Subversion Repositories HelenOS

Rev

Rev 787 | Rev 814 | 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/heap.h>
  33. #include <mm/frame.h>
  34. #include <mm/page.h>
  35. #include <mm/as.h>
  36. #include <arch/asm.h>
  37. #include <arch/faddr.h>
  38. #include <arch/atomic.h>
  39. #include <synch/spinlock.h>
  40. #include <config.h>
  41. #include <context.h>
  42. #include <func.h>
  43. #include <arch.h>
  44. #include <adt/list.h>
  45. #include <panic.h>
  46. #include <typedefs.h>
  47. #include <cpu.h>
  48. #include <print.h>
  49. #include <debug.h>
  50.  
  51. atomic_t nrdy;
  52.  
  53. /** Take actions before new thread runs
  54.  *
  55.  * Perform actions that need to be
  56.  * taken before the newly selected
  57.  * tread is passed control.
  58.  *
  59.  */
  60. void before_thread_runs(void)
  61. {
  62.     before_thread_runs_arch();
  63. #ifdef CONFIG_FPU_LAZY
  64.     if(THREAD==CPU->fpu_owner)
  65.         fpu_enable();
  66.     else
  67.         fpu_disable();
  68. #else
  69.     fpu_enable();
  70.     if (THREAD->fpu_context_exists)
  71.         fpu_context_restore(&(THREAD->saved_fpu_context));
  72.     else {
  73.         fpu_init();
  74.         THREAD->fpu_context_exists=1;
  75.     }
  76. #endif
  77. }
  78.  
  79. #ifdef CONFIG_FPU_LAZY
  80. void scheduler_fpu_lazy_request(void)
  81. {
  82.     fpu_enable();
  83.     if (CPU->fpu_owner != NULL) {  
  84.         fpu_context_save(&CPU->fpu_owner->saved_fpu_context);
  85.         /* don't prevent migration */
  86.         CPU->fpu_owner->fpu_context_engaged=0;
  87.     }
  88.     if (THREAD->fpu_context_exists)
  89.         fpu_context_restore(&THREAD->saved_fpu_context);
  90.     else {
  91.         fpu_init();
  92.         THREAD->fpu_context_exists=1;
  93.     }
  94.     CPU->fpu_owner=THREAD;
  95.     THREAD->fpu_context_engaged = 1;
  96. }
  97. #endif
  98.  
  99. /** Initialize scheduler
  100.  *
  101.  * Initialize kernel scheduler.
  102.  *
  103.  */
  104. void scheduler_init(void)
  105. {
  106. }
  107.  
  108.  
  109. /** Get thread to be scheduled
  110.  *
  111.  * Get the optimal thread to be scheduled
  112.  * according to thread accounting and scheduler
  113.  * policy.
  114.  *
  115.  * @return Thread to be scheduled.
  116.  *
  117.  */
  118. static thread_t *find_best_thread(void)
  119. {
  120.     thread_t *t;
  121.     runq_t *r;
  122.     int i;
  123.  
  124.     ASSERT(CPU != NULL);
  125.  
  126. loop:
  127.     interrupts_enable();
  128.    
  129.     if (atomic_get(&CPU->nrdy) == 0) {
  130.         /*
  131.          * For there was nothing to run, the CPU goes to sleep
  132.          * until a hardware interrupt or an IPI comes.
  133.          * This improves energy saving and hyperthreading.
  134.          */
  135.  
  136.         /*
  137.          * An interrupt might occur right now and wake up a thread.
  138.          * In such case, the CPU will continue to go to sleep
  139.          * even though there is a runnable thread.
  140.          */
  141.  
  142.          cpu_sleep();
  143.          goto loop;
  144.     }
  145.  
  146.     interrupts_disable();
  147.    
  148.     i = 0;
  149.     for (; i<RQ_COUNT; i++) {
  150.         r = &CPU->rq[i];
  151.         spinlock_lock(&r->lock);
  152.         if (r->n == 0) {
  153.             /*
  154.              * If this queue is empty, try a lower-priority queue.
  155.              */
  156.             spinlock_unlock(&r->lock);
  157.             continue;
  158.         }
  159.  
  160.         atomic_dec(&CPU->nrdy);
  161.         atomic_dec(&nrdy);
  162.         r->n--;
  163.  
  164.         /*
  165.          * Take the first thread from the queue.
  166.          */
  167.         t = list_get_instance(r->rq_head.next, thread_t, rq_link);
  168.         list_remove(&t->rq_link);
  169.  
  170.         spinlock_unlock(&r->lock);
  171.  
  172.         spinlock_lock(&t->lock);
  173.         t->cpu = CPU;
  174.  
  175.         t->ticks = us2ticks((i+1)*10000);
  176.         t->priority = i;    /* eventually correct rq index */
  177.  
  178.         /*
  179.          * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.
  180.          */
  181.         t->flags &= ~X_STOLEN;
  182.         spinlock_unlock(&t->lock);
  183.  
  184.         return t;
  185.     }
  186.     goto loop;
  187.  
  188. }
  189.  
  190.  
  191. /** Prevent rq starvation
  192.  *
  193.  * Prevent low priority threads from starving in rq's.
  194.  *
  195.  * When the function decides to relink rq's, it reconnects
  196.  * respective pointers so that in result threads with 'pri'
  197.  * greater or equal 'start' are moved to a higher-priority queue.
  198.  *
  199.  * @param start Threshold priority.
  200.  *
  201.  */
  202. static void relink_rq(int start)
  203. {
  204.     link_t head;
  205.     runq_t *r;
  206.     int i, n;
  207.  
  208.     list_initialize(&head);
  209.     spinlock_lock(&CPU->lock);
  210.     if (CPU->needs_relink > NEEDS_RELINK_MAX) {
  211.         for (i = start; i<RQ_COUNT-1; i++) {
  212.             /* remember and empty rq[i + 1] */
  213.             r = &CPU->rq[i + 1];
  214.             spinlock_lock(&r->lock);
  215.             list_concat(&head, &r->rq_head);
  216.             n = r->n;
  217.             r->n = 0;
  218.             spinlock_unlock(&r->lock);
  219.        
  220.             /* append rq[i + 1] to rq[i] */
  221.             r = &CPU->rq[i];
  222.             spinlock_lock(&r->lock);
  223.             list_concat(&r->rq_head, &head);
  224.             r->n += n;
  225.             spinlock_unlock(&r->lock);
  226.         }
  227.         CPU->needs_relink = 0;
  228.     }
  229.     spinlock_unlock(&CPU->lock);
  230.  
  231. }
  232.  
  233.  
  234. /** Scheduler stack switch wrapper
  235.  *
  236.  * Second part of the scheduler() function
  237.  * using new stack. Handling the actual context
  238.  * switch to a new thread.
  239.  *
  240.  * Assume THREAD->lock is held.
  241.  */
  242. static void scheduler_separated_stack(void)
  243. {
  244.     int priority;
  245.  
  246.     ASSERT(CPU != NULL);
  247.  
  248.     if (THREAD) {
  249.         switch (THREAD->state) {
  250.             case Running:
  251.             THREAD->state = Ready;
  252.             spinlock_unlock(&THREAD->lock);
  253.             thread_ready(THREAD);
  254.             break;
  255.  
  256.             case Exiting:
  257.             thread_destroy(THREAD);
  258.             break;
  259.            
  260.             case Sleeping:
  261.             /*
  262.              * Prefer the thread after it's woken up.
  263.              */
  264.             THREAD->priority = -1;
  265.  
  266.             /*
  267.              * We need to release wq->lock which we locked in waitq_sleep().
  268.              * Address of wq->lock is kept in THREAD->sleep_queue.
  269.              */
  270.             spinlock_unlock(&THREAD->sleep_queue->lock);
  271.  
  272.             /*
  273.              * Check for possible requests for out-of-context invocation.
  274.              */
  275.             if (THREAD->call_me) {
  276.                 THREAD->call_me(THREAD->call_me_with);
  277.                 THREAD->call_me = NULL;
  278.                 THREAD->call_me_with = NULL;
  279.             }
  280.  
  281.             spinlock_unlock(&THREAD->lock);
  282.  
  283.             break;
  284.  
  285.             default:
  286.             /*
  287.              * Entering state is unexpected.
  288.              */
  289.             panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
  290.             break;
  291.         }
  292.         THREAD = NULL;
  293.     }
  294.  
  295.  
  296.     THREAD = find_best_thread();
  297.    
  298.     spinlock_lock(&THREAD->lock);
  299.     priority = THREAD->priority;
  300.     spinlock_unlock(&THREAD->lock);
  301.  
  302.     relink_rq(priority);       
  303.  
  304.     spinlock_lock(&THREAD->lock);  
  305.  
  306.     /*
  307.      * If both the old and the new task are the same, lots of work is avoided.
  308.      */
  309.     if (TASK != THREAD->task) {
  310.         as_t *as1 = NULL;
  311.         as_t *as2;
  312.  
  313.         if (TASK) {
  314.             spinlock_lock(&TASK->lock);
  315.             as1 = TASK->as;
  316.             spinlock_unlock(&TASK->lock);
  317.         }
  318.  
  319.         spinlock_lock(&THREAD->task->lock);
  320.         as2 = THREAD->task->as;
  321.         spinlock_unlock(&THREAD->task->lock);
  322.        
  323.         /*
  324.          * Note that it is possible for two tasks to share one address space.
  325.          */
  326.         if (as1 != as2) {
  327.             /*
  328.              * Both tasks and address spaces are different.
  329.              * Replace the old one with the new one.
  330.              */
  331.             as_install(as2);
  332.         }
  333.         TASK = THREAD->task;   
  334.     }
  335.  
  336.     THREAD->state = Running;
  337.  
  338.     #ifdef SCHEDULER_VERBOSE
  339.     printf("cpu%d: tid %d (priority=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks, CPU->nrdy);
  340.     #endif 
  341.  
  342.     /*
  343.      * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack.
  344.      */
  345.     the_copy(THE, (the_t *) THREAD->kstack);
  346.    
  347.     context_restore(&THREAD->saved_context);
  348.     /* not reached */
  349. }
  350.  
  351.  
  352. /** The scheduler
  353.  *
  354.  * The thread scheduling procedure.
  355.  * Passes control directly to
  356.  * scheduler_separated_stack().
  357.  *
  358.  */
  359. void scheduler(void)
  360. {
  361.     volatile ipl_t ipl;
  362.  
  363.     ASSERT(CPU != NULL);
  364.  
  365.     ipl = interrupts_disable();
  366.  
  367.     if (atomic_get(&haltstate))
  368.         halt();
  369.  
  370.     if (THREAD) {
  371.         spinlock_lock(&THREAD->lock);
  372. #ifndef CONFIG_FPU_LAZY
  373.         fpu_context_save(&(THREAD->saved_fpu_context));
  374. #endif
  375.         if (!context_save(&THREAD->saved_context)) {
  376.             /*
  377.              * This is the place where threads leave scheduler();
  378.              */
  379.             before_thread_runs();
  380.             spinlock_unlock(&THREAD->lock);
  381.             interrupts_restore(THREAD->saved_context.ipl);
  382.             return;
  383.         }
  384.  
  385.         /*
  386.          * Interrupt priority level of preempted thread is recorded here
  387.          * to facilitate scheduler() invocations from interrupts_disable()'d
  388.          * code (e.g. waitq_sleep_timeout()).
  389.          */
  390.         THREAD->saved_context.ipl = ipl;
  391.     }
  392.  
  393.     /*
  394.      * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM
  395.      * and preemption counter. At this point THE could be coming either
  396.      * from THREAD's or CPU's stack.
  397.      */
  398.     the_copy(THE, (the_t *) CPU->stack);
  399.  
  400.     /*
  401.      * We may not keep the old stack.
  402.      * Reason: If we kept the old stack and got blocked, for instance, in
  403.      * find_best_thread(), the old thread could get rescheduled by another
  404.      * CPU and overwrite the part of its own stack that was also used by
  405.      * the scheduler on this CPU.
  406.      *
  407.      * Moreover, we have to bypass the compiler-generated POP sequence
  408.      * which is fooled by SP being set to the very top of the stack.
  409.      * Therefore the scheduler() function continues in
  410.      * scheduler_separated_stack().
  411.      */
  412.     context_save(&CPU->saved_context);
  413.     context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), (__address) CPU->stack, CPU_STACK_SIZE);
  414.     context_restore(&CPU->saved_context);
  415.     /* not reached */
  416. }
  417.  
  418.  
  419.  
  420.  
  421.  
  422. #ifdef CONFIG_SMP
  423. /** Load balancing thread
  424.  *
  425.  * SMP load balancing thread, supervising thread supplies
  426.  * for the CPU it's wired to.
  427.  *
  428.  * @param arg Generic thread argument (unused).
  429.  *
  430.  */
  431. void kcpulb(void *arg)
  432. {
  433.     thread_t *t;
  434.     int count, average, i, j, k = 0;
  435.     ipl_t ipl;
  436.  
  437. loop:
  438.     /*
  439.      * Work in 1s intervals.
  440.      */
  441.     thread_sleep(1);
  442.  
  443. not_satisfied:
  444.     /*
  445.      * Calculate the number of threads that will be migrated/stolen from
  446.      * other CPU's. Note that situation can have changed between two
  447.      * passes. Each time get the most up to date counts.
  448.      */
  449.     average = atomic_get(&nrdy) / config.cpu_active + 1;
  450.     count = average - atomic_get(&CPU->nrdy);
  451.  
  452.     if (count <= 0)
  453.         goto satisfied;
  454.  
  455.     /*
  456.      * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
  457.      */
  458.     for (j=RQ_COUNT-1; j >= 0; j--) {
  459.         for (i=0; i < config.cpu_active; i++) {
  460.             link_t *l;
  461.             runq_t *r;
  462.             cpu_t *cpu;
  463.  
  464.             cpu = &cpus[(i + k) % config.cpu_active];
  465.  
  466.             /*
  467.              * Not interested in ourselves.
  468.              * Doesn't require interrupt disabling for kcpulb is X_WIRED.
  469.              */
  470.             if (CPU == cpu)
  471.                 continue;
  472.             if (atomic_get(&cpu->nrdy) <= average)
  473.                 continue;
  474.  
  475.             ipl = interrupts_disable();
  476.             r = &cpu->rq[j];
  477.             spinlock_lock(&r->lock);
  478.             if (r->n == 0) {
  479.                 spinlock_unlock(&r->lock);
  480.                 interrupts_restore(ipl);
  481.                 continue;
  482.             }
  483.        
  484.             t = NULL;
  485.             l = r->rq_head.prev;    /* search rq from the back */
  486.             while (l != &r->rq_head) {
  487.                 t = list_get_instance(l, thread_t, rq_link);
  488.                 /*
  489.                  * We don't want to steal CPU-wired threads neither threads already stolen.
  490.                  * The latter prevents threads from migrating between CPU's without ever being run.
  491.                  * We don't want to steal threads whose FPU context is still in CPU.
  492.                  */
  493.                 spinlock_lock(&t->lock);
  494.                 if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
  495.                     /*
  496.                      * Remove t from r.
  497.                      */
  498.                     spinlock_unlock(&t->lock);
  499.                    
  500.                     atomic_dec(&cpu->nrdy);
  501.                     atomic_dec(&nrdy);
  502.  
  503.                     r->n--;
  504.                     list_remove(&t->rq_link);
  505.  
  506.                     break;
  507.                 }
  508.                 spinlock_unlock(&t->lock);
  509.                 l = l->prev;
  510.                 t = NULL;
  511.             }
  512.             spinlock_unlock(&r->lock);
  513.  
  514.             if (t) {
  515.                 /*
  516.                  * Ready t on local CPU
  517.                  */
  518.                 spinlock_lock(&t->lock);
  519.                 #ifdef KCPULB_VERBOSE
  520.                 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);
  521.                 #endif
  522.                 t->flags |= X_STOLEN;
  523.                 spinlock_unlock(&t->lock);
  524.    
  525.                 thread_ready(t);
  526.  
  527.                 interrupts_restore(ipl);
  528.    
  529.                 if (--count == 0)
  530.                     goto satisfied;
  531.                    
  532.                 /*
  533.                  * We are not satisfied yet, focus on another CPU next time.
  534.                  */
  535.                 k++;
  536.                
  537.                 continue;
  538.             }
  539.             interrupts_restore(ipl);
  540.         }
  541.     }
  542.  
  543.     if (atomic_get(&CPU->nrdy)) {
  544.         /*
  545.          * Be a little bit light-weight and let migrated threads run.
  546.          */
  547.         scheduler();
  548.     } else {
  549.         /*
  550.          * We failed to migrate a single thread.
  551.          * Give up this turn.
  552.          */
  553.         goto loop;
  554.     }
  555.        
  556.     goto not_satisfied;
  557.  
  558. satisfied:
  559.     goto loop;
  560. }
  561.  
  562. #endif /* CONFIG_SMP */
  563.  
  564.  
  565. /** Print information about threads & scheduler queues */
  566. void sched_print_list(void)
  567. {
  568.     ipl_t ipl;
  569.     int cpu,i;
  570.     runq_t *r;
  571.     thread_t *t;
  572.     link_t *cur;
  573.  
  574.     /* We are going to mess with scheduler structures,
  575.      * let's not be interrupted */
  576.     ipl = interrupts_disable();
  577.     printf("*********** Scheduler dump ***********\n");
  578.     for (cpu=0;cpu < config.cpu_count; cpu++) {
  579.         if (!cpus[cpu].active)
  580.             continue;
  581.         spinlock_lock(&cpus[cpu].lock);
  582.         printf("cpu%d: nrdy: %d needs_relink: %d\n",
  583.                cpus[cpu].id, atomic_get(&cpus[cpu].nrdy), cpus[cpu].needs_relink);
  584.        
  585.         for (i=0; i<RQ_COUNT; i++) {
  586.             r = &cpus[cpu].rq[i];
  587.             spinlock_lock(&r->lock);
  588.             if (!r->n) {
  589.                 spinlock_unlock(&r->lock);
  590.                 continue;
  591.             }
  592.             printf("\tRq %d: ", i);
  593.             for (cur=r->rq_head.next; cur!=&r->rq_head; cur=cur->next) {
  594.                 t = list_get_instance(cur, thread_t, rq_link);
  595.                 printf("%d(%s) ", t->tid,
  596.                        thread_states[t->state]);
  597.             }
  598.             printf("\n");
  599.             spinlock_unlock(&r->lock);
  600.         }
  601.         spinlock_unlock(&cpus[cpu].lock);
  602.     }
  603.    
  604.     interrupts_restore(ipl);
  605. }
  606.