1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* |
3 | * Generic cpu hotunplug interrupt migration code copied from the |
4 | * arch/arm implementation |
5 | * |
6 | * Copyright (C) Russell King |
7 | * |
8 | * This program is free software; you can redistribute it and/or modify |
9 | * it under the terms of the GNU General Public License version 2 as |
10 | * published by the Free Software Foundation. |
11 | */ |
12 | #include <linux/interrupt.h> |
13 | #include <linux/ratelimit.h> |
14 | #include <linux/irq.h> |
15 | #include <linux/sched/isolation.h> |
16 | |
17 | #include "internals.h" |
18 | |
19 | /* For !GENERIC_IRQ_EFFECTIVE_AFF_MASK this looks at general affinity mask */ |
20 | static inline bool irq_needs_fixup(struct irq_data *d) |
21 | { |
22 | const struct cpumask *m = irq_data_get_effective_affinity_mask(d); |
23 | unsigned int cpu = smp_processor_id(); |
24 | |
25 | #ifdef CONFIG_GENERIC_IRQ_EFFECTIVE_AFF_MASK |
26 | /* |
27 | * The cpumask_empty() check is a workaround for interrupt chips, |
28 | * which do not implement effective affinity, but the architecture has |
29 | * enabled the config switch. Use the general affinity mask instead. |
30 | */ |
31 | if (cpumask_empty(srcp: m)) |
32 | m = irq_data_get_affinity_mask(d); |
33 | |
34 | /* |
35 | * Sanity check. If the mask is not empty when excluding the outgoing |
36 | * CPU then it must contain at least one online CPU. The outgoing CPU |
37 | * has been removed from the online mask already. |
38 | */ |
39 | if (cpumask_any_but(mask: m, cpu) < nr_cpu_ids && |
40 | cpumask_any_and(m, cpu_online_mask) >= nr_cpu_ids) { |
41 | /* |
42 | * If this happens then there was a missed IRQ fixup at some |
43 | * point. Warn about it and enforce fixup. |
44 | */ |
45 | pr_warn("Eff. affinity %*pbl of IRQ %u contains only offline CPUs after offlining CPU %u\n" , |
46 | cpumask_pr_args(m), d->irq, cpu); |
47 | return true; |
48 | } |
49 | #endif |
50 | return cpumask_test_cpu(cpu, cpumask: m); |
51 | } |
52 | |
53 | static bool migrate_one_irq(struct irq_desc *desc) |
54 | { |
55 | struct irq_data *d = irq_desc_get_irq_data(desc); |
56 | struct irq_chip *chip = irq_data_get_irq_chip(d); |
57 | bool maskchip = !irq_can_move_pcntxt(data: d) && !irqd_irq_masked(d); |
58 | const struct cpumask *affinity; |
59 | bool brokeaff = false; |
60 | int err; |
61 | |
62 | /* |
63 | * IRQ chip might be already torn down, but the irq descriptor is |
64 | * still in the radix tree. Also if the chip has no affinity setter, |
65 | * nothing can be done here. |
66 | */ |
67 | if (!chip || !chip->irq_set_affinity) { |
68 | pr_debug("IRQ %u: Unable to migrate away\n" , d->irq); |
69 | return false; |
70 | } |
71 | |
72 | /* |
73 | * No move required, if: |
74 | * - Interrupt is per cpu |
75 | * - Interrupt is not started |
76 | * - Affinity mask does not include this CPU. |
77 | * |
78 | * Note: Do not check desc->action as this might be a chained |
79 | * interrupt. |
80 | */ |
81 | if (irqd_is_per_cpu(d) || !irqd_is_started(d) || !irq_needs_fixup(d)) { |
82 | /* |
83 | * If an irq move is pending, abort it if the dying CPU is |
84 | * the sole target. |
85 | */ |
86 | irq_fixup_move_pending(desc, force_clear: false); |
87 | return false; |
88 | } |
89 | |
90 | /* |
91 | * Complete an eventually pending irq move cleanup. If this |
92 | * interrupt was moved in hard irq context, then the vectors need |
93 | * to be cleaned up. It can't wait until this interrupt actually |
94 | * happens and this CPU was involved. |
95 | */ |
96 | irq_force_complete_move(desc); |
97 | |
98 | /* |
99 | * If there is a setaffinity pending, then try to reuse the pending |
100 | * mask, so the last change of the affinity does not get lost. If |
101 | * there is no move pending or the pending mask does not contain |
102 | * any online CPU, use the current affinity mask. |
103 | */ |
104 | if (irq_fixup_move_pending(desc, force_clear: true)) |
105 | affinity = irq_desc_get_pending_mask(desc); |
106 | else |
107 | affinity = irq_data_get_affinity_mask(d); |
108 | |
109 | /* Mask the chip for interrupts which cannot move in process context */ |
110 | if (maskchip && chip->irq_mask) |
111 | chip->irq_mask(d); |
112 | |
113 | if (cpumask_any_and(affinity, cpu_online_mask) >= nr_cpu_ids) { |
114 | /* |
115 | * If the interrupt is managed, then shut it down and leave |
116 | * the affinity untouched. |
117 | */ |
118 | if (irqd_affinity_is_managed(d)) { |
119 | irqd_set_managed_shutdown(d); |
120 | irq_shutdown_and_deactivate(desc); |
121 | return false; |
122 | } |
123 | affinity = cpu_online_mask; |
124 | brokeaff = true; |
125 | } |
126 | /* |
127 | * Do not set the force argument of irq_do_set_affinity() as this |
128 | * disables the masking of offline CPUs from the supplied affinity |
129 | * mask and therefore might keep/reassign the irq to the outgoing |
130 | * CPU. |
131 | */ |
132 | err = irq_do_set_affinity(data: d, dest: affinity, force: false); |
133 | if (err) { |
134 | pr_warn_ratelimited("IRQ%u: set affinity failed(%d).\n" , |
135 | d->irq, err); |
136 | brokeaff = false; |
137 | } |
138 | |
139 | if (maskchip && chip->irq_unmask) |
140 | chip->irq_unmask(d); |
141 | |
142 | return brokeaff; |
143 | } |
144 | |
145 | /** |
146 | * irq_migrate_all_off_this_cpu - Migrate irqs away from offline cpu |
147 | * |
148 | * The current CPU has been marked offline. Migrate IRQs off this CPU. |
149 | * If the affinity settings do not allow other CPUs, force them onto any |
150 | * available CPU. |
151 | * |
152 | * Note: we must iterate over all IRQs, whether they have an attached |
153 | * action structure or not, as we need to get chained interrupts too. |
154 | */ |
155 | void irq_migrate_all_off_this_cpu(void) |
156 | { |
157 | struct irq_desc *desc; |
158 | unsigned int irq; |
159 | |
160 | for_each_active_irq(irq) { |
161 | bool affinity_broken; |
162 | |
163 | desc = irq_to_desc(irq); |
164 | raw_spin_lock(&desc->lock); |
165 | affinity_broken = migrate_one_irq(desc); |
166 | raw_spin_unlock(&desc->lock); |
167 | |
168 | if (affinity_broken) { |
169 | pr_debug_ratelimited("IRQ %u: no longer affine to CPU%u\n" , |
170 | irq, smp_processor_id()); |
171 | } |
172 | } |
173 | } |
174 | |
175 | static bool hk_should_isolate(struct irq_data *data, unsigned int cpu) |
176 | { |
177 | const struct cpumask *hk_mask; |
178 | |
179 | if (!housekeeping_enabled(type: HK_TYPE_MANAGED_IRQ)) |
180 | return false; |
181 | |
182 | hk_mask = housekeeping_cpumask(type: HK_TYPE_MANAGED_IRQ); |
183 | if (cpumask_subset(src1p: irq_data_get_effective_affinity_mask(d: data), src2p: hk_mask)) |
184 | return false; |
185 | |
186 | return cpumask_test_cpu(cpu, cpumask: hk_mask); |
187 | } |
188 | |
189 | static void irq_restore_affinity_of_irq(struct irq_desc *desc, unsigned int cpu) |
190 | { |
191 | struct irq_data *data = irq_desc_get_irq_data(desc); |
192 | const struct cpumask *affinity = irq_data_get_affinity_mask(d: data); |
193 | |
194 | if (!irqd_affinity_is_managed(d: data) || !desc->action || |
195 | !irq_data_get_irq_chip(d: data) || !cpumask_test_cpu(cpu, cpumask: affinity)) |
196 | return; |
197 | |
198 | if (irqd_is_managed_and_shutdown(d: data)) { |
199 | irq_startup(desc, IRQ_RESEND, IRQ_START_COND); |
200 | return; |
201 | } |
202 | |
203 | /* |
204 | * If the interrupt can only be directed to a single target |
205 | * CPU then it is already assigned to a CPU in the affinity |
206 | * mask. No point in trying to move it around unless the |
207 | * isolation mechanism requests to move it to an upcoming |
208 | * housekeeping CPU. |
209 | */ |
210 | if (!irqd_is_single_target(d: data) || hk_should_isolate(data, cpu)) |
211 | irq_set_affinity_locked(data, cpumask: affinity, force: false); |
212 | } |
213 | |
214 | /** |
215 | * irq_affinity_online_cpu - Restore affinity for managed interrupts |
216 | * @cpu: Upcoming CPU for which interrupts should be restored |
217 | */ |
218 | int irq_affinity_online_cpu(unsigned int cpu) |
219 | { |
220 | struct irq_desc *desc; |
221 | unsigned int irq; |
222 | |
223 | irq_lock_sparse(); |
224 | for_each_active_irq(irq) { |
225 | desc = irq_to_desc(irq); |
226 | raw_spin_lock_irq(&desc->lock); |
227 | irq_restore_affinity_of_irq(desc, cpu); |
228 | raw_spin_unlock_irq(&desc->lock); |
229 | } |
230 | irq_unlock_sparse(); |
231 | |
232 | return 0; |
233 | } |
234 | |