Commit | Line | Data |
---|---|---|
6e0534f2 GH |
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) \ | |
984b3f57 | 50 | for_each_set_bit(idx, array, CPUPRI_NR_PRIORITIES) |
6e0534f2 GH |
51 | |
52 | /** | |
53 | * cpupri_find - find the best (lowest-pri) CPU in the system | |
54 | * @cp: The cpupri context | |
55 | * @p: The task | |
13b8bd0a | 56 | * @lowest_mask: A mask to fill in with selected CPUs (or NULL) |
6e0534f2 GH |
57 | * |
58 | * Note: This function returns the recommended CPUs as calculated during the | |
2a61aa40 | 59 | * current invocation. By the time the call returns, the CPUs may have in |
6e0534f2 GH |
60 | * fact changed priorities any number of times. While not ideal, it is not |
61 | * an issue of correctness since the normal rebalancer logic will correct | |
62 | * any discrepancies created by racing against the uncertainty of the current | |
63 | * priority configuration. | |
64 | * | |
65 | * Returns: (int)bool - CPUs were found | |
66 | */ | |
67 | int cpupri_find(struct cpupri *cp, struct task_struct *p, | |
68e74568 | 68 | struct cpumask *lowest_mask) |
6e0534f2 GH |
69 | { |
70 | int idx = 0; | |
71 | int task_pri = convert_prio(p->prio); | |
72 | ||
73 | for_each_cpupri_active(cp->pri_active, idx) { | |
74 | struct cpupri_vec *vec = &cp->pri_to_cpu[idx]; | |
6e0534f2 GH |
75 | |
76 | if (idx >= task_pri) | |
77 | break; | |
78 | ||
68e74568 | 79 | if (cpumask_any_and(&p->cpus_allowed, vec->mask) >= nr_cpu_ids) |
6e0534f2 GH |
80 | continue; |
81 | ||
07903af1 | 82 | if (lowest_mask) { |
13b8bd0a | 83 | cpumask_and(lowest_mask, &p->cpus_allowed, vec->mask); |
07903af1 GH |
84 | |
85 | /* | |
86 | * We have to ensure that we have at least one bit | |
87 | * still set in the array, since the map could have | |
88 | * been concurrently emptied between the first and | |
89 | * second reads of vec->mask. If we hit this | |
90 | * condition, simply act as though we never hit this | |
91 | * priority level and continue on. | |
92 | */ | |
93 | if (cpumask_any(lowest_mask) >= nr_cpu_ids) | |
94 | continue; | |
95 | } | |
96 | ||
6e0534f2 GH |
97 | return 1; |
98 | } | |
99 | ||
100 | return 0; | |
101 | } | |
102 | ||
103 | /** | |
104 | * cpupri_set - update the cpu priority setting | |
105 | * @cp: The cpupri context | |
106 | * @cpu: The target cpu | |
107 | * @pri: The priority (INVALID-RT99) to assign to this CPU | |
108 | * | |
109 | * Note: Assumes cpu_rq(cpu)->lock is locked | |
110 | * | |
111 | * Returns: (void) | |
112 | */ | |
113 | void cpupri_set(struct cpupri *cp, int cpu, int newpri) | |
114 | { | |
115 | int *currpri = &cp->cpu_to_pri[cpu]; | |
116 | int oldpri = *currpri; | |
117 | unsigned long flags; | |
118 | ||
119 | newpri = convert_prio(newpri); | |
120 | ||
121 | BUG_ON(newpri >= CPUPRI_NR_PRIORITIES); | |
122 | ||
123 | if (newpri == oldpri) | |
124 | return; | |
125 | ||
126 | /* | |
127 | * If the cpu was currently mapped to a different value, we | |
c3a2ae3d SR |
128 | * need to map it to the new value then remove the old value. |
129 | * Note, we must add the new value first, otherwise we risk the | |
130 | * cpu being cleared from pri_active, and this cpu could be | |
131 | * missed for a push or pull. | |
6e0534f2 | 132 | */ |
6e0534f2 GH |
133 | if (likely(newpri != CPUPRI_INVALID)) { |
134 | struct cpupri_vec *vec = &cp->pri_to_cpu[newpri]; | |
135 | ||
fe841226 | 136 | raw_spin_lock_irqsave(&vec->lock, flags); |
6e0534f2 | 137 | |
68e74568 | 138 | cpumask_set_cpu(cpu, vec->mask); |
6e0534f2 GH |
139 | vec->count++; |
140 | if (vec->count == 1) | |
141 | set_bit(newpri, cp->pri_active); | |
142 | ||
fe841226 | 143 | raw_spin_unlock_irqrestore(&vec->lock, flags); |
6e0534f2 | 144 | } |
c3a2ae3d SR |
145 | if (likely(oldpri != CPUPRI_INVALID)) { |
146 | struct cpupri_vec *vec = &cp->pri_to_cpu[oldpri]; | |
147 | ||
fe841226 | 148 | raw_spin_lock_irqsave(&vec->lock, flags); |
c3a2ae3d SR |
149 | |
150 | vec->count--; | |
151 | if (!vec->count) | |
152 | clear_bit(oldpri, cp->pri_active); | |
153 | cpumask_clear_cpu(cpu, vec->mask); | |
154 | ||
fe841226 | 155 | raw_spin_unlock_irqrestore(&vec->lock, flags); |
c3a2ae3d | 156 | } |
6e0534f2 GH |
157 | |
158 | *currpri = newpri; | |
159 | } | |
160 | ||
161 | /** | |
162 | * cpupri_init - initialize the cpupri structure | |
163 | * @cp: The cpupri context | |
68e74568 | 164 | * @bootmem: true if allocations need to use bootmem |
6e0534f2 | 165 | * |
68e74568 | 166 | * Returns: -ENOMEM if memory fails. |
6e0534f2 | 167 | */ |
fd5e1b5d | 168 | int cpupri_init(struct cpupri *cp, bool bootmem) |
6e0534f2 | 169 | { |
0fb53029 | 170 | gfp_t gfp = GFP_KERNEL; |
6e0534f2 GH |
171 | int i; |
172 | ||
0fb53029 PE |
173 | if (bootmem) |
174 | gfp = GFP_NOWAIT; | |
175 | ||
6e0534f2 GH |
176 | memset(cp, 0, sizeof(*cp)); |
177 | ||
178 | for (i = 0; i < CPUPRI_NR_PRIORITIES; i++) { | |
179 | struct cpupri_vec *vec = &cp->pri_to_cpu[i]; | |
180 | ||
fe841226 | 181 | raw_spin_lock_init(&vec->lock); |
6e0534f2 | 182 | vec->count = 0; |
0fb53029 | 183 | if (!zalloc_cpumask_var(&vec->mask, gfp)) |
68e74568 | 184 | goto cleanup; |
6e0534f2 GH |
185 | } |
186 | ||
187 | for_each_possible_cpu(i) | |
188 | cp->cpu_to_pri[i] = CPUPRI_INVALID; | |
68e74568 RR |
189 | return 0; |
190 | ||
191 | cleanup: | |
192 | for (i--; i >= 0; i--) | |
193 | free_cpumask_var(cp->pri_to_cpu[i].mask); | |
194 | return -ENOMEM; | |
6e0534f2 GH |
195 | } |
196 | ||
68e74568 RR |
197 | /** |
198 | * cpupri_cleanup - clean up the cpupri structure | |
199 | * @cp: The cpupri context | |
200 | */ | |
201 | void cpupri_cleanup(struct cpupri *cp) | |
202 | { | |
203 | int i; | |
6e0534f2 | 204 | |
68e74568 RR |
205 | for (i = 0; i < CPUPRI_NR_PRIORITIES; i++) |
206 | free_cpumask_var(cp->pri_to_cpu[i].mask); | |
207 | } |