[deliverable/linux.git] / arch / ia64 / kernel / time.c

/*
 * linux/arch/ia64/kernel/time.c
 *
 * Copyright (C) 1998-2003 Hewlett-Packard Co
 *	Stephane Eranian <eranian@hpl.hp.com>
 *	David Mosberger <davidm@hpl.hp.com>
 * Copyright (C) 1999 Don Dugger <don.dugger@intel.com>
 * Copyright (C) 1999-2000 VA Linux Systems
 * Copyright (C) 1999-2000 Walt Drummond <drummond@valinux.com>
 */

#include <linux/cpu.h>
#include <linux/init.h>
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/profile.h>
#include <linux/sched.h>
#include <linux/time.h>
#include <linux/interrupt.h>
#include <linux/efi.h>
#include <linux/timex.h>
#include <linux/timekeeper_internal.h>
#include <linux/platform_device.h>

#include <asm/machvec.h>
#include <asm/delay.h>
#include <asm/hw_irq.h>
#include <asm/paravirt.h>
#include <asm/ptrace.h>
#include <asm/sal.h>
#include <asm/sections.h>

#include "fsyscall_gtod_data.h"

static cycle_t itc_get_cycles(struct clocksource *cs);

struct fsyscall_gtod_data_t fsyscall_gtod_data;

struct itc_jitter_data_t itc_jitter_data;

volatile int time_keeper_id = 0; /* smp_processor_id() of time-keeper */

#ifdef CONFIG_IA64_DEBUG_IRQ

unsigned long last_cli_ip;
EXPORT_SYMBOL(last_cli_ip);

#endif

#ifdef CONFIG_PARAVIRT
/* We need to define a real function for sched_clock, to override the
   weak default version */
unsigned long long sched_clock(void)
{
        return paravirt_sched_clock();
}
#endif

#ifdef CONFIG_PARAVIRT
static void
paravirt_clocksource_resume(struct clocksource *cs)
{
	if (pv_time_ops.clocksource_resume)
		pv_time_ops.clocksource_resume();
}
#endif

static struct clocksource clocksource_itc = {
	.name           = "itc",
	.rating         = 350,
	.read           = itc_get_cycles,
	.mask           = CLOCKSOURCE_MASK(64),
	.flags          = CLOCK_SOURCE_IS_CONTINUOUS,
#ifdef CONFIG_PARAVIRT
	.resume		= paravirt_clocksource_resume,
#endif
};
static struct clocksource *itc_clocksource;

#ifdef CONFIG_VIRT_CPU_ACCOUNTING

#include <linux/kernel_stat.h>

extern cputime_t cycle_to_cputime(u64 cyc);

/*
 * Called from the context switch with interrupts disabled, to charge all
 * accumulated times to the current process, and to prepare accounting on
 * the next process.
 */
void ia64_account_on_switch(struct task_struct *prev, struct task_struct *next)
{
	struct thread_info *pi = task_thread_info(prev);
	struct thread_info *ni = task_thread_info(next);
	cputime_t delta_stime, delta_utime;
	__u64 now;

	now = ia64_get_itc();

	delta_stime = cycle_to_cputime(pi->ac_stime + (now - pi->ac_stamp));
	if (idle_task(smp_processor_id()) != prev)
		account_system_time(prev, 0, delta_stime, delta_stime);
	else
		account_idle_time(delta_stime);

	if (pi->ac_utime) {
		delta_utime = cycle_to_cputime(pi->ac_utime);
		account_user_time(prev, delta_utime, delta_utime);
	}

	pi->ac_stamp = ni->ac_stamp = now;
	ni->ac_stime = ni->ac_utime = 0;
}

/*
 * Account time for a transition between system, hard irq or soft irq state.
 * Note that this function is called with interrupts enabled.
 */
void account_system_vtime(struct task_struct *tsk)
{
	struct thread_info *ti = task_thread_info(tsk);
	unsigned long flags;
	cputime_t delta_stime;
	__u64 now;

	local_irq_save(flags);

	now = ia64_get_itc();

	delta_stime = cycle_to_cputime(ti->ac_stime + (now - ti->ac_stamp));
	if (irq_count() || idle_task(smp_processor_id()) != tsk)
		account_system_time(tsk, 0, delta_stime, delta_stime);
	else
		account_idle_time(delta_stime);
	ti->ac_stime = 0;

	ti->ac_stamp = now;

	local_irq_restore(flags);
}
EXPORT_SYMBOL_GPL(account_system_vtime);

/*
 * Called from the timer interrupt handler to charge accumulated user time
 * to the current process.  Must be called with interrupts disabled.
 */
void account_process_tick(struct task_struct *p, int user_tick)
{
	struct thread_info *ti = task_thread_info(p);
	cputime_t delta_utime;

	if (ti->ac_utime) {
		delta_utime = cycle_to_cputime(ti->ac_utime);
		account_user_time(p, delta_utime, delta_utime);
		ti->ac_utime = 0;
	}
}

#endif /* CONFIG_VIRT_CPU_ACCOUNTING */

static irqreturn_t
timer_interrupt (int irq, void *dev_id)
{
	unsigned long new_itm;

	if (cpu_is_offline(smp_processor_id())) {
		return IRQ_HANDLED;
	}

	platform_timer_interrupt(irq, dev_id);

	new_itm = local_cpu_data->itm_next;

	if (!time_after(ia64_get_itc(), new_itm))
		printk(KERN_ERR "Oops: timer tick before it's due (itc=%lx,itm=%lx)\n",
		       ia64_get_itc(), new_itm);

	profile_tick(CPU_PROFILING);

	if (paravirt_do_steal_accounting(&new_itm))
		goto skip_process_time_accounting;

	while (1) {
		update_process_times(user_mode(get_irq_regs()));

		new_itm += local_cpu_data->itm_delta;

		if (smp_processor_id() == time_keeper_id)
			xtime_update(1);

		local_cpu_data->itm_next = new_itm;

		if (time_after(new_itm, ia64_get_itc()))
			break;

		/*
		 * Allow IPIs to interrupt the timer loop.
		 */
		local_irq_enable();
		local_irq_disable();
	}

skip_process_time_accounting:

	do {
		/*
		 * If we're too close to the next clock tick for
		 * comfort, we increase the safety margin by
		 * intentionally dropping the next tick(s).  We do NOT
		 * update itm.next because that would force us to call
		 * xtime_update() which in turn would let our clock run
		 * too fast (with the potentially devastating effect
		 * of losing monotony of time).
		 */
		while (!time_after(new_itm, ia64_get_itc() + local_cpu_data->itm_delta/2))
			new_itm += local_cpu_data->itm_delta;
		ia64_set_itm(new_itm);
		/* double check, in case we got hit by a (slow) PMI: */
	} while (time_after_eq(ia64_get_itc(), new_itm));
	return IRQ_HANDLED;
}

/*
 * Encapsulate access to the itm structure for SMP.
 */
void
ia64_cpu_local_tick (void)
{
	int cpu = smp_processor_id();
	unsigned long shift = 0, delta;

	/* arrange for the cycle counter to generate a timer interrupt: */
	ia64_set_itv(IA64_TIMER_VECTOR);

	delta = local_cpu_data->itm_delta;
	/*
	 * Stagger the timer tick for each CPU so they don't occur all at (almost) the
	 * same time:
	 */
	if (cpu) {
		unsigned long hi = 1UL << ia64_fls(cpu);
		shift = (2*(cpu - hi) + 1) * delta/hi/2;
	}
	local_cpu_data->itm_next = ia64_get_itc() + delta + shift;
	ia64_set_itm(local_cpu_data->itm_next);
}

static int nojitter;

static int __init nojitter_setup(char *str)
{
	nojitter = 1;
	printk("Jitter checking for ITC timers disabled\n");
	return 1;
}

__setup("nojitter", nojitter_setup);


void __devinit
ia64_init_itm (void)
{
	unsigned long platform_base_freq, itc_freq;
	struct pal_freq_ratio itc_ratio, proc_ratio;
	long status, platform_base_drift, itc_drift;

	/*
	 * According to SAL v2.6, we need to use a SAL call to determine the platform base
	 * frequency and then a PAL call to determine the frequency ratio between the ITC
	 * and the base frequency.
	 */
	status = ia64_sal_freq_base(SAL_FREQ_BASE_PLATFORM,
				    &platform_base_freq, &platform_base_drift);
	if (status != 0) {
		printk(KERN_ERR "SAL_FREQ_BASE_PLATFORM failed: %s\n", ia64_sal_strerror(status));
	} else {
		status = ia64_pal_freq_ratios(&proc_ratio, NULL, &itc_ratio);
		if (status != 0)
			printk(KERN_ERR "PAL_FREQ_RATIOS failed with status=%ld\n", status);
	}
	if (status != 0) {
		/* invent "random" values */
		printk(KERN_ERR
		       "SAL/PAL failed to obtain frequency info---inventing reasonable values\n");
		platform_base_freq = 100000000;
		platform_base_drift = -1;	/* no drift info */
		itc_ratio.num = 3;
		itc_ratio.den = 1;
	}
	if (platform_base_freq < 40000000) {
		printk(KERN_ERR "Platform base frequency %lu bogus---resetting to 75MHz!\n",
		       platform_base_freq);
		platform_base_freq = 75000000;
		platform_base_drift = -1;
	}
	if (!proc_ratio.den)
		proc_ratio.den = 1;	/* avoid division by zero */
	if (!itc_ratio.den)
		itc_ratio.den = 1;	/* avoid division by zero */

	itc_freq = (platform_base_freq*itc_ratio.num)/itc_ratio.den;

	local_cpu_data->itm_delta = (itc_freq + HZ/2) / HZ;
	printk(KERN_DEBUG "CPU %d: base freq=%lu.%03luMHz, ITC ratio=%u/%u, "
	       "ITC freq=%lu.%03luMHz", smp_processor_id(),
	       platform_base_freq / 1000000, (platform_base_freq / 1000) % 1000,
	       itc_ratio.num, itc_ratio.den, itc_freq / 1000000, (itc_freq / 1000) % 1000);

	if (platform_base_drift != -1) {
		itc_drift = platform_base_drift*itc_ratio.num/itc_ratio.den;
		printk("+/-%ldppm\n", itc_drift);
	} else {
		itc_drift = -1;
		printk("\n");
	}

	local_cpu_data->proc_freq = (platform_base_freq*proc_ratio.num)/proc_ratio.den;
	local_cpu_data->itc_freq = itc_freq;
	local_cpu_data->cyc_per_usec = (itc_freq + USEC_PER_SEC/2) / USEC_PER_SEC;
	local_cpu_data->nsec_per_cyc = ((NSEC_PER_SEC<<IA64_NSEC_PER_CYC_SHIFT)
					+ itc_freq/2)/itc_freq;

	if (!(sal_platform_features & IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT)) {
#ifdef CONFIG_SMP
		/* On IA64 in an SMP configuration ITCs are never accurately synchronized.
		 * Jitter compensation requires a cmpxchg which may limit
		 * the scalability of the syscalls for retrieving time.
		 * The ITC synchronization is usually successful to within a few
		 * ITC ticks but this is not a sure thing. If you need to improve
		 * timer performance in SMP situations then boot the kernel with the
		 * "nojitter" option. However, doing so may result in time fluctuating (maybe
		 * even going backward) if the ITC offsets between the individual CPUs
		 * are too large.
		 */
		if (!nojitter)
			itc_jitter_data.itc_jitter = 1;
#endif
	} else
		/*
		 * ITC is drifty and we have not synchronized the ITCs in smpboot.c.
		 * ITC values may fluctuate significantly between processors.
		 * Clock should not be used for hrtimers. Mark itc as only
		 * useful for boot and testing.
		 *
		 * Note that jitter compensation is off! There is no point of
		 * synchronizing ITCs since they may be large differentials
		 * that change over time.
		 *
		 * The only way to fix this would be to repeatedly sync the
		 * ITCs. Until that time we have to avoid ITC.
		 */
		clocksource_itc.rating = 50;

	paravirt_init_missing_ticks_accounting(smp_processor_id());

	/* avoid softlock up message when cpu is unplug and plugged again. */
	touch_softlockup_watchdog();

	/* Setup the CPU local timer tick */
	ia64_cpu_local_tick();

	if (!itc_clocksource) {
		clocksource_register_hz(&clocksource_itc,
						local_cpu_data->itc_freq);
		itc_clocksource = &clocksource_itc;
	}
}

static cycle_t itc_get_cycles(struct clocksource *cs)
{
	unsigned long lcycle, now, ret;

	if (!itc_jitter_data.itc_jitter)
		return get_cycles();

	lcycle = itc_jitter_data.itc_lastcycle;
	now = get_cycles();
	if (lcycle && time_after(lcycle, now))
		return lcycle;

	/*
	 * Keep track of the last timer value returned.
	 * In an SMP environment, you could lose out in contention of
	 * cmpxchg. If so, your cmpxchg returns new value which the
	 * winner of contention updated to. Use the new value instead.
	 */
	ret = cmpxchg(&itc_jitter_data.itc_lastcycle, lcycle, now);
	if (unlikely(ret != lcycle))
		return ret;

	return now;
}


static struct irqaction timer_irqaction = {
	.handler =	timer_interrupt,
	.flags =	IRQF_DISABLED | IRQF_IRQPOLL,
	.name =		"timer"
};

static struct platform_device rtc_efi_dev = {
	.name = "rtc-efi",
	.id = -1,
};

static int __init rtc_init(void)
{
	if (platform_device_register(&rtc_efi_dev) < 0)
		printk(KERN_ERR "unable to register rtc device...\n");

	/* not necessarily an error */
	return 0;
}
module_init(rtc_init);

void read_persistent_clock(struct timespec *ts)
{
	efi_gettimeofday(ts);
}

void __init
time_init (void)
{
	register_percpu_irq(IA64_TIMER_VECTOR, &timer_irqaction);
	ia64_init_itm();
}

/*
 * Generic udelay assumes that if preemption is allowed and the thread
 * migrates to another CPU, that the ITC values are synchronized across
 * all CPUs.
 */
static void
ia64_itc_udelay (unsigned long usecs)
{
	unsigned long start = ia64_get_itc();
	unsigned long end = start + usecs*local_cpu_data->cyc_per_usec;

	while (time_before(ia64_get_itc(), end))
		cpu_relax();
}

void (*ia64_udelay)(unsigned long usecs) = &ia64_itc_udelay;

void
udelay (unsigned long usecs)
{
	(*ia64_udelay)(usecs);
}
EXPORT_SYMBOL(udelay);

/* IA64 doesn't cache the timezone */
void update_vsyscall_tz(void)
{
}

void update_vsyscall(struct timespec *wall, struct timespec *wtm,
			struct clocksource *c, u32 mult)
{
	write_seqcount_begin(&fsyscall_gtod_data.seq);

        /* copy fsyscall clock data */
        fsyscall_gtod_data.clk_mask = c->mask;
        fsyscall_gtod_data.clk_mult = mult;
        fsyscall_gtod_data.clk_shift = c->shift;
        fsyscall_gtod_data.clk_fsys_mmio = c->archdata.fsys_mmio;
        fsyscall_gtod_data.clk_cycle_last = c->cycle_last;

	/* copy kernel time structures */
        fsyscall_gtod_data.wall_time.tv_sec = wall->tv_sec;
        fsyscall_gtod_data.wall_time.tv_nsec = wall->tv_nsec;
	fsyscall_gtod_data.monotonic_time.tv_sec = wtm->tv_sec
							+ wall->tv_sec;
	fsyscall_gtod_data.monotonic_time.tv_nsec = wtm->tv_nsec
							+ wall->tv_nsec;

	/* normalize */
	while (fsyscall_gtod_data.monotonic_time.tv_nsec >= NSEC_PER_SEC) {
		fsyscall_gtod_data.monotonic_time.tv_nsec -= NSEC_PER_SEC;
		fsyscall_gtod_data.monotonic_time.tv_sec++;
	}

	write_seqcount_end(&fsyscall_gtod_data.seq);
}
Commit	Line	Data
1da177e4 LT	1	/*
	2	* linux/arch/ia64/kernel/time.c
	3	*
	4	* Copyright (C) 1998-2003 Hewlett-Packard Co
	5	* Stephane Eranian <eranian@hpl.hp.com>
	6	* David Mosberger <davidm@hpl.hp.com>
	7	* Copyright (C) 1999 Don Dugger <don.dugger@intel.com>
	8	* Copyright (C) 1999-2000 VA Linux Systems
	9	* Copyright (C) 1999-2000 Walt Drummond <drummond@valinux.com>
	10	*/
1da177e4 LT	11
	12	#include <linux/cpu.h>
	13	#include <linux/init.h>
	14	#include <linux/kernel.h>
	15	#include <linux/module.h>
	16	#include <linux/profile.h>
	17	#include <linux/sched.h>
	18	#include <linux/time.h>
	19	#include <linux/interrupt.h>
	20	#include <linux/efi.h>
1da177e4	21	#include <linux/timex.h>
189374ae	22	#include <linux/timekeeper_internal.h>
5e3fd9e5	23	#include <linux/platform_device.h>
1da177e4 LT	24
	25	#include <asm/machvec.h>
	26	#include <asm/delay.h>
	27	#include <asm/hw_irq.h>
00d21d82	28	#include <asm/paravirt.h>
1da177e4 LT	29	#include <asm/ptrace.h>
	30	#include <asm/sal.h>
	31	#include <asm/sections.h>
1da177e4	32
0aa366f3 TL	33	#include "fsyscall_gtod_data.h"
0aa366f3 TL	34
8e19608e	35	static cycle_t itc_get_cycles(struct clocksource *cs);
0aa366f3	36
74a622be	37	struct fsyscall_gtod_data_t fsyscall_gtod_data;
0aa366f3 TL	38
	39	struct itc_jitter_data_t itc_jitter_data;
	40
ff741906	41	volatile int time_keeper_id = 0; /* smp_processor_id() of time-keeper */
1da177e4 LT	42
	43	#ifdef CONFIG_IA64_DEBUG_IRQ
	44
	45	unsigned long last_cli_ip;
	46	EXPORT_SYMBOL(last_cli_ip);
	47
	48	#endif
	49
f927da17 IY	50	#ifdef CONFIG_PARAVIRT
	51	/* We need to define a real function for sched_clock, to override the
	52	weak default version */
	53	unsigned long long sched_clock(void)
	54	{
	55	return paravirt_sched_clock();
	56	}
	57	#endif
	58
00d21d82 IY	59	#ifdef CONFIG_PARAVIRT
00d21d82 IY	60	static void
17622339	61	paravirt_clocksource_resume(struct clocksource *cs)
00d21d82 IY	62	{
	63	if (pv_time_ops.clocksource_resume)
	64	pv_time_ops.clocksource_resume();
	65	}
	66	#endif
	67
0aa366f3	68	static struct clocksource clocksource_itc = {
3eb05676 LZ	69	.name = "itc",
	70	.rating = 350,
	71	.read = itc_get_cycles,
	72	.mask = CLOCKSOURCE_MASK(64),
3eb05676	73	.flags = CLOCK_SOURCE_IS_CONTINUOUS,
00d21d82 IY	74	#ifdef CONFIG_PARAVIRT
	75	.resume = paravirt_clocksource_resume,
	76	#endif
1da177e4	77	};
0aa366f3	78	static struct clocksource *itc_clocksource;
1da177e4	79
b64f34cd HS	80	#ifdef CONFIG_VIRT_CPU_ACCOUNTING
	81
	82	#include <linux/kernel_stat.h>
	83
	84	extern cputime_t cycle_to_cputime(u64 cyc);
	85
	86	/*
	87	* Called from the context switch with interrupts disabled, to charge all
	88	* accumulated times to the current process, and to prepare accounting on
	89	* the next process.
	90	*/
	91	void ia64_account_on_switch(struct task_struct prev, struct task_struct next)
	92	{
	93	struct thread_info *pi = task_thread_info(prev);
	94	struct thread_info *ni = task_thread_info(next);
	95	cputime_t delta_stime, delta_utime;
	96	__u64 now;
	97
	98	now = ia64_get_itc();
	99
	100	delta_stime = cycle_to_cputime(pi->ac_stime + (now - pi->ac_stamp));
79741dd3 MS	101	if (idle_task(smp_processor_id()) != prev)
	102	account_system_time(prev, 0, delta_stime, delta_stime);
	103	else
	104	account_idle_time(delta_stime);
b64f34cd HS	105
	106	if (pi->ac_utime) {
	107	delta_utime = cycle_to_cputime(pi->ac_utime);
457533a7	108	account_user_time(prev, delta_utime, delta_utime);
b64f34cd HS	109	}
	110
	111	pi->ac_stamp = ni->ac_stamp = now;
	112	ni->ac_stime = ni->ac_utime = 0;
	113	}
	114
	115	/*
	116	* Account time for a transition between system, hard irq or soft irq state.
	117	* Note that this function is called with interrupts enabled.
	118	*/
	119	void account_system_vtime(struct task_struct *tsk)
	120	{
	121	struct thread_info *ti = task_thread_info(tsk);
	122	unsigned long flags;
	123	cputime_t delta_stime;
	124	__u64 now;
	125
	126	local_irq_save(flags);
	127
	128	now = ia64_get_itc();
	129
	130	delta_stime = cycle_to_cputime(ti->ac_stime + (now - ti->ac_stamp));
79741dd3 MS	131	if (irq_count() \|\| idle_task(smp_processor_id()) != tsk)
	132	account_system_time(tsk, 0, delta_stime, delta_stime);
	133	else
	134	account_idle_time(delta_stime);
b64f34cd HS	135	ti->ac_stime = 0;
	136
	137	ti->ac_stamp = now;
	138
	139	local_irq_restore(flags);
	140	}
3a677d21	141	EXPORT_SYMBOL_GPL(account_system_vtime);
b64f34cd HS	142
	143	/*
	144	* Called from the timer interrupt handler to charge accumulated user time
	145	* to the current process. Must be called with interrupts disabled.
	146	*/
	147	void account_process_tick(struct task_struct *p, int user_tick)
	148	{
	149	struct thread_info *ti = task_thread_info(p);
	150	cputime_t delta_utime;
	151
	152	if (ti->ac_utime) {
	153	delta_utime = cycle_to_cputime(ti->ac_utime);
457533a7	154	account_user_time(p, delta_utime, delta_utime);
b64f34cd HS	155	ti->ac_utime = 0;
	156	}
	157	}
	158
	159	#endif /* CONFIG_VIRT_CPU_ACCOUNTING */
	160
1da177e4	161	static irqreturn_t
7d12e780	162	timer_interrupt (int irq, void *dev_id)
1da177e4 LT	163	{
	164	unsigned long new_itm;
	165
e7d28253	166	if (cpu_is_offline(smp_processor_id())) {
1da177e4 LT	167	return IRQ_HANDLED;
	168	}
	169
7d12e780	170	platform_timer_interrupt(irq, dev_id);
1da177e4 LT	171
	172	new_itm = local_cpu_data->itm_next;
	173
	174	if (!time_after(ia64_get_itc(), new_itm))
	175	printk(KERN_ERR "Oops: timer tick before it's due (itc=%lx,itm=%lx)\n",
	176	ia64_get_itc(), new_itm);
	177
7d12e780	178	profile_tick(CPU_PROFILING);
1da177e4	179
00d21d82 IY	180	if (paravirt_do_steal_accounting(&new_itm))
	181	goto skip_process_time_accounting;
	182
1da177e4	183	while (1) {
7d12e780	184	update_process_times(user_mode(get_irq_regs()));
1da177e4 LT	185
	186	new_itm += local_cpu_data->itm_delta;
	187
1aabd67d TH	188	if (smp_processor_id() == time_keeper_id)
	189	xtime_update(1);
	190
	191	local_cpu_data->itm_next = new_itm;
1da177e4 LT	192
	193	if (time_after(new_itm, ia64_get_itc()))
	194	break;
accaddb2 JS	195
	196	/*
	197	* Allow IPIs to interrupt the timer loop.
	198	*/
	199	local_irq_enable();
	200	local_irq_disable();
1da177e4 LT	201	}
1da177e4 LT	202
00d21d82 IY	203	skip_process_time_accounting:
00d21d82 IY	204
1da177e4 LT	205	do {
	206	/*
	207	* If we're too close to the next clock tick for
	208	* comfort, we increase the safety margin by
	209	* intentionally dropping the next tick(s). We do NOT
	210	* update itm.next because that would force us to call
1aabd67d	211	* xtime_update() which in turn would let our clock run
1da177e4 LT	212	* too fast (with the potentially devastating effect
	213	* of losing monotony of time).
	214	*/
	215	while (!time_after(new_itm, ia64_get_itc() + local_cpu_data->itm_delta/2))
	216	new_itm += local_cpu_data->itm_delta;
	217	ia64_set_itm(new_itm);
	218	/* double check, in case we got hit by a (slow) PMI: */
	219	} while (time_after_eq(ia64_get_itc(), new_itm));
	220	return IRQ_HANDLED;
	221	}
	222
	223	/*
	224	* Encapsulate access to the itm structure for SMP.
	225	*/
	226	void
	227	ia64_cpu_local_tick (void)
	228	{
	229	int cpu = smp_processor_id();
	230	unsigned long shift = 0, delta;
	231
	232	/* arrange for the cycle counter to generate a timer interrupt: */
	233	ia64_set_itv(IA64_TIMER_VECTOR);
	234
	235	delta = local_cpu_data->itm_delta;
	236	/*
	237	* Stagger the timer tick for each CPU so they don't occur all at (almost) the
	238	* same time:
	239	*/
	240	if (cpu) {
	241	unsigned long hi = 1UL << ia64_fls(cpu);
	242	shift = (2(cpu - hi) + 1) delta/hi/2;
	243	}
	244	local_cpu_data->itm_next = ia64_get_itc() + delta + shift;
	245	ia64_set_itm(local_cpu_data->itm_next);
	246	}
	247
	248	static int nojitter;
	249
	250	static int __init nojitter_setup(char *str)
	251	{
	252	nojitter = 1;
	253	printk("Jitter checking for ITC timers disabled\n");
	254	return 1;
	255	}
	256
	257	__setup("nojitter", nojitter_setup);
	258
	259
	260	void __devinit
	261	ia64_init_itm (void)
	262	{
	263	unsigned long platform_base_freq, itc_freq;
	264	struct pal_freq_ratio itc_ratio, proc_ratio;
	265	long status, platform_base_drift, itc_drift;
	266
	267	/*
	268	* According to SAL v2.6, we need to use a SAL call to determine the platform base
	269	* frequency and then a PAL call to determine the frequency ratio between the ITC
	270	* and the base frequency.
	271	*/
	272	status = ia64_sal_freq_base(SAL_FREQ_BASE_PLATFORM,
	273	&platform_base_freq, &platform_base_drift);
	274	if (status != 0) {
	275	printk(KERN_ERR "SAL_FREQ_BASE_PLATFORM failed: %s\n", ia64_sal_strerror(status));
276	} else {
277	status = ia64_pal_freq_ratios(&proc_ratio, NULL, &itc_ratio);
278	if (status != 0)
279	printk(KERN_ERR "PAL_FREQ_RATIOS failed with status=%ld\n", status);
280	}
281	if (status != 0) {
282	/* invent "random" values */
283	printk(KERN_ERR
284	"SAL/PAL failed to obtain frequency info---inventing reasonable values\n");
285	platform_base_freq = 100000000;
286	platform_base_drift = -1; /* no drift info */
287	itc_ratio.num = 3;
288	itc_ratio.den = 1;
289	}
290	if (platform_base_freq < 40000000) {
291	printk(KERN_ERR "Platform base frequency %lu bogus---resetting to 75MHz!\n",
292	platform_base_freq);
293	platform_base_freq = 75000000;
294	platform_base_drift = -1;
295	}
296	if (!proc_ratio.den)
297	proc_ratio.den = 1; /* avoid division by zero */
298	if (!itc_ratio.den)
299	itc_ratio.den = 1; /* avoid division by zero */
300
301	itc_freq = (platform_base_freq*itc_ratio.num)/itc_ratio.den;
302
303	local_cpu_data->itm_delta = (itc_freq + HZ/2) / HZ;
2ab9391d	304	printk(KERN_DEBUG "CPU %d: base freq=%lu.%03luMHz, ITC ratio=%u/%u, "
1da177e4 LT	305	"ITC freq=%lu.%03luMHz", smp_processor_id(),
	306	platform_base_freq / 1000000, (platform_base_freq / 1000) % 1000,
	307	itc_ratio.num, itc_ratio.den, itc_freq / 1000000, (itc_freq / 1000) % 1000);
	308
	309	if (platform_base_drift != -1) {
	310	itc_drift = platform_base_drift*itc_ratio.num/itc_ratio.den;
	311	printk("+/-%ldppm\n", itc_drift);
	312	} else {
	313	itc_drift = -1;
	314	printk("\n");
	315	}
	316
	317	local_cpu_data->proc_freq = (platform_base_freq*proc_ratio.num)/proc_ratio.den;
	318	local_cpu_data->itc_freq = itc_freq;
	319	local_cpu_data->cyc_per_usec = (itc_freq + USEC_PER_SEC/2) / USEC_PER_SEC;
	320	local_cpu_data->nsec_per_cyc = ((NSEC_PER_SEC<<IA64_NSEC_PER_CYC_SHIFT)
	321	+ itc_freq/2)/itc_freq;
	322
	323	if (!(sal_platform_features & IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT)) {
1da177e4 LT	324	#ifdef CONFIG_SMP
	325	/* On IA64 in an SMP configuration ITCs are never accurately synchronized.
	326	* Jitter compensation requires a cmpxchg which may limit
	327	* the scalability of the syscalls for retrieving time.
	328	* The ITC synchronization is usually successful to within a few
	329	* ITC ticks but this is not a sure thing. If you need to improve
	330	* timer performance in SMP situations then boot the kernel with the
	331	* "nojitter" option. However, doing so may result in time fluctuating (maybe
	332	* even going backward) if the ITC offsets between the individual CPUs
	333	* are too large.
	334	*/
0aa366f3 TL	335	if (!nojitter)
0aa366f3 TL	336	itc_jitter_data.itc_jitter = 1;
1da177e4	337	#endif
b718f91c CL	338	} else
	339	/*
	340	* ITC is drifty and we have not synchronized the ITCs in smpboot.c.
	341	* ITC values may fluctuate significantly between processors.
	342	* Clock should not be used for hrtimers. Mark itc as only
	343	* useful for boot and testing.
	344	*
	345	* Note that jitter compensation is off! There is no point of
	346	* synchronizing ITCs since they may be large differentials
	347	* that change over time.
	348	*
	349	* The only way to fix this would be to repeatedly sync the
	350	* ITCs. Until that time we have to avoid ITC.
	351	*/
	352	clocksource_itc.rating = 50;
1da177e4	353
00d21d82 IY	354	paravirt_init_missing_ticks_accounting(smp_processor_id());
	355
	356	/* avoid softlock up message when cpu is unplug and plugged again. */
	357	touch_softlockup_watchdog();
	358
1da177e4 LT	359	/* Setup the CPU local timer tick */
1da177e4 LT	360	ia64_cpu_local_tick();
0aa366f3 TL	361
0aa366f3 TL	362	if (!itc_clocksource) {
d60c3041 JS	363	clocksource_register_hz(&clocksource_itc,
d60c3041 JS	364	local_cpu_data->itc_freq);
0aa366f3 TL	365	itc_clocksource = &clocksource_itc;
0aa366f3 TL	366	}
1da177e4 LT	367	}
1da177e4 LT	368
8e19608e	369	static cycle_t itc_get_cycles(struct clocksource *cs)
0aa366f3	370	{
e088a4ad	371	unsigned long lcycle, now, ret;
0aa366f3 TL	372
	373	if (!itc_jitter_data.itc_jitter)
	374	return get_cycles();
	375
	376	lcycle = itc_jitter_data.itc_lastcycle;
	377	now = get_cycles();
	378	if (lcycle && time_after(lcycle, now))
	379	return lcycle;
	380
	381	/*
	382	* Keep track of the last timer value returned.
	383	* In an SMP environment, you could lose out in contention of
	384	* cmpxchg. If so, your cmpxchg returns new value which the
	385	* winner of contention updated to. Use the new value instead.
	386	*/
	387	ret = cmpxchg(&itc_jitter_data.itc_lastcycle, lcycle, now);
	388	if (unlikely(ret != lcycle))
	389	return ret;
	390
	391	return now;
	392	}
	393
	394
1da177e4 LT	395	static struct irqaction timer_irqaction = {
1da177e4 LT	396	.handler = timer_interrupt,
d217c265	397	.flags = IRQF_DISABLED \| IRQF_IRQPOLL,
1da177e4 LT	398	.name = "timer"
	399	};
	400
5e3fd9e5	401	static struct platform_device rtc_efi_dev = {
	402	.name = "rtc-efi",
	403	.id = -1,
	404	};
	405
	406	static int __init rtc_init(void)
	407	{
	408	if (platform_device_register(&rtc_efi_dev) < 0)
	409	printk(KERN_ERR "unable to register rtc device...\n");
	410
	411	/* not necessarily an error */
	412	return 0;
	413	}
	414	module_init(rtc_init);
	415
6ffdc577 JS	416	void read_persistent_clock(struct timespec *ts)
	417	{
	418	efi_gettimeofday(ts);
	419	}
	420
1da177e4 LT	421	void __init
	422	time_init (void)
	423	{
	424	register_percpu_irq(IA64_TIMER_VECTOR, &timer_irqaction);
1da177e4	425	ia64_init_itm();
1da177e4	426	}
f5899b5d	427
defbb2c9	428	/*
	429	* Generic udelay assumes that if preemption is allowed and the thread
	430	* migrates to another CPU, that the ITC values are synchronized across
	431	* all CPUs.
	432	*/
	433	static void
	434	ia64_itc_udelay (unsigned long usecs)
f5899b5d	435	{
defbb2c9	436	unsigned long start = ia64_get_itc();
defbb2c9	437	unsigned long end = start + usecs*local_cpu_data->cyc_per_usec;
f5899b5d	438
defbb2c9	439	while (time_before(ia64_get_itc(), end))
	440	cpu_relax();
	441	}
f5899b5d	442
defbb2c9	443	void (*ia64_udelay)(unsigned long usecs) = &ia64_itc_udelay;
f5899b5d	444
defbb2c9	445	void
	446	udelay (unsigned long usecs)
	447	{
	448	(*ia64_udelay)(usecs);
f5899b5d JH	449	}
f5899b5d JH	450	EXPORT_SYMBOL(udelay);
d6e56a2a	451
2c622148 TB	452	/* IA64 doesn't cache the timezone */
	453	void update_vsyscall_tz(void)
	454	{
	455	}
	456
7615856e JS	457	void update_vsyscall(struct timespec wall, struct timespec wtm,
7615856e JS	458	struct clocksource *c, u32 mult)
0aa366f3	459	{
74a622be	460	write_seqcount_begin(&fsyscall_gtod_data.seq);
0aa366f3 TL	461
	462	/* copy fsyscall clock data */
	463	fsyscall_gtod_data.clk_mask = c->mask;
0696b711	464	fsyscall_gtod_data.clk_mult = mult;
0aa366f3	465	fsyscall_gtod_data.clk_shift = c->shift;
574c44fa	466	fsyscall_gtod_data.clk_fsys_mmio = c->archdata.fsys_mmio;
0aa366f3 TL	467	fsyscall_gtod_data.clk_cycle_last = c->cycle_last;
	468
	469	/* copy kernel time structures */
	470	fsyscall_gtod_data.wall_time.tv_sec = wall->tv_sec;
	471	fsyscall_gtod_data.wall_time.tv_nsec = wall->tv_nsec;
7615856e	472	fsyscall_gtod_data.monotonic_time.tv_sec = wtm->tv_sec
0aa366f3	473	+ wall->tv_sec;
7615856e	474	fsyscall_gtod_data.monotonic_time.tv_nsec = wtm->tv_nsec
0aa366f3 TL	475	+ wall->tv_nsec;
	476
	477	/* normalize */
	478	while (fsyscall_gtod_data.monotonic_time.tv_nsec >= NSEC_PER_SEC) {
	479	fsyscall_gtod_data.monotonic_time.tv_nsec -= NSEC_PER_SEC;
	480	fsyscall_gtod_data.monotonic_time.tv_sec++;
	481	}
	482
74a622be	483	write_seqcount_end(&fsyscall_gtod_data.seq);
0aa366f3 TL	484	}
0aa366f3 TL	485