: Monday, March 22, 2021 2:49 PM
>>>> To: Yigit, Ferruh <ferruh.yigit at intel.com>; Lijun Ou <oulijun at huawei.com>; thomas at monjalon.net
>>>> Cc: dev at dpdk.org; linuxarm at openeuler.org; Andrew Rybchenko <arybchenko at solarflare.com>; David Marchand
>>>> <david.marchand at redhat.com>; Ray Kinsella <mdr at ashroe.eu>; Luca Boccassi <bluca at debian.org>
>>>> Subject: Re: [dpdk-dev] [PATCH] ethdev: add queue state when retrieve queue information
>>>>
>>>> On 3/22/21 12:22 PM, Ferruh Yigit wrote:
>>>>> On 3/18/2021 12:25 PM, Lijun Ou wrote:
>>>>>> Currently, upper-layer application could get queue state only
>>>>>> through pointers such as dev->data->tx_queue_state[queue_id],
>>>>>> this is not the recommended way to access it. So this patch
>>>>>> add get queue state when call rte_eth_rx_queue_info_get and
>>>>>> rte_eth_tx_queue_info_get API.
>>>>>>
>>>>>> Note: The hairpin queue is not supported with above
>>>>>> rte_eth_*x_queue_info_get, so the queue state could be
>>>>>> RTE_ETH_QUEUE_STATE_STARTED or RTE_ETH_QUEUE_STATE_STOPPED.
>>>>>> Note: After add queue_state field, the 'struct rte_eth_rxq_info' size
>>>>>> remains 128B, and the 'struct rte_eth_txq_info' size remains 64B, so
>>>>>> it could be ABI compatible.
>>>>>>
>>>>>> Signed-off-by: Chengwen Feng <fengchengwen at huawei.com>
>>>>>> Signed-off-by: Lijun Ou <oulijun at huawei.com>
>>>>>
>>>>> <...>
>>>>>
>>>>>> diff --git a/lib/librte_ethdev/rte_ethdev.h
>>>>>> b/lib/librte_ethdev/rte_ethdev.h
>>>>>> index efda313..3b83c5a 100644
>>>>>> --- a/lib/librte_ethdev/rte_ethdev.h
>>>>>> +++ b/lib/librte_ethdev/rte_ethdev.h
>>>>>> @@ -1591,6 +1591,8 @@ struct rte_eth_rxq_info {
>>>>>>       uint8_t scattered_rx;       /**< scattered packets RX supported. */
>>>>>>       uint16_t nb_desc;           /**< configured number of RXDs. */
>>>>>>       uint16_t rx_buf_size;       /**< hardware receive buffer size. */
>>>>>> +    /**< Queues state: STARTED(1) / STOPPED(0). */
>>>>>> +    uint8_t queue_state;
>>>>>>   } __rte_cache_min_aligned;
>>>>>>     /**
>>>>>> @@ -1600,6 +1602,8 @@ struct rte_eth_rxq_info {
>>>>>>   struct rte_eth_txq_info {
>>>>>>       struct rte_eth_txconf conf; /**< queue config parameters. */
>>>>>>       uint16_t nb_desc;           /**< configured number of TXDs. */
>>>>>> +    /**< Queues state: STARTED(1) / STOPPED(0). */
>>>>>> +    uint8_t queue_state;
>>>>>>   } __rte_cache_min_aligned;
>>>>>>     /* Generic Burst mode flag definition, values can be ORed. */
>>>>>>
>>>>>
>>>>> This is causing an ABI warning [1], but I guess it is safe since the
>>>>> size of the struct is not changing (cache align). Adding a few more
>>>>> people to comment.
>>>>>
>>>>>
>>>>> [1]
>>>>> https://travis-ci.com/github/ovsrobot/dpdk/builds/220497651
>>>>
>>>> Frankly speaking I dislike addition of queue_state as uint8_t.
>>>> IMHO it should be either 'bool started' or enum to support more
>>>> states in the future if we need.
>>>
>>> I think we already have set of defines for it:
>>> lib/librte_ethdev/rte_ethdev_driver.h:925:#define RTE_ETH_QUEUE_STATE_STOPPED 0
>>> lib/librte_ethdev/rte_ethdev_driver.h:926:#define RTE_ETH_QUEUE_STATE_STARTED 1
>>> lib/librte_ethdev/rte_ethdev_driver.h:927:#define RTE_ETH_QUEUE_STATE_HAIRPIN 2
>>>
>>> If we want to publish it, then might be enough just move these macros to rte_ethdev.h or so.
>>>
>>> About uint8_t vs enum - yes, in principle enum would be a bit nicer,
>>> but right now rte_eth_dev_data.(rx|tx)_queue_state[]  itself is an array of uint8_t.
>>> So probably not much point to waste extra 3B in rte_eth_(rxq|txq)_info.
>>> Unless in future will want to change it in struct rte_eth_dev_data too
>>> (or even hide it inside dev private queue data).
>>
>> I forgot about hairpin and bitmask... If so, I think it is
>> sufficient to fix absolutely misleading comment, say
>> that it is a bit mask and think about removal of
>> RTE_ETH_QUEUE_STATE_STOPPED (since it could be
>> stopped+hairpin). May be consider to use uin16_t,
>> since 8 bit is really small bitmask. It still fits in
>> available hole.
> 
> Hmm, as I can read the code - hairpin queue can't be started/stopped by SW,
> and each of the states (stopped/started/hairpin) is mutually exclusive.
> Is that not what was intended when hairpin queues were introduced?
> 

Thanks, yes, you're right. My memory lies to me. If queue state
is not a bit mask, it should be an enum from API point of view.
Rx/Tx queue info structures are control path. I see no point to
save bits here. Clear API is more important on control path.
The only reason here to use uint8_t is to avoid ABI breakage.
I can't judge if it is critical to wait or not.