Resguardo inicial

linux-7.2.3/drivers/net/ethernet/netronome/nfp/abm/cls.c

@@ -0,0 +1,273 @@
+// 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 offsetting 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 offsetting 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_obj(*match);
+		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;
+	}
+}
+
+static LIST_HEAD(nfp_abm_block_cb_list);
+
+int nfp_abm_setup_cls_block(struct net_device *netdev, struct nfp_repr *repr,
+			    struct flow_block_offload *f)
+{
+	return flow_block_cb_setup_simple(f, &nfp_abm_block_cb_list,
+					  nfp_abm_setup_tc_block_cb,
+					  repr, repr, true);
+}

linux-7.2.3/drivers/net/ethernet/netronome/nfp/abm/ctrl.c

@@ -0,0 +1,429 @@
+// 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"
+#include "../nfp_app.h"
+#include "../nfp_abi.h"
+#include "../nfp_main.h"
+#include "../nfp_net.h"
+#include "main.h"
+
+#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_RED_SUPPORT_SYM_NAME	"_abi_nfd_out_red_offload_%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%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 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;
+
+	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 band %d queue %d (+ %d)\n",
+			alink->id, band, queue, alink->queue_base);
+		return err;
+	}
+
+	*res = is_u64 ? val : val32;
+	return 0;
+}
+
+int __nfp_abm_ctrl_set_q_lvl(struct nfp_abm *abm, unsigned int id, u32 val)
+{
+	struct nfp_cpp *cpp = abm->app->cpp;
+	u64 sym_offset;
+	int err;
+
+	__clear_bit(id, abm->threshold_undef);
+	if (abm->thresholds[id] == val)
+		return 0;
+
+	sym_offset = id * NFP_QLVL_STRIDE + NFP_QLVL_THRS;
+	err = __nfp_rtsym_writel(cpp, abm->q_lvls, 4, 0, sym_offset, val);
+	if (err) {
+		nfp_err(cpp,
+			"RED offload setting level failed on subqueue %d\n",
+			id);
+		return err;
+	}
+
+	abm->thresholds[id] = val;
+	return 0;
+}
+
+int nfp_abm_ctrl_set_q_lvl(struct nfp_abm_link *alink, unsigned int band,
+			   unsigned int queue, u32 val)
+{
+	unsigned int threshold;
+
+	threshold = band * NFP_NET_MAX_RX_RINGS + alink->queue_base + queue;
+
+	return __nfp_abm_ctrl_set_q_lvl(alink->abm, threshold, val);
+}
+
+int __nfp_abm_ctrl_set_q_act(struct nfp_abm *abm, unsigned int id,
+			     enum nfp_abm_q_action act)
+{
+	struct nfp_cpp *cpp = abm->app->cpp;
+	u64 sym_offset;
+	int err;
+
+	if (abm->actions[id] == act)
+		return 0;
+
+	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;
+}
+
+static int
+nfp_abm_ctrl_stat_basic(struct nfp_abm_link *alink, unsigned int band,
+			unsigned int queue, unsigned int off, 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;
+		}
+
+		return 0;
+	} 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 band,
+			      unsigned int queue, struct nfp_alink_stats *stats)
+{
+	int err;
+
+	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, 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,
+				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,
+				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,
+				 band, queue, true, &stats->overlimits);
+}
+
+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,
+				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,
+				 band, queue, true, &xstats->ecn_marked);
+}
+
+int nfp_abm_ctrl_qm_enable(struct nfp_abm *abm)
+{
+	return nfp_mbox_cmd(abm->app->pf, NFP_MBOX_PCIE_ABM_ENABLE,
+			    NULL, 0, NULL, 0);
+}
+
+int nfp_abm_ctrl_qm_disable(struct nfp_abm *abm)
+{
+	return nfp_mbox_cmd(abm->app->pf, NFP_MBOX_PCIE_ABM_DISABLE,
+			    NULL, 0, NULL, 0);
+}
+
+int nfp_abm_ctrl_prio_map_update(struct nfp_abm_link *alink, u32 *packed)
+{
+	const u32 cmd = NFP_NET_CFG_MBOX_CMD_PCI_DSCP_PRIOMAP_SET;
+	struct nfp_net *nn = alink->vnic;
+	unsigned int i;
+	int err;
+
+	err = nfp_net_mbox_lock(nn, alink->abm->prio_map_len);
+	if (err)
+		return 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_mbox_reconfig_and_unlock(nn, cmd);
+	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 *
+nfp_abm_ctrl_find_rtsym(struct nfp_pf *pf, const char *name, unsigned int size)
+{
+	const struct nfp_rtsym *sym;
+
+	sym = nfp_rtsym_lookup(pf->rtbl, name);
+	if (!sym) {
+		nfp_err(pf->cpp, "Symbol '%s' not found\n", name);
+		return ERR_PTR(-ENOENT);
+	}
+	if (nfp_rtsym_size(sym) != size) {
+		nfp_err(pf->cpp,
+			"Symbol '%s' wrong size: expected %u got %llu\n",
+			name, size, nfp_rtsym_size(sym));
+		return ERR_PTR(-EINVAL);
+	}
+
+	return sym;
+}
+
+static const struct nfp_rtsym *
+nfp_abm_ctrl_find_q_rtsym(struct nfp_abm *abm, const char *name_fmt,
+			  size_t size)
+{
+	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;
+	int res;
+
+	abm->pf_id = nfp_cppcore_pcie_unit(pf->cpp);
+
+	/* Check if Qdisc offloads are supported */
+	res = nfp_pf_rtsym_read_optional(pf, NFP_RED_SUPPORT_SYM_NAME, 1);
+	if (res < 0)
+		return res;
+	abm->red_support = res;
+
+	/* 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;
+	}
+
+	/* Find level and stat symbols */
+	if (!abm->red_support)
+		return 0;
+
+	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;
+
+	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;
+}

linux-7.2.3/drivers/net/ethernet/netronome/nfp/abm/main.c

@@ -0,0 +1,557 @@
+// SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
+/* Copyright (C) 2018 Netronome Systems, Inc. */
+
+#include <linux/bitfield.h>
+#include <linux/bitmap.h>
+#include <linux/etherdevice.h>
+#include <linux/lockdep.h>
+#include <linux/netdevice.h>
+#include <linux/rcupdate.h>
+#include <linux/rtnetlink.h>
+#include <linux/slab.h>
+
+#include "../nfpcore/nfp.h"
+#include "../nfpcore/nfp_cpp.h"
+#include "../nfpcore/nfp_nsp.h"
+#include "../nfp_app.h"
+#include "../nfp_main.h"
+#include "../nfp_net.h"
+#include "../nfp_net_repr.h"
+#include "../nfp_port.h"
+#include "main.h"
+
+static u32 nfp_abm_portid(enum nfp_repr_type rtype, unsigned int id)
+{
+	return FIELD_PREP(NFP_ABM_PORTID_TYPE, rtype) |
+	       FIELD_PREP(NFP_ABM_PORTID_ID, id);
+}
+
+static int
+nfp_abm_setup_tc(struct nfp_app *app, struct net_device *netdev,
+		 enum tc_setup_type type, void *type_data)
+{
+	struct nfp_repr *repr = netdev_priv(netdev);
+	struct nfp_port *port;
+
+	port = nfp_port_from_netdev(netdev);
+	if (!port || port->type != NFP_PORT_PF_PORT)
+		return -EOPNOTSUPP;
+
+	switch (type) {
+	case TC_SETUP_ROOT_QDISC:
+		return nfp_abm_setup_root(netdev, repr->app_priv, type_data);
+	case TC_SETUP_QDISC_MQ:
+		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;
+	}
+}
+
+static struct net_device *
+nfp_abm_repr_get(struct nfp_app *app, u32 port_id, bool *redir_egress)
+{
+	enum nfp_repr_type rtype;
+	struct nfp_reprs *reprs;
+	u8 port;
+
+	rtype = FIELD_GET(NFP_ABM_PORTID_TYPE, port_id);
+	port = FIELD_GET(NFP_ABM_PORTID_ID, port_id);
+
+	reprs = rcu_dereference(app->reprs[rtype]);
+	if (!reprs)
+		return NULL;
+
+	if (port >= reprs->num_reprs)
+		return NULL;
+
+	return rcu_dereference(reprs->reprs[port]);
+}
+
+static int
+nfp_abm_spawn_repr(struct nfp_app *app, struct nfp_abm_link *alink,
+		   enum nfp_port_type ptype)
+{
+	struct net_device *netdev;
+	enum nfp_repr_type rtype;
+	struct nfp_reprs *reprs;
+	struct nfp_repr *repr;
+	struct nfp_port *port;
+	unsigned int txqs;
+	int err;
+
+	if (ptype == NFP_PORT_PHYS_PORT) {
+		rtype = NFP_REPR_TYPE_PHYS_PORT;
+		txqs = 1;
+	} else {
+		rtype = NFP_REPR_TYPE_PF;
+		txqs = alink->vnic->max_rx_rings;
+	}
+
+	netdev = nfp_repr_alloc_mqs(app, txqs, 1);
+	if (!netdev)
+		return -ENOMEM;
+	repr = netdev_priv(netdev);
+	repr->app_priv = alink;
+
+	port = nfp_port_alloc(app, ptype, netdev);
+	if (IS_ERR(port)) {
+		err = PTR_ERR(port);
+		goto err_free_repr;
+	}
+
+	if (ptype == NFP_PORT_PHYS_PORT) {
+		port->eth_forced = true;
+		err = nfp_port_init_phy_port(app->pf, app, port, alink->id);
+		if (err)
+			goto err_free_port;
+	} else {
+		port->pf_id = alink->abm->pf_id;
+		port->pf_split = app->pf->max_data_vnics > 1;
+		port->pf_split_id = alink->id;
+		port->vnic = alink->vnic->dp.ctrl_bar;
+	}
+
+	SET_NETDEV_DEV(netdev, &alink->vnic->pdev->dev);
+	eth_hw_addr_random(netdev);
+
+	err = nfp_repr_init(app, netdev, nfp_abm_portid(rtype, alink->id),
+			    port, alink->vnic->dp.netdev);
+	if (err)
+		goto err_free_port;
+
+	reprs = nfp_reprs_get_locked(app, rtype);
+	WARN(nfp_repr_get_locked(app, reprs, alink->id), "duplicate repr");
+	rtnl_lock();
+	rcu_assign_pointer(reprs->reprs[alink->id], netdev);
+	rtnl_unlock();
+
+	nfp_info(app->cpp, "%s Port %d Representor(%s) created\n",
+		 ptype == NFP_PORT_PF_PORT ? "PCIe" : "Phys",
+		 alink->id, netdev->name);
+
+	return 0;
+
+err_free_port:
+	nfp_port_free(port);
+err_free_repr:
+	nfp_repr_free(netdev);
+	return err;
+}
+
+static void
+nfp_abm_kill_repr(struct nfp_app *app, struct nfp_abm_link *alink,
+		  enum nfp_repr_type rtype)
+{
+	struct net_device *netdev;
+	struct nfp_reprs *reprs;
+
+	reprs = nfp_reprs_get_locked(app, rtype);
+	netdev = nfp_repr_get_locked(app, reprs, alink->id);
+	if (!netdev)
+		return;
+	rtnl_lock();
+	rcu_assign_pointer(reprs->reprs[alink->id], NULL);
+	rtnl_unlock();
+	synchronize_rcu();
+	/* Cast to make sure nfp_repr_clean_and_free() takes a nfp_repr */
+	nfp_repr_clean_and_free((struct nfp_repr *)netdev_priv(netdev));
+}
+
+static void
+nfp_abm_kill_reprs(struct nfp_abm *abm, struct nfp_abm_link *alink)
+{
+	nfp_abm_kill_repr(abm->app, alink, NFP_REPR_TYPE_PF);
+	nfp_abm_kill_repr(abm->app, alink, NFP_REPR_TYPE_PHYS_PORT);
+}
+
+static void nfp_abm_kill_reprs_all(struct nfp_abm *abm)
+{
+	struct nfp_pf *pf = abm->app->pf;
+	struct nfp_net *nn;
+
+	list_for_each_entry(nn, &pf->vnics, vnic_list)
+		nfp_abm_kill_reprs(abm, (struct nfp_abm_link *)nn->app_priv);
+}
+
+static enum devlink_eswitch_mode nfp_abm_eswitch_mode_get(struct nfp_app *app)
+{
+	struct nfp_abm *abm = app->priv;
+
+	return abm->eswitch_mode;
+}
+
+static int nfp_abm_eswitch_set_legacy(struct nfp_abm *abm)
+{
+	nfp_abm_kill_reprs_all(abm);
+	nfp_abm_ctrl_qm_disable(abm);
+
+	abm->eswitch_mode = DEVLINK_ESWITCH_MODE_LEGACY;
+	return 0;
+}
+
+static void nfp_abm_eswitch_clean_up(struct nfp_abm *abm)
+{
+	if (abm->eswitch_mode != DEVLINK_ESWITCH_MODE_LEGACY)
+		WARN_ON(nfp_abm_eswitch_set_legacy(abm));
+}
+
+static int nfp_abm_eswitch_set_switchdev(struct nfp_abm *abm)
+{
+	struct nfp_app *app = abm->app;
+	struct nfp_pf *pf = app->pf;
+	struct nfp_net *nn;
+	int err;
+
+	if (!abm->red_support)
+		return -EOPNOTSUPP;
+
+	err = nfp_abm_ctrl_qm_enable(abm);
+	if (err)
+		return err;
+
+	list_for_each_entry(nn, &pf->vnics, vnic_list) {
+		struct nfp_abm_link *alink = nn->app_priv;
+
+		err = nfp_abm_spawn_repr(app, alink, NFP_PORT_PHYS_PORT);
+		if (err)
+			goto err_kill_all_reprs;
+
+		err = nfp_abm_spawn_repr(app, alink, NFP_PORT_PF_PORT);
+		if (err)
+			goto err_kill_all_reprs;
+	}
+
+	abm->eswitch_mode = DEVLINK_ESWITCH_MODE_SWITCHDEV;
+	return 0;
+
+err_kill_all_reprs:
+	nfp_abm_kill_reprs_all(abm);
+	nfp_abm_ctrl_qm_disable(abm);
+	return err;
+}
+
+static int nfp_abm_eswitch_mode_set(struct nfp_app *app, u16 mode)
+{
+	struct nfp_abm *abm = app->priv;
+
+	if (abm->eswitch_mode == mode)
+		return 0;
+
+	switch (mode) {
+	case DEVLINK_ESWITCH_MODE_LEGACY:
+		return nfp_abm_eswitch_set_legacy(abm);
+	case DEVLINK_ESWITCH_MODE_SWITCHDEV:
+		return nfp_abm_eswitch_set_switchdev(abm);
+	default:
+		return -EINVAL;
+	}
+}
+
+static void
+nfp_abm_vnic_set_mac(struct nfp_pf *pf, struct nfp_abm *abm, struct nfp_net *nn,
+		     unsigned int id)
+{
+	struct nfp_eth_table_port *eth_port = &pf->eth_tbl->ports[id];
+	u8 mac_addr[ETH_ALEN];
+	struct nfp_nsp *nsp;
+	char hwinfo[32];
+	int err;
+
+	if (id > pf->eth_tbl->count) {
+		nfp_warn(pf->cpp, "No entry for persistent MAC address\n");
+		eth_hw_addr_random(nn->dp.netdev);
+		return;
+	}
+
+	snprintf(hwinfo, sizeof(hwinfo), "eth%u.mac.pf%u",
+		 eth_port->eth_index, abm->pf_id);
+
+	nsp = nfp_nsp_open(pf->cpp);
+	if (IS_ERR(nsp)) {
+		nfp_warn(pf->cpp, "Failed to access the NSP for persistent MAC address: %ld\n",
+			 PTR_ERR(nsp));
+		eth_hw_addr_random(nn->dp.netdev);
+		return;
+	}
+
+	if (!nfp_nsp_has_hwinfo_lookup(nsp)) {
+		nfp_warn(pf->cpp, "NSP doesn't support PF MAC generation\n");
+		eth_hw_addr_random(nn->dp.netdev);
+		nfp_nsp_close(nsp);
+		return;
+	}
+
+	err = nfp_nsp_hwinfo_lookup(nsp, hwinfo, sizeof(hwinfo));
+	nfp_nsp_close(nsp);
+	if (err) {
+		nfp_warn(pf->cpp, "Reading persistent MAC address failed: %d\n",
+			 err);
+		eth_hw_addr_random(nn->dp.netdev);
+		return;
+	}
+
+	if (sscanf(hwinfo, "%02hhx:%02hhx:%02hhx:%02hhx:%02hhx:%02hhx",
+		   &mac_addr[0], &mac_addr[1], &mac_addr[2],
+		   &mac_addr[3], &mac_addr[4], &mac_addr[5]) != 6) {
+		nfp_warn(pf->cpp, "Can't parse persistent MAC address (%s)\n",
+			 hwinfo);
+		eth_hw_addr_random(nn->dp.netdev);
+		return;
+	}
+
+	eth_hw_addr_set(nn->dp.netdev, mac_addr);
+	ether_addr_copy(nn->dp.netdev->perm_addr, mac_addr);
+}
+
+static int
+nfp_abm_vnic_alloc(struct nfp_app *app, struct nfp_net *nn, unsigned int id)
+{
+	struct nfp_eth_table_port *eth_port = &app->pf->eth_tbl->ports[id];
+	struct nfp_abm *abm = app->priv;
+	struct nfp_abm_link *alink;
+	int err;
+
+	alink = kzalloc_obj(*alink);
+	if (!alink)
+		return -ENOMEM;
+	nn->app_priv = alink;
+	alink->abm = abm;
+	alink->vnic = nn;
+	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) {
+		err = -ENOMEM;
+		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_priomap;
+
+	netif_keep_dst(nn->dp.netdev);
+
+	nfp_abm_vnic_set_mac(app->pf, abm, nn, id);
+	INIT_RADIX_TREE(&alink->qdiscs, GFP_KERNEL);
+
+	return 0;
+
+err_free_priomap:
+	kfree(alink->prio_map);
+err_free_alink:
+	kfree(alink);
+	return err;
+}
+
+static void nfp_abm_vnic_free(struct nfp_app *app, struct nfp_net *nn)
+{
+	struct nfp_abm_link *alink = nn->app_priv;
+
+	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)
+{
+	struct nfp_repr *repr = netdev_priv(port->netdev);
+	struct nfp_abm_link *alink;
+	unsigned int i;
+
+	if (port->type != NFP_PORT_PF_PORT)
+		return data;
+	alink = repr->app_priv;
+	for (i = 0; i < alink->vnic->dp.num_r_vecs; i++) {
+		*data++ = nfp_abm_ctrl_stat_non_sto(alink, i);
+		*data++ = nfp_abm_ctrl_stat_sto(alink, i);
+	}
+	return data;
+}
+
+static int
+nfp_abm_port_get_stats_count(struct nfp_app *app, struct nfp_port *port)
+{
+	struct nfp_repr *repr = netdev_priv(port->netdev);
+	struct nfp_abm_link *alink;
+
+	if (port->type != NFP_PORT_PF_PORT)
+		return 0;
+	alink = repr->app_priv;
+	return alink->vnic->dp.num_r_vecs * 2;
+}
+
+static u8 *
+nfp_abm_port_get_stats_strings(struct nfp_app *app, struct nfp_port *port,
+			       u8 *data)
+{
+	struct nfp_repr *repr = netdev_priv(port->netdev);
+	struct nfp_abm_link *alink;
+	unsigned int i;
+
+	if (port->type != NFP_PORT_PF_PORT)
+		return data;
+	alink = repr->app_priv;
+	for (i = 0; i < alink->vnic->dp.num_r_vecs; i++) {
+		ethtool_sprintf(&data, "q%u_no_wait", i);
+		ethtool_sprintf(&data, "q%u_delayed", i);
+	}
+	return data;
+}
+
+static int nfp_abm_fw_init_reset(struct nfp_abm *abm)
+{
+	unsigned int i;
+
+	if (!abm->red_support)
+		return 0;
+
+	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);
+		__nfp_abm_ctrl_set_q_act(abm, i, NFP_ABM_ACT_DROP);
+	}
+
+	return nfp_abm_ctrl_qm_disable(abm);
+}
+
+static int nfp_abm_init(struct nfp_app *app)
+{
+	struct nfp_pf *pf = app->pf;
+	struct nfp_reprs *reprs;
+	struct nfp_abm *abm;
+	int err;
+
+	if (!pf->eth_tbl) {
+		nfp_err(pf->cpp, "ABM NIC requires ETH table\n");
+		return -EINVAL;
+	}
+	if (pf->max_data_vnics != pf->eth_tbl->count) {
+		nfp_err(pf->cpp, "ETH entries don't match vNICs (%d vs %d)\n",
+			pf->max_data_vnics, pf->eth_tbl->count);
+		return -EINVAL;
+	}
+	if (!pf->mac_stats_bar) {
+		nfp_warn(app->cpp, "ABM NIC requires mac_stats symbol\n");
+		return -EINVAL;
+	}
+
+	abm = kzalloc_obj(*abm);
+	if (!abm)
+		return -ENOMEM;
+	app->priv = abm;
+	abm->app = app;
+
+	err = nfp_abm_ctrl_find_addrs(abm);
+	if (err)
+		goto err_free_abm;
+
+	err = -ENOMEM;
+	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;
+
+	abm->thresholds = kvcalloc(abm->num_thresholds,
+				   sizeof(*abm->thresholds), GFP_KERNEL);
+	if (!abm->thresholds)
+		goto err_free_thresh_umap;
+
+	abm->actions = kvcalloc(abm->num_thresholds, sizeof(*abm->actions),
+				GFP_KERNEL);
+	if (!abm->actions)
+		goto err_free_thresh;
+
+	/* We start in legacy mode, make sure advanced queuing is disabled */
+	err = nfp_abm_fw_init_reset(abm);
+	if (err)
+		goto err_free_act;
+
+	err = -ENOMEM;
+	reprs = nfp_reprs_alloc(pf->max_data_vnics);
+	if (!reprs)
+		goto err_free_act;
+	RCU_INIT_POINTER(app->reprs[NFP_REPR_TYPE_PHYS_PORT], reprs);
+
+	reprs = nfp_reprs_alloc(pf->max_data_vnics);
+	if (!reprs)
+		goto err_free_phys;
+	RCU_INIT_POINTER(app->reprs[NFP_REPR_TYPE_PF], reprs);
+
+	return 0;
+
+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:
+	bitmap_free(abm->threshold_undef);
+err_free_abm:
+	kfree(abm);
+	app->priv = NULL;
+	return err;
+}
+
+static void nfp_abm_clean(struct nfp_app *app)
+{
+	struct nfp_abm *abm = app->priv;
+
+	nfp_abm_eswitch_clean_up(abm);
+	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;
+}
+
+const struct nfp_app_type app_abm = {
+	.id		= NFP_APP_ACTIVE_BUFFER_MGMT_NIC,
+	.name		= "abm",
+
+	.init		= nfp_abm_init,
+	.clean		= nfp_abm_clean,
+
+	.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,
+	.port_get_stats_strings	= nfp_abm_port_get_stats_strings,
+
+	.setup_tc	= nfp_abm_setup_tc,
+
+	.eswitch_mode_get	= nfp_abm_eswitch_mode_get,
+	.eswitch_mode_set	= nfp_abm_eswitch_mode_set,
+
+	.dev_get	= nfp_abm_repr_get,
+};

linux-7.2.3/drivers/net/ethernet/netronome/nfp/abm/main.h

@@ -0,0 +1,273 @@
+/* SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) */
+/* Copyright (C) 2018 Netronome Systems, Inc. */
+
+#ifndef __NFP_ABM_H__
+#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.
+ */
+#define NFP_ABM_STATS_REFRESH_IVAL	(2500 * 1000) /* ns */
+
+#define NFP_ABM_LVL_INFINITY		S32_MAX
+
+struct nfp_app;
+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
+ *
+ * @red_support:	is RED offload supported
+ * @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 red_support;
+	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;
+};
+
+/**
+ * struct nfp_alink_stats - ABM NIC statistics
+ * @tx_pkts:		number of TXed packets
+ * @tx_bytes:		number of TXed bytes
+ * @backlog_pkts:	momentary backlog length (packets)
+ * @backlog_bytes:	momentary backlog length (bytes)
+ * @overlimits:		number of ECN marked TXed packets (accumulative)
+ * @drops:		number of tail-dropped packets (accumulative)
+ */
+struct nfp_alink_stats {
+	u64 tx_pkts;
+	u64 tx_bytes;
+	u64 backlog_pkts;
+	u64 backlog_bytes;
+	u64 overlimits;
+	u64 drops;
+};
+
+/**
+ * struct nfp_alink_xstats - extended ABM NIC statistics
+ * @ecn_marked:		number of ECN marked TXed packets
+ * @pdrop:		number of hard drops due to queue limit
+ */
+struct nfp_alink_xstats {
+	u64 ecn_marked;
+	u64 pdrop;
+};
+
+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)
+
+/**
+ * struct nfp_qdisc - tracked TC Qdisc
+ * @netdev:		netdev on which Qdisc was created
+ * @type:		Qdisc type
+ * @handle:		handle of this Qdisc
+ * @parent_handle:	handle of the parent (unreliable if Qdisc was grafted)
+ * @use_cnt:		number of attachment points in the hierarchy
+ * @num_children:	current size of the @children array
+ * @children:		pointers to children
+ *
+ * @params_ok:		parameters of this Qdisc are OK for offload
+ * @offload_mark:	offload refresh state - selected for offload
+ * @offloaded:		Qdisc is currently offloaded to the HW
+ *
+ * @mq:			MQ Qdisc specific parameters and state
+ * @mq.stats:		current stats of the MQ Qdisc
+ * @mq.prev_stats:	previously reported @mq.stats
+ *
+ * @red:		RED Qdisc specific parameters and state
+ * @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;
+	enum nfp_qdisc_type type;
+	u32 handle;
+	u32 parent_handle;
+	unsigned int use_cnt;
+	unsigned int num_children;
+	struct nfp_qdisc **children;
+
+	bool params_ok;
+	bool offload_mark;
+	bool offloaded;
+
+	union {
+		/* NFP_QDISC_MQ */
+		struct {
+			struct nfp_alink_stats stats;
+			struct nfp_alink_stats prev_stats;
+		} mq;
+		/* TC_SETUP_QDISC_RED, TC_SETUP_QDISC_GRED */
+		struct {
+			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;
+	};
+};
+
+/**
+ * struct nfp_abm_link - port tuple of a ABM NIC
+ * @abm:	back pointer to nfp_abm
+ * @vnic:	data vNIC
+ * @id:		id of the data vNIC
+ * @queue_base:	id of base to host queue within PCIe (not QC idx)
+ * @total_queues:	number of PF queues
+ *
+ * @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
+ */
+struct nfp_abm_link {
+	struct nfp_abm *abm;
+	struct nfp_net *vnic;
+	unsigned int id;
+	unsigned int queue_base;
+	unsigned int total_queues;
+
+	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);
+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 flow_block_offload *opt);
+
+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 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 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

linux-7.2.3/drivers/net/ethernet/netronome/nfp/abm/qdisc.c

@@ -0,0 +1,850 @@
+// SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
+/* Copyright (C) 2018 Netronome Systems, Inc. */
+
+#include <linux/rtnetlink.h>
+#include <net/pkt_cls.h>
+#include <net/pkt_sched.h>
+#include <net/red.h>
+
+#include "../nfpcore/nfp_cpp.h"
+#include "../nfp_app.h"
+#include "../nfp_main.h"
+#include "../nfp_net.h"
+#include "../nfp_port.h"
+#include "main.h"
+
+static bool nfp_abm_qdisc_is_red(struct nfp_qdisc *qdisc)
+{
+	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)
+{
+	return qdisc->children[id] &&
+	       qdisc->children[id] != NFP_QDISC_UNTRACKED;
+}
+
+static void *nfp_abm_qdisc_tree_deref_slot(void __rcu **slot)
+{
+	return rtnl_dereference(*slot);
+}
+
+static void
+nfp_abm_stats_propagate(struct nfp_alink_stats *parent,
+			struct nfp_alink_stats *child)
+{
+	parent->tx_pkts		+= child->tx_pkts;
+	parent->tx_bytes	+= child->tx_bytes;
+	parent->backlog_pkts	+= child->backlog_pkts;
+	parent->backlog_bytes	+= child->backlog_bytes;
+	parent->overlimits	+= child->overlimits;
+	parent->drops		+= child->drops;
+}
+
+static void
+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;
+
+	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
+nfp_abm_stats_update_mq(struct nfp_abm_link *alink, struct nfp_qdisc *qdisc)
+{
+	unsigned int i;
+
+	if (qdisc->type != NFP_QDISC_MQ)
+		return;
+
+	for (i = 0; i < alink->total_queues; i++)
+		if (nfp_abm_qdisc_child_valid(qdisc, i))
+			nfp_abm_stats_update_red(alink, qdisc->children[i], i);
+}
+
+static void __nfp_abm_stats_update(struct nfp_abm_link *alink, u64 time_now)
+{
+	alink->last_stats_update = time_now;
+	if (alink->root_qdisc)
+		nfp_abm_stats_update_mq(alink, alink->root_qdisc);
+}
+
+static void nfp_abm_stats_update(struct nfp_abm_link *alink)
+{
+	u64 now;
+
+	/* Limit the frequency of updates - stats of non-leaf qdiscs are a sum
+	 * of all their leafs, so we would read the same stat multiple times
+	 * for every dump.
+	 */
+	now = ktime_get();
+	if (now - alink->last_stats_update < NFP_ABM_STATS_REFRESH_IVAL)
+		return;
+
+	__nfp_abm_stats_update(alink, now);
+}
+
+static void
+nfp_abm_qdisc_unlink_children(struct nfp_qdisc *qdisc,
+			      unsigned int start, unsigned int end)
+{
+	unsigned int i;
+
+	for (i = start; i < end; i++)
+		if (nfp_abm_qdisc_child_valid(qdisc, i)) {
+			qdisc->children[i]->use_cnt--;
+			qdisc->children[i] = NULL;
+		}
+}
+
+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",
+			 qdisc->handle);
+
+	if (!nfp_abm_qdisc_is_red(qdisc))
+		return;
+
+	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, unsigned int band,
+		     unsigned int queue, struct nfp_alink_stats *prev_stats,
+		     struct nfp_alink_xstats *prev_xstats)
+{
+	u64 backlog_pkts, backlog_bytes;
+	int err;
+
+	/* Don't touch the backlog, backlog can only be reset after it has
+	 * been reported back to the tc qdisc stats.
+	 */
+	backlog_pkts = prev_stats->backlog_pkts;
+	backlog_bytes = prev_stats->backlog_bytes;
+
+	err = nfp_abm_ctrl_read_q_stats(alink, band, queue, prev_stats);
+	if (err) {
+		nfp_err(alink->abm->app->cpp,
+			"RED stats init (%d, %d) failed with error %d\n",
+			band, queue, err);
+		return err;
+	}
+
+	err = nfp_abm_ctrl_read_q_xstats(alink, band, queue, prev_xstats);
+	if (err) {
+		nfp_err(alink->abm->app->cpp,
+			"RED xstats init (%d, %d) failed with error %d\n",
+			band, queue, err);
+		return err;
+	}
+
+	prev_stats->backlog_pkts = backlog_pkts;
+	prev_stats->backlog_bytes = backlog_bytes;
+	return 0;
+}
+
+static int
+nfp_abm_stats_init(struct nfp_abm_link *alink, struct nfp_qdisc *qdisc,
+		   unsigned int queue)
+{
+	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)
+{
+	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)
+		if (nfp_abm_stats_init(alink, qdisc, queue))
+			qdisc->offload_mark = false;
+
+	if (!qdisc->offload_mark)
+		return;
+
+	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
+nfp_abm_offload_compile_mq(struct nfp_abm_link *alink, struct nfp_qdisc *qdisc)
+{
+	unsigned int i;
+
+	qdisc->offload_mark = qdisc->type == NFP_QDISC_MQ;
+	if (!qdisc->offload_mark)
+		return;
+
+	for (i = 0; i < alink->total_queues; i++) {
+		struct nfp_qdisc *child = qdisc->children[i];
+
+		if (!nfp_abm_qdisc_child_valid(qdisc, i))
+			continue;
+
+		nfp_abm_offload_compile_red(alink, child, i);
+	}
+}
+
+void nfp_abm_qdisc_offload_update(struct nfp_abm_link *alink)
+{
+	struct nfp_abm *abm = alink->abm;
+	struct radix_tree_iter iter;
+	struct nfp_qdisc *qdisc;
+	void __rcu **slot;
+	size_t i;
+
+	/* Mark all thresholds as unconfigured */
+	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) {
+		qdisc = nfp_abm_qdisc_tree_deref_slot(slot);
+		qdisc->offload_mark = false;
+	}
+
+	if (alink->root_qdisc)
+		nfp_abm_offload_compile_mq(alink, alink->root_qdisc);
+
+	/* Refresh offload status */
+	radix_tree_for_each_slot(slot, &alink->qdiscs, &iter, 0) {
+		qdisc = nfp_abm_qdisc_tree_deref_slot(slot);
+		if (!qdisc->offload_mark && qdisc->offloaded)
+			nfp_abm_qdisc_offload_stop(alink, qdisc);
+		qdisc->offloaded = qdisc->offload_mark;
+	}
+
+	/* Reset the unconfigured thresholds */
+	for (i = 0; i < abm->num_thresholds; i++)
+		if (test_bit(i, abm->threshold_undef))
+			__nfp_abm_ctrl_set_q_lvl(abm, i, NFP_ABM_LVL_INFINITY);
+
+	__nfp_abm_stats_update(alink, ktime_get());
+}
+
+static void
+nfp_abm_qdisc_clear_mq(struct net_device *netdev, struct nfp_abm_link *alink,
+		       struct nfp_qdisc *qdisc)
+{
+	struct radix_tree_iter iter;
+	unsigned int mq_refs = 0;
+	void __rcu **slot;
+
+	if (!qdisc->use_cnt)
+		return;
+	/* MQ doesn't notify well on destruction, we need special handling of
+	 * MQ's children.
+	 */
+	if (qdisc->type == NFP_QDISC_MQ &&
+	    qdisc == alink->root_qdisc &&
+	    netdev->reg_state == NETREG_UNREGISTERING)
+		return;
+
+	/* Count refs held by MQ instances and clear pointers */
+	radix_tree_for_each_slot(slot, &alink->qdiscs, &iter, 0) {
+		struct nfp_qdisc *mq = nfp_abm_qdisc_tree_deref_slot(slot);
+		unsigned int i;
+
+		if (mq->type != NFP_QDISC_MQ || mq->netdev != netdev)
+			continue;
+		for (i = 0; i < mq->num_children; i++)
+			if (mq->children[i] == qdisc) {
+				mq->children[i] = NULL;
+				mq_refs++;
+			}
+	}
+
+	WARN(qdisc->use_cnt != mq_refs, "non-zero qdisc use count: %d (- %d)\n",
+	     qdisc->use_cnt, mq_refs);
+}
+
+static void
+nfp_abm_qdisc_free(struct net_device *netdev, struct nfp_abm_link *alink,
+		   struct nfp_qdisc *qdisc)
+{
+	struct nfp_port *port = nfp_port_from_netdev(netdev);
+
+	if (!qdisc)
+		return;
+	nfp_abm_qdisc_clear_mq(netdev, alink, qdisc);
+	WARN_ON(radix_tree_delete(&alink->qdiscs,
+				  TC_H_MAJ(qdisc->handle)) != qdisc);
+
+	kfree(qdisc->children);
+	kfree(qdisc);
+
+	port->tc_offload_cnt--;
+}
+
+static struct nfp_qdisc *
+nfp_abm_qdisc_alloc(struct net_device *netdev, struct nfp_abm_link *alink,
+		    enum nfp_qdisc_type type, u32 parent_handle, u32 handle,
+		    unsigned int children)
+{
+	struct nfp_port *port = nfp_port_from_netdev(netdev);
+	struct nfp_qdisc *qdisc;
+	int err;
+
+	qdisc = kzalloc_obj(*qdisc);
+	if (!qdisc)
+		return NULL;
+
+	if (children) {
+		qdisc->children = kcalloc(children, sizeof(void *), GFP_KERNEL);
+		if (!qdisc->children)
+			goto err_free_qdisc;
+	}
+
+	qdisc->netdev = netdev;
+	qdisc->type = type;
+	qdisc->parent_handle = parent_handle;
+	qdisc->handle = handle;
+	qdisc->num_children = children;
+
+	err = radix_tree_insert(&alink->qdiscs, TC_H_MAJ(qdisc->handle), qdisc);
+	if (err) {
+		nfp_err(alink->abm->app->cpp,
+			"Qdisc insertion into radix tree failed: %d\n", err);
+		goto err_free_child_tbl;
+	}
+
+	port->tc_offload_cnt++;
+	return qdisc;
+
+err_free_child_tbl:
+	kfree(qdisc->children);
+err_free_qdisc:
+	kfree(qdisc);
+	return NULL;
+}
+
+static struct nfp_qdisc *
+nfp_abm_qdisc_find(struct nfp_abm_link *alink, u32 handle)
+{
+	return radix_tree_lookup(&alink->qdiscs, TC_H_MAJ(handle));
+}
+
+static int
+nfp_abm_qdisc_replace(struct net_device *netdev, struct nfp_abm_link *alink,
+		      enum nfp_qdisc_type type, u32 parent_handle, u32 handle,
+		      unsigned int children, struct nfp_qdisc **qdisc)
+{
+	*qdisc = nfp_abm_qdisc_find(alink, handle);
+	if (*qdisc) {
+		if (WARN_ON((*qdisc)->type != type))
+			return -EINVAL;
+		return 1;
+	}
+
+	*qdisc = nfp_abm_qdisc_alloc(netdev, alink, type, parent_handle, handle,
+				     children);
+	return *qdisc ? 0 : -ENOMEM;
+}
+
+static void
+nfp_abm_qdisc_destroy(struct net_device *netdev, struct nfp_abm_link *alink,
+		      u32 handle)
+{
+	struct nfp_qdisc *qdisc;
+
+	qdisc = nfp_abm_qdisc_find(alink, handle);
+	if (!qdisc)
+		return;
+
+	/* We don't get TC_SETUP_ROOT_QDISC w/ MQ when netdev is unregistered */
+	if (alink->root_qdisc == qdisc)
+		qdisc->use_cnt--;
+
+	nfp_abm_qdisc_unlink_children(qdisc, 0, qdisc->num_children);
+	nfp_abm_qdisc_free(netdev, alink, qdisc);
+
+	if (alink->root_qdisc == qdisc) {
+		alink->root_qdisc = NULL;
+		/* Only root change matters, other changes are acted upon on
+		 * the graft notification.
+		 */
+		nfp_abm_qdisc_offload_update(alink);
+	}
+}
+
+static int
+nfp_abm_qdisc_graft(struct nfp_abm_link *alink, u32 handle, u32 child_handle,
+		    unsigned int id)
+{
+	struct nfp_qdisc *parent, *child;
+
+	parent = nfp_abm_qdisc_find(alink, handle);
+	if (!parent)
+		return 0;
+
+	if (WARN(id >= parent->num_children,
+		 "graft child out of bound %d >= %d\n",
+		 id, parent->num_children))
+		return -EINVAL;
+
+	nfp_abm_qdisc_unlink_children(parent, id, id + 1);
+
+	child = nfp_abm_qdisc_find(alink, child_handle);
+	if (child)
+		child->use_cnt++;
+	else
+		child = NFP_QDISC_UNTRACKED;
+	parent->children[id] = child;
+
+	nfp_abm_qdisc_offload_update(alink);
+
+	return 0;
+}
+
+static void
+nfp_abm_stats_calculate(struct nfp_alink_stats *new,
+			struct nfp_alink_stats *old,
+			struct gnet_stats_basic_sync *bstats,
+			struct gnet_stats_queue *qstats)
+{
+	_bstats_update(bstats, new->tx_bytes - old->tx_bytes,
+		       new->tx_pkts - old->tx_pkts);
+	qstats->qlen += new->backlog_pkts - old->backlog_pkts;
+	qstats->backlog += new->backlog_bytes - old->backlog_bytes;
+	qstats->overlimits += new->overlimits - old->overlimits;
+	qstats->drops += new->drops - old->drops;
+}
+
+static void
+nfp_abm_stats_red_calculate(struct nfp_alink_xstats *new,
+			    struct nfp_alink_xstats *old,
+			    struct red_stats *stats)
+{
+	stats->forced_mark += new->ecn_marked - old->ecn_marked;
+	stats->pdrop += new->pdrop - old->pdrop;
+}
+
+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;
+
+	nfp_abm_stats_update(alink);
+
+	qdisc = nfp_abm_qdisc_find(alink, opt->handle);
+	if (!qdisc || !qdisc->offloaded)
+		return -EOPNOTSUPP;
+
+	nfp_abm_stats_red_calculate(&qdisc->red.band[0].xstats,
+				    &qdisc->red.band[0].prev_xstats,
+				    opt->xstats);
+	qdisc->red.band[0].prev_xstats = qdisc->red.band[0].xstats;
+	return 0;
+}
+
+static int
+nfp_abm_red_stats(struct nfp_abm_link *alink, u32 handle,
+		  struct tc_qopt_offload_stats *stats)
+{
+	struct nfp_qdisc *qdisc;
+
+	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.
+	 */
+
+	nfp_abm_stats_calculate(&qdisc->red.band[0].stats,
+				&qdisc->red.band[0].prev_stats,
+				stats->bstats, stats->qstats);
+	qdisc->red.band[0].prev_stats = qdisc->red.band[0].stats;
+
+	return qdisc->offloaded ? 0 : -EOPNOTSUPP;
+}
+
+static bool
+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 && !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);
+		return false;
+	}
+	if (opt->set.min != opt->set.max) {
+		nfp_warn(cpp, "RED offload failed - unsupported min/max parameters (p:%08x h:%08x)\n",
+			 opt->parent, opt->handle);
+		return false;
+	}
+	if (opt->set.min > NFP_ABM_LVL_INFINITY) {
+		nfp_warn(cpp, "RED offload failed - threshold too large %d > %d (p:%08x h:%08x)\n",
+			 opt->set.min, NFP_ABM_LVL_INFINITY, opt->parent,
+			 opt->handle);
+		return false;
+	}
+
+	return true;
+}
+
+static int
+nfp_abm_red_replace(struct net_device *netdev, struct nfp_abm_link *alink,
+		    struct tc_red_qopt_offload *opt)
+{
+	struct nfp_qdisc *qdisc;
+	int ret;
+
+	ret = nfp_abm_qdisc_replace(netdev, alink, NFP_QDISC_RED, opt->parent,
+				    opt->handle, 1, &qdisc);
+	if (ret < 0)
+		return ret;
+
+	/* If limit != 0 child gets reset */
+	if (opt->set.limit) {
+		if (nfp_abm_qdisc_child_valid(qdisc, 0))
+			qdisc->children[0]->use_cnt--;
+		qdisc->children[0] = NULL;
+	} else {
+		/* Qdisc was just allocated without a limit will use noop_qdisc,
+		 * i.e. a block hole.
+		 */
+		if (!ret)
+			qdisc->children[0] = NFP_QDISC_UNTRACKED;
+	}
+
+	qdisc->params_ok = nfp_abm_red_check_params(alink, opt);
+	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);
+
+	return 0;
+}
+
+int nfp_abm_setup_tc_red(struct net_device *netdev, struct nfp_abm_link *alink,
+			 struct tc_red_qopt_offload *opt)
+{
+	switch (opt->command) {
+	case TC_RED_REPLACE:
+		return nfp_abm_red_replace(netdev, alink, opt);
+	case TC_RED_DESTROY:
+		nfp_abm_qdisc_destroy(netdev, alink, opt->handle);
+		return 0;
+	case TC_RED_STATS:
+		return nfp_abm_red_stats(alink, opt->handle, &opt->stats);
+	case TC_RED_XSTATS:
+		return nfp_abm_red_xstats(alink, opt);
+	case TC_RED_GRAFT:
+		return nfp_abm_qdisc_graft(alink, opt->handle,
+					   opt->child_handle, 0);
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
+static int
+nfp_abm_mq_create(struct net_device *netdev, struct nfp_abm_link *alink,
+		  struct tc_mq_qopt_offload *opt)
+{
+	struct nfp_qdisc *qdisc;
+	int ret;
+
+	ret = nfp_abm_qdisc_replace(netdev, alink, NFP_QDISC_MQ,
+				    TC_H_ROOT, opt->handle, alink->total_queues,
+				    &qdisc);
+	if (ret < 0)
+		return ret;
+
+	qdisc->params_ok = true;
+	qdisc->offloaded = true;
+	nfp_abm_qdisc_offload_update(alink);
+	return 0;
+}
+
+static int
+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, j;
+
+	qdisc = nfp_abm_qdisc_find(alink, handle);
+	if (!qdisc)
+		return -EOPNOTSUPP;
+
+	nfp_abm_stats_update(alink);
+
+	/* MQ stats are summed over the children in the core, so we need
+	 * to add up the unreported child values.
+	 */
+	memset(&qdisc->mq.stats, 0, sizeof(qdisc->mq.stats));
+	memset(&qdisc->mq.prev_stats, 0, sizeof(qdisc->mq.prev_stats));
+
+	for (i = 0; i < qdisc->num_children; i++) {
+		if (!nfp_abm_qdisc_child_valid(qdisc, i))
+			continue;
+
+		if (!nfp_abm_qdisc_is_red(qdisc->children[i]))
+			continue;
+		red = qdisc->children[i];
+
+		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,
+				stats->bstats, stats->qstats);
+
+	return qdisc->offloaded ? 0 : -EOPNOTSUPP;
+}
+
+int nfp_abm_setup_tc_mq(struct net_device *netdev, struct nfp_abm_link *alink,
+			struct tc_mq_qopt_offload *opt)
+{
+	switch (opt->command) {
+	case TC_MQ_CREATE:
+		return nfp_abm_mq_create(netdev, alink, opt);
+	case TC_MQ_DESTROY:
+		nfp_abm_qdisc_destroy(netdev, alink, opt->handle);
+		return 0;
+	case TC_MQ_STATS:
+		return nfp_abm_mq_stats(alink, opt->handle, &opt->stats);
+	case TC_MQ_GRAFT:
+		return nfp_abm_qdisc_graft(alink, opt->handle,
+					   opt->graft_params.child_handle,
+					   opt->graft_params.queue);
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
+int nfp_abm_setup_root(struct net_device *netdev, struct nfp_abm_link *alink,
+		       struct tc_root_qopt_offload *opt)
+{
+	if (opt->ingress)
+		return -EOPNOTSUPP;
+	if (alink->root_qdisc)
+		alink->root_qdisc->use_cnt--;
+	alink->root_qdisc = nfp_abm_qdisc_find(alink, opt->handle);
+	if (alink->root_qdisc)
+		alink->root_qdisc->use_cnt++;
+
+	nfp_abm_qdisc_offload_update(alink);
+
+	return 0;
+}