Some events needs to be handled with a very low latency constraint. If the system is in deep sleep, exit latency from this low level state exceeds sometimes the maximum latency constraint of these events. Before suspending the system, select which events is happening sooner, kernel events or normal events. CPU will be up just before the next event occurs taking into account the exit latency of the current power state Change also the policy event API to take as argument absolute time in HW cycles instead of time in us Signed-off-by: Riadh Ghaddab <rghaddab@baylibre.com>
272 lines
7.5 KiB
C
272 lines
7.5 KiB
C
/*
|
|
* Copyright (c) 2018 Intel Corporation.
|
|
*
|
|
* SPDX-License-Identifier: Apache-2.0
|
|
*/
|
|
|
|
#include <zephyr/device.h>
|
|
#include <zephyr/kernel.h>
|
|
#include <zephyr/kernel_structs.h>
|
|
#include <zephyr/init.h>
|
|
#include <string.h>
|
|
#include <zephyr/drivers/timer/system_timer.h>
|
|
#include <zephyr/pm/device.h>
|
|
#include <zephyr/pm/device_runtime.h>
|
|
#include <zephyr/pm/pm.h>
|
|
#include <zephyr/pm/state.h>
|
|
#include <zephyr/pm/policy.h>
|
|
#include <zephyr/tracing/tracing.h>
|
|
|
|
#include "pm_stats.h"
|
|
#include "device_system_managed.h"
|
|
|
|
#include <zephyr/logging/log.h>
|
|
LOG_MODULE_REGISTER(pm, CONFIG_PM_LOG_LEVEL);
|
|
|
|
static ATOMIC_DEFINE(z_post_ops_required, CONFIG_MP_MAX_NUM_CPUS);
|
|
static sys_slist_t pm_notifiers = SYS_SLIST_STATIC_INIT(&pm_notifiers);
|
|
|
|
/*
|
|
* Properly initialize cpu power states. Do not make assumptions that
|
|
* ACTIVE_STATE is 0
|
|
*/
|
|
#define CPU_PM_STATE_INIT(_, __) \
|
|
{ .state = PM_STATE_ACTIVE }
|
|
static struct pm_state_info z_cpus_pm_state[] = {
|
|
LISTIFY(CONFIG_MP_MAX_NUM_CPUS, CPU_PM_STATE_INIT, (,))
|
|
};
|
|
|
|
static struct pm_state_info z_cpus_pm_forced_state[] = {
|
|
LISTIFY(CONFIG_MP_MAX_NUM_CPUS, CPU_PM_STATE_INIT, (,))
|
|
};
|
|
|
|
static struct k_spinlock pm_forced_state_lock;
|
|
static struct k_spinlock pm_notifier_lock;
|
|
|
|
/*
|
|
* Function called to notify when the system is entering / exiting a
|
|
* power state
|
|
*/
|
|
static inline void pm_state_notify(bool entering_state)
|
|
{
|
|
struct pm_notifier *notifier;
|
|
k_spinlock_key_t pm_notifier_key;
|
|
void (*callback)(enum pm_state state);
|
|
|
|
pm_notifier_key = k_spin_lock(&pm_notifier_lock);
|
|
SYS_SLIST_FOR_EACH_CONTAINER(&pm_notifiers, notifier, _node) {
|
|
if (entering_state) {
|
|
callback = notifier->state_entry;
|
|
} else {
|
|
callback = notifier->state_exit;
|
|
}
|
|
|
|
if (callback) {
|
|
callback(z_cpus_pm_state[_current_cpu->id].state);
|
|
}
|
|
}
|
|
k_spin_unlock(&pm_notifier_lock, pm_notifier_key);
|
|
}
|
|
|
|
static inline int32_t ticks_expiring_sooner(int32_t ticks1, int32_t ticks2)
|
|
{
|
|
/*
|
|
* Ticks are relative numbers that defines the number of ticks
|
|
* until the next event.
|
|
* Its maximum value is K_TICKS_FOREVER ((uint32_t)-1) which is -1
|
|
* when we cast it to (int32_t)
|
|
* We need to find out which one is the closest
|
|
*/
|
|
|
|
__ASSERT(ticks1 >= -1, "ticks1 has unexpected negative value");
|
|
__ASSERT(ticks2 >= -1, "ticks2 has unexpected negative value");
|
|
|
|
if (ticks1 == K_TICKS_FOREVER) {
|
|
return ticks2;
|
|
}
|
|
if (ticks2 == K_TICKS_FOREVER) {
|
|
return ticks1;
|
|
}
|
|
/* At this step ticks1 and ticks2 are positive */
|
|
return MIN(ticks1, ticks2);
|
|
}
|
|
|
|
void pm_system_resume(void)
|
|
{
|
|
uint8_t id = _current_cpu->id;
|
|
|
|
/*
|
|
* This notification is called from the ISR of the event
|
|
* that caused exit from kernel idling after PM operations.
|
|
*
|
|
* Some CPU low power states require enabling of interrupts
|
|
* atomically when entering those states. The wake up from
|
|
* such a state first executes code in the ISR of the interrupt
|
|
* that caused the wake. This hook will be called from the ISR.
|
|
* For such CPU LPS states, do post operations and restores here.
|
|
* The kernel scheduler will get control after the ISR finishes
|
|
* and it may schedule another thread.
|
|
*/
|
|
if (atomic_test_and_clear_bit(z_post_ops_required, id)) {
|
|
#ifdef CONFIG_PM_DEVICE_SYSTEM_MANAGED
|
|
if (atomic_add(&_cpus_active, 1) == 0) {
|
|
if ((z_cpus_pm_state[id].state != PM_STATE_RUNTIME_IDLE) &&
|
|
!z_cpus_pm_state[id].pm_device_disabled) {
|
|
pm_resume_devices();
|
|
}
|
|
}
|
|
#endif
|
|
pm_state_exit_post_ops(z_cpus_pm_state[id].state, z_cpus_pm_state[id].substate_id);
|
|
pm_state_notify(false);
|
|
#ifdef CONFIG_SYS_CLOCK_EXISTS
|
|
sys_clock_idle_exit();
|
|
#endif /* CONFIG_SYS_CLOCK_EXISTS */
|
|
z_cpus_pm_state[id] = (struct pm_state_info){PM_STATE_ACTIVE,
|
|
0, 0};
|
|
}
|
|
}
|
|
|
|
bool pm_state_force(uint8_t cpu, const struct pm_state_info *info)
|
|
{
|
|
k_spinlock_key_t key;
|
|
|
|
__ASSERT(info->state < PM_STATE_COUNT,
|
|
"Invalid power state %d!", info->state);
|
|
|
|
key = k_spin_lock(&pm_forced_state_lock);
|
|
z_cpus_pm_forced_state[cpu] = *info;
|
|
k_spin_unlock(&pm_forced_state_lock, key);
|
|
|
|
return true;
|
|
}
|
|
|
|
bool pm_system_suspend(int32_t kernel_ticks)
|
|
{
|
|
uint8_t id = _current_cpu->id;
|
|
k_spinlock_key_t key;
|
|
int32_t ticks, events_ticks;
|
|
|
|
SYS_PORT_TRACING_FUNC_ENTER(pm, system_suspend, kernel_ticks);
|
|
|
|
/*
|
|
* CPU needs to be fully wake up before the event is triggered.
|
|
* We need to find out first the ticks to the next event
|
|
*/
|
|
events_ticks = pm_policy_next_event_ticks();
|
|
ticks = ticks_expiring_sooner(kernel_ticks, events_ticks);
|
|
|
|
key = k_spin_lock(&pm_forced_state_lock);
|
|
if (z_cpus_pm_forced_state[id].state != PM_STATE_ACTIVE) {
|
|
z_cpus_pm_state[id] = z_cpus_pm_forced_state[id];
|
|
z_cpus_pm_forced_state[id].state = PM_STATE_ACTIVE;
|
|
} else {
|
|
const struct pm_state_info *info;
|
|
|
|
info = pm_policy_next_state(id, ticks);
|
|
if (info != NULL) {
|
|
z_cpus_pm_state[id] = *info;
|
|
} else {
|
|
z_cpus_pm_state[id].state = PM_STATE_ACTIVE;
|
|
}
|
|
}
|
|
k_spin_unlock(&pm_forced_state_lock, key);
|
|
|
|
if (z_cpus_pm_state[id].state == PM_STATE_ACTIVE) {
|
|
LOG_DBG("No PM operations done.");
|
|
SYS_PORT_TRACING_FUNC_EXIT(pm, system_suspend, ticks,
|
|
z_cpus_pm_state[id].state);
|
|
return false;
|
|
}
|
|
|
|
#ifdef CONFIG_PM_DEVICE_SYSTEM_MANAGED
|
|
if (atomic_sub(&_cpus_active, 1) == 1) {
|
|
if ((z_cpus_pm_state[id].state != PM_STATE_RUNTIME_IDLE) &&
|
|
!z_cpus_pm_state[id].pm_device_disabled) {
|
|
if (!pm_suspend_devices()) {
|
|
pm_resume_devices();
|
|
z_cpus_pm_state[id].state = PM_STATE_ACTIVE;
|
|
(void)atomic_add(&_cpus_active, 1);
|
|
SYS_PORT_TRACING_FUNC_EXIT(pm, system_suspend, ticks,
|
|
z_cpus_pm_state[id].state);
|
|
return false;
|
|
}
|
|
}
|
|
}
|
|
#endif
|
|
|
|
if ((z_cpus_pm_state[id].exit_latency_us != 0) &&
|
|
(ticks != K_TICKS_FOREVER)) {
|
|
/*
|
|
* We need to set the timer to interrupt a little bit early to
|
|
* accommodate the time required by the CPU to fully wake up.
|
|
*/
|
|
sys_clock_set_timeout(ticks -
|
|
k_us_to_ticks_ceil32(
|
|
z_cpus_pm_state[id].exit_latency_us),
|
|
true);
|
|
}
|
|
|
|
/*
|
|
* This function runs with interruptions locked but it is
|
|
* expected the SoC to unlock them in
|
|
* pm_state_exit_post_ops() when returning to active
|
|
* state. We don't want to be scheduled out yet, first we need
|
|
* to send a notification about leaving the idle state. So,
|
|
* we lock the scheduler here and unlock just after we have
|
|
* sent the notification in pm_system_resume().
|
|
*/
|
|
k_sched_lock();
|
|
pm_stats_start();
|
|
/* Enter power state */
|
|
pm_state_notify(true);
|
|
atomic_set_bit(z_post_ops_required, id);
|
|
pm_state_set(z_cpus_pm_state[id].state, z_cpus_pm_state[id].substate_id);
|
|
pm_stats_stop();
|
|
|
|
/* Wake up sequence starts here */
|
|
pm_stats_update(z_cpus_pm_state[id].state);
|
|
pm_system_resume();
|
|
k_sched_unlock();
|
|
SYS_PORT_TRACING_FUNC_EXIT(pm, system_suspend, ticks,
|
|
z_cpus_pm_state[id].state);
|
|
|
|
return true;
|
|
}
|
|
|
|
void pm_notifier_register(struct pm_notifier *notifier)
|
|
{
|
|
k_spinlock_key_t pm_notifier_key = k_spin_lock(&pm_notifier_lock);
|
|
|
|
sys_slist_append(&pm_notifiers, ¬ifier->_node);
|
|
k_spin_unlock(&pm_notifier_lock, pm_notifier_key);
|
|
}
|
|
|
|
int pm_notifier_unregister(struct pm_notifier *notifier)
|
|
{
|
|
int ret = -EINVAL;
|
|
k_spinlock_key_t pm_notifier_key;
|
|
|
|
pm_notifier_key = k_spin_lock(&pm_notifier_lock);
|
|
if (sys_slist_find_and_remove(&pm_notifiers, &(notifier->_node))) {
|
|
ret = 0;
|
|
}
|
|
k_spin_unlock(&pm_notifier_lock, pm_notifier_key);
|
|
|
|
return ret;
|
|
}
|
|
|
|
const struct pm_state_info *pm_state_next_get(uint8_t cpu)
|
|
{
|
|
return &z_cpus_pm_state[cpu];
|
|
}
|
|
|
|
void z_pm_save_idle_exit(void)
|
|
{
|
|
/* Some CPU low power states require notification at the ISR
|
|
* to allow any operations that needs to be done before kernel
|
|
* switches task or processes nested interrupts.
|
|
* This can be simply ignored if not required.
|
|
*/
|
|
pm_system_resume();
|
|
}
|