Re: [PATCH v3 1/4] ethdev: introduce ethdev HW desc dump PI

[Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>]

Cc various driver maintainers looking for opinion about API discussion
below

On 6/7/22 16:59, Dongdong Liu wrote:
> Hi Andrew
> 
> Many thanks for your review.
> 
> On 2022/6/2 3:53, Andrew Rybchenko wrote:
>> On 6/1/22 10:49, Min Hu (Connor) wrote:
>>> Added the ethdev HW Rx desc dump API which provides functions for query
>>> HW descriptor from device. HW descriptor info differs in different NICs.
>>> The information demonstrates I/O process which is important for debug.
>>> As the information is different between NICs, the new API is introduced.
>>>
>>> Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
>>> ---
>>>   doc/guides/rel_notes/release_22_07.rst |  7 ++++
>>>   lib/ethdev/ethdev_driver.h             | 42 ++++++++++++++++++++++++
>>>   lib/ethdev/rte_ethdev.c                | 44 ++++++++++++++++++++++++++
>>>   lib/ethdev/rte_ethdev.h                | 44 ++++++++++++++++++++++++++
>>>   lib/ethdev/version.map                 |  2 ++
>>>   5 files changed, 139 insertions(+)
>>>
>>> diff --git a/doc/guides/rel_notes/release_22_07.rst
>>> b/doc/guides/rel_notes/release_22_07.rst
>>> index 8932a1d478..56c675121a 100644
>>> --- a/doc/guides/rel_notes/release_22_07.rst
>>> +++ b/doc/guides/rel_notes/release_22_07.rst
>>> @@ -137,6 +137,13 @@ New Features
>>>     * ``RTE_EVENT_QUEUE_ATTR_WEIGHT``
>>>     * ``RTE_EVENT_QUEUE_ATTR_AFFINITY``
>>>   +* **Added ethdev HW desc dump API, to dump Rx/Tx HW desc info from
>>> device.**
>>> +
>>> +  Added the ethdev HW Rx desc dump API which provides functions for
>>> query
>>> +  HW descriptor from device. HW descriptor info differs in different
>>> NICs.
>>> +  The information demonstrates I/O process which is important for 
>>> debug.
>>> +  As the information is different between NICs, the new API is
>>> introduced.
>>> +
>>>     Removed Items
>>>   -------------
>>> diff --git a/lib/ethdev/ethdev_driver.h b/lib/ethdev/ethdev_driver.h
>>> index 69d9dc21d8..9c1726eb2d 100644
>>> --- a/lib/ethdev/ethdev_driver.h
>>> +++ b/lib/ethdev/ethdev_driver.h
>>> @@ -1073,6 +1073,42 @@ typedef int
>>> (*eth_ip_reassembly_conf_set_t)(struct rte_eth_dev *dev,
>>>    */
>>>   typedef int (*eth_dev_priv_dump_t)(struct rte_eth_dev *dev, FILE
>>> *file);
>>>   +/**
>>> + * @internal
>>> + * Dump ethdev Rx descriptor info to a file.
>>
>> ethdev is not required in the description above. It is an ethdev
>> operation type.
> 
> Will remove it.
>>
>> Is there any limitations on caller context? Should be it the same
>> CPU core (since typically it is the case for per-queue API) which
>> polls the queue? The answer must be in the API function description.
> 
> "[PATCH v3 4/4] app/procinfo: support descriptor dump" shows how to use
> the API. It is used for debug, not a dataplane API,  no special
> limitations on caller context.

Please, be explicit in the ethdev API documentation.

>>
>>> + *
>>> + * @param file
>>> + *   A pointer to a file for output.
>>> + * @param dev
>>> + *   Port (ethdev) handle.
>>> + * @param queue_id
>>> + *   The selected queue.
>>> + * @param desc_id
>>> + *   The selected descriptor.
>>
>> Is it an ID in the ring regardless of the current position or
>> is it offset relative to current position in the ring?
>> It should be clarified in any case.
>> I'd say that it should be an offset to be consistent with
>> descriptor status API.
> 
> It is an ID in the ring regardless of the current position.

IMHO, it is inconvenient since typically it is interesting
what happens nearby current position.

>>
>> It would be useful to be able to dump many descriptor at once.
> This can be done by the appliacation.

Yes, that's true, but easy to do in the driver as well. It would
be especially important if ID semantics changes.

>>
>>> + * @return
>>> + *   Negative errno value on error, zero on success.
>>> + */
>>> +typedef int (*eth_rx_hw_desc_dump_t)(FILE *file, const struct
>>> rte_eth_dev *dev,
>>> +                     uint16_t queue_id, uint16_t desc_id);
>>> +
>>> +/**
>>> + * @internal
>>> + * Dump ethdev Tx descriptor info to a file.
>>> + *
>>> + * @param file
>>> + *   A pointer to a file for output.
>>> + * @param dev
>>> + *   Port (ethdev) handle.
>>> + * @param queue_id
>>> + *   The selected queue.
>>> + * @param desc_id
>>> + *   The selected descriptor.
>>> + * @return
>>> + *   Negative errno value on error, zero on success.
>>> + */
>>> +typedef int (*eth_tx_hw_desc_dump_t)(FILE *file, const struct
>>> rte_eth_dev *dev,
>>> +                     uint16_t queue_id, uint16_t desc_id);
>>> +
>>>   /**
>>>    * @internal A structure containing the functions exported by an
>>> Ethernet driver.
>>>    */
>>> @@ -1283,6 +1319,12 @@ struct eth_dev_ops {
>>>         /** Dump private info from device */
>>>       eth_dev_priv_dump_t eth_dev_priv_dump;
>>> +
>>> +    /** Dump ethdev Rx descriptor info */\\
>>
>> It is an ethdev operations. So, 'ethdev' is not necessary in the
>> description.
> Will fix.
>>
>>> +    eth_rx_hw_desc_dump_t eth_rx_hw_desc_dump;
>>> +
>>> +    /** Dump ethdev Tx descriptor info */
>>> +    eth_tx_hw_desc_dump_t eth_tx_hw_desc_dump;
>>>   };
>>>     /**
>>> diff --git a/lib/ethdev/rte_ethdev.c b/lib/ethdev/rte_ethdev.c
>>> index 46c088dc88..bbd8439fa0 100644
>>> --- a/lib/ethdev/rte_ethdev.c
>>> +++ b/lib/ethdev/rte_ethdev.c
>>> @@ -5876,6 +5876,50 @@ rte_eth_dev_priv_dump(uint16_t port_id, FILE
>>> *file)
>>>       return eth_err(port_id, (*dev->dev_ops->eth_dev_priv_dump)(dev,
>>> file));
>>>   }
>>>   +int
>>> +rte_eth_rx_hw_desc_dump(FILE *file, uint16_t port_id, uint16_t 
>>> queue_id,
>>> +            uint16_t desc_id)
>>> +{
>>> +    struct rte_eth_dev *dev;
>>> +    int ret;
>>> +
>>> +    RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -ENODEV);
>>> +    dev = &rte_eth_devices[port_id];
>>
>> Shouldn't we check file vs null?
> Yes, will do.
>>
>>> +
>>> +    if (queue_id >= dev->data->nb_rx_queues) {
>>> +        RTE_ETHDEV_LOG(ERR, "Invalid Rx queue_id=%u\n", queue_id);
>>> +        return -EINVAL;
>>> +    }
>>
>> Don't we need a check vs queue size?
> 
> This need to be done in pmd as "[PATCH v3 3/4] net/hns3: support Rx/Tx 
> bd dump" shows.

All applicable generic checks must be done on ethdev level to avoid code
duplication and missing checks in PMDs.

Andrew.