[dpdk-dev] [PATCH 1/3 v3] ethdev: add rx offload to drop error packets

[Andrew Rybchenko]

On 10/12/20 2:30 PM, Nipun Gupta wrote:
> 
> 
>> -----Original Message-----
>> From: Andrew Rybchenko <Andrew.Rybchenko at oktetlabs.ru>
>> Sent: Monday, October 12, 2020 1:32 PM
>> To: Nipun Gupta <nipun.gupta at nxp.com>; dev at dpdk.org
>> Cc: thomas at monjalon.net; ferruh.yigit at intel.com; arybchenko at solarflare.com;
>> Hemant Agrawal <hemant.agrawal at nxp.com>; Sachin Saxena
>> <sachin.saxena at nxp.com>; Rohit Raj <rohit.raj at nxp.com>;
>> jerinjacobk at gmail.com; stephen at networkplumber.org; asafp at nvidia.com
>> Subject: Re: [dpdk-dev] [PATCH 1/3 v3] ethdev: add rx offload to drop error
>> packets
>>
>> On 10/9/20 4:13 PM, nipun.gupta at nxp.com wrote:
>>> From: Nipun Gupta <nipun.gupta at nxp.com>
>>>
>>> This change adds a RX offload capability and configuration to
>>> enable hardware to drop the packets in case of any error in the
>>> packets such as L3 checksum error or L4 checksum.
>>>
>>> Signed-off-by: Nipun Gupta <nipun.gupta at nxp.com>
>>> Signed-off-by: Rohit Raj <rohit.raj at nxp.com>
>>> Reviewed-by: Asaf Penso <asafp at nvidia.com>
>>> ---
>>>
>>> v3:
>>>  - Add additional rx_err_drop_offload_capa, which is specific
>>>    capability flag for RX packets error drop offload. Currently
>>>    only 'all' error packet drops are enabled, but can be extended
>>>    to provide capability to drop any specific errors like L1 FCS,
>>>    L3 Checksum etc.
>>>  - Added separate config structure to enable the drop configuration.
>>>  - Updated doc with the new updated option in testbbdev (patch 3/3)
>>>
>>> v2:
>>>  - Add support in DPAA1 driver (patch 2/3)
>>>  - Add support and config parameter in testpmd (patch 3/3)
>>>
>>>  lib/librte_ethdev/rte_ethdev.c |  1 +
>>>  lib/librte_ethdev/rte_ethdev.h | 22 ++++++++++++++++++++++
>>>  2 files changed, 23 insertions(+)
>>>
>>> diff --git a/lib/librte_ethdev/rte_ethdev.c b/lib/librte_ethdev/rte_ethdev.c
>>> index 48d1333b1..be25e947e 100644
>>> --- a/lib/librte_ethdev/rte_ethdev.c
>>> +++ b/lib/librte_ethdev/rte_ethdev.c
>>> @@ -128,6 +128,7 @@ static const struct {
>>>  	RTE_RX_OFFLOAD_BIT2STR(SCTP_CKSUM),
>>>  	RTE_RX_OFFLOAD_BIT2STR(OUTER_UDP_CKSUM),
>>>  	RTE_RX_OFFLOAD_BIT2STR(RSS_HASH),
>>> +	RTE_RX_OFFLOAD_BIT2STR(ERR_PKT_DROP),
>>>  };
>>>
>>>  #undef RTE_RX_OFFLOAD_BIT2STR
>>> diff --git a/lib/librte_ethdev/rte_ethdev.h b/lib/librte_ethdev/rte_ethdev.h
>>> index d2bf74f12..cb968d38a 100644
>>> --- a/lib/librte_ethdev/rte_ethdev.h
>>> +++ b/lib/librte_ethdev/rte_ethdev.h
>>> @@ -1194,6 +1194,16 @@ struct rte_intr_conf {
>>>  	uint32_t rmv:1;
>>>  };
>>>
>>> +/**
>>> + * A structure used to enable/disable error packet drop on RX.
>>> + */
>>> +struct rte_rx_err_pkt_drop_conf {
>>> +	/** enable/disable all RX error packet drop.
>>> +	 * 0 (default) - disable, 1 enable
>>> +	 */
>>> +	uint32_t all:1;
>>
>> "all" is bad. It should be granular and in the same terms
>> as mask in dev_info capability.
> 
> Consider that application do not want to receive any error packets as it will drop all
> the error packets after increasing the error counter maintainer in the application.
> If same functionality can be done by the hardware where hardware maintains the error
> statistics, then why not have the same?
> When hardware does parsing of packet headers and found error during that parsing,
> this is to say to HW - hey drop any such error packets which you found during parsing
> and just increase specific error counter (if present).
> 
> I agree, granular shall also be added, but as our driver does not support them, so I did
> not add those bits here.

There is always precise definition behind "all".
You can report it in dev_info, an application can get it and
use as is (all reported capability bits) in configuration.

> 
>>
>>> +};
>>> +
>>>  /**
>>>   * A structure used to configure an Ethernet port.
>>>   * Depending upon the RX multi-queue mode, extra advanced
>>> @@ -1236,6 +1246,8 @@ struct rte_eth_conf {
>>>  	uint32_t dcb_capability_en;
>>>  	struct rte_fdir_conf fdir_conf; /**< FDIR configuration. DEPRECATED */
>>>  	struct rte_intr_conf intr_conf; /**< Interrupt mode configuration. */
>>> +	struct rte_rx_err_pkt_drop_conf err_pkt_drop_conf;
>>> +	/**< RX error packet drop configuration. */
>>
>> Why not per queue?
> 
> We do not support per queue configuration for error packet drop, but only
> on ethernet basis.
> 
>>
>>>  };
>>>
>>>  /**
>>> @@ -1260,6 +1272,7 @@ struct rte_eth_conf {
>>>  #define DEV_RX_OFFLOAD_SCTP_CKSUM	0x00020000
>>>  #define DEV_RX_OFFLOAD_OUTER_UDP_CKSUM  0x00040000
>>>  #define DEV_RX_OFFLOAD_RSS_HASH		0x00080000
>>> +#define DEV_RX_OFFLOAD_ERR_PKT_DROP	0x00100000
>>>
>>>  #define DEV_RX_OFFLOAD_CHECKSUM (DEV_RX_OFFLOAD_IPV4_CKSUM | \
>>>  				 DEV_RX_OFFLOAD_UDP_CKSUM | \
>>> @@ -1274,6 +1287,13 @@ struct rte_eth_conf {
>>>   * mentioned in rte_rx_offload_names in rte_ethdev.c file.
>>>   */
>>>
>>> +/**
>>> + * RX Error Drop offload config/capabilities of a device. These
>>> + * are valid only when RX capability DEV_RX_OFFLOAD_ERR_PKT_DROP
>>> + * is supported by the device.
>>> + */
>>> +#define DEV_RX_ERR_PKT_DROP_OFFLOAD_ALL		0x00000001
>>> +
>>
>> I strictly dislike "all". It will always be bad defined.
>> It must be granular.
>>
>>>  /**
>>>   * TX offload capabilities of a device.
>>>   */
>>> @@ -1411,6 +1431,8 @@ struct rte_eth_dev_info {
>>>  	/**< Device per-queue RX offload capabilities. */
>>>  	uint64_t tx_queue_offload_capa;
>>>  	/**< Device per-queue TX offload capabilities. */
>>> +	uint64_t rx_err_drop_offload_capa;
>>> +	/**< RX error packet drop offload capabilities. */
>>>  	uint16_t reta_size;
>>>  	/**< Device redirection table size, the total number of entries. */
>>>  	uint8_t hash_key_size; /**< Hash key size in bytes */
>>>
>