]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/commitdiff
ethernet: ti: am65-cpsw-qos: add TAPRIO offload support
authorIvan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Wed, 13 May 2020 13:26:15 +0000 (09:26 -0400)
committerDavid S. Miller <davem@davemloft.net>
Fri, 15 May 2020 00:33:30 +0000 (17:33 -0700)
AM65 CPSW h/w supports Enhanced Scheduled Traffic (EST – defined
in P802.1Qbv/D2.2 that later got included in IEEE 802.1Q-2018)
configuration. EST allows express queue traffic to be scheduled
(placed) on the wire at specific repeatable time intervals. In
Linux kernel, EST configuration is done through tc command and
the taprio scheduler in the net core implements a software only
scheduler (SCH_TAPRIO). If the NIC is capable of EST configuration,
user indicate "flag 2" in the command which is then parsed by
taprio scheduler in net core and indicate that the command is to
be offloaded to h/w. taprio then offloads the command to the
driver by calling ndo_setup_tc() ndo ops. This patch implements
ndo_setup_tc() to offload EST configuration to CPSW h/w.

Currently driver supports only SetGateStates operation. EST
operates on a repeating time interval generated by the CPTS EST
function generator. Each Ethernet port has a global EST fetch
RAM that can be configured as 2 buffers, each of 64 locations
or one large buffer of 128 locations. In 2 buffer configuration,
a ping pong mechanism is used to hold the active schedule (oper)
in one buffer and new (admin) command in the other. Each 22-bit
fetch command consists of a 14-bit fetch count (14 MSB’s) and an
8-bit priority fetch allow (8 LSB’s) that will be applied for the
fetch count time in wireside clocks. Driver process each of the
sched-entry in the offload command and update the fetch RAM.
Driver configures duration in sched-entry into the fetch count
and Gate mask into the priority fetch bits of the RAM. Then
configures the CPTS EST function generator to activate the
schedule. Currently driver supports only 2 buffer configuration
which means driver supports a max cycle time of ~8 msec.

CPSW supports a configurable number of priority queues (up to 8)
and needs to be switched to this mode from the default round
robin mode before EST can be offloaded. User configures
these through ethtool commands (-L for changing number of
queues and --set-priv-flags to disable round robin mode).
Driver doesn't enable EST if pf_p0_rx_ptype_rrobin privat flag
is set. The flag is common for all ports, and so can't be just
overridden by taprio configuration w/o user involvement.
Command fails if pf_p0_rx_ptype_rrobin is already set in the
driver.

Scheds (commands) configuration depends on interface speed so
driver translates the duration to the fetch count based on
link speed. Each schedule can be constructed with several
command entries in fetch RAM  depending on interval. For example
if each sched has timer interval < ~130us on 1000 Mb link then
each sched consumes one command and have 1:1 mapping. When
Ethernet link goes down, driver purge the configuration if link
is down for more than 1 second.

The patch allows to update the timer and scheds memory only if it's
really needed, and skip cases required the user to stop timer by
configuring only shceds memory.

Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/ti/Kconfig
drivers/net/ethernet/ti/Makefile
drivers/net/ethernet/ti/am65-cpsw-ethtool.c
drivers/net/ethernet/ti/am65-cpsw-nuss.c
drivers/net/ethernet/ti/am65-cpsw-nuss.h
drivers/net/ethernet/ti/am65-cpsw-qos.c [new file with mode: 0644]
drivers/net/ethernet/ti/am65-cpsw-qos.h [new file with mode: 0644]

index 988e907e33223b3f6deb226f3c9cda9c441b7ff7..4d4852f00ff7d7878378bac314209f89b8366a3c 100644 (file)
@@ -123,6 +123,15 @@ config TI_K3_AM65_CPTS
          protocol, Ethernet Enhanced Scheduled Traffic Operations (CPTS_ESTFn)
          and PCIe Subsystem Precision Time Measurement (PTM).
 
+config TI_AM65_CPSW_TAS
+       bool "Enable TAS offload in AM65 CPSW"
+       depends on TI_K3_AM65_CPSW_NUSS && NET_SCH_TAPRIO && TI_K3_AM65_CPTS
+       help
+         Say y here to support Time Aware Shaper(TAS) offload in AM65 CPSW.
+         AM65 CPSW hardware supports Enhanced Scheduled Traffic (EST)
+         defined in IEEE 802.1Q 2018. The EST scheduler runs on CPTS and the
+         TAS/EST schedule is updated in the Fetch RAM memory of the CPSW.
+
 config TI_KEYSTONE_NETCP
        tristate "TI Keystone NETCP Core Support"
        select TI_DAVINCI_MDIO
index bf86067f9b4c87fe56aadb75bf1a9c78ae6b15c6..be95512d80b5cd1af364d7e317d93062ffd2018c 100644 (file)
@@ -25,5 +25,5 @@ obj-$(CONFIG_TI_KEYSTONE_NETCP_ETHSS) += keystone_netcp_ethss.o
 keystone_netcp_ethss-y := netcp_ethss.o netcp_sgmii.o netcp_xgbepcsr.o cpsw_ale.o
 
 obj-$(CONFIG_TI_K3_AM65_CPSW_NUSS) += ti-am65-cpsw-nuss.o
-ti-am65-cpsw-nuss-y := am65-cpsw-nuss.o cpsw_sl.o am65-cpsw-ethtool.o cpsw_ale.o k3-cppi-desc-pool.o
+ti-am65-cpsw-nuss-y := am65-cpsw-nuss.o cpsw_sl.o am65-cpsw-ethtool.o cpsw_ale.o k3-cppi-desc-pool.o am65-cpsw-qos.o
 obj-$(CONFIG_TI_K3_AM65_CPTS) += am65-cpts.o
index 23661a6ed426022ab2d9c388a347838166d6e0b1..8c4690f3ebcbdd2f9180dcdea5cb36bdd443a946 100644 (file)
@@ -730,9 +730,17 @@ static u32 am65_cpsw_get_ethtool_priv_flags(struct net_device *ndev)
 static int am65_cpsw_set_ethtool_priv_flags(struct net_device *ndev, u32 flags)
 {
        struct am65_cpsw_common *common = am65_ndev_to_common(ndev);
+       int rrobin;
 
-       common->pf_p0_rx_ptype_rrobin =
-                       !!(flags & AM65_CPSW_PRIV_P0_RX_PTYPE_RROBIN);
+       rrobin = !!(flags & AM65_CPSW_PRIV_P0_RX_PTYPE_RROBIN);
+
+       if (common->est_enabled && rrobin) {
+               netdev_err(ndev,
+                          "p0-rx-ptype-rrobin flag conflicts with QOS\n");
+               return -EINVAL;
+       }
+
+       common->pf_p0_rx_ptype_rrobin = rrobin;
        am65_cpsw_nuss_set_p0_ptype(common);
 
        return 0;
index 8cdbb2b9b13a59011edd50f63a055686bed8da79..4a8229864ae46c34957933d6471e2efaf48a4fde 100644 (file)
 #define AM65_CPSW_XGMII_BASE   0x2100
 #define AM65_CPSW_CPSW_NU_BASE 0x20000
 #define AM65_CPSW_NU_PORTS_BASE        0x1000
+#define AM65_CPSW_NU_FRAM_BASE 0x12000
 #define AM65_CPSW_NU_STATS_BASE        0x1a000
 #define AM65_CPSW_NU_ALE_BASE  0x1e000
 #define AM65_CPSW_NU_CPTS_BASE 0x1d000
 
 #define AM65_CPSW_NU_PORTS_OFFSET      0x1000
 #define AM65_CPSW_NU_STATS_PORT_OFFSET 0x200
+#define AM65_CPSW_NU_FRAM_PORT_OFFSET  0x200
 
 #define AM65_CPSW_MAX_PORTS    8
 
@@ -188,9 +190,11 @@ void am65_cpsw_nuss_adjust_link(struct net_device *ndev)
                cpsw_ale_control_set(common->ale, port->port_id,
                                     ALE_PORT_STATE, ALE_PORT_STATE_FORWARD);
 
+               am65_cpsw_qos_link_up(ndev, phy->speed);
                netif_tx_wake_all_queues(ndev);
        } else {
                int tmo;
+
                /* disable forwarding */
                cpsw_ale_control_set(common->ale, port->port_id,
                                     ALE_PORT_STATE, ALE_PORT_STATE_DISABLE);
@@ -204,6 +208,7 @@ void am65_cpsw_nuss_adjust_link(struct net_device *ndev)
 
                cpsw_sl_ctl_reset(port->slave.mac_sl);
 
+               am65_cpsw_qos_link_down(ndev);
                netif_tx_stop_all_queues(ndev);
        }
 
@@ -1378,6 +1383,7 @@ static const struct net_device_ops am65_cpsw_nuss_netdev_ops_2g = {
        .ndo_vlan_rx_kill_vid   = am65_cpsw_nuss_ndo_slave_kill_vid,
        .ndo_do_ioctl           = am65_cpsw_nuss_ndo_slave_ioctl,
        .ndo_set_features       = am65_cpsw_nuss_ndo_slave_set_features,
+       .ndo_setup_tc           = am65_cpsw_qos_ndo_setup_tc,
 };
 
 static void am65_cpsw_nuss_slave_disable_unused(struct am65_cpsw_port *port)
@@ -1739,6 +1745,9 @@ static int am65_cpsw_nuss_init_slave_ports(struct am65_cpsw_common *common)
                port->stat_base = common->cpsw_base + AM65_CPSW_NU_STATS_BASE +
                                  (AM65_CPSW_NU_STATS_PORT_OFFSET * port_id);
                port->name = of_get_property(port_np, "label", NULL);
+               port->fetch_ram_base =
+                               common->cpsw_base + AM65_CPSW_NU_FRAM_BASE +
+                               (AM65_CPSW_NU_FRAM_PORT_OFFSET * (port_id - 1));
 
                port->disabled = !of_device_is_available(port_np);
                if (port->disabled)
index 8a6382188cb5c2ff408b17b91f36d7242621fc22..9faf4fb1409b9ea79ad4caa6acec57523ad15b5d 100644 (file)
@@ -9,7 +9,9 @@
 #include <linux/kernel.h>
 #include <linux/module.h>
 #include <linux/netdevice.h>
+#include <linux/phy.h>
 #include <linux/platform_device.h>
+#include "am65-cpsw-qos.h"
 
 struct am65_cpts;
 
@@ -38,10 +40,12 @@ struct am65_cpsw_port {
        u32                             port_id;
        void __iomem                    *port_base;
        void __iomem                    *stat_base;
+       void __iomem                    *fetch_ram_base;
        bool                            disabled;
        struct am65_cpsw_slave_data     slave;
        bool                            tx_ts_enabled;
        bool                            rx_ts_enabled;
+       struct am65_cpsw_qos            qos;
 };
 
 struct am65_cpsw_host {
@@ -104,6 +108,7 @@ struct am65_cpsw_common {
        u32                     cpsw_ver;
        bool                    pf_p0_rx_ptype_rrobin;
        struct am65_cpts        *cpts;
+       int                     est_enabled;
 };
 
 struct am65_cpsw_ndev_stats {
diff --git a/drivers/net/ethernet/ti/am65-cpsw-qos.c b/drivers/net/ethernet/ti/am65-cpsw-qos.c
new file mode 100644 (file)
index 0000000..32eac04
--- /dev/null
@@ -0,0 +1,626 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Texas Instruments K3 AM65 Ethernet QoS submodule
+ * Copyright (C) 2020 Texas Instruments Incorporated - http://www.ti.com/
+ *
+ * quality of service module includes:
+ * Enhanced Scheduler Traffic (EST - P802.1Qbv/D2.2)
+ */
+
+#include <linux/pm_runtime.h>
+#include <linux/time.h>
+
+#include "am65-cpsw-nuss.h"
+#include "am65-cpsw-qos.h"
+#include "am65-cpts.h"
+
+#define AM65_CPSW_REG_CTL                      0x004
+#define AM65_CPSW_PN_REG_CTL                   0x004
+#define AM65_CPSW_PN_REG_FIFO_STATUS           0x050
+#define AM65_CPSW_PN_REG_EST_CTL               0x060
+
+/* AM65_CPSW_REG_CTL register fields */
+#define AM65_CPSW_CTL_EST_EN                   BIT(18)
+
+/* AM65_CPSW_PN_REG_CTL register fields */
+#define AM65_CPSW_PN_CTL_EST_PORT_EN           BIT(17)
+
+/* AM65_CPSW_PN_REG_EST_CTL register fields */
+#define AM65_CPSW_PN_EST_ONEBUF                        BIT(0)
+#define AM65_CPSW_PN_EST_BUFSEL                        BIT(1)
+#define AM65_CPSW_PN_EST_TS_EN                 BIT(2)
+#define AM65_CPSW_PN_EST_TS_FIRST              BIT(3)
+#define AM65_CPSW_PN_EST_ONEPRI                        BIT(4)
+#define AM65_CPSW_PN_EST_TS_PRI_MSK            GENMASK(7, 5)
+
+/* AM65_CPSW_PN_REG_FIFO_STATUS register fields */
+#define AM65_CPSW_PN_FST_TX_PRI_ACTIVE_MSK     GENMASK(7, 0)
+#define AM65_CPSW_PN_FST_TX_E_MAC_ALLOW_MSK    GENMASK(15, 8)
+#define AM65_CPSW_PN_FST_EST_CNT_ERR           BIT(16)
+#define AM65_CPSW_PN_FST_EST_ADD_ERR           BIT(17)
+#define AM65_CPSW_PN_FST_EST_BUFACT            BIT(18)
+
+/* EST FETCH COMMAND RAM */
+#define AM65_CPSW_FETCH_RAM_CMD_NUM            0x80
+#define AM65_CPSW_FETCH_CNT_MSK                        GENMASK(21, 8)
+#define AM65_CPSW_FETCH_CNT_MAX                        (AM65_CPSW_FETCH_CNT_MSK >> 8)
+#define AM65_CPSW_FETCH_CNT_OFFSET             8
+#define AM65_CPSW_FETCH_ALLOW_MSK              GENMASK(7, 0)
+#define AM65_CPSW_FETCH_ALLOW_MAX              AM65_CPSW_FETCH_ALLOW_MSK
+
+enum timer_act {
+       TACT_PROG,              /* need program timer */
+       TACT_NEED_STOP,         /* need stop first */
+       TACT_SKIP_PROG,         /* just buffer can be updated */
+};
+
+static int am65_cpsw_port_est_enabled(struct am65_cpsw_port *port)
+{
+       return port->qos.est_oper || port->qos.est_admin;
+}
+
+static void am65_cpsw_est_enable(struct am65_cpsw_common *common, int enable)
+{
+       u32 val;
+
+       val = readl(common->cpsw_base + AM65_CPSW_REG_CTL);
+
+       if (enable)
+               val |= AM65_CPSW_CTL_EST_EN;
+       else
+               val &= ~AM65_CPSW_CTL_EST_EN;
+
+       writel(val, common->cpsw_base + AM65_CPSW_REG_CTL);
+       common->est_enabled = enable;
+}
+
+static void am65_cpsw_port_est_enable(struct am65_cpsw_port *port, int enable)
+{
+       u32 val;
+
+       val = readl(port->port_base + AM65_CPSW_PN_REG_CTL);
+       if (enable)
+               val |= AM65_CPSW_PN_CTL_EST_PORT_EN;
+       else
+               val &= ~AM65_CPSW_PN_CTL_EST_PORT_EN;
+
+       writel(val, port->port_base + AM65_CPSW_PN_REG_CTL);
+}
+
+/* target new EST RAM buffer, actual toggle happens after cycle completion */
+static void am65_cpsw_port_est_assign_buf_num(struct net_device *ndev,
+                                             int buf_num)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       u32 val;
+
+       val = readl(port->port_base + AM65_CPSW_PN_REG_EST_CTL);
+       if (buf_num)
+               val |= AM65_CPSW_PN_EST_BUFSEL;
+       else
+               val &= ~AM65_CPSW_PN_EST_BUFSEL;
+
+       writel(val, port->port_base + AM65_CPSW_PN_REG_EST_CTL);
+}
+
+/* am65_cpsw_port_est_is_swapped() - Indicate if h/w is transitioned
+ * admin -> oper or not
+ *
+ * Return true if already transitioned. i.e oper is equal to admin and buf
+ * numbers match (est_oper->buf match with est_admin->buf).
+ * false if before transition. i.e oper is not equal to admin, (i.e a
+ * previous admin command is waiting to be transitioned to oper state
+ * and est_oper->buf not match with est_oper->buf).
+ */
+static int am65_cpsw_port_est_is_swapped(struct net_device *ndev, int *oper,
+                                        int *admin)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       u32 val;
+
+       val = readl(port->port_base + AM65_CPSW_PN_REG_FIFO_STATUS);
+       *oper = !!(val & AM65_CPSW_PN_FST_EST_BUFACT);
+
+       val = readl(port->port_base + AM65_CPSW_PN_REG_EST_CTL);
+       *admin = !!(val & AM65_CPSW_PN_EST_BUFSEL);
+
+       return *admin == *oper;
+}
+
+/* am65_cpsw_port_est_get_free_buf_num() - Get free buffer number for
+ * Admin to program the new schedule.
+ *
+ * Logic as follows:-
+ * If oper is same as admin, return the other buffer (!oper) as the admin
+ * buffer.  If oper is not the same, driver let the current oper to continue
+ * as it is in the process of transitioning from admin -> oper. So keep the
+ * oper by selecting the same oper buffer by writing to EST_BUFSEL bit in
+ * EST CTL register. In the second iteration they will match and code returns.
+ * The actual buffer to write command is selected later before it is ready
+ * to update the schedule.
+ */
+static int am65_cpsw_port_est_get_free_buf_num(struct net_device *ndev)
+{
+       int oper, admin;
+       int roll = 2;
+
+       while (roll--) {
+               if (am65_cpsw_port_est_is_swapped(ndev, &oper, &admin))
+                       return !oper;
+
+               /* admin is not set, so hinder transition as it's not allowed
+                * to touch memory in-flight, by targeting same oper buf.
+                */
+               am65_cpsw_port_est_assign_buf_num(ndev, oper);
+
+               dev_info(&ndev->dev,
+                        "Prev. EST admin cycle is in transit %d -> %d\n",
+                        oper, admin);
+       }
+
+       return admin;
+}
+
+static void am65_cpsw_admin_to_oper(struct net_device *ndev)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+
+       if (port->qos.est_oper)
+               devm_kfree(&ndev->dev, port->qos.est_oper);
+
+       port->qos.est_oper = port->qos.est_admin;
+       port->qos.est_admin = NULL;
+}
+
+static void am65_cpsw_port_est_get_buf_num(struct net_device *ndev,
+                                          struct am65_cpsw_est *est_new)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       u32 val;
+
+       val = readl(port->port_base + AM65_CPSW_PN_REG_EST_CTL);
+       val &= ~AM65_CPSW_PN_EST_ONEBUF;
+       writel(val, port->port_base + AM65_CPSW_PN_REG_EST_CTL);
+
+       est_new->buf = am65_cpsw_port_est_get_free_buf_num(ndev);
+
+       /* rolled buf num means changed buf while configuring */
+       if (port->qos.est_oper && port->qos.est_admin &&
+           est_new->buf == port->qos.est_oper->buf)
+               am65_cpsw_admin_to_oper(ndev);
+}
+
+static void am65_cpsw_est_set(struct net_device *ndev, int enable)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       struct am65_cpsw_common *common = port->common;
+       int common_enable = 0;
+       int i;
+
+       am65_cpsw_port_est_enable(port, enable);
+
+       for (i = 0; i < common->port_num; i++)
+               common_enable |= am65_cpsw_port_est_enabled(&common->ports[i]);
+
+       common_enable |= enable;
+       am65_cpsw_est_enable(common, common_enable);
+}
+
+/* This update is supposed to be used in any routine before getting real state
+ * of admin -> oper transition, particularly it's supposed to be used in some
+ * generic routine for providing real state to Taprio Qdisc.
+ */
+static void am65_cpsw_est_update_state(struct net_device *ndev)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       int oper, admin;
+
+       if (!port->qos.est_admin)
+               return;
+
+       if (!am65_cpsw_port_est_is_swapped(ndev, &oper, &admin))
+               return;
+
+       am65_cpsw_admin_to_oper(ndev);
+}
+
+/* Fetch command count it's number of bytes in Gigabit mode or nibbles in
+ * 10/100Mb mode. So, having speed and time in ns, recalculate ns to number of
+ * bytes/nibbles that can be sent while transmission on given speed.
+ */
+static int am65_est_cmd_ns_to_cnt(u64 ns, int link_speed)
+{
+       u64 temp;
+
+       temp = ns * link_speed;
+       if (link_speed < SPEED_1000)
+               temp <<= 1;
+
+       return DIV_ROUND_UP(temp, 8 * 1000);
+}
+
+static void __iomem *am65_cpsw_est_set_sched_cmds(void __iomem *addr,
+                                                 int fetch_cnt,
+                                                 int fetch_allow)
+{
+       u32 prio_mask, cmd_fetch_cnt, cmd;
+
+       do {
+               if (fetch_cnt > AM65_CPSW_FETCH_CNT_MAX) {
+                       fetch_cnt -= AM65_CPSW_FETCH_CNT_MAX;
+                       cmd_fetch_cnt = AM65_CPSW_FETCH_CNT_MAX;
+               } else {
+                       cmd_fetch_cnt = fetch_cnt;
+                       /* fetch count can't be less than 16? */
+                       if (cmd_fetch_cnt && cmd_fetch_cnt < 16)
+                               cmd_fetch_cnt = 16;
+
+                       fetch_cnt = 0;
+               }
+
+               prio_mask = fetch_allow & AM65_CPSW_FETCH_ALLOW_MSK;
+               cmd = (cmd_fetch_cnt << AM65_CPSW_FETCH_CNT_OFFSET) | prio_mask;
+
+               writel(cmd, addr);
+               addr += 4;
+       } while (fetch_cnt);
+
+       return addr;
+}
+
+static int am65_cpsw_est_calc_cmd_num(struct net_device *ndev,
+                                     struct tc_taprio_qopt_offload *taprio,
+                                     int link_speed)
+{
+       int i, cmd_cnt, cmd_sum = 0;
+       u32 fetch_cnt;
+
+       for (i = 0; i < taprio->num_entries; i++) {
+               if (taprio->entries[i].command != TC_TAPRIO_CMD_SET_GATES) {
+                       dev_err(&ndev->dev, "Only SET command is supported");
+                       return -EINVAL;
+               }
+
+               fetch_cnt = am65_est_cmd_ns_to_cnt(taprio->entries[i].interval,
+                                                  link_speed);
+
+               cmd_cnt = DIV_ROUND_UP(fetch_cnt, AM65_CPSW_FETCH_CNT_MAX);
+               if (!cmd_cnt)
+                       cmd_cnt++;
+
+               cmd_sum += cmd_cnt;
+
+               if (!fetch_cnt)
+                       break;
+       }
+
+       return cmd_sum;
+}
+
+static int am65_cpsw_est_check_scheds(struct net_device *ndev,
+                                     struct am65_cpsw_est *est_new)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       int cmd_num;
+
+       cmd_num = am65_cpsw_est_calc_cmd_num(ndev, &est_new->taprio,
+                                            port->qos.link_speed);
+       if (cmd_num < 0)
+               return cmd_num;
+
+       if (cmd_num > AM65_CPSW_FETCH_RAM_CMD_NUM / 2) {
+               dev_err(&ndev->dev, "No fetch RAM");
+               return -ENOMEM;
+       }
+
+       return 0;
+}
+
+static void am65_cpsw_est_set_sched_list(struct net_device *ndev,
+                                        struct am65_cpsw_est *est_new)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       u32 fetch_cnt, fetch_allow, all_fetch_allow = 0;
+       void __iomem *ram_addr, *max_ram_addr;
+       struct tc_taprio_sched_entry *entry;
+       int i, ram_size;
+
+       ram_addr = port->fetch_ram_base;
+       ram_size = AM65_CPSW_FETCH_RAM_CMD_NUM * 2;
+       ram_addr += est_new->buf * ram_size;
+
+       max_ram_addr = ram_size + ram_addr;
+       for (i = 0; i < est_new->taprio.num_entries; i++) {
+               entry = &est_new->taprio.entries[i];
+
+               fetch_cnt = am65_est_cmd_ns_to_cnt(entry->interval,
+                                                  port->qos.link_speed);
+               fetch_allow = entry->gate_mask;
+               if (fetch_allow > AM65_CPSW_FETCH_ALLOW_MAX)
+                       dev_dbg(&ndev->dev, "fetch_allow > 8 bits: %d\n",
+                               fetch_allow);
+
+               ram_addr = am65_cpsw_est_set_sched_cmds(ram_addr, fetch_cnt,
+                                                       fetch_allow);
+
+               if (!fetch_cnt && i < est_new->taprio.num_entries - 1) {
+                       dev_info(&ndev->dev,
+                                "next scheds after %d have no impact", i + 1);
+                       break;
+               }
+
+               all_fetch_allow |= fetch_allow;
+       }
+
+       /* end cmd, enabling non-timed queues for potential over cycle time */
+       if (ram_addr < max_ram_addr)
+               writel(~all_fetch_allow & AM65_CPSW_FETCH_ALLOW_MSK, ram_addr);
+}
+
+/**
+ * Enable ESTf periodic output, set cycle start time and interval.
+ */
+static int am65_cpsw_timer_set(struct net_device *ndev,
+                              struct am65_cpsw_est *est_new)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       struct am65_cpsw_common *common = port->common;
+       struct am65_cpts *cpts = common->cpts;
+       struct am65_cpts_estf_cfg cfg;
+
+       cfg.ns_period = est_new->taprio.cycle_time;
+       cfg.ns_start = est_new->taprio.base_time;
+
+       return am65_cpts_estf_enable(cpts, port->port_id - 1, &cfg);
+}
+
+static void am65_cpsw_timer_stop(struct net_device *ndev)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       struct am65_cpts *cpts = port->common->cpts;
+
+       am65_cpts_estf_disable(cpts, port->port_id - 1);
+}
+
+static enum timer_act am65_cpsw_timer_act(struct net_device *ndev,
+                                         struct am65_cpsw_est *est_new)
+{
+       struct tc_taprio_qopt_offload *taprio_oper, *taprio_new;
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       struct am65_cpts *cpts = port->common->cpts;
+       u64 cur_time;
+       s64 diff;
+
+       if (!port->qos.est_oper)
+               return TACT_PROG;
+
+       taprio_new = &est_new->taprio;
+       taprio_oper = &port->qos.est_oper->taprio;
+
+       if (taprio_new->cycle_time != taprio_oper->cycle_time)
+               return TACT_NEED_STOP;
+
+       /* in order to avoid timer reset get base_time form oper taprio */
+       if (!taprio_new->base_time && taprio_oper)
+               taprio_new->base_time = taprio_oper->base_time;
+
+       if (taprio_new->base_time == taprio_oper->base_time)
+               return TACT_SKIP_PROG;
+
+       /* base times are cycle synchronized */
+       diff = taprio_new->base_time - taprio_oper->base_time;
+       diff = diff < 0 ? -diff : diff;
+       if (diff % taprio_new->cycle_time)
+               return TACT_NEED_STOP;
+
+       cur_time = am65_cpts_ns_gettime(cpts);
+       if (taprio_new->base_time <= cur_time + taprio_new->cycle_time)
+               return TACT_SKIP_PROG;
+
+       /* TODO: Admin schedule at future time is not currently supported */
+       return TACT_NEED_STOP;
+}
+
+static void am65_cpsw_stop_est(struct net_device *ndev)
+{
+       am65_cpsw_est_set(ndev, 0);
+       am65_cpsw_timer_stop(ndev);
+}
+
+static void am65_cpsw_purge_est(struct net_device *ndev)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+
+       am65_cpsw_stop_est(ndev);
+
+       if (port->qos.est_admin)
+               devm_kfree(&ndev->dev, port->qos.est_admin);
+
+       if (port->qos.est_oper)
+               devm_kfree(&ndev->dev, port->qos.est_oper);
+
+       port->qos.est_oper = NULL;
+       port->qos.est_admin = NULL;
+}
+
+static int am65_cpsw_configure_taprio(struct net_device *ndev,
+                                     struct am65_cpsw_est *est_new)
+{
+       struct am65_cpsw_common *common = am65_ndev_to_common(ndev);
+       struct am65_cpts *cpts = common->cpts;
+       int ret = 0, tact = TACT_PROG;
+
+       am65_cpsw_est_update_state(ndev);
+
+       if (!est_new->taprio.enable) {
+               am65_cpsw_stop_est(ndev);
+               return ret;
+       }
+
+       ret = am65_cpsw_est_check_scheds(ndev, est_new);
+       if (ret < 0)
+               return ret;
+
+       tact = am65_cpsw_timer_act(ndev, est_new);
+       if (tact == TACT_NEED_STOP) {
+               dev_err(&ndev->dev,
+                       "Can't toggle estf timer, stop taprio first");
+               return -EINVAL;
+       }
+
+       if (tact == TACT_PROG)
+               am65_cpsw_timer_stop(ndev);
+
+       if (!est_new->taprio.base_time)
+               est_new->taprio.base_time = am65_cpts_ns_gettime(cpts);
+
+       am65_cpsw_port_est_get_buf_num(ndev, est_new);
+       am65_cpsw_est_set_sched_list(ndev, est_new);
+       am65_cpsw_port_est_assign_buf_num(ndev, est_new->buf);
+
+       am65_cpsw_est_set(ndev, est_new->taprio.enable);
+
+       if (tact == TACT_PROG) {
+               ret = am65_cpsw_timer_set(ndev, est_new);
+               if (ret) {
+                       dev_err(&ndev->dev, "Failed to set cycle time");
+                       return ret;
+               }
+       }
+
+       return ret;
+}
+
+static void am65_cpsw_cp_taprio(struct tc_taprio_qopt_offload *from,
+                               struct tc_taprio_qopt_offload *to)
+{
+       int i;
+
+       *to = *from;
+       for (i = 0; i < from->num_entries; i++)
+               to->entries[i] = from->entries[i];
+}
+
+static int am65_cpsw_set_taprio(struct net_device *ndev, void *type_data)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       struct tc_taprio_qopt_offload *taprio = type_data;
+       struct am65_cpsw_est *est_new;
+       size_t size;
+       int ret = 0;
+
+       if (taprio->cycle_time_extension) {
+               dev_err(&ndev->dev, "Failed to set cycle time extension");
+               return -EOPNOTSUPP;
+       }
+
+       size = sizeof(struct tc_taprio_sched_entry) * taprio->num_entries +
+              sizeof(struct am65_cpsw_est);
+
+       est_new = devm_kzalloc(&ndev->dev, size, GFP_KERNEL);
+       if (!est_new)
+               return -ENOMEM;
+
+       am65_cpsw_cp_taprio(taprio, &est_new->taprio);
+       ret = am65_cpsw_configure_taprio(ndev, est_new);
+       if (!ret) {
+               if (taprio->enable) {
+                       if (port->qos.est_admin)
+                               devm_kfree(&ndev->dev, port->qos.est_admin);
+
+                       port->qos.est_admin = est_new;
+               } else {
+                       devm_kfree(&ndev->dev, est_new);
+                       am65_cpsw_purge_est(ndev);
+               }
+       } else {
+               devm_kfree(&ndev->dev, est_new);
+       }
+
+       return ret;
+}
+
+static void am65_cpsw_est_link_up(struct net_device *ndev, int link_speed)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       ktime_t cur_time;
+       s64 delta;
+
+       port->qos.link_speed = link_speed;
+       if (!am65_cpsw_port_est_enabled(port))
+               return;
+
+       if (port->qos.link_down_time) {
+               cur_time = ktime_get();
+               delta = ktime_us_delta(cur_time, port->qos.link_down_time);
+               if (delta > USEC_PER_SEC) {
+                       dev_err(&ndev->dev,
+                               "Link has been lost too long, stopping TAS");
+                       goto purge_est;
+               }
+       }
+
+       return;
+
+purge_est:
+       am65_cpsw_purge_est(ndev);
+}
+
+static int am65_cpsw_setup_taprio(struct net_device *ndev, void *type_data)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+       struct am65_cpsw_common *common = port->common;
+
+       if (!IS_ENABLED(CONFIG_TI_AM65_CPSW_TAS))
+               return -ENODEV;
+
+       if (!netif_running(ndev)) {
+               dev_err(&ndev->dev, "interface is down, link speed unknown\n");
+               return -ENETDOWN;
+       }
+
+       if (common->pf_p0_rx_ptype_rrobin) {
+               dev_err(&ndev->dev,
+                       "p0-rx-ptype-rrobin flag conflicts with taprio qdisc\n");
+               return -EINVAL;
+       }
+
+       if (port->qos.link_speed == SPEED_UNKNOWN)
+               return -ENOLINK;
+
+       return am65_cpsw_set_taprio(ndev, type_data);
+}
+
+int am65_cpsw_qos_ndo_setup_tc(struct net_device *ndev, enum tc_setup_type type,
+                              void *type_data)
+{
+       switch (type) {
+       case TC_SETUP_QDISC_TAPRIO:
+               return am65_cpsw_setup_taprio(ndev, type_data);
+       default:
+               return -EOPNOTSUPP;
+       }
+}
+
+void am65_cpsw_qos_link_up(struct net_device *ndev, int link_speed)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+
+       if (!IS_ENABLED(CONFIG_TI_AM65_CPSW_TAS))
+               return;
+
+       am65_cpsw_est_link_up(ndev, link_speed);
+       port->qos.link_down_time = 0;
+}
+
+void am65_cpsw_qos_link_down(struct net_device *ndev)
+{
+       struct am65_cpsw_port *port = am65_ndev_to_port(ndev);
+
+       if (!IS_ENABLED(CONFIG_TI_AM65_CPSW_TAS))
+               return;
+
+       if (!port->qos.link_down_time)
+               port->qos.link_down_time = ktime_get();
+
+       port->qos.link_speed = SPEED_UNKNOWN;
+}
diff --git a/drivers/net/ethernet/ti/am65-cpsw-qos.h b/drivers/net/ethernet/ti/am65-cpsw-qos.h
new file mode 100644 (file)
index 0000000..e8f1b6b
--- /dev/null
@@ -0,0 +1,29 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright (C) 2020 Texas Instruments Incorporated - http://www.ti.com/
+ */
+
+#ifndef AM65_CPSW_QOS_H_
+#define AM65_CPSW_QOS_H_
+
+#include <linux/netdevice.h>
+#include <net/pkt_sched.h>
+
+struct am65_cpsw_est {
+       int buf;
+       /* has to be the last one */
+       struct tc_taprio_qopt_offload taprio;
+};
+
+struct am65_cpsw_qos {
+       struct am65_cpsw_est *est_admin;
+       struct am65_cpsw_est *est_oper;
+       ktime_t link_down_time;
+       int link_speed;
+};
+
+int am65_cpsw_qos_ndo_setup_tc(struct net_device *ndev, enum tc_setup_type type,
+                              void *type_data);
+void am65_cpsw_qos_link_up(struct net_device *ndev, int link_speed);
+void am65_cpsw_qos_link_down(struct net_device *ndev);
+
+#endif /* AM65_CPSW_QOS_H_ */