From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 9E724A04BA; Tue, 22 Sep 2020 13:07:08 +0200 (CEST) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 08B8D1DBC8; Tue, 22 Sep 2020 13:07:08 +0200 (CEST) Received: from huawei.com (szxga07-in.huawei.com [45.249.212.35]) by dpdk.org (Postfix) with ESMTP id 432651DB96 for ; Tue, 22 Sep 2020 13:07:06 +0200 (CEST) Received: from DGGEMS412-HUB.china.huawei.com (unknown [172.30.72.60]) by Forcepoint Email with ESMTP id 4091F425CD762BA7FBD5; Tue, 22 Sep 2020 19:07:04 +0800 (CST) Received: from [10.67.103.128] (10.67.103.128) by DGGEMS412-HUB.china.huawei.com (10.3.19.212) with Microsoft SMTP Server id 14.3.487.0; Tue, 22 Sep 2020 19:06:53 +0800 To: Andrew Rybchenko , CC: , , , References: <1599534347-20430-1-git-send-email-humin29@huawei.com> <1600668838-31498-1-git-send-email-humin29@huawei.com> <1600668838-31498-2-git-send-email-humin29@huawei.com> From: "Min Hu (Connor)" Message-ID: <879dfbdb-d706-f4c9-2d5a-6c6eccbc4ab9@huawei.com> Date: Tue, 22 Sep 2020 19:06:53 +0800 User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:68.0) Gecko/20100101 Thunderbird/68.3.1 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset="utf-8"; format=flowed Content-Transfer-Encoding: 8bit X-Originating-IP: [10.67.103.128] X-CFilter-Loop: Reflected Subject: Re: [dpdk-dev] [PATCH V9 1/3] ethdev: introduce FEC API X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" 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) >>>> Reviewed-by: Wei Hu (Xavier) >>>> Reviewed-by: Chengwen Feng >>>> Reviewed-by: Chengchang Tang >>>> Reviewed-by: Ajit Khaparde >>>> Acked-by: Konstantin Ananyev >>> >>> [snip] >>> >>>> diff --git a/lib/librte_ethdev/rte_ethdev.h >>>> b/lib/librte_ethdev/rte_ethdev.h >>>> index 70295d7..7d5e81b 100644 >>>> --- a/lib/librte_ethdev/rte_ethdev.h >>>> +++ b/lib/librte_ethdev/rte_ethdev.h >>>> @@ -1310,6 +1310,9 @@ struct rte_eth_conf { >>>>   #define RTE_ETH_DEV_FALLBACK_RX_NBQUEUES 1 >>>>   #define RTE_ETH_DEV_FALLBACK_TX_NBQUEUES 1 >>>>   +/* Translate from FEC mode to FEC capa */ >>>> +#define RTE_ETH_FEC_MODE_TO_CAPA(x)    (1U << (x)) >>>> + >>> >>> It should not be so far from rte_eth_fec_mode. Please, add just >>> after it. >> I will fix it in V10, thanks. >>> >>> May be it should be: >>> #define RTE_ETH_FEC_MODE_TO_CAPA(x) (1U << (RTE_ETH_FEC_ ## x)) >>> >> This will be weird， as "x" should be one fec mode。examples >> as follows: >> >> rte_eth_fec_set(uint16_t port_id, enum rte_eth_fec_mode mode) >> { >>     struct rte_eth_dev *dev; >>     uint32_t fec_mode_mask; >>     int ret; >> >>     ret = rte_eth_fec_get_capability(port_id, &fec_mode_mask); >>     if (ret != 0) >>         return ret; >> >>     /* >>      * Check whether the configured mode is within the FEC capability. >>      * If not, the configured mode will not be supported. >>      */ >>     if (!(fec_mode_mask & RTE_ETH_FEC_MODE_TO_CAPA(mode))) { >>         RTE_ETHDEV_LOG(ERR, "unsupported FEC mode = %d, port_id = %u\n", >>                    mode, port_id); >>         return -EINVAL; >>     } > > Got it, makes sense. > >> [snip] >>>>   /** >>>>    * Preferred Rx/Tx port parameters. >>>>    * There are separate instances of this structure for transmission >>>> @@ -1511,6 +1514,24 @@ struct rte_eth_dcb_info { >>>>       struct rte_eth_dcb_tc_queue_mapping tc_queue; >>>>   }; >>>>   +/** >>>> + * This enum indicates the possible (forward error correction)FEC modes >>>> + * of an ethdev port. >>>> + */ >>>> +enum rte_eth_fec_mode { >>>> +    RTE_ETH_FEC_NOFEC = 0,      /**< FEC is off */ >>>> +    RTE_ETH_FEC_AUTO,        /**< FEC autonegotiation modes */ >>>> +    RTE_ETH_FEC_BASER,          /**< FEC using common algorithm */ >>>> +    RTE_ETH_FEC_RS,             /**< FEC using RS algorithm */ >>>> +}; >>>> + >>>> +/* This indicates FEC capabilities */ >>>> +#define RTE_ETH_FEC_CAPA_NOFEC  (1U << RTE_ETH_FEC_NOFEC) >>>> +#define RTE_ETH_FEC_CAPA_AUTO   (1U << RTE_ETH_FEC_AUTO) >>>> +#define RTE_ETH_FEC_CAPA_BASER  (1U << RTE_ETH_FEC_BASER) >>>> +#define RTE_ETH_FEC_CAPA_RS     (1U << RTE_ETH_FEC_RS) >>> >>> Shouldn't RTE_ETH_FEC_MODE_TO_CAPA be used as definition >>> values? >>> I will define a new macro as you suggest in V10, like this: >> /* This macro indicates FEC capa mask*/ >> #define RTE_ETH_FEC_MODE_CAPA_MASK(x)    (1U << (RTE_ETH_FEC_ ## x)) >> >> that macro will replace the follow macros: >> #define RTE_ETH_FEC_CAPA_NOFEC  (1U << RTE_ETH_FEC_NOFEC) >> #define RTE_ETH_FEC_CAPA_AUTO   (1U << RTE_ETH_FEC_AUTO) >> #define RTE_ETH_FEC_CAPA_BASER  (1U << RTE_ETH_FEC_BASER) >> #define RTE_ETH_FEC_CAPA_RS     (1U << RTE_ETH_FEC_RS) >> >> i.e.: >> RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) >> RTE_ETH_FEC_MODE_CAPA_MASK(AUTO) >> RTE_ETH_FEC_MODE_CAPA_MASK(BASER) >> RTE_ETH_FEC_MODE_CAPA_MASK(RS) >> >> thanks for your suggestions. > > Thanks for your work on the patchset. > >> >>>> + >>>> + >>>>   #define RTE_ETH_ALL RTE_MAX_ETHPORTS >>>>     /* Macros to check for valid port */ >>>> @@ -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? > >> >> 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. >> >> >>>> + >>>> +/** >>>> + * @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. >> 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. >>> 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] >>> >>> . >>> > > . >