```// SPDX-License-Identifier: GPL-2.0+
/* Microchip Sparx5 Switch driver
*
* Copyright (c) 2022 Microchip Technology Inc. and its subsidiaries.
*/

#include <net/pkt_cls.h>

#include "sparx5_main.h"
#include "sparx5_qos.h"

/* Calculate new base_time based on cycle_time.
*
* The hardware requires a base_time that is always in the future.
* We define threshold_time as current_time + (2 * cycle_time).
* If base_time is below threshold_time this function recalculates it to be in
* the interval:
* threshold_time <= base_time < (threshold_time + cycle_time)
*
* A very simple algorithm could be like this:
* new_base_time = org_base_time + N * cycle_time
* using the lowest N so (new_base_time >= threshold_time
*/
void sparx5_new_base_time(struct sparx5 *sparx5, const u32 cycle_time,
const ktime_t org_base_time, ktime_t *new_base_time)
{
ktime_t current_time, threshold_time, new_time;
struct timespec64 ts;
u64 nr_of_cycles_p2;
u64 nr_of_cycles;
u64 diff_time;

new_time = org_base_time;

sparx5_ptp_gettime64(&sparx5->phc[SPARX5_PHC_PORT].info, &ts);
current_time = timespec64_to_ktime(ts);
threshold_time = current_time + (2 * cycle_time);
diff_time = threshold_time - new_time;
nr_of_cycles = div_u64(diff_time, cycle_time);
nr_of_cycles_p2 = 1; /* Use 2^0 as start value */

if (new_time >= threshold_time) {
*new_base_time = new_time;
return;
}

/* Calculate the smallest power of 2 (nr_of_cycles_p2)
* that is larger than nr_of_cycles.
*/
while (nr_of_cycles_p2 < nr_of_cycles)
nr_of_cycles_p2 <<= 1; /* Next (higher) power of 2 */

/* Add as big chunks (power of 2 * cycle_time)
* as possible for each power of 2
*/
while (nr_of_cycles_p2) {
if (new_time < threshold_time) {
new_time += cycle_time * nr_of_cycles_p2;
while (new_time < threshold_time)
new_time += cycle_time * nr_of_cycles_p2;
new_time -= cycle_time * nr_of_cycles_p2;
}
nr_of_cycles_p2 >>= 1; /* Next (lower) power of 2 */
}
new_time += cycle_time;
*new_base_time = new_time;
}

/* Max rates for leak groups */
static const u32 spx5_hsch_max_group_rate[SPX5_HSCH_LEAK_GRP_CNT] = {
1048568, /*  1.049 Gbps */
2621420, /*  2.621 Gbps */
10485680, /* 10.486 Gbps */
26214200 /* 26.214 Gbps */
};

static struct sparx5_layer layers[SPX5_HSCH_LAYER_CNT];

static u32 sparx5_lg_get_leak_time(struct sparx5 *sparx5, u32 layer, u32 group)
{
u32 value;

value = spx5_rd(sparx5, HSCH_HSCH_TIMER_CFG(layer, group));
return HSCH_HSCH_TIMER_CFG_LEAK_TIME_GET(value);
}

static void sparx5_lg_set_leak_time(struct sparx5 *sparx5, u32 layer, u32 group,
u32 leak_time)
{
spx5_wr(HSCH_HSCH_TIMER_CFG_LEAK_TIME_SET(leak_time), sparx5,
HSCH_HSCH_TIMER_CFG(layer, group));
}

static u32 sparx5_lg_get_first(struct sparx5 *sparx5, u32 layer, u32 group)
{
u32 value;

value = spx5_rd(sparx5, HSCH_HSCH_LEAK_CFG(layer, group));
return HSCH_HSCH_LEAK_CFG_LEAK_FIRST_GET(value);
}

static u32 sparx5_lg_get_next(struct sparx5 *sparx5, u32 layer, u32 group,
u32 idx)

{
u32 value;

value = spx5_rd(sparx5, HSCH_SE_CONNECT(idx));
}

static u32 sparx5_lg_get_last(struct sparx5 *sparx5, u32 layer, u32 group)
{
u32 itr, next;

itr = sparx5_lg_get_first(sparx5, layer, group);

for (;;) {
next = sparx5_lg_get_next(sparx5, layer, group, itr);
if (itr == next)
return itr;

itr = next;
}
}

static bool sparx5_lg_is_last(struct sparx5 *sparx5, u32 layer, u32 group,
u32 idx)
{
return idx == sparx5_lg_get_next(sparx5, layer, group, idx);
}

static bool sparx5_lg_is_first(struct sparx5 *sparx5, u32 layer, u32 group,
u32 idx)
{
return idx == sparx5_lg_get_first(sparx5, layer, group);
}

static bool sparx5_lg_is_empty(struct sparx5 *sparx5, u32 layer, u32 group)
{
return sparx5_lg_get_leak_time(sparx5, layer, group) == 0;
}

static bool sparx5_lg_is_singular(struct sparx5 *sparx5, u32 layer, u32 group)
{
if (sparx5_lg_is_empty(sparx5, layer, group))
return false;

return sparx5_lg_get_first(sparx5, layer, group) ==
sparx5_lg_get_last(sparx5, layer, group);
}

static void sparx5_lg_enable(struct sparx5 *sparx5, u32 layer, u32 group,
u32 leak_time)
{
sparx5_lg_set_leak_time(sparx5, layer, group, leak_time);
}

static void sparx5_lg_disable(struct sparx5 *sparx5, u32 layer, u32 group)
{
sparx5_lg_set_leak_time(sparx5, layer, group, 0);
}

static int sparx5_lg_get_group_by_index(struct sparx5 *sparx5, u32 layer,
u32 idx, u32 *group)
{
u32 itr, next;
int i;

for (i = 0; i < SPX5_HSCH_LEAK_GRP_CNT; i++) {
if (sparx5_lg_is_empty(sparx5, layer, i))
continue;

itr = sparx5_lg_get_first(sparx5, layer, i);

for (;;) {
next = sparx5_lg_get_next(sparx5, layer, i, itr);

if (itr == idx) {
*group = i;
return 0; /* Found it */
}
if (itr == next)

itr = next;
}
}

return -1;
}

static int sparx5_lg_get_group_by_rate(u32 layer, u32 rate, u32 *group)
{
struct sparx5_layer *l = &layers[layer];
struct sparx5_lg *lg;
u32 i;

for (i = 0; i < SPX5_HSCH_LEAK_GRP_CNT; i++) {
lg = &l->leak_groups[i];
if (rate <= lg->max_rate) {
*group = i;
return 0;
}
}

return -1;
}

static int sparx5_lg_get_adjacent(struct sparx5 *sparx5, u32 layer, u32 group,
u32 idx, u32 *prev, u32 *next, u32 *first)
{
u32 itr;

*first = sparx5_lg_get_first(sparx5, layer, group);
*prev = *first;
*next = *first;
itr = *first;

for (;;) {
*next = sparx5_lg_get_next(sparx5, layer, group, itr);

if (itr == idx)
return 0; /* Found it */

if (itr == *next)

*prev = itr;
itr = *next;
}

return -1;
}

static int sparx5_lg_conf_set(struct sparx5 *sparx5, u32 layer, u32 group,
u32 se_first, u32 idx, u32 idx_next, bool empty)
{
u32 leak_time = layers[layer].leak_groups[group].leak_time;

/* Stop leaking */
sparx5_lg_disable(sparx5, layer, group);

if (empty)
return 0;

/* Select layer */
spx5_rmw(HSCH_HSCH_CFG_CFG_HSCH_LAYER_SET(layer),
HSCH_HSCH_CFG_CFG_HSCH_LAYER, sparx5, HSCH_HSCH_CFG_CFG);

HSCH_SE_CONNECT(idx));

/* Set the first element. */
spx5_rmw(HSCH_HSCH_LEAK_CFG_LEAK_FIRST_SET(se_first),
HSCH_HSCH_LEAK_CFG_LEAK_FIRST, sparx5,
HSCH_HSCH_LEAK_CFG(layer, group));

/* Start leaking */
sparx5_lg_enable(sparx5, layer, group, leak_time);

return 0;
}

static int sparx5_lg_del(struct sparx5 *sparx5, u32 layer, u32 group, u32 idx)
{
u32 first, next, prev;
bool empty = false;

/* idx *must* be present in the leak group */
WARN_ON(sparx5_lg_get_adjacent(sparx5, layer, group, idx, &prev, &next,
&first) < 0);

if (sparx5_lg_is_singular(sparx5, layer, group)) {
empty = true;
} else if (sparx5_lg_is_last(sparx5, layer, group, idx)) {
/* idx is removed, prev is now last */
idx = prev;
next = prev;
} else if (sparx5_lg_is_first(sparx5, layer, group, idx)) {
/* idx is removed and points to itself, first is next */
first = next;
next = idx;
} else {
/* Next is not touched */
idx = prev;
}

return sparx5_lg_conf_set(sparx5, layer, group, first, idx, next,
empty);
}

static int sparx5_lg_add(struct sparx5 *sparx5, u32 layer, u32 new_group,
u32 idx)
{
u32 first, next, old_group;

pr_debug("ADD: layer: %d, new_group: %d, idx: %d", layer, new_group,
idx);

/* Is this SE already shaping ? */
if (sparx5_lg_get_group_by_index(sparx5, layer, idx, &old_group) >= 0) {
if (old_group != new_group) {
/* Delete from old group */
sparx5_lg_del(sparx5, layer, old_group, idx);
} else {
/* Nothing to do here */
return 0;
}
}

first = idx;

if (sparx5_lg_is_empty(sparx5, layer, new_group))
next = idx;
else
next = sparx5_lg_get_first(sparx5, layer, new_group);

return sparx5_lg_conf_set(sparx5, layer, new_group, first, idx, next,
false);
}

static int sparx5_shaper_conf_set(struct sparx5_port *port,
const struct sparx5_shaper *sh, u32 layer,
u32 idx, u32 group)
{
int (*sparx5_lg_action)(struct sparx5 *, u32, u32, u32);
struct sparx5 *sparx5 = port->sparx5;

if (!sh->rate && !sh->burst)
sparx5_lg_action = &sparx5_lg_del;
else

/* Select layer */
spx5_rmw(HSCH_HSCH_CFG_CFG_HSCH_LAYER_SET(layer),
HSCH_HSCH_CFG_CFG_HSCH_LAYER, sparx5, HSCH_HSCH_CFG_CFG);

/* Set frame mode */
spx5_rmw(HSCH_SE_CFG_SE_FRM_MODE_SET(sh->mode), HSCH_SE_CFG_SE_FRM_MODE,
sparx5, HSCH_SE_CFG(idx));

/* Set committed rate and burst */
spx5_wr(HSCH_CIR_CFG_CIR_RATE_SET(sh->rate) |
HSCH_CIR_CFG_CIR_BURST_SET(sh->burst),
sparx5, HSCH_CIR_CFG(idx));

/* This has to be done after the shaper configuration has been set */
sparx5_lg_action(sparx5, layer, group, idx);

return 0;
}

static u32 sparx5_weight_to_hw_cost(u32 weight_min, u32 weight)
{
return ((((SPX5_DWRR_COST_MAX << 4) * weight_min / weight) + 8) >> 4) -
1;
}

static int sparx5_dwrr_conf_set(struct sparx5_port *port,
struct sparx5_dwrr *dwrr)
{
int i;

spx5_rmw(HSCH_HSCH_CFG_CFG_HSCH_LAYER_SET(2) |
HSCH_HSCH_CFG_CFG_CFG_SE_IDX_SET(port->portno),
HSCH_HSCH_CFG_CFG_HSCH_LAYER | HSCH_HSCH_CFG_CFG_CFG_SE_IDX,
port->sparx5, HSCH_HSCH_CFG_CFG);

/* Number of *lower* indexes that are arbitrated dwrr */
spx5_rmw(HSCH_SE_CFG_SE_DWRR_CNT_SET(dwrr->count),
HSCH_SE_CFG_SE_DWRR_CNT, port->sparx5,
HSCH_SE_CFG(port->portno));

for (i = 0; i < dwrr->count; i++) {
spx5_rmw(HSCH_DWRR_ENTRY_DWRR_COST_SET(dwrr->cost[i]),
HSCH_DWRR_ENTRY_DWRR_COST, port->sparx5,
HSCH_DWRR_ENTRY(i));
}

return 0;
}

static int sparx5_leak_groups_init(struct sparx5 *sparx5)
{
struct sparx5_layer *layer;
u32 sys_clk_per_100ps;
struct sparx5_lg *lg;
u32 leak_time_us;
int i, ii;

sys_clk_per_100ps = spx5_rd(sparx5, HSCH_SYS_CLK_PER);

for (i = 0; i < SPX5_HSCH_LAYER_CNT; i++) {
layer = &layers[i];
for (ii = 0; ii < SPX5_HSCH_LEAK_GRP_CNT; ii++) {
lg = &layer->leak_groups[ii];
lg->max_rate = spx5_hsch_max_group_rate[ii];

/* Calculate the leak time in us, to serve a maximum
* rate of 'max_rate' for this group
*/
leak_time_us = (SPX5_SE_RATE_MAX * 1000) / lg->max_rate;

/* Hardware wants leak time in ns */
lg->leak_time = 1000 * leak_time_us;

/* Calculate resolution */
lg->resolution = 1000 / leak_time_us;

/* Maximum number of shapers that can be served by
* this leak group
*/
lg->max_ses = (1000 * leak_time_us) / sys_clk_per_100ps;

/* Example:
* Wanted bandwidth is 100Mbit:
*
* 100 mbps can be served by leak group zero.
*
* leak_time is 125000 ns.
* resolution is: 8
*
* cir          = 100000 / 8 = 12500
* leaks_pr_sec = 125000 / 10^9 = 8000
* bw           = 12500 * 8000 = 10^8 (100 Mbit)
*/

/* Disable by default - this also indicates an empty
* leak group
*/
sparx5_lg_disable(sparx5, i, ii);
}
}

return 0;
}

int sparx5_qos_init(struct sparx5 *sparx5)
{
int ret;

ret = sparx5_leak_groups_init(sparx5);
if (ret < 0)
return ret;

ret = sparx5_dcb_init(sparx5);
if (ret < 0)
return ret;

sparx5_psfp_init(sparx5);

return 0;
}

int sparx5_tc_mqprio_add(struct net_device *ndev, u8 num_tc)
{
int i;

if (num_tc != SPX5_PRIOS) {
netdev_err(ndev, "Only %d traffic classes supported\n",
SPX5_PRIOS);
return -EINVAL;
}

netdev_set_num_tc(ndev, num_tc);

for (i = 0; i < num_tc; i++)
netdev_set_tc_queue(ndev, i, 1, i);

netdev_dbg(ndev, "dev->num_tc %u dev->real_num_tx_queues %u\n",
ndev->num_tc, ndev->real_num_tx_queues);

return 0;
}

int sparx5_tc_mqprio_del(struct net_device *ndev)
{
netdev_reset_tc(ndev);

netdev_dbg(ndev, "dev->num_tc %u dev->real_num_tx_queues %u\n",
ndev->num_tc, ndev->real_num_tx_queues);

return 0;
}

u32 layer, u32 idx)
{
struct sparx5_shaper sh = {
.mode = SPX5_SE_MODE_DATARATE,
.rate = div_u64(params->rate.rate_bytes_ps, 1000) * 8,
.burst = params->max_size,
};
struct sparx5_lg *lg;
u32 group;

/* Find suitable group for this se */
if (sparx5_lg_get_group_by_rate(layer, sh.rate, &group) < 0) {
pr_debug("Could not find leak group for se with rate: %d",
sh.rate);
return -EINVAL;
}

lg = &layers[layer].leak_groups[group];

pr_debug("Found matching group (speed: %d)\n", lg->max_rate);

if (sh.rate < SPX5_SE_RATE_MIN || sh.burst < SPX5_SE_BURST_MIN)
return -EINVAL;

/* Calculate committed rate and burst */
sh.rate = DIV_ROUND_UP(sh.rate, lg->resolution);
sh.burst = DIV_ROUND_UP(sh.burst, SPX5_SE_BURST_UNIT);

if (sh.rate > SPX5_SE_RATE_MAX || sh.burst > SPX5_SE_BURST_MAX)
return -EINVAL;

return sparx5_shaper_conf_set(port, &sh, layer, idx, group);
}

int sparx5_tc_tbf_del(struct sparx5_port *port, u32 layer, u32 idx)
{
struct sparx5_shaper sh = {0};
u32 group;

sparx5_lg_get_group_by_index(port->sparx5, layer, idx, &group);

return sparx5_shaper_conf_set(port, &sh, layer, idx, group);
}

{
struct sparx5_dwrr dwrr = {0};
/* Minimum weight for each iteration */
unsigned int w_min = 100;
int i;

/* Find minimum weight for all dwrr bands */
for (i = 0; i < SPX5_PRIOS; i++) {
if (params->quanta[i] == 0)
continue;
w_min = min(w_min, params->weights[i]);
}

for (i = 0; i < SPX5_PRIOS; i++) {
/* Strict band; skip */
if (params->quanta[i] == 0)
continue;

dwrr.count++;

/* On the sparx5, bands with higher indexes are preferred and
* arbitrated strict. Strict bands are put in the lower indexes,
* by tc, so we reverse the bands here.
*
* Also convert the weight to something the hardware
* understands.
*/
dwrr.cost[SPX5_PRIOS - i - 1] =
sparx5_weight_to_hw_cost(w_min, params->weights[i]);
}

return sparx5_dwrr_conf_set(port, &dwrr);
}

int sparx5_tc_ets_del(struct sparx5_port *port)
{
struct sparx5_dwrr dwrr = {0};

return sparx5_dwrr_conf_set(port, &dwrr);
}```