time.c

来自「底层驱动开发」· C语言 代码 · 共 882 行 · 第 1/2 页

/*
 * 
 * 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/mc146818rtc.h>
#include <linux/time.h>
#include <linux/init.h>
#include <linux/profile.h>
#include <linux/cpu.h>
#include <linux/security.h>

#include <asm/io.h>
#include <asm/processor.h>
#include <asm/nvram.h>
#include <asm/cache.h>
#include <asm/machdep.h>
#ifdef CONFIG_PPC_ISERIES
#include <asm/iSeries/ItLpQueue.h>
#include <asm/iSeries/HvCallXm.h>
#endif
#include <asm/uaccess.h>
#include <asm/time.h>
#include <asm/ppcdebug.h>
#include <asm/prom.h>
#include <asm/sections.h>
#include <asm/systemcfg.h>
#include <asm/firmware.h>

u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;

EXPORT_SYMBOL(jiffies_64);

/* 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

#define XSEC_PER_SEC (1024*1024)

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;
unsigned long tb_to_xs;
unsigned      tb_to_us;
unsigned long processor_freq;
DEFINE_SPINLOCK(rtc_lock);
EXPORT_SYMBOL_GPL(rtc_lock);

unsigned long tb_to_ns_scale;
unsigned long tb_to_ns_shift;

struct gettimeofday_struct do_gtod;

extern unsigned long wall_jiffies;
extern int smp_tb_synchronized;

extern struct timezone sys_tz;

void ppc_adjtimex(void);

static unsigned adjusting_time = 0;

unsigned long ppc_proc_freq;
unsigned long ppc_tb_freq;

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 (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, &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, unsigned long tb_val)
{
	unsigned long sec, usec, tb_ticks;
	unsigned long xsec, tb_xsec;
	struct gettimeofday_vars * temp_varp;
	unsigned long 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;
	tb_xsec = mulhdu( tb_ticks, temp_tb_to_xs );
	xsec = temp_stamp_xsec + tb_xsec;
	sec = xsec / XSEC_PER_SEC;
	xsec -= sec * XSEC_PER_SEC;
	usec = (xsec * USEC_PER_SEC)/XSEC_PER_SEC;

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

void do_gettimeofday(struct timeval *tv)
{
	__do_gettimeofday(tv, get_tb());
}

EXPORT_SYMBOL(do_gettimeofday);

/* Synchronize xtime with do_gettimeofday */ 

static inline void timer_sync_xtime(unsigned long cur_tb)
{
	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;
	}
}

/*
 * 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(unsigned long cur_tb)
{
	struct gettimeofday_vars * temp_varp;
	unsigned temp_idx;
	unsigned long offset, new_stamp_xsec, new_tb_orig_stamp;

	if (((cur_tb - do_gtod.varp->tb_orig_stamp) & 0x80000000u) == 0)
		return;

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

	new_tb_orig_stamp = cur_tb;
	offset = new_tb_orig_stamp - do_gtod.varp->tb_orig_stamp;
	new_stamp_xsec = do_gtod.varp->stamp_xsec + mulhdu(offset, do_gtod.varp->tb_to_xs);

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

	++(systemcfg->tb_update_count);
	smp_wmb();
	systemcfg->tb_orig_stamp = new_tb_orig_stamp;
	systemcfg->stamp_xsec = new_stamp_xsec;
	smp_wmb();
	++(systemcfg->tb_update_count);
}

#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;
				systemcfg->tb_ticks_per_sec = tb_ticks_per_sec;
				systemcfg->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)
 */

unsigned long tb_last_stamp __cacheline_aligned_in_smp;

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

	irq_enter();

	profile_tick(CPU_PROFILING, regs);

	lpaca->lppaca.int_dword.fields.decr_int = 0;

	while (lpaca->next_jiffy_update_tb <= (cur_tb = get_tb())) {
		/*
		 * 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) {
			write_seqlock(&xtime_lock);
			tb_last_stamp = lpaca->next_jiffy_update_tb;
			timer_recalc_offset(lpaca->next_jiffy_update_tb);
			do_timer(regs);
			timer_sync_xtime(lpaca->next_jiffy_update_tb);
			timer_check_rtc();
			write_sequnlock(&xtime_lock);
			if ( adjusting_time && (time_adjust == 0) )
				ppc_adjtimex();
		}
		lpaca->next_jiffy_update_tb += tb_ticks_per_jiffy;
	}
	
	next_dec = lpaca->next_jiffy_update_tb - cur_tb;
	if (next_dec > lpaca->default_decr)
        	next_dec = lpaca->default_decr;
	set_dec(next_dec);

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

	/* 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);
	}

	irq_exit();

	return 1;
}

/*
 * Scheduler clock - returns current time in nanosec units.
 *
 * Note: mulhdu(a, b) (multiply high double unsigned) returns
 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
 * are 64-bit unsigned numbers.
 */
unsigned long long sched_clock(void)
{
	return mulhdu(get_tb(), tb_to_ns_scale) << tb_to_ns_shift;
}

int do_settimeofday(struct timespec *tv)
{
	time_t wtm_sec, new_sec = tv->tv_sec;
	long wtm_nsec, new_nsec = tv->tv_nsec;
	unsigned long flags;
	unsigned long delta_xsec;
	long int tb_delta;
	unsigned long new_xsec;

	if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
		return -EINVAL;

	write_seqlock_irqsave(&xtime_lock, flags);
	/* Updating the RTC is not the job of this code. If the time is
	 * stepped under NTP, the RTC will be update after STA_UNSYNC
	 * is cleared. Tool like clock/hwclock either copy the RTC
	 * to the system time, in which case there is no point in writing
	 * to the RTC again, or write to the RTC but then they don't call
	 * settimeofday to perform this operation.
	 */
#ifdef CONFIG_PPC_ISERIES
	if ( first_settimeofday ) {
		iSeries_tb_recal();
		first_settimeofday = 0;
	}
#endif
	tb_delta = tb_ticks_since(tb_last_stamp);
	tb_delta += (jiffies - wall_jiffies) * tb_ticks_per_jiffy;

	new_nsec -= tb_delta / tb_ticks_per_usec / 1000;

	wtm_sec  = wall_to_monotonic.tv_sec + (xtime.tv_sec - new_sec);
	wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - new_nsec);

 	set_normalized_timespec(&xtime, new_sec, new_nsec);
	set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);

	/* In case of a large backwards jump in time with NTP, we want the 
	 * clock to be updated as soon as the PLL is again in lock.
	 */
	last_rtc_update = new_sec - 658;

	ntp_clear();

	delta_xsec = mulhdu( (tb_last_stamp-do_gtod.varp->tb_orig_stamp),
			     do_gtod.varp->tb_to_xs );