Subversion Repositories HelenOS-historic

Rev

Rev 475 | Rev 501 | 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/vm.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.         #ifdef CONFIG_SMP
  137.         /*
  138.          * If the load balancing thread is not running, wake it up and
  139.          * set CPU-private flag that the kcpulb has been started.
  140.          */
  141.         if (test_and_set(&CPU->kcpulbstarted) == 0) {
  142.             waitq_wakeup(&CPU->kcpulb_wq, 0);
  143.             goto loop;
  144.         }
  145.         #endif /* CONFIG_SMP */
  146.        
  147.         /*
  148.          * For there was nothing to run, the CPU goes to sleep
  149.          * until a hardware interrupt or an IPI comes.
  150.          * This improves energy saving and hyperthreading.
  151.          * On the other hand, several hardware interrupts can be ignored.
  152.          */
  153.          cpu_sleep();
  154.          goto loop;
  155.     }
  156.  
  157.     interrupts_disable();
  158.    
  159.     i = 0;
  160. retry:
  161.     for (; i<RQ_COUNT; i++) {
  162.         r = &CPU->rq[i];
  163.         spinlock_lock(&r->lock);
  164.         if (r->n == 0) {
  165.             /*
  166.              * If this queue is empty, try a lower-priority queue.
  167.              */
  168.             spinlock_unlock(&r->lock);
  169.             continue;
  170.         }
  171.  
  172.         /* avoid deadlock with relink_rq() */
  173.         if (!spinlock_trylock(&CPU->lock)) {
  174.             /*
  175.              * Unlock r and try again.
  176.              */
  177.             spinlock_unlock(&r->lock);
  178.             goto retry;
  179.         }
  180.         CPU->nrdy--;
  181.         spinlock_unlock(&CPU->lock);
  182.  
  183.         atomic_dec(&nrdy);
  184.         r->n--;
  185.  
  186.         /*
  187.          * Take the first thread from the queue.
  188.          */
  189.         t = list_get_instance(r->rq_head.next, thread_t, rq_link);
  190.         list_remove(&t->rq_link);
  191.  
  192.         spinlock_unlock(&r->lock);
  193.  
  194.         spinlock_lock(&t->lock);
  195.         t->cpu = CPU;
  196.  
  197.         t->ticks = us2ticks((i+1)*10000);
  198.         t->priority = i;    /* eventually correct rq index */
  199.  
  200.         /*
  201.          * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.
  202.          */
  203.         t->flags &= ~X_STOLEN;
  204.         spinlock_unlock(&t->lock);
  205.  
  206.         return t;
  207.     }
  208.     goto loop;
  209.  
  210. }
  211.  
  212.  
  213. /** Prevent rq starvation
  214.  *
  215.  * Prevent low priority threads from starving in rq's.
  216.  *
  217.  * When the function decides to relink rq's, it reconnects
  218.  * respective pointers so that in result threads with 'pri'
  219.  * greater or equal 'start' are moved to a higher-priority queue.
  220.  *
  221.  * @param start Threshold priority.
  222.  *
  223.  */
  224. static void relink_rq(int start)
  225. {
  226.     link_t head;
  227.     runq_t *r;
  228.     int i, n;
  229.  
  230.     list_initialize(&head);
  231.     spinlock_lock(&CPU->lock);
  232.     if (CPU->needs_relink > NEEDS_RELINK_MAX) {
  233.         for (i = start; i<RQ_COUNT-1; i++) {
  234.             /* remember and empty rq[i + 1] */
  235.             r = &CPU->rq[i + 1];
  236.             spinlock_lock(&r->lock);
  237.             list_concat(&head, &r->rq_head);
  238.             n = r->n;
  239.             r->n = 0;
  240.             spinlock_unlock(&r->lock);
  241.        
  242.             /* append rq[i + 1] to rq[i] */
  243.             r = &CPU->rq[i];
  244.             spinlock_lock(&r->lock);
  245.             list_concat(&r->rq_head, &head);
  246.             r->n += n;
  247.             spinlock_unlock(&r->lock);
  248.         }
  249.         CPU->needs_relink = 0;
  250.     }
  251.     spinlock_unlock(&CPU->lock);               
  252.  
  253. }
  254.  
  255.  
  256. /** Scheduler stack switch wrapper
  257.  *
  258.  * Second part of the scheduler() function
  259.  * using new stack. Handling the actual context
  260.  * switch to a new thread.
  261.  *
  262.  */
  263. static void scheduler_separated_stack(void)
  264. {
  265.     int priority;
  266.  
  267.     ASSERT(CPU != NULL);
  268.  
  269.     if (THREAD) {
  270.         switch (THREAD->state) {
  271.             case Running:
  272.             THREAD->state = Ready;
  273.             spinlock_unlock(&THREAD->lock);
  274.             thread_ready(THREAD);
  275.             break;
  276.  
  277.             case Exiting:
  278.             frame_free((__address) THREAD->kstack);
  279.             if (THREAD->ustack) {
  280.                 frame_free((__address) THREAD->ustack);
  281.             }
  282.  
  283.             /*
  284.              * Detach from the containing task.
  285.              */
  286.             spinlock_lock(&TASK->lock);
  287.             list_remove(&THREAD->th_link);
  288.             spinlock_unlock(&TASK->lock);
  289.  
  290.             spinlock_unlock(&THREAD->lock);
  291.    
  292.             spinlock_lock(&threads_lock);
  293.             list_remove(&THREAD->threads_link);
  294.             spinlock_unlock(&threads_lock);
  295.  
  296.             spinlock_lock(&CPU->lock);
  297.             if(CPU->fpu_owner==THREAD) CPU->fpu_owner=NULL;
  298.             spinlock_unlock(&CPU->lock);
  299.  
  300.             free(THREAD);
  301.  
  302.             break;
  303.    
  304.             case Sleeping:
  305.             /*
  306.              * Prefer the thread after it's woken up.
  307.              */
  308.             THREAD->priority = -1;
  309.  
  310.             /*
  311.              * We need to release wq->lock which we locked in waitq_sleep().
  312.              * Address of wq->lock is kept in THREAD->sleep_queue.
  313.              */
  314.             spinlock_unlock(&THREAD->sleep_queue->lock);
  315.  
  316.             /*
  317.              * Check for possible requests for out-of-context invocation.
  318.              */
  319.             if (THREAD->call_me) {
  320.                 THREAD->call_me(THREAD->call_me_with);
  321.                 THREAD->call_me = NULL;
  322.                 THREAD->call_me_with = NULL;
  323.             }
  324.  
  325.             spinlock_unlock(&THREAD->lock);
  326.  
  327.             break;
  328.  
  329.             default:
  330.             /*
  331.              * Entering state is unexpected.
  332.              */
  333.             panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
  334.             break;
  335.         }
  336.         THREAD = NULL;
  337.     }
  338.  
  339.  
  340.     THREAD = find_best_thread();
  341.    
  342.     spinlock_lock(&THREAD->lock);
  343.     priority = THREAD->priority;
  344.     spinlock_unlock(&THREAD->lock);
  345.  
  346.     relink_rq(priority);       
  347.  
  348.     spinlock_lock(&THREAD->lock);  
  349.  
  350.     /*
  351.      * If both the old and the new task are the same, lots of work is avoided.
  352.      */
  353.     if (TASK != THREAD->task) {
  354.         vm_t *m1 = NULL;
  355.         vm_t *m2;
  356.  
  357.         if (TASK) {
  358.             spinlock_lock(&TASK->lock);
  359.             m1 = TASK->vm;
  360.             spinlock_unlock(&TASK->lock);
  361.         }
  362.  
  363.         spinlock_lock(&THREAD->task->lock);
  364.         m2 = THREAD->task->vm;
  365.         spinlock_unlock(&THREAD->task->lock);
  366.        
  367.         /*
  368.          * Note that it is possible for two tasks to share one vm mapping.
  369.          */
  370.         if (m1 != m2) {
  371.             /*
  372.              * Both tasks and vm mappings are different.
  373.              * Replace the old one with the new one.
  374.              */
  375.             vm_install(m2);
  376.         }
  377.         TASK = THREAD->task;   
  378.     }
  379.  
  380.     THREAD->state = Running;
  381.  
  382.     #ifdef SCHEDULER_VERBOSE
  383.     printf("cpu%d: tid %d (priority=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks, CPU->nrdy);
  384.     #endif 
  385.  
  386.     /*
  387.      * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack.
  388.      */
  389.     the_copy(THE, (the_t *) THREAD->kstack);
  390.    
  391.     context_restore(&THREAD->saved_context);
  392.     /* not reached */
  393. }
  394.  
  395.  
  396. /** The scheduler
  397.  *
  398.  * The thread scheduling procedure.
  399.  *
  400.  */
  401. void scheduler(void)
  402. {
  403.     volatile ipl_t ipl;
  404.  
  405.     ASSERT(CPU != NULL);
  406.  
  407.     ipl = interrupts_disable();
  408.  
  409.     if (haltstate)
  410.         halt();
  411.  
  412.     if (THREAD) {
  413.         spinlock_lock(&THREAD->lock);
  414. #ifndef CONFIG_FPU_LAZY
  415.         fpu_context_save(&(THREAD->saved_fpu_context));
  416. #endif
  417.         if (!context_save(&THREAD->saved_context)) {
  418.             /*
  419.              * This is the place where threads leave scheduler();
  420.              */
  421.             before_thread_runs();
  422.             spinlock_unlock(&THREAD->lock);
  423.             interrupts_restore(THREAD->saved_context.ipl);
  424.             return;
  425.         }
  426.  
  427.         /*
  428.          * Interrupt priority level of preempted thread is recorded here
  429.          * to facilitate scheduler() invocations from interrupts_disable()'d
  430.          * code (e.g. waitq_sleep_timeout()).
  431.          */
  432.         THREAD->saved_context.ipl = ipl;
  433.     }
  434.  
  435.     /*
  436.      * Through the 'THE' structure, we keep track of THREAD, TASK, CPU
  437.      * and preemption counter. At this point THE could be coming either
  438.      * from THREAD's or CPU's stack.
  439.      */
  440.     the_copy(THE, (the_t *) CPU->stack);
  441.  
  442.     /*
  443.      * We may not keep the old stack.
  444.      * Reason: If we kept the old stack and got blocked, for instance, in
  445.      * find_best_thread(), the old thread could get rescheduled by another
  446.      * CPU and overwrite the part of its own stack that was also used by
  447.      * the scheduler on this CPU.
  448.      *
  449.      * Moreover, we have to bypass the compiler-generated POP sequence
  450.      * which is fooled by SP being set to the very top of the stack.
  451.      * Therefore the scheduler() function continues in
  452.      * scheduler_separated_stack().
  453.      */
  454.     context_save(&CPU->saved_context);
  455.     context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), (__address) CPU->stack, CPU_STACK_SIZE);
  456.     context_restore(&CPU->saved_context);
  457.     /* not reached */
  458. }
  459.  
  460.  
  461.  
  462.  
  463.  
  464. #ifdef CONFIG_SMP
  465. /** Load balancing thread
  466.  *
  467.  * SMP load balancing thread, supervising thread supplies
  468.  * for the CPU it's wired to.
  469.  *
  470.  * @param arg Generic thread argument (unused).
  471.  *
  472.  */
  473. void kcpulb(void *arg)
  474. {
  475.     thread_t *t;
  476.     int count, i, j, k = 0;
  477.     ipl_t ipl;
  478.  
  479. loop:
  480.     /*
  481.      * Sleep until there's some work to do.
  482.      */
  483.     waitq_sleep(&CPU->kcpulb_wq);
  484.  
  485. not_satisfied:
  486.     /*
  487.      * Calculate the number of threads that will be migrated/stolen from
  488.      * other CPU's. Note that situation can have changed between two
  489.      * passes. Each time get the most up to date counts.
  490.      */
  491.     ipl = interrupts_disable();
  492.     spinlock_lock(&CPU->lock);
  493.     count = nrdy / config.cpu_active;
  494.     count -= CPU->nrdy;
  495.     spinlock_unlock(&CPU->lock);
  496.     interrupts_restore(ipl);
  497.  
  498.     if (count <= 0)
  499.         goto satisfied;
  500.  
  501.     /*
  502.      * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
  503.      */
  504.     for (j=RQ_COUNT-1; j >= 0; j--) {
  505.         for (i=0; i < config.cpu_active; i++) {
  506.             link_t *l;
  507.             runq_t *r;
  508.             cpu_t *cpu;
  509.  
  510.             cpu = &cpus[(i + k) % config.cpu_active];
  511.  
  512.             /*
  513.              * Not interested in ourselves.
  514.              * Doesn't require interrupt disabling for kcpulb is X_WIRED.
  515.              */
  516.             if (CPU == cpu)
  517.                 continue;              
  518.  
  519. restart:        ipl = interrupts_disable();
  520.             r = &cpu->rq[j];
  521.             spinlock_lock(&r->lock);
  522.             if (r->n == 0) {
  523.                 spinlock_unlock(&r->lock);
  524.                 interrupts_restore(ipl);
  525.                 continue;
  526.             }
  527.        
  528.             t = NULL;
  529.             l = r->rq_head.prev;    /* search rq from the back */
  530.             while (l != &r->rq_head) {
  531.                 t = list_get_instance(l, thread_t, rq_link);
  532.                 /*
  533.                  * We don't want to steal CPU-wired threads neither threads already stolen.
  534.                  * The latter prevents threads from migrating between CPU's without ever being run.
  535.                  * We don't want to steal threads whose FPU context is still in CPU.
  536.                  */
  537.                 spinlock_lock(&t->lock);
  538.                 if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
  539.                
  540.                     /*
  541.                      * Remove t from r.
  542.                      */
  543.  
  544.                     spinlock_unlock(&t->lock);
  545.                    
  546.                     /*
  547.                      * Here we have to avoid deadlock with relink_rq(),
  548.                      * because it locks cpu and r in a different order than we do.
  549.                      */
  550.                     if (!spinlock_trylock(&cpu->lock)) {
  551.                         /* Release all locks and try again. */
  552.                         spinlock_unlock(&r->lock);
  553.                         interrupts_restore(ipl);
  554.                         goto restart;
  555.                     }
  556.                     cpu->nrdy--;
  557.                     spinlock_unlock(&cpu->lock);
  558.  
  559.                     atomic_dec(&nrdy);
  560.  
  561.                     r->n--;
  562.                     list_remove(&t->rq_link);
  563.  
  564.                     break;
  565.                 }
  566.                 spinlock_unlock(&t->lock);
  567.                 l = l->prev;
  568.                 t = NULL;
  569.             }
  570.             spinlock_unlock(&r->lock);
  571.  
  572.             if (t) {
  573.                 /*
  574.                  * Ready t on local CPU
  575.                  */
  576.                 spinlock_lock(&t->lock);
  577.                 #ifdef KCPULB_VERBOSE
  578.                 printf("kcpulb%d: TID %d -> cpu%d, nrdy=%d, avg=%d\n", CPU->id, t->tid, CPU->id, CPU->nrdy, nrdy / config.cpu_active);
  579.                 #endif
  580.                 t->flags |= X_STOLEN;
  581.                 spinlock_unlock(&t->lock);
  582.    
  583.                 thread_ready(t);
  584.  
  585.                 interrupts_restore(ipl);
  586.    
  587.                 if (--count == 0)
  588.                     goto satisfied;
  589.                    
  590.                 /*
  591.                  * We are not satisfied yet, focus on another CPU next time.
  592.                  */
  593.                 k++;
  594.                
  595.                 continue;
  596.             }
  597.             interrupts_restore(ipl);
  598.         }
  599.     }
  600.  
  601.     if (CPU->nrdy) {
  602.         /*
  603.          * Be a little bit light-weight and let migrated threads run.
  604.          */
  605.         scheduler();
  606.     }
  607.     else {
  608.         /*
  609.          * We failed to migrate a single thread.
  610.          * Something more sophisticated should be done.
  611.          */
  612.         scheduler();
  613.     }
  614.        
  615.     goto not_satisfied;
  616.  
  617. satisfied:
  618.     /*
  619.      * Tell find_best_thread() to wake us up later again.
  620.      */
  621.     CPU->kcpulbstarted = 0;
  622.     goto loop;
  623. }
  624.  
  625. #endif /* CONFIG_SMP */
  626.