1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2016 6WIND S.A.
3 * Copyright 2016 Mellanox Technologies, Ltd
6 #include <netinet/in.h>
14 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
16 #pragma GCC diagnostic ignored "-Wpedantic"
18 #include <infiniband/verbs.h>
20 #pragma GCC diagnostic error "-Wpedantic"
23 #include <rte_common.h>
24 #include <rte_ether.h>
25 #include <rte_ethdev_driver.h>
27 #include <rte_cycles.h>
28 #include <rte_flow_driver.h>
29 #include <rte_malloc.h>
32 #include <mlx5_devx_cmds.h>
34 #include <mlx5_malloc.h>
36 #include "mlx5_defs.h"
38 #include "mlx5_flow.h"
39 #include "mlx5_flow_os.h"
40 #include "mlx5_rxtx.h"
42 /** Device flow drivers. */
43 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
45 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
47 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
48 [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
49 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
50 [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
52 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
53 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
58 MLX5_EXPANSION_ROOT_OUTER,
59 MLX5_EXPANSION_ROOT_ETH_VLAN,
60 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
61 MLX5_EXPANSION_OUTER_ETH,
62 MLX5_EXPANSION_OUTER_ETH_VLAN,
63 MLX5_EXPANSION_OUTER_VLAN,
64 MLX5_EXPANSION_OUTER_IPV4,
65 MLX5_EXPANSION_OUTER_IPV4_UDP,
66 MLX5_EXPANSION_OUTER_IPV4_TCP,
67 MLX5_EXPANSION_OUTER_IPV6,
68 MLX5_EXPANSION_OUTER_IPV6_UDP,
69 MLX5_EXPANSION_OUTER_IPV6_TCP,
71 MLX5_EXPANSION_VXLAN_GPE,
75 MLX5_EXPANSION_ETH_VLAN,
78 MLX5_EXPANSION_IPV4_UDP,
79 MLX5_EXPANSION_IPV4_TCP,
81 MLX5_EXPANSION_IPV6_UDP,
82 MLX5_EXPANSION_IPV6_TCP,
85 /** Supported expansion of items. */
86 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
87 [MLX5_EXPANSION_ROOT] = {
88 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
91 .type = RTE_FLOW_ITEM_TYPE_END,
93 [MLX5_EXPANSION_ROOT_OUTER] = {
94 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
95 MLX5_EXPANSION_OUTER_IPV4,
96 MLX5_EXPANSION_OUTER_IPV6),
97 .type = RTE_FLOW_ITEM_TYPE_END,
99 [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
100 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
101 .type = RTE_FLOW_ITEM_TYPE_END,
103 [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
104 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
105 .type = RTE_FLOW_ITEM_TYPE_END,
107 [MLX5_EXPANSION_OUTER_ETH] = {
108 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
109 MLX5_EXPANSION_OUTER_IPV6,
110 MLX5_EXPANSION_MPLS),
111 .type = RTE_FLOW_ITEM_TYPE_ETH,
114 [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
115 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
116 .type = RTE_FLOW_ITEM_TYPE_ETH,
119 [MLX5_EXPANSION_OUTER_VLAN] = {
120 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
121 MLX5_EXPANSION_OUTER_IPV6),
122 .type = RTE_FLOW_ITEM_TYPE_VLAN,
124 [MLX5_EXPANSION_OUTER_IPV4] = {
125 .next = RTE_FLOW_EXPAND_RSS_NEXT
126 (MLX5_EXPANSION_OUTER_IPV4_UDP,
127 MLX5_EXPANSION_OUTER_IPV4_TCP,
130 MLX5_EXPANSION_IPV6),
131 .type = RTE_FLOW_ITEM_TYPE_IPV4,
132 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
133 ETH_RSS_NONFRAG_IPV4_OTHER,
135 [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
136 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
137 MLX5_EXPANSION_VXLAN_GPE),
138 .type = RTE_FLOW_ITEM_TYPE_UDP,
139 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
141 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
142 .type = RTE_FLOW_ITEM_TYPE_TCP,
143 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
145 [MLX5_EXPANSION_OUTER_IPV6] = {
146 .next = RTE_FLOW_EXPAND_RSS_NEXT
147 (MLX5_EXPANSION_OUTER_IPV6_UDP,
148 MLX5_EXPANSION_OUTER_IPV6_TCP,
150 MLX5_EXPANSION_IPV6),
151 .type = RTE_FLOW_ITEM_TYPE_IPV6,
152 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
153 ETH_RSS_NONFRAG_IPV6_OTHER,
155 [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
156 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
157 MLX5_EXPANSION_VXLAN_GPE),
158 .type = RTE_FLOW_ITEM_TYPE_UDP,
159 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
161 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
162 .type = RTE_FLOW_ITEM_TYPE_TCP,
163 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
165 [MLX5_EXPANSION_VXLAN] = {
166 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
168 MLX5_EXPANSION_IPV6),
169 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
171 [MLX5_EXPANSION_VXLAN_GPE] = {
172 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
174 MLX5_EXPANSION_IPV6),
175 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
177 [MLX5_EXPANSION_GRE] = {
178 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
179 .type = RTE_FLOW_ITEM_TYPE_GRE,
181 [MLX5_EXPANSION_MPLS] = {
182 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
183 MLX5_EXPANSION_IPV6),
184 .type = RTE_FLOW_ITEM_TYPE_MPLS,
186 [MLX5_EXPANSION_ETH] = {
187 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
188 MLX5_EXPANSION_IPV6),
189 .type = RTE_FLOW_ITEM_TYPE_ETH,
191 [MLX5_EXPANSION_ETH_VLAN] = {
192 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
193 .type = RTE_FLOW_ITEM_TYPE_ETH,
195 [MLX5_EXPANSION_VLAN] = {
196 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
197 MLX5_EXPANSION_IPV6),
198 .type = RTE_FLOW_ITEM_TYPE_VLAN,
200 [MLX5_EXPANSION_IPV4] = {
201 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
202 MLX5_EXPANSION_IPV4_TCP),
203 .type = RTE_FLOW_ITEM_TYPE_IPV4,
204 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
205 ETH_RSS_NONFRAG_IPV4_OTHER,
207 [MLX5_EXPANSION_IPV4_UDP] = {
208 .type = RTE_FLOW_ITEM_TYPE_UDP,
209 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
211 [MLX5_EXPANSION_IPV4_TCP] = {
212 .type = RTE_FLOW_ITEM_TYPE_TCP,
213 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
215 [MLX5_EXPANSION_IPV6] = {
216 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
217 MLX5_EXPANSION_IPV6_TCP),
218 .type = RTE_FLOW_ITEM_TYPE_IPV6,
219 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
220 ETH_RSS_NONFRAG_IPV6_OTHER,
222 [MLX5_EXPANSION_IPV6_UDP] = {
223 .type = RTE_FLOW_ITEM_TYPE_UDP,
224 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
226 [MLX5_EXPANSION_IPV6_TCP] = {
227 .type = RTE_FLOW_ITEM_TYPE_TCP,
228 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
232 static const struct rte_flow_ops mlx5_flow_ops = {
233 .validate = mlx5_flow_validate,
234 .create = mlx5_flow_create,
235 .destroy = mlx5_flow_destroy,
236 .flush = mlx5_flow_flush,
237 .isolate = mlx5_flow_isolate,
238 .query = mlx5_flow_query,
239 .dev_dump = mlx5_flow_dev_dump,
240 .get_aged_flows = mlx5_flow_get_aged_flows,
243 /* Convert FDIR request to Generic flow. */
245 struct rte_flow_attr attr;
246 struct rte_flow_item items[4];
247 struct rte_flow_item_eth l2;
248 struct rte_flow_item_eth l2_mask;
250 struct rte_flow_item_ipv4 ipv4;
251 struct rte_flow_item_ipv6 ipv6;
254 struct rte_flow_item_ipv4 ipv4;
255 struct rte_flow_item_ipv6 ipv6;
258 struct rte_flow_item_udp udp;
259 struct rte_flow_item_tcp tcp;
262 struct rte_flow_item_udp udp;
263 struct rte_flow_item_tcp tcp;
265 struct rte_flow_action actions[2];
266 struct rte_flow_action_queue queue;
269 /* Tunnel information. */
270 struct mlx5_flow_tunnel_info {
271 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
272 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
275 static struct mlx5_flow_tunnel_info tunnels_info[] = {
277 .tunnel = MLX5_FLOW_LAYER_VXLAN,
278 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
281 .tunnel = MLX5_FLOW_LAYER_GENEVE,
282 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
285 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
286 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
289 .tunnel = MLX5_FLOW_LAYER_GRE,
290 .ptype = RTE_PTYPE_TUNNEL_GRE,
293 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
294 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
297 .tunnel = MLX5_FLOW_LAYER_MPLS,
298 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
301 .tunnel = MLX5_FLOW_LAYER_NVGRE,
302 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
305 .tunnel = MLX5_FLOW_LAYER_IPIP,
306 .ptype = RTE_PTYPE_TUNNEL_IP,
309 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
310 .ptype = RTE_PTYPE_TUNNEL_IP,
313 .tunnel = MLX5_FLOW_LAYER_GTP,
314 .ptype = RTE_PTYPE_TUNNEL_GTPU,
319 * Translate tag ID to register.
322 * Pointer to the Ethernet device structure.
324 * The feature that request the register.
326 * The request register ID.
328 * Error description in case of any.
331 * The request register on success, a negative errno
332 * value otherwise and rte_errno is set.
335 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
336 enum mlx5_feature_name feature,
338 struct rte_flow_error *error)
340 struct mlx5_priv *priv = dev->data->dev_private;
341 struct mlx5_dev_config *config = &priv->config;
342 enum modify_reg start_reg;
343 bool skip_mtr_reg = false;
346 case MLX5_HAIRPIN_RX:
348 case MLX5_HAIRPIN_TX:
350 case MLX5_METADATA_RX:
351 switch (config->dv_xmeta_en) {
352 case MLX5_XMETA_MODE_LEGACY:
354 case MLX5_XMETA_MODE_META16:
356 case MLX5_XMETA_MODE_META32:
360 case MLX5_METADATA_TX:
362 case MLX5_METADATA_FDB:
363 switch (config->dv_xmeta_en) {
364 case MLX5_XMETA_MODE_LEGACY:
366 case MLX5_XMETA_MODE_META16:
368 case MLX5_XMETA_MODE_META32:
373 switch (config->dv_xmeta_en) {
374 case MLX5_XMETA_MODE_LEGACY:
376 case MLX5_XMETA_MODE_META16:
378 case MLX5_XMETA_MODE_META32:
384 * If meter color and flow match share one register, flow match
385 * should use the meter color register for match.
387 if (priv->mtr_reg_share)
388 return priv->mtr_color_reg;
390 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
393 MLX5_ASSERT(priv->mtr_color_reg != REG_NONE);
394 return priv->mtr_color_reg;
397 * Metadata COPY_MARK register using is in meter suffix sub
398 * flow while with meter. It's safe to share the same register.
400 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
403 * If meter is enable, it will engage the register for color
404 * match and flow match. If meter color match is not using the
405 * REG_C_2, need to skip the REG_C_x be used by meter color
407 * If meter is disable, free to use all available registers.
409 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
410 (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
411 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
412 if (id > (REG_C_7 - start_reg))
413 return rte_flow_error_set(error, EINVAL,
414 RTE_FLOW_ERROR_TYPE_ITEM,
415 NULL, "invalid tag id");
416 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
417 return rte_flow_error_set(error, ENOTSUP,
418 RTE_FLOW_ERROR_TYPE_ITEM,
419 NULL, "unsupported tag id");
421 * This case means meter is using the REG_C_x great than 2.
422 * Take care not to conflict with meter color REG_C_x.
423 * If the available index REG_C_y >= REG_C_x, skip the
426 if (skip_mtr_reg && config->flow_mreg_c
427 [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
428 if (id >= (REG_C_7 - start_reg))
429 return rte_flow_error_set(error, EINVAL,
430 RTE_FLOW_ERROR_TYPE_ITEM,
431 NULL, "invalid tag id");
432 if (config->flow_mreg_c
433 [id + 1 + start_reg - REG_C_0] != REG_NONE)
434 return config->flow_mreg_c
435 [id + 1 + start_reg - REG_C_0];
436 return rte_flow_error_set(error, ENOTSUP,
437 RTE_FLOW_ERROR_TYPE_ITEM,
438 NULL, "unsupported tag id");
440 return config->flow_mreg_c[id + start_reg - REG_C_0];
443 return rte_flow_error_set(error, EINVAL,
444 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
445 NULL, "invalid feature name");
449 * Check extensive flow metadata register support.
452 * Pointer to rte_eth_dev structure.
455 * True if device supports extensive flow metadata register, otherwise false.
458 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
460 struct mlx5_priv *priv = dev->data->dev_private;
461 struct mlx5_dev_config *config = &priv->config;
464 * Having available reg_c can be regarded inclusively as supporting
465 * extensive flow metadata register, which could mean,
466 * - metadata register copy action by modify header.
467 * - 16 modify header actions is supported.
468 * - reg_c's are preserved across different domain (FDB and NIC) on
469 * packet loopback by flow lookup miss.
471 return config->flow_mreg_c[2] != REG_NONE;
475 * Verify the @p item specifications (spec, last, mask) are compatible with the
479 * Item specification.
481 * @p item->mask or flow default bit-masks.
482 * @param[in] nic_mask
483 * Bit-masks covering supported fields by the NIC to compare with user mask.
485 * Bit-masks size in bytes.
487 * Pointer to error structure.
490 * 0 on success, a negative errno value otherwise and rte_errno is set.
493 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
495 const uint8_t *nic_mask,
497 struct rte_flow_error *error)
501 MLX5_ASSERT(nic_mask);
502 for (i = 0; i < size; ++i)
503 if ((nic_mask[i] | mask[i]) != nic_mask[i])
504 return rte_flow_error_set(error, ENOTSUP,
505 RTE_FLOW_ERROR_TYPE_ITEM,
507 "mask enables non supported"
509 if (!item->spec && (item->mask || item->last))
510 return rte_flow_error_set(error, EINVAL,
511 RTE_FLOW_ERROR_TYPE_ITEM, item,
512 "mask/last without a spec is not"
514 if (item->spec && item->last) {
520 for (i = 0; i < size; ++i) {
521 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
522 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
524 ret = memcmp(spec, last, size);
526 return rte_flow_error_set(error, EINVAL,
527 RTE_FLOW_ERROR_TYPE_ITEM,
529 "range is not valid");
535 * Adjust the hash fields according to the @p flow information.
537 * @param[in] dev_flow.
538 * Pointer to the mlx5_flow.
540 * 1 when the hash field is for a tunnel item.
541 * @param[in] layer_types
543 * @param[in] hash_fields
547 * The hash fields that should be used.
550 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
551 int tunnel __rte_unused, uint64_t layer_types,
552 uint64_t hash_fields)
554 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
555 int rss_request_inner = rss_desc->level >= 2;
557 /* Check RSS hash level for tunnel. */
558 if (tunnel && rss_request_inner)
559 hash_fields |= IBV_RX_HASH_INNER;
560 else if (tunnel || rss_request_inner)
563 /* Check if requested layer matches RSS hash fields. */
564 if (!(rss_desc->types & layer_types))
570 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
571 * if several tunnel rules are used on this queue, the tunnel ptype will be
575 * Rx queue to update.
578 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
581 uint32_t tunnel_ptype = 0;
583 /* Look up for the ptype to use. */
584 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
585 if (!rxq_ctrl->flow_tunnels_n[i])
588 tunnel_ptype = tunnels_info[i].ptype;
594 rxq_ctrl->rxq.tunnel = tunnel_ptype;
598 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
602 * Pointer to the Ethernet device structure.
603 * @param[in] dev_handle
604 * Pointer to device flow handle structure.
607 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
608 struct mlx5_flow_handle *dev_handle)
610 struct mlx5_priv *priv = dev->data->dev_private;
611 const int mark = dev_handle->mark;
612 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
613 struct mlx5_hrxq *hrxq;
616 if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
618 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
619 dev_handle->rix_hrxq);
622 for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
623 int idx = hrxq->ind_table->queues[i];
624 struct mlx5_rxq_ctrl *rxq_ctrl =
625 container_of((*priv->rxqs)[idx],
626 struct mlx5_rxq_ctrl, rxq);
629 * To support metadata register copy on Tx loopback,
630 * this must be always enabled (metadata may arive
631 * from other port - not from local flows only.
633 if (priv->config.dv_flow_en &&
634 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
635 mlx5_flow_ext_mreg_supported(dev)) {
636 rxq_ctrl->rxq.mark = 1;
637 rxq_ctrl->flow_mark_n = 1;
639 rxq_ctrl->rxq.mark = 1;
640 rxq_ctrl->flow_mark_n++;
645 /* Increase the counter matching the flow. */
646 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
647 if ((tunnels_info[j].tunnel &
648 dev_handle->layers) ==
649 tunnels_info[j].tunnel) {
650 rxq_ctrl->flow_tunnels_n[j]++;
654 flow_rxq_tunnel_ptype_update(rxq_ctrl);
660 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
663 * Pointer to the Ethernet device structure.
665 * Pointer to flow structure.
668 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
670 struct mlx5_priv *priv = dev->data->dev_private;
672 struct mlx5_flow_handle *dev_handle;
674 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
675 handle_idx, dev_handle, next)
676 flow_drv_rxq_flags_set(dev, dev_handle);
680 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
681 * device flow if no other flow uses it with the same kind of request.
684 * Pointer to Ethernet device.
685 * @param[in] dev_handle
686 * Pointer to the device flow handle structure.
689 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
690 struct mlx5_flow_handle *dev_handle)
692 struct mlx5_priv *priv = dev->data->dev_private;
693 const int mark = dev_handle->mark;
694 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
695 struct mlx5_hrxq *hrxq;
698 if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
700 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
701 dev_handle->rix_hrxq);
704 MLX5_ASSERT(dev->data->dev_started);
705 for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
706 int idx = hrxq->ind_table->queues[i];
707 struct mlx5_rxq_ctrl *rxq_ctrl =
708 container_of((*priv->rxqs)[idx],
709 struct mlx5_rxq_ctrl, rxq);
711 if (priv->config.dv_flow_en &&
712 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
713 mlx5_flow_ext_mreg_supported(dev)) {
714 rxq_ctrl->rxq.mark = 1;
715 rxq_ctrl->flow_mark_n = 1;
717 rxq_ctrl->flow_mark_n--;
718 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
723 /* Decrease the counter matching the flow. */
724 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
725 if ((tunnels_info[j].tunnel &
726 dev_handle->layers) ==
727 tunnels_info[j].tunnel) {
728 rxq_ctrl->flow_tunnels_n[j]--;
732 flow_rxq_tunnel_ptype_update(rxq_ctrl);
738 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
739 * @p flow if no other flow uses it with the same kind of request.
742 * Pointer to Ethernet device.
744 * Pointer to the flow.
747 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
749 struct mlx5_priv *priv = dev->data->dev_private;
751 struct mlx5_flow_handle *dev_handle;
753 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
754 handle_idx, dev_handle, next)
755 flow_drv_rxq_flags_trim(dev, dev_handle);
759 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
762 * Pointer to Ethernet device.
765 flow_rxq_flags_clear(struct rte_eth_dev *dev)
767 struct mlx5_priv *priv = dev->data->dev_private;
770 for (i = 0; i != priv->rxqs_n; ++i) {
771 struct mlx5_rxq_ctrl *rxq_ctrl;
774 if (!(*priv->rxqs)[i])
776 rxq_ctrl = container_of((*priv->rxqs)[i],
777 struct mlx5_rxq_ctrl, rxq);
778 rxq_ctrl->flow_mark_n = 0;
779 rxq_ctrl->rxq.mark = 0;
780 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
781 rxq_ctrl->flow_tunnels_n[j] = 0;
782 rxq_ctrl->rxq.tunnel = 0;
787 * Set the Rx queue dynamic metadata (mask and offset) for a flow
790 * Pointer to the Ethernet device structure.
793 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
795 struct mlx5_priv *priv = dev->data->dev_private;
796 struct mlx5_rxq_data *data;
799 for (i = 0; i != priv->rxqs_n; ++i) {
800 if (!(*priv->rxqs)[i])
802 data = (*priv->rxqs)[i];
803 if (!rte_flow_dynf_metadata_avail()) {
805 data->flow_meta_mask = 0;
806 data->flow_meta_offset = -1;
809 data->flow_meta_mask = rte_flow_dynf_metadata_mask;
810 data->flow_meta_offset = rte_flow_dynf_metadata_offs;
816 * return a pointer to the desired action in the list of actions.
819 * The list of actions to search the action in.
821 * The action to find.
824 * Pointer to the action in the list, if found. NULL otherwise.
826 const struct rte_flow_action *
827 mlx5_flow_find_action(const struct rte_flow_action *actions,
828 enum rte_flow_action_type action)
832 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
833 if (actions->type == action)
839 * Validate the flag action.
841 * @param[in] action_flags
842 * Bit-fields that holds the actions detected until now.
844 * Attributes of flow that includes this action.
846 * Pointer to error structure.
849 * 0 on success, a negative errno value otherwise and rte_errno is set.
852 mlx5_flow_validate_action_flag(uint64_t action_flags,
853 const struct rte_flow_attr *attr,
854 struct rte_flow_error *error)
856 if (action_flags & MLX5_FLOW_ACTION_MARK)
857 return rte_flow_error_set(error, EINVAL,
858 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
859 "can't mark and flag in same flow");
860 if (action_flags & MLX5_FLOW_ACTION_FLAG)
861 return rte_flow_error_set(error, EINVAL,
862 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
864 " actions in same flow");
866 return rte_flow_error_set(error, ENOTSUP,
867 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
868 "flag action not supported for "
874 * Validate the mark action.
877 * Pointer to the queue action.
878 * @param[in] action_flags
879 * Bit-fields that holds the actions detected until now.
881 * Attributes of flow that includes this action.
883 * Pointer to error structure.
886 * 0 on success, a negative errno value otherwise and rte_errno is set.
889 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
890 uint64_t action_flags,
891 const struct rte_flow_attr *attr,
892 struct rte_flow_error *error)
894 const struct rte_flow_action_mark *mark = action->conf;
897 return rte_flow_error_set(error, EINVAL,
898 RTE_FLOW_ERROR_TYPE_ACTION,
900 "configuration cannot be null");
901 if (mark->id >= MLX5_FLOW_MARK_MAX)
902 return rte_flow_error_set(error, EINVAL,
903 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
905 "mark id must in 0 <= id < "
906 RTE_STR(MLX5_FLOW_MARK_MAX));
907 if (action_flags & MLX5_FLOW_ACTION_FLAG)
908 return rte_flow_error_set(error, EINVAL,
909 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
910 "can't flag and mark in same flow");
911 if (action_flags & MLX5_FLOW_ACTION_MARK)
912 return rte_flow_error_set(error, EINVAL,
913 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
914 "can't have 2 mark actions in same"
917 return rte_flow_error_set(error, ENOTSUP,
918 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
919 "mark action not supported for "
925 * Validate the drop action.
927 * @param[in] action_flags
928 * Bit-fields that holds the actions detected until now.
930 * Attributes of flow that includes this action.
932 * Pointer to error structure.
935 * 0 on success, a negative errno value otherwise and rte_errno is set.
938 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
939 const struct rte_flow_attr *attr,
940 struct rte_flow_error *error)
943 return rte_flow_error_set(error, ENOTSUP,
944 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
945 "drop action not supported for "
951 * Validate the queue action.
954 * Pointer to the queue action.
955 * @param[in] action_flags
956 * Bit-fields that holds the actions detected until now.
958 * Pointer to the Ethernet device structure.
960 * Attributes of flow that includes this action.
962 * Pointer to error structure.
965 * 0 on success, a negative errno value otherwise and rte_errno is set.
968 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
969 uint64_t action_flags,
970 struct rte_eth_dev *dev,
971 const struct rte_flow_attr *attr,
972 struct rte_flow_error *error)
974 struct mlx5_priv *priv = dev->data->dev_private;
975 const struct rte_flow_action_queue *queue = action->conf;
977 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
978 return rte_flow_error_set(error, EINVAL,
979 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
980 "can't have 2 fate actions in"
983 return rte_flow_error_set(error, EINVAL,
984 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
985 NULL, "No Rx queues configured");
986 if (queue->index >= priv->rxqs_n)
987 return rte_flow_error_set(error, EINVAL,
988 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
990 "queue index out of range");
991 if (!(*priv->rxqs)[queue->index])
992 return rte_flow_error_set(error, EINVAL,
993 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
995 "queue is not configured");
997 return rte_flow_error_set(error, ENOTSUP,
998 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
999 "queue action not supported for "
1005 * Validate the rss action.
1008 * Pointer to the queue action.
1009 * @param[in] action_flags
1010 * Bit-fields that holds the actions detected until now.
1012 * Pointer to the Ethernet device structure.
1014 * Attributes of flow that includes this action.
1015 * @param[in] item_flags
1016 * Items that were detected.
1018 * Pointer to error structure.
1021 * 0 on success, a negative errno value otherwise and rte_errno is set.
1024 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1025 uint64_t action_flags,
1026 struct rte_eth_dev *dev,
1027 const struct rte_flow_attr *attr,
1028 uint64_t item_flags,
1029 struct rte_flow_error *error)
1031 struct mlx5_priv *priv = dev->data->dev_private;
1032 const struct rte_flow_action_rss *rss = action->conf;
1033 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1036 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1037 return rte_flow_error_set(error, EINVAL,
1038 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1039 "can't have 2 fate actions"
1041 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1042 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1043 return rte_flow_error_set(error, ENOTSUP,
1044 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1046 "RSS hash function not supported");
1047 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1052 return rte_flow_error_set(error, ENOTSUP,
1053 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1055 "tunnel RSS is not supported");
1056 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1057 if (rss->key_len == 0 && rss->key != NULL)
1058 return rte_flow_error_set(error, ENOTSUP,
1059 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1061 "RSS hash key length 0");
1062 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1063 return rte_flow_error_set(error, ENOTSUP,
1064 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1066 "RSS hash key too small");
1067 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1068 return rte_flow_error_set(error, ENOTSUP,
1069 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1071 "RSS hash key too large");
1072 if (rss->queue_num > priv->config.ind_table_max_size)
1073 return rte_flow_error_set(error, ENOTSUP,
1074 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1076 "number of queues too large");
1077 if (rss->types & MLX5_RSS_HF_MASK)
1078 return rte_flow_error_set(error, ENOTSUP,
1079 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1081 "some RSS protocols are not"
1083 if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1084 !(rss->types & ETH_RSS_IP))
1085 return rte_flow_error_set(error, EINVAL,
1086 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1087 "L3 partial RSS requested but L3 RSS"
1088 " type not specified");
1089 if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1090 !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1091 return rte_flow_error_set(error, EINVAL,
1092 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1093 "L4 partial RSS requested but L4 RSS"
1094 " type not specified");
1096 return rte_flow_error_set(error, EINVAL,
1097 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1098 NULL, "No Rx queues configured");
1099 if (!rss->queue_num)
1100 return rte_flow_error_set(error, EINVAL,
1101 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1102 NULL, "No queues configured");
1103 for (i = 0; i != rss->queue_num; ++i) {
1104 if (rss->queue[i] >= priv->rxqs_n)
1105 return rte_flow_error_set
1107 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1108 &rss->queue[i], "queue index out of range");
1109 if (!(*priv->rxqs)[rss->queue[i]])
1110 return rte_flow_error_set
1111 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1112 &rss->queue[i], "queue is not configured");
1115 return rte_flow_error_set(error, ENOTSUP,
1116 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1117 "rss action not supported for "
1119 if (rss->level > 1 && !tunnel)
1120 return rte_flow_error_set(error, EINVAL,
1121 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1122 "inner RSS is not supported for "
1123 "non-tunnel flows");
1124 if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1125 !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1126 return rte_flow_error_set(error, EINVAL,
1127 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1128 "RSS on eCPRI is not supported now");
1134 * Validate the default miss action.
1136 * @param[in] action_flags
1137 * Bit-fields that holds the actions detected until now.
1139 * Pointer to error structure.
1142 * 0 on success, a negative errno value otherwise and rte_errno is set.
1145 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1146 const struct rte_flow_attr *attr,
1147 struct rte_flow_error *error)
1149 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1150 return rte_flow_error_set(error, EINVAL,
1151 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1152 "can't have 2 fate actions in"
1155 return rte_flow_error_set(error, ENOTSUP,
1156 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1157 "default miss action not supported "
1160 return rte_flow_error_set(error, ENOTSUP,
1161 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1162 "only group 0 is supported");
1164 return rte_flow_error_set(error, ENOTSUP,
1165 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1166 NULL, "transfer is not supported");
1171 * Validate the count action.
1174 * Pointer to the Ethernet device structure.
1176 * Attributes of flow that includes this action.
1178 * Pointer to error structure.
1181 * 0 on success, a negative errno value otherwise and rte_errno is set.
1184 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1185 const struct rte_flow_attr *attr,
1186 struct rte_flow_error *error)
1189 return rte_flow_error_set(error, ENOTSUP,
1190 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1191 "count action not supported for "
1197 * Verify the @p attributes will be correctly understood by the NIC and store
1198 * them in the @p flow if everything is correct.
1201 * Pointer to the Ethernet device structure.
1202 * @param[in] attributes
1203 * Pointer to flow attributes
1205 * Pointer to error structure.
1208 * 0 on success, a negative errno value otherwise and rte_errno is set.
1211 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1212 const struct rte_flow_attr *attributes,
1213 struct rte_flow_error *error)
1215 struct mlx5_priv *priv = dev->data->dev_private;
1216 uint32_t priority_max = priv->config.flow_prio - 1;
1218 if (attributes->group)
1219 return rte_flow_error_set(error, ENOTSUP,
1220 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1221 NULL, "groups is not supported");
1222 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1223 attributes->priority >= priority_max)
1224 return rte_flow_error_set(error, ENOTSUP,
1225 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1226 NULL, "priority out of range");
1227 if (attributes->egress)
1228 return rte_flow_error_set(error, ENOTSUP,
1229 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1230 "egress is not supported");
1231 if (attributes->transfer && !priv->config.dv_esw_en)
1232 return rte_flow_error_set(error, ENOTSUP,
1233 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1234 NULL, "transfer is not supported");
1235 if (!attributes->ingress)
1236 return rte_flow_error_set(error, EINVAL,
1237 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1239 "ingress attribute is mandatory");
1244 * Validate ICMP6 item.
1247 * Item specification.
1248 * @param[in] item_flags
1249 * Bit-fields that holds the items detected until now.
1251 * Pointer to error structure.
1254 * 0 on success, a negative errno value otherwise and rte_errno is set.
1257 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1258 uint64_t item_flags,
1259 uint8_t target_protocol,
1260 struct rte_flow_error *error)
1262 const struct rte_flow_item_icmp6 *mask = item->mask;
1263 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1264 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1265 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1266 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1267 MLX5_FLOW_LAYER_OUTER_L4;
1270 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1271 return rte_flow_error_set(error, EINVAL,
1272 RTE_FLOW_ERROR_TYPE_ITEM, item,
1273 "protocol filtering not compatible"
1274 " with ICMP6 layer");
1275 if (!(item_flags & l3m))
1276 return rte_flow_error_set(error, EINVAL,
1277 RTE_FLOW_ERROR_TYPE_ITEM, item,
1278 "IPv6 is mandatory to filter on"
1280 if (item_flags & l4m)
1281 return rte_flow_error_set(error, EINVAL,
1282 RTE_FLOW_ERROR_TYPE_ITEM, item,
1283 "multiple L4 layers not supported");
1285 mask = &rte_flow_item_icmp6_mask;
1286 ret = mlx5_flow_item_acceptable
1287 (item, (const uint8_t *)mask,
1288 (const uint8_t *)&rte_flow_item_icmp6_mask,
1289 sizeof(struct rte_flow_item_icmp6), error);
1296 * Validate ICMP item.
1299 * Item specification.
1300 * @param[in] item_flags
1301 * Bit-fields that holds the items detected until now.
1303 * Pointer to error structure.
1306 * 0 on success, a negative errno value otherwise and rte_errno is set.
1309 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1310 uint64_t item_flags,
1311 uint8_t target_protocol,
1312 struct rte_flow_error *error)
1314 const struct rte_flow_item_icmp *mask = item->mask;
1315 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1316 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1317 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1318 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1319 MLX5_FLOW_LAYER_OUTER_L4;
1322 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1323 return rte_flow_error_set(error, EINVAL,
1324 RTE_FLOW_ERROR_TYPE_ITEM, item,
1325 "protocol filtering not compatible"
1326 " with ICMP layer");
1327 if (!(item_flags & l3m))
1328 return rte_flow_error_set(error, EINVAL,
1329 RTE_FLOW_ERROR_TYPE_ITEM, item,
1330 "IPv4 is mandatory to filter"
1332 if (item_flags & l4m)
1333 return rte_flow_error_set(error, EINVAL,
1334 RTE_FLOW_ERROR_TYPE_ITEM, item,
1335 "multiple L4 layers not supported");
1337 mask = &rte_flow_item_icmp_mask;
1338 ret = mlx5_flow_item_acceptable
1339 (item, (const uint8_t *)mask,
1340 (const uint8_t *)&rte_flow_item_icmp_mask,
1341 sizeof(struct rte_flow_item_icmp), error);
1348 * Validate Ethernet item.
1351 * Item specification.
1352 * @param[in] item_flags
1353 * Bit-fields that holds the items detected until now.
1355 * Pointer to error structure.
1358 * 0 on success, a negative errno value otherwise and rte_errno is set.
1361 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1362 uint64_t item_flags,
1363 struct rte_flow_error *error)
1365 const struct rte_flow_item_eth *mask = item->mask;
1366 const struct rte_flow_item_eth nic_mask = {
1367 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1368 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1369 .type = RTE_BE16(0xffff),
1372 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1373 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1374 MLX5_FLOW_LAYER_OUTER_L2;
1376 if (item_flags & ethm)
1377 return rte_flow_error_set(error, ENOTSUP,
1378 RTE_FLOW_ERROR_TYPE_ITEM, item,
1379 "multiple L2 layers not supported");
1380 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1381 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1382 return rte_flow_error_set(error, EINVAL,
1383 RTE_FLOW_ERROR_TYPE_ITEM, item,
1384 "L2 layer should not follow "
1386 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1387 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1388 return rte_flow_error_set(error, EINVAL,
1389 RTE_FLOW_ERROR_TYPE_ITEM, item,
1390 "L2 layer should not follow VLAN");
1392 mask = &rte_flow_item_eth_mask;
1393 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1394 (const uint8_t *)&nic_mask,
1395 sizeof(struct rte_flow_item_eth),
1401 * Validate VLAN item.
1404 * Item specification.
1405 * @param[in] item_flags
1406 * Bit-fields that holds the items detected until now.
1408 * Ethernet device flow is being created on.
1410 * Pointer to error structure.
1413 * 0 on success, a negative errno value otherwise and rte_errno is set.
1416 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1417 uint64_t item_flags,
1418 struct rte_eth_dev *dev,
1419 struct rte_flow_error *error)
1421 const struct rte_flow_item_vlan *spec = item->spec;
1422 const struct rte_flow_item_vlan *mask = item->mask;
1423 const struct rte_flow_item_vlan nic_mask = {
1424 .tci = RTE_BE16(UINT16_MAX),
1425 .inner_type = RTE_BE16(UINT16_MAX),
1427 uint16_t vlan_tag = 0;
1428 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1430 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1431 MLX5_FLOW_LAYER_INNER_L4) :
1432 (MLX5_FLOW_LAYER_OUTER_L3 |
1433 MLX5_FLOW_LAYER_OUTER_L4);
1434 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1435 MLX5_FLOW_LAYER_OUTER_VLAN;
1437 if (item_flags & vlanm)
1438 return rte_flow_error_set(error, EINVAL,
1439 RTE_FLOW_ERROR_TYPE_ITEM, item,
1440 "multiple VLAN layers not supported");
1441 else if ((item_flags & l34m) != 0)
1442 return rte_flow_error_set(error, EINVAL,
1443 RTE_FLOW_ERROR_TYPE_ITEM, item,
1444 "VLAN cannot follow L3/L4 layer");
1446 mask = &rte_flow_item_vlan_mask;
1447 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1448 (const uint8_t *)&nic_mask,
1449 sizeof(struct rte_flow_item_vlan),
1453 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1454 struct mlx5_priv *priv = dev->data->dev_private;
1456 if (priv->vmwa_context) {
1458 * Non-NULL context means we have a virtual machine
1459 * and SR-IOV enabled, we have to create VLAN interface
1460 * to make hypervisor to setup E-Switch vport
1461 * context correctly. We avoid creating the multiple
1462 * VLAN interfaces, so we cannot support VLAN tag mask.
1464 return rte_flow_error_set(error, EINVAL,
1465 RTE_FLOW_ERROR_TYPE_ITEM,
1467 "VLAN tag mask is not"
1468 " supported in virtual"
1473 vlan_tag = spec->tci;
1474 vlan_tag &= mask->tci;
1477 * From verbs perspective an empty VLAN is equivalent
1478 * to a packet without VLAN layer.
1481 return rte_flow_error_set(error, EINVAL,
1482 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1484 "VLAN cannot be empty");
1489 * Validate IPV4 item.
1492 * Item specification.
1493 * @param[in] item_flags
1494 * Bit-fields that holds the items detected until now.
1495 * @param[in] last_item
1496 * Previous validated item in the pattern items.
1497 * @param[in] ether_type
1498 * Type in the ethernet layer header (including dot1q).
1499 * @param[in] acc_mask
1500 * Acceptable mask, if NULL default internal default mask
1501 * will be used to check whether item fields are supported.
1503 * Pointer to error structure.
1506 * 0 on success, a negative errno value otherwise and rte_errno is set.
1509 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1510 uint64_t item_flags,
1512 uint16_t ether_type,
1513 const struct rte_flow_item_ipv4 *acc_mask,
1514 struct rte_flow_error *error)
1516 const struct rte_flow_item_ipv4 *mask = item->mask;
1517 const struct rte_flow_item_ipv4 *spec = item->spec;
1518 const struct rte_flow_item_ipv4 nic_mask = {
1520 .src_addr = RTE_BE32(0xffffffff),
1521 .dst_addr = RTE_BE32(0xffffffff),
1522 .type_of_service = 0xff,
1523 .next_proto_id = 0xff,
1526 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1527 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1528 MLX5_FLOW_LAYER_OUTER_L3;
1529 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1530 MLX5_FLOW_LAYER_OUTER_L4;
1532 uint8_t next_proto = 0xFF;
1533 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1534 MLX5_FLOW_LAYER_OUTER_VLAN |
1535 MLX5_FLOW_LAYER_INNER_VLAN);
1537 if ((last_item & l2_vlan) && ether_type &&
1538 ether_type != RTE_ETHER_TYPE_IPV4)
1539 return rte_flow_error_set(error, EINVAL,
1540 RTE_FLOW_ERROR_TYPE_ITEM, item,
1541 "IPv4 cannot follow L2/VLAN layer "
1542 "which ether type is not IPv4");
1543 if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1545 next_proto = mask->hdr.next_proto_id &
1546 spec->hdr.next_proto_id;
1547 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1548 return rte_flow_error_set(error, EINVAL,
1549 RTE_FLOW_ERROR_TYPE_ITEM,
1554 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1555 return rte_flow_error_set(error, EINVAL,
1556 RTE_FLOW_ERROR_TYPE_ITEM, item,
1557 "wrong tunnel type - IPv6 specified "
1558 "but IPv4 item provided");
1559 if (item_flags & l3m)
1560 return rte_flow_error_set(error, ENOTSUP,
1561 RTE_FLOW_ERROR_TYPE_ITEM, item,
1562 "multiple L3 layers not supported");
1563 else if (item_flags & l4m)
1564 return rte_flow_error_set(error, EINVAL,
1565 RTE_FLOW_ERROR_TYPE_ITEM, item,
1566 "L3 cannot follow an L4 layer.");
1567 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1568 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1569 return rte_flow_error_set(error, EINVAL,
1570 RTE_FLOW_ERROR_TYPE_ITEM, item,
1571 "L3 cannot follow an NVGRE layer.");
1573 mask = &rte_flow_item_ipv4_mask;
1574 else if (mask->hdr.next_proto_id != 0 &&
1575 mask->hdr.next_proto_id != 0xff)
1576 return rte_flow_error_set(error, EINVAL,
1577 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1578 "partial mask is not supported"
1580 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1581 acc_mask ? (const uint8_t *)acc_mask
1582 : (const uint8_t *)&nic_mask,
1583 sizeof(struct rte_flow_item_ipv4),
1591 * Validate IPV6 item.
1594 * Item specification.
1595 * @param[in] item_flags
1596 * Bit-fields that holds the items detected until now.
1597 * @param[in] last_item
1598 * Previous validated item in the pattern items.
1599 * @param[in] ether_type
1600 * Type in the ethernet layer header (including dot1q).
1601 * @param[in] acc_mask
1602 * Acceptable mask, if NULL default internal default mask
1603 * will be used to check whether item fields are supported.
1605 * Pointer to error structure.
1608 * 0 on success, a negative errno value otherwise and rte_errno is set.
1611 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1612 uint64_t item_flags,
1614 uint16_t ether_type,
1615 const struct rte_flow_item_ipv6 *acc_mask,
1616 struct rte_flow_error *error)
1618 const struct rte_flow_item_ipv6 *mask = item->mask;
1619 const struct rte_flow_item_ipv6 *spec = item->spec;
1620 const struct rte_flow_item_ipv6 nic_mask = {
1623 "\xff\xff\xff\xff\xff\xff\xff\xff"
1624 "\xff\xff\xff\xff\xff\xff\xff\xff",
1626 "\xff\xff\xff\xff\xff\xff\xff\xff"
1627 "\xff\xff\xff\xff\xff\xff\xff\xff",
1628 .vtc_flow = RTE_BE32(0xffffffff),
1632 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1633 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1634 MLX5_FLOW_LAYER_OUTER_L3;
1635 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1636 MLX5_FLOW_LAYER_OUTER_L4;
1638 uint8_t next_proto = 0xFF;
1639 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1640 MLX5_FLOW_LAYER_OUTER_VLAN |
1641 MLX5_FLOW_LAYER_INNER_VLAN);
1643 if ((last_item & l2_vlan) && ether_type &&
1644 ether_type != RTE_ETHER_TYPE_IPV6)
1645 return rte_flow_error_set(error, EINVAL,
1646 RTE_FLOW_ERROR_TYPE_ITEM, item,
1647 "IPv6 cannot follow L2/VLAN layer "
1648 "which ether type is not IPv6");
1649 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1651 next_proto = mask->hdr.proto & spec->hdr.proto;
1652 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1653 return rte_flow_error_set(error, EINVAL,
1654 RTE_FLOW_ERROR_TYPE_ITEM,
1659 if (item_flags & MLX5_FLOW_LAYER_IPIP)
1660 return rte_flow_error_set(error, EINVAL,
1661 RTE_FLOW_ERROR_TYPE_ITEM, item,
1662 "wrong tunnel type - IPv4 specified "
1663 "but IPv6 item provided");
1664 if (item_flags & l3m)
1665 return rte_flow_error_set(error, ENOTSUP,
1666 RTE_FLOW_ERROR_TYPE_ITEM, item,
1667 "multiple L3 layers not supported");
1668 else if (item_flags & l4m)
1669 return rte_flow_error_set(error, EINVAL,
1670 RTE_FLOW_ERROR_TYPE_ITEM, item,
1671 "L3 cannot follow an L4 layer.");
1672 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1673 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1674 return rte_flow_error_set(error, EINVAL,
1675 RTE_FLOW_ERROR_TYPE_ITEM, item,
1676 "L3 cannot follow an NVGRE layer.");
1678 mask = &rte_flow_item_ipv6_mask;
1679 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1680 acc_mask ? (const uint8_t *)acc_mask
1681 : (const uint8_t *)&nic_mask,
1682 sizeof(struct rte_flow_item_ipv6),
1690 * Validate UDP item.
1693 * Item specification.
1694 * @param[in] item_flags
1695 * Bit-fields that holds the items detected until now.
1696 * @param[in] target_protocol
1697 * The next protocol in the previous item.
1698 * @param[in] flow_mask
1699 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1701 * Pointer to error structure.
1704 * 0 on success, a negative errno value otherwise and rte_errno is set.
1707 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1708 uint64_t item_flags,
1709 uint8_t target_protocol,
1710 struct rte_flow_error *error)
1712 const struct rte_flow_item_udp *mask = item->mask;
1713 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1714 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1715 MLX5_FLOW_LAYER_OUTER_L3;
1716 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1717 MLX5_FLOW_LAYER_OUTER_L4;
1720 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1721 return rte_flow_error_set(error, EINVAL,
1722 RTE_FLOW_ERROR_TYPE_ITEM, item,
1723 "protocol filtering not compatible"
1725 if (!(item_flags & l3m))
1726 return rte_flow_error_set(error, EINVAL,
1727 RTE_FLOW_ERROR_TYPE_ITEM, item,
1728 "L3 is mandatory to filter on L4");
1729 if (item_flags & l4m)
1730 return rte_flow_error_set(error, EINVAL,
1731 RTE_FLOW_ERROR_TYPE_ITEM, item,
1732 "multiple L4 layers not supported");
1734 mask = &rte_flow_item_udp_mask;
1735 ret = mlx5_flow_item_acceptable
1736 (item, (const uint8_t *)mask,
1737 (const uint8_t *)&rte_flow_item_udp_mask,
1738 sizeof(struct rte_flow_item_udp), error);
1745 * Validate TCP item.
1748 * Item specification.
1749 * @param[in] item_flags
1750 * Bit-fields that holds the items detected until now.
1751 * @param[in] target_protocol
1752 * The next protocol in the previous item.
1754 * Pointer to error structure.
1757 * 0 on success, a negative errno value otherwise and rte_errno is set.
1760 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1761 uint64_t item_flags,
1762 uint8_t target_protocol,
1763 const struct rte_flow_item_tcp *flow_mask,
1764 struct rte_flow_error *error)
1766 const struct rte_flow_item_tcp *mask = item->mask;
1767 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1768 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1769 MLX5_FLOW_LAYER_OUTER_L3;
1770 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1771 MLX5_FLOW_LAYER_OUTER_L4;
1774 MLX5_ASSERT(flow_mask);
1775 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1776 return rte_flow_error_set(error, EINVAL,
1777 RTE_FLOW_ERROR_TYPE_ITEM, item,
1778 "protocol filtering not compatible"
1780 if (!(item_flags & l3m))
1781 return rte_flow_error_set(error, EINVAL,
1782 RTE_FLOW_ERROR_TYPE_ITEM, item,
1783 "L3 is mandatory to filter on L4");
1784 if (item_flags & l4m)
1785 return rte_flow_error_set(error, EINVAL,
1786 RTE_FLOW_ERROR_TYPE_ITEM, item,
1787 "multiple L4 layers not supported");
1789 mask = &rte_flow_item_tcp_mask;
1790 ret = mlx5_flow_item_acceptable
1791 (item, (const uint8_t *)mask,
1792 (const uint8_t *)flow_mask,
1793 sizeof(struct rte_flow_item_tcp), error);
1800 * Validate VXLAN item.
1803 * Item specification.
1804 * @param[in] item_flags
1805 * Bit-fields that holds the items detected until now.
1806 * @param[in] target_protocol
1807 * The next protocol in the previous item.
1809 * Pointer to error structure.
1812 * 0 on success, a negative errno value otherwise and rte_errno is set.
1815 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1816 uint64_t item_flags,
1817 struct rte_flow_error *error)
1819 const struct rte_flow_item_vxlan *spec = item->spec;
1820 const struct rte_flow_item_vxlan *mask = item->mask;
1825 } id = { .vlan_id = 0, };
1828 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1829 return rte_flow_error_set(error, ENOTSUP,
1830 RTE_FLOW_ERROR_TYPE_ITEM, item,
1831 "multiple tunnel layers not"
1834 * Verify only UDPv4 is present as defined in
1835 * https://tools.ietf.org/html/rfc7348
1837 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1838 return rte_flow_error_set(error, EINVAL,
1839 RTE_FLOW_ERROR_TYPE_ITEM, item,
1840 "no outer UDP layer found");
1842 mask = &rte_flow_item_vxlan_mask;
1843 ret = mlx5_flow_item_acceptable
1844 (item, (const uint8_t *)mask,
1845 (const uint8_t *)&rte_flow_item_vxlan_mask,
1846 sizeof(struct rte_flow_item_vxlan),
1851 memcpy(&id.vni[1], spec->vni, 3);
1852 memcpy(&id.vni[1], mask->vni, 3);
1854 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1855 return rte_flow_error_set(error, ENOTSUP,
1856 RTE_FLOW_ERROR_TYPE_ITEM, item,
1857 "VXLAN tunnel must be fully defined");
1862 * Validate VXLAN_GPE item.
1865 * Item specification.
1866 * @param[in] item_flags
1867 * Bit-fields that holds the items detected until now.
1869 * Pointer to the private data structure.
1870 * @param[in] target_protocol
1871 * The next protocol in the previous item.
1873 * Pointer to error structure.
1876 * 0 on success, a negative errno value otherwise and rte_errno is set.
1879 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1880 uint64_t item_flags,
1881 struct rte_eth_dev *dev,
1882 struct rte_flow_error *error)
1884 struct mlx5_priv *priv = dev->data->dev_private;
1885 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1886 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1891 } id = { .vlan_id = 0, };
1893 if (!priv->config.l3_vxlan_en)
1894 return rte_flow_error_set(error, ENOTSUP,
1895 RTE_FLOW_ERROR_TYPE_ITEM, item,
1896 "L3 VXLAN is not enabled by device"
1897 " parameter and/or not configured in"
1899 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1900 return rte_flow_error_set(error, ENOTSUP,
1901 RTE_FLOW_ERROR_TYPE_ITEM, item,
1902 "multiple tunnel layers not"
1905 * Verify only UDPv4 is present as defined in
1906 * https://tools.ietf.org/html/rfc7348
1908 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1909 return rte_flow_error_set(error, EINVAL,
1910 RTE_FLOW_ERROR_TYPE_ITEM, item,
1911 "no outer UDP layer found");
1913 mask = &rte_flow_item_vxlan_gpe_mask;
1914 ret = mlx5_flow_item_acceptable
1915 (item, (const uint8_t *)mask,
1916 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1917 sizeof(struct rte_flow_item_vxlan_gpe),
1923 return rte_flow_error_set(error, ENOTSUP,
1924 RTE_FLOW_ERROR_TYPE_ITEM,
1926 "VxLAN-GPE protocol"
1928 memcpy(&id.vni[1], spec->vni, 3);
1929 memcpy(&id.vni[1], mask->vni, 3);
1931 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1932 return rte_flow_error_set(error, ENOTSUP,
1933 RTE_FLOW_ERROR_TYPE_ITEM, item,
1934 "VXLAN-GPE tunnel must be fully"
1939 * Validate GRE Key item.
1942 * Item specification.
1943 * @param[in] item_flags
1944 * Bit flags to mark detected items.
1945 * @param[in] gre_item
1946 * Pointer to gre_item
1948 * Pointer to error structure.
1951 * 0 on success, a negative errno value otherwise and rte_errno is set.
1954 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1955 uint64_t item_flags,
1956 const struct rte_flow_item *gre_item,
1957 struct rte_flow_error *error)
1959 const rte_be32_t *mask = item->mask;
1961 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1962 const struct rte_flow_item_gre *gre_spec;
1963 const struct rte_flow_item_gre *gre_mask;
1965 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1966 return rte_flow_error_set(error, ENOTSUP,
1967 RTE_FLOW_ERROR_TYPE_ITEM, item,
1968 "Multiple GRE key not support");
1969 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
1970 return rte_flow_error_set(error, ENOTSUP,
1971 RTE_FLOW_ERROR_TYPE_ITEM, item,
1972 "No preceding GRE header");
1973 if (item_flags & MLX5_FLOW_LAYER_INNER)
1974 return rte_flow_error_set(error, ENOTSUP,
1975 RTE_FLOW_ERROR_TYPE_ITEM, item,
1976 "GRE key following a wrong item");
1977 gre_mask = gre_item->mask;
1979 gre_mask = &rte_flow_item_gre_mask;
1980 gre_spec = gre_item->spec;
1981 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
1982 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
1983 return rte_flow_error_set(error, EINVAL,
1984 RTE_FLOW_ERROR_TYPE_ITEM, item,
1985 "Key bit must be on");
1988 mask = &gre_key_default_mask;
1989 ret = mlx5_flow_item_acceptable
1990 (item, (const uint8_t *)mask,
1991 (const uint8_t *)&gre_key_default_mask,
1992 sizeof(rte_be32_t), error);
1997 * Validate GRE item.
2000 * Item specification.
2001 * @param[in] item_flags
2002 * Bit flags to mark detected items.
2003 * @param[in] target_protocol
2004 * The next protocol in the previous item.
2006 * Pointer to error structure.
2009 * 0 on success, a negative errno value otherwise and rte_errno is set.
2012 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2013 uint64_t item_flags,
2014 uint8_t target_protocol,
2015 struct rte_flow_error *error)
2017 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2018 const struct rte_flow_item_gre *mask = item->mask;
2020 const struct rte_flow_item_gre nic_mask = {
2021 .c_rsvd0_ver = RTE_BE16(0xB000),
2022 .protocol = RTE_BE16(UINT16_MAX),
2025 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2026 return rte_flow_error_set(error, EINVAL,
2027 RTE_FLOW_ERROR_TYPE_ITEM, item,
2028 "protocol filtering not compatible"
2029 " with this GRE layer");
2030 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2031 return rte_flow_error_set(error, ENOTSUP,
2032 RTE_FLOW_ERROR_TYPE_ITEM, item,
2033 "multiple tunnel layers not"
2035 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2036 return rte_flow_error_set(error, ENOTSUP,
2037 RTE_FLOW_ERROR_TYPE_ITEM, item,
2038 "L3 Layer is missing");
2040 mask = &rte_flow_item_gre_mask;
2041 ret = mlx5_flow_item_acceptable
2042 (item, (const uint8_t *)mask,
2043 (const uint8_t *)&nic_mask,
2044 sizeof(struct rte_flow_item_gre), error);
2047 #ifndef HAVE_MLX5DV_DR
2048 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2049 if (spec && (spec->protocol & mask->protocol))
2050 return rte_flow_error_set(error, ENOTSUP,
2051 RTE_FLOW_ERROR_TYPE_ITEM, item,
2052 "without MPLS support the"
2053 " specification cannot be used for"
2061 * Validate Geneve item.
2064 * Item specification.
2065 * @param[in] itemFlags
2066 * Bit-fields that holds the items detected until now.
2068 * Pointer to the private data structure.
2070 * Pointer to error structure.
2073 * 0 on success, a negative errno value otherwise and rte_errno is set.
2077 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2078 uint64_t item_flags,
2079 struct rte_eth_dev *dev,
2080 struct rte_flow_error *error)
2082 struct mlx5_priv *priv = dev->data->dev_private;
2083 const struct rte_flow_item_geneve *spec = item->spec;
2084 const struct rte_flow_item_geneve *mask = item->mask;
2087 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2088 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2089 const struct rte_flow_item_geneve nic_mask = {
2090 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2091 .vni = "\xff\xff\xff",
2092 .protocol = RTE_BE16(UINT16_MAX),
2095 if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2096 return rte_flow_error_set(error, ENOTSUP,
2097 RTE_FLOW_ERROR_TYPE_ITEM, item,
2098 "L3 Geneve is not enabled by device"
2099 " parameter and/or not configured in"
2101 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2102 return rte_flow_error_set(error, ENOTSUP,
2103 RTE_FLOW_ERROR_TYPE_ITEM, item,
2104 "multiple tunnel layers not"
2107 * Verify only UDPv4 is present as defined in
2108 * https://tools.ietf.org/html/rfc7348
2110 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2111 return rte_flow_error_set(error, EINVAL,
2112 RTE_FLOW_ERROR_TYPE_ITEM, item,
2113 "no outer UDP layer found");
2115 mask = &rte_flow_item_geneve_mask;
2116 ret = mlx5_flow_item_acceptable
2117 (item, (const uint8_t *)mask,
2118 (const uint8_t *)&nic_mask,
2119 sizeof(struct rte_flow_item_geneve), error);
2123 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2124 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2125 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2126 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2127 return rte_flow_error_set(error, ENOTSUP,
2128 RTE_FLOW_ERROR_TYPE_ITEM,
2130 "Geneve protocol unsupported"
2131 " fields are being used");
2132 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2133 return rte_flow_error_set
2135 RTE_FLOW_ERROR_TYPE_ITEM,
2137 "Unsupported Geneve options length");
2139 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2140 return rte_flow_error_set
2142 RTE_FLOW_ERROR_TYPE_ITEM, item,
2143 "Geneve tunnel must be fully defined");
2148 * Validate MPLS item.
2151 * Pointer to the rte_eth_dev structure.
2153 * Item specification.
2154 * @param[in] item_flags
2155 * Bit-fields that holds the items detected until now.
2156 * @param[in] prev_layer
2157 * The protocol layer indicated in previous item.
2159 * Pointer to error structure.
2162 * 0 on success, a negative errno value otherwise and rte_errno is set.
2165 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2166 const struct rte_flow_item *item __rte_unused,
2167 uint64_t item_flags __rte_unused,
2168 uint64_t prev_layer __rte_unused,
2169 struct rte_flow_error *error)
2171 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2172 const struct rte_flow_item_mpls *mask = item->mask;
2173 struct mlx5_priv *priv = dev->data->dev_private;
2176 if (!priv->config.mpls_en)
2177 return rte_flow_error_set(error, ENOTSUP,
2178 RTE_FLOW_ERROR_TYPE_ITEM, item,
2179 "MPLS not supported or"
2180 " disabled in firmware"
2182 /* MPLS over IP, UDP, GRE is allowed */
2183 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2184 MLX5_FLOW_LAYER_OUTER_L4_UDP |
2185 MLX5_FLOW_LAYER_GRE)))
2186 return rte_flow_error_set(error, EINVAL,
2187 RTE_FLOW_ERROR_TYPE_ITEM, item,
2188 "protocol filtering not compatible"
2189 " with MPLS layer");
2190 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2191 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2192 !(item_flags & MLX5_FLOW_LAYER_GRE))
2193 return rte_flow_error_set(error, ENOTSUP,
2194 RTE_FLOW_ERROR_TYPE_ITEM, item,
2195 "multiple tunnel layers not"
2198 mask = &rte_flow_item_mpls_mask;
2199 ret = mlx5_flow_item_acceptable
2200 (item, (const uint8_t *)mask,
2201 (const uint8_t *)&rte_flow_item_mpls_mask,
2202 sizeof(struct rte_flow_item_mpls), error);
2207 return rte_flow_error_set(error, ENOTSUP,
2208 RTE_FLOW_ERROR_TYPE_ITEM, item,
2209 "MPLS is not supported by Verbs, please"
2215 * Validate NVGRE item.
2218 * Item specification.
2219 * @param[in] item_flags
2220 * Bit flags to mark detected items.
2221 * @param[in] target_protocol
2222 * The next protocol in the previous item.
2224 * Pointer to error structure.
2227 * 0 on success, a negative errno value otherwise and rte_errno is set.
2230 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2231 uint64_t item_flags,
2232 uint8_t target_protocol,
2233 struct rte_flow_error *error)
2235 const struct rte_flow_item_nvgre *mask = item->mask;
2238 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2239 return rte_flow_error_set(error, EINVAL,
2240 RTE_FLOW_ERROR_TYPE_ITEM, item,
2241 "protocol filtering not compatible"
2242 " with this GRE layer");
2243 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2244 return rte_flow_error_set(error, ENOTSUP,
2245 RTE_FLOW_ERROR_TYPE_ITEM, item,
2246 "multiple tunnel layers not"
2248 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2249 return rte_flow_error_set(error, ENOTSUP,
2250 RTE_FLOW_ERROR_TYPE_ITEM, item,
2251 "L3 Layer is missing");
2253 mask = &rte_flow_item_nvgre_mask;
2254 ret = mlx5_flow_item_acceptable
2255 (item, (const uint8_t *)mask,
2256 (const uint8_t *)&rte_flow_item_nvgre_mask,
2257 sizeof(struct rte_flow_item_nvgre), error);
2264 * Validate eCPRI item.
2267 * Item specification.
2268 * @param[in] item_flags
2269 * Bit-fields that holds the items detected until now.
2270 * @param[in] last_item
2271 * Previous validated item in the pattern items.
2272 * @param[in] ether_type
2273 * Type in the ethernet layer header (including dot1q).
2274 * @param[in] acc_mask
2275 * Acceptable mask, if NULL default internal default mask
2276 * will be used to check whether item fields are supported.
2278 * Pointer to error structure.
2281 * 0 on success, a negative errno value otherwise and rte_errno is set.
2284 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2285 uint64_t item_flags,
2287 uint16_t ether_type,
2288 const struct rte_flow_item_ecpri *acc_mask,
2289 struct rte_flow_error *error)
2291 const struct rte_flow_item_ecpri *mask = item->mask;
2292 const struct rte_flow_item_ecpri nic_mask = {
2296 RTE_BE32(((const struct rte_ecpri_common_hdr) {
2300 .dummy[0] = 0xFFFFFFFF,
2303 const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2304 MLX5_FLOW_LAYER_OUTER_VLAN);
2305 struct rte_flow_item_ecpri mask_lo;
2307 if ((last_item & outer_l2_vlan) && ether_type &&
2308 ether_type != RTE_ETHER_TYPE_ECPRI)
2309 return rte_flow_error_set(error, EINVAL,
2310 RTE_FLOW_ERROR_TYPE_ITEM, item,
2311 "eCPRI cannot follow L2/VLAN layer "
2312 "which ether type is not 0xAEFE.");
2313 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2314 return rte_flow_error_set(error, EINVAL,
2315 RTE_FLOW_ERROR_TYPE_ITEM, item,
2316 "eCPRI with tunnel is not supported "
2318 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2319 return rte_flow_error_set(error, ENOTSUP,
2320 RTE_FLOW_ERROR_TYPE_ITEM, item,
2321 "multiple L3 layers not supported");
2322 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2323 return rte_flow_error_set(error, EINVAL,
2324 RTE_FLOW_ERROR_TYPE_ITEM, item,
2325 "eCPRI cannot follow a TCP layer.");
2326 /* In specification, eCPRI could be over UDP layer. */
2327 else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2328 return rte_flow_error_set(error, EINVAL,
2329 RTE_FLOW_ERROR_TYPE_ITEM, item,
2330 "eCPRI over UDP layer is not yet "
2331 "supported right now.");
2332 /* Mask for type field in common header could be zero. */
2334 mask = &rte_flow_item_ecpri_mask;
2335 mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2336 /* Input mask is in big-endian format. */
2337 if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2338 return rte_flow_error_set(error, EINVAL,
2339 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2340 "partial mask is not supported "
2342 else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2343 return rte_flow_error_set(error, EINVAL,
2344 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2345 "message header mask must be after "
2347 return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2348 acc_mask ? (const uint8_t *)acc_mask
2349 : (const uint8_t *)&nic_mask,
2350 sizeof(struct rte_flow_item_ecpri),
2354 /* Allocate unique ID for the split Q/RSS subflows. */
2356 flow_qrss_get_id(struct rte_eth_dev *dev)
2358 struct mlx5_priv *priv = dev->data->dev_private;
2359 uint32_t qrss_id, ret;
2361 ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2364 MLX5_ASSERT(qrss_id);
2368 /* Free unique ID for the split Q/RSS subflows. */
2370 flow_qrss_free_id(struct rte_eth_dev *dev, uint32_t qrss_id)
2372 struct mlx5_priv *priv = dev->data->dev_private;
2375 mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2379 * Release resource related QUEUE/RSS action split.
2382 * Pointer to Ethernet device.
2384 * Flow to release id's from.
2387 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2388 struct rte_flow *flow)
2390 struct mlx5_priv *priv = dev->data->dev_private;
2391 uint32_t handle_idx;
2392 struct mlx5_flow_handle *dev_handle;
2394 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2395 handle_idx, dev_handle, next)
2396 if (dev_handle->split_flow_id)
2397 flow_qrss_free_id(dev, dev_handle->split_flow_id);
2401 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2402 const struct rte_flow_attr *attr __rte_unused,
2403 const struct rte_flow_item items[] __rte_unused,
2404 const struct rte_flow_action actions[] __rte_unused,
2405 bool external __rte_unused,
2406 int hairpin __rte_unused,
2407 struct rte_flow_error *error)
2409 return rte_flow_error_set(error, ENOTSUP,
2410 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2413 static struct mlx5_flow *
2414 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2415 const struct rte_flow_attr *attr __rte_unused,
2416 const struct rte_flow_item items[] __rte_unused,
2417 const struct rte_flow_action actions[] __rte_unused,
2418 struct rte_flow_error *error)
2420 rte_flow_error_set(error, ENOTSUP,
2421 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2426 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2427 struct mlx5_flow *dev_flow __rte_unused,
2428 const struct rte_flow_attr *attr __rte_unused,
2429 const struct rte_flow_item items[] __rte_unused,
2430 const struct rte_flow_action actions[] __rte_unused,
2431 struct rte_flow_error *error)
2433 return rte_flow_error_set(error, ENOTSUP,
2434 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2438 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2439 struct rte_flow *flow __rte_unused,
2440 struct rte_flow_error *error)
2442 return rte_flow_error_set(error, ENOTSUP,
2443 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2447 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2448 struct rte_flow *flow __rte_unused)
2453 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2454 struct rte_flow *flow __rte_unused)
2459 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2460 struct rte_flow *flow __rte_unused,
2461 const struct rte_flow_action *actions __rte_unused,
2462 void *data __rte_unused,
2463 struct rte_flow_error *error)
2465 return rte_flow_error_set(error, ENOTSUP,
2466 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2469 /* Void driver to protect from null pointer reference. */
2470 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2471 .validate = flow_null_validate,
2472 .prepare = flow_null_prepare,
2473 .translate = flow_null_translate,
2474 .apply = flow_null_apply,
2475 .remove = flow_null_remove,
2476 .destroy = flow_null_destroy,
2477 .query = flow_null_query,
2481 * Select flow driver type according to flow attributes and device
2485 * Pointer to the dev structure.
2487 * Pointer to the flow attributes.
2490 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2492 static enum mlx5_flow_drv_type
2493 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2495 struct mlx5_priv *priv = dev->data->dev_private;
2496 /* The OS can determine first a specific flow type (DV, VERBS) */
2497 enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
2499 if (type != MLX5_FLOW_TYPE_MAX)
2501 /* If no OS specific type - continue with DV/VERBS selection */
2502 if (attr->transfer && priv->config.dv_esw_en)
2503 type = MLX5_FLOW_TYPE_DV;
2504 if (!attr->transfer)
2505 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2506 MLX5_FLOW_TYPE_VERBS;
2510 #define flow_get_drv_ops(type) flow_drv_ops[type]
2513 * Flow driver validation API. This abstracts calling driver specific functions.
2514 * The type of flow driver is determined according to flow attributes.
2517 * Pointer to the dev structure.
2519 * Pointer to the flow attributes.
2521 * Pointer to the list of items.
2522 * @param[in] actions
2523 * Pointer to the list of actions.
2524 * @param[in] external
2525 * This flow rule is created by request external to PMD.
2526 * @param[in] hairpin
2527 * Number of hairpin TX actions, 0 means classic flow.
2529 * Pointer to the error structure.
2532 * 0 on success, a negative errno value otherwise and rte_errno is set.
2535 flow_drv_validate(struct rte_eth_dev *dev,
2536 const struct rte_flow_attr *attr,
2537 const struct rte_flow_item items[],
2538 const struct rte_flow_action actions[],
2539 bool external, int hairpin, struct rte_flow_error *error)
2541 const struct mlx5_flow_driver_ops *fops;
2542 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2544 fops = flow_get_drv_ops(type);
2545 return fops->validate(dev, attr, items, actions, external,
2550 * Flow driver preparation API. This abstracts calling driver specific
2551 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2552 * calculates the size of memory required for device flow, allocates the memory,
2553 * initializes the device flow and returns the pointer.
2556 * This function initializes device flow structure such as dv or verbs in
2557 * struct mlx5_flow. However, it is caller's responsibility to initialize the
2558 * rest. For example, adding returning device flow to flow->dev_flow list and
2559 * setting backward reference to the flow should be done out of this function.
2560 * layers field is not filled either.
2563 * Pointer to the dev structure.
2565 * Pointer to the flow attributes.
2567 * Pointer to the list of items.
2568 * @param[in] actions
2569 * Pointer to the list of actions.
2570 * @param[in] flow_idx
2571 * This memory pool index to the flow.
2573 * Pointer to the error structure.
2576 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
2578 static inline struct mlx5_flow *
2579 flow_drv_prepare(struct rte_eth_dev *dev,
2580 const struct rte_flow *flow,
2581 const struct rte_flow_attr *attr,
2582 const struct rte_flow_item items[],
2583 const struct rte_flow_action actions[],
2585 struct rte_flow_error *error)
2587 const struct mlx5_flow_driver_ops *fops;
2588 enum mlx5_flow_drv_type type = flow->drv_type;
2589 struct mlx5_flow *mlx5_flow = NULL;
2591 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2592 fops = flow_get_drv_ops(type);
2593 mlx5_flow = fops->prepare(dev, attr, items, actions, error);
2595 mlx5_flow->flow_idx = flow_idx;
2600 * Flow driver translation API. This abstracts calling driver specific
2601 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2602 * translates a generic flow into a driver flow. flow_drv_prepare() must
2606 * dev_flow->layers could be filled as a result of parsing during translation
2607 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2608 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2609 * flow->actions could be overwritten even though all the expanded dev_flows
2610 * have the same actions.
2613 * Pointer to the rte dev structure.
2614 * @param[in, out] dev_flow
2615 * Pointer to the mlx5 flow.
2617 * Pointer to the flow attributes.
2619 * Pointer to the list of items.
2620 * @param[in] actions
2621 * Pointer to the list of actions.
2623 * Pointer to the error structure.
2626 * 0 on success, a negative errno value otherwise and rte_errno is set.
2629 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2630 const struct rte_flow_attr *attr,
2631 const struct rte_flow_item items[],
2632 const struct rte_flow_action actions[],
2633 struct rte_flow_error *error)
2635 const struct mlx5_flow_driver_ops *fops;
2636 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2638 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2639 fops = flow_get_drv_ops(type);
2640 return fops->translate(dev, dev_flow, attr, items, actions, error);
2644 * Flow driver apply API. This abstracts calling driver specific functions.
2645 * Parent flow (rte_flow) should have driver type (drv_type). It applies
2646 * translated driver flows on to device. flow_drv_translate() must precede.
2649 * Pointer to Ethernet device structure.
2650 * @param[in, out] flow
2651 * Pointer to flow structure.
2653 * Pointer to error structure.
2656 * 0 on success, a negative errno value otherwise and rte_errno is set.
2659 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2660 struct rte_flow_error *error)
2662 const struct mlx5_flow_driver_ops *fops;
2663 enum mlx5_flow_drv_type type = flow->drv_type;
2665 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2666 fops = flow_get_drv_ops(type);
2667 return fops->apply(dev, flow, error);
2671 * Flow driver remove API. This abstracts calling driver specific functions.
2672 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2673 * on device. All the resources of the flow should be freed by calling
2674 * flow_drv_destroy().
2677 * Pointer to Ethernet device.
2678 * @param[in, out] flow
2679 * Pointer to flow structure.
2682 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2684 const struct mlx5_flow_driver_ops *fops;
2685 enum mlx5_flow_drv_type type = flow->drv_type;
2687 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2688 fops = flow_get_drv_ops(type);
2689 fops->remove(dev, flow);
2693 * Flow driver destroy API. This abstracts calling driver specific functions.
2694 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2695 * on device and releases resources of the flow.
2698 * Pointer to Ethernet device.
2699 * @param[in, out] flow
2700 * Pointer to flow structure.
2703 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2705 const struct mlx5_flow_driver_ops *fops;
2706 enum mlx5_flow_drv_type type = flow->drv_type;
2708 flow_mreg_split_qrss_release(dev, flow);
2709 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2710 fops = flow_get_drv_ops(type);
2711 fops->destroy(dev, flow);
2715 * Get RSS action from the action list.
2717 * @param[in] actions
2718 * Pointer to the list of actions.
2721 * Pointer to the RSS action if exist, else return NULL.
2723 static const struct rte_flow_action_rss*
2724 flow_get_rss_action(const struct rte_flow_action actions[])
2726 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2727 switch (actions->type) {
2728 case RTE_FLOW_ACTION_TYPE_RSS:
2729 return (const struct rte_flow_action_rss *)
2739 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2741 const struct rte_flow_item *item;
2742 unsigned int has_vlan = 0;
2744 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2745 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2751 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2752 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2753 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2754 MLX5_EXPANSION_ROOT_OUTER;
2758 * Get layer flags from the prefix flow.
2760 * Some flows may be split to several subflows, the prefix subflow gets the
2761 * match items and the suffix sub flow gets the actions.
2762 * Some actions need the user defined match item flags to get the detail for
2764 * This function helps the suffix flow to get the item layer flags from prefix
2767 * @param[in] dev_flow
2768 * Pointer the created preifx subflow.
2771 * The layers get from prefix subflow.
2773 static inline uint64_t
2774 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2776 uint64_t layers = 0;
2779 * Layers bits could be localization, but usually the compiler will
2780 * help to do the optimization work for source code.
2781 * If no decap actions, use the layers directly.
2783 if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
2784 return dev_flow->handle->layers;
2785 /* Convert L3 layers with decap action. */
2786 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2787 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2788 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2789 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2790 /* Convert L4 layers with decap action. */
2791 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2792 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2793 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2794 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2799 * Get metadata split action information.
2801 * @param[in] actions
2802 * Pointer to the list of actions.
2804 * Pointer to the return pointer.
2805 * @param[out] qrss_type
2806 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2807 * if no QUEUE/RSS is found.
2808 * @param[out] encap_idx
2809 * Pointer to the index of the encap action if exists, otherwise the last
2813 * Total number of actions.
2816 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2817 const struct rte_flow_action **qrss,
2820 const struct rte_flow_action_raw_encap *raw_encap;
2822 int raw_decap_idx = -1;
2825 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2826 switch (actions->type) {
2827 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2828 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2829 *encap_idx = actions_n;
2831 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2832 raw_decap_idx = actions_n;
2834 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2835 raw_encap = actions->conf;
2836 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2837 *encap_idx = raw_decap_idx != -1 ?
2838 raw_decap_idx : actions_n;
2840 case RTE_FLOW_ACTION_TYPE_QUEUE:
2841 case RTE_FLOW_ACTION_TYPE_RSS:
2849 if (*encap_idx == -1)
2850 *encap_idx = actions_n;
2851 /* Count RTE_FLOW_ACTION_TYPE_END. */
2852 return actions_n + 1;
2856 * Check meter action from the action list.
2858 * @param[in] actions
2859 * Pointer to the list of actions.
2861 * Pointer to the meter exist flag.
2864 * Total number of actions.
2867 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2873 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2874 switch (actions->type) {
2875 case RTE_FLOW_ACTION_TYPE_METER:
2883 /* Count RTE_FLOW_ACTION_TYPE_END. */
2884 return actions_n + 1;
2888 * Check if the flow should be splited due to hairpin.
2889 * The reason for the split is that in current HW we can't
2890 * support encap on Rx, so if a flow have encap we move it
2894 * Pointer to Ethernet device.
2896 * Flow rule attributes.
2897 * @param[in] actions
2898 * Associated actions (list terminated by the END action).
2901 * > 0 the number of actions and the flow should be split,
2902 * 0 when no split required.
2905 flow_check_hairpin_split(struct rte_eth_dev *dev,
2906 const struct rte_flow_attr *attr,
2907 const struct rte_flow_action actions[])
2909 int queue_action = 0;
2912 const struct rte_flow_action_queue *queue;
2913 const struct rte_flow_action_rss *rss;
2914 const struct rte_flow_action_raw_encap *raw_encap;
2918 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2919 switch (actions->type) {
2920 case RTE_FLOW_ACTION_TYPE_QUEUE:
2921 queue = actions->conf;
2924 if (mlx5_rxq_get_type(dev, queue->index) !=
2925 MLX5_RXQ_TYPE_HAIRPIN)
2930 case RTE_FLOW_ACTION_TYPE_RSS:
2931 rss = actions->conf;
2932 if (rss == NULL || rss->queue_num == 0)
2934 if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2935 MLX5_RXQ_TYPE_HAIRPIN)
2940 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2941 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2945 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2946 raw_encap = actions->conf;
2947 if (raw_encap->size >
2948 (sizeof(struct rte_flow_item_eth) +
2949 sizeof(struct rte_flow_item_ipv4)))
2958 if (encap == 1 && queue_action)
2963 /* Declare flow create/destroy prototype in advance. */
2965 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
2966 const struct rte_flow_attr *attr,
2967 const struct rte_flow_item items[],
2968 const struct rte_flow_action actions[],
2969 bool external, struct rte_flow_error *error);
2972 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
2976 * Add a flow of copying flow metadata registers in RX_CP_TBL.
2978 * As mark_id is unique, if there's already a registered flow for the mark_id,
2979 * return by increasing the reference counter of the resource. Otherwise, create
2980 * the resource (mcp_res) and flow.
2983 * - If ingress port is ANY and reg_c[1] is mark_id,
2984 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2986 * For default flow (zero mark_id), flow is like,
2987 * - If ingress port is ANY,
2988 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
2991 * Pointer to Ethernet device.
2993 * ID of MARK action, zero means default flow for META.
2995 * Perform verbose error reporting if not NULL.
2998 * Associated resource on success, NULL otherwise and rte_errno is set.
3000 static struct mlx5_flow_mreg_copy_resource *
3001 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3002 struct rte_flow_error *error)
3004 struct mlx5_priv *priv = dev->data->dev_private;
3005 struct rte_flow_attr attr = {
3006 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3009 struct mlx5_rte_flow_item_tag tag_spec = {
3012 struct rte_flow_item items[] = {
3013 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3015 struct rte_flow_action_mark ftag = {
3018 struct mlx5_flow_action_copy_mreg cp_mreg = {
3022 struct rte_flow_action_jump jump = {
3023 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3025 struct rte_flow_action actions[] = {
3026 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3028 struct mlx5_flow_mreg_copy_resource *mcp_res;
3032 /* Fill the register fileds in the flow. */
3033 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3037 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3041 /* Check if already registered. */
3042 MLX5_ASSERT(priv->mreg_cp_tbl);
3043 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
3045 /* For non-default rule. */
3046 if (mark_id != MLX5_DEFAULT_COPY_ID)
3048 MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
3049 mcp_res->refcnt == 1);
3052 /* Provide the full width of FLAG specific value. */
3053 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3054 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3055 /* Build a new flow. */
3056 if (mark_id != MLX5_DEFAULT_COPY_ID) {
3057 items[0] = (struct rte_flow_item){
3058 .type = (enum rte_flow_item_type)
3059 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3062 items[1] = (struct rte_flow_item){
3063 .type = RTE_FLOW_ITEM_TYPE_END,
3065 actions[0] = (struct rte_flow_action){
3066 .type = (enum rte_flow_action_type)
3067 MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3070 actions[1] = (struct rte_flow_action){
3071 .type = (enum rte_flow_action_type)
3072 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3075 actions[2] = (struct rte_flow_action){
3076 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3079 actions[3] = (struct rte_flow_action){
3080 .type = RTE_FLOW_ACTION_TYPE_END,
3083 /* Default rule, wildcard match. */
3084 attr.priority = MLX5_FLOW_PRIO_RSVD;
3085 items[0] = (struct rte_flow_item){
3086 .type = RTE_FLOW_ITEM_TYPE_END,
3088 actions[0] = (struct rte_flow_action){
3089 .type = (enum rte_flow_action_type)
3090 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3093 actions[1] = (struct rte_flow_action){
3094 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3097 actions[2] = (struct rte_flow_action){
3098 .type = RTE_FLOW_ACTION_TYPE_END,
3101 /* Build a new entry. */
3102 mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3109 * The copy Flows are not included in any list. There
3110 * ones are referenced from other Flows and can not
3111 * be applied, removed, deleted in ardbitrary order
3112 * by list traversing.
3114 mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3115 actions, false, error);
3116 if (!mcp_res->rix_flow)
3119 mcp_res->hlist_ent.key = mark_id;
3120 ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3121 &mcp_res->hlist_ent);
3127 if (mcp_res->rix_flow)
3128 flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3129 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3134 * Release flow in RX_CP_TBL.
3137 * Pointer to Ethernet device.
3139 * Parent flow for wich copying is provided.
3142 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3143 struct rte_flow *flow)
3145 struct mlx5_flow_mreg_copy_resource *mcp_res;
3146 struct mlx5_priv *priv = dev->data->dev_private;
3148 if (!flow->rix_mreg_copy)
3150 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3151 flow->rix_mreg_copy);
3152 if (!mcp_res || !priv->mreg_cp_tbl)
3154 if (flow->copy_applied) {
3155 MLX5_ASSERT(mcp_res->appcnt);
3156 flow->copy_applied = 0;
3158 if (!mcp_res->appcnt) {
3159 struct rte_flow *mcp_flow = mlx5_ipool_get
3160 (priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3164 flow_drv_remove(dev, mcp_flow);
3168 * We do not check availability of metadata registers here,
3169 * because copy resources are not allocated in this case.
3171 if (--mcp_res->refcnt)
3173 MLX5_ASSERT(mcp_res->rix_flow);
3174 flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3175 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3176 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3177 flow->rix_mreg_copy = 0;
3181 * Start flow in RX_CP_TBL.
3184 * Pointer to Ethernet device.
3186 * Parent flow for wich copying is provided.
3189 * 0 on success, a negative errno value otherwise and rte_errno is set.
3192 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3193 struct rte_flow *flow)
3195 struct mlx5_flow_mreg_copy_resource *mcp_res;
3196 struct mlx5_priv *priv = dev->data->dev_private;
3199 if (!flow->rix_mreg_copy || flow->copy_applied)
3201 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3202 flow->rix_mreg_copy);
3205 if (!mcp_res->appcnt) {
3206 struct rte_flow *mcp_flow = mlx5_ipool_get
3207 (priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3211 ret = flow_drv_apply(dev, mcp_flow, NULL);
3217 flow->copy_applied = 1;
3222 * Stop flow in RX_CP_TBL.
3225 * Pointer to Ethernet device.
3227 * Parent flow for wich copying is provided.
3230 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3231 struct rte_flow *flow)
3233 struct mlx5_flow_mreg_copy_resource *mcp_res;
3234 struct mlx5_priv *priv = dev->data->dev_private;
3236 if (!flow->rix_mreg_copy || !flow->copy_applied)
3238 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3239 flow->rix_mreg_copy);
3242 MLX5_ASSERT(mcp_res->appcnt);
3244 flow->copy_applied = 0;
3245 if (!mcp_res->appcnt) {
3246 struct rte_flow *mcp_flow = mlx5_ipool_get
3247 (priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3251 flow_drv_remove(dev, mcp_flow);
3256 * Remove the default copy action from RX_CP_TBL.
3259 * Pointer to Ethernet device.
3262 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3264 struct mlx5_flow_mreg_copy_resource *mcp_res;
3265 struct mlx5_priv *priv = dev->data->dev_private;
3267 /* Check if default flow is registered. */
3268 if (!priv->mreg_cp_tbl)
3270 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3271 MLX5_DEFAULT_COPY_ID);
3274 MLX5_ASSERT(mcp_res->rix_flow);
3275 flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3276 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3277 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3281 * Add the default copy action in in RX_CP_TBL.
3284 * Pointer to Ethernet device.
3286 * Perform verbose error reporting if not NULL.
3289 * 0 for success, negative value otherwise and rte_errno is set.
3292 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3293 struct rte_flow_error *error)
3295 struct mlx5_priv *priv = dev->data->dev_private;
3296 struct mlx5_flow_mreg_copy_resource *mcp_res;
3298 /* Check whether extensive metadata feature is engaged. */
3299 if (!priv->config.dv_flow_en ||
3300 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3301 !mlx5_flow_ext_mreg_supported(dev) ||
3302 !priv->sh->dv_regc0_mask)
3304 mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3311 * Add a flow of copying flow metadata registers in RX_CP_TBL.
3313 * All the flow having Q/RSS action should be split by
3314 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3315 * performs the following,
3316 * - CQE->flow_tag := reg_c[1] (MARK)
3317 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3318 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3319 * but there should be a flow per each MARK ID set by MARK action.
3321 * For the aforementioned reason, if there's a MARK action in flow's action
3322 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3323 * the MARK ID to CQE's flow_tag like,
3324 * - If reg_c[1] is mark_id,
3325 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3327 * For SET_META action which stores value in reg_c[0], as the destination is
3328 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3329 * MARK ID means the default flow. The default flow looks like,
3330 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3333 * Pointer to Ethernet device.
3335 * Pointer to flow structure.
3336 * @param[in] actions
3337 * Pointer to the list of actions.
3339 * Perform verbose error reporting if not NULL.
3342 * 0 on success, negative value otherwise and rte_errno is set.
3345 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3346 struct rte_flow *flow,
3347 const struct rte_flow_action *actions,
3348 struct rte_flow_error *error)
3350 struct mlx5_priv *priv = dev->data->dev_private;
3351 struct mlx5_dev_config *config = &priv->config;
3352 struct mlx5_flow_mreg_copy_resource *mcp_res;
3353 const struct rte_flow_action_mark *mark;
3355 /* Check whether extensive metadata feature is engaged. */
3356 if (!config->dv_flow_en ||
3357 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3358 !mlx5_flow_ext_mreg_supported(dev) ||
3359 !priv->sh->dv_regc0_mask)
3361 /* Find MARK action. */
3362 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3363 switch (actions->type) {
3364 case RTE_FLOW_ACTION_TYPE_FLAG:
3365 mcp_res = flow_mreg_add_copy_action
3366 (dev, MLX5_FLOW_MARK_DEFAULT, error);
3369 flow->rix_mreg_copy = mcp_res->idx;
3370 if (dev->data->dev_started) {
3372 flow->copy_applied = 1;
3375 case RTE_FLOW_ACTION_TYPE_MARK:
3376 mark = (const struct rte_flow_action_mark *)
3379 flow_mreg_add_copy_action(dev, mark->id, error);
3382 flow->rix_mreg_copy = mcp_res->idx;
3383 if (dev->data->dev_started) {
3385 flow->copy_applied = 1;
3395 #define MLX5_MAX_SPLIT_ACTIONS 24
3396 #define MLX5_MAX_SPLIT_ITEMS 24
3399 * Split the hairpin flow.
3400 * Since HW can't support encap on Rx we move the encap to Tx.
3401 * If the count action is after the encap then we also
3402 * move the count action. in this case the count will also measure
3406 * Pointer to Ethernet device.
3407 * @param[in] actions
3408 * Associated actions (list terminated by the END action).
3409 * @param[out] actions_rx
3411 * @param[out] actions_tx
3413 * @param[out] pattern_tx
3414 * The pattern items for the Tx flow.
3415 * @param[out] flow_id
3416 * The flow ID connected to this flow.
3422 flow_hairpin_split(struct rte_eth_dev *dev,
3423 const struct rte_flow_action actions[],
3424 struct rte_flow_action actions_rx[],
3425 struct rte_flow_action actions_tx[],
3426 struct rte_flow_item pattern_tx[],
3429 struct mlx5_priv *priv = dev->data->dev_private;
3430 const struct rte_flow_action_raw_encap *raw_encap;
3431 const struct rte_flow_action_raw_decap *raw_decap;
3432 struct mlx5_rte_flow_action_set_tag *set_tag;
3433 struct rte_flow_action *tag_action;
3434 struct mlx5_rte_flow_item_tag *tag_item;
3435 struct rte_flow_item *item;
3439 mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3440 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3441 switch (actions->type) {
3442 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3443 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3444 rte_memcpy(actions_tx, actions,
3445 sizeof(struct rte_flow_action));
3448 case RTE_FLOW_ACTION_TYPE_COUNT:
3450 rte_memcpy(actions_tx, actions,
3451 sizeof(struct rte_flow_action));
3454 rte_memcpy(actions_rx, actions,
3455 sizeof(struct rte_flow_action));
3459 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3460 raw_encap = actions->conf;
3461 if (raw_encap->size >
3462 (sizeof(struct rte_flow_item_eth) +
3463 sizeof(struct rte_flow_item_ipv4))) {
3464 memcpy(actions_tx, actions,
3465 sizeof(struct rte_flow_action));
3469 rte_memcpy(actions_rx, actions,
3470 sizeof(struct rte_flow_action));
3474 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3475 raw_decap = actions->conf;
3476 if (raw_decap->size <
3477 (sizeof(struct rte_flow_item_eth) +
3478 sizeof(struct rte_flow_item_ipv4))) {
3479 memcpy(actions_tx, actions,
3480 sizeof(struct rte_flow_action));
3483 rte_memcpy(actions_rx, actions,
3484 sizeof(struct rte_flow_action));
3489 rte_memcpy(actions_rx, actions,
3490 sizeof(struct rte_flow_action));
3495 /* Add set meta action and end action for the Rx flow. */
3496 tag_action = actions_rx;
3497 tag_action->type = (enum rte_flow_action_type)
3498 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3500 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3502 set_tag = (void *)actions_rx;
3503 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3504 MLX5_ASSERT(set_tag->id > REG_NONE);
3505 set_tag->data = *flow_id;
3506 tag_action->conf = set_tag;
3507 /* Create Tx item list. */
3508 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3509 addr = (void *)&pattern_tx[2];
3511 item->type = (enum rte_flow_item_type)
3512 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3513 tag_item = (void *)addr;
3514 tag_item->data = *flow_id;
3515 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3516 MLX5_ASSERT(set_tag->id > REG_NONE);
3517 item->spec = tag_item;
3518 addr += sizeof(struct mlx5_rte_flow_item_tag);
3519 tag_item = (void *)addr;
3520 tag_item->data = UINT32_MAX;
3521 tag_item->id = UINT16_MAX;
3522 item->mask = tag_item;
3523 addr += sizeof(struct mlx5_rte_flow_item_tag);
3526 item->type = RTE_FLOW_ITEM_TYPE_END;
3531 * The last stage of splitting chain, just creates the subflow
3532 * without any modification.
3535 * Pointer to Ethernet device.
3537 * Parent flow structure pointer.
3538 * @param[in, out] sub_flow
3539 * Pointer to return the created subflow, may be NULL.
3540 * @param[in] prefix_layers
3541 * Prefix subflow layers, may be 0.
3543 * Flow rule attributes.
3545 * Pattern specification (list terminated by the END pattern item).
3546 * @param[in] actions
3547 * Associated actions (list terminated by the END action).
3548 * @param[in] external
3549 * This flow rule is created by request external to PMD.
3550 * @param[in] flow_idx
3551 * This memory pool index to the flow.
3553 * Perform verbose error reporting if not NULL.
3555 * 0 on success, negative value otherwise
3558 flow_create_split_inner(struct rte_eth_dev *dev,
3559 struct rte_flow *flow,
3560 struct mlx5_flow **sub_flow,
3561 uint64_t prefix_layers,
3562 const struct rte_flow_attr *attr,
3563 const struct rte_flow_item items[],
3564 const struct rte_flow_action actions[],
3565 bool external, uint32_t flow_idx,
3566 struct rte_flow_error *error)
3568 struct mlx5_flow *dev_flow;
3570 dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
3574 dev_flow->flow = flow;
3575 dev_flow->external = external;
3576 /* Subflow object was created, we must include one in the list. */
3577 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3578 dev_flow->handle, next);
3580 * If dev_flow is as one of the suffix flow, some actions in suffix
3581 * flow may need some user defined item layer flags.
3584 dev_flow->handle->layers = prefix_layers;
3586 *sub_flow = dev_flow;
3587 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3591 * Split the meter flow.
3593 * As meter flow will split to three sub flow, other than meter
3594 * action, the other actions make sense to only meter accepts
3595 * the packet. If it need to be dropped, no other additional
3596 * actions should be take.
3598 * One kind of special action which decapsulates the L3 tunnel
3599 * header will be in the prefix sub flow, as not to take the
3600 * L3 tunnel header into account.
3603 * Pointer to Ethernet device.
3605 * Pattern specification (list terminated by the END pattern item).
3606 * @param[out] sfx_items
3607 * Suffix flow match items (list terminated by the END pattern item).
3608 * @param[in] actions
3609 * Associated actions (list terminated by the END action).
3610 * @param[out] actions_sfx
3611 * Suffix flow actions.
3612 * @param[out] actions_pre
3613 * Prefix flow actions.
3614 * @param[out] pattern_sfx
3615 * The pattern items for the suffix flow.
3616 * @param[out] tag_sfx
3617 * Pointer to suffix flow tag.
3623 flow_meter_split_prep(struct rte_eth_dev *dev,
3624 const struct rte_flow_item items[],
3625 struct rte_flow_item sfx_items[],
3626 const struct rte_flow_action actions[],
3627 struct rte_flow_action actions_sfx[],
3628 struct rte_flow_action actions_pre[])
3630 struct rte_flow_action *tag_action = NULL;
3631 struct rte_flow_item *tag_item;
3632 struct mlx5_rte_flow_action_set_tag *set_tag;
3633 struct rte_flow_error error;
3634 const struct rte_flow_action_raw_encap *raw_encap;
3635 const struct rte_flow_action_raw_decap *raw_decap;
3636 struct mlx5_rte_flow_item_tag *tag_spec;
3637 struct mlx5_rte_flow_item_tag *tag_mask;
3639 bool copy_vlan = false;
3641 /* Prepare the actions for prefix and suffix flow. */
3642 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3643 struct rte_flow_action **action_cur = NULL;
3645 switch (actions->type) {
3646 case RTE_FLOW_ACTION_TYPE_METER:
3647 /* Add the extra tag action first. */
3648 tag_action = actions_pre;
3649 tag_action->type = (enum rte_flow_action_type)
3650 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3652 action_cur = &actions_pre;
3654 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3655 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3656 action_cur = &actions_pre;
3658 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3659 raw_encap = actions->conf;
3660 if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3661 action_cur = &actions_pre;
3663 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3664 raw_decap = actions->conf;
3665 if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3666 action_cur = &actions_pre;
3668 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3669 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3676 action_cur = &actions_sfx;
3677 memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3680 /* Add end action to the actions. */
3681 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3682 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3685 set_tag = (void *)actions_pre;
3686 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3688 * Get the id from the qrss_pool to make qrss share the id with meter.
3690 tag_id = flow_qrss_get_id(dev);
3691 set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3693 tag_action->conf = set_tag;
3694 /* Prepare the suffix subflow items. */
3695 tag_item = sfx_items++;
3696 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3697 int item_type = items->type;
3699 switch (item_type) {
3700 case RTE_FLOW_ITEM_TYPE_PORT_ID:
3701 memcpy(sfx_items, items, sizeof(*sfx_items));
3704 case RTE_FLOW_ITEM_TYPE_VLAN:
3706 memcpy(sfx_items, items, sizeof(*sfx_items));
3708 * Convert to internal match item, it is used
3709 * for vlan push and set vid.
3711 sfx_items->type = (enum rte_flow_item_type)
3712 MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3720 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3722 tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3723 tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3724 tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3725 tag_mask = tag_spec + 1;
3726 tag_mask->data = 0xffffff00;
3727 tag_item->type = (enum rte_flow_item_type)
3728 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3729 tag_item->spec = tag_spec;
3730 tag_item->last = NULL;
3731 tag_item->mask = tag_mask;
3736 * Split action list having QUEUE/RSS for metadata register copy.
3738 * Once Q/RSS action is detected in user's action list, the flow action
3739 * should be split in order to copy metadata registers, which will happen in
3741 * - CQE->flow_tag := reg_c[1] (MARK)
3742 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3743 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3744 * This is because the last action of each flow must be a terminal action
3745 * (QUEUE, RSS or DROP).
3747 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3748 * stored and kept in the mlx5_flow structure per each sub_flow.
3750 * The Q/RSS action is replaced with,
3751 * - SET_TAG, setting the allocated flow ID to reg_c[2].
3752 * And the following JUMP action is added at the end,
3753 * - JUMP, to RX_CP_TBL.
3755 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3756 * flow_create_split_metadata() routine. The flow will look like,
3757 * - If flow ID matches (reg_c[2]), perform Q/RSS.
3760 * Pointer to Ethernet device.
3761 * @param[out] split_actions
3762 * Pointer to store split actions to jump to CP_TBL.
3763 * @param[in] actions
3764 * Pointer to the list of original flow actions.
3766 * Pointer to the Q/RSS action.
3767 * @param[in] actions_n
3768 * Number of original actions.
3770 * Perform verbose error reporting if not NULL.
3773 * non-zero unique flow_id on success, otherwise 0 and
3774 * error/rte_error are set.
3777 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3778 struct rte_flow_action *split_actions,
3779 const struct rte_flow_action *actions,
3780 const struct rte_flow_action *qrss,
3781 int actions_n, struct rte_flow_error *error)
3783 struct mlx5_rte_flow_action_set_tag *set_tag;
3784 struct rte_flow_action_jump *jump;
3785 const int qrss_idx = qrss - actions;
3786 uint32_t flow_id = 0;
3790 * Given actions will be split
3791 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3792 * - Add jump to mreg CP_TBL.
3793 * As a result, there will be one more action.
3796 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3797 set_tag = (void *)(split_actions + actions_n);
3799 * If tag action is not set to void(it means we are not the meter
3800 * suffix flow), add the tag action. Since meter suffix flow already
3801 * has the tag added.
3803 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3805 * Allocate the new subflow ID. This one is unique within
3806 * device and not shared with representors. Otherwise,
3807 * we would have to resolve multi-thread access synch
3808 * issue. Each flow on the shared device is appended
3809 * with source vport identifier, so the resulting
3810 * flows will be unique in the shared (by master and
3811 * representors) domain even if they have coinciding
3814 flow_id = flow_qrss_get_id(dev);
3816 return rte_flow_error_set(error, ENOMEM,
3817 RTE_FLOW_ERROR_TYPE_ACTION,
3818 NULL, "can't allocate id "
3819 "for split Q/RSS subflow");
3820 /* Internal SET_TAG action to set flow ID. */
3821 *set_tag = (struct mlx5_rte_flow_action_set_tag){
3824 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3828 /* Construct new actions array. */
3829 /* Replace QUEUE/RSS action. */
3830 split_actions[qrss_idx] = (struct rte_flow_action){
3831 .type = (enum rte_flow_action_type)
3832 MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3836 /* JUMP action to jump to mreg copy table (CP_TBL). */
3837 jump = (void *)(set_tag + 1);
3838 *jump = (struct rte_flow_action_jump){
3839 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3841 split_actions[actions_n - 2] = (struct rte_flow_action){
3842 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3845 split_actions[actions_n - 1] = (struct rte_flow_action){
3846 .type = RTE_FLOW_ACTION_TYPE_END,
3852 * Extend the given action list for Tx metadata copy.
3854 * Copy the given action list to the ext_actions and add flow metadata register
3855 * copy action in order to copy reg_a set by WQE to reg_c[0].
3857 * @param[out] ext_actions
3858 * Pointer to the extended action list.
3859 * @param[in] actions
3860 * Pointer to the list of actions.
3861 * @param[in] actions_n
3862 * Number of actions in the list.
3864 * Perform verbose error reporting if not NULL.
3865 * @param[in] encap_idx
3866 * The encap action inndex.
3869 * 0 on success, negative value otherwise
3872 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3873 struct rte_flow_action *ext_actions,
3874 const struct rte_flow_action *actions,
3875 int actions_n, struct rte_flow_error *error,
3878 struct mlx5_flow_action_copy_mreg *cp_mreg =
3879 (struct mlx5_flow_action_copy_mreg *)
3880 (ext_actions + actions_n + 1);
3883 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3887 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3892 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3893 if (encap_idx == actions_n - 1) {
3894 ext_actions[actions_n - 1] = (struct rte_flow_action){
3895 .type = (enum rte_flow_action_type)
3896 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3899 ext_actions[actions_n] = (struct rte_flow_action){
3900 .type = RTE_FLOW_ACTION_TYPE_END,
3903 ext_actions[encap_idx] = (struct rte_flow_action){
3904 .type = (enum rte_flow_action_type)
3905 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3908 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3909 sizeof(*ext_actions) * (actions_n - encap_idx));
3915 * The splitting for metadata feature.
3917 * - Q/RSS action on NIC Rx should be split in order to pass by
3918 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3919 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3921 * - All the actions on NIC Tx should have a mreg copy action to
3922 * copy reg_a from WQE to reg_c[0].
3925 * Pointer to Ethernet device.
3927 * Parent flow structure pointer.
3928 * @param[in] prefix_layers
3929 * Prefix flow layer flags.
3931 * Flow rule attributes.
3933 * Pattern specification (list terminated by the END pattern item).
3934 * @param[in] actions
3935 * Associated actions (list terminated by the END action).
3936 * @param[in] external
3937 * This flow rule is created by request external to PMD.
3938 * @param[in] flow_idx
3939 * This memory pool index to the flow.
3941 * Perform verbose error reporting if not NULL.
3943 * 0 on success, negative value otherwise
3946 flow_create_split_metadata(struct rte_eth_dev *dev,
3947 struct rte_flow *flow,
3948 uint64_t prefix_layers,
3949 const struct rte_flow_attr *attr,
3950 const struct rte_flow_item items[],
3951 const struct rte_flow_action actions[],
3952 bool external, uint32_t flow_idx,
3953 struct rte_flow_error *error)
3955 struct mlx5_priv *priv = dev->data->dev_private;
3956 struct mlx5_dev_config *config = &priv->config;
3957 const struct rte_flow_action *qrss = NULL;
3958 struct rte_flow_action *ext_actions = NULL;
3959 struct mlx5_flow *dev_flow = NULL;
3960 uint32_t qrss_id = 0;
3967 /* Check whether extensive metadata feature is engaged. */
3968 if (!config->dv_flow_en ||
3969 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3970 !mlx5_flow_ext_mreg_supported(dev))
3971 return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3972 attr, items, actions, external,
3974 actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3977 /* Exclude hairpin flows from splitting. */
3978 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3979 const struct rte_flow_action_queue *queue;
3982 if (mlx5_rxq_get_type(dev, queue->index) ==
3983 MLX5_RXQ_TYPE_HAIRPIN)
3985 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3986 const struct rte_flow_action_rss *rss;
3989 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3990 MLX5_RXQ_TYPE_HAIRPIN)
3995 /* Check if it is in meter suffix table. */
3996 mtr_sfx = attr->group == (attr->transfer ?
3997 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3998 MLX5_FLOW_TABLE_LEVEL_SUFFIX);
4000 * Q/RSS action on NIC Rx should be split in order to pass by
4001 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4002 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4004 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4005 sizeof(struct rte_flow_action_set_tag) +
4006 sizeof(struct rte_flow_action_jump);
4007 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4010 return rte_flow_error_set(error, ENOMEM,
4011 RTE_FLOW_ERROR_TYPE_ACTION,
4012 NULL, "no memory to split "
4015 * If we are the suffix flow of meter, tag already exist.
4016 * Set the tag action to void.
4019 ext_actions[qrss - actions].type =
4020 RTE_FLOW_ACTION_TYPE_VOID;
4022 ext_actions[qrss - actions].type =
4023 (enum rte_flow_action_type)
4024 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4026 * Create the new actions list with removed Q/RSS action
4027 * and appended set tag and jump to register copy table
4028 * (RX_CP_TBL). We should preallocate unique tag ID here
4029 * in advance, because it is needed for set tag action.
4031 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
4032 qrss, actions_n, error);
4033 if (!mtr_sfx && !qrss_id) {
4037 } else if (attr->egress && !attr->transfer) {
4039 * All the actions on NIC Tx should have a metadata register
4040 * copy action to copy reg_a from WQE to reg_c[meta]
4042 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4043 sizeof(struct mlx5_flow_action_copy_mreg);
4044 ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4047 return rte_flow_error_set(error, ENOMEM,
4048 RTE_FLOW_ERROR_TYPE_ACTION,
4049 NULL, "no memory to split "
4051 /* Create the action list appended with copy register. */
4052 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
4053 actions_n, error, encap_idx);
4057 /* Add the unmodified original or prefix subflow. */
4058 ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
4059 items, ext_actions ? ext_actions :
4060 actions, external, flow_idx, error);
4063 MLX5_ASSERT(dev_flow);
4065 const struct rte_flow_attr q_attr = {
4066 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4069 /* Internal PMD action to set register. */
4070 struct mlx5_rte_flow_item_tag q_tag_spec = {
4074 struct rte_flow_item q_items[] = {
4076 .type = (enum rte_flow_item_type)
4077 MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4078 .spec = &q_tag_spec,
4083 .type = RTE_FLOW_ITEM_TYPE_END,
4086 struct rte_flow_action q_actions[] = {
4092 .type = RTE_FLOW_ACTION_TYPE_END,
4095 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4098 * Configure the tag item only if there is no meter subflow.
4099 * Since tag is already marked in the meter suffix subflow
4100 * we can just use the meter suffix items as is.
4103 /* Not meter subflow. */
4104 MLX5_ASSERT(!mtr_sfx);
4106 * Put unique id in prefix flow due to it is destroyed
4107 * after suffix flow and id will be freed after there
4108 * is no actual flows with this id and identifier
4109 * reallocation becomes possible (for example, for
4110 * other flows in other threads).
4112 dev_flow->handle->split_flow_id = qrss_id;
4113 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4117 q_tag_spec.id = ret;
4120 /* Add suffix subflow to execute Q/RSS. */
4121 ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4122 &q_attr, mtr_sfx ? items :
4124 external, flow_idx, error);
4127 /* qrss ID should be freed if failed. */
4129 MLX5_ASSERT(dev_flow);
4134 * We do not destroy the partially created sub_flows in case of error.
4135 * These ones are included into parent flow list and will be destroyed
4136 * by flow_drv_destroy.
4138 flow_qrss_free_id(dev, qrss_id);
4139 mlx5_free(ext_actions);
4144 * The splitting for meter feature.
4146 * - The meter flow will be split to two flows as prefix and
4147 * suffix flow. The packets make sense only it pass the prefix
4150 * - Reg_C_5 is used for the packet to match betweend prefix and
4154 * Pointer to Ethernet device.
4156 * Parent flow structure pointer.
4158 * Flow rule attributes.
4160 * Pattern specification (list terminated by the END pattern item).
4161 * @param[in] actions
4162 * Associated actions (list terminated by the END action).
4163 * @param[in] external
4164 * This flow rule is created by request external to PMD.
4165 * @param[in] flow_idx
4166 * This memory pool index to the flow.
4168 * Perform verbose error reporting if not NULL.
4170 * 0 on success, negative value otherwise
4173 flow_create_split_meter(struct rte_eth_dev *dev,
4174 struct rte_flow *flow,
4175 const struct rte_flow_attr *attr,
4176 const struct rte_flow_item items[],
4177 const struct rte_flow_action actions[],
4178 bool external, uint32_t flow_idx,
4179 struct rte_flow_error *error)
4181 struct mlx5_priv *priv = dev->data->dev_private;
4182 struct rte_flow_action *sfx_actions = NULL;
4183 struct rte_flow_action *pre_actions = NULL;
4184 struct rte_flow_item *sfx_items = NULL;
4185 struct mlx5_flow *dev_flow = NULL;
4186 struct rte_flow_attr sfx_attr = *attr;
4188 uint32_t mtr_tag_id = 0;
4195 actions_n = flow_check_meter_action(actions, &mtr);
4197 /* The five prefix actions: meter, decap, encap, tag, end. */
4198 act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4199 sizeof(struct mlx5_rte_flow_action_set_tag);
4200 /* tag, vlan, port id, end. */
4201 #define METER_SUFFIX_ITEM 4
4202 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4203 sizeof(struct mlx5_rte_flow_item_tag) * 2;
4204 sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
4207 return rte_flow_error_set(error, ENOMEM,
4208 RTE_FLOW_ERROR_TYPE_ACTION,
4209 NULL, "no memory to split "
4211 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4213 pre_actions = sfx_actions + actions_n;
4214 mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4215 actions, sfx_actions,
4221 /* Add the prefix subflow. */
4222 ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4223 items, pre_actions, external,
4229 dev_flow->handle->split_flow_id = mtr_tag_id;
4230 /* Setting the sfx group atrr. */
4231 sfx_attr.group = sfx_attr.transfer ?
4232 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4233 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4235 /* Add the prefix subflow. */
4236 ret = flow_create_split_metadata(dev, flow, dev_flow ?
4237 flow_get_prefix_layer_flags(dev_flow) :
4239 sfx_items ? sfx_items : items,
4240 sfx_actions ? sfx_actions : actions,
4241 external, flow_idx, error);
4244 mlx5_free(sfx_actions);
4249 * Split the flow to subflow set. The splitters might be linked
4250 * in the chain, like this:
4251 * flow_create_split_outer() calls:
4252 * flow_create_split_meter() calls:
4253 * flow_create_split_metadata(meter_subflow_0) calls:
4254 * flow_create_split_inner(metadata_subflow_0)
4255 * flow_create_split_inner(metadata_subflow_1)
4256 * flow_create_split_inner(metadata_subflow_2)
4257 * flow_create_split_metadata(meter_subflow_1) calls:
4258 * flow_create_split_inner(metadata_subflow_0)
4259 * flow_create_split_inner(metadata_subflow_1)
4260 * flow_create_split_inner(metadata_subflow_2)
4262 * This provide flexible way to add new levels of flow splitting.
4263 * The all of successfully created subflows are included to the
4264 * parent flow dev_flow list.
4267 * Pointer to Ethernet device.
4269 * Parent flow structure pointer.
4271 * Flow rule attributes.
4273 * Pattern specification (list terminated by the END pattern item).
4274 * @param[in] actions
4275 * Associated actions (list terminated by the END action).
4276 * @param[in] external
4277 * This flow rule is created by request external to PMD.
4278 * @param[in] flow_idx
4279 * This memory pool index to the flow.
4281 * Perform verbose error reporting if not NULL.
4283 * 0 on success, negative value otherwise
4286 flow_create_split_outer(struct rte_eth_dev *dev,
4287 struct rte_flow *flow,
4288 const struct rte_flow_attr *attr,
4289 const struct rte_flow_item items[],
4290 const struct rte_flow_action actions[],
4291 bool external, uint32_t flow_idx,
4292 struct rte_flow_error *error)
4296 ret = flow_create_split_meter(dev, flow, attr, items,
4297 actions, external, flow_idx, error);
4298 MLX5_ASSERT(ret <= 0);
4303 * Create a flow and add it to @p list.
4306 * Pointer to Ethernet device.
4308 * Pointer to a TAILQ flow list. If this parameter NULL,
4309 * no list insertion occurred, flow is just created,
4310 * this is caller's responsibility to track the
4313 * Flow rule attributes.
4315 * Pattern specification (list terminated by the END pattern item).
4316 * @param[in] actions
4317 * Associated actions (list terminated by the END action).
4318 * @param[in] external
4319 * This flow rule is created by request external to PMD.
4321 * Perform verbose error reporting if not NULL.
4324 * A flow index on success, 0 otherwise and rte_errno is set.
4327 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
4328 const struct rte_flow_attr *attr,
4329 const struct rte_flow_item items[],
4330 const struct rte_flow_action actions[],
4331 bool external, struct rte_flow_error *error)
4333 struct mlx5_priv *priv = dev->data->dev_private;
4334 struct rte_flow *flow = NULL;
4335 struct mlx5_flow *dev_flow;
4336 const struct rte_flow_action_rss *rss;
4338 struct rte_flow_expand_rss buf;
4339 uint8_t buffer[2048];
4342 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4343 uint8_t buffer[2048];
4346 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4347 uint8_t buffer[2048];
4348 } actions_hairpin_tx;
4350 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4351 uint8_t buffer[2048];
4353 struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4354 struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
4355 priv->rss_desc)[!!priv->flow_idx];
4356 const struct rte_flow_action *p_actions_rx = actions;
4360 uint32_t hairpin_id = 0;
4361 struct rte_flow_attr attr_tx = { .priority = 0 };
4364 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4365 ret = flow_drv_validate(dev, attr, items, p_actions_rx,
4366 external, hairpin_flow, error);
4369 if (hairpin_flow > 0) {
4370 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4374 flow_hairpin_split(dev, actions, actions_rx.actions,
4375 actions_hairpin_tx.actions, items_tx.items,
4377 p_actions_rx = actions_rx.actions;
4379 flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
4382 goto error_before_flow;
4384 flow->drv_type = flow_get_drv_type(dev, attr);
4385 if (hairpin_id != 0)
4386 flow->hairpin_flow_id = hairpin_id;
4387 MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4388 flow->drv_type < MLX5_FLOW_TYPE_MAX);
4389 memset(rss_desc, 0, sizeof(*rss_desc));
4390 rss = flow_get_rss_action(p_actions_rx);
4393 * The following information is required by
4394 * mlx5_flow_hashfields_adjust() in advance.
4396 rss_desc->level = rss->level;
4397 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4398 rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
4400 flow->dev_handles = 0;
4401 if (rss && rss->types) {
4402 unsigned int graph_root;
4404 graph_root = find_graph_root(items, rss->level);
4405 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4407 mlx5_support_expansion,
4409 MLX5_ASSERT(ret > 0 &&
4410 (unsigned int)ret < sizeof(expand_buffer.buffer));
4413 buf->entry[0].pattern = (void *)(uintptr_t)items;
4416 * Record the start index when there is a nested call. All sub-flows
4417 * need to be translated before another calling.
4418 * No need to use ping-pong buffer to save memory here.
4420 if (priv->flow_idx) {
4421 MLX5_ASSERT(!priv->flow_nested_idx);
4422 priv->flow_nested_idx = priv->flow_idx;
4424 for (i = 0; i < buf->entries; ++i) {
4426 * The splitter may create multiple dev_flows,
4427 * depending on configuration. In the simplest
4428 * case it just creates unmodified original flow.
4430 ret = flow_create_split_outer(dev, flow, attr,
4431 buf->entry[i].pattern,
4432 p_actions_rx, external, idx,
4437 /* Create the tx flow. */
4439 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4440 attr_tx.ingress = 0;
4442 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4443 actions_hairpin_tx.actions,
4447 dev_flow->flow = flow;
4448 dev_flow->external = 0;
4449 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4450 dev_flow->handle, next);
4451 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4453 actions_hairpin_tx.actions, error);
4458 * Update the metadata register copy table. If extensive
4459 * metadata feature is enabled and registers are supported
4460 * we might create the extra rte_flow for each unique
4461 * MARK/FLAG action ID.
4463 * The table is updated for ingress Flows only, because
4464 * the egress Flows belong to the different device and
4465 * copy table should be updated in peer NIC Rx domain.
4467 if (attr->ingress &&
4468 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4469 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4474 * If the flow is external (from application) OR device is started, then
4475 * the flow will be applied immediately.
4477 if (external || dev->data->dev_started) {
4478 ret = flow_drv_apply(dev, flow, error);
4483 ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
4485 flow_rxq_flags_set(dev, flow);
4486 /* Nested flow creation index recovery. */
4487 priv->flow_idx = priv->flow_nested_idx;
4488 if (priv->flow_nested_idx)
4489 priv->flow_nested_idx = 0;
4493 ret = rte_errno; /* Save rte_errno before cleanup. */
4494 flow_mreg_del_copy_action(dev, flow);
4495 flow_drv_destroy(dev, flow);
4496 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
4497 rte_errno = ret; /* Restore rte_errno. */
4501 mlx5_flow_id_release(priv->sh->flow_id_pool,
4504 priv->flow_idx = priv->flow_nested_idx;
4505 if (priv->flow_nested_idx)
4506 priv->flow_nested_idx = 0;
4511 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4512 * incoming packets to table 1.
4514 * Other flow rules, requested for group n, will be created in
4515 * e-switch table n+1.
4516 * Jump action to e-switch group n will be created to group n+1.
4518 * Used when working in switchdev mode, to utilise advantages of table 1
4522 * Pointer to Ethernet device.
4525 * Pointer to flow on success, NULL otherwise and rte_errno is set.
4528 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4530 const struct rte_flow_attr attr = {
4537 const struct rte_flow_item pattern = {
4538 .type = RTE_FLOW_ITEM_TYPE_END,
4540 struct rte_flow_action_jump jump = {
4543 const struct rte_flow_action actions[] = {
4545 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4549 .type = RTE_FLOW_ACTION_TYPE_END,
4552 struct mlx5_priv *priv = dev->data->dev_private;
4553 struct rte_flow_error error;
4555 return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
4557 actions, false, &error);
4561 * Validate a flow supported by the NIC.
4563 * @see rte_flow_validate()
4567 mlx5_flow_validate(struct rte_eth_dev *dev,
4568 const struct rte_flow_attr *attr,
4569 const struct rte_flow_item items[],
4570 const struct rte_flow_action actions[],
4571 struct rte_flow_error *error)
4575 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4576 return flow_drv_validate(dev, attr, items, actions,
4577 true, hairpin_flow, error);
4583 * @see rte_flow_create()
4587 mlx5_flow_create(struct rte_eth_dev *dev,
4588 const struct rte_flow_attr *attr,
4589 const struct rte_flow_item items[],
4590 const struct rte_flow_action actions[],
4591 struct rte_flow_error *error)
4593 struct mlx5_priv *priv = dev->data->dev_private;
4596 * If the device is not started yet, it is not allowed to created a
4597 * flow from application. PMD default flows and traffic control flows
4600 if (unlikely(!dev->data->dev_started)) {
4601 DRV_LOG(DEBUG, "port %u is not started when "
4602 "inserting a flow", dev->data->port_id);
4603 rte_flow_error_set(error, ENODEV,
4604 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4606 "port not started");
4609 return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
4610 attr, items, actions, true, error);
4614 * Destroy a flow in a list.
4617 * Pointer to Ethernet device.
4619 * Pointer to the Indexed flow list. If this parameter NULL,
4620 * there is no flow removal from the list. Be noted that as
4621 * flow is add to the indexed list, memory of the indexed
4622 * list points to maybe changed as flow destroyed.
4623 * @param[in] flow_idx
4624 * Index of flow to destroy.
4627 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
4630 struct mlx5_priv *priv = dev->data->dev_private;
4631 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
4632 struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
4633 [MLX5_IPOOL_RTE_FLOW], flow_idx);
4638 * Update RX queue flags only if port is started, otherwise it is
4641 if (dev->data->dev_started)
4642 flow_rxq_flags_trim(dev, flow);
4643 if (flow->hairpin_flow_id)
4644 mlx5_flow_id_release(priv->sh->flow_id_pool,
4645 flow->hairpin_flow_id);
4646 flow_drv_destroy(dev, flow);
4648 ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
4649 flow_idx, flow, next);
4650 flow_mreg_del_copy_action(dev, flow);
4652 LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
4653 if (priv_fdir_flow->rix_flow == flow_idx)
4656 if (priv_fdir_flow) {
4657 LIST_REMOVE(priv_fdir_flow, next);
4658 mlx5_free(priv_fdir_flow->fdir);
4659 mlx5_free(priv_fdir_flow);
4662 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
4666 * Destroy all flows.
4669 * Pointer to Ethernet device.
4671 * Pointer to the Indexed flow list.
4673 * If flushing is called avtively.
4676 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
4678 uint32_t num_flushed = 0;
4681 flow_list_destroy(dev, list, *list);
4685 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4686 dev->data->port_id, num_flushed);
4694 * Pointer to Ethernet device.
4696 * Pointer to the Indexed flow list.
4699 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
4701 struct mlx5_priv *priv = dev->data->dev_private;
4702 struct rte_flow *flow = NULL;
4705 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4707 flow_drv_remove(dev, flow);
4708 flow_mreg_stop_copy_action(dev, flow);
4710 flow_mreg_del_default_copy_action(dev);
4711 flow_rxq_flags_clear(dev);
4718 * Pointer to Ethernet device.
4720 * Pointer to the Indexed flow list.
4723 * 0 on success, a negative errno value otherwise and rte_errno is set.
4726 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
4728 struct mlx5_priv *priv = dev->data->dev_private;
4729 struct rte_flow *flow = NULL;
4730 struct rte_flow_error error;
4734 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4735 ret = flow_mreg_add_default_copy_action(dev, &error);
4738 /* Apply Flows created by application. */
4739 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4741 ret = flow_mreg_start_copy_action(dev, flow);
4744 ret = flow_drv_apply(dev, flow, &error);
4747 flow_rxq_flags_set(dev, flow);
4751 ret = rte_errno; /* Save rte_errno before cleanup. */
4752 mlx5_flow_stop(dev, list);
4753 rte_errno = ret; /* Restore rte_errno. */
4758 * Stop all default actions for flows.
4761 * Pointer to Ethernet device.
4764 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4766 flow_mreg_del_default_copy_action(dev);
4767 flow_rxq_flags_clear(dev);
4771 * Start all default actions for flows.
4774 * Pointer to Ethernet device.
4776 * 0 on success, a negative errno value otherwise and rte_errno is set.
4779 mlx5_flow_start_default(struct rte_eth_dev *dev)
4781 struct rte_flow_error error;
4783 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4784 return flow_mreg_add_default_copy_action(dev, &error);
4788 * Allocate intermediate resources for flow creation.
4791 * Pointer to Ethernet device.
4794 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
4796 struct mlx5_priv *priv = dev->data->dev_private;
4798 if (!priv->inter_flows) {
4799 priv->inter_flows = mlx5_malloc(MLX5_MEM_ZERO,
4800 MLX5_NUM_MAX_DEV_FLOWS *
4801 sizeof(struct mlx5_flow) +
4802 (sizeof(struct mlx5_flow_rss_desc) +
4803 sizeof(uint16_t) * UINT16_MAX) * 2, 0,
4805 if (!priv->inter_flows) {
4806 DRV_LOG(ERR, "can't allocate intermediate memory.");
4810 priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
4811 [MLX5_NUM_MAX_DEV_FLOWS];
4812 /* Reset the index. */
4814 priv->flow_nested_idx = 0;
4818 * Free intermediate resources for flows.
4821 * Pointer to Ethernet device.
4824 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
4826 struct mlx5_priv *priv = dev->data->dev_private;
4828 mlx5_free(priv->inter_flows);
4829 priv->inter_flows = NULL;
4833 * Verify the flow list is empty
4836 * Pointer to Ethernet device.
4838 * @return the number of flows not released.
4841 mlx5_flow_verify(struct rte_eth_dev *dev)
4843 struct mlx5_priv *priv = dev->data->dev_private;
4844 struct rte_flow *flow;
4848 ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
4850 DRV_LOG(DEBUG, "port %u flow %p still referenced",
4851 dev->data->port_id, (void *)flow);
4858 * Enable default hairpin egress flow.
4861 * Pointer to Ethernet device.
4866 * 0 on success, a negative errno value otherwise and rte_errno is set.
4869 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4872 struct mlx5_priv *priv = dev->data->dev_private;
4873 const struct rte_flow_attr attr = {
4877 struct mlx5_rte_flow_item_tx_queue queue_spec = {
4880 struct mlx5_rte_flow_item_tx_queue queue_mask = {
4881 .queue = UINT32_MAX,
4883 struct rte_flow_item items[] = {
4885 .type = (enum rte_flow_item_type)
4886 MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4887 .spec = &queue_spec,
4889 .mask = &queue_mask,
4892 .type = RTE_FLOW_ITEM_TYPE_END,
4895 struct rte_flow_action_jump jump = {
4896 .group = MLX5_HAIRPIN_TX_TABLE,
4898 struct rte_flow_action actions[2];
4900 struct rte_flow_error error;
4902 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4903 actions[0].conf = &jump;
4904 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4905 flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4906 &attr, items, actions, false, &error);
4909 "Failed to create ctrl flow: rte_errno(%d),"
4910 " type(%d), message(%s)",
4911 rte_errno, error.type,
4912 error.message ? error.message : " (no stated reason)");
4919 * Enable a control flow configured from the control plane.
4922 * Pointer to Ethernet device.
4924 * An Ethernet flow spec to apply.
4926 * An Ethernet flow mask to apply.
4928 * A VLAN flow spec to apply.
4930 * A VLAN flow mask to apply.
4933 * 0 on success, a negative errno value otherwise and rte_errno is set.
4936 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4937 struct rte_flow_item_eth *eth_spec,
4938 struct rte_flow_item_eth *eth_mask,
4939 struct rte_flow_item_vlan *vlan_spec,
4940 struct rte_flow_item_vlan *vlan_mask)
4942 struct mlx5_priv *priv = dev->data->dev_private;
4943 const struct rte_flow_attr attr = {
4945 .priority = MLX5_FLOW_PRIO_RSVD,
4947 struct rte_flow_item items[] = {
4949 .type = RTE_FLOW_ITEM_TYPE_ETH,
4955 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4956 RTE_FLOW_ITEM_TYPE_END,
4962 .type = RTE_FLOW_ITEM_TYPE_END,
4965 uint16_t queue[priv->reta_idx_n];
4966 struct rte_flow_action_rss action_rss = {
4967 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4969 .types = priv->rss_conf.rss_hf,
4970 .key_len = priv->rss_conf.rss_key_len,
4971 .queue_num = priv->reta_idx_n,
4972 .key = priv->rss_conf.rss_key,
4975 struct rte_flow_action actions[] = {
4977 .type = RTE_FLOW_ACTION_TYPE_RSS,
4978 .conf = &action_rss,
4981 .type = RTE_FLOW_ACTION_TYPE_END,
4985 struct rte_flow_error error;
4988 if (!priv->reta_idx_n || !priv->rxqs_n) {
4991 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
4992 action_rss.types = 0;
4993 for (i = 0; i != priv->reta_idx_n; ++i)
4994 queue[i] = (*priv->reta_idx)[i];
4995 flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4996 &attr, items, actions, false, &error);
5003 * Enable a flow control configured from the control plane.
5006 * Pointer to Ethernet device.
5008 * An Ethernet flow spec to apply.
5010 * An Ethernet flow mask to apply.
5013 * 0 on success, a negative errno value otherwise and rte_errno is set.
5016 mlx5_ctrl_flow(struct rte_eth_dev *dev,
5017 struct rte_flow_item_eth *eth_spec,
5018 struct rte_flow_item_eth *eth_mask)
5020 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
5024 * Create default miss flow rule matching lacp traffic
5027 * Pointer to Ethernet device.
5029 * An Ethernet flow spec to apply.
5032 * 0 on success, a negative errno value otherwise and rte_errno is set.
5035 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
5037 struct mlx5_priv *priv = dev->data->dev_private;
5039 * The LACP matching is done by only using ether type since using
5040 * a multicast dst mac causes kernel to give low priority to this flow.
5042 static const struct rte_flow_item_eth lacp_spec = {
5043 .type = RTE_BE16(0x8809),
5045 static const struct rte_flow_item_eth lacp_mask = {
5048 const struct rte_flow_attr attr = {
5051 struct rte_flow_item items[] = {
5053 .type = RTE_FLOW_ITEM_TYPE_ETH,
5058 .type = RTE_FLOW_ITEM_TYPE_END,
5061 struct rte_flow_action actions[] = {
5063 .type = (enum rte_flow_action_type)
5064 MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
5067 .type = RTE_FLOW_ACTION_TYPE_END,
5070 struct rte_flow_error error;
5071 uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5072 &attr, items, actions, false, &error);
5082 * @see rte_flow_destroy()
5086 mlx5_flow_destroy(struct rte_eth_dev *dev,
5087 struct rte_flow *flow,
5088 struct rte_flow_error *error __rte_unused)
5090 struct mlx5_priv *priv = dev->data->dev_private;
5092 flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
5097 * Destroy all flows.
5099 * @see rte_flow_flush()
5103 mlx5_flow_flush(struct rte_eth_dev *dev,
5104 struct rte_flow_error *error __rte_unused)
5106 struct mlx5_priv *priv = dev->data->dev_private;
5108 mlx5_flow_list_flush(dev, &priv->flows, false);
5115 * @see rte_flow_isolate()
5119 mlx5_flow_isolate(struct rte_eth_dev *dev,
5121 struct rte_flow_error *error)
5123 struct mlx5_priv *priv = dev->data->dev_private;
5125 if (dev->data->dev_started) {
5126 rte_flow_error_set(error, EBUSY,
5127 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5129 "port must be stopped first");
5132 priv->isolated = !!enable;
5134 dev->dev_ops = &mlx5_os_dev_ops_isolate;
5136 dev->dev_ops = &mlx5_os_dev_ops;
5143 * @see rte_flow_query()
5147 flow_drv_query(struct rte_eth_dev *dev,
5149 const struct rte_flow_action *actions,
5151 struct rte_flow_error *error)
5153 struct mlx5_priv *priv = dev->data->dev_private;
5154 const struct mlx5_flow_driver_ops *fops;
5155 struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5156 [MLX5_IPOOL_RTE_FLOW],
5158 enum mlx5_flow_drv_type ftype;
5161 return rte_flow_error_set(error, ENOENT,
5162 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5164 "invalid flow handle");
5166 ftype = flow->drv_type;
5167 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
5168 fops = flow_get_drv_ops(ftype);
5170 return fops->query(dev, flow, actions, data, error);
5176 * @see rte_flow_query()
5180 mlx5_flow_query(struct rte_eth_dev *dev,
5181 struct rte_flow *flow,
5182 const struct rte_flow_action *actions,
5184 struct rte_flow_error *error)
5188 ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
5196 * Convert a flow director filter to a generic flow.
5199 * Pointer to Ethernet device.
5200 * @param fdir_filter
5201 * Flow director filter to add.
5203 * Generic flow parameters structure.
5206 * 0 on success, a negative errno value otherwise and rte_errno is set.
5209 flow_fdir_filter_convert(struct rte_eth_dev *dev,
5210 const struct rte_eth_fdir_filter *fdir_filter,
5211 struct mlx5_fdir *attributes)
5213 struct mlx5_priv *priv = dev->data->dev_private;
5214 const struct rte_eth_fdir_input *input = &fdir_filter->input;
5215 const struct rte_eth_fdir_masks *mask =
5216 &dev->data->dev_conf.fdir_conf.mask;
5218 /* Validate queue number. */
5219 if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
5220 DRV_LOG(ERR, "port %u invalid queue number %d",
5221 dev->data->port_id, fdir_filter->action.rx_queue);
5225 attributes->attr.ingress = 1;
5226 attributes->items[0] = (struct rte_flow_item) {
5227 .type = RTE_FLOW_ITEM_TYPE_ETH,
5228 .spec = &attributes->l2,
5229 .mask = &attributes->l2_mask,
5231 switch (fdir_filter->action.behavior) {
5232 case RTE_ETH_FDIR_ACCEPT:
5233 attributes->actions[0] = (struct rte_flow_action){
5234 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
5235 .conf = &attributes->queue,
5238 case RTE_ETH_FDIR_REJECT:
5239 attributes->actions[0] = (struct rte_flow_action){
5240 .type = RTE_FLOW_ACTION_TYPE_DROP,
5244 DRV_LOG(ERR, "port %u invalid behavior %d",
5246 fdir_filter->action.behavior);
5247 rte_errno = ENOTSUP;
5250 attributes->queue.index = fdir_filter->action.rx_queue;
5252 switch (fdir_filter->input.flow_type) {
5253 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5254 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5255 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5256 attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5257 .src_addr = input->flow.ip4_flow.src_ip,
5258 .dst_addr = input->flow.ip4_flow.dst_ip,
5259 .time_to_live = input->flow.ip4_flow.ttl,
5260 .type_of_service = input->flow.ip4_flow.tos,
5262 attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5263 .src_addr = mask->ipv4_mask.src_ip,
5264 .dst_addr = mask->ipv4_mask.dst_ip,
5265 .time_to_live = mask->ipv4_mask.ttl,
5266 .type_of_service = mask->ipv4_mask.tos,
5267 .next_proto_id = mask->ipv4_mask.proto,
5269 attributes->items[1] = (struct rte_flow_item){
5270 .type = RTE_FLOW_ITEM_TYPE_IPV4,
5271 .spec = &attributes->l3,
5272 .mask = &attributes->l3_mask,
5275 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5276 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5277 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5278 attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5279 .hop_limits = input->flow.ipv6_flow.hop_limits,
5280 .proto = input->flow.ipv6_flow.proto,
5283 memcpy(attributes->l3.ipv6.hdr.src_addr,
5284 input->flow.ipv6_flow.src_ip,
5285 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5286 memcpy(attributes->l3.ipv6.hdr.dst_addr,
5287 input->flow.ipv6_flow.dst_ip,
5288 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5289 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5290 mask->ipv6_mask.src_ip,
5291 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5292 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5293 mask->ipv6_mask.dst_ip,
5294 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5295 attributes->items[1] = (struct rte_flow_item){
5296 .type = RTE_FLOW_ITEM_TYPE_IPV6,
5297 .spec = &attributes->l3,
5298 .mask = &attributes->l3_mask,
5302 DRV_LOG(ERR, "port %u invalid flow type%d",
5303 dev->data->port_id, fdir_filter->input.flow_type);
5304 rte_errno = ENOTSUP;
5308 switch (fdir_filter->input.flow_type) {
5309 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5310 attributes->l4.udp.hdr = (struct rte_udp_hdr){
5311 .src_port = input->flow.udp4_flow.src_port,
5312 .dst_port = input->flow.udp4_flow.dst_port,
5314 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5315 .src_port = mask->src_port_mask,
5316 .dst_port = mask->dst_port_mask,
5318 attributes->items[2] = (struct rte_flow_item){
5319 .type = RTE_FLOW_ITEM_TYPE_UDP,
5320 .spec = &attributes->l4,
5321 .mask = &attributes->l4_mask,
5324 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5325 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5326 .src_port = input->flow.tcp4_flow.src_port,
5327 .dst_port = input->flow.tcp4_flow.dst_port,
5329 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5330 .src_port = mask->src_port_mask,
5331 .dst_port = mask->dst_port_mask,
5333 attributes->items[2] = (struct rte_flow_item){
5334 .type = RTE_FLOW_ITEM_TYPE_TCP,
5335 .spec = &attributes->l4,
5336 .mask = &attributes->l4_mask,
5339 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5340 attributes->l4.udp.hdr = (struct rte_udp_hdr){
5341 .src_port = input->flow.udp6_flow.src_port,
5342 .dst_port = input->flow.udp6_flow.dst_port,
5344 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5345 .src_port = mask->src_port_mask,
5346 .dst_port = mask->dst_port_mask,
5348 attributes->items[2] = (struct rte_flow_item){
5349 .type = RTE_FLOW_ITEM_TYPE_UDP,
5350 .spec = &attributes->l4,
5351 .mask = &attributes->l4_mask,
5354 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5355 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5356 .src_port = input->flow.tcp6_flow.src_port,
5357 .dst_port = input->flow.tcp6_flow.dst_port,
5359 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5360 .src_port = mask->src_port_mask,
5361 .dst_port = mask->dst_port_mask,
5363 attributes->items[2] = (struct rte_flow_item){
5364 .type = RTE_FLOW_ITEM_TYPE_TCP,
5365 .spec = &attributes->l4,
5366 .mask = &attributes->l4_mask,
5369 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5370 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5373 DRV_LOG(ERR, "port %u invalid flow type%d",
5374 dev->data->port_id, fdir_filter->input.flow_type);
5375 rte_errno = ENOTSUP;
5381 #define FLOW_FDIR_CMP(f1, f2, fld) \
5382 memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5385 * Compare two FDIR flows. If items and actions are identical, the two flows are
5389 * Pointer to Ethernet device.
5391 * FDIR flow to compare.
5393 * FDIR flow to compare.
5396 * Zero on match, 1 otherwise.
5399 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5401 if (FLOW_FDIR_CMP(f1, f2, attr) ||
5402 FLOW_FDIR_CMP(f1, f2, l2) ||
5403 FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5404 FLOW_FDIR_CMP(f1, f2, l3) ||
5405 FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5406 FLOW_FDIR_CMP(f1, f2, l4) ||
5407 FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5408 FLOW_FDIR_CMP(f1, f2, actions[0].type))
5410 if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5411 FLOW_FDIR_CMP(f1, f2, queue))
5417 * Search device flow list to find out a matched FDIR flow.
5420 * Pointer to Ethernet device.
5422 * FDIR flow to lookup.
5425 * Index of flow if found, 0 otherwise.
5428 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5430 struct mlx5_priv *priv = dev->data->dev_private;
5431 uint32_t flow_idx = 0;
5432 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5434 MLX5_ASSERT(fdir_flow);
5435 LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5436 if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
5437 DRV_LOG(DEBUG, "port %u found FDIR flow %u",
5438 dev->data->port_id, flow_idx);
5439 flow_idx = priv_fdir_flow->rix_flow;
5447 * Add new flow director filter and store it in list.
5450 * Pointer to Ethernet device.
5451 * @param fdir_filter
5452 * Flow director filter to add.
5455 * 0 on success, a negative errno value otherwise and rte_errno is set.
5458 flow_fdir_filter_add(struct rte_eth_dev *dev,
5459 const struct rte_eth_fdir_filter *fdir_filter)
5461 struct mlx5_priv *priv = dev->data->dev_private;
5462 struct mlx5_fdir *fdir_flow;
5463 struct rte_flow *flow;
5464 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5468 fdir_flow = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*fdir_flow), 0,
5474 ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5477 flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
5482 priv_fdir_flow = mlx5_malloc(MLX5_MEM_ZERO,
5483 sizeof(struct mlx5_fdir_flow),
5485 if (!priv_fdir_flow) {
5489 flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5490 fdir_flow->items, fdir_flow->actions, true,
5492 flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5496 priv_fdir_flow->fdir = fdir_flow;
5497 priv_fdir_flow->rix_flow = flow_idx;
5498 LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
5499 DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5500 dev->data->port_id, (void *)flow);
5503 mlx5_free(priv_fdir_flow);
5504 mlx5_free(fdir_flow);
5509 * Delete specific filter.
5512 * Pointer to Ethernet device.
5513 * @param fdir_filter
5514 * Filter to be deleted.
5517 * 0 on success, a negative errno value otherwise and rte_errno is set.
5520 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5521 const struct rte_eth_fdir_filter *fdir_filter)
5523 struct mlx5_priv *priv = dev->data->dev_private;
5525 struct mlx5_fdir fdir_flow = {
5528 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5531 ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5534 LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5535 /* Find the fdir in priv list */
5536 if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
5539 if (!priv_fdir_flow)
5541 LIST_REMOVE(priv_fdir_flow, next);
5542 flow_idx = priv_fdir_flow->rix_flow;
5543 flow_list_destroy(dev, &priv->flows, flow_idx);
5544 mlx5_free(priv_fdir_flow->fdir);
5545 mlx5_free(priv_fdir_flow);
5546 DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
5547 dev->data->port_id, flow_idx);
5552 * Update queue for specific filter.
5555 * Pointer to Ethernet device.
5556 * @param fdir_filter
5557 * Filter to be updated.
5560 * 0 on success, a negative errno value otherwise and rte_errno is set.
5563 flow_fdir_filter_update(struct rte_eth_dev *dev,
5564 const struct rte_eth_fdir_filter *fdir_filter)
5568 ret = flow_fdir_filter_delete(dev, fdir_filter);
5571 return flow_fdir_filter_add(dev, fdir_filter);
5575 * Flush all filters.
5578 * Pointer to Ethernet device.
5581 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5583 struct mlx5_priv *priv = dev->data->dev_private;
5584 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5586 while (!LIST_EMPTY(&priv->fdir_flows)) {
5587 priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
5588 LIST_REMOVE(priv_fdir_flow, next);
5589 flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
5590 mlx5_free(priv_fdir_flow->fdir);
5591 mlx5_free(priv_fdir_flow);
5596 * Get flow director information.
5599 * Pointer to Ethernet device.
5600 * @param[out] fdir_info
5601 * Resulting flow director information.
5604 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5606 struct rte_eth_fdir_masks *mask =
5607 &dev->data->dev_conf.fdir_conf.mask;
5609 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5610 fdir_info->guarant_spc = 0;
5611 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5612 fdir_info->max_flexpayload = 0;
5613 fdir_info->flow_types_mask[0] = 0;
5614 fdir_info->flex_payload_unit = 0;
5615 fdir_info->max_flex_payload_segment_num = 0;
5616 fdir_info->flex_payload_limit = 0;
5617 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5621 * Deal with flow director operations.
5624 * Pointer to Ethernet device.
5626 * Operation to perform.
5628 * Pointer to operation-specific structure.
5631 * 0 on success, a negative errno value otherwise and rte_errno is set.
5634 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5637 enum rte_fdir_mode fdir_mode =
5638 dev->data->dev_conf.fdir_conf.mode;
5640 if (filter_op == RTE_ETH_FILTER_NOP)
5642 if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5643 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5644 DRV_LOG(ERR, "port %u flow director mode %d not supported",
5645 dev->data->port_id, fdir_mode);
5649 switch (filter_op) {
5650 case RTE_ETH_FILTER_ADD:
5651 return flow_fdir_filter_add(dev, arg);
5652 case RTE_ETH_FILTER_UPDATE:
5653 return flow_fdir_filter_update(dev, arg);
5654 case RTE_ETH_FILTER_DELETE:
5655 return flow_fdir_filter_delete(dev, arg);
5656 case RTE_ETH_FILTER_FLUSH:
5657 flow_fdir_filter_flush(dev);
5659 case RTE_ETH_FILTER_INFO:
5660 flow_fdir_info_get(dev, arg);
5663 DRV_LOG(DEBUG, "port %u unknown operation %u",
5664 dev->data->port_id, filter_op);
5672 * Manage filter operations.
5675 * Pointer to Ethernet device structure.
5676 * @param filter_type
5679 * Operation to perform.
5681 * Pointer to operation-specific structure.
5684 * 0 on success, a negative errno value otherwise and rte_errno is set.
5687 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5688 enum rte_filter_type filter_type,
5689 enum rte_filter_op filter_op,
5692 switch (filter_type) {
5693 case RTE_ETH_FILTER_GENERIC:
5694 if (filter_op != RTE_ETH_FILTER_GET) {
5698 *(const void **)arg = &mlx5_flow_ops;
5700 case RTE_ETH_FILTER_FDIR:
5701 return flow_fdir_ctrl_func(dev, filter_op, arg);
5703 DRV_LOG(ERR, "port %u filter type (%d) not supported",
5704 dev->data->port_id, filter_type);
5705 rte_errno = ENOTSUP;
5712 * Create the needed meter and suffix tables.
5715 * Pointer to Ethernet device.
5717 * Pointer to the flow meter.
5720 * Pointer to table set on success, NULL otherwise.
5722 struct mlx5_meter_domains_infos *
5723 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5724 const struct mlx5_flow_meter *fm)
5726 const struct mlx5_flow_driver_ops *fops;
5728 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5729 return fops->create_mtr_tbls(dev, fm);
5733 * Destroy the meter table set.
5736 * Pointer to Ethernet device.
5738 * Pointer to the meter table set.
5744 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5745 struct mlx5_meter_domains_infos *tbls)
5747 const struct mlx5_flow_driver_ops *fops;
5749 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5750 return fops->destroy_mtr_tbls(dev, tbls);
5754 * Create policer rules.
5757 * Pointer to Ethernet device.
5759 * Pointer to flow meter structure.
5761 * Pointer to flow attributes.
5764 * 0 on success, -1 otherwise.
5767 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5768 struct mlx5_flow_meter *fm,
5769 const struct rte_flow_attr *attr)
5771 const struct mlx5_flow_driver_ops *fops;
5773 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5774 return fops->create_policer_rules(dev, fm, attr);
5778 * Destroy policer rules.
5781 * Pointer to flow meter structure.
5783 * Pointer to flow attributes.
5786 * 0 on success, -1 otherwise.
5789 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5790 struct mlx5_flow_meter *fm,
5791 const struct rte_flow_attr *attr)
5793 const struct mlx5_flow_driver_ops *fops;
5795 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5796 return fops->destroy_policer_rules(dev, fm, attr);
5800 * Allocate a counter.
5803 * Pointer to Ethernet device structure.
5806 * Index to allocated counter on success, 0 otherwise.
5809 mlx5_counter_alloc(struct rte_eth_dev *dev)
5811 const struct mlx5_flow_driver_ops *fops;
5812 struct rte_flow_attr attr = { .transfer = 0 };
5814 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5815 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5816 return fops->counter_alloc(dev);
5819 "port %u counter allocate is not supported.",
5820 dev->data->port_id);
5828 * Pointer to Ethernet device structure.
5830 * Index to counter to be free.
5833 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
5835 const struct mlx5_flow_driver_ops *fops;
5836 struct rte_flow_attr attr = { .transfer = 0 };
5838 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5839 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5840 fops->counter_free(dev, cnt);
5844 "port %u counter free is not supported.",
5845 dev->data->port_id);
5849 * Query counter statistics.
5852 * Pointer to Ethernet device structure.
5854 * Index to counter to query.
5856 * Set to clear counter statistics.
5858 * The counter hits packets number to save.
5860 * The counter hits bytes number to save.
5863 * 0 on success, a negative errno value otherwise.
5866 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
5867 bool clear, uint64_t *pkts, uint64_t *bytes)
5869 const struct mlx5_flow_driver_ops *fops;
5870 struct rte_flow_attr attr = { .transfer = 0 };
5872 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5873 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5874 return fops->counter_query(dev, cnt, clear, pkts, bytes);
5877 "port %u counter query is not supported.",
5878 dev->data->port_id);
5882 #define MLX5_POOL_QUERY_FREQ_US 1000000
5885 * Get number of all validate pools.
5888 * Pointer to mlx5_dev_ctx_shared object.
5891 * The number of all validate pools.
5894 mlx5_get_all_valid_pool_count(struct mlx5_dev_ctx_shared *sh)
5897 uint32_t pools_n = 0;
5899 for (i = 0; i < MLX5_CCONT_TYPE_MAX; ++i)
5900 pools_n += rte_atomic16_read(&sh->cmng.ccont[i].n_valid);
5905 * Set the periodic procedure for triggering asynchronous batch queries for all
5906 * the counter pools.
5909 * Pointer to mlx5_dev_ctx_shared object.
5912 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
5914 uint32_t pools_n, us;
5916 pools_n = mlx5_get_all_valid_pool_count(sh);
5917 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5918 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5919 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5920 sh->cmng.query_thread_on = 0;
5921 DRV_LOG(ERR, "Cannot reinitialize query alarm");
5923 sh->cmng.query_thread_on = 1;
5928 * The periodic procedure for triggering asynchronous batch queries for all the
5929 * counter pools. This function is probably called by the host thread.
5932 * The parameter for the alarm process.
5935 mlx5_flow_query_alarm(void *arg)
5937 struct mlx5_dev_ctx_shared *sh = arg;
5938 struct mlx5_devx_obj *dcs;
5941 uint8_t batch = sh->cmng.batch;
5942 uint8_t age = sh->cmng.age;
5943 uint16_t pool_index = sh->cmng.pool_index;
5944 struct mlx5_pools_container *cont;
5945 struct mlx5_flow_counter_pool *pool;
5946 int cont_loop = MLX5_CCONT_TYPE_MAX;
5948 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5951 cont = MLX5_CNT_CONTAINER(sh, batch, age);
5952 rte_spinlock_lock(&cont->resize_sl);
5954 rte_spinlock_unlock(&cont->resize_sl);
5955 /* Check if all the containers are empty. */
5956 if (unlikely(--cont_loop == 0))
5960 if (batch == 0 && pool_index == 0) {
5962 sh->cmng.batch = batch;
5965 goto next_container;
5967 pool = cont->pools[pool_index];
5968 rte_spinlock_unlock(&cont->resize_sl);
5970 /* There is a pool query in progress. */
5973 LIST_FIRST(&sh->cmng.free_stat_raws);
5975 /* No free counter statistics raw memory. */
5977 dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5979 offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5981 * Identify the counters released between query trigger and query
5982 * handle more effiecntly. The counter released in this gap period
5983 * should wait for a new round of query as the new arrived packets
5984 * will not be taken into account.
5987 ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5989 pool->raw_hw->mem_mng->dm->id,
5991 (pool->raw_hw->data + offset),
5993 (uint64_t)(uintptr_t)pool);
5995 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5996 " %d", pool->min_dcs->id);
5997 pool->raw_hw = NULL;
6000 pool->raw_hw->min_dcs_id = dcs->id;
6001 LIST_REMOVE(pool->raw_hw, next);
6002 sh->cmng.pending_queries++;
6004 if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
6007 if (batch == 0 && pool_index == 0)
6011 sh->cmng.batch = batch;
6012 sh->cmng.pool_index = pool_index;
6014 mlx5_set_query_alarm(sh);
6018 * Check and callback event for new aged flow in the counter pool
6021 * Pointer to mlx5_dev_ctx_shared object.
6023 * Pointer to Current counter pool.
6026 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6027 struct mlx5_flow_counter_pool *pool)
6029 struct mlx5_priv *priv;
6030 struct mlx5_flow_counter *cnt;
6031 struct mlx5_age_info *age_info;
6032 struct mlx5_age_param *age_param;
6033 struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6034 struct mlx5_counter_stats_raw *prev = pool->raw;
6035 uint16_t curr = rte_rdtsc() / (rte_get_tsc_hz() / 10);
6038 for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6039 cnt = MLX5_POOL_GET_CNT(pool, i);
6040 age_param = MLX5_CNT_TO_AGE(cnt);
6041 if (rte_atomic16_read(&age_param->state) != AGE_CANDIDATE)
6043 if (cur->data[i].hits != prev->data[i].hits) {
6044 age_param->expire = curr + age_param->timeout;
6047 if ((uint16_t)(curr - age_param->expire) >= (UINT16_MAX / 2))
6050 * Hold the lock first, or if between the
6051 * state AGE_TMOUT and tailq operation the
6052 * release happened, the release procedure
6053 * may delete a non-existent tailq node.
6055 priv = rte_eth_devices[age_param->port_id].data->dev_private;
6056 age_info = GET_PORT_AGE_INFO(priv);
6057 rte_spinlock_lock(&age_info->aged_sl);
6058 /* If the cpmset fails, release happens. */
6059 if (rte_atomic16_cmpset((volatile uint16_t *)
6064 TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6065 MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6067 rte_spinlock_unlock(&age_info->aged_sl);
6069 for (i = 0; i < sh->max_port; i++) {
6070 age_info = &sh->port[i].age_info;
6071 if (!MLX5_AGE_GET(age_info, MLX5_AGE_EVENT_NEW))
6073 if (MLX5_AGE_GET(age_info, MLX5_AGE_TRIGGER))
6074 _rte_eth_dev_callback_process
6075 (&rte_eth_devices[sh->port[i].devx_ih_port_id],
6076 RTE_ETH_EVENT_FLOW_AGED, NULL);
6077 age_info->flags = 0;
6082 * Handler for the HW respond about ready values from an asynchronous batch
6083 * query. This function is probably called by the host thread.
6086 * The pointer to the shared device context.
6087 * @param[in] async_id
6088 * The Devx async ID.
6090 * The status of the completion.
6093 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6094 uint64_t async_id, int status)
6096 struct mlx5_flow_counter_pool *pool =
6097 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6098 struct mlx5_counter_stats_raw *raw_to_free;
6099 uint8_t age = !!IS_AGE_POOL(pool);
6100 uint8_t query_gen = pool->query_gen ^ 1;
6101 struct mlx5_pools_container *cont =
6102 MLX5_CNT_CONTAINER(sh, !IS_EXT_POOL(pool), age);
6104 if (unlikely(status)) {
6105 raw_to_free = pool->raw_hw;
6107 raw_to_free = pool->raw;
6108 if (IS_AGE_POOL(pool))
6109 mlx5_flow_aging_check(sh, pool);
6110 rte_spinlock_lock(&pool->sl);
6111 pool->raw = pool->raw_hw;
6112 rte_spinlock_unlock(&pool->sl);
6113 /* Be sure the new raw counters data is updated in memory. */
6115 if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6116 rte_spinlock_lock(&cont->csl);
6117 TAILQ_CONCAT(&cont->counters,
6118 &pool->counters[query_gen], next);
6119 rte_spinlock_unlock(&cont->csl);
6122 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6123 pool->raw_hw = NULL;
6124 sh->cmng.pending_queries--;
6128 * Translate the rte_flow group index to HW table value.
6130 * @param[in] attributes
6131 * Pointer to flow attributes
6132 * @param[in] external
6133 * Value is part of flow rule created by request external to PMD.
6135 * rte_flow group index value.
6136 * @param[out] fdb_def_rule
6137 * Whether fdb jump to table 1 is configured.
6141 * Pointer to error structure.
6144 * 0 on success, a negative errno value otherwise and rte_errno is set.
6147 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
6148 uint32_t group, bool fdb_def_rule, uint32_t *table,
6149 struct rte_flow_error *error)
6151 if (attributes->transfer && external && fdb_def_rule) {
6152 if (group == UINT32_MAX)
6153 return rte_flow_error_set
6155 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6157 "group index not supported");
6166 * Discover availability of metadata reg_c's.
6168 * Iteratively use test flows to check availability.
6171 * Pointer to the Ethernet device structure.
6174 * 0 on success, a negative errno value otherwise and rte_errno is set.
6177 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
6179 struct mlx5_priv *priv = dev->data->dev_private;
6180 struct mlx5_dev_config *config = &priv->config;
6181 enum modify_reg idx;
6184 /* reg_c[0] and reg_c[1] are reserved. */
6185 config->flow_mreg_c[n++] = REG_C_0;
6186 config->flow_mreg_c[n++] = REG_C_1;
6187 /* Discover availability of other reg_c's. */
6188 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
6189 struct rte_flow_attr attr = {
6190 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
6191 .priority = MLX5_FLOW_PRIO_RSVD,
6194 struct rte_flow_item items[] = {
6196 .type = RTE_FLOW_ITEM_TYPE_END,
6199 struct rte_flow_action actions[] = {
6201 .type = (enum rte_flow_action_type)
6202 MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
6203 .conf = &(struct mlx5_flow_action_copy_mreg){
6209 .type = RTE_FLOW_ACTION_TYPE_JUMP,
6210 .conf = &(struct rte_flow_action_jump){
6211 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
6215 .type = RTE_FLOW_ACTION_TYPE_END,
6219 struct rte_flow *flow;
6220 struct rte_flow_error error;
6222 if (!config->dv_flow_en)
6224 /* Create internal flow, validation skips copy action. */
6225 flow_idx = flow_list_create(dev, NULL, &attr, items,
6226 actions, false, &error);
6227 flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
6231 if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
6232 config->flow_mreg_c[n++] = idx;
6233 flow_list_destroy(dev, NULL, flow_idx);
6235 for (; n < MLX5_MREG_C_NUM; ++n)
6236 config->flow_mreg_c[n] = REG_NONE;
6241 * Dump flow raw hw data to file
6244 * The pointer to Ethernet device.
6246 * A pointer to a file for output.
6248 * Perform verbose error reporting if not NULL. PMDs initialize this
6249 * structure in case of error only.
6251 * 0 on success, a nagative value otherwise.
6254 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
6256 struct rte_flow_error *error __rte_unused)
6258 struct mlx5_priv *priv = dev->data->dev_private;
6259 struct mlx5_dev_ctx_shared *sh = priv->sh;
6261 return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
6262 sh->tx_domain, file);
6266 * Get aged-out flows.
6269 * Pointer to the Ethernet device structure.
6270 * @param[in] context
6271 * The address of an array of pointers to the aged-out flows contexts.
6272 * @param[in] nb_countexts
6273 * The length of context array pointers.
6275 * Perform verbose error reporting if not NULL. Initialized in case of
6279 * how many contexts get in success, otherwise negative errno value.
6280 * if nb_contexts is 0, return the amount of all aged contexts.
6281 * if nb_contexts is not 0 , return the amount of aged flows reported
6282 * in the context array.
6285 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
6286 uint32_t nb_contexts, struct rte_flow_error *error)
6288 const struct mlx5_flow_driver_ops *fops;
6289 struct rte_flow_attr attr = { .transfer = 0 };
6291 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6292 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6293 return fops->get_aged_flows(dev, contexts, nb_contexts,
6297 "port %u get aged flows is not supported.",
6298 dev->data->port_id);