13 files changed, 1042 insertions, 116 deletions

diff --git a/drivers/net/ethernet/netronome/nfp/Makefile b/drivers/net/ethernet/netronome/nfp/Makefile
index 190e8b56a41f..47c708f08ade 100644
--- a/drivers/net/ethernet/netronome/nfp/Makefile
+++ b/drivers/net/ethernet/netronome/nfp/Makefile
@@ -56,6 +56,7 @@ endif
 
 ifeq ($(CONFIG_NFP_APP_ABM_NIC),y)
 nfp-objs += \
+	    abm/cls.o \
 	    abm/ctrl.o \
 	    abm/qdisc.o \
 	    abm/main.o
diff --git a/drivers/net/ethernet/netronome/nfp/abm/cls.c b/drivers/net/ethernet/netronome/nfp/abm/cls.c
new file mode 100644
index 000000000000..9852080cf454
--- /dev/null
+++ b/drivers/net/ethernet/netronome/nfp/abm/cls.c
@@ -0,0 +1,283 @@
+// SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
+/* Copyright (C) 2018 Netronome Systems, Inc. */
+
+#include <linux/bitfield.h>
+#include <net/pkt_cls.h>
+
+#include "../nfpcore/nfp_cpp.h"
+#include "../nfp_app.h"
+#include "../nfp_net_repr.h"
+#include "main.h"
+
+struct nfp_abm_u32_match {
+	u32 handle;
+	u32 band;
+	u8 mask;
+	u8 val;
+	struct list_head list;
+};
+
+static bool
+nfp_abm_u32_check_knode(struct nfp_abm *abm, struct tc_cls_u32_knode *knode,
+			__be16 proto, struct netlink_ext_ack *extack)
+{
+	struct tc_u32_key *k;
+	unsigned int tos_off;
+
+	if (knode->exts && tcf_exts_has_actions(knode->exts)) {
+		NL_SET_ERR_MSG_MOD(extack, "action offload not supported");
+		return false;
+	}
+	if (knode->link_handle) {
+		NL_SET_ERR_MSG_MOD(extack, "linking not supported");
+		return false;
+	}
+	if (knode->sel->flags != TC_U32_TERMINAL) {
+		NL_SET_ERR_MSG_MOD(extack,
+				   "flags must be equal to TC_U32_TERMINAL");
+		return false;
+	}
+	if (knode->sel->off || knode->sel->offshift || knode->sel->offmask ||
+	    knode->sel->offoff || knode->fshift) {
+		NL_SET_ERR_MSG_MOD(extack, "variable offseting not supported");
+		return false;
+	}
+	if (knode->sel->hoff || knode->sel->hmask) {
+		NL_SET_ERR_MSG_MOD(extack, "hashing not supported");
+		return false;
+	}
+	if (knode->val || knode->mask) {
+		NL_SET_ERR_MSG_MOD(extack, "matching on mark not supported");
+		return false;
+	}
+	if (knode->res && knode->res->class) {
+		NL_SET_ERR_MSG_MOD(extack, "setting non-0 class not supported");
+		return false;
+	}
+	if (knode->res && knode->res->classid >= abm->num_bands) {
+		NL_SET_ERR_MSG_MOD(extack,
+				   "classid higher than number of bands");
+		return false;
+	}
+	if (knode->sel->nkeys != 1) {
+		NL_SET_ERR_MSG_MOD(extack, "exactly one key required");
+		return false;
+	}
+
+	switch (proto) {
+	case htons(ETH_P_IP):
+		tos_off = 16;
+		break;
+	case htons(ETH_P_IPV6):
+		tos_off = 20;
+		break;
+	default:
+		NL_SET_ERR_MSG_MOD(extack, "only IP and IPv6 supported as filter protocol");
+		return false;
+	}
+
+	k = &knode->sel->keys[0];
+	if (k->offmask) {
+		NL_SET_ERR_MSG_MOD(extack, "offset mask - variable offseting not supported");
+		return false;
+	}
+	if (k->off) {
+		NL_SET_ERR_MSG_MOD(extack, "only DSCP fields can be matched");
+		return false;
+	}
+	if (k->val & ~k->mask) {
+		NL_SET_ERR_MSG_MOD(extack, "mask does not cover the key");
+		return false;
+	}
+	if (be32_to_cpu(k->mask) >> tos_off & ~abm->dscp_mask) {
+		NL_SET_ERR_MSG_MOD(extack, "only high DSCP class selector bits can be used");
+		nfp_err(abm->app->cpp,
+			"u32 offload: requested mask %x FW can support only %x\n",
+			be32_to_cpu(k->mask) >> tos_off, abm->dscp_mask);
+		return false;
+	}
+
+	return true;
+}
+
+/* This filter list -> map conversion is O(n * m), we expect single digit or
+ * low double digit number of prios and likewise for the filters.  Also u32
+ * doesn't report stats, so it's really only setup time cost.
+ */
+static unsigned int
+nfp_abm_find_band_for_prio(struct nfp_abm_link *alink, unsigned int prio)
+{
+	struct nfp_abm_u32_match *iter;
+
+	list_for_each_entry(iter, &alink->dscp_map, list)
+		if ((prio & iter->mask) == iter->val)
+			return iter->band;
+
+	return alink->def_band;
+}
+
+static int nfp_abm_update_band_map(struct nfp_abm_link *alink)
+{
+	unsigned int i, bits_per_prio, prios_per_word, base_shift;
+	struct nfp_abm *abm = alink->abm;
+	u32 field_mask;
+
+	alink->has_prio = !list_empty(&alink->dscp_map);
+
+	bits_per_prio = roundup_pow_of_two(order_base_2(abm->num_bands));
+	field_mask = (1 << bits_per_prio) - 1;
+	prios_per_word = sizeof(u32) * BITS_PER_BYTE / bits_per_prio;
+
+	/* FW mask applies from top bits */
+	base_shift = 8 - order_base_2(abm->num_prios);
+
+	for (i = 0; i < abm->num_prios; i++) {
+		unsigned int offset;
+		u32 *word;
+		u8 band;
+
+		word = &alink->prio_map[i / prios_per_word];
+		offset = (i % prios_per_word) * bits_per_prio;
+
+		band = nfp_abm_find_band_for_prio(alink, i << base_shift);
+
+		*word &= ~(field_mask << offset);
+		*word |= band << offset;
+	}
+
+	/* Qdisc offload status may change if has_prio changed */
+	nfp_abm_qdisc_offload_update(alink);
+
+	return nfp_abm_ctrl_prio_map_update(alink, alink->prio_map);
+}
+
+static void
+nfp_abm_u32_knode_delete(struct nfp_abm_link *alink,
+			 struct tc_cls_u32_knode *knode)
+{
+	struct nfp_abm_u32_match *iter;
+
+	list_for_each_entry(iter, &alink->dscp_map, list)
+		if (iter->handle == knode->handle) {
+			list_del(&iter->list);
+			kfree(iter);
+			nfp_abm_update_band_map(alink);
+			return;
+		}
+}
+
+static int
+nfp_abm_u32_knode_replace(struct nfp_abm_link *alink,
+			  struct tc_cls_u32_knode *knode,
+			  __be16 proto, struct netlink_ext_ack *extack)
+{
+	struct nfp_abm_u32_match *match = NULL, *iter;
+	unsigned int tos_off;
+	u8 mask, val;
+	int err;
+
+	if (!nfp_abm_u32_check_knode(alink->abm, knode, proto, extack))
+		goto err_delete;
+
+	tos_off = proto == htons(ETH_P_IP) ? 16 : 20;
+
+	/* Extract the DSCP Class Selector bits */
+	val = be32_to_cpu(knode->sel->keys[0].val) >> tos_off & 0xff;
+	mask = be32_to_cpu(knode->sel->keys[0].mask) >> tos_off & 0xff;
+
+	/* Check if there is no conflicting mapping and find match by handle */
+	list_for_each_entry(iter, &alink->dscp_map, list) {
+		u32 cmask;
+
+		if (iter->handle == knode->handle) {
+			match = iter;
+			continue;
+		}
+
+		cmask = iter->mask & mask;
+		if ((iter->val & cmask) == (val & cmask) &&
+		    iter->band != knode->res->classid) {
+			NL_SET_ERR_MSG_MOD(extack, "conflict with already offloaded filter");
+			goto err_delete;
+		}
+	}
+
+	if (!match) {
+		match = kzalloc(sizeof(*match), GFP_KERNEL);
+		if (!match)
+			return -ENOMEM;
+		list_add(&match->list, &alink->dscp_map);
+	}
+	match->handle = knode->handle;
+	match->band = knode->res->classid;
+	match->mask = mask;
+	match->val = val;
+
+	err = nfp_abm_update_band_map(alink);
+	if (err)
+		goto err_delete;
+
+	return 0;
+
+err_delete:
+	nfp_abm_u32_knode_delete(alink, knode);
+	return -EOPNOTSUPP;
+}
+
+static int nfp_abm_setup_tc_block_cb(enum tc_setup_type type,
+				     void *type_data, void *cb_priv)
+{
+	struct tc_cls_u32_offload *cls_u32 = type_data;
+	struct nfp_repr *repr = cb_priv;
+	struct nfp_abm_link *alink;
+
+	alink = repr->app_priv;
+
+	if (type != TC_SETUP_CLSU32) {
+		NL_SET_ERR_MSG_MOD(cls_u32->common.extack,
+				   "only offload of u32 classifier supported");
+		return -EOPNOTSUPP;
+	}
+	if (!tc_cls_can_offload_and_chain0(repr->netdev, &cls_u32->common))
+		return -EOPNOTSUPP;
+
+	if (cls_u32->common.protocol != htons(ETH_P_IP) &&
+	    cls_u32->common.protocol != htons(ETH_P_IPV6)) {
+		NL_SET_ERR_MSG_MOD(cls_u32->common.extack,
+				   "only IP and IPv6 supported as filter protocol");
+		return -EOPNOTSUPP;
+	}
+
+	switch (cls_u32->command) {
+	case TC_CLSU32_NEW_KNODE:
+	case TC_CLSU32_REPLACE_KNODE:
+		return nfp_abm_u32_knode_replace(alink, &cls_u32->knode,
+						 cls_u32->common.protocol,
+						 cls_u32->common.extack);
+	case TC_CLSU32_DELETE_KNODE:
+		nfp_abm_u32_knode_delete(alink, &cls_u32->knode);
+		return 0;
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
+int nfp_abm_setup_cls_block(struct net_device *netdev, struct nfp_repr *repr,
+			    struct tc_block_offload *f)
+{
+	if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
+		return -EOPNOTSUPP;
+
+	switch (f->command) {
+	case TC_BLOCK_BIND:
+		return tcf_block_cb_register(f->block,
+					     nfp_abm_setup_tc_block_cb,
+					     repr, repr, f->extack);
+	case TC_BLOCK_UNBIND:
+		tcf_block_cb_unregister(f->block, nfp_abm_setup_tc_block_cb,
+					repr);
+		return 0;
+	default:
+		return -EOPNOTSUPP;
+	}
+}
diff --git a/drivers/net/ethernet/netronome/nfp/abm/ctrl.c b/drivers/net/ethernet/netronome/nfp/abm/ctrl.c
index 1629b07f727b..ad6c2a621c7a 100644
--- a/drivers/net/ethernet/netronome/nfp/abm/ctrl.c
+++ b/drivers/net/ethernet/netronome/nfp/abm/ctrl.c
@@ -1,7 +1,9 @@
 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
 /* Copyright (C) 2018 Netronome Systems, Inc. */
 
+#include <linux/bitops.h>
 #include <linux/kernel.h>
+#include <linux/log2.h>
 
 #include "../nfpcore/nfp_cpp.h"
 #include "../nfpcore/nfp_nffw.h"
@@ -11,38 +13,56 @@
 #include "../nfp_net.h"
 #include "main.h"
 
-#define NFP_QLVL_SYM_NAME	"_abi_nfd_out_q_lvls_%u"
+#define NFP_NUM_PRIOS_SYM_NAME	"_abi_pci_dscp_num_prio_%u"
+#define NFP_NUM_BANDS_SYM_NAME	"_abi_pci_dscp_num_band_%u"
+#define NFP_ACT_MASK_SYM_NAME	"_abi_nfd_out_q_actions_%u"
+
+#define NFP_QLVL_SYM_NAME	"_abi_nfd_out_q_lvls_%u%s"
 #define NFP_QLVL_STRIDE		16
 #define NFP_QLVL_BLOG_BYTES	0
 #define NFP_QLVL_BLOG_PKTS	4
 #define NFP_QLVL_THRS		8
+#define NFP_QLVL_ACT		12
 
-#define NFP_QMSTAT_SYM_NAME	"_abi_nfdqm%u_stats"
+#define NFP_QMSTAT_SYM_NAME	"_abi_nfdqm%u_stats%s"
 #define NFP_QMSTAT_STRIDE	32
 #define NFP_QMSTAT_NON_STO	0
 #define NFP_QMSTAT_STO		8
 #define NFP_QMSTAT_DROP		16
 #define NFP_QMSTAT_ECN		24
 
+#define NFP_Q_STAT_SYM_NAME	"_abi_nfd_rxq_stats%u%s"
+#define NFP_Q_STAT_STRIDE	16
+#define NFP_Q_STAT_PKTS		0
+#define NFP_Q_STAT_BYTES	8
+
+#define NFP_NET_ABM_MBOX_CMD		NFP_NET_CFG_MBOX_SIMPLE_CMD
+#define NFP_NET_ABM_MBOX_RET		NFP_NET_CFG_MBOX_SIMPLE_RET
+#define NFP_NET_ABM_MBOX_DATALEN	NFP_NET_CFG_MBOX_SIMPLE_VAL
+#define NFP_NET_ABM_MBOX_RESERVED	(NFP_NET_CFG_MBOX_SIMPLE_VAL + 4)
+#define NFP_NET_ABM_MBOX_DATA		(NFP_NET_CFG_MBOX_SIMPLE_VAL + 8)
+
 static int
 nfp_abm_ctrl_stat(struct nfp_abm_link *alink, const struct nfp_rtsym *sym,
-		  unsigned int stride, unsigned int offset, unsigned int i,
-		  bool is_u64, u64 *res)
+		  unsigned int stride, unsigned int offset, unsigned int band,
+		  unsigned int queue, bool is_u64, u64 *res)
 {
 	struct nfp_cpp *cpp = alink->abm->app->cpp;
 	u64 val, sym_offset;
+	unsigned int qid;
 	u32 val32;
 	int err;
 
-	sym_offset = (alink->queue_base + i) * stride + offset;
+	qid = band * NFP_NET_MAX_RX_RINGS + alink->queue_base + queue;
+
+	sym_offset = qid * stride + offset;
 	if (is_u64)
 		err = __nfp_rtsym_readq(cpp, sym, 3, 0, sym_offset, &val);
 	else
 		err = __nfp_rtsym_readl(cpp, sym, 3, 0, sym_offset, &val32);
 	if (err) {
-		nfp_err(cpp,
-			"RED offload reading stat failed on vNIC %d queue %d\n",
-			alink->id, i);
+		nfp_err(cpp, "RED offload reading stat failed on vNIC %d band %d queue %d (+ %d)\n",
+			alink->id, band, queue, alink->queue_base);
 		return err;
 	}
 
@@ -73,81 +93,156 @@ int __nfp_abm_ctrl_set_q_lvl(struct nfp_abm *abm, unsigned int id, u32 val)
 	return 0;
 }
 
-int nfp_abm_ctrl_set_q_lvl(struct nfp_abm_link *alink, unsigned int queue,
-			   u32 val)
+int nfp_abm_ctrl_set_q_lvl(struct nfp_abm_link *alink, unsigned int band,
+			   unsigned int queue, u32 val)
 {
 	unsigned int threshold;
 
-	threshold = alink->queue_base + queue;
+	threshold = band * NFP_NET_MAX_RX_RINGS + alink->queue_base + queue;
 
 	return __nfp_abm_ctrl_set_q_lvl(alink->abm, threshold, val);
 }
 
-u64 nfp_abm_ctrl_stat_non_sto(struct nfp_abm_link *alink, unsigned int i)
+int __nfp_abm_ctrl_set_q_act(struct nfp_abm *abm, unsigned int id,
+			     enum nfp_abm_q_action act)
 {
-	u64 val;
+	struct nfp_cpp *cpp = abm->app->cpp;
+	u64 sym_offset;
+	int err;
 
-	if (nfp_abm_ctrl_stat(alink, alink->abm->qm_stats, NFP_QMSTAT_STRIDE,
-			      NFP_QMSTAT_NON_STO, i, true, &val))
+	if (abm->actions[id] == act)
 		return 0;
-	return val;
+
+	sym_offset = id * NFP_QLVL_STRIDE + NFP_QLVL_ACT;
+	err = __nfp_rtsym_writel(cpp, abm->q_lvls, 4, 0, sym_offset, act);
+	if (err) {
+		nfp_err(cpp,
+			"RED offload setting action failed on subqueue %d\n",
+			id);
+		return err;
+	}
+
+	abm->actions[id] = act;
+	return 0;
+}
+
+int nfp_abm_ctrl_set_q_act(struct nfp_abm_link *alink, unsigned int band,
+			   unsigned int queue, enum nfp_abm_q_action act)
+{
+	unsigned int qid;
+
+	qid = band * NFP_NET_MAX_RX_RINGS + alink->queue_base + queue;
+
+	return __nfp_abm_ctrl_set_q_act(alink->abm, qid, act);
+}
+
+u64 nfp_abm_ctrl_stat_non_sto(struct nfp_abm_link *alink, unsigned int queue)
+{
+	unsigned int band;
+	u64 val, sum = 0;
+
+	for (band = 0; band < alink->abm->num_bands; band++) {
+		if (nfp_abm_ctrl_stat(alink, alink->abm->qm_stats,
+				      NFP_QMSTAT_STRIDE, NFP_QMSTAT_NON_STO,
+				      band, queue, true, &val))
+			return 0;
+		sum += val;
+	}
+
+	return sum;
+}
+
+u64 nfp_abm_ctrl_stat_sto(struct nfp_abm_link *alink, unsigned int queue)
+{
+	unsigned int band;
+	u64 val, sum = 0;
+
+	for (band = 0; band < alink->abm->num_bands; band++) {
+		if (nfp_abm_ctrl_stat(alink, alink->abm->qm_stats,
+				      NFP_QMSTAT_STRIDE, NFP_QMSTAT_STO,
+				      band, queue, true, &val))
+			return 0;
+		sum += val;
+	}
+
+	return sum;
 }
 
-u64 nfp_abm_ctrl_stat_sto(struct nfp_abm_link *alink, unsigned int i)
+static int
+nfp_abm_ctrl_stat_basic(struct nfp_abm_link *alink, unsigned int band,
+			unsigned int queue, unsigned int off, u64 *val)
 {
-	u64 val;
+	if (!nfp_abm_has_prio(alink->abm)) {
+		if (!band) {
+			unsigned int id = alink->queue_base + queue;
+
+			*val = nn_readq(alink->vnic,
+					NFP_NET_CFG_RXR_STATS(id) + off);
+		} else {
+			*val = 0;
+		}
 
-	if (nfp_abm_ctrl_stat(alink, alink->abm->qm_stats, NFP_QMSTAT_STRIDE,
-			      NFP_QMSTAT_STO, i, true, &val))
 		return 0;
-	return val;
+	} else {
+		return nfp_abm_ctrl_stat(alink, alink->abm->q_stats,
+					 NFP_Q_STAT_STRIDE, off, band, queue,
+					 true, val);
+	}
 }
 
-int nfp_abm_ctrl_read_q_stats(struct nfp_abm_link *alink, unsigned int i,
-			      struct nfp_alink_stats *stats)
+int nfp_abm_ctrl_read_q_stats(struct nfp_abm_link *alink, unsigned int band,
+			      unsigned int queue, struct nfp_alink_stats *stats)
 {
 	int err;
 
-	stats->tx_pkts = nn_readq(alink->vnic, NFP_NET_CFG_RXR_STATS(i));
-	stats->tx_bytes = nn_readq(alink->vnic, NFP_NET_CFG_RXR_STATS(i) + 8);
+	err = nfp_abm_ctrl_stat_basic(alink, band, queue, NFP_Q_STAT_PKTS,
+				      &stats->tx_pkts);
+	if (err)
+		return err;
+
+	err = nfp_abm_ctrl_stat_basic(alink, band, queue, NFP_Q_STAT_BYTES,
+				      &stats->tx_bytes);
+	if (err)
+		return err;
 
-	err = nfp_abm_ctrl_stat(alink, alink->abm->q_lvls,
-				NFP_QLVL_STRIDE, NFP_QLVL_BLOG_BYTES,
-				i, false, &stats->backlog_bytes);
+	err = nfp_abm_ctrl_stat(alink, alink->abm->q_lvls, NFP_QLVL_STRIDE,
+				NFP_QLVL_BLOG_BYTES, band, queue, false,
+				&stats->backlog_bytes);
 	if (err)
 		return err;
 
 	err = nfp_abm_ctrl_stat(alink, alink->abm->q_lvls,
 				NFP_QLVL_STRIDE, NFP_QLVL_BLOG_PKTS,
-				i, false, &stats->backlog_pkts);
+				band, queue, false, &stats->backlog_pkts);
 	if (err)
 		return err;
 
 	err = nfp_abm_ctrl_stat(alink, alink->abm->qm_stats,
 				NFP_QMSTAT_STRIDE, NFP_QMSTAT_DROP,
-				i, true, &stats->drops);
+				band, queue, true, &stats->drops);
 	if (err)
 		return err;
 
 	return nfp_abm_ctrl_stat(alink, alink->abm->qm_stats,
 				 NFP_QMSTAT_STRIDE, NFP_QMSTAT_ECN,
-				 i, true, &stats->overlimits);
+				 band, queue, true, &stats->overlimits);
 }
 
-int nfp_abm_ctrl_read_q_xstats(struct nfp_abm_link *alink, unsigned int i,
+int nfp_abm_ctrl_read_q_xstats(struct nfp_abm_link *alink,
+			       unsigned int band, unsigned int queue,
 			       struct nfp_alink_xstats *xstats)
 {
 	int err;
 
 	err = nfp_abm_ctrl_stat(alink, alink->abm->qm_stats,
 				NFP_QMSTAT_STRIDE, NFP_QMSTAT_DROP,
-				i, true, &xstats->pdrop);
+				band, queue, true, &xstats->pdrop);
 	if (err)
 		return err;
 
 	return nfp_abm_ctrl_stat(alink, alink->abm->qm_stats,
 				 NFP_QMSTAT_STRIDE, NFP_QMSTAT_ECN,
-				 i, true, &xstats->ecn_marked);
+				 band, queue, true, &xstats->ecn_marked);
 }
 
 int nfp_abm_ctrl_qm_enable(struct nfp_abm *abm)
@@ -162,10 +257,64 @@ int nfp_abm_ctrl_qm_disable(struct nfp_abm *abm)
 			    NULL, 0, NULL, 0);
 }
 
-void nfp_abm_ctrl_read_params(struct nfp_abm_link *alink)
+int nfp_abm_ctrl_prio_map_update(struct nfp_abm_link *alink, u32 *packed)
+{
+	struct nfp_net *nn = alink->vnic;
+	unsigned int i;
+	int err;
+
+	/* Write data_len and wipe reserved */
+	nn_writeq(nn, nn->tlv_caps.mbox_off + NFP_NET_ABM_MBOX_DATALEN,
+		  alink->abm->prio_map_len);
+
+	for (i = 0; i < alink->abm->prio_map_len; i += sizeof(u32))
+		nn_writel(nn, nn->tlv_caps.mbox_off + NFP_NET_ABM_MBOX_DATA + i,
+			  packed[i / sizeof(u32)]);
+
+	err = nfp_net_reconfig_mbox(nn,
+				    NFP_NET_CFG_MBOX_CMD_PCI_DSCP_PRIOMAP_SET);
+	if (err)
+		nfp_err(alink->abm->app->cpp,
+			"setting DSCP -> VQ map failed with error %d\n", err);
+	return err;
+}
+
+static int nfp_abm_ctrl_prio_check_params(struct nfp_abm_link *alink)
+{
+	struct nfp_abm *abm = alink->abm;
+	struct nfp_net *nn = alink->vnic;
+	unsigned int min_mbox_sz;
+
+	if (!nfp_abm_has_prio(alink->abm))
+		return 0;
+
+	min_mbox_sz = NFP_NET_ABM_MBOX_DATA + alink->abm->prio_map_len;
+	if (nn->tlv_caps.mbox_len < min_mbox_sz) {
+		nfp_err(abm->app->pf->cpp, "vNIC mailbox too small for prio offload: %u, need: %u\n",
+			nn->tlv_caps.mbox_len,  min_mbox_sz);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+int nfp_abm_ctrl_read_params(struct nfp_abm_link *alink)
 {
 	alink->queue_base = nn_readl(alink->vnic, NFP_NET_CFG_START_RXQ);
 	alink->queue_base /= alink->vnic->stride_rx;
+
+	return nfp_abm_ctrl_prio_check_params(alink);
+}
+
+static unsigned int nfp_abm_ctrl_prio_map_size(struct nfp_abm *abm)
+{
+	unsigned int size;
+
+	size = roundup_pow_of_two(order_base_2(abm->num_bands));
+	size = DIV_ROUND_UP(size * abm->num_prios, BITS_PER_BYTE);
+	size = round_up(size, sizeof(u32));
+
+	return size;
 }
 
 static const struct nfp_rtsym *
@@ -189,33 +338,77 @@ nfp_abm_ctrl_find_rtsym(struct nfp_pf *pf, const char *name, unsigned int size)
 }
 
 static const struct nfp_rtsym *
-nfp_abm_ctrl_find_q_rtsym(struct nfp_pf *pf, const char *name,
-			  unsigned int size)
+nfp_abm_ctrl_find_q_rtsym(struct nfp_abm *abm, const char *name_fmt,
+			  size_t size)
 {
-	return nfp_abm_ctrl_find_rtsym(pf, name, size * NFP_NET_MAX_RX_RINGS);
+	char pf_symbol[64];
+
+	size = array3_size(size, abm->num_bands, NFP_NET_MAX_RX_RINGS);
+	snprintf(pf_symbol, sizeof(pf_symbol), name_fmt,
+		 abm->pf_id, nfp_abm_has_prio(abm) ? "_per_band" : "");
+
+	return nfp_abm_ctrl_find_rtsym(abm->app->pf, pf_symbol, size);
 }
 
 int nfp_abm_ctrl_find_addrs(struct nfp_abm *abm)
 {
 	struct nfp_pf *pf = abm->app->pf;
 	const struct nfp_rtsym *sym;
-	unsigned int pf_id;
-	char pf_symbol[64];
-
-	pf_id =	nfp_cppcore_pcie_unit(pf->cpp);
-	abm->pf_id = pf_id;
+	int res;
+
+	abm->pf_id = nfp_cppcore_pcie_unit(pf->cpp);
+
+	/* Read count of prios and prio bands */
+	res = nfp_pf_rtsym_read_optional(pf, NFP_NUM_BANDS_SYM_NAME, 1);
+	if (res < 0)
+		return res;
+	abm->num_bands = res;
+
+	res = nfp_pf_rtsym_read_optional(pf, NFP_NUM_PRIOS_SYM_NAME, 1);
+	if (res < 0)
+		return res;
+	abm->num_prios = res;
+
+	/* Read available actions */
+	res = nfp_pf_rtsym_read_optional(pf, NFP_ACT_MASK_SYM_NAME,
+					 BIT(NFP_ABM_ACT_MARK_DROP));
+	if (res < 0)
+		return res;
+	abm->action_mask = res;
+
+	abm->prio_map_len = nfp_abm_ctrl_prio_map_size(abm);
+	abm->dscp_mask = GENMASK(7, 8 - order_base_2(abm->num_prios));
+
+	/* Check values are sane, U16_MAX is arbitrarily chosen as max */
+	if (!is_power_of_2(abm->num_bands) || !is_power_of_2(abm->num_prios) ||
+	    abm->num_bands > U16_MAX || abm->num_prios > U16_MAX ||
+	    (abm->num_bands == 1) != (abm->num_prios == 1)) {
+		nfp_err(pf->cpp,
+			"invalid priomap description num bands: %u and num prios: %u\n",
+			abm->num_bands, abm->num_prios);
+		return -EINVAL;
+	}
 
-	snprintf(pf_symbol, sizeof(pf_symbol), NFP_QLVL_SYM_NAME, pf_id);
-	sym = nfp_abm_ctrl_find_q_rtsym(pf, pf_symbol, NFP_QLVL_STRIDE);
+	/* Find level and stat symbols */
+	sym = nfp_abm_ctrl_find_q_rtsym(abm, NFP_QLVL_SYM_NAME,
+					NFP_QLVL_STRIDE);
 	if (IS_ERR(sym))
 		return PTR_ERR(sym);
 	abm->q_lvls = sym;
 
-	snprintf(pf_symbol, sizeof(pf_symbol), NFP_QMSTAT_SYM_NAME, pf_id);
-	sym = nfp_abm_ctrl_find_q_rtsym(pf, pf_symbol, NFP_QMSTAT_STRIDE);
+	sym = nfp_abm_ctrl_find_q_rtsym(abm, NFP_QMSTAT_SYM_NAME,
+					NFP_QMSTAT_STRIDE);
 	if (IS_ERR(sym))
 		return PTR_ERR(sym);
 	abm->qm_stats = sym;
 
+	if (nfp_abm_has_prio(abm)) {
+		sym = nfp_abm_ctrl_find_q_rtsym(abm, NFP_Q_STAT_SYM_NAME,
+						NFP_Q_STAT_STRIDE);
+		if (IS_ERR(sym))
+			return PTR_ERR(sym);
+		abm->q_stats = sym;
+	}
+
 	return 0;
 }
diff --git a/drivers/net/ethernet/netronome/nfp/abm/main.c b/drivers/net/ethernet/netronome/nfp/abm/main.c
index a5732d3bd1b7..7a4d55f794c2 100644
--- a/drivers/net/ethernet/netronome/nfp/abm/main.c
+++ b/drivers/net/ethernet/netronome/nfp/abm/main.c
@@ -44,6 +44,10 @@ nfp_abm_setup_tc(struct nfp_app *app, struct net_device *netdev,
 		return nfp_abm_setup_tc_mq(netdev, repr->app_priv, type_data);
 	case TC_SETUP_QDISC_RED:
 		return nfp_abm_setup_tc_red(netdev, repr->app_priv, type_data);
+	case TC_SETUP_QDISC_GRED:
+		return nfp_abm_setup_tc_gred(netdev, repr->app_priv, type_data);
+	case TC_SETUP_BLOCK:
+		return nfp_abm_setup_cls_block(netdev, repr, type_data);
 	default:
 		return -EOPNOTSUPP;
 	}
@@ -313,21 +317,32 @@ nfp_abm_vnic_alloc(struct nfp_app *app, struct nfp_net *nn, unsigned int id)
 	alink->id = id;
 	alink->total_queues = alink->vnic->max_rx_rings;
 
+	INIT_LIST_HEAD(&alink->dscp_map);
+
+	err = nfp_abm_ctrl_read_params(alink);
+	if (err)
+		goto err_free_alink;
+
+	alink->prio_map = kzalloc(abm->prio_map_len, GFP_KERNEL);
+	if (!alink->prio_map)
+		goto err_free_alink;
+
 	/* This is a multi-host app, make sure MAC/PHY is up, but don't
 	 * make the MAC/PHY state follow the state of any of the ports.
 	 */
 	err = nfp_eth_set_configured(app->cpp, eth_port->index, true);
 	if (err < 0)
-		goto err_free_alink;
+		goto err_free_priomap;
 
 	netif_keep_dst(nn->dp.netdev);
 
 	nfp_abm_vnic_set_mac(app->pf, abm, nn, id);
-	nfp_abm_ctrl_read_params(alink);
 	INIT_RADIX_TREE(&alink->qdiscs, GFP_KERNEL);
 
 	return 0;
 
+err_free_priomap:
+	kfree(alink->prio_map);
 err_free_alink:
 	kfree(alink);
 	return err;
@@ -339,9 +354,19 @@ static void nfp_abm_vnic_free(struct nfp_app *app, struct nfp_net *nn)
 
 	nfp_abm_kill_reprs(alink->abm, alink);
 	WARN(!radix_tree_empty(&alink->qdiscs), "left over qdiscs\n");
+	kfree(alink->prio_map);
 	kfree(alink);
 }
 
+static int nfp_abm_vnic_init(struct nfp_app *app, struct nfp_net *nn)
+{
+	struct nfp_abm_link *alink = nn->app_priv;
+
+	if (nfp_abm_has_prio(alink->abm))
+		return nfp_abm_ctrl_prio_map_update(alink, alink->prio_map);
+	return 0;
+}
+
 static u64 *
 nfp_abm_port_get_stats(struct nfp_app *app, struct nfp_port *port, u64 *data)
 {
@@ -422,7 +447,7 @@ static int nfp_abm_init(struct nfp_app *app)
 		goto err_free_abm;
 
 	err = -ENOMEM;
-	abm->num_thresholds = NFP_NET_MAX_RX_RINGS;
+	abm->num_thresholds = array_size(abm->num_bands, NFP_NET_MAX_RX_RINGS);
 	abm->threshold_undef = bitmap_zalloc(abm->num_thresholds, GFP_KERNEL);
 	if (!abm->threshold_undef)
 		goto err_free_abm;
@@ -431,18 +456,25 @@ static int nfp_abm_init(struct nfp_app *app)
 				   sizeof(*abm->thresholds), GFP_KERNEL);
 	if (!abm->thresholds)
 		goto err_free_thresh_umap;
-	for (i = 0; i < NFP_NET_MAX_RX_RINGS; i++)
+	for (i = 0; i < abm->num_bands * NFP_NET_MAX_RX_RINGS; i++)
 		__nfp_abm_ctrl_set_q_lvl(abm, i, NFP_ABM_LVL_INFINITY);
 
+	abm->actions = kvcalloc(abm->num_thresholds, sizeof(*abm->actions),
+				GFP_KERNEL);
+	if (!abm->actions)
+		goto err_free_thresh;
+	for (i = 0; i < abm->num_bands * NFP_NET_MAX_RX_RINGS; i++)
+		__nfp_abm_ctrl_set_q_act(abm, i, NFP_ABM_ACT_DROP);
+
 	/* We start in legacy mode, make sure advanced queuing is disabled */
 	err = nfp_abm_ctrl_qm_disable(abm);
 	if (err)
-		goto err_free_thresh;
+		goto err_free_act;
 
 	err = -ENOMEM;
 	reprs = nfp_reprs_alloc(pf->max_data_vnics);
 	if (!reprs)
-		goto err_free_thresh;
+		goto err_free_act;
 	RCU_INIT_POINTER(app->reprs[NFP_REPR_TYPE_PHYS_PORT], reprs);
 
 	reprs = nfp_reprs_alloc(pf->max_data_vnics);
@@ -454,6 +486,8 @@ static int nfp_abm_init(struct nfp_app *app)
 
 err_free_phys:
 	nfp_reprs_clean_and_free_by_type(app, NFP_REPR_TYPE_PHYS_PORT);
+err_free_act:
+	kvfree(abm->actions);
 err_free_thresh:
 	kvfree(abm->thresholds);
 err_free_thresh_umap:
@@ -472,6 +506,7 @@ static void nfp_abm_clean(struct nfp_app *app)
 	nfp_reprs_clean_and_free_by_type(app, NFP_REPR_TYPE_PF);
 	nfp_reprs_clean_and_free_by_type(app, NFP_REPR_TYPE_PHYS_PORT);
 	bitmap_free(abm->threshold_undef);
+	kvfree(abm->actions);
 	kvfree(abm->thresholds);
 	kfree(abm);
 	app->priv = NULL;
@@ -486,6 +521,7 @@ const struct nfp_app_type app_abm = {
 
 	.vnic_alloc	= nfp_abm_vnic_alloc,
 	.vnic_free	= nfp_abm_vnic_free,
+	.vnic_init	= nfp_abm_vnic_init,
 
 	.port_get_stats		= nfp_abm_port_get_stats,
 	.port_get_stats_count	= nfp_abm_port_get_stats_count,
diff --git a/drivers/net/ethernet/netronome/nfp/abm/main.h b/drivers/net/ethernet/netronome/nfp/abm/main.h
index 240e2c8683fe..4dcf5881fb4b 100644
--- a/drivers/net/ethernet/netronome/nfp/abm/main.h
+++ b/drivers/net/ethernet/netronome/nfp/abm/main.h
@@ -5,9 +5,11 @@
 #define __NFP_ABM_H__ 1
 
 #include <linux/bits.h>
+#include <linux/list.h>
 #include <linux/radix-tree.h>
 #include <net/devlink.h>
 #include <net/pkt_cls.h>
+#include <net/pkt_sched.h>
 
 /* Dump of 64 PRIOs and 256 REDs seems to take 850us on Xeon v4 @ 2.20GHz;
  * 2.5ms / 400Hz seems more than sufficient for stats resolution.
@@ -22,31 +24,62 @@ struct nfp_net;
 #define NFP_ABM_PORTID_TYPE	GENMASK(23, 16)
 #define NFP_ABM_PORTID_ID	GENMASK(7, 0)
 
+/* The possible actions if thresholds are exceeded */
+enum nfp_abm_q_action {
+	/* mark if ECN capable, otherwise drop */
+	NFP_ABM_ACT_MARK_DROP		= 0,
+	/* mark if ECN capable, otherwise goto QM */
+	NFP_ABM_ACT_MARK_QUEUE		= 1,
+	NFP_ABM_ACT_DROP		= 2,
+	NFP_ABM_ACT_QUEUE		= 3,
+	NFP_ABM_ACT_NOQUEUE		= 4,
+};
+
 /**
  * struct nfp_abm - ABM NIC app structure
  * @app:	back pointer to nfp_app
  * @pf_id:	ID of our PF link
  *
+ * @num_prios:	number of supported DSCP priorities
+ * @num_bands:	number of supported DSCP priority bands
+ * @action_mask:	bitmask of supported actions
+ *
  * @thresholds:		current threshold configuration
  * @threshold_undef:	bitmap of thresholds which have not been set
+ * @actions:		current FW action configuration
  * @num_thresholds:	number of @thresholds and bits in @threshold_undef
  *
+ * @prio_map_len:	computed length of FW priority map (in bytes)
+ * @dscp_mask:		mask FW will apply on DSCP field
+ *
  * @eswitch_mode:	devlink eswitch mode, advanced functions only visible
  *			in switchdev mode
+ *
  * @q_lvls:	queue level control area
  * @qm_stats:	queue statistics symbol
+ * @q_stats:	basic queue statistics (only in per-band case)
  */
 struct nfp_abm {
 	struct nfp_app *app;
 	unsigned int pf_id;
 
+	unsigned int num_prios;
+	unsigned int num_bands;
+	unsigned int action_mask;
+
 	u32 *thresholds;
 	unsigned long *threshold_undef;
+	u8 *actions;
 	size_t num_thresholds;
 
+	unsigned int prio_map_len;
+	u8 dscp_mask;
+
 	enum devlink_eswitch_mode eswitch_mode;
+
 	const struct nfp_rtsym *q_lvls;
 	const struct nfp_rtsym *qm_stats;
+	const struct nfp_rtsym *q_stats;
 };
 
 /**
@@ -81,6 +114,7 @@ enum nfp_qdisc_type {
 	NFP_QDISC_NONE = 0,
 	NFP_QDISC_MQ,
 	NFP_QDISC_RED,
+	NFP_QDISC_GRED,
 };
 
 #define NFP_QDISC_UNTRACKED	((struct nfp_qdisc *)1UL)
@@ -104,11 +138,14 @@ enum nfp_qdisc_type {
  * @mq.prev_stats:	previously reported @mq.stats
  *
  * @red:		RED Qdisc specific parameters and state
- * @red.threshold:	ECN marking threshold
- * @red.stats:		current stats of the RED Qdisc
- * @red.prev_stats:	previously reported @red.stats
- * @red.xstats:		extended stats for RED - current
- * @red.prev_xstats:	extended stats for RED - previously reported
+ * @red.num_bands:	Number of valid entries in the @red.band table
+ * @red.band:		Per-band array of RED instances
+ * @red.band.ecn:		ECN marking is enabled (rather than drop)
+ * @red.band.threshold:		ECN marking threshold
+ * @red.band.stats:		current stats of the RED Qdisc
+ * @red.band.prev_stats:	previously reported @red.stats
+ * @red.band.xstats:		extended stats for RED - current
+ * @red.band.prev_xstats:	extended stats for RED - previously reported
  */
 struct nfp_qdisc {
 	struct net_device *netdev;
@@ -129,13 +166,18 @@ struct nfp_qdisc {
 			struct nfp_alink_stats stats;
 			struct nfp_alink_stats prev_stats;
 		} mq;
-		/* TC_SETUP_QDISC_RED */
+		/* TC_SETUP_QDISC_RED, TC_SETUP_QDISC_GRED */
 		struct {
-			u32 threshold;
-			struct nfp_alink_stats stats;
-			struct nfp_alink_stats prev_stats;
-			struct nfp_alink_xstats xstats;
-			struct nfp_alink_xstats prev_xstats;
+			unsigned int num_bands;
+
+			struct {
+				bool ecn;
+				u32 threshold;
+				struct nfp_alink_stats stats;
+				struct nfp_alink_stats prev_stats;
+				struct nfp_alink_xstats xstats;
+				struct nfp_alink_xstats prev_xstats;
+			} band[MAX_DPs];
 		} red;
 	};
 };
@@ -150,6 +192,12 @@ struct nfp_qdisc {
  *
  * @last_stats_update:	ktime of last stats update
  *
+ * @prio_map:		current map of priorities
+ * @has_prio:		@prio_map is valid
+ *
+ * @def_band:		default band to use
+ * @dscp_map:		list of DSCP to band mappings
+ *
  * @root_qdisc:	pointer to the current root of the Qdisc hierarchy
  * @qdiscs:	all qdiscs recorded by major part of the handle
  */
@@ -162,10 +210,31 @@ struct nfp_abm_link {
 
 	u64 last_stats_update;
 
+	u32 *prio_map;
+	bool has_prio;
+
+	u8 def_band;
+	struct list_head dscp_map;
+
 	struct nfp_qdisc *root_qdisc;
 	struct radix_tree_root qdiscs;
 };
 
+static inline bool nfp_abm_has_prio(struct nfp_abm *abm)
+{
+	return abm->num_bands > 1;
+}
+
+static inline bool nfp_abm_has_drop(struct nfp_abm *abm)
+{
+	return abm->action_mask & BIT(NFP_ABM_ACT_DROP);
+}
+
+static inline bool nfp_abm_has_mark(struct nfp_abm *abm)
+{
+	return abm->action_mask & BIT(NFP_ABM_ACT_MARK_DROP);
+}
+
 void nfp_abm_qdisc_offload_update(struct nfp_abm_link *alink);
 int nfp_abm_setup_root(struct net_device *netdev, struct nfp_abm_link *alink,
 		       struct tc_root_qopt_offload *opt);
@@ -173,18 +242,30 @@ int nfp_abm_setup_tc_red(struct net_device *netdev, struct nfp_abm_link *alink,
 			 struct tc_red_qopt_offload *opt);
 int nfp_abm_setup_tc_mq(struct net_device *netdev, struct nfp_abm_link *alink,
 			struct tc_mq_qopt_offload *opt);
+int nfp_abm_setup_tc_gred(struct net_device *netdev, struct nfp_abm_link *alink,
+			  struct tc_gred_qopt_offload *opt);
+int nfp_abm_setup_cls_block(struct net_device *netdev, struct nfp_repr *repr,
+			    struct tc_block_offload *opt);
 
-void nfp_abm_ctrl_read_params(struct nfp_abm_link *alink);
+int nfp_abm_ctrl_read_params(struct nfp_abm_link *alink);
 int nfp_abm_ctrl_find_addrs(struct nfp_abm *abm);
 int __nfp_abm_ctrl_set_q_lvl(struct nfp_abm *abm, unsigned int id, u32 val);
-int nfp_abm_ctrl_set_q_lvl(struct nfp_abm_link *alink, unsigned int queue,
-			   u32 val);
-int nfp_abm_ctrl_read_q_stats(struct nfp_abm_link *alink, unsigned int i,
+int nfp_abm_ctrl_set_q_lvl(struct nfp_abm_link *alink, unsigned int band,
+			   unsigned int queue, u32 val);
+int __nfp_abm_ctrl_set_q_act(struct nfp_abm *abm, unsigned int id,
+			     enum nfp_abm_q_action act);
+int nfp_abm_ctrl_set_q_act(struct nfp_abm_link *alink, unsigned int band,
+			   unsigned int queue, enum nfp_abm_q_action act);
+int nfp_abm_ctrl_read_q_stats(struct nfp_abm_link *alink,
+			      unsigned int band, unsigned int queue,
 			      struct nfp_alink_stats *stats);
-int nfp_abm_ctrl_read_q_xstats(struct nfp_abm_link *alink, unsigned int i,
+int nfp_abm_ctrl_read_q_xstats(struct nfp_abm_link *alink,
+			       unsigned int band, unsigned int queue,
 			       struct nfp_alink_xstats *xstats);
 u64 nfp_abm_ctrl_stat_non_sto(struct nfp_abm_link *alink, unsigned int i);
 u64 nfp_abm_ctrl_stat_sto(struct nfp_abm_link *alink, unsigned int i);
 int nfp_abm_ctrl_qm_enable(struct nfp_abm *abm);
 int nfp_abm_ctrl_qm_disable(struct nfp_abm *abm);
+void nfp_abm_prio_map_update(struct nfp_abm *abm);
+int nfp_abm_ctrl_prio_map_update(struct nfp_abm_link *alink, u32 *packed);
 #endif
diff --git a/drivers/net/ethernet/netronome/nfp/abm/qdisc.c b/drivers/net/ethernet/netronome/nfp/abm/qdisc.c
index 16c4afe3a37f..2473fb5f75e5 100644
--- a/drivers/net/ethernet/netronome/nfp/abm/qdisc.c
+++ b/drivers/net/ethernet/netronome/nfp/abm/qdisc.c
@@ -15,7 +15,7 @@
 
 static bool nfp_abm_qdisc_is_red(struct nfp_qdisc *qdisc)
 {
-	return qdisc->type == NFP_QDISC_RED;
+	return qdisc->type == NFP_QDISC_RED || qdisc->type == NFP_QDISC_GRED;
 }
 
 static bool nfp_abm_qdisc_child_valid(struct nfp_qdisc *qdisc, unsigned int id)
@@ -46,20 +46,25 @@ nfp_abm_stats_update_red(struct nfp_abm_link *alink, struct nfp_qdisc *qdisc,
 			 unsigned int queue)
 {
 	struct nfp_cpp *cpp = alink->abm->app->cpp;
+	unsigned int i;
 	int err;
 
 	if (!qdisc->offloaded)
 		return;
 
-	err = nfp_abm_ctrl_read_q_stats(alink, queue, &qdisc->red.stats);
-	if (err)
-		nfp_err(cpp, "RED stats (%d) read failed with error %d\n",
-			queue, err);
-
-	err = nfp_abm_ctrl_read_q_xstats(alink, queue, &qdisc->red.xstats);
-	if (err)
-		nfp_err(cpp, "RED xstats (%d) read failed with error %d\n",
-			queue, err);
+	for (i = 0; i < qdisc->red.num_bands; i++) {
+		err = nfp_abm_ctrl_read_q_stats(alink, i, queue,
+						&qdisc->red.band[i].stats);
+		if (err)
+			nfp_err(cpp, "RED stats (%d, %d) read failed with error %d\n",
+				i, queue, err);
+
+		err = nfp_abm_ctrl_read_q_xstats(alink, i, queue,
+						 &qdisc->red.band[i].xstats);
+		if (err)
+			nfp_err(cpp, "RED xstats (%d, %d) read failed with error %d\n",
+				i, queue, err);
+	}
 }
 
 static void
@@ -113,6 +118,8 @@ nfp_abm_qdisc_unlink_children(struct nfp_qdisc *qdisc,
 static void
 nfp_abm_qdisc_offload_stop(struct nfp_abm_link *alink, struct nfp_qdisc *qdisc)
 {
+	unsigned int i;
+
 	/* Don't complain when qdisc is getting unlinked */
 	if (qdisc->use_cnt)
 		nfp_warn(alink->abm->app->cpp, "Offload of '%08x' stopped\n",
@@ -121,12 +128,14 @@ nfp_abm_qdisc_offload_stop(struct nfp_abm_link *alink, struct nfp_qdisc *qdisc)
 	if (!nfp_abm_qdisc_is_red(qdisc))
 		return;
 
-	qdisc->red.stats.backlog_pkts = 0;
-	qdisc->red.stats.backlog_bytes = 0;
+	for (i = 0; i < qdisc->red.num_bands; i++) {
+		qdisc->red.band[i].stats.backlog_pkts = 0;
+		qdisc->red.band[i].stats.backlog_bytes = 0;
+	}
 }
 
 static int
-__nfp_abm_stats_init(struct nfp_abm_link *alink,
+__nfp_abm_stats_init(struct nfp_abm_link *alink, unsigned int band,
 		     unsigned int queue, struct nfp_alink_stats *prev_stats,
 		     struct nfp_alink_xstats *prev_xstats)
 {
@@ -139,19 +148,19 @@ __nfp_abm_stats_init(struct nfp_abm_link *alink,
 	backlog_pkts = prev_stats->backlog_pkts;
 	backlog_bytes = prev_stats->backlog_bytes;
 
-	err = nfp_abm_ctrl_read_q_stats(alink, queue, prev_stats);
+	err = nfp_abm_ctrl_read_q_stats(alink, band, queue, prev_stats);
 	if (err) {
 		nfp_err(alink->abm->app->cpp,
-			"RED stats init (%d) failed with error %d\n",
-			queue, err);
+			"RED stats init (%d, %d) failed with error %d\n",
+			band, queue, err);
 		return err;
 	}
 
-	err = nfp_abm_ctrl_read_q_xstats(alink, queue, prev_xstats);
+	err = nfp_abm_ctrl_read_q_xstats(alink, band, queue, prev_xstats);
 	if (err) {
 		nfp_err(alink->abm->app->cpp,
-			"RED xstats init (%d) failed with error %d\n",
-			queue, err);
+			"RED xstats init (%d, %d) failed with error %d\n",
+			band, queue, err);
 		return err;
 	}
 
@@ -164,19 +173,36 @@ static int
 nfp_abm_stats_init(struct nfp_abm_link *alink, struct nfp_qdisc *qdisc,
 		   unsigned int queue)
 {
-	return __nfp_abm_stats_init(alink, queue,
-				    &qdisc->red.prev_stats,
-				    &qdisc->red.prev_xstats);
+	unsigned int i;
+	int err;
+
+	for (i = 0; i < qdisc->red.num_bands; i++) {
+		err = __nfp_abm_stats_init(alink, i, queue,
+					   &qdisc->red.band[i].prev_stats,
+					   &qdisc->red.band[i].prev_xstats);
+		if (err)
+			return err;
+	}
+
+	return 0;
 }
 
 static void
 nfp_abm_offload_compile_red(struct nfp_abm_link *alink, struct nfp_qdisc *qdisc,
 			    unsigned int queue)
 {
-	qdisc->offload_mark = qdisc->type == NFP_QDISC_RED &&
-			      qdisc->params_ok &&
-			      qdisc->use_cnt == 1 &&
-			      !qdisc->children[0];
+	bool good_red, good_gred;
+	unsigned int i;
+
+	good_red = qdisc->type == NFP_QDISC_RED &&
+		   qdisc->params_ok &&
+		   qdisc->use_cnt == 1 &&
+		   !alink->has_prio &&
+		   !qdisc->children[0];
+	good_gred = qdisc->type == NFP_QDISC_GRED &&
+		    qdisc->params_ok &&
+		    qdisc->use_cnt == 1;
+	qdisc->offload_mark = good_red || good_gred;
 
 	/* If we are starting offload init prev_stats */
 	if (qdisc->offload_mark && !qdisc->offloaded)
@@ -186,7 +212,15 @@ nfp_abm_offload_compile_red(struct nfp_abm_link *alink, struct nfp_qdisc *qdisc,
 	if (!qdisc->offload_mark)
 		return;
 
-	nfp_abm_ctrl_set_q_lvl(alink, queue, qdisc->red.threshold);
+	for (i = 0; i < alink->abm->num_bands; i++) {
+		enum nfp_abm_q_action act;
+
+		nfp_abm_ctrl_set_q_lvl(alink, i, queue,
+				       qdisc->red.band[i].threshold);
+		act = qdisc->red.band[i].ecn ?
+			NFP_ABM_ACT_MARK_DROP : NFP_ABM_ACT_DROP;
+		nfp_abm_ctrl_set_q_act(alink, i, queue, act);
+	}
 }
 
 static void
@@ -217,8 +251,10 @@ void nfp_abm_qdisc_offload_update(struct nfp_abm_link *alink)
 	size_t i;
 
 	/* Mark all thresholds as unconfigured */
-	__bitmap_set(abm->threshold_undef,
-		     alink->queue_base, alink->total_queues);
+	for (i = 0; i < abm->num_bands; i++)
+		__bitmap_set(abm->threshold_undef,
+			     i * NFP_NET_MAX_RX_RINGS + alink->queue_base,
+			     alink->total_queues);
 
 	/* Clear offload marks */
 	radix_tree_for_each_slot(slot, &alink->qdiscs, &iter, 0) {
@@ -312,9 +348,11 @@ nfp_abm_qdisc_alloc(struct net_device *netdev, struct nfp_abm_link *alink,
 	if (!qdisc)
 		return NULL;
 
-	qdisc->children = kcalloc(children, sizeof(void *), GFP_KERNEL);
-	if (!qdisc->children)
-		goto err_free_qdisc;
+	if (children) {
+		qdisc->children = kcalloc(children, sizeof(void *), GFP_KERNEL);
+		if (!qdisc->children)
+			goto err_free_qdisc;
+	}
 
 	qdisc->netdev = netdev;
 	qdisc->type = type;
@@ -441,6 +479,144 @@ nfp_abm_stats_red_calculate(struct nfp_alink_xstats *new,
 }
 
 static int
+nfp_abm_gred_stats(struct nfp_abm_link *alink, u32 handle,
+		   struct tc_gred_qopt_offload_stats *stats)
+{
+	struct nfp_qdisc *qdisc;
+	unsigned int i;
+
+	nfp_abm_stats_update(alink);
+
+	qdisc = nfp_abm_qdisc_find(alink, handle);
+	if (!qdisc)
+		return -EOPNOTSUPP;
+	/* If the qdisc offload has stopped we may need to adjust the backlog
+	 * counters back so carry on even if qdisc is not currently offloaded.
+	 */
+
+	for (i = 0; i < qdisc->red.num_bands; i++) {
+		if (!stats->xstats[i])
+			continue;
+
+		nfp_abm_stats_calculate(&qdisc->red.band[i].stats,
+					&qdisc->red.band[i].prev_stats,
+					&stats->bstats[i], &stats->qstats[i]);
+		qdisc->red.band[i].prev_stats = qdisc->red.band[i].stats;
+
+		nfp_abm_stats_red_calculate(&qdisc->red.band[i].xstats,
+					    &qdisc->red.band[i].prev_xstats,
+					    stats->xstats[i]);
+		qdisc->red.band[i].prev_xstats = qdisc->red.band[i].xstats;
+	}
+
+	return qdisc->offloaded ? 0 : -EOPNOTSUPP;
+}
+
+static bool
+nfp_abm_gred_check_params(struct nfp_abm_link *alink,
+			  struct tc_gred_qopt_offload *opt)
+{
+	struct nfp_cpp *cpp = alink->abm->app->cpp;
+	struct nfp_abm *abm = alink->abm;
+	unsigned int i;
+
+	if (opt->set.grio_on || opt->set.wred_on) {
+		nfp_warn(cpp, "GRED offload failed - GRIO and WRED not supported (p:%08x h:%08x)\n",
+			 opt->parent, opt->handle);
+		return false;
+	}
+	if (opt->set.dp_def != alink->def_band) {
+		nfp_warn(cpp, "GRED offload failed - default band must be %d (p:%08x h:%08x)\n",
+			 alink->def_band, opt->parent, opt->handle);
+		return false;
+	}
+	if (opt->set.dp_cnt != abm->num_bands) {
+		nfp_warn(cpp, "GRED offload failed - band count must be %d (p:%08x h:%08x)\n",
+			 abm->num_bands, opt->parent, opt->handle);
+		return false;
+	}
+
+	for (i = 0; i < abm->num_bands; i++) {
+		struct tc_gred_vq_qopt_offload_params *band = &opt->set.tab[i];
+
+		if (!band->present)
+			return false;
+		if (!band->is_ecn && !nfp_abm_has_drop(abm)) {
+			nfp_warn(cpp, "GRED offload failed - drop is not supported (ECN option required) (p:%08x h:%08x vq:%d)\n",
+				 opt->parent, opt->handle, i);
+			return false;
+		}
+		if (band->is_ecn && !nfp_abm_has_mark(abm)) {
+			nfp_warn(cpp, "GRED offload failed - ECN marking not supported (p:%08x h:%08x vq:%d)\n",
+				 opt->parent, opt->handle, i);
+			return false;
+		}
+		if (band->is_harddrop) {
+			nfp_warn(cpp, "GRED offload failed - harddrop is not supported (p:%08x h:%08x vq:%d)\n",
+				 opt->parent, opt->handle, i);
+			return false;
+		}
+		if (band->min != band->max) {
+			nfp_warn(cpp, "GRED offload failed - threshold mismatch (p:%08x h:%08x vq:%d)\n",
+				 opt->parent, opt->handle, i);
+			return false;
+		}
+		if (band->min > S32_MAX) {
+			nfp_warn(cpp, "GRED offload failed - threshold too large %d > %d (p:%08x h:%08x vq:%d)\n",
+				 band->min, S32_MAX, opt->parent, opt->handle,
+				 i);
+			return false;
+		}
+	}
+
+	return true;
+}
+
+static int
+nfp_abm_gred_replace(struct net_device *netdev, struct nfp_abm_link *alink,
+		     struct tc_gred_qopt_offload *opt)
+{
+	struct nfp_qdisc *qdisc;
+	unsigned int i;
+	int ret;
+
+	ret = nfp_abm_qdisc_replace(netdev, alink, NFP_QDISC_GRED, opt->parent,
+				    opt->handle, 0, &qdisc);
+	if (ret < 0)
+		return ret;
+
+	qdisc->params_ok = nfp_abm_gred_check_params(alink, opt);
+	if (qdisc->params_ok) {
+		qdisc->red.num_bands = opt->set.dp_cnt;
+		for (i = 0; i < qdisc->red.num_bands; i++) {
+			qdisc->red.band[i].ecn = opt->set.tab[i].is_ecn;
+			qdisc->red.band[i].threshold = opt->set.tab[i].min;
+		}
+	}
+
+	if (qdisc->use_cnt)
+		nfp_abm_qdisc_offload_update(alink);
+
+	return 0;
+}
+
+int nfp_abm_setup_tc_gred(struct net_device *netdev, struct nfp_abm_link *alink,
+			  struct tc_gred_qopt_offload *opt)
+{
+	switch (opt->command) {
+	case TC_GRED_REPLACE:
+		return nfp_abm_gred_replace(netdev, alink, opt);
+	case TC_GRED_DESTROY:
+		nfp_abm_qdisc_destroy(netdev, alink, opt->handle);
+		return 0;
+	case TC_GRED_STATS:
+		return nfp_abm_gred_stats(alink, opt->handle, &opt->stats);
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
+static int
 nfp_abm_red_xstats(struct nfp_abm_link *alink, struct tc_red_qopt_offload *opt)
 {
 	struct nfp_qdisc *qdisc;
@@ -451,10 +627,10 @@ nfp_abm_red_xstats(struct nfp_abm_link *alink, struct tc_red_qopt_offload *opt)
 	if (!qdisc || !qdisc->offloaded)
 		return -EOPNOTSUPP;
 
-	nfp_abm_stats_red_calculate(&qdisc->red.xstats,
-				    &qdisc->red.prev_xstats,
+	nfp_abm_stats_red_calculate(&qdisc->red.band[0].xstats,
+				    &qdisc->red.band[0].prev_xstats,
 				    opt->xstats);
-	qdisc->red.prev_xstats = qdisc->red.xstats;
+	qdisc->red.band[0].prev_xstats = qdisc->red.band[0].xstats;
 	return 0;
 }
 
@@ -473,10 +649,10 @@ nfp_abm_red_stats(struct nfp_abm_link *alink, u32 handle,
 	 * counters back so carry on even if qdisc is not currently offloaded.
 	 */
 
-	nfp_abm_stats_calculate(&qdisc->red.stats,
-				&qdisc->red.prev_stats,
+	nfp_abm_stats_calculate(&qdisc->red.band[0].stats,
+				&qdisc->red.band[0].prev_stats,
 				stats->bstats, stats->qstats);
-	qdisc->red.prev_stats = qdisc->red.stats;
+	qdisc->red.band[0].prev_stats = qdisc->red.band[0].stats;
 
 	return qdisc->offloaded ? 0 : -EOPNOTSUPP;
 }
@@ -486,12 +662,18 @@ nfp_abm_red_check_params(struct nfp_abm_link *alink,
 			 struct tc_red_qopt_offload *opt)
 {
 	struct nfp_cpp *cpp = alink->abm->app->cpp;
+	struct nfp_abm *abm = alink->abm;
 
-	if (!opt->set.is_ecn) {
+	if (!opt->set.is_ecn && !nfp_abm_has_drop(abm)) {
 		nfp_warn(cpp, "RED offload failed - drop is not supported (ECN option required) (p:%08x h:%08x)\n",
 			 opt->parent, opt->handle);
 		return false;
 	}
+	if (opt->set.is_ecn && !nfp_abm_has_mark(abm)) {
+		nfp_warn(cpp, "RED offload failed - ECN marking not supported (p:%08x h:%08x)\n",
+			 opt->parent, opt->handle);
+		return false;
+	}
 	if (opt->set.is_harddrop) {
 		nfp_warn(cpp, "RED offload failed - harddrop is not supported (p:%08x h:%08x)\n",
 			 opt->parent, opt->handle);
@@ -538,8 +720,11 @@ nfp_abm_red_replace(struct net_device *netdev, struct nfp_abm_link *alink,
 	}
 
 	qdisc->params_ok = nfp_abm_red_check_params(alink, opt);
-	if (qdisc->params_ok)
-		qdisc->red.threshold = opt->set.min;
+	if (qdisc->params_ok) {
+		qdisc->red.num_bands = 1;
+		qdisc->red.band[0].ecn = opt->set.is_ecn;
+		qdisc->red.band[0].threshold = opt->set.min;
+	}
 
 	if (qdisc->use_cnt == 1)
 		nfp_abm_qdisc_offload_update(alink);
@@ -592,7 +777,7 @@ nfp_abm_mq_stats(struct nfp_abm_link *alink, u32 handle,
 		 struct tc_qopt_offload_stats *stats)
 {
 	struct nfp_qdisc *qdisc, *red;
-	unsigned int i;
+	unsigned int i, j;
 
 	qdisc = nfp_abm_qdisc_find(alink, handle);
 	if (!qdisc)
@@ -614,10 +799,12 @@ nfp_abm_mq_stats(struct nfp_abm_link *alink, u32 handle,
 			continue;
 		red = qdisc->children[i];
 
-		nfp_abm_stats_propagate(&qdisc->mq.stats,
-					&red->red.stats);
-		nfp_abm_stats_propagate(&qdisc->mq.prev_stats,
-					&red->red.prev_stats);
+		for (j = 0; j < red->red.num_bands; j++) {
+			nfp_abm_stats_propagate(&qdisc->mq.stats,
+						&red->red.band[j].stats);
+			nfp_abm_stats_propagate(&qdisc->mq.prev_stats,
+						&red->red.band[j].prev_stats);
+		}
 	}
 
 	nfp_abm_stats_calculate(&qdisc->mq.stats, &qdisc->mq.prev_stats,
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net.h b/drivers/net/ethernet/netronome/nfp/nfp_net.h
index dda02fefc806..bb3dbd74583b 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net.h
@@ -868,6 +868,7 @@ unsigned int nfp_net_rss_key_sz(struct nfp_net *nn);
 void nfp_net_rss_write_itbl(struct nfp_net *nn);
 void nfp_net_rss_write_key(struct nfp_net *nn);
 void nfp_net_coalesce_write_cfg(struct nfp_net *nn);
+int nfp_net_reconfig_mbox(struct nfp_net *nn, u32 mbox_cmd);
 
 unsigned int
 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries,
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index a0343f25068a..9aa6265bf4de 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -279,7 +279,7 @@ int nfp_net_reconfig(struct nfp_net *nn, u32 update)
  *
  * Return: Negative errno on error, 0 on success
  */
-static int nfp_net_reconfig_mbox(struct nfp_net *nn, u32 mbox_cmd)
+int nfp_net_reconfig_mbox(struct nfp_net *nn, u32 mbox_cmd)
 {
 	u32 mbox = nn->tlv_caps.mbox_off;
 	int ret;
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h b/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h
index d7c8518ac952..ccec69944de5 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h
@@ -397,6 +397,8 @@
 #define NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_ADD 1
 #define NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_KILL 2
 
+#define NFP_NET_CFG_MBOX_CMD_PCI_DSCP_PRIOMAP_SET	5
+
 /**
  * VLAN filtering using general use mailbox
  * %NFP_NET_CFG_VLAN_FILTER:		Base address of VLAN filter mailbox
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 086e64d88597..4b4207ebd5c0 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -846,6 +846,7 @@ enum tc_setup_type {
 	TC_SETUP_QDISC_MQ,
 	TC_SETUP_QDISC_ETF,
 	TC_SETUP_ROOT_QDISC,
+	TC_SETUP_QDISC_GRED,
 };
 
 /* These structures hold the attributes of bpf state that are being passed
diff --git a/include/net/pkt_cls.h b/include/net/pkt_cls.h
index c497ada7f591..ea191d8cfcc9 100644
--- a/include/net/pkt_cls.h
+++ b/include/net/pkt_cls.h
@@ -643,6 +643,7 @@ struct tc_cls_common_offload {
 
 struct tc_cls_u32_knode {
 	struct tcf_exts *exts;
+	struct tcf_result *res;
 	struct tc_u32_sel *sel;
 	u32 handle;
 	u32 val;
@@ -868,6 +869,50 @@ struct tc_red_qopt_offload {
 	};
 };
 
+enum tc_gred_command {
+	TC_GRED_REPLACE,
+	TC_GRED_DESTROY,
+	TC_GRED_STATS,
+};
+
+struct tc_gred_vq_qopt_offload_params {
+	bool present;
+	u32 limit;
+	u32 prio;
+	u32 min;
+	u32 max;
+	bool is_ecn;
+	bool is_harddrop;
+	u32 probability;
+	/* Only need backlog, see struct tc_prio_qopt_offload_params */
+	u32 *backlog;
+};
+
+struct tc_gred_qopt_offload_params {
+	bool grio_on;
+	bool wred_on;
+	unsigned int dp_cnt;
+	unsigned int dp_def;
+	struct gnet_stats_queue *qstats;
+	struct tc_gred_vq_qopt_offload_params tab[MAX_DPs];
+};
+
+struct tc_gred_qopt_offload_stats {
+	struct gnet_stats_basic_packed bstats[MAX_DPs];
+	struct gnet_stats_queue qstats[MAX_DPs];
+	struct red_stats *xstats[MAX_DPs];
+};
+
+struct tc_gred_qopt_offload {
+	enum tc_gred_command command;
+	u32 handle;
+	u32 parent;
+	union {
+		struct tc_gred_qopt_offload_params set;
+		struct tc_gred_qopt_offload_stats stats;
+	};
+};
+
 enum tc_prio_command {
 	TC_PRIO_REPLACE,
 	TC_PRIO_DESTROY,
diff --git a/net/sched/cls_u32.c b/net/sched/cls_u32.c
index 4b28fd44576d..4c54bc440798 100644
--- a/net/sched/cls_u32.c
+++ b/net/sched/cls_u32.c
@@ -558,6 +558,7 @@ static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 	cls_u32.knode.mask = 0;
 #endif
 	cls_u32.knode.sel = &n->sel;
+	cls_u32.knode.res = &n->res;
 	cls_u32.knode.exts = &n->exts;
 	if (n->ht_down)
 		cls_u32.knode.link_handle = ht->handle;
@@ -1206,6 +1207,7 @@ static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 		cls_u32.knode.mask = 0;
 #endif
 		cls_u32.knode.sel = &n->sel;
+		cls_u32.knode.res = &n->res;
 		cls_u32.knode.exts = &n->exts;
 		if (n->ht_down)
 			cls_u32.knode.link_handle = ht->handle;
diff --git a/net/sched/sch_gred.c b/net/sched/sch_gred.c
index 8b8c325f48bc..234afbf9115b 100644
--- a/net/sched/sch_gred.c
+++ b/net/sched/sch_gred.c
@@ -23,6 +23,7 @@
 #include <linux/types.h>
 #include <linux/kernel.h>
 #include <linux/skbuff.h>
+#include <net/pkt_cls.h>
 #include <net/pkt_sched.h>
 #include <net/red.h>
 
@@ -311,6 +312,92 @@ static void gred_reset(struct Qdisc *sch)
 	}
 }
 
+static void gred_offload(struct Qdisc *sch, enum tc_gred_command command)
+{
+	struct gred_sched *table = qdisc_priv(sch);
+	struct net_device *dev = qdisc_dev(sch);
+	struct tc_gred_qopt_offload opt = {
+		.command	= command,
+		.handle		= sch->handle,
+		.parent		= sch->parent,
+	};
+
+	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
+		return;
+
+	if (command == TC_GRED_REPLACE) {
+		unsigned int i;
+
+		opt.set.grio_on = gred_rio_mode(table);
+		opt.set.wred_on = gred_wred_mode(table);
+		opt.set.dp_cnt = table->DPs;
+		opt.set.dp_def = table->def;
+
+		for (i = 0; i < table->DPs; i++) {
+			struct gred_sched_data *q = table->tab[i];
+
+			if (!q)
+				continue;
+			opt.set.tab[i].present = true;
+			opt.set.tab[i].limit = q->limit;
+			opt.set.tab[i].prio = q->prio;
+			opt.set.tab[i].min = q->parms.qth_min >> q->parms.Wlog;
+			opt.set.tab[i].max = q->parms.qth_max >> q->parms.Wlog;
+			opt.set.tab[i].is_ecn = gred_use_ecn(q);
+			opt.set.tab[i].is_harddrop = gred_use_harddrop(q);
+			opt.set.tab[i].probability = q->parms.max_P;
+			opt.set.tab[i].backlog = &q->backlog;
+		}
+		opt.set.qstats = &sch->qstats;
+	}
+
+	dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_GRED, &opt);
+}
+
+static int gred_offload_dump_stats(struct Qdisc *sch)
+{
+	struct gred_sched *table = qdisc_priv(sch);
+	struct tc_gred_qopt_offload *hw_stats;
+	unsigned int i;
+	int ret;
+
+	hw_stats = kzalloc(sizeof(*hw_stats), GFP_KERNEL);
+	if (!hw_stats)
+		return -ENOMEM;
+
+	hw_stats->command = TC_GRED_STATS;
+	hw_stats->handle = sch->handle;
+	hw_stats->parent = sch->parent;
+
+	for (i = 0; i < MAX_DPs; i++)
+		if (table->tab[i])
+			hw_stats->stats.xstats[i] = &table->tab[i]->stats;
+
+	ret = qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_GRED, hw_stats);
+	/* Even if driver returns failure adjust the stats - in case offload
+	 * ended but driver still wants to adjust the values.
+	 */
+	for (i = 0; i < MAX_DPs; i++) {
+		if (!table->tab[i])
+			continue;
+		table->tab[i]->packetsin += hw_stats->stats.bstats[i].packets;
+		table->tab[i]->bytesin += hw_stats->stats.bstats[i].bytes;
+		table->tab[i]->backlog += hw_stats->stats.qstats[i].backlog;
+
+		_bstats_update(&sch->bstats,
+			       hw_stats->stats.bstats[i].bytes,
+			       hw_stats->stats.bstats[i].packets);
+		sch->qstats.qlen += hw_stats->stats.qstats[i].qlen;
+		sch->qstats.backlog += hw_stats->stats.qstats[i].backlog;
+		sch->qstats.drops += hw_stats->stats.qstats[i].drops;
+		sch->qstats.requeues += hw_stats->stats.qstats[i].requeues;
+		sch->qstats.overlimits += hw_stats->stats.qstats[i].overlimits;
+	}
+
+	kfree(hw_stats);
+	return ret;
+}
+
 static inline void gred_destroy_vq(struct gred_sched_data *q)
 {
 	kfree(q);
@@ -385,6 +472,7 @@ static int gred_change_table_def(struct Qdisc *sch, struct nlattr *dps,
 		}
 	}
 
+	gred_offload(sch, TC_GRED_REPLACE);
 	return 0;
 }
 
@@ -630,6 +718,8 @@ static int gred_change(struct Qdisc *sch, struct nlattr *opt,
 
 	sch_tree_unlock(sch);
 	kfree(prealloc);
+
+	gred_offload(sch, TC_GRED_REPLACE);
 	return 0;
 
 err_unlock_free:
@@ -679,6 +769,9 @@ static int gred_dump(struct Qdisc *sch, struct sk_buff *skb)
 		.flags	= table->red_flags,
 	};
 
+	if (gred_offload_dump_stats(sch))
+		goto nla_put_failure;
+
 	opts = nla_nest_start(skb, TCA_OPTIONS);
 	if (opts == NULL)
 		goto nla_put_failure;
@@ -815,6 +908,7 @@ static void gred_destroy(struct Qdisc *sch)
 		if (table->tab[i])
 			gred_destroy_vq(table->tab[i]);
 	}
+	gred_offload(sch, TC_GRED_DESTROY);
 }
 
 static struct Qdisc_ops gred_qdisc_ops __read_mostly = {