Subversion Repositories HelenOS-historic

Rev

Rev 1702 | Rev 1708 | Go to most recent revision | Only display areas with differences | Ignore whitespace | Details | Blame | Last modification | View Log | RSS feed

Rev 1702 Rev 1705
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
 /**
-
 
30
 * @defgroup proc Proc
-
 
31
 * @ingroup kernel
-
 
32
 * @{
-
 
33
 * @}
-
 
34
 */
-
 
35
 
29
 
36
 /** @addtogroup genericproc generic
30
 /** @addtogroup genericproc
37
 * @ingroup proc
-
 
38
 * @{
31
 * @{
39
 */
32
 */
40
 
33
 
41
/**
34
/**
42
 * @file
35
 * @file
43
 * @brief   Scheduler and load balancing.
36
 * @brief   Scheduler and load balancing.
44
 *
37
 *
45
 * This file contains the scheduler and kcpulb kernel thread which
38
 * This file contains the scheduler and kcpulb kernel thread which
46
 * performs load-balancing of per-CPU run queues.
39
 * performs load-balancing of per-CPU run queues.
47
 */
40
 */
48
 
41
 
49
#include <proc/scheduler.h>
42
#include <proc/scheduler.h>
50
#include <proc/thread.h>
43
#include <proc/thread.h>
51
#include <proc/task.h>
44
#include <proc/task.h>
52
#include <mm/frame.h>
45
#include <mm/frame.h>
53
#include <mm/page.h>
46
#include <mm/page.h>
54
#include <mm/as.h>
47
#include <mm/as.h>
55
#include <time/delay.h>
48
#include <time/delay.h>
56
#include <arch/asm.h>
49
#include <arch/asm.h>
57
#include <arch/faddr.h>
50
#include <arch/faddr.h>
58
#include <atomic.h>
51
#include <atomic.h>
59
#include <synch/spinlock.h>
52
#include <synch/spinlock.h>
60
#include <config.h>
53
#include <config.h>
61
#include <context.h>
54
#include <context.h>
62
#include <func.h>
55
#include <func.h>
63
#include <arch.h>
56
#include <arch.h>
64
#include <adt/list.h>
57
#include <adt/list.h>
65
#include <panic.h>
58
#include <panic.h>
66
#include <typedefs.h>
59
#include <typedefs.h>
67
#include <cpu.h>
60
#include <cpu.h>
68
#include <print.h>
61
#include <print.h>
69
#include <debug.h>
62
#include <debug.h>
70
 
63
 
71
static void before_task_runs(void);
64
static void before_task_runs(void);
72
static void before_thread_runs(void);
65
static void before_thread_runs(void);
73
static void after_thread_ran(void);
66
static void after_thread_ran(void);
74
static void scheduler_separated_stack(void);
67
static void scheduler_separated_stack(void);
75
 
68
 
76
atomic_t nrdy;  /**< Number of ready threads in the system. */
69
atomic_t nrdy;  /**< Number of ready threads in the system. */
77
 
70
 
78
/** Carry out actions before new task runs. */
71
/** Carry out actions before new task runs. */
79
void before_task_runs(void)
72
void before_task_runs(void)
80
{
73
{
81
    before_task_runs_arch();
74
    before_task_runs_arch();
82
}
75
}
83
 
76
 
84
/** Take actions before new thread runs.
77
/** Take actions before new thread runs.
85
 *
78
 *
86
 * Perform actions that need to be
79
 * Perform actions that need to be
87
 * taken before the newly selected
80
 * taken before the newly selected
88
 * tread is passed control.
81
 * tread is passed control.
89
 *
82
 *
90
 * THREAD->lock is locked on entry
83
 * THREAD->lock is locked on entry
91
 *
84
 *
92
 */
85
 */
93
void before_thread_runs(void)
86
void before_thread_runs(void)
94
{
87
{
95
    before_thread_runs_arch();
88
    before_thread_runs_arch();
96
#ifdef CONFIG_FPU_LAZY
89
#ifdef CONFIG_FPU_LAZY
97
    if(THREAD==CPU->fpu_owner)
90
    if(THREAD==CPU->fpu_owner)
98
        fpu_enable();
91
        fpu_enable();
99
    else
92
    else
100
        fpu_disable();
93
        fpu_disable();
101
#else
94
#else
102
    fpu_enable();
95
    fpu_enable();
103
    if (THREAD->fpu_context_exists)
96
    if (THREAD->fpu_context_exists)
104
        fpu_context_restore(THREAD->saved_fpu_context);
97
        fpu_context_restore(THREAD->saved_fpu_context);
105
    else {
98
    else {
106
        fpu_init();
99
        fpu_init();
107
        THREAD->fpu_context_exists=1;
100
        THREAD->fpu_context_exists=1;
108
    }
101
    }
109
#endif
102
#endif
110
}
103
}
111
 
104
 
112
/** Take actions after THREAD had run.
105
/** Take actions after THREAD had run.
113
 *
106
 *
114
 * Perform actions that need to be
107
 * Perform actions that need to be
115
 * taken after the running thread
108
 * taken after the running thread
116
 * had been preempted by the scheduler.
109
 * had been preempted by the scheduler.
117
 *
110
 *
118
 * THREAD->lock is locked on entry
111
 * THREAD->lock is locked on entry
119
 *
112
 *
120
 */
113
 */
121
void after_thread_ran(void)
114
void after_thread_ran(void)
122
{
115
{
123
    after_thread_ran_arch();
116
    after_thread_ran_arch();
124
}
117
}
125
 
118
 
126
#ifdef CONFIG_FPU_LAZY
119
#ifdef CONFIG_FPU_LAZY
127
void scheduler_fpu_lazy_request(void)
120
void scheduler_fpu_lazy_request(void)
128
{
121
{
129
restart:
122
restart:
130
    fpu_enable();
123
    fpu_enable();
131
    spinlock_lock(&CPU->lock);
124
    spinlock_lock(&CPU->lock);
132
 
125
 
133
    /* Save old context */
126
    /* Save old context */
134
    if (CPU->fpu_owner != NULL) {  
127
    if (CPU->fpu_owner != NULL) {  
135
        spinlock_lock(&CPU->fpu_owner->lock);
128
        spinlock_lock(&CPU->fpu_owner->lock);
136
        fpu_context_save(CPU->fpu_owner->saved_fpu_context);
129
        fpu_context_save(CPU->fpu_owner->saved_fpu_context);
137
        /* don't prevent migration */
130
        /* don't prevent migration */
138
        CPU->fpu_owner->fpu_context_engaged=0;
131
        CPU->fpu_owner->fpu_context_engaged=0;
139
        spinlock_unlock(&CPU->fpu_owner->lock);
132
        spinlock_unlock(&CPU->fpu_owner->lock);
140
        CPU->fpu_owner = NULL;
133
        CPU->fpu_owner = NULL;
141
    }
134
    }
142
 
135
 
143
    spinlock_lock(&THREAD->lock);
136
    spinlock_lock(&THREAD->lock);
144
    if (THREAD->fpu_context_exists) {
137
    if (THREAD->fpu_context_exists) {
145
        fpu_context_restore(THREAD->saved_fpu_context);
138
        fpu_context_restore(THREAD->saved_fpu_context);
146
    } else {
139
    } else {
147
        /* Allocate FPU context */
140
        /* Allocate FPU context */
148
        if (!THREAD->saved_fpu_context) {
141
        if (!THREAD->saved_fpu_context) {
149
            /* Might sleep */
142
            /* Might sleep */
150
            spinlock_unlock(&THREAD->lock);
143
            spinlock_unlock(&THREAD->lock);
151
            spinlock_unlock(&CPU->lock);
144
            spinlock_unlock(&CPU->lock);
152
            THREAD->saved_fpu_context = slab_alloc(fpu_context_slab,
145
            THREAD->saved_fpu_context = slab_alloc(fpu_context_slab,
153
                                   0);
146
                                   0);
154
            /* We may have switched CPUs during slab_alloc */
147
            /* We may have switched CPUs during slab_alloc */
155
            goto restart;
148
            goto restart;
156
        }
149
        }
157
        fpu_init();
150
        fpu_init();
158
        THREAD->fpu_context_exists=1;
151
        THREAD->fpu_context_exists=1;
159
    }
152
    }
160
    CPU->fpu_owner=THREAD;
153
    CPU->fpu_owner=THREAD;
161
    THREAD->fpu_context_engaged = 1;
154
    THREAD->fpu_context_engaged = 1;
162
    spinlock_unlock(&THREAD->lock);
155
    spinlock_unlock(&THREAD->lock);
163
 
156
 
164
    spinlock_unlock(&CPU->lock);
157
    spinlock_unlock(&CPU->lock);
165
}
158
}
166
#endif
159
#endif
167
 
160
 
168
/** Initialize scheduler
161
/** Initialize scheduler
169
 *
162
 *
170
 * Initialize kernel scheduler.
163
 * Initialize kernel scheduler.
171
 *
164
 *
172
 */
165
 */
173
void scheduler_init(void)
166
void scheduler_init(void)
174
{
167
{
175
}
168
}
176
 
169
 
177
/** Get thread to be scheduled
170
/** Get thread to be scheduled
178
 *
171
 *
179
 * Get the optimal thread to be scheduled
172
 * Get the optimal thread to be scheduled
180
 * according to thread accounting and scheduler
173
 * according to thread accounting and scheduler
181
 * policy.
174
 * policy.
182
 *
175
 *
183
 * @return Thread to be scheduled.
176
 * @return Thread to be scheduled.
184
 *
177
 *
185
 */
178
 */
186
static thread_t *find_best_thread(void)
179
static thread_t *find_best_thread(void)
187
{
180
{
188
    thread_t *t;
181
    thread_t *t;
189
    runq_t *r;
182
    runq_t *r;
190
    int i;
183
    int i;
191
 
184
 
192
    ASSERT(CPU != NULL);
185
    ASSERT(CPU != NULL);
193
 
186
 
194
loop:
187
loop:
195
    interrupts_enable();
188
    interrupts_enable();
196
   
189
   
197
    if (atomic_get(&CPU->nrdy) == 0) {
190
    if (atomic_get(&CPU->nrdy) == 0) {
198
        /*
191
        /*
199
         * For there was nothing to run, the CPU goes to sleep
192
         * For there was nothing to run, the CPU goes to sleep
200
         * until a hardware interrupt or an IPI comes.
193
         * until a hardware interrupt or an IPI comes.
201
         * This improves energy saving and hyperthreading.
194
         * This improves energy saving and hyperthreading.
202
         */
195
         */
203
 
196
 
204
        /*
197
        /*
205
         * An interrupt might occur right now and wake up a thread.
198
         * An interrupt might occur right now and wake up a thread.
206
         * In such case, the CPU will continue to go to sleep
199
         * In such case, the CPU will continue to go to sleep
207
         * even though there is a runnable thread.
200
         * even though there is a runnable thread.
208
         */
201
         */
209
 
202
 
210
         cpu_sleep();
203
         cpu_sleep();
211
         goto loop;
204
         goto loop;
212
    }
205
    }
213
 
206
 
214
    interrupts_disable();
207
    interrupts_disable();
215
   
208
   
216
    for (i = 0; i<RQ_COUNT; i++) {
209
    for (i = 0; i<RQ_COUNT; i++) {
217
        r = &CPU->rq[i];
210
        r = &CPU->rq[i];
218
        spinlock_lock(&r->lock);
211
        spinlock_lock(&r->lock);
219
        if (r->n == 0) {
212
        if (r->n == 0) {
220
            /*
213
            /*
221
             * If this queue is empty, try a lower-priority queue.
214
             * If this queue is empty, try a lower-priority queue.
222
             */
215
             */
223
            spinlock_unlock(&r->lock);
216
            spinlock_unlock(&r->lock);
224
            continue;
217
            continue;
225
        }
218
        }
226
 
219
 
227
        atomic_dec(&CPU->nrdy);
220
        atomic_dec(&CPU->nrdy);
228
        atomic_dec(&nrdy);
221
        atomic_dec(&nrdy);
229
        r->n--;
222
        r->n--;
230
 
223
 
231
        /*
224
        /*
232
         * Take the first thread from the queue.
225
         * Take the first thread from the queue.
233
         */
226
         */
234
        t = list_get_instance(r->rq_head.next, thread_t, rq_link);
227
        t = list_get_instance(r->rq_head.next, thread_t, rq_link);
235
        list_remove(&t->rq_link);
228
        list_remove(&t->rq_link);
236
 
229
 
237
        spinlock_unlock(&r->lock);
230
        spinlock_unlock(&r->lock);
238
 
231
 
239
        spinlock_lock(&t->lock);
232
        spinlock_lock(&t->lock);
240
        t->cpu = CPU;
233
        t->cpu = CPU;
241
 
234
 
242
        t->ticks = us2ticks((i+1)*10000);
235
        t->ticks = us2ticks((i+1)*10000);
243
        t->priority = i;    /* correct rq index */
236
        t->priority = i;    /* correct rq index */
244
 
237
 
245
        /*
238
        /*
246
         * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.
239
         * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.
247
         */
240
         */
248
        t->flags &= ~X_STOLEN;
241
        t->flags &= ~X_STOLEN;
249
        spinlock_unlock(&t->lock);
242
        spinlock_unlock(&t->lock);
250
 
243
 
251
        return t;
244
        return t;
252
    }
245
    }
253
    goto loop;
246
    goto loop;
254
 
247
 
255
}
248
}
256
 
249
 
257
/** Prevent rq starvation
250
/** Prevent rq starvation
258
 *
251
 *
259
 * Prevent low priority threads from starving in rq's.
252
 * Prevent low priority threads from starving in rq's.
260
 *
253
 *
261
 * When the function decides to relink rq's, it reconnects
254
 * When the function decides to relink rq's, it reconnects
262
 * respective pointers so that in result threads with 'pri'
255
 * respective pointers so that in result threads with 'pri'
263
 * greater or equal @start are moved to a higher-priority queue.
256
 * greater or equal @start are moved to a higher-priority queue.
264
 *
257
 *
265
 * @param start Threshold priority.
258
 * @param start Threshold priority.
266
 *
259
 *
267
 */
260
 */
268
static void relink_rq(int start)
261
static void relink_rq(int start)
269
{
262
{
270
    link_t head;
263
    link_t head;
271
    runq_t *r;
264
    runq_t *r;
272
    int i, n;
265
    int i, n;
273
 
266
 
274
    list_initialize(&head);
267
    list_initialize(&head);
275
    spinlock_lock(&CPU->lock);
268
    spinlock_lock(&CPU->lock);
276
    if (CPU->needs_relink > NEEDS_RELINK_MAX) {
269
    if (CPU->needs_relink > NEEDS_RELINK_MAX) {
277
        for (i = start; i<RQ_COUNT-1; i++) {
270
        for (i = start; i<RQ_COUNT-1; i++) {
278
            /* remember and empty rq[i + 1] */
271
            /* remember and empty rq[i + 1] */
279
            r = &CPU->rq[i + 1];
272
            r = &CPU->rq[i + 1];
280
            spinlock_lock(&r->lock);
273
            spinlock_lock(&r->lock);
281
            list_concat(&head, &r->rq_head);
274
            list_concat(&head, &r->rq_head);
282
            n = r->n;
275
            n = r->n;
283
            r->n = 0;
276
            r->n = 0;
284
            spinlock_unlock(&r->lock);
277
            spinlock_unlock(&r->lock);
285
       
278
       
286
            /* append rq[i + 1] to rq[i] */
279
            /* append rq[i + 1] to rq[i] */
287
            r = &CPU->rq[i];
280
            r = &CPU->rq[i];
288
            spinlock_lock(&r->lock);
281
            spinlock_lock(&r->lock);
289
            list_concat(&r->rq_head, &head);
282
            list_concat(&r->rq_head, &head);
290
            r->n += n;
283
            r->n += n;
291
            spinlock_unlock(&r->lock);
284
            spinlock_unlock(&r->lock);
292
        }
285
        }
293
        CPU->needs_relink = 0;
286
        CPU->needs_relink = 0;
294
    }
287
    }
295
    spinlock_unlock(&CPU->lock);
288
    spinlock_unlock(&CPU->lock);
296
 
289
 
297
}
290
}
298
 
291
 
299
/** The scheduler
292
/** The scheduler
300
 *
293
 *
301
 * The thread scheduling procedure.
294
 * The thread scheduling procedure.
302
 * Passes control directly to
295
 * Passes control directly to
303
 * scheduler_separated_stack().
296
 * scheduler_separated_stack().
304
 *
297
 *
305
 */
298
 */
306
void scheduler(void)
299
void scheduler(void)
307
{
300
{
308
    volatile ipl_t ipl;
301
    volatile ipl_t ipl;
309
 
302
 
310
    ASSERT(CPU != NULL);
303
    ASSERT(CPU != NULL);
311
 
304
 
312
    ipl = interrupts_disable();
305
    ipl = interrupts_disable();
313
 
306
 
314
    if (atomic_get(&haltstate))
307
    if (atomic_get(&haltstate))
315
        halt();
308
        halt();
316
   
309
   
317
    if (THREAD) {
310
    if (THREAD) {
318
        spinlock_lock(&THREAD->lock);
311
        spinlock_lock(&THREAD->lock);
319
#ifndef CONFIG_FPU_LAZY
312
#ifndef CONFIG_FPU_LAZY
320
        fpu_context_save(THREAD->saved_fpu_context);
313
        fpu_context_save(THREAD->saved_fpu_context);
321
#endif
314
#endif
322
        if (!context_save(&THREAD->saved_context)) {
315
        if (!context_save(&THREAD->saved_context)) {
323
            /*
316
            /*
324
             * This is the place where threads leave scheduler();
317
             * This is the place where threads leave scheduler();
325
             */
318
             */
326
            spinlock_unlock(&THREAD->lock);
319
            spinlock_unlock(&THREAD->lock);
327
            interrupts_restore(THREAD->saved_context.ipl);
320
            interrupts_restore(THREAD->saved_context.ipl);
328
           
321
           
329
            return;
322
            return;
330
        }
323
        }
331
 
324
 
332
        /*
325
        /*
333
         * Interrupt priority level of preempted thread is recorded here
326
         * Interrupt priority level of preempted thread is recorded here
334
         * to facilitate scheduler() invocations from interrupts_disable()'d
327
         * to facilitate scheduler() invocations from interrupts_disable()'d
335
         * code (e.g. waitq_sleep_timeout()).
328
         * code (e.g. waitq_sleep_timeout()).
336
         */
329
         */
337
        THREAD->saved_context.ipl = ipl;
330
        THREAD->saved_context.ipl = ipl;
338
    }
331
    }
339
 
332
 
340
    /*
333
    /*
341
     * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM
334
     * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM
342
     * and preemption counter. At this point THE could be coming either
335
     * and preemption counter. At this point THE could be coming either
343
     * from THREAD's or CPU's stack.
336
     * from THREAD's or CPU's stack.
344
     */
337
     */
345
    the_copy(THE, (the_t *) CPU->stack);
338
    the_copy(THE, (the_t *) CPU->stack);
346
 
339
 
347
    /*
340
    /*
348
     * We may not keep the old stack.
341
     * We may not keep the old stack.
349
     * Reason: If we kept the old stack and got blocked, for instance, in
342
     * Reason: If we kept the old stack and got blocked, for instance, in
350
     * find_best_thread(), the old thread could get rescheduled by another
343
     * find_best_thread(), the old thread could get rescheduled by another
351
     * CPU and overwrite the part of its own stack that was also used by
344
     * CPU and overwrite the part of its own stack that was also used by
352
     * the scheduler on this CPU.
345
     * the scheduler on this CPU.
353
     *
346
     *
354
     * Moreover, we have to bypass the compiler-generated POP sequence
347
     * Moreover, we have to bypass the compiler-generated POP sequence
355
     * which is fooled by SP being set to the very top of the stack.
348
     * which is fooled by SP being set to the very top of the stack.
356
     * Therefore the scheduler() function continues in
349
     * Therefore the scheduler() function continues in
357
     * scheduler_separated_stack().
350
     * scheduler_separated_stack().
358
     */
351
     */
359
    context_save(&CPU->saved_context);
352
    context_save(&CPU->saved_context);
360
    context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), (__address) CPU->stack, CPU_STACK_SIZE);
353
    context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), (__address) CPU->stack, CPU_STACK_SIZE);
361
    context_restore(&CPU->saved_context);
354
    context_restore(&CPU->saved_context);
362
    /* not reached */
355
    /* not reached */
363
}
356
}
364
 
357
 
365
/** Scheduler stack switch wrapper
358
/** Scheduler stack switch wrapper
366
 *
359
 *
367
 * Second part of the scheduler() function
360
 * Second part of the scheduler() function
368
 * using new stack. Handling the actual context
361
 * using new stack. Handling the actual context
369
 * switch to a new thread.
362
 * switch to a new thread.
370
 *
363
 *
371
 * Assume THREAD->lock is held.
364
 * Assume THREAD->lock is held.
372
 */
365
 */
373
void scheduler_separated_stack(void)
366
void scheduler_separated_stack(void)
374
{
367
{
375
    int priority;
368
    int priority;
376
   
369
   
377
    ASSERT(CPU != NULL);
370
    ASSERT(CPU != NULL);
378
   
371
   
379
    if (THREAD) {
372
    if (THREAD) {
380
        /* must be run after the switch to scheduler stack */
373
        /* must be run after the switch to scheduler stack */
381
        after_thread_ran();
374
        after_thread_ran();
382
 
375
 
383
        switch (THREAD->state) {
376
        switch (THREAD->state) {
384
            case Running:
377
            case Running:
385
            spinlock_unlock(&THREAD->lock);
378
            spinlock_unlock(&THREAD->lock);
386
            thread_ready(THREAD);
379
            thread_ready(THREAD);
387
            break;
380
            break;
388
 
381
 
389
            case Exiting:
382
            case Exiting:
390
repeat:
383
repeat:
391
                if (THREAD->detached) {
384
                if (THREAD->detached) {
392
                thread_destroy(THREAD);
385
                thread_destroy(THREAD);
393
            } else {
386
            } else {
394
                /*
387
                /*
395
                 * The thread structure is kept allocated until somebody
388
                 * The thread structure is kept allocated until somebody
396
                 * calls thread_detach() on it.
389
                 * calls thread_detach() on it.
397
                 */
390
                 */
398
                if (!spinlock_trylock(&THREAD->join_wq.lock)) {
391
                if (!spinlock_trylock(&THREAD->join_wq.lock)) {
399
                    /*
392
                    /*
400
                     * Avoid deadlock.
393
                     * Avoid deadlock.
401
                     */
394
                     */
402
                    spinlock_unlock(&THREAD->lock);
395
                    spinlock_unlock(&THREAD->lock);
403
                    delay(10);
396
                    delay(10);
404
                    spinlock_lock(&THREAD->lock);
397
                    spinlock_lock(&THREAD->lock);
405
                    goto repeat;
398
                    goto repeat;
406
                }
399
                }
407
                _waitq_wakeup_unsafe(&THREAD->join_wq, false);
400
                _waitq_wakeup_unsafe(&THREAD->join_wq, false);
408
                spinlock_unlock(&THREAD->join_wq.lock);
401
                spinlock_unlock(&THREAD->join_wq.lock);
409
               
402
               
410
                THREAD->state = Undead;
403
                THREAD->state = Undead;
411
                spinlock_unlock(&THREAD->lock);
404
                spinlock_unlock(&THREAD->lock);
412
            }
405
            }
413
            break;
406
            break;
414
           
407
           
415
            case Sleeping:
408
            case Sleeping:
416
            /*
409
            /*
417
             * Prefer the thread after it's woken up.
410
             * Prefer the thread after it's woken up.
418
             */
411
             */
419
            THREAD->priority = -1;
412
            THREAD->priority = -1;
420
 
413
 
421
            /*
414
            /*
422
             * We need to release wq->lock which we locked in waitq_sleep().
415
             * We need to release wq->lock which we locked in waitq_sleep().
423
             * Address of wq->lock is kept in THREAD->sleep_queue.
416
             * Address of wq->lock is kept in THREAD->sleep_queue.
424
             */
417
             */
425
            spinlock_unlock(&THREAD->sleep_queue->lock);
418
            spinlock_unlock(&THREAD->sleep_queue->lock);
426
 
419
 
427
            /*
420
            /*
428
             * Check for possible requests for out-of-context invocation.
421
             * Check for possible requests for out-of-context invocation.
429
             */
422
             */
430
            if (THREAD->call_me) {
423
            if (THREAD->call_me) {
431
                THREAD->call_me(THREAD->call_me_with);
424
                THREAD->call_me(THREAD->call_me_with);
432
                THREAD->call_me = NULL;
425
                THREAD->call_me = NULL;
433
                THREAD->call_me_with = NULL;
426
                THREAD->call_me_with = NULL;
434
            }
427
            }
435
 
428
 
436
            spinlock_unlock(&THREAD->lock);
429
            spinlock_unlock(&THREAD->lock);
437
 
430
 
438
            break;
431
            break;
439
 
432
 
440
            default:
433
            default:
441
            /*
434
            /*
442
             * Entering state is unexpected.
435
             * Entering state is unexpected.
443
             */
436
             */
444
            panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
437
            panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
445
            break;
438
            break;
446
        }
439
        }
447
 
440
 
448
        THREAD = NULL;
441
        THREAD = NULL;
449
    }
442
    }
450
 
443
 
451
    THREAD = find_best_thread();
444
    THREAD = find_best_thread();
452
   
445
   
453
    spinlock_lock(&THREAD->lock);
446
    spinlock_lock(&THREAD->lock);
454
    priority = THREAD->priority;
447
    priority = THREAD->priority;
455
    spinlock_unlock(&THREAD->lock);
448
    spinlock_unlock(&THREAD->lock);
456
 
449
 
457
    relink_rq(priority);       
450
    relink_rq(priority);       
458
 
451
 
459
    /*
452
    /*
460
     * If both the old and the new task are the same, lots of work is avoided.
453
     * If both the old and the new task are the same, lots of work is avoided.
461
     */
454
     */
462
    if (TASK != THREAD->task) {
455
    if (TASK != THREAD->task) {
463
        as_t *as1 = NULL;
456
        as_t *as1 = NULL;
464
        as_t *as2;
457
        as_t *as2;
465
 
458
 
466
        if (TASK) {
459
        if (TASK) {
467
            spinlock_lock(&TASK->lock);
460
            spinlock_lock(&TASK->lock);
468
            as1 = TASK->as;
461
            as1 = TASK->as;
469
            spinlock_unlock(&TASK->lock);
462
            spinlock_unlock(&TASK->lock);
470
        }
463
        }
471
 
464
 
472
        spinlock_lock(&THREAD->task->lock);
465
        spinlock_lock(&THREAD->task->lock);
473
        as2 = THREAD->task->as;
466
        as2 = THREAD->task->as;
474
        spinlock_unlock(&THREAD->task->lock);
467
        spinlock_unlock(&THREAD->task->lock);
475
       
468
       
476
        /*
469
        /*
477
         * Note that it is possible for two tasks to share one address space.
470
         * Note that it is possible for two tasks to share one address space.
478
         */
471
         */
479
        if (as1 != as2) {
472
        if (as1 != as2) {
480
            /*
473
            /*
481
             * Both tasks and address spaces are different.
474
             * Both tasks and address spaces are different.
482
             * Replace the old one with the new one.
475
             * Replace the old one with the new one.
483
             */
476
             */
484
            as_switch(as1, as2);
477
            as_switch(as1, as2);
485
        }
478
        }
486
        TASK = THREAD->task;
479
        TASK = THREAD->task;
487
        before_task_runs();
480
        before_task_runs();
488
    }
481
    }
489
 
482
 
490
    spinlock_lock(&THREAD->lock);  
483
    spinlock_lock(&THREAD->lock);  
491
    THREAD->state = Running;
484
    THREAD->state = Running;
492
 
485
 
493
#ifdef SCHEDULER_VERBOSE
486
#ifdef SCHEDULER_VERBOSE
494
    printf("cpu%d: tid %d (priority=%d,ticks=%lld,nrdy=%ld)\n", CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks, atomic_get(&CPU->nrdy));
487
    printf("cpu%d: tid %d (priority=%d,ticks=%lld,nrdy=%ld)\n", CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks, atomic_get(&CPU->nrdy));
495
#endif  
488
#endif  
496
 
489
 
497
    /*
490
    /*
498
     * Some architectures provide late kernel PA2KA(identity)
491
     * Some architectures provide late kernel PA2KA(identity)
499
     * mapping in a page fault handler. However, the page fault
492
     * mapping in a page fault handler. However, the page fault
500
     * handler uses the kernel stack of the running thread and
493
     * handler uses the kernel stack of the running thread and
501
     * therefore cannot be used to map it. The kernel stack, if
494
     * therefore cannot be used to map it. The kernel stack, if
502
     * necessary, is to be mapped in before_thread_runs(). This
495
     * necessary, is to be mapped in before_thread_runs(). This
503
     * function must be executed before the switch to the new stack.
496
     * function must be executed before the switch to the new stack.
504
     */
497
     */
505
    before_thread_runs();
498
    before_thread_runs();
506
 
499
 
507
    /*
500
    /*
508
     * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack.
501
     * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack.
509
     */
502
     */
510
    the_copy(THE, (the_t *) THREAD->kstack);
503
    the_copy(THE, (the_t *) THREAD->kstack);
511
   
504
   
512
    context_restore(&THREAD->saved_context);
505
    context_restore(&THREAD->saved_context);
513
    /* not reached */
506
    /* not reached */
514
}
507
}
515
 
508
 
516
#ifdef CONFIG_SMP
509
#ifdef CONFIG_SMP
517
/** Load balancing thread
510
/** Load balancing thread
518
 *
511
 *
519
 * SMP load balancing thread, supervising thread supplies
512
 * SMP load balancing thread, supervising thread supplies
520
 * for the CPU it's wired to.
513
 * for the CPU it's wired to.
521
 *
514
 *
522
 * @param arg Generic thread argument (unused).
515
 * @param arg Generic thread argument (unused).
523
 *
516
 *
524
 */
517
 */
525
void kcpulb(void *arg)
518
void kcpulb(void *arg)
526
{
519
{
527
    thread_t *t;
520
    thread_t *t;
528
    int count, average, i, j, k = 0;
521
    int count, average, i, j, k = 0;
529
    ipl_t ipl;
522
    ipl_t ipl;
530
 
523
 
531
    /*
524
    /*
532
     * Detach kcpulb as nobody will call thread_join_timeout() on it.
525
     * Detach kcpulb as nobody will call thread_join_timeout() on it.
533
     */
526
     */
534
    thread_detach(THREAD);
527
    thread_detach(THREAD);
535
   
528
   
536
loop:
529
loop:
537
    /*
530
    /*
538
     * Work in 1s intervals.
531
     * Work in 1s intervals.
539
     */
532
     */
540
    thread_sleep(1);
533
    thread_sleep(1);
541
 
534
 
542
not_satisfied:
535
not_satisfied:
543
    /*
536
    /*
544
     * Calculate the number of threads that will be migrated/stolen from
537
     * Calculate the number of threads that will be migrated/stolen from
545
     * other CPU's. Note that situation can have changed between two
538
     * other CPU's. Note that situation can have changed between two
546
     * passes. Each time get the most up to date counts.
539
     * passes. Each time get the most up to date counts.
547
     */
540
     */
548
    average = atomic_get(&nrdy) / config.cpu_active + 1;
541
    average = atomic_get(&nrdy) / config.cpu_active + 1;
549
    count = average - atomic_get(&CPU->nrdy);
542
    count = average - atomic_get(&CPU->nrdy);
550
 
543
 
551
    if (count <= 0)
544
    if (count <= 0)
552
        goto satisfied;
545
        goto satisfied;
553
 
546
 
554
    /*
547
    /*
555
     * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
548
     * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
556
     */
549
     */
557
    for (j=RQ_COUNT-1; j >= 0; j--) {
550
    for (j=RQ_COUNT-1; j >= 0; j--) {
558
        for (i=0; i < config.cpu_active; i++) {
551
        for (i=0; i < config.cpu_active; i++) {
559
            link_t *l;
552
            link_t *l;
560
            runq_t *r;
553
            runq_t *r;
561
            cpu_t *cpu;
554
            cpu_t *cpu;
562
 
555
 
563
            cpu = &cpus[(i + k) % config.cpu_active];
556
            cpu = &cpus[(i + k) % config.cpu_active];
564
 
557
 
565
            /*
558
            /*
566
             * Not interested in ourselves.
559
             * Not interested in ourselves.
567
             * Doesn't require interrupt disabling for kcpulb is X_WIRED.
560
             * Doesn't require interrupt disabling for kcpulb is X_WIRED.
568
             */
561
             */
569
            if (CPU == cpu)
562
            if (CPU == cpu)
570
                continue;
563
                continue;
571
            if (atomic_get(&cpu->nrdy) <= average)
564
            if (atomic_get(&cpu->nrdy) <= average)
572
                continue;
565
                continue;
573
 
566
 
574
            ipl = interrupts_disable();
567
            ipl = interrupts_disable();
575
            r = &cpu->rq[j];
568
            r = &cpu->rq[j];
576
            spinlock_lock(&r->lock);
569
            spinlock_lock(&r->lock);
577
            if (r->n == 0) {
570
            if (r->n == 0) {
578
                spinlock_unlock(&r->lock);
571
                spinlock_unlock(&r->lock);
579
                interrupts_restore(ipl);
572
                interrupts_restore(ipl);
580
                continue;
573
                continue;
581
            }
574
            }
582
       
575
       
583
            t = NULL;
576
            t = NULL;
584
            l = r->rq_head.prev;    /* search rq from the back */
577
            l = r->rq_head.prev;    /* search rq from the back */
585
            while (l != &r->rq_head) {
578
            while (l != &r->rq_head) {
586
                t = list_get_instance(l, thread_t, rq_link);
579
                t = list_get_instance(l, thread_t, rq_link);
587
                /*
580
                /*
588
                 * We don't want to steal CPU-wired threads neither threads already stolen.
581
                 * We don't want to steal CPU-wired threads neither threads already stolen.
589
                 * The latter prevents threads from migrating between CPU's without ever being run.
582
                 * The latter prevents threads from migrating between CPU's without ever being run.
590
                 * We don't want to steal threads whose FPU context is still in CPU.
583
                 * We don't want to steal threads whose FPU context is still in CPU.
591
                 */
584
                 */
592
                spinlock_lock(&t->lock);
585
                spinlock_lock(&t->lock);
593
                if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
586
                if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
594
                    /*
587
                    /*
595
                     * Remove t from r.
588
                     * Remove t from r.
596
                     */
589
                     */
597
                    spinlock_unlock(&t->lock);
590
                    spinlock_unlock(&t->lock);
598
                   
591
                   
599
                    atomic_dec(&cpu->nrdy);
592
                    atomic_dec(&cpu->nrdy);
600
                    atomic_dec(&nrdy);
593
                    atomic_dec(&nrdy);
601
 
594
 
602
                    r->n--;
595
                    r->n--;
603
                    list_remove(&t->rq_link);
596
                    list_remove(&t->rq_link);
604
 
597
 
605
                    break;
598
                    break;
606
                }
599
                }
607
                spinlock_unlock(&t->lock);
600
                spinlock_unlock(&t->lock);
608
                l = l->prev;
601
                l = l->prev;
609
                t = NULL;
602
                t = NULL;
610
            }
603
            }
611
            spinlock_unlock(&r->lock);
604
            spinlock_unlock(&r->lock);
612
 
605
 
613
            if (t) {
606
            if (t) {
614
                /*
607
                /*
615
                 * Ready t on local CPU
608
                 * Ready t on local CPU
616
                 */
609
                 */
617
                spinlock_lock(&t->lock);
610
                spinlock_lock(&t->lock);
618
#ifdef KCPULB_VERBOSE
611
#ifdef KCPULB_VERBOSE
619
                printf("kcpulb%d: TID %d -> cpu%d, nrdy=%ld, avg=%nd\n", CPU->id, t->tid, CPU->id, atomic_get(&CPU->nrdy), atomic_get(&nrdy) / config.cpu_active);
612
                printf("kcpulb%d: TID %d -> cpu%d, nrdy=%ld, avg=%nd\n", CPU->id, t->tid, CPU->id, atomic_get(&CPU->nrdy), atomic_get(&nrdy) / config.cpu_active);
620
#endif
613
#endif
621
                t->flags |= X_STOLEN;
614
                t->flags |= X_STOLEN;
622
                t->state = Entering;
615
                t->state = Entering;
623
                spinlock_unlock(&t->lock);
616
                spinlock_unlock(&t->lock);
624
   
617
   
625
                thread_ready(t);
618
                thread_ready(t);
626
 
619
 
627
                interrupts_restore(ipl);
620
                interrupts_restore(ipl);
628
   
621
   
629
                if (--count == 0)
622
                if (--count == 0)
630
                    goto satisfied;
623
                    goto satisfied;
631
                   
624
                   
632
                /*
625
                /*
633
                 * We are not satisfied yet, focus on another CPU next time.
626
                 * We are not satisfied yet, focus on another CPU next time.
634
                 */
627
                 */
635
                k++;
628
                k++;
636
               
629
               
637
                continue;
630
                continue;
638
            }
631
            }
639
            interrupts_restore(ipl);
632
            interrupts_restore(ipl);
640
        }
633
        }
641
    }
634
    }
642
 
635
 
643
    if (atomic_get(&CPU->nrdy)) {
636
    if (atomic_get(&CPU->nrdy)) {
644
        /*
637
        /*
645
         * Be a little bit light-weight and let migrated threads run.
638
         * Be a little bit light-weight and let migrated threads run.
646
         */
639
         */
647
        scheduler();
640
        scheduler();
648
    } else {
641
    } else {
649
        /*
642
        /*
650
         * We failed to migrate a single thread.
643
         * We failed to migrate a single thread.
651
         * Give up this turn.
644
         * Give up this turn.
652
         */
645
         */
653
        goto loop;
646
        goto loop;
654
    }
647
    }
655
       
648
       
656
    goto not_satisfied;
649
    goto not_satisfied;
657
 
650
 
658
satisfied:
651
satisfied:
659
    goto loop;
652
    goto loop;
660
}
653
}
661
 
654
 
662
#endif /* CONFIG_SMP */
655
#endif /* CONFIG_SMP */
663
 
656
 
664
 
657
 
665
/** Print information about threads & scheduler queues */
658
/** Print information about threads & scheduler queues */
666
void sched_print_list(void)
659
void sched_print_list(void)
667
{
660
{
668
    ipl_t ipl;
661
    ipl_t ipl;
669
    int cpu,i;
662
    int cpu,i;
670
    runq_t *r;
663
    runq_t *r;
671
    thread_t *t;
664
    thread_t *t;
672
    link_t *cur;
665
    link_t *cur;
673
 
666
 
674
    /* We are going to mess with scheduler structures,
667
    /* We are going to mess with scheduler structures,
675
     * let's not be interrupted */
668
     * let's not be interrupted */
676
    ipl = interrupts_disable();
669
    ipl = interrupts_disable();
677
    for (cpu=0;cpu < config.cpu_count; cpu++) {
670
    for (cpu=0;cpu < config.cpu_count; cpu++) {
678
 
671
 
679
        if (!cpus[cpu].active)
672
        if (!cpus[cpu].active)
680
            continue;
673
            continue;
681
 
674
 
682
        spinlock_lock(&cpus[cpu].lock);
675
        spinlock_lock(&cpus[cpu].lock);
683
        printf("cpu%d: address=%p, nrdy=%ld, needs_relink=%ld\n",
676
        printf("cpu%d: address=%p, nrdy=%ld, needs_relink=%ld\n",
684
               cpus[cpu].id, &cpus[cpu], atomic_get(&cpus[cpu].nrdy), cpus[cpu].needs_relink);
677
               cpus[cpu].id, &cpus[cpu], atomic_get(&cpus[cpu].nrdy), cpus[cpu].needs_relink);
685
       
678
       
686
        for (i=0; i<RQ_COUNT; i++) {
679
        for (i=0; i<RQ_COUNT; i++) {
687
            r = &cpus[cpu].rq[i];
680
            r = &cpus[cpu].rq[i];
688
            spinlock_lock(&r->lock);
681
            spinlock_lock(&r->lock);
689
            if (!r->n) {
682
            if (!r->n) {
690
                spinlock_unlock(&r->lock);
683
                spinlock_unlock(&r->lock);
691
                continue;
684
                continue;
692
            }
685
            }
693
            printf("\trq[%d]: ", i);
686
            printf("\trq[%d]: ", i);
694
            for (cur=r->rq_head.next; cur!=&r->rq_head; cur=cur->next) {
687
            for (cur=r->rq_head.next; cur!=&r->rq_head; cur=cur->next) {
695
                t = list_get_instance(cur, thread_t, rq_link);
688
                t = list_get_instance(cur, thread_t, rq_link);
696
                printf("%d(%s) ", t->tid,
689
                printf("%d(%s) ", t->tid,
697
                       thread_states[t->state]);
690
                       thread_states[t->state]);
698
            }
691
            }
699
            printf("\n");
692
            printf("\n");
700
            spinlock_unlock(&r->lock);
693
            spinlock_unlock(&r->lock);
701
        }
694
        }
702
        spinlock_unlock(&cpus[cpu].lock);
695
        spinlock_unlock(&cpus[cpu].lock);
703
    }
696
    }
704
   
697
   
705
    interrupts_restore(ipl);
698
    interrupts_restore(ipl);
706
}
699
}
707
 
700
 
708
 /** @}
701
 /** @}
709
 */
702
 */
710
 
703
 
711
 
704