Subversion Repositories HelenOS

Rev

Rev 898 | Rev 907 | Go to most recent revision | Blame | Compare with Previous | Last modification | View Log | Download | RSS feed

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