[dpdk-dev] [PATCH V9 1/3] ethdev: introduce FEC API

Andrew Rybchenko arybchenko at solarflare.com
Tue Sep 22 14:18:57 CEST 2020


On 9/22/20 2:06 PM, Min Hu (Connor) wrote:
> 
> 
> 在 2020/9/22 16:02, Andrew Rybchenko 写道:
>> On 9/22/20 7:58 AM, Min Hu (Connor) wrote:
>>>
>>>
>>> 在 2020/9/21 21:39, Andrew Rybchenko 写道:
>>>> On 9/21/20 9:13 AM, Min Hu (Connor) wrote:
>>>>> This patch adds Forward error correction(FEC) support for ethdev.
>>>>> Introduce APIs which support query and config FEC information in
>>>>> hardware.
>>>>>
>>>>> Signed-off-by: Min Hu (Connor) <humin29 at huawei.com>
>>>>> Reviewed-by: Wei Hu (Xavier) <xavier.huwei at huawei.com>
>>>>> Reviewed-by: Chengwen Feng <fengchengwen at huawei.com>
>>>>> Reviewed-by: Chengchang Tang <tangchengchang at huawei.com>
>>>>> Reviewed-by: Ajit Khaparde <ajit.khaparde at broadcom.com>
>>>>> Acked-by: Konstantin Ananyev <konstantin.ananyev at intel.com>

[snip]

>>>>> @@ -3328,6 +3349,70 @@ int  rte_eth_led_on(uint16_t port_id);
>>>>>    int  rte_eth_led_off(uint16_t port_id);
>>>>>      /**
>>>>> + * @warning
>>>>> + * @b EXPERIMENTAL: this API may change, or be removed, without
>>>>> prior notice
>>>>> + *
>>>>> + * Get Forward Error Correction(FEC) capability.
>>>>> + *
>>>>> + * @param port_id
>>>>> + *   The port identifier of the Ethernet device.
>>>>> + * @param fec_cap
>>>>> + *   returns the FEC capability from the device, as follows:
>>>>> + *   RTE_ETH_FEC_CAPA_NOFEC
>>>>> + *   RTE_ETH_FEC_CAPA_AUTO
>>>>> + *   RTE_ETH_FEC_CAPA_BASER
>>>>> + *   RTE_ETH_FEC_CAPA_RS
>>>>> + * @return
>>>>> + *   - (0) if successful.
>>>>> + *   - (-ENOTSUP) if underlying hardware OR driver doesn't support.
>>>>> + *     that operation.
>>>>> + *   - (-EIO) if device is removed.
>>>>> + *   - (-ENODEV)  if *port_id* invalid.
>>>>> + */
>>>>> +__rte_experimental
>>>>> +int rte_eth_fec_get_capability(uint16_t port_id, uint32_t *fec_cap);
>>>>
>>>> The API does not allow to report capabilities per link speed:
>>>> which FEC mode is supported at which link speed?
>>>>
>>>> What about something like:
>>>>
>>>> struct rte_eth_fec_capa {
>>>>     uint32_t speed; /**< Link speed (see ETH_SPEED_NUM_*) */
>>>>     uint32_t capa;  /**< FEC capabilities bitmask (see
>>>> RTE_FEC_CAPA_*) */
>>>> };
>>>>
>>>> __rte_experimental
>>>> int rte_eth_fec_get_capability(uint16_t port_id, uint32_t *num, struct
>>>> rte_eth_fec_capa *speed_capa);
>>>>
>>>> where:
>>>>    - num is in/out with a number of elements in an array
>>>>    - speed_capa is out only with per-speed capabilities
>>>>
>>> There is no need to report capabilities per link speed, because
>>> relastionship between the link speed and fec mode is fixed. The
>>> infomations can be referred to in official documents or internet.
>>
>> Should an application download documents and search for it? :)
> 
> OK, I will report capabilities per link speed in V11.
> By the way,
>>>> where:
>>>>    - num is in/out with a number of elements in an array
> 
> could you describe "num" more detailedly, how to use this value?

On input, num should specify a number of elements in speed_capa
array provided by the caller to get FEC capabilities.
If the number is insufficient to, error should be returned and
the number should contain required number of elements.
If sufficient, on output driver should return a number of
filled in array elements.

>>
>>>
>>> A ethernet port may have various link speed in diffrent situations(for
>>> example, optical module with different speed is used). But we do not
>>> care about capabilities per link speed. We only care about FEC capa of
>>> the ethernet device at a specific moment, because set FEC mode also
>>> depend on the current FEC capa.
>>
>> Capabilities should not be for a specific moment. Capabilities
>> should be fixed and stable (if transceiver is not replaced).
>> Capabilities should not depend on a link speed or link status.
>> Otherwise an application can't use it in a reliable way.
>>
>>>
>>> By the way, we can also get link speed of the device by API
>>> "rte_eth_link_get" in the same time.
>>>
>>> thanks.
>>>
>>>>> +
>>>>> +/**
>>>>> + * @warning
>>>>> + * @b EXPERIMENTAL: this API may change, or be removed, without
>>>>> prior notice
>>>>> + *
>>>>> + * Get current Forward Error Correction(FEC) mode.
>>>>> + *
>>>>> + * @param port_id
>>>>> + *   The port identifier of the Ethernet device.
>>>>> + * @param mode
>>>>> + *   returns the FEC mode from the device.
>>>>> + * @return
>>>>> + *   - (0) if successful.
>>>>> + *   - (-ENOTSUP) if underlying hardware OR driver doesn't support.
>>>>> + *     that operation.
>>>>> + *   - (-EIO) if device is removed.
>>>>> + *   - (-ENODEV)  if *port_id* invalid.
>>>>> + */
>>>>> +__rte_experimental
>>>>> +int rte_eth_fec_get(uint16_t port_id, enum rte_eth_fec_mode *mode);
>>>>
>>>> Please, specify what should be reported if link is down.
>>>> E.g. if set to RS, but link is down.
>>>>
>>>> Does AUTO make sense here?
>>>>
>>> OK, I will add the information in the function header comment:
>>> when link down,None AUTO mode(RS, BASER. NOFEC) keep as it is when link
>>> up, AUTO mode will change from rs,baser to nofec when quering the mode.
>>
>> I'll take a look at the patch, above text is hardly readable.

> (1). If the current mode of device is one of these modes:
> RS, BASER. NOFEC.
> when link up, for example, the mode is RS. when the device is linked
> down, the mode is always RS.

> (2). If the current mode of device is AUTO:
> when the device is linked down, the mode varies in this order:
> rs->baser->nofec, until Auto-negotiation success (link shoud be
> up first).
> 
> But this is defined in our hardware. I think the first feature(1) are
> common and can be adopted.

Sorry, I don't understand.
What abgot if link is down and AUTO is enabled, AUTO is
returned, otherwise, configured FEC mode is returned.
If link is up, current FEC mode is returned (i.e. not AUTO,
either NOFEC, or RS, or BASER).

>>>>> +
>>>>> +/**
>>>>> + * @warning
>>>>> + * @b EXPERIMENTAL: this API may change, or be removed, without
>>>>> prior notice
>>>>> + *
>>>>> + * Set Forward Error Correction(FEC) mode.
>>>>> + *
>>>>> + * @param port_id
>>>>> + *   The port identifier of the Ethernet device.
>>>>> + * @param mode
>>>>> + *   the FEC mode.
>>>>> + * @return
>>>>> + *   - (0) if successful.
>>>>> + *   - (-EINVAL) if the FEC mode is not valid.
>>>>> + *   - (-ENOTSUP) if underlying hardware OR driver doesn't support.
>>>>> + *   - (-EIO) if device is removed.
>>>>> + *   - (-ENODEV)  if *port_id* invalid.
>>>>> + */
>>>>> +__rte_experimental
>>>>> +int rte_eth_fec_set(uint16_t port_id, enum rte_eth_fec_mode mode);
>>>>
>>>> It does not allow to tweak autoneg facilities.
>>>> E.g. "I know that RS is buggy, so I want to exclude it from
>>>> auto-negotiation".
>>>> So, I suggest to change mode to capa bitmask.
>>>> If AUTO is set, other bits may be set and specify allowed
>>>> options. E.g. AUTO|RS|BASER will require FEC, i.e. NOFEC is
>>> The two FEC modes cannot be configured for hardware at the same time,
>>> including AUTO and other FEC modes. This is determined by Hardware
>>> itself.
>>
>> Which HW? Yours? If so, it does not matter. The patch adds
>> generic API. My comments are not abstract thoughts. There
>> are requirements and capabilities behind.
>
> yes, it is in my HW. But I think the feature of FEC will exist in other
> HW: the two FEC modes cannot be configured for hardware at the same time.
> By the way, if set two FEC mode in our HW, the result will be unknown.
> I also test X710 nic device, it does not support that feature.
> I do not support that solutions. thanks.

I'm not trying to say that two FEC modes could be running
simultaneously. I'm trying to say that in the future a PHY
could support more than one FEC mode and autonegotiation
could make a choice which FEC mode to use.
E.g.
NOFEC, FOO and BAR modes supported
set (AUTO|FOO|BAR) will require either FOO or BAR to be
negotiated and does not allow NOFEC.

>>> Thanks.
>>>
>>>> not allowed. If just RS, it means that auto-negotiation is
>>>> disabled and RS must be used.
>>>> If AUTO is unset, only one bit may be set in capabilities.
>>>> Since we don't do it per speed, I think it is safe to ignore
>>>> unsupported mode bits. I.e. do not return error if unsupported
>>>> capa is requested to together with AUTO, however it could be
> Why? if the mode is unsupported (not in capa),why we can configure
> the the mode to the device? Because this is unreaonable. Also,
> the configure will not be ineffective, and the hardware will return
> error back to the driver.

I agree that requested mode should be in capabilities for at
least some speed, but not required to be applicable to
currently negotiated and running speed. May be it is obvious.
I.e. if I set AUTO|NOFEC|RS when capabilities are
25G(NOFEC,AUTO,BASER) and 100G(NOFEC,AUTO,RS), it will
enforce NOFEC for 25G (since BASER is disabled) and allow
either NOFEC or RS for 100G.

>>>> a problem if no modes are allowed for negotiated link speed.
>>>> Thoughts are welcome.
>>>>
>>>>> +
>>>>> +/**
>>>>>     * Get current status of the Ethernet link flow control for
>>>>> Ethernet device
>>>>>     *
>>>>>     * @param port_id
>>>>
>>>> [snip]
>>>>
>>>> .
>>>>
>>
>> .
>>



More information about the dev mailing list