Release 4.14 arch/x86/xen/time.c

// SPDX-License-Identifier: GPL-2.0
/*
 * Xen time implementation.
 *
 * This is implemented in terms of a clocksource driver which uses
 * the hypervisor clock as a nanosecond timebase, and a clockevent
 * driver which uses the hypervisor's timer mechanism.
 *
 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
 */
#include <linux/kernel.h>
#include <linux/interrupt.h>
#include <linux/clocksource.h>
#include <linux/clockchips.h>
#include <linux/gfp.h>
#include <linux/slab.h>
#include <linux/pvclock_gtod.h>
#include <linux/timekeeper_internal.h>

#include <asm/pvclock.h>
#include <asm/xen/hypervisor.h>
#include <asm/xen/hypercall.h>

#include <xen/events.h>
#include <xen/features.h>
#include <xen/interface/xen.h>
#include <xen/interface/vcpu.h>

#include "xen-ops.h"

/* Xen may fire a timer up to this many ns early */

#define TIMER_SLOP	100000

/* Get the TSC speed from Xen */


static unsigned long xen_tsc_khz(void)
{
	struct pvclock_vcpu_time_info *info =
		&HYPERVISOR_shared_info->vcpu_info[0].time;

	return pvclock_tsc_khz(info);
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
25
83.33%
1
20.00%
Glauber de Oliveira Costa
2
6.67%
1
20.00%
Gerd Hoffmann
1
3.33%
1
20.00%
Alok N Kataria
1
3.33%
1
20.00%
Stefano Stabellini
1
3.33%
1
20.00%
Total
30
100.00%
5
100.00%



u64 xen_clocksource_read(void)
{
        struct pvclock_vcpu_time_info *src;
	u64 ret;

	preempt_disable_notrace();
	src = &__this_cpu_read(xen_vcpu)->time;
	ret = pvclock_clocksource_read(src);
	preempt_enable_notrace();
	return ret;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
30
73.17%
3
42.86%
Gerd Hoffmann
5
12.20%
1
14.29%
Boris Ostrovsky
3
7.32%
1
14.29%
Thomas Gleixner
2
4.88%
1
14.29%
Christoph Lameter
1
2.44%
1
14.29%
Total
41
100.00%
7
100.00%



static u64 xen_clocksource_get_cycles(struct clocksource *cs)
{
	return xen_clocksource_read();
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Magnus Damm
14
93.33%
1
50.00%
Thomas Gleixner
1
6.67%
1
50.00%
Total
15
100.00%
2
100.00%



static void xen_read_wallclock(struct timespec *ts)
{
	struct shared_info *s = HYPERVISOR_shared_info;
	struct pvclock_wall_clock *wall_clock = &(s->wc);
        struct pvclock_vcpu_time_info *vcpu_time;

	vcpu_time = &get_cpu_var(xen_vcpu)->time;
	pvclock_read_wallclock(wall_clock, vcpu_time, ts);
	put_cpu_var(xen_vcpu);
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
33
55.93%
1
50.00%
Gerd Hoffmann
26
44.07%
1
50.00%
Total
59
100.00%
2
100.00%



static void xen_get_wallclock(struct timespec *now)
{
	xen_read_wallclock(now);
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
9
56.25%
1
33.33%
David Vrabel
6
37.50%
1
33.33%
Stefano Stabellini
1
6.25%
1
33.33%
Total
16
100.00%
3
100.00%



static int xen_set_wallclock(const struct timespec *now)
{
	return -1;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
11
68.75%
1
33.33%
David Vrabel
4
25.00%
1
33.33%
Stefano Stabellini
1
6.25%
1
33.33%
Total
16
100.00%
3
100.00%



static int xen_pvclock_gtod_notify(struct notifier_block *nb,
				   unsigned long was_set, void *priv)
{
	/* Protected by the calling core code serialization */
	static struct timespec64 next_sync;

	struct xen_platform_op op;
	struct timespec64 now;
	struct timekeeper *tk = priv;
	static bool settime64_supported = true;
	int ret;

	now.tv_sec = tk->xtime_sec;
	now.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);

	/*
         * We only take the expensive HV call when the clock was set
         * or when the 11 minutes RTC synchronization time elapsed.
         */
	if (!was_set && timespec64_compare(&now, &next_sync) < 0)
		return NOTIFY_OK;

again:
	if (settime64_supported) {
		op.cmd = XENPF_settime64;
		op.u.settime64.mbz = 0;
		op.u.settime64.secs = now.tv_sec;
		op.u.settime64.nsecs = now.tv_nsec;
		op.u.settime64.system_time = xen_clocksource_read();
	} else {
		op.cmd = XENPF_settime32;
		op.u.settime32.secs = now.tv_sec;
		op.u.settime32.nsecs = now.tv_nsec;
		op.u.settime32.system_time = xen_clocksource_read();
	}

	ret = HYPERVISOR_platform_op(&op);

	if (ret == -ENOSYS && settime64_supported) {
		settime64_supported = false;
		goto again;
	}
	if (ret < 0)
		return NOTIFY_BAD;

	/*
         * Move the next drift compensation time 11 minutes
         * ahead. That's emulating the sync_cmos_clock() update for
         * the hardware RTC.
         */
	next_sync = now;
	next_sync.tv_sec += 11 * 60;

	return NOTIFY_OK;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Stefano Stabellini
141
55.95%
4
50.00%
Jeremy Fitzhardinge
57
22.62%
2
25.00%
David Vrabel
54
21.43%
2
25.00%
Total
252
100.00%
8
100.00%


static struct notifier_block xen_pvclock_gtod_notifier = {
	.notifier_call = xen_pvclock_gtod_notify,
};


static struct clocksource xen_clocksource __read_mostly = {
	.name = "xen",
	.rating = 400,
	.read = xen_clocksource_get_cycles,
	.mask = ~0,
	.flags = CLOCK_SOURCE_IS_CONTINUOUS,
};

/*
   Xen clockevent implementation

   Xen has two clockevent implementations:

   The old timer_op one works with all released versions of Xen prior
   to version 3.0.4.  This version of the hypervisor provides a
   single-shot timer with nanosecond resolution.  However, sharing the
   same event channel is a 100Hz tick which is delivered while the
   vcpu is running.  We don't care about or use this tick, but it will
   cause the core time code to think the timer fired too soon, and
   will end up resetting it each time.  It could be filtered, but
   doing so has complications when the ktime clocksource is not yet
   the xen clocksource (ie, at boot time).

   The new vcpu_op-based timer interface allows the tick timer period
   to be changed or turned off.  The tick timer is not useful as a
   periodic timer because events are only delivered to running vcpus.
   The one-shot timer can report when a timeout is in the past, so
   set_next_event is capable of returning -ETIME when appropriate.
   This interface is used when available.
*/


/*
  Get a hypervisor absolute time.  In theory we could maintain an
  offset between the kernel's time and the hypervisor's time, and
  apply that to a kernel's absolute timeout.  Unfortunately the
  hypervisor and kernel times can drift even if the kernel is using
  the Xen clocksource, because ntp can warp the kernel's clocksource.
*/


static s64 get_abs_timeout(unsigned long delta)
{
	return xen_clocksource_read() + delta;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
16
100.00%
1
100.00%
Total
16
100.00%
1
100.00%



static int xen_timerop_shutdown(struct clock_event_device *evt)
{
	/* cancel timeout */
	HYPERVISOR_set_timer_op(0);

	return 0;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
14
70.00%
1
50.00%
Viresh Kumar
6
30.00%
1
50.00%
Total
20
100.00%
2
100.00%



static int xen_timerop_set_next_event(unsigned long delta,
				      struct clock_event_device *evt)
{
	WARN_ON(!clockevent_state_oneshot(evt));

	if (HYPERVISOR_set_timer_op(get_abs_timeout(delta)) < 0)
		BUG();

	/* We may have missed the deadline, but there's no real way of
           knowing for sure.  If the event was in the past, then we'll
           get an immediate interrupt. */

	return 0;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
39
90.70%
1
50.00%
Viresh Kumar
4
9.30%
1
50.00%
Total
43
100.00%
2
100.00%


static const struct clock_event_device xen_timerop_clockevent = {
	.name			= "xen",
	.features		= CLOCK_EVT_FEAT_ONESHOT,

	.max_delta_ns		= 0xffffffff,
	.max_delta_ticks	= 0xffffffff,
	.min_delta_ns		= TIMER_SLOP,
	.min_delta_ticks	= TIMER_SLOP,

	.mult			= 1,
	.shift			= 0,
	.rating			= 500,

	.set_state_shutdown	= xen_timerop_shutdown,
	.set_next_event		= xen_timerop_set_next_event,
};



static int xen_vcpuop_shutdown(struct clock_event_device *evt)
{
	int cpu = smp_processor_id();

	if (HYPERVISOR_vcpu_op(VCPUOP_stop_singleshot_timer, xen_vcpu_nr(cpu),
			       NULL) ||
	    HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer, xen_vcpu_nr(cpu),
			       NULL))
		BUG();

	return 0;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
29
59.18%
1
33.33%
Viresh Kumar
14
28.57%
1
33.33%
Vitaly Kuznetsov
6
12.24%
1
33.33%
Total
49
100.00%
3
100.00%



static int xen_vcpuop_set_oneshot(struct clock_event_device *evt)
{
	int cpu = smp_processor_id();

	if (HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer, xen_vcpu_nr(cpu),
			       NULL))
		BUG();

	return 0;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Viresh Kumar
19
51.35%
1
33.33%
Jeremy Fitzhardinge
15
40.54%
1
33.33%
Vitaly Kuznetsov
3
8.11%
1
33.33%
Total
37
100.00%
3
100.00%



static int xen_vcpuop_set_next_event(unsigned long delta,
				     struct clock_event_device *evt)
{
	int cpu = smp_processor_id();
	struct vcpu_set_singleshot_timer single;
	int ret;

	WARN_ON(!clockevent_state_oneshot(evt));

	single.timeout_abs_ns = get_abs_timeout(delta);
	/* Get an event anyway, even if the timeout is already expired */
	single.flags = 0;

	ret = HYPERVISOR_vcpu_op(VCPUOP_set_singleshot_timer, xen_vcpu_nr(cpu),
				 &single);
	BUG_ON(ret != 0);

	return ret;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
69
88.46%
1
25.00%
Viresh Kumar
4
5.13%
1
25.00%
Vitaly Kuznetsov
3
3.85%
1
25.00%
Stefano Stabellini
2
2.56%
1
25.00%
Total
78
100.00%
4
100.00%


static const struct clock_event_device xen_vcpuop_clockevent = {
	.name = "xen",
	.features = CLOCK_EVT_FEAT_ONESHOT,

	.max_delta_ns = 0xffffffff,
	.max_delta_ticks = 0xffffffff,
	.min_delta_ns = TIMER_SLOP,
	.min_delta_ticks = TIMER_SLOP,

	.mult = 1,
	.shift = 0,
	.rating = 500,

	.set_state_shutdown = xen_vcpuop_shutdown,
	.set_state_oneshot = xen_vcpuop_set_oneshot,
	.set_next_event = xen_vcpuop_set_next_event,
};


static const struct clock_event_device *xen_clockevent =
	&xen_timerop_clockevent;


struct xen_clock_event_device {
	
struct clock_event_device evt;
	
char name[16];
};
static DEFINE_PER_CPU(struct xen_clock_event_device, xen_clock_events) = { .evt.irq = -1 };



static irqreturn_t xen_timer_interrupt(int irq, void *dev_id)
{
	struct clock_event_device *evt = this_cpu_ptr(&xen_clock_events.evt);
	irqreturn_t ret;

	ret = IRQ_NONE;
	if (evt->event_handler) {
		evt->event_handler(evt);
		ret = IRQ_HANDLED;
	}

	return ret;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
50
90.91%
1
33.33%
Christoph Lameter
3
5.45%
1
33.33%
Konrad Rzeszutek Wilk
2
3.64%
1
33.33%
Total
55
100.00%
3
100.00%



void xen_teardown_timer(int cpu)
{
	struct clock_event_device *evt;
	evt = &per_cpu(xen_clock_events, cpu).evt;

	if (evt->irq >= 0) {
		unbind_from_irqhandler(evt->irq, NULL);
		evt->irq = -1;
	}
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Konrad Rzeszutek Wilk
51
100.00%
1
100.00%
Total
51
100.00%
1
100.00%



void xen_setup_timer(int cpu)
{
	struct xen_clock_event_device *xevt = &per_cpu(xen_clock_events, cpu);
	struct clock_event_device *evt = &xevt->evt;
	int irq;

	WARN(evt->irq >= 0, "IRQ%d for CPU%d is already allocated\n", evt->irq, cpu);
	if (evt->irq >= 0)
		xen_teardown_timer(cpu);

	printk(KERN_INFO "installing Xen timer for CPU %d\n", cpu);

	snprintf(xevt->name, sizeof(xevt->name), "timer%d", cpu);

	irq = bind_virq_to_irqhandler(VIRQ_TIMER, cpu, xen_timer_interrupt,
				      IRQF_PERCPU|IRQF_NOBALANCING|IRQF_TIMER|
				      IRQF_FORCE_RESUME|IRQF_EARLY_RESUME,
				      xevt->name, NULL);
	(void)xen_set_irq_priority(irq, XEN_IRQ_PRIORITY_MAX);

	memcpy(evt, xen_clockevent, sizeof(*evt));

	evt->cpumask = cpumask_of(cpu);
	evt->irq = irq;
}


Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
72
46.45%
2
18.18%
Konrad Rzeszutek Wilk
38
24.52%
3
27.27%
Vitaly Kuznetsov
28
18.06%
1
9.09%
David Vrabel
12
7.74%
2
18.18%
Ian Campbell
4
2.58%
2
18.18%
Rusty Russell
1
0.65%
1
9.09%
Total
155
100.00%
11
100.00%



void xen_setup_cpu_clockevents(void)
{
	clockevents_register_device(this_cpu_ptr(&xen_clock_events.evt));
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
13
72.22%
2
50.00%
Christoph Lameter
3
16.67%
1
25.00%
Konrad Rzeszutek Wilk
2
11.11%
1
25.00%
Total
18
100.00%
4
100.00%



void xen_timer_resume(void)
{
	int cpu;

	pvclock_resume();

	if (xen_clockevent != &xen_vcpuop_clockevent)
		return;

	for_each_online_cpu(cpu) {
		if (HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer,
				       xen_vcpu_nr(cpu), NULL))
			BUG();
	}
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
41
93.18%
2
66.67%
Vitaly Kuznetsov
3
6.82%
1
33.33%
Total
44
100.00%
3
100.00%


static const struct pv_time_ops xen_time_ops __initconst = {
	.sched_clock = xen_clocksource_read,
	.steal_clock = xen_steal_clock,
};



static void __init xen_time_init(void)
{
	int cpu = smp_processor_id();
	struct timespec tp;

	/* As Dom0 is never moved, no penalty on using TSC there */
	if (xen_initial_domain())
		xen_clocksource.rating = 275;

	clocksource_register_hz(&xen_clocksource, NSEC_PER_SEC);

	if (HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer, xen_vcpu_nr(cpu),
			       NULL) == 0) {
		/* Successfully turned off 100Hz tick, so we have the
                   vcpuop-based timer interface */
		printk(KERN_DEBUG "Xen: using vcpuop timer interface\n");
		xen_clockevent = &xen_vcpuop_clockevent;
	}

	/* Set initial system time with full resolution */
	xen_read_wallclock(&tp);
	do_settimeofday(&tp);

	setup_force_cpu_cap(X86_FEATURE_TSC);

	xen_setup_runstate_info(cpu);
	xen_setup_timer(cpu);
	xen_setup_cpu_clockevents();

	xen_time_setup_guest();

	if (xen_initial_domain())
		pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier);
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
64
56.14%
3
23.08%
Imre Palik
12
10.53%
1
7.69%
David Vrabel
11
9.65%
1
7.69%
John Stultz
10
8.77%
2
15.38%
Ian Campbell
5
4.39%
1
7.69%
Andi Kleen
4
3.51%
1
7.69%
Vitaly Kuznetsov
3
2.63%
1
7.69%
Juergen Gross
3
2.63%
1
7.69%
Stefano Stabellini
1
0.88%
1
7.69%
Daniel Kiper
1
0.88%
1
7.69%
Total
114
100.00%
13
100.00%



void __ref xen_init_time_ops(void)
{
	pv_time_ops = xen_time_ops;

	x86_init.timers.timer_init = xen_time_init;
	x86_init.timers.setup_percpu_clockev = x86_init_noop;
	x86_cpuinit.setup_percpu_clockev = x86_init_noop;

	x86_platform.calibrate_tsc = xen_tsc_khz;
	x86_platform.get_wallclock = xen_get_wallclock;
	/* Dom0 uses the native method to set the hardware RTC. */
	if (!xen_initial_domain())
		x86_platform.set_wallclock = xen_set_wallclock;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Stefano Stabellini
51
86.44%
1
33.33%
David Vrabel
7
11.86%
1
33.33%
Boris Ostrovsky
1
1.69%
1
33.33%
Total
59
100.00%
3
100.00%

#ifdef CONFIG_XEN_PVHVM


static void xen_hvm_setup_cpu_clockevents(void)
{
	int cpu = smp_processor_id();
	xen_setup_runstate_info(cpu);
	/*
         * xen_setup_timer(cpu) - snprintf is bad in atomic context. Hence
         * doing it xen_hvm_cpu_notify (which gets called by smp_init during
         * early bootup and also during CPU hotplug events).
         */
	xen_setup_cpu_clockevents();
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Stefano Stabellini
22
95.65%
1
50.00%
Konrad Rzeszutek Wilk
1
4.35%
1
50.00%
Total
23
100.00%
2
100.00%



void __init xen_hvm_init_time_ops(void)
{
	/*
         * vector callback is needed otherwise we cannot receive interrupts
         * on cpu > 0 and at this point we don't know how many cpus are
         * available.
         */
	if (!xen_have_vector_callback)
		return;

	if (!xen_feature(XENFEAT_hvm_safe_pvclock)) {
		printk(KERN_INFO "Xen doesn't support pvclock on HVM,"
				"disable pv timer\n");
		return;
	}

	pv_time_ops = xen_time_ops;
	x86_init.timers.setup_percpu_clockev = xen_time_init;
	x86_cpuinit.setup_percpu_clockev = xen_hvm_setup_cpu_clockevents;

	x86_platform.calibrate_tsc = xen_tsc_khz;
	x86_platform.get_wallclock = xen_get_wallclock;
	x86_platform.set_wallclock = xen_set_wallclock;
}
Contributors
Person
Tokens
Prop
Commits
CommitProp
Stefano Stabellini
61
88.41%
1
33.33%
Boris Ostrovsky
7
10.14%
1
33.33%
Daniel Kiper
1
1.45%
1
33.33%
Total
69
100.00%
3
100.00%

#endif
Overall Contributors
Person
Tokens
Prop
Commits
CommitProp
Jeremy Fitzhardinge
779
49.84%
8
15.69%
Stefano Stabellini
304
19.45%
7
13.73%
Konrad Rzeszutek Wilk
120
7.68%
5
9.80%
David Vrabel
109
6.97%
5
9.80%
Viresh Kumar
56
3.58%
1
1.96%
Vitaly Kuznetsov
49
3.13%
2
3.92%
Gerd Hoffmann
35
2.24%
1
1.96%
Nicolai Stange
20
1.28%
1
1.96%
Magnus Damm
15
0.96%
1
1.96%
Imre Palik
12
0.77%
1
1.96%
Boris Ostrovsky
11
0.70%
3
5.88%
John Stultz
10
0.64%
2
3.92%
Ian Campbell
9
0.58%
3
5.88%
Juergen Gross
8
0.51%
2
3.92%
Christoph Lameter
7
0.45%
1
1.96%
Andi Kleen
4
0.26%
1
1.96%
Thomas Gleixner
3
0.19%
1
1.96%
Tejun Heo
3
0.19%
1
1.96%
Daniel Kiper
3
0.19%
1
1.96%
Glauber de Oliveira Costa
2
0.13%
1
1.96%
Alok N Kataria
2
0.13%
1
1.96%
Rusty Russell
1
0.06%
1
1.96%
Greg Kroah-Hartman
1
0.06%
1
1.96%
Total
1563
100.00%
51
100.00%
Directory: arch/x86/xen

Information contained on this website is for historical information purposes only and does not indicate or represent copyright ownership.