net: dsa: sja1105: implement tc-gate using time-triggered virtual links

Restrict the TTEthernet hardware support on this switch to operate as
closely as possible to IEEE 802.1Qci as possible. This means that it can
perform PTP-time-based ingress admission control on streams identified
by {DMAC, VID, PCP}, which is useful when trying to ensure the
determinism of traffic scheduled via IEEE 802.1Qbv.

The oddity comes from the fact that in hardware (and in TTEthernet at
large), virtual links always need a full-blown action, including not
only the type of policing, but also the list of destination ports. So in
practice, a single tc-gate action will result in all packets getting
dropped. Additional actions (either "trap" or "redirect") need to be
specified in the same filter rule such that the conforming packets are
actually forwarded somewhere.

Apart from the VL Lookup, Policing and Forwarding tables which need to
be programmed for each flow (virtual link), the Schedule engine also
needs to be told to open/close the admission gates for each individual
virtual link. A fairly accurate (and detailed) description of how that
works is already present in sja1105_tas.c, since it is already used to
trigger the egress gates for the tc-taprio offload (IEEE 802.1Qbv). Key
point here, we remember that the schedule engine supports 8
"subschedules" (execution threads that iterate through the global
schedule in parallel, and that no 2 hardware threads must execute a
schedule entry at the same time). For tc-taprio, each egress port used
one of these 8 subschedules, leaving a total of 4 subschedules unused.
In principle we could have allocated 1 subschedule for the tc-gate
offload of each ingress port, but actually the schedules of all virtual
links installed on each ingress port would have needed to be merged
together, before they could have been programmed to hardware. So
simplify our life and just merge the entire tc-gate configuration, for
all virtual links on all ingress ports, into a single subschedule. Be
sure to check that against the usual hardware scheduling conflicts, and
program it to hardware alongside any tc-taprio subschedule that may be
present.

The following scenarios were tested:

1. Quantitative testing:

   tc qdisc add dev swp2 clsact
   tc filter add dev swp2 ingress flower skip_sw \
           dst_mac 42:be:24:9b:76:20 \
           action gate index 1 base-time 0 \
           sched-entry OPEN 1200 -1 -1 \
           sched-entry CLOSE 1200 -1 -1 \
           action trap

   ping 192.168.1.2 -f
   PING 192.168.1.2 (192.168.1.2) 56(84) bytes of data.
   .............................
   --- 192.168.1.2 ping statistics ---
   948 packets transmitted, 467 received, 50.7384% packet loss, time 9671ms

2. Qualitative testing (with a phase-aligned schedule - the clocks are
   synchronized by ptp4l, not shown here):

   Receiver (sja1105):

   tc qdisc add dev swp2 clsact
   now=$(phc_ctl /dev/ptp1 get | awk '/clock time is/ {print $5}') && \
           sec=$(echo $now | awk -F. '{print $1}') && \
           base_time="$(((sec + 2) * 1000000000))" && \
           echo "base time ${base_time}"
   tc filter add dev swp2 ingress flower skip_sw \
           dst_mac 42:be:24:9b:76:20 \
           action gate base-time ${base_time} \
           sched-entry OPEN  60000 -1 -1 \
           sched-entry CLOSE 40000 -1 -1 \
           action trap

   Sender (enetc):
   now=$(phc_ctl /dev/ptp0 get | awk '/clock time is/ {print $5}') && \
           sec=$(echo $now | awk -F. '{print $1}') && \
           base_time="$(((sec + 2) * 1000000000))" && \
           echo "base time ${base_time}"
   tc qdisc add dev eno0 parent root taprio \
           num_tc 8 \
           map 0 1 2 3 4 5 6 7 \
           queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 \
           base-time ${base_time} \
           sched-entry S 01  50000 \
           sched-entry S 00  50000 \
           flags 2

   ping -A 192.168.1.1
   PING 192.168.1.1 (192.168.1.1): 56 data bytes
   ...
   ^C
   --- 192.168.1.1 ping statistics ---
   1425 packets transmitted, 1424 packets received, 0% packet loss
   round-trip min/avg/max = 0.322/0.361/0.990 ms

   And just for comparison, with the tc-taprio schedule deleted:

   ping -A 192.168.1.1
   PING 192.168.1.1 (192.168.1.1): 56 data bytes
   ...
   ^C
   --- 192.168.1.1 ping statistics ---
   33 packets transmitted, 19 packets received, 42% packet loss
   round-trip min/avg/max = 0.336/0.464/0.597 ms

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 112 insertions, 15 deletions

diff --git a/drivers/net/dsa/sja1105/sja1105_tas.c b/drivers/net/dsa/sja1105/sja1105_tas.c
index 77e547b4cd89..3aa1a8b5f766 100644
--- a/drivers/net/dsa/sja1105/sja1105_tas.c
+++ b/drivers/net/dsa/sja1105/sja1105_tas.c
@@ -7,7 +7,6 @@
 #define SJA1105_TAS_CLKSRC_STANDALONE	1
 #define SJA1105_TAS_CLKSRC_AS6802	2
 #define SJA1105_TAS_CLKSRC_PTP		3
-#define SJA1105_TAS_MAX_DELTA		BIT(19)
 #define SJA1105_GATE_MASK		GENMASK_ULL(SJA1105_NUM_TC - 1, 0)
 
 #define work_to_sja1105_tas(d) \
@@ -15,22 +14,10 @@
 #define tas_to_sja1105(d) \
 	container_of((d), struct sja1105_private, tas_data)
 
-/* This is not a preprocessor macro because the "ns" argument may or may not be
- * s64 at caller side. This ensures it is properly type-cast before div_s64.
- */
-static s64 ns_to_sja1105_delta(s64 ns)
-{
-	return div_s64(ns, 200);
-}
-
-static s64 sja1105_delta_to_ns(s64 delta)
-{
-	return delta * 200;
-}
-
 static int sja1105_tas_set_runtime_params(struct sja1105_private *priv)
 {
 	struct sja1105_tas_data *tas_data = &priv->tas_data;
+	struct sja1105_gating_config *gating_cfg = &tas_data->gating_cfg;
 	struct dsa_switch *ds = priv->ds;
 	s64 earliest_base_time = S64_MAX;
 	s64 latest_base_time = 0;
@@ -59,6 +46,19 @@ static int sja1105_tas_set_runtime_params(struct sja1105_private *priv)
 		}
 	}
 
+	if (!list_empty(&gating_cfg->entries)) {
+		tas_data->enabled = true;
+
+		if (max_cycle_time < gating_cfg->cycle_time)
+			max_cycle_time = gating_cfg->cycle_time;
+		if (latest_base_time < gating_cfg->base_time)
+			latest_base_time = gating_cfg->base_time;
+		if (earliest_base_time > gating_cfg->base_time) {
+			earliest_base_time = gating_cfg->base_time;
+			its_cycle_time = gating_cfg->cycle_time;
+		}
+	}
+
 	if (!tas_data->enabled)
 		return 0;
 
@@ -155,13 +155,14 @@ static int sja1105_tas_set_runtime_params(struct sja1105_private *priv)
  *  their "subschedule end index" (subscheind) equal to the last valid
  *  subschedule's end index (in this case 5).
  */
-static int sja1105_init_scheduling(struct sja1105_private *priv)
+int sja1105_init_scheduling(struct sja1105_private *priv)
 {
 	struct sja1105_schedule_entry_points_entry *schedule_entry_points;
 	struct sja1105_schedule_entry_points_params_entry
 					*schedule_entry_points_params;
 	struct sja1105_schedule_params_entry *schedule_params;
 	struct sja1105_tas_data *tas_data = &priv->tas_data;
+	struct sja1105_gating_config *gating_cfg = &tas_data->gating_cfg;
 	struct sja1105_schedule_entry *schedule;
 	struct sja1105_table *table;
 	int schedule_start_idx;
@@ -213,6 +214,11 @@ static int sja1105_init_scheduling(struct sja1105_private *priv)
 		}
 	}
 
+	if (!list_empty(&gating_cfg->entries)) {
+		num_entries += gating_cfg->num_entries;
+		num_cycles++;
+	}
+
 	/* Nothing to do */
 	if (!num_cycles)
 		return 0;
@@ -312,6 +318,42 @@ static int sja1105_init_scheduling(struct sja1105_private *priv)
 		cycle++;
 	}
 
+	if (!list_empty(&gating_cfg->entries)) {
+		struct sja1105_gate_entry *e;
+
+		/* Relative base time */
+		s64 rbt;
+
+		schedule_start_idx = k;
+		schedule_end_idx = k + gating_cfg->num_entries - 1;
+		rbt = future_base_time(gating_cfg->base_time,
+				       gating_cfg->cycle_time,
+				       tas_data->earliest_base_time);
+		rbt -= tas_data->earliest_base_time;
+		entry_point_delta = ns_to_sja1105_delta(rbt) + 1;
+
+		schedule_entry_points[cycle].subschindx = cycle;
+		schedule_entry_points[cycle].delta = entry_point_delta;
+		schedule_entry_points[cycle].address = schedule_start_idx;
+
+		for (i = cycle; i < 8; i++)
+			schedule_params->subscheind[i] = schedule_end_idx;
+
+		list_for_each_entry(e, &gating_cfg->entries, list) {
+			schedule[k].delta = ns_to_sja1105_delta(e->interval);
+			schedule[k].destports = e->rule->vl.destports;
+			schedule[k].setvalid = true;
+			schedule[k].txen = true;
+			schedule[k].vlindex = e->rule->vl.sharindx;
+			schedule[k].winstindex = e->rule->vl.sharindx;
+			if (e->gate_state) /* Gate open */
+				schedule[k].winst = true;
+			else /* Gate closed */
+				schedule[k].winend = true;
+			k++;
+		}
+	}
+
 	return 0;
 }
 
@@ -415,6 +457,54 @@ sja1105_tas_check_conflicts(struct sja1105_private *priv, int port,
 	return false;
 }
 
+/* Check the tc-taprio configuration on @port for conflicts with the tc-gate
+ * global subschedule. If @port is -1, check it against all ports.
+ * To reuse the sja1105_tas_check_conflicts logic without refactoring it,
+ * convert the gating configuration to a dummy tc-taprio offload structure.
+ */
+bool sja1105_gating_check_conflicts(struct sja1105_private *priv, int port,
+				    struct netlink_ext_ack *extack)
+{
+	struct sja1105_gating_config *gating_cfg = &priv->tas_data.gating_cfg;
+	size_t num_entries = gating_cfg->num_entries;
+	struct tc_taprio_qopt_offload *dummy;
+	struct sja1105_gate_entry *e;
+	bool conflict;
+	int i = 0;
+
+	if (list_empty(&gating_cfg->entries))
+		return false;
+
+	dummy = kzalloc(sizeof(struct tc_taprio_sched_entry) * num_entries +
+			sizeof(struct tc_taprio_qopt_offload), GFP_KERNEL);
+	if (!dummy) {
+		NL_SET_ERR_MSG_MOD(extack, "Failed to allocate memory");
+		return true;
+	}
+
+	dummy->num_entries = num_entries;
+	dummy->base_time = gating_cfg->base_time;
+	dummy->cycle_time = gating_cfg->cycle_time;
+
+	list_for_each_entry(e, &gating_cfg->entries, list)
+		dummy->entries[i++].interval = e->interval;
+
+	if (port != -1) {
+		conflict = sja1105_tas_check_conflicts(priv, port, dummy);
+	} else {
+		for (port = 0; port < SJA1105_NUM_PORTS; port++) {
+			conflict = sja1105_tas_check_conflicts(priv, port,
+							       dummy);
+			if (conflict)
+				break;
+		}
+	}
+
+	kfree(dummy);
+
+	return conflict;
+}
+
 int sja1105_setup_tc_taprio(struct dsa_switch *ds, int port,
 			    struct tc_taprio_qopt_offload *admin)
 {
@@ -473,6 +563,11 @@ int sja1105_setup_tc_taprio(struct dsa_switch *ds, int port,
 			return -ERANGE;
 	}
 
+	if (sja1105_gating_check_conflicts(priv, port, NULL)) {
+		dev_err(ds->dev, "Conflict with tc-gate schedule\n");
+		return -ERANGE;
+	}
+
 	tas_data->offload[port] = taprio_offload_get(admin);
 
 	rc = sja1105_init_scheduling(priv);
@@ -779,6 +874,8 @@ void sja1105_tas_setup(struct dsa_switch *ds)
 	INIT_WORK(&tas_data->tas_work, sja1105_tas_state_machine);
 	tas_data->state = SJA1105_TAS_STATE_DISABLED;
 	tas_data->last_op = SJA1105_PTP_NONE;
+
+	INIT_LIST_HEAD(&tas_data->gating_cfg.entries);
 }
 
 void sja1105_tas_teardown(struct dsa_switch *ds)