Re: [dpdk-dev] [PATCH v7 1/7] ethdev: add set ptype function

[Andrew Rybchenko <arybchenko@solarflare.com>]

On 10/2/19 4:37 PM, Andrew Rybchenko wrote:
> Hi,
>
> looks good, just few comments below.
>
> Many thanks for working on it,
> Andrew.
>
> On 10/2/19 6:47 AM, pbhagavatula@marvell.com wrote:
>> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>
>> Add `rte_eth_dev_set_supported_ptypes` function that will allow the
>> application to inform the PMD the packet types it is interested in.
>> Based on the ptypes set PMDs can optimize their Rx path.
>>
>> -If application doesn’t want any ptype information it can call
>> `rte_eth_dev_set_supported_ptypes(ethdev_id, RTE_PTYPE_UNKNOWN, NULL, 
>> 0)`
>> and PMD may skip packet type processing and set rte_mbuf::packet_type to
>> RTE_PTYPE_UNKNOWN.
>>
>> -If application doesn’t call `rte_eth_dev_set_supported_ptypes` PMD can
>> return `rte_mbuf::packet_type` with `rte_eth_dev_get_supported_ptypes`.
>>
>> -If application is interested only in L2/L3 layer, it can inform the PMD
>> to update `rte_mbuf::packet_type` with L2/L3 ptype by calling
>> `rte_eth_dev_set_supported_ptypes(ethdev_id,
>>         RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK, NULL, 0)`.
>>
>> Suggested-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
>> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> [snip]
>
>> diff --git a/lib/librte_ethdev/rte_ethdev.c 
>> b/lib/librte_ethdev/rte_ethdev.c
>> index 17d183e1f..b1588fe7a 100644
>> --- a/lib/librte_ethdev/rte_ethdev.c
>> +++ b/lib/librte_ethdev/rte_ethdev.c
>> @@ -2602,6 +2602,56 @@ rte_eth_dev_get_supported_ptypes(uint16_t 
>> port_id, uint32_t ptype_mask,
>>       return j;
>>   }
>>   +int
>> +rte_eth_dev_set_supported_ptypes(uint16_t port_id, uint32_t ptype_mask,
>> +                 uint32_t *set_ptypes, unsigned int num)
>> +{
>> +    unsigned int i, j;
>> +    struct rte_eth_dev *dev;
>> +    const uint32_t *all_ptypes;
>> +
>> +    RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -ENODEV);
>> +    dev = &rte_eth_devices[port_id];
>> + RTE_FUNC_PTR_OR_ERR_RET(*dev->dev_ops->dev_supported_ptypes_get, 0);
>> + RTE_FUNC_PTR_OR_ERR_RET(*dev->dev_ops->dev_supported_ptypes_set, 0);
>
> When 0 is returned above, we should set set_ptypes. So, below
> check num vs set_types should be done before callbacks check and
> I think it is OK to do
>       set_ptypes[0] = RTE_PTYPE_UNKNOWN;

of course if num > 0

> just after the check. It will allow to simplify code below and will
> make it set correctly even if 0 returned because of no callbacks.
>
>> +
>> +    if (num > 0 && set_ptypes == NULL)
>> +        return -EINVAL;
>> +
>> +    if (ptype_mask == 0) {
>> +        if (num > 0)
>> +            set_ptypes[0] = RTE_PTYPE_UNKNOWN;
>> +
>> +        return (*dev->dev_ops->dev_supported_ptypes_set)(dev,
>> +                ptype_mask);
>> +    }
>> +
>> +    all_ptypes = (*dev->dev_ops->dev_supported_ptypes_get)(dev);
>> +    if (all_ptypes == NULL) {
>> +        if (num > 0)
>> +            set_ptypes[0] = RTE_PTYPE_UNKNOWN;
>> +
>> +        return 0;
>> +    }
>> +
>> +    for (i = 0, j = 0; set_ptypes != NULL &&
>> +                (all_ptypes[i] != RTE_PTYPE_UNKNOWN); ++i) {
>> +        if (ptype_mask & all_ptypes[i]) {
>> +            if (j < num - 1) {
>> +                set_ptypes[j] = all_ptypes[i];
>> +                j++;
>> +                continue;
>> +            }
>> +            break;
>
> I'd like to repeat my question about insufficient space to return
> set_ptypes. Do we need to signal it somehow? If no, it should be
> explained why in the comments here.
>
>> +        }
>> +    }
>> +
>> +    if (set_ptypes != NULL)
>> +        set_ptypes[j] = RTE_PTYPE_UNKNOWN;
>> +
>> +    return (*dev->dev_ops->dev_supported_ptypes_set)(dev, ptype_mask);
>> +}
>> +
>>   void
>>   rte_eth_macaddr_get(uint16_t port_id, struct rte_ether_addr *mac_addr)
>>   {
>> diff --git a/lib/librte_ethdev/rte_ethdev.h 
>> b/lib/librte_ethdev/rte_ethdev.h
>> index d9871782e..c577a9172 100644
>> --- a/lib/librte_ethdev/rte_ethdev.h
>> +++ b/lib/librte_ethdev/rte_ethdev.h
>> @@ -2431,6 +2431,42 @@ int rte_eth_dev_fw_version_get(uint16_t port_id,
>>    */
>>   int rte_eth_dev_get_supported_ptypes(uint16_t port_id, uint32_t 
>> ptype_mask,
>>                        uint32_t *ptypes, int num);
>> +/**
>> + * @warning
>> + * @b EXPERIMENTAL: this API may change without prior notice.
>> + *
>> + * Inform Ethernet device of the packet types classification in which
>> + * the recipient is interested.
>> + *
>> + * Application can use this function to set only specific ptypes 
>> that it's
>> + * interested. This information can be used by the PMD to optimize 
>> Rx path.
>> + *
>> + * The function accepts an array `set_ptypes` allocated by the 
>> caller to
>> + * store the packet types set by the driver, the last element of the 
>> array
>> + * is set to RTE_PTYPE_UNKNOWN. The size of the `set_ptype` array 
>> should be
>> + * `rte_eth_dev_get_supported_ptypes() + 1` else it might only be 
>> filled
>> + * partially.
>> + *
>> + * @param port_id
>> + *   The port identifier of the Ethernet device.
>> + * @param ptype_mask
>> + *   The ptype family that application is interested in.
>> + * @param set_ptypes
>> + *   An array pointer to store set packet types, allocated by 
>> caller. The
>> + *   function marks the end of array with RTE_PTYPE_UNKNOWN.
>> + * @param num
>> + *   Size of the array pointed by param ptypes.
>> + *   Should be rte_eth_dev_get_supported_ptypes() + 1 to accommodate 
>> the
>> + *   set ptypes.
>> + * @return
>> + *   - (0) if Success.
>> + *   - (-ENODEV) if *port_id* invalid.
>> + *   - (-EINVAL) if *ptype_mask* is invalid (or) set_ptypes is NULL and
>> + *     num > 0.
>> + */
>> +__rte_experimental
>> +int rte_eth_dev_set_supported_ptypes(uint16_t port_id, uint32_t 
>> ptype_mask,
>> +                     uint32_t *set_ptypes, unsigned int num);
>>     /**
>>    * Retrieve the MTU of an Ethernet device.
>> diff --git a/lib/librte_ethdev/rte_ethdev_core.h 
>> b/lib/librte_ethdev/rte_ethdev_core.h
>> index 2922d5b7c..93bc34480 100644
>> --- a/lib/librte_ethdev/rte_ethdev_core.h
>> +++ b/lib/librte_ethdev/rte_ethdev_core.h
>> @@ -110,6 +110,10 @@ typedef void (*eth_dev_infos_get_t)(struct 
>> rte_eth_dev *dev,
>>   typedef const uint32_t *(*eth_dev_supported_ptypes_get_t)(struct 
>> rte_eth_dev *dev);
>>   /**< @internal Get supported ptypes of an Ethernet device. */
>>   +typedef uint32_t (*eth_dev_supported_ptypes_set_t)(struct 
>> rte_eth_dev *dev,
>> +                          uint32_t ptype_mask);
>> +/**< @internal Inform device about packet types in which the 
>> recipient is interested. */
>> +
>
> Please, take a look at promiscuous mode callback and let's put
> more verbose description here which better document interface
> to drivers. I think ptyep_mask description should refer to corresponding
> defines to be used.
>
> [snip]
>