Subversion Repositories HelenOS

Rev

Rev 823 | Rev 897 | 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. atomic_t nrdy;
  51.  
  52. /** Take actions before new thread runs
  53.  *
  54.  * Perform actions that need to be
  55.  * taken before the newly selected
  56.  * tread is passed control.
  57.  *
  58.  * THREAD->lock is locked on entry
  59.  *
  60.  */
  61. void before_thread_runs(void)
  62. {
  63.     before_thread_runs_arch();
  64. #ifdef CONFIG_FPU_LAZY
  65.     if(THREAD==CPU->fpu_owner)
  66.         fpu_enable();
  67.     else
  68.         fpu_disable();
  69. #else
  70.     fpu_enable();
  71.     if (THREAD->fpu_context_exists)
  72.         fpu_context_restore(&(THREAD->saved_fpu_context));
  73.     else {
  74.         fpu_init(&(THREAD->saved_fpu_context));
  75.         THREAD->fpu_context_exists=1;
  76.     }
  77. #endif
  78. }
  79.  
  80. #ifdef CONFIG_FPU_LAZY
  81. void scheduler_fpu_lazy_request(void)
  82. {
  83.     fpu_enable();
  84.     spinlock_lock(&CPU->lock);
  85.  
  86.     /* Save old context */
  87.     if (CPU->fpu_owner != NULL) {  
  88.         spinlock_lock(&CPU->fpu_owner->lock);
  89.         fpu_context_save(&CPU->fpu_owner->saved_fpu_context);
  90.         /* don't prevent migration */
  91.         CPU->fpu_owner->fpu_context_engaged=0;
  92.         spinlock_unlock(&CPU->fpu_owner->lock);
  93.     }
  94.  
  95.     spinlock_lock(&THREAD->lock);
  96.     if (THREAD->fpu_context_exists)
  97.         fpu_context_restore(&THREAD->saved_fpu_context);
  98.     else {
  99.         fpu_init(&(THREAD->saved_fpu_context));
  100.         THREAD->fpu_context_exists=1;
  101.     }
  102.     CPU->fpu_owner=THREAD;
  103.     THREAD->fpu_context_engaged = 1;
  104.  
  105.     spinlock_unlock(&THREAD->lock);
  106.     spinlock_unlock(&CPU->lock);
  107. }
  108. #endif
  109.  
  110. /** Initialize scheduler
  111.  *
  112.  * Initialize kernel scheduler.
  113.  *
  114.  */
  115. void scheduler_init(void)
  116. {
  117. }
  118.  
  119.  
  120. /** Get thread to be scheduled
  121.  *
  122.  * Get the optimal thread to be scheduled
  123.  * according to thread accounting and scheduler
  124.  * policy.
  125.  *
  126.  * @return Thread to be scheduled.
  127.  *
  128.  */
  129. static thread_t *find_best_thread(void)
  130. {
  131.     thread_t *t;
  132.     runq_t *r;
  133.     int i;
  134.  
  135.     ASSERT(CPU != NULL);
  136.  
  137. loop:
  138.     interrupts_enable();
  139.    
  140.     if (atomic_get(&CPU->nrdy) == 0) {
  141.         /*
  142.          * For there was nothing to run, the CPU goes to sleep
  143.          * until a hardware interrupt or an IPI comes.
  144.          * This improves energy saving and hyperthreading.
  145.          */
  146.  
  147.         /*
  148.          * An interrupt might occur right now and wake up a thread.
  149.          * In such case, the CPU will continue to go to sleep
  150.          * even though there is a runnable thread.
  151.          */
  152.  
  153.          cpu_sleep();
  154.          goto loop;
  155.     }
  156.  
  157.     interrupts_disable();
  158.    
  159.     i = 0;
  160.     for (; i<RQ_COUNT; i++) {
  161.         r = &CPU->rq[i];
  162.         spinlock_lock(&r->lock);
  163.         if (r->n == 0) {
  164.             /*
  165.              * If this queue is empty, try a lower-priority queue.
  166.              */
  167.             spinlock_unlock(&r->lock);
  168.             continue;
  169.         }
  170.  
  171.         atomic_dec(&CPU->nrdy);
  172.         atomic_dec(&nrdy);
  173.         r->n--;
  174.  
  175.         /*
  176.          * Take the first thread from the queue.
  177.          */
  178.         t = list_get_instance(r->rq_head.next, thread_t, rq_link);
  179.         list_remove(&t->rq_link);
  180.  
  181.         spinlock_unlock(&r->lock);
  182.  
  183.         spinlock_lock(&t->lock);
  184.         t->cpu = CPU;
  185.  
  186.         t->ticks = us2ticks((i+1)*10000);
  187.         t->priority = i;    /* eventually correct rq index */
  188.  
  189.         /*
  190.          * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.
  191.          */
  192.         t->flags &= ~X_STOLEN;
  193.         spinlock_unlock(&t->lock);
  194.  
  195.         return t;
  196.     }
  197.     goto loop;
  198.  
  199. }
  200.  
  201.  
  202. /** Prevent rq starvation
  203.  *
  204.  * Prevent low priority threads from starving in rq's.
  205.  *
  206.  * When the function decides to relink rq's, it reconnects
  207.  * respective pointers so that in result threads with 'pri'
  208.  * greater or equal 'start' are moved to a higher-priority queue.
  209.  *
  210.  * @param start Threshold priority.
  211.  *
  212.  */
  213. static void relink_rq(int start)
  214. {
  215.     link_t head;
  216.     runq_t *r;
  217.     int i, n;
  218.  
  219.     list_initialize(&head);
  220.     spinlock_lock(&CPU->lock);
  221.     if (CPU->needs_relink > NEEDS_RELINK_MAX) {
  222.         for (i = start; i<RQ_COUNT-1; i++) {
  223.             /* remember and empty rq[i + 1] */
  224.             r = &CPU->rq[i + 1];
  225.             spinlock_lock(&r->lock);
  226.             list_concat(&head, &r->rq_head);
  227.             n = r->n;
  228.             r->n = 0;
  229.             spinlock_unlock(&r->lock);
  230.        
  231.             /* append rq[i + 1] to rq[i] */
  232.             r = &CPU->rq[i];
  233.             spinlock_lock(&r->lock);
  234.             list_concat(&r->rq_head, &head);
  235.             r->n += n;
  236.             spinlock_unlock(&r->lock);
  237.         }
  238.         CPU->needs_relink = 0;
  239.     }
  240.     spinlock_unlock(&CPU->lock);
  241.  
  242. }
  243.  
  244.  
  245. /** Scheduler stack switch wrapper
  246.  *
  247.  * Second part of the scheduler() function
  248.  * using new stack. Handling the actual context
  249.  * switch to a new thread.
  250.  *
  251.  * Assume THREAD->lock is held.
  252.  */
  253. static void scheduler_separated_stack(void)
  254. {
  255.     int priority;
  256.  
  257.     ASSERT(CPU != NULL);
  258.  
  259.     if (THREAD) {
  260.         switch (THREAD->state) {
  261.             case Running:
  262.             THREAD->state = Ready;
  263.             spinlock_unlock(&THREAD->lock);
  264.             thread_ready(THREAD);
  265.             break;
  266.  
  267.             case Exiting:
  268.             thread_destroy(THREAD);
  269.             break;
  270.            
  271.             case Sleeping:
  272.             /*
  273.              * Prefer the thread after it's woken up.
  274.              */
  275.             THREAD->priority = -1;
  276.  
  277.             /*
  278.              * We need to release wq->lock which we locked in waitq_sleep().
  279.              * Address of wq->lock is kept in THREAD->sleep_queue.
  280.              */
  281.             spinlock_unlock(&THREAD->sleep_queue->lock);
  282.  
  283.             /*
  284.              * Check for possible requests for out-of-context invocation.
  285.              */
  286.             if (THREAD->call_me) {
  287.                 THREAD->call_me(THREAD->call_me_with);
  288.                 THREAD->call_me = NULL;
  289.                 THREAD->call_me_with = NULL;
  290.             }
  291.  
  292.             spinlock_unlock(&THREAD->lock);
  293.  
  294.             break;
  295.  
  296.             default:
  297.             /*
  298.              * Entering state is unexpected.
  299.              */
  300.             panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
  301.             break;
  302.         }
  303.         THREAD = NULL;
  304.     }
  305.  
  306.  
  307.     THREAD = find_best_thread();
  308.    
  309.     spinlock_lock(&THREAD->lock);
  310.     priority = THREAD->priority;
  311.     spinlock_unlock(&THREAD->lock);
  312.  
  313.     relink_rq(priority);       
  314.  
  315.     spinlock_lock(&THREAD->lock);  
  316.  
  317.     /*
  318.      * If both the old and the new task are the same, lots of work is avoided.
  319.      */
  320.     if (TASK != THREAD->task) {
  321.         as_t *as1 = NULL;
  322.         as_t *as2;
  323.  
  324.         if (TASK) {
  325.             spinlock_lock(&TASK->lock);
  326.             as1 = TASK->as;
  327.             spinlock_unlock(&TASK->lock);
  328.         }
  329.  
  330.         spinlock_lock(&THREAD->task->lock);
  331.         as2 = THREAD->task->as;
  332.         spinlock_unlock(&THREAD->task->lock);
  333.        
  334.         /*
  335.          * Note that it is possible for two tasks to share one address space.
  336.          */
  337.         if (as1 != as2) {
  338.             /*
  339.              * Both tasks and address spaces are different.
  340.              * Replace the old one with the new one.
  341.              */
  342.             as_switch(as1, as2);
  343.         }
  344.         TASK = THREAD->task;   
  345.     }
  346.  
  347.     THREAD->state = Running;
  348.  
  349.     #ifdef SCHEDULER_VERBOSE
  350.     printf("cpu%d: tid %d (priority=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks, atomic_get(&CPU->nrdy));
  351.     #endif 
  352.  
  353.     /*
  354.      * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack.
  355.      */
  356.     the_copy(THE, (the_t *) THREAD->kstack);
  357.    
  358.     context_restore(&THREAD->saved_context);
  359.     /* not reached */
  360. }
  361.  
  362.  
  363. /** The scheduler
  364.  *
  365.  * The thread scheduling procedure.
  366.  * Passes control directly to
  367.  * scheduler_separated_stack().
  368.  *
  369.  */
  370. void scheduler(void)
  371. {
  372.     volatile ipl_t ipl;
  373.  
  374.     ASSERT(CPU != NULL);
  375.  
  376.     ipl = interrupts_disable();
  377.  
  378.     if (atomic_get(&haltstate))
  379.         halt();
  380.  
  381.     if (THREAD) {
  382.         spinlock_lock(&THREAD->lock);
  383. #ifndef CONFIG_FPU_LAZY
  384.         fpu_context_save(&(THREAD->saved_fpu_context));
  385. #endif
  386.         if (!context_save(&THREAD->saved_context)) {
  387.             /*
  388.              * This is the place where threads leave scheduler();
  389.              */
  390.             before_thread_runs();
  391.             spinlock_unlock(&THREAD->lock);
  392.             interrupts_restore(THREAD->saved_context.ipl);
  393.             return;
  394.         }
  395.  
  396.         /*
  397.          * Interrupt priority level of preempted thread is recorded here
  398.          * to facilitate scheduler() invocations from interrupts_disable()'d
  399.          * code (e.g. waitq_sleep_timeout()).
  400.          */
  401.         THREAD->saved_context.ipl = ipl;
  402.     }
  403.  
  404.     /*
  405.      * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM
  406.      * and preemption counter. At this point THE could be coming either
  407.      * from THREAD's or CPU's stack.
  408.      */
  409.     the_copy(THE, (the_t *) CPU->stack);
  410.  
  411.     /*
  412.      * We may not keep the old stack.
  413.      * Reason: If we kept the old stack and got blocked, for instance, in
  414.      * find_best_thread(), the old thread could get rescheduled by another
  415.      * CPU and overwrite the part of its own stack that was also used by
  416.      * the scheduler on this CPU.
  417.      *
  418.      * Moreover, we have to bypass the compiler-generated POP sequence
  419.      * which is fooled by SP being set to the very top of the stack.
  420.      * Therefore the scheduler() function continues in
  421.      * scheduler_separated_stack().
  422.      */
  423.     context_save(&CPU->saved_context);
  424.     context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), (__address) CPU->stack, CPU_STACK_SIZE);
  425.     context_restore(&CPU->saved_context);
  426.     /* not reached */
  427. }
  428.  
  429.  
  430.  
  431.  
  432.  
  433. #ifdef CONFIG_SMP
  434. /** Load balancing thread
  435.  *
  436.  * SMP load balancing thread, supervising thread supplies
  437.  * for the CPU it's wired to.
  438.  *
  439.  * @param arg Generic thread argument (unused).
  440.  *
  441.  */
  442. void kcpulb(void *arg)
  443. {
  444.     thread_t *t;
  445.     int count, average, i, j, k = 0;
  446.     ipl_t ipl;
  447.  
  448. loop:
  449.     /*
  450.      * Work in 1s intervals.
  451.      */
  452.     thread_sleep(1);
  453.  
  454. not_satisfied:
  455.     /*
  456.      * Calculate the number of threads that will be migrated/stolen from
  457.      * other CPU's. Note that situation can have changed between two
  458.      * passes. Each time get the most up to date counts.
  459.      */
  460.     average = atomic_get(&nrdy) / config.cpu_active + 1;
  461.     count = average - atomic_get(&CPU->nrdy);
  462.  
  463.     if (count <= 0)
  464.         goto satisfied;
  465.  
  466.     /*
  467.      * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
  468.      */
  469.     for (j=RQ_COUNT-1; j >= 0; j--) {
  470.         for (i=0; i < config.cpu_active; i++) {
  471.             link_t *l;
  472.             runq_t *r;
  473.             cpu_t *cpu;
  474.  
  475.             cpu = &cpus[(i + k) % config.cpu_active];
  476.  
  477.             /*
  478.              * Not interested in ourselves.
  479.              * Doesn't require interrupt disabling for kcpulb is X_WIRED.
  480.              */
  481.             if (CPU == cpu)
  482.                 continue;
  483.             if (atomic_get(&cpu->nrdy) <= average)
  484.                 continue;
  485.  
  486.             ipl = interrupts_disable();
  487.             r = &cpu->rq[j];
  488.             spinlock_lock(&r->lock);
  489.             if (r->n == 0) {
  490.                 spinlock_unlock(&r->lock);
  491.                 interrupts_restore(ipl);
  492.                 continue;
  493.             }
  494.        
  495.             t = NULL;
  496.             l = r->rq_head.prev;    /* search rq from the back */
  497.             while (l != &r->rq_head) {
  498.                 t = list_get_instance(l, thread_t, rq_link);
  499.                 /*
  500.                  * We don't want to steal CPU-wired threads neither threads already stolen.
  501.                  * The latter prevents threads from migrating between CPU's without ever being run.
  502.                  * We don't want to steal threads whose FPU context is still in CPU.
  503.                  */
  504.                 spinlock_lock(&t->lock);
  505.                 if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
  506.                     /*
  507.                      * Remove t from r.
  508.                      */
  509.                     spinlock_unlock(&t->lock);
  510.                    
  511.                     atomic_dec(&cpu->nrdy);
  512.                     atomic_dec(&nrdy);
  513.  
  514.                     r->n--;
  515.                     list_remove(&t->rq_link);
  516.  
  517.                     break;
  518.                 }
  519.                 spinlock_unlock(&t->lock);
  520.                 l = l->prev;
  521.                 t = NULL;
  522.             }
  523.             spinlock_unlock(&r->lock);
  524.  
  525.             if (t) {
  526.                 /*
  527.                  * Ready t on local CPU
  528.                  */
  529.                 spinlock_lock(&t->lock);
  530.                 #ifdef KCPULB_VERBOSE
  531.                 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);
  532.                 #endif
  533.                 t->flags |= X_STOLEN;
  534.                 spinlock_unlock(&t->lock);
  535.    
  536.                 thread_ready(t);
  537.  
  538.                 interrupts_restore(ipl);
  539.    
  540.                 if (--count == 0)
  541.                     goto satisfied;
  542.                    
  543.                 /*
  544.                  * We are not satisfied yet, focus on another CPU next time.
  545.                  */
  546.                 k++;
  547.                
  548.                 continue;
  549.             }
  550.             interrupts_restore(ipl);
  551.         }
  552.     }
  553.  
  554.     if (atomic_get(&CPU->nrdy)) {
  555.         /*
  556.          * Be a little bit light-weight and let migrated threads run.
  557.          */
  558.         scheduler();
  559.     } else {
  560.         /*
  561.          * We failed to migrate a single thread.
  562.          * Give up this turn.
  563.          */
  564.         goto loop;
  565.     }
  566.        
  567.     goto not_satisfied;
  568.  
  569. satisfied:
  570.     goto loop;
  571. }
  572.  
  573. #endif /* CONFIG_SMP */
  574.  
  575.  
  576. /** Print information about threads & scheduler queues */
  577. void sched_print_list(void)
  578. {
  579.     ipl_t ipl;
  580.     int cpu,i;
  581.     runq_t *r;
  582.     thread_t *t;
  583.     link_t *cur;
  584.  
  585.     /* We are going to mess with scheduler structures,
  586.      * let's not be interrupted */
  587.     ipl = interrupts_disable();
  588.     printf("*********** Scheduler dump ***********\n");
  589.     for (cpu=0;cpu < config.cpu_count; cpu++) {
  590.         if (!cpus[cpu].active)
  591.             continue;
  592.         spinlock_lock(&cpus[cpu].lock);
  593.         printf("cpu%d: nrdy: %d needs_relink: %d\n",
  594.                cpus[cpu].id, atomic_get(&cpus[cpu].nrdy), cpus[cpu].needs_relink);
  595.        
  596.         for (i=0; i<RQ_COUNT; i++) {
  597.             r = &cpus[cpu].rq[i];
  598.             spinlock_lock(&r->lock);
  599.             if (!r->n) {
  600.                 spinlock_unlock(&r->lock);
  601.                 continue;
  602.             }
  603.             printf("\tRq %d: ", i);
  604.             for (cur=r->rq_head.next; cur!=&r->rq_head; cur=cur->next) {
  605.                 t = list_get_instance(cur, thread_t, rq_link);
  606.                 printf("%d(%s) ", t->tid,
  607.                        thread_states[t->state]);
  608.             }
  609.             printf("\n");
  610.             spinlock_unlock(&r->lock);
  611.         }
  612.         spinlock_unlock(&cpus[cpu].lock);
  613.     }
  614.    
  615.     interrupts_restore(ipl);
  616. }
  617.