Subversion Repositories HelenOS-historic

Rev

Rev 788 | Rev 827 | 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.  */
  59. void before_thread_runs(void)
  60. {
  61.     before_thread_runs_arch();
  62. #ifdef CONFIG_FPU_LAZY
  63.     if(THREAD==CPU->fpu_owner)
  64.         fpu_enable();
  65.     else
  66.         fpu_disable();
  67. #else
  68.     fpu_enable();
  69.     if (THREAD->fpu_context_exists)
  70.         fpu_context_restore(&(THREAD->saved_fpu_context));
  71.     else {
  72.         fpu_init();
  73.         THREAD->fpu_context_exists=1;
  74.     }
  75. #endif
  76. }
  77.  
  78. #ifdef CONFIG_FPU_LAZY
  79. void scheduler_fpu_lazy_request(void)
  80. {
  81.     fpu_enable();
  82.     if (CPU->fpu_owner != NULL) {  
  83.         fpu_context_save(&CPU->fpu_owner->saved_fpu_context);
  84.         /* don't prevent migration */
  85.         CPU->fpu_owner->fpu_context_engaged=0;
  86.     }
  87.     if (THREAD->fpu_context_exists)
  88.         fpu_context_restore(&THREAD->saved_fpu_context);
  89.     else {
  90.         fpu_init();
  91.         THREAD->fpu_context_exists=1;
  92.     }
  93.     CPU->fpu_owner=THREAD;
  94.     THREAD->fpu_context_engaged = 1;
  95. }
  96. #endif
  97.  
  98. /** Initialize scheduler
  99.  *
  100.  * Initialize kernel scheduler.
  101.  *
  102.  */
  103. void scheduler_init(void)
  104. {
  105. }
  106.  
  107.  
  108. /** Get thread to be scheduled
  109.  *
  110.  * Get the optimal thread to be scheduled
  111.  * according to thread accounting and scheduler
  112.  * policy.
  113.  *
  114.  * @return Thread to be scheduled.
  115.  *
  116.  */
  117. static thread_t *find_best_thread(void)
  118. {
  119.     thread_t *t;
  120.     runq_t *r;
  121.     int i;
  122.  
  123.     ASSERT(CPU != NULL);
  124.  
  125. loop:
  126.     interrupts_enable();
  127.    
  128.     if (atomic_get(&CPU->nrdy) == 0) {
  129.         /*
  130.          * For there was nothing to run, the CPU goes to sleep
  131.          * until a hardware interrupt or an IPI comes.
  132.          * This improves energy saving and hyperthreading.
  133.          */
  134.  
  135.         /*
  136.          * An interrupt might occur right now and wake up a thread.
  137.          * In such case, the CPU will continue to go to sleep
  138.          * even though there is a runnable thread.
  139.          */
  140.  
  141.          cpu_sleep();
  142.          goto loop;
  143.     }
  144.  
  145.     interrupts_disable();
  146.    
  147.     i = 0;
  148.     for (; i<RQ_COUNT; i++) {
  149.         r = &CPU->rq[i];
  150.         spinlock_lock(&r->lock);
  151.         if (r->n == 0) {
  152.             /*
  153.              * If this queue is empty, try a lower-priority queue.
  154.              */
  155.             spinlock_unlock(&r->lock);
  156.             continue;
  157.         }
  158.  
  159.         atomic_dec(&CPU->nrdy);
  160.         atomic_dec(&nrdy);
  161.         r->n--;
  162.  
  163.         /*
  164.          * Take the first thread from the queue.
  165.          */
  166.         t = list_get_instance(r->rq_head.next, thread_t, rq_link);
  167.         list_remove(&t->rq_link);
  168.  
  169.         spinlock_unlock(&r->lock);
  170.  
  171.         spinlock_lock(&t->lock);
  172.         t->cpu = CPU;
  173.  
  174.         t->ticks = us2ticks((i+1)*10000);
  175.         t->priority = i;    /* eventually correct rq index */
  176.  
  177.         /*
  178.          * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.
  179.          */
  180.         t->flags &= ~X_STOLEN;
  181.         spinlock_unlock(&t->lock);
  182.  
  183.         return t;
  184.     }
  185.     goto loop;
  186.  
  187. }
  188.  
  189.  
  190. /** Prevent rq starvation
  191.  *
  192.  * Prevent low priority threads from starving in rq's.
  193.  *
  194.  * When the function decides to relink rq's, it reconnects
  195.  * respective pointers so that in result threads with 'pri'
  196.  * greater or equal 'start' are moved to a higher-priority queue.
  197.  *
  198.  * @param start Threshold priority.
  199.  *
  200.  */
  201. static void relink_rq(int start)
  202. {
  203.     link_t head;
  204.     runq_t *r;
  205.     int i, n;
  206.  
  207.     list_initialize(&head);
  208.     spinlock_lock(&CPU->lock);
  209.     if (CPU->needs_relink > NEEDS_RELINK_MAX) {
  210.         for (i = start; i<RQ_COUNT-1; i++) {
  211.             /* remember and empty rq[i + 1] */
  212.             r = &CPU->rq[i + 1];
  213.             spinlock_lock(&r->lock);
  214.             list_concat(&head, &r->rq_head);
  215.             n = r->n;
  216.             r->n = 0;
  217.             spinlock_unlock(&r->lock);
  218.        
  219.             /* append rq[i + 1] to rq[i] */
  220.             r = &CPU->rq[i];
  221.             spinlock_lock(&r->lock);
  222.             list_concat(&r->rq_head, &head);
  223.             r->n += n;
  224.             spinlock_unlock(&r->lock);
  225.         }
  226.         CPU->needs_relink = 0;
  227.     }
  228.     spinlock_unlock(&CPU->lock);
  229.  
  230. }
  231.  
  232.  
  233. /** Scheduler stack switch wrapper
  234.  *
  235.  * Second part of the scheduler() function
  236.  * using new stack. Handling the actual context
  237.  * switch to a new thread.
  238.  *
  239.  * Assume THREAD->lock is held.
  240.  */
  241. static void scheduler_separated_stack(void)
  242. {
  243.     int priority;
  244.  
  245.     ASSERT(CPU != NULL);
  246.  
  247.     if (THREAD) {
  248.         switch (THREAD->state) {
  249.             case Running:
  250.             THREAD->state = Ready;
  251.             spinlock_unlock(&THREAD->lock);
  252.             thread_ready(THREAD);
  253.             break;
  254.  
  255.             case Exiting:
  256.             thread_destroy(THREAD);
  257.             break;
  258.            
  259.             case Sleeping:
  260.             /*
  261.              * Prefer the thread after it's woken up.
  262.              */
  263.             THREAD->priority = -1;
  264.  
  265.             /*
  266.              * We need to release wq->lock which we locked in waitq_sleep().
  267.              * Address of wq->lock is kept in THREAD->sleep_queue.
  268.              */
  269.             spinlock_unlock(&THREAD->sleep_queue->lock);
  270.  
  271.             /*
  272.              * Check for possible requests for out-of-context invocation.
  273.              */
  274.             if (THREAD->call_me) {
  275.                 THREAD->call_me(THREAD->call_me_with);
  276.                 THREAD->call_me = NULL;
  277.                 THREAD->call_me_with = NULL;
  278.             }
  279.  
  280.             spinlock_unlock(&THREAD->lock);
  281.  
  282.             break;
  283.  
  284.             default:
  285.             /*
  286.              * Entering state is unexpected.
  287.              */
  288.             panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
  289.             break;
  290.         }
  291.         THREAD = NULL;
  292.     }
  293.  
  294.  
  295.     THREAD = find_best_thread();
  296.    
  297.     spinlock_lock(&THREAD->lock);
  298.     priority = THREAD->priority;
  299.     spinlock_unlock(&THREAD->lock);
  300.  
  301.     relink_rq(priority);       
  302.  
  303.     spinlock_lock(&THREAD->lock);  
  304.  
  305.     /*
  306.      * If both the old and the new task are the same, lots of work is avoided.
  307.      */
  308.     if (TASK != THREAD->task) {
  309.         as_t *as1 = NULL;
  310.         as_t *as2;
  311.  
  312.         if (TASK) {
  313.             spinlock_lock(&TASK->lock);
  314.             as1 = TASK->as;
  315.             spinlock_unlock(&TASK->lock);
  316.         }
  317.  
  318.         spinlock_lock(&THREAD->task->lock);
  319.         as2 = THREAD->task->as;
  320.         spinlock_unlock(&THREAD->task->lock);
  321.        
  322.         /*
  323.          * Note that it is possible for two tasks to share one address space.
  324.          */
  325.         if (as1 != as2) {
  326.             /*
  327.              * Both tasks and address spaces are different.
  328.              * Replace the old one with the new one.
  329.              */
  330.             as_install(as2);
  331.         }
  332.         TASK = THREAD->task;   
  333.     }
  334.  
  335.     THREAD->state = Running;
  336.  
  337.     #ifdef SCHEDULER_VERBOSE
  338.     printf("cpu%d: tid %d (priority=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks, CPU->nrdy);
  339.     #endif 
  340.  
  341.     /*
  342.      * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack.
  343.      */
  344.     the_copy(THE, (the_t *) THREAD->kstack);
  345.    
  346.     context_restore(&THREAD->saved_context);
  347.     /* not reached */
  348. }
  349.  
  350.  
  351. /** The scheduler
  352.  *
  353.  * The thread scheduling procedure.
  354.  * Passes control directly to
  355.  * scheduler_separated_stack().
  356.  *
  357.  */
  358. void scheduler(void)
  359. {
  360.     volatile ipl_t ipl;
  361.  
  362.     ASSERT(CPU != NULL);
  363.  
  364.     ipl = interrupts_disable();
  365.  
  366.     if (atomic_get(&haltstate))
  367.         halt();
  368.  
  369.     if (THREAD) {
  370.         spinlock_lock(&THREAD->lock);
  371. #ifndef CONFIG_FPU_LAZY
  372.         fpu_context_save(&(THREAD->saved_fpu_context));
  373. #endif
  374.         if (!context_save(&THREAD->saved_context)) {
  375.             /*
  376.              * This is the place where threads leave scheduler();
  377.              */
  378.             before_thread_runs();
  379.             spinlock_unlock(&THREAD->lock);
  380.             interrupts_restore(THREAD->saved_context.ipl);
  381.             return;
  382.         }
  383.  
  384.         /*
  385.          * Interrupt priority level of preempted thread is recorded here
  386.          * to facilitate scheduler() invocations from interrupts_disable()'d
  387.          * code (e.g. waitq_sleep_timeout()).
  388.          */
  389.         THREAD->saved_context.ipl = ipl;
  390.     }
  391.  
  392.     /*
  393.      * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM
  394.      * and preemption counter. At this point THE could be coming either
  395.      * from THREAD's or CPU's stack.
  396.      */
  397.     the_copy(THE, (the_t *) CPU->stack);
  398.  
  399.     /*
  400.      * We may not keep the old stack.
  401.      * Reason: If we kept the old stack and got blocked, for instance, in
  402.      * find_best_thread(), the old thread could get rescheduled by another
  403.      * CPU and overwrite the part of its own stack that was also used by
  404.      * the scheduler on this CPU.
  405.      *
  406.      * Moreover, we have to bypass the compiler-generated POP sequence
  407.      * which is fooled by SP being set to the very top of the stack.
  408.      * Therefore the scheduler() function continues in
  409.      * scheduler_separated_stack().
  410.      */
  411.     context_save(&CPU->saved_context);
  412.     context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), (__address) CPU->stack, CPU_STACK_SIZE);
  413.     context_restore(&CPU->saved_context);
  414.     /* not reached */
  415. }
  416.  
  417.  
  418.  
  419.  
  420.  
  421. #ifdef CONFIG_SMP
  422. /** Load balancing thread
  423.  *
  424.  * SMP load balancing thread, supervising thread supplies
  425.  * for the CPU it's wired to.
  426.  *
  427.  * @param arg Generic thread argument (unused).
  428.  *
  429.  */
  430. void kcpulb(void *arg)
  431. {
  432.     thread_t *t;
  433.     int count, average, i, j, k = 0;
  434.     ipl_t ipl;
  435.  
  436. loop:
  437.     /*
  438.      * Work in 1s intervals.
  439.      */
  440.     thread_sleep(1);
  441.  
  442. not_satisfied:
  443.     /*
  444.      * Calculate the number of threads that will be migrated/stolen from
  445.      * other CPU's. Note that situation can have changed between two
  446.      * passes. Each time get the most up to date counts.
  447.      */
  448.     average = atomic_get(&nrdy) / config.cpu_active + 1;
  449.     count = average - atomic_get(&CPU->nrdy);
  450.  
  451.     if (count <= 0)
  452.         goto satisfied;
  453.  
  454.     /*
  455.      * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
  456.      */
  457.     for (j=RQ_COUNT-1; j >= 0; j--) {
  458.         for (i=0; i < config.cpu_active; i++) {
  459.             link_t *l;
  460.             runq_t *r;
  461.             cpu_t *cpu;
  462.  
  463.             cpu = &cpus[(i + k) % config.cpu_active];
  464.  
  465.             /*
  466.              * Not interested in ourselves.
  467.              * Doesn't require interrupt disabling for kcpulb is X_WIRED.
  468.              */
  469.             if (CPU == cpu)
  470.                 continue;
  471.             if (atomic_get(&cpu->nrdy) <= average)
  472.                 continue;
  473.  
  474.             ipl = interrupts_disable();
  475.             r = &cpu->rq[j];
  476.             spinlock_lock(&r->lock);
  477.             if (r->n == 0) {
  478.                 spinlock_unlock(&r->lock);
  479.                 interrupts_restore(ipl);
  480.                 continue;
  481.             }
  482.        
  483.             t = NULL;
  484.             l = r->rq_head.prev;    /* search rq from the back */
  485.             while (l != &r->rq_head) {
  486.                 t = list_get_instance(l, thread_t, rq_link);
  487.                 /*
  488.                  * We don't want to steal CPU-wired threads neither threads already stolen.
  489.                  * The latter prevents threads from migrating between CPU's without ever being run.
  490.                  * We don't want to steal threads whose FPU context is still in CPU.
  491.                  */
  492.                 spinlock_lock(&t->lock);
  493.                 if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
  494.                     /*
  495.                      * Remove t from r.
  496.                      */
  497.                     spinlock_unlock(&t->lock);
  498.                    
  499.                     atomic_dec(&cpu->nrdy);
  500.                     atomic_dec(&nrdy);
  501.  
  502.                     r->n--;
  503.                     list_remove(&t->rq_link);
  504.  
  505.                     break;
  506.                 }
  507.                 spinlock_unlock(&t->lock);
  508.                 l = l->prev;
  509.                 t = NULL;
  510.             }
  511.             spinlock_unlock(&r->lock);
  512.  
  513.             if (t) {
  514.                 /*
  515.                  * Ready t on local CPU
  516.                  */
  517.                 spinlock_lock(&t->lock);
  518.                 #ifdef KCPULB_VERBOSE
  519.                 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);
  520.                 #endif
  521.                 t->flags |= X_STOLEN;
  522.                 spinlock_unlock(&t->lock);
  523.    
  524.                 thread_ready(t);
  525.  
  526.                 interrupts_restore(ipl);
  527.    
  528.                 if (--count == 0)
  529.                     goto satisfied;
  530.                    
  531.                 /*
  532.                  * We are not satisfied yet, focus on another CPU next time.
  533.                  */
  534.                 k++;
  535.                
  536.                 continue;
  537.             }
  538.             interrupts_restore(ipl);
  539.         }
  540.     }
  541.  
  542.     if (atomic_get(&CPU->nrdy)) {
  543.         /*
  544.          * Be a little bit light-weight and let migrated threads run.
  545.          */
  546.         scheduler();
  547.     } else {
  548.         /*
  549.          * We failed to migrate a single thread.
  550.          * Give up this turn.
  551.          */
  552.         goto loop;
  553.     }
  554.        
  555.     goto not_satisfied;
  556.  
  557. satisfied:
  558.     goto loop;
  559. }
  560.  
  561. #endif /* CONFIG_SMP */
  562.  
  563.  
  564. /** Print information about threads & scheduler queues */
  565. void sched_print_list(void)
  566. {
  567.     ipl_t ipl;
  568.     int cpu,i;
  569.     runq_t *r;
  570.     thread_t *t;
  571.     link_t *cur;
  572.  
  573.     /* We are going to mess with scheduler structures,
  574.      * let's not be interrupted */
  575.     ipl = interrupts_disable();
  576.     printf("*********** Scheduler dump ***********\n");
  577.     for (cpu=0;cpu < config.cpu_count; cpu++) {
  578.         if (!cpus[cpu].active)
  579.             continue;
  580.         spinlock_lock(&cpus[cpu].lock);
  581.         printf("cpu%d: nrdy: %d needs_relink: %d\n",
  582.                cpus[cpu].id, atomic_get(&cpus[cpu].nrdy), cpus[cpu].needs_relink);
  583.        
  584.         for (i=0; i<RQ_COUNT; i++) {
  585.             r = &cpus[cpu].rq[i];
  586.             spinlock_lock(&r->lock);
  587.             if (!r->n) {
  588.                 spinlock_unlock(&r->lock);
  589.                 continue;
  590.             }
  591.             printf("\tRq %d: ", i);
  592.             for (cur=r->rq_head.next; cur!=&r->rq_head; cur=cur->next) {
  593.                 t = list_get_instance(cur, thread_t, rq_link);
  594.                 printf("%d(%s) ", t->tid,
  595.                        thread_states[t->state]);
  596.             }
  597.             printf("\n");
  598.             spinlock_unlock(&r->lock);
  599.         }
  600.         spinlock_unlock(&cpus[cpu].lock);
  601.     }
  602.    
  603.     interrupts_restore(ipl);
  604. }
  605.