net: ethernet: ti: am65-cpsw: add mqprio qdisc offload in channel mode
authorGrygorii Strashko <grygorii.strashko@ti.com>
Tue, 19 Dec 2023 10:58:03 +0000 (12:58 +0200)
committerDavid S. Miller <davem@davemloft.net>
Sat, 23 Dec 2023 01:01:19 +0000 (01:01 +0000)
This patch adds MQPRIO Qdisc offload in full 'channel' mode which allows
not only setting up pri:tc mapping, but also configuring TX shapers
(rate-limiting) on external port FIFOs.

The MQPRIO Qdisc offload is expected to work with or without VLAN/priority
tagged packets.

The CPSW external Port FIFO has 8 Priority queues. The rate-limit can be
set for each of these priority queues. Which Priority queue a packet is
assigned to depends on PN_REG_TX_PRI_MAP register which maps header
priority to switch priority.

The header priority of a packet is assigned via the RX_PRI_MAP_REG which
maps packet priority to header priority.

The packet priority is either the VLAN priority (for VLAN tagged packets)
or the thread/channel offset.

For simplicity, we assign the same priority queue to all queues of a
Traffic Class so it can be rate-limited correctly.

Configuration example:
 ethtool -L eth1 tx 5
 ethtool --set-priv-flags eth1 p0-rx-ptype-rrobin off

 tc qdisc add dev eth1 parent root handle 100: mqprio num_tc 3 \
 map 0 0 1 2 0 0 0 0 0 0 0 0 0 0 0 0 \
 queues 1@0 1@1 1@2 hw 1 mode channel \
 shaper bw_rlimit min_rate 0 100mbit 200mbit max_rate 0 101mbit 202mbit

 tc qdisc replace dev eth2 handle 100: parent root mqprio num_tc 1 \
 map 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 queues 1@0 hw 1

 ip link add link eth1 name eth1.100 type vlan id 100
 ip link set eth1.100 type vlan egress 0:0 1:1 2:2 3:3 4:4 5:5 6:6 7:7

In the above example two ports share the same TX CPPI queue 0 for low
priority traffic. 3 traffic classes are defined for eth1 and mapped to:
TC0 - low priority, TX CPPI queue 0 -> ext Port 1 fifo0, no rate limit
TC1 - prio 2, TX CPPI queue 1 -> ext Port 1 fifo1, CIR=100Mbit/s, EIR=1Mbit/s
TC2 - prio 3, TX CPPI queue 2 -> ext Port 1 fifo2, CIR=200Mbit/s, EIR=2Mbit/s

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/ti/Kconfig
drivers/net/ethernet/ti/am65-cpsw-nuss.c
drivers/net/ethernet/ti/am65-cpsw-qos.c
drivers/net/ethernet/ti/am65-cpsw-qos.h

index 49cd96c4f53227543d1a4cc5947252a7f5d47b3d..f76532443298dc01af6f38fb79af4fbdbdb6dd45 100644 (file)
@@ -139,7 +139,8 @@ config TI_AM65_CPSW_QOS
        depends on TI_K3_AM65_CPSW_NUSS && NET_SCH_TAPRIO && TI_K3_AM65_CPTS
        help
          This option enables QoS offload features in AM65 CPSW like
-         Time Aware Shaper (TAS) / Enhanced Scheduled Traffic (EST).
+         Time Aware Shaper (TAS) / Enhanced Scheduled Traffic (EST)
+         and MQPRIO qdisc offload.
          The EST scheduler runs on CPTS and the TAS/EST schedule is
          updated in the Fetch RAM memory of the CPSW.
 
index 7651f90f51f2478521549db9aac8f7c6c456abe5..2e374745f2390d6699309a3bf83350f107f66a62 100644 (file)
@@ -2127,6 +2127,9 @@ static int am65_cpsw_nuss_init_slave_ports(struct am65_cpsw_common *common)
                                dev_err(dev, "Use random MAC address\n");
                        }
                }
+
+               /* Reset all Queue priorities to 0 */
+               writel(0, port->port_base + AM65_CPSW_PN_REG_TX_PRI_MAP);
        }
        of_node_put(node);
 
index 9f0a05e763d1d45921cc2ae524fd08fbc9fbb4d3..7790d739e74e332265fa92afee3f31c94bc9a7d1 100644 (file)
@@ -7,7 +7,9 @@
  */
 
 #include <linux/pm_runtime.h>
+#include <linux/math.h>
 #include <linux/time.h>
+#include <linux/units.h>
 #include <net/pkt_cls.h>
 
 #include "am65-cpsw-nuss.h"
@@ -15,6 +17,8 @@
 #include "am65-cpts.h"
 #include "cpsw_ale.h"
 
+#define TO_MBPS(x)     DIV_ROUND_UP((x), BYTES_PER_MBIT)
+
 enum timer_act {
        TACT_PROG,              /* need program timer */
        TACT_NEED_STOP,         /* need stop first */
@@ -31,6 +35,232 @@ am65_cpsw_qos_tx_rate_calc(u32 rate_mbps, unsigned long bus_freq)
        return ir;
 }
 
+static void am65_cpsw_tx_pn_shaper_reset(struct am65_cpsw_port *port)
+{
+       int prio;
+
+       for (prio = 0; prio < AM65_CPSW_PN_FIFO_PRIO_NUM; prio++) {
+               writel(0, port->port_base + AM65_CPSW_PN_REG_PRI_CIR(prio));
+               writel(0, port->port_base + AM65_CPSW_PN_REG_PRI_EIR(prio));
+       }
+}
+
+static void am65_cpsw_tx_pn_shaper_apply(struct am65_cpsw_port *port)
+{
+       struct am65_cpsw_mqprio *p_mqprio = &port->qos.mqprio;
+       struct am65_cpsw_common *common = port->common;
+       struct tc_mqprio_qopt_offload *mqprio;
+       bool enable, shaper_susp = false;
+       u32 rate_mbps;
+       int tc, prio;
+
+       mqprio = &p_mqprio->mqprio_hw;
+       /* takes care of no link case as well */
+       if (p_mqprio->max_rate_total > port->qos.link_speed)
+               shaper_susp = true;
+
+       am65_cpsw_tx_pn_shaper_reset(port);
+
+       enable = p_mqprio->shaper_en && !shaper_susp;
+       if (!enable)
+               return;
+
+       /* Rate limit is specified per Traffic Class but
+        * for CPSW, rate limit can be applied per priority
+        * at port FIFO.
+        *
+        * We have assigned the same priority (TCn) to all queues
+        * of a Traffic Class so they share the same shaper
+        * bandwidth.
+        */
+       for (tc = 0; tc < mqprio->qopt.num_tc; tc++) {
+               prio = tc;
+
+               rate_mbps = TO_MBPS(mqprio->min_rate[tc]);
+               rate_mbps = am65_cpsw_qos_tx_rate_calc(rate_mbps,
+                                                      common->bus_freq);
+               writel(rate_mbps,
+                      port->port_base + AM65_CPSW_PN_REG_PRI_CIR(prio));
+
+               rate_mbps = 0;
+
+               if (mqprio->max_rate[tc]) {
+                       rate_mbps = mqprio->max_rate[tc] - mqprio->min_rate[tc];
+                       rate_mbps = TO_MBPS(rate_mbps);
+                       rate_mbps = am65_cpsw_qos_tx_rate_calc(rate_mbps,
+                                                              common->bus_freq);
+               }
+
+               writel(rate_mbps,
+                      port->port_base + AM65_CPSW_PN_REG_PRI_EIR(prio));
+       }
+}
+
+static int am65_cpsw_mqprio_verify_shaper(struct am65_cpsw_port *port,
+                                         struct tc_mqprio_qopt_offload *mqprio)
+{
+       struct am65_cpsw_mqprio *p_mqprio = &port->qos.mqprio;
+       struct netlink_ext_ack *extack = mqprio->extack;
+       u64 min_rate_total = 0, max_rate_total = 0;
+       u32 min_rate_msk = 0, max_rate_msk = 0;
+       bool has_min_rate, has_max_rate;
+       int num_tc, i;
+
+       if (!(mqprio->flags & TC_MQPRIO_F_SHAPER))
+               return 0;
+
+       if (mqprio->shaper != TC_MQPRIO_SHAPER_BW_RATE)
+               return 0;
+
+       has_min_rate = !!(mqprio->flags & TC_MQPRIO_F_MIN_RATE);
+       has_max_rate = !!(mqprio->flags & TC_MQPRIO_F_MAX_RATE);
+
+       if (!has_min_rate && has_max_rate) {
+               NL_SET_ERR_MSG_MOD(extack, "min_rate is required with max_rate");
+               return -EOPNOTSUPP;
+       }
+
+       if (!has_min_rate)
+               return 0;
+
+       num_tc = mqprio->qopt.num_tc;
+
+       for (i = num_tc - 1; i >= 0; i--) {
+               u32 ch_msk;
+
+               if (mqprio->min_rate[i])
+                       min_rate_msk |= BIT(i);
+               min_rate_total +=  mqprio->min_rate[i];
+
+               if (has_max_rate) {
+                       if (mqprio->max_rate[i])
+                               max_rate_msk |= BIT(i);
+                       max_rate_total +=  mqprio->max_rate[i];
+
+                       if (!mqprio->min_rate[i] && mqprio->max_rate[i]) {
+                               NL_SET_ERR_MSG_FMT_MOD(extack,
+                                                      "TX tc%d rate max>0 but min=0",
+                                                      i);
+                               return -EINVAL;
+                       }
+
+                       if (mqprio->max_rate[i] &&
+                           mqprio->max_rate[i] < mqprio->min_rate[i]) {
+                               NL_SET_ERR_MSG_FMT_MOD(extack,
+                                                      "TX tc%d rate min(%llu)>max(%llu)",
+                                                      i, mqprio->min_rate[i],
+                                                      mqprio->max_rate[i]);
+                               return -EINVAL;
+                       }
+               }
+
+               ch_msk = GENMASK(num_tc - 1, i);
+               if ((min_rate_msk & BIT(i)) && (min_rate_msk ^ ch_msk)) {
+                       NL_SET_ERR_MSG_FMT_MOD(extack,
+                                              "Min rate must be set sequentially hi->lo tx_rate_msk%x",
+                                              min_rate_msk);
+                       return -EINVAL;
+               }
+
+               if ((max_rate_msk & BIT(i)) && (max_rate_msk ^ ch_msk)) {
+                       NL_SET_ERR_MSG_FMT_MOD(extack,
+                                              "Max rate must be set sequentially hi->lo tx_rate_msk%x",
+                                              max_rate_msk);
+                       return -EINVAL;
+               }
+       }
+
+       min_rate_total = TO_MBPS(min_rate_total);
+       max_rate_total = TO_MBPS(max_rate_total);
+
+       p_mqprio->shaper_en = true;
+       p_mqprio->max_rate_total = max_t(u64, min_rate_total, max_rate_total);
+
+       return 0;
+}
+
+static void am65_cpsw_reset_tc_mqprio(struct net_device *ndev)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       struct am65_cpsw_mqprio *p_mqprio = &port->qos.mqprio;
+
+       p_mqprio->shaper_en = false;
+       p_mqprio->max_rate_total = 0;
+
+       am65_cpsw_tx_pn_shaper_reset(port);
+       netdev_reset_tc(ndev);
+
+       /* Reset all Queue priorities to 0 */
+       writel(0, port->port_base + AM65_CPSW_PN_REG_TX_PRI_MAP);
+}
+
+static int am65_cpsw_setup_mqprio(struct net_device *ndev, void *type_data)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       struct am65_cpsw_mqprio *p_mqprio = &port->qos.mqprio;
+       struct tc_mqprio_qopt_offload *mqprio = type_data;
+       struct am65_cpsw_common *common = port->common;
+       struct tc_mqprio_qopt *qopt = &mqprio->qopt;
+       int i, tc, offset, count, prio, ret;
+       u8 num_tc = qopt->num_tc;
+       u32 tx_prio_map = 0;
+
+       memcpy(&p_mqprio->mqprio_hw, mqprio, sizeof(*mqprio));
+
+       ret = pm_runtime_get_sync(common->dev);
+       if (ret < 0) {
+               pm_runtime_put_noidle(common->dev);
+               return ret;
+       }
+
+       if (!num_tc) {
+               am65_cpsw_reset_tc_mqprio(ndev);
+               ret = 0;
+               goto exit_put;
+       }
+
+       ret = am65_cpsw_mqprio_verify_shaper(port, mqprio);
+       if (ret)
+               goto exit_put;
+
+       netdev_set_num_tc(ndev, num_tc);
+
+       /* Multiple Linux priorities can map to a Traffic Class
+        * A Traffic Class can have multiple contiguous Queues,
+        * Queues get mapped to Channels (thread_id),
+        *      if not VLAN tagged, thread_id is used as packet_priority
+        *      if VLAN tagged. VLAN priority is used as packet_priority
+        * packet_priority gets mapped to header_priority in p0_rx_pri_map,
+        * header_priority gets mapped to switch_priority in pn_tx_pri_map.
+        * As p0_rx_pri_map is left at defaults (0x76543210), we can
+        * assume that Queue_n gets mapped to header_priority_n. We can then
+        * set the switch priority in pn_tx_pri_map.
+        */
+
+       for (tc = 0; tc < num_tc; tc++) {
+               prio = tc;
+
+               /* For simplicity we assign the same priority (TCn) to
+                * all queues of a Traffic Class.
+                */
+               for (i = qopt->offset[tc]; i < qopt->offset[tc] + qopt->count[tc]; i++)
+                       tx_prio_map |= prio << (4 * i);
+
+               count = qopt->count[tc];
+               offset = qopt->offset[tc];
+               netdev_set_tc_queue(ndev, tc, count, offset);
+       }
+
+       writel(tx_prio_map, port->port_base + AM65_CPSW_PN_REG_TX_PRI_MAP);
+
+       am65_cpsw_tx_pn_shaper_apply(port);
+
+exit_put:
+       pm_runtime_put(common->dev);
+
+       return ret;
+}
+
 static int am65_cpsw_port_est_enabled(struct am65_cpsw_port *port)
 {
        return port->qos.est_oper || port->qos.est_admin;
@@ -414,6 +644,8 @@ static void am65_cpsw_taprio_destroy(struct net_device *ndev)
 
        port->qos.est_oper = NULL;
        port->qos.est_admin = NULL;
+
+       am65_cpsw_reset_tc_mqprio(ndev);
 }
 
 static void am65_cpsw_cp_taprio(struct tc_taprio_qopt_offload *from,
@@ -462,6 +694,10 @@ static int am65_cpsw_taprio_replace(struct net_device *ndev,
        if (!est_new)
                return -ENOMEM;
 
+       ret = am65_cpsw_setup_mqprio(ndev, &taprio->mqprio);
+       if (ret)
+               return ret;
+
        am65_cpsw_cp_taprio(taprio, &est_new->taprio);
 
        am65_cpsw_est_update_state(ndev);
@@ -505,6 +741,7 @@ static int am65_cpsw_taprio_replace(struct net_device *ndev,
        return 0;
 
 fail:
+       am65_cpsw_reset_tc_mqprio(ndev);
        devm_kfree(&ndev->dev, est_new);
        return ret;
 }
@@ -515,7 +752,6 @@ static void am65_cpsw_est_link_up(struct net_device *ndev, int link_speed)
        ktime_t cur_time;
        s64 delta;
 
-       port->qos.link_speed = link_speed;
        if (!am65_cpsw_port_est_enabled(port))
                return;
 
@@ -559,6 +795,14 @@ static int am65_cpsw_tc_query_caps(struct net_device *ndev, void *type_data)
        struct tc_query_caps_base *base = type_data;
 
        switch (base->type) {
+       case TC_SETUP_QDISC_MQPRIO: {
+               struct tc_mqprio_caps *caps = base->caps;
+
+               caps->validate_queue_counts = true;
+
+               return 0;
+       }
+
        case TC_SETUP_QDISC_TAPRIO: {
                struct tc_taprio_caps *caps = base->caps;
 
@@ -857,6 +1101,8 @@ int am65_cpsw_qos_ndo_setup_tc(struct net_device *ndev, enum tc_setup_type type,
                return am65_cpsw_tc_query_caps(ndev, type_data);
        case TC_SETUP_QDISC_TAPRIO:
                return am65_cpsw_setup_taprio(ndev, type_data);
+       case TC_SETUP_QDISC_MQPRIO:
+               return am65_cpsw_setup_mqprio(ndev, type_data);
        case TC_SETUP_BLOCK:
                return am65_cpsw_qos_setup_tc_block(ndev, type_data);
        default:
@@ -868,6 +1114,9 @@ void am65_cpsw_qos_link_up(struct net_device *ndev, int link_speed)
 {
        struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
 
+       port->qos.link_speed = link_speed;
+       am65_cpsw_tx_pn_shaper_apply(port);
+
        am65_cpsw_est_link_up(ndev, link_speed);
        port->qos.link_down_time = 0;
 }
@@ -876,8 +1125,9 @@ void am65_cpsw_qos_link_down(struct net_device *ndev)
 {
        struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
 
+       port->qos.link_speed = SPEED_UNKNOWN;
+       am65_cpsw_tx_pn_shaper_apply(port);
+
        if (!port->qos.link_down_time)
                port->qos.link_down_time = ktime_get();
-
-       port->qos.link_speed = SPEED_UNKNOWN;
 }
index e3259f27baadbfc225cb4f1792ceed35b18b6712..b54c34c0aa7f59ce1baa0fae4738efc849b1d72a 100644 (file)
@@ -9,6 +9,7 @@
 #include <net/pkt_sched.h>
 
 struct am65_cpsw_common;
+struct am65_cpsw_port;
 
 struct am65_cpsw_est {
        int buf;
@@ -16,6 +17,12 @@ struct am65_cpsw_est {
        struct tc_taprio_qopt_offload taprio;
 };
 
+struct am65_cpsw_mqprio {
+       struct tc_mqprio_qopt_offload mqprio_hw;
+       u64 max_rate_total;
+       bool shaper_en;
+};
+
 struct am65_cpsw_ale_ratelimit {
        unsigned long cookie;
        u64 rate_packet_ps;
@@ -26,6 +33,7 @@ struct am65_cpsw_qos {
        struct am65_cpsw_est *est_oper;
        ktime_t link_down_time;
        int link_speed;
+       struct am65_cpsw_mqprio mqprio;
 
        struct am65_cpsw_ale_ratelimit ale_bc_ratelimit;
        struct am65_cpsw_ale_ratelimit ale_mc_ratelimit;
@@ -36,6 +44,15 @@ struct am65_cpsw_qos {
 #define AM65_CPSW_PN_REG_FIFO_STATUS           0x050
 #define AM65_CPSW_PN_REG_EST_CTL               0x060
 #define AM65_CPSW_PN_REG_PRI_CIR(pri)          (0x140 + 4 * (pri))
+#define AM65_CPSW_P0_REG_PRI_EIR(pri)          (0x160 + 4 * (pri))
+
+#define AM65_CPSW_PN_REG_CTL                   0x004
+#define AM65_CPSW_PN_REG_TX_PRI_MAP            0x018
+#define AM65_CPSW_PN_REG_RX_PRI_MAP            0x020
+#define AM65_CPSW_PN_REG_FIFO_STATUS           0x050
+#define AM65_CPSW_PN_REG_EST_CTL               0x060
+#define AM65_CPSW_PN_REG_PRI_CIR(pri)          (0x140 + 4 * (pri))
+#define AM65_CPSW_PN_REG_PRI_EIR(pri)          (0x160 + 4 * (pri))
 
 /* AM65_CPSW_REG_CTL register fields */
 #define AM65_CPSW_CTL_EST_EN                   BIT(18)
@@ -66,6 +83,9 @@ struct am65_cpsw_qos {
 #define AM65_CPSW_FETCH_ALLOW_MSK              GENMASK(7, 0)
 #define AM65_CPSW_FETCH_ALLOW_MAX              AM65_CPSW_FETCH_ALLOW_MSK
 
+/* number of priority queues per port FIFO */
+#define AM65_CPSW_PN_FIFO_PRIO_NUM             8
+
 #if IS_ENABLED(CONFIG_TI_AM65_CPSW_QOS)
 int am65_cpsw_qos_ndo_setup_tc(struct net_device *ndev, enum tc_setup_type type,
                               void *type_data);