[linux-2.6-block.git] / arch / x86 / kernel / tsc.c

#include <linux/kernel.h>
#include <linux/sched.h>
#include <linux/init.h>
#include <linux/module.h>
#include <linux/timer.h>
#include <linux/acpi_pmtmr.h>
#include <linux/cpufreq.h>
#include <linux/dmi.h>
#include <linux/delay.h>
#include <linux/clocksource.h>
#include <linux/percpu.h>

#include <asm/hpet.h>
#include <asm/timer.h>
#include <asm/vgtod.h>
#include <asm/time.h>
#include <asm/delay.h>

unsigned int cpu_khz;           /* TSC clocks / usec, not used here */
EXPORT_SYMBOL(cpu_khz);
unsigned int tsc_khz;
EXPORT_SYMBOL(tsc_khz);

/*
 * TSC can be unstable due to cpufreq or due to unsynced TSCs
 */
static int tsc_unstable;

/* native_sched_clock() is called before tsc_init(), so
   we must start with the TSC soft disabled to prevent
   erroneous rdtsc usage on !cpu_has_tsc processors */
static int tsc_disabled = -1;

/*
 * Scheduler clock - returns current time in nanosec units.
 */
u64 native_sched_clock(void)
{
	u64 this_offset;

	/*
	 * Fall back to jiffies if there's no TSC available:
	 * ( But note that we still use it if the TSC is marked
	 *   unstable. We do this because unlike Time Of Day,
	 *   the scheduler clock tolerates small errors and it's
	 *   very important for it to be as fast as the platform
	 *   can achive it. )
	 */
	if (unlikely(tsc_disabled)) {
		/* No locking but a rare wrong value is not a big deal: */
		return (jiffies_64 - INITIAL_JIFFIES) * (1000000000 / HZ);
	}

	/* read the Time Stamp Counter: */
	rdtscll(this_offset);

	/* return the value in ns */
	return cycles_2_ns(this_offset);
}

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

int check_tsc_unstable(void)
{
	return tsc_unstable;
}
EXPORT_SYMBOL_GPL(check_tsc_unstable);

#ifdef CONFIG_X86_TSC
int __init notsc_setup(char *str)
{
	printk(KERN_WARNING "notsc: Kernel compiled with CONFIG_X86_TSC, "
			"cannot disable TSC completely.\n");
	tsc_disabled = 1;
	return 1;
}
#else
/*
 * disable flag for tsc. Takes effect by clearing the TSC cpu flag
 * in cpu/common.c
 */
int __init notsc_setup(char *str)
{
	setup_clear_cpu_cap(X86_FEATURE_TSC);
	return 1;
}
#endif

__setup("notsc", notsc_setup);

#define MAX_RETRIES     5
#define SMI_TRESHOLD    50000

/*
 * Read TSC and the reference counters. Take care of SMI disturbance
 */
static u64 __init tsc_read_refs(u64 *pm, u64 *hpet)
{
	u64 t1, t2;
	int i;

	for (i = 0; i < MAX_RETRIES; i++) {
		t1 = get_cycles();
		if (hpet)
			*hpet = hpet_readl(HPET_COUNTER) & 0xFFFFFFFF;
		else
			*pm = acpi_pm_read_early();
		t2 = get_cycles();
		if ((t2 - t1) < SMI_TRESHOLD)
			return t2;
	}
	return ULLONG_MAX;
}

/**
 * tsc_calibrate - calibrate the tsc on boot
 */
static unsigned int __init tsc_calibrate(void)
{
	unsigned long flags;
	u64 tsc1, tsc2, tr1, tr2, delta, pm1, pm2, hpet1, hpet2;
	int hpet = is_hpet_enabled();
	unsigned int tsc_khz_val = 0;

	local_irq_save(flags);

	tsc1 = tsc_read_refs(&pm1, hpet ? &hpet1 : NULL);

	outb((inb(0x61) & ~0x02) | 0x01, 0x61);

	outb(0xb0, 0x43);
	outb((CLOCK_TICK_RATE / (1000 / 50)) & 0xff, 0x42);
	outb((CLOCK_TICK_RATE / (1000 / 50)) >> 8, 0x42);
	tr1 = get_cycles();
	while ((inb(0x61) & 0x20) == 0);
	tr2 = get_cycles();

	tsc2 = tsc_read_refs(&pm2, hpet ? &hpet2 : NULL);

	local_irq_restore(flags);

	/*
	 * Preset the result with the raw and inaccurate PIT
	 * calibration value
	 */
	delta = (tr2 - tr1);
	do_div(delta, 50);
	tsc_khz_val = delta;

	/* hpet or pmtimer available ? */
	if (!hpet && !pm1 && !pm2) {
		printk(KERN_INFO "TSC calibrated against PIT\n");
		goto out;
	}

	/* Check, whether the sampling was disturbed by an SMI */
	if (tsc1 == ULLONG_MAX || tsc2 == ULLONG_MAX) {
		printk(KERN_WARNING "TSC calibration disturbed by SMI, "
				"using PIT calibration result\n");
		goto out;
	}

	tsc2 = (tsc2 - tsc1) * 1000000LL;

	if (hpet) {
		printk(KERN_INFO "TSC calibrated against HPET\n");
		if (hpet2 < hpet1)
			hpet2 += 0x100000000ULL;
		hpet2 -= hpet1;
		tsc1 = ((u64)hpet2 * hpet_readl(HPET_PERIOD));
		do_div(tsc1, 1000000);
	} else {
		printk(KERN_INFO "TSC calibrated against PM_TIMER\n");
		if (pm2 < pm1)
			pm2 += (u64)ACPI_PM_OVRRUN;
		pm2 -= pm1;
		tsc1 = pm2 * 1000000000LL;
		do_div(tsc1, PMTMR_TICKS_PER_SEC);
	}

	do_div(tsc2, tsc1);
	tsc_khz_val = tsc2;

out:
	return tsc_khz_val;
}

unsigned long native_calculate_cpu_khz(void)
{
	return tsc_calibrate();
}

#ifdef CONFIG_X86_32
/* Only called from the Powernow K7 cpu freq driver */
int recalibrate_cpu_khz(void)
{
#ifndef CONFIG_SMP
	unsigned long cpu_khz_old = cpu_khz;

	if (cpu_has_tsc) {
		cpu_khz = calculate_cpu_khz();
		tsc_khz = cpu_khz;
		cpu_data(0).loops_per_jiffy =
			cpufreq_scale(cpu_data(0).loops_per_jiffy,
					cpu_khz_old, cpu_khz);
		return 0;
	} else
		return -ENODEV;
#else
	return -ENODEV;
#endif
}

EXPORT_SYMBOL(recalibrate_cpu_khz);

#endif /* CONFIG_X86_32 */

/* Accelerators for sched_clock()
 * 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_khz * 10^3))
 *              ns = cycles * (10^6 / cpu_khz)
 *
 *      Then we use scaling math (suggested by george@mvista.com) to get:
 *              ns = cycles * (10^6 * SC / cpu_khz) / SC
 *              ns = cycles * cyc2ns_scale / SC
 *
 *      And since SC is a constant power of two, we can convert the div
 *  into a shift.
 *
 *  We can use khz divisor instead of mhz to keep a better precision, since
 *  cyc2ns_scale is limited to 10^6 * 2^10, which fits in 32 bits.
 *  (mathieu.desnoyers@polymtl.ca)
 *
 *                      -johnstul@us.ibm.com "math is hard, lets go shopping!"
 */

DEFINE_PER_CPU(unsigned long, cyc2ns);

static void set_cyc2ns_scale(unsigned long cpu_khz, int cpu)
{
	unsigned long long tsc_now, ns_now;
	unsigned long flags, *scale;

	local_irq_save(flags);
	sched_clock_idle_sleep_event();

	scale = &per_cpu(cyc2ns, cpu);

	rdtscll(tsc_now);
	ns_now = __cycles_2_ns(tsc_now);

	if (cpu_khz)
		*scale = (NSEC_PER_MSEC << CYC2NS_SCALE_FACTOR)/cpu_khz;

	sched_clock_idle_wakeup_event(0);
	local_irq_restore(flags);
}

#ifdef CONFIG_CPU_FREQ

/* Frequency scaling support. Adjust the TSC based timer when the cpu frequency
 * changes.
 *
 * RED-PEN: On SMP we assume all CPUs run with the same frequency.  It's
 * not that important because current Opteron setups do not support
 * scaling on SMP anyroads.
 *
 * Should fix up last_tsc too. Currently gettimeofday in the
 * first tick after the change will be slightly wrong.
 */

static unsigned int  ref_freq;
static unsigned long loops_per_jiffy_ref;
static unsigned long tsc_khz_ref;

static int time_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
				void *data)
{
	struct cpufreq_freqs *freq = data;
	unsigned long *lpj, dummy;

	if (cpu_has(&cpu_data(freq->cpu), X86_FEATURE_CONSTANT_TSC))
		return 0;

	lpj = &dummy;
	if (!(freq->flags & CPUFREQ_CONST_LOOPS))
#ifdef CONFIG_SMP
		lpj = &cpu_data(freq->cpu).loops_per_jiffy;
#else
	lpj = &boot_cpu_data.loops_per_jiffy;
#endif

	if (!ref_freq) {
		ref_freq = freq->old;
		loops_per_jiffy_ref = *lpj;
		tsc_khz_ref = tsc_khz;
	}
	if ((val == CPUFREQ_PRECHANGE  && freq->old < freq->new) ||
			(val == CPUFREQ_POSTCHANGE && freq->old > freq->new) ||
			(val == CPUFREQ_RESUMECHANGE)) {
		*lpj = 	cpufreq_scale(loops_per_jiffy_ref, ref_freq, freq->new);

		tsc_khz = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
		if (!(freq->flags & CPUFREQ_CONST_LOOPS))
			mark_tsc_unstable("cpufreq changes");
	}

	set_cyc2ns_scale(tsc_khz_ref, freq->cpu);

	return 0;
}

static struct notifier_block time_cpufreq_notifier_block = {
	.notifier_call  = time_cpufreq_notifier
};

static int __init cpufreq_tsc(void)
{
	cpufreq_register_notifier(&time_cpufreq_notifier_block,
				CPUFREQ_TRANSITION_NOTIFIER);
	return 0;
}

core_initcall(cpufreq_tsc);

#endif /* CONFIG_CPU_FREQ */

/* clocksource code */

static struct clocksource clocksource_tsc;

/*
 * We compare the TSC to the cycle_last value in the clocksource
 * structure to avoid a nasty time-warp. This can be observed in a
 * very small window right after one CPU updated cycle_last under
 * xtime/vsyscall_gtod lock and the other CPU reads a TSC value which
 * is smaller than the cycle_last reference value due to a TSC which
 * is slighty behind. This delta is nowhere else observable, but in
 * that case it results in a forward time jump in the range of hours
 * due to the unsigned delta calculation of the time keeping core
 * code, which is necessary to support wrapping clocksources like pm
 * timer.
 */
static cycle_t read_tsc(void)
{
	cycle_t ret = (cycle_t)get_cycles();

	return ret >= clocksource_tsc.cycle_last ?
		ret : clocksource_tsc.cycle_last;
}

static cycle_t __vsyscall_fn vread_tsc(void)
{
	cycle_t ret = (cycle_t)vget_cycles();

	return ret >= __vsyscall_gtod_data.clock.cycle_last ?
		ret : __vsyscall_gtod_data.clock.cycle_last;
}

static struct clocksource clocksource_tsc = {
	.name                   = "tsc",
	.rating                 = 300,
	.read                   = read_tsc,
	.mask                   = CLOCKSOURCE_MASK(64),
	.shift                  = 22,
	.flags                  = CLOCK_SOURCE_IS_CONTINUOUS |
				  CLOCK_SOURCE_MUST_VERIFY,
#ifdef CONFIG_X86_64
	.vread                  = vread_tsc,
#endif
};

void mark_tsc_unstable(char *reason)
{
	if (!tsc_unstable) {
		tsc_unstable = 1;
		printk("Marking TSC unstable due to %s\n", reason);
		/* Change only the rating, when not registered */
		if (clocksource_tsc.mult)
			clocksource_change_rating(&clocksource_tsc, 0);
		else
			clocksource_tsc.rating = 0;
	}
}

EXPORT_SYMBOL_GPL(mark_tsc_unstable);

static int __init dmi_mark_tsc_unstable(const struct dmi_system_id *d)
{
	printk(KERN_NOTICE "%s detected: marking TSC unstable.\n",
			d->ident);
	tsc_unstable = 1;
	return 0;
}

/* List of systems that have known TSC problems */
static struct dmi_system_id __initdata bad_tsc_dmi_table[] = {
	{
		.callback = dmi_mark_tsc_unstable,
		.ident = "IBM Thinkpad 380XD",
		.matches = {
			DMI_MATCH(DMI_BOARD_VENDOR, "IBM"),
			DMI_MATCH(DMI_BOARD_NAME, "2635FA0"),
		},
	},
	{}
};

/*
 * Geode_LX - the OLPC CPU has a possibly a very reliable TSC
 */
#ifdef CONFIG_MGEODE_LX
/* RTSC counts during suspend */
#define RTSC_SUSP 0x100

static void __init check_geode_tsc_reliable(void)
{
	unsigned long res_low, res_high;

	rdmsr_safe(MSR_GEODE_BUSCONT_CONF0, &res_low, &res_high);
	if (res_low & RTSC_SUSP)
		clocksource_tsc.flags &= ~CLOCK_SOURCE_MUST_VERIFY;
}
#else
static inline void check_geode_tsc_reliable(void) { }
#endif

/*
 * Make an educated guess if the TSC is trustworthy and synchronized
 * over all CPUs.
 */
__cpuinit int unsynchronized_tsc(void)
{
	if (!cpu_has_tsc || tsc_unstable)
		return 1;

#ifdef CONFIG_SMP
	if (apic_is_clustered_box())
		return 1;
#endif

	if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
		return 0;
	/*
	 * Intel systems are normally all synchronized.
	 * Exceptions must mark TSC as unstable:
	 */
	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
		/* assume multi socket systems are not synchronized: */
		if (num_possible_cpus() > 1)
			tsc_unstable = 1;
	}

	return tsc_unstable;
}

static void __init init_tsc_clocksource(void)
{
	clocksource_tsc.mult = clocksource_khz2mult(tsc_khz,
			clocksource_tsc.shift);
	/* lower the rating if we already know its unstable: */
	if (check_tsc_unstable()) {
		clocksource_tsc.rating = 0;
		clocksource_tsc.flags &= ~CLOCK_SOURCE_IS_CONTINUOUS;
	}
	clocksource_register(&clocksource_tsc);
}

void __init tsc_init(void)
{
	u64 lpj;
	int cpu;

	if (!cpu_has_tsc)
		return;

	cpu_khz = calculate_cpu_khz();
	tsc_khz = cpu_khz;

	if (!cpu_khz) {
		mark_tsc_unstable("could not calculate TSC khz");
		return;
	}

#ifdef CONFIG_X86_64
	if (cpu_has(&boot_cpu_data, X86_FEATURE_CONSTANT_TSC) &&
			(boot_cpu_data.x86_vendor == X86_VENDOR_AMD))
		cpu_khz = calibrate_cpu();
#endif

	lpj = ((u64)tsc_khz * 1000);
	do_div(lpj, HZ);
	lpj_fine = lpj;

	printk("Detected %lu.%03lu MHz processor.\n",
			(unsigned long)cpu_khz / 1000,
			(unsigned long)cpu_khz % 1000);

	/*
	 * Secondary CPUs do not run through tsc_init(), so set up
	 * all the scale factors for all CPUs, assuming the same
	 * speed as the bootup CPU. (cpufreq notifiers will fix this
	 * up if their speed diverges)
	 */
	for_each_possible_cpu(cpu)
		set_cyc2ns_scale(cpu_khz, cpu);

	if (tsc_disabled > 0)
		return;

	/* now allow native_sched_clock() to use rdtsc */
	tsc_disabled = 0;

	use_tsc_delay();
	/* Check and install the TSC clocksource */
	dmi_check_system(bad_tsc_dmi_table);

	if (unsynchronized_tsc())
		mark_tsc_unstable("TSCs unsynchronized");

	check_geode_tsc_reliable();
	init_tsc_clocksource();
}
Commit	Line	Data
bfc0f594	1	#include <linux/kernel.h>
0ef95533 AK	2	#include <linux/sched.h>
	3	#include <linux/init.h>
	4	#include <linux/module.h>
	5	#include <linux/timer.h>
bfc0f594	6	#include <linux/acpi_pmtmr.h>
2dbe06fa	7	#include <linux/cpufreq.h>
8fbbc4b4 AK	8	#include <linux/dmi.h>
	9	#include <linux/delay.h>
	10	#include <linux/clocksource.h>
	11	#include <linux/percpu.h>
bfc0f594 AK	12
bfc0f594 AK	13	#include <asm/hpet.h>
8fbbc4b4 AK	14	#include <asm/timer.h>
	15	#include <asm/vgtod.h>
	16	#include <asm/time.h>
	17	#include <asm/delay.h>
0ef95533 AK	18
	19	unsigned int cpu_khz; /* TSC clocks / usec, not used here */
	20	EXPORT_SYMBOL(cpu_khz);
	21	unsigned int tsc_khz;
	22	EXPORT_SYMBOL(tsc_khz);
	23
	24	/*
	25	* TSC can be unstable due to cpufreq or due to unsynced TSCs
	26	*/
8fbbc4b4	27	static int tsc_unstable;
0ef95533 AK	28
	29	/* native_sched_clock() is called before tsc_init(), so
	30	we must start with the TSC soft disabled to prevent
	31	erroneous rdtsc usage on !cpu_has_tsc processors */
8fbbc4b4	32	static int tsc_disabled = -1;
0ef95533 AK	33
	34	/*
	35	* Scheduler clock - returns current time in nanosec units.
	36	*/
	37	u64 native_sched_clock(void)
	38	{
	39	u64 this_offset;
	40
	41	/*
	42	* Fall back to jiffies if there's no TSC available:
	43	* ( But note that we still use it if the TSC is marked
	44	* unstable. We do this because unlike Time Of Day,
	45	* the scheduler clock tolerates small errors and it's
	46	* very important for it to be as fast as the platform
	47	* can achive it. )
	48	*/
	49	if (unlikely(tsc_disabled)) {
	50	/* No locking but a rare wrong value is not a big deal: */
	51	return (jiffies_64 - INITIAL_JIFFIES) * (1000000000 / HZ);
	52	}
	53
	54	/* read the Time Stamp Counter: */
	55	rdtscll(this_offset);
	56
	57	/* return the value in ns */
	58	return cycles_2_ns(this_offset);
	59	}
	60
	61	/* We need to define a real function for sched_clock, to override the
	62	weak default version */
	63	#ifdef CONFIG_PARAVIRT
	64	unsigned long long sched_clock(void)
	65	{
	66	return paravirt_sched_clock();
	67	}
	68	#else
	69	unsigned long long
	70	sched_clock(void) __attribute__((alias("native_sched_clock")));
	71	#endif
	72
	73	int check_tsc_unstable(void)
	74	{
	75	return tsc_unstable;
	76	}
	77	EXPORT_SYMBOL_GPL(check_tsc_unstable);
	78
	79	#ifdef CONFIG_X86_TSC
	80	int __init notsc_setup(char *str)
	81	{
	82	printk(KERN_WARNING "notsc: Kernel compiled with CONFIG_X86_TSC, "
	83	"cannot disable TSC completely.\n");
	84	tsc_disabled = 1;
	85	return 1;
	86	}
	87	#else
	88	/*
	89	* disable flag for tsc. Takes effect by clearing the TSC cpu flag
	90	* in cpu/common.c
	91	*/
	92	int __init notsc_setup(char *str)
	93	{
	94	setup_clear_cpu_cap(X86_FEATURE_TSC);
	95	return 1;
	96	}
97	#endif
98
99	__setup("notsc", notsc_setup);
bfc0f594 AK	100
	101	#define MAX_RETRIES 5
	102	#define SMI_TRESHOLD 50000
	103
	104	/*
	105	* Read TSC and the reference counters. Take care of SMI disturbance
	106	*/
	107	static u64 __init tsc_read_refs(u64 pm, u64 hpet)
	108	{
	109	u64 t1, t2;
	110	int i;
	111
	112	for (i = 0; i < MAX_RETRIES; i++) {
	113	t1 = get_cycles();
	114	if (hpet)
	115	*hpet = hpet_readl(HPET_COUNTER) & 0xFFFFFFFF;
	116	else
	117	*pm = acpi_pm_read_early();
	118	t2 = get_cycles();
	119	if ((t2 - t1) < SMI_TRESHOLD)
	120	return t2;
	121	}
	122	return ULLONG_MAX;
	123	}
	124
	125	/**
	126	* tsc_calibrate - calibrate the tsc on boot
	127	*/
	128	static unsigned int __init tsc_calibrate(void)
	129	{
	130	unsigned long flags;
	131	u64 tsc1, tsc2, tr1, tr2, delta, pm1, pm2, hpet1, hpet2;
	132	int hpet = is_hpet_enabled();
	133	unsigned int tsc_khz_val = 0;
	134
	135	local_irq_save(flags);
	136
	137	tsc1 = tsc_read_refs(&pm1, hpet ? &hpet1 : NULL);
	138
	139	outb((inb(0x61) & ~0x02) \| 0x01, 0x61);
	140
	141	outb(0xb0, 0x43);
	142	outb((CLOCK_TICK_RATE / (1000 / 50)) & 0xff, 0x42);
	143	outb((CLOCK_TICK_RATE / (1000 / 50)) >> 8, 0x42);
	144	tr1 = get_cycles();
	145	while ((inb(0x61) & 0x20) == 0);
	146	tr2 = get_cycles();
	147
	148	tsc2 = tsc_read_refs(&pm2, hpet ? &hpet2 : NULL);
	149
	150	local_irq_restore(flags);
	151
	152	/*
	153	* Preset the result with the raw and inaccurate PIT
	154	* calibration value
	155	*/
	156	delta = (tr2 - tr1);
	157	do_div(delta, 50);
	158	tsc_khz_val = delta;
	159
	160	/* hpet or pmtimer available ? */
	161	if (!hpet && !pm1 && !pm2) {
	162	printk(KERN_INFO "TSC calibrated against PIT\n");
	163	goto out;
164	}
165
166	/* Check, whether the sampling was disturbed by an SMI */
167	if (tsc1 == ULLONG_MAX \|\| tsc2 == ULLONG_MAX) {
168	printk(KERN_WARNING "TSC calibration disturbed by SMI, "
169	"using PIT calibration result\n");
170	goto out;
171	}
172
173	tsc2 = (tsc2 - tsc1) * 1000000LL;
174
175	if (hpet) {
176	printk(KERN_INFO "TSC calibrated against HPET\n");
177	if (hpet2 < hpet1)
178	hpet2 += 0x100000000ULL;
179	hpet2 -= hpet1;
180	tsc1 = ((u64)hpet2 * hpet_readl(HPET_PERIOD));
181	do_div(tsc1, 1000000);
182	} else {
183	printk(KERN_INFO "TSC calibrated against PM_TIMER\n");
184	if (pm2 < pm1)
185	pm2 += (u64)ACPI_PM_OVRRUN;
186	pm2 -= pm1;
187	tsc1 = pm2 * 1000000000LL;
188	do_div(tsc1, PMTMR_TICKS_PER_SEC);
189	}
190
191	do_div(tsc2, tsc1);
192	tsc_khz_val = tsc2;
193
194	out:
195	return tsc_khz_val;
196	}
197
198	unsigned long native_calculate_cpu_khz(void)
199	{
200	return tsc_calibrate();
201	}
202
203	#ifdef CONFIG_X86_32
204	/* Only called from the Powernow K7 cpu freq driver */
205	int recalibrate_cpu_khz(void)
206	{
207	#ifndef CONFIG_SMP
208	unsigned long cpu_khz_old = cpu_khz;
209
210	if (cpu_has_tsc) {
211	cpu_khz = calculate_cpu_khz();
212	tsc_khz = cpu_khz;
213	cpu_data(0).loops_per_jiffy =
214	cpufreq_scale(cpu_data(0).loops_per_jiffy,
215	cpu_khz_old, cpu_khz);
216	return 0;
217	} else
218	return -ENODEV;
219	#else
220	return -ENODEV;
221	#endif
222	}
223
224	EXPORT_SYMBOL(recalibrate_cpu_khz);
225
226	#endif /* CONFIG_X86_32 */
2dbe06fa AK	227
	228	/* Accelerators for sched_clock()
	229	* convert from cycles(64bits) => nanoseconds (64bits)
	230	* basic equation:
	231	* ns = cycles / (freq / ns_per_sec)
	232	* ns = cycles * (ns_per_sec / freq)
	233	* ns = cycles * (10^9 / (cpu_khz * 10^3))
	234	* ns = cycles * (10^6 / cpu_khz)
	235	*
	236	* Then we use scaling math (suggested by george@mvista.com) to get:
	237	* ns = cycles * (10^6 * SC / cpu_khz) / SC
	238	* ns = cycles * cyc2ns_scale / SC
	239	*
	240	* And since SC is a constant power of two, we can convert the div
	241	* into a shift.
	242	*
	243	* We can use khz divisor instead of mhz to keep a better precision, since
	244	* cyc2ns_scale is limited to 10^6 * 2^10, which fits in 32 bits.
	245	* (mathieu.desnoyers@polymtl.ca)
	246	*
	247	* -johnstul@us.ibm.com "math is hard, lets go shopping!"
	248	*/
	249
	250	DEFINE_PER_CPU(unsigned long, cyc2ns);
	251
8fbbc4b4	252	static void set_cyc2ns_scale(unsigned long cpu_khz, int cpu)
2dbe06fa AK	253	{
	254	unsigned long long tsc_now, ns_now;
	255	unsigned long flags, *scale;
	256
	257	local_irq_save(flags);
	258	sched_clock_idle_sleep_event();
	259
	260	scale = &per_cpu(cyc2ns, cpu);
	261
	262	rdtscll(tsc_now);
	263	ns_now = __cycles_2_ns(tsc_now);
	264
	265	if (cpu_khz)
	266	*scale = (NSEC_PER_MSEC << CYC2NS_SCALE_FACTOR)/cpu_khz;
	267
	268	sched_clock_idle_wakeup_event(0);
	269	local_irq_restore(flags);
	270	}
	271
	272	#ifdef CONFIG_CPU_FREQ
	273
	274	/* Frequency scaling support. Adjust the TSC based timer when the cpu frequency
	275	* changes.
	276	*
	277	* RED-PEN: On SMP we assume all CPUs run with the same frequency. It's
	278	* not that important because current Opteron setups do not support
	279	* scaling on SMP anyroads.
	280	*
	281	* Should fix up last_tsc too. Currently gettimeofday in the
	282	* first tick after the change will be slightly wrong.
	283	*/
	284
	285	static unsigned int ref_freq;
	286	static unsigned long loops_per_jiffy_ref;
	287	static unsigned long tsc_khz_ref;
	288
	289	static int time_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
	290	void *data)
	291	{
	292	struct cpufreq_freqs *freq = data;
	293	unsigned long *lpj, dummy;
	294
	295	if (cpu_has(&cpu_data(freq->cpu), X86_FEATURE_CONSTANT_TSC))
	296	return 0;
	297
	298	lpj = &dummy;
	299	if (!(freq->flags & CPUFREQ_CONST_LOOPS))
	300	#ifdef CONFIG_SMP
	301	lpj = &cpu_data(freq->cpu).loops_per_jiffy;
	302	#else
	303	lpj = &boot_cpu_data.loops_per_jiffy;
	304	#endif
	305
	306	if (!ref_freq) {
	307	ref_freq = freq->old;
	308	loops_per_jiffy_ref = *lpj;
	309	tsc_khz_ref = tsc_khz;
	310	}
	311	if ((val == CPUFREQ_PRECHANGE && freq->old < freq->new) \|\|
	312	(val == CPUFREQ_POSTCHANGE && freq->old > freq->new) \|\|
	313	(val == CPUFREQ_RESUMECHANGE)) {
	314	*lpj = cpufreq_scale(loops_per_jiffy_ref, ref_freq, freq->new);
	315
	316	tsc_khz = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
317	if (!(freq->flags & CPUFREQ_CONST_LOOPS))
318	mark_tsc_unstable("cpufreq changes");
319	}
320
321	set_cyc2ns_scale(tsc_khz_ref, freq->cpu);
322
323	return 0;
324	}
325
326	static struct notifier_block time_cpufreq_notifier_block = {
327	.notifier_call = time_cpufreq_notifier
328	};
329
330	static int __init cpufreq_tsc(void)
331	{
332	cpufreq_register_notifier(&time_cpufreq_notifier_block,
333	CPUFREQ_TRANSITION_NOTIFIER);
334	return 0;
335	}
336
337	core_initcall(cpufreq_tsc);
338
339	#endif /* CONFIG_CPU_FREQ */
8fbbc4b4 AK	340
	341	/* clocksource code */
	342
	343	static struct clocksource clocksource_tsc;
	344
	345	/*
	346	* We compare the TSC to the cycle_last value in the clocksource
	347	* structure to avoid a nasty time-warp. This can be observed in a
	348	* very small window right after one CPU updated cycle_last under
	349	* xtime/vsyscall_gtod lock and the other CPU reads a TSC value which
	350	* is smaller than the cycle_last reference value due to a TSC which
	351	* is slighty behind. This delta is nowhere else observable, but in
	352	* that case it results in a forward time jump in the range of hours
	353	* due to the unsigned delta calculation of the time keeping core
	354	* code, which is necessary to support wrapping clocksources like pm
	355	* timer.
	356	*/
	357	static cycle_t read_tsc(void)
	358	{
	359	cycle_t ret = (cycle_t)get_cycles();
	360
	361	return ret >= clocksource_tsc.cycle_last ?
	362	ret : clocksource_tsc.cycle_last;
	363	}
	364
	365	static cycle_t __vsyscall_fn vread_tsc(void)
	366	{
	367	cycle_t ret = (cycle_t)vget_cycles();
	368
	369	return ret >= __vsyscall_gtod_data.clock.cycle_last ?
	370	ret : __vsyscall_gtod_data.clock.cycle_last;
	371	}
	372
	373	static struct clocksource clocksource_tsc = {
	374	.name = "tsc",
	375	.rating = 300,
	376	.read = read_tsc,
	377	.mask = CLOCKSOURCE_MASK(64),
	378	.shift = 22,
	379	.flags = CLOCK_SOURCE_IS_CONTINUOUS \|
	380	CLOCK_SOURCE_MUST_VERIFY,
	381	#ifdef CONFIG_X86_64
	382	.vread = vread_tsc,
	383	#endif
	384	};
	385
	386	void mark_tsc_unstable(char *reason)
	387	{
	388	if (!tsc_unstable) {
	389	tsc_unstable = 1;
	390	printk("Marking TSC unstable due to %s\n", reason);
	391	/* Change only the rating, when not registered */
	392	if (clocksource_tsc.mult)
	393	clocksource_change_rating(&clocksource_tsc, 0);
	394	else
	395	clocksource_tsc.rating = 0;
	396	}
	397	}
	398
	399	EXPORT_SYMBOL_GPL(mark_tsc_unstable);
	400
	401	static int __init dmi_mark_tsc_unstable(const struct dmi_system_id *d)
	402	{
	403	printk(KERN_NOTICE "%s detected: marking TSC unstable.\n",
404	d->ident);
405	tsc_unstable = 1;
406	return 0;
407	}
408
409	/* List of systems that have known TSC problems */
410	static struct dmi_system_id __initdata bad_tsc_dmi_table[] = {
411	{
412	.callback = dmi_mark_tsc_unstable,
413	.ident = "IBM Thinkpad 380XD",
414	.matches = {
415	DMI_MATCH(DMI_BOARD_VENDOR, "IBM"),
416	DMI_MATCH(DMI_BOARD_NAME, "2635FA0"),
417	},
418	},
419	{}
420	};
421
422	/*
423	* Geode_LX - the OLPC CPU has a possibly a very reliable TSC
424	*/
425	#ifdef CONFIG_MGEODE_LX
426	/* RTSC counts during suspend */
427	#define RTSC_SUSP 0x100
428
429	static void __init check_geode_tsc_reliable(void)
430	{
431	unsigned long res_low, res_high;
432
433	rdmsr_safe(MSR_GEODE_BUSCONT_CONF0, &res_low, &res_high);
434	if (res_low & RTSC_SUSP)
435	clocksource_tsc.flags &= ~CLOCK_SOURCE_MUST_VERIFY;
436	}
437	#else
438	static inline void check_geode_tsc_reliable(void) { }
439	#endif
440
441	/*
442	* Make an educated guess if the TSC is trustworthy and synchronized
443	* over all CPUs.
444	*/
445	__cpuinit int unsynchronized_tsc(void)
446	{
447	if (!cpu_has_tsc \|\| tsc_unstable)
448	return 1;
449
450	#ifdef CONFIG_SMP
451	if (apic_is_clustered_box())
452	return 1;
453	#endif
454
455	if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
456	return 0;
457	/*
458	* Intel systems are normally all synchronized.
459	* Exceptions must mark TSC as unstable:
460	*/
461	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
462	/* assume multi socket systems are not synchronized: */
463	if (num_possible_cpus() > 1)
464	tsc_unstable = 1;
465	}
466
467	return tsc_unstable;
468	}
469
470	static void __init init_tsc_clocksource(void)
471	{
472	clocksource_tsc.mult = clocksource_khz2mult(tsc_khz,
473	clocksource_tsc.shift);
474	/* lower the rating if we already know its unstable: */
475	if (check_tsc_unstable()) {
476	clocksource_tsc.rating = 0;
477	clocksource_tsc.flags &= ~CLOCK_SOURCE_IS_CONTINUOUS;
478	}
479	clocksource_register(&clocksource_tsc);
480	}
481
482	void __init tsc_init(void)
483	{
484	u64 lpj;
485	int cpu;
486
487	if (!cpu_has_tsc)
488	return;
489
490	cpu_khz = calculate_cpu_khz();
491	tsc_khz = cpu_khz;
492
493	if (!cpu_khz) {
494	mark_tsc_unstable("could not calculate TSC khz");
495	return;
496	}
497
498	#ifdef CONFIG_X86_64
499	if (cpu_has(&boot_cpu_data, X86_FEATURE_CONSTANT_TSC) &&
500	(boot_cpu_data.x86_vendor == X86_VENDOR_AMD))
501	cpu_khz = calibrate_cpu();
502	#endif
503
504	lpj = ((u64)tsc_khz * 1000);
505	do_div(lpj, HZ);
506	lpj_fine = lpj;
507
508	printk("Detected %lu.%03lu MHz processor.\n",
509	(unsigned long)cpu_khz / 1000,
510	(unsigned long)cpu_khz % 1000);
511
512	/*
513	* Secondary CPUs do not run through tsc_init(), so set up
514	* all the scale factors for all CPUs, assuming the same
515	* speed as the bootup CPU. (cpufreq notifiers will fix this
516	* up if their speed diverges)
517	*/
518	for_each_possible_cpu(cpu)
519	set_cyc2ns_scale(cpu_khz, cpu);
520
521	if (tsc_disabled > 0)
522	return;
523
524	/* now allow native_sched_clock() to use rdtsc */
525	tsc_disabled = 0;
526
527	use_tsc_delay();
528	/* Check and install the TSC clocksource */
529	dmi_check_system(bad_tsc_dmi_table);
530
531	if (unsynchronized_tsc())
532	mark_tsc_unstable("TSCs unsynchronized");
533
534	check_geode_tsc_reliable();
535	init_tsc_clocksource();
536	}
537