time.c

linux-2.6.15.6

/*
 * Common time routines among all ppc machines.
 *
 * Written by Cort Dougan ([e-mail]) to merge
 * Paul Mackerras' version and mine for PReP and Pmac.
 * MPC8xx/MBX changes by Dan Malek ([e-mail]).
 * Converted for 64-bit by Mike Corrigan ([e-mail])
 *
 * First round of bugfixes by Gabriel Paubert ([e-mail])
 * to make clock more stable (2.4.0-test5). The only thing
 * that this code assumes is that the timebases have been synchronized
 * by firmware on SMP and are never stopped (never do sleep
 * on SMP then, nap and doze are OK).
 * 
 * Speeded up do_gettimeofday by getting rid of references to
 * xtime (which required locks for consistency). ([e-mail])
 *
 * TODO (not necessarily in this file):
 * - improve precision and reproducibility of timebase frequency
 * measurement at boot time. (for iSeries, we calibrate the timebase
 * against the Titan chip's clock.)
 * - for astronomical applications: add a new function to get
 * non ambiguous timestamps even around leap seconds. This needs
 * a new timestamp format and a good name.
 *
 * 1997-09-10  Updated NTP code according to technical memorandum Jan '96
 *             "A Kernel Model for Precision Timekeeping" by Dave Mills
 *
 *      This program is free software; you can redistribute it and/or
 *      modify it under the terms of the GNU General Public License
 *      as published by the Free Software Foundation; either version
 *      2 of the License, or (at your option) any later version.
 */

#include <linux/config.h>
#include <linux/errno.h>
#include <linux/module.h>
#include <linux/sched.h>
#include <linux/kernel.h>
#include <linux/param.h>
#include <linux/string.h>
#include <linux/mm.h>
#include <linux/interrupt.h>
#include <linux/timex.h>
#include <linux/kernel_stat.h>
#include <linux/time.h>
#include <linux/init.h>
#include <linux/profile.h>
#include <linux/cpu.h>
#include <linux/security.h>
#include <linux/percpu.h>
#include <linux/rtc.h>

#include <asm/io.h>
#include <asm/processor.h>
#include <asm/nvram.h>
#include <asm/cache.h>
#include <asm/machdep.h>
#include <asm/uaccess.h>
#include <asm/time.h>
#include <asm/prom.h>
#include <asm/irq.h>
#include <asm/div64.h>
#include <asm/smp.h>
#include <asm/vdso_datapage.h>
#ifdef CONFIG_PPC64
#include <asm/firmware.h>
#endif
#ifdef CONFIG_PPC_ISERIES
#include <asm/iseries/it_lp_queue.h>
#include <asm/iseries/hv_call_xm.h>
#endif
#include <asm/smp.h>

/* keep track of when we need to update the rtc */
time_t last_rtc_update;
extern int piranha_simulator;
#ifdef CONFIG_PPC_ISERIES
unsigned long iSeries_recal_titan = 0;
unsigned long iSeries_recal_tb = 0; 
static unsigned long first_settimeofday = 1;
#endif

/* The decrementer counts down by 128 every 128ns on a 601. */
#define DECREMENTER_COUNT_601	(1000000000 / HZ)

#define XSEC_PER_SEC (1024*1024)

#ifdef CONFIG_PPC64
#define SCALE_XSEC(xsec, max)	(((xsec) * max) / XSEC_PER_SEC)
#else
/* compute ((xsec << 12) * max) >> 32 */
#define SCALE_XSEC(xsec, max)	mulhwu((xsec) << 12, max)
#endif

unsigned long tb_ticks_per_jiffy;
unsigned long tb_ticks_per_usec = 100; /* sane default */
EXPORT_SYMBOL(tb_ticks_per_usec);
unsigned long tb_ticks_per_sec;
u64 tb_to_xs;
unsigned tb_to_us;
unsigned long processor_freq;
DEFINE_SPINLOCK(rtc_lock);
EXPORT_SYMBOL_GPL(rtc_lock);

u64 tb_to_ns_scale;
unsigned tb_to_ns_shift;

struct gettimeofday_struct do_gtod;

extern unsigned long wall_jiffies;

extern struct timezone sys_tz;
static long timezone_offset;

void ppc_adjtimex(void);

static unsigned adjusting_time = 0;

unsigned long ppc_proc_freq;
unsigned long ppc_tb_freq;

u64 tb_last_jiffy __cacheline_aligned_in_smp;
unsigned long tb_last_stamp;

/*
 * Note that on ppc32 this only stores the bottom 32 bits of
 * the timebase value, but that's enough to tell when a jiffy
 * has passed.
 */
DEFINE_PER_CPU(unsigned long, last_jiffy);

void __delay(unsigned long loops)
{
	unsigned long start;
	int diff;

	if (__USE_RTC()) {
		start = get_rtcl();
		do {
			/* the RTCL register wraps at 1000000000 */
			diff = get_rtcl() - start;
			if (diff < 0)
				diff += 1000000000;
		} while (diff < loops);
	} else {
		start = get_tbl();
		while (get_tbl() - start < loops)
			HMT_low();
		HMT_medium();
	}
}
EXPORT_SYMBOL(__delay);

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

static __inline__ void timer_check_rtc(void)
{
        /*
         * update the rtc when needed, this should be performed on the
         * right fraction of a second. Half or full second ?
         * Full second works on mk48t59 clocks, others need testing.
         * Note that this update is basically only used through 
         * the adjtimex system calls. Setting the HW clock in
         * any other way is a /dev/rtc and userland business.
         * This is still wrong by -0.5/+1.5 jiffies because of the
         * timer interrupt resolution and possible delay, but here we 
         * hit a quantization limit which can only be solved by higher
         * resolution timers and decoupling time management from timer
         * interrupts. This is also wrong on the clocks
         * which require being written at the half second boundary.
         * We should have an rtc call that only sets the minutes and
         * seconds like on Intel to avoid problems with non UTC clocks.
         */
        if (ppc_md.set_rtc_time && ntp_synced() &&
	    xtime.tv_sec - last_rtc_update >= 659 &&
	    abs((xtime.tv_nsec/1000) - (1000000-1000000/HZ)) < 500000/HZ &&
	    jiffies - wall_jiffies == 1) {
		struct rtc_time tm;
		to_tm(xtime.tv_sec + 1 + timezone_offset, &tm);
		tm.tm_year -= 1900;
		tm.tm_mon -= 1;
		if (ppc_md.set_rtc_time(&tm) == 0)
			last_rtc_update = xtime.tv_sec + 1;
		else
			/* Try again one minute later */
			last_rtc_update += 60;
        }
}

/*
 * This version of gettimeofday has microsecond resolution.
 */
static inline void __do_gettimeofday(struct timeval *tv, u64 tb_val)
{
	unsigned long sec, usec;
	u64 tb_ticks, xsec;
	struct gettimeofday_vars *temp_varp;
	u64 temp_tb_to_xs, temp_stamp_xsec;

	/*
	 * These calculations are faster (gets rid of divides)
	 * if done in units of 1/2^20 rather than microseconds.
	 * The conversion to microseconds at the end is done
	 * without a divide (and in fact, without a multiply)
	 */
	temp_varp = do_gtod.varp;
	tb_ticks = tb_val - temp_varp->tb_orig_stamp;
	temp_tb_to_xs = temp_varp->tb_to_xs;
	temp_stamp_xsec = temp_varp->stamp_xsec;
	xsec = temp_stamp_xsec + mulhdu(tb_ticks, temp_tb_to_xs);
	sec = xsec / XSEC_PER_SEC;
	usec = (unsigned long)xsec & (XSEC_PER_SEC - 1);
	usec = SCALE_XSEC(usec, 1000000);

	tv->tv_sec = sec;
	tv->tv_usec = usec;
}

void do_gettimeofday(struct timeval *tv)
{
	if (__USE_RTC()) {
		/* do this the old way */
		unsigned long flags, seq;
		unsigned int sec, nsec, usec, lost;

		do {
			seq = read_seqbegin_irqsave(&xtime_lock, flags);
			sec = xtime.tv_sec;
			nsec = xtime.tv_nsec + tb_ticks_since(tb_last_stamp);
			lost = jiffies - wall_jiffies;
		} while (read_seqretry_irqrestore(&xtime_lock, seq, flags));
		usec = nsec / 1000 + lost * (1000000 / HZ);
		while (usec >= 1000000) {
			usec -= 1000000;
			++sec;
		}
		tv->tv_sec = sec;
		tv->tv_usec = usec;
		return;
	}
	__do_gettimeofday(tv, get_tb());
}

EXPORT_SYMBOL(do_gettimeofday);

/* Synchronize xtime with do_gettimeofday */ 

static inline void timer_sync_xtime(unsigned long cur_tb)
{
#ifdef CONFIG_PPC64
	/* why do we do this? */
	struct timeval my_tv;

	__do_gettimeofday(&my_tv, cur_tb);

	if (xtime.tv_sec <= my_tv.tv_sec) {
		xtime.tv_sec = my_tv.tv_sec;
		xtime.tv_nsec = my_tv.tv_usec * 1000;
	}
#endif
}

/*
 * There are two copies of tb_to_xs and stamp_xsec so that no
 * lock is needed to access and use these values in
 * do_gettimeofday.  We alternate the copies and as long as a
 * reasonable time elapses between changes, there will never
 * be inconsistent values.  ntpd has a minimum of one minute
 * between updates.
 */
static inline void update_gtod(u64 new_tb_stamp, u64 new_stamp_xsec,
			       u64 new_tb_to_xs)
{
	unsigned temp_idx;
	struct gettimeofday_vars *temp_varp;

	temp_idx = (do_gtod.var_idx == 0);
	temp_varp = &do_gtod.vars[temp_idx];

	temp_varp->tb_to_xs = new_tb_to_xs;
	temp_varp->tb_orig_stamp = new_tb_stamp;
	temp_varp->stamp_xsec = new_stamp_xsec;
	smp_mb();
	do_gtod.varp = temp_varp;
	do_gtod.var_idx = temp_idx;

	/*
	 * tb_update_count is used to allow the userspace gettimeofday code
	 * to assure itself that it sees a consistent view of the tb_to_xs and
	 * stamp_xsec variables.  It reads the tb_update_count, then reads
	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
	 * the two values of tb_update_count match and are even then the
	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
	 * loops back and reads them again until this criteria is met.
	 */
	++(vdso_data->tb_update_count);
	smp_wmb();
	vdso_data->tb_orig_stamp = new_tb_stamp;
	vdso_data->stamp_xsec = new_stamp_xsec;
	vdso_data->tb_to_xs = new_tb_to_xs;
	vdso_data->wtom_clock_sec = wall_to_monotonic.tv_sec;
	vdso_data->wtom_clock_nsec = wall_to_monotonic.tv_nsec;
	smp_wmb();
	++(vdso_data->tb_update_count);
}

/*
 * When the timebase - tb_orig_stamp gets too big, we do a manipulation
 * between tb_orig_stamp and stamp_xsec. The goal here is to keep the
 * difference tb - tb_orig_stamp small enough to always fit inside a
 * 32 bits number. This is a requirement of our fast 32 bits userland
 * implementation in the vdso. If we "miss" a call to this function
 * (interrupt latency, CPU locked in a spinlock, ...) and we end up
 * with a too big difference, then the vdso will fallback to calling
 * the syscall
 */
static __inline__ void timer_recalc_offset(u64 cur_tb)
{
	unsigned long offset;
	u64 new_stamp_xsec;

	if (__USE_RTC())
		return;
	offset = cur_tb - do_gtod.varp->tb_orig_stamp;
	if ((offset & 0x80000000u) == 0)
		return;
	new_stamp_xsec = do_gtod.varp->stamp_xsec
		+ mulhdu(offset, do_gtod.varp->tb_to_xs);
	update_gtod(cur_tb, new_stamp_xsec, do_gtod.varp->tb_to_xs);
}

#ifdef CONFIG_SMP
unsigned long profile_pc(struct pt_regs *regs)
{
	unsigned long pc = instruction_pointer(regs);

	if (in_lock_functions(pc))
		return regs->link;

	return pc;
}
EXPORT_SYMBOL(profile_pc);
#endif

#ifdef CONFIG_PPC_ISERIES

/* 
 * This function recalibrates the timebase based on the 49-bit time-of-day
 * value in the Titan chip.  The Titan is much more accurate than the value
 * returned by the service processor for the timebase frequency.  
 */

static void iSeries_tb_recal(void)
{
	struct div_result divres;
	unsigned long titan, tb;
	tb = get_tb();
	titan = HvCallXm_loadTod();
	if ( iSeries_recal_titan ) {
		unsigned long tb_ticks = tb - iSeries_recal_tb;
		unsigned long titan_usec = (titan - iSeries_recal_titan) >> 12;
		unsigned long new_tb_ticks_per_sec   = (tb_ticks * USEC_PER_SEC)/titan_usec;
		unsigned long new_tb_ticks_per_jiffy = (new_tb_ticks_per_sec+(HZ/2))/HZ;
		long tick_diff = new_tb_ticks_per_jiffy - tb_ticks_per_jiffy;
		char sign = '+';		
		/* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */
		new_tb_ticks_per_sec = new_tb_ticks_per_jiffy * HZ;

		if ( tick_diff < 0 ) {
			tick_diff = -tick_diff;
			sign = '-';
		}
		if ( tick_diff ) {
			if ( tick_diff < tb_ticks_per_jiffy/25 ) {
				printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n",
						new_tb_ticks_per_jiffy, sign, tick_diff );
				tb_ticks_per_jiffy = new_tb_ticks_per_jiffy;
				tb_ticks_per_sec   = new_tb_ticks_per_sec;
				div128_by_32( XSEC_PER_SEC, 0, tb_ticks_per_sec, &divres );
				do_gtod.tb_ticks_per_sec = tb_ticks_per_sec;
				tb_to_xs = divres.result_low;
				do_gtod.varp->tb_to_xs = tb_to_xs;
				vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
				vdso_data->tb_to_xs = tb_to_xs;
			}
			else {
				printk( "Titan recalibrate: FAILED (difference > 4 percent)\n"
					"                   new tb_ticks_per_jiffy = %lu\n"
					"                   old tb_ticks_per_jiffy = %lu\n",
					new_tb_ticks_per_jiffy, tb_ticks_per_jiffy );
			}
		}
	}
	iSeries_recal_titan = titan;
	iSeries_recal_tb = tb;
}
#endif

/*
 * For iSeries shared processors, we have to let the hypervisor
 * set the hardware decrementer.  We set a virtual decrementer
 * in the lppaca and call the hypervisor if the virtual
 * decrementer is less than the current value in the hardware
 * decrementer. (almost always the new decrementer value will
 * be greater than the current hardware decementer so the hypervisor
 * call will not be needed)
 */

/*
 * timer_interrupt - gets called when the decrementer overflows,
 * with interrupts disabled.
 */
void timer_interrupt(struct pt_regs * regs)
{
	int next_dec;
	int cpu = smp_processor_id();
	unsigned long ticks;

#ifdef CONFIG_PPC32
	if (atomic_read(&ppc_n_lost_interrupts) != 0)
		do_IRQ(regs);
#endif

	irq_enter();

	profile_tick(CPU_PROFILING, regs);

#ifdef CONFIG_PPC_ISERIES
	get_paca()->lppaca.int_dword.fields.decr_int = 0;
#endif

	while ((ticks = tb_ticks_since(per_cpu(last_jiffy, cpu)))
	       >= tb_ticks_per_jiffy) {
		/* Update last_jiffy */
		per_cpu(last_jiffy, cpu) += tb_ticks_per_jiffy;
		/* Handle RTCL overflow on 601 */
		if (__USE_RTC() && per_cpu(last_jiffy, cpu) >= 1000000000)
			per_cpu(last_jiffy, cpu) -= 1000000000;

		/*
		 * We cannot disable the decrementer, so in the period
		 * between this cpu's being marked offline in cpu_online_map
		 * and calling stop-self, it is taking timer interrupts.
		 * Avoid calling into the scheduler rebalancing code if this
		 * is the case.
		 */
		if (!cpu_is_offline(cpu))
			update_process_times(user_mode(regs));

		/*
		 * No need to check whether cpu is offline here; boot_cpuid
		 * should have been fixed up by now.
		 */
		if (cpu != boot_cpuid)
			continue;

		write_seqlock(&xtime_lock);
		tb_last_jiffy += tb_ticks_per_jiffy;
		tb_last_stamp = per_cpu(last_jiffy, cpu);
		timer_recalc_offset(tb_last_jiffy);
		do_timer(regs);
		timer_sync_xtime(tb_last_jiffy);
		timer_check_rtc();
		write_sequnlock(&xtime_lock);
		if (adjusting_time && (time_adjust == 0))
			ppc_adjtimex();
	}
	
	next_dec = tb_ticks_per_jiffy - ticks;
	set_dec(next_dec);

#ifdef CONFIG_PPC_ISERIES
	if (hvlpevent_is_pending())
		process_hvlpevents(regs);
#endif

#ifdef CONFIG_PPC64
	/* collect purr register values often, for accurate calculations */
	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
		cu->current_tb = mfspr(SPRN_PURR);
	}
#endif

	irq_exit();
}

void wakeup_decrementer(void)
{
	int i;

	set_dec(tb_ticks_per_jiffy);
	/*
	 * We don't expect this to be called on a machine with a 601,
	 * so using get_tbl is fine.
	 */
	tb_last_stamp = tb_last_jiffy = get_tb();
	for_each_cpu(i)
		per_cpu(last_jiffy, i) = tb_last_stamp;
}

#ifdef CONFIG_SMP
void __init smp_space_timers(unsigned int max_cpus)
{
	int i;
	unsigned long offset = tb_ticks_per_jiffy / max_cpus;
	unsigned long previous_tb = per_cpu(last_jiffy, boot_cpuid);

	/* make sure tb > per_cpu(last_jiffy, cpu) for all cpus always */
	previous_tb -= tb_ticks_per_jiffy;
	for_each_cpu(i) {