1
2
3
4
5
6
7
8
9
10
11#include <linux/kernel.h>
12#include <linux/interrupt.h>
13#include <linux/clocksource.h>
14#include <linux/clockchips.h>
15#include <linux/gfp.h>
16#include <linux/slab.h>
17#include <linux/pvclock_gtod.h>
18#include <linux/timekeeper_internal.h>
19
20#include <asm/pvclock.h>
21#include <asm/xen/hypervisor.h>
22#include <asm/xen/hypercall.h>
23
24#include <xen/events.h>
25#include <xen/features.h>
26#include <xen/interface/xen.h>
27#include <xen/interface/vcpu.h>
28
29#include "xen-ops.h"
30
31
32#define TIMER_SLOP 100000
33
34static u64 xen_sched_clock_offset __read_mostly;
35
36
37static unsigned long xen_tsc_khz(void)
38{
39 struct pvclock_vcpu_time_info *info =
40 &HYPERVISOR_shared_info->vcpu_info[0].time;
41
42 return pvclock_tsc_khz(info);
43}
44
45static u64 xen_clocksource_read(void)
46{
47 struct pvclock_vcpu_time_info *src;
48 u64 ret;
49
50 preempt_disable_notrace();
51 src = &__this_cpu_read(xen_vcpu)->time;
52 ret = pvclock_clocksource_read(src);
53 preempt_enable_notrace();
54 return ret;
55}
56
57static u64 xen_clocksource_get_cycles(struct clocksource *cs)
58{
59 return xen_clocksource_read();
60}
61
62static u64 xen_sched_clock(void)
63{
64 return xen_clocksource_read() - xen_sched_clock_offset;
65}
66
67static void xen_read_wallclock(struct timespec64 *ts)
68{
69 struct shared_info *s = HYPERVISOR_shared_info;
70 struct pvclock_wall_clock *wall_clock = &(s->wc);
71 struct pvclock_vcpu_time_info *vcpu_time;
72
73 vcpu_time = &get_cpu_var(xen_vcpu)->time;
74 pvclock_read_wallclock(wall_clock, vcpu_time, ts);
75 put_cpu_var(xen_vcpu);
76}
77
78static void xen_get_wallclock(struct timespec64 *now)
79{
80 xen_read_wallclock(now);
81}
82
83static int xen_set_wallclock(const struct timespec64 *now)
84{
85 return -ENODEV;
86}
87
88static int xen_pvclock_gtod_notify(struct notifier_block *nb,
89 unsigned long was_set, void *priv)
90{
91
92 static struct timespec64 next_sync;
93
94 struct xen_platform_op op;
95 struct timespec64 now;
96 struct timekeeper *tk = priv;
97 static bool settime64_supported = true;
98 int ret;
99
100 now.tv_sec = tk->xtime_sec;
101 now.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
102
103
104
105
106
107 if (!was_set && timespec64_compare(&now, &next_sync) < 0)
108 return NOTIFY_OK;
109
110again:
111 if (settime64_supported) {
112 op.cmd = XENPF_settime64;
113 op.u.settime64.mbz = 0;
114 op.u.settime64.secs = now.tv_sec;
115 op.u.settime64.nsecs = now.tv_nsec;
116 op.u.settime64.system_time = xen_clocksource_read();
117 } else {
118 op.cmd = XENPF_settime32;
119 op.u.settime32.secs = now.tv_sec;
120 op.u.settime32.nsecs = now.tv_nsec;
121 op.u.settime32.system_time = xen_clocksource_read();
122 }
123
124 ret = HYPERVISOR_platform_op(&op);
125
126 if (ret == -ENOSYS && settime64_supported) {
127 settime64_supported = false;
128 goto again;
129 }
130 if (ret < 0)
131 return NOTIFY_BAD;
132
133
134
135
136
137
138 next_sync = now;
139 next_sync.tv_sec += 11 * 60;
140
141 return NOTIFY_OK;
142}
143
144static struct notifier_block xen_pvclock_gtod_notifier = {
145 .notifier_call = xen_pvclock_gtod_notify,
146};
147
148static struct clocksource xen_clocksource __read_mostly = {
149 .name = "xen",
150 .rating = 400,
151 .read = xen_clocksource_get_cycles,
152 .mask = ~0,
153 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
154};
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187static s64 get_abs_timeout(unsigned long delta)
188{
189 return xen_clocksource_read() + delta;
190}
191
192static int xen_timerop_shutdown(struct clock_event_device *evt)
193{
194
195 HYPERVISOR_set_timer_op(0);
196
197 return 0;
198}
199
200static int xen_timerop_set_next_event(unsigned long delta,
201 struct clock_event_device *evt)
202{
203 WARN_ON(!clockevent_state_oneshot(evt));
204
205 if (HYPERVISOR_set_timer_op(get_abs_timeout(delta)) < 0)
206 BUG();
207
208
209
210
211
212 return 0;
213}
214
215static const struct clock_event_device xen_timerop_clockevent = {
216 .name = "xen",
217 .features = CLOCK_EVT_FEAT_ONESHOT,
218
219 .max_delta_ns = 0xffffffff,
220 .max_delta_ticks = 0xffffffff,
221 .min_delta_ns = TIMER_SLOP,
222 .min_delta_ticks = TIMER_SLOP,
223
224 .mult = 1,
225 .shift = 0,
226 .rating = 500,
227
228 .set_state_shutdown = xen_timerop_shutdown,
229 .set_next_event = xen_timerop_set_next_event,
230};
231
232static int xen_vcpuop_shutdown(struct clock_event_device *evt)
233{
234 int cpu = smp_processor_id();
235
236 if (HYPERVISOR_vcpu_op(VCPUOP_stop_singleshot_timer, xen_vcpu_nr(cpu),
237 NULL) ||
238 HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer, xen_vcpu_nr(cpu),
239 NULL))
240 BUG();
241
242 return 0;
243}
244
245static int xen_vcpuop_set_oneshot(struct clock_event_device *evt)
246{
247 int cpu = smp_processor_id();
248
249 if (HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer, xen_vcpu_nr(cpu),
250 NULL))
251 BUG();
252
253 return 0;
254}
255
256static int xen_vcpuop_set_next_event(unsigned long delta,
257 struct clock_event_device *evt)
258{
259 int cpu = smp_processor_id();
260 struct vcpu_set_singleshot_timer single;
261 int ret;
262
263 WARN_ON(!clockevent_state_oneshot(evt));
264
265 single.timeout_abs_ns = get_abs_timeout(delta);
266
267 single.flags = 0;
268
269 ret = HYPERVISOR_vcpu_op(VCPUOP_set_singleshot_timer, xen_vcpu_nr(cpu),
270 &single);
271 BUG_ON(ret != 0);
272
273 return ret;
274}
275
276static const struct clock_event_device xen_vcpuop_clockevent = {
277 .name = "xen",
278 .features = CLOCK_EVT_FEAT_ONESHOT,
279
280 .max_delta_ns = 0xffffffff,
281 .max_delta_ticks = 0xffffffff,
282 .min_delta_ns = TIMER_SLOP,
283 .min_delta_ticks = TIMER_SLOP,
284
285 .mult = 1,
286 .shift = 0,
287 .rating = 500,
288
289 .set_state_shutdown = xen_vcpuop_shutdown,
290 .set_state_oneshot = xen_vcpuop_set_oneshot,
291 .set_next_event = xen_vcpuop_set_next_event,
292};
293
294static const struct clock_event_device *xen_clockevent =
295 &xen_timerop_clockevent;
296
297struct xen_clock_event_device {
298 struct clock_event_device evt;
299 char name[16];
300};
301static DEFINE_PER_CPU(struct xen_clock_event_device, xen_clock_events) = { .evt.irq = -1 };
302
303static irqreturn_t xen_timer_interrupt(int irq, void *dev_id)
304{
305 struct clock_event_device *evt = this_cpu_ptr(&xen_clock_events.evt);
306 irqreturn_t ret;
307
308 ret = IRQ_NONE;
309 if (evt->event_handler) {
310 evt->event_handler(evt);
311 ret = IRQ_HANDLED;
312 }
313
314 return ret;
315}
316
317void xen_teardown_timer(int cpu)
318{
319 struct clock_event_device *evt;
320 evt = &per_cpu(xen_clock_events, cpu).evt;
321
322 if (evt->irq >= 0) {
323 unbind_from_irqhandler(evt->irq, NULL);
324 evt->irq = -1;
325 }
326}
327
328void xen_setup_timer(int cpu)
329{
330 struct xen_clock_event_device *xevt = &per_cpu(xen_clock_events, cpu);
331 struct clock_event_device *evt = &xevt->evt;
332 int irq;
333
334 WARN(evt->irq >= 0, "IRQ%d for CPU%d is already allocated\n", evt->irq, cpu);
335 if (evt->irq >= 0)
336 xen_teardown_timer(cpu);
337
338 printk(KERN_INFO "installing Xen timer for CPU %d\n", cpu);
339
340 snprintf(xevt->name, sizeof(xevt->name), "timer%d", cpu);
341
342 irq = bind_virq_to_irqhandler(VIRQ_TIMER, cpu, xen_timer_interrupt,
343 IRQF_PERCPU|IRQF_NOBALANCING|IRQF_TIMER|
344 IRQF_FORCE_RESUME|IRQF_EARLY_RESUME,
345 xevt->name, NULL);
346 (void)xen_set_irq_priority(irq, XEN_IRQ_PRIORITY_MAX);
347
348 memcpy(evt, xen_clockevent, sizeof(*evt));
349
350 evt->cpumask = cpumask_of(cpu);
351 evt->irq = irq;
352}
353
354
355void xen_setup_cpu_clockevents(void)
356{
357 clockevents_register_device(this_cpu_ptr(&xen_clock_events.evt));
358}
359
360void xen_timer_resume(void)
361{
362 int cpu;
363
364 pvclock_resume();
365
366 if (xen_clockevent != &xen_vcpuop_clockevent)
367 return;
368
369 for_each_online_cpu(cpu) {
370 if (HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer,
371 xen_vcpu_nr(cpu), NULL))
372 BUG();
373 }
374}
375
376static const struct pv_time_ops xen_time_ops __initconst = {
377 .sched_clock = xen_sched_clock,
378 .steal_clock = xen_steal_clock,
379};
380
381static struct pvclock_vsyscall_time_info *xen_clock __read_mostly;
382
383void xen_save_time_memory_area(void)
384{
385 struct vcpu_register_time_memory_area t;
386 int ret;
387
388 if (!xen_clock)
389 return;
390
391 t.addr.v = NULL;
392
393 ret = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_time_memory_area, 0, &t);
394 if (ret != 0)
395 pr_notice("Cannot save secondary vcpu_time_info (err %d)",
396 ret);
397 else
398 clear_page(xen_clock);
399}
400
401void xen_restore_time_memory_area(void)
402{
403 struct vcpu_register_time_memory_area t;
404 int ret;
405
406 if (!xen_clock)
407 return;
408
409 t.addr.v = &xen_clock->pvti;
410
411 ret = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_time_memory_area, 0, &t);
412
413
414
415
416
417
418
419
420
421 if (ret != 0)
422 pr_notice("Cannot restore secondary vcpu_time_info (err %d)",
423 ret);
424}
425
426static void xen_setup_vsyscall_time_info(void)
427{
428 struct vcpu_register_time_memory_area t;
429 struct pvclock_vsyscall_time_info *ti;
430 int ret;
431
432 ti = (struct pvclock_vsyscall_time_info *)get_zeroed_page(GFP_KERNEL);
433 if (!ti)
434 return;
435
436 t.addr.v = &ti->pvti;
437
438 ret = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_time_memory_area, 0, &t);
439 if (ret) {
440 pr_notice("xen: VCLOCK_PVCLOCK not supported (err %d)\n", ret);
441 free_page((unsigned long)ti);
442 return;
443 }
444
445
446
447
448
449
450 if (!(ti->pvti.flags & PVCLOCK_TSC_STABLE_BIT)) {
451 t.addr.v = NULL;
452 ret = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_time_memory_area,
453 0, &t);
454 if (!ret)
455 free_page((unsigned long)ti);
456
457 pr_notice("xen: VCLOCK_PVCLOCK not supported (tsc unstable)\n");
458 return;
459 }
460
461 xen_clock = ti;
462 pvclock_set_pvti_cpu0_va(xen_clock);
463
464 xen_clocksource.archdata.vclock_mode = VCLOCK_PVCLOCK;
465}
466
467static void __init xen_time_init(void)
468{
469 struct pvclock_vcpu_time_info *pvti;
470 int cpu = smp_processor_id();
471 struct timespec64 tp;
472
473
474 if (xen_initial_domain())
475 xen_clocksource.rating = 275;
476
477 clocksource_register_hz(&xen_clocksource, NSEC_PER_SEC);
478
479 if (HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer, xen_vcpu_nr(cpu),
480 NULL) == 0) {
481
482
483 printk(KERN_DEBUG "Xen: using vcpuop timer interface\n");
484 xen_clockevent = &xen_vcpuop_clockevent;
485 }
486
487
488 xen_read_wallclock(&tp);
489 do_settimeofday64(&tp);
490
491 setup_force_cpu_cap(X86_FEATURE_TSC);
492
493
494
495
496
497 pvti = &__this_cpu_read(xen_vcpu)->time;
498 if (pvti->flags & PVCLOCK_TSC_STABLE_BIT) {
499 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
500 xen_setup_vsyscall_time_info();
501 }
502
503 xen_setup_runstate_info(cpu);
504 xen_setup_timer(cpu);
505 xen_setup_cpu_clockevents();
506
507 xen_time_setup_guest();
508
509 if (xen_initial_domain())
510 pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier);
511}
512
513void __init xen_init_time_ops(void)
514{
515 xen_sched_clock_offset = xen_clocksource_read();
516 pv_ops.time = xen_time_ops;
517
518 x86_init.timers.timer_init = xen_time_init;
519 x86_init.timers.setup_percpu_clockev = x86_init_noop;
520 x86_cpuinit.setup_percpu_clockev = x86_init_noop;
521
522 x86_platform.calibrate_tsc = xen_tsc_khz;
523 x86_platform.get_wallclock = xen_get_wallclock;
524
525 if (!xen_initial_domain())
526 x86_platform.set_wallclock = xen_set_wallclock;
527}
528
529#ifdef CONFIG_XEN_PVHVM
530static void xen_hvm_setup_cpu_clockevents(void)
531{
532 int cpu = smp_processor_id();
533 xen_setup_runstate_info(cpu);
534
535
536
537
538
539 xen_setup_cpu_clockevents();
540}
541
542void __init xen_hvm_init_time_ops(void)
543{
544
545
546
547
548
549 if (!xen_have_vector_callback)
550 return;
551
552 if (!xen_feature(XENFEAT_hvm_safe_pvclock)) {
553 pr_info("Xen doesn't support pvclock on HVM, disable pv timer");
554 return;
555 }
556
557 xen_sched_clock_offset = xen_clocksource_read();
558 pv_ops.time = xen_time_ops;
559 x86_init.timers.setup_percpu_clockev = xen_time_init;
560 x86_cpuinit.setup_percpu_clockev = xen_hvm_setup_cpu_clockevents;
561
562 x86_platform.calibrate_tsc = xen_tsc_khz;
563 x86_platform.get_wallclock = xen_get_wallclock;
564 x86_platform.set_wallclock = xen_set_wallclock;
565}
566#endif
567