On 9/29/2017 10:16 AM, Wei Zhao wrote:
> This feature enable queue regions configuration for RSS in PF,
> so that different traffic classes or different packet
> classification types can be separated to different queues in
> different queue regions.This patch can set queue region range,
> it include queue number in a region and the index of first queue.
> This patch enable mapping between different priorities (UP) and
> different traffic classes.It also enable mapping between a region
> index and a sepcific flowtype(PCTYPE).It also provide the solution
> of flush all configuration about queue region the above described.
>
> Signed-off-by: Wei Zhao <wei.zh...@intel.com>
> ---
<...>
> @@ -541,6 +547,34 @@ struct i40e_ethertype_rule {
> struct rte_hash *hash_table;
> };
>
> +/* queue region info */
> +struct i40e_region_info {
> + /* the region id for this configuration */
> + uint8_t region_id;
> + /* the start queue index for this region */
> + uint8_t queue_start_index;
> + /* the total queue number of this queue region */
> + uint8_t queue_num;
> + /* the total number of user priority for this region */
> + uint8_t user_priority_num;
> + /* the packet's user priority for this region */
> + uint8_t user_priority[I40E_MAX_USER_PRIORITY];
> + /* the total number of flowtype for this region */
> + uint8_t flowtype_num;
> + /**
> + * the pctype or hardware flowtype of packet,
> + * the specific index for each type has been defined
> + * in file i40e_type.h as enum i40e_filter_pctype.
> + */
> + uint8_t hw_flowtype[I40E_FILTER_PCTYPE_MAX];
> +};
> +
> +struct i40e_queue_region_info {
What do you think renaming:
i40e_region_info -> i40e_queue_region_info
i40e_queue_region_info -> i40e_queue_regions
> + /* the total number of queue region for this port */
> + uint16_t queue_region_number;
> + struct i40e_region_info region[I40E_REGION_MAX_INDEX + 1];
> +};
> +
> /* Tunnel filter number HW supports */
> #define I40E_MAX_TUNNEL_FILTER_NUM 400
>
<...>
> +static int
> +i40e_queue_region_set_flowtype(struct i40e_pf *pf,
> + struct rte_i40e_rss_region_conf *rss_region_conf)
> +{
> + int32_t ret = -EINVAL;
> + struct i40e_queue_region_info *info = &pf->queue_region;
> + uint16_t i, j;
> + uint16_t region_index, flowtype_index;
> +
> + /**
This is doxygen sytle comment, only nedeed for public headers.
> + * For the pctype or hardware flowtype of packet,
> + * the specific index for each type has been defined
> + * in file i40e_type.h as enum i40e_filter_pctype.
> + */
> +
<...>
> + }
> + }
> +}
Missing line.
> +static int
> +i40e_queue_region_set_user_priority(struct i40e_pf *pf,
> + struct rte_i40e_rss_region_conf *rss_region_conf)
> +{
> + struct i40e_queue_region_info *info = &pf->queue_region;
> + int32_t ret = -EINVAL;
> + uint16_t i, j, region_index;
> +
<...>
> +static int
> +i40e_queue_region_display_all_info(struct i40e_pf *pf, uint16_t port_id)
> +{
> + uint16_t i, j;
> + struct i40e_queue_region_info *info = &pf->queue_region;
> + static const char *queue_region_info_stats_border = "-------";
> +
> + if (!info->queue_region_number)
> + PMD_DRV_LOG(ERR, "there is no has been region set before");
> +
> + printf("\n %s All queue region info for port=%2d %s",
> + queue_region_info_stats_border, port_id,
> + queue_region_info_stats_border);
printf ? Please use driver logging functions.
> + printf("\n queue_region_number: %-14u \n", info->queue_region_number);
> +
> + for (i = 0; i < info->queue_region_number; i++) {
> + printf("\n region_id: %-14u queue_number: %-14u "
> + "queue_start_index: %-14u \n",
> + info->region[i].region_id,
> + info->region[i].queue_num,
> + info->region[i].queue_start_index);
> +
> + printf(" user_priority_num is %-14u :",
> + info->region[i].user_priority_num);
> + for (j = 0; j < info->region[i].user_priority_num; j++)
> + printf(" %-14u ", info->region[i].user_priority[j]);
> +
> + printf("\n flowtype_num is %-14u :",
> + info->region[i].flowtype_num);
> + for (j = 0; j < info->region[i].flowtype_num; j++)
> + printf(" %-14u ", info->region[i].hw_flowtype[j]);
> + }
> +
> + printf("\n\n");
> + return 0;
> +}
> +
> +int rte_pmd_i40e_rss_queue_region_conf(uint16_t port_id,
> + struct rte_i40e_rss_region_conf *rss_region_conf)
> +{
> + struct rte_eth_dev *dev = &rte_eth_devices[port_id];
> + struct i40e_pf *pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
> + struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
> + enum rte_pmd_i40e_queue_region_op op_type = rss_region_conf->op;
> + int32_t ret;
> +
> + RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -ENODEV);
> +
> + if (!is_i40e_supported(dev))
> + return -ENOTSUP;
> +
> + if (!(!i40e_queue_region_pf_check_rss(pf)))
> + return -ENOTSUP;
> +
> + /**
> + * This queue region feature only support pf by now. It should
> + * be called after dev_start, and will be clear after dev_stop.
> + * "RTE_PMD_I40E_RSS_QUEUE_REGION_ALL_FLUSH_ON"
> + * is just an enable function which server for other configuration,
> + * it is for all configuration about queue region from up layer,
> + * at first will only keep in DPDK softwarestored in driver,
> + * only after "FLUSH_ON", it commit all configuration to HW.
> + * Because PMD had to set hardware configuration at a time, so
> + * it will record all up layer command at first.
> + * "RTE_PMD_I40E_RSS_QUEUE_REGION_ALL_FLUSH_OFF" is
> + * just clean all configuration about queue region just now,
> + * and restore all to DPDK i40e driver default
> + * config when start up.
> + */
> +
> + switch (op_type) {
> + case RTE_PMD_I40E_QUEUE_REGION_SET:
> + ret = i40e_queue_region_set_region(pf, rss_region_conf);
> + break;
> + case RTE_PMD_I40E_REGION_FLOWTYPE_SET:
> + ret = i40e_queue_region_set_flowtype(pf, rss_region_conf);
> + break;
> + case RTE_PMD_I40E_USER_PRIORITY_REGION_SET:
> + ret = i40e_queue_region_set_user_priority(pf, rss_region_conf);
> + break;
> + case RTE_PMD_I40E_RSS_QUEUE_REGION_ALL_FLUSH_ON:
> + ret = i40e_flush_queue_region_all_conf(dev, hw, pf, 1);
> + break;
> + case RTE_PMD_I40E_RSS_QUEUE_REGION_ALL_FLUSH_OFF:
> + ret = i40e_flush_queue_region_all_conf(dev, hw, pf, 0);
> + break;
> + case RTE_PMD_I40E_RSS_QUEUE_REGION_INFO_GET:
> + ret = i40e_queue_region_display_all_info(pf, port_id);
Displaying is not always very useful in API level.
What do you think get fills the rss_region_conf with current device
settings and return back to the caller?
> + break;
> + default:
> + PMD_DRV_LOG(WARNING, "op type (%d) not supported",
> + op_type);
> + ret = -EINVAL;
> + }
> +
> + I40E_WRITE_FLUSH(hw);
> +
> + return ret;
> +}
> diff --git a/drivers/net/i40e/rte_pmd_i40e.h b/drivers/net/i40e/rte_pmd_i40e.h
> index 155b7e8..2219318 100644
> --- a/drivers/net/i40e/rte_pmd_i40e.h
> +++ b/drivers/net/i40e/rte_pmd_i40e.h
> @@ -91,6 +91,33 @@ enum rte_pmd_i40e_package_info {
> RTE_PMD_I40E_PKG_INFO_MAX = 0xFFFFFFFF
> };
>
> +/**
> + * Option types of queue region.
> + */
> +enum rte_pmd_i40e_queue_region_op {
> + RTE_PMD_I40E_REGION_UNDEFINED,
> + RTE_PMD_I40E_QUEUE_REGION_SET, /**< add queue region set */
> + RTE_PMD_I40E_REGION_FLOWTYPE_SET, /**< add pf region pctype set */
> + /*** add queue region user priority set */
> + RTE_PMD_I40E_USER_PRIORITY_REGION_SET,
> + /**
> + * ALL configuration about queue region from up layer
> + * at first will only keep in DPDK softwarestored in driver,
> + * only after " FLUSH_ON ", it commit all configuration to HW.
> + * Because PMD had to set hardware configuration at a time, so
> + * it will record all up layer command at first.
> + */
> + RTE_PMD_I40E_RSS_QUEUE_REGION_ALL_FLUSH_ON,
> + /**
> + * "FLUSH_OFF " is just clean all configuration about queue
> + * region just now, and restore all to DPDK i40e driver default
> + * config when start up.
> + */
> + RTE_PMD_I40E_RSS_QUEUE_REGION_ALL_FLUSH_OFF,
> + RTE_PMD_I40E_RSS_QUEUE_REGION_INFO_GET,
> + RTE_PMD_I40E_QUEUE_REGION_OP_MAX
These are in public header now.
Can you please use same namespace for all ops.
It can be "RTE_PMD_I40E_RSS_QUEUE_REGION_"
> +};
> +
> #define RTE_PMD_I40E_DDP_NAME_SIZE 32
>
> /**
> @@ -146,6 +173,27 @@ struct rte_pmd_i40e_ptype_mapping {
> };
>
> /**
> + * Queue region related information.
> + */
> +struct rte_i40e_rss_region_conf {
Can you please be consistant in using public names.
There are:
rss_region
queue_region
rss_queue_region
Also there is confusion between using:
rte_i40e_
rte_pmd_i40e_
> + /*** the region id for this configuration */
> + uint8_t region_id;
> + /** the pctype or hardware flowtype of packet,
> + * the specific index for each type has been defined
> + * in file i40e_type.h as enum i40e_filter_pctype.
> + */
> + uint8_t hw_flowtype;
> + /*** the start queue index for this region */
> + uint8_t queue_start_index;
> + /*** the total queue number of this queue region */
> + uint8_t queue_num;
> + /*** the packet's user priority for this region */
> + uint8_t user_priority;
> + /*** Option types of queue region */
> + enum rte_pmd_i40e_queue_region_op op;
> +};
> +
> +/**
> * Notify VF when PF link status changes.
> *
> * @param port
> @@ -657,4 +705,16 @@ int rte_pmd_i40e_ptype_mapping_replace(uint8_t port,
> int rte_pmd_i40e_add_vf_mac_addr(uint8_t port, uint16_t vf_id,
> struct ether_addr *mac_addr);
>
> +/**
> + * Do RSS queue region configuration for that port as
> + * the command option type
> + *
> + * @param port
> + * pointer id for that port device
port id ?
> + * @param conf_ptr
> + * pointer to the struct that contain all the
> + * region configuration parameters
> + */
> +int rte_pmd_i40e_rss_queue_region_conf(uint16_t port_id,
By defining port_id uint16_t, this patches becomes dependent to the
port_id update patchset, please clarify this in cover letter.
> + struct rte_i40e_rss_region_conf *rss_region_conf);
> #endif /* _PMD_I40E_H_ */
> diff --git a/drivers/net/i40e/rte_pmd_i40e_version.map
> b/drivers/net/i40e/rte_pmd_i40e_version.map
> index ef8882b..29d6b74 100644
> --- a/drivers/net/i40e/rte_pmd_i40e_version.map
> +++ b/drivers/net/i40e/rte_pmd_i40e_version.map
> @@ -50,5 +50,6 @@ DPDK_17.11 {
> global:
>
> rte_pmd_i40e_add_vf_mac_addr;
> + rte_pmd_i40e_rss_queue_region_conf;
>
> } DPDK_17.08;
>
