Subversion Repositories HelenOS

Rev

Rev 227 | Rev 248 | 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 <cpu.h>
  33. #include <mm/vm.h>
  34. #include <config.h>
  35. #include <context.h>
  36. #include <func.h>
  37. #include <arch.h>
  38. #include <arch/asm.h>
  39. #include <list.h>
  40. #include <panic.h>
  41. #include <typedefs.h>
  42. #include <mm/page.h>
  43. #include <synch/spinlock.h>
  44. #include <arch/faddr.h>
  45. #include <arch/atomic.h>
  46. #include <print.h>
  47. #include <mm/frame.h>
  48. #include <mm/heap.h>
  49. #include <debug.h>
  50.  
  51. volatile count_t nrdy;
  52.  
  53.  
  54. /** Take actions before new thread runs
  55.  *
  56.  * Perform actions that need to be
  57.  * taken before the newly selected
  58.  * tread is passed control.
  59.  *
  60.  */
  61. void before_thread_runs(void)
  62. {
  63.     before_thread_runs_arch();
  64.     fpu_context_restore(&(THREAD->saved_fpu_context));
  65. }
  66.  
  67.  
  68. /** Initialize scheduler
  69.  *
  70.  * Initialize kernel scheduler.
  71.  *
  72.  */
  73. void scheduler_init(void)
  74. {
  75. }
  76.  
  77.  
  78. /** Get thread to be scheduled
  79.  *
  80.  * Get the optimal thread to be scheduled
  81.  * according to thread accounting and scheduler
  82.  * policy.
  83.  *
  84.  * @return Thread to be scheduled.
  85.  *
  86.  */
  87. struct thread *find_best_thread(void)
  88. {
  89.     thread_t *t;
  90.     runq_t *r;
  91.     int i, n;
  92.  
  93.     ASSERT(CPU != NULL);
  94.  
  95. loop:
  96.     cpu_priority_high();
  97.  
  98.     spinlock_lock(&CPU->lock);
  99.     n = CPU->nrdy;
  100.     spinlock_unlock(&CPU->lock);
  101.  
  102.     cpu_priority_low();
  103.    
  104.     if (n == 0) {
  105.         #ifdef __SMP__
  106.         /*
  107.          * If the load balancing thread is not running, wake it up and
  108.          * set CPU-private flag that the kcpulb has been started.
  109.          */
  110.         if (test_and_set(&CPU->kcpulbstarted) == 0) {
  111.             waitq_wakeup(&CPU->kcpulb_wq, 0);
  112.             goto loop;
  113.         }
  114.         #endif /* __SMP__ */
  115.        
  116.         /*
  117.          * For there was nothing to run, the CPU goes to sleep
  118.          * until a hardware interrupt or an IPI comes.
  119.          * This improves energy saving and hyperthreading.
  120.          * On the other hand, several hardware interrupts can be ignored.
  121.          */
  122.          cpu_sleep();
  123.          goto loop;
  124.     }
  125.  
  126.     cpu_priority_high();
  127.    
  128.     i = 0;
  129. retry:
  130.     for (; i<RQ_COUNT; i++) {
  131.         r = &CPU->rq[i];
  132.         spinlock_lock(&r->lock);
  133.         if (r->n == 0) {
  134.             /*
  135.              * If this queue is empty, try a lower-priority queue.
  136.              */
  137.             spinlock_unlock(&r->lock);
  138.             continue;
  139.         }
  140.  
  141.         /* avoid deadlock with relink_rq() */
  142.         if (!spinlock_trylock(&CPU->lock)) {
  143.             /*
  144.              * Unlock r and try again.
  145.              */
  146.             spinlock_unlock(&r->lock);
  147.             goto retry;
  148.         }
  149.         CPU->nrdy--;
  150.         spinlock_unlock(&CPU->lock);
  151.  
  152.         atomic_dec(&nrdy);
  153.         r->n--;
  154.  
  155.         /*
  156.          * Take the first thread from the queue.
  157.          */
  158.         t = list_get_instance(r->rq_head.next, thread_t, rq_link);
  159.         list_remove(&t->rq_link);
  160.  
  161.         spinlock_unlock(&r->lock);
  162.  
  163.         spinlock_lock(&t->lock);
  164.         t->cpu = CPU;
  165.  
  166.         t->ticks = us2ticks((i+1)*10000);
  167.         t->pri = i; /* eventually correct rq index */
  168.  
  169.         /*
  170.          * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.
  171.          */
  172.         t->flags &= ~X_STOLEN;
  173.         spinlock_unlock(&t->lock);
  174.  
  175.         return t;
  176.     }
  177.     goto loop;
  178.  
  179. }
  180.  
  181.  
  182. /** Prevent rq starvation
  183.  *
  184.  * Prevent low priority threads from starving in rq's.
  185.  *
  186.  * When the function decides to relink rq's, it reconnects
  187.  * respective pointers so that in result threads with 'pri'
  188.  * greater or equal 'start' are moved to a higher-priority queue.
  189.  *
  190.  * @param start Threshold priority.
  191.  *
  192.  */
  193. void relink_rq(int start)
  194. {
  195.     link_t head;
  196.     runq_t *r;
  197.     int i, n;
  198.  
  199.     list_initialize(&head);
  200.     spinlock_lock(&CPU->lock);
  201.     if (CPU->needs_relink > NEEDS_RELINK_MAX) {
  202.         for (i = start; i<RQ_COUNT-1; i++) {
  203.             /* remember and empty rq[i + 1] */
  204.             r = &CPU->rq[i + 1];
  205.             spinlock_lock(&r->lock);
  206.             list_concat(&head, &r->rq_head);
  207.             n = r->n;
  208.             r->n = 0;
  209.             spinlock_unlock(&r->lock);
  210.        
  211.             /* append rq[i + 1] to rq[i] */
  212.             r = &CPU->rq[i];
  213.             spinlock_lock(&r->lock);
  214.             list_concat(&r->rq_head, &head);
  215.             r->n += n;
  216.             spinlock_unlock(&r->lock);
  217.         }
  218.         CPU->needs_relink = 0;
  219.     }
  220.     spinlock_unlock(&CPU->lock);               
  221.  
  222. }
  223.  
  224.  
  225. /** The scheduler
  226.  *
  227.  * The thread scheduling procedure.
  228.  *
  229.  */
  230. void scheduler(void)
  231. {
  232.     volatile pri_t pri;
  233.  
  234.     ASSERT(CPU != NULL);
  235.  
  236.     pri = cpu_priority_high();
  237.  
  238.     if (haltstate)
  239.         halt();
  240.  
  241.     if (THREAD) {
  242.         spinlock_lock(&THREAD->lock);
  243.         fpu_context_save(&(THREAD->saved_fpu_context));
  244.         if (!context_save(&THREAD->saved_context)) {
  245.             /*
  246.              * This is the place where threads leave scheduler();
  247.              */
  248.             before_thread_runs();
  249.             spinlock_unlock(&THREAD->lock);
  250.             cpu_priority_restore(THREAD->saved_context.pri);
  251.             return;
  252.         }
  253.  
  254.         /*
  255.          * CPU priority of preempted thread is recorded here
  256.          * to facilitate scheduler() invocations from
  257.          * cpu_priority_high()'ed code (e.g. waitq_sleep_timeout()).
  258.          */
  259.         THREAD->saved_context.pri = pri;
  260.     }
  261.  
  262.     /*
  263.      * Through the 'THE' structure, we keep track of THREAD, TASK, CPU
  264.      * and preemption counter. At this point THE could be coming either
  265.      * from THREAD's or CPU's stack.
  266.      */
  267.     the_copy(THE, (the_t *) CPU->stack);
  268.  
  269.     /*
  270.      * We may not keep the old stack.
  271.      * Reason: If we kept the old stack and got blocked, for instance, in
  272.      * find_best_thread(), the old thread could get rescheduled by another
  273.      * CPU and overwrite the part of its own stack that was also used by
  274.      * the scheduler on this CPU.
  275.      *
  276.      * Moreover, we have to bypass the compiler-generated POP sequence
  277.      * which is fooled by SP being set to the very top of the stack.
  278.      * Therefore the scheduler() function continues in
  279.      * scheduler_separated_stack().
  280.      */
  281.     context_save(&CPU->saved_context);
  282.     context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), CPU->stack, CPU_STACK_SIZE);
  283.     context_restore(&CPU->saved_context);
  284.     /* not reached */
  285. }
  286.  
  287.  
  288. /** Scheduler stack switch wrapper
  289.  *
  290.  * Second part of the scheduler() function
  291.  * using new stack. Handling the actual context
  292.  * switch to a new thread.
  293.  *
  294.  */
  295. void scheduler_separated_stack(void)
  296. {
  297.     int priority;
  298.  
  299.     ASSERT(CPU != NULL);
  300.  
  301.     if (THREAD) {
  302.         switch (THREAD->state) {
  303.             case Running:
  304.             THREAD->state = Ready;
  305.             spinlock_unlock(&THREAD->lock);
  306.             thread_ready(THREAD);
  307.             break;
  308.  
  309.             case Exiting:
  310.             frame_free((__address) THREAD->kstack);
  311.             if (THREAD->ustack) {
  312.                 frame_free((__address) THREAD->ustack);
  313.             }
  314.  
  315.             /*
  316.              * Detach from the containing task.
  317.              */
  318.             spinlock_lock(&TASK->lock);
  319.             list_remove(&THREAD->th_link);
  320.             spinlock_unlock(&TASK->lock);
  321.  
  322.             spinlock_unlock(&THREAD->lock);
  323.    
  324.             spinlock_lock(&threads_lock);
  325.             list_remove(&THREAD->threads_link);
  326.             spinlock_unlock(&threads_lock);
  327.  
  328.             spinlock_lock(&CPU->lock);
  329.             if(CPU->fpu_owner==THREAD) CPU->fpu_owner=NULL;
  330.             spinlock_unlock(&CPU->lock);
  331.  
  332.             free(THREAD);
  333.  
  334.             break;
  335.    
  336.             case Sleeping:
  337.             /*
  338.              * Prefer the thread after it's woken up.
  339.              */
  340.             THREAD->pri = -1;
  341.  
  342.             /*
  343.              * We need to release wq->lock which we locked in waitq_sleep().
  344.              * Address of wq->lock is kept in THREAD->sleep_queue.
  345.              */
  346.             spinlock_unlock(&THREAD->sleep_queue->lock);
  347.  
  348.             /*
  349.              * Check for possible requests for out-of-context invocation.
  350.              */
  351.             if (THREAD->call_me) {
  352.                 THREAD->call_me(THREAD->call_me_with);
  353.                 THREAD->call_me = NULL;
  354.                 THREAD->call_me_with = NULL;
  355.             }
  356.  
  357.             spinlock_unlock(&THREAD->lock);
  358.  
  359.             break;
  360.  
  361.             default:
  362.             /*
  363.              * Entering state is unexpected.
  364.              */
  365.             panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
  366.             break;
  367.         }
  368.         THREAD = NULL;
  369.     }
  370.  
  371.  
  372.     THREAD = find_best_thread();
  373.    
  374.     spinlock_lock(&THREAD->lock);
  375.     priority = THREAD->pri;
  376.     spinlock_unlock(&THREAD->lock);
  377.  
  378.     relink_rq(priority);       
  379.  
  380.     spinlock_lock(&THREAD->lock);  
  381.  
  382.     /*
  383.      * If both the old and the new task are the same, lots of work is avoided.
  384.      */
  385.     if (TASK != THREAD->task) {
  386.         vm_t *m1 = NULL;
  387.         vm_t *m2;
  388.  
  389.         if (TASK) {
  390.             spinlock_lock(&TASK->lock);
  391.             m1 = TASK->vm;
  392.             spinlock_unlock(&TASK->lock);
  393.         }
  394.  
  395.         spinlock_lock(&THREAD->task->lock);
  396.         m2 = THREAD->task->vm;
  397.         spinlock_unlock(&THREAD->task->lock);
  398.        
  399.         /*
  400.          * Note that it is possible for two tasks to share one vm mapping.
  401.          */
  402.         if (m1 != m2) {
  403.             /*
  404.              * Both tasks and vm mappings are different.
  405.              * Replace the old one with the new one.
  406.              */
  407.             vm_install(m2);
  408.         }
  409.         TASK = THREAD->task;   
  410.     }
  411.  
  412.     THREAD->state = Running;
  413.  
  414.     #ifdef SCHEDULER_VERBOSE
  415.     printf("cpu%d: tid %d (pri=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->pri, THREAD->ticks, CPU->nrdy);
  416.     #endif 
  417.  
  418.     /*
  419.      * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack.
  420.      */
  421.     the_copy(THE, (the_t *) THREAD->kstack);
  422.    
  423.     context_restore(&THREAD->saved_context);
  424.     /* not reached */
  425. }
  426.  
  427.  
  428. #ifdef __SMP__
  429. /** Load balancing thread
  430.  *
  431.  * SMP load balancing thread, supervising thread supplies
  432.  * for the CPU it's wired to.
  433.  *
  434.  * @param arg Generic thread argument (unused).
  435.  *
  436.  */
  437. void kcpulb(void *arg)
  438. {
  439.     thread_t *t;
  440.     int count, i, j, k = 0;
  441.     pri_t pri;
  442.  
  443. loop:
  444.     /*
  445.      * Sleep until there's some work to do.
  446.      */
  447.     waitq_sleep(&CPU->kcpulb_wq);
  448.  
  449. not_satisfied:
  450.     /*
  451.      * Calculate the number of threads that will be migrated/stolen from
  452.      * other CPU's. Note that situation can have changed between two
  453.      * passes. Each time get the most up to date counts.
  454.      */
  455.     pri = cpu_priority_high();
  456.     spinlock_lock(&CPU->lock);
  457.     count = nrdy / config.cpu_active;
  458.     count -= CPU->nrdy;
  459.     spinlock_unlock(&CPU->lock);
  460.     cpu_priority_restore(pri);
  461.  
  462.     if (count <= 0)
  463.         goto satisfied;
  464.  
  465.     /*
  466.      * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
  467.      */
  468.     for (j=RQ_COUNT-1; j >= 0; j--) {
  469.         for (i=0; i < config.cpu_active; i++) {
  470.             link_t *l;
  471.             runq_t *r;
  472.             cpu_t *cpu;
  473.  
  474.             cpu = &cpus[(i + k) % config.cpu_active];
  475.  
  476.             /*
  477.              * Not interested in ourselves.
  478.              * Doesn't require interrupt disabling for kcpulb is X_WIRED.
  479.              */
  480.             if (CPU == cpu)
  481.                 continue;              
  482.  
  483. restart:        pri = cpu_priority_high();
  484.             r = &cpu->rq[j];
  485.             spinlock_lock(&r->lock);
  486.             if (r->n == 0) {
  487.                 spinlock_unlock(&r->lock);
  488.                 cpu_priority_restore(pri);
  489.                 continue;
  490.             }
  491.        
  492.             t = NULL;
  493.             l = r->rq_head.prev;    /* search rq from the back */
  494.             while (l != &r->rq_head) {
  495.                 t = list_get_instance(l, thread_t, rq_link);
  496.                 /*
  497.                  * We don't want to steal CPU-wired threads neither threads already stolen.
  498.                  * The latter prevents threads from migrating between CPU's without ever being run.
  499.                  * We don't want to steal threads whose FPU context is still in CPU.
  500.                  */
  501.                 spinlock_lock(&t->lock);
  502.                 if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
  503.                
  504.                     /*
  505.                      * Remove t from r.
  506.                      */
  507.  
  508.                     spinlock_unlock(&t->lock);
  509.                    
  510.                     /*
  511.                      * Here we have to avoid deadlock with relink_rq(),
  512.                      * because it locks cpu and r in a different order than we do.
  513.                      */
  514.                     if (!spinlock_trylock(&cpu->lock)) {
  515.                         /* Release all locks and try again. */
  516.                         spinlock_unlock(&r->lock);
  517.                         cpu_priority_restore(pri);
  518.                         goto restart;
  519.                     }
  520.                     cpu->nrdy--;
  521.                     spinlock_unlock(&cpu->lock);
  522.  
  523.                     atomic_dec(&nrdy);
  524.  
  525.                     r->n--;
  526.                     list_remove(&t->rq_link);
  527.  
  528.                     break;
  529.                 }
  530.                 spinlock_unlock(&t->lock);
  531.                 l = l->prev;
  532.                 t = NULL;
  533.             }
  534.             spinlock_unlock(&r->lock);
  535.  
  536.             if (t) {
  537.                 /*
  538.                  * Ready t on local CPU
  539.                  */
  540.                 spinlock_lock(&t->lock);
  541.                 #ifdef KCPULB_VERBOSE
  542.                 printf("kcpulb%d: TID %d -> cpu%d, nrdy=%d, avg=%d\n", CPU->id, t->tid, CPU->id, CPU->nrdy, nrdy / config.cpu_active);
  543.                 #endif
  544.                 t->flags |= X_STOLEN;
  545.                 spinlock_unlock(&t->lock);
  546.    
  547.                 thread_ready(t);
  548.  
  549.                 cpu_priority_restore(pri);
  550.    
  551.                 if (--count == 0)
  552.                     goto satisfied;
  553.                    
  554.                 /*
  555.                  * We are not satisfied yet, focus on another CPU next time.
  556.                  */
  557.                 k++;
  558.                
  559.                 continue;
  560.             }
  561.             cpu_priority_restore(pri);
  562.         }
  563.     }
  564.  
  565.     if (CPU->nrdy) {
  566.         /*
  567.          * Be a little bit light-weight and let migrated threads run.
  568.          */
  569.         scheduler();
  570.     }
  571.     else {
  572.         /*
  573.          * We failed to migrate a single thread.
  574.          * Something more sophisticated should be done.
  575.          */
  576.         scheduler();
  577.     }
  578.        
  579.     goto not_satisfied;
  580.  
  581. satisfied:
  582.     /*
  583.      * Tell find_best_thread() to wake us up later again.
  584.      */
  585.     CPU->kcpulbstarted = 0;
  586.     goto loop;
  587. }
  588.  
  589. #endif /* __SMP__ */
  590.