Subversion Repositories HelenOS

Rev

Rev 57 | Rev 73 | 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.  
  45. #ifdef __SMP__
  46. #include <arch/smp/atomic.h>
  47. #endif /* __SMP__ */
  48.  
  49. /*
  50.  * NOTE ON ATOMIC READS:
  51.  * Some architectures cannot read __u32 atomically.
  52.  * For that reason, all accesses to nrdy and the likes must be protected by spinlock.
  53.  */
  54.  
  55. spinlock_t nrdylock;
  56. volatile int nrdy;
  57.  
  58. void before_thread_runs(void)
  59. {
  60.     before_thread_runs_arch();
  61.     fpu_context_restore(&(THREAD->saved_fpu_context));
  62. }
  63.  
  64.  
  65. void scheduler_init(void)
  66. {
  67.     spinlock_initialize(&nrdylock);
  68. }
  69.  
  70. /* cpu_priority_high()'d */
  71. struct thread *find_best_thread(void)
  72. {
  73.     thread_t *t;
  74.     runq_t *r;
  75.     int i, n;
  76.  
  77. loop:
  78.     cpu_priority_high();
  79.  
  80.     spinlock_lock(&CPU->lock);
  81.     n = CPU->nrdy;
  82.     spinlock_unlock(&CPU->lock);
  83.  
  84.     cpu_priority_low();
  85.    
  86.     if (n == 0) {
  87.         #ifdef __SMP__
  88.         /*
  89.          * If the load balancing thread is not running, wake it up and
  90.          * set CPU-private flag that the kcpulb has been started.
  91.          */
  92.         if (test_and_set(&CPU->kcpulbstarted) == 0) {
  93.                 waitq_wakeup(&CPU->kcpulb_wq, 0);
  94.             goto loop;
  95.         }
  96.         #endif /* __SMP__ */
  97.        
  98.         /*
  99.          * For there was nothing to run, the CPU goes to sleep
  100.          * until a hardware interrupt or an IPI comes.
  101.          * This improves energy saving and hyperthreading.
  102.          * On the other hand, several hardware interrupts can be ignored.
  103.          */
  104.          cpu_sleep();
  105.          goto loop;
  106.     }
  107.  
  108.     cpu_priority_high();
  109.  
  110.     for (i = 0; i<RQ_COUNT; i++) {
  111.         r = &CPU->rq[i];
  112.         spinlock_lock(&r->lock);
  113.         if (r->n == 0) {
  114.             /*
  115.              * If this queue is empty, try a lower-priority queue.
  116.              */
  117.             spinlock_unlock(&r->lock);
  118.             continue;
  119.         }
  120.    
  121.         spinlock_lock(&nrdylock);
  122.         nrdy--;
  123.         spinlock_unlock(&nrdylock);    
  124.  
  125.         spinlock_lock(&CPU->lock);
  126.         CPU->nrdy--;
  127.         spinlock_unlock(&CPU->lock);
  128.  
  129.         r->n--;
  130.  
  131.         /*
  132.          * Take the first thread from the queue.
  133.          */
  134.         t = list_get_instance(r->rq_head.next, thread_t, rq_link);
  135.         list_remove(&t->rq_link);
  136.  
  137.         spinlock_unlock(&r->lock);
  138.  
  139.         spinlock_lock(&t->lock);
  140.         t->cpu = CPU;
  141.  
  142.         t->ticks = us2ticks((i+1)*10000);
  143.         t->pri = i; /* eventually correct rq index */
  144.  
  145.         /*
  146.          * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.
  147.          */
  148.         t->flags &= ~X_STOLEN;
  149.         spinlock_unlock(&t->lock);
  150.  
  151.         return t;
  152.     }
  153.     goto loop;
  154.  
  155. }
  156.  
  157. /*
  158.  * This function prevents low priority threads from starving in rq's.
  159.  * When it decides to relink rq's, it reconnects respective pointers
  160.  * so that in result threads with 'pri' greater or equal 'start' are
  161.  * moved to a higher-priority queue.
  162.  */
  163. void relink_rq(int start)
  164. {
  165.     link_t head;
  166.     runq_t *r;
  167.     int i, n;
  168.  
  169.     list_initialize(&head);
  170.     spinlock_lock(&CPU->lock);
  171.     if (CPU->needs_relink > NEEDS_RELINK_MAX) {
  172.         for (i = start; i<RQ_COUNT-1; i++) {
  173.             /* remember and empty rq[i + 1] */
  174.             r = &CPU->rq[i + 1];
  175.             spinlock_lock(&r->lock);
  176.             list_concat(&head, &r->rq_head);
  177.             n = r->n;
  178.             r->n = 0;
  179.             spinlock_unlock(&r->lock);
  180.        
  181.             /* append rq[i + 1] to rq[i] */
  182.             r = &CPU->rq[i];
  183.             spinlock_lock(&r->lock);
  184.             list_concat(&r->rq_head, &head);
  185.             r->n += n;
  186.             spinlock_unlock(&r->lock);
  187.         }
  188.         CPU->needs_relink = 0;
  189.     }
  190.     spinlock_unlock(&CPU->lock);               
  191.  
  192. }
  193.  
  194. /*
  195.  * The scheduler.
  196.  */
  197. void scheduler(void)
  198. {
  199.     volatile pri_t pri;
  200.  
  201.     pri = cpu_priority_high();
  202.  
  203.     if (haltstate)
  204.         halt();
  205.  
  206.     if (THREAD) {
  207.         spinlock_lock(&THREAD->lock);
  208.         fpu_context_save(&(THREAD->saved_fpu_context));
  209.         if (!context_save(&THREAD->saved_context)) {
  210.             /*
  211.              * This is the place where threads leave scheduler();
  212.              */
  213.             before_thread_runs();
  214.                 spinlock_unlock(&THREAD->lock);
  215.             cpu_priority_restore(THREAD->saved_context.pri);
  216.             return;
  217.         }
  218.         THREAD->saved_context.pri = pri;
  219.     }
  220.  
  221.     /*
  222.      * We may not keep the old stack.
  223.      * Reason: If we kept the old stack and got blocked, for instance, in
  224.      * find_best_thread(), the old thread could get rescheduled by another
  225.      * CPU and overwrite the part of its own stack that was also used by
  226.      * the scheduler on this CPU.
  227.      *
  228.      * Moreover, we have to bypass the compiler-generated POP sequence
  229.      * which is fooled by SP being set to the very top of the stack.
  230.      * Therefore the scheduler() function continues in
  231.      * scheduler_separated_stack().
  232.      */
  233.     context_save(&CPU->saved_context);
  234.     CPU->saved_context.sp = (__address) &CPU->stack[CPU_STACK_SIZE-8];
  235.     CPU->saved_context.pc = (__address) scheduler_separated_stack;
  236.     context_restore(&CPU->saved_context);
  237.     /* not reached */
  238. }
  239.  
  240. void scheduler_separated_stack(void)
  241. {
  242.     int priority;
  243.  
  244.     if (THREAD) {
  245.         switch (THREAD->state) {
  246.             case Running:
  247.                 THREAD->state = Ready;
  248.                 spinlock_unlock(&THREAD->lock);
  249.                 thread_ready(THREAD);
  250.                 break;
  251.  
  252.             case Exiting:
  253.                 frame_free((__address) THREAD->kstack);
  254.                 if (THREAD->ustack) {
  255.                     frame_free((__address) THREAD->ustack);
  256.                 }
  257.                
  258.                 /*
  259.                  * Detach from the containing task.
  260.                  */
  261.                 spinlock_lock(&TASK->lock);
  262.                 list_remove(&THREAD->th_link);
  263.                 spinlock_unlock(&TASK->lock);
  264.  
  265.                 spinlock_unlock(&THREAD->lock);
  266.                
  267.                 spinlock_lock(&threads_lock);
  268.                 list_remove(&THREAD->threads_link);
  269.                 spinlock_unlock(&threads_lock);
  270.                
  271.                 free(THREAD);
  272.                
  273.                 break;
  274.                
  275.             case Sleeping:
  276.                 /*
  277.                  * Prefer the thread after it's woken up.
  278.                  */
  279.                 THREAD->pri = -1;
  280.  
  281.                 /*
  282.                  * We need to release wq->lock which we locked in waitq_sleep().
  283.                  * Address of wq->lock is kept in THREAD->sleep_queue.
  284.                  */
  285.                 spinlock_unlock(&THREAD->sleep_queue->lock);
  286.  
  287.                 /*
  288.                  * Check for possible requests for out-of-context invocation.
  289.                  */
  290.                 if (THREAD->call_me) {
  291.                     THREAD->call_me(THREAD->call_me_with);
  292.                     THREAD->call_me = NULL;
  293.                     THREAD->call_me_with = NULL;
  294.                 }
  295.  
  296.                 spinlock_unlock(&THREAD->lock);
  297.                
  298.                 break;
  299.  
  300.             default:
  301.                 /*
  302.                  * Entering state is unexpected.
  303.                  */
  304.                 panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
  305.                 break;
  306.         }
  307.         THREAD = NULL;
  308.     }
  309.    
  310.     THREAD = find_best_thread();
  311.    
  312.     spinlock_lock(&THREAD->lock);
  313.     priority = THREAD->pri;
  314.     spinlock_unlock(&THREAD->lock);
  315.    
  316.     relink_rq(priority);       
  317.  
  318.     spinlock_lock(&THREAD->lock);  
  319.  
  320.     /*
  321.      * If both the old and the new task are the same, lots of work is avoided.
  322.      */
  323.     if (TASK != THREAD->task) {
  324.         vm_t *m1 = NULL;
  325.         vm_t *m2;
  326.  
  327.         if (TASK) {
  328.             spinlock_lock(&TASK->lock);
  329.             m1 = TASK->vm;
  330.             spinlock_unlock(&TASK->lock);
  331.         }
  332.  
  333.         spinlock_lock(&THREAD->task->lock);
  334.         m2 = THREAD->task->vm;
  335.         spinlock_unlock(&THREAD->task->lock);
  336.        
  337.         /*
  338.          * Note that it is possible for two tasks to share one vm mapping.
  339.          */
  340.         if (m1 != m2) {
  341.             /*
  342.              * Both tasks and vm mappings are different.
  343.              * Replace the old one with the new one.
  344.              */
  345.             if (m1) {
  346.                 vm_uninstall(m1);
  347.             }
  348.             vm_install(m2);
  349.         }
  350.         TASK = THREAD->task;   
  351.     }
  352.  
  353.     THREAD->state = Running;
  354.  
  355.     #ifdef SCHEDULER_VERBOSE
  356.     printf("cpu%d: tid %d (pri=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->pri, THREAD->ticks, CPU->nrdy);
  357.     #endif 
  358.  
  359.     context_restore(&THREAD->saved_context);
  360.     /* not reached */
  361. }
  362.  
  363. #ifdef __SMP__
  364. /*
  365.  * This is the load balancing thread.
  366.  * It supervises thread supplies for the CPU it's wired to.
  367.  */
  368. void kcpulb(void *arg)
  369. {
  370.     thread_t *t;
  371.     int count, i, j, k = 0;
  372.     pri_t pri;
  373.  
  374. loop:
  375.     /*
  376.      * Sleep until there's some work to do.
  377.      */
  378.     waitq_sleep(&CPU->kcpulb_wq);
  379.  
  380. not_satisfied:
  381.     /*
  382.      * Calculate the number of threads that will be migrated/stolen from
  383.      * other CPU's. Note that situation can have changed between two
  384.      * passes. Each time get the most up to date counts.
  385.      */
  386.     pri = cpu_priority_high();
  387.     spinlock_lock(&CPU->lock);
  388.     count = nrdy / config.cpu_active;
  389.     count -= CPU->nrdy;
  390.     spinlock_unlock(&CPU->lock);
  391.     cpu_priority_restore(pri);
  392.  
  393.     if (count <= 0)
  394.         goto satisfied;
  395.  
  396.     /*
  397.      * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
  398.      */
  399.     for (j=RQ_COUNT-1; j >= 0; j--) {
  400.         for (i=0; i < config.cpu_active; i++) {
  401.             link_t *l;
  402.             runq_t *r;
  403.             cpu_t *cpu;
  404.  
  405.             cpu = &cpus[(i + k) % config.cpu_active];
  406.             r = &cpu->rq[j];
  407.  
  408.             /*
  409.              * Not interested in ourselves.
  410.              * Doesn't require interrupt disabling for kcpulb is X_WIRED.
  411.              */
  412.             if (CPU == cpu)
  413.                 continue;
  414.  
  415. restart:        pri = cpu_priority_high();
  416.             spinlock_lock(&r->lock);
  417.             if (r->n == 0) {
  418.                 spinlock_unlock(&r->lock);
  419.                 cpu_priority_restore(pri);
  420.                 continue;
  421.             }
  422.        
  423.             t = NULL;
  424.             l = r->rq_head.prev;    /* search rq from the back */
  425.             while (l != &r->rq_head) {
  426.                 t = list_get_instance(l, thread_t, rq_link);
  427.                 /*
  428.                      * We don't want to steal CPU-wired threads neither threads already stolen.
  429.                  * The latter prevents threads from migrating between CPU's without ever being run.
  430.                      */
  431.                 spinlock_lock(&t->lock);
  432.                 if (!(t->flags & (X_WIRED | X_STOLEN))) {
  433.                     /*
  434.                      * Remove t from r.
  435.                      */
  436.  
  437.                     spinlock_unlock(&t->lock);
  438.                    
  439.                     /*
  440.                      * Here we have to avoid deadlock with relink_rq(),
  441.                      * because it locks cpu and r in a different order than we do.
  442.                      */
  443.                     if (!spinlock_trylock(&cpu->lock)) {
  444.                         /* Release all locks and try again. */
  445.                         spinlock_unlock(&r->lock);
  446.                         cpu_priority_restore(pri);
  447.                         goto restart;
  448.                     }
  449.                     cpu->nrdy--;
  450.                     spinlock_unlock(&cpu->lock);
  451.  
  452.                     spinlock_lock(&nrdylock);
  453.                     nrdy--;
  454.                     spinlock_unlock(&nrdylock);                
  455.  
  456.                         r->n--;
  457.                     list_remove(&t->rq_link);
  458.  
  459.                     break;
  460.                 }
  461.                 spinlock_unlock(&t->lock);
  462.                 l = l->prev;
  463.                 t = NULL;
  464.             }
  465.             spinlock_unlock(&r->lock);
  466.  
  467.             if (t) {
  468.                 /*
  469.                  * Ready t on local CPU
  470.                  */
  471.                 spinlock_lock(&t->lock);
  472.                 #ifdef KCPULB_VERBOSE
  473.                 printf("kcpulb%d: TID %d -> cpu%d, nrdy=%d, avg=%d\n", CPU->id, t->tid, CPU->id, CPU->nrdy, nrdy / config.cpu_active);
  474.                 #endif
  475.                 t->flags |= X_STOLEN;
  476.                 spinlock_unlock(&t->lock);
  477.    
  478.                 thread_ready(t);
  479.  
  480.                 cpu_priority_restore(pri);
  481.    
  482.                 if (--count == 0)
  483.                     goto satisfied;
  484.                    
  485.                 /*
  486.                              * We are not satisfied yet, focus on another CPU next time.
  487.                  */
  488.                 k++;
  489.                
  490.                 continue;
  491.             }
  492.             cpu_priority_restore(pri);
  493.         }
  494.     }
  495.  
  496.     if (CPU->nrdy) {
  497.         /*
  498.          * Be a little bit light-weight and let migrated threads run.
  499.          */
  500.         scheduler();
  501.     }
  502.     else {
  503.         /*
  504.          * We failed to migrate a single thread.
  505.          * Something more sophisticated should be done.
  506.          */
  507.         scheduler();
  508.     }
  509.        
  510.     goto not_satisfied;
  511.    
  512. satisfied:
  513.     /*
  514.      * Tell find_best_thread() to wake us up later again.
  515.      */
  516.     CPU->kcpulbstarted = 0;
  517.     goto loop;
  518. }
  519.  
  520. #endif /* __SMP__ */
  521.