[mirror_ubuntu-zesty-kernel.git] / arch / x86 / xen / time.c

/*
 * 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/kernel_stat.h>
#include <linux/math64.h>

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

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

#include "xen-ops.h"

#define XEN_SHIFT 22

/* Xen may fire a timer up to this many ns early */
#define TIMER_SLOP	100000
#define NS_PER_TICK	(1000000000LL / HZ)

/* runstate info updated by Xen */
static DEFINE_PER_CPU(struct vcpu_runstate_info, runstate);

/* snapshots of runstate info */
static DEFINE_PER_CPU(struct vcpu_runstate_info, runstate_snapshot);

/* unused ns of stolen and blocked time */
static DEFINE_PER_CPU(u64, residual_stolen);
static DEFINE_PER_CPU(u64, residual_blocked);

/* return an consistent snapshot of 64-bit time/counter value */
static u64 get64(const u64 *p)
{
	u64 ret;

	if (BITS_PER_LONG < 64) {
		u32 *p32 = (u32 *)p;
		u32 h, l;

		/*
		 * Read high then low, and then make sure high is
		 * still the same; this will only loop if low wraps
		 * and carries into high.
		 * XXX some clean way to make this endian-proof?
		 */
		do {
			h = p32[1];
			barrier();
			l = p32[0];
			barrier();
		} while (p32[1] != h);

		ret = (((u64)h) << 32) | l;
	} else
		ret = *p;

	return ret;
}

/*
 * Runstate accounting
 */
static void get_runstate_snapshot(struct vcpu_runstate_info *res)
{
	u64 state_time;
	struct vcpu_runstate_info *state;

	BUG_ON(preemptible());

	state = &__get_cpu_var(runstate);

	/*
	 * The runstate info is always updated by the hypervisor on
	 * the current CPU, so there's no need to use anything
	 * stronger than a compiler barrier when fetching it.
	 */
	do {
		state_time = get64(&state->state_entry_time);
		barrier();
		*res = *state;
		barrier();
	} while (get64(&state->state_entry_time) != state_time);
}

/* return true when a vcpu could run but has no real cpu to run on */
bool xen_vcpu_stolen(int vcpu)
{
	return per_cpu(runstate, vcpu).state == RUNSTATE_runnable;
}

static void setup_runstate_info(int cpu)
{
	struct vcpu_register_runstate_memory_area area;

	area.addr.v = &per_cpu(runstate, cpu);

	if (HYPERVISOR_vcpu_op(VCPUOP_register_runstate_memory_area,
			       cpu, &area))
		BUG();
}

static void do_stolen_accounting(void)
{
	struct vcpu_runstate_info state;
	struct vcpu_runstate_info *snap;
	s64 blocked, runnable, offline, stolen;
	cputime_t ticks;

	get_runstate_snapshot(&state);

	WARN_ON(state.state != RUNSTATE_running);

	snap = &__get_cpu_var(runstate_snapshot);

	/* work out how much time the VCPU has not been runn*ing*  */
	blocked = state.time[RUNSTATE_blocked] - snap->time[RUNSTATE_blocked];
	runnable = state.time[RUNSTATE_runnable] - snap->time[RUNSTATE_runnable];
	offline = state.time[RUNSTATE_offline] - snap->time[RUNSTATE_offline];

	*snap = state;

	/* Add the appropriate number of ticks of stolen time,
	   including any left-overs from last time. */
	stolen = runnable + offline + __get_cpu_var(residual_stolen);

	if (stolen < 0)
		stolen = 0;

	ticks = iter_div_u64_rem(stolen, NS_PER_TICK, &stolen);
	__get_cpu_var(residual_stolen) = stolen;
	account_steal_ticks(ticks);

	/* Add the appropriate number of ticks of blocked time,
	   including any left-overs from last time. */
	blocked += __get_cpu_var(residual_blocked);

	if (blocked < 0)
		blocked = 0;

	ticks = iter_div_u64_rem(blocked, NS_PER_TICK, &blocked);
	__get_cpu_var(residual_blocked) = blocked;
	account_idle_ticks(ticks);
}

/*
 * Xen sched_clock implementation.  Returns the number of unstolen
 * nanoseconds, which is nanoseconds the VCPU spent in RUNNING+BLOCKED
 * states.
 */
unsigned long long xen_sched_clock(void)
{
	struct vcpu_runstate_info state;
	cycle_t now;
	u64 ret;
	s64 offset;

	/*
	 * Ideally sched_clock should be called on a per-cpu basis
	 * anyway, so preempt should already be disabled, but that's
	 * not current practice at the moment.
	 */
	preempt_disable();

	now = xen_clocksource_read();

	get_runstate_snapshot(&state);

	WARN_ON(state.state != RUNSTATE_running);

	offset = now - state.state_entry_time;
	if (offset < 0)
		offset = 0;

	ret = state.time[RUNSTATE_blocked] +
		state.time[RUNSTATE_running] +
		offset;

	preempt_enable();

	return ret;
}


/* Get the TSC speed from Xen */
unsigned long xen_tsc_khz(void)
{
	struct pvclock_vcpu_time_info *info =
		&HYPERVISOR_shared_info->vcpu_info[0].time;

	return pvclock_tsc_khz(info);
}

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

	src = &get_cpu_var(xen_vcpu)->time;
	ret = pvclock_clocksource_read(src);
	put_cpu_var(xen_vcpu);
	return ret;
}

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

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);
}

unsigned long xen_get_wallclock(void)
{
	struct timespec ts;

	xen_read_wallclock(&ts);
	return ts.tv_sec;
}

int xen_set_wallclock(unsigned long now)
{
	/* do nothing for domU */
	return -1;
}

static struct clocksource xen_clocksource __read_mostly = {
	.name = "xen",
	.rating = 400,
	.read = xen_clocksource_get_cycles,
	.mask = ~0,
	.mult = 1<<XEN_SHIFT,		/* time directly in nanoseconds */
	.shift = XEN_SHIFT,
	.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;
}

static void xen_timerop_set_mode(enum clock_event_mode mode,
				 struct clock_event_device *evt)
{
	switch (mode) {
	case CLOCK_EVT_MODE_PERIODIC:
		/* unsupported */
		WARN_ON(1);
		break;

	case CLOCK_EVT_MODE_ONESHOT:
	case CLOCK_EVT_MODE_RESUME:
		break;

	case CLOCK_EVT_MODE_UNUSED:
	case CLOCK_EVT_MODE_SHUTDOWN:
		HYPERVISOR_set_timer_op(0);  /* cancel timeout */
		break;
	}
}

static int xen_timerop_set_next_event(unsigned long delta,
				      struct clock_event_device *evt)
{
	WARN_ON(evt->mode != CLOCK_EVT_MODE_ONESHOT);

	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;
}

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

	.max_delta_ns = 0xffffffff,
	.min_delta_ns = TIMER_SLOP,

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

	.set_mode = xen_timerop_set_mode,
	.set_next_event = xen_timerop_set_next_event,
};


static void xen_vcpuop_set_mode(enum clock_event_mode mode,
				struct clock_event_device *evt)
{
	int cpu = smp_processor_id();

	switch (mode) {
	case CLOCK_EVT_MODE_PERIODIC:
		WARN_ON(1);	/* unsupported */
		break;

	case CLOCK_EVT_MODE_ONESHOT:
		if (HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer, cpu, NULL))
			BUG();
		break;

	case CLOCK_EVT_MODE_UNUSED:
	case CLOCK_EVT_MODE_SHUTDOWN:
		if (HYPERVISOR_vcpu_op(VCPUOP_stop_singleshot_timer, cpu, NULL) ||
		    HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer, cpu, NULL))
			BUG();
		break;
	case CLOCK_EVT_MODE_RESUME:
		break;
	}
}

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(evt->mode != CLOCK_EVT_MODE_ONESHOT);

	single.timeout_abs_ns = get_abs_timeout(delta);
	single.flags = VCPU_SSHOTTMR_future;

	ret = HYPERVISOR_vcpu_op(VCPUOP_set_singleshot_timer, cpu, &single);

	BUG_ON(ret != 0 && ret != -ETIME);

	return ret;
}

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

	.max_delta_ns = 0xffffffff,
	.min_delta_ns = TIMER_SLOP,

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

	.set_mode = xen_vcpuop_set_mode,
	.set_next_event = xen_vcpuop_set_next_event,
};

static const struct clock_event_device *xen_clockevent =
	&xen_timerop_clockevent;
static DEFINE_PER_CPU(struct clock_event_device, xen_clock_events);

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

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

	do_stolen_accounting();

	return ret;
}

void xen_setup_timer(int cpu)
{
	const char *name;
	struct clock_event_device *evt;
	int irq;

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

	name = kasprintf(GFP_KERNEL, "timer%d", cpu);
	if (!name)
		name = "<timer kasprintf failed>";

	irq = bind_virq_to_irqhandler(VIRQ_TIMER, cpu, xen_timer_interrupt,
				      IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
				      name, NULL);

	evt = &per_cpu(xen_clock_events, cpu);
	memcpy(evt, xen_clockevent, sizeof(*evt));

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

	setup_runstate_info(cpu);
}

void xen_teardown_timer(int cpu)
{
	struct clock_event_device *evt;
	BUG_ON(cpu == 0);
	evt = &per_cpu(xen_clock_events, cpu);
	unbind_from_irqhandler(evt->irq, NULL);
}

void xen_setup_cpu_clockevents(void)
{
	BUG_ON(preemptible());

	clockevents_register_device(&__get_cpu_var(xen_clock_events));
}

void xen_timer_resume(void)
{
	int cpu;

	if (xen_clockevent != &xen_vcpuop_clockevent)
		return;

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

__init void xen_time_init(void)
{
	int cpu = smp_processor_id();

	clocksource_register(&xen_clocksource);

	if (HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer, 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(&xtime);
	set_normalized_timespec(&wall_to_monotonic,
				-xtime.tv_sec, -xtime.tv_nsec);

	setup_force_cpu_cap(X86_FEATURE_TSC);

	xen_setup_timer(cpu);
	xen_setup_cpu_clockevents();
}
Commit	Line	Data
15c84731 JF	1	/*
	2	* Xen time implementation.
	3	*
	4	* This is implemented in terms of a clocksource driver which uses
	5	* the hypervisor clock as a nanosecond timebase, and a clockevent
	6	* driver which uses the hypervisor's timer mechanism.
	7	*
	8	* Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
	9	*/
	10	#include <linux/kernel.h>
	11	#include <linux/interrupt.h>
	12	#include <linux/clocksource.h>
	13	#include <linux/clockchips.h>
f91a8b44	14	#include <linux/kernel_stat.h>
f595ec96	15	#include <linux/math64.h>
15c84731	16
1c7b67f7	17	#include <asm/pvclock.h>
15c84731 JF	18	#include <asm/xen/hypervisor.h>
	19	#include <asm/xen/hypercall.h>
	20
	21	#include <xen/events.h>
	22	#include <xen/interface/xen.h>
	23	#include <xen/interface/vcpu.h>
	24
	25	#include "xen-ops.h"
	26
	27	#define XEN_SHIFT 22
	28
	29	/* Xen may fire a timer up to this many ns early */
	30	#define TIMER_SLOP 100000
f91a8b44	31	#define NS_PER_TICK (1000000000LL / HZ)
15c84731	32
f91a8b44 JF	33	/* runstate info updated by Xen */
	34	static DEFINE_PER_CPU(struct vcpu_runstate_info, runstate);
	35
	36	/* snapshots of runstate info */
	37	static DEFINE_PER_CPU(struct vcpu_runstate_info, runstate_snapshot);
	38
	39	/* unused ns of stolen and blocked time */
	40	static DEFINE_PER_CPU(u64, residual_stolen);
	41	static DEFINE_PER_CPU(u64, residual_blocked);
	42
	43	/* return an consistent snapshot of 64-bit time/counter value */
	44	static u64 get64(const u64 *p)
	45	{
	46	u64 ret;
	47
	48	if (BITS_PER_LONG < 64) {
	49	u32 p32 = (u32 )p;
	50	u32 h, l;
	51
	52	/*
	53	* Read high then low, and then make sure high is
	54	* still the same; this will only loop if low wraps
	55	* and carries into high.
	56	* XXX some clean way to make this endian-proof?
	57	*/
	58	do {
	59	h = p32[1];
	60	barrier();
	61	l = p32[0];
	62	barrier();
	63	} while (p32[1] != h);
	64
	65	ret = (((u64)h) << 32) \| l;
	66	} else
	67	ret = *p;
	68
	69	return ret;
	70	}
	71
	72	/*
	73	* Runstate accounting
	74	*/
	75	static void get_runstate_snapshot(struct vcpu_runstate_info *res)
	76	{
	77	u64 state_time;
	78	struct vcpu_runstate_info *state;
	79
f120f13e	80	BUG_ON(preemptible());
f91a8b44 JF	81
	82	state = &__get_cpu_var(runstate);
	83
	84	/*
	85	* The runstate info is always updated by the hypervisor on
	86	* the current CPU, so there's no need to use anything
	87	* stronger than a compiler barrier when fetching it.
	88	*/
	89	do {
	90	state_time = get64(&state->state_entry_time);
	91	barrier();
	92	res = state;
	93	barrier();
	94	} while (get64(&state->state_entry_time) != state_time);
f91a8b44 JF	95	}
f91a8b44 JF	96
f0d73394 JF	97	/* return true when a vcpu could run but has no real cpu to run on */
	98	bool xen_vcpu_stolen(int vcpu)
	99	{
	100	return per_cpu(runstate, vcpu).state == RUNSTATE_runnable;
	101	}
	102
f91a8b44 JF	103	static void setup_runstate_info(int cpu)
	104	{
	105	struct vcpu_register_runstate_memory_area area;
	106
	107	area.addr.v = &per_cpu(runstate, cpu);
	108
	109	if (HYPERVISOR_vcpu_op(VCPUOP_register_runstate_memory_area,
	110	cpu, &area))
	111	BUG();
	112	}
	113
	114	static void do_stolen_accounting(void)
	115	{
	116	struct vcpu_runstate_info state;
	117	struct vcpu_runstate_info *snap;
	118	s64 blocked, runnable, offline, stolen;
	119	cputime_t ticks;
	120
	121	get_runstate_snapshot(&state);
	122
	123	WARN_ON(state.state != RUNSTATE_running);
	124
	125	snap = &__get_cpu_var(runstate_snapshot);
	126
	127	/* work out how much time the VCPU has not been running */
	128	blocked = state.time[RUNSTATE_blocked] - snap->time[RUNSTATE_blocked];
	129	runnable = state.time[RUNSTATE_runnable] - snap->time[RUNSTATE_runnable];
	130	offline = state.time[RUNSTATE_offline] - snap->time[RUNSTATE_offline];
	131
	132	*snap = state;
	133
	134	/* Add the appropriate number of ticks of stolen time,
79741dd3	135	including any left-overs from last time. */
f91a8b44 JF	136	stolen = runnable + offline + __get_cpu_var(residual_stolen);
	137
	138	if (stolen < 0)
	139	stolen = 0;
	140
f595ec96	141	ticks = iter_div_u64_rem(stolen, NS_PER_TICK, &stolen);
f91a8b44	142	__get_cpu_var(residual_stolen) = stolen;
79741dd3	143	account_steal_ticks(ticks);
f91a8b44 JF	144
f91a8b44 JF	145	/* Add the appropriate number of ticks of blocked time,
79741dd3	146	including any left-overs from last time. */
f91a8b44 JF	147	blocked += __get_cpu_var(residual_blocked);
	148
	149	if (blocked < 0)
	150	blocked = 0;
	151
f595ec96	152	ticks = iter_div_u64_rem(blocked, NS_PER_TICK, &blocked);
f91a8b44	153	__get_cpu_var(residual_blocked) = blocked;
79741dd3	154	account_idle_ticks(ticks);
f91a8b44 JF	155	}
f91a8b44 JF	156
ab550288 JF	157	/*
	158	* Xen sched_clock implementation. Returns the number of unstolen
	159	* nanoseconds, which is nanoseconds the VCPU spent in RUNNING+BLOCKED
	160	* states.
	161	*/
	162	unsigned long long xen_sched_clock(void)
	163	{
	164	struct vcpu_runstate_info state;
f120f13e JF	165	cycle_t now;
f120f13e JF	166	u64 ret;
ab550288 JF	167	s64 offset;
ab550288 JF	168
f120f13e JF	169	/*
	170	* Ideally sched_clock should be called on a per-cpu basis
	171	* anyway, so preempt should already be disabled, but that's
	172	* not current practice at the moment.
	173	*/
	174	preempt_disable();
	175
	176	now = xen_clocksource_read();
	177
ab550288 JF	178	get_runstate_snapshot(&state);
	179
	180	WARN_ON(state.state != RUNSTATE_running);
	181
	182	offset = now - state.state_entry_time;
	183	if (offset < 0)
	184	offset = 0;
	185
f120f13e	186	ret = state.time[RUNSTATE_blocked] +
ab550288 JF	187	state.time[RUNSTATE_running] +
ab550288 JF	188	offset;
f120f13e JF	189
	190	preempt_enable();
	191
	192	return ret;
ab550288	193	}
f91a8b44 JF	194
f91a8b44 JF	195
e93ef949 AK	196	/* Get the TSC speed from Xen */
e93ef949 AK	197	unsigned long xen_tsc_khz(void)
15c84731	198	{
3807f345	199	struct pvclock_vcpu_time_info *info =
15c84731 JF	200	&HYPERVISOR_shared_info->vcpu_info[0].time;
15c84731 JF	201
3807f345	202	return pvclock_tsc_khz(info);
15c84731 JF	203	}
15c84731 JF	204
ee7686bc	205	cycle_t xen_clocksource_read(void)
15c84731	206	{
1c7b67f7	207	struct pvclock_vcpu_time_info *src;
15c84731	208	cycle_t ret;
15c84731	209
1c7b67f7 GH	210	src = &get_cpu_var(xen_vcpu)->time;
	211	ret = pvclock_clocksource_read(src);
	212	put_cpu_var(xen_vcpu);
15c84731 JF	213	return ret;
	214	}
	215
8e19608e MD	216	static cycle_t xen_clocksource_get_cycles(struct clocksource *cs)
	217	{
	218	return xen_clocksource_read();
	219	}
	220
15c84731 JF	221	static void xen_read_wallclock(struct timespec *ts)
15c84731 JF	222	{
1c7b67f7 GH	223	struct shared_info *s = HYPERVISOR_shared_info;
	224	struct pvclock_wall_clock *wall_clock = &(s->wc);
	225	struct pvclock_vcpu_time_info *vcpu_time;
15c84731	226
1c7b67f7 GH	227	vcpu_time = &get_cpu_var(xen_vcpu)->time;
	228	pvclock_read_wallclock(wall_clock, vcpu_time, ts);
	229	put_cpu_var(xen_vcpu);
15c84731 JF	230	}
	231
	232	unsigned long xen_get_wallclock(void)
	233	{
	234	struct timespec ts;
	235
	236	xen_read_wallclock(&ts);
15c84731 JF	237	return ts.tv_sec;
	238	}
	239
	240	int xen_set_wallclock(unsigned long now)
	241	{
	242	/* do nothing for domU */
	243	return -1;
	244	}
	245
	246	static struct clocksource xen_clocksource __read_mostly = {
	247	.name = "xen",
	248	.rating = 400,
8e19608e	249	.read = xen_clocksource_get_cycles,
15c84731 JF	250	.mask = ~0,
	251	.mult = 1<<XEN_SHIFT, /* time directly in nanoseconds */
	252	.shift = XEN_SHIFT,
	253	.flags = CLOCK_SOURCE_IS_CONTINUOUS,
	254	};
	255
	256	/*
	257	Xen clockevent implementation
	258
	259	Xen has two clockevent implementations:
	260
	261	The old timer_op one works with all released versions of Xen prior
	262	to version 3.0.4. This version of the hypervisor provides a
	263	single-shot timer with nanosecond resolution. However, sharing the
	264	same event channel is a 100Hz tick which is delivered while the
	265	vcpu is running. We don't care about or use this tick, but it will
	266	cause the core time code to think the timer fired too soon, and
	267	will end up resetting it each time. It could be filtered, but
	268	doing so has complications when the ktime clocksource is not yet
	269	the xen clocksource (ie, at boot time).
	270
	271	The new vcpu_op-based timer interface allows the tick timer period
	272	to be changed or turned off. The tick timer is not useful as a
	273	periodic timer because events are only delivered to running vcpus.
	274	The one-shot timer can report when a timeout is in the past, so
	275	set_next_event is capable of returning -ETIME when appropriate.
	276	This interface is used when available.
	277	*/
	278
	279
	280	/*
	281	Get a hypervisor absolute time. In theory we could maintain an
	282	offset between the kernel's time and the hypervisor's time, and
	283	apply that to a kernel's absolute timeout. Unfortunately the
	284	hypervisor and kernel times can drift even if the kernel is using
	285	the Xen clocksource, because ntp can warp the kernel's clocksource.
	286	*/
	287	static s64 get_abs_timeout(unsigned long delta)
	288	{
	289	return xen_clocksource_read() + delta;
	290	}
	291
	292	static void xen_timerop_set_mode(enum clock_event_mode mode,
	293	struct clock_event_device *evt)
	294	{
	295	switch (mode) {
	296	case CLOCK_EVT_MODE_PERIODIC:
	297	/* unsupported */
	298	WARN_ON(1);
	299	break;
	300
	301	case CLOCK_EVT_MODE_ONESHOT:
18de5bc4	302	case CLOCK_EVT_MODE_RESUME:
15c84731 JF	303	break;
	304
	305	case CLOCK_EVT_MODE_UNUSED:
	306	case CLOCK_EVT_MODE_SHUTDOWN:
	307	HYPERVISOR_set_timer_op(0); /* cancel timeout */
	308	break;
	309	}
	310	}
	311
	312	static int xen_timerop_set_next_event(unsigned long delta,
	313	struct clock_event_device *evt)
	314	{
	315	WARN_ON(evt->mode != CLOCK_EVT_MODE_ONESHOT);
	316
	317	if (HYPERVISOR_set_timer_op(get_abs_timeout(delta)) < 0)
	318	BUG();
	319
	320	/* We may have missed the deadline, but there's no real way of
	321	knowing for sure. If the event was in the past, then we'll
	322	get an immediate interrupt. */
	323
	324	return 0;
	325	}
	326
	327	static const struct clock_event_device xen_timerop_clockevent = {
	328	.name = "xen",
	329	.features = CLOCK_EVT_FEAT_ONESHOT,
	330
	331	.max_delta_ns = 0xffffffff,
	332	.min_delta_ns = TIMER_SLOP,
	333
	334	.mult = 1,
	335	.shift = 0,
	336	.rating = 500,
	337
	338	.set_mode = xen_timerop_set_mode,
	339	.set_next_event = xen_timerop_set_next_event,
	340	};
	341
	342
	343
	344	static void xen_vcpuop_set_mode(enum clock_event_mode mode,
	345	struct clock_event_device *evt)
	346	{
	347	int cpu = smp_processor_id();
	348
	349	switch (mode) {
	350	case CLOCK_EVT_MODE_PERIODIC:
	351	WARN_ON(1); /* unsupported */
	352	break;
	353
	354	case CLOCK_EVT_MODE_ONESHOT:
	355	if (HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer, cpu, NULL))
	356	BUG();
	357	break;
	358
	359	case CLOCK_EVT_MODE_UNUSED:
	360	case CLOCK_EVT_MODE_SHUTDOWN:
	361	if (HYPERVISOR_vcpu_op(VCPUOP_stop_singleshot_timer, cpu, NULL) \|\|
	362	HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer, cpu, NULL))
	363	BUG();
	364	break;
18de5bc4 TG	365	case CLOCK_EVT_MODE_RESUME:
18de5bc4 TG	366	break;
15c84731 JF	367	}
	368	}
	369
	370	static int xen_vcpuop_set_next_event(unsigned long delta,
	371	struct clock_event_device *evt)
	372	{
	373	int cpu = smp_processor_id();
	374	struct vcpu_set_singleshot_timer single;
	375	int ret;
	376
	377	WARN_ON(evt->mode != CLOCK_EVT_MODE_ONESHOT);
	378
	379	single.timeout_abs_ns = get_abs_timeout(delta);
	380	single.flags = VCPU_SSHOTTMR_future;
	381
	382	ret = HYPERVISOR_vcpu_op(VCPUOP_set_singleshot_timer, cpu, &single);
	383
	384	BUG_ON(ret != 0 && ret != -ETIME);
	385
	386	return ret;
	387	}
	388
	389	static const struct clock_event_device xen_vcpuop_clockevent = {
	390	.name = "xen",
	391	.features = CLOCK_EVT_FEAT_ONESHOT,
	392
	393	.max_delta_ns = 0xffffffff,
	394	.min_delta_ns = TIMER_SLOP,
	395
	396	.mult = 1,
	397	.shift = 0,
	398	.rating = 500,
	399
	400	.set_mode = xen_vcpuop_set_mode,
	401	.set_next_event = xen_vcpuop_set_next_event,
	402	};
	403
	404	static const struct clock_event_device *xen_clockevent =
	405	&xen_timerop_clockevent;
	406	static DEFINE_PER_CPU(struct clock_event_device, xen_clock_events);
	407
	408	static irqreturn_t xen_timer_interrupt(int irq, void *dev_id)
	409	{
	410	struct clock_event_device *evt = &__get_cpu_var(xen_clock_events);
	411	irqreturn_t ret;
	412
	413	ret = IRQ_NONE;
	414	if (evt->event_handler) {
	415	evt->event_handler(evt);
	416	ret = IRQ_HANDLED;
	417	}
	418
f91a8b44 JF	419	do_stolen_accounting();
f91a8b44 JF	420
15c84731 JF	421	return ret;
	422	}
	423
f87e4cac	424	void xen_setup_timer(int cpu)
15c84731 JF	425	{
	426	const char *name;
	427	struct clock_event_device *evt;
	428	int irq;
	429
	430	printk(KERN_INFO "installing Xen timer for CPU %d\n", cpu);
	431
	432	name = kasprintf(GFP_KERNEL, "timer%d", cpu);
	433	if (!name)
	434	name = "<timer kasprintf failed>";
	435
	436	irq = bind_virq_to_irqhandler(VIRQ_TIMER, cpu, xen_timer_interrupt,
	437	IRQF_DISABLED\|IRQF_PERCPU\|IRQF_NOBALANCING,
	438	name, NULL);
	439
f87e4cac	440	evt = &per_cpu(xen_clock_events, cpu);
15c84731 JF	441	memcpy(evt, xen_clockevent, sizeof(*evt));
15c84731 JF	442
320ab2b0	443	evt->cpumask = cpumask_of(cpu);
15c84731	444	evt->irq = irq;
15c84731	445
f91a8b44	446	setup_runstate_info(cpu);
f87e4cac JF	447	}
f87e4cac JF	448
d68d82af AN	449	void xen_teardown_timer(int cpu)
	450	{
	451	struct clock_event_device *evt;
	452	BUG_ON(cpu == 0);
	453	evt = &per_cpu(xen_clock_events, cpu);
	454	unbind_from_irqhandler(evt->irq, NULL);
	455	}
	456
f87e4cac JF	457	void xen_setup_cpu_clockevents(void)
	458	{
	459	BUG_ON(preemptible());
f91a8b44	460
f87e4cac	461	clockevents_register_device(&__get_cpu_var(xen_clock_events));
15c84731 JF	462	}
15c84731 JF	463
d07af1f0 JF	464	void xen_timer_resume(void)
	465	{
	466	int cpu;
	467
	468	if (xen_clockevent != &xen_vcpuop_clockevent)
	469	return;
	470
	471	for_each_online_cpu(cpu) {
	472	if (HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer, cpu, NULL))
	473	BUG();
	474	}
	475	}
	476
15c84731 JF	477	__init void xen_time_init(void)
	478	{
	479	int cpu = smp_processor_id();
	480
15c84731 JF	481	clocksource_register(&xen_clocksource);
	482
	483	if (HYPERVISOR_vcpu_op(VCPUOP_stop_periodic_timer, cpu, NULL) == 0) {
f91a8b44	484	/* Successfully turned off 100Hz tick, so we have the
15c84731 JF	485	vcpuop-based timer interface */
	486	printk(KERN_DEBUG "Xen: using vcpuop timer interface\n");
	487	xen_clockevent = &xen_vcpuop_clockevent;
	488	}
	489
	490	/* Set initial system time with full resolution */
	491	xen_read_wallclock(&xtime);
	492	set_normalized_timespec(&wall_to_monotonic,
	493	-xtime.tv_sec, -xtime.tv_nsec);
	494
404ee5b1	495	setup_force_cpu_cap(X86_FEATURE_TSC);
15c84731 JF	496
15c84731 JF	497	xen_setup_timer(cpu);
f87e4cac	498	xen_setup_cpu_clockevents();
15c84731	499	}