time.c 9.0 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34
/*
 *  linux/arch/parisc/kernel/time.c
 *
 *  Copyright (C) 1991, 1992, 1995  Linus Torvalds
 *  Modifications for ARM (C) 1994, 1995, 1996,1997 Russell King
 *  Copyright (C) 1999 SuSE GmbH, (Philipp Rumpf, prumpf@tux.org)
 *
 * 1994-07-02  Alan Modra
 *             fixed set_rtc_mmss, fixed time.year for >= 2000, new mktime
 * 1998-12-20  Updated NTP code according to technical memorandum Jan '96
 *             "A Kernel Model for Precision Timekeeping" by Dave Mills
 */
#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/time.h>
#include <linux/init.h>
#include <linux/smp.h>
#include <linux/profile.h>

#include <asm/uaccess.h>
#include <asm/io.h>
#include <asm/irq.h>
#include <asm/param.h>
#include <asm/pdc.h>
#include <asm/led.h>

#include <linux/timex.h>

35
static unsigned long clocktick __read_mostly;	/* timer cycles per tick */
L
Linus Torvalds 已提交
36 37 38 39 40

#ifdef CONFIG_SMP
extern void smp_do_timer(struct pt_regs *regs);
#endif

41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57
/*
 * We keep time on PA-RISC Linux by using the Interval Timer which is
 * a pair of registers; one is read-only and one is write-only; both
 * accessed through CR16.  The read-only register is 32 or 64 bits wide,
 * and increments by 1 every CPU clock tick.  The architecture only
 * guarantees us a rate between 0.5 and 2, but all implementations use a
 * rate of 1.  The write-only register is 32-bits wide.  When the lowest
 * 32 bits of the read-only register compare equal to the write-only
 * register, it raises a maskable external interrupt.  Each processor has
 * an Interval Timer of its own and they are not synchronised.  
 *
 * We want to generate an interrupt every 1/HZ seconds.  So we program
 * CR16 to interrupt every @clocktick cycles.  The it_value in cpu_data
 * is programmed with the intended time of the next tick.  We can be
 * held off for an arbitrarily long period of time by interrupts being
 * disabled, so we may miss one or more ticks.
 */
L
Linus Torvalds 已提交
58 59
irqreturn_t timer_interrupt(int irq, void *dev_id, struct pt_regs *regs)
{
60 61
	unsigned long now;
	unsigned long next_tick;
62
	unsigned long cycles_elapsed, ticks_elapsed;
63 64
	unsigned long cycles_remainder;
	unsigned int cpu = smp_processor_id();
L
Linus Torvalds 已提交
65

66
	/* gcc can optimize for "read-only" case with a local clocktick */
67
	unsigned long cpt = clocktick;
68

L
Linus Torvalds 已提交
69 70
	profile_tick(CPU_PROFILING, regs);

71
	/* Initialize next_tick to the expected tick time. */
L
Linus Torvalds 已提交
72 73
	next_tick = cpu_data[cpu].it_value;

74 75 76
	/* Get current interval timer.
	 * CR16 reads as 64 bits in CPU wide mode.
	 * CR16 reads as 32 bits in CPU narrow mode.
L
Linus Torvalds 已提交
77
	 */
78
	now = mfctl(16);
L
Linus Torvalds 已提交
79

80 81
	cycles_elapsed = now - next_tick;

82 83 84
	if ((cycles_elapsed >> 5) < cpt) {
		/* use "cheap" math (add/subtract) instead
		 * of the more expensive div/mul method
85
		 */
86
		cycles_remainder = cycles_elapsed;
87
		ticks_elapsed = 1;
88 89
		while (cycles_remainder > cpt) {
			cycles_remainder -= cpt;
90
			ticks_elapsed++;
91
		}
92
	} else {
93
		cycles_remainder = cycles_elapsed % cpt;
94
		ticks_elapsed = 1 + cycles_elapsed / cpt;
95
	}
96 97 98 99 100 101 102 103

	/* Can we differentiate between "early CR16" (aka Scenario 1) and
	 * "long delay" (aka Scenario 3)? I don't think so.
	 *
	 * We expected timer_interrupt to be delivered at least a few hundred
	 * cycles after the IT fires. But it's arbitrary how much time passes
	 * before we call it "late". I've picked one second.
	 */
104
	if (ticks_elapsed > HZ) {
105
		/* Scenario 3: very long delay?  bad in any case */
106
		printk (KERN_CRIT "timer_interrupt(CPU %d): delayed!"
107
			" cycles %lX rem %lX "
108 109
			" next/now %lX/%lX\n",
			cpu,
110
			cycles_elapsed, cycles_remainder,
111 112 113
			next_tick, now );
	}

114 115
	/* convert from "division remainder" to "remainder of clock tick" */
	cycles_remainder = cpt - cycles_remainder;
116 117 118 119 120

	/* Determine when (in CR16 cycles) next IT interrupt will fire.
	 * We want IT to fire modulo clocktick even if we miss/skip some.
	 * But those interrupts don't in fact get delivered that regularly.
	 */
121 122 123
	next_tick = now + cycles_remainder;

	cpu_data[cpu].it_value = next_tick;
124 125

	/* Skip one clocktick on purpose if we are likely to miss next_tick.
126 127 128 129 130 131
	 * We want to avoid the new next_tick being less than CR16.
	 * If that happened, itimer wouldn't fire until CR16 wrapped.
	 * We'll catch the tick we missed on the tick after that.
	 */
	if (!(cycles_remainder >> 13))
		next_tick += cpt;
132 133 134

	/* Program the IT when to deliver the next interrupt. */
        /* Only bottom 32-bits of next_tick are written to cr16.  */
135
	mtctl(next_tick, 16);
L
Linus Torvalds 已提交
136

137 138 139

	/* Done mucking with unreliable delivery of interrupts.
	 * Go do system house keeping.
140
	 */
L
Linus Torvalds 已提交
141
#ifdef CONFIG_SMP
142
	smp_do_timer(regs);
L
Linus Torvalds 已提交
143
#else
144
	update_process_times(user_mode(regs));
L
Linus Torvalds 已提交
145
#endif
146 147
	if (cpu == 0) {
		write_seqlock(&xtime_lock);
148
		do_timer(ticks_elapsed);
149
		write_sequnlock(&xtime_lock);
L
Linus Torvalds 已提交
150
	}
151

L
Linus Torvalds 已提交
152 153 154 155 156 157 158
	/* check soft power switch status */
	if (cpu == 0 && !atomic_read(&power_tasklet.count))
		tasklet_schedule(&power_tasklet);

	return IRQ_HANDLED;
}

159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176

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

	if (regs->gr[0] & PSW_N)
		pc -= 4;

#ifdef CONFIG_SMP
	if (in_lock_functions(pc))
		pc = regs->gr[2];
#endif

	return pc;
}
EXPORT_SYMBOL(profile_pc);


L
Linus Torvalds 已提交
177 178
/*
 * Return the number of micro-seconds that elapsed since the last
A
Atsushi Nemoto 已提交
179
 * update to wall time (aka xtime).  The xtime_lock
L
Linus Torvalds 已提交
180 181
 * must be at least read-locked when calling this routine.
 */
182
static inline unsigned long gettimeoffset (void)
L
Linus Torvalds 已提交
183 184 185 186 187 188 189
{
#ifndef CONFIG_SMP
	/*
	 * FIXME: This won't work on smp because jiffies are updated by cpu 0.
	 *    Once parisc-linux learns the cr16 difference between processors,
	 *    this could be made to work.
	 */
190 191 192 193 194
	unsigned long now;
	unsigned long prev_tick;
	unsigned long next_tick;
	unsigned long elapsed_cycles;
	unsigned long usec;
195
	unsigned long cpuid = smp_processor_id();
196
	unsigned long cpt = clocktick;
L
Linus Torvalds 已提交
197

198
	next_tick = cpu_data[cpuid].it_value;
199
	now = mfctl(16);	/* Read the hardware interval timer.  */
L
Linus Torvalds 已提交
200

201
	prev_tick = next_tick - cpt;
202 203 204 205

	/* Assume Scenario 1: "now" is later than prev_tick.  */
	elapsed_cycles = now - prev_tick;

206 207 208 209 210 211 212 213 214 215 216 217
/* aproximate HZ with shifts. Intended math is "(elapsed/clocktick) > HZ" */
#if HZ == 1000
	if (elapsed_cycles > (cpt << 10) )
#elif HZ == 250
	if (elapsed_cycles > (cpt << 8) )
#elif HZ == 100
	if (elapsed_cycles > (cpt << 7) )
#else
#warn WTF is HZ set to anyway?
	if (elapsed_cycles > (HZ * cpt) )
#endif
	{
218
		/* Scenario 3: clock ticks are missing. */
219 220 221 222
		printk (KERN_CRIT "gettimeoffset(CPU %ld): missing %ld ticks!"
			" cycles %lX prev/now/next %lX/%lX/%lX  clock %lX\n",
			cpuid, elapsed_cycles / cpt,
			elapsed_cycles, prev_tick, now, next_tick, cpt);
223 224 225 226 227
	}

	/* FIXME: Can we improve the precision? Not with PAGE0. */
	usec = (elapsed_cycles * 10000) / PAGE0->mem_10msec;
	return usec;
L
Linus Torvalds 已提交
228 229 230 231 232 233 234 235 236 237
#else
	return 0;
#endif
}

void
do_gettimeofday (struct timeval *tv)
{
	unsigned long flags, seq, usec, sec;

238
	/* Hold xtime_lock and adjust timeval.  */
L
Linus Torvalds 已提交
239 240 241 242 243 244 245
	do {
		seq = read_seqbegin_irqsave(&xtime_lock, flags);
		usec = gettimeoffset();
		sec = xtime.tv_sec;
		usec += (xtime.tv_nsec / 1000);
	} while (read_seqretry_irqrestore(&xtime_lock, seq, flags));

246
	/* Move adjusted usec's into sec's.  */
247 248
	while (usec >= USEC_PER_SEC) {
		usec -= USEC_PER_SEC;
L
Linus Torvalds 已提交
249 250 251
		++sec;
	}

252
	/* Return adjusted result.  */
L
Linus Torvalds 已提交
253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284
	tv->tv_sec = sec;
	tv->tv_usec = usec;
}

EXPORT_SYMBOL(do_gettimeofday);

int
do_settimeofday (struct timespec *tv)
{
	time_t wtm_sec, sec = tv->tv_sec;
	long wtm_nsec, nsec = tv->tv_nsec;

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

	write_seqlock_irq(&xtime_lock);
	{
		/*
		 * This is revolting. We need to set "xtime"
		 * correctly. However, the value in this location is
		 * the value at the most recent update of wall time.
		 * Discover what correction gettimeofday would have
		 * done, and then undo it!
		 */
		nsec -= gettimeoffset() * 1000;

		wtm_sec  = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
		wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);

		set_normalized_timespec(&xtime, sec, nsec);
		set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);

J
john stultz 已提交
285
		ntp_clear();
L
Linus Torvalds 已提交
286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303
	}
	write_sequnlock_irq(&xtime_lock);
	clock_was_set();
	return 0;
}
EXPORT_SYMBOL(do_settimeofday);

/*
 * XXX: We can do better than this.
 * Returns nanoseconds
 */

unsigned long long sched_clock(void)
{
	return (unsigned long long)jiffies * (1000000000 / HZ);
}


304 305 306 307 308 309 310 311 312 313
void __init start_cpu_itimer(void)
{
	unsigned int cpu = smp_processor_id();
	unsigned long next_tick = mfctl(16) + clocktick;

	mtctl(next_tick, 16);		/* kick off Interval Timer (CR16) */

	cpu_data[cpu].it_value = next_tick;
}

L
Linus Torvalds 已提交
314 315 316 317 318 319
void __init time_init(void)
{
	static struct pdc_tod tod_data;

	clocktick = (100 * PAGE0->mem_10msec) / HZ;

320
	start_cpu_itimer();	/* get CPU 0 started */
L
Linus Torvalds 已提交
321 322 323 324 325 326 327 328 329 330 331 332 333 334 335

	if(pdc_tod_read(&tod_data) == 0) {
		write_seqlock_irq(&xtime_lock);
		xtime.tv_sec = tod_data.tod_sec;
		xtime.tv_nsec = tod_data.tod_usec * 1000;
		set_normalized_timespec(&wall_to_monotonic,
		                        -xtime.tv_sec, -xtime.tv_nsec);
		write_sequnlock_irq(&xtime_lock);
	} else {
		printk(KERN_ERR "Error reading tod clock\n");
	        xtime.tv_sec = 0;
		xtime.tv_nsec = 0;
	}
}