net/softnic: add TM hierarchy related ops
authorJasvinder Singh <jasvinder.singh@intel.com>
Tue, 10 Oct 2017 10:18:17 +0000 (11:18 +0100)
committerFerruh Yigit <ferruh.yigit@intel.com>
Thu, 12 Oct 2017 00:52:48 +0000 (01:52 +0100)
Implement ethdev TM hierarchy related APIs in SoftNIC PMD.

Signed-off-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
Signed-off-by: Jasvinder Singh <jasvinder.singh@intel.com>
Acked-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
drivers/net/softnic/rte_eth_softnic_internals.h
drivers/net/softnic/rte_eth_softnic_tm.c

index 75d9387..1f75806 100644 (file)
@@ -138,6 +138,36 @@ enum tm_node_level {
        TM_NODE_LEVEL_MAX,
 };
 
+/* TM Shaper Profile */
+struct tm_shaper_profile {
+       TAILQ_ENTRY(tm_shaper_profile) node;
+       uint32_t shaper_profile_id;
+       uint32_t n_users;
+       struct rte_tm_shaper_params params;
+};
+
+TAILQ_HEAD(tm_shaper_profile_list, tm_shaper_profile);
+
+/* TM Shared Shaper */
+struct tm_shared_shaper {
+       TAILQ_ENTRY(tm_shared_shaper) node;
+       uint32_t shared_shaper_id;
+       uint32_t n_users;
+       uint32_t shaper_profile_id;
+};
+
+TAILQ_HEAD(tm_shared_shaper_list, tm_shared_shaper);
+
+/* TM WRED Profile */
+struct tm_wred_profile {
+       TAILQ_ENTRY(tm_wred_profile) node;
+       uint32_t wred_profile_id;
+       uint32_t n_users;
+       struct rte_tm_wred_params params;
+};
+
+TAILQ_HEAD(tm_wred_profile_list, tm_wred_profile);
+
 /* TM Node */
 struct tm_node {
        TAILQ_ENTRY(tm_node) node;
@@ -147,6 +177,8 @@ struct tm_node {
        uint32_t weight;
        uint32_t level;
        struct tm_node *parent_node;
+       struct tm_shaper_profile *shaper_profile;
+       struct tm_wred_profile *wred_profile;
        struct rte_tm_node_params params;
        struct rte_tm_node_stats stats;
        uint32_t n_children;
@@ -156,8 +188,16 @@ TAILQ_HEAD(tm_node_list, tm_node);
 
 /* TM Hierarchy Specification */
 struct tm_hierarchy {
+       struct tm_shaper_profile_list shaper_profiles;
+       struct tm_shared_shaper_list shared_shapers;
+       struct tm_wred_profile_list wred_profiles;
        struct tm_node_list nodes;
 
+       uint32_t n_shaper_profiles;
+       uint32_t n_shared_shapers;
+       uint32_t n_wred_profiles;
+       uint32_t n_nodes;
+
        uint32_t n_tm_nodes[TM_NODE_LEVEL_MAX];
 };
 
@@ -170,6 +210,7 @@ struct tm_internals {
         *      sense to keep the hierarchy frozen after the port is started.
         */
        struct tm_hierarchy h;
+       int hierarchy_frozen;
 
        /** Blueprints */
        struct tm_params params;
index 73274d4..682cc4d 100644 (file)
@@ -40,7 +40,9 @@
 #include "rte_eth_softnic_internals.h"
 #include "rte_eth_softnic.h"
 
-#define BYTES_IN_MBPS (1000 * 1000 / 8)
+#define BYTES_IN_MBPS          (1000 * 1000 / 8)
+#define SUBPORT_TC_PERIOD      10
+#define PIPE_TC_PERIOD         40
 
 int
 tm_params_check(struct pmd_params *params, uint32_t hard_rate)
@@ -86,6 +88,79 @@ tm_params_check(struct pmd_params *params, uint32_t hard_rate)
        return 0;
 }
 
+static void
+tm_hierarchy_init(struct pmd_internals *p)
+{
+       memset(&p->soft.tm.h, 0, sizeof(p->soft.tm.h));
+
+       /* Initialize shaper profile list */
+       TAILQ_INIT(&p->soft.tm.h.shaper_profiles);
+
+       /* Initialize shared shaper list */
+       TAILQ_INIT(&p->soft.tm.h.shared_shapers);
+
+       /* Initialize wred profile list */
+       TAILQ_INIT(&p->soft.tm.h.wred_profiles);
+
+       /* Initialize TM node list */
+       TAILQ_INIT(&p->soft.tm.h.nodes);
+}
+
+static void
+tm_hierarchy_uninit(struct pmd_internals *p)
+{
+       /* Remove all nodes*/
+       for ( ; ; ) {
+               struct tm_node *tm_node;
+
+               tm_node = TAILQ_FIRST(&p->soft.tm.h.nodes);
+               if (tm_node == NULL)
+                       break;
+
+               TAILQ_REMOVE(&p->soft.tm.h.nodes, tm_node, node);
+               free(tm_node);
+       }
+
+       /* Remove all WRED profiles */
+       for ( ; ; ) {
+               struct tm_wred_profile *wred_profile;
+
+               wred_profile = TAILQ_FIRST(&p->soft.tm.h.wred_profiles);
+               if (wred_profile == NULL)
+                       break;
+
+               TAILQ_REMOVE(&p->soft.tm.h.wred_profiles, wred_profile, node);
+               free(wred_profile);
+       }
+
+       /* Remove all shared shapers */
+       for ( ; ; ) {
+               struct tm_shared_shaper *shared_shaper;
+
+               shared_shaper = TAILQ_FIRST(&p->soft.tm.h.shared_shapers);
+               if (shared_shaper == NULL)
+                       break;
+
+               TAILQ_REMOVE(&p->soft.tm.h.shared_shapers, shared_shaper, node);
+               free(shared_shaper);
+       }
+
+       /* Remove all shaper profiles */
+       for ( ; ; ) {
+               struct tm_shaper_profile *shaper_profile;
+
+               shaper_profile = TAILQ_FIRST(&p->soft.tm.h.shaper_profiles);
+               if (shaper_profile == NULL)
+                       break;
+
+               TAILQ_REMOVE(&p->soft.tm.h.shaper_profiles,
+                       shaper_profile, node);
+               free(shaper_profile);
+       }
+
+       memset(&p->soft.tm.h, 0, sizeof(p->soft.tm.h));
+}
+
 int
 tm_init(struct pmd_internals *p,
        struct pmd_params *params,
@@ -112,12 +187,15 @@ tm_init(struct pmd_internals *p,
                return -ENOMEM;
        }
 
+       tm_hierarchy_init(p);
+
        return 0;
 }
 
 void
 tm_free(struct pmd_internals *p)
 {
+       tm_hierarchy_uninit(p);
        rte_free(p->soft.tm.pkts_enq);
        rte_free(p->soft.tm.pkts_deq);
 }
@@ -129,6 +207,10 @@ tm_start(struct pmd_internals *p)
        uint32_t n_subports, subport_id;
        int status;
 
+       /* Is hierarchy frozen? */
+       if (p->soft.tm.hierarchy_frozen == 0)
+               return -1;
+
        /* Port */
        p->soft.tm.sched = rte_sched_port_config(&t->port_params);
        if (p->soft.tm.sched == NULL)
@@ -178,6 +260,51 @@ tm_stop(struct pmd_internals *p)
 {
        if (p->soft.tm.sched)
                rte_sched_port_free(p->soft.tm.sched);
+
+       /* Unfreeze hierarchy */
+       p->soft.tm.hierarchy_frozen = 0;
+}
+
+static struct tm_shaper_profile *
+tm_shaper_profile_search(struct rte_eth_dev *dev, uint32_t shaper_profile_id)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_shaper_profile_list *spl = &p->soft.tm.h.shaper_profiles;
+       struct tm_shaper_profile *sp;
+
+       TAILQ_FOREACH(sp, spl, node)
+               if (shaper_profile_id == sp->shaper_profile_id)
+                       return sp;
+
+       return NULL;
+}
+
+static struct tm_shared_shaper *
+tm_shared_shaper_search(struct rte_eth_dev *dev, uint32_t shared_shaper_id)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_shared_shaper_list *ssl = &p->soft.tm.h.shared_shapers;
+       struct tm_shared_shaper *ss;
+
+       TAILQ_FOREACH(ss, ssl, node)
+               if (shared_shaper_id == ss->shared_shaper_id)
+                       return ss;
+
+       return NULL;
+}
+
+static struct tm_wred_profile *
+tm_wred_profile_search(struct rte_eth_dev *dev, uint32_t wred_profile_id)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_wred_profile_list *wpl = &p->soft.tm.h.wred_profiles;
+       struct tm_wred_profile *wp;
+
+       TAILQ_FOREACH(wp, wpl, node)
+               if (wred_profile_id == wp->wred_profile_id)
+                       return wp;
+
+       return NULL;
 }
 
 static struct tm_node *
@@ -194,6 +321,94 @@ tm_node_search(struct rte_eth_dev *dev, uint32_t node_id)
        return NULL;
 }
 
+static struct tm_node *
+tm_root_node_present(struct rte_eth_dev *dev)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_node_list *nl = &p->soft.tm.h.nodes;
+       struct tm_node *n;
+
+       TAILQ_FOREACH(n, nl, node)
+               if (n->parent_node_id == RTE_TM_NODE_ID_NULL)
+                       return n;
+
+       return NULL;
+}
+
+static uint32_t
+tm_node_subport_id(struct rte_eth_dev *dev, struct tm_node *subport_node)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_node_list *nl = &p->soft.tm.h.nodes;
+       struct tm_node *ns;
+       uint32_t subport_id;
+
+       subport_id = 0;
+       TAILQ_FOREACH(ns, nl, node) {
+               if (ns->level != TM_NODE_LEVEL_SUBPORT)
+                       continue;
+
+               if (ns->node_id == subport_node->node_id)
+                       return subport_id;
+
+               subport_id++;
+       }
+
+       return UINT32_MAX;
+}
+
+static uint32_t
+tm_node_pipe_id(struct rte_eth_dev *dev, struct tm_node *pipe_node)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_node_list *nl = &p->soft.tm.h.nodes;
+       struct tm_node *np;
+       uint32_t pipe_id;
+
+       pipe_id = 0;
+       TAILQ_FOREACH(np, nl, node) {
+               if (np->level != TM_NODE_LEVEL_PIPE ||
+                       np->parent_node_id != pipe_node->parent_node_id)
+                       continue;
+
+               if (np->node_id == pipe_node->node_id)
+                       return pipe_id;
+
+               pipe_id++;
+       }
+
+       return UINT32_MAX;
+}
+
+static uint32_t
+tm_node_tc_id(struct rte_eth_dev *dev __rte_unused, struct tm_node *tc_node)
+{
+       return tc_node->priority;
+}
+
+static uint32_t
+tm_node_queue_id(struct rte_eth_dev *dev, struct tm_node *queue_node)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_node_list *nl = &p->soft.tm.h.nodes;
+       struct tm_node *nq;
+       uint32_t queue_id;
+
+       queue_id = 0;
+       TAILQ_FOREACH(nq, nl, node) {
+               if (nq->level != TM_NODE_LEVEL_QUEUE ||
+                       nq->parent_node_id != queue_node->parent_node_id)
+                       continue;
+
+               if (nq->node_id == queue_node->node_id)
+                       return queue_id;
+
+               queue_id++;
+       }
+
+       return UINT32_MAX;
+}
+
 static uint32_t
 tm_level_get_max_nodes(struct rte_eth_dev *dev, enum tm_node_level level)
 {
@@ -219,6 +434,35 @@ tm_level_get_max_nodes(struct rte_eth_dev *dev, enum tm_node_level level)
        }
 }
 
+/* Traffic manager node type get */
+static int
+pmd_tm_node_type_get(struct rte_eth_dev *dev,
+       uint32_t node_id,
+       int *is_leaf,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+
+       if (is_leaf == NULL)
+               return -rte_tm_error_set(error,
+                  EINVAL,
+                  RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                  NULL,
+                  rte_strerror(EINVAL));
+
+       if (node_id == RTE_TM_NODE_ID_NULL ||
+               (tm_node_search(dev, node_id) == NULL))
+               return -rte_tm_error_set(error,
+                  EINVAL,
+                  RTE_TM_ERROR_TYPE_NODE_ID,
+                  NULL,
+                  rte_strerror(EINVAL));
+
+       *is_leaf = node_id < p->params.soft.tm.nb_queues;
+
+       return 0;
+}
+
 #ifdef RTE_SCHED_RED
 #define WRED_SUPPORTED                                         1
 #else
@@ -674,8 +918,2535 @@ pmd_tm_node_capabilities_get(struct rte_eth_dev *dev __rte_unused,
        return 0;
 }
 
-const struct rte_tm_ops pmd_tm_ops = {
-       .capabilities_get = pmd_tm_capabilities_get,
-       .level_capabilities_get = pmd_tm_level_capabilities_get,
-       .node_capabilities_get = pmd_tm_node_capabilities_get,
+static int
+shaper_profile_check(struct rte_eth_dev *dev,
+       uint32_t shaper_profile_id,
+       struct rte_tm_shaper_params *profile,
+       struct rte_tm_error *error)
+{
+       struct tm_shaper_profile *sp;
+
+       /* Shaper profile ID must not be NONE. */
+       if (shaper_profile_id == RTE_TM_SHAPER_PROFILE_ID_NONE)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_SHAPER_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Shaper profile must not exist. */
+       sp = tm_shaper_profile_search(dev, shaper_profile_id);
+       if (sp)
+               return -rte_tm_error_set(error,
+                       EEXIST,
+                       RTE_TM_ERROR_TYPE_SHAPER_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EEXIST));
+
+       /* Profile must not be NULL. */
+       if (profile == NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_SHAPER_PROFILE,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Peak rate: non-zero, 32-bit */
+       if (profile->peak.rate == 0 ||
+               profile->peak.rate >= UINT32_MAX)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_SHAPER_PROFILE_PEAK_RATE,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Peak size: non-zero, 32-bit */
+       if (profile->peak.size == 0 ||
+               profile->peak.size >= UINT32_MAX)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_SHAPER_PROFILE_PEAK_SIZE,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Dual-rate profiles are not supported. */
+       if (profile->committed.rate != 0)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_SHAPER_PROFILE_COMMITTED_RATE,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Packet length adjust: 24 bytes */
+       if (profile->pkt_length_adjust != RTE_TM_ETH_FRAMING_OVERHEAD_FCS)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_SHAPER_PROFILE_PKT_ADJUST_LEN,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       return 0;
+}
+
+/* Traffic manager shaper profile add */
+static int
+pmd_tm_shaper_profile_add(struct rte_eth_dev *dev,
+       uint32_t shaper_profile_id,
+       struct rte_tm_shaper_params *profile,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_shaper_profile_list *spl = &p->soft.tm.h.shaper_profiles;
+       struct tm_shaper_profile *sp;
+       int status;
+
+       /* Check input params */
+       status = shaper_profile_check(dev, shaper_profile_id, profile, error);
+       if (status)
+               return status;
+
+       /* Memory allocation */
+       sp = calloc(1, sizeof(struct tm_shaper_profile));
+       if (sp == NULL)
+               return -rte_tm_error_set(error,
+                       ENOMEM,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(ENOMEM));
+
+       /* Fill in */
+       sp->shaper_profile_id = shaper_profile_id;
+       memcpy(&sp->params, profile, sizeof(sp->params));
+
+       /* Add to list */
+       TAILQ_INSERT_TAIL(spl, sp, node);
+       p->soft.tm.h.n_shaper_profiles++;
+
+       return 0;
+}
+
+/* Traffic manager shaper profile delete */
+static int
+pmd_tm_shaper_profile_delete(struct rte_eth_dev *dev,
+       uint32_t shaper_profile_id,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_shaper_profile *sp;
+
+       /* Check existing */
+       sp = tm_shaper_profile_search(dev, shaper_profile_id);
+       if (sp == NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_SHAPER_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Check unused */
+       if (sp->n_users)
+               return -rte_tm_error_set(error,
+                       EBUSY,
+                       RTE_TM_ERROR_TYPE_SHAPER_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EBUSY));
+
+       /* Remove from list */
+       TAILQ_REMOVE(&p->soft.tm.h.shaper_profiles, sp, node);
+       p->soft.tm.h.n_shaper_profiles--;
+       free(sp);
+
+       return 0;
+}
+
+static struct tm_node *
+tm_shared_shaper_get_tc(struct rte_eth_dev *dev,
+       struct tm_shared_shaper *ss)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_node_list *nl = &p->soft.tm.h.nodes;
+       struct tm_node *n;
+
+       /* Subport: each TC uses shared shaper  */
+       TAILQ_FOREACH(n, nl, node) {
+               if (n->level != TM_NODE_LEVEL_TC ||
+                       n->params.n_shared_shapers == 0 ||
+                       n->params.shared_shaper_id[0] != ss->shared_shaper_id)
+                       continue;
+
+               return n;
+       }
+
+       return NULL;
+}
+
+static int
+update_subport_tc_rate(struct rte_eth_dev *dev,
+       struct tm_node *nt,
+       struct tm_shared_shaper *ss,
+       struct tm_shaper_profile *sp_new)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       uint32_t tc_id = tm_node_tc_id(dev, nt);
+
+       struct tm_node *np = nt->parent_node;
+
+       struct tm_node *ns = np->parent_node;
+       uint32_t subport_id = tm_node_subport_id(dev, ns);
+
+       struct rte_sched_subport_params subport_params;
+
+       struct tm_shaper_profile *sp_old = tm_shaper_profile_search(dev,
+               ss->shaper_profile_id);
+
+       /* Derive new subport configuration. */
+       memcpy(&subport_params,
+               &p->soft.tm.params.subport_params[subport_id],
+               sizeof(subport_params));
+       subport_params.tc_rate[tc_id] = sp_new->params.peak.rate;
+
+       /* Update the subport configuration. */
+       if (rte_sched_subport_config(p->soft.tm.sched,
+               subport_id, &subport_params))
+               return -1;
+
+       /* Commit changes. */
+       sp_old->n_users--;
+
+       ss->shaper_profile_id = sp_new->shaper_profile_id;
+       sp_new->n_users++;
+
+       memcpy(&p->soft.tm.params.subport_params[subport_id],
+               &subport_params,
+               sizeof(subport_params));
+
+       return 0;
+}
+
+/* Traffic manager shared shaper add/update */
+static int
+pmd_tm_shared_shaper_add_update(struct rte_eth_dev *dev,
+       uint32_t shared_shaper_id,
+       uint32_t shaper_profile_id,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_shared_shaper *ss;
+       struct tm_shaper_profile *sp;
+       struct tm_node *nt;
+
+       /* Shaper profile must be valid. */
+       sp = tm_shaper_profile_search(dev, shaper_profile_id);
+       if (sp == NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_SHAPER_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /**
+        * Add new shared shaper
+        */
+       ss = tm_shared_shaper_search(dev, shared_shaper_id);
+       if (ss == NULL) {
+               struct tm_shared_shaper_list *ssl =
+                       &p->soft.tm.h.shared_shapers;
+
+               /* Hierarchy must not be frozen */
+               if (p->soft.tm.hierarchy_frozen)
+                       return -rte_tm_error_set(error,
+                               EBUSY,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EBUSY));
+
+               /* Memory allocation */
+               ss = calloc(1, sizeof(struct tm_shared_shaper));
+               if (ss == NULL)
+                       return -rte_tm_error_set(error,
+                               ENOMEM,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(ENOMEM));
+
+               /* Fill in */
+               ss->shared_shaper_id = shared_shaper_id;
+               ss->shaper_profile_id = shaper_profile_id;
+
+               /* Add to list */
+               TAILQ_INSERT_TAIL(ssl, ss, node);
+               p->soft.tm.h.n_shared_shapers++;
+
+               return 0;
+       }
+
+       /**
+        * Update existing shared shaper
+        */
+       /* Hierarchy must be frozen (run-time update) */
+       if (p->soft.tm.hierarchy_frozen == 0)
+               return -rte_tm_error_set(error,
+                       EBUSY,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(EBUSY));
+
+
+       /* Propagate change. */
+       nt = tm_shared_shaper_get_tc(dev, ss);
+       if (update_subport_tc_rate(dev, nt, ss, sp))
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       return 0;
+}
+
+/* Traffic manager shared shaper delete */
+static int
+pmd_tm_shared_shaper_delete(struct rte_eth_dev *dev,
+       uint32_t shared_shaper_id,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_shared_shaper *ss;
+
+       /* Check existing */
+       ss = tm_shared_shaper_search(dev, shared_shaper_id);
+       if (ss == NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_SHARED_SHAPER_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Check unused */
+       if (ss->n_users)
+               return -rte_tm_error_set(error,
+                       EBUSY,
+                       RTE_TM_ERROR_TYPE_SHARED_SHAPER_ID,
+                       NULL,
+                       rte_strerror(EBUSY));
+
+       /* Remove from list */
+       TAILQ_REMOVE(&p->soft.tm.h.shared_shapers, ss, node);
+       p->soft.tm.h.n_shared_shapers--;
+       free(ss);
+
+       return 0;
+}
+
+static int
+wred_profile_check(struct rte_eth_dev *dev,
+       uint32_t wred_profile_id,
+       struct rte_tm_wred_params *profile,
+       struct rte_tm_error *error)
+{
+       struct tm_wred_profile *wp;
+       enum rte_tm_color color;
+
+       /* WRED profile ID must not be NONE. */
+       if (wred_profile_id == RTE_TM_WRED_PROFILE_ID_NONE)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_WRED_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* WRED profile must not exist. */
+       wp = tm_wred_profile_search(dev, wred_profile_id);
+       if (wp)
+               return -rte_tm_error_set(error,
+                       EEXIST,
+                       RTE_TM_ERROR_TYPE_WRED_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EEXIST));
+
+       /* Profile must not be NULL. */
+       if (profile == NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_WRED_PROFILE,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* min_th <= max_th, max_th > 0  */
+       for (color = RTE_TM_GREEN; color < RTE_TM_COLORS; color++) {
+               uint16_t min_th = profile->red_params[color].min_th;
+               uint16_t max_th = profile->red_params[color].max_th;
+
+               if (min_th > max_th || max_th == 0)
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_WRED_PROFILE,
+                               NULL,
+                               rte_strerror(EINVAL));
+       }
+
+       return 0;
+}
+
+/* Traffic manager WRED profile add */
+static int
+pmd_tm_wred_profile_add(struct rte_eth_dev *dev,
+       uint32_t wred_profile_id,
+       struct rte_tm_wred_params *profile,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_wred_profile_list *wpl = &p->soft.tm.h.wred_profiles;
+       struct tm_wred_profile *wp;
+       int status;
+
+       /* Check input params */
+       status = wred_profile_check(dev, wred_profile_id, profile, error);
+       if (status)
+               return status;
+
+       /* Memory allocation */
+       wp = calloc(1, sizeof(struct tm_wred_profile));
+       if (wp == NULL)
+               return -rte_tm_error_set(error,
+                       ENOMEM,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(ENOMEM));
+
+       /* Fill in */
+       wp->wred_profile_id = wred_profile_id;
+       memcpy(&wp->params, profile, sizeof(wp->params));
+
+       /* Add to list */
+       TAILQ_INSERT_TAIL(wpl, wp, node);
+       p->soft.tm.h.n_wred_profiles++;
+
+       return 0;
+}
+
+/* Traffic manager WRED profile delete */
+static int
+pmd_tm_wred_profile_delete(struct rte_eth_dev *dev,
+       uint32_t wred_profile_id,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_wred_profile *wp;
+
+       /* Check existing */
+       wp = tm_wred_profile_search(dev, wred_profile_id);
+       if (wp == NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_WRED_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Check unused */
+       if (wp->n_users)
+               return -rte_tm_error_set(error,
+                       EBUSY,
+                       RTE_TM_ERROR_TYPE_WRED_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EBUSY));
+
+       /* Remove from list */
+       TAILQ_REMOVE(&p->soft.tm.h.wred_profiles, wp, node);
+       p->soft.tm.h.n_wred_profiles--;
+       free(wp);
+
+       return 0;
+}
+
+static int
+node_add_check_port(struct rte_eth_dev *dev,
+       uint32_t node_id,
+       uint32_t parent_node_id __rte_unused,
+       uint32_t priority,
+       uint32_t weight,
+       uint32_t level_id __rte_unused,
+       struct rte_tm_node_params *params,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_shaper_profile *sp = tm_shaper_profile_search(dev,
+               params->shaper_profile_id);
+
+       /* node type: non-leaf */
+       if (node_id < p->params.soft.tm.nb_queues)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Priority must be 0 */
+       if (priority != 0)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PRIORITY,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Weight must be 1 */
+       if (weight != 1)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_WEIGHT,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Shaper must be valid.
+        * Shaper profile peak rate must fit the configured port rate.
+        */
+       if (params->shaper_profile_id == RTE_TM_SHAPER_PROFILE_ID_NONE ||
+               sp == NULL ||
+               sp->params.peak.rate > p->params.soft.tm.rate)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_SHAPER_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* No shared shapers */
+       if (params->n_shared_shapers != 0)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_N_SHARED_SHAPERS,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Number of SP priorities must be 1 */
+       if (params->nonleaf.n_sp_priorities != 1)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_N_SP_PRIORITIES,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Stats */
+       if (params->stats_mask & ~STATS_MASK_DEFAULT)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_STATS,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       return 0;
+}
+
+static int
+node_add_check_subport(struct rte_eth_dev *dev,
+       uint32_t node_id,
+       uint32_t parent_node_id __rte_unused,
+       uint32_t priority,
+       uint32_t weight,
+       uint32_t level_id __rte_unused,
+       struct rte_tm_node_params *params,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+
+       /* node type: non-leaf */
+       if (node_id < p->params.soft.tm.nb_queues)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Priority must be 0 */
+       if (priority != 0)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PRIORITY,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Weight must be 1 */
+       if (weight != 1)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_WEIGHT,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Shaper must be valid */
+       if (params->shaper_profile_id == RTE_TM_SHAPER_PROFILE_ID_NONE ||
+               (!tm_shaper_profile_search(dev, params->shaper_profile_id)))
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_SHAPER_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* No shared shapers */
+       if (params->n_shared_shapers != 0)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_N_SHARED_SHAPERS,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Number of SP priorities must be 1 */
+       if (params->nonleaf.n_sp_priorities != 1)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_N_SP_PRIORITIES,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Stats */
+       if (params->stats_mask & ~STATS_MASK_DEFAULT)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_STATS,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       return 0;
+}
+
+static int
+node_add_check_pipe(struct rte_eth_dev *dev,
+       uint32_t node_id,
+       uint32_t parent_node_id __rte_unused,
+       uint32_t priority,
+       uint32_t weight __rte_unused,
+       uint32_t level_id __rte_unused,
+       struct rte_tm_node_params *params,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+
+       /* node type: non-leaf */
+       if (node_id < p->params.soft.tm.nb_queues)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Priority must be 0 */
+       if (priority != 0)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PRIORITY,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Shaper must be valid */
+       if (params->shaper_profile_id == RTE_TM_SHAPER_PROFILE_ID_NONE ||
+               (!tm_shaper_profile_search(dev, params->shaper_profile_id)))
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_SHAPER_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* No shared shapers */
+       if (params->n_shared_shapers != 0)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_N_SHARED_SHAPERS,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Number of SP priorities must be 4 */
+       if (params->nonleaf.n_sp_priorities !=
+               RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_N_SP_PRIORITIES,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* WFQ mode must be byte mode */
+       if (params->nonleaf.wfq_weight_mode != NULL &&
+               params->nonleaf.wfq_weight_mode[0] != 0 &&
+               params->nonleaf.wfq_weight_mode[1] != 0 &&
+               params->nonleaf.wfq_weight_mode[2] != 0 &&
+               params->nonleaf.wfq_weight_mode[3] != 0)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_WFQ_WEIGHT_MODE,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Stats */
+       if (params->stats_mask & ~STATS_MASK_DEFAULT)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_STATS,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       return 0;
+}
+
+static int
+node_add_check_tc(struct rte_eth_dev *dev,
+       uint32_t node_id,
+       uint32_t parent_node_id __rte_unused,
+       uint32_t priority __rte_unused,
+       uint32_t weight,
+       uint32_t level_id __rte_unused,
+       struct rte_tm_node_params *params,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+
+       /* node type: non-leaf */
+       if (node_id < p->params.soft.tm.nb_queues)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Weight must be 1 */
+       if (weight != 1)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_WEIGHT,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Shaper must be valid */
+       if (params->shaper_profile_id == RTE_TM_SHAPER_PROFILE_ID_NONE ||
+               (!tm_shaper_profile_search(dev, params->shaper_profile_id)))
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_SHAPER_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Single valid shared shaper */
+       if (params->n_shared_shapers > 1)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_N_SHARED_SHAPERS,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       if (params->n_shared_shapers == 1 &&
+               (params->shared_shaper_id == NULL ||
+               (!tm_shared_shaper_search(dev, params->shared_shaper_id[0]))))
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_SHARED_SHAPER_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Number of priorities must be 1 */
+       if (params->nonleaf.n_sp_priorities != 1)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_N_SP_PRIORITIES,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Stats */
+       if (params->stats_mask & ~STATS_MASK_DEFAULT)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_STATS,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       return 0;
+}
+
+static int
+node_add_check_queue(struct rte_eth_dev *dev,
+       uint32_t node_id,
+       uint32_t parent_node_id __rte_unused,
+       uint32_t priority,
+       uint32_t weight __rte_unused,
+       uint32_t level_id __rte_unused,
+       struct rte_tm_node_params *params,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+
+       /* node type: leaf */
+       if (node_id >= p->params.soft.tm.nb_queues)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Priority must be 0 */
+       if (priority != 0)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PRIORITY,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* No shaper */
+       if (params->shaper_profile_id != RTE_TM_SHAPER_PROFILE_ID_NONE)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_SHAPER_PROFILE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* No shared shapers */
+       if (params->n_shared_shapers != 0)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_N_SHARED_SHAPERS,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Congestion management must not be head drop */
+       if (params->leaf.cman == RTE_TM_CMAN_HEAD_DROP)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_CMAN,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Congestion management set to WRED */
+       if (params->leaf.cman == RTE_TM_CMAN_WRED) {
+               uint32_t wred_profile_id = params->leaf.wred.wred_profile_id;
+               struct tm_wred_profile *wp = tm_wred_profile_search(dev,
+                       wred_profile_id);
+
+               /* WRED profile (for private WRED context) must be valid */
+               if (wred_profile_id == RTE_TM_WRED_PROFILE_ID_NONE ||
+                       wp == NULL)
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_NODE_PARAMS_WRED_PROFILE_ID,
+                               NULL,
+                               rte_strerror(EINVAL));
+
+               /* No shared WRED contexts */
+               if (params->leaf.wred.n_shared_wred_contexts != 0)
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_NODE_PARAMS_N_SHARED_WRED_CONTEXTS,
+                               NULL,
+                               rte_strerror(EINVAL));
+       }
+
+       /* Stats */
+       if (params->stats_mask & ~STATS_MASK_QUEUE)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS_STATS,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       return 0;
+}
+
+static int
+node_add_check(struct rte_eth_dev *dev,
+       uint32_t node_id,
+       uint32_t parent_node_id,
+       uint32_t priority,
+       uint32_t weight,
+       uint32_t level_id,
+       struct rte_tm_node_params *params,
+       struct rte_tm_error *error)
+{
+       struct tm_node *pn;
+       uint32_t level;
+       int status;
+
+       /* node_id, parent_node_id:
+        *    -node_id must not be RTE_TM_NODE_ID_NULL
+        *    -node_id must not be in use
+        *    -root node add (parent_node_id is RTE_TM_NODE_ID_NULL):
+        *        -root node must not exist
+        *    -non-root node add (parent_node_id is not RTE_TM_NODE_ID_NULL):
+        *        -parent_node_id must be valid
+        */
+       if (node_id == RTE_TM_NODE_ID_NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       if (tm_node_search(dev, node_id))
+               return -rte_tm_error_set(error,
+                       EEXIST,
+                       RTE_TM_ERROR_TYPE_NODE_ID,
+                       NULL,
+                       rte_strerror(EEXIST));
+
+       if (parent_node_id == RTE_TM_NODE_ID_NULL) {
+               pn = NULL;
+               if (tm_root_node_present(dev))
+                       return -rte_tm_error_set(error,
+                               EEXIST,
+                               RTE_TM_ERROR_TYPE_NODE_PARENT_NODE_ID,
+                               NULL,
+                               rte_strerror(EEXIST));
+       } else {
+               pn = tm_node_search(dev, parent_node_id);
+               if (pn == NULL)
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_NODE_PARENT_NODE_ID,
+                               NULL,
+                               rte_strerror(EINVAL));
+       }
+
+       /* priority: must be 0 .. 3 */
+       if (priority >= RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PRIORITY,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* weight: must be 1 .. 255 */
+       if (weight == 0 || weight >= UINT8_MAX)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_WEIGHT,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* level_id: if valid, then
+        *    -root node add (parent_node_id is RTE_TM_NODE_ID_NULL):
+        *        -level_id must be zero
+        *    -non-root node add (parent_node_id is not RTE_TM_NODE_ID_NULL):
+        *        -level_id must be parent level ID plus one
+        */
+       level = (pn == NULL) ? 0 : pn->level + 1;
+       if (level_id != RTE_TM_NODE_LEVEL_ID_ANY && level_id != level)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_LEVEL_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* params: must not be NULL */
+       if (params == NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARAMS,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* params: per level checks */
+       switch (level) {
+       case TM_NODE_LEVEL_PORT:
+               status = node_add_check_port(dev, node_id,
+                       parent_node_id, priority, weight, level_id,
+                       params, error);
+               if (status)
+                       return status;
+               break;
+
+       case TM_NODE_LEVEL_SUBPORT:
+               status = node_add_check_subport(dev, node_id,
+                       parent_node_id, priority, weight, level_id,
+                       params, error);
+               if (status)
+                       return status;
+               break;
+
+       case TM_NODE_LEVEL_PIPE:
+               status = node_add_check_pipe(dev, node_id,
+                       parent_node_id, priority, weight, level_id,
+                       params, error);
+               if (status)
+                       return status;
+               break;
+
+       case TM_NODE_LEVEL_TC:
+               status = node_add_check_tc(dev, node_id,
+                       parent_node_id, priority, weight, level_id,
+                       params, error);
+               if (status)
+                       return status;
+               break;
+
+       case TM_NODE_LEVEL_QUEUE:
+               status = node_add_check_queue(dev, node_id,
+                       parent_node_id, priority, weight, level_id,
+                       params, error);
+               if (status)
+                       return status;
+               break;
+
+       default:
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_LEVEL_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+       }
+
+       return 0;
+}
+
+/* Traffic manager node add */
+static int
+pmd_tm_node_add(struct rte_eth_dev *dev,
+       uint32_t node_id,
+       uint32_t parent_node_id,
+       uint32_t priority,
+       uint32_t weight,
+       uint32_t level_id,
+       struct rte_tm_node_params *params,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_node_list *nl = &p->soft.tm.h.nodes;
+       struct tm_node *n;
+       uint32_t i;
+       int status;
+
+       /* Checks */
+       if (p->soft.tm.hierarchy_frozen)
+               return -rte_tm_error_set(error,
+                       EBUSY,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(EBUSY));
+
+       status = node_add_check(dev, node_id, parent_node_id, priority, weight,
+               level_id, params, error);
+       if (status)
+               return status;
+
+       /* Memory allocation */
+       n = calloc(1, sizeof(struct tm_node));
+       if (n == NULL)
+               return -rte_tm_error_set(error,
+                       ENOMEM,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(ENOMEM));
+
+       /* Fill in */
+       n->node_id = node_id;
+       n->parent_node_id = parent_node_id;
+       n->priority = priority;
+       n->weight = weight;
+
+       if (parent_node_id != RTE_TM_NODE_ID_NULL) {
+               n->parent_node = tm_node_search(dev, parent_node_id);
+               n->level = n->parent_node->level + 1;
+       }
+
+       if (params->shaper_profile_id != RTE_TM_SHAPER_PROFILE_ID_NONE)
+               n->shaper_profile = tm_shaper_profile_search(dev,
+                       params->shaper_profile_id);
+
+       if (n->level == TM_NODE_LEVEL_QUEUE &&
+               params->leaf.cman == RTE_TM_CMAN_WRED)
+               n->wred_profile = tm_wred_profile_search(dev,
+                       params->leaf.wred.wred_profile_id);
+
+       memcpy(&n->params, params, sizeof(n->params));
+
+       /* Add to list */
+       TAILQ_INSERT_TAIL(nl, n, node);
+       p->soft.tm.h.n_nodes++;
+
+       /* Update dependencies */
+       if (n->parent_node)
+               n->parent_node->n_children++;
+
+       if (n->shaper_profile)
+               n->shaper_profile->n_users++;
+
+       for (i = 0; i < params->n_shared_shapers; i++) {
+               struct tm_shared_shaper *ss;
+
+               ss = tm_shared_shaper_search(dev, params->shared_shaper_id[i]);
+               ss->n_users++;
+       }
+
+       if (n->wred_profile)
+               n->wred_profile->n_users++;
+
+       p->soft.tm.h.n_tm_nodes[n->level]++;
+
+       return 0;
+}
+
+/* Traffic manager node delete */
+static int
+pmd_tm_node_delete(struct rte_eth_dev *dev,
+       uint32_t node_id,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_node *n;
+       uint32_t i;
+
+       /* Check hierarchy changes are currently allowed */
+       if (p->soft.tm.hierarchy_frozen)
+               return -rte_tm_error_set(error,
+                       EBUSY,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(EBUSY));
+
+       /* Check existing */
+       n = tm_node_search(dev, node_id);
+       if (n == NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Check unused */
+       if (n->n_children)
+               return -rte_tm_error_set(error,
+                       EBUSY,
+                       RTE_TM_ERROR_TYPE_NODE_ID,
+                       NULL,
+                       rte_strerror(EBUSY));
+
+       /* Update dependencies */
+       p->soft.tm.h.n_tm_nodes[n->level]--;
+
+       if (n->wred_profile)
+               n->wred_profile->n_users--;
+
+       for (i = 0; i < n->params.n_shared_shapers; i++) {
+               struct tm_shared_shaper *ss;
+
+               ss = tm_shared_shaper_search(dev,
+                               n->params.shared_shaper_id[i]);
+               ss->n_users--;
+       }
+
+       if (n->shaper_profile)
+               n->shaper_profile->n_users--;
+
+       if (n->parent_node)
+               n->parent_node->n_children--;
+
+       /* Remove from list */
+       TAILQ_REMOVE(&p->soft.tm.h.nodes, n, node);
+       p->soft.tm.h.n_nodes--;
+       free(n);
+
+       return 0;
+}
+
+
+static void
+pipe_profile_build(struct rte_eth_dev *dev,
+       struct tm_node *np,
+       struct rte_sched_pipe_params *pp)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_hierarchy *h = &p->soft.tm.h;
+       struct tm_node_list *nl = &h->nodes;
+       struct tm_node *nt, *nq;
+
+       memset(pp, 0, sizeof(*pp));
+
+       /* Pipe */
+       pp->tb_rate = np->shaper_profile->params.peak.rate;
+       pp->tb_size = np->shaper_profile->params.peak.size;
+
+       /* Traffic Class (TC) */
+       pp->tc_period = PIPE_TC_PERIOD;
+
+#ifdef RTE_SCHED_SUBPORT_TC_OV
+       pp->tc_ov_weight = np->weight;
+#endif
+
+       TAILQ_FOREACH(nt, nl, node) {
+               uint32_t queue_id = 0;
+
+               if (nt->level != TM_NODE_LEVEL_TC ||
+                       nt->parent_node_id != np->node_id)
+                       continue;
+
+               pp->tc_rate[nt->priority] =
+                       nt->shaper_profile->params.peak.rate;
+
+               /* Queue */
+               TAILQ_FOREACH(nq, nl, node) {
+                       uint32_t pipe_queue_id;
+
+                       if (nq->level != TM_NODE_LEVEL_QUEUE ||
+                               nq->parent_node_id != nt->node_id)
+                               continue;
+
+                       pipe_queue_id = nt->priority *
+                               RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS + queue_id;
+                       pp->wrr_weights[pipe_queue_id] = nq->weight;
+
+                       queue_id++;
+               }
+       }
+}
+
+static int
+pipe_profile_free_exists(struct rte_eth_dev *dev,
+       uint32_t *pipe_profile_id)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_params *t = &p->soft.tm.params;
+
+       if (t->n_pipe_profiles < RTE_SCHED_PIPE_PROFILES_PER_PORT) {
+               *pipe_profile_id = t->n_pipe_profiles;
+               return 1;
+       }
+
+       return 0;
+}
+
+static int
+pipe_profile_exists(struct rte_eth_dev *dev,
+       struct rte_sched_pipe_params *pp,
+       uint32_t *pipe_profile_id)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_params *t = &p->soft.tm.params;
+       uint32_t i;
+
+       for (i = 0; i < t->n_pipe_profiles; i++)
+               if (memcmp(&t->pipe_profiles[i], pp, sizeof(*pp)) == 0) {
+                       if (pipe_profile_id)
+                               *pipe_profile_id = i;
+                       return 1;
+               }
+
+       return 0;
+}
+
+static void
+pipe_profile_install(struct rte_eth_dev *dev,
+       struct rte_sched_pipe_params *pp,
+       uint32_t pipe_profile_id)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_params *t = &p->soft.tm.params;
+
+       memcpy(&t->pipe_profiles[pipe_profile_id], pp, sizeof(*pp));
+       t->n_pipe_profiles++;
+}
+
+static void
+pipe_profile_mark(struct rte_eth_dev *dev,
+       uint32_t subport_id,
+       uint32_t pipe_id,
+       uint32_t pipe_profile_id)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_hierarchy *h = &p->soft.tm.h;
+       struct tm_params *t = &p->soft.tm.params;
+       uint32_t n_pipes_per_subport, pos;
+
+       n_pipes_per_subport = h->n_tm_nodes[TM_NODE_LEVEL_PIPE] /
+               h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT];
+       pos = subport_id * n_pipes_per_subport + pipe_id;
+
+       t->pipe_to_profile[pos] = pipe_profile_id;
+}
+
+static struct rte_sched_pipe_params *
+pipe_profile_get(struct rte_eth_dev *dev, struct tm_node *np)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_hierarchy *h = &p->soft.tm.h;
+       struct tm_params *t = &p->soft.tm.params;
+       uint32_t n_pipes_per_subport = h->n_tm_nodes[TM_NODE_LEVEL_PIPE] /
+               h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT];
+
+       uint32_t subport_id = tm_node_subport_id(dev, np->parent_node);
+       uint32_t pipe_id = tm_node_pipe_id(dev, np);
+
+       uint32_t pos = subport_id * n_pipes_per_subport + pipe_id;
+       uint32_t pipe_profile_id = t->pipe_to_profile[pos];
+
+       return &t->pipe_profiles[pipe_profile_id];
+}
+
+static int
+pipe_profiles_generate(struct rte_eth_dev *dev)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_hierarchy *h = &p->soft.tm.h;
+       struct tm_node_list *nl = &h->nodes;
+       struct tm_node *ns, *np;
+       uint32_t subport_id;
+
+       /* Objective: Fill in the following fields in struct tm_params:
+        *    - pipe_profiles
+        *    - n_pipe_profiles
+        *    - pipe_to_profile
+        */
+
+       subport_id = 0;
+       TAILQ_FOREACH(ns, nl, node) {
+               uint32_t pipe_id;
+
+               if (ns->level != TM_NODE_LEVEL_SUBPORT)
+                       continue;
+
+               pipe_id = 0;
+               TAILQ_FOREACH(np, nl, node) {
+                       struct rte_sched_pipe_params pp;
+                       uint32_t pos;
+
+                       if (np->level != TM_NODE_LEVEL_PIPE ||
+                               np->parent_node_id != ns->node_id)
+                               continue;
+
+                       pipe_profile_build(dev, np, &pp);
+
+                       if (!pipe_profile_exists(dev, &pp, &pos)) {
+                               if (!pipe_profile_free_exists(dev, &pos))
+                                       return -1;
+
+                               pipe_profile_install(dev, &pp, pos);
+                       }
+
+                       pipe_profile_mark(dev, subport_id, pipe_id, pos);
+
+                       pipe_id++;
+               }
+
+               subport_id++;
+       }
+
+       return 0;
+}
+
+static struct tm_wred_profile *
+tm_tc_wred_profile_get(struct rte_eth_dev *dev, uint32_t tc_id)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_hierarchy *h = &p->soft.tm.h;
+       struct tm_node_list *nl = &h->nodes;
+       struct tm_node *nq;
+
+       TAILQ_FOREACH(nq, nl, node) {
+               if (nq->level != TM_NODE_LEVEL_QUEUE ||
+                       nq->parent_node->priority != tc_id)
+                       continue;
+
+               return nq->wred_profile;
+       }
+
+       return NULL;
+}
+
+#ifdef RTE_SCHED_RED
+
+static void
+wred_profiles_set(struct rte_eth_dev *dev)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct rte_sched_port_params *pp = &p->soft.tm.params.port_params;
+       uint32_t tc_id;
+       enum rte_tm_color color;
+
+       for (tc_id = 0; tc_id < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; tc_id++)
+               for (color = RTE_TM_GREEN; color < RTE_TM_COLORS; color++) {
+                       struct rte_red_params *dst =
+                               &pp->red_params[tc_id][color];
+                       struct tm_wred_profile *src_wp =
+                               tm_tc_wred_profile_get(dev, tc_id);
+                       struct rte_tm_red_params *src =
+                               &src_wp->params.red_params[color];
+
+                       memcpy(dst, src, sizeof(*dst));
+               }
+}
+
+#else
+
+#define wred_profiles_set(dev)
+
+#endif
+
+static struct tm_shared_shaper *
+tm_tc_shared_shaper_get(struct rte_eth_dev *dev, struct tm_node *tc_node)
+{
+       return (tc_node->params.n_shared_shapers) ?
+               tm_shared_shaper_search(dev,
+                       tc_node->params.shared_shaper_id[0]) :
+               NULL;
+}
+
+static struct tm_shared_shaper *
+tm_subport_tc_shared_shaper_get(struct rte_eth_dev *dev,
+       struct tm_node *subport_node,
+       uint32_t tc_id)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_node_list *nl = &p->soft.tm.h.nodes;
+       struct tm_node *n;
+
+       TAILQ_FOREACH(n, nl, node) {
+               if (n->level != TM_NODE_LEVEL_TC ||
+                       n->parent_node->parent_node_id !=
+                               subport_node->node_id ||
+                       n->priority != tc_id)
+                       continue;
+
+               return tm_tc_shared_shaper_get(dev, n);
+       }
+
+       return NULL;
+}
+
+static int
+hierarchy_commit_check(struct rte_eth_dev *dev, struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_hierarchy *h = &p->soft.tm.h;
+       struct tm_node_list *nl = &h->nodes;
+       struct tm_shared_shaper_list *ssl = &h->shared_shapers;
+       struct tm_wred_profile_list *wpl = &h->wred_profiles;
+       struct tm_node *nr = tm_root_node_present(dev), *ns, *np, *nt, *nq;
+       struct tm_shared_shaper *ss;
+
+       uint32_t n_pipes_per_subport;
+
+       /* Root node exists. */
+       if (nr == NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_LEVEL_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* There is at least one subport, max is not exceeded. */
+       if (nr->n_children == 0 || nr->n_children > TM_MAX_SUBPORTS)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_LEVEL_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* There is at least one pipe. */
+       if (h->n_tm_nodes[TM_NODE_LEVEL_PIPE] == 0)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_LEVEL_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Number of pipes is the same for all subports. Maximum number of pipes
+        * per subport is not exceeded.
+        */
+       n_pipes_per_subport = h->n_tm_nodes[TM_NODE_LEVEL_PIPE] /
+               h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT];
+
+       if (n_pipes_per_subport > TM_MAX_PIPES_PER_SUBPORT)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       TAILQ_FOREACH(ns, nl, node) {
+               if (ns->level != TM_NODE_LEVEL_SUBPORT)
+                       continue;
+
+               if (ns->n_children != n_pipes_per_subport)
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+       }
+
+       /* Each pipe has exactly 4 TCs, with exactly one TC for each priority */
+       TAILQ_FOREACH(np, nl, node) {
+               uint32_t mask = 0, mask_expected =
+                       RTE_LEN2MASK(RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE,
+                               uint32_t);
+
+               if (np->level != TM_NODE_LEVEL_PIPE)
+                       continue;
+
+               if (np->n_children != RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE)
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+
+               TAILQ_FOREACH(nt, nl, node) {
+                       if (nt->level != TM_NODE_LEVEL_TC ||
+                               nt->parent_node_id != np->node_id)
+                               continue;
+
+                       mask |= 1 << nt->priority;
+               }
+
+               if (mask != mask_expected)
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+       }
+
+       /* Each TC has exactly 4 packet queues. */
+       TAILQ_FOREACH(nt, nl, node) {
+               if (nt->level != TM_NODE_LEVEL_TC)
+                       continue;
+
+               if (nt->n_children != RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS)
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+       }
+
+       /**
+        * Shared shapers:
+        *    -For each TC #i, all pipes in the same subport use the same
+        *     shared shaper (or no shared shaper) for their TC#i.
+        *    -Each shared shaper needs to have at least one user. All its
+        *     users have to be TC nodes with the same priority and the same
+        *     subport.
+        */
+       TAILQ_FOREACH(ns, nl, node) {
+               struct tm_shared_shaper *s[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
+               uint32_t id;
+
+               if (ns->level != TM_NODE_LEVEL_SUBPORT)
+                       continue;
+
+               for (id = 0; id < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; id++)
+                       s[id] = tm_subport_tc_shared_shaper_get(dev, ns, id);
+
+               TAILQ_FOREACH(nt, nl, node) {
+                       struct tm_shared_shaper *subport_ss, *tc_ss;
+
+                       if (nt->level != TM_NODE_LEVEL_TC ||
+                               nt->parent_node->parent_node_id !=
+                                       ns->node_id)
+                               continue;
+
+                       subport_ss = s[nt->priority];
+                       tc_ss = tm_tc_shared_shaper_get(dev, nt);
+
+                       if (subport_ss == NULL && tc_ss == NULL)
+                               continue;
+
+                       if ((subport_ss == NULL && tc_ss != NULL) ||
+                               (subport_ss != NULL && tc_ss == NULL) ||
+                               subport_ss->shared_shaper_id !=
+                                       tc_ss->shared_shaper_id)
+                               return -rte_tm_error_set(error,
+                                       EINVAL,
+                                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                                       NULL,
+                                       rte_strerror(EINVAL));
+               }
+       }
+
+       TAILQ_FOREACH(ss, ssl, node) {
+               struct tm_node *nt_any = tm_shared_shaper_get_tc(dev, ss);
+               uint32_t n_users = 0;
+
+               if (nt_any != NULL)
+                       TAILQ_FOREACH(nt, nl, node) {
+                               if (nt->level != TM_NODE_LEVEL_TC ||
+                                       nt->priority != nt_any->priority ||
+                                       nt->parent_node->parent_node_id !=
+                                       nt_any->parent_node->parent_node_id)
+                                       continue;
+
+                               n_users++;
+                       }
+
+               if (ss->n_users == 0 || ss->n_users != n_users)
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+       }
+
+       /* Not too many pipe profiles. */
+       if (pipe_profiles_generate(dev))
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /**
+        * WRED (when used, i.e. at least one WRED profile defined):
+        *    -Each WRED profile must have at least one user.
+        *    -All leaf nodes must have their private WRED context enabled.
+        *    -For each TC #i, all leaf nodes must use the same WRED profile
+        *     for their private WRED context.
+        */
+       if (h->n_wred_profiles) {
+               struct tm_wred_profile *wp;
+               struct tm_wred_profile *w[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
+               uint32_t id;
+
+               TAILQ_FOREACH(wp, wpl, node)
+                       if (wp->n_users == 0)
+                               return -rte_tm_error_set(error,
+                                       EINVAL,
+                                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                                       NULL,
+                                       rte_strerror(EINVAL));
+
+               for (id = 0; id < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; id++) {
+                       w[id] = tm_tc_wred_profile_get(dev, id);
+
+                       if (w[id] == NULL)
+                               return -rte_tm_error_set(error,
+                                       EINVAL,
+                                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                                       NULL,
+                                       rte_strerror(EINVAL));
+               }
+
+               TAILQ_FOREACH(nq, nl, node) {
+                       uint32_t id;
+
+                       if (nq->level != TM_NODE_LEVEL_QUEUE)
+                               continue;
+
+                       id = nq->parent_node->priority;
+
+                       if (nq->wred_profile == NULL ||
+                               nq->wred_profile->wred_profile_id !=
+                                       w[id]->wred_profile_id)
+                               return -rte_tm_error_set(error,
+                                       EINVAL,
+                                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                                       NULL,
+                                       rte_strerror(EINVAL));
+               }
+       }
+
+       return 0;
+}
+
+static void
+hierarchy_blueprints_create(struct rte_eth_dev *dev)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_params *t = &p->soft.tm.params;
+       struct tm_hierarchy *h = &p->soft.tm.h;
+
+       struct tm_node_list *nl = &h->nodes;
+       struct tm_node *root = tm_root_node_present(dev), *n;
+
+       uint32_t subport_id;
+
+       t->port_params = (struct rte_sched_port_params) {
+               .name = dev->data->name,
+               .socket = dev->data->numa_node,
+               .rate = root->shaper_profile->params.peak.rate,
+               .mtu = dev->data->mtu,
+               .frame_overhead =
+                       root->shaper_profile->params.pkt_length_adjust,
+               .n_subports_per_port = root->n_children,
+               .n_pipes_per_subport = h->n_tm_nodes[TM_NODE_LEVEL_PIPE] /
+                       h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT],
+               .qsize = {p->params.soft.tm.qsize[0],
+                       p->params.soft.tm.qsize[1],
+                       p->params.soft.tm.qsize[2],
+                       p->params.soft.tm.qsize[3],
+               },
+               .pipe_profiles = t->pipe_profiles,
+               .n_pipe_profiles = t->n_pipe_profiles,
+       };
+
+       wred_profiles_set(dev);
+
+       subport_id = 0;
+       TAILQ_FOREACH(n, nl, node) {
+               uint64_t tc_rate[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
+               uint32_t i;
+
+               if (n->level != TM_NODE_LEVEL_SUBPORT)
+                       continue;
+
+               for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+                       struct tm_shared_shaper *ss;
+                       struct tm_shaper_profile *sp;
+
+                       ss = tm_subport_tc_shared_shaper_get(dev, n, i);
+                       sp = (ss) ? tm_shaper_profile_search(dev,
+                               ss->shaper_profile_id) :
+                               n->shaper_profile;
+                       tc_rate[i] = sp->params.peak.rate;
+               }
+
+               t->subport_params[subport_id] =
+                       (struct rte_sched_subport_params) {
+                               .tb_rate = n->shaper_profile->params.peak.rate,
+                               .tb_size = n->shaper_profile->params.peak.size,
+
+                               .tc_rate = {tc_rate[0],
+                                       tc_rate[1],
+                                       tc_rate[2],
+                                       tc_rate[3],
+                       },
+                       .tc_period = SUBPORT_TC_PERIOD,
+               };
+
+               subport_id++;
+       }
+}
+
+/* Traffic manager hierarchy commit */
+static int
+pmd_tm_hierarchy_commit(struct rte_eth_dev *dev,
+       int clear_on_fail,
+       struct rte_tm_error *error)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       int status;
+
+       /* Checks */
+       if (p->soft.tm.hierarchy_frozen)
+               return -rte_tm_error_set(error,
+                       EBUSY,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(EBUSY));
+
+       status = hierarchy_commit_check(dev, error);
+       if (status) {
+               if (clear_on_fail) {
+                       tm_hierarchy_uninit(p);
+                       tm_hierarchy_init(p);
+               }
+
+               return status;
+       }
+
+       /* Create blueprints */
+       hierarchy_blueprints_create(dev);
+
+       /* Freeze hierarchy */
+       p->soft.tm.hierarchy_frozen = 1;
+
+       return 0;
+}
+
+#ifdef RTE_SCHED_SUBPORT_TC_OV
+
+static int
+update_pipe_weight(struct rte_eth_dev *dev, struct tm_node *np, uint32_t weight)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       uint32_t pipe_id = tm_node_pipe_id(dev, np);
+
+       struct tm_node *ns = np->parent_node;
+       uint32_t subport_id = tm_node_subport_id(dev, ns);
+
+       struct rte_sched_pipe_params *profile0 = pipe_profile_get(dev, np);
+       struct rte_sched_pipe_params profile1;
+       uint32_t pipe_profile_id;
+
+       /* Derive new pipe profile. */
+       memcpy(&profile1, profile0, sizeof(profile1));
+       profile1.tc_ov_weight = (uint8_t)weight;
+
+       /* Since implementation does not allow adding more pipe profiles after
+        * port configuration, the pipe configuration can be successfully
+        * updated only if the new profile is also part of the existing set of
+        * pipe profiles.
+        */
+       if (pipe_profile_exists(dev, &profile1, &pipe_profile_id) == 0)
+               return -1;
+
+       /* Update the pipe profile used by the current pipe. */
+       if (rte_sched_pipe_config(p->soft.tm.sched, subport_id, pipe_id,
+               (int32_t)pipe_profile_id))
+               return -1;
+
+       /* Commit changes. */
+       pipe_profile_mark(dev, subport_id, pipe_id, pipe_profile_id);
+       np->weight = weight;
+
+       return 0;
+}
+
+#endif
+
+static int
+update_queue_weight(struct rte_eth_dev *dev,
+       struct tm_node *nq, uint32_t weight)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       uint32_t queue_id = tm_node_queue_id(dev, nq);
+
+       struct tm_node *nt = nq->parent_node;
+       uint32_t tc_id = tm_node_tc_id(dev, nt);
+
+       struct tm_node *np = nt->parent_node;
+       uint32_t pipe_id = tm_node_pipe_id(dev, np);
+
+       struct tm_node *ns = np->parent_node;
+       uint32_t subport_id = tm_node_subport_id(dev, ns);
+
+       uint32_t pipe_queue_id =
+               tc_id * RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS + queue_id;
+
+       struct rte_sched_pipe_params *profile0 = pipe_profile_get(dev, np);
+       struct rte_sched_pipe_params profile1;
+       uint32_t pipe_profile_id;
+
+       /* Derive new pipe profile. */
+       memcpy(&profile1, profile0, sizeof(profile1));
+       profile1.wrr_weights[pipe_queue_id] = (uint8_t)weight;
+
+       /* Since implementation does not allow adding more pipe profiles after
+        * port configuration, the pipe configuration can be successfully
+        * updated only if the new profile is also part of the existing set
+        * of pipe profiles.
+        */
+       if (pipe_profile_exists(dev, &profile1, &pipe_profile_id) == 0)
+               return -1;
+
+       /* Update the pipe profile used by the current pipe. */
+       if (rte_sched_pipe_config(p->soft.tm.sched, subport_id, pipe_id,
+               (int32_t)pipe_profile_id))
+               return -1;
+
+       /* Commit changes. */
+       pipe_profile_mark(dev, subport_id, pipe_id, pipe_profile_id);
+       nq->weight = weight;
+
+       return 0;
+}
+
+/* Traffic manager node parent update */
+static int
+pmd_tm_node_parent_update(struct rte_eth_dev *dev,
+       uint32_t node_id,
+       uint32_t parent_node_id,
+       uint32_t priority,
+       uint32_t weight,
+       struct rte_tm_error *error)
+{
+       struct tm_node *n;
+
+       /* Port must be started and TM used. */
+       if (dev->data->dev_started == 0 && (tm_used(dev) == 0))
+               return -rte_tm_error_set(error,
+                       EBUSY,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(EBUSY));
+
+       /* Node must be valid */
+       n = tm_node_search(dev, node_id);
+       if (n == NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Parent node must be the same */
+       if (n->parent_node_id != parent_node_id)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PARENT_NODE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Priority must be the same */
+       if (n->priority != priority)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_PRIORITY,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* weight: must be 1 .. 255 */
+       if (weight == 0 || weight >= UINT8_MAX)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_WEIGHT,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       switch (n->level) {
+       case TM_NODE_LEVEL_PORT:
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_WEIGHT,
+                       NULL,
+                       rte_strerror(EINVAL));
+               /* fall-through */
+       case TM_NODE_LEVEL_SUBPORT:
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_WEIGHT,
+                       NULL,
+                       rte_strerror(EINVAL));
+               /* fall-through */
+       case TM_NODE_LEVEL_PIPE:
+#ifdef RTE_SCHED_SUBPORT_TC_OV
+               if (update_pipe_weight(dev, n, weight))
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+               return 0;
+#else
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_WEIGHT,
+                       NULL,
+                       rte_strerror(EINVAL));
+#endif
+               /* fall-through */
+       case TM_NODE_LEVEL_TC:
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_WEIGHT,
+                       NULL,
+                       rte_strerror(EINVAL));
+               /* fall-through */
+       case TM_NODE_LEVEL_QUEUE:
+               /* fall-through */
+       default:
+               if (update_queue_weight(dev, n, weight))
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+               return 0;
+       }
+}
+
+static int
+update_subport_rate(struct rte_eth_dev *dev,
+       struct tm_node *ns,
+       struct tm_shaper_profile *sp)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       uint32_t subport_id = tm_node_subport_id(dev, ns);
+
+       struct rte_sched_subport_params subport_params;
+
+       /* Derive new subport configuration. */
+       memcpy(&subport_params,
+               &p->soft.tm.params.subport_params[subport_id],
+               sizeof(subport_params));
+       subport_params.tb_rate = sp->params.peak.rate;
+       subport_params.tb_size = sp->params.peak.size;
+
+       /* Update the subport configuration. */
+       if (rte_sched_subport_config(p->soft.tm.sched, subport_id,
+               &subport_params))
+               return -1;
+
+       /* Commit changes. */
+       ns->shaper_profile->n_users--;
+
+       ns->shaper_profile = sp;
+       ns->params.shaper_profile_id = sp->shaper_profile_id;
+       sp->n_users++;
+
+       memcpy(&p->soft.tm.params.subport_params[subport_id],
+               &subport_params,
+               sizeof(subport_params));
+
+       return 0;
+}
+
+static int
+update_pipe_rate(struct rte_eth_dev *dev,
+       struct tm_node *np,
+       struct tm_shaper_profile *sp)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       uint32_t pipe_id = tm_node_pipe_id(dev, np);
+
+       struct tm_node *ns = np->parent_node;
+       uint32_t subport_id = tm_node_subport_id(dev, ns);
+
+       struct rte_sched_pipe_params *profile0 = pipe_profile_get(dev, np);
+       struct rte_sched_pipe_params profile1;
+       uint32_t pipe_profile_id;
+
+       /* Derive new pipe profile. */
+       memcpy(&profile1, profile0, sizeof(profile1));
+       profile1.tb_rate = sp->params.peak.rate;
+       profile1.tb_size = sp->params.peak.size;
+
+       /* Since implementation does not allow adding more pipe profiles after
+        * port configuration, the pipe configuration can be successfully
+        * updated only if the new profile is also part of the existing set of
+        * pipe profiles.
+        */
+       if (pipe_profile_exists(dev, &profile1, &pipe_profile_id) == 0)
+               return -1;
+
+       /* Update the pipe profile used by the current pipe. */
+       if (rte_sched_pipe_config(p->soft.tm.sched, subport_id, pipe_id,
+               (int32_t)pipe_profile_id))
+               return -1;
+
+       /* Commit changes. */
+       pipe_profile_mark(dev, subport_id, pipe_id, pipe_profile_id);
+       np->shaper_profile->n_users--;
+       np->shaper_profile = sp;
+       np->params.shaper_profile_id = sp->shaper_profile_id;
+       sp->n_users++;
+
+       return 0;
+}
+
+static int
+update_tc_rate(struct rte_eth_dev *dev,
+       struct tm_node *nt,
+       struct tm_shaper_profile *sp)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       uint32_t tc_id = tm_node_tc_id(dev, nt);
+
+       struct tm_node *np = nt->parent_node;
+       uint32_t pipe_id = tm_node_pipe_id(dev, np);
+
+       struct tm_node *ns = np->parent_node;
+       uint32_t subport_id = tm_node_subport_id(dev, ns);
+
+       struct rte_sched_pipe_params *profile0 = pipe_profile_get(dev, np);
+       struct rte_sched_pipe_params profile1;
+       uint32_t pipe_profile_id;
+
+       /* Derive new pipe profile. */
+       memcpy(&profile1, profile0, sizeof(profile1));
+       profile1.tc_rate[tc_id] = sp->params.peak.rate;
+
+       /* Since implementation does not allow adding more pipe profiles after
+        * port configuration, the pipe configuration can be successfully
+        * updated only if the new profile is also part of the existing set of
+        * pipe profiles.
+        */
+       if (pipe_profile_exists(dev, &profile1, &pipe_profile_id) == 0)
+               return -1;
+
+       /* Update the pipe profile used by the current pipe. */
+       if (rte_sched_pipe_config(p->soft.tm.sched, subport_id, pipe_id,
+               (int32_t)pipe_profile_id))
+               return -1;
+
+       /* Commit changes. */
+       pipe_profile_mark(dev, subport_id, pipe_id, pipe_profile_id);
+       nt->shaper_profile->n_users--;
+       nt->shaper_profile = sp;
+       nt->params.shaper_profile_id = sp->shaper_profile_id;
+       sp->n_users++;
+
+       return 0;
+}
+
+/* Traffic manager node shaper update */
+static int
+pmd_tm_node_shaper_update(struct rte_eth_dev *dev,
+       uint32_t node_id,
+       uint32_t shaper_profile_id,
+       struct rte_tm_error *error)
+{
+       struct tm_node *n;
+       struct tm_shaper_profile *sp;
+
+       /* Port must be started and TM used. */
+       if (dev->data->dev_started == 0 && (tm_used(dev) == 0))
+               return -rte_tm_error_set(error,
+                       EBUSY,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(EBUSY));
+
+       /* Node must be valid */
+       n = tm_node_search(dev, node_id);
+       if (n == NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       /* Shaper profile must be valid. */
+       sp = tm_shaper_profile_search(dev, shaper_profile_id);
+       if (sp == NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_SHAPER_PROFILE,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       switch (n->level) {
+       case TM_NODE_LEVEL_PORT:
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(EINVAL));
+               /* fall-through */
+       case TM_NODE_LEVEL_SUBPORT:
+               if (update_subport_rate(dev, n, sp))
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+               return 0;
+               /* fall-through */
+       case TM_NODE_LEVEL_PIPE:
+               if (update_pipe_rate(dev, n, sp))
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+               return 0;
+               /* fall-through */
+       case TM_NODE_LEVEL_TC:
+               if (update_tc_rate(dev, n, sp))
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+               return 0;
+               /* fall-through */
+       case TM_NODE_LEVEL_QUEUE:
+               /* fall-through */
+       default:
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(EINVAL));
+       }
+}
+
+static inline uint32_t
+tm_port_queue_id(struct rte_eth_dev *dev,
+       uint32_t port_subport_id,
+       uint32_t subport_pipe_id,
+       uint32_t pipe_tc_id,
+       uint32_t tc_queue_id)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_hierarchy *h = &p->soft.tm.h;
+       uint32_t n_pipes_per_subport = h->n_tm_nodes[TM_NODE_LEVEL_PIPE] /
+                       h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT];
+
+       uint32_t port_pipe_id =
+               port_subport_id * n_pipes_per_subport + subport_pipe_id;
+       uint32_t port_tc_id =
+               port_pipe_id * RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE + pipe_tc_id;
+       uint32_t port_queue_id =
+               port_tc_id * RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS + tc_queue_id;
+
+       return port_queue_id;
+}
+
+static int
+read_port_stats(struct rte_eth_dev *dev,
+       struct tm_node *nr,
+       struct rte_tm_node_stats *stats,
+       uint64_t *stats_mask,
+       int clear)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct tm_hierarchy *h = &p->soft.tm.h;
+       uint32_t n_subports_per_port = h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT];
+       uint32_t subport_id;
+
+       for (subport_id = 0; subport_id < n_subports_per_port; subport_id++) {
+               struct rte_sched_subport_stats s;
+               uint32_t tc_ov, id;
+
+               /* Stats read */
+               int status = rte_sched_subport_read_stats(
+                       p->soft.tm.sched,
+                       subport_id,
+                       &s,
+                       &tc_ov);
+               if (status)
+                       return status;
+
+               /* Stats accumulate */
+               for (id = 0; id < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; id++) {
+                       nr->stats.n_pkts +=
+                               s.n_pkts_tc[id] - s.n_pkts_tc_dropped[id];
+                       nr->stats.n_bytes +=
+                               s.n_bytes_tc[id] - s.n_bytes_tc_dropped[id];
+                       nr->stats.leaf.n_pkts_dropped[RTE_TM_GREEN] +=
+                               s.n_pkts_tc_dropped[id];
+                       nr->stats.leaf.n_bytes_dropped[RTE_TM_GREEN] +=
+                               s.n_bytes_tc_dropped[id];
+               }
+       }
+
+       /* Stats copy */
+       if (stats)
+               memcpy(stats, &nr->stats, sizeof(*stats));
+
+       if (stats_mask)
+               *stats_mask = STATS_MASK_DEFAULT;
+
+       /* Stats clear */
+       if (clear)
+               memset(&nr->stats, 0, sizeof(nr->stats));
+
+       return 0;
+}
+
+static int
+read_subport_stats(struct rte_eth_dev *dev,
+       struct tm_node *ns,
+       struct rte_tm_node_stats *stats,
+       uint64_t *stats_mask,
+       int clear)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       uint32_t subport_id = tm_node_subport_id(dev, ns);
+       struct rte_sched_subport_stats s;
+       uint32_t tc_ov, tc_id;
+
+       /* Stats read */
+       int status = rte_sched_subport_read_stats(
+               p->soft.tm.sched,
+               subport_id,
+               &s,
+               &tc_ov);
+       if (status)
+               return status;
+
+       /* Stats accumulate */
+       for (tc_id = 0; tc_id < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; tc_id++) {
+               ns->stats.n_pkts +=
+                       s.n_pkts_tc[tc_id] - s.n_pkts_tc_dropped[tc_id];
+               ns->stats.n_bytes +=
+                       s.n_bytes_tc[tc_id] - s.n_bytes_tc_dropped[tc_id];
+               ns->stats.leaf.n_pkts_dropped[RTE_TM_GREEN] +=
+                       s.n_pkts_tc_dropped[tc_id];
+               ns->stats.leaf.n_bytes_dropped[RTE_TM_GREEN] +=
+                       s.n_bytes_tc_dropped[tc_id];
+       }
+
+       /* Stats copy */
+       if (stats)
+               memcpy(stats, &ns->stats, sizeof(*stats));
+
+       if (stats_mask)
+               *stats_mask = STATS_MASK_DEFAULT;
+
+       /* Stats clear */
+       if (clear)
+               memset(&ns->stats, 0, sizeof(ns->stats));
+
+       return 0;
+}
+
+static int
+read_pipe_stats(struct rte_eth_dev *dev,
+       struct tm_node *np,
+       struct rte_tm_node_stats *stats,
+       uint64_t *stats_mask,
+       int clear)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+
+       uint32_t pipe_id = tm_node_pipe_id(dev, np);
+
+       struct tm_node *ns = np->parent_node;
+       uint32_t subport_id = tm_node_subport_id(dev, ns);
+
+       uint32_t i;
+
+       /* Stats read */
+       for (i = 0; i < RTE_SCHED_QUEUES_PER_PIPE; i++) {
+               struct rte_sched_queue_stats s;
+               uint16_t qlen;
+
+               uint32_t qid = tm_port_queue_id(dev,
+                       subport_id,
+                       pipe_id,
+                       i / RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS,
+                       i % RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS);
+
+               int status = rte_sched_queue_read_stats(
+                       p->soft.tm.sched,
+                       qid,
+                       &s,
+                       &qlen);
+               if (status)
+                       return status;
+
+               /* Stats accumulate */
+               np->stats.n_pkts += s.n_pkts - s.n_pkts_dropped;
+               np->stats.n_bytes += s.n_bytes - s.n_bytes_dropped;
+               np->stats.leaf.n_pkts_dropped[RTE_TM_GREEN] += s.n_pkts_dropped;
+               np->stats.leaf.n_bytes_dropped[RTE_TM_GREEN] +=
+                       s.n_bytes_dropped;
+               np->stats.leaf.n_pkts_queued = qlen;
+       }
+
+       /* Stats copy */
+       if (stats)
+               memcpy(stats, &np->stats, sizeof(*stats));
+
+       if (stats_mask)
+               *stats_mask = STATS_MASK_DEFAULT;
+
+       /* Stats clear */
+       if (clear)
+               memset(&np->stats, 0, sizeof(np->stats));
+
+       return 0;
+}
+
+static int
+read_tc_stats(struct rte_eth_dev *dev,
+       struct tm_node *nt,
+       struct rte_tm_node_stats *stats,
+       uint64_t *stats_mask,
+       int clear)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+
+       uint32_t tc_id = tm_node_tc_id(dev, nt);
+
+       struct tm_node *np = nt->parent_node;
+       uint32_t pipe_id = tm_node_pipe_id(dev, np);
+
+       struct tm_node *ns = np->parent_node;
+       uint32_t subport_id = tm_node_subport_id(dev, ns);
+
+       uint32_t i;
+
+       /* Stats read */
+       for (i = 0; i < RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS; i++) {
+               struct rte_sched_queue_stats s;
+               uint16_t qlen;
+
+               uint32_t qid = tm_port_queue_id(dev,
+                       subport_id,
+                       pipe_id,
+                       tc_id,
+                       i);
+
+               int status = rte_sched_queue_read_stats(
+                       p->soft.tm.sched,
+                       qid,
+                       &s,
+                       &qlen);
+               if (status)
+                       return status;
+
+               /* Stats accumulate */
+               nt->stats.n_pkts += s.n_pkts - s.n_pkts_dropped;
+               nt->stats.n_bytes += s.n_bytes - s.n_bytes_dropped;
+               nt->stats.leaf.n_pkts_dropped[RTE_TM_GREEN] += s.n_pkts_dropped;
+               nt->stats.leaf.n_bytes_dropped[RTE_TM_GREEN] +=
+                       s.n_bytes_dropped;
+               nt->stats.leaf.n_pkts_queued = qlen;
+       }
+
+       /* Stats copy */
+       if (stats)
+               memcpy(stats, &nt->stats, sizeof(*stats));
+
+       if (stats_mask)
+               *stats_mask = STATS_MASK_DEFAULT;
+
+       /* Stats clear */
+       if (clear)
+               memset(&nt->stats, 0, sizeof(nt->stats));
+
+       return 0;
+}
+
+static int
+read_queue_stats(struct rte_eth_dev *dev,
+       struct tm_node *nq,
+       struct rte_tm_node_stats *stats,
+       uint64_t *stats_mask,
+       int clear)
+{
+       struct pmd_internals *p = dev->data->dev_private;
+       struct rte_sched_queue_stats s;
+       uint16_t qlen;
+
+       uint32_t queue_id = tm_node_queue_id(dev, nq);
+
+       struct tm_node *nt = nq->parent_node;
+       uint32_t tc_id = tm_node_tc_id(dev, nt);
+
+       struct tm_node *np = nt->parent_node;
+       uint32_t pipe_id = tm_node_pipe_id(dev, np);
+
+       struct tm_node *ns = np->parent_node;
+       uint32_t subport_id = tm_node_subport_id(dev, ns);
+
+       /* Stats read */
+       uint32_t qid = tm_port_queue_id(dev,
+               subport_id,
+               pipe_id,
+               tc_id,
+               queue_id);
+
+       int status = rte_sched_queue_read_stats(
+               p->soft.tm.sched,
+               qid,
+               &s,
+               &qlen);
+       if (status)
+               return status;
+
+       /* Stats accumulate */
+       nq->stats.n_pkts += s.n_pkts - s.n_pkts_dropped;
+       nq->stats.n_bytes += s.n_bytes - s.n_bytes_dropped;
+       nq->stats.leaf.n_pkts_dropped[RTE_TM_GREEN] += s.n_pkts_dropped;
+       nq->stats.leaf.n_bytes_dropped[RTE_TM_GREEN] +=
+               s.n_bytes_dropped;
+       nq->stats.leaf.n_pkts_queued = qlen;
+
+       /* Stats copy */
+       if (stats)
+               memcpy(stats, &nq->stats, sizeof(*stats));
+
+       if (stats_mask)
+               *stats_mask = STATS_MASK_QUEUE;
+
+       /* Stats clear */
+       if (clear)
+               memset(&nq->stats, 0, sizeof(nq->stats));
+
+       return 0;
+}
+
+/* Traffic manager read stats counters for specific node */
+static int
+pmd_tm_node_stats_read(struct rte_eth_dev *dev,
+       uint32_t node_id,
+       struct rte_tm_node_stats *stats,
+       uint64_t *stats_mask,
+       int clear,
+       struct rte_tm_error *error)
+{
+       struct tm_node *n;
+
+       /* Port must be started and TM used. */
+       if (dev->data->dev_started == 0 && (tm_used(dev) == 0))
+               return -rte_tm_error_set(error,
+                       EBUSY,
+                       RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       rte_strerror(EBUSY));
+
+       /* Node must be valid */
+       n = tm_node_search(dev, node_id);
+       if (n == NULL)
+               return -rte_tm_error_set(error,
+                       EINVAL,
+                       RTE_TM_ERROR_TYPE_NODE_ID,
+                       NULL,
+                       rte_strerror(EINVAL));
+
+       switch (n->level) {
+       case TM_NODE_LEVEL_PORT:
+               if (read_port_stats(dev, n, stats, stats_mask, clear))
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+               return 0;
+
+       case TM_NODE_LEVEL_SUBPORT:
+               if (read_subport_stats(dev, n, stats, stats_mask, clear))
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+               return 0;
+
+       case TM_NODE_LEVEL_PIPE:
+               if (read_pipe_stats(dev, n, stats, stats_mask, clear))
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+               return 0;
+
+       case TM_NODE_LEVEL_TC:
+               if (read_tc_stats(dev, n, stats, stats_mask, clear))
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+               return 0;
+
+       case TM_NODE_LEVEL_QUEUE:
+       default:
+               if (read_queue_stats(dev, n, stats, stats_mask, clear))
+                       return -rte_tm_error_set(error,
+                               EINVAL,
+                               RTE_TM_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               rte_strerror(EINVAL));
+               return 0;
+       }
+}
+
+const struct rte_tm_ops pmd_tm_ops = {
+       .node_type_get = pmd_tm_node_type_get,
+       .capabilities_get = pmd_tm_capabilities_get,
+       .level_capabilities_get = pmd_tm_level_capabilities_get,
+       .node_capabilities_get = pmd_tm_node_capabilities_get,
+
+       .wred_profile_add = pmd_tm_wred_profile_add,
+       .wred_profile_delete = pmd_tm_wred_profile_delete,
+       .shared_wred_context_add_update = NULL,
+       .shared_wred_context_delete = NULL,
+
+       .shaper_profile_add = pmd_tm_shaper_profile_add,
+       .shaper_profile_delete = pmd_tm_shaper_profile_delete,
+       .shared_shaper_add_update = pmd_tm_shared_shaper_add_update,
+       .shared_shaper_delete = pmd_tm_shared_shaper_delete,
+
+       .node_add = pmd_tm_node_add,
+       .node_delete = pmd_tm_node_delete,
+       .node_suspend = NULL,
+       .node_resume = NULL,
+       .hierarchy_commit = pmd_tm_hierarchy_commit,
+
+       .node_parent_update = pmd_tm_node_parent_update,
+       .node_shaper_update = pmd_tm_node_shaper_update,
+       .node_shared_shaper_update = NULL,
+       .node_stats_update = NULL,
+       .node_wfq_weight_mode_update = NULL,
+       .node_cman_update = NULL,
+       .node_wred_context_update = NULL,
+       .node_shared_wred_context_update = NULL,
+
+       .node_stats_read = pmd_tm_node_stats_read,
 };