1 /*
2  * Copyright (c) 2018 Intel Corporation
3  *
4  * SPDX-License-Identifier: Apache-2.0
5  */
6 #ifndef ZEPHYR_KERNEL_INCLUDE_KSWAP_H_
7 #define ZEPHYR_KERNEL_INCLUDE_KSWAP_H_
8 
9 #include <ksched.h>
10 #include <zephyr/spinlock.h>
11 #include <zephyr/sys/barrier.h>
12 #include <kernel_arch_func.h>
13 
14 #ifdef CONFIG_STACK_SENTINEL
15 extern void z_check_stack_sentinel(void);
16 #else
17 #define z_check_stack_sentinel() /**/
18 #endif /* CONFIG_STACK_SENTINEL */
19 
20 extern struct k_spinlock _sched_spinlock;
21 
22 /* In SMP, the irq_lock() is a spinlock which is implicitly released
23  * and reacquired on context switch to preserve the existing
24  * semantics.  This means that whenever we are about to return to a
25  * thread (via either z_swap() or interrupt/exception return!) we need
26  * to restore the lock state to whatever the thread's counter
27  * expects.
28  */
29 void z_smp_release_global_lock(struct k_thread *thread);
30 
31 /* context switching and scheduling-related routines */
32 #ifdef CONFIG_USE_SWITCH
33 
34 /* Spin, with the scheduler lock held (!), on a thread that is known
35  * (!!) to have released the lock and be on a path where it will
36  * deterministically (!!!) reach arch_switch() in very small constant
37  * time.
38  *
39  * This exists to treat an unavoidable SMP race when threads swap --
40  * their thread record is in the queue (and visible to other CPUs)
41  * before arch_switch() finishes saving state.  We must spin for the
42  * switch handle before entering a new thread.  See docs on
43  * arch_switch().
44  *
45  * Stated differently: there's a chicken and egg bug with the question
46  * of "is a thread running or not?".  The thread needs to mark itself
47  * "not running" from its own context, but at that moment it obviously
48  * is still running until it reaches arch_switch()!  Locking can't
49  * treat this because the scheduler lock can't be released by the
50  * switched-to thread, which is going to (obviously) be running its
51  * own code and doesn't know it was switched out.
52  */
z_sched_switch_spin(struct k_thread * thread)53 static inline void z_sched_switch_spin(struct k_thread *thread)
54 {
55 #ifdef CONFIG_SMP
56 	volatile void **shp = (void *)&thread->switch_handle;
57 
58 	while (*shp == NULL) {
59 		arch_spin_relax();
60 	}
61 	/* Read barrier: don't allow any subsequent loads in the
62 	 * calling code to reorder before we saw switch_handle go
63 	 * non-null.
64 	 */
65 	barrier_dmem_fence_full();
66 #endif /* CONFIG_SMP */
67 }
68 
69 /* New style context switching.  arch_switch() is a lower level
70  * primitive that doesn't know about the scheduler or return value.
71  * Needed for SMP, where the scheduler requires spinlocking that we
72  * don't want to have to do in per-architecture assembly.
73  *
74  * Note that is_spinlock is a compile-time construct which will be
75  * optimized out when this function is expanded.
76  */
do_swap(unsigned int key,struct k_spinlock * lock,bool is_spinlock)77 static ALWAYS_INLINE unsigned int do_swap(unsigned int key,
78 					  struct k_spinlock *lock,
79 					  bool is_spinlock)
80 {
81 	struct k_thread *new_thread, *old_thread;
82 
83 #ifdef CONFIG_SPIN_VALIDATE
84 	/* Make sure the key acts to unmask interrupts, if it doesn't,
85 	 * then we are context switching out of a nested lock
86 	 * (i.e. breaking the lock of someone up the stack) which is
87 	 * forbidden!  The sole exception are dummy threads used
88 	 * during initialization (where we start with interrupts
89 	 * masked and switch away to begin scheduling) and the case of
90 	 * a dead current thread that was just aborted (where the
91 	 * damage was already done by the abort anyway).
92 	 *
93 	 * (Note that this is disabled on ARM64, where system calls
94 	 * can sometimes run with interrupts masked in ways that don't
95 	 * represent lock state.  See #35307)
96 	 */
97 # ifndef CONFIG_ARM64
98 	__ASSERT(arch_irq_unlocked(key) ||
99 		 _current->base.thread_state & (_THREAD_DUMMY | _THREAD_DEAD),
100 		 "Context switching while holding lock!");
101 # endif /* CONFIG_ARM64 */
102 #endif /* CONFIG_SPIN_VALIDATE */
103 
104 	old_thread = _current;
105 
106 	z_check_stack_sentinel();
107 
108 	old_thread->swap_retval = -EAGAIN;
109 
110 	/* We always take the scheduler spinlock if we don't already
111 	 * have it.  We "release" other spinlocks here.  But we never
112 	 * drop the interrupt lock.
113 	 */
114 	if (is_spinlock && lock != NULL && lock != &_sched_spinlock) {
115 		k_spin_release(lock);
116 	}
117 	if (!is_spinlock || lock != &_sched_spinlock) {
118 		(void) k_spin_lock(&_sched_spinlock);
119 	}
120 
121 	new_thread = z_swap_next_thread();
122 
123 	if (new_thread != old_thread) {
124 		z_sched_usage_switch(new_thread);
125 
126 #ifdef CONFIG_SMP
127 		new_thread->base.cpu = arch_curr_cpu()->id;
128 
129 		if (!is_spinlock) {
130 			z_smp_release_global_lock(new_thread);
131 		}
132 #endif /* CONFIG_SMP */
133 		z_thread_mark_switched_out();
134 		z_sched_switch_spin(new_thread);
135 		z_current_thread_set(new_thread);
136 
137 #ifdef CONFIG_TIMESLICING
138 		z_reset_time_slice(new_thread);
139 #endif /* CONFIG_TIMESLICING */
140 
141 #ifdef CONFIG_SPIN_VALIDATE
142 		z_spin_lock_set_owner(&_sched_spinlock);
143 #endif /* CONFIG_SPIN_VALIDATE */
144 
145 		arch_cohere_stacks(old_thread, NULL, new_thread);
146 
147 #ifdef CONFIG_SMP
148 		/* Now add _current back to the run queue, once we are
149 		 * guaranteed to reach the context switch in finite
150 		 * time.  See z_sched_switch_spin().
151 		 */
152 		z_requeue_current(old_thread);
153 #endif /* CONFIG_SMP */
154 		void *newsh = new_thread->switch_handle;
155 
156 		if (IS_ENABLED(CONFIG_SMP)) {
157 			/* Active threads must have a null here.  And
158 			 * it must be seen before the scheduler lock
159 			 * is released!
160 			 */
161 			new_thread->switch_handle = NULL;
162 			barrier_dmem_fence_full(); /* write barrier */
163 		}
164 		k_spin_release(&_sched_spinlock);
165 		arch_switch(newsh, &old_thread->switch_handle);
166 	} else {
167 		k_spin_release(&_sched_spinlock);
168 	}
169 
170 	if (is_spinlock) {
171 		arch_irq_unlock(key);
172 	} else {
173 		irq_unlock(key);
174 	}
175 
176 	return _current->swap_retval;
177 }
178 
z_swap_irqlock(unsigned int key)179 static inline int z_swap_irqlock(unsigned int key)
180 {
181 	return do_swap(key, NULL, false);
182 }
183 
z_swap(struct k_spinlock * lock,k_spinlock_key_t key)184 static inline int z_swap(struct k_spinlock *lock, k_spinlock_key_t key)
185 {
186 	return do_swap(key.key, lock, true);
187 }
188 
z_swap_unlocked(void)189 static inline void z_swap_unlocked(void)
190 {
191 	(void) do_swap(arch_irq_lock(), NULL, true);
192 }
193 
194 #else /* !CONFIG_USE_SWITCH */
195 
196 extern int arch_swap(unsigned int key);
197 
z_sched_switch_spin(struct k_thread * thread)198 static inline void z_sched_switch_spin(struct k_thread *thread)
199 {
200 	ARG_UNUSED(thread);
201 }
202 
z_swap_irqlock(unsigned int key)203 static inline int z_swap_irqlock(unsigned int key)
204 {
205 	int ret;
206 	z_check_stack_sentinel();
207 	ret = arch_swap(key);
208 	return ret;
209 }
210 
211 /* If !USE_SWITCH, then spinlocks are guaranteed degenerate as we
212  * can't be in SMP.  The k_spin_release() call is just for validation
213  * handling.
214  */
z_swap(struct k_spinlock * lock,k_spinlock_key_t key)215 static ALWAYS_INLINE int z_swap(struct k_spinlock *lock, k_spinlock_key_t key)
216 {
217 	k_spin_release(lock);
218 	return z_swap_irqlock(key.key);
219 }
220 
z_swap_unlocked(void)221 static inline void z_swap_unlocked(void)
222 {
223 	(void) z_swap_irqlock(arch_irq_lock());
224 }
225 
226 #endif /* !CONFIG_USE_SWITCH */
227 
228 /**
229  * Set up a "dummy" thread, used at early initialization to launch the
230  * first thread on a CPU.
231  *
232  * Needs to set enough fields such that the context switching code can
233  * use it to properly store state, which will just be discarded.
234  *
235  * The memory of the dummy thread can be completely uninitialized.
236  */
237 void z_dummy_thread_init(struct k_thread *dummy_thread);
238 
239 #endif /* ZEPHYR_KERNEL_INCLUDE_KSWAP_H_ */
240