2 * Copyright (C) 2012 ARM Ltd.
3 * Author: Marc Zyngier <marc.zyngier@arm.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 #include <linux/cpu.h>
20 #include <linux/of_irq.h>
21 #include <linux/kvm.h>
22 #include <linux/kvm_host.h>
23 #include <linux/interrupt.h>
25 #include <clocksource/arm_arch_timer.h>
26 #include <asm/arch_timer.h>
28 #include <kvm/arm_vgic.h>
29 #include <kvm/arm_arch_timer.h>
33 static struct timecounter *timecounter;
34 static struct workqueue_struct *wqueue;
35 static unsigned int host_vtimer_irq;
37 static cycle_t kvm_phys_timer_read(void)
39 return timecounter->cc->read(timecounter->cc);
42 static bool timer_is_armed(struct arch_timer_cpu *timer)
47 /* timer_arm: as in "arm the timer", not as in ARM the company */
48 static void timer_arm(struct arch_timer_cpu *timer, u64 ns)
51 hrtimer_start(&timer->timer, ktime_add_ns(ktime_get(), ns),
55 static void timer_disarm(struct arch_timer_cpu *timer)
57 if (timer_is_armed(timer)) {
58 hrtimer_cancel(&timer->timer);
59 cancel_work_sync(&timer->expired);
64 static irqreturn_t kvm_arch_timer_handler(int irq, void *dev_id)
66 struct kvm_vcpu *vcpu = *(struct kvm_vcpu **)dev_id;
69 * We disable the timer in the world switch and let it be
70 * handled by kvm_timer_sync_hwstate(). Getting a timer
71 * interrupt at this point is a sure sign of some major
74 pr_warn("Unexpected interrupt %d on vcpu %p\n", irq, vcpu);
79 * Work function for handling the backup timer that we schedule when a vcpu is
80 * no longer running, but had a timer programmed to fire in the future.
82 static void kvm_timer_inject_irq_work(struct work_struct *work)
84 struct kvm_vcpu *vcpu;
86 vcpu = container_of(work, struct kvm_vcpu, arch.timer_cpu.expired);
87 vcpu->arch.timer_cpu.armed = false;
89 WARN_ON(!kvm_timer_should_fire(vcpu));
92 * If the vcpu is blocked we want to wake it up so that it will see
93 * the timer has expired when entering the guest.
98 static u64 kvm_timer_compute_delta(struct kvm_vcpu *vcpu)
102 cval = vcpu->arch.timer_cpu.cntv_cval;
103 now = kvm_phys_timer_read() - vcpu->kvm->arch.timer.cntvoff;
108 ns = cyclecounter_cyc2ns(timecounter->cc,
118 static enum hrtimer_restart kvm_timer_expire(struct hrtimer *hrt)
120 struct arch_timer_cpu *timer;
121 struct kvm_vcpu *vcpu;
124 timer = container_of(hrt, struct arch_timer_cpu, timer);
125 vcpu = container_of(timer, struct kvm_vcpu, arch.timer_cpu);
128 * Check that the timer has really expired from the guest's
129 * PoV (NTP on the host may have forced it to expire
130 * early). If we should have slept longer, restart it.
132 ns = kvm_timer_compute_delta(vcpu);
134 hrtimer_forward_now(hrt, ns_to_ktime(ns));
135 return HRTIMER_RESTART;
138 queue_work(wqueue, &timer->expired);
139 return HRTIMER_NORESTART;
142 static bool kvm_timer_irq_can_fire(struct kvm_vcpu *vcpu)
144 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
146 return !(timer->cntv_ctl & ARCH_TIMER_CTRL_IT_MASK) &&
147 (timer->cntv_ctl & ARCH_TIMER_CTRL_ENABLE);
150 bool kvm_timer_should_fire(struct kvm_vcpu *vcpu)
152 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
155 if (!kvm_timer_irq_can_fire(vcpu))
158 cval = timer->cntv_cval;
159 now = kvm_phys_timer_read() - vcpu->kvm->arch.timer.cntvoff;
164 static void kvm_timer_update_irq(struct kvm_vcpu *vcpu, bool new_level)
167 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
169 BUG_ON(!vgic_initialized(vcpu->kvm));
171 timer->irq.level = new_level;
172 trace_kvm_timer_update_irq(vcpu->vcpu_id, timer->map->virt_irq,
174 ret = kvm_vgic_inject_mapped_irq(vcpu->kvm, vcpu->vcpu_id,
181 * Check if there was a change in the timer state (should we raise or lower
182 * the line level to the GIC).
184 static int kvm_timer_update_state(struct kvm_vcpu *vcpu)
186 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
189 * If userspace modified the timer registers via SET_ONE_REG before
190 * the vgic was initialized, we mustn't set the timer->irq.level value
191 * because the guest would never see the interrupt. Instead wait
192 * until we call this function from kvm_timer_flush_hwstate.
194 if (!vgic_initialized(vcpu->kvm))
197 if (kvm_timer_should_fire(vcpu) != timer->irq.level)
198 kvm_timer_update_irq(vcpu, !timer->irq.level);
204 * Schedule the background timer before calling kvm_vcpu_block, so that this
205 * thread is removed from its waitqueue and made runnable when there's a timer
206 * interrupt to handle.
208 void kvm_timer_schedule(struct kvm_vcpu *vcpu)
210 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
212 BUG_ON(timer_is_armed(timer));
215 * No need to schedule a background timer if the guest timer has
216 * already expired, because kvm_vcpu_block will return before putting
217 * the thread to sleep.
219 if (kvm_timer_should_fire(vcpu))
223 * If the timer is not capable of raising interrupts (disabled or
224 * masked), then there's no more work for us to do.
226 if (!kvm_timer_irq_can_fire(vcpu))
229 /* The timer has not yet expired, schedule a background timer */
230 timer_arm(timer, kvm_timer_compute_delta(vcpu));
233 void kvm_timer_unschedule(struct kvm_vcpu *vcpu)
235 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
240 * kvm_timer_flush_hwstate - prepare to move the virt timer to the cpu
241 * @vcpu: The vcpu pointer
243 * Check if the virtual timer has expired while we were running in the host,
244 * and inject an interrupt if that was the case.
246 void kvm_timer_flush_hwstate(struct kvm_vcpu *vcpu)
248 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
252 if (kvm_timer_update_state(vcpu))
256 * If we enter the guest with the virtual input level to the VGIC
257 * asserted, then we have already told the VGIC what we need to, and
258 * we don't need to exit from the guest until the guest deactivates
259 * the already injected interrupt, so therefore we should set the
260 * hardware active state to prevent unnecessary exits from the guest.
262 * Also, if we enter the guest with the virtual timer interrupt active,
263 * then it must be active on the physical distributor, because we set
264 * the HW bit and the guest must be able to deactivate the virtual and
265 * physical interrupt at the same time.
267 * Conversely, if the virtual input level is deasserted and the virtual
268 * interrupt is not active, then always clear the hardware active state
269 * to ensure that hardware interrupts from the timer triggers a guest
272 if (timer->irq.level || kvm_vgic_map_is_active(vcpu, timer->map))
277 ret = irq_set_irqchip_state(timer->map->irq,
278 IRQCHIP_STATE_ACTIVE,
284 * kvm_timer_sync_hwstate - sync timer state from cpu
285 * @vcpu: The vcpu pointer
287 * Check if the virtual timer has expired while we were running in the guest,
288 * and inject an interrupt if that was the case.
290 void kvm_timer_sync_hwstate(struct kvm_vcpu *vcpu)
292 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
294 BUG_ON(timer_is_armed(timer));
297 * The guest could have modified the timer registers or the timer
298 * could have expired, update the timer state.
300 kvm_timer_update_state(vcpu);
303 int kvm_timer_vcpu_reset(struct kvm_vcpu *vcpu,
304 const struct kvm_irq_level *irq)
306 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
307 struct irq_phys_map *map;
310 * The vcpu timer irq number cannot be determined in
311 * kvm_timer_vcpu_init() because it is called much before
312 * kvm_vcpu_set_target(). To handle this, we determine
313 * vcpu timer irq number when the vcpu is reset.
315 timer->irq.irq = irq->irq;
318 * The bits in CNTV_CTL are architecturally reset to UNKNOWN for ARMv8
319 * and to 0 for ARMv7. We provide an implementation that always
320 * resets the timer to be disabled and unmasked and is compliant with
321 * the ARMv7 architecture.
324 kvm_timer_update_state(vcpu);
327 * Tell the VGIC that the virtual interrupt is tied to a
328 * physical interrupt. We do that once per VCPU.
330 map = kvm_vgic_map_phys_irq(vcpu, irq->irq, host_vtimer_irq);
331 if (WARN_ON(IS_ERR(map)))
338 void kvm_timer_vcpu_init(struct kvm_vcpu *vcpu)
340 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
342 INIT_WORK(&timer->expired, kvm_timer_inject_irq_work);
343 hrtimer_init(&timer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
344 timer->timer.function = kvm_timer_expire;
347 static void kvm_timer_init_interrupt(void *info)
349 enable_percpu_irq(host_vtimer_irq, 0);
352 int kvm_arm_timer_set_reg(struct kvm_vcpu *vcpu, u64 regid, u64 value)
354 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
357 case KVM_REG_ARM_TIMER_CTL:
358 timer->cntv_ctl = value;
360 case KVM_REG_ARM_TIMER_CNT:
361 vcpu->kvm->arch.timer.cntvoff = kvm_phys_timer_read() - value;
363 case KVM_REG_ARM_TIMER_CVAL:
364 timer->cntv_cval = value;
370 kvm_timer_update_state(vcpu);
374 u64 kvm_arm_timer_get_reg(struct kvm_vcpu *vcpu, u64 regid)
376 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
379 case KVM_REG_ARM_TIMER_CTL:
380 return timer->cntv_ctl;
381 case KVM_REG_ARM_TIMER_CNT:
382 return kvm_phys_timer_read() - vcpu->kvm->arch.timer.cntvoff;
383 case KVM_REG_ARM_TIMER_CVAL:
384 return timer->cntv_cval;
389 static int kvm_timer_cpu_notify(struct notifier_block *self,
390 unsigned long action, void *cpu)
394 case CPU_STARTING_FROZEN:
395 kvm_timer_init_interrupt(NULL);
398 case CPU_DYING_FROZEN:
399 disable_percpu_irq(host_vtimer_irq);
406 static struct notifier_block kvm_timer_cpu_nb = {
407 .notifier_call = kvm_timer_cpu_notify,
410 static const struct of_device_id arch_timer_of_match[] = {
411 { .compatible = "arm,armv7-timer", },
412 { .compatible = "arm,armv8-timer", },
416 int kvm_timer_hyp_init(void)
418 struct device_node *np;
422 timecounter = arch_timer_get_timecounter();
426 np = of_find_matching_node(NULL, arch_timer_of_match);
428 kvm_err("kvm_arch_timer: can't find DT node\n");
432 ppi = irq_of_parse_and_map(np, 2);
434 kvm_err("kvm_arch_timer: no virtual timer interrupt\n");
439 err = request_percpu_irq(ppi, kvm_arch_timer_handler,
440 "kvm guest timer", kvm_get_running_vcpus());
442 kvm_err("kvm_arch_timer: can't request interrupt %d (%d)\n",
447 host_vtimer_irq = ppi;
449 err = __register_cpu_notifier(&kvm_timer_cpu_nb);
451 kvm_err("Cannot register timer CPU notifier\n");
455 wqueue = create_singlethread_workqueue("kvm_arch_timer");
461 kvm_info("%s IRQ%d\n", np->name, ppi);
462 on_each_cpu(kvm_timer_init_interrupt, NULL, 1);
466 free_percpu_irq(ppi, kvm_get_running_vcpus());
472 void kvm_timer_vcpu_terminate(struct kvm_vcpu *vcpu)
474 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
478 kvm_vgic_unmap_phys_irq(vcpu, timer->map);
481 void kvm_timer_enable(struct kvm *kvm)
483 if (kvm->arch.timer.enabled)
487 * There is a potential race here between VCPUs starting for the first
488 * time, which may be enabling the timer multiple times. That doesn't
489 * hurt though, because we're just setting a variable to the same
490 * variable that it already was. The important thing is that all
491 * VCPUs have the enabled variable set, before entering the guest, if
492 * the arch timers are enabled.
494 if (timecounter && wqueue)
495 kvm->arch.timer.enabled = 1;
498 void kvm_timer_init(struct kvm *kvm)
500 kvm->arch.timer.cntvoff = kvm_phys_timer_read();