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
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
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 	ARG_UNUSED(lock);
82 	struct k_thread *new_thread, *old_thread;
83 
84 #ifdef CONFIG_SPIN_VALIDATE
85 	/* Make sure the key acts to unmask interrupts, if it doesn't,
86 	 * then we are context switching out of a nested lock
87 	 * (i.e. breaking the lock of someone up the stack) which is
88 	 * forbidden!  The sole exception are dummy threads used
89 	 * during initialization (where we start with interrupts
90 	 * masked and switch away to begin scheduling) and the case of
91 	 * a dead current thread that was just aborted (where the
92 	 * damage was already done by the abort anyway).
93 	 *
94 	 * (Note that this is disabled on ARM64, where system calls
95 	 * can sometimes run with interrupts masked in ways that don't
96 	 * represent lock state.  See #35307)
97 	 */
98 # ifndef CONFIG_ARM64
99 	__ASSERT(arch_irq_unlocked(key) ||
100 		 _current->base.thread_state & (_THREAD_DUMMY | _THREAD_DEAD),
101 		 "Context switching while holding lock!");
102 # endif
103 #endif
104 
105 	old_thread = _current;
106 
107 	z_check_stack_sentinel();
108 
109 	old_thread->swap_retval = -EAGAIN;
110 
111 	/* We always take the scheduler spinlock if we don't already
112 	 * have it.  We "release" other spinlocks here.  But we never
113 	 * drop the interrupt lock.
114 	 */
115 	if (is_spinlock && lock != NULL && lock != &sched_spinlock) {
116 		k_spin_release(lock);
117 	}
118 	if (!is_spinlock || lock != &sched_spinlock) {
119 		(void) k_spin_lock(&sched_spinlock);
120 	}
121 
122 	new_thread = z_swap_next_thread();
123 
124 	if (new_thread != old_thread) {
125 		z_sched_usage_switch(new_thread);
126 
127 #ifdef CONFIG_SMP
128 		_current_cpu->swap_ok = 0;
129 		new_thread->base.cpu = arch_curr_cpu()->id;
130 
131 		if (!is_spinlock) {
132 			z_smp_release_global_lock(new_thread);
133 		}
134 #endif
135 		z_thread_mark_switched_out();
136 		z_sched_switch_spin(new_thread);
137 		_current_cpu->current = new_thread;
138 
139 #ifdef CONFIG_TIMESLICING
140 		z_reset_time_slice(new_thread);
141 #endif
142 
143 #ifdef CONFIG_SPIN_VALIDATE
144 		z_spin_lock_set_owner(&sched_spinlock);
145 #endif
146 
147 		arch_cohere_stacks(old_thread, NULL, new_thread);
148 
149 #ifdef CONFIG_SMP
150 		/* Now add _current back to the run queue, once we are
151 		 * guaranteed to reach the context switch in finite
152 		 * time.  See z_sched_switch_spin().
153 		 */
154 		z_requeue_current(old_thread);
155 #endif
156 		void *newsh = new_thread->switch_handle;
157 
158 		if (IS_ENABLED(CONFIG_SMP)) {
159 			/* Active threads must have a null here.  And
160 			 * it must be seen before the scheduler lock
161 			 * is released!
162 			 */
163 			new_thread->switch_handle = NULL;
164 			barrier_dmem_fence_full(); /* write barrier */
165 		}
166 		k_spin_release(&sched_spinlock);
167 		arch_switch(newsh, &old_thread->switch_handle);
168 	} else {
169 		k_spin_release(&sched_spinlock);
170 	}
171 
172 	if (is_spinlock) {
173 		arch_irq_unlock(key);
174 	} else {
175 		irq_unlock(key);
176 	}
177 
178 	return _current->swap_retval;
179 }
180 
z_swap_irqlock(unsigned int key)181 static inline int z_swap_irqlock(unsigned int key)
182 {
183 	return do_swap(key, NULL, false);
184 }
185 
z_swap(struct k_spinlock * lock,k_spinlock_key_t key)186 static inline int z_swap(struct k_spinlock *lock, k_spinlock_key_t key)
187 {
188 	return do_swap(key.key, lock, true);
189 }
190 
z_swap_unlocked(void)191 static inline void z_swap_unlocked(void)
192 {
193 	(void) do_swap(arch_irq_lock(), NULL, true);
194 }
195 
196 #else /* !CONFIG_USE_SWITCH */
197 
198 extern int arch_swap(unsigned int key);
199 
z_sched_switch_spin(struct k_thread * thread)200 static inline void z_sched_switch_spin(struct k_thread *thread)
201 {
202 	ARG_UNUSED(thread);
203 }
204 
z_swap_irqlock(unsigned int key)205 static inline int z_swap_irqlock(unsigned int key)
206 {
207 	int ret;
208 	z_check_stack_sentinel();
209 	ret = arch_swap(key);
210 	return ret;
211 }
212 
213 /* If !USE_SWITCH, then spinlocks are guaranteed degenerate as we
214  * can't be in SMP.  The k_spin_release() call is just for validation
215  * handling.
216  */
z_swap(struct k_spinlock * lock,k_spinlock_key_t key)217 static ALWAYS_INLINE int z_swap(struct k_spinlock *lock, k_spinlock_key_t key)
218 {
219 	k_spin_release(lock);
220 	return z_swap_irqlock(key.key);
221 }
222 
z_swap_unlocked(void)223 static inline void z_swap_unlocked(void)
224 {
225 	(void) z_swap_irqlock(arch_irq_lock());
226 }
227 
228 #endif /* !CONFIG_USE_SWITCH */
229 
230 /**
231  * Set up a "dummy" thread, used at early initialization to launch the
232  * first thread on a CPU.
233  *
234  * Needs to set enough fields such that the context switching code can
235  * use it to properly store state, which will just be discarded.
236  *
237  * The memory of the dummy thread can be completely uninitialized.
238  */
z_dummy_thread_init(struct k_thread * dummy_thread)239 static inline void z_dummy_thread_init(struct k_thread *dummy_thread)
240 {
241 	dummy_thread->base.thread_state = _THREAD_DUMMY;
242 #ifdef CONFIG_SCHED_CPU_MASK
243 	dummy_thread->base.cpu_mask = -1;
244 #endif
245 	dummy_thread->base.user_options = K_ESSENTIAL;
246 #ifdef CONFIG_THREAD_STACK_INFO
247 	dummy_thread->stack_info.start = 0U;
248 	dummy_thread->stack_info.size = 0U;
249 #endif
250 #ifdef CONFIG_USERSPACE
251 	dummy_thread->mem_domain_info.mem_domain = &k_mem_domain_default;
252 #endif
253 #if (CONFIG_HEAP_MEM_POOL_SIZE > 0)
254 	k_thread_system_pool_assign(dummy_thread);
255 #else
256 	dummy_thread->resource_pool = NULL;
257 #endif
258 
259 #ifdef CONFIG_TIMESLICE_PER_THREAD
260 	dummy_thread->base.slice_ticks = 0;
261 #endif
262 
263 	_current_cpu->current = dummy_thread;
264 }
265 #endif /* ZEPHYR_KERNEL_INCLUDE_KSWAP_H_ */
266