Rev 906 | Rev 1007 | Go to most recent revision | Only display areas with differences | Ignore whitespace | Details | Blame | Last modification | View Log | RSS feed
Rev 906 | Rev 907 | ||
---|---|---|---|
1 | /* |
1 | /* |
2 | * Copyright (C) 2001-2004 Jakub Jermar |
2 | * Copyright (C) 2001-2004 Jakub Jermar |
3 | * All rights reserved. |
3 | * All rights reserved. |
4 | * |
4 | * |
5 | * Redistribution and use in source and binary forms, with or without |
5 | * Redistribution and use in source and binary forms, with or without |
6 | * modification, are permitted provided that the following conditions |
6 | * modification, are permitted provided that the following conditions |
7 | * are met: |
7 | * are met: |
8 | * |
8 | * |
9 | * - Redistributions of source code must retain the above copyright |
9 | * - Redistributions of source code must retain the above copyright |
10 | * notice, this list of conditions and the following disclaimer. |
10 | * notice, this list of conditions and the following disclaimer. |
11 | * - Redistributions in binary form must reproduce the above copyright |
11 | * - Redistributions in binary form must reproduce the above copyright |
12 | * notice, this list of conditions and the following disclaimer in the |
12 | * notice, this list of conditions and the following disclaimer in the |
13 | * documentation and/or other materials provided with the distribution. |
13 | * documentation and/or other materials provided with the distribution. |
14 | * - The name of the author may not be used to endorse or promote products |
14 | * - The name of the author may not be used to endorse or promote products |
15 | * derived from this software without specific prior written permission. |
15 | * derived from this software without specific prior written permission. |
16 | * |
16 | * |
17 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR |
17 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR |
18 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES |
18 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES |
19 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. |
19 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. |
20 | * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, |
20 | * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, |
21 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT |
21 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT |
22 | * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
22 | * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
23 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
23 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
24 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
24 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
25 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF |
25 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF |
26 | * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
26 | * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
27 | */ |
27 | */ |
28 | 28 | ||
29 | #include <proc/scheduler.h> |
29 | #include <proc/scheduler.h> |
30 | #include <proc/thread.h> |
30 | #include <proc/thread.h> |
31 | #include <proc/task.h> |
31 | #include <proc/task.h> |
32 | #include <mm/frame.h> |
32 | #include <mm/frame.h> |
33 | #include <mm/page.h> |
33 | #include <mm/page.h> |
34 | #include <mm/as.h> |
34 | #include <mm/as.h> |
35 | #include <arch/asm.h> |
35 | #include <arch/asm.h> |
36 | #include <arch/faddr.h> |
36 | #include <arch/faddr.h> |
37 | #include <arch/atomic.h> |
37 | #include <arch/atomic.h> |
38 | #include <synch/spinlock.h> |
38 | #include <synch/spinlock.h> |
39 | #include <config.h> |
39 | #include <config.h> |
40 | #include <context.h> |
40 | #include <context.h> |
41 | #include <func.h> |
41 | #include <func.h> |
42 | #include <arch.h> |
42 | #include <arch.h> |
43 | #include <adt/list.h> |
43 | #include <adt/list.h> |
44 | #include <panic.h> |
44 | #include <panic.h> |
45 | #include <typedefs.h> |
45 | #include <typedefs.h> |
46 | #include <cpu.h> |
46 | #include <cpu.h> |
47 | #include <print.h> |
47 | #include <print.h> |
48 | #include <debug.h> |
48 | #include <debug.h> |
49 | 49 | ||
50 | static void scheduler_separated_stack(void); |
50 | static void scheduler_separated_stack(void); |
51 | 51 | ||
52 | atomic_t nrdy; /**< Number of ready threads in the system. */ |
52 | atomic_t nrdy; /**< Number of ready threads in the system. */ |
53 | 53 | ||
54 | /** Take actions before new thread runs. |
54 | /** Take actions before new thread runs. |
55 | * |
55 | * |
56 | * Perform actions that need to be |
56 | * Perform actions that need to be |
57 | * taken before the newly selected |
57 | * taken before the newly selected |
58 | * tread is passed control. |
58 | * tread is passed control. |
59 | * |
59 | * |
60 | * THREAD->lock is locked on entry |
60 | * THREAD->lock is locked on entry |
61 | * |
61 | * |
62 | */ |
62 | */ |
63 | void before_thread_runs(void) |
63 | void before_thread_runs(void) |
64 | { |
64 | { |
65 | before_thread_runs_arch(); |
65 | before_thread_runs_arch(); |
66 | #ifdef CONFIG_FPU_LAZY |
66 | #ifdef CONFIG_FPU_LAZY |
67 | if(THREAD==CPU->fpu_owner) |
67 | if(THREAD==CPU->fpu_owner) |
68 | fpu_enable(); |
68 | fpu_enable(); |
69 | else |
69 | else |
70 | fpu_disable(); |
70 | fpu_disable(); |
71 | #else |
71 | #else |
72 | fpu_enable(); |
72 | fpu_enable(); |
73 | if (THREAD->fpu_context_exists) |
73 | if (THREAD->fpu_context_exists) |
74 | fpu_context_restore(THREAD->saved_fpu_context); |
74 | fpu_context_restore(THREAD->saved_fpu_context); |
75 | else { |
75 | else { |
76 | fpu_init(); |
76 | fpu_init(); |
77 | THREAD->fpu_context_exists=1; |
77 | THREAD->fpu_context_exists=1; |
78 | } |
78 | } |
79 | #endif |
79 | #endif |
80 | } |
80 | } |
81 | 81 | ||
82 | /** Take actions after THREAD had run. |
82 | /** Take actions after THREAD had run. |
83 | * |
83 | * |
84 | * Perform actions that need to be |
84 | * Perform actions that need to be |
85 | * taken after the running thread |
85 | * taken after the running thread |
86 | * had been preempted by the scheduler. |
86 | * had been preempted by the scheduler. |
87 | * |
87 | * |
88 | * THREAD->lock is locked on entry |
88 | * THREAD->lock is locked on entry |
89 | * |
89 | * |
90 | */ |
90 | */ |
91 | void after_thread_ran(void) |
91 | void after_thread_ran(void) |
92 | { |
92 | { |
93 | after_thread_ran_arch(); |
93 | after_thread_ran_arch(); |
94 | } |
94 | } |
95 | 95 | ||
96 | #ifdef CONFIG_FPU_LAZY |
96 | #ifdef CONFIG_FPU_LAZY |
97 | void scheduler_fpu_lazy_request(void) |
97 | void scheduler_fpu_lazy_request(void) |
98 | { |
98 | { |
- | 99 | restart: |
|
99 | fpu_enable(); |
100 | fpu_enable(); |
100 | spinlock_lock(&CPU->lock); |
101 | spinlock_lock(&CPU->lock); |
101 | 102 | ||
102 | /* Save old context */ |
103 | /* Save old context */ |
103 | if (CPU->fpu_owner != NULL) { |
104 | if (CPU->fpu_owner != NULL) { |
104 | spinlock_lock(&CPU->fpu_owner->lock); |
105 | spinlock_lock(&CPU->fpu_owner->lock); |
105 | fpu_context_save(CPU->fpu_owner->saved_fpu_context); |
106 | fpu_context_save(CPU->fpu_owner->saved_fpu_context); |
106 | /* don't prevent migration */ |
107 | /* don't prevent migration */ |
107 | CPU->fpu_owner->fpu_context_engaged=0; |
108 | CPU->fpu_owner->fpu_context_engaged=0; |
108 | spinlock_unlock(&CPU->fpu_owner->lock); |
109 | spinlock_unlock(&CPU->fpu_owner->lock); |
- | 110 | CPU->fpu_owner = NULL; |
|
109 | } |
111 | } |
110 | 112 | ||
111 | spinlock_lock(&THREAD->lock); |
113 | spinlock_lock(&THREAD->lock); |
112 | if (THREAD->fpu_context_exists) { |
114 | if (THREAD->fpu_context_exists) { |
113 | fpu_context_restore(THREAD->saved_fpu_context); |
115 | fpu_context_restore(THREAD->saved_fpu_context); |
114 | } else { |
116 | } else { |
115 | /* Allocate FPU context */ |
117 | /* Allocate FPU context */ |
116 | if (!THREAD->saved_fpu_context) { |
118 | if (!THREAD->saved_fpu_context) { |
117 | /* Might sleep */ |
119 | /* Might sleep */ |
118 | spinlock_unlock(&THREAD->lock); |
120 | spinlock_unlock(&THREAD->lock); |
- | 121 | spinlock_unlock(&CPU->lock); |
|
119 | THREAD->saved_fpu_context = slab_alloc(fpu_context_slab, |
122 | THREAD->saved_fpu_context = slab_alloc(fpu_context_slab, |
120 | 0); |
123 | 0); |
121 | spinlock_lock(&THREAD->lock); |
124 | /* We may have switched CPUs during slab_alloc */ |
- | 125 | goto restart; |
|
122 | } |
126 | } |
123 | fpu_init(); |
127 | fpu_init(); |
124 | THREAD->fpu_context_exists=1; |
128 | THREAD->fpu_context_exists=1; |
125 | } |
129 | } |
126 | CPU->fpu_owner=THREAD; |
130 | CPU->fpu_owner=THREAD; |
127 | THREAD->fpu_context_engaged = 1; |
131 | THREAD->fpu_context_engaged = 1; |
128 | spinlock_unlock(&THREAD->lock); |
132 | spinlock_unlock(&THREAD->lock); |
129 | 133 | ||
130 | spinlock_unlock(&CPU->lock); |
134 | spinlock_unlock(&CPU->lock); |
131 | } |
135 | } |
132 | #endif |
136 | #endif |
133 | 137 | ||
134 | /** Initialize scheduler |
138 | /** Initialize scheduler |
135 | * |
139 | * |
136 | * Initialize kernel scheduler. |
140 | * Initialize kernel scheduler. |
137 | * |
141 | * |
138 | */ |
142 | */ |
139 | void scheduler_init(void) |
143 | void scheduler_init(void) |
140 | { |
144 | { |
141 | } |
145 | } |
142 | 146 | ||
143 | /** Get thread to be scheduled |
147 | /** Get thread to be scheduled |
144 | * |
148 | * |
145 | * Get the optimal thread to be scheduled |
149 | * Get the optimal thread to be scheduled |
146 | * according to thread accounting and scheduler |
150 | * according to thread accounting and scheduler |
147 | * policy. |
151 | * policy. |
148 | * |
152 | * |
149 | * @return Thread to be scheduled. |
153 | * @return Thread to be scheduled. |
150 | * |
154 | * |
151 | */ |
155 | */ |
152 | static thread_t *find_best_thread(void) |
156 | static thread_t *find_best_thread(void) |
153 | { |
157 | { |
154 | thread_t *t; |
158 | thread_t *t; |
155 | runq_t *r; |
159 | runq_t *r; |
156 | int i; |
160 | int i; |
157 | 161 | ||
158 | ASSERT(CPU != NULL); |
162 | ASSERT(CPU != NULL); |
159 | 163 | ||
160 | loop: |
164 | loop: |
161 | interrupts_enable(); |
165 | interrupts_enable(); |
162 | 166 | ||
163 | if (atomic_get(&CPU->nrdy) == 0) { |
167 | if (atomic_get(&CPU->nrdy) == 0) { |
164 | /* |
168 | /* |
165 | * For there was nothing to run, the CPU goes to sleep |
169 | * For there was nothing to run, the CPU goes to sleep |
166 | * until a hardware interrupt or an IPI comes. |
170 | * until a hardware interrupt or an IPI comes. |
167 | * This improves energy saving and hyperthreading. |
171 | * This improves energy saving and hyperthreading. |
168 | */ |
172 | */ |
169 | 173 | ||
170 | /* |
174 | /* |
171 | * An interrupt might occur right now and wake up a thread. |
175 | * An interrupt might occur right now and wake up a thread. |
172 | * In such case, the CPU will continue to go to sleep |
176 | * In such case, the CPU will continue to go to sleep |
173 | * even though there is a runnable thread. |
177 | * even though there is a runnable thread. |
174 | */ |
178 | */ |
175 | 179 | ||
176 | cpu_sleep(); |
180 | cpu_sleep(); |
177 | goto loop; |
181 | goto loop; |
178 | } |
182 | } |
179 | 183 | ||
180 | interrupts_disable(); |
184 | interrupts_disable(); |
181 | 185 | ||
182 | for (i = 0; i<RQ_COUNT; i++) { |
186 | for (i = 0; i<RQ_COUNT; i++) { |
183 | r = &CPU->rq[i]; |
187 | r = &CPU->rq[i]; |
184 | spinlock_lock(&r->lock); |
188 | spinlock_lock(&r->lock); |
185 | if (r->n == 0) { |
189 | if (r->n == 0) { |
186 | /* |
190 | /* |
187 | * If this queue is empty, try a lower-priority queue. |
191 | * If this queue is empty, try a lower-priority queue. |
188 | */ |
192 | */ |
189 | spinlock_unlock(&r->lock); |
193 | spinlock_unlock(&r->lock); |
190 | continue; |
194 | continue; |
191 | } |
195 | } |
192 | 196 | ||
193 | atomic_dec(&CPU->nrdy); |
197 | atomic_dec(&CPU->nrdy); |
194 | atomic_dec(&nrdy); |
198 | atomic_dec(&nrdy); |
195 | r->n--; |
199 | r->n--; |
196 | 200 | ||
197 | /* |
201 | /* |
198 | * Take the first thread from the queue. |
202 | * Take the first thread from the queue. |
199 | */ |
203 | */ |
200 | t = list_get_instance(r->rq_head.next, thread_t, rq_link); |
204 | t = list_get_instance(r->rq_head.next, thread_t, rq_link); |
201 | list_remove(&t->rq_link); |
205 | list_remove(&t->rq_link); |
202 | 206 | ||
203 | spinlock_unlock(&r->lock); |
207 | spinlock_unlock(&r->lock); |
204 | 208 | ||
205 | spinlock_lock(&t->lock); |
209 | spinlock_lock(&t->lock); |
206 | t->cpu = CPU; |
210 | t->cpu = CPU; |
207 | 211 | ||
208 | t->ticks = us2ticks((i+1)*10000); |
212 | t->ticks = us2ticks((i+1)*10000); |
209 | t->priority = i; /* correct rq index */ |
213 | t->priority = i; /* correct rq index */ |
210 | 214 | ||
211 | /* |
215 | /* |
212 | * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge. |
216 | * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge. |
213 | */ |
217 | */ |
214 | t->flags &= ~X_STOLEN; |
218 | t->flags &= ~X_STOLEN; |
215 | spinlock_unlock(&t->lock); |
219 | spinlock_unlock(&t->lock); |
216 | 220 | ||
217 | return t; |
221 | return t; |
218 | } |
222 | } |
219 | goto loop; |
223 | goto loop; |
220 | 224 | ||
221 | } |
225 | } |
222 | 226 | ||
223 | /** Prevent rq starvation |
227 | /** Prevent rq starvation |
224 | * |
228 | * |
225 | * Prevent low priority threads from starving in rq's. |
229 | * Prevent low priority threads from starving in rq's. |
226 | * |
230 | * |
227 | * When the function decides to relink rq's, it reconnects |
231 | * When the function decides to relink rq's, it reconnects |
228 | * respective pointers so that in result threads with 'pri' |
232 | * respective pointers so that in result threads with 'pri' |
229 | * greater or equal 'start' are moved to a higher-priority queue. |
233 | * greater or equal 'start' are moved to a higher-priority queue. |
230 | * |
234 | * |
231 | * @param start Threshold priority. |
235 | * @param start Threshold priority. |
232 | * |
236 | * |
233 | */ |
237 | */ |
234 | static void relink_rq(int start) |
238 | static void relink_rq(int start) |
235 | { |
239 | { |
236 | link_t head; |
240 | link_t head; |
237 | runq_t *r; |
241 | runq_t *r; |
238 | int i, n; |
242 | int i, n; |
239 | 243 | ||
240 | list_initialize(&head); |
244 | list_initialize(&head); |
241 | spinlock_lock(&CPU->lock); |
245 | spinlock_lock(&CPU->lock); |
242 | if (CPU->needs_relink > NEEDS_RELINK_MAX) { |
246 | if (CPU->needs_relink > NEEDS_RELINK_MAX) { |
243 | for (i = start; i<RQ_COUNT-1; i++) { |
247 | for (i = start; i<RQ_COUNT-1; i++) { |
244 | /* remember and empty rq[i + 1] */ |
248 | /* remember and empty rq[i + 1] */ |
245 | r = &CPU->rq[i + 1]; |
249 | r = &CPU->rq[i + 1]; |
246 | spinlock_lock(&r->lock); |
250 | spinlock_lock(&r->lock); |
247 | list_concat(&head, &r->rq_head); |
251 | list_concat(&head, &r->rq_head); |
248 | n = r->n; |
252 | n = r->n; |
249 | r->n = 0; |
253 | r->n = 0; |
250 | spinlock_unlock(&r->lock); |
254 | spinlock_unlock(&r->lock); |
251 | 255 | ||
252 | /* append rq[i + 1] to rq[i] */ |
256 | /* append rq[i + 1] to rq[i] */ |
253 | r = &CPU->rq[i]; |
257 | r = &CPU->rq[i]; |
254 | spinlock_lock(&r->lock); |
258 | spinlock_lock(&r->lock); |
255 | list_concat(&r->rq_head, &head); |
259 | list_concat(&r->rq_head, &head); |
256 | r->n += n; |
260 | r->n += n; |
257 | spinlock_unlock(&r->lock); |
261 | spinlock_unlock(&r->lock); |
258 | } |
262 | } |
259 | CPU->needs_relink = 0; |
263 | CPU->needs_relink = 0; |
260 | } |
264 | } |
261 | spinlock_unlock(&CPU->lock); |
265 | spinlock_unlock(&CPU->lock); |
262 | 266 | ||
263 | } |
267 | } |
264 | 268 | ||
265 | /** The scheduler |
269 | /** The scheduler |
266 | * |
270 | * |
267 | * The thread scheduling procedure. |
271 | * The thread scheduling procedure. |
268 | * Passes control directly to |
272 | * Passes control directly to |
269 | * scheduler_separated_stack(). |
273 | * scheduler_separated_stack(). |
270 | * |
274 | * |
271 | */ |
275 | */ |
272 | void scheduler(void) |
276 | void scheduler(void) |
273 | { |
277 | { |
274 | volatile ipl_t ipl; |
278 | volatile ipl_t ipl; |
275 | 279 | ||
276 | ASSERT(CPU != NULL); |
280 | ASSERT(CPU != NULL); |
277 | 281 | ||
278 | ipl = interrupts_disable(); |
282 | ipl = interrupts_disable(); |
279 | 283 | ||
280 | if (atomic_get(&haltstate)) |
284 | if (atomic_get(&haltstate)) |
281 | halt(); |
285 | halt(); |
282 | 286 | ||
283 | if (THREAD) { |
287 | if (THREAD) { |
284 | spinlock_lock(&THREAD->lock); |
288 | spinlock_lock(&THREAD->lock); |
285 | #ifndef CONFIG_FPU_LAZY |
289 | #ifndef CONFIG_FPU_LAZY |
286 | fpu_context_save(THREAD->saved_fpu_context); |
290 | fpu_context_save(THREAD->saved_fpu_context); |
287 | #endif |
291 | #endif |
288 | if (!context_save(&THREAD->saved_context)) { |
292 | if (!context_save(&THREAD->saved_context)) { |
289 | /* |
293 | /* |
290 | * This is the place where threads leave scheduler(); |
294 | * This is the place where threads leave scheduler(); |
291 | */ |
295 | */ |
292 | spinlock_unlock(&THREAD->lock); |
296 | spinlock_unlock(&THREAD->lock); |
293 | interrupts_restore(THREAD->saved_context.ipl); |
297 | interrupts_restore(THREAD->saved_context.ipl); |
294 | return; |
298 | return; |
295 | } |
299 | } |
296 | 300 | ||
297 | /* |
301 | /* |
298 | * Interrupt priority level of preempted thread is recorded here |
302 | * Interrupt priority level of preempted thread is recorded here |
299 | * to facilitate scheduler() invocations from interrupts_disable()'d |
303 | * to facilitate scheduler() invocations from interrupts_disable()'d |
300 | * code (e.g. waitq_sleep_timeout()). |
304 | * code (e.g. waitq_sleep_timeout()). |
301 | */ |
305 | */ |
302 | THREAD->saved_context.ipl = ipl; |
306 | THREAD->saved_context.ipl = ipl; |
303 | } |
307 | } |
304 | 308 | ||
305 | /* |
309 | /* |
306 | * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM |
310 | * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM |
307 | * and preemption counter. At this point THE could be coming either |
311 | * and preemption counter. At this point THE could be coming either |
308 | * from THREAD's or CPU's stack. |
312 | * from THREAD's or CPU's stack. |
309 | */ |
313 | */ |
310 | the_copy(THE, (the_t *) CPU->stack); |
314 | the_copy(THE, (the_t *) CPU->stack); |
311 | 315 | ||
312 | /* |
316 | /* |
313 | * We may not keep the old stack. |
317 | * We may not keep the old stack. |
314 | * Reason: If we kept the old stack and got blocked, for instance, in |
318 | * Reason: If we kept the old stack and got blocked, for instance, in |
315 | * find_best_thread(), the old thread could get rescheduled by another |
319 | * find_best_thread(), the old thread could get rescheduled by another |
316 | * CPU and overwrite the part of its own stack that was also used by |
320 | * CPU and overwrite the part of its own stack that was also used by |
317 | * the scheduler on this CPU. |
321 | * the scheduler on this CPU. |
318 | * |
322 | * |
319 | * Moreover, we have to bypass the compiler-generated POP sequence |
323 | * Moreover, we have to bypass the compiler-generated POP sequence |
320 | * which is fooled by SP being set to the very top of the stack. |
324 | * which is fooled by SP being set to the very top of the stack. |
321 | * Therefore the scheduler() function continues in |
325 | * Therefore the scheduler() function continues in |
322 | * scheduler_separated_stack(). |
326 | * scheduler_separated_stack(). |
323 | */ |
327 | */ |
324 | context_save(&CPU->saved_context); |
328 | context_save(&CPU->saved_context); |
325 | context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), (__address) CPU->stack, CPU_STACK_SIZE); |
329 | context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), (__address) CPU->stack, CPU_STACK_SIZE); |
326 | context_restore(&CPU->saved_context); |
330 | context_restore(&CPU->saved_context); |
327 | /* not reached */ |
331 | /* not reached */ |
328 | } |
332 | } |
329 | 333 | ||
330 | /** Scheduler stack switch wrapper |
334 | /** Scheduler stack switch wrapper |
331 | * |
335 | * |
332 | * Second part of the scheduler() function |
336 | * Second part of the scheduler() function |
333 | * using new stack. Handling the actual context |
337 | * using new stack. Handling the actual context |
334 | * switch to a new thread. |
338 | * switch to a new thread. |
335 | * |
339 | * |
336 | * Assume THREAD->lock is held. |
340 | * Assume THREAD->lock is held. |
337 | */ |
341 | */ |
338 | void scheduler_separated_stack(void) |
342 | void scheduler_separated_stack(void) |
339 | { |
343 | { |
340 | int priority; |
344 | int priority; |
341 | 345 | ||
342 | ASSERT(CPU != NULL); |
346 | ASSERT(CPU != NULL); |
343 | 347 | ||
344 | if (THREAD) { |
348 | if (THREAD) { |
345 | /* must be run after the switch to scheduler stack */ |
349 | /* must be run after the switch to scheduler stack */ |
346 | after_thread_ran(); |
350 | after_thread_ran(); |
347 | 351 | ||
348 | switch (THREAD->state) { |
352 | switch (THREAD->state) { |
349 | case Running: |
353 | case Running: |
350 | THREAD->state = Ready; |
354 | THREAD->state = Ready; |
351 | spinlock_unlock(&THREAD->lock); |
355 | spinlock_unlock(&THREAD->lock); |
352 | thread_ready(THREAD); |
356 | thread_ready(THREAD); |
353 | break; |
357 | break; |
354 | 358 | ||
355 | case Exiting: |
359 | case Exiting: |
356 | thread_destroy(THREAD); |
360 | thread_destroy(THREAD); |
357 | break; |
361 | break; |
358 | 362 | ||
359 | case Sleeping: |
363 | case Sleeping: |
360 | /* |
364 | /* |
361 | * Prefer the thread after it's woken up. |
365 | * Prefer the thread after it's woken up. |
362 | */ |
366 | */ |
363 | THREAD->priority = -1; |
367 | THREAD->priority = -1; |
364 | 368 | ||
365 | /* |
369 | /* |
366 | * We need to release wq->lock which we locked in waitq_sleep(). |
370 | * We need to release wq->lock which we locked in waitq_sleep(). |
367 | * Address of wq->lock is kept in THREAD->sleep_queue. |
371 | * Address of wq->lock is kept in THREAD->sleep_queue. |
368 | */ |
372 | */ |
369 | spinlock_unlock(&THREAD->sleep_queue->lock); |
373 | spinlock_unlock(&THREAD->sleep_queue->lock); |
370 | 374 | ||
371 | /* |
375 | /* |
372 | * Check for possible requests for out-of-context invocation. |
376 | * Check for possible requests for out-of-context invocation. |
373 | */ |
377 | */ |
374 | if (THREAD->call_me) { |
378 | if (THREAD->call_me) { |
375 | THREAD->call_me(THREAD->call_me_with); |
379 | THREAD->call_me(THREAD->call_me_with); |
376 | THREAD->call_me = NULL; |
380 | THREAD->call_me = NULL; |
377 | THREAD->call_me_with = NULL; |
381 | THREAD->call_me_with = NULL; |
378 | } |
382 | } |
379 | 383 | ||
380 | spinlock_unlock(&THREAD->lock); |
384 | spinlock_unlock(&THREAD->lock); |
381 | 385 | ||
382 | break; |
386 | break; |
383 | 387 | ||
384 | default: |
388 | default: |
385 | /* |
389 | /* |
386 | * Entering state is unexpected. |
390 | * Entering state is unexpected. |
387 | */ |
391 | */ |
388 | panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]); |
392 | panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]); |
389 | break; |
393 | break; |
390 | } |
394 | } |
391 | 395 | ||
392 | THREAD = NULL; |
396 | THREAD = NULL; |
393 | } |
397 | } |
394 | 398 | ||
395 | THREAD = find_best_thread(); |
399 | THREAD = find_best_thread(); |
396 | 400 | ||
397 | spinlock_lock(&THREAD->lock); |
401 | spinlock_lock(&THREAD->lock); |
398 | priority = THREAD->priority; |
402 | priority = THREAD->priority; |
399 | spinlock_unlock(&THREAD->lock); |
403 | spinlock_unlock(&THREAD->lock); |
400 | 404 | ||
401 | relink_rq(priority); |
405 | relink_rq(priority); |
402 | 406 | ||
403 | spinlock_lock(&THREAD->lock); |
407 | spinlock_lock(&THREAD->lock); |
404 | 408 | ||
405 | /* |
409 | /* |
406 | * If both the old and the new task are the same, lots of work is avoided. |
410 | * If both the old and the new task are the same, lots of work is avoided. |
407 | */ |
411 | */ |
408 | if (TASK != THREAD->task) { |
412 | if (TASK != THREAD->task) { |
409 | as_t *as1 = NULL; |
413 | as_t *as1 = NULL; |
410 | as_t *as2; |
414 | as_t *as2; |
411 | 415 | ||
412 | if (TASK) { |
416 | if (TASK) { |
413 | spinlock_lock(&TASK->lock); |
417 | spinlock_lock(&TASK->lock); |
414 | as1 = TASK->as; |
418 | as1 = TASK->as; |
415 | spinlock_unlock(&TASK->lock); |
419 | spinlock_unlock(&TASK->lock); |
416 | } |
420 | } |
417 | 421 | ||
418 | spinlock_lock(&THREAD->task->lock); |
422 | spinlock_lock(&THREAD->task->lock); |
419 | as2 = THREAD->task->as; |
423 | as2 = THREAD->task->as; |
420 | spinlock_unlock(&THREAD->task->lock); |
424 | spinlock_unlock(&THREAD->task->lock); |
421 | 425 | ||
422 | /* |
426 | /* |
423 | * Note that it is possible for two tasks to share one address space. |
427 | * Note that it is possible for two tasks to share one address space. |
424 | */ |
428 | */ |
425 | if (as1 != as2) { |
429 | if (as1 != as2) { |
426 | /* |
430 | /* |
427 | * Both tasks and address spaces are different. |
431 | * Both tasks and address spaces are different. |
428 | * Replace the old one with the new one. |
432 | * Replace the old one with the new one. |
429 | */ |
433 | */ |
430 | as_switch(as1, as2); |
434 | as_switch(as1, as2); |
431 | } |
435 | } |
432 | TASK = THREAD->task; |
436 | TASK = THREAD->task; |
433 | } |
437 | } |
434 | 438 | ||
435 | THREAD->state = Running; |
439 | THREAD->state = Running; |
436 | 440 | ||
437 | #ifdef SCHEDULER_VERBOSE |
441 | #ifdef SCHEDULER_VERBOSE |
438 | printf("cpu%d: tid %d (priority=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks, atomic_get(&CPU->nrdy)); |
442 | printf("cpu%d: tid %d (priority=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks, atomic_get(&CPU->nrdy)); |
439 | #endif |
443 | #endif |
440 | 444 | ||
441 | /* |
445 | /* |
442 | * Some architectures provide late kernel PA2KA(identity) |
446 | * Some architectures provide late kernel PA2KA(identity) |
443 | * mapping in a page fault handler. However, the page fault |
447 | * mapping in a page fault handler. However, the page fault |
444 | * handler uses the kernel stack of the running thread and |
448 | * handler uses the kernel stack of the running thread and |
445 | * therefore cannot be used to map it. The kernel stack, if |
449 | * therefore cannot be used to map it. The kernel stack, if |
446 | * necessary, is to be mapped in before_thread_runs(). This |
450 | * necessary, is to be mapped in before_thread_runs(). This |
447 | * function must be executed before the switch to the new stack. |
451 | * function must be executed before the switch to the new stack. |
448 | */ |
452 | */ |
449 | before_thread_runs(); |
453 | before_thread_runs(); |
450 | 454 | ||
451 | /* |
455 | /* |
452 | * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack. |
456 | * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack. |
453 | */ |
457 | */ |
454 | the_copy(THE, (the_t *) THREAD->kstack); |
458 | the_copy(THE, (the_t *) THREAD->kstack); |
455 | 459 | ||
456 | context_restore(&THREAD->saved_context); |
460 | context_restore(&THREAD->saved_context); |
457 | /* not reached */ |
461 | /* not reached */ |
458 | } |
462 | } |
459 | 463 | ||
460 | #ifdef CONFIG_SMP |
464 | #ifdef CONFIG_SMP |
461 | /** Load balancing thread |
465 | /** Load balancing thread |
462 | * |
466 | * |
463 | * SMP load balancing thread, supervising thread supplies |
467 | * SMP load balancing thread, supervising thread supplies |
464 | * for the CPU it's wired to. |
468 | * for the CPU it's wired to. |
465 | * |
469 | * |
466 | * @param arg Generic thread argument (unused). |
470 | * @param arg Generic thread argument (unused). |
467 | * |
471 | * |
468 | */ |
472 | */ |
469 | void kcpulb(void *arg) |
473 | void kcpulb(void *arg) |
470 | { |
474 | { |
471 | thread_t *t; |
475 | thread_t *t; |
472 | int count, average, i, j, k = 0; |
476 | int count, average, i, j, k = 0; |
473 | ipl_t ipl; |
477 | ipl_t ipl; |
474 | 478 | ||
475 | loop: |
479 | loop: |
476 | /* |
480 | /* |
477 | * Work in 1s intervals. |
481 | * Work in 1s intervals. |
478 | */ |
482 | */ |
479 | thread_sleep(1); |
483 | thread_sleep(1); |
480 | 484 | ||
481 | not_satisfied: |
485 | not_satisfied: |
482 | /* |
486 | /* |
483 | * Calculate the number of threads that will be migrated/stolen from |
487 | * Calculate the number of threads that will be migrated/stolen from |
484 | * other CPU's. Note that situation can have changed between two |
488 | * other CPU's. Note that situation can have changed between two |
485 | * passes. Each time get the most up to date counts. |
489 | * passes. Each time get the most up to date counts. |
486 | */ |
490 | */ |
487 | average = atomic_get(&nrdy) / config.cpu_active + 1; |
491 | average = atomic_get(&nrdy) / config.cpu_active + 1; |
488 | count = average - atomic_get(&CPU->nrdy); |
492 | count = average - atomic_get(&CPU->nrdy); |
489 | 493 | ||
490 | if (count <= 0) |
494 | if (count <= 0) |
491 | goto satisfied; |
495 | goto satisfied; |
492 | 496 | ||
493 | /* |
497 | /* |
494 | * Searching least priority queues on all CPU's first and most priority queues on all CPU's last. |
498 | * Searching least priority queues on all CPU's first and most priority queues on all CPU's last. |
495 | */ |
499 | */ |
496 | for (j=RQ_COUNT-1; j >= 0; j--) { |
500 | for (j=RQ_COUNT-1; j >= 0; j--) { |
497 | for (i=0; i < config.cpu_active; i++) { |
501 | for (i=0; i < config.cpu_active; i++) { |
498 | link_t *l; |
502 | link_t *l; |
499 | runq_t *r; |
503 | runq_t *r; |
500 | cpu_t *cpu; |
504 | cpu_t *cpu; |
501 | 505 | ||
502 | cpu = &cpus[(i + k) % config.cpu_active]; |
506 | cpu = &cpus[(i + k) % config.cpu_active]; |
503 | 507 | ||
504 | /* |
508 | /* |
505 | * Not interested in ourselves. |
509 | * Not interested in ourselves. |
506 | * Doesn't require interrupt disabling for kcpulb is X_WIRED. |
510 | * Doesn't require interrupt disabling for kcpulb is X_WIRED. |
507 | */ |
511 | */ |
508 | if (CPU == cpu) |
512 | if (CPU == cpu) |
509 | continue; |
513 | continue; |
510 | if (atomic_get(&cpu->nrdy) <= average) |
514 | if (atomic_get(&cpu->nrdy) <= average) |
511 | continue; |
515 | continue; |
512 | 516 | ||
513 | ipl = interrupts_disable(); |
517 | ipl = interrupts_disable(); |
514 | r = &cpu->rq[j]; |
518 | r = &cpu->rq[j]; |
515 | spinlock_lock(&r->lock); |
519 | spinlock_lock(&r->lock); |
516 | if (r->n == 0) { |
520 | if (r->n == 0) { |
517 | spinlock_unlock(&r->lock); |
521 | spinlock_unlock(&r->lock); |
518 | interrupts_restore(ipl); |
522 | interrupts_restore(ipl); |
519 | continue; |
523 | continue; |
520 | } |
524 | } |
521 | 525 | ||
522 | t = NULL; |
526 | t = NULL; |
523 | l = r->rq_head.prev; /* search rq from the back */ |
527 | l = r->rq_head.prev; /* search rq from the back */ |
524 | while (l != &r->rq_head) { |
528 | while (l != &r->rq_head) { |
525 | t = list_get_instance(l, thread_t, rq_link); |
529 | t = list_get_instance(l, thread_t, rq_link); |
526 | /* |
530 | /* |
527 | * We don't want to steal CPU-wired threads neither threads already stolen. |
531 | * We don't want to steal CPU-wired threads neither threads already stolen. |
528 | * The latter prevents threads from migrating between CPU's without ever being run. |
532 | * The latter prevents threads from migrating between CPU's without ever being run. |
529 | * We don't want to steal threads whose FPU context is still in CPU. |
533 | * We don't want to steal threads whose FPU context is still in CPU. |
530 | */ |
534 | */ |
531 | spinlock_lock(&t->lock); |
535 | spinlock_lock(&t->lock); |
532 | if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) { |
536 | if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) { |
533 | /* |
537 | /* |
534 | * Remove t from r. |
538 | * Remove t from r. |
535 | */ |
539 | */ |
536 | spinlock_unlock(&t->lock); |
540 | spinlock_unlock(&t->lock); |
537 | 541 | ||
538 | atomic_dec(&cpu->nrdy); |
542 | atomic_dec(&cpu->nrdy); |
539 | atomic_dec(&nrdy); |
543 | atomic_dec(&nrdy); |
540 | 544 | ||
541 | r->n--; |
545 | r->n--; |
542 | list_remove(&t->rq_link); |
546 | list_remove(&t->rq_link); |
543 | 547 | ||
544 | break; |
548 | break; |
545 | } |
549 | } |
546 | spinlock_unlock(&t->lock); |
550 | spinlock_unlock(&t->lock); |
547 | l = l->prev; |
551 | l = l->prev; |
548 | t = NULL; |
552 | t = NULL; |
549 | } |
553 | } |
550 | spinlock_unlock(&r->lock); |
554 | spinlock_unlock(&r->lock); |
551 | 555 | ||
552 | if (t) { |
556 | if (t) { |
553 | /* |
557 | /* |
554 | * Ready t on local CPU |
558 | * Ready t on local CPU |
555 | */ |
559 | */ |
556 | spinlock_lock(&t->lock); |
560 | spinlock_lock(&t->lock); |
557 | #ifdef KCPULB_VERBOSE |
561 | #ifdef KCPULB_VERBOSE |
558 | printf("kcpulb%d: TID %d -> cpu%d, nrdy=%d, avg=%d\n", CPU->id, t->tid, CPU->id, atomic_get(&CPU->nrdy), atomic_get(&nrdy) / config.cpu_active); |
562 | printf("kcpulb%d: TID %d -> cpu%d, nrdy=%d, avg=%d\n", CPU->id, t->tid, CPU->id, atomic_get(&CPU->nrdy), atomic_get(&nrdy) / config.cpu_active); |
559 | #endif |
563 | #endif |
560 | t->flags |= X_STOLEN; |
564 | t->flags |= X_STOLEN; |
561 | spinlock_unlock(&t->lock); |
565 | spinlock_unlock(&t->lock); |
562 | 566 | ||
563 | thread_ready(t); |
567 | thread_ready(t); |
564 | 568 | ||
565 | interrupts_restore(ipl); |
569 | interrupts_restore(ipl); |
566 | 570 | ||
567 | if (--count == 0) |
571 | if (--count == 0) |
568 | goto satisfied; |
572 | goto satisfied; |
569 | 573 | ||
570 | /* |
574 | /* |
571 | * We are not satisfied yet, focus on another CPU next time. |
575 | * We are not satisfied yet, focus on another CPU next time. |
572 | */ |
576 | */ |
573 | k++; |
577 | k++; |
574 | 578 | ||
575 | continue; |
579 | continue; |
576 | } |
580 | } |
577 | interrupts_restore(ipl); |
581 | interrupts_restore(ipl); |
578 | } |
582 | } |
579 | } |
583 | } |
580 | 584 | ||
581 | if (atomic_get(&CPU->nrdy)) { |
585 | if (atomic_get(&CPU->nrdy)) { |
582 | /* |
586 | /* |
583 | * Be a little bit light-weight and let migrated threads run. |
587 | * Be a little bit light-weight and let migrated threads run. |
584 | */ |
588 | */ |
585 | scheduler(); |
589 | scheduler(); |
586 | } else { |
590 | } else { |
587 | /* |
591 | /* |
588 | * We failed to migrate a single thread. |
592 | * We failed to migrate a single thread. |
589 | * Give up this turn. |
593 | * Give up this turn. |
590 | */ |
594 | */ |
591 | goto loop; |
595 | goto loop; |
592 | } |
596 | } |
593 | 597 | ||
594 | goto not_satisfied; |
598 | goto not_satisfied; |
595 | 599 | ||
596 | satisfied: |
600 | satisfied: |
597 | goto loop; |
601 | goto loop; |
598 | } |
602 | } |
599 | 603 | ||
600 | #endif /* CONFIG_SMP */ |
604 | #endif /* CONFIG_SMP */ |
601 | 605 | ||
602 | 606 | ||
603 | /** Print information about threads & scheduler queues */ |
607 | /** Print information about threads & scheduler queues */ |
604 | void sched_print_list(void) |
608 | void sched_print_list(void) |
605 | { |
609 | { |
606 | ipl_t ipl; |
610 | ipl_t ipl; |
607 | int cpu,i; |
611 | int cpu,i; |
608 | runq_t *r; |
612 | runq_t *r; |
609 | thread_t *t; |
613 | thread_t *t; |
610 | link_t *cur; |
614 | link_t *cur; |
611 | 615 | ||
612 | /* We are going to mess with scheduler structures, |
616 | /* We are going to mess with scheduler structures, |
613 | * let's not be interrupted */ |
617 | * let's not be interrupted */ |
614 | ipl = interrupts_disable(); |
618 | ipl = interrupts_disable(); |
615 | printf("Scheduler dump:\n"); |
619 | printf("Scheduler dump:\n"); |
616 | for (cpu=0;cpu < config.cpu_count; cpu++) { |
620 | for (cpu=0;cpu < config.cpu_count; cpu++) { |
617 | 621 | ||
618 | if (!cpus[cpu].active) |
622 | if (!cpus[cpu].active) |
619 | continue; |
623 | continue; |
620 | 624 | ||
621 | spinlock_lock(&cpus[cpu].lock); |
625 | spinlock_lock(&cpus[cpu].lock); |
622 | printf("cpu%d: nrdy: %d, needs_relink: %d\n", |
626 | printf("cpu%d: nrdy: %d, needs_relink: %d\n", |
623 | cpus[cpu].id, atomic_get(&cpus[cpu].nrdy), cpus[cpu].needs_relink); |
627 | cpus[cpu].id, atomic_get(&cpus[cpu].nrdy), cpus[cpu].needs_relink); |
624 | 628 | ||
625 | for (i=0; i<RQ_COUNT; i++) { |
629 | for (i=0; i<RQ_COUNT; i++) { |
626 | r = &cpus[cpu].rq[i]; |
630 | r = &cpus[cpu].rq[i]; |
627 | spinlock_lock(&r->lock); |
631 | spinlock_lock(&r->lock); |
628 | if (!r->n) { |
632 | if (!r->n) { |
629 | spinlock_unlock(&r->lock); |
633 | spinlock_unlock(&r->lock); |
630 | continue; |
634 | continue; |
631 | } |
635 | } |
632 | printf("\trq[%d]: ", i); |
636 | printf("\trq[%d]: ", i); |
633 | for (cur=r->rq_head.next; cur!=&r->rq_head; cur=cur->next) { |
637 | for (cur=r->rq_head.next; cur!=&r->rq_head; cur=cur->next) { |
634 | t = list_get_instance(cur, thread_t, rq_link); |
638 | t = list_get_instance(cur, thread_t, rq_link); |
635 | printf("%d(%s) ", t->tid, |
639 | printf("%d(%s) ", t->tid, |
636 | thread_states[t->state]); |
640 | thread_states[t->state]); |
637 | } |
641 | } |
638 | printf("\n"); |
642 | printf("\n"); |
639 | spinlock_unlock(&r->lock); |
643 | spinlock_unlock(&r->lock); |
640 | } |
644 | } |
641 | spinlock_unlock(&cpus[cpu].lock); |
645 | spinlock_unlock(&cpus[cpu].lock); |
642 | } |
646 | } |
643 | 647 | ||
644 | interrupts_restore(ipl); |
648 | interrupts_restore(ipl); |
645 | } |
649 | } |
646 | 650 |