xref: /linux/kernel/sched/cpupri.c (revision 17cfcb68af3bc7d5e8ae08779b1853310a2949f3)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  kernel/sched/cpupri.c
4  *
5  *  CPU priority management
6  *
7  *  Copyright (C) 2007-2008 Novell
8  *
9  *  Author: Gregory Haskins <ghaskins@novell.com>
10  *
11  *  This code tracks the priority of each CPU so that global migration
12  *  decisions are easy to calculate.  Each CPU can be in a state as follows:
13  *
14  *                 (INVALID), IDLE, NORMAL, RT1, ... RT99
15  *
16  *  going from the lowest priority to the highest.  CPUs in the INVALID state
17  *  are not eligible for routing.  The system maintains this state with
18  *  a 2 dimensional bitmap (the first for priority class, the second for CPUs
19  *  in that class).  Therefore a typical application without affinity
20  *  restrictions can find a suitable CPU with O(1) complexity (e.g. two bit
21  *  searches).  For tasks with affinity restrictions, the algorithm has a
22  *  worst case complexity of O(min(102, nr_domcpus)), though the scenario that
23  *  yields the worst case search is fairly contrived.
24  */
25 #include "sched.h"
26 
27 /* Convert between a 140 based task->prio, and our 102 based cpupri */
28 static int convert_prio(int prio)
29 {
30 	int cpupri;
31 
32 	if (prio == CPUPRI_INVALID)
33 		cpupri = CPUPRI_INVALID;
34 	else if (prio == MAX_PRIO)
35 		cpupri = CPUPRI_IDLE;
36 	else if (prio >= MAX_RT_PRIO)
37 		cpupri = CPUPRI_NORMAL;
38 	else
39 		cpupri = MAX_RT_PRIO - prio + 1;
40 
41 	return cpupri;
42 }
43 
44 /**
45  * cpupri_find - find the best (lowest-pri) CPU in the system
46  * @cp: The cpupri context
47  * @p: The task
48  * @lowest_mask: A mask to fill in with selected CPUs (or NULL)
49  *
50  * Note: This function returns the recommended CPUs as calculated during the
51  * current invocation.  By the time the call returns, the CPUs may have in
52  * fact changed priorities any number of times.  While not ideal, it is not
53  * an issue of correctness since the normal rebalancer logic will correct
54  * any discrepancies created by racing against the uncertainty of the current
55  * priority configuration.
56  *
57  * Return: (int)bool - CPUs were found
58  */
59 int cpupri_find(struct cpupri *cp, struct task_struct *p,
60 		struct cpumask *lowest_mask)
61 {
62 	int idx = 0;
63 	int task_pri = convert_prio(p->prio);
64 
65 	BUG_ON(task_pri >= CPUPRI_NR_PRIORITIES);
66 
67 	for (idx = 0; idx < task_pri; idx++) {
68 		struct cpupri_vec *vec  = &cp->pri_to_cpu[idx];
69 		int skip = 0;
70 
71 		if (!atomic_read(&(vec)->count))
72 			skip = 1;
73 		/*
74 		 * When looking at the vector, we need to read the counter,
75 		 * do a memory barrier, then read the mask.
76 		 *
77 		 * Note: This is still all racey, but we can deal with it.
78 		 *  Ideally, we only want to look at masks that are set.
79 		 *
80 		 *  If a mask is not set, then the only thing wrong is that we
81 		 *  did a little more work than necessary.
82 		 *
83 		 *  If we read a zero count but the mask is set, because of the
84 		 *  memory barriers, that can only happen when the highest prio
85 		 *  task for a run queue has left the run queue, in which case,
86 		 *  it will be followed by a pull. If the task we are processing
87 		 *  fails to find a proper place to go, that pull request will
88 		 *  pull this task if the run queue is running at a lower
89 		 *  priority.
90 		 */
91 		smp_rmb();
92 
93 		/* Need to do the rmb for every iteration */
94 		if (skip)
95 			continue;
96 
97 		if (cpumask_any_and(p->cpus_ptr, vec->mask) >= nr_cpu_ids)
98 			continue;
99 
100 		if (lowest_mask) {
101 			cpumask_and(lowest_mask, p->cpus_ptr, vec->mask);
102 
103 			/*
104 			 * We have to ensure that we have at least one bit
105 			 * still set in the array, since the map could have
106 			 * been concurrently emptied between the first and
107 			 * second reads of vec->mask.  If we hit this
108 			 * condition, simply act as though we never hit this
109 			 * priority level and continue on.
110 			 */
111 			if (cpumask_any(lowest_mask) >= nr_cpu_ids)
112 				continue;
113 		}
114 
115 		return 1;
116 	}
117 
118 	return 0;
119 }
120 
121 /**
122  * cpupri_set - update the CPU priority setting
123  * @cp: The cpupri context
124  * @cpu: The target CPU
125  * @newpri: The priority (INVALID-RT99) to assign to this CPU
126  *
127  * Note: Assumes cpu_rq(cpu)->lock is locked
128  *
129  * Returns: (void)
130  */
131 void cpupri_set(struct cpupri *cp, int cpu, int newpri)
132 {
133 	int *currpri = &cp->cpu_to_pri[cpu];
134 	int oldpri = *currpri;
135 	int do_mb = 0;
136 
137 	newpri = convert_prio(newpri);
138 
139 	BUG_ON(newpri >= CPUPRI_NR_PRIORITIES);
140 
141 	if (newpri == oldpri)
142 		return;
143 
144 	/*
145 	 * If the CPU was currently mapped to a different value, we
146 	 * need to map it to the new value then remove the old value.
147 	 * Note, we must add the new value first, otherwise we risk the
148 	 * cpu being missed by the priority loop in cpupri_find.
149 	 */
150 	if (likely(newpri != CPUPRI_INVALID)) {
151 		struct cpupri_vec *vec = &cp->pri_to_cpu[newpri];
152 
153 		cpumask_set_cpu(cpu, vec->mask);
154 		/*
155 		 * When adding a new vector, we update the mask first,
156 		 * do a write memory barrier, and then update the count, to
157 		 * make sure the vector is visible when count is set.
158 		 */
159 		smp_mb__before_atomic();
160 		atomic_inc(&(vec)->count);
161 		do_mb = 1;
162 	}
163 	if (likely(oldpri != CPUPRI_INVALID)) {
164 		struct cpupri_vec *vec  = &cp->pri_to_cpu[oldpri];
165 
166 		/*
167 		 * Because the order of modification of the vec->count
168 		 * is important, we must make sure that the update
169 		 * of the new prio is seen before we decrement the
170 		 * old prio. This makes sure that the loop sees
171 		 * one or the other when we raise the priority of
172 		 * the run queue. We don't care about when we lower the
173 		 * priority, as that will trigger an rt pull anyway.
174 		 *
175 		 * We only need to do a memory barrier if we updated
176 		 * the new priority vec.
177 		 */
178 		if (do_mb)
179 			smp_mb__after_atomic();
180 
181 		/*
182 		 * When removing from the vector, we decrement the counter first
183 		 * do a memory barrier and then clear the mask.
184 		 */
185 		atomic_dec(&(vec)->count);
186 		smp_mb__after_atomic();
187 		cpumask_clear_cpu(cpu, vec->mask);
188 	}
189 
190 	*currpri = newpri;
191 }
192 
193 /**
194  * cpupri_init - initialize the cpupri structure
195  * @cp: The cpupri context
196  *
197  * Return: -ENOMEM on memory allocation failure.
198  */
199 int cpupri_init(struct cpupri *cp)
200 {
201 	int i;
202 
203 	for (i = 0; i < CPUPRI_NR_PRIORITIES; i++) {
204 		struct cpupri_vec *vec = &cp->pri_to_cpu[i];
205 
206 		atomic_set(&vec->count, 0);
207 		if (!zalloc_cpumask_var(&vec->mask, GFP_KERNEL))
208 			goto cleanup;
209 	}
210 
211 	cp->cpu_to_pri = kcalloc(nr_cpu_ids, sizeof(int), GFP_KERNEL);
212 	if (!cp->cpu_to_pri)
213 		goto cleanup;
214 
215 	for_each_possible_cpu(i)
216 		cp->cpu_to_pri[i] = CPUPRI_INVALID;
217 
218 	return 0;
219 
220 cleanup:
221 	for (i--; i >= 0; i--)
222 		free_cpumask_var(cp->pri_to_cpu[i].mask);
223 	return -ENOMEM;
224 }
225 
226 /**
227  * cpupri_cleanup - clean up the cpupri structure
228  * @cp: The cpupri context
229  */
230 void cpupri_cleanup(struct cpupri *cp)
231 {
232 	int i;
233 
234 	kfree(cp->cpu_to_pri);
235 	for (i = 0; i < CPUPRI_NR_PRIORITIES; i++)
236 		free_cpumask_var(cp->pri_to_cpu[i].mask);
237 }
238