Linux 4.8-rc8
[linux/fpc-iii.git] / virt / kvm / arm / vgic / vgic.c
blobe83b7fe4baaed1e1c5bc5f7fd520995002186b61
1 /*
2 * Copyright (C) 2015, 2016 ARM Ltd.
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program. If not, see <http://www.gnu.org/licenses/>.
17 #include <linux/kvm.h>
18 #include <linux/kvm_host.h>
19 #include <linux/list_sort.h>
21 #include "vgic.h"
23 #define CREATE_TRACE_POINTS
24 #include "../trace.h"
26 #ifdef CONFIG_DEBUG_SPINLOCK
27 #define DEBUG_SPINLOCK_BUG_ON(p) BUG_ON(p)
28 #else
29 #define DEBUG_SPINLOCK_BUG_ON(p)
30 #endif
32 struct vgic_global __section(.hyp.text) kvm_vgic_global_state;
35 * Locking order is always:
36 * its->cmd_lock (mutex)
37 * its->its_lock (mutex)
38 * vgic_cpu->ap_list_lock
39 * kvm->lpi_list_lock
40 * vgic_irq->irq_lock
42 * If you need to take multiple locks, always take the upper lock first,
43 * then the lower ones, e.g. first take the its_lock, then the irq_lock.
44 * If you are already holding a lock and need to take a higher one, you
45 * have to drop the lower ranking lock first and re-aquire it after having
46 * taken the upper one.
48 * When taking more than one ap_list_lock at the same time, always take the
49 * lowest numbered VCPU's ap_list_lock first, so:
50 * vcpuX->vcpu_id < vcpuY->vcpu_id:
51 * spin_lock(vcpuX->arch.vgic_cpu.ap_list_lock);
52 * spin_lock(vcpuY->arch.vgic_cpu.ap_list_lock);
56 * Iterate over the VM's list of mapped LPIs to find the one with a
57 * matching interrupt ID and return a reference to the IRQ structure.
59 static struct vgic_irq *vgic_get_lpi(struct kvm *kvm, u32 intid)
61 struct vgic_dist *dist = &kvm->arch.vgic;
62 struct vgic_irq *irq = NULL;
64 spin_lock(&dist->lpi_list_lock);
66 list_for_each_entry(irq, &dist->lpi_list_head, lpi_list) {
67 if (irq->intid != intid)
68 continue;
71 * This increases the refcount, the caller is expected to
72 * call vgic_put_irq() later once it's finished with the IRQ.
74 vgic_get_irq_kref(irq);
75 goto out_unlock;
77 irq = NULL;
79 out_unlock:
80 spin_unlock(&dist->lpi_list_lock);
82 return irq;
86 * This looks up the virtual interrupt ID to get the corresponding
87 * struct vgic_irq. It also increases the refcount, so any caller is expected
88 * to call vgic_put_irq() once it's finished with this IRQ.
90 struct vgic_irq *vgic_get_irq(struct kvm *kvm, struct kvm_vcpu *vcpu,
91 u32 intid)
93 /* SGIs and PPIs */
94 if (intid <= VGIC_MAX_PRIVATE)
95 return &vcpu->arch.vgic_cpu.private_irqs[intid];
97 /* SPIs */
98 if (intid <= VGIC_MAX_SPI)
99 return &kvm->arch.vgic.spis[intid - VGIC_NR_PRIVATE_IRQS];
101 /* LPIs */
102 if (intid >= VGIC_MIN_LPI)
103 return vgic_get_lpi(kvm, intid);
105 WARN(1, "Looking up struct vgic_irq for reserved INTID");
106 return NULL;
110 * We can't do anything in here, because we lack the kvm pointer to
111 * lock and remove the item from the lpi_list. So we keep this function
112 * empty and use the return value of kref_put() to trigger the freeing.
114 static void vgic_irq_release(struct kref *ref)
118 void vgic_put_irq(struct kvm *kvm, struct vgic_irq *irq)
120 struct vgic_dist *dist = &kvm->arch.vgic;
122 if (irq->intid < VGIC_MIN_LPI)
123 return;
125 spin_lock(&dist->lpi_list_lock);
126 if (!kref_put(&irq->refcount, vgic_irq_release)) {
127 spin_unlock(&dist->lpi_list_lock);
128 return;
131 list_del(&irq->lpi_list);
132 dist->lpi_list_count--;
133 spin_unlock(&dist->lpi_list_lock);
135 kfree(irq);
139 * kvm_vgic_target_oracle - compute the target vcpu for an irq
141 * @irq: The irq to route. Must be already locked.
143 * Based on the current state of the interrupt (enabled, pending,
144 * active, vcpu and target_vcpu), compute the next vcpu this should be
145 * given to. Return NULL if this shouldn't be injected at all.
147 * Requires the IRQ lock to be held.
149 static struct kvm_vcpu *vgic_target_oracle(struct vgic_irq *irq)
151 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&irq->irq_lock));
153 /* If the interrupt is active, it must stay on the current vcpu */
154 if (irq->active)
155 return irq->vcpu ? : irq->target_vcpu;
158 * If the IRQ is not active but enabled and pending, we should direct
159 * it to its configured target VCPU.
160 * If the distributor is disabled, pending interrupts shouldn't be
161 * forwarded.
163 if (irq->enabled && irq->pending) {
164 if (unlikely(irq->target_vcpu &&
165 !irq->target_vcpu->kvm->arch.vgic.enabled))
166 return NULL;
168 return irq->target_vcpu;
171 /* If neither active nor pending and enabled, then this IRQ should not
172 * be queued to any VCPU.
174 return NULL;
178 * The order of items in the ap_lists defines how we'll pack things in LRs as
179 * well, the first items in the list being the first things populated in the
180 * LRs.
182 * A hard rule is that active interrupts can never be pushed out of the LRs
183 * (and therefore take priority) since we cannot reliably trap on deactivation
184 * of IRQs and therefore they have to be present in the LRs.
186 * Otherwise things should be sorted by the priority field and the GIC
187 * hardware support will take care of preemption of priority groups etc.
189 * Return negative if "a" sorts before "b", 0 to preserve order, and positive
190 * to sort "b" before "a".
192 static int vgic_irq_cmp(void *priv, struct list_head *a, struct list_head *b)
194 struct vgic_irq *irqa = container_of(a, struct vgic_irq, ap_list);
195 struct vgic_irq *irqb = container_of(b, struct vgic_irq, ap_list);
196 bool penda, pendb;
197 int ret;
199 spin_lock(&irqa->irq_lock);
200 spin_lock_nested(&irqb->irq_lock, SINGLE_DEPTH_NESTING);
202 if (irqa->active || irqb->active) {
203 ret = (int)irqb->active - (int)irqa->active;
204 goto out;
207 penda = irqa->enabled && irqa->pending;
208 pendb = irqb->enabled && irqb->pending;
210 if (!penda || !pendb) {
211 ret = (int)pendb - (int)penda;
212 goto out;
215 /* Both pending and enabled, sort by priority */
216 ret = irqa->priority - irqb->priority;
217 out:
218 spin_unlock(&irqb->irq_lock);
219 spin_unlock(&irqa->irq_lock);
220 return ret;
223 /* Must be called with the ap_list_lock held */
224 static void vgic_sort_ap_list(struct kvm_vcpu *vcpu)
226 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
228 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&vgic_cpu->ap_list_lock));
230 list_sort(NULL, &vgic_cpu->ap_list_head, vgic_irq_cmp);
234 * Only valid injection if changing level for level-triggered IRQs or for a
235 * rising edge.
237 static bool vgic_validate_injection(struct vgic_irq *irq, bool level)
239 switch (irq->config) {
240 case VGIC_CONFIG_LEVEL:
241 return irq->line_level != level;
242 case VGIC_CONFIG_EDGE:
243 return level;
246 return false;
250 * Check whether an IRQ needs to (and can) be queued to a VCPU's ap list.
251 * Do the queuing if necessary, taking the right locks in the right order.
252 * Returns true when the IRQ was queued, false otherwise.
254 * Needs to be entered with the IRQ lock already held, but will return
255 * with all locks dropped.
257 bool vgic_queue_irq_unlock(struct kvm *kvm, struct vgic_irq *irq)
259 struct kvm_vcpu *vcpu;
261 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&irq->irq_lock));
263 retry:
264 vcpu = vgic_target_oracle(irq);
265 if (irq->vcpu || !vcpu) {
267 * If this IRQ is already on a VCPU's ap_list, then it
268 * cannot be moved or modified and there is no more work for
269 * us to do.
271 * Otherwise, if the irq is not pending and enabled, it does
272 * not need to be inserted into an ap_list and there is also
273 * no more work for us to do.
275 spin_unlock(&irq->irq_lock);
276 return false;
280 * We must unlock the irq lock to take the ap_list_lock where
281 * we are going to insert this new pending interrupt.
283 spin_unlock(&irq->irq_lock);
285 /* someone can do stuff here, which we re-check below */
287 spin_lock(&vcpu->arch.vgic_cpu.ap_list_lock);
288 spin_lock(&irq->irq_lock);
291 * Did something change behind our backs?
293 * There are two cases:
294 * 1) The irq lost its pending state or was disabled behind our
295 * backs and/or it was queued to another VCPU's ap_list.
296 * 2) Someone changed the affinity on this irq behind our
297 * backs and we are now holding the wrong ap_list_lock.
299 * In both cases, drop the locks and retry.
302 if (unlikely(irq->vcpu || vcpu != vgic_target_oracle(irq))) {
303 spin_unlock(&irq->irq_lock);
304 spin_unlock(&vcpu->arch.vgic_cpu.ap_list_lock);
306 spin_lock(&irq->irq_lock);
307 goto retry;
311 * Grab a reference to the irq to reflect the fact that it is
312 * now in the ap_list.
314 vgic_get_irq_kref(irq);
315 list_add_tail(&irq->ap_list, &vcpu->arch.vgic_cpu.ap_list_head);
316 irq->vcpu = vcpu;
318 spin_unlock(&irq->irq_lock);
319 spin_unlock(&vcpu->arch.vgic_cpu.ap_list_lock);
321 kvm_vcpu_kick(vcpu);
323 return true;
326 static int vgic_update_irq_pending(struct kvm *kvm, int cpuid,
327 unsigned int intid, bool level,
328 bool mapped_irq)
330 struct kvm_vcpu *vcpu;
331 struct vgic_irq *irq;
332 int ret;
334 trace_vgic_update_irq_pending(cpuid, intid, level);
336 ret = vgic_lazy_init(kvm);
337 if (ret)
338 return ret;
340 vcpu = kvm_get_vcpu(kvm, cpuid);
341 if (!vcpu && intid < VGIC_NR_PRIVATE_IRQS)
342 return -EINVAL;
344 irq = vgic_get_irq(kvm, vcpu, intid);
345 if (!irq)
346 return -EINVAL;
348 if (irq->hw != mapped_irq) {
349 vgic_put_irq(kvm, irq);
350 return -EINVAL;
353 spin_lock(&irq->irq_lock);
355 if (!vgic_validate_injection(irq, level)) {
356 /* Nothing to see here, move along... */
357 spin_unlock(&irq->irq_lock);
358 vgic_put_irq(kvm, irq);
359 return 0;
362 if (irq->config == VGIC_CONFIG_LEVEL) {
363 irq->line_level = level;
364 irq->pending = level || irq->soft_pending;
365 } else {
366 irq->pending = true;
369 vgic_queue_irq_unlock(kvm, irq);
370 vgic_put_irq(kvm, irq);
372 return 0;
376 * kvm_vgic_inject_irq - Inject an IRQ from a device to the vgic
377 * @kvm: The VM structure pointer
378 * @cpuid: The CPU for PPIs
379 * @intid: The INTID to inject a new state to.
380 * @level: Edge-triggered: true: to trigger the interrupt
381 * false: to ignore the call
382 * Level-sensitive true: raise the input signal
383 * false: lower the input signal
385 * The VGIC is not concerned with devices being active-LOW or active-HIGH for
386 * level-sensitive interrupts. You can think of the level parameter as 1
387 * being HIGH and 0 being LOW and all devices being active-HIGH.
389 int kvm_vgic_inject_irq(struct kvm *kvm, int cpuid, unsigned int intid,
390 bool level)
392 return vgic_update_irq_pending(kvm, cpuid, intid, level, false);
395 int kvm_vgic_inject_mapped_irq(struct kvm *kvm, int cpuid, unsigned int intid,
396 bool level)
398 return vgic_update_irq_pending(kvm, cpuid, intid, level, true);
401 int kvm_vgic_map_phys_irq(struct kvm_vcpu *vcpu, u32 virt_irq, u32 phys_irq)
403 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, virt_irq);
405 BUG_ON(!irq);
407 spin_lock(&irq->irq_lock);
409 irq->hw = true;
410 irq->hwintid = phys_irq;
412 spin_unlock(&irq->irq_lock);
413 vgic_put_irq(vcpu->kvm, irq);
415 return 0;
418 int kvm_vgic_unmap_phys_irq(struct kvm_vcpu *vcpu, unsigned int virt_irq)
420 struct vgic_irq *irq;
422 if (!vgic_initialized(vcpu->kvm))
423 return -EAGAIN;
425 irq = vgic_get_irq(vcpu->kvm, vcpu, virt_irq);
426 BUG_ON(!irq);
428 spin_lock(&irq->irq_lock);
430 irq->hw = false;
431 irq->hwintid = 0;
433 spin_unlock(&irq->irq_lock);
434 vgic_put_irq(vcpu->kvm, irq);
436 return 0;
440 * vgic_prune_ap_list - Remove non-relevant interrupts from the list
442 * @vcpu: The VCPU pointer
444 * Go over the list of "interesting" interrupts, and prune those that we
445 * won't have to consider in the near future.
447 static void vgic_prune_ap_list(struct kvm_vcpu *vcpu)
449 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
450 struct vgic_irq *irq, *tmp;
452 retry:
453 spin_lock(&vgic_cpu->ap_list_lock);
455 list_for_each_entry_safe(irq, tmp, &vgic_cpu->ap_list_head, ap_list) {
456 struct kvm_vcpu *target_vcpu, *vcpuA, *vcpuB;
458 spin_lock(&irq->irq_lock);
460 BUG_ON(vcpu != irq->vcpu);
462 target_vcpu = vgic_target_oracle(irq);
464 if (!target_vcpu) {
466 * We don't need to process this interrupt any
467 * further, move it off the list.
469 list_del(&irq->ap_list);
470 irq->vcpu = NULL;
471 spin_unlock(&irq->irq_lock);
474 * This vgic_put_irq call matches the
475 * vgic_get_irq_kref in vgic_queue_irq_unlock,
476 * where we added the LPI to the ap_list. As
477 * we remove the irq from the list, we drop
478 * also drop the refcount.
480 vgic_put_irq(vcpu->kvm, irq);
481 continue;
484 if (target_vcpu == vcpu) {
485 /* We're on the right CPU */
486 spin_unlock(&irq->irq_lock);
487 continue;
490 /* This interrupt looks like it has to be migrated. */
492 spin_unlock(&irq->irq_lock);
493 spin_unlock(&vgic_cpu->ap_list_lock);
496 * Ensure locking order by always locking the smallest
497 * ID first.
499 if (vcpu->vcpu_id < target_vcpu->vcpu_id) {
500 vcpuA = vcpu;
501 vcpuB = target_vcpu;
502 } else {
503 vcpuA = target_vcpu;
504 vcpuB = vcpu;
507 spin_lock(&vcpuA->arch.vgic_cpu.ap_list_lock);
508 spin_lock_nested(&vcpuB->arch.vgic_cpu.ap_list_lock,
509 SINGLE_DEPTH_NESTING);
510 spin_lock(&irq->irq_lock);
513 * If the affinity has been preserved, move the
514 * interrupt around. Otherwise, it means things have
515 * changed while the interrupt was unlocked, and we
516 * need to replay this.
518 * In all cases, we cannot trust the list not to have
519 * changed, so we restart from the beginning.
521 if (target_vcpu == vgic_target_oracle(irq)) {
522 struct vgic_cpu *new_cpu = &target_vcpu->arch.vgic_cpu;
524 list_del(&irq->ap_list);
525 irq->vcpu = target_vcpu;
526 list_add_tail(&irq->ap_list, &new_cpu->ap_list_head);
529 spin_unlock(&irq->irq_lock);
530 spin_unlock(&vcpuB->arch.vgic_cpu.ap_list_lock);
531 spin_unlock(&vcpuA->arch.vgic_cpu.ap_list_lock);
532 goto retry;
535 spin_unlock(&vgic_cpu->ap_list_lock);
538 static inline void vgic_process_maintenance_interrupt(struct kvm_vcpu *vcpu)
540 if (kvm_vgic_global_state.type == VGIC_V2)
541 vgic_v2_process_maintenance(vcpu);
542 else
543 vgic_v3_process_maintenance(vcpu);
546 static inline void vgic_fold_lr_state(struct kvm_vcpu *vcpu)
548 if (kvm_vgic_global_state.type == VGIC_V2)
549 vgic_v2_fold_lr_state(vcpu);
550 else
551 vgic_v3_fold_lr_state(vcpu);
554 /* Requires the irq_lock to be held. */
555 static inline void vgic_populate_lr(struct kvm_vcpu *vcpu,
556 struct vgic_irq *irq, int lr)
558 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&irq->irq_lock));
560 if (kvm_vgic_global_state.type == VGIC_V2)
561 vgic_v2_populate_lr(vcpu, irq, lr);
562 else
563 vgic_v3_populate_lr(vcpu, irq, lr);
566 static inline void vgic_clear_lr(struct kvm_vcpu *vcpu, int lr)
568 if (kvm_vgic_global_state.type == VGIC_V2)
569 vgic_v2_clear_lr(vcpu, lr);
570 else
571 vgic_v3_clear_lr(vcpu, lr);
574 static inline void vgic_set_underflow(struct kvm_vcpu *vcpu)
576 if (kvm_vgic_global_state.type == VGIC_V2)
577 vgic_v2_set_underflow(vcpu);
578 else
579 vgic_v3_set_underflow(vcpu);
582 /* Requires the ap_list_lock to be held. */
583 static int compute_ap_list_depth(struct kvm_vcpu *vcpu)
585 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
586 struct vgic_irq *irq;
587 int count = 0;
589 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&vgic_cpu->ap_list_lock));
591 list_for_each_entry(irq, &vgic_cpu->ap_list_head, ap_list) {
592 spin_lock(&irq->irq_lock);
593 /* GICv2 SGIs can count for more than one... */
594 if (vgic_irq_is_sgi(irq->intid) && irq->source)
595 count += hweight8(irq->source);
596 else
597 count++;
598 spin_unlock(&irq->irq_lock);
600 return count;
603 /* Requires the VCPU's ap_list_lock to be held. */
604 static void vgic_flush_lr_state(struct kvm_vcpu *vcpu)
606 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
607 struct vgic_irq *irq;
608 int count = 0;
610 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&vgic_cpu->ap_list_lock));
612 if (compute_ap_list_depth(vcpu) > kvm_vgic_global_state.nr_lr) {
613 vgic_set_underflow(vcpu);
614 vgic_sort_ap_list(vcpu);
617 list_for_each_entry(irq, &vgic_cpu->ap_list_head, ap_list) {
618 spin_lock(&irq->irq_lock);
620 if (unlikely(vgic_target_oracle(irq) != vcpu))
621 goto next;
624 * If we get an SGI with multiple sources, try to get
625 * them in all at once.
627 do {
628 vgic_populate_lr(vcpu, irq, count++);
629 } while (irq->source && count < kvm_vgic_global_state.nr_lr);
631 next:
632 spin_unlock(&irq->irq_lock);
634 if (count == kvm_vgic_global_state.nr_lr)
635 break;
638 vcpu->arch.vgic_cpu.used_lrs = count;
640 /* Nuke remaining LRs */
641 for ( ; count < kvm_vgic_global_state.nr_lr; count++)
642 vgic_clear_lr(vcpu, count);
645 /* Sync back the hardware VGIC state into our emulation after a guest's run. */
646 void kvm_vgic_sync_hwstate(struct kvm_vcpu *vcpu)
648 vgic_process_maintenance_interrupt(vcpu);
649 vgic_fold_lr_state(vcpu);
650 vgic_prune_ap_list(vcpu);
653 /* Flush our emulation state into the GIC hardware before entering the guest. */
654 void kvm_vgic_flush_hwstate(struct kvm_vcpu *vcpu)
656 spin_lock(&vcpu->arch.vgic_cpu.ap_list_lock);
657 vgic_flush_lr_state(vcpu);
658 spin_unlock(&vcpu->arch.vgic_cpu.ap_list_lock);
661 int kvm_vgic_vcpu_pending_irq(struct kvm_vcpu *vcpu)
663 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
664 struct vgic_irq *irq;
665 bool pending = false;
667 if (!vcpu->kvm->arch.vgic.enabled)
668 return false;
670 spin_lock(&vgic_cpu->ap_list_lock);
672 list_for_each_entry(irq, &vgic_cpu->ap_list_head, ap_list) {
673 spin_lock(&irq->irq_lock);
674 pending = irq->pending && irq->enabled;
675 spin_unlock(&irq->irq_lock);
677 if (pending)
678 break;
681 spin_unlock(&vgic_cpu->ap_list_lock);
683 return pending;
686 void vgic_kick_vcpus(struct kvm *kvm)
688 struct kvm_vcpu *vcpu;
689 int c;
692 * We've injected an interrupt, time to find out who deserves
693 * a good kick...
695 kvm_for_each_vcpu(c, vcpu, kvm) {
696 if (kvm_vgic_vcpu_pending_irq(vcpu))
697 kvm_vcpu_kick(vcpu);
701 bool kvm_vgic_map_is_active(struct kvm_vcpu *vcpu, unsigned int virt_irq)
703 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, virt_irq);
704 bool map_is_active;
706 spin_lock(&irq->irq_lock);
707 map_is_active = irq->hw && irq->active;
708 spin_unlock(&irq->irq_lock);
709 vgic_put_irq(vcpu->kvm, irq);
711 return map_is_active;