zephyr/kernel/timeout.c
Andy Ross 887e1abace kernel/timeout: Fix timeout "sooner" computation
There was an edge case in the timeout handling (exposed by, but not
strictly related to, the recent timeslice fix): the next_timeout()
computation would include time slice expiration as a clamp on the
result, but this would be invoked also on the z_set_timeout_expiry()
path which gets hooked on entry to a new thread which is needed to set
the timeout in the first place.  So if no other timer interrupt was
scheduled, it was possible to miss the first timeslice interrupt after
thread scheduling.

The explanation is much longer than the fix (use <= as the comparator
instead of <).

In practice this was only being hit in the existing test suite on
riscv miv running under renode using non-default clock rates.

Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
2021-02-02 17:58:40 -05:00

327 lines
6.9 KiB
C

/*
* Copyright (c) 2018 Intel Corporation
*
* SPDX-License-Identifier: Apache-2.0
*/
#include <kernel.h>
#include <spinlock.h>
#include <ksched.h>
#include <timeout_q.h>
#include <syscall_handler.h>
#include <drivers/timer/system_timer.h>
#include <sys_clock.h>
#define LOCKED(lck) for (k_spinlock_key_t __i = {}, \
__key = k_spin_lock(lck); \
__i.key == 0; \
k_spin_unlock(lck, __key), __i.key = 1)
static uint64_t curr_tick;
static sys_dlist_t timeout_list = SYS_DLIST_STATIC_INIT(&timeout_list);
static struct k_spinlock timeout_lock;
#define MAX_WAIT (IS_ENABLED(CONFIG_SYSTEM_CLOCK_SLOPPY_IDLE) \
? K_TICKS_FOREVER : INT_MAX)
/* Cycles left to process in the currently-executing z_clock_announce() */
static int announce_remaining;
#if defined(CONFIG_TIMER_READS_ITS_FREQUENCY_AT_RUNTIME)
int z_clock_hw_cycles_per_sec = CONFIG_SYS_CLOCK_HW_CYCLES_PER_SEC;
#ifdef CONFIG_USERSPACE
static inline int z_vrfy_z_clock_hw_cycles_per_sec_runtime_get(void)
{
return z_impl_z_clock_hw_cycles_per_sec_runtime_get();
}
#include <syscalls/z_clock_hw_cycles_per_sec_runtime_get_mrsh.c>
#endif /* CONFIG_USERSPACE */
#endif /* CONFIG_TIMER_READS_ITS_FREQUENCY_AT_RUNTIME */
static struct _timeout *first(void)
{
sys_dnode_t *t = sys_dlist_peek_head(&timeout_list);
return t == NULL ? NULL : CONTAINER_OF(t, struct _timeout, node);
}
static struct _timeout *next(struct _timeout *t)
{
sys_dnode_t *n = sys_dlist_peek_next(&timeout_list, &t->node);
return n == NULL ? NULL : CONTAINER_OF(n, struct _timeout, node);
}
static void remove_timeout(struct _timeout *t)
{
if (next(t) != NULL) {
next(t)->dticks += t->dticks;
}
sys_dlist_remove(&t->node);
}
static int32_t elapsed(void)
{
return announce_remaining == 0 ? z_clock_elapsed() : 0U;
}
static int32_t next_timeout(void)
{
struct _timeout *to = first();
int32_t ticks_elapsed = elapsed();
int32_t ret = to == NULL ? MAX_WAIT
: CLAMP(to->dticks - ticks_elapsed, 0, MAX_WAIT);
#ifdef CONFIG_TIMESLICING
if (_current_cpu->slice_ticks && _current_cpu->slice_ticks < ret) {
ret = _current_cpu->slice_ticks;
}
#endif
return ret;
}
void z_add_timeout(struct _timeout *to, _timeout_func_t fn,
k_timeout_t timeout)
{
if (K_TIMEOUT_EQ(timeout, K_FOREVER)) {
return;
}
#ifdef KERNEL_COHERENCE
__ASSERT_NO_MSG(arch_mem_coherent(to));
#endif
k_ticks_t ticks = timeout.ticks + 1;
if (IS_ENABLED(CONFIG_TIMEOUT_64BIT) && Z_TICK_ABS(ticks) >= 0) {
ticks = Z_TICK_ABS(ticks) - (curr_tick + elapsed());
}
__ASSERT(!sys_dnode_is_linked(&to->node), "");
to->fn = fn;
ticks = MAX(1, ticks);
LOCKED(&timeout_lock) {
struct _timeout *t;
to->dticks = ticks + elapsed();
for (t = first(); t != NULL; t = next(t)) {
if (t->dticks > to->dticks) {
t->dticks -= to->dticks;
sys_dlist_insert(&t->node, &to->node);
break;
}
to->dticks -= t->dticks;
}
if (t == NULL) {
sys_dlist_append(&timeout_list, &to->node);
}
if (to == first()) {
#if CONFIG_TIMESLICING
/*
* This is not ideal, since it does not
* account the time elapsed since the the
* last announcement, and slice_ticks is based
* on that. It means the that time remaining for
* the next announcement can be lesser than
* slice_ticks.
*/
int32_t next_time = next_timeout();
if (next_time == 0 ||
_current_cpu->slice_ticks != next_time) {
z_clock_set_timeout(next_time, false);
}
#else
z_clock_set_timeout(next_timeout(), false);
#endif /* CONFIG_TIMESLICING */
}
}
}
int z_abort_timeout(struct _timeout *to)
{
int ret = -EINVAL;
LOCKED(&timeout_lock) {
if (sys_dnode_is_linked(&to->node)) {
remove_timeout(to);
ret = 0;
}
}
return ret;
}
/* must be locked */
static k_ticks_t timeout_rem(const struct _timeout *timeout)
{
k_ticks_t ticks = 0;
if (z_is_inactive_timeout(timeout)) {
return 0;
}
for (struct _timeout *t = first(); t != NULL; t = next(t)) {
ticks += t->dticks;
if (timeout == t) {
break;
}
}
return ticks - elapsed();
}
k_ticks_t z_timeout_remaining(const struct _timeout *timeout)
{
k_ticks_t ticks = 0;
LOCKED(&timeout_lock) {
ticks = timeout_rem(timeout);
}
return ticks;
}
k_ticks_t z_timeout_expires(const struct _timeout *timeout)
{
k_ticks_t ticks = 0;
LOCKED(&timeout_lock) {
ticks = curr_tick + timeout_rem(timeout);
}
return ticks;
}
int32_t z_get_next_timeout_expiry(void)
{
int32_t ret = (int32_t) K_TICKS_FOREVER;
LOCKED(&timeout_lock) {
ret = next_timeout();
}
return ret;
}
void z_set_timeout_expiry(int32_t ticks, bool is_idle)
{
LOCKED(&timeout_lock) {
int next_to = next_timeout();
bool sooner = (next_to == K_TICKS_FOREVER)
|| (ticks <= next_to);
bool imminent = next_to <= 1;
/* Only set new timeouts when they are sooner than
* what we have. Also don't try to set a timeout when
* one is about to expire: drivers have internal logic
* that will bump the timeout to the "next" tick if
* it's not considered to be settable as directed.
* SMP can't use this optimization though: we don't
* know when context switches happen until interrupt
* exit and so can't get the timeslicing clamp folded
* in.
*/
if (!imminent && (sooner || IS_ENABLED(CONFIG_SMP))) {
z_clock_set_timeout(ticks, is_idle);
}
}
}
void z_clock_announce(int32_t ticks)
{
#ifdef CONFIG_TIMESLICING
z_time_slice(ticks);
#endif
k_spinlock_key_t key = k_spin_lock(&timeout_lock);
announce_remaining = ticks;
while (first() != NULL && first()->dticks <= announce_remaining) {
struct _timeout *t = first();
int dt = t->dticks;
curr_tick += dt;
announce_remaining -= dt;
t->dticks = 0;
remove_timeout(t);
k_spin_unlock(&timeout_lock, key);
t->fn(t);
key = k_spin_lock(&timeout_lock);
}
if (first() != NULL) {
first()->dticks -= announce_remaining;
}
curr_tick += announce_remaining;
announce_remaining = 0;
z_clock_set_timeout(next_timeout(), false);
k_spin_unlock(&timeout_lock, key);
}
int64_t z_tick_get(void)
{
uint64_t t = 0U;
LOCKED(&timeout_lock) {
t = curr_tick + z_clock_elapsed();
}
return t;
}
uint32_t z_tick_get_32(void)
{
#ifdef CONFIG_TICKLESS_KERNEL
return (uint32_t)z_tick_get();
#else
return (uint32_t)curr_tick;
#endif
}
int64_t z_impl_k_uptime_ticks(void)
{
return z_tick_get();
}
#ifdef CONFIG_USERSPACE
static inline int64_t z_vrfy_k_uptime_ticks(void)
{
return z_impl_k_uptime_ticks();
}
#include <syscalls/k_uptime_ticks_mrsh.c>
#endif
/* Returns the uptime expiration (relative to an unlocked "now"!) of a
* timeout object. When used correctly, this should be called once,
* synchronously with the user passing a new timeout value. It should
* not be used iteratively to adjust a timeout.
*/
uint64_t z_timeout_end_calc(k_timeout_t timeout)
{
k_ticks_t dt;
if (K_TIMEOUT_EQ(timeout, K_FOREVER)) {
return UINT64_MAX;
} else if (K_TIMEOUT_EQ(timeout, K_NO_WAIT)) {
return z_tick_get();
}
dt = timeout.ticks;
if (IS_ENABLED(CONFIG_TIMEOUT_64BIT) && Z_TICK_ABS(dt) >= 0) {
return Z_TICK_ABS(dt);
}
return z_tick_get() + MAX(1, dt);
}