kernel/sched_cpupri.c

   1 /*
   2  *  kernel/sched_cpupri.c
   3  *
   4  *  CPU priority management
   5  *
   6  *  Copyright (C) 2007-2008 Novell
   7  *
   8  *  Author: Gregory Haskins <ghaskins@novell.com>
   9  *
  10  *  This code tracks the priority of each CPU so that global migration
  11  *  decisions are easy to calculate.  Each CPU can be in a state as follows:
  12  *
  13  *                 (INVALID), IDLE, NORMAL, RT1, ... RT99
  14  *
  15  *  going from the lowest priority to the highest.  CPUs in the INVALID state
  16  *  are not eligible for routing.  The system maintains this state with
  17  *  a 2 dimensional bitmap (the first for priority class, the second for cpus
  18  *  in that class).  Therefore a typical application without affinity
  19  *  restrictions can find a suitable CPU with O(1) complexity (e.g. two bit
  20  *  searches).  For tasks with affinity restrictions, the algorithm has a
  21  *  worst case complexity of O(min(102, nr_domcpus)), though the scenario that
  22  *  yields the worst case search is fairly contrived.
  23  *
  24  *  This program is free software; you can redistribute it and/or
  25  *  modify it under the terms of the GNU General Public License
  26  *  as published by the Free Software Foundation; version 2
  27  *  of the License.
  28  */
  29
  30 #include "sched_cpupri.h"
  31
  32 /* Convert between a 140 based task->prio, and our 102 based cpupri */
  33 static int convert_prio(int prio)
  34 {
  35         int cpupri;
  36
  37         if (prio == CPUPRI_INVALID)
  38                 cpupri = CPUPRI_INVALID;
  39         else if (prio == MAX_PRIO)
  40                 cpupri = CPUPRI_IDLE;
  41         else if (prio >= MAX_RT_PRIO)
  42                 cpupri = CPUPRI_NORMAL;
  43         else
  44                 cpupri = MAX_RT_PRIO - prio + 1;
  45
  46         return cpupri;
  47 }
  48
  49 #define for_each_cpupri_active(array, idx)                    \
  50   for (idx = find_first_bit(array, CPUPRI_NR_PRIORITIES);     \
  51        idx < CPUPRI_NR_PRIORITIES;                            \
  52        idx = find_next_bit(array, CPUPRI_NR_PRIORITIES, idx+1))
  53
  54 /**
  55  * cpupri_find - find the best (lowest-pri) CPU in the system
  56  * @cp: The cpupri context
  57  * @p: The task
  58  * @lowest_mask: A mask to fill in with selected CPUs (or NULL)
  59  *
  60  * Note: This function returns the recommended CPUs as calculated during the
  61  * current invokation.  By the time the call returns, the CPUs may have in
  62  * fact changed priorities any number of times.  While not ideal, it is not
  63  * an issue of correctness since the normal rebalancer logic will correct
  64  * any discrepancies created by racing against the uncertainty of the current
  65  * priority configuration.
  66  *
  67  * Returns: (int)bool - CPUs were found
  68  */
  69 int cpupri_find(struct cpupri *cp, struct task_struct *p,
  70                 struct cpumask *lowest_mask)
  71 {
  72         int                  idx      = 0;
  73         int                  task_pri = convert_prio(p->prio);
  74
  75         for_each_cpupri_active(cp->pri_active, idx) {
  76                 struct cpupri_vec *vec  = &cp->pri_to_cpu[idx];
  77
  78                 if (idx >= task_pri)
  79                         break;
  80
  81                 if (cpumask_any_and(&p->cpus_allowed, vec->mask) >= nr_cpu_ids)
  82                         continue;
  83
  84                 if (lowest_mask)
  85                         cpumask_and(lowest_mask, &p->cpus_allowed, vec->mask);
  86                 return 1;
  87         }
  88
  89         return 0;
  90 }
  91
  92 /**
  93  * cpupri_set - update the cpu priority setting
  94  * @cp: The cpupri context
  95  * @cpu: The target cpu
  96  * @pri: The priority (INVALID-RT99) to assign to this CPU
  97  *
  98  * Note: Assumes cpu_rq(cpu)->lock is locked
  99  *
 100  * Returns: (void)
 101  */
 102 void cpupri_set(struct cpupri *cp, int cpu, int newpri)
 103 {
 104         int                 *currpri = &cp->cpu_to_pri[cpu];
 105         int                  oldpri  = *currpri;
 106         unsigned long        flags;
 107
 108         newpri = convert_prio(newpri);
 109
 110         BUG_ON(newpri >= CPUPRI_NR_PRIORITIES);
 111
 112         if (newpri == oldpri)
 113                 return;
 114
 115         /*
 116          * If the cpu was currently mapped to a different value, we
 117          * first need to unmap the old value
 118          */
 119         if (likely(oldpri != CPUPRI_INVALID)) {
 120                 struct cpupri_vec *vec  = &cp->pri_to_cpu[oldpri];
 121
 122                 spin_lock_irqsave(&vec->lock, flags);
 123
 124                 vec->count--;
 125                 if (!vec->count)
 126                         clear_bit(oldpri, cp->pri_active);
 127                 cpumask_clear_cpu(cpu, vec->mask);
 128
 129                 spin_unlock_irqrestore(&vec->lock, flags);
 130         }
 131
 132         if (likely(newpri != CPUPRI_INVALID)) {
 133                 struct cpupri_vec *vec = &cp->pri_to_cpu[newpri];
 134
 135                 spin_lock_irqsave(&vec->lock, flags);
 136
 137                 cpumask_set_cpu(cpu, vec->mask);
 138                 vec->count++;
 139                 if (vec->count == 1)
 140                         set_bit(newpri, cp->pri_active);
 141
 142                 spin_unlock_irqrestore(&vec->lock, flags);
 143         }
 144
 145         *currpri = newpri;
 146 }
 147
 148 /**
 149  * cpupri_init - initialize the cpupri structure
 150  * @cp: The cpupri context
 151  * @bootmem: true if allocations need to use bootmem
 152  *
 153  * Returns: -ENOMEM if memory fails.
 154  */
 155 int __init_refok cpupri_init(struct cpupri *cp, bool bootmem)
 156 {
 157         int i;
 158
 159         memset(cp, 0, sizeof(*cp));
 160
 161         for (i = 0; i < CPUPRI_NR_PRIORITIES; i++) {
 162                 struct cpupri_vec *vec = &cp->pri_to_cpu[i];
 163
 164                 spin_lock_init(&vec->lock);
 165                 vec->count = 0;
 166                 if (bootmem)
 167                         alloc_bootmem_cpumask_var(&vec->mask);
 168                 else if (!alloc_cpumask_var(&vec->mask, GFP_KERNEL))
 169                         goto cleanup;
 170         }
 171
 172         for_each_possible_cpu(i)
 173                 cp->cpu_to_pri[i] = CPUPRI_INVALID;
 174         return 0;
 175
 176 cleanup:
 177         for (i--; i >= 0; i--)
 178                 free_cpumask_var(cp->pri_to_cpu[i].mask);
 179         return -ENOMEM;
 180 }
 181
 182 /**
 183  * cpupri_cleanup - clean up the cpupri structure
 184  * @cp: The cpupri context
 185  */
 186 void cpupri_cleanup(struct cpupri *cp)
 187 {
 188         int i;
 189
 190         for (i = 0; i < CPUPRI_NR_PRIORITIES; i++)
 191                 free_cpumask_var(cp->pri_to_cpu[i].mask);
 192 }