Subversion Repositories HelenOS

Rev

Rev 775 | Rev 783 | 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 <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, n;
  123.  
  124.     ASSERT(CPU != NULL);
  125.  
  126. loop:
  127.     interrupts_disable();
  128.  
  129.     spinlock_lock(&CPU->lock);
  130.     n = CPU->nrdy;
  131.     spinlock_unlock(&CPU->lock);
  132.  
  133.     interrupts_enable();
  134.    
  135.     if (n == 0) {
  136.         /*
  137.          * For there was nothing to run, the CPU goes to sleep
  138.          * until a hardware interrupt or an IPI comes.
  139.          * This improves energy saving and hyperthreading.
  140.          */
  141.          cpu_sleep();
  142.          goto loop;
  143.     }
  144.  
  145.     interrupts_disable();
  146.    
  147.     i = 0;
  148. retry:
  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.         /* avoid deadlock with relink_rq() */
  161.         if (!spinlock_trylock(&CPU->lock)) {
  162.             /*
  163.              * Unlock r and try again.
  164.              */
  165.             spinlock_unlock(&r->lock);
  166.             goto retry;
  167.         }
  168.         CPU->nrdy--;
  169.         spinlock_unlock(&CPU->lock);
  170.  
  171.         atomic_dec(&nrdy);
  172.         r->n--;
  173.  
  174.         /*
  175.          * Take the first thread from the queue.
  176.          */
  177.         t = list_get_instance(r->rq_head.next, thread_t, rq_link);
  178.         list_remove(&t->rq_link);
  179.  
  180.         spinlock_unlock(&r->lock);
  181.  
  182.         spinlock_lock(&t->lock);
  183.         t->cpu = CPU;
  184.  
  185.         t->ticks = us2ticks((i+1)*10000);
  186.         t->priority = i;    /* eventually correct rq index */
  187.  
  188.         /*
  189.          * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.
  190.          */
  191.         t->flags &= ~X_STOLEN;
  192.         spinlock_unlock(&t->lock);
  193.  
  194.         return t;
  195.     }
  196.     goto loop;
  197.  
  198. }
  199.  
  200.  
  201. /** Prevent rq starvation
  202.  *
  203.  * Prevent low priority threads from starving in rq's.
  204.  *
  205.  * When the function decides to relink rq's, it reconnects
  206.  * respective pointers so that in result threads with 'pri'
  207.  * greater or equal 'start' are moved to a higher-priority queue.
  208.  *
  209.  * @param start Threshold priority.
  210.  *
  211.  */
  212. static void relink_rq(int start)
  213. {
  214.     link_t head;
  215.     runq_t *r;
  216.     int i, n;
  217.  
  218.     list_initialize(&head);
  219.     spinlock_lock(&CPU->lock);
  220.     if (CPU->needs_relink > NEEDS_RELINK_MAX) {
  221.         for (i = start; i<RQ_COUNT-1; i++) {
  222.             /* remember and empty rq[i + 1] */
  223.             r = &CPU->rq[i + 1];
  224.             spinlock_lock(&r->lock);
  225.             list_concat(&head, &r->rq_head);
  226.             n = r->n;
  227.             r->n = 0;
  228.             spinlock_unlock(&r->lock);
  229.        
  230.             /* append rq[i + 1] to rq[i] */
  231.             r = &CPU->rq[i];
  232.             spinlock_lock(&r->lock);
  233.             list_concat(&r->rq_head, &head);
  234.             r->n += n;
  235.             spinlock_unlock(&r->lock);
  236.         }
  237.         CPU->needs_relink = 0;
  238.     }
  239.     spinlock_unlock(&CPU->lock);               
  240.  
  241. }
  242.  
  243.  
  244. /** Scheduler stack switch wrapper
  245.  *
  246.  * Second part of the scheduler() function
  247.  * using new stack. Handling the actual context
  248.  * switch to a new thread.
  249.  *
  250.  */
  251. static void scheduler_separated_stack(void)
  252. {
  253.     int priority;
  254.  
  255.     ASSERT(CPU != NULL);
  256.  
  257.     if (THREAD) {
  258.         switch (THREAD->state) {
  259.             case Running:
  260.             THREAD->state = Ready;
  261.             spinlock_unlock(&THREAD->lock);
  262.             thread_ready(THREAD);
  263.             break;
  264.  
  265.             case Exiting:
  266.             frame_free((__address) THREAD->kstack);
  267.             if (THREAD->ustack) {
  268.                 frame_free((__address) THREAD->ustack);
  269.             }
  270.  
  271.             /*
  272.              * Detach from the containing task.
  273.              */
  274.             spinlock_lock(&TASK->lock);
  275.             list_remove(&THREAD->th_link);
  276.             spinlock_unlock(&TASK->lock);
  277.  
  278.             spinlock_unlock(&THREAD->lock);
  279.    
  280.             spinlock_lock(&threads_lock);
  281.             list_remove(&THREAD->threads_link);
  282.             spinlock_unlock(&threads_lock);
  283.  
  284.             spinlock_lock(&CPU->lock);
  285.             if(CPU->fpu_owner==THREAD)
  286.                 CPU->fpu_owner=NULL;
  287.             spinlock_unlock(&CPU->lock);
  288.  
  289.             free(THREAD);
  290.  
  291.             break;
  292.    
  293.             case Sleeping:
  294.             /*
  295.              * Prefer the thread after it's woken up.
  296.              */
  297.             THREAD->priority = -1;
  298.  
  299.             /*
  300.              * We need to release wq->lock which we locked in waitq_sleep().
  301.              * Address of wq->lock is kept in THREAD->sleep_queue.
  302.              */
  303.             spinlock_unlock(&THREAD->sleep_queue->lock);
  304.  
  305.             /*
  306.              * Check for possible requests for out-of-context invocation.
  307.              */
  308.             if (THREAD->call_me) {
  309.                 THREAD->call_me(THREAD->call_me_with);
  310.                 THREAD->call_me = NULL;
  311.                 THREAD->call_me_with = NULL;
  312.             }
  313.  
  314.             spinlock_unlock(&THREAD->lock);
  315.  
  316.             break;
  317.  
  318.             default:
  319.             /*
  320.              * Entering state is unexpected.
  321.              */
  322.             panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
  323.             break;
  324.         }
  325.         THREAD = NULL;
  326.     }
  327.  
  328.  
  329.     THREAD = find_best_thread();
  330.    
  331.     spinlock_lock(&THREAD->lock);
  332.     priority = THREAD->priority;
  333.     spinlock_unlock(&THREAD->lock);
  334.  
  335.     relink_rq(priority);       
  336.  
  337.     spinlock_lock(&THREAD->lock);  
  338.  
  339.     /*
  340.      * If both the old and the new task are the same, lots of work is avoided.
  341.      */
  342.     if (TASK != THREAD->task) {
  343.         as_t *as1 = NULL;
  344.         as_t *as2;
  345.  
  346.         if (TASK) {
  347.             spinlock_lock(&TASK->lock);
  348.             as1 = TASK->as;
  349.             spinlock_unlock(&TASK->lock);
  350.         }
  351.  
  352.         spinlock_lock(&THREAD->task->lock);
  353.         as2 = THREAD->task->as;
  354.         spinlock_unlock(&THREAD->task->lock);
  355.        
  356.         /*
  357.          * Note that it is possible for two tasks to share one address space.
  358.          */
  359.         if (as1 != as2) {
  360.             /*
  361.              * Both tasks and address spaces are different.
  362.              * Replace the old one with the new one.
  363.              */
  364.             as_install(as2);
  365.         }
  366.         TASK = THREAD->task;   
  367.     }
  368.  
  369.     THREAD->state = Running;
  370.  
  371.     #ifdef SCHEDULER_VERBOSE
  372.     printf("cpu%d: tid %d (priority=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks, CPU->nrdy);
  373.     #endif 
  374.  
  375.     /*
  376.      * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack.
  377.      */
  378.     the_copy(THE, (the_t *) THREAD->kstack);
  379.    
  380.     context_restore(&THREAD->saved_context);
  381.     /* not reached */
  382. }
  383.  
  384.  
  385. /** The scheduler
  386.  *
  387.  * The thread scheduling procedure.
  388.  * Passes control directly to
  389.  * scheduler_separated_stack().
  390.  *
  391.  */
  392. void scheduler(void)
  393. {
  394.     volatile ipl_t ipl;
  395.  
  396.     ASSERT(CPU != NULL);
  397.  
  398.     ipl = interrupts_disable();
  399.  
  400.     if (atomic_get(&haltstate))
  401.         halt();
  402.  
  403.     if (THREAD) {
  404.         spinlock_lock(&THREAD->lock);
  405. #ifndef CONFIG_FPU_LAZY
  406.         fpu_context_save(&(THREAD->saved_fpu_context));
  407. #endif
  408.         if (!context_save(&THREAD->saved_context)) {
  409.             /*
  410.              * This is the place where threads leave scheduler();
  411.              */
  412.             before_thread_runs();
  413.             spinlock_unlock(&THREAD->lock);
  414.             interrupts_restore(THREAD->saved_context.ipl);
  415.             return;
  416.         }
  417.  
  418.         /*
  419.          * Interrupt priority level of preempted thread is recorded here
  420.          * to facilitate scheduler() invocations from interrupts_disable()'d
  421.          * code (e.g. waitq_sleep_timeout()).
  422.          */
  423.         THREAD->saved_context.ipl = ipl;
  424.     }
  425.  
  426.     /*
  427.      * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM
  428.      * and preemption counter. At this point THE could be coming either
  429.      * from THREAD's or CPU's stack.
  430.      */
  431.     the_copy(THE, (the_t *) CPU->stack);
  432.  
  433.     /*
  434.      * We may not keep the old stack.
  435.      * Reason: If we kept the old stack and got blocked, for instance, in
  436.      * find_best_thread(), the old thread could get rescheduled by another
  437.      * CPU and overwrite the part of its own stack that was also used by
  438.      * the scheduler on this CPU.
  439.      *
  440.      * Moreover, we have to bypass the compiler-generated POP sequence
  441.      * which is fooled by SP being set to the very top of the stack.
  442.      * Therefore the scheduler() function continues in
  443.      * scheduler_separated_stack().
  444.      */
  445.     context_save(&CPU->saved_context);
  446.     context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), (__address) CPU->stack, CPU_STACK_SIZE);
  447.     context_restore(&CPU->saved_context);
  448.     /* not reached */
  449. }
  450.  
  451.  
  452.  
  453.  
  454.  
  455. #ifdef CONFIG_SMP
  456. /** Load balancing thread
  457.  *
  458.  * SMP load balancing thread, supervising thread supplies
  459.  * for the CPU it's wired to.
  460.  *
  461.  * @param arg Generic thread argument (unused).
  462.  *
  463.  */
  464. void kcpulb(void *arg)
  465. {
  466.     thread_t *t;
  467.     int count, i, j, k = 0;
  468.     ipl_t ipl;
  469.  
  470. loop:
  471.     /*
  472.      * Work in 1s intervals.
  473.      */
  474.     thread_sleep(1);
  475.  
  476. not_satisfied:
  477.     /*
  478.      * Calculate the number of threads that will be migrated/stolen from
  479.      * other CPU's. Note that situation can have changed between two
  480.      * passes. Each time get the most up to date counts.
  481.      */
  482.     ipl = interrupts_disable();
  483.     spinlock_lock(&CPU->lock);
  484.     count = atomic_get(&nrdy) / config.cpu_active;
  485.     count -= CPU->nrdy;
  486.     spinlock_unlock(&CPU->lock);
  487.     interrupts_restore(ipl);
  488.  
  489.     if (count <= 0)
  490.         goto satisfied;
  491.  
  492.     /*
  493.      * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
  494.      */
  495.     for (j=RQ_COUNT-1; j >= 0; j--) {
  496.         for (i=0; i < config.cpu_active; i++) {
  497.             link_t *l;
  498.             runq_t *r;
  499.             cpu_t *cpu;
  500.  
  501.             cpu = &cpus[(i + k) % config.cpu_active];
  502.  
  503.             /*
  504.              * Not interested in ourselves.
  505.              * Doesn't require interrupt disabling for kcpulb is X_WIRED.
  506.              */
  507.             if (CPU == cpu)
  508.                 continue;              
  509.  
  510. restart:        ipl = interrupts_disable();
  511.             r = &cpu->rq[j];
  512.             spinlock_lock(&r->lock);
  513.             if (r->n == 0) {
  514.                 spinlock_unlock(&r->lock);
  515.                 interrupts_restore(ipl);
  516.                 continue;
  517.             }
  518.        
  519.             t = NULL;
  520.             l = r->rq_head.prev;    /* search rq from the back */
  521.             while (l != &r->rq_head) {
  522.                 t = list_get_instance(l, thread_t, rq_link);
  523.                 /*
  524.                  * We don't want to steal CPU-wired threads neither threads already stolen.
  525.                  * The latter prevents threads from migrating between CPU's without ever being run.
  526.                  * We don't want to steal threads whose FPU context is still in CPU.
  527.                  */
  528.                 spinlock_lock(&t->lock);
  529.                 if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
  530.                
  531.                     /*
  532.                      * Remove t from r.
  533.                      */
  534.  
  535.                     spinlock_unlock(&t->lock);
  536.                    
  537.                     /*
  538.                      * Here we have to avoid deadlock with relink_rq(),
  539.                      * because it locks cpu and r in a different order than we do.
  540.                      */
  541.                     if (!spinlock_trylock(&cpu->lock)) {
  542.                         /* Release all locks and try again. */
  543.                         spinlock_unlock(&r->lock);
  544.                         interrupts_restore(ipl);
  545.                         goto restart;
  546.                     }
  547.                     cpu->nrdy--;
  548.                     spinlock_unlock(&cpu->lock);
  549.  
  550.                     atomic_dec(&nrdy);
  551.  
  552.                     r->n--;
  553.                     list_remove(&t->rq_link);
  554.  
  555.                     break;
  556.                 }
  557.                 spinlock_unlock(&t->lock);
  558.                 l = l->prev;
  559.                 t = NULL;
  560.             }
  561.             spinlock_unlock(&r->lock);
  562.  
  563.             if (t) {
  564.                 /*
  565.                  * Ready t on local CPU
  566.                  */
  567.                 spinlock_lock(&t->lock);
  568.                 #ifdef KCPULB_VERBOSE
  569.                 printf("kcpulb%d: TID %d -> cpu%d, nrdy=%d, avg=%d\n", CPU->id, t->tid, CPU->id, CPU->nrdy, atomic_get(&nrdy) / config.cpu_active);
  570.                 #endif
  571.                 t->flags |= X_STOLEN;
  572.                 spinlock_unlock(&t->lock);
  573.    
  574.                 thread_ready(t);
  575.  
  576.                 interrupts_restore(ipl);
  577.    
  578.                 if (--count == 0)
  579.                     goto satisfied;
  580.                    
  581.                 /*
  582.                  * We are not satisfied yet, focus on another CPU next time.
  583.                  */
  584.                 k++;
  585.                
  586.                 continue;
  587.             }
  588.             interrupts_restore(ipl);
  589.         }
  590.     }
  591.  
  592.     if (CPU->nrdy) {
  593.         /*
  594.          * Be a little bit light-weight and let migrated threads run.
  595.          */
  596.         scheduler();
  597.     } else {
  598.         /*
  599.          * We failed to migrate a single thread.
  600.          * Give up this turn.
  601.          */
  602.         goto loop;
  603.     }
  604.        
  605.     goto not_satisfied;
  606.  
  607. satisfied:
  608.     goto loop;
  609. }
  610.  
  611. #endif /* CONFIG_SMP */
  612.  
  613.  
  614. /** Print information about threads & scheduler queues */
  615. void sched_print_list(void)
  616. {
  617.     ipl_t ipl;
  618.     int cpu,i;
  619.     runq_t *r;
  620.     thread_t *t;
  621.     link_t *cur;
  622.  
  623.     /* We are going to mess with scheduler structures,
  624.      * let's not be interrupted */
  625.     ipl = interrupts_disable();
  626.     printf("*********** Scheduler dump ***********\n");
  627.     for (cpu=0;cpu < config.cpu_count; cpu++) {
  628.         if (!cpus[cpu].active)
  629.             continue;
  630.         spinlock_lock(&cpus[cpu].lock);
  631.         printf("cpu%d: nrdy: %d needs_relink: %d\n",
  632.                cpus[cpu].id, cpus[cpu].nrdy, cpus[cpu].needs_relink);
  633.        
  634.         for (i=0; i<RQ_COUNT; i++) {
  635.             r = &cpus[cpu].rq[i];
  636.             spinlock_lock(&r->lock);
  637.             if (!r->n) {
  638.                 spinlock_unlock(&r->lock);
  639.                 continue;
  640.             }
  641.             printf("\tRq %d: ", i);
  642.             for (cur=r->rq_head.next; cur!=&r->rq_head; cur=cur->next) {
  643.                 t = list_get_instance(cur, thread_t, rq_link);
  644.                 printf("%d(%s) ", t->tid,
  645.                        thread_states[t->state]);
  646.             }
  647.             printf("\n");
  648.             spinlock_unlock(&r->lock);
  649.         }
  650.         spinlock_unlock(&cpus[cpu].lock);
  651.     }
  652.    
  653.     interrupts_restore(ipl);
  654. }
  655.