net/octeontx2: add flow operations
authorKiran Kumar K <kirankumark@marvell.com>
Sat, 29 Jun 2019 10:48:16 +0000 (16:18 +0530)
committerFerruh Yigit <ferruh.yigit@intel.com>
Thu, 4 Jul 2019 23:52:02 +0000 (01:52 +0200)
Adding the initial flow ops like flow_create and flow_validate.
These will be used to alloc and write flow rule to device and
validate the flow rule.

Signed-off-by: Kiran Kumar K <kirankumark@marvell.com>
Signed-off-by: Vivek Sharma <viveksharma@marvell.com>
drivers/net/octeontx2/Makefile
drivers/net/octeontx2/meson.build
drivers/net/octeontx2/otx2_flow.c [new file with mode: 0644]

index 636af65..c6a3e92 100644 (file)
@@ -35,6 +35,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_OCTEONTX2_PMD) += \
        otx2_rss.c      \
        otx2_mac.c      \
        otx2_ptp.c      \
+       otx2_flow.c     \
        otx2_link.c     \
        otx2_stats.c    \
        otx2_lookup.c   \
index f608c49..f0e03bf 100644 (file)
@@ -7,6 +7,7 @@ sources = files(
                'otx2_rss.c',
                'otx2_mac.c',
                'otx2_ptp.c',
+               'otx2_flow.c',
                'otx2_link.c',
                'otx2_stats.c',
                'otx2_lookup.c',
diff --git a/drivers/net/octeontx2/otx2_flow.c b/drivers/net/octeontx2/otx2_flow.c
new file mode 100644 (file)
index 0000000..896aef0
--- /dev/null
@@ -0,0 +1,451 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(C) 2019 Marvell International Ltd.
+ */
+
+#include "otx2_ethdev.h"
+#include "otx2_flow.h"
+
+static int
+flow_program_npc(struct otx2_parse_state *pst, struct otx2_mbox *mbox,
+                struct otx2_npc_flow_info *flow_info)
+{
+       /* This is non-LDATA part in search key */
+       uint64_t key_data[2] = {0ULL, 0ULL};
+       uint64_t key_mask[2] = {0ULL, 0ULL};
+       int intf = pst->flow->nix_intf;
+       int key_len, bit = 0, index;
+       int off, idx, data_off = 0;
+       uint8_t lid, mask, data;
+       uint16_t layer_info;
+       uint64_t lt, flags;
+
+
+       /* Skip till Layer A data start */
+       while (bit < NPC_PARSE_KEX_S_LA_OFFSET) {
+               if (flow_info->keyx_supp_nmask[intf] & (1 << bit))
+                       data_off++;
+               bit++;
+       }
+
+       /* Each bit represents 1 nibble */
+       data_off *= 4;
+
+       index = 0;
+       for (lid = 0; lid < NPC_MAX_LID; lid++) {
+               /* Offset in key */
+               off = NPC_PARSE_KEX_S_LID_OFFSET(lid);
+               lt = pst->lt[lid] & 0xf;
+               flags = pst->flags[lid] & 0xff;
+
+               /* NPC_LAYER_KEX_S */
+               layer_info = ((flow_info->keyx_supp_nmask[intf] >> off) & 0x7);
+
+               if (layer_info) {
+                       for (idx = 0; idx <= 2 ; idx++) {
+                               if (layer_info & (1 << idx)) {
+                                       if (idx == 2)
+                                               data = lt;
+                                       else if (idx == 1)
+                                               data = ((flags >> 4) & 0xf);
+                                       else
+                                               data = (flags & 0xf);
+
+                                       if (data_off >= 64) {
+                                               data_off = 0;
+                                               index++;
+                                       }
+                                       key_data[index] |= ((uint64_t)data <<
+                                                           data_off);
+                                       mask = 0xf;
+                                       if (lt == 0)
+                                               mask = 0;
+                                       key_mask[index] |= ((uint64_t)mask <<
+                                                           data_off);
+                                       data_off += 4;
+                               }
+                       }
+               }
+       }
+
+       otx2_npc_dbg("Npc prog key data0: 0x%" PRIx64 ", data1: 0x%" PRIx64,
+                    key_data[0], key_data[1]);
+
+       /* Copy this into mcam string */
+       key_len = (pst->npc->keyx_len[intf] + 7) / 8;
+       otx2_npc_dbg("Key_len  = %d", key_len);
+       memcpy(pst->flow->mcam_data, key_data, key_len);
+       memcpy(pst->flow->mcam_mask, key_mask, key_len);
+
+       otx2_npc_dbg("Final flow data");
+       for (idx = 0; idx < OTX2_MAX_MCAM_WIDTH_DWORDS; idx++) {
+               otx2_npc_dbg("data[%d]: 0x%" PRIx64 ", mask[%d]: 0x%" PRIx64,
+                            idx, pst->flow->mcam_data[idx],
+                            idx, pst->flow->mcam_mask[idx]);
+       }
+
+       /*
+        * Now we have mcam data and mask formatted as
+        * [Key_len/4 nibbles][0 or 1 nibble hole][data]
+        * hole is present if key_len is odd number of nibbles.
+        * mcam data must be split into 64 bits + 48 bits segments
+        * for each back W0, W1.
+        */
+
+       return otx2_flow_mcam_alloc_and_write(pst->flow, mbox, pst, flow_info);
+}
+
+static int
+flow_parse_attr(struct rte_eth_dev *eth_dev,
+               const struct rte_flow_attr *attr,
+               struct rte_flow_error *error,
+               struct rte_flow *flow)
+{
+       struct otx2_eth_dev *dev = eth_dev->data->dev_private;
+       const char *errmsg = NULL;
+
+       if (attr == NULL)
+               errmsg = "Attribute can't be empty";
+       else if (attr->group)
+               errmsg = "Groups are not supported";
+       else if (attr->priority >= dev->npc_flow.flow_max_priority)
+               errmsg = "Priority should be with in specified range";
+       else if ((!attr->egress && !attr->ingress) ||
+                (attr->egress && attr->ingress))
+               errmsg = "Exactly one of ingress or egress must be set";
+
+       if (errmsg != NULL) {
+               rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ATTR,
+                                  attr, errmsg);
+               return -ENOTSUP;
+       }
+
+       if (attr->ingress)
+               flow->nix_intf = OTX2_INTF_RX;
+       else
+               flow->nix_intf = OTX2_INTF_TX;
+
+       flow->priority = attr->priority;
+       return 0;
+}
+
+static inline int
+flow_get_free_rss_grp(struct rte_bitmap *bmap,
+                     uint32_t size, uint32_t *pos)
+{
+       for (*pos = 0; *pos < size; ++*pos) {
+               if (!rte_bitmap_get(bmap, *pos))
+                       break;
+       }
+
+       return *pos < size ? 0 : -1;
+}
+
+static int
+flow_configure_rss_action(struct otx2_eth_dev *dev,
+                         const struct rte_flow_action_rss *rss,
+                         uint8_t *alg_idx, uint32_t *rss_grp,
+                         int mcam_index)
+{
+       struct otx2_npc_flow_info *flow_info = &dev->npc_flow;
+       uint16_t reta[NIX_RSS_RETA_SIZE_MAX];
+       uint32_t flowkey_cfg, grp_aval, i;
+       uint16_t *ind_tbl = NULL;
+       uint8_t flowkey_algx;
+       int rc;
+
+       rc = flow_get_free_rss_grp(flow_info->rss_grp_entries,
+                                  flow_info->rss_grps, &grp_aval);
+       /* RSS group :0 is not usable for flow rss action */
+       if (rc < 0 || grp_aval == 0)
+               return -ENOSPC;
+
+       *rss_grp = grp_aval;
+
+       otx2_nix_rss_set_key(dev, (uint8_t *)(uintptr_t)rss->key,
+                            rss->key_len);
+
+       /* If queue count passed in the rss action is less than
+        * HW configured reta size, replicate rss action reta
+        * across HW reta table.
+        */
+       if (dev->rss_info.rss_size > rss->queue_num) {
+               ind_tbl = reta;
+
+               for (i = 0; i < (dev->rss_info.rss_size / rss->queue_num); i++)
+                       memcpy(reta + i * rss->queue_num, rss->queue,
+                              sizeof(uint16_t) * rss->queue_num);
+
+               i = dev->rss_info.rss_size % rss->queue_num;
+               if (i)
+                       memcpy(&reta[dev->rss_info.rss_size] - i,
+                              rss->queue, i * sizeof(uint16_t));
+       } else {
+               ind_tbl = (uint16_t *)(uintptr_t)rss->queue;
+       }
+
+       rc = otx2_nix_rss_tbl_init(dev, *rss_grp, ind_tbl);
+       if (rc) {
+               otx2_err("Failed to init rss table rc = %d", rc);
+               return rc;
+       }
+
+       flowkey_cfg = otx2_rss_ethdev_to_nix(dev, rss->types, rss->level);
+
+       rc = otx2_rss_set_hf(dev, flowkey_cfg, &flowkey_algx,
+                            *rss_grp, mcam_index);
+       if (rc) {
+               otx2_err("Failed to set rss hash function rc = %d", rc);
+               return rc;
+       }
+
+       *alg_idx = flowkey_algx;
+
+       rte_bitmap_set(flow_info->rss_grp_entries, *rss_grp);
+
+       return 0;
+}
+
+
+static int
+flow_program_rss_action(struct rte_eth_dev *eth_dev,
+                       const struct rte_flow_action actions[],
+                       struct rte_flow *flow)
+{
+       struct otx2_eth_dev *dev = eth_dev->data->dev_private;
+       const struct rte_flow_action_rss *rss;
+       uint32_t rss_grp;
+       uint8_t alg_idx;
+       int rc;
+
+       for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
+               if (actions->type == RTE_FLOW_ACTION_TYPE_RSS) {
+                       rss = (const struct rte_flow_action_rss *)actions->conf;
+
+                       rc = flow_configure_rss_action(dev,
+                                                      rss, &alg_idx, &rss_grp,
+                                                      flow->mcam_id);
+                       if (rc)
+                               return rc;
+
+                       flow->npc_action |=
+                               ((uint64_t)(alg_idx & NIX_RSS_ACT_ALG_MASK) <<
+                                NIX_RSS_ACT_ALG_OFFSET) |
+                               ((uint64_t)(rss_grp & NIX_RSS_ACT_GRP_MASK) <<
+                                NIX_RSS_ACT_GRP_OFFSET);
+               }
+       }
+       return 0;
+}
+
+static int
+flow_parse_meta_items(__rte_unused struct otx2_parse_state *pst)
+{
+       otx2_npc_dbg("Meta Item");
+       return 0;
+}
+
+/*
+ * Parse function of each layer:
+ *  - Consume one or more patterns that are relevant.
+ *  - Update parse_state
+ *  - Set parse_state.pattern = last item consumed
+ *  - Set appropriate error code/message when returning error.
+ */
+typedef int (*flow_parse_stage_func_t)(struct otx2_parse_state *pst);
+
+static int
+flow_parse_pattern(struct rte_eth_dev *dev,
+                  const struct rte_flow_item pattern[],
+                  struct rte_flow_error *error,
+                  struct rte_flow *flow,
+                  struct otx2_parse_state *pst)
+{
+       flow_parse_stage_func_t parse_stage_funcs[] = {
+               flow_parse_meta_items,
+               otx2_flow_parse_la,
+               otx2_flow_parse_lb,
+               otx2_flow_parse_lc,
+               otx2_flow_parse_ld,
+               otx2_flow_parse_le,
+               otx2_flow_parse_lf,
+               otx2_flow_parse_lg,
+               otx2_flow_parse_lh,
+       };
+       struct otx2_eth_dev *hw = dev->data->dev_private;
+       uint8_t layer = 0;
+       int key_offset;
+       int rc;
+
+       if (pattern == NULL) {
+               rte_flow_error_set(error, EINVAL,
+                                  RTE_FLOW_ERROR_TYPE_ITEM_NUM, NULL,
+                                  "pattern is NULL");
+               return -EINVAL;
+       }
+
+       memset(pst, 0, sizeof(*pst));
+       pst->npc = &hw->npc_flow;
+       pst->error = error;
+       pst->flow = flow;
+
+       /* Use integral byte offset */
+       key_offset = pst->npc->keyx_len[flow->nix_intf];
+       key_offset = (key_offset + 7) / 8;
+
+       /* Location where LDATA would begin */
+       pst->mcam_data = (uint8_t *)flow->mcam_data;
+       pst->mcam_mask = (uint8_t *)flow->mcam_mask;
+
+       while (pattern->type != RTE_FLOW_ITEM_TYPE_END &&
+              layer < RTE_DIM(parse_stage_funcs)) {
+               otx2_npc_dbg("Pattern type = %d", pattern->type);
+
+               /* Skip place-holders */
+               pattern = otx2_flow_skip_void_and_any_items(pattern);
+
+               pst->pattern = pattern;
+               otx2_npc_dbg("Is tunnel = %d, layer = %d", pst->tunnel, layer);
+               rc = parse_stage_funcs[layer](pst);
+               if (rc != 0)
+                       return -rte_errno;
+
+               layer++;
+
+               /*
+                * Parse stage function sets pst->pattern to
+                * 1 past the last item it consumed.
+                */
+               pattern = pst->pattern;
+
+               if (pst->terminate)
+                       break;
+       }
+
+       /* Skip trailing place-holders */
+       pattern = otx2_flow_skip_void_and_any_items(pattern);
+
+       /* Are there more items than what we can handle? */
+       if (pattern->type != RTE_FLOW_ITEM_TYPE_END) {
+               rte_flow_error_set(error, ENOTSUP,
+                                  RTE_FLOW_ERROR_TYPE_ITEM, pattern,
+                                  "unsupported item in the sequence");
+               return -ENOTSUP;
+       }
+
+       return 0;
+}
+
+static int
+flow_parse_rule(struct rte_eth_dev *dev,
+               const struct rte_flow_attr *attr,
+               const struct rte_flow_item pattern[],
+               const struct rte_flow_action actions[],
+               struct rte_flow_error *error,
+               struct rte_flow *flow,
+               struct otx2_parse_state *pst)
+{
+       int err;
+
+       /* Check attributes */
+       err = flow_parse_attr(dev, attr, error, flow);
+       if (err)
+               return err;
+
+       /* Check actions */
+       err = otx2_flow_parse_actions(dev, attr, actions, error, flow);
+       if (err)
+               return err;
+
+       /* Check pattern */
+       err = flow_parse_pattern(dev, pattern, error, flow, pst);
+       if (err)
+               return err;
+
+       /* Check for overlaps? */
+       return 0;
+}
+
+static int
+otx2_flow_validate(struct rte_eth_dev *dev,
+                  const struct rte_flow_attr *attr,
+                  const struct rte_flow_item pattern[],
+                  const struct rte_flow_action actions[],
+                  struct rte_flow_error *error)
+{
+       struct otx2_parse_state parse_state;
+       struct rte_flow flow;
+
+       memset(&flow, 0, sizeof(flow));
+       return flow_parse_rule(dev, attr, pattern, actions, error, &flow,
+                              &parse_state);
+}
+
+static struct rte_flow *
+otx2_flow_create(struct rte_eth_dev *dev,
+                const struct rte_flow_attr *attr,
+                const struct rte_flow_item pattern[],
+                const struct rte_flow_action actions[],
+                struct rte_flow_error *error)
+{
+       struct otx2_eth_dev *hw = dev->data->dev_private;
+       struct otx2_parse_state parse_state;
+       struct otx2_mbox *mbox = hw->mbox;
+       struct rte_flow *flow, *flow_iter;
+       struct otx2_flow_list *list;
+       int rc;
+
+       flow = rte_zmalloc("otx2_rte_flow", sizeof(*flow), 0);
+       if (flow == NULL) {
+               rte_flow_error_set(error, ENOMEM,
+                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                                  NULL,
+                                  "Memory allocation failed");
+               return NULL;
+       }
+       memset(flow, 0, sizeof(*flow));
+
+       rc = flow_parse_rule(dev, attr, pattern, actions, error, flow,
+                            &parse_state);
+       if (rc != 0)
+               goto err_exit;
+
+       rc = flow_program_npc(&parse_state, mbox, &hw->npc_flow);
+       if (rc != 0) {
+               rte_flow_error_set(error, EIO,
+                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                                  NULL,
+                                  "Failed to insert filter");
+               goto err_exit;
+       }
+
+       rc = flow_program_rss_action(dev, actions, flow);
+       if (rc != 0) {
+               rte_flow_error_set(error, EIO,
+                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                                  NULL,
+                                  "Failed to program rss action");
+               goto err_exit;
+       }
+
+
+       list = &hw->npc_flow.flow_list[flow->priority];
+       /* List in ascending order of mcam entries */
+       TAILQ_FOREACH(flow_iter, list, next) {
+               if (flow_iter->mcam_id > flow->mcam_id) {
+                       TAILQ_INSERT_BEFORE(flow_iter, flow, next);
+                       return flow;
+               }
+       }
+
+       TAILQ_INSERT_TAIL(list, flow, next);
+       return flow;
+
+err_exit:
+       rte_free(flow);
+       return NULL;
+}
+
+const struct rte_flow_ops otx2_flow_ops = {
+       .validate = otx2_flow_validate,
+       .create = otx2_flow_create,
+};