zephyr/kernel/smp.c
Andy Ross fd340ebf31 sched: Optimize dummy thread usage on SMP
Nicolas Pitre points out that since these thread structs are just
dummies for the context swtiching, they can be presumed to be "write
only" and thus there's no point in having one per CPU, everyone can
share the same one.

The only gotcha is that we never really documented (nor really have a
place to document) that rule, so it's not theoretically impossible for
an architecture to read back what it might have written underneath
arch_switch().  Leave this in a separate commit for bisection
purposes, but the risk seems very low.

Signed-off-by: Andy Ross <andyross@google.com>
2024-05-02 13:55:03 -04:00

250 lines
5.6 KiB
C

/* Copyright (c) 2022 Intel corporation
* SPDX-License-Identifier: Apache-2.0
*/
#include <zephyr/kernel.h>
#include <zephyr/kernel_structs.h>
#include <zephyr/kernel/smp.h>
#include <zephyr/spinlock.h>
#include <kswap.h>
#include <kernel_internal.h>
static atomic_t global_lock;
/**
* Flag to tell recently powered up CPU to start
* initialization routine.
*
* 0 to tell powered up CPU to wait.
* 1 to tell powered up CPU to continue initialization.
*/
static atomic_t cpu_start_flag;
/**
* Flag to tell caller that the target CPU is now
* powered up and ready to be initialized.
*
* 0 if target CPU is not yet ready.
* 1 if target CPU has powered up and ready to be initialized.
*/
static atomic_t ready_flag;
/**
* Struct holding the function to be called before handing off
* to schedule and its argument.
*/
static struct cpu_start_cb {
/**
* Function to be called before handing off to scheduler.
* Can be NULL.
*/
smp_init_fn fn;
/** Argument to @ref cpu_start_fn.fn. */
void *arg;
/** Invoke scheduler after CPU has started if true. */
bool invoke_sched;
#ifdef CONFIG_SYS_CLOCK_EXISTS
/** True if smp_timer_init() needs to be called. */
bool reinit_timer;
#endif /* CONFIG_SYS_CLOCK_EXISTS */
} cpu_start_fn;
static struct k_spinlock cpu_start_lock;
unsigned int z_smp_global_lock(void)
{
unsigned int key = arch_irq_lock();
if (!_current->base.global_lock_count) {
while (!atomic_cas(&global_lock, 0, 1)) {
}
}
_current->base.global_lock_count++;
return key;
}
void z_smp_global_unlock(unsigned int key)
{
if (_current->base.global_lock_count != 0U) {
_current->base.global_lock_count--;
if (!_current->base.global_lock_count) {
(void)atomic_clear(&global_lock);
}
}
arch_irq_unlock(key);
}
/* Called from within z_swap(), so assumes lock already held */
void z_smp_release_global_lock(struct k_thread *thread)
{
if (!thread->base.global_lock_count) {
(void)atomic_clear(&global_lock);
}
}
/* Tiny delay that relaxes bus traffic to avoid spamming a shared
* memory bus looking at an atomic variable
*/
static inline void local_delay(void)
{
for (volatile int i = 0; i < 1000; i++) {
}
}
static void wait_for_start_signal(atomic_t *start_flag)
{
/* Wait for the signal to begin scheduling */
while (!atomic_get(start_flag)) {
local_delay();
}
}
static inline void smp_init_top(void *arg)
{
struct cpu_start_cb csc = arg ? *(struct cpu_start_cb *)arg : (struct cpu_start_cb){0};
/* Let start_cpu() know that this CPU has powered up. */
(void)atomic_set(&ready_flag, 1);
/* Wait for the CPU start caller to signal that
* we can start initialization.
*/
wait_for_start_signal(&cpu_start_flag);
if ((arg == NULL) || csc.invoke_sched) {
/* Initialize the dummy thread struct so that
* the scheduler can schedule actual threads to run.
*/
z_dummy_thread_init(&_thread_dummy);
}
#ifdef CONFIG_SYS_CLOCK_EXISTS
if ((arg == NULL) || csc.reinit_timer) {
smp_timer_init();
}
#endif /* CONFIG_SYS_CLOCK_EXISTS */
/* Do additional initialization steps if needed. */
if (csc.fn != NULL) {
csc.fn(csc.arg);
}
if ((arg != NULL) && !csc.invoke_sched) {
/* Don't invoke scheduler. */
return;
}
/* Let scheduler decide what thread to run next. */
z_swap_unlocked();
CODE_UNREACHABLE; /* LCOV_EXCL_LINE */
}
static void start_cpu(int id, struct cpu_start_cb *csc)
{
/* Clear the ready flag so the newly powered up CPU can
* signal that it has powered up.
*/
(void)atomic_clear(&ready_flag);
/* Power up the CPU */
arch_cpu_start(id, z_interrupt_stacks[id], CONFIG_ISR_STACK_SIZE,
smp_init_top, csc);
/* Wait until the newly powered up CPU to signal that
* it has powered up.
*/
while (!atomic_get(&ready_flag)) {
local_delay();
}
}
void k_smp_cpu_start(int id, smp_init_fn fn, void *arg)
{
k_spinlock_key_t key = k_spin_lock(&cpu_start_lock);
cpu_start_fn.fn = fn;
cpu_start_fn.arg = arg;
cpu_start_fn.invoke_sched = true;
#ifdef CONFIG_SYS_CLOCK_EXISTS
cpu_start_fn.reinit_timer = true;
#endif /* CONFIG_SYS_CLOCK_EXISTS */
/* We are only starting one CPU so we do not need to synchronize
* across all CPUs using the start_flag. So just set it to 1.
*/
(void)atomic_set(&cpu_start_flag, 1); /* async, don't care */
/* Initialize various CPU structs related to this CPU. */
z_init_cpu(id);
/* Start the CPU! */
start_cpu(id, &cpu_start_fn);
k_spin_unlock(&cpu_start_lock, key);
}
void k_smp_cpu_resume(int id, smp_init_fn fn, void *arg,
bool reinit_timer, bool invoke_sched)
{
k_spinlock_key_t key = k_spin_lock(&cpu_start_lock);
cpu_start_fn.fn = fn;
cpu_start_fn.arg = arg;
cpu_start_fn.invoke_sched = invoke_sched;
#ifdef CONFIG_SYS_CLOCK_EXISTS
cpu_start_fn.reinit_timer = reinit_timer;
#else
ARG_UNUSED(reinit_timer);
#endif /* CONFIG_SYS_CLOCK_EXISTS */
/* We are only starting one CPU so we do not need to synchronize
* across all CPUs using the start_flag. So just set it to 1.
*/
(void)atomic_set(&cpu_start_flag, 1);
/* Start the CPU! */
start_cpu(id, &cpu_start_fn);
k_spin_unlock(&cpu_start_lock, key);
}
void z_smp_init(void)
{
/* We are powering up all CPUs and we want to synchronize their
* entry into scheduler. So set the start flag to 0 here.
*/
(void)atomic_clear(&cpu_start_flag);
/* Just start CPUs one by one. */
unsigned int num_cpus = arch_num_cpus();
for (int i = 1; i < num_cpus; i++) {
z_init_cpu(i);
start_cpu(i, NULL);
}
/* Let loose those CPUs so they can start scheduling
* threads to run.
*/
(void)atomic_set(&cpu_start_flag, 1);
}
bool z_smp_cpu_mobile(void)
{
unsigned int k = arch_irq_lock();
bool pinned = arch_is_in_isr() || !arch_irq_unlocked(k);
arch_irq_unlock(k);
return !pinned;
}