Merge branch 'pps-policing'

Simon Horman says:

====================
net/sched: act_police: add support for packet-per-second policing

This series enhances the TC policer action implementation to allow a
policer action instance to enforce a rate-limit based on
packets-per-second, configurable using a packet-per-second rate and burst
parameters.

In the hope of aiding review this is broken up into three patches.

* [PATCH 1/3] flow_offload: add support for packet-per-second policing

  Add support for this feature to the flow_offload API that is used to allow
  programming flows, including TC rules and their actions, into hardware.

* [PATCH 2/3] flow_offload: reject configuration of packet-per-second policing in offload drivers

  Teach all exiting users of the flow_offload API that allow offload of
  policer action instances to reject offload if packet-per-second rate
  limiting is configured: none support it at this time

* [PATCH 3/3] net/sched: act_police: add support for packet-per-second policing

  With the above ground-work in place add the new feature to the TC policer
  action itself

With the above in place the feature may be used.

As follow-ups we plan to provide:
* Corresponding updates to iproute2
* Corresponding self tests (which depend on the iproute2 changes)
* Hardware offload support for the NFP driver

Key changes since v2:
* Added patches 1 and 2, which makes adding patch 3 safe for existing
  hardware offload of the policer action
* Re-worked patch 3 so that a TC policer action instance may be configured
  for packet-per-second or byte-per-second rate limiting, but not both.
* Corrected kdoc usage
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/dsa/sja1105/sja1105_flower.c

@@ -322,6 +322,12 @@ int sja1105_cls_flower_add(struct dsa_switch *ds, int port,
 	flow_action_for_each(i, act, &rule->action) {
 		switch (act->id) {
 		case FLOW_ACTION_POLICE:
+			if (act->police.rate_pkt_ps) {
+				NL_SET_ERR_MSG_MOD(extack,
+						   "QoS offload not support packets per second");
+				goto out;
+			}
+
 			rc = sja1105_flower_policer(priv, port, extack, cookie,
 						    &key,
 						    act->police.rate_bytes_ps,

drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_matchall.c

@@ -48,6 +48,11 @@ static int cxgb4_matchall_egress_validate(struct net_device *dev,
 	flow_action_for_each(i, entry, actions) {
 		switch (entry->id) {
 		case FLOW_ACTION_POLICE:
+			if (entry->police.rate_pkt_ps) {
+				NL_SET_ERR_MSG_MOD(extack,
+						   "QoS offload not support packets per second");
+				return -EOPNOTSUPP;
+			}
 			/* Convert bytes per second to bits per second */
 			if (entry->police.rate_bytes_ps * 8 > max_link_rate) {
 				NL_SET_ERR_MSG_MOD(extack,
@@ -145,7 +150,11 @@ static int cxgb4_matchall_alloc_tc(struct net_device *dev,
 	flow_action_for_each(i, entry, &cls->rule->action)
 		if (entry->id == FLOW_ACTION_POLICE)
 			break;
-
+	if (entry->police.rate_pkt_ps) {
+		NL_SET_ERR_MSG_MOD(extack,
+				   "QoS offload not support packets per second");
+		return -EOPNOTSUPP;
+	}
 	/* Convert from bytes per second to Kbps */
 	p.u.params.maxrate = div_u64(entry->police.rate_bytes_ps * 8, 1000);
 	p.u.params.channel = pi->tx_chan;

drivers/net/ethernet/freescale/enetc/enetc_qos.c

@@ -1221,6 +1221,11 @@ static int enetc_psfp_parse_clsflower(struct enetc_ndev_priv *priv,
 
 	/* Flow meter and max frame size */
 	if (entryp) {
+		if (entryp->police.rate_pkt_ps) {
+			NL_SET_ERR_MSG_MOD(extack, "QoS offload not support packets per second");
+			err = -EOPNOTSUPP;
+			goto free_sfi;
+		}
 		if (entryp->police.burst) {
 			fmi = kzalloc(sizeof(*fmi), GFP_KERNEL);
 			if (!fmi) {

drivers/net/ethernet/mellanox/mlx5/core/en_tc.c

@@ -4538,6 +4538,10 @@ static int scan_tc_matchall_fdb_actions(struct mlx5e_priv *priv,
 	flow_action_for_each(i, act, flow_action) {
 		switch (act->id) {
 		case FLOW_ACTION_POLICE:
+			if (act->police.rate_pkt_ps) {
+				NL_SET_ERR_MSG_MOD(extack, "QoS offload not support packets per second");
+				return -EOPNOTSUPP;
+			}
 			err = apply_police_params(priv, act->police.rate_bytes_ps, extack);
 			if (err)
 				return err;

drivers/net/ethernet/mellanox/mlxsw/spectrum_flower.c

@@ -190,6 +190,11 @@ static int mlxsw_sp_flower_parse_actions(struct mlxsw_sp *mlxsw_sp,
 				return -EOPNOTSUPP;
 			}
 
+			if (act->police.rate_pkt_ps) {
+				NL_SET_ERR_MSG_MOD(extack, "QoS offload not support packets per second");
+				return -EOPNOTSUPP;
+			}
+
 			/* The kernel might adjust the requested burst size so
 			 * that it is not exactly a power of two. Re-adjust it
 			 * here since the hardware only supports burst sizes

drivers/net/ethernet/mscc/ocelot_flower.c

@@ -220,6 +220,11 @@ static int ocelot_flower_parse_action(struct ocelot *ocelot, int port,
 						   "Last action must be GOTO");
 				return -EOPNOTSUPP;
 			}
+			if (a->police.rate_pkt_ps) {
+				NL_SET_ERR_MSG_MOD(extack,
+						   "QoS offload not support packets per second");
+				return -EOPNOTSUPP;
+			}
 			filter->action.police_ena = true;
 			rate = a->police.rate_bytes_ps;
 			filter->action.pol.rate = div_u64(rate, 1000) * 8;

drivers/net/ethernet/mscc/ocelot_net.c

@@ -251,6 +251,12 @@ static int ocelot_setup_tc_cls_matchall(struct ocelot_port_private *priv,
 			return -EEXIST;
 		}
 
+		if (action->police.rate_pkt_ps) {
+			NL_SET_ERR_MSG_MOD(extack,
+					   "QoS offload not support packets per second");
+			return -EOPNOTSUPP;
+		}
+
 		pol.rate = (u32)div_u64(action->police.rate_bytes_ps, 1000) * 8;
 		pol.burst = action->police.burst;
 

drivers/net/ethernet/netronome/nfp/flower/qos_conf.c

@@ -104,6 +104,11 @@ nfp_flower_install_rate_limiter(struct nfp_app *app, struct net_device *netdev,
 		return -EOPNOTSUPP;
 	}
 
+	if (action->police.rate_pkt_ps) {
+		NL_SET_ERR_MSG_MOD(extack, "unsupported offload: qos rate limit offload not support packets per second");
+		return -EOPNOTSUPP;
+	}
+
 	rate = action->police.rate_bytes_ps;
 	burst = action->police.burst;
 	netdev_port_id = nfp_repr_get_port_id(netdev);

include/net/flow_offload.h

@@ -234,6 +234,8 @@ struct flow_action_entry {
 			u32			index;
 			u32			burst;
 			u64			rate_bytes_ps;
+			u64			burst_pkt;
+			u64			rate_pkt_ps;
 			u32			mtu;
 		} police;
 		struct {				/* FLOW_ACTION_CT */

include/net/sch_generic.h

@@ -1242,6 +1242,20 @@ static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
 	res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
 }
 
+struct psched_pktrate {
+	u64	rate_pkts_ps; /* packets per second */
+	u32	mult;
+	u8	shift;
+};
+
+static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
+				  unsigned int pkt_num)
+{
+	return ((u64)pkt_num * r->mult) >> r->shift;
+}
+
+void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
+
 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
  * The fast path only needs to access filter list and to update stats
  */

include/net/tc_act/tc_police.h

@@ -10,10 +10,13 @@ struct tcf_police_params {
 	s64			tcfp_burst;
 	u32			tcfp_mtu;
 	s64			tcfp_mtu_ptoks;
+	s64			tcfp_pkt_burst;
 	struct psched_ratecfg	rate;
 	bool			rate_present;
 	struct psched_ratecfg	peak;
 	bool			peak_present;
+	struct psched_pktrate	ppsrate;
+	bool			pps_present;
 	struct rcu_head rcu;
 };
 
@@ -24,6 +27,7 @@ struct tcf_police {
 	spinlock_t		tcfp_lock ____cacheline_aligned_in_smp;
 	s64			tcfp_toks;
 	s64			tcfp_ptoks;
+	s64			tcfp_pkttoks;
 	s64			tcfp_t_c;
 };
 
@@ -97,6 +101,54 @@ static inline u32 tcf_police_burst(const struct tc_action *act)
 	return burst;
 }
 
+static inline u64 tcf_police_rate_pkt_ps(const struct tc_action *act)
+{
+	struct tcf_police *police = to_police(act);
+	struct tcf_police_params *params;
+
+	params = rcu_dereference_protected(police->params,
+					   lockdep_is_held(&police->tcf_lock));
+	return params->ppsrate.rate_pkts_ps;
+}
+
+static inline u32 tcf_police_burst_pkt(const struct tc_action *act)
+{
+	struct tcf_police *police = to_police(act);
+	struct tcf_police_params *params;
+	u32 burst;
+
+	params = rcu_dereference_protected(police->params,
+					   lockdep_is_held(&police->tcf_lock));
+
+	/*
+	 *  "rate" pkts     "burst" nanoseconds
+	 *  ------------ *  -------------------
+	 *    1 second          2^6 ticks
+	 *
+	 * ------------------------------------
+	 *        NSEC_PER_SEC nanoseconds
+	 *        ------------------------
+	 *              2^6 ticks
+	 *
+	 *    "rate" pkts    "burst" nanoseconds            2^6 ticks
+	 *  = ------------ * ------------------- * ------------------------
+	 *      1 second          2^6 ticks        NSEC_PER_SEC nanoseconds
+	 *
+	 *   "rate" * "burst"
+	 * = ---------------- pkts/nanosecond
+	 *    NSEC_PER_SEC^2
+	 *
+	 *
+	 *   "rate" * "burst"
+	 * = ---------------- pkts/second
+	 *     NSEC_PER_SEC
+	 */
+	burst = div_u64(params->tcfp_pkt_burst * params->ppsrate.rate_pkts_ps,
+			NSEC_PER_SEC);
+
+	return burst;
+}
+
 static inline u32 tcf_police_tcfp_mtu(const struct tc_action *act)
 {
 	struct tcf_police *police = to_police(act);

include/uapi/linux/pkt_cls.h

@@ -190,6 +190,8 @@ enum {
 	TCA_POLICE_PAD,
 	TCA_POLICE_RATE64,
 	TCA_POLICE_PEAKRATE64,
+	TCA_POLICE_PKTRATE64,
+	TCA_POLICE_PKTBURST64,
 	__TCA_POLICE_MAX
 #define TCA_POLICE_RESULT TCA_POLICE_RESULT
 };

net/sched/act_police.c

@@ -42,6 +42,8 @@ static const struct nla_policy police_policy[TCA_POLICE_MAX + 1] = {
 	[TCA_POLICE_RESULT]	= { .type = NLA_U32 },
 	[TCA_POLICE_RATE64]     = { .type = NLA_U64 },
 	[TCA_POLICE_PEAKRATE64] = { .type = NLA_U64 },
+	[TCA_POLICE_PKTRATE64]  = { .type = NLA_U64, .min = 1 },
+	[TCA_POLICE_PKTBURST64] = { .type = NLA_U64, .min = 1 },
 };
 
 static int tcf_police_init(struct net *net, struct nlattr *nla,
@@ -61,6 +63,7 @@ static int tcf_police_init(struct net *net, struct nlattr *nla,
 	bool exists = false;
 	u32 index;
 	u64 rate64, prate64;
+	u64 pps, ppsburst;
 
 	if (nla == NULL)
 		return -EINVAL;
@@ -142,6 +145,21 @@ static int tcf_police_init(struct net *net, struct nlattr *nla,
 		}
 	}
 
+	if ((tb[TCA_POLICE_PKTRATE64] && !tb[TCA_POLICE_PKTBURST64]) ||
+	    (!tb[TCA_POLICE_PKTRATE64] && tb[TCA_POLICE_PKTBURST64])) {
+		NL_SET_ERR_MSG(extack,
+			       "Both or neither packet-per-second burst and rate must be provided");
+		err = -EINVAL;
+		goto failure;
+	}
+
+	if (tb[TCA_POLICE_PKTRATE64] && R_tab) {
+		NL_SET_ERR_MSG(extack,
+			       "packet-per-second and byte-per-second rate limits not allowed in same action");
+		err = -EINVAL;
+		goto failure;
+	}
+
 	new = kzalloc(sizeof(*new), GFP_KERNEL);
 	if (unlikely(!new)) {
 		err = -ENOMEM;
@@ -183,6 +201,14 @@ static int tcf_police_init(struct net *net, struct nlattr *nla,
 	if (tb[TCA_POLICE_AVRATE])
 		new->tcfp_ewma_rate = nla_get_u32(tb[TCA_POLICE_AVRATE]);
 
+	if (tb[TCA_POLICE_PKTRATE64]) {
+		pps = nla_get_u64(tb[TCA_POLICE_PKTRATE64]);
+		ppsburst = nla_get_u64(tb[TCA_POLICE_PKTBURST64]);
+		new->pps_present = true;
+		new->tcfp_pkt_burst = PSCHED_TICKS2NS(ppsburst);
+		psched_ppscfg_precompute(&new->ppsrate, pps);
+	}
+
 	spin_lock_bh(&police->tcf_lock);
 	spin_lock_bh(&police->tcfp_lock);
 	police->tcfp_t_c = ktime_get_ns();
@@ -217,8 +243,8 @@ static int tcf_police_act(struct sk_buff *skb, const struct tc_action *a,
 			  struct tcf_result *res)
 {
 	struct tcf_police *police = to_police(a);
+	s64 now, toks, ppstoks = 0, ptoks = 0;
 	struct tcf_police_params *p;
-	s64 now, toks, ptoks = 0;
 	int ret;
 
 	tcf_lastuse_update(&police->tcf_tm);
@@ -236,7 +262,7 @@ static int tcf_police_act(struct sk_buff *skb, const struct tc_action *a,
 	}
 
 	if (qdisc_pkt_len(skb) <= p->tcfp_mtu) {
-		if (!p->rate_present) {
+		if (!p->rate_present && !p->pps_present) {
 			ret = p->tcfp_result;
 			goto end;
 		}
@@ -251,14 +277,23 @@ static int tcf_police_act(struct sk_buff *skb, const struct tc_action *a,
 			ptoks -= (s64)psched_l2t_ns(&p->peak,
 						    qdisc_pkt_len(skb));
 		}
-		toks += police->tcfp_toks;
-		if (toks > p->tcfp_burst)
-			toks = p->tcfp_burst;
-		toks -= (s64)psched_l2t_ns(&p->rate, qdisc_pkt_len(skb));
-		if ((toks|ptoks) >= 0) {
+		if (p->rate_present) {
+			toks += police->tcfp_toks;
+			if (toks > p->tcfp_burst)
+				toks = p->tcfp_burst;
+			toks -= (s64)psched_l2t_ns(&p->rate, qdisc_pkt_len(skb));
+		} else if (p->pps_present) {
+			ppstoks = min_t(s64, now - police->tcfp_t_c, p->tcfp_pkt_burst);
+			ppstoks += police->tcfp_pkttoks;
+			if (ppstoks > p->tcfp_pkt_burst)
+				ppstoks = p->tcfp_pkt_burst;
+			ppstoks -= (s64)psched_pkt2t_ns(&p->ppsrate, 1);
+		}
+		if ((toks | ptoks | ppstoks) >= 0) {
 			police->tcfp_t_c = now;
 			police->tcfp_toks = toks;
 			police->tcfp_ptoks = ptoks;
+			police->tcfp_pkttoks = ppstoks;
 			spin_unlock_bh(&police->tcfp_lock);
 			ret = p->tcfp_result;
 			goto inc_drops;
@@ -331,6 +366,16 @@ static int tcf_police_dump(struct sk_buff *skb, struct tc_action *a,
 				      TCA_POLICE_PAD))
 			goto nla_put_failure;
 	}
+	if (p->pps_present) {
+		if (nla_put_u64_64bit(skb, TCA_POLICE_PKTRATE64,
+				      police->params->ppsrate.rate_pkts_ps,
+				      TCA_POLICE_PAD))
+			goto nla_put_failure;
+		if (nla_put_u64_64bit(skb, TCA_POLICE_PKTBURST64,
+				      PSCHED_NS2TICKS(p->tcfp_pkt_burst),
+				      TCA_POLICE_PAD))
+			goto nla_put_failure;
+	}
 	if (nla_put(skb, TCA_POLICE_TBF, sizeof(opt), &opt))
 		goto nla_put_failure;
 	if (p->tcfp_result &&

net/sched/cls_api.c

@@ -3661,6 +3661,9 @@ int tc_setup_flow_action(struct flow_action *flow_action,
 			entry->police.burst = tcf_police_burst(act);
 			entry->police.rate_bytes_ps =
 				tcf_police_rate_bytes_ps(act);
+			entry->police.burst_pkt = tcf_police_burst_pkt(act);
+			entry->police.rate_pkt_ps =
+				tcf_police_rate_pkt_ps(act);
 			entry->police.mtu = tcf_police_tcfp_mtu(act);
 			entry->police.index = act->tcfa_index;
 		} else if (is_tcf_ct(act)) {

net/sched/sch_generic.c

@@ -1325,6 +1325,48 @@ void dev_shutdown(struct net_device *dev)
 	WARN_ON(timer_pending(&dev->watchdog_timer));
 }
 
+/**
+ * psched_ratecfg_precompute__() - Pre-compute values for reciprocal division
+ * @rate:   Rate to compute reciprocal division values of
+ * @mult:   Multiplier for reciprocal division
+ * @shift:  Shift for reciprocal division
+ *
+ * The multiplier and shift for reciprocal division by rate are stored
+ * in mult and shift.
+ *
+ * The deal here is to replace a divide by a reciprocal one
+ * in fast path (a reciprocal divide is a multiply and a shift)
+ *
+ * Normal formula would be :
+ *  time_in_ns = (NSEC_PER_SEC * len) / rate_bps
+ *
+ * We compute mult/shift to use instead :
+ *  time_in_ns = (len * mult) >> shift;
+ *
+ * We try to get the highest possible mult value for accuracy,
+ * but have to make sure no overflows will ever happen.
+ *
+ * reciprocal_value() is not used here it doesn't handle 64-bit values.
+ */
+static void psched_ratecfg_precompute__(u64 rate, u32 *mult, u8 *shift)
+{
+	u64 factor = NSEC_PER_SEC;
+
+	*mult = 1;
+	*shift = 0;
+
+	if (rate <= 0)
+		return;
+
+	for (;;) {
+		*mult = div64_u64(factor, rate);
+		if (*mult & (1U << 31) || factor & (1ULL << 63))
+			break;
+		factor <<= 1;
+		(*shift)++;
+	}
+}
+
 void psched_ratecfg_precompute(struct psched_ratecfg *r,
 			       const struct tc_ratespec *conf,
 			       u64 rate64)
@@ -1333,34 +1375,17 @@ void psched_ratecfg_precompute(struct psched_ratecfg *r,
 	r->overhead = conf->overhead;
 	r->rate_bytes_ps = max_t(u64, conf->rate, rate64);
 	r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK);
-	r->mult = 1;
-	/*
-	 * The deal here is to replace a divide by a reciprocal one
-	 * in fast path (a reciprocal divide is a multiply and a shift)
-	 *
-	 * Normal formula would be :
-	 *  time_in_ns = (NSEC_PER_SEC * len) / rate_bps
-	 *
-	 * We compute mult/shift to use instead :
-	 *  time_in_ns = (len * mult) >> shift;
-	 *
-	 * We try to get the highest possible mult value for accuracy,
-	 * but have to make sure no overflows will ever happen.
-	 */
-	if (r->rate_bytes_ps > 0) {
-		u64 factor = NSEC_PER_SEC;
-
-		for (;;) {
-			r->mult = div64_u64(factor, r->rate_bytes_ps);
-			if (r->mult & (1U << 31) || factor & (1ULL << 63))
-				break;
-			factor <<= 1;
-			r->shift++;
-		}
-	}
+	psched_ratecfg_precompute__(r->rate_bytes_ps, &r->mult, &r->shift);
 }
 EXPORT_SYMBOL(psched_ratecfg_precompute);
 
+void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64)
+{
+	r->rate_pkts_ps = pktrate64;
+	psched_ratecfg_precompute__(r->rate_pkts_ps, &r->mult, &r->shift);
+}
+EXPORT_SYMBOL(psched_ppscfg_precompute);
+
 static void mini_qdisc_rcu_func(struct rcu_head *head)
 {
 }