OpenCloudOS-Kernel/drivers/net/ethernet/ti/am65-cpsw-qos.c

625 lines
16 KiB
C
Raw Normal View History

ethernet: ti: am65-cpsw-qos: add TAPRIO offload support 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>
2020-05-13 21:26:15 +08:00
// 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);
}
/*
ethernet: ti: am65-cpsw-qos: add TAPRIO offload support 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>
2020-05-13 21:26:15 +08:00
* 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;
int ret = 0;
if (taprio->cycle_time_extension) {
dev_err(&ndev->dev, "Failed to set cycle time extension");
return -EOPNOTSUPP;
}
est_new = devm_kzalloc(&ndev->dev,
struct_size(est_new, taprio.entries, taprio->num_entries),
GFP_KERNEL);
ethernet: ti: am65-cpsw-qos: add TAPRIO offload support 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>
2020-05-13 21:26:15 +08:00
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;
}