1/*
2 * Copyright (C) 2012 ARM Ltd.
3 * Author: Marc Zyngier <marc.zyngier@arm.com>
4 *
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.
8 *
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.
13 *
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
17 */
18
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>
24
25#include <clocksource/arm_arch_timer.h>
26#include <asm/arch_timer.h>
27
28#include <kvm/arm_vgic.h>
29#include <kvm/arm_arch_timer.h>
30
31#include "trace.h"
32
33static struct timecounter *timecounter;
34static struct workqueue_struct *wqueue;
35static unsigned int host_vtimer_irq;
36
37static cycle_t kvm_phys_timer_read(void)
38{
39	return timecounter->cc->read(timecounter->cc);
40}
41
42static bool timer_is_armed(struct arch_timer_cpu *timer)
43{
44	return timer->armed;
45}
46
47/* timer_arm: as in "arm the timer", not as in ARM the company */
48static void timer_arm(struct arch_timer_cpu *timer, u64 ns)
49{
50	timer->armed = true;
51	hrtimer_start(&timer->timer, ktime_add_ns(ktime_get(), ns),
52		      HRTIMER_MODE_ABS);
53}
54
55static void timer_disarm(struct arch_timer_cpu *timer)
56{
57	if (timer_is_armed(timer)) {
58		hrtimer_cancel(&timer->timer);
59		cancel_work_sync(&timer->expired);
60		timer->armed = false;
61	}
62}
63
64static irqreturn_t kvm_arch_timer_handler(int irq, void *dev_id)
65{
66	struct kvm_vcpu *vcpu = *(struct kvm_vcpu **)dev_id;
67
68	/*
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
72	 * breakage.
73	 */
74	pr_warn("Unexpected interrupt %d on vcpu %p\n", irq, vcpu);
75	return IRQ_HANDLED;
76}
77
78/*
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.
81 */
82static void kvm_timer_inject_irq_work(struct work_struct *work)
83{
84	struct kvm_vcpu *vcpu;
85
86	vcpu = container_of(work, struct kvm_vcpu, arch.timer_cpu.expired);
87	vcpu->arch.timer_cpu.armed = false;
88
89	WARN_ON(!kvm_timer_should_fire(vcpu));
90
91	/*
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.
94	 */
95	kvm_vcpu_kick(vcpu);
96}
97
98static u64 kvm_timer_compute_delta(struct kvm_vcpu *vcpu)
99{
100	cycle_t cval, now;
101
102	cval = vcpu->arch.timer_cpu.cntv_cval;
103	now = kvm_phys_timer_read() - vcpu->kvm->arch.timer.cntvoff;
104
105	if (now < cval) {
106		u64 ns;
107
108		ns = cyclecounter_cyc2ns(timecounter->cc,
109					 cval - now,
110					 timecounter->mask,
111					 &timecounter->frac);
112		return ns;
113	}
114
115	return 0;
116}
117
118static enum hrtimer_restart kvm_timer_expire(struct hrtimer *hrt)
119{
120	struct arch_timer_cpu *timer;
121	struct kvm_vcpu *vcpu;
122	u64 ns;
123
124	timer = container_of(hrt, struct arch_timer_cpu, timer);
125	vcpu = container_of(timer, struct kvm_vcpu, arch.timer_cpu);
126
127	/*
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.
131	 */
132	ns = kvm_timer_compute_delta(vcpu);
133	if (unlikely(ns)) {
134		hrtimer_forward_now(hrt, ns_to_ktime(ns));
135		return HRTIMER_RESTART;
136	}
137
138	queue_work(wqueue, &timer->expired);
139	return HRTIMER_NORESTART;
140}
141
142static bool kvm_timer_irq_can_fire(struct kvm_vcpu *vcpu)
143{
144	struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
145
146	return !(timer->cntv_ctl & ARCH_TIMER_CTRL_IT_MASK) &&
147		(timer->cntv_ctl & ARCH_TIMER_CTRL_ENABLE);
148}
149
150bool kvm_timer_should_fire(struct kvm_vcpu *vcpu)
151{
152	struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
153	cycle_t cval, now;
154
155	if (!kvm_timer_irq_can_fire(vcpu))
156		return false;
157
158	cval = timer->cntv_cval;
159	now = kvm_phys_timer_read() - vcpu->kvm->arch.timer.cntvoff;
160
161	return cval <= now;
162}
163
164static void kvm_timer_update_irq(struct kvm_vcpu *vcpu, bool new_level)
165{
166	int ret;
167	struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
168
169	BUG_ON(!vgic_initialized(vcpu->kvm));
170
171	timer->irq.level = new_level;
172	trace_kvm_timer_update_irq(vcpu->vcpu_id, timer->map->virt_irq,
173				   timer->irq.level);
174	ret = kvm_vgic_inject_mapped_irq(vcpu->kvm, vcpu->vcpu_id,
175					 timer->map,
176					 timer->irq.level);
177	WARN_ON(ret);
178}
179
180/*
181 * Check if there was a change in the timer state (should we raise or lower
182 * the line level to the GIC).
183 */
184static int kvm_timer_update_state(struct kvm_vcpu *vcpu)
185{
186	struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
187
188	/*
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.
193	 */
194	if (!vgic_initialized(vcpu->kvm))
195		return -ENODEV;
196
197	if (kvm_timer_should_fire(vcpu) != timer->irq.level)
198		kvm_timer_update_irq(vcpu, !timer->irq.level);
199
200	return 0;
201}
202
203/*
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.
207 */
208void kvm_timer_schedule(struct kvm_vcpu *vcpu)
209{
210	struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
211
212	BUG_ON(timer_is_armed(timer));
213
214	/*
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.
218	 */
219	if (kvm_timer_should_fire(vcpu))
220		return;
221
222	/*
223	 * If the timer is not capable of raising interrupts (disabled or
224	 * masked), then there's no more work for us to do.
225	 */
226	if (!kvm_timer_irq_can_fire(vcpu))
227		return;
228
229	/*  The timer has not yet expired, schedule a background timer */
230	timer_arm(timer, kvm_timer_compute_delta(vcpu));
231}
232
233void kvm_timer_unschedule(struct kvm_vcpu *vcpu)
234{
235	struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
236	timer_disarm(timer);
237}
238
239/**
240 * kvm_timer_flush_hwstate - prepare to move the virt timer to the cpu
241 * @vcpu: The vcpu pointer
242 *
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.
245 */
246void kvm_timer_flush_hwstate(struct kvm_vcpu *vcpu)
247{
248	struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
249	bool phys_active;
250	int ret;
251
252	if (kvm_timer_update_state(vcpu))
253		return;
254
255	/*
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.
261	*
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.
266	*
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
270	* exit.
271	*/
272	if (timer->irq.level || kvm_vgic_map_is_active(vcpu, timer->map))
273		phys_active = true;
274	else
275		phys_active = false;
276
277	ret = irq_set_irqchip_state(timer->map->irq,
278				    IRQCHIP_STATE_ACTIVE,
279				    phys_active);
280	WARN_ON(ret);
281}
282
283/**
284 * kvm_timer_sync_hwstate - sync timer state from cpu
285 * @vcpu: The vcpu pointer
286 *
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.
289 */
290void kvm_timer_sync_hwstate(struct kvm_vcpu *vcpu)
291{
292	struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
293
294	BUG_ON(timer_is_armed(timer));
295
296	/*
297	 * The guest could have modified the timer registers or the timer
298	 * could have expired, update the timer state.
299	 */
300	kvm_timer_update_state(vcpu);
301}
302
303int kvm_timer_vcpu_reset(struct kvm_vcpu *vcpu,
304			 const struct kvm_irq_level *irq)
305{
306	struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
307	struct irq_phys_map *map;
308
309	/*
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.
314	 */
315	timer->irq.irq = irq->irq;
316
317	/*
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.
322	 */
323	timer->cntv_ctl = 0;
324	kvm_timer_update_state(vcpu);
325
326	/*
327	 * Tell the VGIC that the virtual interrupt is tied to a
328	 * physical interrupt. We do that once per VCPU.
329	 */
330	map = kvm_vgic_map_phys_irq(vcpu, irq->irq, host_vtimer_irq);
331	if (WARN_ON(IS_ERR(map)))
332		return PTR_ERR(map);
333
334	timer->map = map;
335	return 0;
336}
337
338void kvm_timer_vcpu_init(struct kvm_vcpu *vcpu)
339{
340	struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
341
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;
345}
346
347static void kvm_timer_init_interrupt(void *info)
348{
349	enable_percpu_irq(host_vtimer_irq, 0);
350}
351
352int kvm_arm_timer_set_reg(struct kvm_vcpu *vcpu, u64 regid, u64 value)
353{
354	struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
355
356	switch (regid) {
357	case KVM_REG_ARM_TIMER_CTL:
358		timer->cntv_ctl = value;
359		break;
360	case KVM_REG_ARM_TIMER_CNT:
361		vcpu->kvm->arch.timer.cntvoff = kvm_phys_timer_read() - value;
362		break;
363	case KVM_REG_ARM_TIMER_CVAL:
364		timer->cntv_cval = value;
365		break;
366	default:
367		return -1;
368	}
369
370	kvm_timer_update_state(vcpu);
371	return 0;
372}
373
374u64 kvm_arm_timer_get_reg(struct kvm_vcpu *vcpu, u64 regid)
375{
376	struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
377
378	switch (regid) {
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;
385	}
386	return (u64)-1;
387}
388
389static int kvm_timer_cpu_notify(struct notifier_block *self,
390				unsigned long action, void *cpu)
391{
392	switch (action) {
393	case CPU_STARTING:
394	case CPU_STARTING_FROZEN:
395		kvm_timer_init_interrupt(NULL);
396		break;
397	case CPU_DYING:
398	case CPU_DYING_FROZEN:
399		disable_percpu_irq(host_vtimer_irq);
400		break;
401	}
402
403	return NOTIFY_OK;
404}
405
406static struct notifier_block kvm_timer_cpu_nb = {
407	.notifier_call = kvm_timer_cpu_notify,
408};
409
410static const struct of_device_id arch_timer_of_match[] = {
411	{ .compatible	= "arm,armv7-timer",	},
412	{ .compatible	= "arm,armv8-timer",	},
413	{},
414};
415
416int kvm_timer_hyp_init(void)
417{
418	struct device_node *np;
419	unsigned int ppi;
420	int err;
421
422	timecounter = arch_timer_get_timecounter();
423	if (!timecounter)
424		return -ENODEV;
425
426	np = of_find_matching_node(NULL, arch_timer_of_match);
427	if (!np) {
428		kvm_err("kvm_arch_timer: can't find DT node\n");
429		return -ENODEV;
430	}
431
432	ppi = irq_of_parse_and_map(np, 2);
433	if (!ppi) {
434		kvm_err("kvm_arch_timer: no virtual timer interrupt\n");
435		err = -EINVAL;
436		goto out;
437	}
438
439	err = request_percpu_irq(ppi, kvm_arch_timer_handler,
440				 "kvm guest timer", kvm_get_running_vcpus());
441	if (err) {
442		kvm_err("kvm_arch_timer: can't request interrupt %d (%d)\n",
443			ppi, err);
444		goto out;
445	}
446
447	host_vtimer_irq = ppi;
448
449	err = __register_cpu_notifier(&kvm_timer_cpu_nb);
450	if (err) {
451		kvm_err("Cannot register timer CPU notifier\n");
452		goto out_free;
453	}
454
455	wqueue = create_singlethread_workqueue("kvm_arch_timer");
456	if (!wqueue) {
457		err = -ENOMEM;
458		goto out_free;
459	}
460
461	kvm_info("%s IRQ%d\n", np->name, ppi);
462	on_each_cpu(kvm_timer_init_interrupt, NULL, 1);
463
464	goto out;
465out_free:
466	free_percpu_irq(ppi, kvm_get_running_vcpus());
467out:
468	of_node_put(np);
469	return err;
470}
471
472void kvm_timer_vcpu_terminate(struct kvm_vcpu *vcpu)
473{
474	struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
475
476	timer_disarm(timer);
477	if (timer->map)
478		kvm_vgic_unmap_phys_irq(vcpu, timer->map);
479}
480
481void kvm_timer_enable(struct kvm *kvm)
482{
483	if (kvm->arch.timer.enabled)
484		return;
485
486	/*
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.
493	 */
494	if (timecounter && wqueue)
495		kvm->arch.timer.enabled = 1;
496}
497
498void kvm_timer_init(struct kvm *kvm)
499{
500	kvm->arch.timer.cntvoff = kvm_phys_timer_read();
501}
502