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>
23 #define CREATE_TRACE_POINTS
26 #ifdef CONFIG_DEBUG_SPINLOCK
27 #define DEBUG_SPINLOCK_BUG_ON(p) BUG_ON(p)
29 #define DEBUG_SPINLOCK_BUG_ON(p)
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
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)
71 * This increases the refcount, the caller is expected to
72 * call vgic_put_irq() later once it's finished with the IRQ.
74 kref_get(&irq->refcount);
80 spin_unlock(&dist->lpi_list_lock);
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,
94 if (intid <= VGIC_MAX_PRIVATE)
95 return &vcpu->arch.vgic_cpu.private_irqs[intid];
98 if (intid <= VGIC_MAX_SPI)
99 return &kvm->arch.vgic.spis[intid - VGIC_NR_PRIVATE_IRQS];
102 if (intid >= VGIC_MIN_LPI)
103 return vgic_get_lpi(kvm, intid);
105 WARN(1, "Looking up struct vgic_irq for reserved INTID");
109 static void vgic_get_irq_kref(struct vgic_irq *irq)
111 if (irq->intid < VGIC_MIN_LPI)
114 kref_get(&irq->refcount);
118 * We can't do anything in here, because we lack the kvm pointer to
119 * lock and remove the item from the lpi_list. So we keep this function
120 * empty and use the return value of kref_put() to trigger the freeing.
122 static void vgic_irq_release(struct kref *ref)
126 void vgic_put_irq(struct kvm *kvm, struct vgic_irq *irq)
128 struct vgic_dist *dist;
130 if (irq->intid < VGIC_MIN_LPI)
133 if (!kref_put(&irq->refcount, vgic_irq_release))
136 dist = &kvm->arch.vgic;
138 spin_lock(&dist->lpi_list_lock);
139 list_del(&irq->lpi_list);
140 dist->lpi_list_count--;
141 spin_unlock(&dist->lpi_list_lock);
147 * kvm_vgic_target_oracle - compute the target vcpu for an irq
149 * @irq: The irq to route. Must be already locked.
151 * Based on the current state of the interrupt (enabled, pending,
152 * active, vcpu and target_vcpu), compute the next vcpu this should be
153 * given to. Return NULL if this shouldn't be injected at all.
155 * Requires the IRQ lock to be held.
157 static struct kvm_vcpu *vgic_target_oracle(struct vgic_irq *irq)
159 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&irq->irq_lock));
161 /* If the interrupt is active, it must stay on the current vcpu */
163 return irq->vcpu ? : irq->target_vcpu;
166 * If the IRQ is not active but enabled and pending, we should direct
167 * it to its configured target VCPU.
168 * If the distributor is disabled, pending interrupts shouldn't be
171 if (irq->enabled && irq->pending) {
172 if (unlikely(irq->target_vcpu &&
173 !irq->target_vcpu->kvm->arch.vgic.enabled))
176 return irq->target_vcpu;
179 /* If neither active nor pending and enabled, then this IRQ should not
180 * be queued to any VCPU.
186 * The order of items in the ap_lists defines how we'll pack things in LRs as
187 * well, the first items in the list being the first things populated in the
190 * A hard rule is that active interrupts can never be pushed out of the LRs
191 * (and therefore take priority) since we cannot reliably trap on deactivation
192 * of IRQs and therefore they have to be present in the LRs.
194 * Otherwise things should be sorted by the priority field and the GIC
195 * hardware support will take care of preemption of priority groups etc.
197 * Return negative if "a" sorts before "b", 0 to preserve order, and positive
198 * to sort "b" before "a".
200 static int vgic_irq_cmp(void *priv, struct list_head *a, struct list_head *b)
202 struct vgic_irq *irqa = container_of(a, struct vgic_irq, ap_list);
203 struct vgic_irq *irqb = container_of(b, struct vgic_irq, ap_list);
207 spin_lock(&irqa->irq_lock);
208 spin_lock_nested(&irqb->irq_lock, SINGLE_DEPTH_NESTING);
210 if (irqa->active || irqb->active) {
211 ret = (int)irqb->active - (int)irqa->active;
215 penda = irqa->enabled && irqa->pending;
216 pendb = irqb->enabled && irqb->pending;
218 if (!penda || !pendb) {
219 ret = (int)pendb - (int)penda;
223 /* Both pending and enabled, sort by priority */
224 ret = irqa->priority - irqb->priority;
226 spin_unlock(&irqb->irq_lock);
227 spin_unlock(&irqa->irq_lock);
231 /* Must be called with the ap_list_lock held */
232 static void vgic_sort_ap_list(struct kvm_vcpu *vcpu)
234 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
236 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&vgic_cpu->ap_list_lock));
238 list_sort(NULL, &vgic_cpu->ap_list_head, vgic_irq_cmp);
242 * Only valid injection if changing level for level-triggered IRQs or for a
245 static bool vgic_validate_injection(struct vgic_irq *irq, bool level)
247 switch (irq->config) {
248 case VGIC_CONFIG_LEVEL:
249 return irq->line_level != level;
250 case VGIC_CONFIG_EDGE:
258 * Check whether an IRQ needs to (and can) be queued to a VCPU's ap list.
259 * Do the queuing if necessary, taking the right locks in the right order.
260 * Returns true when the IRQ was queued, false otherwise.
262 * Needs to be entered with the IRQ lock already held, but will return
263 * with all locks dropped.
265 bool vgic_queue_irq_unlock(struct kvm *kvm, struct vgic_irq *irq)
267 struct kvm_vcpu *vcpu;
269 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&irq->irq_lock));
272 vcpu = vgic_target_oracle(irq);
273 if (irq->vcpu || !vcpu) {
275 * If this IRQ is already on a VCPU's ap_list, then it
276 * cannot be moved or modified and there is no more work for
279 * Otherwise, if the irq is not pending and enabled, it does
280 * not need to be inserted into an ap_list and there is also
281 * no more work for us to do.
283 spin_unlock(&irq->irq_lock);
288 * We must unlock the irq lock to take the ap_list_lock where
289 * we are going to insert this new pending interrupt.
291 spin_unlock(&irq->irq_lock);
293 /* someone can do stuff here, which we re-check below */
295 spin_lock(&vcpu->arch.vgic_cpu.ap_list_lock);
296 spin_lock(&irq->irq_lock);
299 * Did something change behind our backs?
301 * There are two cases:
302 * 1) The irq lost its pending state or was disabled behind our
303 * backs and/or it was queued to another VCPU's ap_list.
304 * 2) Someone changed the affinity on this irq behind our
305 * backs and we are now holding the wrong ap_list_lock.
307 * In both cases, drop the locks and retry.
310 if (unlikely(irq->vcpu || vcpu != vgic_target_oracle(irq))) {
311 spin_unlock(&irq->irq_lock);
312 spin_unlock(&vcpu->arch.vgic_cpu.ap_list_lock);
314 spin_lock(&irq->irq_lock);
319 * Grab a reference to the irq to reflect the fact that it is
320 * now in the ap_list.
322 vgic_get_irq_kref(irq);
323 list_add_tail(&irq->ap_list, &vcpu->arch.vgic_cpu.ap_list_head);
326 spin_unlock(&irq->irq_lock);
327 spin_unlock(&vcpu->arch.vgic_cpu.ap_list_lock);
334 static int vgic_update_irq_pending(struct kvm *kvm, int cpuid,
335 unsigned int intid, bool level,
338 struct kvm_vcpu *vcpu;
339 struct vgic_irq *irq;
342 trace_vgic_update_irq_pending(cpuid, intid, level);
344 ret = vgic_lazy_init(kvm);
348 vcpu = kvm_get_vcpu(kvm, cpuid);
349 if (!vcpu && intid < VGIC_NR_PRIVATE_IRQS)
352 irq = vgic_get_irq(kvm, vcpu, intid);
356 if (irq->hw != mapped_irq) {
357 vgic_put_irq(kvm, irq);
361 spin_lock(&irq->irq_lock);
363 if (!vgic_validate_injection(irq, level)) {
364 /* Nothing to see here, move along... */
365 spin_unlock(&irq->irq_lock);
366 vgic_put_irq(kvm, irq);
370 if (irq->config == VGIC_CONFIG_LEVEL) {
371 irq->line_level = level;
372 irq->pending = level || irq->soft_pending;
377 vgic_queue_irq_unlock(kvm, irq);
378 vgic_put_irq(kvm, irq);
384 * kvm_vgic_inject_irq - Inject an IRQ from a device to the vgic
385 * @kvm: The VM structure pointer
386 * @cpuid: The CPU for PPIs
387 * @intid: The INTID to inject a new state to.
388 * @level: Edge-triggered: true: to trigger the interrupt
389 * false: to ignore the call
390 * Level-sensitive true: raise the input signal
391 * false: lower the input signal
393 * The VGIC is not concerned with devices being active-LOW or active-HIGH for
394 * level-sensitive interrupts. You can think of the level parameter as 1
395 * being HIGH and 0 being LOW and all devices being active-HIGH.
397 int kvm_vgic_inject_irq(struct kvm *kvm, int cpuid, unsigned int intid,
400 return vgic_update_irq_pending(kvm, cpuid, intid, level, false);
403 int kvm_vgic_inject_mapped_irq(struct kvm *kvm, int cpuid, unsigned int intid,
406 return vgic_update_irq_pending(kvm, cpuid, intid, level, true);
409 int kvm_vgic_map_phys_irq(struct kvm_vcpu *vcpu, u32 virt_irq, u32 phys_irq)
411 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, virt_irq);
415 spin_lock(&irq->irq_lock);
418 irq->hwintid = phys_irq;
420 spin_unlock(&irq->irq_lock);
421 vgic_put_irq(vcpu->kvm, irq);
426 int kvm_vgic_unmap_phys_irq(struct kvm_vcpu *vcpu, unsigned int virt_irq)
428 struct vgic_irq *irq;
430 if (!vgic_initialized(vcpu->kvm))
433 irq = vgic_get_irq(vcpu->kvm, vcpu, virt_irq);
436 spin_lock(&irq->irq_lock);
441 spin_unlock(&irq->irq_lock);
442 vgic_put_irq(vcpu->kvm, irq);
448 * vgic_prune_ap_list - Remove non-relevant interrupts from the list
450 * @vcpu: The VCPU pointer
452 * Go over the list of "interesting" interrupts, and prune those that we
453 * won't have to consider in the near future.
455 static void vgic_prune_ap_list(struct kvm_vcpu *vcpu)
457 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
458 struct vgic_irq *irq, *tmp;
461 spin_lock(&vgic_cpu->ap_list_lock);
463 list_for_each_entry_safe(irq, tmp, &vgic_cpu->ap_list_head, ap_list) {
464 struct kvm_vcpu *target_vcpu, *vcpuA, *vcpuB;
466 spin_lock(&irq->irq_lock);
468 BUG_ON(vcpu != irq->vcpu);
470 target_vcpu = vgic_target_oracle(irq);
474 * We don't need to process this interrupt any
475 * further, move it off the list.
477 list_del(&irq->ap_list);
479 spin_unlock(&irq->irq_lock);
482 * This vgic_put_irq call matches the
483 * vgic_get_irq_kref in vgic_queue_irq_unlock,
484 * where we added the LPI to the ap_list. As
485 * we remove the irq from the list, we drop
486 * also drop the refcount.
488 vgic_put_irq(vcpu->kvm, irq);
492 if (target_vcpu == vcpu) {
493 /* We're on the right CPU */
494 spin_unlock(&irq->irq_lock);
498 /* This interrupt looks like it has to be migrated. */
500 spin_unlock(&irq->irq_lock);
501 spin_unlock(&vgic_cpu->ap_list_lock);
504 * Ensure locking order by always locking the smallest
507 if (vcpu->vcpu_id < target_vcpu->vcpu_id) {
515 spin_lock(&vcpuA->arch.vgic_cpu.ap_list_lock);
516 spin_lock_nested(&vcpuB->arch.vgic_cpu.ap_list_lock,
517 SINGLE_DEPTH_NESTING);
518 spin_lock(&irq->irq_lock);
521 * If the affinity has been preserved, move the
522 * interrupt around. Otherwise, it means things have
523 * changed while the interrupt was unlocked, and we
524 * need to replay this.
526 * In all cases, we cannot trust the list not to have
527 * changed, so we restart from the beginning.
529 if (target_vcpu == vgic_target_oracle(irq)) {
530 struct vgic_cpu *new_cpu = &target_vcpu->arch.vgic_cpu;
532 list_del(&irq->ap_list);
533 irq->vcpu = target_vcpu;
534 list_add_tail(&irq->ap_list, &new_cpu->ap_list_head);
537 spin_unlock(&irq->irq_lock);
538 spin_unlock(&vcpuB->arch.vgic_cpu.ap_list_lock);
539 spin_unlock(&vcpuA->arch.vgic_cpu.ap_list_lock);
543 spin_unlock(&vgic_cpu->ap_list_lock);
546 static inline void vgic_process_maintenance_interrupt(struct kvm_vcpu *vcpu)
548 if (kvm_vgic_global_state.type == VGIC_V2)
549 vgic_v2_process_maintenance(vcpu);
551 vgic_v3_process_maintenance(vcpu);
554 static inline void vgic_fold_lr_state(struct kvm_vcpu *vcpu)
556 if (kvm_vgic_global_state.type == VGIC_V2)
557 vgic_v2_fold_lr_state(vcpu);
559 vgic_v3_fold_lr_state(vcpu);
562 /* Requires the irq_lock to be held. */
563 static inline void vgic_populate_lr(struct kvm_vcpu *vcpu,
564 struct vgic_irq *irq, int lr)
566 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&irq->irq_lock));
568 if (kvm_vgic_global_state.type == VGIC_V2)
569 vgic_v2_populate_lr(vcpu, irq, lr);
571 vgic_v3_populate_lr(vcpu, irq, lr);
574 static inline void vgic_clear_lr(struct kvm_vcpu *vcpu, int lr)
576 if (kvm_vgic_global_state.type == VGIC_V2)
577 vgic_v2_clear_lr(vcpu, lr);
579 vgic_v3_clear_lr(vcpu, lr);
582 static inline void vgic_set_underflow(struct kvm_vcpu *vcpu)
584 if (kvm_vgic_global_state.type == VGIC_V2)
585 vgic_v2_set_underflow(vcpu);
587 vgic_v3_set_underflow(vcpu);
590 /* Requires the ap_list_lock to be held. */
591 static int compute_ap_list_depth(struct kvm_vcpu *vcpu)
593 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
594 struct vgic_irq *irq;
597 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&vgic_cpu->ap_list_lock));
599 list_for_each_entry(irq, &vgic_cpu->ap_list_head, ap_list) {
600 spin_lock(&irq->irq_lock);
601 /* GICv2 SGIs can count for more than one... */
602 if (vgic_irq_is_sgi(irq->intid) && irq->source)
603 count += hweight8(irq->source);
606 spin_unlock(&irq->irq_lock);
611 /* Requires the VCPU's ap_list_lock to be held. */
612 static void vgic_flush_lr_state(struct kvm_vcpu *vcpu)
614 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
615 struct vgic_irq *irq;
618 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&vgic_cpu->ap_list_lock));
620 if (compute_ap_list_depth(vcpu) > kvm_vgic_global_state.nr_lr) {
621 vgic_set_underflow(vcpu);
622 vgic_sort_ap_list(vcpu);
625 list_for_each_entry(irq, &vgic_cpu->ap_list_head, ap_list) {
626 spin_lock(&irq->irq_lock);
628 if (unlikely(vgic_target_oracle(irq) != vcpu))
632 * If we get an SGI with multiple sources, try to get
633 * them in all at once.
636 vgic_populate_lr(vcpu, irq, count++);
637 } while (irq->source && count < kvm_vgic_global_state.nr_lr);
640 spin_unlock(&irq->irq_lock);
642 if (count == kvm_vgic_global_state.nr_lr)
646 vcpu->arch.vgic_cpu.used_lrs = count;
648 /* Nuke remaining LRs */
649 for ( ; count < kvm_vgic_global_state.nr_lr; count++)
650 vgic_clear_lr(vcpu, count);
653 /* Sync back the hardware VGIC state into our emulation after a guest's run. */
654 void kvm_vgic_sync_hwstate(struct kvm_vcpu *vcpu)
656 vgic_process_maintenance_interrupt(vcpu);
657 vgic_fold_lr_state(vcpu);
658 vgic_prune_ap_list(vcpu);
661 /* Flush our emulation state into the GIC hardware before entering the guest. */
662 void kvm_vgic_flush_hwstate(struct kvm_vcpu *vcpu)
664 spin_lock(&vcpu->arch.vgic_cpu.ap_list_lock);
665 vgic_flush_lr_state(vcpu);
666 spin_unlock(&vcpu->arch.vgic_cpu.ap_list_lock);
669 int kvm_vgic_vcpu_pending_irq(struct kvm_vcpu *vcpu)
671 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
672 struct vgic_irq *irq;
673 bool pending = false;
675 if (!vcpu->kvm->arch.vgic.enabled)
678 spin_lock(&vgic_cpu->ap_list_lock);
680 list_for_each_entry(irq, &vgic_cpu->ap_list_head, ap_list) {
681 spin_lock(&irq->irq_lock);
682 pending = irq->pending && irq->enabled;
683 spin_unlock(&irq->irq_lock);
689 spin_unlock(&vgic_cpu->ap_list_lock);
694 void vgic_kick_vcpus(struct kvm *kvm)
696 struct kvm_vcpu *vcpu;
700 * We've injected an interrupt, time to find out who deserves
703 kvm_for_each_vcpu(c, vcpu, kvm) {
704 if (kvm_vgic_vcpu_pending_irq(vcpu))
709 bool kvm_vgic_map_is_active(struct kvm_vcpu *vcpu, unsigned int virt_irq)
711 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, virt_irq);
714 spin_lock(&irq->irq_lock);
715 map_is_active = irq->hw && irq->active;
716 spin_unlock(&irq->irq_lock);
717 vgic_put_irq(vcpu->kvm, irq);
719 return map_is_active;
722 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi)
724 if (vgic_has_its(kvm))
725 return vgic_its_inject_msi(kvm, msi);