[deliverable/linux.git] / arch / i386 / kernel / timers / timer_hpet.c

/*
 * This code largely moved from arch/i386/kernel/time.c.
 * See comments there for proper credits.
 */

#include <linux/spinlock.h>
#include <linux/init.h>
#include <linux/timex.h>
#include <linux/errno.h>
#include <linux/string.h>
#include <linux/jiffies.h>

#include <asm/timer.h>
#include <asm/io.h>
#include <asm/processor.h>

#include "io_ports.h"
#include "mach_timer.h"
#include <asm/hpet.h>

static unsigned long __read_mostly hpet_usec_quotient;	/* convert hpet clks to usec */
static unsigned long tsc_hpet_quotient;		/* convert tsc to hpet clks */
static unsigned long hpet_last; 	/* hpet counter value at last tick*/
static unsigned long last_tsc_low;	/* lsb 32 bits of Time Stamp Counter */
static unsigned long last_tsc_high; 	/* msb 32 bits of Time Stamp Counter */
static unsigned long long monotonic_base;
static seqlock_t monotonic_lock = SEQLOCK_UNLOCKED;

/* convert from cycles(64bits) => nanoseconds (64bits)
 *  basic equation:
 *		ns = cycles / (freq / ns_per_sec)
 *		ns = cycles * (ns_per_sec / freq)
 *		ns = cycles * (10^9 / (cpu_mhz * 10^6))
 *		ns = cycles * (10^3 / cpu_mhz)
 *
 *	Then we use scaling math (suggested by george@mvista.com) to get:
 *		ns = cycles * (10^3 * SC / cpu_mhz) / SC
 *		ns = cycles * cyc2ns_scale / SC
 *
 *	And since SC is a constant power of two, we can convert the div
 *  into a shift.
 *			-johnstul@us.ibm.com "math is hard, lets go shopping!"
 */
static unsigned long cyc2ns_scale;
#define CYC2NS_SCALE_FACTOR 10 /* 2^10, carefully chosen */

static inline void set_cyc2ns_scale(unsigned long cpu_mhz)
{
	cyc2ns_scale = (1000 << CYC2NS_SCALE_FACTOR)/cpu_mhz;
}

static inline unsigned long long cycles_2_ns(unsigned long long cyc)
{
	return (cyc * cyc2ns_scale) >> CYC2NS_SCALE_FACTOR;
}

static unsigned long long monotonic_clock_hpet(void)
{
	unsigned long long last_offset, this_offset, base;
	unsigned seq;

	/* atomically read monotonic base & last_offset */
	do {
		seq = read_seqbegin(&monotonic_lock);
		last_offset = ((unsigned long long)last_tsc_high<<32)|last_tsc_low;
		base = monotonic_base;
	} while (read_seqretry(&monotonic_lock, seq));

	/* Read the Time Stamp Counter */
	rdtscll(this_offset);

	/* return the value in ns */
	return base + cycles_2_ns(this_offset - last_offset);
}

static unsigned long get_offset_hpet(void)
{
	register unsigned long eax, edx;

	eax = hpet_readl(HPET_COUNTER);
	eax -= hpet_last;	/* hpet delta */
	eax = min(hpet_tick, eax);
	/*
         * Time offset = (hpet delta) * ( usecs per HPET clock )
	 *             = (hpet delta) * ( usecs per tick / HPET clocks per tick)
	 *             = (hpet delta) * ( hpet_usec_quotient ) / (2^32)
	 *
	 * Where,
	 * hpet_usec_quotient = (2^32 * usecs per tick)/HPET clocks per tick
	 *
	 * Using a mull instead of a divl saves some cycles in critical path.
         */
	ASM_MUL64_REG(eax, edx, hpet_usec_quotient, eax);

	/* our adjusted time offset in microseconds */
	return edx;
}

static void mark_offset_hpet(void)
{
	unsigned long long this_offset, last_offset;
	unsigned long offset;

	write_seqlock(&monotonic_lock);
	last_offset = ((unsigned long long)last_tsc_high<<32)|last_tsc_low;
	rdtsc(last_tsc_low, last_tsc_high);

	if (hpet_use_timer)
		offset = hpet_readl(HPET_T0_CMP) - hpet_tick;
	else
		offset = hpet_readl(HPET_COUNTER);
	if (unlikely(((offset - hpet_last) >= (2*hpet_tick)) && (hpet_last != 0))) {
		int lost_ticks = ((offset - hpet_last) / hpet_tick) - 1;
		jiffies_64 += lost_ticks;
	}
	hpet_last = offset;

	/* update the monotonic base value */
	this_offset = ((unsigned long long)last_tsc_high<<32)|last_tsc_low;
	monotonic_base += cycles_2_ns(this_offset - last_offset);
	write_sequnlock(&monotonic_lock);
}

static void delay_hpet(unsigned long loops)
{
	unsigned long hpet_start, hpet_end;
	unsigned long eax;

	/* loops is the number of cpu cycles. Convert it to hpet clocks */
	ASM_MUL64_REG(eax, loops, tsc_hpet_quotient, loops);

	hpet_start = hpet_readl(HPET_COUNTER);
	do {
		rep_nop();
		hpet_end = hpet_readl(HPET_COUNTER);
	} while ((hpet_end - hpet_start) < (loops));
}

static struct timer_opts timer_hpet;

static int __init init_hpet(char* override)
{
	unsigned long result, remain;

	/* check clock override */
	if (override[0] && strncmp(override,"hpet",4))
		return -ENODEV;

	if (!is_hpet_enabled())
		return -ENODEV;

	printk("Using HPET for gettimeofday\n");
	if (cpu_has_tsc) {
		unsigned long tsc_quotient = calibrate_tsc_hpet(&tsc_hpet_quotient);
		if (tsc_quotient) {
			/* report CPU clock rate in Hz.
			 * The formula is (10^6 * 2^32) / (2^32 * 1 / (clocks/us)) =
			 * clock/second. Our precision is about 100 ppm.
			 */
			{	unsigned long eax=0, edx=1000;
				ASM_DIV64_REG(cpu_khz, edx, tsc_quotient,
						eax, edx);
				printk("Detected %u.%03u MHz processor.\n",
					cpu_khz / 1000, cpu_khz % 1000);
			}
			set_cyc2ns_scale(cpu_khz/1000);
		}
		/* set this only when cpu_has_tsc */
		timer_hpet.read_timer = read_timer_tsc;
	}

	/*
	 * Math to calculate hpet to usec multiplier
	 * Look for the comments at get_offset_hpet()
	 */
	ASM_DIV64_REG(result, remain, hpet_tick, 0, KERNEL_TICK_USEC);
	if (remain > (hpet_tick >> 1))
		result++; /* rounding the result */
	hpet_usec_quotient = result;

	return 0;
}

static int hpet_resume(void)
{
	write_seqlock(&monotonic_lock);
	/* Assume this is the last mark offset time */
	rdtsc(last_tsc_low, last_tsc_high);

	if (hpet_use_timer)
		hpet_last = hpet_readl(HPET_T0_CMP) - hpet_tick;
	else
		hpet_last = hpet_readl(HPET_COUNTER);
	write_sequnlock(&monotonic_lock);
	return 0;
}
/************************************************************/

/* tsc timer_opts struct */
static struct timer_opts timer_hpet __read_mostly = {
	.name = 		"hpet",
	.mark_offset =		mark_offset_hpet,
	.get_offset =		get_offset_hpet,
	.monotonic_clock =	monotonic_clock_hpet,
	.delay = 		delay_hpet,
	.resume	=		hpet_resume,
};

struct init_timer_opts __initdata timer_hpet_init = {
	.init =	init_hpet,
	.opts = &timer_hpet,
};
Commit	Line	Data
1da177e4 LT	1	/*
	2	* This code largely moved from arch/i386/kernel/time.c.
	3	* See comments there for proper credits.
	4	*/
	5
	6	#include <linux/spinlock.h>
	7	#include <linux/init.h>
	8	#include <linux/timex.h>
	9	#include <linux/errno.h>
	10	#include <linux/string.h>
	11	#include <linux/jiffies.h>
	12
	13	#include <asm/timer.h>
	14	#include <asm/io.h>
	15	#include <asm/processor.h>
	16
	17	#include "io_ports.h"
	18	#include "mach_timer.h"
	19	#include <asm/hpet.h>
	20
6c036527	21	static unsigned long __read_mostly hpet_usec_quotient; /* convert hpet clks to usec */
1da177e4 LT	22	static unsigned long tsc_hpet_quotient; /* convert tsc to hpet clks */
	23	static unsigned long hpet_last; /* hpet counter value at last tick*/
	24	static unsigned long last_tsc_low; /* lsb 32 bits of Time Stamp Counter */
	25	static unsigned long last_tsc_high; /* msb 32 bits of Time Stamp Counter */
	26	static unsigned long long monotonic_base;
	27	static seqlock_t monotonic_lock = SEQLOCK_UNLOCKED;
	28
	29	/* convert from cycles(64bits) => nanoseconds (64bits)
	30	* basic equation:
	31	* ns = cycles / (freq / ns_per_sec)
	32	* ns = cycles * (ns_per_sec / freq)
	33	* ns = cycles * (10^9 / (cpu_mhz * 10^6))
	34	* ns = cycles * (10^3 / cpu_mhz)
	35	*
	36	* Then we use scaling math (suggested by george@mvista.com) to get:
	37	* ns = cycles * (10^3 * SC / cpu_mhz) / SC
	38	* ns = cycles * cyc2ns_scale / SC
	39	*
	40	* And since SC is a constant power of two, we can convert the div
	41	* into a shift.
	42	* -johnstul@us.ibm.com "math is hard, lets go shopping!"
	43	*/
	44	static unsigned long cyc2ns_scale;
	45	#define CYC2NS_SCALE_FACTOR 10 /* 2^10, carefully chosen */
	46
	47	static inline void set_cyc2ns_scale(unsigned long cpu_mhz)
	48	{
	49	cyc2ns_scale = (1000 << CYC2NS_SCALE_FACTOR)/cpu_mhz;
	50	}
	51
	52	static inline unsigned long long cycles_2_ns(unsigned long long cyc)
	53	{
	54	return (cyc * cyc2ns_scale) >> CYC2NS_SCALE_FACTOR;
	55	}
	56
	57	static unsigned long long monotonic_clock_hpet(void)
	58	{
	59	unsigned long long last_offset, this_offset, base;
	60	unsigned seq;
	61
	62	/* atomically read monotonic base & last_offset */
	63	do {
	64	seq = read_seqbegin(&monotonic_lock);
	65	last_offset = ((unsigned long long)last_tsc_high<<32)\|last_tsc_low;
	66	base = monotonic_base;
	67	} while (read_seqretry(&monotonic_lock, seq));
	68
	69	/* Read the Time Stamp Counter */
	70	rdtscll(this_offset);
	71
	72	/* return the value in ns */
	73	return base + cycles_2_ns(this_offset - last_offset);
	74	}
	75
	76	static unsigned long get_offset_hpet(void)
	77	{
	78	register unsigned long eax, edx;
	79
	80	eax = hpet_readl(HPET_COUNTER);
	81	eax -= hpet_last; /* hpet delta */
35492df5	82	eax = min(hpet_tick, eax);
1da177e4 LT	83	/*
	84	* Time offset = (hpet delta) * ( usecs per HPET clock )
	85	* = (hpet delta) * ( usecs per tick / HPET clocks per tick)
	86	* = (hpet delta) * ( hpet_usec_quotient ) / (2^32)
	87	*
	88	* Where,
	89	* hpet_usec_quotient = (2^32 * usecs per tick)/HPET clocks per tick
	90	*
	91	* Using a mull instead of a divl saves some cycles in critical path.
	92	*/
	93	ASM_MUL64_REG(eax, edx, hpet_usec_quotient, eax);
	94
	95	/* our adjusted time offset in microseconds */
	96	return edx;
	97	}
	98
	99	static void mark_offset_hpet(void)
	100	{
	101	unsigned long long this_offset, last_offset;
	102	unsigned long offset;
	103
	104	write_seqlock(&monotonic_lock);
	105	last_offset = ((unsigned long long)last_tsc_high<<32)\|last_tsc_low;
	106	rdtsc(last_tsc_low, last_tsc_high);
	107
35492df5	108	if (hpet_use_timer)
	109	offset = hpet_readl(HPET_T0_CMP) - hpet_tick;
	110	else
	111	offset = hpet_readl(HPET_COUNTER);
	112	if (unlikely(((offset - hpet_last) >= (2*hpet_tick)) && (hpet_last != 0))) {
	113	int lost_ticks = ((offset - hpet_last) / hpet_tick) - 1;
1da177e4 LT	114	jiffies_64 += lost_ticks;
	115	}
	116	hpet_last = offset;
	117
	118	/* update the monotonic base value */
	119	this_offset = ((unsigned long long)last_tsc_high<<32)\|last_tsc_low;
	120	monotonic_base += cycles_2_ns(this_offset - last_offset);
	121	write_sequnlock(&monotonic_lock);
	122	}
	123
	124	static void delay_hpet(unsigned long loops)
	125	{
	126	unsigned long hpet_start, hpet_end;
	127	unsigned long eax;
	128
	129	/* loops is the number of cpu cycles. Convert it to hpet clocks */
	130	ASM_MUL64_REG(eax, loops, tsc_hpet_quotient, loops);
	131
	132	hpet_start = hpet_readl(HPET_COUNTER);
	133	do {
	134	rep_nop();
	135	hpet_end = hpet_readl(HPET_COUNTER);
	136	} while ((hpet_end - hpet_start) < (loops));
	137	}
	138
4116c527 VP	139	static struct timer_opts timer_hpet;
4116c527 VP	140
1da177e4 LT	141	static int __init init_hpet(char* override)
	142	{
	143	unsigned long result, remain;
	144
	145	/* check clock override */
	146	if (override[0] && strncmp(override,"hpet",4))
	147	return -ENODEV;
	148
	149	if (!is_hpet_enabled())
	150	return -ENODEV;
	151
	152	printk("Using HPET for gettimeofday\n");
	153	if (cpu_has_tsc) {
	154	unsigned long tsc_quotient = calibrate_tsc_hpet(&tsc_hpet_quotient);
	155	if (tsc_quotient) {
	156	/* report CPU clock rate in Hz.
	157	* The formula is (10^6 * 2^32) / (2^32 * 1 / (clocks/us)) =
	158	* clock/second. Our precision is about 100 ppm.
	159	*/
	160	{ unsigned long eax=0, edx=1000;
	161	ASM_DIV64_REG(cpu_khz, edx, tsc_quotient,
	162	eax, edx);
a3a255e7	163	printk("Detected %u.%03u MHz processor.\n",
1da177e4 LT	164	cpu_khz / 1000, cpu_khz % 1000);
	165	}
	166	set_cyc2ns_scale(cpu_khz/1000);
	167	}
4116c527 VP	168	/* set this only when cpu_has_tsc */
4116c527 VP	169	timer_hpet.read_timer = read_timer_tsc;
1da177e4 LT	170	}
	171
	172	/*
	173	* Math to calculate hpet to usec multiplier
	174	* Look for the comments at get_offset_hpet()
	175	*/
	176	ASM_DIV64_REG(result, remain, hpet_tick, 0, KERNEL_TICK_USEC);
	177	if (remain > (hpet_tick >> 1))
	178	result++; /* rounding the result */
	179	hpet_usec_quotient = result;
	180
	181	return 0;
	182	}
	183
c3c433e4 SL	184	static int hpet_resume(void)
	185	{
	186	write_seqlock(&monotonic_lock);
	187	/* Assume this is the last mark offset time */
	188	rdtsc(last_tsc_low, last_tsc_high);
	189
	190	if (hpet_use_timer)
	191	hpet_last = hpet_readl(HPET_T0_CMP) - hpet_tick;
	192	else
	193	hpet_last = hpet_readl(HPET_COUNTER);
	194	write_sequnlock(&monotonic_lock);
	195	return 0;
	196	}
1da177e4 LT	197	/************************************************************/
	198
	199	/* tsc timer_opts struct */
6c036527	200	static struct timer_opts timer_hpet __read_mostly = {
1da177e4 LT	201	.name = "hpet",
	202	.mark_offset = mark_offset_hpet,
	203	.get_offset = get_offset_hpet,
	204	.monotonic_clock = monotonic_clock_hpet,
	205	.delay = delay_hpet,
c3c433e4	206	.resume = hpet_resume,
1da177e4 LT	207	};
	208
	209	struct init_timer_opts __initdata timer_hpet_init = {
	210	.init = init_hpet,
	211	.opts = &timer_hpet,
	212	};