]>
git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - virt/kvm/arm/vgic/vgic.c
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
= {.gicv3_cpuif
= STATIC_KEY_FALSE_INIT
,};
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 vgic_get_irq_kref(irq
);
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");
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
)
125 spin_lock(&dist
->lpi_list_lock
);
126 if (!kref_put(&irq
->refcount
, vgic_irq_release
)) {
127 spin_unlock(&dist
->lpi_list_lock
);
131 list_del(&irq
->lpi_list
);
132 dist
->lpi_list_count
--;
133 spin_unlock(&dist
->lpi_list_lock
);
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 */
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
163 if (irq
->enabled
&& irq
->pending
) {
164 if (unlikely(irq
->target_vcpu
&&
165 !irq
->target_vcpu
->kvm
->arch
.vgic
.enabled
))
168 return irq
->target_vcpu
;
171 /* If neither active nor pending and enabled, then this IRQ should not
172 * be queued to any VCPU.
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
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
);
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
;
207 penda
= irqa
->enabled
&& irqa
->pending
;
208 pendb
= irqb
->enabled
&& irqb
->pending
;
210 if (!penda
|| !pendb
) {
211 ret
= (int)pendb
- (int)penda
;
215 /* Both pending and enabled, sort by priority */
216 ret
= irqa
->priority
- irqb
->priority
;
218 spin_unlock(&irqb
->irq_lock
);
219 spin_unlock(&irqa
->irq_lock
);
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
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
:
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
));
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
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
);
278 * We have to kick the VCPU here, because we could be
279 * queueing an edge-triggered interrupt for which we
280 * get no EOI maintenance interrupt. In that case,
281 * while the IRQ is already on the VCPU's AP list, the
282 * VCPU could have EOI'ed the original interrupt and
283 * won't see this one until it exits for some other
292 * We must unlock the irq lock to take the ap_list_lock where
293 * we are going to insert this new pending interrupt.
295 spin_unlock(&irq
->irq_lock
);
297 /* someone can do stuff here, which we re-check below */
299 spin_lock(&vcpu
->arch
.vgic_cpu
.ap_list_lock
);
300 spin_lock(&irq
->irq_lock
);
303 * Did something change behind our backs?
305 * There are two cases:
306 * 1) The irq lost its pending state or was disabled behind our
307 * backs and/or it was queued to another VCPU's ap_list.
308 * 2) Someone changed the affinity on this irq behind our
309 * backs and we are now holding the wrong ap_list_lock.
311 * In both cases, drop the locks and retry.
314 if (unlikely(irq
->vcpu
|| vcpu
!= vgic_target_oracle(irq
))) {
315 spin_unlock(&irq
->irq_lock
);
316 spin_unlock(&vcpu
->arch
.vgic_cpu
.ap_list_lock
);
318 spin_lock(&irq
->irq_lock
);
323 * Grab a reference to the irq to reflect the fact that it is
324 * now in the ap_list.
326 vgic_get_irq_kref(irq
);
327 list_add_tail(&irq
->ap_list
, &vcpu
->arch
.vgic_cpu
.ap_list_head
);
330 spin_unlock(&irq
->irq_lock
);
331 spin_unlock(&vcpu
->arch
.vgic_cpu
.ap_list_lock
);
338 static int vgic_update_irq_pending(struct kvm
*kvm
, int cpuid
,
339 unsigned int intid
, bool level
,
342 struct kvm_vcpu
*vcpu
;
343 struct vgic_irq
*irq
;
346 trace_vgic_update_irq_pending(cpuid
, intid
, level
);
348 ret
= vgic_lazy_init(kvm
);
352 vcpu
= kvm_get_vcpu(kvm
, cpuid
);
353 if (!vcpu
&& intid
< VGIC_NR_PRIVATE_IRQS
)
356 irq
= vgic_get_irq(kvm
, vcpu
, intid
);
360 if (irq
->hw
!= mapped_irq
) {
361 vgic_put_irq(kvm
, irq
);
365 spin_lock(&irq
->irq_lock
);
367 if (!vgic_validate_injection(irq
, level
)) {
368 /* Nothing to see here, move along... */
369 spin_unlock(&irq
->irq_lock
);
370 vgic_put_irq(kvm
, irq
);
374 if (irq
->config
== VGIC_CONFIG_LEVEL
) {
375 irq
->line_level
= level
;
376 irq
->pending
= level
|| irq
->soft_pending
;
381 vgic_queue_irq_unlock(kvm
, irq
);
382 vgic_put_irq(kvm
, irq
);
388 * kvm_vgic_inject_irq - Inject an IRQ from a device to the vgic
389 * @kvm: The VM structure pointer
390 * @cpuid: The CPU for PPIs
391 * @intid: The INTID to inject a new state to.
392 * @level: Edge-triggered: true: to trigger the interrupt
393 * false: to ignore the call
394 * Level-sensitive true: raise the input signal
395 * false: lower the input signal
397 * The VGIC is not concerned with devices being active-LOW or active-HIGH for
398 * level-sensitive interrupts. You can think of the level parameter as 1
399 * being HIGH and 0 being LOW and all devices being active-HIGH.
401 int kvm_vgic_inject_irq(struct kvm
*kvm
, int cpuid
, unsigned int intid
,
404 return vgic_update_irq_pending(kvm
, cpuid
, intid
, level
, false);
407 int kvm_vgic_inject_mapped_irq(struct kvm
*kvm
, int cpuid
, unsigned int intid
,
410 return vgic_update_irq_pending(kvm
, cpuid
, intid
, level
, true);
413 int kvm_vgic_map_phys_irq(struct kvm_vcpu
*vcpu
, u32 virt_irq
, u32 phys_irq
)
415 struct vgic_irq
*irq
= vgic_get_irq(vcpu
->kvm
, vcpu
, virt_irq
);
419 spin_lock(&irq
->irq_lock
);
422 irq
->hwintid
= phys_irq
;
424 spin_unlock(&irq
->irq_lock
);
425 vgic_put_irq(vcpu
->kvm
, irq
);
430 int kvm_vgic_unmap_phys_irq(struct kvm_vcpu
*vcpu
, unsigned int virt_irq
)
432 struct vgic_irq
*irq
;
434 if (!vgic_initialized(vcpu
->kvm
))
437 irq
= vgic_get_irq(vcpu
->kvm
, vcpu
, virt_irq
);
440 spin_lock(&irq
->irq_lock
);
445 spin_unlock(&irq
->irq_lock
);
446 vgic_put_irq(vcpu
->kvm
, irq
);
452 * vgic_prune_ap_list - Remove non-relevant interrupts from the list
454 * @vcpu: The VCPU pointer
456 * Go over the list of "interesting" interrupts, and prune those that we
457 * won't have to consider in the near future.
459 static void vgic_prune_ap_list(struct kvm_vcpu
*vcpu
)
461 struct vgic_cpu
*vgic_cpu
= &vcpu
->arch
.vgic_cpu
;
462 struct vgic_irq
*irq
, *tmp
;
465 spin_lock(&vgic_cpu
->ap_list_lock
);
467 list_for_each_entry_safe(irq
, tmp
, &vgic_cpu
->ap_list_head
, ap_list
) {
468 struct kvm_vcpu
*target_vcpu
, *vcpuA
, *vcpuB
;
470 spin_lock(&irq
->irq_lock
);
472 BUG_ON(vcpu
!= irq
->vcpu
);
474 target_vcpu
= vgic_target_oracle(irq
);
478 * We don't need to process this interrupt any
479 * further, move it off the list.
481 list_del(&irq
->ap_list
);
483 spin_unlock(&irq
->irq_lock
);
486 * This vgic_put_irq call matches the
487 * vgic_get_irq_kref in vgic_queue_irq_unlock,
488 * where we added the LPI to the ap_list. As
489 * we remove the irq from the list, we drop
490 * also drop the refcount.
492 vgic_put_irq(vcpu
->kvm
, irq
);
496 if (target_vcpu
== vcpu
) {
497 /* We're on the right CPU */
498 spin_unlock(&irq
->irq_lock
);
502 /* This interrupt looks like it has to be migrated. */
504 spin_unlock(&irq
->irq_lock
);
505 spin_unlock(&vgic_cpu
->ap_list_lock
);
508 * Ensure locking order by always locking the smallest
511 if (vcpu
->vcpu_id
< target_vcpu
->vcpu_id
) {
519 spin_lock(&vcpuA
->arch
.vgic_cpu
.ap_list_lock
);
520 spin_lock_nested(&vcpuB
->arch
.vgic_cpu
.ap_list_lock
,
521 SINGLE_DEPTH_NESTING
);
522 spin_lock(&irq
->irq_lock
);
525 * If the affinity has been preserved, move the
526 * interrupt around. Otherwise, it means things have
527 * changed while the interrupt was unlocked, and we
528 * need to replay this.
530 * In all cases, we cannot trust the list not to have
531 * changed, so we restart from the beginning.
533 if (target_vcpu
== vgic_target_oracle(irq
)) {
534 struct vgic_cpu
*new_cpu
= &target_vcpu
->arch
.vgic_cpu
;
536 list_del(&irq
->ap_list
);
537 irq
->vcpu
= target_vcpu
;
538 list_add_tail(&irq
->ap_list
, &new_cpu
->ap_list_head
);
541 spin_unlock(&irq
->irq_lock
);
542 spin_unlock(&vcpuB
->arch
.vgic_cpu
.ap_list_lock
);
543 spin_unlock(&vcpuA
->arch
.vgic_cpu
.ap_list_lock
);
547 spin_unlock(&vgic_cpu
->ap_list_lock
);
550 static inline void vgic_process_maintenance_interrupt(struct kvm_vcpu
*vcpu
)
552 if (kvm_vgic_global_state
.type
== VGIC_V2
)
553 vgic_v2_process_maintenance(vcpu
);
555 vgic_v3_process_maintenance(vcpu
);
558 static inline void vgic_fold_lr_state(struct kvm_vcpu
*vcpu
)
560 if (kvm_vgic_global_state
.type
== VGIC_V2
)
561 vgic_v2_fold_lr_state(vcpu
);
563 vgic_v3_fold_lr_state(vcpu
);
566 /* Requires the irq_lock to be held. */
567 static inline void vgic_populate_lr(struct kvm_vcpu
*vcpu
,
568 struct vgic_irq
*irq
, int lr
)
570 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&irq
->irq_lock
));
572 if (kvm_vgic_global_state
.type
== VGIC_V2
)
573 vgic_v2_populate_lr(vcpu
, irq
, lr
);
575 vgic_v3_populate_lr(vcpu
, irq
, lr
);
578 static inline void vgic_clear_lr(struct kvm_vcpu
*vcpu
, int lr
)
580 if (kvm_vgic_global_state
.type
== VGIC_V2
)
581 vgic_v2_clear_lr(vcpu
, lr
);
583 vgic_v3_clear_lr(vcpu
, lr
);
586 static inline void vgic_set_underflow(struct kvm_vcpu
*vcpu
)
588 if (kvm_vgic_global_state
.type
== VGIC_V2
)
589 vgic_v2_set_underflow(vcpu
);
591 vgic_v3_set_underflow(vcpu
);
594 /* Requires the ap_list_lock to be held. */
595 static int compute_ap_list_depth(struct kvm_vcpu
*vcpu
)
597 struct vgic_cpu
*vgic_cpu
= &vcpu
->arch
.vgic_cpu
;
598 struct vgic_irq
*irq
;
601 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&vgic_cpu
->ap_list_lock
));
603 list_for_each_entry(irq
, &vgic_cpu
->ap_list_head
, ap_list
) {
604 spin_lock(&irq
->irq_lock
);
605 /* GICv2 SGIs can count for more than one... */
606 if (vgic_irq_is_sgi(irq
->intid
) && irq
->source
)
607 count
+= hweight8(irq
->source
);
610 spin_unlock(&irq
->irq_lock
);
615 /* Requires the VCPU's ap_list_lock to be held. */
616 static void vgic_flush_lr_state(struct kvm_vcpu
*vcpu
)
618 struct vgic_cpu
*vgic_cpu
= &vcpu
->arch
.vgic_cpu
;
619 struct vgic_irq
*irq
;
622 DEBUG_SPINLOCK_BUG_ON(!spin_is_locked(&vgic_cpu
->ap_list_lock
));
624 if (compute_ap_list_depth(vcpu
) > kvm_vgic_global_state
.nr_lr
) {
625 vgic_set_underflow(vcpu
);
626 vgic_sort_ap_list(vcpu
);
629 list_for_each_entry(irq
, &vgic_cpu
->ap_list_head
, ap_list
) {
630 spin_lock(&irq
->irq_lock
);
632 if (unlikely(vgic_target_oracle(irq
) != vcpu
))
636 * If we get an SGI with multiple sources, try to get
637 * them in all at once.
640 vgic_populate_lr(vcpu
, irq
, count
++);
641 } while (irq
->source
&& count
< kvm_vgic_global_state
.nr_lr
);
644 spin_unlock(&irq
->irq_lock
);
646 if (count
== kvm_vgic_global_state
.nr_lr
)
650 vcpu
->arch
.vgic_cpu
.used_lrs
= count
;
652 /* Nuke remaining LRs */
653 for ( ; count
< kvm_vgic_global_state
.nr_lr
; count
++)
654 vgic_clear_lr(vcpu
, count
);
657 /* Sync back the hardware VGIC state into our emulation after a guest's run. */
658 void kvm_vgic_sync_hwstate(struct kvm_vcpu
*vcpu
)
660 if (unlikely(!vgic_initialized(vcpu
->kvm
)))
663 vgic_process_maintenance_interrupt(vcpu
);
664 vgic_fold_lr_state(vcpu
);
665 vgic_prune_ap_list(vcpu
);
668 /* Flush our emulation state into the GIC hardware before entering the guest. */
669 void kvm_vgic_flush_hwstate(struct kvm_vcpu
*vcpu
)
671 if (unlikely(!vgic_initialized(vcpu
->kvm
)))
674 spin_lock(&vcpu
->arch
.vgic_cpu
.ap_list_lock
);
675 vgic_flush_lr_state(vcpu
);
676 spin_unlock(&vcpu
->arch
.vgic_cpu
.ap_list_lock
);
679 int kvm_vgic_vcpu_pending_irq(struct kvm_vcpu
*vcpu
)
681 struct vgic_cpu
*vgic_cpu
= &vcpu
->arch
.vgic_cpu
;
682 struct vgic_irq
*irq
;
683 bool pending
= false;
685 if (!vcpu
->kvm
->arch
.vgic
.enabled
)
688 spin_lock(&vgic_cpu
->ap_list_lock
);
690 list_for_each_entry(irq
, &vgic_cpu
->ap_list_head
, ap_list
) {
691 spin_lock(&irq
->irq_lock
);
692 pending
= irq
->pending
&& irq
->enabled
;
693 spin_unlock(&irq
->irq_lock
);
699 spin_unlock(&vgic_cpu
->ap_list_lock
);
704 void vgic_kick_vcpus(struct kvm
*kvm
)
706 struct kvm_vcpu
*vcpu
;
710 * We've injected an interrupt, time to find out who deserves
713 kvm_for_each_vcpu(c
, vcpu
, kvm
) {
714 if (kvm_vgic_vcpu_pending_irq(vcpu
))
719 bool kvm_vgic_map_is_active(struct kvm_vcpu
*vcpu
, unsigned int virt_irq
)
721 struct vgic_irq
*irq
= vgic_get_irq(vcpu
->kvm
, vcpu
, virt_irq
);
724 spin_lock(&irq
->irq_lock
);
725 map_is_active
= irq
->hw
&& irq
->active
;
726 spin_unlock(&irq
->irq_lock
);
727 vgic_put_irq(vcpu
->kvm
, irq
);
729 return map_is_active
;