2 * kernel/sched_cpupri.c
4 * CPU priority management
6 * Copyright (C) 2007-2008 Novell
8 * Author: Gregory Haskins <ghaskins@novell.com>
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:
13 * (INVALID), IDLE, NORMAL, RT1, ... RT99
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.
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
30 #include <linux/gfp.h>
31 #include "sched_cpupri.h"
33 /* Convert between a 140 based task->prio, and our 102 based cpupri */
34 static int convert_prio(int prio)
38 if (prio == CPUPRI_INVALID)
39 cpupri = CPUPRI_INVALID;
40 else if (prio == MAX_PRIO)
42 else if (prio >= MAX_RT_PRIO)
43 cpupri = CPUPRI_NORMAL;
45 cpupri = MAX_RT_PRIO - prio + 1;
51 * cpupri_find - find the best (lowest-pri) CPU in the system
52 * @cp: The cpupri context
54 * @lowest_mask: A mask to fill in with selected CPUs (or NULL)
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.
63 * Returns: (int)bool - CPUs were found
65 int cpupri_find(struct cpupri *cp, struct task_struct *p,
66 struct cpumask *lowest_mask)
69 int task_pri = convert_prio(p->prio);
71 if (task_pri >= MAX_RT_PRIO)
74 for (idx = 0; idx < task_pri; idx++) {
75 struct cpupri_vec *vec = &cp->pri_to_cpu[idx];
77 if (!atomic_read(&(vec)->count))
80 * When looking at the vector, we need to read the counter,
81 * do a memory barrier, then read the mask.
83 * Note: This is still all racey, but we can deal with it.
84 * Ideally, we only want to look at masks that are set.
86 * If a mask is not set, then the only thing wrong is that we
87 * did a little more work than necessary.
89 * If we read a zero count but the mask is set, because of the
90 * memory barriers, that can only happen when the highest prio
91 * task for a run queue has left the run queue, in which case,
92 * it will be followed by a pull. If the task we are processing
93 * fails to find a proper place to go, that pull request will
94 * pull this task if the run queue is running at a lower
99 if (cpumask_any_and(&p->cpus_allowed, vec->mask) >= nr_cpu_ids)
103 cpumask_and(lowest_mask, &p->cpus_allowed, vec->mask);
106 * We have to ensure that we have at least one bit
107 * still set in the array, since the map could have
108 * been concurrently emptied between the first and
109 * second reads of vec->mask. If we hit this
110 * condition, simply act as though we never hit this
111 * priority level and continue on.
113 if (cpumask_any(lowest_mask) >= nr_cpu_ids)
124 * cpupri_set - update the cpu priority setting
125 * @cp: The cpupri context
126 * @cpu: The target cpu
127 * @pri: The priority (INVALID-RT99) to assign to this CPU
129 * Note: Assumes cpu_rq(cpu)->lock is locked
133 void cpupri_set(struct cpupri *cp, int cpu, int newpri)
135 int *currpri = &cp->cpu_to_pri[cpu];
136 int oldpri = *currpri;
138 newpri = convert_prio(newpri);
140 BUG_ON(newpri >= CPUPRI_NR_PRIORITIES);
142 if (newpri == oldpri)
146 * If the cpu was currently mapped to a different value, we
147 * need to map it to the new value then remove the old value.
148 * Note, we must add the new value first, otherwise we risk the
149 * cpu being cleared from pri_active, and this cpu could be
150 * missed for a push or pull.
152 if (likely(newpri != CPUPRI_INVALID)) {
153 struct cpupri_vec *vec = &cp->pri_to_cpu[newpri];
155 cpumask_set_cpu(cpu, vec->mask);
157 * When adding a new vector, we update the mask first,
158 * do a write memory barrier, and then update the count, to
159 * make sure the vector is visible when count is set.
162 atomic_inc(&(vec)->count);
164 if (likely(oldpri != CPUPRI_INVALID)) {
165 struct cpupri_vec *vec = &cp->pri_to_cpu[oldpri];
168 * When removing from the vector, we decrement the counter first
169 * do a memory barrier and then clear the mask.
171 atomic_dec(&(vec)->count);
173 cpumask_clear_cpu(cpu, vec->mask);
180 * cpupri_init - initialize the cpupri structure
181 * @cp: The cpupri context
182 * @bootmem: true if allocations need to use bootmem
184 * Returns: -ENOMEM if memory fails.
186 int cpupri_init(struct cpupri *cp)
190 memset(cp, 0, sizeof(*cp));
192 for (i = 0; i < CPUPRI_NR_PRIORITIES; i++) {
193 struct cpupri_vec *vec = &cp->pri_to_cpu[i];
195 atomic_set(&vec->count, 0);
196 if (!zalloc_cpumask_var(&vec->mask, GFP_KERNEL))
200 for_each_possible_cpu(i)
201 cp->cpu_to_pri[i] = CPUPRI_INVALID;
205 for (i--; i >= 0; i--)
206 free_cpumask_var(cp->pri_to_cpu[i].mask);
211 * cpupri_cleanup - clean up the cpupri structure
212 * @cp: The cpupri context
214 void cpupri_cleanup(struct cpupri *cp)
218 for (i = 0; i < CPUPRI_NR_PRIORITIES; i++)
219 free_cpumask_var(cp->pri_to_cpu[i].mask);