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 <linux/gfp.h> 31 #include "cpupri.h" 32 33 /* Convert between a 140 based task->prio, and our 102 based cpupri */ 34 static int convert_prio(int prio) 35 { 36 int cpupri; 37 38 if (prio == CPUPRI_INVALID) 39 cpupri = CPUPRI_INVALID; 40 else if (prio == MAX_PRIO) 41 cpupri = CPUPRI_IDLE; 42 else if (prio >= MAX_RT_PRIO) 43 cpupri = CPUPRI_NORMAL; 44 else 45 cpupri = MAX_RT_PRIO - prio + 1; 46 47 return cpupri; 48 } 49 50 /** 51 * cpupri_find - find the best (lowest-pri) CPU in the system 52 * @cp: The cpupri context 53 * @p: The task 54 * @lowest_mask: A mask to fill in with selected CPUs (or NULL) 55 * 56 * Note: This function returns the recommended CPUs as calculated during the 57 * current invocation. By the time the call returns, the CPUs may have in 58 * fact changed priorities any number of times. While not ideal, it is not 59 * an issue of correctness since the normal rebalancer logic will correct 60 * any discrepancies created by racing against the uncertainty of the current 61 * priority configuration. 62 * 63 * Returns: (int)bool - CPUs were found 64 */ 65 int cpupri_find(struct cpupri *cp, struct task_struct *p, 66 struct cpumask *lowest_mask) 67 { 68 int idx = 0; 69 int task_pri = convert_prio(p->prio); 70 71 if (task_pri >= MAX_RT_PRIO) 72 return 0; 73 74 for (idx = 0; idx < task_pri; idx++) { 75 struct cpupri_vec *vec = &cp->pri_to_cpu[idx]; 76 int skip = 0; 77 78 if (!atomic_read(&(vec)->count)) 79 skip = 1; 80 /* 81 * When looking at the vector, we need to read the counter, 82 * do a memory barrier, then read the mask. 83 * 84 * Note: This is still all racey, but we can deal with it. 85 * Ideally, we only want to look at masks that are set. 86 * 87 * If a mask is not set, then the only thing wrong is that we 88 * did a little more work than necessary. 89 * 90 * If we read a zero count but the mask is set, because of the 91 * memory barriers, that can only happen when the highest prio 92 * task for a run queue has left the run queue, in which case, 93 * it will be followed by a pull. If the task we are processing 94 * fails to find a proper place to go, that pull request will 95 * pull this task if the run queue is running at a lower 96 * priority. 97 */ 98 smp_rmb(); 99 100 /* Need to do the rmb for every iteration */ 101 if (skip) 102 continue; 103 104 if (cpumask_any_and(&p->cpus_allowed, vec->mask) >= nr_cpu_ids) 105 continue; 106 107 if (lowest_mask) { 108 cpumask_and(lowest_mask, &p->cpus_allowed, vec->mask); 109 110 /* 111 * We have to ensure that we have at least one bit 112 * still set in the array, since the map could have 113 * been concurrently emptied between the first and 114 * second reads of vec->mask. If we hit this 115 * condition, simply act as though we never hit this 116 * priority level and continue on. 117 */ 118 if (cpumask_any(lowest_mask) >= nr_cpu_ids) 119 continue; 120 } 121 122 return 1; 123 } 124 125 return 0; 126 } 127 128 /** 129 * cpupri_set - update the cpu priority setting 130 * @cp: The cpupri context 131 * @cpu: The target cpu 132 * @newpri: The priority (INVALID-RT99) to assign to this CPU 133 * 134 * Note: Assumes cpu_rq(cpu)->lock is locked 135 * 136 * Returns: (void) 137 */ 138 void cpupri_set(struct cpupri *cp, int cpu, int newpri) 139 { 140 int *currpri = &cp->cpu_to_pri[cpu]; 141 int oldpri = *currpri; 142 int do_mb = 0; 143 144 newpri = convert_prio(newpri); 145 146 BUG_ON(newpri >= CPUPRI_NR_PRIORITIES); 147 148 if (newpri == oldpri) 149 return; 150 151 /* 152 * If the cpu was currently mapped to a different value, we 153 * need to map it to the new value then remove the old value. 154 * Note, we must add the new value first, otherwise we risk the 155 * cpu being missed by the priority loop in cpupri_find. 156 */ 157 if (likely(newpri != CPUPRI_INVALID)) { 158 struct cpupri_vec *vec = &cp->pri_to_cpu[newpri]; 159 160 cpumask_set_cpu(cpu, vec->mask); 161 /* 162 * When adding a new vector, we update the mask first, 163 * do a write memory barrier, and then update the count, to 164 * make sure the vector is visible when count is set. 165 */ 166 smp_mb__before_atomic_inc(); 167 atomic_inc(&(vec)->count); 168 do_mb = 1; 169 } 170 if (likely(oldpri != CPUPRI_INVALID)) { 171 struct cpupri_vec *vec = &cp->pri_to_cpu[oldpri]; 172 173 /* 174 * Because the order of modification of the vec->count 175 * is important, we must make sure that the update 176 * of the new prio is seen before we decrement the 177 * old prio. This makes sure that the loop sees 178 * one or the other when we raise the priority of 179 * the run queue. We don't care about when we lower the 180 * priority, as that will trigger an rt pull anyway. 181 * 182 * We only need to do a memory barrier if we updated 183 * the new priority vec. 184 */ 185 if (do_mb) 186 smp_mb__after_atomic_inc(); 187 188 /* 189 * When removing from the vector, we decrement the counter first 190 * do a memory barrier and then clear the mask. 191 */ 192 atomic_dec(&(vec)->count); 193 smp_mb__after_atomic_inc(); 194 cpumask_clear_cpu(cpu, vec->mask); 195 } 196 197 *currpri = newpri; 198 } 199 200 /** 201 * cpupri_init - initialize the cpupri structure 202 * @cp: The cpupri context 203 * 204 * Returns: -ENOMEM if memory fails. 205 */ 206 int cpupri_init(struct cpupri *cp) 207 { 208 int i; 209 210 memset(cp, 0, sizeof(*cp)); 211 212 for (i = 0; i < CPUPRI_NR_PRIORITIES; i++) { 213 struct cpupri_vec *vec = &cp->pri_to_cpu[i]; 214 215 atomic_set(&vec->count, 0); 216 if (!zalloc_cpumask_var(&vec->mask, GFP_KERNEL)) 217 goto cleanup; 218 } 219 220 for_each_possible_cpu(i) 221 cp->cpu_to_pri[i] = CPUPRI_INVALID; 222 return 0; 223 224 cleanup: 225 for (i--; i >= 0; i--) 226 free_cpumask_var(cp->pri_to_cpu[i].mask); 227 return -ENOMEM; 228 } 229 230 /** 231 * cpupri_cleanup - clean up the cpupri structure 232 * @cp: The cpupri context 233 */ 234 void cpupri_cleanup(struct cpupri *cp) 235 { 236 int i; 237 238 for (i = 0; i < CPUPRI_NR_PRIORITIES; i++) 239 free_cpumask_var(cp->pri_to_cpu[i].mask); 240 } 241