Subversion Repositories HelenOS

Rev

Rev 907 | Rev 1062 | 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. static void scheduler_separated_stack(void);
  51.  
  52. atomic_t nrdy;  /**< Number of ready threads in the system. */
  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.  * THREAD->lock is locked on entry
  61.  *
  62.  */
  63. void before_thread_runs(void)
  64. {
  65.     before_thread_runs_arch();
  66. #ifdef CONFIG_FPU_LAZY
  67.     if(THREAD==CPU->fpu_owner)
  68.         fpu_enable();
  69.     else
  70.         fpu_disable();
  71. #else
  72.     fpu_enable();
  73.     if (THREAD->fpu_context_exists)
  74.         fpu_context_restore(THREAD->saved_fpu_context);
  75.     else {
  76.         fpu_init();
  77.         THREAD->fpu_context_exists=1;
  78.     }
  79. #endif
  80. }
  81.  
  82. /** Take actions after THREAD had run.
  83.  *
  84.  * Perform actions that need to be
  85.  * taken after the running thread
  86.  * had been preempted by the scheduler.
  87.  *
  88.  * THREAD->lock is locked on entry
  89.  *
  90.  */
  91. void after_thread_ran(void)
  92. {
  93.     after_thread_ran_arch();
  94. }
  95.  
  96. #ifdef CONFIG_FPU_LAZY
  97. void scheduler_fpu_lazy_request(void)
  98. {
  99. restart:
  100.     fpu_enable();
  101.     spinlock_lock(&CPU->lock);
  102.  
  103.     /* Save old context */
  104.     if (CPU->fpu_owner != NULL) {  
  105.         spinlock_lock(&CPU->fpu_owner->lock);
  106.         fpu_context_save(CPU->fpu_owner->saved_fpu_context);
  107.         /* don't prevent migration */
  108.         CPU->fpu_owner->fpu_context_engaged=0;
  109.         spinlock_unlock(&CPU->fpu_owner->lock);
  110.         CPU->fpu_owner = NULL;
  111.     }
  112.  
  113.     spinlock_lock(&THREAD->lock);
  114.     if (THREAD->fpu_context_exists) {
  115.         fpu_context_restore(THREAD->saved_fpu_context);
  116.     } else {
  117.         /* Allocate FPU context */
  118.         if (!THREAD->saved_fpu_context) {
  119.             /* Might sleep */
  120.             spinlock_unlock(&THREAD->lock);
  121.             spinlock_unlock(&CPU->lock);
  122.             THREAD->saved_fpu_context = slab_alloc(fpu_context_slab,
  123.                                    0);
  124.             /* We may have switched CPUs during slab_alloc */
  125.             goto restart;
  126.         }
  127.         fpu_init();
  128.         THREAD->fpu_context_exists=1;
  129.     }
  130.     CPU->fpu_owner=THREAD;
  131.     THREAD->fpu_context_engaged = 1;
  132.     spinlock_unlock(&THREAD->lock);
  133.  
  134.     spinlock_unlock(&CPU->lock);
  135. }
  136. #endif
  137.  
  138. /** Initialize scheduler
  139.  *
  140.  * Initialize kernel scheduler.
  141.  *
  142.  */
  143. void scheduler_init(void)
  144. {
  145. }
  146.  
  147. /** Get thread to be scheduled
  148.  *
  149.  * Get the optimal thread to be scheduled
  150.  * according to thread accounting and scheduler
  151.  * policy.
  152.  *
  153.  * @return Thread to be scheduled.
  154.  *
  155.  */
  156. static thread_t *find_best_thread(void)
  157. {
  158.     thread_t *t;
  159.     runq_t *r;
  160.     int i;
  161.  
  162.     ASSERT(CPU != NULL);
  163.  
  164. loop:
  165.     interrupts_enable();
  166.    
  167.     if (atomic_get(&CPU->nrdy) == 0) {
  168.         /*
  169.          * For there was nothing to run, the CPU goes to sleep
  170.          * until a hardware interrupt or an IPI comes.
  171.          * This improves energy saving and hyperthreading.
  172.          */
  173.  
  174.         /*
  175.          * An interrupt might occur right now and wake up a thread.
  176.          * In such case, the CPU will continue to go to sleep
  177.          * even though there is a runnable thread.
  178.          */
  179.  
  180.          cpu_sleep();
  181.          goto loop;
  182.     }
  183.  
  184.     interrupts_disable();
  185.    
  186.     for (i = 0; i<RQ_COUNT; i++) {
  187.         r = &CPU->rq[i];
  188.         spinlock_lock(&r->lock);
  189.         if (r->n == 0) {
  190.             /*
  191.              * If this queue is empty, try a lower-priority queue.
  192.              */
  193.             spinlock_unlock(&r->lock);
  194.             continue;
  195.         }
  196.  
  197.         atomic_dec(&CPU->nrdy);
  198.         atomic_dec(&nrdy);
  199.         r->n--;
  200.  
  201.         /*
  202.          * Take the first thread from the queue.
  203.          */
  204.         t = list_get_instance(r->rq_head.next, thread_t, rq_link);
  205.         list_remove(&t->rq_link);
  206.  
  207.         spinlock_unlock(&r->lock);
  208.  
  209.         spinlock_lock(&t->lock);
  210.         t->cpu = CPU;
  211.  
  212.         t->ticks = us2ticks((i+1)*10000);
  213.         t->priority = i;    /* correct rq index */
  214.  
  215.         /*
  216.          * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.
  217.          */
  218.         t->flags &= ~X_STOLEN;
  219.         spinlock_unlock(&t->lock);
  220.  
  221.         return t;
  222.     }
  223.     goto loop;
  224.  
  225. }
  226.  
  227. /** Prevent rq starvation
  228.  *
  229.  * Prevent low priority threads from starving in rq's.
  230.  *
  231.  * When the function decides to relink rq's, it reconnects
  232.  * respective pointers so that in result threads with 'pri'
  233.  * greater or equal 'start' are moved to a higher-priority queue.
  234.  *
  235.  * @param start Threshold priority.
  236.  *
  237.  */
  238. static void relink_rq(int start)
  239. {
  240.     link_t head;
  241.     runq_t *r;
  242.     int i, n;
  243.  
  244.     list_initialize(&head);
  245.     spinlock_lock(&CPU->lock);
  246.     if (CPU->needs_relink > NEEDS_RELINK_MAX) {
  247.         for (i = start; i<RQ_COUNT-1; i++) {
  248.             /* remember and empty rq[i + 1] */
  249.             r = &CPU->rq[i + 1];
  250.             spinlock_lock(&r->lock);
  251.             list_concat(&head, &r->rq_head);
  252.             n = r->n;
  253.             r->n = 0;
  254.             spinlock_unlock(&r->lock);
  255.        
  256.             /* append rq[i + 1] to rq[i] */
  257.             r = &CPU->rq[i];
  258.             spinlock_lock(&r->lock);
  259.             list_concat(&r->rq_head, &head);
  260.             r->n += n;
  261.             spinlock_unlock(&r->lock);
  262.         }
  263.         CPU->needs_relink = 0;
  264.     }
  265.     spinlock_unlock(&CPU->lock);
  266.  
  267. }
  268.  
  269. /** The scheduler
  270.  *
  271.  * The thread scheduling procedure.
  272.  * Passes control directly to
  273.  * scheduler_separated_stack().
  274.  *
  275.  */
  276. void scheduler(void)
  277. {
  278.     volatile ipl_t ipl;
  279.  
  280.     ASSERT(CPU != NULL);
  281.  
  282.     ipl = interrupts_disable();
  283.  
  284.     if (atomic_get(&haltstate))
  285.         halt();
  286.    
  287.     if (THREAD) {
  288.         spinlock_lock(&THREAD->lock);
  289. #ifndef CONFIG_FPU_LAZY
  290.         fpu_context_save(THREAD->saved_fpu_context);
  291. #endif
  292.         if (!context_save(&THREAD->saved_context)) {
  293.             /*
  294.              * This is the place where threads leave scheduler();
  295.              */
  296.             spinlock_unlock(&THREAD->lock);
  297.             interrupts_restore(THREAD->saved_context.ipl);
  298.            
  299.             return;
  300.         }
  301.  
  302.         /*
  303.          * Interrupt priority level of preempted thread is recorded here
  304.          * to facilitate scheduler() invocations from interrupts_disable()'d
  305.          * code (e.g. waitq_sleep_timeout()).
  306.          */
  307.         THREAD->saved_context.ipl = ipl;
  308.     }
  309.  
  310.     /*
  311.      * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM
  312.      * and preemption counter. At this point THE could be coming either
  313.      * from THREAD's or CPU's stack.
  314.      */
  315.     the_copy(THE, (the_t *) CPU->stack);
  316.  
  317.     /*
  318.      * We may not keep the old stack.
  319.      * Reason: If we kept the old stack and got blocked, for instance, in
  320.      * find_best_thread(), the old thread could get rescheduled by another
  321.      * CPU and overwrite the part of its own stack that was also used by
  322.      * the scheduler on this CPU.
  323.      *
  324.      * Moreover, we have to bypass the compiler-generated POP sequence
  325.      * which is fooled by SP being set to the very top of the stack.
  326.      * Therefore the scheduler() function continues in
  327.      * scheduler_separated_stack().
  328.      */
  329.     context_save(&CPU->saved_context);
  330.     context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), (__address) CPU->stack, CPU_STACK_SIZE);
  331.     context_restore(&CPU->saved_context);
  332.     /* not reached */
  333. }
  334.  
  335. /** Scheduler stack switch wrapper
  336.  *
  337.  * Second part of the scheduler() function
  338.  * using new stack. Handling the actual context
  339.  * switch to a new thread.
  340.  *
  341.  * Assume THREAD->lock is held.
  342.  */
  343. void scheduler_separated_stack(void)
  344. {
  345.     int priority;
  346.    
  347.     ASSERT(CPU != NULL);
  348.    
  349.     if (THREAD) {
  350.         /* must be run after the switch to scheduler stack */
  351.         after_thread_ran();
  352.  
  353.         switch (THREAD->state) {
  354.             case Running:
  355.             THREAD->state = Ready;
  356.             spinlock_unlock(&THREAD->lock);
  357.             thread_ready(THREAD);
  358.             break;
  359.  
  360.             case Exiting:
  361.             thread_destroy(THREAD);
  362.             break;
  363.            
  364.             case Sleeping:
  365.             /*
  366.              * Prefer the thread after it's woken up.
  367.              */
  368.             THREAD->priority = -1;
  369.  
  370.             /*
  371.              * We need to release wq->lock which we locked in waitq_sleep().
  372.              * Address of wq->lock is kept in THREAD->sleep_queue.
  373.              */
  374.             spinlock_unlock(&THREAD->sleep_queue->lock);
  375.  
  376.             /*
  377.              * Check for possible requests for out-of-context invocation.
  378.              */
  379.             if (THREAD->call_me) {
  380.                 THREAD->call_me(THREAD->call_me_with);
  381.                 THREAD->call_me = NULL;
  382.                 THREAD->call_me_with = NULL;
  383.             }
  384.  
  385.             spinlock_unlock(&THREAD->lock);
  386.  
  387.             break;
  388.  
  389.             default:
  390.             /*
  391.              * Entering state is unexpected.
  392.              */
  393.             panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
  394.             break;
  395.         }
  396.  
  397.         THREAD = NULL;
  398.     }
  399.  
  400.     THREAD = find_best_thread();
  401.    
  402.     spinlock_lock(&THREAD->lock);
  403.     priority = THREAD->priority;
  404.     spinlock_unlock(&THREAD->lock);
  405.  
  406.     relink_rq(priority);       
  407.  
  408.     spinlock_lock(&THREAD->lock);  
  409.  
  410.     /*
  411.      * If both the old and the new task are the same, lots of work is avoided.
  412.      */
  413.     if (TASK != THREAD->task) {
  414.         as_t *as1 = NULL;
  415.         as_t *as2;
  416.  
  417.         if (TASK) {
  418.             spinlock_lock(&TASK->lock);
  419.             as1 = TASK->as;
  420.             spinlock_unlock(&TASK->lock);
  421.         }
  422.  
  423.         spinlock_lock(&THREAD->task->lock);
  424.         as2 = THREAD->task->as;
  425.         spinlock_unlock(&THREAD->task->lock);
  426.        
  427.         /*
  428.          * Note that it is possible for two tasks to share one address space.
  429.          */
  430.         if (as1 != as2) {
  431.             /*
  432.              * Both tasks and address spaces are different.
  433.              * Replace the old one with the new one.
  434.              */
  435.             as_switch(as1, as2);
  436.         }
  437.         TASK = THREAD->task;
  438.     }
  439.  
  440.     THREAD->state = Running;
  441.  
  442. #ifdef SCHEDULER_VERBOSE
  443.     printf("cpu%d: tid %d (priority=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks, atomic_get(&CPU->nrdy));
  444. #endif 
  445.  
  446.     /*
  447.      * Some architectures provide late kernel PA2KA(identity)
  448.      * mapping in a page fault handler. However, the page fault
  449.      * handler uses the kernel stack of the running thread and
  450.      * therefore cannot be used to map it. The kernel stack, if
  451.      * necessary, is to be mapped in before_thread_runs(). This
  452.      * function must be executed before the switch to the new stack.
  453.      */
  454.     before_thread_runs();
  455.  
  456.     /*
  457.      * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack.
  458.      */
  459.     the_copy(THE, (the_t *) THREAD->kstack);
  460.    
  461.     context_restore(&THREAD->saved_context);
  462.     /* not reached */
  463. }
  464.  
  465. #ifdef CONFIG_SMP
  466. /** Load balancing thread
  467.  *
  468.  * SMP load balancing thread, supervising thread supplies
  469.  * for the CPU it's wired to.
  470.  *
  471.  * @param arg Generic thread argument (unused).
  472.  *
  473.  */
  474. void kcpulb(void *arg)
  475. {
  476.     thread_t *t;
  477.     int count, average, i, j, k = 0;
  478.     ipl_t ipl;
  479.  
  480. loop:
  481.     /*
  482.      * Work in 1s intervals.
  483.      */
  484.     thread_sleep(1);
  485.  
  486. not_satisfied:
  487.     /*
  488.      * Calculate the number of threads that will be migrated/stolen from
  489.      * other CPU's. Note that situation can have changed between two
  490.      * passes. Each time get the most up to date counts.
  491.      */
  492.     average = atomic_get(&nrdy) / config.cpu_active + 1;
  493.     count = average - atomic_get(&CPU->nrdy);
  494.  
  495.     if (count <= 0)
  496.         goto satisfied;
  497.  
  498.     /*
  499.      * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
  500.      */
  501.     for (j=RQ_COUNT-1; j >= 0; j--) {
  502.         for (i=0; i < config.cpu_active; i++) {
  503.             link_t *l;
  504.             runq_t *r;
  505.             cpu_t *cpu;
  506.  
  507.             cpu = &cpus[(i + k) % config.cpu_active];
  508.  
  509.             /*
  510.              * Not interested in ourselves.
  511.              * Doesn't require interrupt disabling for kcpulb is X_WIRED.
  512.              */
  513.             if (CPU == cpu)
  514.                 continue;
  515.             if (atomic_get(&cpu->nrdy) <= average)
  516.                 continue;
  517.  
  518.             ipl = interrupts_disable();
  519.             r = &cpu->rq[j];
  520.             spinlock_lock(&r->lock);
  521.             if (r->n == 0) {
  522.                 spinlock_unlock(&r->lock);
  523.                 interrupts_restore(ipl);
  524.                 continue;
  525.             }
  526.        
  527.             t = NULL;
  528.             l = r->rq_head.prev;    /* search rq from the back */
  529.             while (l != &r->rq_head) {
  530.                 t = list_get_instance(l, thread_t, rq_link);
  531.                 /*
  532.                  * We don't want to steal CPU-wired threads neither threads already stolen.
  533.                  * The latter prevents threads from migrating between CPU's without ever being run.
  534.                  * We don't want to steal threads whose FPU context is still in CPU.
  535.                  */
  536.                 spinlock_lock(&t->lock);
  537.                 if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
  538.                     /*
  539.                      * Remove t from r.
  540.                      */
  541.                     spinlock_unlock(&t->lock);
  542.                    
  543.                     atomic_dec(&cpu->nrdy);
  544.                     atomic_dec(&nrdy);
  545.  
  546.                     r->n--;
  547.                     list_remove(&t->rq_link);
  548.  
  549.                     break;
  550.                 }
  551.                 spinlock_unlock(&t->lock);
  552.                 l = l->prev;
  553.                 t = NULL;
  554.             }
  555.             spinlock_unlock(&r->lock);
  556.  
  557.             if (t) {
  558.                 /*
  559.                  * Ready t on local CPU
  560.                  */
  561.                 spinlock_lock(&t->lock);
  562. #ifdef KCPULB_VERBOSE
  563.                 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);
  564. #endif
  565.                 t->flags |= X_STOLEN;
  566.                 spinlock_unlock(&t->lock);
  567.    
  568.                 thread_ready(t);
  569.  
  570.                 interrupts_restore(ipl);
  571.    
  572.                 if (--count == 0)
  573.                     goto satisfied;
  574.                    
  575.                 /*
  576.                  * We are not satisfied yet, focus on another CPU next time.
  577.                  */
  578.                 k++;
  579.                
  580.                 continue;
  581.             }
  582.             interrupts_restore(ipl);
  583.         }
  584.     }
  585.  
  586.     if (atomic_get(&CPU->nrdy)) {
  587.         /*
  588.          * Be a little bit light-weight and let migrated threads run.
  589.          */
  590.         scheduler();
  591.     } else {
  592.         /*
  593.          * We failed to migrate a single thread.
  594.          * Give up this turn.
  595.          */
  596.         goto loop;
  597.     }
  598.        
  599.     goto not_satisfied;
  600.  
  601. satisfied:
  602.     goto loop;
  603. }
  604.  
  605. #endif /* CONFIG_SMP */
  606.  
  607.  
  608. /** Print information about threads & scheduler queues */
  609. void sched_print_list(void)
  610. {
  611.     ipl_t ipl;
  612.     int cpu,i;
  613.     runq_t *r;
  614.     thread_t *t;
  615.     link_t *cur;
  616.  
  617.     /* We are going to mess with scheduler structures,
  618.      * let's not be interrupted */
  619.     ipl = interrupts_disable();
  620.     printf("Scheduler dump:\n");
  621.     for (cpu=0;cpu < config.cpu_count; cpu++) {
  622.  
  623.         if (!cpus[cpu].active)
  624.             continue;
  625.  
  626.         spinlock_lock(&cpus[cpu].lock);
  627.         printf("cpu%d: nrdy: %d, needs_relink: %d\n",
  628.                cpus[cpu].id, atomic_get(&cpus[cpu].nrdy), cpus[cpu].needs_relink);
  629.        
  630.         for (i=0; i<RQ_COUNT; i++) {
  631.             r = &cpus[cpu].rq[i];
  632.             spinlock_lock(&r->lock);
  633.             if (!r->n) {
  634.                 spinlock_unlock(&r->lock);
  635.                 continue;
  636.             }
  637.             printf("\trq[%d]: ", i);
  638.             for (cur=r->rq_head.next; cur!=&r->rq_head; cur=cur->next) {
  639.                 t = list_get_instance(cur, thread_t, rq_link);
  640.                 printf("%d(%s) ", t->tid,
  641.                        thread_states[t->state]);
  642.             }
  643.             printf("\n");
  644.             spinlock_unlock(&r->lock);
  645.         }
  646.         spinlock_unlock(&cpus[cpu].lock);
  647.     }
  648.    
  649.     interrupts_restore(ipl);
  650. }
  651.