1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* |
3 | * irq_comm.c: Common API for in kernel interrupt controller |
4 | * Copyright (c) 2007, Intel Corporation. |
5 | * |
6 | * Authors: |
7 | * Yaozu (Eddie) Dong <Eddie.dong@intel.com> |
8 | * |
9 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
10 | */ |
11 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
12 | |
13 | #include <linux/kvm_host.h> |
14 | #include <linux/slab.h> |
15 | #include <linux/export.h> |
16 | #include <linux/rculist.h> |
17 | |
18 | #include <trace/events/kvm.h> |
19 | |
20 | #include "irq.h" |
21 | |
22 | #include "ioapic.h" |
23 | |
24 | #include "lapic.h" |
25 | |
26 | #include "hyperv.h" |
27 | #include "x86.h" |
28 | #include "xen.h" |
29 | |
30 | static int kvm_set_pic_irq(struct kvm_kernel_irq_routing_entry *e, |
31 | struct kvm *kvm, int irq_source_id, int level, |
32 | bool line_status) |
33 | { |
34 | struct kvm_pic *pic = kvm->arch.vpic; |
35 | return kvm_pic_set_irq(pic, irq: e->irqchip.pin, irq_source_id, level); |
36 | } |
37 | |
38 | static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, |
39 | struct kvm *kvm, int irq_source_id, int level, |
40 | bool line_status) |
41 | { |
42 | struct kvm_ioapic *ioapic = kvm->arch.vioapic; |
43 | return kvm_ioapic_set_irq(ioapic, irq: e->irqchip.pin, irq_source_id, level, |
44 | line_status); |
45 | } |
46 | |
47 | int kvm_irq_delivery_to_apic(struct kvm *kvm, struct kvm_lapic *src, |
48 | struct kvm_lapic_irq *irq, struct dest_map *dest_map) |
49 | { |
50 | int r = -1; |
51 | struct kvm_vcpu *vcpu, *lowest = NULL; |
52 | unsigned long i, dest_vcpu_bitmap[BITS_TO_LONGS(KVM_MAX_VCPUS)]; |
53 | unsigned int dest_vcpus = 0; |
54 | |
55 | if (kvm_irq_delivery_to_apic_fast(kvm, src, irq, r: &r, dest_map)) |
56 | return r; |
57 | |
58 | if (irq->dest_mode == APIC_DEST_PHYSICAL && |
59 | irq->dest_id == 0xff && kvm_lowest_prio_delivery(irq)) { |
60 | pr_info("apic: phys broadcast and lowest prio\n" ); |
61 | irq->delivery_mode = APIC_DM_FIXED; |
62 | } |
63 | |
64 | memset(dest_vcpu_bitmap, 0, sizeof(dest_vcpu_bitmap)); |
65 | |
66 | kvm_for_each_vcpu(i, vcpu, kvm) { |
67 | if (!kvm_apic_present(vcpu)) |
68 | continue; |
69 | |
70 | if (!kvm_apic_match_dest(vcpu, source: src, shorthand: irq->shorthand, |
71 | dest: irq->dest_id, dest_mode: irq->dest_mode)) |
72 | continue; |
73 | |
74 | if (!kvm_lowest_prio_delivery(irq)) { |
75 | if (r < 0) |
76 | r = 0; |
77 | r += kvm_apic_set_irq(vcpu, irq, dest_map); |
78 | } else if (kvm_apic_sw_enabled(apic: vcpu->arch.apic)) { |
79 | if (!kvm_vector_hashing_enabled()) { |
80 | if (!lowest) |
81 | lowest = vcpu; |
82 | else if (kvm_apic_compare_prio(vcpu1: vcpu, vcpu2: lowest) < 0) |
83 | lowest = vcpu; |
84 | } else { |
85 | __set_bit(i, dest_vcpu_bitmap); |
86 | dest_vcpus++; |
87 | } |
88 | } |
89 | } |
90 | |
91 | if (dest_vcpus != 0) { |
92 | int idx = kvm_vector_to_index(vector: irq->vector, dest_vcpus, |
93 | bitmap: dest_vcpu_bitmap, KVM_MAX_VCPUS); |
94 | |
95 | lowest = kvm_get_vcpu(kvm, i: idx); |
96 | } |
97 | |
98 | if (lowest) |
99 | r = kvm_apic_set_irq(vcpu: lowest, irq, dest_map); |
100 | |
101 | return r; |
102 | } |
103 | |
104 | void kvm_set_msi_irq(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e, |
105 | struct kvm_lapic_irq *irq) |
106 | { |
107 | struct msi_msg msg = { .address_lo = e->msi.address_lo, |
108 | .address_hi = e->msi.address_hi, |
109 | .data = e->msi.data }; |
110 | |
111 | trace_kvm_msi_set_irq(address: msg.address_lo | (kvm->arch.x2apic_format ? |
112 | (u64)msg.address_hi << 32 : 0), data: msg.data); |
113 | |
114 | irq->dest_id = x86_msi_msg_get_destid(msg: &msg, extid: kvm->arch.x2apic_format); |
115 | irq->vector = msg.arch_data.vector; |
116 | irq->dest_mode = kvm_lapic_irq_dest_mode(dest_mode_logical: msg.arch_addr_lo.dest_mode_logical); |
117 | irq->trig_mode = msg.arch_data.is_level; |
118 | irq->delivery_mode = msg.arch_data.delivery_mode << 8; |
119 | irq->msi_redir_hint = msg.arch_addr_lo.redirect_hint; |
120 | irq->level = 1; |
121 | irq->shorthand = APIC_DEST_NOSHORT; |
122 | } |
123 | EXPORT_SYMBOL_GPL(kvm_set_msi_irq); |
124 | |
125 | static inline bool kvm_msi_route_invalid(struct kvm *kvm, |
126 | struct kvm_kernel_irq_routing_entry *e) |
127 | { |
128 | return kvm->arch.x2apic_format && (e->msi.address_hi & 0xff); |
129 | } |
130 | |
131 | int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, |
132 | struct kvm *kvm, int irq_source_id, int level, bool line_status) |
133 | { |
134 | struct kvm_lapic_irq irq; |
135 | |
136 | if (kvm_msi_route_invalid(kvm, e)) |
137 | return -EINVAL; |
138 | |
139 | if (!level) |
140 | return -1; |
141 | |
142 | kvm_set_msi_irq(kvm, e, &irq); |
143 | |
144 | return kvm_irq_delivery_to_apic(kvm, NULL, irq: &irq, NULL); |
145 | } |
146 | |
147 | #ifdef CONFIG_KVM_HYPERV |
148 | static int kvm_hv_set_sint(struct kvm_kernel_irq_routing_entry *e, |
149 | struct kvm *kvm, int irq_source_id, int level, |
150 | bool line_status) |
151 | { |
152 | if (!level) |
153 | return -1; |
154 | |
155 | return kvm_hv_synic_set_irq(kvm, vcpu_id: e->hv_sint.vcpu, sint: e->hv_sint.sint); |
156 | } |
157 | #endif |
158 | |
159 | int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e, |
160 | struct kvm *kvm, int irq_source_id, int level, |
161 | bool line_status) |
162 | { |
163 | struct kvm_lapic_irq irq; |
164 | int r; |
165 | |
166 | switch (e->type) { |
167 | #ifdef CONFIG_KVM_HYPERV |
168 | case KVM_IRQ_ROUTING_HV_SINT: |
169 | return kvm_hv_set_sint(e, kvm, irq_source_id, level, |
170 | line_status); |
171 | #endif |
172 | |
173 | case KVM_IRQ_ROUTING_MSI: |
174 | if (kvm_msi_route_invalid(kvm, e)) |
175 | return -EINVAL; |
176 | |
177 | kvm_set_msi_irq(kvm, e, &irq); |
178 | |
179 | if (kvm_irq_delivery_to_apic_fast(kvm, NULL, irq: &irq, r: &r, NULL)) |
180 | return r; |
181 | break; |
182 | |
183 | #ifdef CONFIG_KVM_XEN |
184 | case KVM_IRQ_ROUTING_XEN_EVTCHN: |
185 | if (!level) |
186 | return -1; |
187 | |
188 | return kvm_xen_set_evtchn_fast(xe: &e->xen_evtchn, kvm); |
189 | #endif |
190 | default: |
191 | break; |
192 | } |
193 | |
194 | return -EWOULDBLOCK; |
195 | } |
196 | |
197 | int kvm_request_irq_source_id(struct kvm *kvm) |
198 | { |
199 | unsigned long *bitmap = &kvm->arch.irq_sources_bitmap; |
200 | int irq_source_id; |
201 | |
202 | mutex_lock(&kvm->irq_lock); |
203 | irq_source_id = find_first_zero_bit(addr: bitmap, BITS_PER_LONG); |
204 | |
205 | if (irq_source_id >= BITS_PER_LONG) { |
206 | pr_warn("exhausted allocatable IRQ sources!\n" ); |
207 | irq_source_id = -EFAULT; |
208 | goto unlock; |
209 | } |
210 | |
211 | ASSERT(irq_source_id != KVM_USERSPACE_IRQ_SOURCE_ID); |
212 | ASSERT(irq_source_id != KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID); |
213 | set_bit(nr: irq_source_id, addr: bitmap); |
214 | unlock: |
215 | mutex_unlock(lock: &kvm->irq_lock); |
216 | |
217 | return irq_source_id; |
218 | } |
219 | |
220 | void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id) |
221 | { |
222 | ASSERT(irq_source_id != KVM_USERSPACE_IRQ_SOURCE_ID); |
223 | ASSERT(irq_source_id != KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID); |
224 | |
225 | mutex_lock(&kvm->irq_lock); |
226 | if (irq_source_id < 0 || |
227 | irq_source_id >= BITS_PER_LONG) { |
228 | pr_err("IRQ source ID out of range!\n" ); |
229 | goto unlock; |
230 | } |
231 | clear_bit(nr: irq_source_id, addr: &kvm->arch.irq_sources_bitmap); |
232 | if (!irqchip_kernel(kvm)) |
233 | goto unlock; |
234 | |
235 | kvm_ioapic_clear_all(ioapic: kvm->arch.vioapic, irq_source_id); |
236 | kvm_pic_clear_all(pic: kvm->arch.vpic, irq_source_id); |
237 | unlock: |
238 | mutex_unlock(lock: &kvm->irq_lock); |
239 | } |
240 | |
241 | void kvm_register_irq_mask_notifier(struct kvm *kvm, int irq, |
242 | struct kvm_irq_mask_notifier *kimn) |
243 | { |
244 | mutex_lock(&kvm->irq_lock); |
245 | kimn->irq = irq; |
246 | hlist_add_head_rcu(n: &kimn->link, h: &kvm->arch.mask_notifier_list); |
247 | mutex_unlock(lock: &kvm->irq_lock); |
248 | } |
249 | |
250 | void kvm_unregister_irq_mask_notifier(struct kvm *kvm, int irq, |
251 | struct kvm_irq_mask_notifier *kimn) |
252 | { |
253 | mutex_lock(&kvm->irq_lock); |
254 | hlist_del_rcu(n: &kimn->link); |
255 | mutex_unlock(lock: &kvm->irq_lock); |
256 | synchronize_srcu(ssp: &kvm->irq_srcu); |
257 | } |
258 | |
259 | void kvm_fire_mask_notifiers(struct kvm *kvm, unsigned irqchip, unsigned pin, |
260 | bool mask) |
261 | { |
262 | struct kvm_irq_mask_notifier *kimn; |
263 | int idx, gsi; |
264 | |
265 | idx = srcu_read_lock(ssp: &kvm->irq_srcu); |
266 | gsi = kvm_irq_map_chip_pin(kvm, irqchip, pin); |
267 | if (gsi != -1) |
268 | hlist_for_each_entry_rcu(kimn, &kvm->arch.mask_notifier_list, link) |
269 | if (kimn->irq == gsi) |
270 | kimn->func(kimn, mask); |
271 | srcu_read_unlock(ssp: &kvm->irq_srcu, idx); |
272 | } |
273 | |
274 | bool kvm_arch_can_set_irq_routing(struct kvm *kvm) |
275 | { |
276 | return irqchip_in_kernel(kvm); |
277 | } |
278 | |
279 | int kvm_set_routing_entry(struct kvm *kvm, |
280 | struct kvm_kernel_irq_routing_entry *e, |
281 | const struct kvm_irq_routing_entry *ue) |
282 | { |
283 | /* We can't check irqchip_in_kernel() here as some callers are |
284 | * currently initializing the irqchip. Other callers should therefore |
285 | * check kvm_arch_can_set_irq_routing() before calling this function. |
286 | */ |
287 | switch (ue->type) { |
288 | case KVM_IRQ_ROUTING_IRQCHIP: |
289 | if (irqchip_split(kvm)) |
290 | return -EINVAL; |
291 | e->irqchip.pin = ue->u.irqchip.pin; |
292 | switch (ue->u.irqchip.irqchip) { |
293 | case KVM_IRQCHIP_PIC_SLAVE: |
294 | e->irqchip.pin += PIC_NUM_PINS / 2; |
295 | fallthrough; |
296 | case KVM_IRQCHIP_PIC_MASTER: |
297 | if (ue->u.irqchip.pin >= PIC_NUM_PINS / 2) |
298 | return -EINVAL; |
299 | e->set = kvm_set_pic_irq; |
300 | break; |
301 | case KVM_IRQCHIP_IOAPIC: |
302 | if (ue->u.irqchip.pin >= KVM_IOAPIC_NUM_PINS) |
303 | return -EINVAL; |
304 | e->set = kvm_set_ioapic_irq; |
305 | break; |
306 | default: |
307 | return -EINVAL; |
308 | } |
309 | e->irqchip.irqchip = ue->u.irqchip.irqchip; |
310 | break; |
311 | case KVM_IRQ_ROUTING_MSI: |
312 | e->set = kvm_set_msi; |
313 | e->msi.address_lo = ue->u.msi.address_lo; |
314 | e->msi.address_hi = ue->u.msi.address_hi; |
315 | e->msi.data = ue->u.msi.data; |
316 | |
317 | if (kvm_msi_route_invalid(kvm, e)) |
318 | return -EINVAL; |
319 | break; |
320 | #ifdef CONFIG_KVM_HYPERV |
321 | case KVM_IRQ_ROUTING_HV_SINT: |
322 | e->set = kvm_hv_set_sint; |
323 | e->hv_sint.vcpu = ue->u.hv_sint.vcpu; |
324 | e->hv_sint.sint = ue->u.hv_sint.sint; |
325 | break; |
326 | #endif |
327 | #ifdef CONFIG_KVM_XEN |
328 | case KVM_IRQ_ROUTING_XEN_EVTCHN: |
329 | return kvm_xen_setup_evtchn(kvm, e, ue); |
330 | #endif |
331 | default: |
332 | return -EINVAL; |
333 | } |
334 | |
335 | return 0; |
336 | } |
337 | |
338 | bool kvm_intr_is_single_vcpu(struct kvm *kvm, struct kvm_lapic_irq *irq, |
339 | struct kvm_vcpu **dest_vcpu) |
340 | { |
341 | int r = 0; |
342 | unsigned long i; |
343 | struct kvm_vcpu *vcpu; |
344 | |
345 | if (kvm_intr_is_single_vcpu_fast(kvm, irq, dest_vcpu)) |
346 | return true; |
347 | |
348 | kvm_for_each_vcpu(i, vcpu, kvm) { |
349 | if (!kvm_apic_present(vcpu)) |
350 | continue; |
351 | |
352 | if (!kvm_apic_match_dest(vcpu, NULL, shorthand: irq->shorthand, |
353 | dest: irq->dest_id, dest_mode: irq->dest_mode)) |
354 | continue; |
355 | |
356 | if (++r == 2) |
357 | return false; |
358 | |
359 | *dest_vcpu = vcpu; |
360 | } |
361 | |
362 | return r == 1; |
363 | } |
364 | EXPORT_SYMBOL_GPL(kvm_intr_is_single_vcpu); |
365 | |
366 | #define IOAPIC_ROUTING_ENTRY(irq) \ |
367 | { .gsi = irq, .type = KVM_IRQ_ROUTING_IRQCHIP, \ |
368 | .u.irqchip = { .irqchip = KVM_IRQCHIP_IOAPIC, .pin = (irq) } } |
369 | #define ROUTING_ENTRY1(irq) IOAPIC_ROUTING_ENTRY(irq) |
370 | |
371 | #define PIC_ROUTING_ENTRY(irq) \ |
372 | { .gsi = irq, .type = KVM_IRQ_ROUTING_IRQCHIP, \ |
373 | .u.irqchip = { .irqchip = SELECT_PIC(irq), .pin = (irq) % 8 } } |
374 | #define ROUTING_ENTRY2(irq) \ |
375 | IOAPIC_ROUTING_ENTRY(irq), PIC_ROUTING_ENTRY(irq) |
376 | |
377 | static const struct kvm_irq_routing_entry default_routing[] = { |
378 | ROUTING_ENTRY2(0), ROUTING_ENTRY2(1), |
379 | ROUTING_ENTRY2(2), ROUTING_ENTRY2(3), |
380 | ROUTING_ENTRY2(4), ROUTING_ENTRY2(5), |
381 | ROUTING_ENTRY2(6), ROUTING_ENTRY2(7), |
382 | ROUTING_ENTRY2(8), ROUTING_ENTRY2(9), |
383 | ROUTING_ENTRY2(10), ROUTING_ENTRY2(11), |
384 | ROUTING_ENTRY2(12), ROUTING_ENTRY2(13), |
385 | ROUTING_ENTRY2(14), ROUTING_ENTRY2(15), |
386 | ROUTING_ENTRY1(16), ROUTING_ENTRY1(17), |
387 | ROUTING_ENTRY1(18), ROUTING_ENTRY1(19), |
388 | ROUTING_ENTRY1(20), ROUTING_ENTRY1(21), |
389 | ROUTING_ENTRY1(22), ROUTING_ENTRY1(23), |
390 | }; |
391 | |
392 | int kvm_setup_default_irq_routing(struct kvm *kvm) |
393 | { |
394 | return kvm_set_irq_routing(kvm, entries: default_routing, |
395 | ARRAY_SIZE(default_routing), flags: 0); |
396 | } |
397 | |
398 | static const struct kvm_irq_routing_entry empty_routing[] = {}; |
399 | |
400 | int kvm_setup_empty_irq_routing(struct kvm *kvm) |
401 | { |
402 | return kvm_set_irq_routing(kvm, entries: empty_routing, nr: 0, flags: 0); |
403 | } |
404 | |
405 | void kvm_arch_post_irq_routing_update(struct kvm *kvm) |
406 | { |
407 | if (!irqchip_split(kvm)) |
408 | return; |
409 | kvm_make_scan_ioapic_request(kvm); |
410 | } |
411 | |
412 | void kvm_scan_ioapic_routes(struct kvm_vcpu *vcpu, |
413 | ulong *ioapic_handled_vectors) |
414 | { |
415 | struct kvm *kvm = vcpu->kvm; |
416 | struct kvm_kernel_irq_routing_entry *entry; |
417 | struct kvm_irq_routing_table *table; |
418 | u32 i, nr_ioapic_pins; |
419 | int idx; |
420 | |
421 | idx = srcu_read_lock(ssp: &kvm->irq_srcu); |
422 | table = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu); |
423 | nr_ioapic_pins = min_t(u32, table->nr_rt_entries, |
424 | kvm->arch.nr_reserved_ioapic_pins); |
425 | for (i = 0; i < nr_ioapic_pins; ++i) { |
426 | hlist_for_each_entry(entry, &table->map[i], link) { |
427 | struct kvm_lapic_irq irq; |
428 | |
429 | if (entry->type != KVM_IRQ_ROUTING_MSI) |
430 | continue; |
431 | |
432 | kvm_set_msi_irq(vcpu->kvm, entry, &irq); |
433 | |
434 | if (irq.trig_mode && |
435 | (kvm_apic_match_dest(vcpu, NULL, APIC_DEST_NOSHORT, |
436 | dest: irq.dest_id, dest_mode: irq.dest_mode) || |
437 | kvm_apic_pending_eoi(vcpu, vector: irq.vector))) |
438 | __set_bit(irq.vector, ioapic_handled_vectors); |
439 | } |
440 | } |
441 | srcu_read_unlock(ssp: &kvm->irq_srcu, idx); |
442 | } |
443 | |
444 | void kvm_arch_irq_routing_update(struct kvm *kvm) |
445 | { |
446 | #ifdef CONFIG_KVM_HYPERV |
447 | kvm_hv_irq_routing_update(kvm); |
448 | #endif |
449 | } |
450 | |