WebSVN – HelenOS-historic – Diff – /kernel/trunk/generic/src/proc/scheduler.c


/*
 * Copyright (C) 2001-2004 Jakub Jermar
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * - Redistributions of source code must retain the above copyright
 *   notice, this list of conditions and the following disclaimer.
 * - Redistributions in binary form must reproduce the above copyright
 *   notice, this list of conditions and the following disclaimer in the
 *   documentation and/or other materials provided with the distribution.
 * - The name of the author may not be used to endorse or promote products
 *   derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
#include <proc/scheduler.h>
#include <proc/thread.h>
#include <proc/task.h>
#include <cpu.h>
#include <mm/vm.h>
#include <config.h>
#include <context.h>
#include <func.h>
#include <arch.h>
#include <arch/asm.h>
#include <list.h>
#include <panic.h>
#include <typedefs.h>
#include <mm/page.h>
#include <synch/spinlock.h>
#include <arch/faddr.h>
#include <arch/atomic.h>
 
volatile int nrdy;
 
 
/** Take actions before new thread runs
 *
 * Perform actions that need to be
 * taken before the newly selected
 * tread is passed control.
 *
 */
void before_thread_runs(void)
{
    before_thread_runs_arch();
    fpu_context_restore(&(THREAD->saved_fpu_context));
}
 
 
/** Initialize scheduler
 *
 * Initialize kernel scheduler.
 *
 */
void scheduler_init(void)
{
}
 
 
/** Get thread to be scheduled
 *
 * Get the optimal thread to be scheduled
 * according to thread accounting and scheduler
 * policy.
 *
 * @return Thread to be scheduled.
 *
 */
struct thread *find_best_thread(void)
{
    thread_t *t;
    runq_t *r;
    int i, n;
 
loop:
    cpu_priority_high();
 
    spinlock_lock(&CPU->lock);
    n = CPU->nrdy;
    spinlock_unlock(&CPU->lock);
 
    cpu_priority_low();
   
    if (n == 0) {
        #ifdef __SMP__
        /*
         * If the load balancing thread is not running, wake it up and
         * set CPU-private flag that the kcpulb has been started.
         */
        if (test_and_set(&CPU->kcpulbstarted) == 0) {
            waitq_wakeup(&CPU->kcpulb_wq, 0);
            goto loop;
        }
        #endif /* __SMP__ */
       
        /*
         * For there was nothing to run, the CPU goes to sleep
         * until a hardware interrupt or an IPI comes.
         * This improves energy saving and hyperthreading.
         * On the other hand, several hardware interrupts can be ignored.
         */
         cpu_sleep();
         goto loop;
    }
 
    cpu_priority_high();
   
    i = 0;
retry:
    for (; i<RQ_COUNT; i++) {
        r = &CPU->rq[i];
        spinlock_lock(&r->lock);
        if (r->n == 0) {
            /*
             * If this queue is empty, try a lower-priority queue.
             */
            spinlock_unlock(&r->lock);
            continue;
        }
   
        /* avoid deadlock with relink_rq() */
        if (!spinlock_trylock(&CPU->lock)) {
            /*
             * Unlock r and try again.
             */
            spinlock_unlock(&r->lock);
            goto retry;
        }
        CPU->nrdy--;
        spinlock_unlock(&CPU->lock);
 
        atomic_dec(&nrdy);
        r->n--;
 
        /*
         * Take the first thread from the queue.
         */
        t = list_get_instance(r->rq_head.next, thread_t, rq_link);
        list_remove(&t->rq_link);
 
        spinlock_unlock(&r->lock);
 
        spinlock_lock(&t->lock);
        t->cpu = CPU;
 
        t->ticks = us2ticks((i+1)*10000);
        t->pri = i; /* eventually correct rq index */
 
        /*
         * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.
         */
        t->flags &= ~X_STOLEN;
        spinlock_unlock(&t->lock);
 
        return t;
    }
    goto loop;
 
}
 
 
/** Prevent rq starvation
 *
 * Prevent low priority threads from starving in rq's.
 *
 * When the function decides to relink rq's, it reconnects
 * respective pointers so that in result threads with 'pri'
 * greater or equal 'start' are moved to a higher-priority queue.
 *
 * @param start Threshold priority.
 *
 */
void relink_rq(int start)
{
    link_t head;
    runq_t *r;
    int i, n;
 
    list_initialize(&head);
    spinlock_lock(&CPU->lock);
    if (CPU->needs_relink > NEEDS_RELINK_MAX) {
        for (i = start; i<RQ_COUNT-1; i++) {
            /* remember and empty rq[i + 1] */
            r = &CPU->rq[i + 1];
            spinlock_lock(&r->lock);
            list_concat(&head, &r->rq_head);
            n = r->n;
            r->n = 0;
            spinlock_unlock(&r->lock);
       
            /* append rq[i + 1] to rq[i] */
            r = &CPU->rq[i];
            spinlock_lock(&r->lock);
            list_concat(&r->rq_head, &head);
            r->n += n;
            spinlock_unlock(&r->lock);
        }
        CPU->needs_relink = 0;
    }
    spinlock_unlock(&CPU->lock);               
 
}
 
 
/** The scheduler
 *
 * The thread scheduling procedure.
 *
 */
void scheduler(void)
{
    volatile pri_t pri;
 
    pri = cpu_priority_high();
 
    if (haltstate)
        halt();
 
    if (THREAD) {
        spinlock_lock(&THREAD->lock);
        fpu_context_save(&(THREAD->saved_fpu_context));
        if (!context_save(&THREAD->saved_context)) {
            /*
             * This is the place where threads leave scheduler();
             */
            before_thread_runs();
            spinlock_unlock(&THREAD->lock);
            cpu_priority_restore(THREAD->saved_context.pri);
            return;
        }
        THREAD->saved_context.pri = pri;
    }
 
    /*
     * We may not keep the old stack.
     * Reason: If we kept the old stack and got blocked, for instance, in
     * find_best_thread(), the old thread could get rescheduled by another
     * CPU and overwrite the part of its own stack that was also used by
     * the scheduler on this CPU.
     *
     * Moreover, we have to bypass the compiler-generated POP sequence
     * which is fooled by SP being set to the very top of the stack.
     * Therefore the scheduler() function continues in
     * scheduler_separated_stack().
     */
    context_save(&CPU->saved_context);
    context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), CPU->stack, CPU_STACK_SIZE);
    context_restore(&CPU->saved_context);
    /* not reached */
}
 
 
/** Scheduler stack switch wrapper
 *
 * Second part of the scheduler() function
 * using new stack. Handling the actual context
 * switch to a new thread.
 *
 */
void scheduler_separated_stack(void)
{
    int priority;
 
    if (THREAD) {
        switch (THREAD->state) {
            case Running:
            THREAD->state = Ready;
            spinlock_unlock(&THREAD->lock);
            thread_ready(THREAD);
            break;
 
            case Exiting:
            frame_free((__address) THREAD->kstack);
            if (THREAD->ustack) {
                frame_free((__address) THREAD->ustack);
            }
 
            /*
             * Detach from the containing task.
             */
            spinlock_lock(&TASK->lock);
            list_remove(&THREAD->th_link);
            spinlock_unlock(&TASK->lock);
 
            spinlock_unlock(&THREAD->lock);
   
            spinlock_lock(&threads_lock);
            list_remove(&THREAD->threads_link);
            spinlock_unlock(&threads_lock);
 
            spinlock_lock(&CPU->lock);
            if(CPU->fpu_owner==THREAD) CPU->fpu_owner=NULL;
            spinlock_unlock(&CPU->lock);
 
 
            free(THREAD);
 
            break;
   
            case Sleeping:
            /*
             * Prefer the thread after it's woken up.
             */
            THREAD->pri = -1;
 
            /*
             * We need to release wq->lock which we locked in waitq_sleep().
             * Address of wq->lock is kept in THREAD->sleep_queue.
             */
            spinlock_unlock(&THREAD->sleep_queue->lock);
 
            /*
             * Check for possible requests for out-of-context invocation.
             */
            if (THREAD->call_me) {
                THREAD->call_me(THREAD->call_me_with);
                THREAD->call_me = NULL;
                THREAD->call_me_with = NULL;
            }
 
            spinlock_unlock(&THREAD->lock);
 
            break;
 
            default:
            /*
             * Entering state is unexpected.
             */
            panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
            break;
        }
        THREAD = NULL;
    }
 
    THREAD = find_best_thread();
   
    spinlock_lock(&THREAD->lock);
    priority = THREAD->pri;
    spinlock_unlock(&THREAD->lock);
   
    relink_rq(priority);       
 
    spinlock_lock(&THREAD->lock);  
 
    /*
     * If both the old and the new task are the same, lots of work is avoided.
     */
    if (TASK != THREAD->task) {
        vm_t *m1 = NULL;
        vm_t *m2;
 
        if (TASK) {
            spinlock_lock(&TASK->lock);
            m1 = TASK->vm;
            spinlock_unlock(&TASK->lock);
        }
 
        spinlock_lock(&THREAD->task->lock);
        m2 = THREAD->task->vm;
        spinlock_unlock(&THREAD->task->lock);
       
        /*
         * Note that it is possible for two tasks to share one vm mapping.
         */
        if (m1 != m2) {
            /*
             * Both tasks and vm mappings are different.
             * Replace the old one with the new one.
             */
            if (m1) {
                vm_uninstall(m1);
            }
            vm_install(m2);
        }
        TASK = THREAD->task;   
    }
 
    THREAD->state = Running;
 
    #ifdef SCHEDULER_VERBOSE
    printf("cpu%d: tid %d (pri=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->pri, THREAD->ticks, CPU->nrdy);
    #endif  
 
    context_restore(&THREAD->saved_context);
    /* not reached */
}
 
 
#ifdef __SMP__
/** Load balancing thread
 *
 * SMP load balancing thread, supervising thread supplies
 * for the CPU it's wired to.
 *
 * @param arg Generic thread argument (unused).
 *
 */
void kcpulb(void *arg)
{
    thread_t *t;
    int count, i, j, k = 0;
    pri_t pri;
 
loop:
    /*
     * Sleep until there's some work to do.
     */
    waitq_sleep(&CPU->kcpulb_wq);
 
not_satisfied:
    /*
     * Calculate the number of threads that will be migrated/stolen from
     * other CPU's. Note that situation can have changed between two
     * passes. Each time get the most up to date counts.
     */
    pri = cpu_priority_high();
    spinlock_lock(&CPU->lock);
    count = nrdy / config.cpu_active;
    count -= CPU->nrdy;
    spinlock_unlock(&CPU->lock);
    cpu_priority_restore(pri);
 
    if (count <= 0)
        goto satisfied;
 
    /*
     * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
     */
    for (j=RQ_COUNT-1; j >= 0; j--) {
        for (i=0; i < config.cpu_active; i++) {
            link_t *l;
            runq_t *r;
            cpu_t *cpu;
 
            cpu = &cpus[(i + k) % config.cpu_active];
 
            /*
             * Not interested in ourselves.
             * Doesn't require interrupt disabling for kcpulb is X_WIRED.
             */
            if (CPU == cpu)
                continue;              
 
restart:        pri = cpu_priority_high();
            r = &cpu->rq[j];
            spinlock_lock(&r->lock);
            if (r->n == 0) {
                spinlock_unlock(&r->lock);
                cpu_priority_restore(pri);
                continue;
            }
       
            t = NULL;
            l = r->rq_head.prev;    /* search rq from the back */
            while (l != &r->rq_head) {
                t = list_get_instance(l, thread_t, rq_link);
                /*
                 * We don't want to steal CPU-wired threads neither threads already stolen.
                 * The latter prevents threads from migrating between CPU's without ever being run.
                 * We don't want to steal threads whose FPU context is still in CPU.
                 */
                spinlock_lock(&t->lock);
                if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
               
                    /*
                     * Remove t from r.
                     */
 
                    spinlock_unlock(&t->lock);
                   
                    /*
                     * Here we have to avoid deadlock with relink_rq(),
                     * because it locks cpu and r in a different order than we do.
                     */
                    if (!spinlock_trylock(&cpu->lock)) {
                        /* Release all locks and try again. */
                        spinlock_unlock(&r->lock);
                        cpu_priority_restore(pri);
                        goto restart;
                    }
                    cpu->nrdy--;
                    spinlock_unlock(&cpu->lock);
 
                    atomic_dec(&nrdy);
 
                    r->n--;
                    list_remove(&t->rq_link);
 
                    break;
                }
                spinlock_unlock(&t->lock);
                l = l->prev;
                t = NULL;
            }
            spinlock_unlock(&r->lock);
 
            if (t) {
                /*
                 * Ready t on local CPU
                 */
                spinlock_lock(&t->lock);
                #ifdef KCPULB_VERBOSE
                printf("kcpulb%d: TID %d -> cpu%d, nrdy=%d, avg=%d\n", CPU->id, t->tid, CPU->id, CPU->nrdy, nrdy / config.cpu_active);
                #endif
                t->flags |= X_STOLEN;
                spinlock_unlock(&t->lock);
   
                thread_ready(t);
 
                cpu_priority_restore(pri);
   
                if (--count == 0)
                    goto satisfied;
                   
                /*
                 * We are not satisfied yet, focus on another CPU next time.
                 */
                k++;
               
                continue;
            }
            cpu_priority_restore(pri);
        }
    }
 
    if (CPU->nrdy) {
        /*
         * Be a little bit light-weight and let migrated threads run.
         */
        scheduler();
    }
    else {
        /*
         * We failed to migrate a single thread.
         * Something more sophisticated should be done.
         */
        scheduler();
    }
       
    goto not_satisfied;
 
satisfied:
    /*
     * Tell find_best_thread() to wake us up later again.
     */
    CPU->kcpulbstarted = 0;
    goto loop;
}
 
#endif /* __SMP__ */
 

Rev 118	Rev 125
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 <cpu.h>	32	#include <cpu.h>
33	#include <mm/vm.h>	33	#include <mm/vm.h>
34	#include <config.h>	34	#include <config.h>
35	#include <context.h>	35	#include <context.h>
36	#include <func.h>	36	#include <func.h>
37	#include <arch.h>	37	#include <arch.h>
38	#include <arch/asm.h>	38	#include <arch/asm.h>
39	#include <list.h>	39	#include <list.h>
40	#include <panic.h>	40	#include <panic.h>
41	#include <typedefs.h>	41	#include <typedefs.h>
42	#include <mm/page.h>	42	#include <mm/page.h>
43	#include <synch/spinlock.h>	43	#include <synch/spinlock.h>
44	#include <arch/faddr.h>	44	#include <arch/faddr.h>
45	#include <arch/atomic.h>	45	#include <arch/atomic.h>
46		46
47	volatile int nrdy;	47	volatile int nrdy;
48		48
49		49
50	/** Take actions before new thread runs	50	/** Take actions before new thread runs
51	*	51	*
52	* Perform actions that need to be	52	* Perform actions that need to be
53	* taken before the newly selected	53	* taken before the newly selected
54	* tread is passed control.	54	* tread is passed control.
55	*	55	*
56	*/	56	*/
57	void before_thread_runs(void)	57	void before_thread_runs(void)
58	{	58	{
59	before_thread_runs_arch();	59	before_thread_runs_arch();
60	fpu_context_restore(&(THREAD->saved_fpu_context));	60	fpu_context_restore(&(THREAD->saved_fpu_context));
61	}	61	}
62		62
63		63
64	/** Initialize scheduler	64	/** Initialize scheduler
65	*	65	*
66	* Initialize kernel scheduler.	66	* Initialize kernel scheduler.
67	*	67	*
68	*/	68	*/
69	void scheduler_init(void)	69	void scheduler_init(void)
70	{	70	{
71	}	71	}
72		72
73		73
74	/** Get thread to be scheduled	74	/** Get thread to be scheduled
75	*	75	*
76	* Get the optimal thread to be scheduled	76	* Get the optimal thread to be scheduled
77	* according to thread accounting and scheduler	77	* according to thread accounting and scheduler
78	* policy.	78	* policy.
79	*	79	*
80	* @return Thread to be scheduled.	80	* @return Thread to be scheduled.
81	*	81	*
82	*/	82	*/
83	struct thread *find_best_thread(void)	83	struct thread *find_best_thread(void)
84	{	84	{
85	thread_t *t;	85	thread_t *t;
86	runq_t *r;	86	runq_t *r;
87	int i, n;	87	int i, n;
88		88
89	loop:	89	loop:
90	cpu_priority_high();	90	cpu_priority_high();
91		91
92	spinlock_lock(&CPU->lock);	92	spinlock_lock(&CPU->lock);
93	n = CPU->nrdy;	93	n = CPU->nrdy;
94	spinlock_unlock(&CPU->lock);	94	spinlock_unlock(&CPU->lock);
95		95
96	cpu_priority_low();	96	cpu_priority_low();
97		97
98	if (n == 0) {	98	if (n == 0) {
99	#ifdef __SMP__	99	#ifdef __SMP__
100	/*	100	/*
101	* If the load balancing thread is not running, wake it up and	101	* If the load balancing thread is not running, wake it up and
102	* set CPU-private flag that the kcpulb has been started.	102	* set CPU-private flag that the kcpulb has been started.
103	*/	103	*/
104	if (test_and_set(&CPU->kcpulbstarted) == 0) {	104	if (test_and_set(&CPU->kcpulbstarted) == 0) {
105	waitq_wakeup(&CPU->kcpulb_wq, 0);	105	waitq_wakeup(&CPU->kcpulb_wq, 0);
106	goto loop;	106	goto loop;
107	}	107	}
108	#endif /* __SMP__ */	108	#endif /* __SMP__ */
109		109
110	/*	110	/*
111	* For there was nothing to run, the CPU goes to sleep	111	* For there was nothing to run, the CPU goes to sleep
112	* until a hardware interrupt or an IPI comes.	112	* until a hardware interrupt or an IPI comes.
113	* This improves energy saving and hyperthreading.	113	* This improves energy saving and hyperthreading.
114	* On the other hand, several hardware interrupts can be ignored.	114	* On the other hand, several hardware interrupts can be ignored.
115	*/	115	*/
116	cpu_sleep();	116	cpu_sleep();
117	goto loop;	117	goto loop;
118	}	118	}
119		119
120	cpu_priority_high();	120	cpu_priority_high();
121		121
122	i = 0;	122	i = 0;
123	retry:	123	retry:
124	for (; i<RQ_COUNT; i++) {	124	for (; i<RQ_COUNT; i++) {
125	r = &CPU->rq[i];	125	r = &CPU->rq[i];
126	spinlock_lock(&r->lock);	126	spinlock_lock(&r->lock);
127	if (r->n == 0) {	127	if (r->n == 0) {
128	/*	128	/*
129	* If this queue is empty, try a lower-priority queue.	129	* If this queue is empty, try a lower-priority queue.
130	*/	130	*/
131	spinlock_unlock(&r->lock);	131	spinlock_unlock(&r->lock);
132	continue;	132	continue;
133	}	133	}
134		134
135	/* avoid deadlock with relink_rq() */	135	/* avoid deadlock with relink_rq() */
136	if (!spinlock_trylock(&CPU->lock)) {	136	if (!spinlock_trylock(&CPU->lock)) {
137	/*	137	/*
138	* Unlock r and try again.	138	* Unlock r and try again.
139	*/	139	*/
140	spinlock_unlock(&r->lock);	140	spinlock_unlock(&r->lock);
141	goto retry;	141	goto retry;
142	}	142	}
143	CPU->nrdy--;	143	CPU->nrdy--;
144	spinlock_unlock(&CPU->lock);	144	spinlock_unlock(&CPU->lock);
145		145
146	atomic_dec(&nrdy);	146	atomic_dec(&nrdy);
147	r->n--;	147	r->n--;
148		148
149	/*	149	/*
150	* Take the first thread from the queue.	150	* Take the first thread from the queue.
151	*/	151	*/
152	t = list_get_instance(r->rq_head.next, thread_t, rq_link);	152	t = list_get_instance(r->rq_head.next, thread_t, rq_link);
153	list_remove(&t->rq_link);	153	list_remove(&t->rq_link);
154		154
155	spinlock_unlock(&r->lock);	155	spinlock_unlock(&r->lock);
156		156
157	spinlock_lock(&t->lock);	157	spinlock_lock(&t->lock);
158	t->cpu = CPU;	158	t->cpu = CPU;
159		159
160	t->ticks = us2ticks((i+1)*10000);	160	t->ticks = us2ticks((i+1)*10000);
161	t->pri = i; /* eventually correct rq index */	161	t->pri = i; /* eventually correct rq index */
162		162
163	/*	163	/*
164	* Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.	164	* Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.
165	*/	165	*/
166	t->flags &= ~X_STOLEN;	166	t->flags &= ~X_STOLEN;
167	spinlock_unlock(&t->lock);	167	spinlock_unlock(&t->lock);
168		168
169	return t;	169	return t;
170	}	170	}
171	goto loop;	171	goto loop;
172		172
173	}	173	}
174		174
175		175
176	/** Prevent rq starvation	176	/** Prevent rq starvation
177	*	177	*
178	* Prevent low priority threads from starving in rq's.	178	* Prevent low priority threads from starving in rq's.
179	*	179	*
180	* When the function decides to relink rq's, it reconnects	180	* When the function decides to relink rq's, it reconnects
181	* respective pointers so that in result threads with 'pri'	181	* respective pointers so that in result threads with 'pri'
182	* greater or equal 'start' are moved to a higher-priority queue.	182	* greater or equal 'start' are moved to a higher-priority queue.
183	*	183	*
184	* @param start Threshold priority.	184	* @param start Threshold priority.
185	*	185	*
186	*/	186	*/
187	void relink_rq(int start)	187	void relink_rq(int start)
188	{	188	{
189	link_t head;	189	link_t head;
190	runq_t *r;	190	runq_t *r;
191	int i, n;	191	int i, n;
192		192
193	list_initialize(&head);	193	list_initialize(&head);
194	spinlock_lock(&CPU->lock);	194	spinlock_lock(&CPU->lock);
195	if (CPU->needs_relink > NEEDS_RELINK_MAX) {	195	if (CPU->needs_relink > NEEDS_RELINK_MAX) {
196	for (i = start; i<RQ_COUNT-1; i++) {	196	for (i = start; i<RQ_COUNT-1; i++) {
197	/* remember and empty rq[i + 1] */	197	/* remember and empty rq[i + 1] */
198	r = &CPU->rq[i + 1];	198	r = &CPU->rq[i + 1];
199	spinlock_lock(&r->lock);	199	spinlock_lock(&r->lock);
200	list_concat(&head, &r->rq_head);	200	list_concat(&head, &r->rq_head);
201	n = r->n;	201	n = r->n;
202	r->n = 0;	202	r->n = 0;
203	spinlock_unlock(&r->lock);	203	spinlock_unlock(&r->lock);
204		204
205	/* append rq[i + 1] to rq[i] */	205	/* append rq[i + 1] to rq[i] */
206	r = &CPU->rq[i];	206	r = &CPU->rq[i];
207	spinlock_lock(&r->lock);	207	spinlock_lock(&r->lock);
208	list_concat(&r->rq_head, &head);	208	list_concat(&r->rq_head, &head);
209	r->n += n;	209	r->n += n;
210	spinlock_unlock(&r->lock);	210	spinlock_unlock(&r->lock);
211	}	211	}
212	CPU->needs_relink = 0;	212	CPU->needs_relink = 0;
213	}	213	}
214	spinlock_unlock(&CPU->lock);	214	spinlock_unlock(&CPU->lock);
215		215
216	}	216	}
217		217
218		218
219	/** The scheduler	219	/** The scheduler
220	*	220	*
221	* The thread scheduling procedure.	221	* The thread scheduling procedure.
222	*	222	*
223	*/	223	*/
224	void scheduler(void)	224	void scheduler(void)
225	{	225	{
226	volatile pri_t pri;	226	volatile pri_t pri;
227		227
228	pri = cpu_priority_high();	228	pri = cpu_priority_high();
229		229
230	if (haltstate)	230	if (haltstate)
231	halt();	231	halt();
232		232
233	if (THREAD) {	233	if (THREAD) {
234	spinlock_lock(&THREAD->lock);	234	spinlock_lock(&THREAD->lock);
235	fpu_context_save(&(THREAD->saved_fpu_context));	235	fpu_context_save(&(THREAD->saved_fpu_context));
236	if (!context_save(&THREAD->saved_context)) {	236	if (!context_save(&THREAD->saved_context)) {
237	/*	237	/*
238	* This is the place where threads leave scheduler();	238	* This is the place where threads leave scheduler();
239	*/	239	*/
240	before_thread_runs();	240	before_thread_runs();
241	spinlock_unlock(&THREAD->lock);	241	spinlock_unlock(&THREAD->lock);
242	cpu_priority_restore(THREAD->saved_context.pri);	242	cpu_priority_restore(THREAD->saved_context.pri);
243	return;	243	return;
244	}	244	}
245	THREAD->saved_context.pri = pri;	245	THREAD->saved_context.pri = pri;
246	}	246	}
247		247
248	/*	248	/*
249	* We may not keep the old stack.	249	* We may not keep the old stack.
250	* Reason: If we kept the old stack and got blocked, for instance, in	250	* Reason: If we kept the old stack and got blocked, for instance, in
251	* find_best_thread(), the old thread could get rescheduled by another	251	* find_best_thread(), the old thread could get rescheduled by another
252	* CPU and overwrite the part of its own stack that was also used by	252	* CPU and overwrite the part of its own stack that was also used by
253	* the scheduler on this CPU.	253	* the scheduler on this CPU.
254	*	254	*
255	* Moreover, we have to bypass the compiler-generated POP sequence	255	* Moreover, we have to bypass the compiler-generated POP sequence
256	* which is fooled by SP being set to the very top of the stack.	256	* which is fooled by SP being set to the very top of the stack.
257	* Therefore the scheduler() function continues in	257	* Therefore the scheduler() function continues in
258	* scheduler_separated_stack().	258	* scheduler_separated_stack().
259	*/	259	*/
260	context_save(&CPU->saved_context);	260	context_save(&CPU->saved_context);
261	context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), CPU->stack, CPU_STACK_SIZE);	261	context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), CPU->stack, CPU_STACK_SIZE);
262	context_restore(&CPU->saved_context);	262	context_restore(&CPU->saved_context);
263	/* not reached */	263	/* not reached */
264	}	264	}
265		265
266		266
267	/** Scheduler stack switch wrapper	267	/** Scheduler stack switch wrapper
268	*	268	*
269	* Second part of the scheduler() function	269	* Second part of the scheduler() function
270	* using new stack. Handling the actual context	270	* using new stack. Handling the actual context
271	* switch to a new thread.	271	* switch to a new thread.
272	*	272	*
273	*/	273	*/
274	void scheduler_separated_stack(void)	274	void scheduler_separated_stack(void)
275	{	275	{
276	int priority;	276	int priority;
277		277
278	if (THREAD) {	278	if (THREAD) {
279	switch (THREAD->state) {	279	switch (THREAD->state) {
280	case Running:	280	case Running:
281	THREAD->state = Ready;	281	THREAD->state = Ready;
282	spinlock_unlock(&THREAD->lock);	282	spinlock_unlock(&THREAD->lock);
283	thread_ready(THREAD);	283	thread_ready(THREAD);
284	break;	284	break;
285		285
286	case Exiting:	286	case Exiting:
287	frame_free((__address) THREAD->kstack);	287	frame_free((__address) THREAD->kstack);
288	if (THREAD->ustack) {	288	if (THREAD->ustack) {
289	frame_free((__address) THREAD->ustack);	289	frame_free((__address) THREAD->ustack);
290	}	290	}
291		291
292	/*	292	/*
293	* Detach from the containing task.	293	* Detach from the containing task.
294	*/	294	*/
295	spinlock_lock(&TASK->lock);	295	spinlock_lock(&TASK->lock);
296	list_remove(&THREAD->th_link);	296	list_remove(&THREAD->th_link);
297	spinlock_unlock(&TASK->lock);	297	spinlock_unlock(&TASK->lock);
298		298
299	spinlock_unlock(&THREAD->lock);	299	spinlock_unlock(&THREAD->lock);
300		300
301	spinlock_lock(&threads_lock);	301	spinlock_lock(&threads_lock);
302	list_remove(&THREAD->threads_link);	302	list_remove(&THREAD->threads_link);
303	spinlock_unlock(&threads_lock);	303	spinlock_unlock(&threads_lock);
304		304
305	spinlock_lock(&CPU->lock);	305	spinlock_lock(&CPU->lock);
306	if(CPU->fpu_owner==THREAD) CPU->fpu_owner=NULL;	306	if(CPU->fpu_owner==THREAD) CPU->fpu_owner=NULL;
307	spinlock_unlock(&CPU->lock);	307	spinlock_unlock(&CPU->lock);
308		308
309		-
310	free(THREAD);	309	free(THREAD);
311		310
312	break;	311	break;
313		312
314	case Sleeping:	313	case Sleeping:
315	/*	314	/*
316	* Prefer the thread after it's woken up.	315	* Prefer the thread after it's woken up.
317	*/	316	*/
318	THREAD->pri = -1;	317	THREAD->pri = -1;
319		318
320	/*	319	/*
321	* We need to release wq->lock which we locked in waitq_sleep().	320	* We need to release wq->lock which we locked in waitq_sleep().
322	* Address of wq->lock is kept in THREAD->sleep_queue.	321	* Address of wq->lock is kept in THREAD->sleep_queue.
323	*/	322	*/
324	spinlock_unlock(&THREAD->sleep_queue->lock);	323	spinlock_unlock(&THREAD->sleep_queue->lock);
325		324
326	/*	325	/*
327	* Check for possible requests for out-of-context invocation.	326	* Check for possible requests for out-of-context invocation.
328	*/	327	*/
329	if (THREAD->call_me) {	328	if (THREAD->call_me) {
330	THREAD->call_me(THREAD->call_me_with);	329	THREAD->call_me(THREAD->call_me_with);
331	THREAD->call_me = NULL;	330	THREAD->call_me = NULL;
332	THREAD->call_me_with = NULL;	331	THREAD->call_me_with = NULL;
333	}	332	}
334		333
335	spinlock_unlock(&THREAD->lock);	334	spinlock_unlock(&THREAD->lock);
336		335
337	break;	336	break;
338		337
339	default:	338	default:
340	/*	339	/*
341	* Entering state is unexpected.	340	* Entering state is unexpected.
342	*/	341	*/
343	panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);	342	panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
344	break;	343	break;
345	}	344	}
346	THREAD = NULL;	345	THREAD = NULL;
347	}	346	}
348		347
349	THREAD = find_best_thread();	348	THREAD = find_best_thread();
350		349
351	spinlock_lock(&THREAD->lock);	350	spinlock_lock(&THREAD->lock);
352	priority = THREAD->pri;	351	priority = THREAD->pri;
353	spinlock_unlock(&THREAD->lock);	352	spinlock_unlock(&THREAD->lock);
354		353
355	relink_rq(priority);	354	relink_rq(priority);
356		355
357	spinlock_lock(&THREAD->lock);	356	spinlock_lock(&THREAD->lock);
358		357
359	/*	358	/*
360	* If both the old and the new task are the same, lots of work is avoided.	359	* If both the old and the new task are the same, lots of work is avoided.
361	*/	360	*/
362	if (TASK != THREAD->task) {	361	if (TASK != THREAD->task) {
363	vm_t *m1 = NULL;	362	vm_t *m1 = NULL;
364	vm_t *m2;	363	vm_t *m2;
365		364
366	if (TASK) {	365	if (TASK) {
367	spinlock_lock(&TASK->lock);	366	spinlock_lock(&TASK->lock);
368	m1 = TASK->vm;	367	m1 = TASK->vm;
369	spinlock_unlock(&TASK->lock);	368	spinlock_unlock(&TASK->lock);
370	}	369	}
371		370
372	spinlock_lock(&THREAD->task->lock);	371	spinlock_lock(&THREAD->task->lock);
373	m2 = THREAD->task->vm;	372	m2 = THREAD->task->vm;
374	spinlock_unlock(&THREAD->task->lock);	373	spinlock_unlock(&THREAD->task->lock);
375		374
376	/*	375	/*
377	* Note that it is possible for two tasks to share one vm mapping.	376	* Note that it is possible for two tasks to share one vm mapping.
378	*/	377	*/
379	if (m1 != m2) {	378	if (m1 != m2) {
380	/*	379	/*
381	* Both tasks and vm mappings are different.	380	* Both tasks and vm mappings are different.
382	* Replace the old one with the new one.	381	* Replace the old one with the new one.
383	*/	382	*/
384	if (m1) {	383	if (m1) {
385	vm_uninstall(m1);	384	vm_uninstall(m1);
386	}	385	}
387	vm_install(m2);	386	vm_install(m2);
388	}	387	}
389	TASK = THREAD->task;	388	TASK = THREAD->task;
390	}	389	}
391		390
392	THREAD->state = Running;	391	THREAD->state = Running;
393		392
394	#ifdef SCHEDULER_VERBOSE	393	#ifdef SCHEDULER_VERBOSE
395	printf("cpu%d: tid %d (pri=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->pri, THREAD->ticks, CPU->nrdy);	394	printf("cpu%d: tid %d (pri=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->pri, THREAD->ticks, CPU->nrdy);
396	#endif	395	#endif
397		396
398	context_restore(&THREAD->saved_context);	397	context_restore(&THREAD->saved_context);
399	/* not reached */	398	/* not reached */
400	}	399	}
401		400
402		401
403	#ifdef __SMP__	402	#ifdef __SMP__
404	/** Load balancing thread	403	/** Load balancing thread
405	*	404	*
406	* SMP load balancing thread, supervising thread supplies	405	* SMP load balancing thread, supervising thread supplies
407	* for the CPU it's wired to.	406	* for the CPU it's wired to.
408	*	407	*
409	* @param arg Generic thread argument (unused).	408	* @param arg Generic thread argument (unused).
410	*	409	*
411	*/	410	*/
412	void kcpulb(void *arg)	411	void kcpulb(void *arg)
413	{	412	{
414	thread_t *t;	413	thread_t *t;
415	int count, i, j, k = 0;	414	int count, i, j, k = 0;
416	pri_t pri;	415	pri_t pri;
417		416
418	loop:	417	loop:
419	/*	418	/*
420	* Sleep until there's some work to do.	419	* Sleep until there's some work to do.
421	*/	420	*/
422	waitq_sleep(&CPU->kcpulb_wq);	421	waitq_sleep(&CPU->kcpulb_wq);
423		422
424	not_satisfied:	423	not_satisfied:
425	/*	424	/*
426	* Calculate the number of threads that will be migrated/stolen from	425	* Calculate the number of threads that will be migrated/stolen from
427	* other CPU's. Note that situation can have changed between two	426	* other CPU's. Note that situation can have changed between two
428	* passes. Each time get the most up to date counts.	427	* passes. Each time get the most up to date counts.
429	*/	428	*/
430	pri = cpu_priority_high();	429	pri = cpu_priority_high();
431	spinlock_lock(&CPU->lock);	430	spinlock_lock(&CPU->lock);
432	count = nrdy / config.cpu_active;	431	count = nrdy / config.cpu_active;
433	count -= CPU->nrdy;	432	count -= CPU->nrdy;
434	spinlock_unlock(&CPU->lock);	433	spinlock_unlock(&CPU->lock);
435	cpu_priority_restore(pri);	434	cpu_priority_restore(pri);
436		435
437	if (count <= 0)	436	if (count <= 0)
438	goto satisfied;	437	goto satisfied;
439		438
440	/*	439	/*
441	* Searching least priority queues on all CPU's first and most priority queues on all CPU's last.	440	* Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
442	*/	441	*/
443	for (j=RQ_COUNT-1; j >= 0; j--) {	442	for (j=RQ_COUNT-1; j >= 0; j--) {
444	for (i=0; i < config.cpu_active; i++) {	443	for (i=0; i < config.cpu_active; i++) {
445	link_t *l;	444	link_t *l;
446	runq_t *r;	445	runq_t *r;
447	cpu_t *cpu;	446	cpu_t *cpu;
448		447
449	cpu = &cpus[(i + k) % config.cpu_active];	448	cpu = &cpus[(i + k) % config.cpu_active];
450		449
451	/*	450	/*
452	* Not interested in ourselves.	451	* Not interested in ourselves.
453	* Doesn't require interrupt disabling for kcpulb is X_WIRED.	452	* Doesn't require interrupt disabling for kcpulb is X_WIRED.
454	*/	453	*/
455	if (CPU == cpu)	454	if (CPU == cpu)
456	continue;	455	continue;
457		456
458	restart: pri = cpu_priority_high();	457	restart: pri = cpu_priority_high();
459	r = &cpu->rq[j];	458	r = &cpu->rq[j];
460	spinlock_lock(&r->lock);	459	spinlock_lock(&r->lock);
461	if (r->n == 0) {	460	if (r->n == 0) {
462	spinlock_unlock(&r->lock);	461	spinlock_unlock(&r->lock);
463	cpu_priority_restore(pri);	462	cpu_priority_restore(pri);
464	continue;	463	continue;
465	}	464	}
466		465
467	t = NULL;	466	t = NULL;
468	l = r->rq_head.prev; /* search rq from the back */	467	l = r->rq_head.prev; /* search rq from the back */
469	while (l != &r->rq_head) {	468	while (l != &r->rq_head) {
470	t = list_get_instance(l, thread_t, rq_link);	469	t = list_get_instance(l, thread_t, rq_link);
471	/*	470	/*
472	* We don't want to steal CPU-wired threads neither threads already stolen.	471	* We don't want to steal CPU-wired threads neither threads already stolen.
473	* The latter prevents threads from migrating between CPU's without ever being run.	472	* The latter prevents threads from migrating between CPU's without ever being run.
474	* We don't want to steal threads whose FPU context is still in CPU.	473	* We don't want to steal threads whose FPU context is still in CPU.
475	*/	474	*/
476	spinlock_lock(&t->lock);	475	spinlock_lock(&t->lock);
477	if ( (!(t->flags & (X_WIRED \| X_STOLEN))) && (!(t->fpu_context_engaged)) ) {	476	if ( (!(t->flags & (X_WIRED \| X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
478		477
479	/*	478	/*
480	* Remove t from r.	479	* Remove t from r.
481	*/	480	*/
482		481
483	spinlock_unlock(&t->lock);	482	spinlock_unlock(&t->lock);
484		483
485	/*	484	/*
486	* Here we have to avoid deadlock with relink_rq(),	485	* Here we have to avoid deadlock with relink_rq(),
487	* because it locks cpu and r in a different order than we do.	486	* because it locks cpu and r in a different order than we do.
488	*/	487	*/
489	if (!spinlock_trylock(&cpu->lock)) {	488	if (!spinlock_trylock(&cpu->lock)) {
490	/* Release all locks and try again. */	489	/* Release all locks and try again. */
491	spinlock_unlock(&r->lock);	490	spinlock_unlock(&r->lock);
492	cpu_priority_restore(pri);	491	cpu_priority_restore(pri);
493	goto restart;	492	goto restart;
494	}	493	}
495	cpu->nrdy--;	494	cpu->nrdy--;
496	spinlock_unlock(&cpu->lock);	495	spinlock_unlock(&cpu->lock);
497		496
498	atomic_dec(&nrdy);	497	atomic_dec(&nrdy);
499		498
500	r->n--;	499	r->n--;

Subversion Repositories HelenOS-historic

(root)/kernel/trunk/generic/src/proc/scheduler.c @ 1757 – Rev 118 → 125