net/mlx5: support match on IPv4 TTL and IPv6 HLIM
[dpdk.git] / drivers / net / octeontx2 / otx2_flow_utils.c
index 6078a82..14625c9 100644 (file)
@@ -5,6 +5,22 @@
 #include "otx2_ethdev.h"
 #include "otx2_flow.h"
 
+static int
+flow_mcam_alloc_counter(struct otx2_mbox *mbox, uint16_t *ctr)
+{
+       struct npc_mcam_alloc_counter_req *req;
+       struct npc_mcam_alloc_counter_rsp *rsp;
+       int rc;
+
+       req = otx2_mbox_alloc_msg_npc_mcam_alloc_counter(mbox);
+       req->count = 1;
+       otx2_mbox_msg_send(mbox, 0);
+       rc = otx2_mbox_get_rsp(mbox, 0, (void *)&rsp);
+
+       *ctr = rsp->cntr_list[0];
+       return rc;
+}
+
 int
 otx2_flow_mcam_free_counter(struct otx2_mbox *mbox, uint16_t ctr_id)
 {
@@ -108,13 +124,37 @@ flow_mem_is_zero(const void *mem, int len)
        return 1;
 }
 
+static void
+flow_set_hw_mask(struct otx2_flow_item_info *info,
+                struct npc_xtract_info *xinfo,
+                char *hw_mask)
+{
+       int max_off, offset;
+       int j;
+
+       if (xinfo->enable == 0)
+               return;
+
+       if (xinfo->hdr_off < info->hw_hdr_len)
+               return;
+
+       max_off = xinfo->hdr_off + xinfo->len - info->hw_hdr_len;
+
+       if (max_off > info->len)
+               max_off = info->len;
+
+       offset = xinfo->hdr_off - info->hw_hdr_len;
+       for (j = offset; j < max_off; j++)
+               hw_mask[j] = 0xff;
+}
+
 void
 otx2_flow_get_hw_supp_mask(struct otx2_parse_state *pst,
                           struct otx2_flow_item_info *info, int lid, int lt)
 {
-       struct npc_xtract_info *xinfo;
+       struct npc_xtract_info *xinfo, *lfinfo;
        char *hw_mask = info->hw_mask;
-       int max_off, offset;
+       int lf_cfg;
        int i, j;
        int intf;
 
@@ -123,21 +163,106 @@ otx2_flow_get_hw_supp_mask(struct otx2_parse_state *pst,
        memset(hw_mask, 0, info->len);
 
        for (i = 0; i < NPC_MAX_LD; i++) {
-               if (xinfo[i].hdr_off < info->hw_hdr_len)
-                       continue;
+               flow_set_hw_mask(info, &xinfo[i], hw_mask);
+       }
 
-               max_off = xinfo[i].hdr_off + xinfo[i].len - info->hw_hdr_len;
+       for (i = 0; i < NPC_MAX_LD; i++) {
 
-               if (xinfo[i].enable == 0)
+               if (xinfo[i].flags_enable == 0)
                        continue;
 
-               if (max_off > info->len)
-                       max_off = info->len;
+               lf_cfg = pst->npc->prx_lfcfg[i].i;
+               if (lf_cfg == lid) {
+                       for (j = 0; j < NPC_MAX_LFL; j++) {
+                               lfinfo = pst->npc->prx_fxcfg[intf]
+                                       [i][j].xtract;
+                               flow_set_hw_mask(info, &lfinfo[0], hw_mask);
+                       }
+               }
+       }
+}
+
+static int
+flow_update_extraction_data(struct otx2_parse_state *pst,
+                           struct otx2_flow_item_info *info,
+                           struct npc_xtract_info *xinfo)
+{
+       uint8_t int_info_mask[NPC_MAX_EXTRACT_DATA_LEN];
+       uint8_t int_info[NPC_MAX_EXTRACT_DATA_LEN];
+       struct npc_xtract_info *x;
+       int k, idx, hdr_off;
+       int len = 0;
+
+       x = xinfo;
+       len = x->len;
+       hdr_off = x->hdr_off;
+
+       if (hdr_off < info->hw_hdr_len)
+               return 0;
+
+       if (x->enable == 0)
+               return 0;
+
+       otx2_npc_dbg("x->hdr_off = %d, len = %d, info->len = %d,"
+                    "x->key_off = %d", x->hdr_off, len, info->len,
+                    x->key_off);
+
+       hdr_off -= info->hw_hdr_len;
 
-               offset = xinfo[i].hdr_off - info->hw_hdr_len;
-               for (j = offset; j < max_off; j++)
-                       hw_mask[j] = 0xff;
+       if (hdr_off + len > info->len)
+               len = info->len - hdr_off;
+
+       /* Check for over-write of previous layer */
+       if (!flow_mem_is_zero(pst->mcam_mask + x->key_off,
+                             len)) {
+               /* Cannot support this data match */
+               rte_flow_error_set(pst->error, ENOTSUP,
+                                  RTE_FLOW_ERROR_TYPE_ITEM,
+                                  pst->pattern,
+                                  "Extraction unsupported");
+               return -rte_errno;
+       }
+
+       len = flow_check_copysz((OTX2_MAX_MCAM_WIDTH_DWORDS * 8)
+                               - x->key_off,
+                               len);
+       if (len < 0) {
+               rte_flow_error_set(pst->error, ENOTSUP,
+                                  RTE_FLOW_ERROR_TYPE_ITEM,
+                                  pst->pattern,
+                                  "Internal Error");
+               return -rte_errno;
        }
+
+       /* Need to reverse complete structure so that dest addr is at
+        * MSB so as to program the MCAM using mcam_data & mcam_mask
+        * arrays
+        */
+       flow_prep_mcam_ldata(int_info,
+                            (const uint8_t *)info->spec + hdr_off,
+                            x->len);
+       flow_prep_mcam_ldata(int_info_mask,
+                            (const uint8_t *)info->mask + hdr_off,
+                            x->len);
+
+       otx2_npc_dbg("Spec: ");
+       for (k = 0; k < info->len; k++)
+               otx2_npc_dbg("0x%.2x ",
+                            ((const uint8_t *)info->spec)[k]);
+
+       otx2_npc_dbg("Int_info: ");
+       for (k = 0; k < info->len; k++)
+               otx2_npc_dbg("0x%.2x ", int_info[k]);
+
+       memcpy(pst->mcam_mask + x->key_off, int_info_mask, len);
+       memcpy(pst->mcam_data + x->key_off, int_info, len);
+
+       otx2_npc_dbg("Parse state mcam data & mask");
+       for (idx = 0; idx < len ; idx++)
+               otx2_npc_dbg("data[%d]: 0x%x, mask[%d]: 0x%x", idx,
+                            *(pst->mcam_data + idx + x->key_off), idx,
+                            *(pst->mcam_mask + idx + x->key_off));
+       return 0;
 }
 
 int
@@ -145,12 +270,10 @@ otx2_flow_update_parse_state(struct otx2_parse_state *pst,
                             struct otx2_flow_item_info *info, int lid, int lt,
                             uint8_t flags)
 {
-       uint8_t int_info_mask[NPC_MAX_EXTRACT_DATA_LEN];
-       uint8_t int_info[NPC_MAX_EXTRACT_DATA_LEN];
        struct npc_lid_lt_xtract_info *xinfo;
-       int len = 0;
-       int intf;
-       int i;
+       struct npc_xtract_info *lfinfo;
+       int intf, lf_cfg;
+       int i, j, rc = 0;
 
        otx2_npc_dbg("Parse state function info mask total %s",
                     (const uint8_t *)info->mask);
@@ -165,93 +288,35 @@ otx2_flow_update_parse_state(struct otx2_parse_state *pst,
        if (xinfo->is_terminating)
                pst->terminate = 1;
 
-       /* Need to check if flags are supported but in latest
-        * KPU profile, flags are used as enumeration! No way,
-        * it can be validated unless MBOX is changed to return
-        * set of valid values out of 2**8 possible values.
-        */
-       if (info->spec == NULL) {       /* Nothing to match */
+       if (info->spec == NULL) {
                otx2_npc_dbg("Info spec NULL");
                goto done;
        }
 
-       /* Copy spec and mask into mcam match string, mask.
-        * Since both RTE FLOW and OTX2 MCAM use network-endianness
-        * for data, we are saved from nasty conversions.
-        */
        for (i = 0; i < NPC_MAX_LD; i++) {
-               struct npc_xtract_info *x;
-               int k, idx, hdr_off;
-
-               x = &xinfo->xtract[i];
-               len = x->len;
-               hdr_off = x->hdr_off;
-
-               if (hdr_off < info->hw_hdr_len)
-                       continue;
+               rc = flow_update_extraction_data(pst, info, &xinfo->xtract[i]);
+               if (rc != 0)
+                       return rc;
+       }
 
-               if (x->enable == 0)
+       for (i = 0; i < NPC_MAX_LD; i++) {
+               if (xinfo->xtract[i].flags_enable == 0)
                        continue;
 
-               otx2_npc_dbg("x->hdr_off = %d, len = %d, info->len = %d,"
-                             "x->key_off = %d", x->hdr_off, len, info->len,
-                             x->key_off);
-
-               hdr_off -= info->hw_hdr_len;
-
-               if (hdr_off + len > info->len)
-                       len = info->len - hdr_off;
-
-               /* Check for over-write of previous layer */
-               if (!flow_mem_is_zero(pst->mcam_mask + x->key_off,
-                                     len)) {
-                       /* Cannot support this data match */
-                       rte_flow_error_set(pst->error, ENOTSUP,
-                                          RTE_FLOW_ERROR_TYPE_ITEM,
-                                          pst->pattern,
-                                          "Extraction unsupported");
-                       return -rte_errno;
+               lf_cfg = pst->npc->prx_lfcfg[i].i;
+               if (lf_cfg == lid) {
+                       for (j = 0; j < NPC_MAX_LFL; j++) {
+                               lfinfo = pst->npc->prx_fxcfg[intf]
+                                       [i][j].xtract;
+                               rc = flow_update_extraction_data(pst, info,
+                                                                &lfinfo[0]);
+                               if (rc != 0)
+                                       return rc;
+
+                               if (lfinfo[0].enable)
+                                       pst->flags[lid] = j;
+                       }
                }
-
-               len = flow_check_copysz((OTX2_MAX_MCAM_WIDTH_DWORDS * 8)
-                                       - x->key_off,
-                                       len);
-               if (len < 0) {
-                       rte_flow_error_set(pst->error, ENOTSUP,
-                                          RTE_FLOW_ERROR_TYPE_ITEM,
-                                          pst->pattern,
-                                          "Internal Error");
-                       return -rte_errno;
-               }
-
-               /* Need to reverse complete structure so that dest addr is at
-                * MSB so as to program the MCAM using mcam_data & mcam_mask
-                * arrays
-                */
-               flow_prep_mcam_ldata(int_info,
-                                    (const uint8_t *)info->spec + hdr_off,
-                                    x->len);
-               flow_prep_mcam_ldata(int_info_mask,
-                                    (const uint8_t *)info->mask + hdr_off,
-                                    x->len);
-
-               otx2_npc_dbg("Spec: ");
-               for (k = 0; k < info->len; k++)
-                       otx2_npc_dbg("0x%.2x ",
-                                    ((const uint8_t *)info->spec)[k]);
-
-               otx2_npc_dbg("Int_info: ");
-               for (k = 0; k < info->len; k++)
-                       otx2_npc_dbg("0x%.2x ", int_info[k]);
-
-               memcpy(pst->mcam_mask + x->key_off, int_info_mask, len);
-               memcpy(pst->mcam_data + x->key_off, int_info, len);
-
-               otx2_npc_dbg("Parse state mcam data & mask");
-               for (idx = 0; idx < len ; idx++)
-                       otx2_npc_dbg("data[%d]: 0x%x, mask[%d]: 0x%x", idx,
-                                    *(pst->mcam_data + idx + x->key_off), idx,
-                                    *(pst->mcam_mask + idx + x->key_off));
        }
 
 done:
@@ -385,3 +450,510 @@ otx2_flow_keyx_compress(uint64_t *data, uint32_t nibble_mask)
        data[1] = cdata[1];
 }
 
+static int
+flow_first_set_bit(uint64_t slab)
+{
+       int num = 0;
+
+       if ((slab & 0xffffffff) == 0) {
+               num += 32;
+               slab >>= 32;
+       }
+       if ((slab & 0xffff) == 0) {
+               num += 16;
+               slab >>= 16;
+       }
+       if ((slab & 0xff) == 0) {
+               num += 8;
+               slab >>= 8;
+       }
+       if ((slab & 0xf) == 0) {
+               num += 4;
+               slab >>= 4;
+       }
+       if ((slab & 0x3) == 0) {
+               num += 2;
+               slab >>= 2;
+       }
+       if ((slab & 0x1) == 0)
+               num += 1;
+
+       return num;
+}
+
+static int
+flow_shift_lv_ent(struct otx2_mbox *mbox, struct rte_flow *flow,
+                 struct otx2_npc_flow_info *flow_info,
+                 uint32_t old_ent, uint32_t new_ent)
+{
+       struct npc_mcam_shift_entry_req *req;
+       struct npc_mcam_shift_entry_rsp *rsp;
+       struct otx2_flow_list *list;
+       struct rte_flow *flow_iter;
+       int rc = 0;
+
+       otx2_npc_dbg("Old ent:%u new ent:%u priority:%u", old_ent, new_ent,
+                    flow->priority);
+
+       list = &flow_info->flow_list[flow->priority];
+
+       /* Old entry is disabled & it's contents are moved to new_entry,
+        * new entry is enabled finally.
+        */
+       req = otx2_mbox_alloc_msg_npc_mcam_shift_entry(mbox);
+       req->curr_entry[0] = old_ent;
+       req->new_entry[0] = new_ent;
+       req->shift_count = 1;
+
+       otx2_mbox_msg_send(mbox, 0);
+       rc = otx2_mbox_get_rsp(mbox, 0, (void *)&rsp);
+       if (rc)
+               return rc;
+
+       /* Remove old node from list */
+       TAILQ_FOREACH(flow_iter, list, next) {
+               if (flow_iter->mcam_id == old_ent)
+                       TAILQ_REMOVE(list, flow_iter, next);
+       }
+
+       /* Insert node with new mcam id at right place */
+       TAILQ_FOREACH(flow_iter, list, next) {
+               if (flow_iter->mcam_id > new_ent)
+                       TAILQ_INSERT_BEFORE(flow_iter, flow, next);
+       }
+       return rc;
+}
+
+/* Exchange all required entries with a given priority level */
+static int
+flow_shift_ent(struct otx2_mbox *mbox, struct rte_flow *flow,
+              struct otx2_npc_flow_info *flow_info,
+              struct npc_mcam_alloc_entry_rsp *rsp, int dir, int prio_lvl)
+{
+       struct rte_bitmap *fr_bmp, *fr_bmp_rev, *lv_bmp, *lv_bmp_rev, *bmp;
+       uint32_t e_fr = 0, e_lv = 0, e, e_id = 0, mcam_entries;
+       uint64_t fr_bit_pos = 0, lv_bit_pos = 0, bit_pos = 0;
+       /* Bit position within the slab */
+       uint32_t sl_fr_bit_off = 0, sl_lv_bit_off = 0;
+       /* Overall bit position of the start of slab */
+       /* free & live entry index */
+       int rc_fr = 0, rc_lv = 0, rc = 0, idx = 0;
+       struct otx2_mcam_ents_info *ent_info;
+       /* free & live bitmap slab */
+       uint64_t sl_fr = 0, sl_lv = 0, *sl;
+
+       fr_bmp = flow_info->free_entries[prio_lvl];
+       fr_bmp_rev = flow_info->free_entries_rev[prio_lvl];
+       lv_bmp = flow_info->live_entries[prio_lvl];
+       lv_bmp_rev = flow_info->live_entries_rev[prio_lvl];
+       ent_info = &flow_info->flow_entry_info[prio_lvl];
+       mcam_entries = flow_info->mcam_entries;
+
+
+       /* New entries allocated are always contiguous, but older entries
+        * already in free/live bitmap can be non-contiguous: so return
+        * shifted entries should be in non-contiguous format.
+        */
+       while (idx <= rsp->count) {
+               if (!sl_fr && !sl_lv) {
+                       /* Lower index elements to be exchanged */
+                       if (dir < 0) {
+                               rc_fr = rte_bitmap_scan(fr_bmp, &e_fr, &sl_fr);
+                               rc_lv = rte_bitmap_scan(lv_bmp, &e_lv, &sl_lv);
+                               otx2_npc_dbg("Fwd slab rc fr %u rc lv %u "
+                                            "e_fr %u e_lv %u", rc_fr, rc_lv,
+                                             e_fr, e_lv);
+                       } else {
+                               rc_fr = rte_bitmap_scan(fr_bmp_rev,
+                                                       &sl_fr_bit_off,
+                                                       &sl_fr);
+                               rc_lv = rte_bitmap_scan(lv_bmp_rev,
+                                                       &sl_lv_bit_off,
+                                                       &sl_lv);
+
+                               otx2_npc_dbg("Rev slab rc fr %u rc lv %u "
+                                            "e_fr %u e_lv %u", rc_fr, rc_lv,
+                                             e_fr, e_lv);
+                       }
+               }
+
+               if (rc_fr) {
+                       fr_bit_pos = flow_first_set_bit(sl_fr);
+                       e_fr = sl_fr_bit_off + fr_bit_pos;
+                       otx2_npc_dbg("Fr_bit_pos 0x%" PRIx64, fr_bit_pos);
+               } else {
+                       e_fr = ~(0);
+               }
+
+               if (rc_lv) {
+                       lv_bit_pos = flow_first_set_bit(sl_lv);
+                       e_lv = sl_lv_bit_off + lv_bit_pos;
+                       otx2_npc_dbg("Lv_bit_pos 0x%" PRIx64, lv_bit_pos);
+               } else {
+                       e_lv = ~(0);
+               }
+
+               /* First entry is from free_bmap */
+               if (e_fr < e_lv) {
+                       bmp = fr_bmp;
+                       e = e_fr;
+                       sl = &sl_fr;
+                       bit_pos = fr_bit_pos;
+                       if (dir > 0)
+                               e_id = mcam_entries - e - 1;
+                       else
+                               e_id = e;
+                       otx2_npc_dbg("Fr e %u e_id %u", e, e_id);
+               } else {
+                       bmp = lv_bmp;
+                       e = e_lv;
+                       sl = &sl_lv;
+                       bit_pos = lv_bit_pos;
+                       if (dir > 0)
+                               e_id = mcam_entries - e - 1;
+                       else
+                               e_id = e;
+
+                       otx2_npc_dbg("Lv e %u e_id %u", e, e_id);
+                       if (idx < rsp->count)
+                               rc =
+                                 flow_shift_lv_ent(mbox, flow,
+                                                   flow_info, e_id,
+                                                   rsp->entry + idx);
+               }
+
+               rte_bitmap_clear(bmp, e);
+               rte_bitmap_set(bmp, rsp->entry + idx);
+               /* Update entry list, use non-contiguous
+                * list now.
+                */
+               rsp->entry_list[idx] = e_id;
+               *sl &= ~(1 << bit_pos);
+
+               /* Update min & max entry identifiers in current
+                * priority level.
+                */
+               if (dir < 0) {
+                       ent_info->max_id = rsp->entry + idx;
+                       ent_info->min_id = e_id;
+               } else {
+                       ent_info->max_id = e_id;
+                       ent_info->min_id = rsp->entry;
+               }
+
+               idx++;
+       }
+       return rc;
+}
+
+/* Validate if newly allocated entries lie in the correct priority zone
+ * since NPC_MCAM_LOWER_PRIO & NPC_MCAM_HIGHER_PRIO don't ensure zone accuracy.
+ * If not properly aligned, shift entries to do so
+ */
+static int
+flow_validate_and_shift_prio_ent(struct otx2_mbox *mbox, struct rte_flow *flow,
+                                struct otx2_npc_flow_info *flow_info,
+                                struct npc_mcam_alloc_entry_rsp *rsp,
+                                int req_prio)
+{
+       int prio_idx = 0, rc = 0, needs_shift = 0, idx, prio = flow->priority;
+       struct otx2_mcam_ents_info *info = flow_info->flow_entry_info;
+       int dir = (req_prio == NPC_MCAM_HIGHER_PRIO) ? 1 : -1;
+       uint32_t tot_ent = 0;
+
+       otx2_npc_dbg("Dir %d, priority = %d", dir, prio);
+
+       if (dir < 0)
+               prio_idx = flow_info->flow_max_priority - 1;
+
+       /* Only live entries needs to be shifted, free entries can just be
+        * moved by bits manipulation.
+        */
+
+       /* For dir = -1(NPC_MCAM_LOWER_PRIO), when shifting,
+        * NPC_MAX_PREALLOC_ENT are exchanged with adjoining higher priority
+        * level entries(lower indexes).
+        *
+        * For dir = +1(NPC_MCAM_HIGHER_PRIO), during shift,
+        * NPC_MAX_PREALLOC_ENT are exchanged with adjoining lower priority
+        * level entries(higher indexes) with highest indexes.
+        */
+       do {
+               tot_ent = info[prio_idx].free_ent + info[prio_idx].live_ent;
+
+               if (dir < 0 && prio_idx != prio &&
+                   rsp->entry > info[prio_idx].max_id && tot_ent) {
+                       otx2_npc_dbg("Rsp entry %u prio idx %u "
+                                    "max id %u", rsp->entry, prio_idx,
+                                     info[prio_idx].max_id);
+
+                       needs_shift = 1;
+               } else if ((dir > 0) && (prio_idx != prio) &&
+                    (rsp->entry < info[prio_idx].min_id) && tot_ent) {
+                       otx2_npc_dbg("Rsp entry %u prio idx %u "
+                                    "min id %u", rsp->entry, prio_idx,
+                                     info[prio_idx].min_id);
+                       needs_shift = 1;
+               }
+
+               otx2_npc_dbg("Needs_shift = %d", needs_shift);
+               if (needs_shift) {
+                       needs_shift = 0;
+                       rc = flow_shift_ent(mbox, flow, flow_info, rsp, dir,
+                                           prio_idx);
+               } else {
+                       for (idx = 0; idx < rsp->count; idx++)
+                               rsp->entry_list[idx] = rsp->entry + idx;
+               }
+       } while ((prio_idx != prio) && (prio_idx += dir));
+
+       return rc;
+}
+
+static int
+flow_find_ref_entry(struct otx2_npc_flow_info *flow_info, int *prio,
+                   int prio_lvl)
+{
+       struct otx2_mcam_ents_info *info = flow_info->flow_entry_info;
+       int step = 1;
+
+       while (step < flow_info->flow_max_priority) {
+               if (((prio_lvl + step) < flow_info->flow_max_priority) &&
+                   info[prio_lvl + step].live_ent) {
+                       *prio = NPC_MCAM_HIGHER_PRIO;
+                       return info[prio_lvl + step].min_id;
+               }
+
+               if (((prio_lvl - step) >= 0) &&
+                   info[prio_lvl - step].live_ent) {
+                       otx2_npc_dbg("Prio_lvl %u live %u", prio_lvl - step,
+                                    info[prio_lvl - step].live_ent);
+                       *prio = NPC_MCAM_LOWER_PRIO;
+                       return info[prio_lvl - step].max_id;
+               }
+               step++;
+       }
+       *prio = NPC_MCAM_ANY_PRIO;
+       return 0;
+}
+
+static int
+flow_fill_entry_cache(struct otx2_mbox *mbox, struct rte_flow *flow,
+                     struct otx2_npc_flow_info *flow_info, uint32_t *free_ent)
+{
+       struct rte_bitmap *free_bmp, *free_bmp_rev, *live_bmp, *live_bmp_rev;
+       struct npc_mcam_alloc_entry_rsp rsp_local;
+       struct npc_mcam_alloc_entry_rsp *rsp_cmd;
+       struct npc_mcam_alloc_entry_req *req;
+       struct npc_mcam_alloc_entry_rsp *rsp;
+       struct otx2_mcam_ents_info *info;
+       uint16_t ref_ent, idx;
+       int rc, prio;
+
+       info = &flow_info->flow_entry_info[flow->priority];
+       free_bmp = flow_info->free_entries[flow->priority];
+       free_bmp_rev = flow_info->free_entries_rev[flow->priority];
+       live_bmp = flow_info->live_entries[flow->priority];
+       live_bmp_rev = flow_info->live_entries_rev[flow->priority];
+
+       ref_ent = flow_find_ref_entry(flow_info, &prio, flow->priority);
+
+       req = otx2_mbox_alloc_msg_npc_mcam_alloc_entry(mbox);
+       req->contig = 1;
+       req->count = flow_info->flow_prealloc_size;
+       req->priority = prio;
+       req->ref_entry = ref_ent;
+
+       otx2_npc_dbg("Fill cache ref entry %u prio %u", ref_ent, prio);
+
+       otx2_mbox_msg_send(mbox, 0);
+       rc = otx2_mbox_get_rsp(mbox, 0, (void *)&rsp_cmd);
+       if (rc)
+               return rc;
+
+       rsp = &rsp_local;
+       memcpy(rsp, rsp_cmd, sizeof(*rsp));
+
+       otx2_npc_dbg("Alloc entry %u count %u , prio = %d", rsp->entry,
+                    rsp->count, prio);
+
+       /* Non-first ent cache fill */
+       if (prio != NPC_MCAM_ANY_PRIO) {
+               flow_validate_and_shift_prio_ent(mbox, flow, flow_info, rsp,
+                                                prio);
+       } else {
+               /* Copy into response entry list */
+               for (idx = 0; idx < rsp->count; idx++)
+                       rsp->entry_list[idx] = rsp->entry + idx;
+       }
+
+       otx2_npc_dbg("Fill entry cache rsp count %u", rsp->count);
+       /* Update free entries, reverse free entries list,
+        * min & max entry ids.
+        */
+       for (idx = 0; idx < rsp->count; idx++) {
+               if (unlikely(rsp->entry_list[idx] < info->min_id))
+                       info->min_id = rsp->entry_list[idx];
+
+               if (unlikely(rsp->entry_list[idx] > info->max_id))
+                       info->max_id = rsp->entry_list[idx];
+
+               /* Skip entry to be returned, not to be part of free
+                * list.
+                */
+               if (prio == NPC_MCAM_HIGHER_PRIO) {
+                       if (unlikely(idx == (rsp->count - 1))) {
+                               *free_ent = rsp->entry_list[idx];
+                               continue;
+                       }
+               } else {
+                       if (unlikely(!idx)) {
+                               *free_ent = rsp->entry_list[idx];
+                               continue;
+                       }
+               }
+               info->free_ent++;
+               rte_bitmap_set(free_bmp, rsp->entry_list[idx]);
+               rte_bitmap_set(free_bmp_rev, flow_info->mcam_entries -
+                              rsp->entry_list[idx] - 1);
+
+               otx2_npc_dbg("Final rsp entry %u rsp entry rev %u",
+                            rsp->entry_list[idx],
+               flow_info->mcam_entries - rsp->entry_list[idx] - 1);
+       }
+
+       otx2_npc_dbg("Cache free entry %u, rev = %u", *free_ent,
+                    flow_info->mcam_entries - *free_ent - 1);
+       info->live_ent++;
+       rte_bitmap_set(live_bmp, *free_ent);
+       rte_bitmap_set(live_bmp_rev, flow_info->mcam_entries - *free_ent - 1);
+
+       return 0;
+}
+
+static int
+flow_check_preallocated_entry_cache(struct otx2_mbox *mbox,
+                                   struct rte_flow *flow,
+                                   struct otx2_npc_flow_info *flow_info)
+{
+       struct rte_bitmap *free, *free_rev, *live, *live_rev;
+       uint32_t pos = 0, free_ent = 0, mcam_entries;
+       struct otx2_mcam_ents_info *info;
+       uint64_t slab = 0;
+       int rc;
+
+       otx2_npc_dbg("Flow priority %u", flow->priority);
+
+       info = &flow_info->flow_entry_info[flow->priority];
+
+       free_rev = flow_info->free_entries_rev[flow->priority];
+       free = flow_info->free_entries[flow->priority];
+       live_rev = flow_info->live_entries_rev[flow->priority];
+       live = flow_info->live_entries[flow->priority];
+       mcam_entries = flow_info->mcam_entries;
+
+       if (info->free_ent) {
+               rc = rte_bitmap_scan(free, &pos, &slab);
+               if (rc) {
+                       /* Get free_ent from free entry bitmap */
+                       free_ent = pos + __builtin_ctzll(slab);
+                       otx2_npc_dbg("Allocated from cache entry %u", free_ent);
+                       /* Remove from free bitmaps and add to live ones */
+                       rte_bitmap_clear(free, free_ent);
+                       rte_bitmap_set(live, free_ent);
+                       rte_bitmap_clear(free_rev,
+                                        mcam_entries - free_ent - 1);
+                       rte_bitmap_set(live_rev,
+                                      mcam_entries - free_ent - 1);
+
+                       info->free_ent--;
+                       info->live_ent++;
+                       return free_ent;
+               }
+
+               otx2_npc_dbg("No free entry:its a mess");
+               return -1;
+       }
+
+       rc = flow_fill_entry_cache(mbox, flow, flow_info, &free_ent);
+       if (rc)
+               return rc;
+
+       return free_ent;
+}
+
+int
+otx2_flow_mcam_alloc_and_write(struct rte_flow *flow, struct otx2_mbox *mbox,
+                              __rte_unused struct otx2_parse_state *pst,
+                              struct otx2_npc_flow_info *flow_info)
+{
+       int use_ctr = (flow->ctr_id == NPC_COUNTER_NONE ? 0 : 1);
+       struct npc_mcam_write_entry_req *req;
+       struct mbox_msghdr *rsp;
+       uint16_t ctr = ~(0);
+       int rc, idx;
+       int entry;
+
+       if (use_ctr) {
+               rc = flow_mcam_alloc_counter(mbox, &ctr);
+               if (rc)
+                       return rc;
+       }
+
+       entry = flow_check_preallocated_entry_cache(mbox, flow, flow_info);
+       if (entry < 0) {
+               otx2_err("Prealloc failed");
+               otx2_flow_mcam_free_counter(mbox, ctr);
+               return NPC_MCAM_ALLOC_FAILED;
+       }
+       req = otx2_mbox_alloc_msg_npc_mcam_write_entry(mbox);
+       req->set_cntr = use_ctr;
+       req->cntr = ctr;
+       req->entry = entry;
+       otx2_npc_dbg("Alloc & write entry %u", entry);
+
+       req->intf =
+               (flow->nix_intf == OTX2_INTF_RX) ? NPC_MCAM_RX : NPC_MCAM_TX;
+       req->enable_entry = 1;
+       req->entry_data.action = flow->npc_action;
+
+       /*
+        * DPDK sets vtag action on per interface basis, not
+        * per flow basis. It is a matter of how we decide to support
+        * this pmd specific behavior. There are two ways:
+        *      1. Inherit the vtag action from the one configured
+        *         for this interface. This can be read from the
+        *         vtag_action configured for default mcam entry of
+        *         this pf_func.
+        *      2. Do not support vtag action with rte_flow.
+        *
+        * Second approach is used now.
+        */
+       req->entry_data.vtag_action = 0ULL;
+
+       for (idx = 0; idx < OTX2_MAX_MCAM_WIDTH_DWORDS; idx++) {
+               req->entry_data.kw[idx] = flow->mcam_data[idx];
+               req->entry_data.kw_mask[idx] = flow->mcam_mask[idx];
+       }
+
+       if (flow->nix_intf == OTX2_INTF_RX) {
+               req->entry_data.kw[0] |= flow_info->channel;
+               req->entry_data.kw_mask[0] |=  (BIT_ULL(12) - 1);
+       } else {
+               uint16_t pf_func = (flow->npc_action >> 4) & 0xffff;
+
+               pf_func = htons(pf_func);
+               req->entry_data.kw[0] |= ((uint64_t)pf_func << 32);
+               req->entry_data.kw_mask[0] |= ((uint64_t)0xffff << 32);
+       }
+
+       otx2_mbox_msg_send(mbox, 0);
+       rc = otx2_mbox_get_rsp(mbox, 0, (void *)&rsp);
+       if (rc != 0)
+               return rc;
+
+       flow->mcam_id = entry;
+       if (use_ctr)
+               flow->ctr_id = ctr;
+       return 0;
+}