Re: [dpdk-dev] [RFC v2 6/6] net/ixgbe: use flow sample to re-realize mirror rule

[Ferruh Yigit <ferruh.yigit@intel.com>]

On 11/3/2020 8:28 AM, Steve Yang wrote:
> When set follow sample rule's ratio equal to one, its behavior is same
> as mirror-rule, so we can use "flow create * pattern * actions sample *"
> to replace old "set port * mirror-rule *" command now.
> 
> The example of mirror rule command mapping to flow management command:
> (in below command, port 0 is PF and port 1-2 is VF):
> 1) ingress: pf => pf
>          set port 0 mirror-rule 0 uplink-mirror dst-pool 2 on
>          or
>          flow create 0 ingress pattern pf / end \
>                                actions sample ratio 1 / port_id id  0 / end
> 2) egress: pf => pf
>          set port 0 mirror-rule 0 downlink-mirror dst-pool 2 on
>          or
>          flow create 0 egress pattern pf / end \
>                               actions sample ratio 1 / port_id id  0 / end
> 3) ingress: pf => vf 2
>          set port 0 mirror-rule 0 uplink-mirror dst-pool 1 on
>          or
>          flow create 0 ingress pattern pf / end \
>                                actions sample ratio 1 / port_id id 2 / end
> 4) egress: pf => vf 2
>          set port 0 mirror-rule 0 downlink-mirror dst-pool 1 on
>          or
>          flow create 0 egress pattern pf / end \
>                               actions sample ratio 1 / port_id id 2 / end
> 5) ingress: vf 0,1 => pf
>          set port 0 mirror-rule 0 pool-mirror-up 0x3 dst-pool 2 on
>          or
>          flow create 0 ingress pattern vf id is 0 / end \
>                                actions sample ratio 1 / port_id id 0 / end
>          flow create 0 ingress pattern vf id is 1 / end \
>                                actions sample ratio 1 / port_id id 0 / end
> 6) ingress: vf 0 => vf 1
>          set port 0 mirror-rule 0 pool-mirror-up 0x1 dst-pool 1 on
>          or
>          flow create 0 ingress pattern vf id is 0 / end \
>                                actions sample ratio 1 / port_id id 2 / end
> 7) ingress: vlan 4,6 => vf 1
>          rx_vlan add 4 port 0 vf 0xf
>          rx_vlan add 6 port 0 vf 0xf
>          set port 0 mirror-rule 0 vlan-mirror 4,6 dst-pool 1 on
>          or
>          rx_vlan add 4 port 0 vf 0xf
>          rx_vlan add 6 port 0 vf 0xf
>          flow create 0 ingress pattern vlan vid is 4 / end \
>                                actions sample ratio 1 / port_id id 2 / end
>          flow create 0 ingress pattern vlan vid is 6 / end \
>                                actions sample ratio 1 / port_id id 2 / end
> 
> Signed-off-by: Steve Yang <stevex.yang@intel.com>

<...>

> +static int
> +ixgbe_config_mirror_filter_add(struct rte_eth_dev *dev,
> +			       struct ixgbe_flow_mirror_conf *mirror_conf)
> +{
> +	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
> +	uint32_t mr_ctl, vlvf;
> +	uint32_t mp_lsb = 0;
> +	uint32_t mv_msb = 0;
> +	uint32_t mv_lsb = 0;
> +	uint32_t mp_msb = 0;
> +	uint8_t i = 0;
> +	int reg_index = 0;
> +	uint64_t vlan_mask = 0;
> +
> +	const uint8_t pool_mask_offset = 32;
> +	const uint8_t vlan_mask_offset = 32;
> +	const uint8_t dst_pool_offset = 8;
> +	const uint8_t rule_mr_offset  = 4;
> +	const uint8_t mirror_rule_mask = 0x0F;
> +
> +	struct ixgbe_hw *hw =
> +		IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
> +	struct ixgbe_filter_info *filter_info =
> +		IXGBE_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
> +	struct ixgbe_mirror_conf_ele *it;
> +	int8_t rule_id;
> +	uint8_t mirror_type = 0;
> +
> +	if (ixgbe_vt_check(hw) < 0)
> +		return -ENOTSUP;
> +
> +	if (IXGBE_INVALID_MIRROR_TYPE(mirror_conf->rule_type)) {
> +		PMD_DRV_LOG(ERR, "unsupported mirror type 0x%x.",
> +			    mirror_conf->rule_type);
> +		return -EINVAL;
> +	}
> +
> +	TAILQ_FOREACH(it, &filter_mirror_list, entries) {
> +		if (it->filter_info.rule_type == mirror_conf->rule_type &&
> +		    it->filter_info.dst_pool == mirror_conf->dst_pool &&
> +		    it->filter_info.pool_mask == mirror_conf->pool_mask &&
> +		    it->filter_info.vlan_mask == mirror_conf->vlan_mask &&
> +		    !memcmp(it->filter_info.vlan_id, mirror_conf->vlan_id,
> +		    ETH_MIRROR_MAX_VLANS * sizeof(mirror_conf->vlan_id[0]))) {
> +			PMD_DRV_LOG(ERR, "mirror rule exists.");
> +			return -EEXIST;
> +		}
> +	}


Can you please help me to understand this.
There is a per device 'ixgbe_filter_info->mirror_filters',
also there is a driver global 'ixgbe_mirror_filter_list',

When a new filter added, above checks the 'ixgbe_mirror_filter_list' for all 
filters, but what if driver is driving multiple devices and we want to add same 
rule for each device, won't above checks be wrong?

Btw, I am not clear why there are two separate per-device and driver global list 
exists, I can see that is a common patter for filters not specific to this one, 
can you please explain?

<...>

> +static void
> +ixgbe_clear_all_mirror_filter(struct rte_eth_dev *dev)
> +{
> +	struct ixgbe_filter_info *filter_info =
> +		IXGBE_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
> +	int i;
> +
> +	for (i = 0; i < IXGBE_MAX_MIRROR_RULES; i++) {
> +		if (filter_info->mirror_mask & (1 << i)) {
> +			ixgbe_config_mirror_filter_del(dev,
> +				&filter_info->mirror_filters[i]);
> +		}
> +	}

This function is clearing all device filters, but not removing them from global 
list, is this intentional?

> +}
> +
>   void
>   ixgbe_filterlist_init(void)
>   {
> @@ -3185,6 +3365,7 @@ ixgbe_filterlist_init(void)
>   	TAILQ_INIT(&filter_fdir_list);
>   	TAILQ_INIT(&filter_l2_tunnel_list);
>   	TAILQ_INIT(&filter_rss_list);
> +	TAILQ_INIT(&filter_mirror_list);
>   	TAILQ_INIT(&ixgbe_flow_list);
>   }
>   
> @@ -3198,6 +3379,7 @@ ixgbe_filterlist_flush(void)
>   	struct ixgbe_fdir_rule_ele *fdir_rule_ptr;
>   	struct ixgbe_flow_mem *ixgbe_flow_mem_ptr;
>   	struct ixgbe_rss_conf_ele *rss_filter_ptr;
> +	struct ixgbe_mirror_conf_ele *mirror_filter_ptr;
>   
>   	while ((ntuple_filter_ptr = TAILQ_FIRST(&filter_ntuple_list))) {
>   		TAILQ_REMOVE(&filter_ntuple_list,
> @@ -3241,6 +3423,13 @@ ixgbe_filterlist_flush(void)
>   		rte_free(rss_filter_ptr);
>   	}
>   
> +	while ((mirror_filter_ptr = TAILQ_FIRST(&filter_mirror_list))) {
> +		TAILQ_REMOVE(&filter_mirror_list,
> +			     mirror_filter_ptr,
> +			     entries);
> +		rte_free(mirror_filter_ptr);
> +	}

This block removes filters from global list, but it doesn't remove them from 
per-device list, and it doesn't remove them from device, so the filtering still 
be active in the device after this call, is this intentional?

<...>

> +	memset(&mirror_conf, 0, sizeof(struct ixgbe_flow_mirror_conf));
> +	ret = ixgbe_parse_sample_filter(dev, attr, pattern,
> +				     actions, &mirror_conf, error);
> +	if (!ret) {
> +		/* Just support mirror behavior */
> +		ret = ixgbe_config_mirror_filter_add(dev, &mirror_conf);
> +		if (ret) {
> +			PMD_DRV_LOG(ERR, "failed to add mirror filter");
> +			goto out;
> +		}
> +
> +		mirror_filter_ptr = rte_zmalloc("ixgbe_mirror_filter",
> +				sizeof(struct ixgbe_mirror_conf_ele), 0);
> +		if (!mirror_filter_ptr) {
> +			PMD_DRV_LOG(ERR, "failed to allocate memory");
> +			goto out;

Should rollback the above filter addition on the error?